Merge git://git.kernel.org/pub/scm/linux/kernel/git/rusty/linux-2.6-lguest

* git://git.kernel.org/pub/scm/linux/kernel/git/rusty/linux-2.6-lguest: (45 commits)
  Use "struct boot_params" in example launcher
  Loading bzImage directly.
  Revert lguest magic and use hook in head.S
  Update lguest documentation to reflect the new virtual block device name.
  generalize lgread_u32/lgwrite_u32.
  Example launcher handle guests not being ready for input
  Update example launcher for virtio
  Lguest support for Virtio
  Remove old lguest I/O infrrasructure.
  Remove old lguest bus and drivers.
  Virtio helper routines for a descriptor ringbuffer implementation
  Module autoprobing support for virtio drivers.
  Virtio console driver
  Block driver using virtio.
  Net driver using virtio
  Virtio interface
  Boot with virtual == physical to get closer to native Linux.
  Allow guest to specify syscall vector to use.
  Rename "cr3" to "gpgdir" to avoid x86-specific naming.
  Pagetables to use normal kernel types
  ...

Documentation/lguest/Makefile

@@ -1,28 +1,8 @@
 # This creates the demonstration utility "lguest" which runs a Linux guest.
-
-# For those people that have a separate object dir, look there for .config
-KBUILD_OUTPUT := ../..
-ifdef O
-  ifeq ("$(origin O)", "command line")
-    KBUILD_OUTPUT := $(O)
-  endif
-endif
-# We rely on CONFIG_PAGE_OFFSET to know where to put lguest binary.
-include $(KBUILD_OUTPUT)/.config
-LGUEST_GUEST_TOP := ($(CONFIG_PAGE_OFFSET) - 0x08000000)
-
-CFLAGS:=-Wall -Wmissing-declarations -Wmissing-prototypes -O3 -Wl,-T,lguest.lds
+CFLAGS:=-Wall -Wmissing-declarations -Wmissing-prototypes -O3 -I../../include
 LDLIBS:=-lz
-# Removing this works for some versions of ld.so (eg. Ubuntu Feisty) and
-# not others (eg. FC7).
-LDFLAGS+=-static
-all: lguest.lds lguest
 
-# The linker script on x86 is so complex the only way of creating one
-# which will link our binary in the right place is to mangle the
-# default one.
-lguest.lds:
-	$(LD) --verbose | awk '/^==========/ { PRINT=1; next; } /SIZEOF_HEADERS/ { gsub(/0x[0-9A-F]*/, "$(LGUEST_GUEST_TOP)") } { if (PRINT) print $$0; }' > $@
+all: lguest
 
 clean:
-	rm -f lguest.lds lguest
+	rm -f lguest

Documentation/lguest/lguest.txt

@@ -6,7 +6,7 @@ Lguest is designed to be a minimal hypervisor for the Linux kernel, for
 Linux developers and users to experiment with virtualization with the
 minimum of complexity.  Nonetheless, it should have sufficient
 features to make it useful for specific tasks, and, of course, you are
-encouraged to fork and enhance it.
+encouraged to fork and enhance it (see drivers/lguest/README).
 
 Features:
 
@@ -23,19 +23,30 @@ Developer features:
 
 Running Lguest:
 
-- Lguest runs the same kernel as guest and host.  You can configure
-  them differently, but usually it's easiest not to.
+- The easiest way to run lguest is to use same kernel as guest and host.
+  You can configure them differently, but usually it's easiest not to.
 
   You will need to configure your kernel with the following options:
 
-  CONFIG_HIGHMEM64G=n ("High Memory Support" "64GB")[1]
-  CONFIG_TUN=y/m ("Universal TUN/TAP device driver support")
-  CONFIG_EXPERIMENTAL=y ("Prompt for development and/or incomplete code/drivers")
-  CONFIG_PARAVIRT=y ("Paravirtualization support (EXPERIMENTAL)")
-  CONFIG_LGUEST=y/m ("Linux hypervisor example code")
-
-  and I recommend:
-  CONFIG_HZ=100 ("Timer frequency")[2]
+  "General setup":
+     "Prompt for development and/or incomplete code/drivers" = Y
+        (CONFIG_EXPERIMENTAL=y)
+
+  "Processor type and features":
+     "Paravirtualized guest support" = Y
+        "Lguest guest support" = Y
+     "High Memory Support" = off/4GB
+     "Alignment value to which kernel should be aligned" = 0x100000
+        (CONFIG_PARAVIRT=y, CONFIG_LGUEST_GUEST=y, CONFIG_HIGHMEM64G=n and
+         CONFIG_PHYSICAL_ALIGN=0x100000)
+
+  "Device Drivers":
+     "Network device support"
+        "Universal TUN/TAP device driver support" = M/Y
+           (CONFIG_TUN=m)
+     "Virtualization"
+        "Linux hypervisor example code" = M/Y
+           (CONFIG_LGUEST=m)
 
 - A tool called "lguest" is available in this directory: type "make"
   to build it.  If you didn't build your kernel in-tree, use "make
@@ -51,14 +62,17 @@ Running Lguest:
 	  dd if=/dev/zero of=rootfile bs=1M count=2048
 	  qemu -cdrom image.iso -hda rootfile -net user -net nic -boot d
 
+  Make sure that you install a getty on /dev/hvc0 if you want to log in on the
+  console!
+
 - "modprobe lg" if you built it as a module.
 
 - Run an lguest as root:
 
-      Documentation/lguest/lguest 64m vmlinux --tunnet=192.168.19.1 --block=rootfile root=/dev/lgba
+      Documentation/lguest/lguest 64 vmlinux --tunnet=192.168.19.1 --block=rootfile root=/dev/vda
 
    Explanation:
-    64m: the amount of memory to use.
+    64: the amount of memory to use, in MB.
 
     vmlinux: the kernel image found in the top of your build directory.  You
        can also use a standard bzImage.
@@ -66,10 +80,10 @@ Running Lguest:
     --tunnet=192.168.19.1: configures a "tap" device for networking with this
        IP address.
 
-    --block=rootfile: a file or block device which becomes /dev/lgba
+    --block=rootfile: a file or block device which becomes /dev/vda
        inside the guest.
 
-    root=/dev/lgba: this (and anything else on the command line) are
+    root=/dev/vda: this (and anything else on the command line) are
        kernel boot parameters.
 
 - Configuring networking.  I usually have the host masquerade, using
@@ -99,31 +113,7 @@ Running Lguest:
   "--sharenet=<filename>": any two guests using the same file are on
   the same network.  This file is created if it does not exist.
 
-Lguest I/O model:
-
-Lguest uses a simplified DMA model plus shared memory for I/O.  Guests
-can communicate with each other if they share underlying memory
-(usually by the lguest program mmaping the same file), but they can
-use any non-shared memory to communicate with the lguest process.
-
-Guests can register DMA buffers at any key (must be a valid physical
-address) using the LHCALL_BIND_DMA(key, dmabufs, num<<8|irq)
-hypercall.  "dmabufs" is the physical address of an array of "num"
-"struct lguest_dma": each contains a used_len, and an array of
-physical addresses and lengths.  When a transfer occurs, the
-"used_len" field of one of the buffers which has used_len 0 will be
-set to the length transferred and the irq will fire.
+There is a helpful mailing list at http://ozlabs.org/mailman/listinfo/lguest
 
-Using an irq value of 0 unbinds the dma buffers.
-
-To send DMA, the LHCALL_SEND_DMA(key, dma_physaddr) hypercall is used,
-and the bytes used is written to the used_len field.  This can be 0 if
-noone else has bound a DMA buffer to that key or some other error.
-DMA buffers bound by the same guest are ignored.
-
-Cheers!
+Good luck!
 Rusty Russell rusty@rustcorp.com.au.
-
-[1] These are on various places on the TODO list, waiting for you to
-    get annoyed enough at the limitation to fix it.
-[2] Lguest is not yet tickless when idle.  See [1].

arch/i386/Kconfig

@@ -227,28 +227,40 @@ config SCHED_NO_NO_OMIT_FRAME_POINTER
 	  If in doubt, say "Y".
 
 config PARAVIRT
-	bool "Paravirtualization support (EXPERIMENTAL)"
-	depends on EXPERIMENTAL
+	bool
 	depends on !(X86_VISWS || X86_VOYAGER)
 	help
-	  Paravirtualization is a way of running multiple instances of
-	  Linux on the same machine, under a hypervisor.  This option
-	  changes the kernel so it can modify itself when it is run
-	  under a hypervisor, improving performance significantly.
-	  However, when run without a hypervisor the kernel is
-	  theoretically slower.  If in doubt, say N.
+	  This changes the kernel so it can modify itself when it is run
+	  under a hypervisor, potentially improving performance significantly
+	  over full virtualization.  However, when run without a hypervisor
+	  the kernel is theoretically slower and slightly larger.
+
+menuconfig PARAVIRT_GUEST
+	bool "Paravirtualized guest support"
+	help
+	  Say Y here to get to see options related to running Linux under
+	  various hypervisors.  This option alone does not add any kernel code.
+
+	  If you say N, all options in this submenu will be skipped and disabled.
+
+if PARAVIRT_GUEST
 
 source "arch/x86/xen/Kconfig"
 
 config VMI
-	bool "VMI Paravirt-ops support"
-	depends on PARAVIRT
+	bool "VMI Guest support"
+	select PARAVIRT
+	depends on !(X86_VISWS || X86_VOYAGER)
 	help
 	  VMI provides a paravirtualized interface to the VMware ESX server
 	  (it could be used by other hypervisors in theory too, but is not
 	  at the moment), by linking the kernel to a GPL-ed ROM module
 	  provided by the hypervisor.
 
+source "arch/x86/lguest/Kconfig"
+
+endif
+
 config ACPI_SRAT
 	bool
 	default y

arch/i386/Makefile

@@ -99,6 +99,9 @@ core-$(CONFIG_X86_ES7000)	:= arch/x86/mach-es7000/
 # Xen paravirtualization support
 core-$(CONFIG_XEN)		+= arch/x86/xen/
 
+# lguest paravirtualization support
+core-$(CONFIG_LGUEST_GUEST)	+= arch/x86/lguest/
+
 # default subarch .h files
 mflags-y += -Iinclude/asm-x86/mach-default
 

arch/x86/kernel/asm-offsets_32.c

@@ -136,6 +136,7 @@ void foo(void)
 #ifdef CONFIG_LGUEST_GUEST
 	BLANK();
 	OFFSET(LGUEST_DATA_irq_enabled, lguest_data, irq_enabled);
+	OFFSET(LGUEST_DATA_pgdir, lguest_data, pgdir);
 	OFFSET(LGUEST_PAGES_host_gdt_desc, lguest_pages, state.host_gdt_desc);
 	OFFSET(LGUEST_PAGES_host_idt_desc, lguest_pages, state.host_idt_desc);
 	OFFSET(LGUEST_PAGES_host_cr3, lguest_pages, state.host_cr3);

arch/x86/lguest/Kconfig

@@ -0,0 +1,14 @@
+config LGUEST_GUEST
+	bool "Lguest guest support"
+	select PARAVIRT
+	depends on !X86_PAE
+	select VIRTIO
+	select VIRTIO_RING
+	select VIRTIO_CONSOLE
+	help
+	  Lguest is a tiny in-kernel hypervisor.  Selecting this will
+	  allow your kernel to boot under lguest.  This option will increase
+	  your kernel size by about 6k.  If in doubt, say N.
+
+	  If you say Y here, make sure you say Y (or M) to the virtio block
+	  and net drivers which lguest needs.

arch/x86/lguest/Makefile

@@ -0,0 +1 @@
+obj-y		:= i386_head.o boot.o

drivers/lguest/lguest.c → arch/x86/lguest/boot.c

@@ -55,7 +55,7 @@
 #include <linux/clockchips.h>
 #include <linux/lguest.h>
 #include <linux/lguest_launcher.h>
-#include <linux/lguest_bus.h>
+#include <linux/virtio_console.h>
 #include <asm/paravirt.h>
 #include <asm/param.h>
 #include <asm/page.h>
@@ -65,6 +65,7 @@
 #include <asm/e820.h>
 #include <asm/mce.h>
 #include <asm/io.h>
+#include <asm/i387.h>
 
 /*G:010 Welcome to the Guest!
  *
@@ -85,9 +86,10 @@ struct lguest_data lguest_data = {
 	.hcall_status = { [0 ... LHCALL_RING_SIZE-1] = 0xFF },
 	.noirq_start = (u32)lguest_noirq_start,
 	.noirq_end = (u32)lguest_noirq_end,
+	.kernel_address = PAGE_OFFSET,
 	.blocked_interrupts = { 1 }, /* Block timer interrupts */
+	.syscall_vec = SYSCALL_VECTOR,
 };
-struct lguest_device_desc *lguest_devices;
 static cycle_t clock_base;
 
 /*G:035 Notice the lazy_hcall() above, rather than hcall().  This is our first
@@ -146,10 +148,10 @@ void async_hcall(unsigned long call,
 		/* Table full, so do normal hcall which will flush table. */
 		hcall(call, arg1, arg2, arg3);
 	} else {
-		lguest_data.hcalls[next_call].eax = call;
-		lguest_data.hcalls[next_call].edx = arg1;
-		lguest_data.hcalls[next_call].ebx = arg2;
-		lguest_data.hcalls[next_call].ecx = arg3;
+		lguest_data.hcalls[next_call].arg0 = call;
+		lguest_data.hcalls[next_call].arg1 = arg1;
+		lguest_data.hcalls[next_call].arg2 = arg2;
+		lguest_data.hcalls[next_call].arg3 = arg3;
 		/* Arguments must all be written before we mark it to go */
 		wmb();
 		lguest_data.hcall_status[next_call] = 0;
@@ -160,46 +162,6 @@ void async_hcall(unsigned long call,
 }
 /*:*/
 
-/* Wrappers for the SEND_DMA and BIND_DMA hypercalls.  This is mainly because
- * Jeff Garzik complained that __pa() should never appear in drivers, and this
- * helps remove most of them.   But also, it wraps some ugliness. */
-void lguest_send_dma(unsigned long key, struct lguest_dma *dma)
-{
-	/* The hcall might not write this if something goes wrong */
-	dma->used_len = 0;
-	hcall(LHCALL_SEND_DMA, key, __pa(dma), 0);
-}
-
-int lguest_bind_dma(unsigned long key, struct lguest_dma *dmas,
-		    unsigned int num, u8 irq)
-{
-	/* This is the only hypercall which actually wants 5 arguments, and we
-	 * only support 4.  Fortunately the interrupt number is always less
-	 * than 256, so we can pack it with the number of dmas in the final
-	 * argument.  */
-	if (!hcall(LHCALL_BIND_DMA, key, __pa(dmas), (num << 8) | irq))
-		return -ENOMEM;
-	return 0;
-}
-
-/* Unbinding is the same hypercall as binding, but with 0 num & irq. */
-void lguest_unbind_dma(unsigned long key, struct lguest_dma *dmas)
-{
-	hcall(LHCALL_BIND_DMA, key, __pa(dmas), 0);
-}
-
-/* For guests, device memory can be used as normal memory, so we cast away the
- * __iomem to quieten sparse. */
-void *lguest_map(unsigned long phys_addr, unsigned long pages)
-{
-	return (__force void *)ioremap(phys_addr, PAGE_SIZE*pages);
-}
-
-void lguest_unmap(void *addr)
-{
-	iounmap((__force void __iomem *)addr);
-}
-
 /*G:033
  * Here are our first native-instruction replacements: four functions for
  * interrupt control.
@@ -680,6 +642,7 @@ static struct clocksource lguest_clock = {
 	.mask		= CLOCKSOURCE_MASK(64),
 	.mult		= 1 << 22,
 	.shift		= 22,
+	.flags		= CLOCK_SOURCE_IS_CONTINUOUS,
 };
 
 /* The "scheduler clock" is just our real clock, adjusted to start at zero */
@@ -761,11 +724,9 @@ static void lguest_time_init(void)
 	 * the TSC, otherwise it's a dumb nanosecond-resolution clock.  Either
 	 * way, the "rating" is initialized so high that it's always chosen
 	 * over any other clocksource. */
-	if (lguest_data.tsc_khz) {
+	if (lguest_data.tsc_khz)
 		lguest_clock.mult = clocksource_khz2mult(lguest_data.tsc_khz,
 							 lguest_clock.shift);
-		lguest_clock.flags = CLOCK_SOURCE_IS_CONTINUOUS;
-	}
 	clock_base = lguest_clock_read();
 	clocksource_register(&lguest_clock);
 
@@ -889,6 +850,23 @@ static __init char *lguest_memory_setup(void)
 	return "LGUEST";
 }
 
+/* Before virtqueues are set up, we use LHCALL_NOTIFY on normal memory to
+ * produce console output. */
+static __init int early_put_chars(u32 vtermno, const char *buf, int count)
+{
+	char scratch[17];
+	unsigned int len = count;
+
+	if (len > sizeof(scratch) - 1)
+		len = sizeof(scratch) - 1;
+	scratch[len] = '\0';
+	memcpy(scratch, buf, len);
+	hcall(LHCALL_NOTIFY, __pa(scratch), 0, 0);
+
+	/* This routine returns the number of bytes actually written. */
+	return len;
+}
+
 /*G:050
  * Patching (Powerfully Placating Performance Pedants)
  *
@@ -950,18 +928,8 @@ static unsigned lguest_patch(u8 type, u16 clobber, void *ibuf,
 /*G:030 Once we get to lguest_init(), we know we're a Guest.  The pv_ops
  * structures in the kernel provide points for (almost) every routine we have
  * to override to avoid privileged instructions. */
-__init void lguest_init(void *boot)
+__init void lguest_init(void)
 {
-	/* Copy boot parameters first: the Launcher put the physical location
-	 * in %esi, and head.S converted that to a virtual address and handed
-	 * it to us.  We use "__memcpy" because "memcpy" sometimes tries to do
-	 * tricky things to go faster, and we're not ready for that. */
-	__memcpy(&boot_params, boot, PARAM_SIZE);
-	/* The boot parameters also tell us where the command-line is: save
-	 * that, too. */
-	__memcpy(boot_command_line, __va(boot_params.hdr.cmd_line_ptr),
-	       COMMAND_LINE_SIZE);
-
 	/* We're under lguest, paravirt is enabled, and we're running at
 	 * privilege level 1, not 0 as normal. */
 	pv_info.name = "lguest";
@@ -1033,11 +1001,7 @@ __init void lguest_init(void *boot)
 
 	/*G:070 Now we've seen all the paravirt_ops, we return to
 	 * lguest_init() where the rest of the fairly chaotic boot setup
-	 * occurs.
-	 *
-	 * The Host expects our first hypercall to tell it where our "struct
-	 * lguest_data" is, so we do that first. */
-	hcall(LHCALL_LGUEST_INIT, __pa(&lguest_data), 0, 0);
+	 * occurs. */
 
 	/* The native boot code sets up initial page tables immediately after
 	 * the kernel itself, and sets init_pg_tables_end so they're not
@@ -1050,11 +1014,6 @@ __init void lguest_init(void *boot)
 	 * the normal data segment to get through booting. */
 	asm volatile ("mov %0, %%fs" : : "r" (__KERNEL_DS) : "memory");
 
-	/* Clear the part of the kernel data which is expected to be zero.
-	 * Normally it will be anyway, but if we're loading from a bzImage with
-	 * CONFIG_RELOCATALE=y, the relocations will be sitting here. */
-	memset(__bss_start, 0, __bss_stop - __bss_start);
-
 	/* The Host uses the top of the Guest's virtual address space for the
 	 * Host<->Guest Switcher, and it tells us how much it needs in
 	 * lguest_data.reserve_mem, set up on the LGUEST_INIT hypercall. */
@@ -1092,6 +1051,9 @@ __init void lguest_init(void *boot)
 	 * adapted for lguest's use. */
 	add_preferred_console("hvc", 0, NULL);
 
+	/* Register our very early console. */
+	virtio_cons_early_init(early_put_chars);
+
 	/* Last of all, we set the power management poweroff hook to point to
 	 * the Guest routine to power off. */
 	pm_power_off = lguest_power_off;

drivers/lguest/lguest_asm.S → arch/x86/lguest/i386_head.S

@@ -1,25 +1,47 @@
 #include <linux/linkage.h>
 #include <linux/lguest.h>
+#include <asm/lguest_hcall.h>
 #include <asm/asm-offsets.h>
 #include <asm/thread_info.h>
 #include <asm/processor-flags.h>
 
-/*G:020 This is where we begin: we have a magic signature which the launcher
- * looks for.  The plan is that the Linux boot protocol will be extended with a
- * "platform type" field which will guide us here from the normal entry point,
- * but for the moment this suffices.  The normal boot code uses %esi for the
- * boot header, so we do too.  We convert it to a virtual address by adding
- * PAGE_OFFSET, and hand it to lguest_init() as its argument (ie. %eax).
+/*G:020 This is where we begin: head.S notes that the boot header's platform
+ * type field is "1" (lguest), so calls us here.  The boot header is in %esi.
+ *
+ * WARNING: be very careful here!  We're running at addresses equal to physical
+ * addesses (around 0), not above PAGE_OFFSET as most code expectes
+ * (eg. 0xC0000000).  Jumps are relative, so they're OK, but we can't touch any
+ * data.
  *
  * The .section line puts this code in .init.text so it will be discarded after
  * boot. */
 .section .init.text, "ax", @progbits
-.ascii "GenuineLguest"
-	/* Set up initial stack. */
- 	movl $(init_thread_union+THREAD_SIZE),%esp
-	movl %esi, %eax
-	addl $__PAGE_OFFSET, %eax
-	jmp lguest_init
+ENTRY(lguest_entry)
+	/* Make initial hypercall now, so we can set up the pagetables. */
+	movl $LHCALL_LGUEST_INIT, %eax
+	movl $lguest_data - __PAGE_OFFSET, %edx
+	int $LGUEST_TRAP_ENTRY
+
+	/* The Host put the toplevel pagetable in lguest_data.pgdir.  The movsl
+	 * instruction uses %esi implicitly. */
+	movl lguest_data - __PAGE_OFFSET + LGUEST_DATA_pgdir, %esi
+
+	/* Copy first 32 entries of page directory to __PAGE_OFFSET entries.
+	 * This means the first 128M of kernel memory will be mapped at
+	 * PAGE_OFFSET where the kernel expects to run.  This will get it far
+	 * enough through boot to switch to its own pagetables. */
+	movl $32, %ecx
+	movl %esi, %edi
+	addl $((__PAGE_OFFSET >> 22) * 4), %edi
+	rep
+	movsl
+
+	/* Set up the initial stack so we can run C code. */
+	movl $(init_thread_union+THREAD_SIZE),%esp
+
+	/* Jumps are relative, and we're running __PAGE_OFFSET too low at the
+	 * moment. */
+	jmp lguest_init+__PAGE_OFFSET
 
 /*G:055 We create a macro which puts the assembler code between lgstart_ and
  * lgend_ markers.  These templates are put in the .text section: they can't be

arch/x86/xen/Kconfig

@@ -3,8 +3,9 @@
 #
 
 config XEN
-	bool "Enable support for Xen hypervisor"
-	depends on PARAVIRT && X86_CMPXCHG && X86_TSC && !NEED_MULTIPLE_NODES
+	bool "Xen guest support"
+	select PARAVIRT
+	depends on X86_CMPXCHG && X86_TSC && !NEED_MULTIPLE_NODES && !(X86_VISWS || X86_VOYAGER)
 	help
 	  This is the Linux Xen port.  Enabling this will allow the
 	  kernel to boot in a paravirtualized environment under the

drivers/Kconfig

@@ -94,5 +94,5 @@ source "drivers/kvm/Kconfig"
 
 source "drivers/uio/Kconfig"
 
-source "drivers/lguest/Kconfig"
+source "drivers/virtio/Kconfig"
 endmenu

drivers/Makefile

@@ -91,3 +91,4 @@ obj-$(CONFIG_HID)		+= hid/
 obj-$(CONFIG_PPC_PS3)		+= ps3/
 obj-$(CONFIG_OF)		+= of/
 obj-$(CONFIG_SSB)		+= ssb/
+obj-$(CONFIG_VIRTIO)		+= virtio/

drivers/block/Kconfig

@@ -425,4 +425,10 @@ config XEN_BLKDEV_FRONTEND
 	  block device driver.  It communicates with a back-end driver
 	  in another domain which drives the actual block device.
 
+config VIRTIO_BLK
+	tristate "Virtio block driver (EXPERIMENTAL)"
+	depends on EXPERIMENTAL && VIRTIO
+	---help---
+	  This is the virtual block driver for lguest.  Say Y or M.
+
 endif # BLK_DEV

drivers/block/Makefile

@@ -25,10 +25,10 @@ obj-$(CONFIG_SUNVDC)		+= sunvdc.o
 obj-$(CONFIG_BLK_DEV_UMEM)	+= umem.o
 obj-$(CONFIG_BLK_DEV_NBD)	+= nbd.o
 obj-$(CONFIG_BLK_DEV_CRYPTOLOOP) += cryptoloop.o
+obj-$(CONFIG_VIRTIO_BLK)	+= virtio_blk.o
 
 obj-$(CONFIG_VIODASD)		+= viodasd.o
 obj-$(CONFIG_BLK_DEV_SX8)	+= sx8.o
 obj-$(CONFIG_BLK_DEV_UB)	+= ub.o
 
 obj-$(CONFIG_XEN_BLKDEV_FRONTEND)	+= xen-blkfront.o
-obj-$(CONFIG_LGUEST_BLOCK)	+= lguest_blk.o

drivers/block/lguest_blk.c

@@ -1,421 +0,0 @@
-/*D:400
- * The Guest block driver
- *
- * This is a simple block driver, which appears as /dev/lgba, lgbb, lgbc etc.
- * The mechanism is simple: we place the information about the request in the
- * device page, then use SEND_DMA (containing the data for a write, or an empty
- * "ping" DMA for a read).
- :*/
-/* Copyright 2006 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation
- *
- * This program is free software; you can redistribute it and/or modify
- * it under the terms of the GNU General Public License as published by
- * the Free Software Foundation; either version 2 of the License, or
- * (at your option) any later version.
- *
- * This program is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- * GNU General Public License for more details.
- *
- * You should have received a copy of the GNU General Public License
- * along with this program; if not, write to the Free Software
- * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
- */
-//#define DEBUG
-#include <linux/init.h>
-#include <linux/types.h>
-#include <linux/blkdev.h>
-#include <linux/interrupt.h>
-#include <linux/lguest_bus.h>
-
-static char next_block_index = 'a';
-
-/*D:420 Here is the structure which holds all the information we need about
- * each Guest block device.
- *
- * I'm sure at this stage, you're wondering "hey, where was the adventure I was
- * promised?" and thinking "Rusty sucks, I shall say nasty things about him on
- * my blog".  I think Real adventures have boring bits, too, and you're in the
- * middle of one.  But it gets better.  Just not quite yet. */
-struct blockdev
-{
-	/* The block queue infrastructure wants a spinlock: it is held while it
-	 * calls our block request function.  We grab it in our interrupt
-	 * handler so the responses don't mess with new requests. */
-	spinlock_t lock;
-
-	/* The disk structure registered with kernel. */
-	struct gendisk *disk;
-
-	/* The major device number for this disk, and the interrupt.  We only
-	 * really keep them here for completeness; we'd need them if we
-	 * supported device unplugging. */
-	int major;
-	int irq;
-
-	/* The physical address of this device's memory page */
-	unsigned long phys_addr;
-	/* The mapped memory page for convenient acces. */
-	struct lguest_block_page *lb_page;
-
-	/* We only have a single request outstanding at a time: this is it. */
-	struct lguest_dma dma;
-	struct request *req;
-};
-
-/*D:495 We originally used end_request() throughout the driver, but it turns
- * out that end_request() is deprecated, and doesn't actually end the request
- * (which seems like a good reason to deprecate it!).  It simply ends the first
- * bio.  So if we had 3 bios in a "struct request" we would do all 3,
- * end_request(), do 2, end_request(), do 1 and end_request(): twice as much
- * work as we needed to do.
- *
- * This reinforced to me that I do not understand the block layer.
- *
- * Nonetheless, Jens Axboe gave me this nice helper to end all chunks of a
- * request.  This improved disk speed by 130%. */
-static void end_entire_request(struct request *req, int uptodate)
-{
-	if (end_that_request_first(req, uptodate, req->hard_nr_sectors))
-		BUG();
-	add_disk_randomness(req->rq_disk);
-	blkdev_dequeue_request(req);
-	end_that_request_last(req, uptodate);
-}
-
-/* I'm told there are only two stories in the world worth telling: love and
- * hate.  So there used to be a love scene here like this:
- *
- *  Launcher:	We could make beautiful I/O together, you and I.
- *  Guest:	My, that's a big disk!
- *
- * Unfortunately, it was just too raunchy for our otherwise-gentle tale. */
-
-/*D:490 This is the interrupt handler, called when a block read or write has
- * been completed for us. */
-static irqreturn_t lgb_irq(int irq, void *_bd)
-{
-	/* We handed our "struct blockdev" as the argument to request_irq(), so
-	 * it is passed through to us here.  This tells us which device we're
-	 * dealing with in case we have more than one. */
-	struct blockdev *bd = _bd;
-	unsigned long flags;
-
-	/* We weren't doing anything?  Strange, but could happen if we shared
-	 * interrupts (we don't!). */
-	if (!bd->req) {
-		pr_debug("No work!\n");
-		return IRQ_NONE;
-	}
-
-	/* Not done yet?  That's equally strange. */
-	if (!bd->lb_page->result) {
-		pr_debug("No result!\n");
-		return IRQ_NONE;
-	}
-
-	/* We have to grab the lock before ending the request. */
-	spin_lock_irqsave(&bd->lock, flags);
-	/* "result" is 1 for success, 2 for failure: end_entire_request() wants
-	 * to know whether this succeeded or not. */
-	end_entire_request(bd->req, bd->lb_page->result == 1);
-	/* Clear out request, it's done. */
-	bd->req = NULL;
-	/* Reset incoming DMA for next time. */
-	bd->dma.used_len = 0;
-	/* Ready for more reads or writes */
-	blk_start_queue(bd->disk->queue);
-	spin_unlock_irqrestore(&bd->lock, flags);
-
-	/* The interrupt was for us, we dealt with it. */
-	return IRQ_HANDLED;
-}
-
-/*D:480 The block layer's "struct request" contains a number of "struct bio"s,
- * each of which contains "struct bio_vec"s, each of which contains a page, an
- * offset and a length.
- *
- * Fortunately there are iterators to help us walk through the "struct
- * request".  Even more fortunately, there were plenty of places to steal the
- * code from.  We pack the "struct request" into our "struct lguest_dma" and
- * return the total length. */
-static unsigned int req_to_dma(struct request *req, struct lguest_dma *dma)
-{
-	unsigned int i = 0, len = 0;
-	struct req_iterator iter;
-	struct bio_vec *bvec;
-
-	rq_for_each_segment(bvec, req, iter) {
-		/* We told the block layer not to give us too many. */
-		BUG_ON(i == LGUEST_MAX_DMA_SECTIONS);
-		/* If we had a zero-length segment, it would look like
-		 * the end of the data referred to by the "struct
-		 * lguest_dma", so make sure that doesn't happen. */
-		BUG_ON(!bvec->bv_len);
-		/* Convert page & offset to a physical address */
-		dma->addr[i] = page_to_phys(bvec->bv_page)
-			+ bvec->bv_offset;
-		dma->len[i] = bvec->bv_len;
-		len += bvec->bv_len;
-		i++;
-	}
-	/* If the array isn't full, we mark the end with a 0 length */
-	if (i < LGUEST_MAX_DMA_SECTIONS)
-		dma->len[i] = 0;
-	return len;
-}
-
-/* This creates an empty DMA, useful for prodding the Host without sending data
- * (ie. when we want to do a read) */
-static void empty_dma(struct lguest_dma *dma)
-{
-	dma->len[0] = 0;
-}
-
-/*D:470 Setting up a request is fairly easy: */
-static void setup_req(struct blockdev *bd,
-		      int type, struct request *req, struct lguest_dma *dma)
-{
-	/* The type is 1 (write) or 0 (read). */
-	bd->lb_page->type = type;
-	/* The sector on disk where the read or write starts. */
-	bd->lb_page->sector = req->sector;
-	/* The result is initialized to 0 (unfinished). */
-	bd->lb_page->result = 0;
-	/* The current request (so we can end it in the interrupt handler). */
-	bd->req = req;
-	/* The number of bytes: returned as a side-effect of req_to_dma(),
-	 * which packs the block layer's "struct request" into our "struct
-	 * lguest_dma" */
-	bd->lb_page->bytes = req_to_dma(req, dma);
-}
-
-/*D:450 Write is pretty straightforward: we pack the request into a "struct
- * lguest_dma", then use SEND_DMA to send the request. */
-static void do_write(struct blockdev *bd, struct request *req)
-{
-	struct lguest_dma send;
-
-	pr_debug("lgb: WRITE sector %li\n", (long)req->sector);
-	setup_req(bd, 1, req, &send);
-
-	lguest_send_dma(bd->phys_addr, &send);
-}
-
-/* Read is similar to write, except we pack the request into our receive
- * "struct lguest_dma" and send through an empty DMA just to tell the Host that
- * there's a request pending. */
-static void do_read(struct blockdev *bd, struct request *req)
-{
-	struct lguest_dma ping;
-
-	pr_debug("lgb: READ sector %li\n", (long)req->sector);
-	setup_req(bd, 0, req, &bd->dma);
-
-	empty_dma(&ping);
-	lguest_send_dma(bd->phys_addr, &ping);
-}
-
-/*D:440 This where requests come in: we get handed the request queue and are
- * expected to pull a "struct request" off it until we've finished them or
- * we're waiting for a reply: */
-static void do_lgb_request(struct request_queue *q)
-{
-	struct blockdev *bd;
-	struct request *req;
-
-again:
-	/* This sometimes returns NULL even on the very first time around.  I
-	 * wonder if it's something to do with letting elves handle the request
-	 * queue... */
-	req = elv_next_request(q);
-	if (!req)
-		return;
-
-	/* We attached the struct blockdev to the disk: get it back */
-	bd = req->rq_disk->private_data;
-	/* Sometimes we get repeated requests after blk_stop_queue(), but we
-	 * can only handle one at a time. */
-	if (bd->req)
-		return;
-
-	/* We only do reads and writes: no tricky business! */
-	if (!blk_fs_request(req)) {
-		pr_debug("Got non-command 0x%08x\n", req->cmd_type);
-		req->errors++;
-		end_entire_request(req, 0);
-		goto again;
-	}
-
-	if (rq_data_dir(req) == WRITE)
-		do_write(bd, req);
-	else
-		do_read(bd, req);
-
-	/* We've put out the request, so stop any more coming in until we get
-	 * an interrupt, which takes us to lgb_irq() to re-enable the queue. */
-	blk_stop_queue(q);
-}
-
-/*D:430 This is the "struct block_device_operations" we attach to the disk at
- * the end of lguestblk_probe().  It doesn't seem to want much. */
-static struct block_device_operations lguestblk_fops = {
-	.owner = THIS_MODULE,
-};
-
-/*D:425 Setting up a disk device seems to involve a lot of code.  I'm not sure
- * quite why.  I do know that the IDE code sent two or three of the maintainers
- * insane, perhaps this is the fringe of the same disease?
- *
- * As in the console code, the probe function gets handed the generic
- * lguest_device from lguest_bus.c: */
-static int lguestblk_probe(struct lguest_device *lgdev)
-{
-	struct blockdev *bd;
-	int err;
-	int irqflags = IRQF_SHARED;
-
-	/* First we allocate our own "struct blockdev" and initialize the easy
-	 * fields. */
-	bd = kmalloc(sizeof(*bd), GFP_KERNEL);
-	if (!bd)
-		return -ENOMEM;
-
-	spin_lock_init(&bd->lock);
-	bd->irq = lgdev_irq(lgdev);
-	bd->req = NULL;
-	bd->dma.used_len = 0;
-	bd->dma.len[0] = 0;
-	/* The descriptor in the lguest_devices array provided by the Host
-	 * gives the Guest the physical page number of the device's page. */
-	bd->phys_addr = (lguest_devices[lgdev->index].pfn << PAGE_SHIFT);
-
-	/* We use lguest_map() to get a pointer to the device page */
-	bd->lb_page = lguest_map(bd->phys_addr, 1);
-	if (!bd->lb_page) {
-		err = -ENOMEM;
-		goto out_free_bd;
-	}
-
-	/* We need a major device number: 0 means "assign one dynamically". */
-	bd->major = register_blkdev(0, "lguestblk");
-	if (bd->major < 0) {
-		err = bd->major;
-		goto out_unmap;
-	}
-
-	/* This allocates a "struct gendisk" where we pack all the information
-	 * about the disk which the rest of Linux sees.  The argument is the
-	 * number of minor devices desired: we need one minor for the main
-	 * disk, and one for each partition.  Of course, we can't possibly know
-	 * how many partitions are on the disk (add_disk does that).
-	 */
-	bd->disk = alloc_disk(16);
-	if (!bd->disk) {
-		err = -ENOMEM;
-		goto out_unregister_blkdev;
-	}
-
-	/* Every disk needs a queue for requests to come in: we set up the
-	 * queue with a callback function (the core of our driver) and the lock
-	 * to use. */
-	bd->disk->queue = blk_init_queue(do_lgb_request, &bd->lock);
-	if (!bd->disk->queue) {
-		err = -ENOMEM;
-		goto out_put_disk;
-	}
-
-	/* We can only handle a certain number of pointers in our SEND_DMA
-	 * call, so we set that with blk_queue_max_hw_segments().  This is not
-	 * to be confused with blk_queue_max_phys_segments() of course!  I
-	 * know, who could possibly confuse the two?
-	 *
-	 * Well, it's simple to tell them apart: this one seems to work and the
-	 * other one didn't. */
-	blk_queue_max_hw_segments(bd->disk->queue, LGUEST_MAX_DMA_SECTIONS);
-
-	/* Due to technical limitations of our Host (and simple coding) we
-	 * can't have a single buffer which crosses a page boundary.  Tell it
-	 * here.  This means that our maximum request size is 16
-	 * (LGUEST_MAX_DMA_SECTIONS) pages. */
-	blk_queue_segment_boundary(bd->disk->queue, PAGE_SIZE-1);
-
-	/* We name our disk: this becomes the device name when udev does its
-	 * magic thing and creates the device node, such as /dev/lgba.
-	 * next_block_index is a global which starts at 'a'.  Unfortunately
-	 * this simple increment logic means that the 27th disk will be called
-	 * "/dev/lgb{".  In that case, I recommend having at least 29 disks, so
-	 * your /dev directory will be balanced. */
-	sprintf(bd->disk->disk_name, "lgb%c", next_block_index++);
-
-	/* We look to the device descriptor again to see if this device's
-	 * interrupts are expected to be random.  If they are, we tell the irq
-	 * subsystem.  At the moment this bit is always set. */
-	if (lguest_devices[lgdev->index].features & LGUEST_DEVICE_F_RANDOMNESS)
-		irqflags |= IRQF_SAMPLE_RANDOM;
-
-	/* Now we have the name and irqflags, we can request the interrupt; we
-	 * give it the "struct blockdev" we have set up to pass to lgb_irq()
-	 * when there is an interrupt. */
-	err = request_irq(bd->irq, lgb_irq, irqflags, bd->disk->disk_name, bd);
-	if (err)
-		goto out_cleanup_queue;
-
-	/* We bind our one-entry DMA pool to the key for this block device so
-	 * the Host can reply to our requests.  The key is equal to the
-	 * physical address of the device's page, which is conveniently
-	 * unique. */
-	err = lguest_bind_dma(bd->phys_addr, &bd->dma, 1, bd->irq);
-	if (err)
-		goto out_free_irq;
-
-	/* We finish our disk initialization and add the disk to the system. */
-	bd->disk->major = bd->major;
-	bd->disk->first_minor = 0;
-	bd->disk->private_data = bd;
-	bd->disk->fops = &lguestblk_fops;
-	/* This is initialized to the disk size by the Launcher. */
-	set_capacity(bd->disk, bd->lb_page->num_sectors);
-	add_disk(bd->disk);
-
-	printk(KERN_INFO "%s: device %i at major %d\n",
-	       bd->disk->disk_name, lgdev->index, bd->major);
-
-	/* We don't need to keep the "struct blockdev" around, but if we ever
-	 * implemented device removal, we'd need this. */
-	lgdev->private = bd;
-	return 0;
-
-out_free_irq:
-	free_irq(bd->irq, bd);
-out_cleanup_queue:
-	blk_cleanup_queue(bd->disk->queue);
-out_put_disk:
-	put_disk(bd->disk);
-out_unregister_blkdev:
-	unregister_blkdev(bd->major, "lguestblk");
-out_unmap:
-	lguest_unmap(bd->lb_page);
-out_free_bd:
-	kfree(bd);
-	return err;
-}
-
-/*D:410 The boilerplate code for registering the lguest block driver is just
- * like the console: */
-static struct lguest_driver lguestblk_drv = {
-	.name = "lguestblk",
-	.owner = THIS_MODULE,
-	.device_type = LGUEST_DEVICE_T_BLOCK,
-	.probe = lguestblk_probe,
-};
-
-static __init int lguestblk_init(void)
-{
-	return register_lguest_driver(&lguestblk_drv);
-}
-module_init(lguestblk_init);
-
-MODULE_DESCRIPTION("Lguest block driver");
-MODULE_LICENSE("GPL");

drivers/block/virtio_blk.c

@@ -0,0 +1,308 @@
+//#define DEBUG
+#include <linux/spinlock.h>
+#include <linux/blkdev.h>
+#include <linux/hdreg.h>
+#include <linux/virtio.h>
+#include <linux/virtio_blk.h>
+#include <linux/virtio_blk.h>
+
+static unsigned char virtblk_index = 'a';
+struct virtio_blk
+{
+	spinlock_t lock;
+
+	struct virtio_device *vdev;
+	struct virtqueue *vq;
+
+	/* The disk structure for the kernel. */
+	struct gendisk *disk;
+
+	/* Request tracking. */
+	struct list_head reqs;
+
+	mempool_t *pool;
+
+	/* Scatterlist: can be too big for stack. */
+	struct scatterlist sg[3+MAX_PHYS_SEGMENTS];
+};
+
+struct virtblk_req
+{
+	struct list_head list;
+	struct request *req;
+	struct virtio_blk_outhdr out_hdr;
+	struct virtio_blk_inhdr in_hdr;
+};
+
+static bool blk_done(struct virtqueue *vq)
+{
+	struct virtio_blk *vblk = vq->vdev->priv;
+	struct virtblk_req *vbr;
+	unsigned int len;
+	unsigned long flags;
+
+	spin_lock_irqsave(&vblk->lock, flags);
+	while ((vbr = vblk->vq->vq_ops->get_buf(vblk->vq, &len)) != NULL) {
+		int uptodate;
+		switch (vbr->in_hdr.status) {
+		case VIRTIO_BLK_S_OK:
+			uptodate = 1;
+			break;
+		case VIRTIO_BLK_S_UNSUPP:
+			uptodate = -ENOTTY;
+			break;
+		default:
+			uptodate = 0;
+			break;
+		}
+
+		end_dequeued_request(vbr->req, uptodate);
+		list_del(&vbr->list);
+		mempool_free(vbr, vblk->pool);
+	}
+	/* In case queue is stopped waiting for more buffers. */
+	blk_start_queue(vblk->disk->queue);
+	spin_unlock_irqrestore(&vblk->lock, flags);
+	return true;
+}
+
+static bool do_req(struct request_queue *q, struct virtio_blk *vblk,
+		   struct request *req)
+{
+	unsigned long num, out, in;
+	struct virtblk_req *vbr;
+
+	vbr = mempool_alloc(vblk->pool, GFP_ATOMIC);
+	if (!vbr)
+		/* When another request finishes we'll try again. */
+		return false;
+
+	vbr->req = req;
+	if (blk_fs_request(vbr->req)) {
+		vbr->out_hdr.type = 0;
+		vbr->out_hdr.sector = vbr->req->sector;
+		vbr->out_hdr.ioprio = vbr->req->ioprio;
+	} else if (blk_pc_request(vbr->req)) {
+		vbr->out_hdr.type = VIRTIO_BLK_T_SCSI_CMD;
+		vbr->out_hdr.sector = 0;
+		vbr->out_hdr.ioprio = vbr->req->ioprio;
+	} else {
+		/* We don't put anything else in the queue. */
+		BUG();
+	}
+
+	if (blk_barrier_rq(vbr->req))
+		vbr->out_hdr.type |= VIRTIO_BLK_T_BARRIER;
+
+	/* We have to zero this, otherwise blk_rq_map_sg gets upset. */
+	memset(vblk->sg, 0, sizeof(vblk->sg));
+	sg_set_buf(&vblk->sg[0], &vbr->out_hdr, sizeof(vbr->out_hdr));
+	num = blk_rq_map_sg(q, vbr->req, vblk->sg+1);
+	sg_set_buf(&vblk->sg[num+1], &vbr->in_hdr, sizeof(vbr->in_hdr));
+
+	if (rq_data_dir(vbr->req) == WRITE) {
+		vbr->out_hdr.type |= VIRTIO_BLK_T_OUT;
+		out = 1 + num;
+		in = 1;
+	} else {
+		vbr->out_hdr.type |= VIRTIO_BLK_T_IN;
+		out = 1;
+		in = 1 + num;
+	}
+
+	if (vblk->vq->vq_ops->add_buf(vblk->vq, vblk->sg, out, in, vbr)) {
+		mempool_free(vbr, vblk->pool);
+		return false;
+	}
+
+	list_add_tail(&vbr->list, &vblk->reqs);
+	return true;
+}
+
+static void do_virtblk_request(struct request_queue *q)
+{
+	struct virtio_blk *vblk = NULL;
+	struct request *req;
+	unsigned int issued = 0;
+
+	while ((req = elv_next_request(q)) != NULL) {
+		vblk = req->rq_disk->private_data;
+		BUG_ON(req->nr_phys_segments > ARRAY_SIZE(vblk->sg));
+
+		/* If this request fails, stop queue and wait for something to
+		   finish to restart it. */
+		if (!do_req(q, vblk, req)) {
+			blk_stop_queue(q);
+			break;
+		}
+		blkdev_dequeue_request(req);
+		issued++;
+	}
+
+	if (issued)
+		vblk->vq->vq_ops->kick(vblk->vq);
+}
+
+static int virtblk_ioctl(struct inode *inode, struct file *filp,
+			 unsigned cmd, unsigned long data)
+{
+	return scsi_cmd_ioctl(filp, inode->i_bdev->bd_disk->queue,
+			      inode->i_bdev->bd_disk, cmd,
+			      (void __user *)data);
+}
+
+static struct block_device_operations virtblk_fops = {
+	.ioctl = virtblk_ioctl,
+	.owner = THIS_MODULE,
+};
+
+static int virtblk_probe(struct virtio_device *vdev)
+{
+	struct virtio_blk *vblk;
+	int err, major;
+	void *token;
+	unsigned int len;
+	u64 cap;
+	u32 v;
+
+	vdev->priv = vblk = kmalloc(sizeof(*vblk), GFP_KERNEL);
+	if (!vblk) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	INIT_LIST_HEAD(&vblk->reqs);
+	spin_lock_init(&vblk->lock);
+	vblk->vdev = vdev;
+
+	/* We expect one virtqueue, for output. */
+	vblk->vq = vdev->config->find_vq(vdev, blk_done);
+	if (IS_ERR(vblk->vq)) {
+		err = PTR_ERR(vblk->vq);
+		goto out_free_vblk;
+	}
+
+	vblk->pool = mempool_create_kmalloc_pool(1,sizeof(struct virtblk_req));
+	if (!vblk->pool) {
+		err = -ENOMEM;
+		goto out_free_vq;
+	}
+
+	major = register_blkdev(0, "virtblk");
+	if (major < 0) {
+		err = major;
+		goto out_mempool;
+	}
+
+	/* FIXME: How many partitions?  How long is a piece of string? */
+	vblk->disk = alloc_disk(1 << 4);
+	if (!vblk->disk) {
+		err = -ENOMEM;
+		goto out_unregister_blkdev;
+	}
+
+	vblk->disk->queue = blk_init_queue(do_virtblk_request, &vblk->lock);
+	if (!vblk->disk->queue) {
+		err = -ENOMEM;
+		goto out_put_disk;
+	}
+
+	sprintf(vblk->disk->disk_name, "vd%c", virtblk_index++);
+	vblk->disk->major = major;
+	vblk->disk->first_minor = 0;
+	vblk->disk->private_data = vblk;
+	vblk->disk->fops = &virtblk_fops;
+
+	/* If barriers are supported, tell block layer that queue is ordered */
+	token = vdev->config->find(vdev, VIRTIO_CONFIG_BLK_F, &len);
+	if (virtio_use_bit(vdev, token, len, VIRTIO_BLK_F_BARRIER))
+		blk_queue_ordered(vblk->disk->queue, QUEUE_ORDERED_TAG, NULL);
+
+	err = virtio_config_val(vdev, VIRTIO_CONFIG_BLK_F_CAPACITY, &cap);
+	if (err) {
+		dev_err(&vdev->dev, "Bad/missing capacity in config\n");
+		goto out_put_disk;
+	}
+
+	/* If capacity is too big, truncate with warning. */
+	if ((sector_t)cap != cap) {
+		dev_warn(&vdev->dev, "Capacity %llu too large: truncating\n",
+			 (unsigned long long)cap);
+		cap = (sector_t)-1;
+	}
+	set_capacity(vblk->disk, cap);
+
+	err = virtio_config_val(vdev, VIRTIO_CONFIG_BLK_F_SIZE_MAX, &v);
+	if (!err)
+		blk_queue_max_segment_size(vblk->disk->queue, v);
+	else if (err != -ENOENT) {
+		dev_err(&vdev->dev, "Bad SIZE_MAX in config\n");
+		goto out_put_disk;
+	}
+
+	err = virtio_config_val(vdev, VIRTIO_CONFIG_BLK_F_SEG_MAX, &v);
+	if (!err)
+		blk_queue_max_hw_segments(vblk->disk->queue, v);
+	else if (err != -ENOENT) {
+		dev_err(&vdev->dev, "Bad SEG_MAX in config\n");
+		goto out_put_disk;
+	}
+
+	add_disk(vblk->disk);
+	return 0;
+
+out_put_disk:
+	put_disk(vblk->disk);
+out_unregister_blkdev:
+	unregister_blkdev(major, "virtblk");
+out_mempool:
+	mempool_destroy(vblk->pool);
+out_free_vq:
+	vdev->config->del_vq(vblk->vq);
+out_free_vblk:
+	kfree(vblk);
+out:
+	return err;
+}
+
+static void virtblk_remove(struct virtio_device *vdev)
+{
+	struct virtio_blk *vblk = vdev->priv;
+	int major = vblk->disk->major;
+
+	BUG_ON(!list_empty(&vblk->reqs));
+	blk_cleanup_queue(vblk->disk->queue);
+	put_disk(vblk->disk);
+	unregister_blkdev(major, "virtblk");
+	mempool_destroy(vblk->pool);
+	kfree(vblk);
+}
+
+static struct virtio_device_id id_table[] = {
+	{ VIRTIO_ID_BLOCK, VIRTIO_DEV_ANY_ID },
+	{ 0 },
+};
+
+static struct virtio_driver virtio_blk = {
+	.driver.name =	KBUILD_MODNAME,
+	.driver.owner =	THIS_MODULE,
+	.id_table =	id_table,
+	.probe =	virtblk_probe,
+	.remove =	__devexit_p(virtblk_remove),
+};
+
+static int __init init(void)
+{
+	return register_virtio_driver(&virtio_blk);
+}
+
+static void __exit fini(void)
+{
+	unregister_virtio_driver(&virtio_blk);
+}
+module_init(init);
+module_exit(fini);
+
+MODULE_DEVICE_TABLE(virtio, id_table);
+MODULE_DESCRIPTION("Virtio block driver");
+MODULE_LICENSE("GPL");

drivers/char/Kconfig

@@ -613,6 +613,10 @@ config HVC_XEN
 	help
 	  Xen virtual console device driver
 
+config VIRTIO_CONSOLE
+	bool
+	select HVC_DRIVER
+
 config HVCS
 	tristate "IBM Hypervisor Virtual Console Server support"
 	depends on PPC_PSERIES

drivers/char/Makefile

@@ -42,7 +42,6 @@ obj-$(CONFIG_SYNCLINK_GT)	+= synclink_gt.o
 obj-$(CONFIG_N_HDLC)		+= n_hdlc.o
 obj-$(CONFIG_AMIGA_BUILTIN_SERIAL) += amiserial.o
 obj-$(CONFIG_SX)		+= sx.o generic_serial.o
-obj-$(CONFIG_LGUEST_GUEST)	+= hvc_lguest.o
 obj-$(CONFIG_RIO)		+= rio/ generic_serial.o
 obj-$(CONFIG_HVC_CONSOLE)	+= hvc_vio.o hvsi.o
 obj-$(CONFIG_HVC_ISERIES)	+= hvc_iseries.o
@@ -50,6 +49,7 @@ obj-$(CONFIG_HVC_RTAS)		+= hvc_rtas.o
 obj-$(CONFIG_HVC_BEAT)		+= hvc_beat.o
 obj-$(CONFIG_HVC_DRIVER)	+= hvc_console.o
 obj-$(CONFIG_HVC_XEN)		+= hvc_xen.o
+obj-$(CONFIG_VIRTIO_CONSOLE)	+= virtio_console.o
 obj-$(CONFIG_RAW_DRIVER)	+= raw.o
 obj-$(CONFIG_SGI_SNSC)		+= snsc.o snsc_event.o
 obj-$(CONFIG_MSPEC)		+= mspec.o

drivers/char/hvc_lguest.c

@@ -1,177 +0,0 @@
-/*D:300
- * The Guest console driver
- *
- * This is a trivial console driver: we use lguest's DMA mechanism to send
- * bytes out, and register a DMA buffer to receive bytes in.  It is assumed to
- * be present and available from the very beginning of boot.
- *
- * Writing console drivers is one of the few remaining Dark Arts in Linux.
- * Fortunately for us, the path of virtual consoles has been well-trodden by
- * the PowerPC folks, who wrote "hvc_console.c" to generically support any
- * virtual console.  We use that infrastructure which only requires us to write
- * the basic put_chars and get_chars functions and call the right register
- * functions.
- :*/
-
-/*M:002 The console can be flooded: while the Guest is processing input the
- * Host can send more.  Buffering in the Host could alleviate this, but it is a
- * difficult problem in general. :*/
-/* Copyright (C) 2006 Rusty Russell, IBM Corporation
- *
- * This program is free software; you can redistribute it and/or modify
- * it under the terms of the GNU General Public License as published by
- * the Free Software Foundation; either version 2 of the License, or
- * (at your option) any later version.
- *
- * This program is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- * GNU General Public License for more details.
- *
- * You should have received a copy of the GNU General Public License
- * along with this program; if not, write to the Free Software
- * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
- */
-#include <linux/err.h>
-#include <linux/init.h>
-#include <linux/lguest_bus.h>
-#include <asm/paravirt.h>
-#include "hvc_console.h"
-
-/*D:340 This is our single console input buffer, with associated "struct
- * lguest_dma" referring to it.  Note the 0-terminated length array, and the
- * use of physical address for the buffer itself. */
-static char inbuf[256];
-static struct lguest_dma cons_input = { .used_len = 0,
-					.addr[0] = __pa(inbuf),
-					.len[0] = sizeof(inbuf),
-					.len[1] = 0 };
-
-/*D:310 The put_chars() callback is pretty straightforward.
- *
- * First we put the pointer and length in a "struct lguest_dma": we only have
- * one pointer, so we set the second length to 0.  Then we use SEND_DMA to send
- * the data to (Host) buffers attached to the console key.  Usually a device's
- * key is a physical address within the device's memory, but because the
- * console device doesn't have any associated physical memory, we use the
- * LGUEST_CONSOLE_DMA_KEY constant (aka 0). */
-static int put_chars(u32 vtermno, const char *buf, int count)
-{
-	struct lguest_dma dma;
-
-	/* FIXME: DMA buffers in a "struct lguest_dma" are not allowed
-	 * to go over page boundaries.  This never seems to happen,
-	 * but if it did we'd need to fix this code. */
-	dma.len[0] = count;
-	dma.len[1] = 0;
-	dma.addr[0] = __pa(buf);
-
-	lguest_send_dma(LGUEST_CONSOLE_DMA_KEY, &dma);
-	/* We're expected to return the amount of data we wrote: all of it. */
-	return count;
-}
-
-/*D:350 get_chars() is the callback from the hvc_console infrastructure when
- * an interrupt is received.
- *
- * Firstly we see if our buffer has been filled: if not, we return.  The rest
- * of the code deals with the fact that the hvc_console() infrastructure only
- * asks us for 16 bytes at a time.  We keep a "cons_offset" variable for
- * partially-read buffers. */
-static int get_chars(u32 vtermno, char *buf, int count)
-{
-	static int cons_offset;
-
-	/* Nothing left to see here... */
-	if (!cons_input.used_len)
-		return 0;
-
-	/* You want more than we have to give?  Well, try wanting less! */
-	if (cons_input.used_len - cons_offset < count)
-		count = cons_input.used_len - cons_offset;
-
-	/* Copy across to their buffer and increment offset. */
-	memcpy(buf, inbuf + cons_offset, count);
-	cons_offset += count;
-
-	/* Finished?  Zero offset, and reset cons_input so Host will use it
-	 * again. */
-	if (cons_offset == cons_input.used_len) {
-		cons_offset = 0;
-		cons_input.used_len = 0;
-	}
-	return count;
-}
-/*:*/
-
-static struct hv_ops lguest_cons = {
-	.get_chars = get_chars,
-	.put_chars = put_chars,
-};
-
-/*D:320 Console drivers are initialized very early so boot messages can go
- * out.  At this stage, the console is output-only.  Our driver checks we're a
- * Guest, and if so hands hvc_instantiate() the console number (0), priority
- * (0), and the struct hv_ops containing the put_chars() function. */
-static int __init cons_init(void)
-{
-	if (strcmp(pv_info.name, "lguest") != 0)
-		return 0;
-
-	return hvc_instantiate(0, 0, &lguest_cons);
-}
-console_initcall(cons_init);
-
-/*D:370 To set up and manage our virtual console, we call hvc_alloc() and
- * stash the result in the private pointer of the "struct lguest_device".
- * Since we never remove the console device we never need this pointer again,
- * but using ->private is considered good form, and you never know who's going
- * to copy your driver.
- *
- * Once the console is set up, we bind our input buffer ready for input. */
-static int lguestcons_probe(struct lguest_device *lgdev)
-{
-	int err;
-
-	/* The first argument of hvc_alloc() is the virtual console number, so
-	 * we use zero.  The second argument is the interrupt number.
-	 *
-	 * The third argument is a "struct hv_ops" containing the put_chars()
-	 * and get_chars() pointers.  The final argument is the output buffer
-	 * size: we use 256 and expect the Host to have room for us to send
-	 * that much. */
-	lgdev->private = hvc_alloc(0, lgdev_irq(lgdev), &lguest_cons, 256);
-	if (IS_ERR(lgdev->private))
-		return PTR_ERR(lgdev->private);
-
-	/* We bind a single DMA buffer at key LGUEST_CONSOLE_DMA_KEY.
-	 * "cons_input" is that statically-initialized global DMA buffer we saw
-	 * above, and we also give the interrupt we want. */
-	err = lguest_bind_dma(LGUEST_CONSOLE_DMA_KEY, &cons_input, 1,
-			      lgdev_irq(lgdev));
-	if (err)
-		printk("lguest console: failed to bind buffer.\n");
-	return err;
-}
-/* Note the use of lgdev_irq() for the interrupt number.  We tell hvc_alloc()
- * to expect input when this interrupt is triggered, and then tell
- * lguest_bind_dma() that is the interrupt to send us when input comes in. */
-
-/*D:360 From now on the console driver follows standard Guest driver form:
- * register_lguest_driver() registers the device type and probe function, and
- * the probe function sets up the device.
- *
- * The standard "struct lguest_driver": */
-static struct lguest_driver lguestcons_drv = {
-	.name = "lguestcons",
-	.owner = THIS_MODULE,
-	.device_type = LGUEST_DEVICE_T_CONSOLE,
-	.probe = lguestcons_probe,
-};
-
-/* The standard init function */
-static int __init hvc_lguest_init(void)
-{
-	return register_lguest_driver(&lguestcons_drv);
-}
-module_init(hvc_lguest_init);

drivers/char/virtio_console.c

@@ -0,0 +1,225 @@
+/*D:300
+ * The Guest console driver
+ *
+ * Writing console drivers is one of the few remaining Dark Arts in Linux.
+ * Fortunately for us, the path of virtual consoles has been well-trodden by
+ * the PowerPC folks, who wrote "hvc_console.c" to generically support any
+ * virtual console.  We use that infrastructure which only requires us to write
+ * the basic put_chars and get_chars functions and call the right register
+ * functions.
+ :*/
+
+/*M:002 The console can be flooded: while the Guest is processing input the
+ * Host can send more.  Buffering in the Host could alleviate this, but it is a
+ * difficult problem in general. :*/
+/* Copyright (C) 2006, 2007 Rusty Russell, IBM Corporation
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+#include <linux/err.h>
+#include <linux/init.h>
+#include <linux/virtio.h>
+#include <linux/virtio_console.h>
+#include "hvc_console.h"
+
+/*D:340 These represent our input and output console queues, and the virtio
+ * operations for them. */
+static struct virtqueue *in_vq, *out_vq;
+static struct virtio_device *vdev;
+
+/* This is our input buffer, and how much data is left in it. */
+static unsigned int in_len;
+static char *in, *inbuf;
+
+/* The operations for our console. */
+static struct hv_ops virtio_cons;
+
+/*D:310 The put_chars() callback is pretty straightforward.
+ *
+ * We turn the characters into a scatter-gather list, add it to the output
+ * queue and then kick the Host.  Then we sit here waiting for it to finish:
+ * inefficient in theory, but in practice implementations will do it
+ * immediately (lguest's Launcher does). */
+static int put_chars(u32 vtermno, const char *buf, int count)
+{
+	struct scatterlist sg[1];
+	unsigned int len;
+
+	/* This is a convenient routine to initialize a single-elem sg list */
+	sg_init_one(sg, buf, count);
+
+	/* add_buf wants a token to identify this buffer: we hand it any
+	 * non-NULL pointer, since there's only ever one buffer. */
+	if (out_vq->vq_ops->add_buf(out_vq, sg, 1, 0, (void *)1) == 0) {
+		/* Tell Host to go! */
+		out_vq->vq_ops->kick(out_vq);
+		/* Chill out until it's done with the buffer. */
+		while (!out_vq->vq_ops->get_buf(out_vq, &len))
+			cpu_relax();
+	}
+
+	/* We're expected to return the amount of data we wrote: all of it. */
+	return count;
+}
+
+/* Create a scatter-gather list representing our input buffer and put it in the
+ * queue. */
+static void add_inbuf(void)
+{
+	struct scatterlist sg[1];
+	sg_init_one(sg, inbuf, PAGE_SIZE);
+
+	/* We should always be able to add one buffer to an empty queue. */
+	if (in_vq->vq_ops->add_buf(in_vq, sg, 0, 1, inbuf) != 0)
+		BUG();
+	in_vq->vq_ops->kick(in_vq);
+}
+
+/*D:350 get_chars() is the callback from the hvc_console infrastructure when
+ * an interrupt is received.
+ *
+ * Most of the code deals with the fact that the hvc_console() infrastructure
+ * only asks us for 16 bytes at a time.  We keep in_offset and in_used fields
+ * for partially-filled buffers. */
+static int get_chars(u32 vtermno, char *buf, int count)
+{
+	/* If we don't have an input queue yet, we can't get input. */
+	BUG_ON(!in_vq);
+
+	/* No buffer?  Try to get one. */
+	if (!in_len) {
+		in = in_vq->vq_ops->get_buf(in_vq, &in_len);
+		if (!in)
+			return 0;
+	}
+
+	/* You want more than we have to give?  Well, try wanting less! */
+	if (in_len < count)
+		count = in_len;
+
+	/* Copy across to their buffer and increment offset. */
+	memcpy(buf, in, count);
+	in += count;
+	in_len -= count;
+
+	/* Finished?  Re-register buffer so Host will use it again. */
+	if (in_len == 0)
+		add_inbuf();
+
+	return count;
+}
+/*:*/
+
+/*D:320 Console drivers are initialized very early so boot messages can go out,
+ * so we do things slightly differently from the generic virtio initialization
+ * of the net and block drivers.
+ *
+ * At this stage, the console is output-only.  It's too early to set up a
+ * virtqueue, so we let the drivers do some boutique early-output thing. */
+int __init virtio_cons_early_init(int (*put_chars)(u32, const char *, int))
+{
+	virtio_cons.put_chars = put_chars;
+	return hvc_instantiate(0, 0, &virtio_cons);
+}
+
+/*D:370 Once we're further in boot, we get probed like any other virtio device.
+ * At this stage we set up the output virtqueue.
+ *
+ * To set up and manage our virtual console, we call hvc_alloc().  Since we
+ * never remove the console device we never need this pointer again.
+ *
+ * Finally we put our input buffer in the input queue, ready to receive. */
+static int virtcons_probe(struct virtio_device *dev)
+{
+	int err;
+	struct hvc_struct *hvc;
+
+	vdev = dev;
+
+	/* This is the scratch page we use to receive console input */
+	inbuf = kmalloc(PAGE_SIZE, GFP_KERNEL);
+	if (!inbuf) {
+		err = -ENOMEM;
+		goto fail;
+	}
+
+	/* Find the input queue. */
+	/* FIXME: This is why we want to wean off hvc: we do nothing
+	 * when input comes in. */
+	in_vq = vdev->config->find_vq(vdev, NULL);
+	if (IS_ERR(in_vq)) {
+		err = PTR_ERR(in_vq);
+		goto free;
+	}
+
+	out_vq = vdev->config->find_vq(vdev, NULL);
+	if (IS_ERR(out_vq)) {
+		err = PTR_ERR(out_vq);
+		goto free_in_vq;
+	}
+
+	/* Start using the new console output. */
+	virtio_cons.get_chars = get_chars;
+	virtio_cons.put_chars = put_chars;
+
+	/* The first argument of hvc_alloc() is the virtual console number, so
+	 * we use zero.  The second argument is the interrupt number; we
+	 * currently leave this as zero: it would be better not to use the
+	 * hvc mechanism and fix this (FIXME!).
+	 *
+	 * The third argument is a "struct hv_ops" containing the put_chars()
+	 * and get_chars() pointers.  The final argument is the output buffer
+	 * size: we can do any size, so we put PAGE_SIZE here. */
+	hvc = hvc_alloc(0, 0, &virtio_cons, PAGE_SIZE);
+	if (IS_ERR(hvc)) {
+		err = PTR_ERR(hvc);
+		goto free_out_vq;
+	}
+
+	/* Register the input buffer the first time. */
+	add_inbuf();
+	return 0;
+
+free_out_vq:
+	vdev->config->del_vq(out_vq);
+free_in_vq:
+	vdev->config->del_vq(in_vq);
+free:
+	kfree(inbuf);
+fail:
+	return err;
+}
+
+static struct virtio_device_id id_table[] = {
+	{ VIRTIO_ID_CONSOLE, VIRTIO_DEV_ANY_ID },
+	{ 0 },
+};
+
+static struct virtio_driver virtio_console = {
+	.driver.name =	KBUILD_MODNAME,
+	.driver.owner =	THIS_MODULE,
+	.id_table =	id_table,
+	.probe =	virtcons_probe,
+};
+
+static int __init init(void)
+{
+	return register_virtio_driver(&virtio_console);
+}
+module_init(init);
+
+MODULE_DEVICE_TABLE(virtio, id_table);
+MODULE_DESCRIPTION("Virtio console driver");
+MODULE_LICENSE("GPL");

drivers/kvm/Kconfig

@@ -47,4 +47,8 @@ config KVM_AMD
 	  Provides support for KVM on AMD processors equipped with the AMD-V
 	  (SVM) extensions.
 
+# OK, it's a little counter-intuitive to do this, but it puts it neatly under
+# the virtualization menu.
+source drivers/lguest/Kconfig
+
 endif # VIRTUALIZATION

drivers/lguest/Kconfig

@@ -1,7 +1,6 @@
 config LGUEST
 	tristate "Linux hypervisor example code"
-	depends on X86 && PARAVIRT && EXPERIMENTAL && !X86_PAE && FUTEX
-	select LGUEST_GUEST
+	depends on X86_32 && EXPERIMENTAL && !X86_PAE && FUTEX && !(X86_VISWS || X86_VOYAGER)
 	select HVC_DRIVER
 	---help---
 	  This is a very simple module which allows you to run
@@ -18,13 +17,3 @@ config LGUEST_GUEST
 	  The guest needs code built-in, even if the host has lguest
 	  support as a module.  The drivers are tiny, so we build them
 	  in too.
-
-config LGUEST_NET
-	tristate
-	default y
-	depends on LGUEST_GUEST && NET
-
-config LGUEST_BLOCK
-	tristate
-	default y
-	depends on LGUEST_GUEST && BLOCK

drivers/lguest/Makefile

@@ -1,10 +1,12 @@
-# Guest requires the paravirt_ops replacement and the bus driver.
-obj-$(CONFIG_LGUEST_GUEST) += lguest.o lguest_asm.o lguest_bus.o
+# Guest requires the device configuration and probing code.
+obj-$(CONFIG_LGUEST_GUEST) += lguest_device.o
 
 # Host requires the other files, which can be a module.
 obj-$(CONFIG_LGUEST)	+= lg.o
-lg-y := core.o hypercalls.o page_tables.o interrupts_and_traps.o \
-	segments.o io.o lguest_user.o switcher.o
+lg-y = core.o hypercalls.o page_tables.o interrupts_and_traps.o \
+	segments.o lguest_user.o
+
+lg-$(CONFIG_X86_32) += x86/switcher_32.o x86/core.o
 
 Preparation Preparation!: PREFIX=P
 Guest: PREFIX=G

drivers/lguest/core.c

@@ -11,58 +11,20 @@
 #include <linux/vmalloc.h>
 #include <linux/cpu.h>
 #include <linux/freezer.h>
+#include <linux/highmem.h>
 #include <asm/paravirt.h>
-#include <asm/desc.h>
 #include <asm/pgtable.h>
 #include <asm/uaccess.h>
 #include <asm/poll.h>
-#include <asm/highmem.h>
 #include <asm/asm-offsets.h>
-#include <asm/i387.h>
 #include "lg.h"
 
-/* Found in switcher.S */
-extern char start_switcher_text[], end_switcher_text[], switch_to_guest[];
-extern unsigned long default_idt_entries[];
-
-/* Every guest maps the core switcher code. */
-#define SHARED_SWITCHER_PAGES \
-	DIV_ROUND_UP(end_switcher_text - start_switcher_text, PAGE_SIZE)
-/* Pages for switcher itself, then two pages per cpu */
-#define TOTAL_SWITCHER_PAGES (SHARED_SWITCHER_PAGES + 2 * NR_CPUS)
-
-/* We map at -4M for ease of mapping into the guest (one PTE page). */
-#define SWITCHER_ADDR 0xFFC00000
 
 static struct vm_struct *switcher_vma;
 static struct page **switcher_page;
 
-static int cpu_had_pge;
-static struct {
-	unsigned long offset;
-	unsigned short segment;
-} lguest_entry;
-
 /* This One Big lock protects all inter-guest data structures. */
 DEFINE_MUTEX(lguest_lock);
-static DEFINE_PER_CPU(struct lguest *, last_guest);
-
-/* FIXME: Make dynamic. */
-#define MAX_LGUEST_GUESTS 16
-struct lguest lguests[MAX_LGUEST_GUESTS];
-
-/* Offset from where switcher.S was compiled to where we've copied it */
-static unsigned long switcher_offset(void)
-{
-	return SWITCHER_ADDR - (unsigned long)start_switcher_text;
-}
-
-/* This cpu's struct lguest_pages. */
-static struct lguest_pages *lguest_pages(unsigned int cpu)
-{
-	return &(((struct lguest_pages *)
-		  (SWITCHER_ADDR + SHARED_SWITCHER_PAGES*PAGE_SIZE))[cpu]);
-}
 
 /*H:010 We need to set up the Switcher at a high virtual address.  Remember the
  * Switcher is a few hundred bytes of assembler code which actually changes the
@@ -73,9 +35,7 @@ static struct lguest_pages *lguest_pages(unsigned int cpu)
  * Host since it will be running as the switchover occurs.
  *
  * Trying to map memory at a particular address is an unusual thing to do, so
- * it's not a simple one-liner.  We also set up the per-cpu parts of the
- * Switcher here.
- */
+ * it's not a simple one-liner. */
 static __init int map_switcher(void)
 {
 	int i, err;
@@ -132,90 +92,11 @@ static __init int map_switcher(void)
 		goto free_vma;
 	}
 
-	/* Now the switcher is mapped at the right address, we can't fail!
-	 * Copy in the compiled-in Switcher code (from switcher.S). */
+	/* Now the Switcher is mapped at the right address, we can't fail!
+	 * Copy in the compiled-in Switcher code (from <arch>_switcher.S). */
 	memcpy(switcher_vma->addr, start_switcher_text,
 	       end_switcher_text - start_switcher_text);
 
-	/* Most of the switcher.S doesn't care that it's been moved; on Intel,
-	 * jumps are relative, and it doesn't access any references to external
-	 * code or data.
-	 *
-	 * The only exception is the interrupt handlers in switcher.S: their
-	 * addresses are placed in a table (default_idt_entries), so we need to
-	 * update the table with the new addresses.  switcher_offset() is a
-	 * convenience function which returns the distance between the builtin
-	 * switcher code and the high-mapped copy we just made. */
-	for (i = 0; i < IDT_ENTRIES; i++)
-		default_idt_entries[i] += switcher_offset();
-
-	/*
-	 * Set up the Switcher's per-cpu areas.
-	 *
-	 * Each CPU gets two pages of its own within the high-mapped region
-	 * (aka. "struct lguest_pages").  Much of this can be initialized now,
-	 * but some depends on what Guest we are running (which is set up in
-	 * copy_in_guest_info()).
-	 */
-	for_each_possible_cpu(i) {
-		/* lguest_pages() returns this CPU's two pages. */
-		struct lguest_pages *pages = lguest_pages(i);
-		/* This is a convenience pointer to make the code fit one
-		 * statement to a line. */
-		struct lguest_ro_state *state = &pages->state;
-
-		/* The Global Descriptor Table: the Host has a different one
-		 * for each CPU.  We keep a descriptor for the GDT which says
-		 * where it is and how big it is (the size is actually the last
-		 * byte, not the size, hence the "-1"). */
-		state->host_gdt_desc.size = GDT_SIZE-1;
-		state->host_gdt_desc.address = (long)get_cpu_gdt_table(i);
-
-		/* All CPUs on the Host use the same Interrupt Descriptor
-		 * Table, so we just use store_idt(), which gets this CPU's IDT
-		 * descriptor. */
-		store_idt(&state->host_idt_desc);
-
-		/* The descriptors for the Guest's GDT and IDT can be filled
-		 * out now, too.  We copy the GDT & IDT into ->guest_gdt and
-		 * ->guest_idt before actually running the Guest. */
-		state->guest_idt_desc.size = sizeof(state->guest_idt)-1;
-		state->guest_idt_desc.address = (long)&state->guest_idt;
-		state->guest_gdt_desc.size = sizeof(state->guest_gdt)-1;
-		state->guest_gdt_desc.address = (long)&state->guest_gdt;
-
-		/* We know where we want the stack to be when the Guest enters
-		 * the switcher: in pages->regs.  The stack grows upwards, so
-		 * we start it at the end of that structure. */
-		state->guest_tss.esp0 = (long)(&pages->regs + 1);
-		/* And this is the GDT entry to use for the stack: we keep a
-		 * couple of special LGUEST entries. */
-		state->guest_tss.ss0 = LGUEST_DS;
-
-		/* x86 can have a finegrained bitmap which indicates what I/O
-		 * ports the process can use.  We set it to the end of our
-		 * structure, meaning "none". */
-		state->guest_tss.io_bitmap_base = sizeof(state->guest_tss);
-
-		/* Some GDT entries are the same across all Guests, so we can
-		 * set them up now. */
-		setup_default_gdt_entries(state);
-		/* Most IDT entries are the same for all Guests, too.*/
-		setup_default_idt_entries(state, default_idt_entries);
-
-		/* The Host needs to be able to use the LGUEST segments on this
-		 * CPU, too, so put them in the Host GDT. */
-		get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT;
-		get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT;
-	}
-
-	/* In the Switcher, we want the %cs segment register to use the
-	 * LGUEST_CS GDT entry: we've put that in the Host and Guest GDTs, so
-	 * it will be undisturbed when we switch.  To change %cs and jump we
-	 * need this structure to feed to Intel's "lcall" instruction. */
-	lguest_entry.offset = (long)switch_to_guest + switcher_offset();
-	lguest_entry.segment = LGUEST_CS;
-
 	printk(KERN_INFO "lguest: mapped switcher at %p\n",
 	       switcher_vma->addr);
 	/* And we succeeded... */
@@ -247,86 +128,12 @@ static void unmap_switcher(void)
 		__free_pages(switcher_page[i], 0);
 }
 
-/*H:130 Our Guest is usually so well behaved; it never tries to do things it
- * isn't allowed to.  Unfortunately, Linux's paravirtual infrastructure isn't
- * quite complete, because it doesn't contain replacements for the Intel I/O
- * instructions.  As a result, the Guest sometimes fumbles across one during
- * the boot process as it probes for various things which are usually attached
- * to a PC.
- *
- * When the Guest uses one of these instructions, we get trap #13 (General
- * Protection Fault) and come here.  We see if it's one of those troublesome
- * instructions and skip over it.  We return true if we did. */
-static int emulate_insn(struct lguest *lg)
-{
-	u8 insn;
-	unsigned int insnlen = 0, in = 0, shift = 0;
-	/* The eip contains the *virtual* address of the Guest's instruction:
-	 * guest_pa just subtracts the Guest's page_offset. */
-	unsigned long physaddr = guest_pa(lg, lg->regs->eip);
-
-	/* The guest_pa() function only works for Guest kernel addresses, but
-	 * that's all we're trying to do anyway. */
-	if (lg->regs->eip < lg->page_offset)
-		return 0;
-
-	/* Decoding x86 instructions is icky. */
-	lgread(lg, &insn, physaddr, 1);
-
-	/* 0x66 is an "operand prefix".  It means it's using the upper 16 bits
-	   of the eax register. */
-	if (insn == 0x66) {
-		shift = 16;
-		/* The instruction is 1 byte so far, read the next byte. */
-		insnlen = 1;
-		lgread(lg, &insn, physaddr + insnlen, 1);
-	}
-
-	/* We can ignore the lower bit for the moment and decode the 4 opcodes
-	 * we need to emulate. */
-	switch (insn & 0xFE) {
-	case 0xE4: /* in     <next byte>,%al */
-		insnlen += 2;
-		in = 1;
-		break;
-	case 0xEC: /* in     (%dx),%al */
-		insnlen += 1;
-		in = 1;
-		break;
-	case 0xE6: /* out    %al,<next byte> */
-		insnlen += 2;
-		break;
-	case 0xEE: /* out    %al,(%dx) */
-		insnlen += 1;
-		break;
-	default:
-		/* OK, we don't know what this is, can't emulate. */
-		return 0;
-	}
-
-	/* If it was an "IN" instruction, they expect the result to be read
-	 * into %eax, so we change %eax.  We always return all-ones, which
-	 * traditionally means "there's nothing there". */
-	if (in) {
-		/* Lower bit tells is whether it's a 16 or 32 bit access */
-		if (insn & 0x1)
-			lg->regs->eax = 0xFFFFFFFF;
-		else
-			lg->regs->eax |= (0xFFFF << shift);
-	}
-	/* Finally, we've "done" the instruction, so move past it. */
-	lg->regs->eip += insnlen;
-	/* Success! */
-	return 1;
-}
-/*:*/
-
 /*L:305
  * Dealing With Guest Memory.
  *
  * When the Guest gives us (what it thinks is) a physical address, we can use
- * the normal copy_from_user() & copy_to_user() on that address: remember,
- * Guest physical == Launcher virtual.
+ * the normal copy_from_user() & copy_to_user() on the corresponding place in
+ * the memory region allocated by the Launcher.
  *
  * But we can't trust the Guest: it might be trying to access the Launcher
  * code.  We have to check that the range is below the pfn_limit the Launcher
@@ -338,148 +145,27 @@ int lguest_address_ok(const struct lguest *lg,
 	return (addr+len) / PAGE_SIZE < lg->pfn_limit && (addr+len >= addr);
 }
 
-/* This is a convenient routine to get a 32-bit value from the Guest (a very
- * common operation).  Here we can see how useful the kill_lguest() routine we
- * met in the Launcher can be: we return a random value (0) instead of needing
- * to return an error. */
-u32 lgread_u32(struct lguest *lg, unsigned long addr)
-{
-	u32 val = 0;
-
-	/* Don't let them access lguest binary. */
-	if (!lguest_address_ok(lg, addr, sizeof(val))
-	    || get_user(val, (u32 __user *)addr) != 0)
-		kill_guest(lg, "bad read address %#lx", addr);
-	return val;
-}
-
-/* Same thing for writing a value. */
-void lgwrite_u32(struct lguest *lg, unsigned long addr, u32 val)
-{
-	if (!lguest_address_ok(lg, addr, sizeof(val))
-	    || put_user(val, (u32 __user *)addr) != 0)
-		kill_guest(lg, "bad write address %#lx", addr);
-}
-
-/* This routine is more generic, and copies a range of Guest bytes into a
- * buffer.  If the copy_from_user() fails, we fill the buffer with zeroes, so
- * the caller doesn't end up using uninitialized kernel memory. */
-void lgread(struct lguest *lg, void *b, unsigned long addr, unsigned bytes)
+/* This routine copies memory from the Guest.  Here we can see how useful the
+ * kill_lguest() routine we met in the Launcher can be: we return a random
+ * value (all zeroes) instead of needing to return an error. */
+void __lgread(struct lguest *lg, void *b, unsigned long addr, unsigned bytes)
 {
 	if (!lguest_address_ok(lg, addr, bytes)
-	    || copy_from_user(b, (void __user *)addr, bytes) != 0) {
+	    || copy_from_user(b, lg->mem_base + addr, bytes) != 0) {
 		/* copy_from_user should do this, but as we rely on it... */
 		memset(b, 0, bytes);
 		kill_guest(lg, "bad read address %#lx len %u", addr, bytes);
 	}
 }
 
-/* Similarly, our generic routine to copy into a range of Guest bytes. */
-void lgwrite(struct lguest *lg, unsigned long addr, const void *b,
-	     unsigned bytes)
+/* This is the write (copy into guest) version. */
+void __lgwrite(struct lguest *lg, unsigned long addr, const void *b,
+	       unsigned bytes)
 {
 	if (!lguest_address_ok(lg, addr, bytes)
-	    || copy_to_user((void __user *)addr, b, bytes) != 0)
+	    || copy_to_user(lg->mem_base + addr, b, bytes) != 0)
 		kill_guest(lg, "bad write address %#lx len %u", addr, bytes);
 }
-/* (end of memory access helper routines) :*/
-
-static void set_ts(void)
-{
-	u32 cr0;
-
-	cr0 = read_cr0();
-	if (!(cr0 & 8))
-		write_cr0(cr0|8);
-}
-
-/*S:010
- * We are getting close to the Switcher.
- *
- * Remember that each CPU has two pages which are visible to the Guest when it
- * runs on that CPU.  This has to contain the state for that Guest: we copy the
- * state in just before we run the Guest.
- *
- * Each Guest has "changed" flags which indicate what has changed in the Guest
- * since it last ran.  We saw this set in interrupts_and_traps.c and
- * segments.c.
- */
-static void copy_in_guest_info(struct lguest *lg, struct lguest_pages *pages)
-{
-	/* Copying all this data can be quite expensive.  We usually run the
-	 * same Guest we ran last time (and that Guest hasn't run anywhere else
-	 * meanwhile).  If that's not the case, we pretend everything in the
-	 * Guest has changed. */
-	if (__get_cpu_var(last_guest) != lg || lg->last_pages != pages) {
-		__get_cpu_var(last_guest) = lg;
-		lg->last_pages = pages;
-		lg->changed = CHANGED_ALL;
-	}
-
-	/* These copies are pretty cheap, so we do them unconditionally: */
-	/* Save the current Host top-level page directory. */
-	pages->state.host_cr3 = __pa(current->mm->pgd);
-	/* Set up the Guest's page tables to see this CPU's pages (and no
-	 * other CPU's pages). */
-	map_switcher_in_guest(lg, pages);
-	/* Set up the two "TSS" members which tell the CPU what stack to use
-	 * for traps which do directly into the Guest (ie. traps at privilege
-	 * level 1). */
-	pages->state.guest_tss.esp1 = lg->esp1;
-	pages->state.guest_tss.ss1 = lg->ss1;
-
-	/* Copy direct-to-Guest trap entries. */
-	if (lg->changed & CHANGED_IDT)
-		copy_traps(lg, pages->state.guest_idt, default_idt_entries);
-
-	/* Copy all GDT entries which the Guest can change. */
-	if (lg->changed & CHANGED_GDT)
-		copy_gdt(lg, pages->state.guest_gdt);
-	/* If only the TLS entries have changed, copy them. */
-	else if (lg->changed & CHANGED_GDT_TLS)
-		copy_gdt_tls(lg, pages->state.guest_gdt);
-
-	/* Mark the Guest as unchanged for next time. */
-	lg->changed = 0;
-}
-
-/* Finally: the code to actually call into the Switcher to run the Guest. */
-static void run_guest_once(struct lguest *lg, struct lguest_pages *pages)
-{
-	/* This is a dummy value we need for GCC's sake. */
-	unsigned int clobber;
-
-	/* Copy the guest-specific information into this CPU's "struct
-	 * lguest_pages". */
-	copy_in_guest_info(lg, pages);
-
-	/* Set the trap number to 256 (impossible value).  If we fault while
-	 * switching to the Guest (bad segment registers or bug), this will
-	 * cause us to abort the Guest. */
-	lg->regs->trapnum = 256;
-
-	/* Now: we push the "eflags" register on the stack, then do an "lcall".
-	 * This is how we change from using the kernel code segment to using
-	 * the dedicated lguest code segment, as well as jumping into the
-	 * Switcher.
-	 *
-	 * The lcall also pushes the old code segment (KERNEL_CS) onto the
-	 * stack, then the address of this call.  This stack layout happens to
-	 * exactly match the stack of an interrupt... */
-	asm volatile("pushf; lcall *lguest_entry"
-		     /* This is how we tell GCC that %eax ("a") and %ebx ("b")
-		      * are changed by this routine.  The "=" means output. */
-		     : "=a"(clobber), "=b"(clobber)
-		     /* %eax contains the pages pointer.  ("0" refers to the
-		      * 0-th argument above, ie "a").  %ebx contains the
-		      * physical address of the Guest's top-level page
-		      * directory. */
-		     : "0"(pages), "1"(__pa(lg->pgdirs[lg->pgdidx].pgdir))
-		     /* We tell gcc that all these registers could change,
-		      * which means we don't have to save and restore them in
-		      * the Switcher. */
-		     : "memory", "%edx", "%ecx", "%edi", "%esi");
-}
 /*:*/
 
 /*H:030 Let's jump straight to the the main loop which runs the Guest.
@@ -489,22 +175,16 @@ int run_guest(struct lguest *lg, unsigned long __user *user)
 {
 	/* We stop running once the Guest is dead. */
 	while (!lg->dead) {
-		/* We need to initialize this, otherwise gcc complains.  It's
-		 * not (yet) clever enough to see that it's initialized when we
-		 * need it. */
-		unsigned int cr2 = 0; /* Damn gcc */
-
-		/* First we run any hypercalls the Guest wants done: either in
-		 * the hypercall ring in "struct lguest_data", or directly by
-		 * using int 31 (LGUEST_TRAP_ENTRY). */
-		do_hypercalls(lg);
-		/* It's possible the Guest did a SEND_DMA hypercall to the
+		/* First we run any hypercalls the Guest wants done. */
+		if (lg->hcall)
+			do_hypercalls(lg);
+
+		/* It's possible the Guest did a NOTIFY hypercall to the
 		 * Launcher, in which case we return from the read() now. */
-		if (lg->dma_is_pending) {
-			if (put_user(lg->pending_dma, user) ||
-			    put_user(lg->pending_key, user+1))
+		if (lg->pending_notify) {
+			if (put_user(lg->pending_notify, user))
 				return -EFAULT;
-			return sizeof(unsigned long)*2;
+			return sizeof(lg->pending_notify);
 		}
 
 		/* Check for signals */
@@ -542,144 +222,20 @@ int run_guest(struct lguest *lg, unsigned long __user *user)
 		 * the "Do Not Disturb" sign: */
 		local_irq_disable();
 
-		/* Remember the awfully-named TS bit?  If the Guest has asked
-		 * to set it we set it now, so we can trap and pass that trap
-		 * to the Guest if it uses the FPU. */
-		if (lg->ts)
-			set_ts();
-
-		/* SYSENTER is an optimized way of doing system calls.  We
-		 * can't allow it because it always jumps to privilege level 0.
-		 * A normal Guest won't try it because we don't advertise it in
-		 * CPUID, but a malicious Guest (or malicious Guest userspace
-		 * program) could, so we tell the CPU to disable it before
-		 * running the Guest. */
-		if (boot_cpu_has(X86_FEATURE_SEP))
-			wrmsr(MSR_IA32_SYSENTER_CS, 0, 0);
-
-		/* Now we actually run the Guest.  It will pop back out when
-		 * something interesting happens, and we can examine its
-		 * registers to see what it was doing. */
-		run_guest_once(lg, lguest_pages(raw_smp_processor_id()));
-
-		/* The "regs" pointer contains two extra entries which are not
-		 * really registers: a trap number which says what interrupt or
-		 * trap made the switcher code come back, and an error code
-		 * which some traps set.  */
-
-		/* If the Guest page faulted, then the cr2 register will tell
-		 * us the bad virtual address.  We have to grab this now,
-		 * because once we re-enable interrupts an interrupt could
-		 * fault and thus overwrite cr2, or we could even move off to a
-		 * different CPU. */
-		if (lg->regs->trapnum == 14)
-			cr2 = read_cr2();
-		/* Similarly, if we took a trap because the Guest used the FPU,
-		 * we have to restore the FPU it expects to see. */
-		else if (lg->regs->trapnum == 7)
-			math_state_restore();
-
-		/* Restore SYSENTER if it's supposed to be on. */
-		if (boot_cpu_has(X86_FEATURE_SEP))
-			wrmsr(MSR_IA32_SYSENTER_CS, __KERNEL_CS, 0);
+		/* Actually run the Guest until something happens. */
+		lguest_arch_run_guest(lg);
 
 		/* Now we're ready to be interrupted or moved to other CPUs */
 		local_irq_enable();
 
-		/* OK, so what happened? */
-		switch (lg->regs->trapnum) {
-		case 13: /* We've intercepted a GPF. */
-			/* Check if this was one of those annoying IN or OUT
-			 * instructions which we need to emulate.  If so, we
-			 * just go back into the Guest after we've done it. */
-			if (lg->regs->errcode == 0) {
-				if (emulate_insn(lg))
-					continue;
-			}
-			break;
-		case 14: /* We've intercepted a page fault. */
-			/* The Guest accessed a virtual address that wasn't
-			 * mapped.  This happens a lot: we don't actually set
-			 * up most of the page tables for the Guest at all when
-			 * we start: as it runs it asks for more and more, and
-			 * we set them up as required. In this case, we don't
-			 * even tell the Guest that the fault happened.
-			 *
-			 * The errcode tells whether this was a read or a
-			 * write, and whether kernel or userspace code. */
-			if (demand_page(lg, cr2, lg->regs->errcode))
-				continue;
-
-			/* OK, it's really not there (or not OK): the Guest
-			 * needs to know.  We write out the cr2 value so it
-			 * knows where the fault occurred.
-			 *
-			 * Note that if the Guest were really messed up, this
-			 * could happen before it's done the INITIALIZE
-			 * hypercall, so lg->lguest_data will be NULL, so
-			 * &lg->lguest_data->cr2 will be address 8.  Writing
-			 * into that address won't hurt the Host at all,
-			 * though. */
-			if (put_user(cr2, &lg->lguest_data->cr2))
-				kill_guest(lg, "Writing cr2");
-			break;
-		case 7: /* We've intercepted a Device Not Available fault. */
-			/* If the Guest doesn't want to know, we already
-			 * restored the Floating Point Unit, so we just
-			 * continue without telling it. */
-			if (!lg->ts)
-				continue;
-			break;
-		case 32 ... 255:
-			/* These values mean a real interrupt occurred, in
-			 * which case the Host handler has already been run.
-			 * We just do a friendly check if another process
-			 * should now be run, then fall through to loop
-			 * around: */
-			cond_resched();
-		case LGUEST_TRAP_ENTRY: /* Handled at top of loop */
-			continue;
-		}
-
-		/* If we get here, it's a trap the Guest wants to know
-		 * about. */
-		if (deliver_trap(lg, lg->regs->trapnum))
-			continue;
-
-		/* If the Guest doesn't have a handler (either it hasn't
-		 * registered any yet, or it's one of the faults we don't let
-		 * it handle), it dies with a cryptic error message. */
-		kill_guest(lg, "unhandled trap %li at %#lx (%#lx)",
-			   lg->regs->trapnum, lg->regs->eip,
-			   lg->regs->trapnum == 14 ? cr2 : lg->regs->errcode);
+		/* Now we deal with whatever happened to the Guest. */
+		lguest_arch_handle_trap(lg);
 	}
+
 	/* The Guest is dead => "No such file or directory" */
 	return -ENOENT;
 }
 
-/* Now we can look at each of the routines this calls, in increasing order of
- * complexity: do_hypercalls(), emulate_insn(), maybe_do_interrupt(),
- * deliver_trap() and demand_page().  After all those, we'll be ready to
- * examine the Switcher, and our philosophical understanding of the Host/Guest
- * duality will be complete. :*/
-
-int find_free_guest(void)
-{
-	unsigned int i;
-	for (i = 0; i < MAX_LGUEST_GUESTS; i++)
-		if (!lguests[i].tsk)
-			return i;
-	return -1;
-}
-
-static void adjust_pge(void *on)
-{
-	if (on)
-		write_cr4(read_cr4() | X86_CR4_PGE);
-	else
-		write_cr4(read_cr4() & ~X86_CR4_PGE);
-}
-
 /*H:000
  * Welcome to the Host!
  *
@@ -701,72 +257,50 @@ static int __init init(void)
 	/* First we put the Switcher up in very high virtual memory. */
 	err = map_switcher();
 	if (err)
-		return err;
+		goto out;
 
 	/* Now we set up the pagetable implementation for the Guests. */
 	err = init_pagetables(switcher_page, SHARED_SWITCHER_PAGES);
-	if (err) {
-		unmap_switcher();
-		return err;
-	}
+	if (err)
+		goto unmap;
 
-	/* The I/O subsystem needs some things initialized. */
-	lguest_io_init();
+	/* We might need to reserve an interrupt vector. */
+	err = init_interrupts();
+	if (err)
+		goto free_pgtables;
 
 	/* /dev/lguest needs to be registered. */
 	err = lguest_device_init();
-	if (err) {
-		free_pagetables();
-		unmap_switcher();
-		return err;
-	}
+	if (err)
+		goto free_interrupts;
 
-	/* Finally, we need to turn off "Page Global Enable".  PGE is an
-	 * optimization where page table entries are specially marked to show
-	 * they never change.  The Host kernel marks all the kernel pages this
-	 * way because it's always present, even when userspace is running.
-	 *
-	 * Lguest breaks this: unbeknownst to the rest of the Host kernel, we
-	 * switch to the Guest kernel.  If you don't disable this on all CPUs,
-	 * you'll get really weird bugs that you'll chase for two days.
-	 *
-	 * I used to turn PGE off every time we switched to the Guest and back
-	 * on when we return, but that slowed the Switcher down noticibly. */
-
-	/* We don't need the complexity of CPUs coming and going while we're
-	 * doing this. */
-	lock_cpu_hotplug();
-	if (cpu_has_pge) { /* We have a broader idea of "global". */
-		/* Remember that this was originally set (for cleanup). */
-		cpu_had_pge = 1;
-		/* adjust_pge is a helper function which sets or unsets the PGE
-		 * bit on its CPU, depending on the argument (0 == unset). */
-		on_each_cpu(adjust_pge, (void *)0, 0, 1);
-		/* Turn off the feature in the global feature set. */
-		clear_bit(X86_FEATURE_PGE, boot_cpu_data.x86_capability);
-	}
-	unlock_cpu_hotplug();
+	/* Finally we do some architecture-specific setup. */
+	lguest_arch_host_init();
 
 	/* All good! */
 	return 0;
+
+free_interrupts:
+	free_interrupts();
+free_pgtables:
+	free_pagetables();
+unmap:
+	unmap_switcher();
+out:
+	return err;
 }
 
 /* Cleaning up is just the same code, backwards.  With a little French. */
 static void __exit fini(void)
 {
 	lguest_device_remove();
+	free_interrupts();
 	free_pagetables();
 	unmap_switcher();
 
-	/* If we had PGE before we started, turn it back on now. */
-	lock_cpu_hotplug();
-	if (cpu_had_pge) {
-		set_bit(X86_FEATURE_PGE, boot_cpu_data.x86_capability);
-		/* adjust_pge's argument "1" means set PGE. */
-		on_each_cpu(adjust_pge, (void *)1, 0, 1);
-	}
-	unlock_cpu_hotplug();
+	lguest_arch_host_fini();
 }
+/*:*/
 
 /* The Host side of lguest can be a module.  This is a nice way for people to
  * play with it.  */

drivers/lguest/hypercalls.c

@@ -25,17 +25,13 @@
 #include <linux/mm.h>
 #include <asm/page.h>
 #include <asm/pgtable.h>
-#include <irq_vectors.h>
 #include "lg.h"
 
-/*H:120 This is the core hypercall routine: where the Guest gets what it
- * wants.  Or gets killed.  Or, in the case of LHCALL_CRASH, both.
- *
- * Remember from the Guest: %eax == which call to make, and the arguments are
- * packed into %edx, %ebx and %ecx if needed. */
-static void do_hcall(struct lguest *lg, struct lguest_regs *regs)
+/*H:120 This is the core hypercall routine: where the Guest gets what it wants.
+ * Or gets killed.  Or, in the case of LHCALL_CRASH, both. */
+static void do_hcall(struct lguest *lg, struct hcall_args *args)
 {
-	switch (regs->eax) {
+	switch (args->arg0) {
 	case LHCALL_FLUSH_ASYNC:
 		/* This call does nothing, except by breaking out of the Guest
 		 * it makes us process all the asynchronous hypercalls. */
@@ -51,7 +47,7 @@ static void do_hcall(struct lguest *lg, struct lguest_regs *regs)
 		char msg[128];
 		/* If the lgread fails, it will call kill_guest() itself; the
 		 * kill_guest() with the message will be ignored. */
-		lgread(lg, msg, regs->edx, sizeof(msg));
+		__lgread(lg, msg, args->arg1, sizeof(msg));
 		msg[sizeof(msg)-1] = '\0';
 		kill_guest(lg, "CRASH: %s", msg);
 		break;
@@ -59,67 +55,49 @@ static void do_hcall(struct lguest *lg, struct lguest_regs *regs)
 	case LHCALL_FLUSH_TLB:
 		/* FLUSH_TLB comes in two flavors, depending on the
 		 * argument: */
-		if (regs->edx)
+		if (args->arg1)
 			guest_pagetable_clear_all(lg);
 		else
 			guest_pagetable_flush_user(lg);
 		break;
-	case LHCALL_BIND_DMA:
-		/* BIND_DMA really wants four arguments, but it's the only call
-		 * which does.  So the Guest packs the number of buffers and
-		 * the interrupt number into the final argument, and we decode
-		 * it here.  This can legitimately fail, since we currently
-		 * place a limit on the number of DMA pools a Guest can have.
-		 * So we return true or false from this call. */
-		regs->eax = bind_dma(lg, regs->edx, regs->ebx,
-				     regs->ecx >> 8, regs->ecx & 0xFF);
-		break;
 
 	/* All these calls simply pass the arguments through to the right
 	 * routines. */
-	case LHCALL_SEND_DMA:
-		send_dma(lg, regs->edx, regs->ebx);
-		break;
-	case LHCALL_LOAD_GDT:
-		load_guest_gdt(lg, regs->edx, regs->ebx);
-		break;
-	case LHCALL_LOAD_IDT_ENTRY:
-		load_guest_idt_entry(lg, regs->edx, regs->ebx, regs->ecx);
-		break;
 	case LHCALL_NEW_PGTABLE:
-		guest_new_pagetable(lg, regs->edx);
+		guest_new_pagetable(lg, args->arg1);
 		break;
 	case LHCALL_SET_STACK:
-		guest_set_stack(lg, regs->edx, regs->ebx, regs->ecx);
+		guest_set_stack(lg, args->arg1, args->arg2, args->arg3);
 		break;
 	case LHCALL_SET_PTE:
-		guest_set_pte(lg, regs->edx, regs->ebx, mkgpte(regs->ecx));
+		guest_set_pte(lg, args->arg1, args->arg2, __pte(args->arg3));
 		break;
 	case LHCALL_SET_PMD:
-		guest_set_pmd(lg, regs->edx, regs->ebx);
-		break;
-	case LHCALL_LOAD_TLS:
-		guest_load_tls(lg, regs->edx);
+		guest_set_pmd(lg, args->arg1, args->arg2);
 		break;
 	case LHCALL_SET_CLOCKEVENT:
-		guest_set_clockevent(lg, regs->edx);
+		guest_set_clockevent(lg, args->arg1);
 		break;
-
 	case LHCALL_TS:
 		/* This sets the TS flag, as we saw used in run_guest(). */
-		lg->ts = regs->edx;
+		lg->ts = args->arg1;
 		break;
 	case LHCALL_HALT:
 		/* Similarly, this sets the halted flag for run_guest(). */
 		lg->halted = 1;
 		break;
+	case LHCALL_NOTIFY:
+		lg->pending_notify = args->arg1;
+		break;
 	default:
-		kill_guest(lg, "Bad hypercall %li\n", regs->eax);
+		if (lguest_arch_do_hcall(lg, args))
+			kill_guest(lg, "Bad hypercall %li\n", args->arg0);
 	}
 }
+/*:*/
 
-/* Asynchronous hypercalls are easy: we just look in the array in the Guest's
- * "struct lguest_data" and see if there are any new ones marked "ready".
+/*H:124 Asynchronous hypercalls are easy: we just look in the array in the
+ * Guest's "struct lguest_data" to see if any new ones are marked "ready".
  *
  * We are careful to do these in order: obviously we respect the order the
  * Guest put them in the ring, but we also promise the Guest that they will
@@ -134,10 +112,9 @@ static void do_async_hcalls(struct lguest *lg)
 	if (copy_from_user(&st, &lg->lguest_data->hcall_status, sizeof(st)))
 		return;
 
-
 	/* We process "struct lguest_data"s hcalls[] ring once. */
 	for (i = 0; i < ARRAY_SIZE(st); i++) {
-		struct lguest_regs regs;
+		struct hcall_args args;
 		/* We remember where we were up to from last time.  This makes
 		 * sure that the hypercalls are done in the order the Guest
 		 * places them in the ring. */
@@ -152,18 +129,16 @@ static void do_async_hcalls(struct lguest *lg)
 		if (++lg->next_hcall == LHCALL_RING_SIZE)
 			lg->next_hcall = 0;
 
-		/* We copy the hypercall arguments into a fake register
-		 * structure.  This makes life simple for do_hcall(). */
-		if (get_user(regs.eax, &lg->lguest_data->hcalls[n].eax)
-		    || get_user(regs.edx, &lg->lguest_data->hcalls[n].edx)
-		    || get_user(regs.ecx, &lg->lguest_data->hcalls[n].ecx)
-		    || get_user(regs.ebx, &lg->lguest_data->hcalls[n].ebx)) {
+		/* Copy the hypercall arguments into a local copy of
+		 * the hcall_args struct. */
+		if (copy_from_user(&args, &lg->lguest_data->hcalls[n],
+				   sizeof(struct hcall_args))) {
 			kill_guest(lg, "Fetching async hypercalls");
 			break;
 		}
 
 		/* Do the hypercall, same as a normal one. */
-		do_hcall(lg, &regs);
+		do_hcall(lg, &args);
 
 		/* Mark the hypercall done. */
 		if (put_user(0xFF, &lg->lguest_data->hcall_status[n])) {
@@ -171,9 +146,9 @@ static void do_async_hcalls(struct lguest *lg)
 			break;
 		}
 
- 		/* Stop doing hypercalls if we've just done a DMA to the
-		 * Launcher: it needs to service this first. */
-		if (lg->dma_is_pending)
+		/* Stop doing hypercalls if they want to notify the Launcher:
+		 * it needs to service this first. */
+		if (lg->pending_notify)
 			break;
 	}
 }
@@ -182,76 +157,35 @@ static void do_async_hcalls(struct lguest *lg)
  * Guest makes a hypercall, we end up here to set things up: */
 static void initialize(struct lguest *lg)
 {
-	u32 tsc_speed;
 
 	/* You can't do anything until you're initialized.  The Guest knows the
 	 * rules, so we're unforgiving here. */
-	if (lg->regs->eax != LHCALL_LGUEST_INIT) {
-		kill_guest(lg, "hypercall %li before LGUEST_INIT",
-			   lg->regs->eax);
+	if (lg->hcall->arg0 != LHCALL_LGUEST_INIT) {
+		kill_guest(lg, "hypercall %li before INIT", lg->hcall->arg0);
 		return;
 	}
 
-	/* We insist that the Time Stamp Counter exist and doesn't change with
-	 * cpu frequency.  Some devious chip manufacturers decided that TSC
-	 * changes could be handled in software.  I decided that time going
-	 * backwards might be good for benchmarks, but it's bad for users.
-	 *
-	 * We also insist that the TSC be stable: the kernel detects unreliable
-	 * TSCs for its own purposes, and we use that here. */
-	if (boot_cpu_has(X86_FEATURE_CONSTANT_TSC) && !check_tsc_unstable())
-		tsc_speed = tsc_khz;
-	else
-		tsc_speed = 0;
-
-	/* The pointer to the Guest's "struct lguest_data" is the only
-	 * argument. */
-	lg->lguest_data = (struct lguest_data __user *)lg->regs->edx;
-	/* If we check the address they gave is OK now, we can simply
-	 * copy_to_user/from_user from now on rather than using lgread/lgwrite.
-	 * I put this in to show that I'm not immune to writing stupid
-	 * optimizations. */
-	if (!lguest_address_ok(lg, lg->regs->edx, sizeof(*lg->lguest_data))) {
+	if (lguest_arch_init_hypercalls(lg))
 		kill_guest(lg, "bad guest page %p", lg->lguest_data);
-		return;
-	}
+
 	/* The Guest tells us where we're not to deliver interrupts by putting
 	 * the range of addresses into "struct lguest_data". */
 	if (get_user(lg->noirq_start, &lg->lguest_data->noirq_start)
-	    || get_user(lg->noirq_end, &lg->lguest_data->noirq_end)
-	    /* We tell the Guest that it can't use the top 4MB of virtual
-	     * addresses used by the Switcher. */
-	    || put_user(4U*1024*1024, &lg->lguest_data->reserve_mem)
-	    || put_user(tsc_speed, &lg->lguest_data->tsc_khz)
-	    /* We also give the Guest a unique id, as used in lguest_net.c. */
-	    || put_user(lg->guestid, &lg->lguest_data->guestid))
+	    || get_user(lg->noirq_end, &lg->lguest_data->noirq_end))
 		kill_guest(lg, "bad guest page %p", lg->lguest_data);
 
 	/* We write the current time into the Guest's data page once now. */
 	write_timestamp(lg);
 
+	/* page_tables.c will also do some setup. */
+	page_table_guest_data_init(lg);
+
 	/* This is the one case where the above accesses might have been the
 	 * first write to a Guest page.  This may have caused a copy-on-write
 	 * fault, but the Guest might be referring to the old (read-only)
 	 * page. */
 	guest_pagetable_clear_all(lg);
 }
-/* Now we've examined the hypercall code; our Guest can make requests.  There
- * is one other way we can do things for the Guest, as we see in
- * emulate_insn(). */
-
-/*H:110 Tricky point: we mark the hypercall as "done" once we've done it.
- * Normally we don't need to do this: the Guest will run again and update the
- * trap number before we come back around the run_guest() loop to
- * do_hypercalls().
- *
- * However, if we are signalled or the Guest sends DMA to the Launcher, that
- * loop will exit without running the Guest.  When it comes back it would try
- * to re-run the hypercall. */
-static void clear_hcall(struct lguest *lg)
-{
-	lg->regs->trapnum = 255;
-}
 
 /*H:100
  * Hypercalls
@@ -261,16 +195,12 @@ static void clear_hcall(struct lguest *lg)
  */
 void do_hypercalls(struct lguest *lg)
 {
-	/* Not initialized yet? */
+	/* Not initialized yet?  This hypercall must do it. */
 	if (unlikely(!lg->lguest_data)) {
-		/* Did the Guest make a hypercall?  We might have come back for
-		 * some other reason (an interrupt, a different trap). */
-		if (lg->regs->trapnum == LGUEST_TRAP_ENTRY) {
-			/* Set up the "struct lguest_data" */
-			initialize(lg);
-			/* The hypercall is done. */
-			clear_hcall(lg);
-		}
+		/* Set up the "struct lguest_data" */
+		initialize(lg);
+		/* Hcall is done. */
+		lg->hcall = NULL;
 		return;
 	}
 
@@ -280,12 +210,21 @@ void do_hypercalls(struct lguest *lg)
 	do_async_hcalls(lg);
 
 	/* If we stopped reading the hypercall ring because the Guest did a
-	 * SEND_DMA to the Launcher, we want to return now.  Otherwise if the
-	 * Guest asked us to do a hypercall, we do it. */
-	if (!lg->dma_is_pending && lg->regs->trapnum == LGUEST_TRAP_ENTRY) {
-		do_hcall(lg, lg->regs);
-		/* The hypercall is done. */
-		clear_hcall(lg);
+	 * NOTIFY to the Launcher, we want to return now.  Otherwise we do
+	 * the hypercall. */
+	if (!lg->pending_notify) {
+		do_hcall(lg, lg->hcall);
+		/* Tricky point: we reset the hcall pointer to mark the
+		 * hypercall as "done".  We use the hcall pointer rather than
+		 * the trap number to indicate a hypercall is pending.
+		 * Normally it doesn't matter: the Guest will run again and
+		 * update the trap number before we come back here.
+		 *
+		 * However, if we are signalled or the Guest sends DMA to the
+		 * Launcher, the run_guest() loop will exit without running the
+		 * Guest.  When it comes back it would try to re-run the
+		 * hypercall. */
+		lg->hcall = NULL;
 	}
 }
 
@@ -295,6 +234,6 @@ void write_timestamp(struct lguest *lg)
 {
 	struct timespec now;
 	ktime_get_real_ts(&now);
-	if (put_user(now, &lg->lguest_data->time))
+	if (copy_to_user(&lg->lguest_data->time, &now, sizeof(struct timespec)))
 		kill_guest(lg, "Writing timestamp");
 }

drivers/lguest/interrupts_and_traps.c

@@ -12,8 +12,14 @@
  * them first, so we also have a way of "reflecting" them into the Guest as if
  * they had been delivered to it directly. :*/
 #include <linux/uaccess.h>
+#include <linux/interrupt.h>
+#include <linux/module.h>
 #include "lg.h"
 
+/* Allow Guests to use a non-128 (ie. non-Linux) syscall trap. */
+static unsigned int syscall_vector = SYSCALL_VECTOR;
+module_param(syscall_vector, uint, 0444);
+
 /* The address of the interrupt handler is split into two bits: */
 static unsigned long idt_address(u32 lo, u32 hi)
 {
@@ -39,7 +45,7 @@ static void push_guest_stack(struct lguest *lg, unsigned long *gstack, u32 val)
 {
 	/* Stack grows upwards: move stack then write value. */
 	*gstack -= 4;
-	lgwrite_u32(lg, *gstack, val);
+	lgwrite(lg, *gstack, u32, val);
 }
 
 /*H:210 The set_guest_interrupt() routine actually delivers the interrupt or
@@ -56,8 +62,9 @@ static void push_guest_stack(struct lguest *lg, unsigned long *gstack, u32 val)
  * it). */
 static void set_guest_interrupt(struct lguest *lg, u32 lo, u32 hi, int has_err)
 {
-	unsigned long gstack;
+	unsigned long gstack, origstack;
 	u32 eflags, ss, irq_enable;
+	unsigned long virtstack;
 
 	/* There are two cases for interrupts: one where the Guest is already
 	 * in the kernel, and a more complex one where the Guest is in
@@ -65,8 +72,10 @@ static void set_guest_interrupt(struct lguest *lg, u32 lo, u32 hi, int has_err)
 	if ((lg->regs->ss&0x3) != GUEST_PL) {
 		/* The Guest told us their kernel stack with the SET_STACK
 		 * hypercall: both the virtual address and the segment */
-		gstack = guest_pa(lg, lg->esp1);
+		virtstack = lg->esp1;
 		ss = lg->ss1;
+
+		origstack = gstack = guest_pa(lg, virtstack);
 		/* We push the old stack segment and pointer onto the new
 		 * stack: when the Guest does an "iret" back from the interrupt
 		 * handler the CPU will notice they're dropping privilege
@@ -75,8 +84,10 @@ static void set_guest_interrupt(struct lguest *lg, u32 lo, u32 hi, int has_err)
 		push_guest_stack(lg, &gstack, lg->regs->esp);
 	} else {
 		/* We're staying on the same Guest (kernel) stack. */
-		gstack = guest_pa(lg, lg->regs->esp);
+		virtstack = lg->regs->esp;
 		ss = lg->regs->ss;
+
+		origstack = gstack = guest_pa(lg, virtstack);
 	}
 
 	/* Remember that we never let the Guest actually disable interrupts, so
@@ -102,7 +113,7 @@ static void set_guest_interrupt(struct lguest *lg, u32 lo, u32 hi, int has_err)
 	/* Now we've pushed all the old state, we change the stack, the code
 	 * segment and the address to execute. */
 	lg->regs->ss = ss;
-	lg->regs->esp = gstack + lg->page_offset;
+	lg->regs->esp = virtstack + (gstack - origstack);
 	lg->regs->cs = (__KERNEL_CS|GUEST_PL);
 	lg->regs->eip = idt_address(lo, hi);
 
@@ -165,7 +176,7 @@ void maybe_do_interrupt(struct lguest *lg)
 	/* Look at the IDT entry the Guest gave us for this interrupt.  The
 	 * first 32 (FIRST_EXTERNAL_VECTOR) entries are for traps, so we skip
 	 * over them. */
-	idt = &lg->idt[FIRST_EXTERNAL_VECTOR+irq];
+	idt = &lg->arch.idt[FIRST_EXTERNAL_VECTOR+irq];
 	/* If they don't have a handler (yet?), we just ignore it */
 	if (idt_present(idt->a, idt->b)) {
 		/* OK, mark it no longer pending and deliver it. */
@@ -183,6 +194,47 @@ void maybe_do_interrupt(struct lguest *lg)
 	 * timer interrupt. */
 	write_timestamp(lg);
 }
+/*:*/
+
+/* Linux uses trap 128 for system calls.  Plan9 uses 64, and Ron Minnich sent
+ * me a patch, so we support that too.  It'd be a big step for lguest if half
+ * the Plan 9 user base were to start using it.
+ *
+ * Actually now I think of it, it's possible that Ron *is* half the Plan 9
+ * userbase.  Oh well. */
+static bool could_be_syscall(unsigned int num)
+{
+	/* Normal Linux SYSCALL_VECTOR or reserved vector? */
+	return num == SYSCALL_VECTOR || num == syscall_vector;
+}
+
+/* The syscall vector it wants must be unused by Host. */
+bool check_syscall_vector(struct lguest *lg)
+{
+	u32 vector;
+
+	if (get_user(vector, &lg->lguest_data->syscall_vec))
+		return false;
+
+	return could_be_syscall(vector);
+}
+
+int init_interrupts(void)
+{
+	/* If they want some strange system call vector, reserve it now */
+	if (syscall_vector != SYSCALL_VECTOR
+	    && test_and_set_bit(syscall_vector, used_vectors)) {
+		printk("lg: couldn't reserve syscall %u\n", syscall_vector);
+		return -EBUSY;
+	}
+	return 0;
+}
+
+void free_interrupts(void)
+{
+	if (syscall_vector != SYSCALL_VECTOR)
+		clear_bit(syscall_vector, used_vectors);
+}
 
 /*H:220 Now we've got the routines to deliver interrupts, delivering traps
  * like page fault is easy.  The only trick is that Intel decided that some
@@ -197,14 +249,14 @@ int deliver_trap(struct lguest *lg, unsigned int num)
 {
 	/* Trap numbers are always 8 bit, but we set an impossible trap number
 	 * for traps inside the Switcher, so check that here. */
-	if (num >= ARRAY_SIZE(lg->idt))
+	if (num >= ARRAY_SIZE(lg->arch.idt))
 		return 0;
 
 	/* Early on the Guest hasn't set the IDT entries (or maybe it put a
 	 * bogus one in): if we fail here, the Guest will be killed. */
-	if (!idt_present(lg->idt[num].a, lg->idt[num].b))
+	if (!idt_present(lg->arch.idt[num].a, lg->arch.idt[num].b))
 		return 0;
-	set_guest_interrupt(lg, lg->idt[num].a, lg->idt[num].b, has_err(num));
+	set_guest_interrupt(lg, lg->arch.idt[num].a, lg->arch.idt[num].b, has_err(num));
 	return 1;
 }
 
@@ -218,28 +270,20 @@ int deliver_trap(struct lguest *lg, unsigned int num)
  * system calls down from 1750ns to 270ns.  Plus, if lguest didn't do it, all
  * the other hypervisors would tease it.
  *
- * This routine determines if a trap can be delivered directly. */
-static int direct_trap(const struct lguest *lg,
-		       const struct desc_struct *trap,
-		       unsigned int num)
+ * This routine indicates if a particular trap number could be delivered
+ * directly. */
+static int direct_trap(unsigned int num)
 {
 	/* Hardware interrupts don't go to the Guest at all (except system
 	 * call). */
-	if (num >= FIRST_EXTERNAL_VECTOR && num != SYSCALL_VECTOR)
+	if (num >= FIRST_EXTERNAL_VECTOR && !could_be_syscall(num))
 		return 0;
 
 	/* The Host needs to see page faults (for shadow paging and to save the
 	 * fault address), general protection faults (in/out emulation) and
 	 * device not available (TS handling), and of course, the hypercall
 	 * trap. */
-	if (num == 14 || num == 13 || num == 7 || num == LGUEST_TRAP_ENTRY)
-		return 0;
-
-	/* Only trap gates (type 15) can go direct to the Guest.  Interrupt
-	 * gates (type 14) disable interrupts as they are entered, which we
-	 * never let the Guest do.  Not present entries (type 0x0) also can't
-	 * go direct, of course 8) */
-	return idt_type(trap->a, trap->b) == 0xF;
+	return num != 14 && num != 13 && num != 7 && num != LGUEST_TRAP_ENTRY;
 }
 /*:*/
 
@@ -348,15 +392,11 @@ void load_guest_idt_entry(struct lguest *lg, unsigned int num, u32 lo, u32 hi)
 	 * to copy this again. */
 	lg->changed |= CHANGED_IDT;
 
-	/* The IDT which we keep in "struct lguest" only contains 32 entries
-	 * for the traps and LGUEST_IRQS (32) entries for interrupts.  We
-	 * ignore attempts to set handlers for higher interrupt numbers, except
-	 * for the system call "interrupt" at 128: we have a special IDT entry
-	 * for that. */
-	if (num < ARRAY_SIZE(lg->idt))
-		set_trap(lg, &lg->idt[num], num, lo, hi);
-	else if (num == SYSCALL_VECTOR)
-		set_trap(lg, &lg->syscall_idt, num, lo, hi);
+	/* Check that the Guest doesn't try to step outside the bounds. */
+	if (num >= ARRAY_SIZE(lg->arch.idt))
+		kill_guest(lg, "Setting idt entry %u", num);
+	else
+		set_trap(lg, &lg->arch.idt[num], num, lo, hi);
 }
 
 /* The default entry for each interrupt points into the Switcher routines which
@@ -399,20 +439,21 @@ void copy_traps(const struct lguest *lg, struct desc_struct *idt,
 
 	/* We can simply copy the direct traps, otherwise we use the default
 	 * ones in the Switcher: they will return to the Host. */
-	for (i = 0; i < FIRST_EXTERNAL_VECTOR; i++) {
-		if (direct_trap(lg, &lg->idt[i], i))
-			idt[i] = lg->idt[i];
+	for (i = 0; i < ARRAY_SIZE(lg->arch.idt); i++) {
+		/* If no Guest can ever override this trap, leave it alone. */
+		if (!direct_trap(i))
+			continue;
+
+		/* Only trap gates (type 15) can go direct to the Guest.
+		 * Interrupt gates (type 14) disable interrupts as they are
+		 * entered, which we never let the Guest do.  Not present
+		 * entries (type 0x0) also can't go direct, of course. */
+		if (idt_type(lg->arch.idt[i].a, lg->arch.idt[i].b) == 0xF)
+			idt[i] = lg->arch.idt[i];
 		else
+			/* Reset it to the default. */
 			default_idt_entry(&idt[i], i, def[i]);
 	}
-
-	/* Don't forget the system call trap!  The IDT entries for other
-	 * interupts never change, so no need to copy them. */
-	i = SYSCALL_VECTOR;
-	if (direct_trap(lg, &lg->syscall_idt, i))
-		idt[i] = lg->syscall_idt;
-	else
-		default_idt_entry(&idt[i], i, def[i]);
 }
 
 void guest_set_clockevent(struct lguest *lg, unsigned long delta)

drivers/lguest/io.c

@@ -1,626 +0,0 @@
-/*P:300 The I/O mechanism in lguest is simple yet flexible, allowing the Guest
- * to talk to the Launcher or directly to another Guest.  It uses familiar
- * concepts of DMA and interrupts, plus some neat code stolen from
- * futexes... :*/
-
-/* Copyright (C) 2006 Rusty Russell IBM Corporation
- *
- *  This program is free software; you can redistribute it and/or modify
- *  it under the terms of the GNU General Public License as published by
- *  the Free Software Foundation; either version 2 of the License, or
- *  (at your option) any later version.
- *
- *  This program is distributed in the hope that it will be useful,
- *  but WITHOUT ANY WARRANTY; without even the implied warranty of
- *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- *  GNU General Public License for more details.
- *
- *  You should have received a copy of the GNU General Public License
- *  along with this program; if not, write to the Free Software
- *  Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301 USA
- */
-#include <linux/types.h>
-#include <linux/futex.h>
-#include <linux/jhash.h>
-#include <linux/mm.h>
-#include <linux/highmem.h>
-#include <linux/uaccess.h>
-#include "lg.h"
-
-/*L:300
- * I/O
- *
- * Getting data in and out of the Guest is quite an art.  There are numerous
- * ways to do it, and they all suck differently.  We try to keep things fairly
- * close to "real" hardware so our Guest's drivers don't look like an alien
- * visitation in the middle of the Linux code, and yet make sure that Guests
- * can talk directly to other Guests, not just the Launcher.
- *
- * To do this, the Guest gives us a key when it binds or sends DMA buffers.
- * The key corresponds to a "physical" address inside the Guest (ie. a virtual
- * address inside the Launcher process).  We don't, however, use this key
- * directly.
- *
- * We want Guests which share memory to be able to DMA to each other: two
- * Launchers can mmap memory the same file, then the Guests can communicate.
- * Fortunately, the futex code provides us with a way to get a "union
- * futex_key" corresponding to the memory lying at a virtual address: if the
- * two processes share memory, the "union futex_key" for that memory will match
- * even if the memory is mapped at different addresses in each.  So we always
- * convert the keys to "union futex_key"s to compare them.
- *
- * Before we dive into this though, we need to look at another set of helper
- * routines used throughout the Host kernel code to access Guest memory.
- :*/
-static struct list_head dma_hash[61];
-
-/* An unfortunate side effect of the Linux double-linked list implementation is
- * that there's no good way to statically initialize an array of linked
- * lists. */
-void lguest_io_init(void)
-{
-	unsigned int i;
-
-	for (i = 0; i < ARRAY_SIZE(dma_hash); i++)
-		INIT_LIST_HEAD(&dma_hash[i]);
-}
-
-/* FIXME: allow multi-page lengths. */
-static int check_dma_list(struct lguest *lg, const struct lguest_dma *dma)
-{
-	unsigned int i;
-
-	for (i = 0; i < LGUEST_MAX_DMA_SECTIONS; i++) {
-		if (!dma->len[i])
-			return 1;
-		if (!lguest_address_ok(lg, dma->addr[i], dma->len[i]))
-			goto kill;
-		if (dma->len[i] > PAGE_SIZE)
-			goto kill;
-		/* We could do over a page, but is it worth it? */
-		if ((dma->addr[i] % PAGE_SIZE) + dma->len[i] > PAGE_SIZE)
-			goto kill;
-	}
-	return 1;
-
-kill:
-	kill_guest(lg, "bad DMA entry: %u@%#lx", dma->len[i], dma->addr[i]);
-	return 0;
-}
-
-/*L:330 This is our hash function, using the wonderful Jenkins hash.
- *
- * The futex key is a union with three parts: an unsigned long word, a pointer,
- * and an int "offset".  We could use jhash_2words() which takes three u32s.
- * (Ok, the hash functions are great: the naming sucks though).
- *
- * It's nice to be portable to 64-bit platforms, so we use the more generic
- * jhash2(), which takes an array of u32, the number of u32s, and an initial
- * u32 to roll in.  This is uglier, but breaks down to almost the same code on
- * 32-bit platforms like this one.
- *
- * We want a position in the array, so we modulo ARRAY_SIZE(dma_hash) (ie. 61).
- */
-static unsigned int hash(const union futex_key *key)
-{
-	return jhash2((u32*)&key->both.word,
-		      (sizeof(key->both.word)+sizeof(key->both.ptr))/4,
-		      key->both.offset)
-		% ARRAY_SIZE(dma_hash);
-}
-
-/* This is a convenience routine to compare two keys.  It's a much bemoaned C
- * weakness that it doesn't allow '==' on structures or unions, so we have to
- * open-code it like this. */
-static inline int key_eq(const union futex_key *a, const union futex_key *b)
-{
-	return (a->both.word == b->both.word
-		&& a->both.ptr == b->both.ptr
-		&& a->both.offset == b->both.offset);
-}
-
-/*L:360 OK, when we need to actually free up a Guest's DMA array we do several
- * things, so we have a convenient function to do it.
- *
- * The caller must hold a read lock on dmainfo owner's current->mm->mmap_sem
- * for the drop_futex_key_refs(). */
-static void unlink_dma(struct lguest_dma_info *dmainfo)
-{
-	/* You locked this too, right? */
-	BUG_ON(!mutex_is_locked(&lguest_lock));
-	/* This is how we know that the entry is free. */
-	dmainfo->interrupt = 0;
-	/* Remove it from the hash table. */
-	list_del(&dmainfo->list);
-	/* Drop the references we were holding (to the inode or mm). */
-	drop_futex_key_refs(&dmainfo->key);
-}
-
-/*L:350 This is the routine which we call when the Guest asks to unregister a
- * DMA array attached to a given key.  Returns true if the array was found. */
-static int unbind_dma(struct lguest *lg,
-		      const union futex_key *key,
-		      unsigned long dmas)
-{
-	int i, ret = 0;
-
-	/* We don't bother with the hash table, just look through all this
-	 * Guest's DMA arrays. */
-	for (i = 0; i < LGUEST_MAX_DMA; i++) {
-		/* In theory it could have more than one array on the same key,
-		 * or one array on multiple keys, so we check both */
-		if (key_eq(key, &lg->dma[i].key) && dmas == lg->dma[i].dmas) {
-			unlink_dma(&lg->dma[i]);
-			ret = 1;
-			break;
-		}
-	}
-	return ret;
-}
-
-/*L:340 BIND_DMA: this is the hypercall which sets up an array of "struct
- * lguest_dma" for receiving I/O.
- *
- * The Guest wants to bind an array of "struct lguest_dma"s to a particular key
- * to receive input.  This only happens when the Guest is setting up a new
- * device, so it doesn't have to be very fast.
- *
- * It returns 1 on a successful registration (it can fail if we hit the limit
- * of registrations for this Guest).
- */
-int bind_dma(struct lguest *lg,
-	     unsigned long ukey, unsigned long dmas, u16 numdmas, u8 interrupt)
-{
-	unsigned int i;
-	int ret = 0;
-	union futex_key key;
-	/* Futex code needs the mmap_sem. */
-	struct rw_semaphore *fshared = &current->mm->mmap_sem;
-
-	/* Invalid interrupt?  (We could kill the guest here). */
-	if (interrupt >= LGUEST_IRQS)
-		return 0;
-
-	/* We need to grab the Big Lguest Lock, because other Guests may be
-	 * trying to look through this Guest's DMAs to send something while
-	 * we're doing this. */
-	mutex_lock(&lguest_lock);
-	down_read(fshared);
-	if (get_futex_key((u32 __user *)ukey, fshared, &key) != 0) {
-		kill_guest(lg, "bad dma key %#lx", ukey);
-		goto unlock;
-	}
-
-	/* We want to keep this key valid once we drop mmap_sem, so we have to
-	 * hold a reference. */
-	get_futex_key_refs(&key);
-
-	/* If the Guest specified an interrupt of 0, that means they want to
-	 * unregister this array of "struct lguest_dma"s. */
-	if (interrupt == 0)
-		ret = unbind_dma(lg, &key, dmas);
-	else {
-		/* Look through this Guest's dma array for an unused entry. */
-		for (i = 0; i < LGUEST_MAX_DMA; i++) {
-			/* If the interrupt is non-zero, the entry is already
-			 * used. */
-			if (lg->dma[i].interrupt)
-				continue;
-
-			/* OK, a free one!  Fill on our details. */
-			lg->dma[i].dmas = dmas;
-			lg->dma[i].num_dmas = numdmas;
-			lg->dma[i].next_dma = 0;
-			lg->dma[i].key = key;
-			lg->dma[i].guestid = lg->guestid;
-			lg->dma[i].interrupt = interrupt;
-
-			/* Now we add it to the hash table: the position
-			 * depends on the futex key that we got. */
-			list_add(&lg->dma[i].list, &dma_hash[hash(&key)]);
-			/* Success! */
-			ret = 1;
-			goto unlock;
-		}
-	}
-	/* If we didn't find a slot to put the key in, drop the reference
-	 * again. */
-	drop_futex_key_refs(&key);
-unlock:
-	/* Unlock and out. */
- 	up_read(fshared);
-	mutex_unlock(&lguest_lock);
-	return ret;
-}
-
-/*L:385 Note that our routines to access a different Guest's memory are called
- * lgread_other() and lgwrite_other(): these names emphasize that they are only
- * used when the Guest is *not* the current Guest.
- *
- * The interface for copying from another process's memory is called
- * access_process_vm(), with a final argument of 0 for a read, and 1 for a
- * write.
- *
- * We need lgread_other() to read the destination Guest's "struct lguest_dma"
- * array. */
-static int lgread_other(struct lguest *lg,
-			void *buf, u32 addr, unsigned bytes)
-{
-	if (!lguest_address_ok(lg, addr, bytes)
-	    || access_process_vm(lg->tsk, addr, buf, bytes, 0) != bytes) {
-		memset(buf, 0, bytes);
-		kill_guest(lg, "bad address in registered DMA struct");
-		return 0;
-	}
-	return 1;
-}
-
-/* "lgwrite()" to another Guest: used to update the destination "used_len" once
- * we've transferred data into the buffer. */
-static int lgwrite_other(struct lguest *lg, u32 addr,
-			 const void *buf, unsigned bytes)
-{
-	if (!lguest_address_ok(lg, addr, bytes)
-	    || (access_process_vm(lg->tsk, addr, (void *)buf, bytes, 1)
-		!= bytes)) {
-		kill_guest(lg, "bad address writing to registered DMA");
-		return 0;
-	}
-	return 1;
-}
-
-/*L:400 This is the generic engine which copies from a source "struct
- * lguest_dma" from this Guest into another Guest's "struct lguest_dma".  The
- * destination Guest's pages have already been mapped, as contained in the
- * pages array.
- *
- * If you're wondering if there's a nice "copy from one process to another"
- * routine, so was I.  But Linux isn't really set up to copy between two
- * unrelated processes, so we have to write it ourselves.
- */
-static u32 copy_data(struct lguest *srclg,
-		     const struct lguest_dma *src,
-		     const struct lguest_dma *dst,
-		     struct page *pages[])
-{
-	unsigned int totlen, si, di, srcoff, dstoff;
-	void *maddr = NULL;
-
-	/* We return the total length transferred. */
-	totlen = 0;
-
-	/* We keep indexes into the source and destination "struct lguest_dma",
-	 * and an offset within each region. */
-	si = di = 0;
-	srcoff = dstoff = 0;
-
-	/* We loop until the source or destination is exhausted. */
-	while (si < LGUEST_MAX_DMA_SECTIONS && src->len[si]
-	       && di < LGUEST_MAX_DMA_SECTIONS && dst->len[di]) {
-		/* We can only transfer the rest of the src buffer, or as much
-		 * as will fit into the destination buffer. */
-		u32 len = min(src->len[si] - srcoff, dst->len[di] - dstoff);
-
-		/* For systems using "highmem" we need to use kmap() to access
-		 * the page we want.  We often use the same page over and over,
-		 * so rather than kmap() it on every loop, we set the maddr
-		 * pointer to NULL when we need to move to the next
-		 * destination page. */
-		if (!maddr)
-			maddr = kmap(pages[di]);
-
-		/* Copy directly from (this Guest's) source address to the
-		 * destination Guest's kmap()ed buffer.  Note that maddr points
-		 * to the start of the page: we need to add the offset of the
-		 * destination address and offset within the buffer. */
-
-		/* FIXME: This is not completely portable.  I looked at
-		 * copy_to_user_page(), and some arch's seem to need special
-		 * flushes.  x86 is fine. */
-		if (copy_from_user(maddr + (dst->addr[di] + dstoff)%PAGE_SIZE,
-				   (void __user *)src->addr[si], len) != 0) {
-			/* If a copy failed, it's the source's fault. */
-			kill_guest(srclg, "bad address in sending DMA");
-			totlen = 0;
-			break;
-		}
-
-		/* Increment the total and src & dst offsets */
-		totlen += len;
-		srcoff += len;
-		dstoff += len;
-
-		/* Presumably we reached the end of the src or dest buffers: */
-		if (srcoff == src->len[si]) {
-			/* Move to the next buffer at offset 0 */
-			si++;
-			srcoff = 0;
-		}
-		if (dstoff == dst->len[di]) {
-			/* We need to unmap that destination page and reset
-			 * maddr ready for the next one. */
-			kunmap(pages[di]);
-			maddr = NULL;
-			di++;
-			dstoff = 0;
-		}
-	}
-
-	/* If we still had a page mapped at the end, unmap now. */
-	if (maddr)
-		kunmap(pages[di]);
-
-	return totlen;
-}
-
-/*L:390 This is how we transfer a "struct lguest_dma" from the source Guest
- * (the current Guest which called SEND_DMA) to another Guest. */
-static u32 do_dma(struct lguest *srclg, const struct lguest_dma *src,
-		  struct lguest *dstlg, const struct lguest_dma *dst)
-{
-	int i;
-	u32 ret;
-	struct page *pages[LGUEST_MAX_DMA_SECTIONS];
-
-	/* We check that both source and destination "struct lguest_dma"s are
-	 * within the bounds of the source and destination Guests */
-	if (!check_dma_list(dstlg, dst) || !check_dma_list(srclg, src))
-		return 0;
-
-	/* We need to map the pages which correspond to each parts of
-	 * destination buffer. */
-	for (i = 0; i < LGUEST_MAX_DMA_SECTIONS; i++) {
-		if (dst->len[i] == 0)
-			break;
-		/* get_user_pages() is a complicated function, especially since
-		 * we only want a single page.  But it works, and returns the
-		 * number of pages.  Note that we're holding the destination's
-		 * mmap_sem, as get_user_pages() requires. */
-		if (get_user_pages(dstlg->tsk, dstlg->mm,
-				   dst->addr[i], 1, 1, 1, pages+i, NULL)
-		    != 1) {
-			/* This means the destination gave us a bogus buffer */
-			kill_guest(dstlg, "Error mapping DMA pages");
-			ret = 0;
-			goto drop_pages;
-		}
-	}
-
-	/* Now copy the data until we run out of src or dst. */
-	ret = copy_data(srclg, src, dst, pages);
-
-drop_pages:
-	while (--i >= 0)
-		put_page(pages[i]);
-	return ret;
-}
-
-/*L:380 Transferring data from one Guest to another is not as simple as I'd
- * like.  We've found the "struct lguest_dma_info" bound to the same address as
- * the send, we need to copy into it.
- *
- * This function returns true if the destination array was empty. */
-static int dma_transfer(struct lguest *srclg,
-			unsigned long udma,
-			struct lguest_dma_info *dst)
-{
-	struct lguest_dma dst_dma, src_dma;
-	struct lguest *dstlg;
-	u32 i, dma = 0;
-
-	/* From the "struct lguest_dma_info" we found in the hash, grab the
-	 * Guest. */
-	dstlg = &lguests[dst->guestid];
-	/* Read in the source "struct lguest_dma" handed to SEND_DMA. */
-	lgread(srclg, &src_dma, udma, sizeof(src_dma));
-
-	/* We need the destination's mmap_sem, and we already hold the source's
-	 * mmap_sem for the futex key lookup.  Normally this would suggest that
-	 * we could deadlock if the destination Guest was trying to send to
-	 * this source Guest at the same time, which is another reason that all
-	 * I/O is done under the big lguest_lock. */
-	down_read(&dstlg->mm->mmap_sem);
-
-	/* Look through the destination DMA array for an available buffer. */
-	for (i = 0; i < dst->num_dmas; i++) {
-		/* We keep a "next_dma" pointer which often helps us avoid
-		 * looking at lots of previously-filled entries. */
-		dma = (dst->next_dma + i) % dst->num_dmas;
-		if (!lgread_other(dstlg, &dst_dma,
-				  dst->dmas + dma * sizeof(struct lguest_dma),
-				  sizeof(dst_dma))) {
-			goto fail;
-		}
-		if (!dst_dma.used_len)
-			break;
-	}
-
-	/* If we found a buffer, we do the actual data copy. */
-	if (i != dst->num_dmas) {
-		unsigned long used_lenp;
-		unsigned int ret;
-
-		ret = do_dma(srclg, &src_dma, dstlg, &dst_dma);
-		/* Put used length in the source "struct lguest_dma"'s used_len
-		 * field.  It's a little tricky to figure out where that is,
-		 * though. */
-		lgwrite_u32(srclg,
-			    udma+offsetof(struct lguest_dma, used_len), ret);
-		/* Tranferring 0 bytes is OK if the source buffer was empty. */
-		if (ret == 0 && src_dma.len[0] != 0)
-			goto fail;
-
-		/* The destination Guest might be running on a different CPU:
-		 * we have to make sure that it will see the "used_len" field
-		 * change to non-zero *after* it sees the data we copied into
-		 * the buffer.  Hence a write memory barrier. */
-		wmb();
-		/* Figuring out where the destination's used_len field for this
-		 * "struct lguest_dma" in the array is also a little ugly. */
-		used_lenp = dst->dmas
-			+ dma * sizeof(struct lguest_dma)
-			+ offsetof(struct lguest_dma, used_len);
-		lgwrite_other(dstlg, used_lenp, &ret, sizeof(ret));
-		/* Move the cursor for next time. */
-		dst->next_dma++;
-	}
- 	up_read(&dstlg->mm->mmap_sem);
-
-	/* We trigger the destination interrupt, even if the destination was
-	 * empty and we didn't transfer anything: this gives them a chance to
-	 * wake up and refill. */
-	set_bit(dst->interrupt, dstlg->irqs_pending);
-	/* Wake up the destination process. */
-	wake_up_process(dstlg->tsk);
-	/* If we passed the last "struct lguest_dma", the receive had no
-	 * buffers left. */
-	return i == dst->num_dmas;
-
-fail:
-	up_read(&dstlg->mm->mmap_sem);
-	return 0;
-}
-
-/*L:370 This is the counter-side to the BIND_DMA hypercall; the SEND_DMA
- * hypercall.  We find out who's listening, and send to them. */
-void send_dma(struct lguest *lg, unsigned long ukey, unsigned long udma)
-{
-	union futex_key key;
-	int empty = 0;
-	struct rw_semaphore *fshared = &current->mm->mmap_sem;
-
-again:
-	mutex_lock(&lguest_lock);
-	down_read(fshared);
-	/* Get the futex key for the key the Guest gave us */
-	if (get_futex_key((u32 __user *)ukey, fshared, &key) != 0) {
-		kill_guest(lg, "bad sending DMA key");
-		goto unlock;
-	}
-	/* Since the key must be a multiple of 4, the futex key uses the lower
-	 * bit of the "offset" field (which would always be 0) to indicate a
-	 * mapping which is shared with other processes (ie. Guests). */
-	if (key.shared.offset & 1) {
-		struct lguest_dma_info *i;
-		/* Look through the hash for other Guests. */
-		list_for_each_entry(i, &dma_hash[hash(&key)], list) {
-			/* Don't send to ourselves. */
-			if (i->guestid == lg->guestid)
-				continue;
-			if (!key_eq(&key, &i->key))
-				continue;
-
-			/* If dma_transfer() tells us the destination has no
-			 * available buffers, we increment "empty". */
-			empty += dma_transfer(lg, udma, i);
-			break;
-		}
-		/* If the destination is empty, we release our locks and
-		 * give the destination Guest a brief chance to restock. */
-		if (empty == 1) {
-			/* Give any recipients one chance to restock. */
-			up_read(&current->mm->mmap_sem);
-			mutex_unlock(&lguest_lock);
-			/* Next time, we won't try again. */
-			empty++;
-			goto again;
-		}
-	} else {
-		/* Private mapping: Guest is sending to its Launcher.  We set
-		 * the "dma_is_pending" flag so that the main loop will exit
-		 * and the Launcher's read() from /dev/lguest will return. */
-		lg->dma_is_pending = 1;
-		lg->pending_dma = udma;
-		lg->pending_key = ukey;
-	}
-unlock:
-	up_read(fshared);
-	mutex_unlock(&lguest_lock);
-}
-/*:*/
-
-void release_all_dma(struct lguest *lg)
-{
-	unsigned int i;
-
-	BUG_ON(!mutex_is_locked(&lguest_lock));
-
-	down_read(&lg->mm->mmap_sem);
-	for (i = 0; i < LGUEST_MAX_DMA; i++) {
-		if (lg->dma[i].interrupt)
-			unlink_dma(&lg->dma[i]);
-	}
-	up_read(&lg->mm->mmap_sem);
-}
-
-/*M:007 We only return a single DMA buffer to the Launcher, but it would be
- * more efficient to return a pointer to the entire array of DMA buffers, which
- * it can cache and choose one whenever it wants.
- *
- * Currently the Launcher uses a write to /dev/lguest, and the return value is
- * the address of the DMA structure with the interrupt number placed in
- * dma->used_len.  If we wanted to return the entire array, we need to return
- * the address, array size and interrupt number: this seems to require an
- * ioctl(). :*/
-
-/*L:320 This routine looks for a DMA buffer registered by the Guest on the
- * given key (using the BIND_DMA hypercall). */
-unsigned long get_dma_buffer(struct lguest *lg,
-			     unsigned long ukey, unsigned long *interrupt)
-{
-	unsigned long ret = 0;
-	union futex_key key;
-	struct lguest_dma_info *i;
-	struct rw_semaphore *fshared = &current->mm->mmap_sem;
-
-	/* Take the Big Lguest Lock to stop other Guests sending this Guest DMA
-	 * at the same time. */
-	mutex_lock(&lguest_lock);
-	/* To match between Guests sharing the same underlying memory we steal
-	 * code from the futex infrastructure.  This requires that we hold the
-	 * "mmap_sem" for our process (the Launcher), and pass it to the futex
-	 * code. */
-	down_read(fshared);
-
-	/* This can fail if it's not a valid address, or if the address is not
-	 * divisible by 4 (the futex code needs that, we don't really). */
-	if (get_futex_key((u32 __user *)ukey, fshared, &key) != 0) {
-		kill_guest(lg, "bad registered DMA buffer");
-		goto unlock;
-	}
-	/* Search the hash table for matching entries (the Launcher can only
-	 * send to its own Guest for the moment, so the entry must be for this
-	 * Guest) */
-	list_for_each_entry(i, &dma_hash[hash(&key)], list) {
-		if (key_eq(&key, &i->key) && i->guestid == lg->guestid) {
-			unsigned int j;
-			/* Look through the registered DMA array for an
-			 * available buffer. */
-			for (j = 0; j < i->num_dmas; j++) {
-				struct lguest_dma dma;
-
-				ret = i->dmas + j * sizeof(struct lguest_dma);
-				lgread(lg, &dma, ret, sizeof(dma));
-				if (dma.used_len == 0)
-					break;
-			}
-			/* Store the interrupt the Guest wants when the buffer
-			 * is used. */
-			*interrupt = i->interrupt;
-			break;
-		}
-	}
-unlock:
-	up_read(fshared);
-	mutex_unlock(&lguest_lock);
-	return ret;
-}
-/*:*/
-
-/*L:410 This really has completed the Launcher.  Not only have we now finished
- * the longest chapter in our journey, but this also means we are over halfway
- * through!
- *
- * Enough prevaricating around the bush: it is time for us to dive into the
- * core of the Host, in "make Host".
- */

drivers/lguest/lg.h

@@ -1,119 +1,25 @@
 #ifndef _LGUEST_H
 #define _LGUEST_H
 
-#include <asm/desc.h>
-
-#define GDT_ENTRY_LGUEST_CS	10
-#define GDT_ENTRY_LGUEST_DS	11
-#define LGUEST_CS		(GDT_ENTRY_LGUEST_CS * 8)
-#define LGUEST_DS		(GDT_ENTRY_LGUEST_DS * 8)
-
 #ifndef __ASSEMBLY__
 #include <linux/types.h>
 #include <linux/init.h>
 #include <linux/stringify.h>
-#include <linux/binfmts.h>
-#include <linux/futex.h>
 #include <linux/lguest.h>
 #include <linux/lguest_launcher.h>
 #include <linux/wait.h>
 #include <linux/err.h>
 #include <asm/semaphore.h>
-#include "irq_vectors.h"
-
-#define GUEST_PL 1
 
-struct lguest_regs
-{
-	/* Manually saved part. */
-	unsigned long ebx, ecx, edx;
-	unsigned long esi, edi, ebp;
-	unsigned long gs;
-	unsigned long eax;
-	unsigned long fs, ds, es;
-	unsigned long trapnum, errcode;
-	/* Trap pushed part */
-	unsigned long eip;
-	unsigned long cs;
-	unsigned long eflags;
-	unsigned long esp;
-	unsigned long ss;
-};
+#include <asm/lguest.h>
 
 void free_pagetables(void);
 int init_pagetables(struct page **switcher_page, unsigned int pages);
 
-/* Full 4G segment descriptors, suitable for CS and DS. */
-#define FULL_EXEC_SEGMENT ((struct desc_struct){0x0000ffff, 0x00cf9b00})
-#define FULL_SEGMENT ((struct desc_struct){0x0000ffff, 0x00cf9300})
-
-struct lguest_dma_info
-{
-	struct list_head list;
-	union futex_key key;
-	unsigned long dmas;
-	u16 next_dma;
-	u16 num_dmas;
-	u16 guestid;
-	u8 interrupt; 	/* 0 when not registered */
-};
-
-/*H:310 The page-table code owes a great debt of gratitude to Andi Kleen.  He
- * reviewed the original code which used "u32" for all page table entries, and
- * insisted that it would be far clearer with explicit typing.  I thought it
- * was overkill, but he was right: it is much clearer than it was before.
- *
- * We have separate types for the Guest's ptes & pgds and the shadow ptes &
- * pgds.  There's already a Linux type for these (pte_t and pgd_t) but they
- * change depending on kernel config options (PAE). */
-
-/* Each entry is identical: lower 12 bits of flags and upper 20 bits for the
- * "page frame number" (0 == first physical page, etc).  They are different
- * types so the compiler will warn us if we mix them improperly. */
-typedef union {
-	struct { unsigned flags:12, pfn:20; };
-	struct { unsigned long val; } raw;
-} spgd_t;
-typedef union {
-	struct { unsigned flags:12, pfn:20; };
-	struct { unsigned long val; } raw;
-} spte_t;
-typedef union {
-	struct { unsigned flags:12, pfn:20; };
-	struct { unsigned long val; } raw;
-} gpgd_t;
-typedef union {
-	struct { unsigned flags:12, pfn:20; };
-	struct { unsigned long val; } raw;
-} gpte_t;
-
-/* We have two convenient macros to convert a "raw" value as handed to us by
- * the Guest into the correct Guest PGD or PTE type. */
-#define mkgpte(_val) ((gpte_t){.raw.val = _val})
-#define mkgpgd(_val) ((gpgd_t){.raw.val = _val})
-/*:*/
-
 struct pgdir
 {
-	unsigned long cr3;
-	spgd_t *pgdir;
-};
-
-/* This is a guest-specific page (mapped ro) into the guest. */
-struct lguest_ro_state
-{
-	/* Host information we need to restore when we switch back. */
-	u32 host_cr3;
-	struct Xgt_desc_struct host_idt_desc;
-	struct Xgt_desc_struct host_gdt_desc;
-	u32 host_sp;
-
-	/* Fields which are used when guest is running. */
-	struct Xgt_desc_struct guest_idt_desc;
-	struct Xgt_desc_struct guest_gdt_desc;
-	struct i386_hw_tss guest_tss;
-	struct desc_struct guest_idt[IDT_ENTRIES];
-	struct desc_struct guest_gdt[GDT_ENTRIES];
+	unsigned long gpgdir;
+	pgd_t *pgdir;
 };
 
 /* We have two pages shared with guests, per cpu.  */
@@ -141,9 +47,11 @@ struct lguest
 	struct lguest_data __user *lguest_data;
 	struct task_struct *tsk;
 	struct mm_struct *mm; 	/* == tsk->mm, but that becomes NULL on exit */
-	u16 guestid;
 	u32 pfn_limit;
-	u32 page_offset;
+	/* This provides the offset to the base of guest-physical
+	 * memory in the Launcher. */
+	void __user *mem_base;
+	unsigned long kernel_address;
 	u32 cr2;
 	int halted;
 	int ts;
@@ -151,6 +59,9 @@ struct lguest
 	u32 esp1;
 	u8 ss1;
 
+	/* If a hypercall was asked for, this points to the arguments. */
+	struct hcall_args *hcall;
+
 	/* Do we need to stop what we're doing and return to userspace? */
 	int break_out;
 	wait_queue_head_t break_wq;
@@ -167,24 +78,15 @@ struct lguest
 	struct task_struct *wake;
 
 	unsigned long noirq_start, noirq_end;
-	int dma_is_pending;
-	unsigned long pending_dma; /* struct lguest_dma */
-	unsigned long pending_key; /* address they're sending to */
+	unsigned long pending_notify; /* pfn from LHCALL_NOTIFY */
 
 	unsigned int stack_pages;
 	u32 tsc_khz;
 
-	struct lguest_dma_info dma[LGUEST_MAX_DMA];
-
 	/* Dead? */
 	const char *dead;
 
-	/* The GDT entries copied into lguest_ro_state when running. */
-	struct desc_struct gdt[GDT_ENTRIES];
-
-	/* The IDT entries: some copied into lguest_ro_state when running. */
-	struct desc_struct idt[FIRST_EXTERNAL_VECTOR+LGUEST_IRQS];
-	struct desc_struct syscall_idt;
+	struct lguest_arch arch;
 
 	/* Virtual clock device */
 	struct hrtimer hrt;
@@ -193,19 +95,38 @@ struct lguest
 	DECLARE_BITMAP(irqs_pending, LGUEST_IRQS);
 };
 
-extern struct lguest lguests[];
 extern struct mutex lguest_lock;
 
 /* core.c: */
-u32 lgread_u32(struct lguest *lg, unsigned long addr);
-void lgwrite_u32(struct lguest *lg, unsigned long addr, u32 val);
-void lgread(struct lguest *lg, void *buf, unsigned long addr, unsigned len);
-void lgwrite(struct lguest *lg, unsigned long, const void *buf, unsigned len);
-int find_free_guest(void);
 int lguest_address_ok(const struct lguest *lg,
 		      unsigned long addr, unsigned long len);
+void __lgread(struct lguest *, void *, unsigned long, unsigned);
+void __lgwrite(struct lguest *, unsigned long, const void *, unsigned);
+
+/*L:306 Using memory-copy operations like that is usually inconvient, so we
+ * have the following helper macros which read and write a specific type (often
+ * an unsigned long).
+ *
+ * This reads into a variable of the given type then returns that. */
+#define lgread(lg, addr, type)						\
+	({ type _v; __lgread((lg), &_v, (addr), sizeof(_v)); _v; })
+
+/* This checks that the variable is of the given type, then writes it out. */
+#define lgwrite(lg, addr, type, val)				\
+	do {							\
+		typecheck(type, val);				\
+		__lgwrite((lg), (addr), &(val), sizeof(val));	\
+	} while(0)
+/* (end of memory access helper routines) :*/
+
 int run_guest(struct lguest *lg, unsigned long __user *user);
 
+/* Helper macros to obtain the first 12 or the last 20 bits, this is only the
+ * first step in the migration to the kernel types.  pte_pfn is already defined
+ * in the kernel. */
+#define pgd_flags(x)	(pgd_val(x) & ~PAGE_MASK)
+#define pte_flags(x)	(pte_val(x) & ~PAGE_MASK)
+#define pgd_pfn(x)	(pgd_val(x) >> PAGE_SHIFT)
 
 /* interrupts_and_traps.c: */
 void maybe_do_interrupt(struct lguest *lg);
@@ -219,6 +140,9 @@ void copy_traps(const struct lguest *lg, struct desc_struct *idt,
 		const unsigned long *def);
 void guest_set_clockevent(struct lguest *lg, unsigned long delta);
 void init_clockdev(struct lguest *lg);
+bool check_syscall_vector(struct lguest *lg);
+int init_interrupts(void);
+void free_interrupts(void);
 
 /* segments.c: */
 void setup_default_gdt_entries(struct lguest_ro_state *state);
@@ -232,28 +156,33 @@ void copy_gdt_tls(const struct lguest *lg, struct desc_struct *gdt);
 int init_guest_pagetable(struct lguest *lg, unsigned long pgtable);
 void free_guest_pagetable(struct lguest *lg);
 void guest_new_pagetable(struct lguest *lg, unsigned long pgtable);
-void guest_set_pmd(struct lguest *lg, unsigned long cr3, u32 i);
+void guest_set_pmd(struct lguest *lg, unsigned long gpgdir, u32 i);
 void guest_pagetable_clear_all(struct lguest *lg);
 void guest_pagetable_flush_user(struct lguest *lg);
-void guest_set_pte(struct lguest *lg, unsigned long cr3,
-		   unsigned long vaddr, gpte_t val);
+void guest_set_pte(struct lguest *lg, unsigned long gpgdir,
+		   unsigned long vaddr, pte_t val);
 void map_switcher_in_guest(struct lguest *lg, struct lguest_pages *pages);
 int demand_page(struct lguest *info, unsigned long cr2, int errcode);
 void pin_page(struct lguest *lg, unsigned long vaddr);
+unsigned long guest_pa(struct lguest *lg, unsigned long vaddr);
+void page_table_guest_data_init(struct lguest *lg);
+
+/* <arch>/core.c: */
+void lguest_arch_host_init(void);
+void lguest_arch_host_fini(void);
+void lguest_arch_run_guest(struct lguest *lg);
+void lguest_arch_handle_trap(struct lguest *lg);
+int lguest_arch_init_hypercalls(struct lguest *lg);
+int lguest_arch_do_hcall(struct lguest *lg, struct hcall_args *args);
+void lguest_arch_setup_regs(struct lguest *lg, unsigned long start);
+
+/* <arch>/switcher.S: */
+extern char start_switcher_text[], end_switcher_text[], switch_to_guest[];
 
 /* lguest_user.c: */
 int lguest_device_init(void);
 void lguest_device_remove(void);
 
-/* io.c: */
-void lguest_io_init(void);
-int bind_dma(struct lguest *lg,
-	     unsigned long key, unsigned long udma, u16 numdmas, u8 interrupt);
-void send_dma(struct lguest *info, unsigned long key, unsigned long udma);
-void release_all_dma(struct lguest *lg);
-unsigned long get_dma_buffer(struct lguest *lg, unsigned long key,
-			     unsigned long *interrupt);
-
 /* hypercalls.c: */
 void do_hypercalls(struct lguest *lg);
 void write_timestamp(struct lguest *lg);
@@ -292,9 +221,5 @@ do {								\
 } while(0)
 /* (End of aside) :*/
 
-static inline unsigned long guest_pa(struct lguest *lg, unsigned long vaddr)
-{
-	return vaddr - lg->page_offset;
-}
 #endif	/* __ASSEMBLY__ */
 #endif	/* _LGUEST_H */

drivers/lguest/lguest_bus.c

@@ -1,218 +0,0 @@
-/*P:050 Lguest guests use a very simple bus for devices.  It's a simple array
- * of device descriptors contained just above the top of normal memory.  The
- * lguest bus is 80% tedious boilerplate code. :*/
-#include <linux/init.h>
-#include <linux/bootmem.h>
-#include <linux/lguest_bus.h>
-#include <asm/io.h>
-#include <asm/paravirt.h>
-
-static ssize_t type_show(struct device *_dev,
-                         struct device_attribute *attr, char *buf)
-{
-	struct lguest_device *dev = container_of(_dev,struct lguest_device,dev);
-	return sprintf(buf, "%hu", lguest_devices[dev->index].type);
-}
-static ssize_t features_show(struct device *_dev,
-                             struct device_attribute *attr, char *buf)
-{
-	struct lguest_device *dev = container_of(_dev,struct lguest_device,dev);
-	return sprintf(buf, "%hx", lguest_devices[dev->index].features);
-}
-static ssize_t pfn_show(struct device *_dev,
-			 struct device_attribute *attr, char *buf)
-{
-	struct lguest_device *dev = container_of(_dev,struct lguest_device,dev);
-	return sprintf(buf, "%u", lguest_devices[dev->index].pfn);
-}
-static ssize_t status_show(struct device *_dev,
-                           struct device_attribute *attr, char *buf)
-{
-	struct lguest_device *dev = container_of(_dev,struct lguest_device,dev);
-	return sprintf(buf, "%hx", lguest_devices[dev->index].status);
-}
-static ssize_t status_store(struct device *_dev, struct device_attribute *attr,
-                            const char *buf, size_t count)
-{
-	struct lguest_device *dev = container_of(_dev,struct lguest_device,dev);
-	if (sscanf(buf, "%hi", &lguest_devices[dev->index].status) != 1)
-		return -EINVAL;
-	return count;
-}
-static struct device_attribute lguest_dev_attrs[] = {
-	__ATTR_RO(type),
-	__ATTR_RO(features),
-	__ATTR_RO(pfn),
-	__ATTR(status, 0644, status_show, status_store),
-	__ATTR_NULL
-};
-
-/*D:130 The generic bus infrastructure requires a function which says whether a
- * device matches a driver.  For us, it is simple: "struct lguest_driver"
- * contains a "device_type" field which indicates what type of device it can
- * handle, so we just cast the args and compare: */
-static int lguest_dev_match(struct device *_dev, struct device_driver *_drv)
-{
-	struct lguest_device *dev = container_of(_dev,struct lguest_device,dev);
-	struct lguest_driver *drv = container_of(_drv,struct lguest_driver,drv);
-
-	return (drv->device_type == lguest_devices[dev->index].type);
-}
-/*:*/
-
-struct lguest_bus {
-	struct bus_type bus;
-	struct device dev;
-};
-
-static struct lguest_bus lguest_bus = {
-	.bus = {
-		.name  = "lguest",
-		.match = lguest_dev_match,
-		.dev_attrs = lguest_dev_attrs,
-	},
-	.dev = {
-		.parent = NULL,
-		.bus_id = "lguest",
-	}
-};
-
-/*D:140 This is the callback which occurs once the bus infrastructure matches
- * up a device and driver, ie. in response to add_lguest_device() calling
- * device_register(), or register_lguest_driver() calling driver_register().
- *
- * At the moment it's always the latter: the devices are added first, since
- * scan_devices() is called from a "core_initcall", and the drivers themselves
- * called later as a normal "initcall".  But it would work the other way too.
- *
- * So now we have the happy couple, we add the status bit to indicate that we
- * found a driver.  If the driver truly loves the device, it will return
- * happiness from its probe function (ok, perhaps this wasn't my greatest
- * analogy), and we set the final "driver ok" bit so the Host sees it's all
- * green. */
-static int lguest_dev_probe(struct device *_dev)
-{
-	int ret;
-	struct lguest_device*dev = container_of(_dev,struct lguest_device,dev);
-	struct lguest_driver*drv = container_of(dev->dev.driver,
-						struct lguest_driver, drv);
-
-	lguest_devices[dev->index].status |= LGUEST_DEVICE_S_DRIVER;
-	ret = drv->probe(dev);
-	if (ret == 0)
-		lguest_devices[dev->index].status |= LGUEST_DEVICE_S_DRIVER_OK;
-	return ret;
-}
-
-/* The last part of the bus infrastructure is the function lguest drivers use
- * to register themselves.  Firstly, we do nothing if there's no lguest bus
- * (ie. this is not a Guest), otherwise we fill in the embedded generic "struct
- * driver" fields and call the generic driver_register(). */
-int register_lguest_driver(struct lguest_driver *drv)
-{
-	if (!lguest_devices)
-		return 0;
-
-	drv->drv.bus = &lguest_bus.bus;
-	drv->drv.name = drv->name;
-	drv->drv.owner = drv->owner;
-	drv->drv.probe = lguest_dev_probe;
-
-	return driver_register(&drv->drv);
-}
-
-/* At the moment we build all the drivers into the kernel because they're so
- * simple: 8144 bytes for all three of them as I type this.  And as the console
- * really needs to be built in, it's actually only 3527 bytes for the network
- * and block drivers.
- *
- * If they get complex it will make sense for them to be modularized, so we
- * need to explicitly export the symbol.
- *
- * I don't think non-GPL modules make sense, so it's a GPL-only export.
- */
-EXPORT_SYMBOL_GPL(register_lguest_driver);
-
-/*D:120 This is the core of the lguest bus: actually adding a new device.
- * It's a separate function because it's neater that way, and because an
- * earlier version of the code supported hotplug and unplug.  They were removed
- * early on because they were never used.
- *
- * As Andrew Tridgell says, "Untested code is buggy code".
- *
- * It's worth reading this carefully: we start with an index into the array of
- * "struct lguest_device_desc"s indicating the device which is new: */
-static void add_lguest_device(unsigned int index)
-{
-	struct lguest_device *new;
-
-	/* Each "struct lguest_device_desc" has a "status" field, which the
-	 * Guest updates as the device is probed.  In the worst case, the Host
-	 * can look at these bits to tell what part of device setup failed,
-	 * even if the console isn't available. */
-	lguest_devices[index].status |= LGUEST_DEVICE_S_ACKNOWLEDGE;
-	new = kmalloc(sizeof(struct lguest_device), GFP_KERNEL);
-	if (!new) {
-		printk(KERN_EMERG "Cannot allocate lguest device %u\n", index);
-		lguest_devices[index].status |= LGUEST_DEVICE_S_FAILED;
-		return;
-	}
-
-	/* The "struct lguest_device" setup is pretty straight-forward example
-	 * code. */
-	new->index = index;
-	new->private = NULL;
-	memset(&new->dev, 0, sizeof(new->dev));
-	new->dev.parent = &lguest_bus.dev;
-	new->dev.bus = &lguest_bus.bus;
-	sprintf(new->dev.bus_id, "%u", index);
-
-	/* device_register() causes the bus infrastructure to look for a
-	 * matching driver. */
-	if (device_register(&new->dev) != 0) {
-		printk(KERN_EMERG "Cannot register lguest device %u\n", index);
-		lguest_devices[index].status |= LGUEST_DEVICE_S_FAILED;
-		kfree(new);
-	}
-}
-
-/*D:110 scan_devices() simply iterates through the device array.  The type 0
- * is reserved to mean "no device", and anything else means we have found a
- * device: add it. */
-static void scan_devices(void)
-{
-	unsigned int i;
-
-	for (i = 0; i < LGUEST_MAX_DEVICES; i++)
-		if (lguest_devices[i].type)
-			add_lguest_device(i);
-}
-
-/*D:100 Fairly early in boot, lguest_bus_init() is called to set up the lguest
- * bus.  We check that we are a Guest by checking paravirt_ops.name: there are
- * other ways of checking, but this seems most obvious to me.
- *
- * So we can access the array of "struct lguest_device_desc"s easily, we map
- * that memory and store the pointer in the global "lguest_devices".  Then we
- * register the bus with the core.  Doing two registrations seems clunky to me,
- * but it seems to be the correct sysfs incantation.
- *
- * Finally we call scan_devices() which adds all the devices found in the
- * "struct lguest_device_desc" array. */
-static int __init lguest_bus_init(void)
-{
-	if (strcmp(pv_info.name, "lguest") != 0)
-		return 0;
-
-	/* Devices are in a single page above top of "normal" mem */
-	lguest_devices = lguest_map(max_pfn<<PAGE_SHIFT, 1);
-
-	if (bus_register(&lguest_bus.bus) != 0
-	    || device_register(&lguest_bus.dev) != 0)
-		panic("lguest bus registration failed");
-
-	scan_devices();
-	return 0;
-}
-/* Do this after core stuff, before devices. */
-postcore_initcall(lguest_bus_init);

drivers/lguest/lguest_device.c

@@ -0,0 +1,373 @@
+/*P:050 Lguest guests use a very simple method to describe devices.  It's a
+ * series of device descriptors contained just above the top of normal
+ * memory.
+ *
+ * We use the standard "virtio" device infrastructure, which provides us with a
+ * console, a network and a block driver.  Each one expects some configuration
+ * information and a "virtqueue" mechanism to send and receive data. :*/
+#include <linux/init.h>
+#include <linux/bootmem.h>
+#include <linux/lguest_launcher.h>
+#include <linux/virtio.h>
+#include <linux/virtio_config.h>
+#include <linux/interrupt.h>
+#include <linux/virtio_ring.h>
+#include <linux/err.h>
+#include <asm/io.h>
+#include <asm/paravirt.h>
+#include <asm/lguest_hcall.h>
+
+/* The pointer to our (page) of device descriptions. */
+static void *lguest_devices;
+
+/* Unique numbering for lguest devices. */
+static unsigned int dev_index;
+
+/* For Guests, device memory can be used as normal memory, so we cast away the
+ * __iomem to quieten sparse. */
+static inline void *lguest_map(unsigned long phys_addr, unsigned long pages)
+{
+	return (__force void *)ioremap(phys_addr, PAGE_SIZE*pages);
+}
+
+static inline void lguest_unmap(void *addr)
+{
+	iounmap((__force void __iomem *)addr);
+}
+
+/*D:100 Each lguest device is just a virtio device plus a pointer to its entry
+ * in the lguest_devices page. */
+struct lguest_device {
+	struct virtio_device vdev;
+
+	/* The entry in the lguest_devices page for this device. */
+	struct lguest_device_desc *desc;
+};
+
+/* Since the virtio infrastructure hands us a pointer to the virtio_device all
+ * the time, it helps to have a curt macro to get a pointer to the struct
+ * lguest_device it's enclosed in.  */
+#define to_lgdev(vdev) container_of(vdev, struct lguest_device, vdev)
+
+/*D:130
+ * Device configurations
+ *
+ * The configuration information for a device consists of a series of fields.
+ * The device will look for these fields during setup.
+ *
+ * For us these fields come immediately after that device's descriptor in the
+ * lguest_devices page.
+ *
+ * Each field starts with a "type" byte, a "length" byte, then that number of
+ * bytes of configuration information.  The device descriptor tells us the
+ * total configuration length so we know when we've reached the last field. */
+
+/* type + length bytes */
+#define FHDR_LEN 2
+
+/* This finds the first field of a given type for a device's configuration. */
+static void *lg_find(struct virtio_device *vdev, u8 type, unsigned int *len)
+{
+	struct lguest_device_desc *desc = to_lgdev(vdev)->desc;
+	int i;
+
+	for (i = 0; i < desc->config_len; i += FHDR_LEN + desc->config[i+1]) {
+		if (desc->config[i] == type) {
+			/* Mark it used, so Host can know we looked at it, and
+			 * also so we won't find the same one twice. */
+			desc->config[i] |= 0x80;
+			/* Remember, the second byte is the length. */
+			*len = desc->config[i+1];
+			/* We return a pointer to the field header. */
+			return desc->config + i;
+		}
+	}
+
+	/* Not found: return NULL for failure. */
+	return NULL;
+}
+
+/* Once they've found a field, getting a copy of it is easy. */
+static void lg_get(struct virtio_device *vdev, void *token,
+		   void *buf, unsigned len)
+{
+	/* Check they didn't ask for more than the length of the field! */
+	BUG_ON(len > ((u8 *)token)[1]);
+	memcpy(buf, token + FHDR_LEN, len);
+}
+
+/* Setting the contents is also trivial. */
+static void lg_set(struct virtio_device *vdev, void *token,
+		   const void *buf, unsigned len)
+{
+	BUG_ON(len > ((u8 *)token)[1]);
+	memcpy(token + FHDR_LEN, buf, len);
+}
+
+/* The operations to get and set the status word just access the status field
+ * of the device descriptor. */
+static u8 lg_get_status(struct virtio_device *vdev)
+{
+	return to_lgdev(vdev)->desc->status;
+}
+
+static void lg_set_status(struct virtio_device *vdev, u8 status)
+{
+	to_lgdev(vdev)->desc->status = status;
+}
+
+/*
+ * Virtqueues
+ *
+ * The other piece of infrastructure virtio needs is a "virtqueue": a way of
+ * the Guest device registering buffers for the other side to read from or
+ * write into (ie. send and receive buffers).  Each device can have multiple
+ * virtqueues: for example the console has one queue for sending and one for
+ * receiving.
+ *
+ * Fortunately for us, a very fast shared-memory-plus-descriptors virtqueue
+ * already exists in virtio_ring.c.  We just need to connect it up.
+ *
+ * We start with the information we need to keep about each virtqueue.
+ */
+
+/*D:140 This is the information we remember about each virtqueue. */
+struct lguest_vq_info
+{
+	/* A copy of the information contained in the device config. */
+	struct lguest_vqconfig config;
+
+	/* The address where we mapped the virtio ring, so we can unmap it. */
+	void *pages;
+};
+
+/* When the virtio_ring code wants to prod the Host, it calls us here and we
+ * make a hypercall.  We hand the page number of the virtqueue so the Host
+ * knows which virtqueue we're talking about. */
+static void lg_notify(struct virtqueue *vq)
+{
+	/* We store our virtqueue information in the "priv" pointer of the
+	 * virtqueue structure. */
+	struct lguest_vq_info *lvq = vq->priv;
+
+	hcall(LHCALL_NOTIFY, lvq->config.pfn << PAGE_SHIFT, 0, 0);
+}
+
+/* This routine finds the first virtqueue described in the configuration of
+ * this device and sets it up.
+ *
+ * This is kind of an ugly duckling.  It'd be nicer to have a standard
+ * representation of a virtqueue in the configuration space, but it seems that
+ * everyone wants to do it differently.  The KVM guys want the Guest to
+ * allocate its own pages and tell the Host where they are, but for lguest it's
+ * simpler for the Host to simply tell us where the pages are.
+ *
+ * So we provide devices with a "find virtqueue and set it up" function. */
+static struct virtqueue *lg_find_vq(struct virtio_device *vdev,
+				    bool (*callback)(struct virtqueue *vq))
+{
+	struct lguest_vq_info *lvq;
+	struct virtqueue *vq;
+	unsigned int len;
+	void *token;
+	int err;
+
+	/* Look for a field of the correct type to mark a virtqueue.  Note that
+	 * if this succeeds, then the type will be changed so it won't be found
+	 * again, and future lg_find_vq() calls will find the next
+	 * virtqueue (if any). */
+	token = vdev->config->find(vdev, VIRTIO_CONFIG_F_VIRTQUEUE, &len);
+	if (!token)
+		return ERR_PTR(-ENOENT);
+
+	lvq = kmalloc(sizeof(*lvq), GFP_KERNEL);
+	if (!lvq)
+		return ERR_PTR(-ENOMEM);
+
+	/* Note: we could use a configuration space inside here, just like we
+	 * do for the device.  This would allow expansion in future, because
+	 * our configuration system is designed to be expansible.  But this is
+	 * way easier. */
+	if (len != sizeof(lvq->config)) {
+		dev_err(&vdev->dev, "Unexpected virtio config len %u\n", len);
+		err = -EIO;
+		goto free_lvq;
+	}
+	/* Make a copy of the "struct lguest_vqconfig" field.  We need a copy
+	 * because the config space might not be aligned correctly. */
+	vdev->config->get(vdev, token, &lvq->config, sizeof(lvq->config));
+
+	/* Figure out how many pages the ring will take, and map that memory */
+	lvq->pages = lguest_map((unsigned long)lvq->config.pfn << PAGE_SHIFT,
+				DIV_ROUND_UP(vring_size(lvq->config.num),
+					     PAGE_SIZE));
+	if (!lvq->pages) {
+		err = -ENOMEM;
+		goto free_lvq;
+	}
+
+	/* OK, tell virtio_ring.c to set up a virtqueue now we know its size
+	 * and we've got a pointer to its pages. */
+	vq = vring_new_virtqueue(lvq->config.num, vdev, lvq->pages,
+				 lg_notify, callback);
+	if (!vq) {
+		err = -ENOMEM;
+		goto unmap;
+	}
+
+	/* Tell the interrupt for this virtqueue to go to the virtio_ring
+	 * interrupt handler. */
+	/* FIXME: We used to have a flag for the Host to tell us we could use
+	 * the interrupt as a source of randomness: it'd be nice to have that
+	 * back.. */
+	err = request_irq(lvq->config.irq, vring_interrupt, IRQF_SHARED,
+			  vdev->dev.bus_id, vq);
+	if (err)
+		goto destroy_vring;
+
+	/* Last of all we hook up our 'struct lguest_vq_info" to the
+	 * virtqueue's priv pointer. */
+	vq->priv = lvq;
+	return vq;
+
+destroy_vring:
+	vring_del_virtqueue(vq);
+unmap:
+	lguest_unmap(lvq->pages);
+free_lvq:
+	kfree(lvq);
+	return ERR_PTR(err);
+}
+/*:*/
+
+/* Cleaning up a virtqueue is easy */
+static void lg_del_vq(struct virtqueue *vq)
+{
+	struct lguest_vq_info *lvq = vq->priv;
+
+	/* Tell virtio_ring.c to free the virtqueue. */
+	vring_del_virtqueue(vq);
+	/* Unmap the pages containing the ring. */
+	lguest_unmap(lvq->pages);
+	/* Free our own queue information. */
+	kfree(lvq);
+}
+
+/* The ops structure which hooks everything together. */
+static struct virtio_config_ops lguest_config_ops = {
+	.find = lg_find,
+	.get = lg_get,
+	.set = lg_set,
+	.get_status = lg_get_status,
+	.set_status = lg_set_status,
+	.find_vq = lg_find_vq,
+	.del_vq = lg_del_vq,
+};
+
+/* The root device for the lguest virtio devices.  This makes them appear as
+ * /sys/devices/lguest/0,1,2 not /sys/devices/0,1,2. */
+static struct device lguest_root = {
+	.parent = NULL,
+	.bus_id = "lguest",
+};
+
+/*D:120 This is the core of the lguest bus: actually adding a new device.
+ * It's a separate function because it's neater that way, and because an
+ * earlier version of the code supported hotplug and unplug.  They were removed
+ * early on because they were never used.
+ *
+ * As Andrew Tridgell says, "Untested code is buggy code".
+ *
+ * It's worth reading this carefully: we start with a pointer to the new device
+ * descriptor in the "lguest_devices" page. */
+static void add_lguest_device(struct lguest_device_desc *d)
+{
+	struct lguest_device *ldev;
+
+	ldev = kzalloc(sizeof(*ldev), GFP_KERNEL);
+	if (!ldev) {
+		printk(KERN_EMERG "Cannot allocate lguest dev %u\n",
+		       dev_index++);
+		return;
+	}
+
+	/* This devices' parent is the lguest/ dir. */
+	ldev->vdev.dev.parent = &lguest_root;
+	/* We have a unique device index thanks to the dev_index counter. */
+	ldev->vdev.index = dev_index++;
+	/* The device type comes straight from the descriptor.  There's also a
+	 * device vendor field in the virtio_device struct, which we leave as
+	 * 0. */
+	ldev->vdev.id.device = d->type;
+	/* We have a simple set of routines for querying the device's
+	 * configuration information and setting its status. */
+	ldev->vdev.config = &lguest_config_ops;
+	/* And we remember the device's descriptor for lguest_config_ops. */
+	ldev->desc = d;
+
+	/* register_virtio_device() sets up the generic fields for the struct
+	 * virtio_device and calls device_register().  This makes the bus
+	 * infrastructure look for a matching driver. */
+	if (register_virtio_device(&ldev->vdev) != 0) {
+		printk(KERN_ERR "Failed to register lguest device %u\n",
+		       ldev->vdev.index);
+		kfree(ldev);
+	}
+}
+
+/*D:110 scan_devices() simply iterates through the device page.  The type 0 is
+ * reserved to mean "end of devices". */
+static void scan_devices(void)
+{
+	unsigned int i;
+	struct lguest_device_desc *d;
+
+	/* We start at the page beginning, and skip over each entry. */
+	for (i = 0; i < PAGE_SIZE; i += sizeof(*d) + d->config_len) {
+		d = lguest_devices + i;
+
+		/* Once we hit a zero, stop. */
+		if (d->type == 0)
+			break;
+
+		add_lguest_device(d);
+	}
+}
+
+/*D:105 Fairly early in boot, lguest_devices_init() is called to set up the
+ * lguest device infrastructure.  We check that we are a Guest by checking
+ * pv_info.name: there are other ways of checking, but this seems most
+ * obvious to me.
+ *
+ * So we can access the "struct lguest_device_desc"s easily, we map that memory
+ * and store the pointer in the global "lguest_devices".  Then we register a
+ * root device from which all our devices will hang (this seems to be the
+ * correct sysfs incantation).
+ *
+ * Finally we call scan_devices() which adds all the devices found in the
+ * lguest_devices page. */
+static int __init lguest_devices_init(void)
+{
+	if (strcmp(pv_info.name, "lguest") != 0)
+		return 0;
+
+	if (device_register(&lguest_root) != 0)
+		panic("Could not register lguest root");
+
+	/* Devices are in a single page above top of "normal" mem */
+	lguest_devices = lguest_map(max_pfn<<PAGE_SHIFT, 1);
+
+	scan_devices();
+	return 0;
+}
+/* We do this after core stuff, but before the drivers. */
+postcore_initcall(lguest_devices_init);
+
+/*D:150 At this point in the journey we used to now wade through the lguest
+ * devices themselves: net, block and console.  Since they're all now virtio
+ * devices rather than lguest-specific, I've decided to ignore them.  Mostly,
+ * they're kind of boring.  But this does mean you'll never experience the
+ * thrill of reading the forbidden love scene buried deep in the block driver.
+ *
+ * "make Launcher" beckons, where we answer questions like "Where do Guests
+ * come from?", and "What do you do when someone asks for optimization?". */

drivers/lguest/lguest_user.c

@@ -1,73 +1,17 @@
 /*P:200 This contains all the /dev/lguest code, whereby the userspace launcher
  * controls and communicates with the Guest.  For example, the first write will
- * tell us the memory size, pagetable, entry point and kernel address offset.
- * A read will run the Guest until a signal is pending (-EINTR), or the Guest
- * does a DMA out to the Launcher.  Writes are also used to get a DMA buffer
- * registered by the Guest and to send the Guest an interrupt. :*/
+ * tell us the Guest's memory layout, pagetable, entry point and kernel address
+ * offset.  A read will run the Guest until something happens, such as a signal
+ * or the Guest doing a NOTIFY out to the Launcher. :*/
 #include <linux/uaccess.h>
 #include <linux/miscdevice.h>
 #include <linux/fs.h>
 #include "lg.h"
 
-/*L:030 setup_regs() doesn't really belong in this file, but it gives us an
- * early glimpse deeper into the Host so it's worth having here.
- *
- * Most of the Guest's registers are left alone: we used get_zeroed_page() to
- * allocate the structure, so they will be 0. */
-static void setup_regs(struct lguest_regs *regs, unsigned long start)
-{
-	/* There are four "segment" registers which the Guest needs to boot:
-	 * The "code segment" register (cs) refers to the kernel code segment
-	 * __KERNEL_CS, and the "data", "extra" and "stack" segment registers
-	 * refer to the kernel data segment __KERNEL_DS.
-	 *
-	 * The privilege level is packed into the lower bits.  The Guest runs
-	 * at privilege level 1 (GUEST_PL).*/
-	regs->ds = regs->es = regs->ss = __KERNEL_DS|GUEST_PL;
-	regs->cs = __KERNEL_CS|GUEST_PL;
-
-	/* The "eflags" register contains miscellaneous flags.  Bit 1 (0x002)
-	 * is supposed to always be "1".  Bit 9 (0x200) controls whether
-	 * interrupts are enabled.  We always leave interrupts enabled while
-	 * running the Guest. */
-	regs->eflags = 0x202;
-
-	/* The "Extended Instruction Pointer" register says where the Guest is
-	 * running. */
-	regs->eip = start;
-
-	/* %esi points to our boot information, at physical address 0, so don't
-	 * touch it. */
-}
-
-/*L:310 To send DMA into the Guest, the Launcher needs to be able to ask for a
- * DMA buffer.  This is done by writing LHREQ_GETDMA and the key to
- * /dev/lguest. */
-static long user_get_dma(struct lguest *lg, const u32 __user *input)
-{
-	unsigned long key, udma, irq;
-
-	/* Fetch the key they wrote to us. */
-	if (get_user(key, input) != 0)
-		return -EFAULT;
-	/* Look for a free Guest DMA buffer bound to that key. */
-	udma = get_dma_buffer(lg, key, &irq);
-	if (!udma)
-		return -ENOENT;
-
-	/* We need to tell the Launcher what interrupt the Guest expects after
-	 * the buffer is filled.  We stash it in udma->used_len. */
-	lgwrite_u32(lg, udma + offsetof(struct lguest_dma, used_len), irq);
-
-	/* The (guest-physical) address of the DMA buffer is returned from
-	 * the write(). */
-	return udma;
-}
-
 /*L:315 To force the Guest to stop running and return to the Launcher, the
  * Waker sets writes LHREQ_BREAK and the value "1" to /dev/lguest.  The
  * Launcher then writes LHREQ_BREAK and "0" to release the Waker. */
-static int break_guest_out(struct lguest *lg, const u32 __user *input)
+static int break_guest_out(struct lguest *lg, const unsigned long __user *input)
 {
 	unsigned long on;
 
@@ -90,9 +34,9 @@ static int break_guest_out(struct lguest *lg, const u32 __user *input)
 
 /*L:050 Sending an interrupt is done by writing LHREQ_IRQ and an interrupt
  * number to /dev/lguest. */
-static int user_send_irq(struct lguest *lg, const u32 __user *input)
+static int user_send_irq(struct lguest *lg, const unsigned long __user *input)
 {
-	u32 irq;
+	unsigned long irq;
 
 	if (get_user(irq, input) != 0)
 		return -EFAULT;
@@ -133,17 +77,19 @@ static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o)
 		return len;
 	}
 
-	/* If we returned from read() last time because the Guest sent DMA,
+	/* If we returned from read() last time because the Guest notified,
 	 * clear the flag. */
-	if (lg->dma_is_pending)
-		lg->dma_is_pending = 0;
+	if (lg->pending_notify)
+		lg->pending_notify = 0;
 
 	/* Run the Guest until something interesting happens. */
 	return run_guest(lg, (unsigned long __user *)user);
 }
 
-/*L:020 The initialization write supplies 4 32-bit values (in addition to the
- * 32-bit LHREQ_INITIALIZE value).  These are:
+/*L:020 The initialization write supplies 4 pointer sized (32 or 64 bit)
+ * values (in addition to the LHREQ_INITIALIZE value).  These are:
+ *
+ * base: The start of the Guest-physical memory inside the Launcher memory.
  *
  * pfnlimit: The highest (Guest-physical) page number the Guest should be
  * allowed to access.  The Launcher has to live in Guest memory, so it sets
@@ -153,23 +99,17 @@ static ssize_t read(struct file *file, char __user *user, size_t size,loff_t*o)
  * pagetables (which are set up by the Launcher).
  *
  * start: The first instruction to execute ("eip" in x86-speak).
- *
- * page_offset: The PAGE_OFFSET constant in the Guest kernel.  We should
- * probably wean the code off this, but it's a very useful constant!  Any
- * address above this is within the Guest kernel, and any kernel address can
- * quickly converted from physical to virtual by adding PAGE_OFFSET.  It's
- * 0xC0000000 (3G) by default, but it's configurable at kernel build time.
  */
-static int initialize(struct file *file, const u32 __user *input)
+static int initialize(struct file *file, const unsigned long __user *input)
 {
 	/* "struct lguest" contains everything we (the Host) know about a
 	 * Guest. */
 	struct lguest *lg;
-	int err, i;
-	u32 args[4];
+	int err;
+	unsigned long args[4];
 
-	/* We grab the Big Lguest lock, which protects the global array
-	 * "lguests" and multiple simultaneous initializations. */
+	/* We grab the Big Lguest lock, which protects against multiple
+	 * simultaneous initializations. */
 	mutex_lock(&lguest_lock);
 	/* You can't initialize twice!  Close the device and start again... */
 	if (file->private_data) {
@@ -182,20 +122,15 @@ static int initialize(struct file *file, const u32 __user *input)
 		goto unlock;
 	}
 
-	/* Find an unused guest. */
-	i = find_free_guest();
-	if (i < 0) {
-		err = -ENOSPC;
+	lg = kzalloc(sizeof(*lg), GFP_KERNEL);
+	if (!lg) {
+		err = -ENOMEM;
 		goto unlock;
 	}
-	/* OK, we have an index into the "lguest" array: "lg" is a convenient
-	 * pointer. */
-	lg = &lguests[i];
 
 	/* Populate the easy fields of our "struct lguest" */
-	lg->guestid = i;
-	lg->pfn_limit = args[0];
-	lg->page_offset = args[3];
+	lg->mem_base = (void __user *)(long)args[0];
+	lg->pfn_limit = args[1];
 
 	/* We need a complete page for the Guest registers: they are accessible
 	 * to the Guest and we can only grant it access to whole pages. */
@@ -210,17 +145,13 @@ static int initialize(struct file *file, const u32 __user *input)
 	/* Initialize the Guest's shadow page tables, using the toplevel
 	 * address the Launcher gave us.  This allocates memory, so can
 	 * fail. */
-	err = init_guest_pagetable(lg, args[1]);
+	err = init_guest_pagetable(lg, args[2]);
 	if (err)
 		goto free_regs;
 
 	/* Now we initialize the Guest's registers, handing it the start
 	 * address. */
-	setup_regs(lg->regs, args[2]);
-
-	/* There are a couple of GDT entries the Guest expects when first
-	 * booting. */
-	setup_guest_gdt(lg);
+	lguest_arch_setup_regs(lg, args[3]);
 
 	/* The timer for lguest's clock needs initialization. */
 	init_clockdev(lg);
@@ -260,18 +191,19 @@ unlock:
 /*L:010 The first operation the Launcher does must be a write.  All writes
  * start with a 32 bit number: for the first write this must be
  * LHREQ_INITIALIZE to set up the Guest.  After that the Launcher can use
- * writes of other values to get DMA buffers and send interrupts. */
-static ssize_t write(struct file *file, const char __user *input,
+ * writes of other values to send interrupts. */
+static ssize_t write(struct file *file, const char __user *in,
 		     size_t size, loff_t *off)
 {
 	/* Once the guest is initialized, we hold the "struct lguest" in the
 	 * file private data. */
 	struct lguest *lg = file->private_data;
-	u32 req;
+	const unsigned long __user *input = (const unsigned long __user *)in;
+	unsigned long req;
 
 	if (get_user(req, input) != 0)
 		return -EFAULT;
-	input += sizeof(req);
+	input++;
 
 	/* If you haven't initialized, you must do that first. */
 	if (req != LHREQ_INITIALIZE && !lg)
@@ -287,13 +219,11 @@ static ssize_t write(struct file *file, const char __user *input,
 
 	switch (req) {
 	case LHREQ_INITIALIZE:
-		return initialize(file, (const u32 __user *)input);
-	case LHREQ_GETDMA:
-		return user_get_dma(lg, (const u32 __user *)input);
+		return initialize(file, input);
 	case LHREQ_IRQ:
-		return user_send_irq(lg, (const u32 __user *)input);
+		return user_send_irq(lg, input);
 	case LHREQ_BREAK:
-		return break_guest_out(lg, (const u32 __user *)input);
+		return break_guest_out(lg, input);
 	default:
 		return -EINVAL;
 	}
@@ -319,8 +249,6 @@ static int close(struct inode *inode, struct file *file)
 	mutex_lock(&lguest_lock);
 	/* Cancels the hrtimer set via LHCALL_SET_CLOCKEVENT. */
 	hrtimer_cancel(&lg->hrt);
-	/* Free any DMA buffers the Guest had bound. */
-	release_all_dma(lg);
 	/* Free up the shadow page tables for the Guest. */
 	free_guest_pagetable(lg);
 	/* Now all the memory cleanups are done, it's safe to release the

drivers/lguest/page_tables.c

@@ -13,6 +13,7 @@
 #include <linux/random.h>
 #include <linux/percpu.h>
 #include <asm/tlbflush.h>
+#include <asm/uaccess.h>
 #include "lg.h"
 
 /*M:008 We hold reference to pages, which prevents them from being swapped.
@@ -44,44 +45,32 @@
  *  (vii) Setting up the page tables initially.
  :*/
 
-/* Pages a 4k long, and each page table entry is 4 bytes long, giving us 1024
- * (or 2^10) entries per page. */
-#define PTES_PER_PAGE_SHIFT 10
-#define PTES_PER_PAGE (1 << PTES_PER_PAGE_SHIFT)
 
 /* 1024 entries in a page table page maps 1024 pages: 4MB.  The Switcher is
  * conveniently placed at the top 4MB, so it uses a separate, complete PTE
  * page.  */
-#define SWITCHER_PGD_INDEX (PTES_PER_PAGE - 1)
+#define SWITCHER_PGD_INDEX (PTRS_PER_PGD - 1)
 
 /* We actually need a separate PTE page for each CPU.  Remember that after the
  * Switcher code itself comes two pages for each CPU, and we don't want this
  * CPU's guest to see the pages of any other CPU. */
-static DEFINE_PER_CPU(spte_t *, switcher_pte_pages);
+static DEFINE_PER_CPU(pte_t *, switcher_pte_pages);
 #define switcher_pte_page(cpu) per_cpu(switcher_pte_pages, cpu)
 
 /*H:320 With our shadow and Guest types established, we need to deal with
  * them: the page table code is curly enough to need helper functions to keep
  * it clear and clean.
  *
- * The first helper takes a virtual address, and says which entry in the top
- * level page table deals with that address.  Since each top level entry deals
- * with 4M, this effectively divides by 4M. */
-static unsigned vaddr_to_pgd_index(unsigned long vaddr)
-{
-	return vaddr >> (PAGE_SHIFT + PTES_PER_PAGE_SHIFT);
-}
-
-/* There are two functions which return pointers to the shadow (aka "real")
+ * There are two functions which return pointers to the shadow (aka "real")
  * page tables.
  *
  * spgd_addr() takes the virtual address and returns a pointer to the top-level
  * page directory entry for that address.  Since we keep track of several page
  * tables, the "i" argument tells us which one we're interested in (it's
  * usually the current one). */
-static spgd_t *spgd_addr(struct lguest *lg, u32 i, unsigned long vaddr)
+static pgd_t *spgd_addr(struct lguest *lg, u32 i, unsigned long vaddr)
 {
-	unsigned int index = vaddr_to_pgd_index(vaddr);
+	unsigned int index = pgd_index(vaddr);
 
 	/* We kill any Guest trying to touch the Switcher addresses. */
 	if (index >= SWITCHER_PGD_INDEX) {
@@ -95,28 +84,28 @@ static spgd_t *spgd_addr(struct lguest *lg, u32 i, unsigned long vaddr)
 /* This routine then takes the PGD entry given above, which contains the
  * address of the PTE page.  It then returns a pointer to the PTE entry for the
  * given address. */
-static spte_t *spte_addr(struct lguest *lg, spgd_t spgd, unsigned long vaddr)
+static pte_t *spte_addr(struct lguest *lg, pgd_t spgd, unsigned long vaddr)
 {
-	spte_t *page = __va(spgd.pfn << PAGE_SHIFT);
+	pte_t *page = __va(pgd_pfn(spgd) << PAGE_SHIFT);
 	/* You should never call this if the PGD entry wasn't valid */
-	BUG_ON(!(spgd.flags & _PAGE_PRESENT));
-	return &page[(vaddr >> PAGE_SHIFT) % PTES_PER_PAGE];
+	BUG_ON(!(pgd_flags(spgd) & _PAGE_PRESENT));
+	return &page[(vaddr >> PAGE_SHIFT) % PTRS_PER_PTE];
 }
 
 /* These two functions just like the above two, except they access the Guest
  * page tables.  Hence they return a Guest address. */
 static unsigned long gpgd_addr(struct lguest *lg, unsigned long vaddr)
 {
-	unsigned int index = vaddr >> (PAGE_SHIFT + PTES_PER_PAGE_SHIFT);
-	return lg->pgdirs[lg->pgdidx].cr3 + index * sizeof(gpgd_t);
+	unsigned int index = vaddr >> (PGDIR_SHIFT);
+	return lg->pgdirs[lg->pgdidx].gpgdir + index * sizeof(pgd_t);
 }
 
 static unsigned long gpte_addr(struct lguest *lg,
-			       gpgd_t gpgd, unsigned long vaddr)
+			       pgd_t gpgd, unsigned long vaddr)
 {
-	unsigned long gpage = gpgd.pfn << PAGE_SHIFT;
-	BUG_ON(!(gpgd.flags & _PAGE_PRESENT));
-	return gpage + ((vaddr>>PAGE_SHIFT) % PTES_PER_PAGE) * sizeof(gpte_t);
+	unsigned long gpage = pgd_pfn(gpgd) << PAGE_SHIFT;
+	BUG_ON(!(pgd_flags(gpgd) & _PAGE_PRESENT));
+	return gpage + ((vaddr>>PAGE_SHIFT) % PTRS_PER_PTE) * sizeof(pte_t);
 }
 
 /*H:350 This routine takes a page number given by the Guest and converts it to
@@ -149,53 +138,55 @@ static unsigned long get_pfn(unsigned long virtpfn, int write)
  * entry can be a little tricky.  The flags are (almost) the same, but the
  * Guest PTE contains a virtual page number: the CPU needs the real page
  * number. */
-static spte_t gpte_to_spte(struct lguest *lg, gpte_t gpte, int write)
+static pte_t gpte_to_spte(struct lguest *lg, pte_t gpte, int write)
 {
-	spte_t spte;
-	unsigned long pfn;
+	unsigned long pfn, base, flags;
 
 	/* The Guest sets the global flag, because it thinks that it is using
 	 * PGE.  We only told it to use PGE so it would tell us whether it was
 	 * flushing a kernel mapping or a userspace mapping.  We don't actually
 	 * use the global bit, so throw it away. */
-	spte.flags = (gpte.flags & ~_PAGE_GLOBAL);
+	flags = (pte_flags(gpte) & ~_PAGE_GLOBAL);
+
+	/* The Guest's pages are offset inside the Launcher. */
+	base = (unsigned long)lg->mem_base / PAGE_SIZE;
 
 	/* We need a temporary "unsigned long" variable to hold the answer from
 	 * get_pfn(), because it returns 0xFFFFFFFF on failure, which wouldn't
 	 * fit in spte.pfn.  get_pfn() finds the real physical number of the
 	 * page, given the virtual number. */
-	pfn = get_pfn(gpte.pfn, write);
+	pfn = get_pfn(base + pte_pfn(gpte), write);
 	if (pfn == -1UL) {
-		kill_guest(lg, "failed to get page %u", gpte.pfn);
+		kill_guest(lg, "failed to get page %lu", pte_pfn(gpte));
 		/* When we destroy the Guest, we'll go through the shadow page
 		 * tables and release_pte() them.  Make sure we don't think
 		 * this one is valid! */
-		spte.flags = 0;
+		flags = 0;
 	}
-	/* Now we assign the page number, and our shadow PTE is complete. */
-	spte.pfn = pfn;
-	return spte;
+	/* Now we assemble our shadow PTE from the page number and flags. */
+	return pfn_pte(pfn, __pgprot(flags));
 }
 
 /*H:460 And to complete the chain, release_pte() looks like this: */
-static void release_pte(spte_t pte)
+static void release_pte(pte_t pte)
 {
 	/* Remember that get_user_pages() took a reference to the page, in
 	 * get_pfn()?  We have to put it back now. */
-	if (pte.flags & _PAGE_PRESENT)
-		put_page(pfn_to_page(pte.pfn));
+	if (pte_flags(pte) & _PAGE_PRESENT)
+		put_page(pfn_to_page(pte_pfn(pte)));
 }
 /*:*/
 
-static void check_gpte(struct lguest *lg, gpte_t gpte)
+static void check_gpte(struct lguest *lg, pte_t gpte)
 {
-	if ((gpte.flags & (_PAGE_PWT|_PAGE_PSE)) || gpte.pfn >= lg->pfn_limit)
+	if ((pte_flags(gpte) & (_PAGE_PWT|_PAGE_PSE))
+	    || pte_pfn(gpte) >= lg->pfn_limit)
 		kill_guest(lg, "bad page table entry");
 }
 
-static void check_gpgd(struct lguest *lg, gpgd_t gpgd)
+static void check_gpgd(struct lguest *lg, pgd_t gpgd)
 {
-	if ((gpgd.flags & ~_PAGE_TABLE) || gpgd.pfn >= lg->pfn_limit)
+	if ((pgd_flags(gpgd) & ~_PAGE_TABLE) || pgd_pfn(gpgd) >= lg->pfn_limit)
 		kill_guest(lg, "bad page directory entry");
 }
 
@@ -211,21 +202,21 @@ static void check_gpgd(struct lguest *lg, gpgd_t gpgd)
  * true. */
 int demand_page(struct lguest *lg, unsigned long vaddr, int errcode)
 {
-	gpgd_t gpgd;
-	spgd_t *spgd;
+	pgd_t gpgd;
+	pgd_t *spgd;
 	unsigned long gpte_ptr;
-	gpte_t gpte;
-	spte_t *spte;
+	pte_t gpte;
+	pte_t *spte;
 
 	/* First step: get the top-level Guest page table entry. */
-	gpgd = mkgpgd(lgread_u32(lg, gpgd_addr(lg, vaddr)));
+	gpgd = lgread(lg, gpgd_addr(lg, vaddr), pgd_t);
 	/* Toplevel not present?  We can't map it in. */
-	if (!(gpgd.flags & _PAGE_PRESENT))
+	if (!(pgd_flags(gpgd) & _PAGE_PRESENT))
 		return 0;
 
 	/* Now look at the matching shadow entry. */
 	spgd = spgd_addr(lg, lg->pgdidx, vaddr);
-	if (!(spgd->flags & _PAGE_PRESENT)) {
+	if (!(pgd_flags(*spgd) & _PAGE_PRESENT)) {
 		/* No shadow entry: allocate a new shadow PTE page. */
 		unsigned long ptepage = get_zeroed_page(GFP_KERNEL);
 		/* This is not really the Guest's fault, but killing it is
@@ -238,34 +229,35 @@ int demand_page(struct lguest *lg, unsigned long vaddr, int errcode)
 		check_gpgd(lg, gpgd);
 		/* And we copy the flags to the shadow PGD entry.  The page
 		 * number in the shadow PGD is the page we just allocated. */
-		spgd->raw.val = (__pa(ptepage) | gpgd.flags);
+		*spgd = __pgd(__pa(ptepage) | pgd_flags(gpgd));
 	}
 
 	/* OK, now we look at the lower level in the Guest page table: keep its
 	 * address, because we might update it later. */
 	gpte_ptr = gpte_addr(lg, gpgd, vaddr);
-	gpte = mkgpte(lgread_u32(lg, gpte_ptr));
+	gpte = lgread(lg, gpte_ptr, pte_t);
 
 	/* If this page isn't in the Guest page tables, we can't page it in. */
-	if (!(gpte.flags & _PAGE_PRESENT))
+	if (!(pte_flags(gpte) & _PAGE_PRESENT))
 		return 0;
 
 	/* Check they're not trying to write to a page the Guest wants
 	 * read-only (bit 2 of errcode == write). */
-	if ((errcode & 2) && !(gpte.flags & _PAGE_RW))
+	if ((errcode & 2) && !(pte_flags(gpte) & _PAGE_RW))
 		return 0;
 
 	/* User access to a kernel page? (bit 3 == user access) */
-	if ((errcode & 4) && !(gpte.flags & _PAGE_USER))
+	if ((errcode & 4) && !(pte_flags(gpte) & _PAGE_USER))
 		return 0;
 
 	/* Check that the Guest PTE flags are OK, and the page number is below
 	 * the pfn_limit (ie. not mapping the Launcher binary). */
 	check_gpte(lg, gpte);
 	/* Add the _PAGE_ACCESSED and (for a write) _PAGE_DIRTY flag */
-	gpte.flags |= _PAGE_ACCESSED;
+	gpte = pte_mkyoung(gpte);
+
 	if (errcode & 2)
-		gpte.flags |= _PAGE_DIRTY;
+		gpte = pte_mkdirty(gpte);
 
 	/* Get the pointer to the shadow PTE entry we're going to set. */
 	spte = spte_addr(lg, *spgd, vaddr);
@@ -275,21 +267,18 @@ int demand_page(struct lguest *lg, unsigned long vaddr, int errcode)
 
 	/* If this is a write, we insist that the Guest page is writable (the
 	 * final arg to gpte_to_spte()). */
-	if (gpte.flags & _PAGE_DIRTY)
+	if (pte_dirty(gpte))
 		*spte = gpte_to_spte(lg, gpte, 1);
-	else {
+	else
 		/* If this is a read, don't set the "writable" bit in the page
 		 * table entry, even if the Guest says it's writable.  That way
 		 * we come back here when a write does actually ocur, so we can
 		 * update the Guest's _PAGE_DIRTY flag. */
-		gpte_t ro_gpte = gpte;
-		ro_gpte.flags &= ~_PAGE_RW;
-		*spte = gpte_to_spte(lg, ro_gpte, 0);
-	}
+		*spte = gpte_to_spte(lg, pte_wrprotect(gpte), 0);
 
 	/* Finally, we write the Guest PTE entry back: we've set the
 	 * _PAGE_ACCESSED and maybe the _PAGE_DIRTY flags. */
-	lgwrite_u32(lg, gpte_ptr, gpte.raw.val);
+	lgwrite(lg, gpte_ptr, pte_t, gpte);
 
 	/* We succeeded in mapping the page! */
 	return 1;
@@ -305,17 +294,18 @@ int demand_page(struct lguest *lg, unsigned long vaddr, int errcode)
  * mapped by the shadow page tables, and is it writable? */
 static int page_writable(struct lguest *lg, unsigned long vaddr)
 {
-	spgd_t *spgd;
+	pgd_t *spgd;
 	unsigned long flags;
 
 	/* Look at the top level entry: is it present? */
 	spgd = spgd_addr(lg, lg->pgdidx, vaddr);
-	if (!(spgd->flags & _PAGE_PRESENT))
+	if (!(pgd_flags(*spgd) & _PAGE_PRESENT))
 		return 0;
 
 	/* Check the flags on the pte entry itself: it must be present and
 	 * writable. */
-	flags = spte_addr(lg, *spgd, vaddr)->flags;
+	flags = pte_flags(*(spte_addr(lg, *spgd, vaddr)));
+
 	return (flags & (_PAGE_PRESENT|_PAGE_RW)) == (_PAGE_PRESENT|_PAGE_RW);
 }
 
@@ -329,22 +319,22 @@ void pin_page(struct lguest *lg, unsigned long vaddr)
 }
 
 /*H:450 If we chase down the release_pgd() code, it looks like this: */
-static void release_pgd(struct lguest *lg, spgd_t *spgd)
+static void release_pgd(struct lguest *lg, pgd_t *spgd)
 {
 	/* If the entry's not present, there's nothing to release. */
-	if (spgd->flags & _PAGE_PRESENT) {
+	if (pgd_flags(*spgd) & _PAGE_PRESENT) {
 		unsigned int i;
 		/* Converting the pfn to find the actual PTE page is easy: turn
 		 * the page number into a physical address, then convert to a
 		 * virtual address (easy for kernel pages like this one). */
-		spte_t *ptepage = __va(spgd->pfn << PAGE_SHIFT);
+		pte_t *ptepage = __va(pgd_pfn(*spgd) << PAGE_SHIFT);
 		/* For each entry in the page, we might need to release it. */
-		for (i = 0; i < PTES_PER_PAGE; i++)
+		for (i = 0; i < PTRS_PER_PTE; i++)
 			release_pte(ptepage[i]);
 		/* Now we can free the page of PTEs */
 		free_page((long)ptepage);
 		/* And zero out the PGD entry we we never release it twice. */
-		spgd->raw.val = 0;
+		*spgd = __pgd(0);
 	}
 }
 
@@ -356,7 +346,7 @@ static void flush_user_mappings(struct lguest *lg, int idx)
 {
 	unsigned int i;
 	/* Release every pgd entry up to the kernel's address. */
-	for (i = 0; i < vaddr_to_pgd_index(lg->page_offset); i++)
+	for (i = 0; i < pgd_index(lg->kernel_address); i++)
 		release_pgd(lg, lg->pgdirs[idx].pgdir + i);
 }
 
@@ -369,6 +359,25 @@ void guest_pagetable_flush_user(struct lguest *lg)
 }
 /*:*/
 
+/* We walk down the guest page tables to get a guest-physical address */
+unsigned long guest_pa(struct lguest *lg, unsigned long vaddr)
+{
+	pgd_t gpgd;
+	pte_t gpte;
+
+	/* First step: get the top-level Guest page table entry. */
+	gpgd = lgread(lg, gpgd_addr(lg, vaddr), pgd_t);
+	/* Toplevel not present?  We can't map it in. */
+	if (!(pgd_flags(gpgd) & _PAGE_PRESENT))
+		kill_guest(lg, "Bad address %#lx", vaddr);
+
+	gpte = lgread(lg, gpte_addr(lg, gpgd, vaddr), pte_t);
+	if (!(pte_flags(gpte) & _PAGE_PRESENT))
+		kill_guest(lg, "Bad address %#lx", vaddr);
+
+	return pte_pfn(gpte) * PAGE_SIZE | (vaddr & ~PAGE_MASK);
+}
+
 /* We keep several page tables.  This is a simple routine to find the page
  * table (if any) corresponding to this top-level address the Guest has given
  * us. */
@@ -376,7 +385,7 @@ static unsigned int find_pgdir(struct lguest *lg, unsigned long pgtable)
 {
 	unsigned int i;
 	for (i = 0; i < ARRAY_SIZE(lg->pgdirs); i++)
-		if (lg->pgdirs[i].cr3 == pgtable)
+		if (lg->pgdirs[i].gpgdir == pgtable)
 			break;
 	return i;
 }
@@ -385,7 +394,7 @@ static unsigned int find_pgdir(struct lguest *lg, unsigned long pgtable)
  * allocate a new one (and so the kernel parts are not there), we set
  * blank_pgdir. */
 static unsigned int new_pgdir(struct lguest *lg,
-			      unsigned long cr3,
+			      unsigned long gpgdir,
 			      int *blank_pgdir)
 {
 	unsigned int next;
@@ -395,7 +404,7 @@ static unsigned int new_pgdir(struct lguest *lg,
 	next = random32() % ARRAY_SIZE(lg->pgdirs);
 	/* If it's never been allocated at all before, try now. */
 	if (!lg->pgdirs[next].pgdir) {
-		lg->pgdirs[next].pgdir = (spgd_t *)get_zeroed_page(GFP_KERNEL);
+		lg->pgdirs[next].pgdir = (pgd_t *)get_zeroed_page(GFP_KERNEL);
 		/* If the allocation fails, just keep using the one we have */
 		if (!lg->pgdirs[next].pgdir)
 			next = lg->pgdidx;
@@ -405,7 +414,7 @@ static unsigned int new_pgdir(struct lguest *lg,
 			*blank_pgdir = 1;
 	}
 	/* Record which Guest toplevel this shadows. */
-	lg->pgdirs[next].cr3 = cr3;
+	lg->pgdirs[next].gpgdir = gpgdir;
 	/* Release all the non-kernel mappings. */
 	flush_user_mappings(lg, next);
 
@@ -472,26 +481,27 @@ void guest_pagetable_clear_all(struct lguest *lg)
  * they set _PAGE_DIRTY then we can put a writable PTE entry in immediately.
  */
 static void do_set_pte(struct lguest *lg, int idx,
-		       unsigned long vaddr, gpte_t gpte)
+		       unsigned long vaddr, pte_t gpte)
 {
 	/* Look up the matching shadow page directot entry. */
-	spgd_t *spgd = spgd_addr(lg, idx, vaddr);
+	pgd_t *spgd = spgd_addr(lg, idx, vaddr);
 
 	/* If the top level isn't present, there's no entry to update. */
-	if (spgd->flags & _PAGE_PRESENT) {
+	if (pgd_flags(*spgd) & _PAGE_PRESENT) {
 		/* Otherwise, we start by releasing the existing entry. */
-		spte_t *spte = spte_addr(lg, *spgd, vaddr);
+		pte_t *spte = spte_addr(lg, *spgd, vaddr);
 		release_pte(*spte);
 
 		/* If they're setting this entry as dirty or accessed, we might
 		 * as well put that entry they've given us in now.  This shaves
 		 * 10% off a copy-on-write micro-benchmark. */
-		if (gpte.flags & (_PAGE_DIRTY | _PAGE_ACCESSED)) {
+		if (pte_flags(gpte) & (_PAGE_DIRTY | _PAGE_ACCESSED)) {
 			check_gpte(lg, gpte);
-			*spte = gpte_to_spte(lg, gpte, gpte.flags&_PAGE_DIRTY);
+			*spte = gpte_to_spte(lg, gpte,
+					     pte_flags(gpte) & _PAGE_DIRTY);
 		} else
 			/* Otherwise we can demand_page() it in later. */
-			spte->raw.val = 0;
+			*spte = __pte(0);
 	}
 }
 
@@ -506,18 +516,18 @@ static void do_set_pte(struct lguest *lg, int idx,
  * The benefit is that when we have to track a new page table, we can copy keep
  * all the kernel mappings.  This speeds up context switch immensely. */
 void guest_set_pte(struct lguest *lg,
-		   unsigned long cr3, unsigned long vaddr, gpte_t gpte)
+		   unsigned long gpgdir, unsigned long vaddr, pte_t gpte)
 {
 	/* Kernel mappings must be changed on all top levels.  Slow, but
 	 * doesn't happen often. */
-	if (vaddr >= lg->page_offset) {
+	if (vaddr >= lg->kernel_address) {
 		unsigned int i;
 		for (i = 0; i < ARRAY_SIZE(lg->pgdirs); i++)
 			if (lg->pgdirs[i].pgdir)
 				do_set_pte(lg, i, vaddr, gpte);
 	} else {
 		/* Is this page table one we have a shadow for? */
-		int pgdir = find_pgdir(lg, cr3);
+		int pgdir = find_pgdir(lg, gpgdir);
 		if (pgdir != ARRAY_SIZE(lg->pgdirs))
 			/* If so, do the update. */
 			do_set_pte(lg, pgdir, vaddr, gpte);
@@ -538,7 +548,7 @@ void guest_set_pte(struct lguest *lg,
  *
  * So with that in mind here's our code to to update a (top-level) PGD entry:
  */
-void guest_set_pmd(struct lguest *lg, unsigned long cr3, u32 idx)
+void guest_set_pmd(struct lguest *lg, unsigned long gpgdir, u32 idx)
 {
 	int pgdir;
 
@@ -548,7 +558,7 @@ void guest_set_pmd(struct lguest *lg, unsigned long cr3, u32 idx)
 		return;
 
 	/* If they're talking about a page table we have a shadow for... */
-	pgdir = find_pgdir(lg, cr3);
+	pgdir = find_pgdir(lg, gpgdir);
 	if (pgdir < ARRAY_SIZE(lg->pgdirs))
 		/* ... throw it away. */
 		release_pgd(lg, lg->pgdirs[pgdir].pgdir + idx);
@@ -560,21 +570,34 @@ void guest_set_pmd(struct lguest *lg, unsigned long cr3, u32 idx)
  * its first page table is.  We set some things up here: */
 int init_guest_pagetable(struct lguest *lg, unsigned long pgtable)
 {
-	/* In flush_user_mappings() we loop from 0 to
-	 * "vaddr_to_pgd_index(lg->page_offset)".  This assumes it won't hit
-	 * the Switcher mappings, so check that now. */
-	if (vaddr_to_pgd_index(lg->page_offset) >= SWITCHER_PGD_INDEX)
-		return -EINVAL;
 	/* We start on the first shadow page table, and give it a blank PGD
 	 * page. */
 	lg->pgdidx = 0;
-	lg->pgdirs[lg->pgdidx].cr3 = pgtable;
-	lg->pgdirs[lg->pgdidx].pgdir = (spgd_t*)get_zeroed_page(GFP_KERNEL);
+	lg->pgdirs[lg->pgdidx].gpgdir = pgtable;
+	lg->pgdirs[lg->pgdidx].pgdir = (pgd_t*)get_zeroed_page(GFP_KERNEL);
 	if (!lg->pgdirs[lg->pgdidx].pgdir)
 		return -ENOMEM;
 	return 0;
 }
 
+/* When the Guest calls LHCALL_LGUEST_INIT we do more setup. */
+void page_table_guest_data_init(struct lguest *lg)
+{
+	/* We get the kernel address: above this is all kernel memory. */
+	if (get_user(lg->kernel_address, &lg->lguest_data->kernel_address)
+	    /* We tell the Guest that it can't use the top 4MB of virtual
+	     * addresses used by the Switcher. */
+	    || put_user(4U*1024*1024, &lg->lguest_data->reserve_mem)
+	    || put_user(lg->pgdirs[lg->pgdidx].gpgdir,&lg->lguest_data->pgdir))
+		kill_guest(lg, "bad guest page %p", lg->lguest_data);
+
+	/* In flush_user_mappings() we loop from 0 to
+	 * "pgd_index(lg->kernel_address)".  This assumes it won't hit the
+	 * Switcher mappings, so check that now. */
+	if (pgd_index(lg->kernel_address) >= SWITCHER_PGD_INDEX)
+		kill_guest(lg, "bad kernel address %#lx", lg->kernel_address);
+}
+
 /* When a Guest dies, our cleanup is fairly simple. */
 void free_guest_pagetable(struct lguest *lg)
 {
@@ -594,14 +617,14 @@ void free_guest_pagetable(struct lguest *lg)
  * for each CPU already set up, we just need to hook them in. */
 void map_switcher_in_guest(struct lguest *lg, struct lguest_pages *pages)
 {
-	spte_t *switcher_pte_page = __get_cpu_var(switcher_pte_pages);
-	spgd_t switcher_pgd;
-	spte_t regs_pte;
+	pte_t *switcher_pte_page = __get_cpu_var(switcher_pte_pages);
+	pgd_t switcher_pgd;
+	pte_t regs_pte;
 
 	/* Make the last PGD entry for this Guest point to the Switcher's PTE
 	 * page for this CPU (with appropriate flags). */
-	switcher_pgd.pfn = __pa(switcher_pte_page) >> PAGE_SHIFT;
-	switcher_pgd.flags = _PAGE_KERNEL;
+	switcher_pgd = __pgd(__pa(switcher_pte_page) | _PAGE_KERNEL);
+
 	lg->pgdirs[lg->pgdidx].pgdir[SWITCHER_PGD_INDEX] = switcher_pgd;
 
 	/* We also change the Switcher PTE page.  When we're running the Guest,
@@ -611,10 +634,8 @@ void map_switcher_in_guest(struct lguest *lg, struct lguest_pages *pages)
 	 * CPU's "struct lguest_pages": if we make sure the Guest's register
 	 * page is already mapped there, we don't have to copy them out
 	 * again. */
-	regs_pte.pfn = __pa(lg->regs_page) >> PAGE_SHIFT;
-	regs_pte.flags = _PAGE_KERNEL;
-	switcher_pte_page[(unsigned long)pages/PAGE_SIZE%PTES_PER_PAGE]
-		= regs_pte;
+	regs_pte = pfn_pte (__pa(lg->regs_page) >> PAGE_SHIFT, __pgprot(_PAGE_KERNEL));
+	switcher_pte_page[(unsigned long)pages/PAGE_SIZE%PTRS_PER_PTE] = regs_pte;
 }
 /*:*/
 
@@ -635,24 +656,25 @@ static __init void populate_switcher_pte_page(unsigned int cpu,
 					      unsigned int pages)
 {
 	unsigned int i;
-	spte_t *pte = switcher_pte_page(cpu);
+	pte_t *pte = switcher_pte_page(cpu);
 
 	/* The first entries are easy: they map the Switcher code. */
 	for (i = 0; i < pages; i++) {
-		pte[i].pfn = page_to_pfn(switcher_page[i]);
-		pte[i].flags = _PAGE_PRESENT|_PAGE_ACCESSED;
+		pte[i] = mk_pte(switcher_page[i],
+				__pgprot(_PAGE_PRESENT|_PAGE_ACCESSED));
 	}
 
 	/* The only other thing we map is this CPU's pair of pages. */
 	i = pages + cpu*2;
 
 	/* First page (Guest registers) is writable from the Guest */
-	pte[i].pfn = page_to_pfn(switcher_page[i]);
-	pte[i].flags = _PAGE_PRESENT|_PAGE_ACCESSED|_PAGE_RW;
+	pte[i] = pfn_pte(page_to_pfn(switcher_page[i]),
+			 __pgprot(_PAGE_PRESENT|_PAGE_ACCESSED|_PAGE_RW));
+
 	/* The second page contains the "struct lguest_ro_state", and is
 	 * read-only. */
-	pte[i+1].pfn = page_to_pfn(switcher_page[i+1]);
-	pte[i+1].flags = _PAGE_PRESENT|_PAGE_ACCESSED;
+	pte[i+1] = pfn_pte(page_to_pfn(switcher_page[i+1]),
+			   __pgprot(_PAGE_PRESENT|_PAGE_ACCESSED));
 }
 
 /*H:510 At boot or module load time, init_pagetables() allocates and populates
@@ -662,7 +684,7 @@ __init int init_pagetables(struct page **switcher_page, unsigned int pages)
 	unsigned int i;
 
 	for_each_possible_cpu(i) {
-		switcher_pte_page(i) = (spte_t *)get_zeroed_page(GFP_KERNEL);
+		switcher_pte_page(i) = (pte_t *)get_zeroed_page(GFP_KERNEL);
 		if (!switcher_pte_page(i)) {
 			free_switcher_pte_pages();
 			return -ENOMEM;

drivers/lguest/segments.c

@@ -73,14 +73,14 @@ static void fixup_gdt_table(struct lguest *lg, unsigned start, unsigned end)
 		/* Segment descriptors contain a privilege level: the Guest is
 		 * sometimes careless and leaves this as 0, even though it's
 		 * running at privilege level 1.  If so, we fix it here. */
-		if ((lg->gdt[i].b & 0x00006000) == 0)
-			lg->gdt[i].b |= (GUEST_PL << 13);
+		if ((lg->arch.gdt[i].b & 0x00006000) == 0)
+			lg->arch.gdt[i].b |= (GUEST_PL << 13);
 
 		/* Each descriptor has an "accessed" bit.  If we don't set it
 		 * now, the CPU will try to set it when the Guest first loads
 		 * that entry into a segment register.  But the GDT isn't
 		 * writable by the Guest, so bad things can happen. */
-		lg->gdt[i].b |= 0x00000100;
+		lg->arch.gdt[i].b |= 0x00000100;
 	}
 }
 
@@ -106,12 +106,12 @@ void setup_default_gdt_entries(struct lguest_ro_state *state)
 void setup_guest_gdt(struct lguest *lg)
 {
 	/* Start with full 0-4G segments... */
-	lg->gdt[GDT_ENTRY_KERNEL_CS] = FULL_EXEC_SEGMENT;
-	lg->gdt[GDT_ENTRY_KERNEL_DS] = FULL_SEGMENT;
+	lg->arch.gdt[GDT_ENTRY_KERNEL_CS] = FULL_EXEC_SEGMENT;
+	lg->arch.gdt[GDT_ENTRY_KERNEL_DS] = FULL_SEGMENT;
 	/* ...except the Guest is allowed to use them, so set the privilege
 	 * level appropriately in the flags. */
-	lg->gdt[GDT_ENTRY_KERNEL_CS].b |= (GUEST_PL << 13);
-	lg->gdt[GDT_ENTRY_KERNEL_DS].b |= (GUEST_PL << 13);
+	lg->arch.gdt[GDT_ENTRY_KERNEL_CS].b |= (GUEST_PL << 13);
+	lg->arch.gdt[GDT_ENTRY_KERNEL_DS].b |= (GUEST_PL << 13);
 }
 
 /* Like the IDT, we never simply use the GDT the Guest gives us.  We set up the
@@ -126,7 +126,7 @@ void copy_gdt_tls(const struct lguest *lg, struct desc_struct *gdt)
 	unsigned int i;
 
 	for (i = GDT_ENTRY_TLS_MIN; i <= GDT_ENTRY_TLS_MAX; i++)
-		gdt[i] = lg->gdt[i];
+		gdt[i] = lg->arch.gdt[i];
 }
 
 /* This is the full version */
@@ -138,7 +138,7 @@ void copy_gdt(const struct lguest *lg, struct desc_struct *gdt)
 	 * replaced.  See ignored_gdt() above. */
 	for (i = 0; i < GDT_ENTRIES; i++)
 		if (!ignored_gdt(i))
-			gdt[i] = lg->gdt[i];
+			gdt[i] = lg->arch.gdt[i];
 }
 
 /* This is where the Guest asks us to load a new GDT (LHCALL_LOAD_GDT). */
@@ -146,12 +146,12 @@ void load_guest_gdt(struct lguest *lg, unsigned long table, u32 num)
 {
 	/* We assume the Guest has the same number of GDT entries as the
 	 * Host, otherwise we'd have to dynamically allocate the Guest GDT. */
-	if (num > ARRAY_SIZE(lg->gdt))
+	if (num > ARRAY_SIZE(lg->arch.gdt))
 		kill_guest(lg, "too many gdt entries %i", num);
 
 	/* We read the whole thing in, then fix it up. */
-	lgread(lg, lg->gdt, table, num * sizeof(lg->gdt[0]));
-	fixup_gdt_table(lg, 0, ARRAY_SIZE(lg->gdt));
+	__lgread(lg, lg->arch.gdt, table, num * sizeof(lg->arch.gdt[0]));
+	fixup_gdt_table(lg, 0, ARRAY_SIZE(lg->arch.gdt));
 	/* Mark that the GDT changed so the core knows it has to copy it again,
 	 * even if the Guest is run on the same CPU. */
 	lg->changed |= CHANGED_GDT;
@@ -159,9 +159,9 @@ void load_guest_gdt(struct lguest *lg, unsigned long table, u32 num)
 
 void guest_load_tls(struct lguest *lg, unsigned long gtls)
 {
-	struct desc_struct *tls = &lg->gdt[GDT_ENTRY_TLS_MIN];
+	struct desc_struct *tls = &lg->arch.gdt[GDT_ENTRY_TLS_MIN];
 
-	lgread(lg, tls, gtls, sizeof(*tls)*GDT_ENTRY_TLS_ENTRIES);
+	__lgread(lg, tls, gtls, sizeof(*tls)*GDT_ENTRY_TLS_ENTRIES);
 	fixup_gdt_table(lg, GDT_ENTRY_TLS_MIN, GDT_ENTRY_TLS_MAX+1);
 	lg->changed |= CHANGED_GDT_TLS;
 }

drivers/lguest/x86/core.c

@@ -0,0 +1,577 @@
+/*
+ * Copyright (C) 2006, Rusty Russell <rusty@rustcorp.com.au> IBM Corporation.
+ * Copyright (C) 2007, Jes Sorensen <jes@sgi.com> SGI.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or
+ * NON INFRINGEMENT.  See the GNU General Public License for more
+ * details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
+ */
+#include <linux/kernel.h>
+#include <linux/start_kernel.h>
+#include <linux/string.h>
+#include <linux/console.h>
+#include <linux/screen_info.h>
+#include <linux/irq.h>
+#include <linux/interrupt.h>
+#include <linux/clocksource.h>
+#include <linux/clockchips.h>
+#include <linux/cpu.h>
+#include <linux/lguest.h>
+#include <linux/lguest_launcher.h>
+#include <asm/paravirt.h>
+#include <asm/param.h>
+#include <asm/page.h>
+#include <asm/pgtable.h>
+#include <asm/desc.h>
+#include <asm/setup.h>
+#include <asm/lguest.h>
+#include <asm/uaccess.h>
+#include <asm/i387.h>
+#include "../lg.h"
+
+static int cpu_had_pge;
+
+static struct {
+	unsigned long offset;
+	unsigned short segment;
+} lguest_entry;
+
+/* Offset from where switcher.S was compiled to where we've copied it */
+static unsigned long switcher_offset(void)
+{
+	return SWITCHER_ADDR - (unsigned long)start_switcher_text;
+}
+
+/* This cpu's struct lguest_pages. */
+static struct lguest_pages *lguest_pages(unsigned int cpu)
+{
+	return &(((struct lguest_pages *)
+		  (SWITCHER_ADDR + SHARED_SWITCHER_PAGES*PAGE_SIZE))[cpu]);
+}
+
+static DEFINE_PER_CPU(struct lguest *, last_guest);
+
+/*S:010
+ * We are getting close to the Switcher.
+ *
+ * Remember that each CPU has two pages which are visible to the Guest when it
+ * runs on that CPU.  This has to contain the state for that Guest: we copy the
+ * state in just before we run the Guest.
+ *
+ * Each Guest has "changed" flags which indicate what has changed in the Guest
+ * since it last ran.  We saw this set in interrupts_and_traps.c and
+ * segments.c.
+ */
+static void copy_in_guest_info(struct lguest *lg, struct lguest_pages *pages)
+{
+	/* Copying all this data can be quite expensive.  We usually run the
+	 * same Guest we ran last time (and that Guest hasn't run anywhere else
+	 * meanwhile).  If that's not the case, we pretend everything in the
+	 * Guest has changed. */
+	if (__get_cpu_var(last_guest) != lg || lg->last_pages != pages) {
+		__get_cpu_var(last_guest) = lg;
+		lg->last_pages = pages;
+		lg->changed = CHANGED_ALL;
+	}
+
+	/* These copies are pretty cheap, so we do them unconditionally: */
+	/* Save the current Host top-level page directory. */
+	pages->state.host_cr3 = __pa(current->mm->pgd);
+	/* Set up the Guest's page tables to see this CPU's pages (and no
+	 * other CPU's pages). */
+	map_switcher_in_guest(lg, pages);
+	/* Set up the two "TSS" members which tell the CPU what stack to use
+	 * for traps which do directly into the Guest (ie. traps at privilege
+	 * level 1). */
+	pages->state.guest_tss.esp1 = lg->esp1;
+	pages->state.guest_tss.ss1 = lg->ss1;
+
+	/* Copy direct-to-Guest trap entries. */
+	if (lg->changed & CHANGED_IDT)
+		copy_traps(lg, pages->state.guest_idt, default_idt_entries);
+
+	/* Copy all GDT entries which the Guest can change. */
+	if (lg->changed & CHANGED_GDT)
+		copy_gdt(lg, pages->state.guest_gdt);
+	/* If only the TLS entries have changed, copy them. */
+	else if (lg->changed & CHANGED_GDT_TLS)
+		copy_gdt_tls(lg, pages->state.guest_gdt);
+
+	/* Mark the Guest as unchanged for next time. */
+	lg->changed = 0;
+}
+
+/* Finally: the code to actually call into the Switcher to run the Guest. */
+static void run_guest_once(struct lguest *lg, struct lguest_pages *pages)
+{
+	/* This is a dummy value we need for GCC's sake. */
+	unsigned int clobber;
+
+	/* Copy the guest-specific information into this CPU's "struct
+	 * lguest_pages". */
+	copy_in_guest_info(lg, pages);
+
+	/* Set the trap number to 256 (impossible value).  If we fault while
+	 * switching to the Guest (bad segment registers or bug), this will
+	 * cause us to abort the Guest. */
+	lg->regs->trapnum = 256;
+
+	/* Now: we push the "eflags" register on the stack, then do an "lcall".
+	 * This is how we change from using the kernel code segment to using
+	 * the dedicated lguest code segment, as well as jumping into the
+	 * Switcher.
+	 *
+	 * The lcall also pushes the old code segment (KERNEL_CS) onto the
+	 * stack, then the address of this call.  This stack layout happens to
+	 * exactly match the stack of an interrupt... */
+	asm volatile("pushf; lcall *lguest_entry"
+		     /* This is how we tell GCC that %eax ("a") and %ebx ("b")
+		      * are changed by this routine.  The "=" means output. */
+		     : "=a"(clobber), "=b"(clobber)
+		     /* %eax contains the pages pointer.  ("0" refers to the
+		      * 0-th argument above, ie "a").  %ebx contains the
+		      * physical address of the Guest's top-level page
+		      * directory. */
+		     : "0"(pages), "1"(__pa(lg->pgdirs[lg->pgdidx].pgdir))
+		     /* We tell gcc that all these registers could change,
+		      * which means we don't have to save and restore them in
+		      * the Switcher. */
+		     : "memory", "%edx", "%ecx", "%edi", "%esi");
+}
+/*:*/
+
+/*H:040 This is the i386-specific code to setup and run the Guest.  Interrupts
+ * are disabled: we own the CPU. */
+void lguest_arch_run_guest(struct lguest *lg)
+{
+	/* Remember the awfully-named TS bit?  If the Guest has asked
+	 * to set it we set it now, so we can trap and pass that trap
+	 * to the Guest if it uses the FPU. */
+	if (lg->ts)
+		lguest_set_ts();
+
+	/* SYSENTER is an optimized way of doing system calls.  We
+	 * can't allow it because it always jumps to privilege level 0.
+	 * A normal Guest won't try it because we don't advertise it in
+	 * CPUID, but a malicious Guest (or malicious Guest userspace
+	 * program) could, so we tell the CPU to disable it before
+	 * running the Guest. */
+	if (boot_cpu_has(X86_FEATURE_SEP))
+		wrmsr(MSR_IA32_SYSENTER_CS, 0, 0);
+
+	/* Now we actually run the Guest.  It will pop back out when
+	 * something interesting happens, and we can examine its
+	 * registers to see what it was doing. */
+	run_guest_once(lg, lguest_pages(raw_smp_processor_id()));
+
+	/* The "regs" pointer contains two extra entries which are not
+	 * really registers: a trap number which says what interrupt or
+	 * trap made the switcher code come back, and an error code
+	 * which some traps set.  */
+
+	/* If the Guest page faulted, then the cr2 register will tell
+	 * us the bad virtual address.  We have to grab this now,
+	 * because once we re-enable interrupts an interrupt could
+	 * fault and thus overwrite cr2, or we could even move off to a
+	 * different CPU. */
+	if (lg->regs->trapnum == 14)
+		lg->arch.last_pagefault = read_cr2();
+	/* Similarly, if we took a trap because the Guest used the FPU,
+	 * we have to restore the FPU it expects to see. */
+	else if (lg->regs->trapnum == 7)
+		math_state_restore();
+
+	/* Restore SYSENTER if it's supposed to be on. */
+	if (boot_cpu_has(X86_FEATURE_SEP))
+		wrmsr(MSR_IA32_SYSENTER_CS, __KERNEL_CS, 0);
+}
+
+/*H:130 Our Guest is usually so well behaved; it never tries to do things it
+ * isn't allowed to.  Unfortunately, Linux's paravirtual infrastructure isn't
+ * quite complete, because it doesn't contain replacements for the Intel I/O
+ * instructions.  As a result, the Guest sometimes fumbles across one during
+ * the boot process as it probes for various things which are usually attached
+ * to a PC.
+ *
+ * When the Guest uses one of these instructions, we get trap #13 (General
+ * Protection Fault) and come here.  We see if it's one of those troublesome
+ * instructions and skip over it.  We return true if we did. */
+static int emulate_insn(struct lguest *lg)
+{
+	u8 insn;
+	unsigned int insnlen = 0, in = 0, shift = 0;
+	/* The eip contains the *virtual* address of the Guest's instruction:
+	 * guest_pa just subtracts the Guest's page_offset. */
+	unsigned long physaddr = guest_pa(lg, lg->regs->eip);
+
+	/* This must be the Guest kernel trying to do something, not userspace!
+	 * The bottom two bits of the CS segment register are the privilege
+	 * level. */
+	if ((lg->regs->cs & 3) != GUEST_PL)
+		return 0;
+
+	/* Decoding x86 instructions is icky. */
+	insn = lgread(lg, physaddr, u8);
+
+	/* 0x66 is an "operand prefix".  It means it's using the upper 16 bits
+	   of the eax register. */
+	if (insn == 0x66) {
+		shift = 16;
+		/* The instruction is 1 byte so far, read the next byte. */
+		insnlen = 1;
+		insn = lgread(lg, physaddr + insnlen, u8);
+	}
+
+	/* We can ignore the lower bit for the moment and decode the 4 opcodes
+	 * we need to emulate. */
+	switch (insn & 0xFE) {
+	case 0xE4: /* in     <next byte>,%al */
+		insnlen += 2;
+		in = 1;
+		break;
+	case 0xEC: /* in     (%dx),%al */
+		insnlen += 1;
+		in = 1;
+		break;
+	case 0xE6: /* out    %al,<next byte> */
+		insnlen += 2;
+		break;
+	case 0xEE: /* out    %al,(%dx) */
+		insnlen += 1;
+		break;
+	default:
+		/* OK, we don't know what this is, can't emulate. */
+		return 0;
+	}
+
+	/* If it was an "IN" instruction, they expect the result to be read
+	 * into %eax, so we change %eax.  We always return all-ones, which
+	 * traditionally means "there's nothing there". */
+	if (in) {
+		/* Lower bit tells is whether it's a 16 or 32 bit access */
+		if (insn & 0x1)
+			lg->regs->eax = 0xFFFFFFFF;
+		else
+			lg->regs->eax |= (0xFFFF << shift);
+	}
+	/* Finally, we've "done" the instruction, so move past it. */
+	lg->regs->eip += insnlen;
+	/* Success! */
+	return 1;
+}
+
+/*H:050 Once we've re-enabled interrupts, we look at why the Guest exited. */
+void lguest_arch_handle_trap(struct lguest *lg)
+{
+	switch (lg->regs->trapnum) {
+	case 13: /* We've intercepted a GPF. */
+		 /* Check if this was one of those annoying IN or OUT
+		  * instructions which we need to emulate.  If so, we
+		  * just go back into the Guest after we've done it. */
+		if (lg->regs->errcode == 0) {
+			if (emulate_insn(lg))
+				return;
+		}
+		break;
+	case 14: /* We've intercepted a page fault. */
+		 /* The Guest accessed a virtual address that wasn't
+		  * mapped.  This happens a lot: we don't actually set
+		  * up most of the page tables for the Guest at all when
+		  * we start: as it runs it asks for more and more, and
+		  * we set them up as required. In this case, we don't
+		  * even tell the Guest that the fault happened.
+		  *
+		  * The errcode tells whether this was a read or a
+		  * write, and whether kernel or userspace code. */
+		if (demand_page(lg, lg->arch.last_pagefault, lg->regs->errcode))
+			return;
+
+		 /* OK, it's really not there (or not OK): the Guest
+		  * needs to know.  We write out the cr2 value so it
+		  * knows where the fault occurred.
+		  *
+		  * Note that if the Guest were really messed up, this
+		  * could happen before it's done the INITIALIZE
+		  * hypercall, so lg->lguest_data will be NULL */
+		if (lg->lguest_data &&
+		    put_user(lg->arch.last_pagefault, &lg->lguest_data->cr2))
+			kill_guest(lg, "Writing cr2");
+		break;
+	case 7: /* We've intercepted a Device Not Available fault. */
+		/* If the Guest doesn't want to know, we already
+		 * restored the Floating Point Unit, so we just
+		 * continue without telling it. */
+		if (!lg->ts)
+			return;
+		break;
+	case 32 ... 255:
+		/* These values mean a real interrupt occurred, in which case
+		 * the Host handler has already been run.  We just do a
+		 * friendly check if another process should now be run, then
+		 * return to run the Guest again */
+		cond_resched();
+		return;
+	case LGUEST_TRAP_ENTRY:
+		/* Our 'struct hcall_args' maps directly over our regs: we set
+		 * up the pointer now to indicate a hypercall is pending. */
+		lg->hcall = (struct hcall_args *)lg->regs;
+		return;
+	}
+
+	/* We didn't handle the trap, so it needs to go to the Guest. */
+	if (!deliver_trap(lg, lg->regs->trapnum))
+		/* If the Guest doesn't have a handler (either it hasn't
+		 * registered any yet, or it's one of the faults we don't let
+		 * it handle), it dies with a cryptic error message. */
+		kill_guest(lg, "unhandled trap %li at %#lx (%#lx)",
+			   lg->regs->trapnum, lg->regs->eip,
+			   lg->regs->trapnum == 14 ? lg->arch.last_pagefault
+			   : lg->regs->errcode);
+}
+
+/* Now we can look at each of the routines this calls, in increasing order of
+ * complexity: do_hypercalls(), emulate_insn(), maybe_do_interrupt(),
+ * deliver_trap() and demand_page().  After all those, we'll be ready to
+ * examine the Switcher, and our philosophical understanding of the Host/Guest
+ * duality will be complete. :*/
+static void adjust_pge(void *on)
+{
+	if (on)
+		write_cr4(read_cr4() | X86_CR4_PGE);
+	else
+		write_cr4(read_cr4() & ~X86_CR4_PGE);
+}
+
+/*H:020 Now the Switcher is mapped and every thing else is ready, we need to do
+ * some more i386-specific initialization. */
+void __init lguest_arch_host_init(void)
+{
+	int i;
+
+	/* Most of the i386/switcher.S doesn't care that it's been moved; on
+	 * Intel, jumps are relative, and it doesn't access any references to
+	 * external code or data.
+	 *
+	 * The only exception is the interrupt handlers in switcher.S: their
+	 * addresses are placed in a table (default_idt_entries), so we need to
+	 * update the table with the new addresses.  switcher_offset() is a
+	 * convenience function which returns the distance between the builtin
+	 * switcher code and the high-mapped copy we just made. */
+	for (i = 0; i < IDT_ENTRIES; i++)
+		default_idt_entries[i] += switcher_offset();
+
+	/*
+	 * Set up the Switcher's per-cpu areas.
+	 *
+	 * Each CPU gets two pages of its own within the high-mapped region
+	 * (aka. "struct lguest_pages").  Much of this can be initialized now,
+	 * but some depends on what Guest we are running (which is set up in
+	 * copy_in_guest_info()).
+	 */
+	for_each_possible_cpu(i) {
+		/* lguest_pages() returns this CPU's two pages. */
+		struct lguest_pages *pages = lguest_pages(i);
+		/* This is a convenience pointer to make the code fit one
+		 * statement to a line. */
+		struct lguest_ro_state *state = &pages->state;
+
+		/* The Global Descriptor Table: the Host has a different one
+		 * for each CPU.  We keep a descriptor for the GDT which says
+		 * where it is and how big it is (the size is actually the last
+		 * byte, not the size, hence the "-1"). */
+		state->host_gdt_desc.size = GDT_SIZE-1;
+		state->host_gdt_desc.address = (long)get_cpu_gdt_table(i);
+
+		/* All CPUs on the Host use the same Interrupt Descriptor
+		 * Table, so we just use store_idt(), which gets this CPU's IDT
+		 * descriptor. */
+		store_idt(&state->host_idt_desc);
+
+		/* The descriptors for the Guest's GDT and IDT can be filled
+		 * out now, too.  We copy the GDT & IDT into ->guest_gdt and
+		 * ->guest_idt before actually running the Guest. */
+		state->guest_idt_desc.size = sizeof(state->guest_idt)-1;
+		state->guest_idt_desc.address = (long)&state->guest_idt;
+		state->guest_gdt_desc.size = sizeof(state->guest_gdt)-1;
+		state->guest_gdt_desc.address = (long)&state->guest_gdt;
+
+		/* We know where we want the stack to be when the Guest enters
+		 * the switcher: in pages->regs.  The stack grows upwards, so
+		 * we start it at the end of that structure. */
+		state->guest_tss.esp0 = (long)(&pages->regs + 1);
+		/* And this is the GDT entry to use for the stack: we keep a
+		 * couple of special LGUEST entries. */
+		state->guest_tss.ss0 = LGUEST_DS;
+
+		/* x86 can have a finegrained bitmap which indicates what I/O
+		 * ports the process can use.  We set it to the end of our
+		 * structure, meaning "none". */
+		state->guest_tss.io_bitmap_base = sizeof(state->guest_tss);
+
+		/* Some GDT entries are the same across all Guests, so we can
+		 * set them up now. */
+		setup_default_gdt_entries(state);
+		/* Most IDT entries are the same for all Guests, too.*/
+		setup_default_idt_entries(state, default_idt_entries);
+
+		/* The Host needs to be able to use the LGUEST segments on this
+		 * CPU, too, so put them in the Host GDT. */
+		get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_CS] = FULL_EXEC_SEGMENT;
+		get_cpu_gdt_table(i)[GDT_ENTRY_LGUEST_DS] = FULL_SEGMENT;
+	}
+
+	/* In the Switcher, we want the %cs segment register to use the
+	 * LGUEST_CS GDT entry: we've put that in the Host and Guest GDTs, so
+	 * it will be undisturbed when we switch.  To change %cs and jump we
+	 * need this structure to feed to Intel's "lcall" instruction. */
+	lguest_entry.offset = (long)switch_to_guest + switcher_offset();
+	lguest_entry.segment = LGUEST_CS;
+
+	/* Finally, we need to turn off "Page Global Enable".  PGE is an
+	 * optimization where page table entries are specially marked to show
+	 * they never change.  The Host kernel marks all the kernel pages this
+	 * way because it's always present, even when userspace is running.
+	 *
+	 * Lguest breaks this: unbeknownst to the rest of the Host kernel, we
+	 * switch to the Guest kernel.  If you don't disable this on all CPUs,
+	 * you'll get really weird bugs that you'll chase for two days.
+	 *
+	 * I used to turn PGE off every time we switched to the Guest and back
+	 * on when we return, but that slowed the Switcher down noticibly. */
+
+	/* We don't need the complexity of CPUs coming and going while we're
+	 * doing this. */
+	lock_cpu_hotplug();
+	if (cpu_has_pge) { /* We have a broader idea of "global". */
+		/* Remember that this was originally set (for cleanup). */
+		cpu_had_pge = 1;
+		/* adjust_pge is a helper function which sets or unsets the PGE
+		 * bit on its CPU, depending on the argument (0 == unset). */
+		on_each_cpu(adjust_pge, (void *)0, 0, 1);
+		/* Turn off the feature in the global feature set. */
+		clear_bit(X86_FEATURE_PGE, boot_cpu_data.x86_capability);
+	}
+	unlock_cpu_hotplug();
+};
+/*:*/
+
+void __exit lguest_arch_host_fini(void)
+{
+	/* If we had PGE before we started, turn it back on now. */
+	lock_cpu_hotplug();
+	if (cpu_had_pge) {
+		set_bit(X86_FEATURE_PGE, boot_cpu_data.x86_capability);
+		/* adjust_pge's argument "1" means set PGE. */
+		on_each_cpu(adjust_pge, (void *)1, 0, 1);
+	}
+	unlock_cpu_hotplug();
+}
+
+
+/*H:122 The i386-specific hypercalls simply farm out to the right functions. */
+int lguest_arch_do_hcall(struct lguest *lg, struct hcall_args *args)
+{
+	switch (args->arg0) {
+	case LHCALL_LOAD_GDT:
+		load_guest_gdt(lg, args->arg1, args->arg2);
+		break;
+	case LHCALL_LOAD_IDT_ENTRY:
+		load_guest_idt_entry(lg, args->arg1, args->arg2, args->arg3);
+		break;
+	case LHCALL_LOAD_TLS:
+		guest_load_tls(lg, args->arg1);
+		break;
+	default:
+		/* Bad Guest.  Bad! */
+		return -EIO;
+	}
+	return 0;
+}
+
+/*H:126 i386-specific hypercall initialization: */
+int lguest_arch_init_hypercalls(struct lguest *lg)
+{
+	u32 tsc_speed;
+
+	/* The pointer to the Guest's "struct lguest_data" is the only
+	 * argument.  We check that address now. */
+	if (!lguest_address_ok(lg, lg->hcall->arg1, sizeof(*lg->lguest_data)))
+		return -EFAULT;
+
+	/* Having checked it, we simply set lg->lguest_data to point straight
+	 * into the Launcher's memory at the right place and then use
+	 * copy_to_user/from_user from now on, instead of lgread/write.  I put
+	 * this in to show that I'm not immune to writing stupid
+	 * optimizations. */
+	lg->lguest_data = lg->mem_base + lg->hcall->arg1;
+
+	/* We insist that the Time Stamp Counter exist and doesn't change with
+	 * cpu frequency.  Some devious chip manufacturers decided that TSC
+	 * changes could be handled in software.  I decided that time going
+	 * backwards might be good for benchmarks, but it's bad for users.
+	 *
+	 * We also insist that the TSC be stable: the kernel detects unreliable
+	 * TSCs for its own purposes, and we use that here. */
+	if (boot_cpu_has(X86_FEATURE_CONSTANT_TSC) && !check_tsc_unstable())
+		tsc_speed = tsc_khz;
+	else
+		tsc_speed = 0;
+	if (put_user(tsc_speed, &lg->lguest_data->tsc_khz))
+		return -EFAULT;
+
+	/* The interrupt code might not like the system call vector. */
+	if (!check_syscall_vector(lg))
+		kill_guest(lg, "bad syscall vector");
+
+	return 0;
+}
+/* Now we've examined the hypercall code; our Guest can make requests.  There
+ * is one other way we can do things for the Guest, as we see in
+ * emulate_insn(). :*/
+
+/*L:030 lguest_arch_setup_regs()
+ *
+ * Most of the Guest's registers are left alone: we used get_zeroed_page() to
+ * allocate the structure, so they will be 0. */
+void lguest_arch_setup_regs(struct lguest *lg, unsigned long start)
+{
+	struct lguest_regs *regs = lg->regs;
+
+	/* There are four "segment" registers which the Guest needs to boot:
+	 * The "code segment" register (cs) refers to the kernel code segment
+	 * __KERNEL_CS, and the "data", "extra" and "stack" segment registers
+	 * refer to the kernel data segment __KERNEL_DS.
+	 *
+	 * The privilege level is packed into the lower bits.  The Guest runs
+	 * at privilege level 1 (GUEST_PL).*/
+	regs->ds = regs->es = regs->ss = __KERNEL_DS|GUEST_PL;
+	regs->cs = __KERNEL_CS|GUEST_PL;
+
+	/* The "eflags" register contains miscellaneous flags.  Bit 1 (0x002)
+	 * is supposed to always be "1".  Bit 9 (0x200) controls whether
+	 * interrupts are enabled.  We always leave interrupts enabled while
+	 * running the Guest. */
+	regs->eflags = 0x202;
+
+	/* The "Extended Instruction Pointer" register says where the Guest is
+	 * running. */
+	regs->eip = start;
+
+	/* %esi points to our boot information, at physical address 0, so don't
+	 * touch it. */
+	/* There are a couple of GDT entries the Guest expects when first
+	 * booting. */
+
+	setup_guest_gdt(lg);
+}

drivers/lguest/switcher.S → drivers/lguest/x86/switcher_32.S

@@ -48,7 +48,8 @@
 #include <linux/linkage.h>
 #include <asm/asm-offsets.h>
 #include <asm/page.h>
-#include "lg.h"
+#include <asm/segment.h>
+#include <asm/lguest.h>
 
 // We mark the start of the code to copy
 // It's placed in .text tho it's never run here
@@ -132,6 +133,7 @@ ENTRY(switch_to_guest)
 	// The Guest's register page has been mapped
 	// Writable onto our %esp (stack) --
 	// We can simply pop off all Guest regs.
+	popl	%eax
 	popl	%ebx
 	popl	%ecx
 	popl	%edx
@@ -139,7 +141,6 @@ ENTRY(switch_to_guest)
 	popl	%edi
 	popl	%ebp
 	popl	%gs
-	popl	%eax
 	popl	%fs
 	popl	%ds
 	popl	%es
@@ -167,7 +168,6 @@ ENTRY(switch_to_guest)
 	pushl	%es;							\
 	pushl	%ds;							\
 	pushl	%fs;							\
-	pushl	%eax;							\
 	pushl	%gs;							\
 	pushl	%ebp;							\
 	pushl	%edi;							\
@@ -175,6 +175,7 @@ ENTRY(switch_to_guest)
 	pushl	%edx;							\
 	pushl	%ecx;							\
 	pushl	%ebx;							\
+	pushl	%eax;							\
 	/* Our stack and our code are using segments			\
 	 * Set in the TSS and IDT					\
 	 * Yet if we were to touch data we'd use			\

drivers/net/Kconfig

@@ -3100,4 +3100,10 @@ config NETPOLL_TRAP
 config NET_POLL_CONTROLLER
 	def_bool NETPOLL
 
+config VIRTIO_NET
+	tristate "Virtio network driver (EXPERIMENTAL)"
+	depends on EXPERIMENTAL && VIRTIO
+	---help---
+	  This is the virtual network driver for lguest.  Say Y or M.
+
 endif # NETDEVICES

drivers/net/Makefile

@@ -183,7 +183,6 @@ obj-$(CONFIG_ZORRO8390) += zorro8390.o
 obj-$(CONFIG_HPLANCE) += hplance.o 7990.o
 obj-$(CONFIG_MVME147_NET) += mvme147.o 7990.o
 obj-$(CONFIG_EQUALIZER) += eql.o
-obj-$(CONFIG_LGUEST_NET) += lguest_net.o
 obj-$(CONFIG_MIPS_JAZZ_SONIC) += jazzsonic.o
 obj-$(CONFIG_MIPS_AU1X00_ENET) += au1000_eth.o
 obj-$(CONFIG_MIPS_SIM_NET) += mipsnet.o
@@ -243,3 +242,4 @@ obj-$(CONFIG_FS_ENET) += fs_enet/
 
 obj-$(CONFIG_NETXEN_NIC) += netxen/
 obj-$(CONFIG_NIU) += niu.o
+obj-$(CONFIG_VIRTIO_NET) += virtio_net.o

drivers/net/lguest_net.c

@@ -1,555 +0,0 @@
-/*D:500
- * The Guest network driver.
- *
- * This is very simple a virtual network driver, and our last Guest driver.
- * The only trick is that it can talk directly to multiple other recipients
- * (ie. other Guests on the same network).  It can also be used with only the
- * Host on the network.
- :*/
-
-/* Copyright 2006 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation
- *
- * This program is free software; you can redistribute it and/or modify
- * it under the terms of the GNU General Public License as published by
- * the Free Software Foundation; either version 2 of the License, or
- * (at your option) any later version.
- *
- * This program is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- * GNU General Public License for more details.
- *
- * You should have received a copy of the GNU General Public License
- * along with this program; if not, write to the Free Software
- * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
- */
-//#define DEBUG
-#include <linux/netdevice.h>
-#include <linux/etherdevice.h>
-#include <linux/module.h>
-#include <linux/mm_types.h>
-#include <linux/io.h>
-#include <linux/lguest_bus.h>
-
-#define SHARED_SIZE		PAGE_SIZE
-#define MAX_LANS		4
-#define NUM_SKBS		8
-
-/*M:011 Network code master Jeff Garzik points out numerous shortcomings in
- * this driver if it aspires to greatness.
- *
- * Firstly, it doesn't use "NAPI": the networking's New API, and is poorer for
- * it.  As he says "NAPI means system-wide load leveling, across multiple
- * network interfaces.  Lack of NAPI can mean competition at higher loads."
- *
- * He also points out that we don't implement set_mac_address, so users cannot
- * change the devices hardware address.  When I asked why one would want to:
- * "Bonding, and situations where you /do/ want the MAC address to "leak" out
- * of the host onto the wider net."
- *
- * Finally, he would like module unloading: "It is not unrealistic to think of
- * [un|re|]loading the net support module in an lguest guest.  And, adding
- * module support makes the programmer more responsible, because they now have
- * to learn to clean up after themselves.  Any driver that cannot clean up
- * after itself is an incomplete driver in my book."
- :*/
-
-/*D:530 The "struct lguestnet_info" contains all the information we need to
- * know about the network device. */
-struct lguestnet_info
-{
-	/* The mapped device page(s) (an array of "struct lguest_net"). */
-	struct lguest_net *peer;
-	/* The physical address of the device page(s) */
-	unsigned long peer_phys;
-	/* The size of the device page(s). */
-	unsigned long mapsize;
-
-	/* The lguest_device I come from */
-	struct lguest_device *lgdev;
-
-	/* My peerid (ie. my slot in the array). */
-	unsigned int me;
-
-	/* Receive queue: the network packets waiting to be filled. */
-	struct sk_buff *skb[NUM_SKBS];
-	struct lguest_dma dma[NUM_SKBS];
-};
-/*:*/
-
-/* How many bytes left in this page. */
-static unsigned int rest_of_page(void *data)
-{
-	return PAGE_SIZE - ((unsigned long)data % PAGE_SIZE);
-}
-
-/*D:570 Each peer (ie. Guest or Host) on the network binds their receive
- * buffers to a different key: we simply use the physical address of the
- * device's memory page plus the peer number.  The Host insists that all keys
- * be a multiple of 4, so we multiply the peer number by 4. */
-static unsigned long peer_key(struct lguestnet_info *info, unsigned peernum)
-{
-	return info->peer_phys + 4 * peernum;
-}
-
-/* This is the routine which sets up a "struct lguest_dma" to point to a
- * network packet, similar to req_to_dma() in lguest_blk.c.  The structure of a
- * "struct sk_buff" has grown complex over the years: it consists of a "head"
- * linear section pointed to by "skb->data", and possibly an array of
- * "fragments" in the case of a non-linear packet.
- *
- * Our receive buffers don't use fragments at all but outgoing skbs might, so
- * we handle it. */
-static void skb_to_dma(const struct sk_buff *skb, unsigned int headlen,
-		       struct lguest_dma *dma)
-{
-	unsigned int i, seg;
-
-	/* First, we put the linear region into the "struct lguest_dma".  Each
-	 * entry can't go over a page boundary, so even though all our packets
-	 * are 1514 bytes or less, we might need to use two entries here: */
-	for (i = seg = 0; i < headlen; seg++, i += rest_of_page(skb->data+i)) {
-		dma->addr[seg] = virt_to_phys(skb->data + i);
-		dma->len[seg] = min((unsigned)(headlen - i),
-				    rest_of_page(skb->data + i));
-	}
-
-	/* Now we handle the fragments: at least they're guaranteed not to go
-	 * over a page.  skb_shinfo(skb) returns a pointer to the structure
-	 * which tells us about the number of fragments and the fragment
-	 * array. */
-	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++, seg++) {
-		const skb_frag_t *f = &skb_shinfo(skb)->frags[i];
-		/* Should not happen with MTU less than 64k - 2 * PAGE_SIZE. */
-		if (seg == LGUEST_MAX_DMA_SECTIONS) {
-			/* We will end up sending a truncated packet should
-			 * this ever happen.  Plus, a cool log message! */
-			printk("Woah dude!  Megapacket!\n");
-			break;
-		}
-		dma->addr[seg] = page_to_phys(f->page) + f->page_offset;
-		dma->len[seg] = f->size;
-	}
-
-	/* If after all that we didn't use the entire "struct lguest_dma"
-	 * array, we terminate it with a 0 length. */
-	if (seg < LGUEST_MAX_DMA_SECTIONS)
-		dma->len[seg] = 0;
-}
-
-/*
- * Packet transmission.
- *
- * Our packet transmission is a little unusual.  A real network card would just
- * send out the packet and leave the receivers to decide if they're interested.
- * Instead, we look through the network device memory page and see if any of
- * the ethernet addresses match the packet destination, and if so we send it to
- * that Guest.
- *
- * This is made a little more complicated in two cases.  The first case is
- * broadcast packets: for that we send the packet to all Guests on the network,
- * one at a time.  The second case is "promiscuous" mode, where a Guest wants
- * to see all the packets on the network.  We need a way for the Guest to tell
- * us it wants to see all packets, so it sets the "multicast" bit on its
- * published MAC address, which is never valid in a real ethernet address.
- */
-#define PROMISC_BIT		0x01
-
-/* This is the callback which is summoned whenever the network device's
- * multicast or promiscuous state changes.  If the card is in promiscuous mode,
- * we advertise that in our ethernet address in the device's memory.  We do the
- * same if Linux wants any or all multicast traffic.  */
-static void lguestnet_set_multicast(struct net_device *dev)
-{
-	struct lguestnet_info *info = netdev_priv(dev);
-
-	if ((dev->flags & (IFF_PROMISC|IFF_ALLMULTI)) || dev->mc_count)
-		info->peer[info->me].mac[0] |= PROMISC_BIT;
-	else
-		info->peer[info->me].mac[0] &= ~PROMISC_BIT;
-}
-
-/* A simple test function to see if a peer wants to see all packets.*/
-static int promisc(struct lguestnet_info *info, unsigned int peer)
-{
-	return info->peer[peer].mac[0] & PROMISC_BIT;
-}
-
-/* Another simple function to see if a peer's advertised ethernet address
- * matches a packet's destination ethernet address. */
-static int mac_eq(const unsigned char mac[ETH_ALEN],
-		  struct lguestnet_info *info, unsigned int peer)
-{
-	/* Ignore multicast bit, which peer turns on to mean promisc. */
-	if ((info->peer[peer].mac[0] & (~PROMISC_BIT)) != mac[0])
-		return 0;
-	return memcmp(mac+1, info->peer[peer].mac+1, ETH_ALEN-1) == 0;
-}
-
-/* This is the function which actually sends a packet once we've decided a
- * peer wants it: */
-static void transfer_packet(struct net_device *dev,
-			    struct sk_buff *skb,
-			    unsigned int peernum)
-{
-	struct lguestnet_info *info = netdev_priv(dev);
-	struct lguest_dma dma;
-
-	/* We use our handy "struct lguest_dma" packing function to prepare
-	 * the skb for sending. */
-	skb_to_dma(skb, skb_headlen(skb), &dma);
-	pr_debug("xfer length %04x (%u)\n", htons(skb->len), skb->len);
-
-	/* This is the actual send call which copies the packet. */
-	lguest_send_dma(peer_key(info, peernum), &dma);
-
-	/* Check that the entire packet was transmitted.  If not, it could mean
-	 * that the other Guest registered a short receive buffer, but this
-	 * driver should never do that.  More likely, the peer is dead. */
-	if (dma.used_len != skb->len) {
-		dev->stats.tx_carrier_errors++;
-		pr_debug("Bad xfer to peer %i: %i of %i (dma %p/%i)\n",
-			 peernum, dma.used_len, skb->len,
-			 (void *)dma.addr[0], dma.len[0]);
-	} else {
-		/* On success we update the stats. */
-		dev->stats.tx_bytes += skb->len;
-		dev->stats.tx_packets++;
-	}
-}
-
-/* Another helper function to tell is if a slot in the device memory is unused.
- * Since we always set the Local Assignment bit in the ethernet address, the
- * first byte can never be 0. */
-static int unused_peer(const struct lguest_net peer[], unsigned int num)
-{
-	return peer[num].mac[0] == 0;
-}
-
-/* Finally, here is the routine which handles an outgoing packet.  It's called
- * "start_xmit" for traditional reasons. */
-static int lguestnet_start_xmit(struct sk_buff *skb, struct net_device *dev)
-{
-	unsigned int i;
-	int broadcast;
-	struct lguestnet_info *info = netdev_priv(dev);
-	/* Extract the destination ethernet address from the packet. */
-	const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
-	DECLARE_MAC_BUF(mac);
-
-	pr_debug("%s: xmit %s\n", dev->name, print_mac(mac, dest));
-
-	/* If it's a multicast packet, we broadcast to everyone.  That's not
-	 * very efficient, but there are very few applications which actually
-	 * use multicast, which is a shame really.
-	 *
-	 * As etherdevice.h points out: "By definition the broadcast address is
-	 * also a multicast address."  So we don't have to test for broadcast
-	 * packets separately. */
-	broadcast = is_multicast_ether_addr(dest);
-
-	/* Look through all the published ethernet addresses to see if we
-	 * should send this packet. */
-	for (i = 0; i < info->mapsize/sizeof(struct lguest_net); i++) {
-		/* We don't send to ourselves (we actually can't SEND_DMA to
-		 * ourselves anyway), and don't send to unused slots.*/
-		if (i == info->me || unused_peer(info->peer, i))
-			continue;
-
-		/* If it's broadcast we send it.  If they want every packet we
-		 * send it.  If the destination matches their address we send
-		 * it.  Otherwise we go to the next peer. */
-		if (!broadcast && !promisc(info, i) && !mac_eq(dest, info, i))
-			continue;
-
-		pr_debug("lguestnet %s: sending from %i to %i\n",
-			 dev->name, info->me, i);
-		/* Our routine which actually does the transfer. */
-		transfer_packet(dev, skb, i);
-	}
-
-	/* An xmit routine is expected to dispose of the packet, so we do. */
-	dev_kfree_skb(skb);
-
-	/* As per kernel convention, 0 means success.  This is why I love
-	 * networking: even if we never sent to anyone, that's still
-	 * success! */
-	return 0;
-}
-
-/*D:560
- * Packet receiving.
- *
- * First, here's a helper routine which fills one of our array of receive
- * buffers: */
-static int fill_slot(struct net_device *dev, unsigned int slot)
-{
-	struct lguestnet_info *info = netdev_priv(dev);
-
-	/* We can receive ETH_DATA_LEN (1500) byte packets, plus a standard
-	 * ethernet header of ETH_HLEN (14) bytes. */
-	info->skb[slot] = netdev_alloc_skb(dev, ETH_HLEN + ETH_DATA_LEN);
-	if (!info->skb[slot]) {
-		printk("%s: could not fill slot %i\n", dev->name, slot);
-		return -ENOMEM;
-	}
-
-	/* skb_to_dma() is a helper which sets up the "struct lguest_dma" to
-	 * point to the data in the skb: we also use it for sending out a
-	 * packet. */
-	skb_to_dma(info->skb[slot], ETH_HLEN + ETH_DATA_LEN, &info->dma[slot]);
-
-	/* This is a Write Memory Barrier: it ensures that the entry in the
-	 * receive buffer array is written *before* we set the "used_len" entry
-	 * to 0.  If the Host were looking at the receive buffer array from a
-	 * different CPU, it could potentially see "used_len = 0" and not see
-	 * the updated receive buffer information.  This would be a horribly
-	 * nasty bug, so make sure the compiler and CPU know this has to happen
-	 * first. */
-	wmb();
-	/* Writing 0 to "used_len" tells the Host it can use this receive
-	 * buffer now. */
-	info->dma[slot].used_len = 0;
-	return 0;
-}
-
-/* This is the actual receive routine.  When we receive an interrupt from the
- * Host to tell us a packet has been delivered, we arrive here: */
-static irqreturn_t lguestnet_rcv(int irq, void *dev_id)
-{
-	struct net_device *dev = dev_id;
-	struct lguestnet_info *info = netdev_priv(dev);
-	unsigned int i, done = 0;
-
-	/* Look through our entire receive array for an entry which has data
-	 * in it. */
-	for (i = 0; i < ARRAY_SIZE(info->dma); i++) {
-		unsigned int length;
-		struct sk_buff *skb;
-
-		length = info->dma[i].used_len;
-		if (length == 0)
-			continue;
-
-		/* We've found one!  Remember the skb (we grabbed the length
-		 * above), and immediately refill the slot we've taken it
-		 * from. */
-		done++;
-		skb = info->skb[i];
-		fill_slot(dev, i);
-
-		/* This shouldn't happen: micropackets could be sent by a
-		 * badly-behaved Guest on the network, but the Host will never
-		 * stuff more data in the buffer than the buffer length. */
-		if (length < ETH_HLEN || length > ETH_HLEN + ETH_DATA_LEN) {
-			pr_debug(KERN_WARNING "%s: unbelievable skb len: %i\n",
-				 dev->name, length);
-			dev_kfree_skb(skb);
-			continue;
-		}
-
-		/* skb_put(), what a great function!  I've ranted about this
-		 * function before (http://lkml.org/lkml/1999/9/26/24).  You
-		 * call it after you've added data to the end of an skb (in
-		 * this case, it was the Host which wrote the data). */
-		skb_put(skb, length);
-
-		/* The ethernet header contains a protocol field: we use the
-		 * standard helper to extract it, and place the result in
-		 * skb->protocol.  The helper also sets up skb->pkt_type and
-		 * eats up the ethernet header from the front of the packet. */
-		skb->protocol = eth_type_trans(skb, dev);
-
-		/* If this device doesn't need checksums for sending, we also
-		 * don't need to check the packets when they come in. */
-		if (dev->features & NETIF_F_NO_CSUM)
-			skb->ip_summed = CHECKSUM_UNNECESSARY;
-
-		/* As a last resort for debugging the driver or the lguest I/O
-		 * subsystem, you can uncomment the "#define DEBUG" at the top
-		 * of this file, which turns all the pr_debug() into printk()
-		 * and floods the logs. */
-		pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
-			 ntohs(skb->protocol), skb->len, skb->pkt_type);
-
-		/* Update the packet and byte counts (visible from ifconfig,
-		 * and good for debugging). */
-		dev->stats.rx_bytes += skb->len;
-		dev->stats.rx_packets++;
-
-		/* Hand our fresh network packet into the stack's "network
-		 * interface receive" routine.  That will free the packet
-		 * itself when it's finished. */
-		netif_rx(skb);
-	}
-
-	/* If we found any packets, we assume the interrupt was for us. */
-	return done ? IRQ_HANDLED : IRQ_NONE;
-}
-
-/*D:550 This is where we start: when the device is brought up by dhcpd or
- * ifconfig.  At this point we advertise our MAC address to the rest of the
- * network, and register receive buffers ready for incoming packets. */
-static int lguestnet_open(struct net_device *dev)
-{
-	int i;
-	struct lguestnet_info *info = netdev_priv(dev);
-
-	/* Copy our MAC address into the device page, so others on the network
-	 * can find us. */
-	memcpy(info->peer[info->me].mac, dev->dev_addr, ETH_ALEN);
-
-	/* We might already be in promisc mode (dev->flags & IFF_PROMISC).  Our
-	 * set_multicast callback handles this already, so we call it now. */
-	lguestnet_set_multicast(dev);
-
-	/* Allocate packets and put them into our "struct lguest_dma" array.
-	 * If we fail to allocate all the packets we could still limp along,
-	 * but it's a sign of real stress so we should probably give up now. */
-	for (i = 0; i < ARRAY_SIZE(info->dma); i++) {
-		if (fill_slot(dev, i) != 0)
-			goto cleanup;
-	}
-
-	/* Finally we tell the Host where our array of "struct lguest_dma"
-	 * receive buffers is, binding it to the key corresponding to the
-	 * device's physical memory plus our peerid. */
-	if (lguest_bind_dma(peer_key(info,info->me), info->dma,
-			    NUM_SKBS, lgdev_irq(info->lgdev)) != 0)
-		goto cleanup;
-	return 0;
-
-cleanup:
-	while (--i >= 0)
-		dev_kfree_skb(info->skb[i]);
-	return -ENOMEM;
-}
-/*:*/
-
-/* The close routine is called when the device is no longer in use: we clean up
- * elegantly. */
-static int lguestnet_close(struct net_device *dev)
-{
-	unsigned int i;
-	struct lguestnet_info *info = netdev_priv(dev);
-
-	/* Clear all trace of our existence out of the device memory by setting
-	 * the slot which held our MAC address to 0 (unused). */
-	memset(&info->peer[info->me], 0, sizeof(info->peer[info->me]));
-
-	/* Unregister our array of receive buffers */
-	lguest_unbind_dma(peer_key(info, info->me), info->dma);
-	for (i = 0; i < ARRAY_SIZE(info->dma); i++)
-		dev_kfree_skb(info->skb[i]);
-	return 0;
-}
-
-/*D:510 The network device probe function is basically a standard ethernet
- * device setup.  It reads the "struct lguest_device_desc" and sets the "struct
- * net_device".  Oh, the line-by-line excitement!  Let's skip over it. :*/
-static int lguestnet_probe(struct lguest_device *lgdev)
-{
-	int err, irqf = IRQF_SHARED;
-	struct net_device *dev;
-	struct lguestnet_info *info;
-	struct lguest_device_desc *desc = &lguest_devices[lgdev->index];
-
-	pr_debug("lguest_net: probing for device %i\n", lgdev->index);
-
-	dev = alloc_etherdev(sizeof(struct lguestnet_info));
-	if (!dev)
-		return -ENOMEM;
-
-	/* Ethernet defaults with some changes */
-	ether_setup(dev);
-	dev->set_mac_address = NULL;
-
-	dev->dev_addr[0] = 0x02; /* set local assignment bit (IEEE802) */
-	dev->dev_addr[1] = 0x00;
-	memcpy(&dev->dev_addr[2], &lguest_data.guestid, 2);
-	dev->dev_addr[4] = 0x00;
-	dev->dev_addr[5] = 0x00;
-
-	dev->open = lguestnet_open;
-	dev->stop = lguestnet_close;
-	dev->hard_start_xmit = lguestnet_start_xmit;
-
-	/* We don't actually support multicast yet, but turning on/off
-	 * promisc also calls dev->set_multicast_list. */
-	dev->set_multicast_list = lguestnet_set_multicast;
-	SET_NETDEV_DEV(dev, &lgdev->dev);
-
-	/* The network code complains if you have "scatter-gather" capability
-	 * if you don't also handle checksums (it seem that would be
-	 * "illogical").  So we use a lie of omission and don't tell it that we
-	 * can handle scattered packets unless we also don't want checksums,
-	 * even though to us they're completely independent. */
-	if (desc->features & LGUEST_NET_F_NOCSUM)
-		dev->features = NETIF_F_SG|NETIF_F_NO_CSUM;
-
-	info = netdev_priv(dev);
-	info->mapsize = PAGE_SIZE * desc->num_pages;
-	info->peer_phys = ((unsigned long)desc->pfn << PAGE_SHIFT);
-	info->lgdev = lgdev;
-	info->peer = lguest_map(info->peer_phys, desc->num_pages);
-	if (!info->peer) {
-		err = -ENOMEM;
-		goto free;
-	}
-
-	/* This stores our peerid (upper bits reserved for future). */
-	info->me = (desc->features & (info->mapsize-1));
-
-	err = register_netdev(dev);
-	if (err) {
-		pr_debug("lguestnet: registering device failed\n");
-		goto unmap;
-	}
-
-	if (lguest_devices[lgdev->index].features & LGUEST_DEVICE_F_RANDOMNESS)
-		irqf |= IRQF_SAMPLE_RANDOM;
-	if (request_irq(lgdev_irq(lgdev), lguestnet_rcv, irqf, "lguestnet",
-			dev) != 0) {
-		pr_debug("lguestnet: cannot get irq %i\n", lgdev_irq(lgdev));
-		goto unregister;
-	}
-
-	pr_debug("lguestnet: registered device %s\n", dev->name);
-	/* Finally, we put the "struct net_device" in the generic "struct
-	 * lguest_device"s private pointer.  Again, it's not necessary, but
-	 * makes sure the cool kernel kids don't tease us. */
-	lgdev->private = dev;
-	return 0;
-
-unregister:
-	unregister_netdev(dev);
-unmap:
-	lguest_unmap(info->peer);
-free:
-	free_netdev(dev);
-	return err;
-}
-
-static struct lguest_driver lguestnet_drv = {
-	.name = "lguestnet",
-	.owner = THIS_MODULE,
-	.device_type = LGUEST_DEVICE_T_NET,
-	.probe = lguestnet_probe,
-};
-
-static __init int lguestnet_init(void)
-{
-	return register_lguest_driver(&lguestnet_drv);
-}
-module_init(lguestnet_init);
-
-MODULE_DESCRIPTION("Lguest network driver");
-MODULE_LICENSE("GPL");
-
-/*D:580
- * This is the last of the Drivers, and with this we have covered the many and
- * wonderous and fine (and boring) details of the Guest.
- *
- * "make Launcher" beckons, where we answer questions like "Where do Guests
- * come from?", and "What do you do when someone asks for optimization?"
- */

drivers/net/virtio_net.c

@@ -0,0 +1,435 @@
+/* A simple network driver using virtio.
+ *
+ * Copyright 2007 Rusty Russell <rusty@rustcorp.com.au> IBM Corporation
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ */
+//#define DEBUG
+#include <linux/netdevice.h>
+#include <linux/etherdevice.h>
+#include <linux/module.h>
+#include <linux/virtio.h>
+#include <linux/virtio_net.h>
+#include <linux/scatterlist.h>
+
+/* FIXME: MTU in config. */
+#define MAX_PACKET_LEN (ETH_HLEN+ETH_DATA_LEN)
+
+struct virtnet_info
+{
+	struct virtio_device *vdev;
+	struct virtqueue *rvq, *svq;
+	struct net_device *dev;
+	struct napi_struct napi;
+
+	/* Number of input buffers, and max we've ever had. */
+	unsigned int num, max;
+
+	/* Receive & send queues. */
+	struct sk_buff_head recv;
+	struct sk_buff_head send;
+};
+
+static inline struct virtio_net_hdr *skb_vnet_hdr(struct sk_buff *skb)
+{
+	return (struct virtio_net_hdr *)skb->cb;
+}
+
+static inline void vnet_hdr_to_sg(struct scatterlist *sg, struct sk_buff *skb)
+{
+	sg_init_one(sg, skb_vnet_hdr(skb), sizeof(struct virtio_net_hdr));
+}
+
+static bool skb_xmit_done(struct virtqueue *rvq)
+{
+	struct virtnet_info *vi = rvq->vdev->priv;
+
+	/* In case we were waiting for output buffers. */
+	netif_wake_queue(vi->dev);
+	return true;
+}
+
+static void receive_skb(struct net_device *dev, struct sk_buff *skb,
+			unsigned len)
+{
+	struct virtio_net_hdr *hdr = skb_vnet_hdr(skb);
+
+	if (unlikely(len < sizeof(struct virtio_net_hdr) + ETH_HLEN)) {
+		pr_debug("%s: short packet %i\n", dev->name, len);
+		dev->stats.rx_length_errors++;
+		goto drop;
+	}
+	len -= sizeof(struct virtio_net_hdr);
+	BUG_ON(len > MAX_PACKET_LEN);
+
+	skb_trim(skb, len);
+	skb->protocol = eth_type_trans(skb, dev);
+	pr_debug("Receiving skb proto 0x%04x len %i type %i\n",
+		 ntohs(skb->protocol), skb->len, skb->pkt_type);
+	dev->stats.rx_bytes += skb->len;
+	dev->stats.rx_packets++;
+
+	if (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {
+		pr_debug("Needs csum!\n");
+		skb->ip_summed = CHECKSUM_PARTIAL;
+		skb->csum_start = hdr->csum_start;
+		skb->csum_offset = hdr->csum_offset;
+		if (skb->csum_start > skb->len - 2
+		    || skb->csum_offset > skb->len - 2) {
+			if (net_ratelimit())
+				printk(KERN_WARNING "%s: csum=%u/%u len=%u\n",
+				       dev->name, skb->csum_start,
+				       skb->csum_offset, skb->len);
+			goto frame_err;
+		}
+	}
+
+	if (hdr->gso_type != VIRTIO_NET_HDR_GSO_NONE) {
+		pr_debug("GSO!\n");
+		switch (hdr->gso_type) {
+		case VIRTIO_NET_HDR_GSO_TCPV4:
+			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV4;
+			break;
+		case VIRTIO_NET_HDR_GSO_TCPV4_ECN:
+			skb_shinfo(skb)->gso_type = SKB_GSO_TCP_ECN;
+			break;
+		case VIRTIO_NET_HDR_GSO_UDP:
+			skb_shinfo(skb)->gso_type = SKB_GSO_UDP;
+			break;
+		case VIRTIO_NET_HDR_GSO_TCPV6:
+			skb_shinfo(skb)->gso_type = SKB_GSO_TCPV6;
+			break;
+		default:
+			if (net_ratelimit())
+				printk(KERN_WARNING "%s: bad gso type %u.\n",
+				       dev->name, hdr->gso_type);
+			goto frame_err;
+		}
+
+		skb_shinfo(skb)->gso_size = hdr->gso_size;
+		if (skb_shinfo(skb)->gso_size == 0) {
+			if (net_ratelimit())
+				printk(KERN_WARNING "%s: zero gso size.\n",
+				       dev->name);
+			goto frame_err;
+		}
+
+		/* Header must be checked, and gso_segs computed. */
+		skb_shinfo(skb)->gso_type |= SKB_GSO_DODGY;
+		skb_shinfo(skb)->gso_segs = 0;
+	}
+
+	netif_receive_skb(skb);
+	return;
+
+frame_err:
+	dev->stats.rx_frame_errors++;
+drop:
+	dev_kfree_skb(skb);
+}
+
+static void try_fill_recv(struct virtnet_info *vi)
+{
+	struct sk_buff *skb;
+	struct scatterlist sg[1+MAX_SKB_FRAGS];
+	int num, err;
+
+	for (;;) {
+		skb = netdev_alloc_skb(vi->dev, MAX_PACKET_LEN);
+		if (unlikely(!skb))
+			break;
+
+		skb_put(skb, MAX_PACKET_LEN);
+		vnet_hdr_to_sg(sg, skb);
+		num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
+		skb_queue_head(&vi->recv, skb);
+
+		err = vi->rvq->vq_ops->add_buf(vi->rvq, sg, 0, num, skb);
+		if (err) {
+			skb_unlink(skb, &vi->recv);
+			kfree_skb(skb);
+			break;
+		}
+		vi->num++;
+	}
+	if (unlikely(vi->num > vi->max))
+		vi->max = vi->num;
+	vi->rvq->vq_ops->kick(vi->rvq);
+}
+
+static bool skb_recv_done(struct virtqueue *rvq)
+{
+	struct virtnet_info *vi = rvq->vdev->priv;
+	netif_rx_schedule(vi->dev, &vi->napi);
+	/* Suppress further interrupts. */
+	return false;
+}
+
+static int virtnet_poll(struct napi_struct *napi, int budget)
+{
+	struct virtnet_info *vi = container_of(napi, struct virtnet_info, napi);
+	struct sk_buff *skb = NULL;
+	unsigned int len, received = 0;
+
+again:
+	while (received < budget &&
+	       (skb = vi->rvq->vq_ops->get_buf(vi->rvq, &len)) != NULL) {
+		__skb_unlink(skb, &vi->recv);
+		receive_skb(vi->dev, skb, len);
+		vi->num--;
+		received++;
+	}
+
+	/* FIXME: If we oom and completely run out of inbufs, we need
+	 * to start a timer trying to fill more. */
+	if (vi->num < vi->max / 2)
+		try_fill_recv(vi);
+
+	/* All done? */
+	if (!skb) {
+		netif_rx_complete(vi->dev, napi);
+		if (unlikely(!vi->rvq->vq_ops->restart(vi->rvq))
+		    && netif_rx_reschedule(vi->dev, napi))
+			goto again;
+	}
+
+	return received;
+}
+
+static void free_old_xmit_skbs(struct virtnet_info *vi)
+{
+	struct sk_buff *skb;
+	unsigned int len;
+
+	while ((skb = vi->svq->vq_ops->get_buf(vi->svq, &len)) != NULL) {
+		pr_debug("Sent skb %p\n", skb);
+		__skb_unlink(skb, &vi->send);
+		vi->dev->stats.tx_bytes += len;
+		vi->dev->stats.tx_packets++;
+		kfree_skb(skb);
+	}
+}
+
+static int start_xmit(struct sk_buff *skb, struct net_device *dev)
+{
+	struct virtnet_info *vi = netdev_priv(dev);
+	int num, err;
+	struct scatterlist sg[1+MAX_SKB_FRAGS];
+	struct virtio_net_hdr *hdr;
+	const unsigned char *dest = ((struct ethhdr *)skb->data)->h_dest;
+	DECLARE_MAC_BUF(mac);
+
+	pr_debug("%s: xmit %p %s\n", dev->name, skb, print_mac(mac, dest));
+
+	free_old_xmit_skbs(vi);
+
+	/* Encode metadata header at front. */
+	hdr = skb_vnet_hdr(skb);
+	if (skb->ip_summed == CHECKSUM_PARTIAL) {
+		hdr->flags = VIRTIO_NET_HDR_F_NEEDS_CSUM;
+		hdr->csum_start = skb->csum_start - skb_headroom(skb);
+		hdr->csum_offset = skb->csum_offset;
+	} else {
+		hdr->flags = 0;
+		hdr->csum_offset = hdr->csum_start = 0;
+	}
+
+	if (skb_is_gso(skb)) {
+		hdr->gso_size = skb_shinfo(skb)->gso_size;
+		if (skb_shinfo(skb)->gso_type & SKB_GSO_TCP_ECN)
+			hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4_ECN;
+		else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
+			hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV4;
+		else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
+			hdr->gso_type = VIRTIO_NET_HDR_GSO_TCPV6;
+		else if (skb_shinfo(skb)->gso_type & SKB_GSO_UDP)
+			hdr->gso_type = VIRTIO_NET_HDR_GSO_UDP;
+		else
+			BUG();
+	} else {
+		hdr->gso_type = VIRTIO_NET_HDR_GSO_NONE;
+		hdr->gso_size = 0;
+	}
+
+	vnet_hdr_to_sg(sg, skb);
+	num = skb_to_sgvec(skb, sg+1, 0, skb->len) + 1;
+	__skb_queue_head(&vi->send, skb);
+	err = vi->svq->vq_ops->add_buf(vi->svq, sg, num, 0, skb);
+	if (err) {
+		pr_debug("%s: virtio not prepared to send\n", dev->name);
+		skb_unlink(skb, &vi->send);
+		netif_stop_queue(dev);
+		return NETDEV_TX_BUSY;
+	}
+	vi->svq->vq_ops->kick(vi->svq);
+
+	return 0;
+}
+
+static int virtnet_open(struct net_device *dev)
+{
+	struct virtnet_info *vi = netdev_priv(dev);
+
+	try_fill_recv(vi);
+
+	/* If we didn't even get one input buffer, we're useless. */
+	if (vi->num == 0)
+		return -ENOMEM;
+
+	napi_enable(&vi->napi);
+	return 0;
+}
+
+static int virtnet_close(struct net_device *dev)
+{
+	struct virtnet_info *vi = netdev_priv(dev);
+	struct sk_buff *skb;
+
+	napi_disable(&vi->napi);
+
+	/* networking core has neutered skb_xmit_done/skb_recv_done, so don't
+	 * worry about races vs. get(). */
+	vi->rvq->vq_ops->shutdown(vi->rvq);
+	while ((skb = __skb_dequeue(&vi->recv)) != NULL) {
+		kfree_skb(skb);
+		vi->num--;
+	}
+	vi->svq->vq_ops->shutdown(vi->svq);
+	while ((skb = __skb_dequeue(&vi->send)) != NULL)
+		kfree_skb(skb);
+
+	BUG_ON(vi->num != 0);
+	return 0;
+}
+
+static int virtnet_probe(struct virtio_device *vdev)
+{
+	int err;
+	unsigned int len;
+	struct net_device *dev;
+	struct virtnet_info *vi;
+	void *token;
+
+	/* Allocate ourselves a network device with room for our info */
+	dev = alloc_etherdev(sizeof(struct virtnet_info));
+	if (!dev)
+		return -ENOMEM;
+
+	/* Set up network device as normal. */
+	ether_setup(dev);
+	dev->open = virtnet_open;
+	dev->stop = virtnet_close;
+	dev->hard_start_xmit = start_xmit;
+	dev->features = NETIF_F_HIGHDMA;
+	SET_NETDEV_DEV(dev, &vdev->dev);
+
+	/* Do we support "hardware" checksums? */
+	token = vdev->config->find(vdev, VIRTIO_CONFIG_NET_F, &len);
+	if (virtio_use_bit(vdev, token, len, VIRTIO_NET_F_NO_CSUM)) {
+		/* This opens up the world of extra features. */
+		dev->features |= NETIF_F_HW_CSUM|NETIF_F_SG|NETIF_F_FRAGLIST;
+		if (virtio_use_bit(vdev, token, len, VIRTIO_NET_F_TSO4))
+			dev->features |= NETIF_F_TSO;
+		if (virtio_use_bit(vdev, token, len, VIRTIO_NET_F_UFO))
+			dev->features |= NETIF_F_UFO;
+		if (virtio_use_bit(vdev, token, len, VIRTIO_NET_F_TSO4_ECN))
+			dev->features |= NETIF_F_TSO_ECN;
+		if (virtio_use_bit(vdev, token, len, VIRTIO_NET_F_TSO6))
+			dev->features |= NETIF_F_TSO6;
+	}
+
+	/* Configuration may specify what MAC to use.  Otherwise random. */
+	token = vdev->config->find(vdev, VIRTIO_CONFIG_NET_MAC_F, &len);
+	if (token) {
+		dev->addr_len = len;
+		vdev->config->get(vdev, token, dev->dev_addr, len);
+	} else
+		random_ether_addr(dev->dev_addr);
+
+	/* Set up our device-specific information */
+	vi = netdev_priv(dev);
+	netif_napi_add(dev, &vi->napi, virtnet_poll, 16);
+	vi->dev = dev;
+	vi->vdev = vdev;
+
+	/* We expect two virtqueues, receive then send. */
+	vi->rvq = vdev->config->find_vq(vdev, skb_recv_done);
+	if (IS_ERR(vi->rvq)) {
+		err = PTR_ERR(vi->rvq);
+		goto free;
+	}
+
+	vi->svq = vdev->config->find_vq(vdev, skb_xmit_done);
+	if (IS_ERR(vi->svq)) {
+		err = PTR_ERR(vi->svq);
+		goto free_recv;
+	}
+
+	/* Initialize our empty receive and send queues. */
+	skb_queue_head_init(&vi->recv);
+	skb_queue_head_init(&vi->send);
+
+	err = register_netdev(dev);
+	if (err) {
+		pr_debug("virtio_net: registering device failed\n");
+		goto free_send;
+	}
+	pr_debug("virtnet: registered device %s\n", dev->name);
+	vdev->priv = vi;
+	return 0;
+
+free_send:
+	vdev->config->del_vq(vi->svq);
+free_recv:
+	vdev->config->del_vq(vi->rvq);
+free:
+	free_netdev(dev);
+	return err;
+}
+
+static void virtnet_remove(struct virtio_device *vdev)
+{
+	unregister_netdev(vdev->priv);
+	free_netdev(vdev->priv);
+}
+
+static struct virtio_device_id id_table[] = {
+	{ VIRTIO_ID_NET, VIRTIO_DEV_ANY_ID },
+	{ 0 },
+};
+
+static struct virtio_driver virtio_net = {
+	.driver.name =	KBUILD_MODNAME,
+	.driver.owner =	THIS_MODULE,
+	.id_table =	id_table,
+	.probe =	virtnet_probe,
+	.remove =	__devexit_p(virtnet_remove),
+};
+
+static int __init init(void)
+{
+	return register_virtio_driver(&virtio_net);
+}
+
+static void __exit fini(void)
+{
+	unregister_virtio_driver(&virtio_net);
+}
+module_init(init);
+module_exit(fini);
+
+MODULE_DEVICE_TABLE(virtio, id_table);
+MODULE_DESCRIPTION("Virtio network driver");
+MODULE_LICENSE("GPL");

drivers/virtio/Kconfig

@@ -0,0 +1,8 @@
+# Virtio always gets selected by whoever wants it.
+config VIRTIO
+	bool
+
+# Similarly the virtio ring implementation.
+config VIRTIO_RING
+	bool
+	depends on VIRTIO

drivers/virtio/Makefile

@@ -0,0 +1,2 @@
+obj-$(CONFIG_VIRTIO) += virtio.o
+obj-$(CONFIG_VIRTIO_RING) += virtio_ring.o

drivers/virtio/config.c

@@ -0,0 +1,13 @@
+/* Configuration space parsing helpers for virtio.
+ *
+ * The configuration is [type][len][... len bytes ...] fields.
+ *
+ * Copyright 2007 Rusty Russell, IBM Corporation.
+ * GPL v2 or later.
+ */
+#include <linux/err.h>
+#include <linux/virtio.h>
+#include <linux/virtio_config.h>
+#include <linux/bug.h>
+#include <asm/system.h>
+

drivers/virtio/virtio.c

@@ -0,0 +1,189 @@
+#include <linux/virtio.h>
+#include <linux/spinlock.h>
+#include <linux/virtio_config.h>
+
+static ssize_t device_show(struct device *_d,
+			   struct device_attribute *attr, char *buf)
+{
+	struct virtio_device *dev = container_of(_d,struct virtio_device,dev);
+	return sprintf(buf, "%hu", dev->id.device);
+}
+static ssize_t vendor_show(struct device *_d,
+			   struct device_attribute *attr, char *buf)
+{
+	struct virtio_device *dev = container_of(_d,struct virtio_device,dev);
+	return sprintf(buf, "%hu", dev->id.vendor);
+}
+static ssize_t status_show(struct device *_d,
+			   struct device_attribute *attr, char *buf)
+{
+	struct virtio_device *dev = container_of(_d,struct virtio_device,dev);
+	return sprintf(buf, "0x%08x", dev->config->get_status(dev));
+}
+static ssize_t modalias_show(struct device *_d,
+			     struct device_attribute *attr, char *buf)
+{
+	struct virtio_device *dev = container_of(_d,struct virtio_device,dev);
+
+	return sprintf(buf, "virtio:d%08Xv%08X\n",
+		       dev->id.device, dev->id.vendor);
+}
+static struct device_attribute virtio_dev_attrs[] = {
+	__ATTR_RO(device),
+	__ATTR_RO(vendor),
+	__ATTR_RO(status),
+	__ATTR_RO(modalias),
+	__ATTR_NULL
+};
+
+static inline int virtio_id_match(const struct virtio_device *dev,
+				  const struct virtio_device_id *id)
+{
+	if (id->device != dev->id.device)
+		return 0;
+
+	return id->vendor == VIRTIO_DEV_ANY_ID || id->vendor != dev->id.vendor;
+}
+
+/* This looks through all the IDs a driver claims to support.  If any of them
+ * match, we return 1 and the kernel will call virtio_dev_probe(). */
+static int virtio_dev_match(struct device *_dv, struct device_driver *_dr)
+{
+	unsigned int i;
+	struct virtio_device *dev = container_of(_dv,struct virtio_device,dev);
+	const struct virtio_device_id *ids;
+
+	ids = container_of(_dr, struct virtio_driver, driver)->id_table;
+	for (i = 0; ids[i].device; i++)
+		if (virtio_id_match(dev, &ids[i]))
+			return 1;
+	return 0;
+}
+
+static int virtio_uevent(struct device *_dv, struct kobj_uevent_env *env)
+{
+	struct virtio_device *dev = container_of(_dv,struct virtio_device,dev);
+
+	return add_uevent_var(env, "MODALIAS=virtio:d%08Xv%08X",
+			      dev->id.device, dev->id.vendor);
+}
+
+static struct bus_type virtio_bus = {
+	.name  = "virtio",
+	.match = virtio_dev_match,
+	.dev_attrs = virtio_dev_attrs,
+	.uevent = virtio_uevent,
+};
+
+static void add_status(struct virtio_device *dev, unsigned status)
+{
+	dev->config->set_status(dev, dev->config->get_status(dev) | status);
+}
+
+static int virtio_dev_probe(struct device *_d)
+{
+	int err;
+	struct virtio_device *dev = container_of(_d,struct virtio_device,dev);
+	struct virtio_driver *drv = container_of(dev->dev.driver,
+						 struct virtio_driver, driver);
+
+	add_status(dev, VIRTIO_CONFIG_S_DRIVER);
+	err = drv->probe(dev);
+	if (err)
+		add_status(dev, VIRTIO_CONFIG_S_FAILED);
+	else
+		add_status(dev, VIRTIO_CONFIG_S_DRIVER_OK);
+	return err;
+}
+
+int register_virtio_driver(struct virtio_driver *driver)
+{
+	driver->driver.bus = &virtio_bus;
+	driver->driver.probe = virtio_dev_probe;
+	return driver_register(&driver->driver);
+}
+EXPORT_SYMBOL_GPL(register_virtio_driver);
+
+void unregister_virtio_driver(struct virtio_driver *driver)
+{
+	driver_unregister(&driver->driver);
+}
+EXPORT_SYMBOL_GPL(unregister_virtio_driver);
+
+int register_virtio_device(struct virtio_device *dev)
+{
+	int err;
+
+	dev->dev.bus = &virtio_bus;
+	sprintf(dev->dev.bus_id, "%u", dev->index);
+
+	/* Acknowledge that we've seen the device. */
+	add_status(dev, VIRTIO_CONFIG_S_ACKNOWLEDGE);
+
+	/* device_register() causes the bus infrastructure to look for a
+	 * matching driver. */
+	err = device_register(&dev->dev);
+	if (err)
+		add_status(dev, VIRTIO_CONFIG_S_FAILED);
+	return err;
+}
+EXPORT_SYMBOL_GPL(register_virtio_device);
+
+void unregister_virtio_device(struct virtio_device *dev)
+{
+	device_unregister(&dev->dev);
+}
+EXPORT_SYMBOL_GPL(unregister_virtio_device);
+
+int __virtio_config_val(struct virtio_device *vdev,
+			u8 type, void *val, size_t size)
+{
+	void *token;
+	unsigned int len;
+
+	token = vdev->config->find(vdev, type, &len);
+	if (!token)
+		return -ENOENT;
+
+	if (len != size)
+		return -EIO;
+
+	vdev->config->get(vdev, token, val, size);
+	return 0;
+}
+EXPORT_SYMBOL_GPL(__virtio_config_val);
+
+int virtio_use_bit(struct virtio_device *vdev,
+		   void *token, unsigned int len, unsigned int bitnum)
+{
+	unsigned long bits[16];
+
+	/* This makes it convenient to pass-through find() results. */
+	if (!token)
+		return 0;
+
+	/* bit not in range of this bitfield? */
+	if (bitnum * 8 >= len / 2)
+		return 0;
+
+	/* Giant feature bitfields are silly. */
+	BUG_ON(len > sizeof(bits));
+	vdev->config->get(vdev, token, bits, len);
+
+	if (!test_bit(bitnum, bits))
+		return 0;
+
+	/* Set acknowledge bit, and write it back. */
+	set_bit(bitnum + len * 8 / 2, bits);
+	vdev->config->set(vdev, token, bits, len);
+	return 1;
+}
+EXPORT_SYMBOL_GPL(virtio_use_bit);
+
+static int virtio_init(void)
+{
+	if (bus_register(&virtio_bus) != 0)
+		panic("virtio bus registration failed");
+	return 0;
+}
+core_initcall(virtio_init);

drivers/virtio/virtio_ring.c

@@ -0,0 +1,313 @@
+/* Virtio ring implementation.
+ *
+ *  Copyright 2007 Rusty Russell IBM Corporation
+ *
+ *  This program is free software; you can redistribute it and/or modify
+ *  it under the terms of the GNU General Public License as published by
+ *  the Free Software Foundation; either version 2 of the License, or
+ *  (at your option) any later version.
+ *
+ *  This program is distributed in the hope that it will be useful,
+ *  but WITHOUT ANY WARRANTY; without even the implied warranty of
+ *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ *  GNU General Public License for more details.
+ *
+ *  You should have received a copy of the GNU General Public License
+ *  along with this program; if not, write to the Free Software
+ *  Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA
+ */
+#include <linux/virtio.h>
+#include <linux/virtio_ring.h>
+#include <linux/device.h>
+
+#ifdef DEBUG
+/* For development, we want to crash whenever the ring is screwed. */
+#define BAD_RING(vq, fmt...)			\
+	do { dev_err(&vq->vq.vdev->dev, fmt); BUG(); } while(0)
+#define START_USE(vq) \
+	do { if ((vq)->in_use) panic("in_use = %i\n", (vq)->in_use); (vq)->in_use = __LINE__; mb(); } while(0)
+#define END_USE(vq) \
+	do { BUG_ON(!(vq)->in_use); (vq)->in_use = 0; mb(); } while(0)
+#else
+#define BAD_RING(vq, fmt...)			\
+	do { dev_err(&vq->vq.vdev->dev, fmt); (vq)->broken = true; } while(0)
+#define START_USE(vq)
+#define END_USE(vq)
+#endif
+
+struct vring_virtqueue
+{
+	struct virtqueue vq;
+
+	/* Actual memory layout for this queue */
+	struct vring vring;
+
+	/* Other side has made a mess, don't try any more. */
+	bool broken;
+
+	/* Number of free buffers */
+	unsigned int num_free;
+	/* Head of free buffer list. */
+	unsigned int free_head;
+	/* Number we've added since last sync. */
+	unsigned int num_added;
+
+	/* Last used index we've seen. */
+	unsigned int last_used_idx;
+
+	/* How to notify other side. FIXME: commonalize hcalls! */
+	void (*notify)(struct virtqueue *vq);
+
+#ifdef DEBUG
+	/* They're supposed to lock for us. */
+	unsigned int in_use;
+#endif
+
+	/* Tokens for callbacks. */
+	void *data[];
+};
+
+#define to_vvq(_vq) container_of(_vq, struct vring_virtqueue, vq)
+
+static int vring_add_buf(struct virtqueue *_vq,
+			 struct scatterlist sg[],
+			 unsigned int out,
+			 unsigned int in,
+			 void *data)
+{
+	struct vring_virtqueue *vq = to_vvq(_vq);
+	unsigned int i, avail, head, uninitialized_var(prev);
+
+	BUG_ON(data == NULL);
+	BUG_ON(out + in > vq->vring.num);
+	BUG_ON(out + in == 0);
+
+	START_USE(vq);
+
+	if (vq->num_free < out + in) {
+		pr_debug("Can't add buf len %i - avail = %i\n",
+			 out + in, vq->num_free);
+		END_USE(vq);
+		return -ENOSPC;
+	}
+
+	/* We're about to use some buffers from the free list. */
+	vq->num_free -= out + in;
+
+	head = vq->free_head;
+	for (i = vq->free_head; out; i = vq->vring.desc[i].next, out--) {
+		vq->vring.desc[i].flags = VRING_DESC_F_NEXT;
+		vq->vring.desc[i].addr = (page_to_pfn(sg_page(sg))<<PAGE_SHIFT)
+			+ sg->offset;
+		vq->vring.desc[i].len = sg->length;
+		prev = i;
+		sg++;
+	}
+	for (; in; i = vq->vring.desc[i].next, in--) {
+		vq->vring.desc[i].flags = VRING_DESC_F_NEXT|VRING_DESC_F_WRITE;
+		vq->vring.desc[i].addr = (page_to_pfn(sg_page(sg))<<PAGE_SHIFT)
+			+ sg->offset;
+		vq->vring.desc[i].len = sg->length;
+		prev = i;
+		sg++;
+	}
+	/* Last one doesn't continue. */
+	vq->vring.desc[prev].flags &= ~VRING_DESC_F_NEXT;
+
+	/* Update free pointer */
+	vq->free_head = i;
+
+	/* Set token. */
+	vq->data[head] = data;
+
+	/* Put entry in available array (but don't update avail->idx until they
+	 * do sync).  FIXME: avoid modulus here? */
+	avail = (vq->vring.avail->idx + vq->num_added++) % vq->vring.num;
+	vq->vring.avail->ring[avail] = head;
+
+	pr_debug("Added buffer head %i to %p\n", head, vq);
+	END_USE(vq);
+	return 0;
+}
+
+static void vring_kick(struct virtqueue *_vq)
+{
+	struct vring_virtqueue *vq = to_vvq(_vq);
+	START_USE(vq);
+	/* Descriptors and available array need to be set before we expose the
+	 * new available array entries. */
+	wmb();
+
+	vq->vring.avail->idx += vq->num_added;
+	vq->num_added = 0;
+
+	/* Need to update avail index before checking if we should notify */
+	mb();
+
+	if (!(vq->vring.used->flags & VRING_USED_F_NO_NOTIFY))
+		/* Prod other side to tell it about changes. */
+		vq->notify(&vq->vq);
+
+	END_USE(vq);
+}
+
+static void detach_buf(struct vring_virtqueue *vq, unsigned int head)
+{
+	unsigned int i;
+
+	/* Clear data ptr. */
+	vq->data[head] = NULL;
+
+	/* Put back on free list: find end */
+	i = head;
+	while (vq->vring.desc[i].flags & VRING_DESC_F_NEXT) {
+		i = vq->vring.desc[i].next;
+		vq->num_free++;
+	}
+
+	vq->vring.desc[i].next = vq->free_head;
+	vq->free_head = head;
+	/* Plus final descriptor */
+	vq->num_free++;
+}
+
+/* FIXME: We need to tell other side about removal, to synchronize. */
+static void vring_shutdown(struct virtqueue *_vq)
+{
+	struct vring_virtqueue *vq = to_vvq(_vq);
+	unsigned int i;
+
+	for (i = 0; i < vq->vring.num; i++)
+		detach_buf(vq, i);
+}
+
+static inline bool more_used(const struct vring_virtqueue *vq)
+{
+	return vq->last_used_idx != vq->vring.used->idx;
+}
+
+static void *vring_get_buf(struct virtqueue *_vq, unsigned int *len)
+{
+	struct vring_virtqueue *vq = to_vvq(_vq);
+	void *ret;
+	unsigned int i;
+
+	START_USE(vq);
+
+	if (!more_used(vq)) {
+		pr_debug("No more buffers in queue\n");
+		END_USE(vq);
+		return NULL;
+	}
+
+	i = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].id;
+	*len = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].len;
+
+	if (unlikely(i >= vq->vring.num)) {
+		BAD_RING(vq, "id %u out of range\n", i);
+		return NULL;
+	}
+	if (unlikely(!vq->data[i])) {
+		BAD_RING(vq, "id %u is not a head!\n", i);
+		return NULL;
+	}
+
+	/* detach_buf clears data, so grab it now. */
+	ret = vq->data[i];
+	detach_buf(vq, i);
+	vq->last_used_idx++;
+	END_USE(vq);
+	return ret;
+}
+
+static bool vring_restart(struct virtqueue *_vq)
+{
+	struct vring_virtqueue *vq = to_vvq(_vq);
+
+	START_USE(vq);
+	BUG_ON(!(vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT));
+
+	/* We optimistically turn back on interrupts, then check if there was
+	 * more to do. */
+	vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
+	mb();
+	if (unlikely(more_used(vq))) {
+		vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
+		END_USE(vq);
+		return false;
+	}
+
+	END_USE(vq);
+	return true;
+}
+
+irqreturn_t vring_interrupt(int irq, void *_vq)
+{
+	struct vring_virtqueue *vq = to_vvq(_vq);
+
+	if (!more_used(vq)) {
+		pr_debug("virtqueue interrupt with no work for %p\n", vq);
+		return IRQ_NONE;
+	}
+
+	if (unlikely(vq->broken))
+		return IRQ_HANDLED;
+
+	pr_debug("virtqueue callback for %p (%p)\n", vq, vq->vq.callback);
+	if (vq->vq.callback && !vq->vq.callback(&vq->vq))
+		vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
+
+	return IRQ_HANDLED;
+}
+
+static struct virtqueue_ops vring_vq_ops = {
+	.add_buf = vring_add_buf,
+	.get_buf = vring_get_buf,
+	.kick = vring_kick,
+	.restart = vring_restart,
+	.shutdown = vring_shutdown,
+};
+
+struct virtqueue *vring_new_virtqueue(unsigned int num,
+				      struct virtio_device *vdev,
+				      void *pages,
+				      void (*notify)(struct virtqueue *),
+				      bool (*callback)(struct virtqueue *))
+{
+	struct vring_virtqueue *vq;
+	unsigned int i;
+
+	vq = kmalloc(sizeof(*vq) + sizeof(void *)*num, GFP_KERNEL);
+	if (!vq)
+		return NULL;
+
+	vring_init(&vq->vring, num, pages);
+	vq->vq.callback = callback;
+	vq->vq.vdev = vdev;
+	vq->vq.vq_ops = &vring_vq_ops;
+	vq->notify = notify;
+	vq->broken = false;
+	vq->last_used_idx = 0;
+	vq->num_added = 0;
+#ifdef DEBUG
+	vq->in_use = false;
+#endif
+
+	/* No callback?  Tell other side not to bother us. */
+	if (!callback)
+		vq->vring.avail->flags |= VRING_AVAIL_F_NO_INTERRUPT;
+
+	/* Put everything in free lists. */
+	vq->num_free = num;
+	vq->free_head = 0;
+	for (i = 0; i < num-1; i++)
+		vq->vring.desc[i].next = i+1;
+
+	return &vq->vq;
+}
+
+void vring_del_virtqueue(struct virtqueue *vq)
+{
+	kfree(to_vvq(vq));
+}
+

include/asm-x86/Kbuild

@@ -1,6 +1,7 @@
 include include/asm-generic/Kbuild.asm
 
 header-y += boot.h
+header-y += bootparam.h
 header-y += debugreg.h
 header-y += ldt.h
 header-y += msr-index.h
@@ -14,8 +15,10 @@ unifdef-y += a.out_32.h
 unifdef-y += a.out_64.h
 unifdef-y += byteorder_32.h
 unifdef-y += byteorder_64.h
+unifdef-y += e820.h
 unifdef-y += elf_32.h
 unifdef-y += elf_64.h
+unifdef-y += ist.h
 unifdef-y += mce.h
 unifdef-y += msgbuf_32.h
 unifdef-y += msgbuf_64.h

include/asm-x86/bootparam.h

@@ -10,85 +10,85 @@
 #include <video/edid.h>
 
 struct setup_header {
-	u8	setup_sects;
-	u16	root_flags;
-	u32	syssize;
-	u16	ram_size;
+	__u8	setup_sects;
+	__u16	root_flags;
+	__u32	syssize;
+	__u16	ram_size;
 #define RAMDISK_IMAGE_START_MASK	0x07FF
 #define RAMDISK_PROMPT_FLAG		0x8000
 #define RAMDISK_LOAD_FLAG		0x4000
-	u16	vid_mode;
-	u16	root_dev;
-	u16	boot_flag;
-	u16	jump;
-	u32	header;
-	u16	version;
-	u32	realmode_swtch;
-	u16	start_sys;
-	u16	kernel_version;
-	u8	type_of_loader;
-	u8	loadflags;
+	__u16	vid_mode;
+	__u16	root_dev;
+	__u16	boot_flag;
+	__u16	jump;
+	__u32	header;
+	__u16	version;
+	__u32	realmode_swtch;
+	__u16	start_sys;
+	__u16	kernel_version;
+	__u8	type_of_loader;
+	__u8	loadflags;
 #define LOADED_HIGH	(1<<0)
 #define KEEP_SEGMENTS	(1<<6)
 #define CAN_USE_HEAP	(1<<7)
-	u16	setup_move_size;
-	u32	code32_start;
-	u32	ramdisk_image;
-	u32	ramdisk_size;
-	u32	bootsect_kludge;
-	u16	heap_end_ptr;
-	u16	_pad1;
-	u32	cmd_line_ptr;
-	u32	initrd_addr_max;
-	u32	kernel_alignment;
-	u8	relocatable_kernel;
-	u8	_pad2[3];
-	u32	cmdline_size;
-	u32	hardware_subarch;
-	u64	hardware_subarch_data;
+	__u16	setup_move_size;
+	__u32	code32_start;
+	__u32	ramdisk_image;
+	__u32	ramdisk_size;
+	__u32	bootsect_kludge;
+	__u16	heap_end_ptr;
+	__u16	_pad1;
+	__u32	cmd_line_ptr;
+	__u32	initrd_addr_max;
+	__u32	kernel_alignment;
+	__u8	relocatable_kernel;
+	__u8	_pad2[3];
+	__u32	cmdline_size;
+	__u32	hardware_subarch;
+	__u64	hardware_subarch_data;
 } __attribute__((packed));
 
 struct sys_desc_table {
-	u16 length;
-	u8  table[14];
+	__u16 length;
+	__u8  table[14];
 };
 
 struct efi_info {
-	u32 _pad1;
-	u32 efi_systab;
-	u32 efi_memdesc_size;
-	u32 efi_memdesc_version;
-	u32 efi_memmap;
-	u32 efi_memmap_size;
-	u32 _pad2[2];
+	__u32 _pad1;
+	__u32 efi_systab;
+	__u32 efi_memdesc_size;
+	__u32 efi_memdesc_version;
+	__u32 efi_memmap;
+	__u32 efi_memmap_size;
+	__u32 _pad2[2];
 };
 
 /* The so-called "zeropage" */
 struct boot_params {
 	struct screen_info screen_info;			/* 0x000 */
 	struct apm_bios_info apm_bios_info;		/* 0x040 */
-	u8  _pad2[12];					/* 0x054 */
+	__u8  _pad2[12];				/* 0x054 */
 	struct ist_info ist_info;			/* 0x060 */
-	u8  _pad3[16];					/* 0x070 */
-	u8  hd0_info[16];	/* obsolete! */		/* 0x080 */
-	u8  hd1_info[16];	/* obsolete! */		/* 0x090 */
+	__u8  _pad3[16];				/* 0x070 */
+	__u8  hd0_info[16];	/* obsolete! */		/* 0x080 */
+	__u8  hd1_info[16];	/* obsolete! */		/* 0x090 */
 	struct sys_desc_table sys_desc_table;		/* 0x0a0 */
-	u8  _pad4[144];					/* 0x0b0 */
+	__u8  _pad4[144];				/* 0x0b0 */
 	struct edid_info edid_info;			/* 0x140 */
 	struct efi_info efi_info;			/* 0x1c0 */
-	u32 alt_mem_k;					/* 0x1e0 */
-	u32 scratch;		/* Scratch field! */	/* 0x1e4 */
-	u8  e820_entries;				/* 0x1e8 */
-	u8  eddbuf_entries;				/* 0x1e9 */
-	u8  edd_mbr_sig_buf_entries;			/* 0x1ea */
-	u8  _pad6[6];					/* 0x1eb */
+	__u32 alt_mem_k;				/* 0x1e0 */
+	__u32 scratch;		/* Scratch field! */	/* 0x1e4 */
+	__u8  e820_entries;				/* 0x1e8 */
+	__u8  eddbuf_entries;				/* 0x1e9 */
+	__u8  edd_mbr_sig_buf_entries;			/* 0x1ea */
+	__u8  _pad6[6];					/* 0x1eb */
 	struct setup_header hdr;    /* setup header */	/* 0x1f1 */
-	u8  _pad7[0x290-0x1f1-sizeof(struct setup_header)];
-	u32 edd_mbr_sig_buffer[EDD_MBR_SIG_MAX];	/* 0x290 */
+	__u8  _pad7[0x290-0x1f1-sizeof(struct setup_header)];
+	__u32 edd_mbr_sig_buffer[EDD_MBR_SIG_MAX];	/* 0x290 */
 	struct e820entry e820_map[E820MAX];		/* 0x2d0 */
-	u8  _pad8[48];					/* 0xcd0 */
+	__u8  _pad8[48];				/* 0xcd0 */
 	struct edd_info eddbuf[EDDMAXNR];		/* 0xd00 */
-	u8  _pad9[276];					/* 0xeec */
+	__u8  _pad9[276];				/* 0xeec */
 } __attribute__((packed));
 
 #endif /* _ASM_BOOTPARAM_H */

include/asm-x86/e820.h

@@ -1,5 +1,33 @@
+#ifndef __ASM_E820_H
+#define __ASM_E820_H
+#define E820MAP	0x2d0		/* our map */
+#define E820MAX	128		/* number of entries in E820MAP */
+#define E820NR	0x1e8		/* # entries in E820MAP */
+
+#define E820_RAM	1
+#define E820_RESERVED	2
+#define E820_ACPI	3
+#define E820_NVS	4
+
+#ifndef __ASSEMBLY__
+struct e820entry {
+	__u64 addr;	/* start of memory segment */
+	__u64 size;	/* size of memory segment */
+	__u32 type;	/* type of memory segment */
+} __attribute__((packed));
+
+struct e820map {
+	__u32 nr_map;
+	struct e820entry map[E820MAX];
+};
+#endif /* __ASSEMBLY__ */
+
+#ifdef __KERNEL__
 #ifdef CONFIG_X86_32
 # include "e820_32.h"
 #else
 # include "e820_64.h"
 #endif
+#endif /* __KERNEL__ */
+
+#endif  /* __ASM_E820_H */

include/asm-x86/e820_32.h

@@ -12,30 +12,10 @@
 #ifndef __E820_HEADER
 #define __E820_HEADER
 
-#define E820MAP	0x2d0		/* our map */
-#define E820MAX	128		/* number of entries in E820MAP */
-#define E820NR	0x1e8		/* # entries in E820MAP */
-
-#define E820_RAM	1
-#define E820_RESERVED	2
-#define E820_ACPI	3
-#define E820_NVS	4
-
 #define HIGH_MEMORY	(1024*1024)
 
 #ifndef __ASSEMBLY__
 
-struct e820entry {
-	u64 addr;	/* start of memory segment */
-	u64 size;	/* size of memory segment */
-	u32 type;	/* type of memory segment */
-} __attribute__((packed));
-
-struct e820map {
-	u32 nr_map;
-	struct e820entry map[E820MAX];
-};
-
 extern struct e820map e820;
 
 extern int e820_all_mapped(unsigned long start, unsigned long end,
@@ -56,5 +36,4 @@ static inline void e820_mark_nosave_regions(void)
 #endif
 
 #endif/*!__ASSEMBLY__*/
-
 #endif/*__E820_HEADER*/

include/asm-x86/e820_64.h

@@ -11,27 +11,7 @@
 #ifndef __E820_HEADER
 #define __E820_HEADER
 
-#define E820MAP	0x2d0		/* our map */
-#define E820MAX	128		/* number of entries in E820MAP */
-#define E820NR	0x1e8		/* # entries in E820MAP */
-
-#define E820_RAM	1
-#define E820_RESERVED	2
-#define E820_ACPI	3
-#define E820_NVS	4
-
 #ifndef __ASSEMBLY__
-struct e820entry {
-	u64 addr;	/* start of memory segment */
-	u64 size;	/* size of memory segment */
-	u32 type;	/* type of memory segment */
-} __attribute__((packed));
-
-struct e820map {
-	u32 nr_map;
-	struct e820entry map[E820MAX];
-};
-
 extern unsigned long find_e820_area(unsigned long start, unsigned long end, 
 				    unsigned size);
 extern void add_memory_region(unsigned long start, unsigned long size, 

include/asm-x86/ist.h

@@ -17,17 +17,17 @@
  */
 
 
-#ifdef __KERNEL__
-
 #include <linux/types.h>
 
 struct ist_info {
-	u32 signature;
-	u32 command;
-	u32 event;
-	u32 perf_level;
+	__u32 signature;
+	__u32 command;
+	__u32 event;
+	__u32 perf_level;
 };
 
+#ifdef __KERNEL__
+
 extern struct ist_info ist_info;
 
 #endif	/* __KERNEL__ */

include/asm-x86/lguest.h

@@ -0,0 +1,86 @@
+#ifndef _X86_LGUEST_H
+#define _X86_LGUEST_H
+
+#define GDT_ENTRY_LGUEST_CS	10
+#define GDT_ENTRY_LGUEST_DS	11
+#define LGUEST_CS		(GDT_ENTRY_LGUEST_CS * 8)
+#define LGUEST_DS		(GDT_ENTRY_LGUEST_DS * 8)
+
+#ifndef __ASSEMBLY__
+#include <asm/desc.h>
+
+#define GUEST_PL 1
+
+/* Every guest maps the core switcher code. */
+#define SHARED_SWITCHER_PAGES \
+	DIV_ROUND_UP(end_switcher_text - start_switcher_text, PAGE_SIZE)
+/* Pages for switcher itself, then two pages per cpu */
+#define TOTAL_SWITCHER_PAGES (SHARED_SWITCHER_PAGES + 2 * NR_CPUS)
+
+/* We map at -4M for ease of mapping into the guest (one PTE page). */
+#define SWITCHER_ADDR 0xFFC00000
+
+/* Found in switcher.S */
+extern unsigned long default_idt_entries[];
+
+struct lguest_regs
+{
+	/* Manually saved part. */
+	unsigned long eax, ebx, ecx, edx;
+	unsigned long esi, edi, ebp;
+	unsigned long gs;
+	unsigned long fs, ds, es;
+	unsigned long trapnum, errcode;
+	/* Trap pushed part */
+	unsigned long eip;
+	unsigned long cs;
+	unsigned long eflags;
+	unsigned long esp;
+	unsigned long ss;
+};
+
+/* This is a guest-specific page (mapped ro) into the guest. */
+struct lguest_ro_state
+{
+	/* Host information we need to restore when we switch back. */
+	u32 host_cr3;
+	struct Xgt_desc_struct host_idt_desc;
+	struct Xgt_desc_struct host_gdt_desc;
+	u32 host_sp;
+
+	/* Fields which are used when guest is running. */
+	struct Xgt_desc_struct guest_idt_desc;
+	struct Xgt_desc_struct guest_gdt_desc;
+	struct i386_hw_tss guest_tss;
+	struct desc_struct guest_idt[IDT_ENTRIES];
+	struct desc_struct guest_gdt[GDT_ENTRIES];
+};
+
+struct lguest_arch
+{
+	/* The GDT entries copied into lguest_ro_state when running. */
+	struct desc_struct gdt[GDT_ENTRIES];
+
+	/* The IDT entries: some copied into lguest_ro_state when running. */
+	struct desc_struct idt[IDT_ENTRIES];
+
+	/* The address of the last guest-visible pagefault (ie. cr2). */
+	unsigned long last_pagefault;
+};
+
+static inline void lguest_set_ts(void)
+{
+	u32 cr0;
+
+	cr0 = read_cr0();
+	if (!(cr0 & 8))
+		write_cr0(cr0|8);
+}
+
+/* Full 4G segment descriptors, suitable for CS and DS. */
+#define FULL_EXEC_SEGMENT ((struct desc_struct){0x0000ffff, 0x00cf9b00})
+#define FULL_SEGMENT ((struct desc_struct){0x0000ffff, 0x00cf9300})
+
+#endif /* __ASSEMBLY__ */
+
+#endif

include/asm-x86/lguest_hcall.h

@@ -0,0 +1,71 @@
+/* Architecture specific portion of the lguest hypercalls */
+#ifndef _X86_LGUEST_HCALL_H
+#define _X86_LGUEST_HCALL_H
+
+#define LHCALL_FLUSH_ASYNC	0
+#define LHCALL_LGUEST_INIT	1
+#define LHCALL_CRASH		2
+#define LHCALL_LOAD_GDT		3
+#define LHCALL_NEW_PGTABLE	4
+#define LHCALL_FLUSH_TLB	5
+#define LHCALL_LOAD_IDT_ENTRY	6
+#define LHCALL_SET_STACK	7
+#define LHCALL_TS		8
+#define LHCALL_SET_CLOCKEVENT	9
+#define LHCALL_HALT		10
+#define LHCALL_SET_PTE		14
+#define LHCALL_SET_PMD		15
+#define LHCALL_LOAD_TLS		16
+#define LHCALL_NOTIFY		17
+
+/*G:031 First, how does our Guest contact the Host to ask for privileged
+ * operations?  There are two ways: the direct way is to make a "hypercall",
+ * to make requests of the Host Itself.
+ *
+ * Our hypercall mechanism uses the highest unused trap code (traps 32 and
+ * above are used by real hardware interrupts).  Seventeen hypercalls are
+ * available: the hypercall number is put in the %eax register, and the
+ * arguments (when required) are placed in %edx, %ebx and %ecx.  If a return
+ * value makes sense, it's returned in %eax.
+ *
+ * Grossly invalid calls result in Sudden Death at the hands of the vengeful
+ * Host, rather than returning failure.  This reflects Winston Churchill's
+ * definition of a gentleman: "someone who is only rude intentionally". */
+#define LGUEST_TRAP_ENTRY 0x1F
+
+#ifndef __ASSEMBLY__
+#include <asm/hw_irq.h>
+
+static inline unsigned long
+hcall(unsigned long call,
+      unsigned long arg1, unsigned long arg2, unsigned long arg3)
+{
+	/* "int" is the Intel instruction to trigger a trap. */
+	asm volatile("int $" __stringify(LGUEST_TRAP_ENTRY)
+		       /* The call is in %eax (aka "a"), and can be replaced */
+		     : "=a"(call)
+		       /* The other arguments are in %eax, %edx, %ebx & %ecx */
+		     : "a"(call), "d"(arg1), "b"(arg2), "c"(arg3)
+		       /* "memory" means this might write somewhere in memory.
+			* This isn't true for all calls, but it's safe to tell
+			* gcc that it might happen so it doesn't get clever. */
+		     : "memory");
+	return call;
+}
+/*:*/
+
+void async_hcall(unsigned long call,
+		 unsigned long arg1, unsigned long arg2, unsigned long arg3);
+
+/* Can't use our min() macro here: needs to be a constant */
+#define LGUEST_IRQS (NR_IRQS < 32 ? NR_IRQS: 32)
+
+#define LHCALL_RING_SIZE 64
+struct hcall_args
+{
+	/* These map directly onto eax, ebx, ecx, edx in struct lguest_regs */
+	unsigned long arg0, arg2, arg3, arg1;
+};
+
+#endif /* !__ASSEMBLY__ */
+#endif	/* _I386_LGUEST_HCALL_H */

include/linux/Kbuild

@@ -186,6 +186,7 @@ unifdef-y += cyclades.h
 unifdef-y += dccp.h
 unifdef-y += dirent.h
 unifdef-y += dlm.h
+unifdef-y += edd.h
 unifdef-y += elfcore.h
 unifdef-y += errno.h
 unifdef-y += errqueue.h
@@ -306,6 +307,7 @@ unifdef-y += rtc.h
 unifdef-y += rtnetlink.h
 unifdef-y += scc.h
 unifdef-y += sched.h
+unifdef-y += screen_info.h
 unifdef-y += sdla.h
 unifdef-y += selinux_netlink.h
 unifdef-y += sem.h
@@ -341,6 +343,9 @@ unifdef-y += user.h
 unifdef-y += utsname.h
 unifdef-y += videodev2.h
 unifdef-y += videodev.h
+unifdef-y += virtio_config.h
+unifdef-y += virtio_blk.h
+unifdef-y += virtio_net.h
 unifdef-y += wait.h
 unifdef-y += wanrouter.h
 unifdef-y += watchdog.h

include/linux/apm_bios.h

@@ -16,29 +16,29 @@
  * General Public License for more details.
  */
 
-typedef unsigned short	apm_event_t;
-typedef unsigned short	apm_eventinfo_t;
+#include <linux/types.h>
+
+struct apm_bios_info {
+	__u16	version;
+	__u16	cseg;
+	__u32	offset;
+	__u16	cseg_16;
+	__u16	dseg;
+	__u16	flags;
+	__u16	cseg_len;
+	__u16	cseg_16_len;
+	__u16	dseg_len;
+};
 
 #ifdef __KERNEL__
 
-#include <linux/types.h>
+typedef unsigned short	apm_event_t;
+typedef unsigned short	apm_eventinfo_t;
 
 #define APM_CS		(GDT_ENTRY_APMBIOS_BASE * 8)
 #define APM_CS_16	(APM_CS + 8)
 #define APM_DS		(APM_CS_16 + 8)
 
-struct apm_bios_info {
-	u16	version;
-	u16	cseg;
-	u32	offset;
-	u16	cseg_16;
-	u16	dseg;
-	u16	flags;
-	u16	cseg_len;
-	u16	cseg_16_len;
-	u16	dseg_len;
-};
-
 /* Results of APM Installation Check */
 #define APM_16_BIT_SUPPORT	0x0001
 #define APM_32_BIT_SUPPORT	0x0002

include/linux/edd.h

@@ -67,113 +67,113 @@
 #define EDD_INFO_USE_INT13_FN50                (1 << 7)
 
 struct edd_device_params {
-	u16 length;
-	u16 info_flags;
-	u32 num_default_cylinders;
-	u32 num_default_heads;
-	u32 sectors_per_track;
-	u64 number_of_sectors;
-	u16 bytes_per_sector;
-	u32 dpte_ptr;		/* 0xFFFFFFFF for our purposes */
-	u16 key;		/* = 0xBEDD */
-	u8 device_path_info_length;	/* = 44 */
-	u8 reserved2;
-	u16 reserved3;
-	u8 host_bus_type[4];
-	u8 interface_type[8];
+	__u16 length;
+	__u16 info_flags;
+	__u32 num_default_cylinders;
+	__u32 num_default_heads;
+	__u32 sectors_per_track;
+	__u64 number_of_sectors;
+	__u16 bytes_per_sector;
+	__u32 dpte_ptr;		/* 0xFFFFFFFF for our purposes */
+	__u16 key;		/* = 0xBEDD */
+	__u8 device_path_info_length;	/* = 44 */
+	__u8 reserved2;
+	__u16 reserved3;
+	__u8 host_bus_type[4];
+	__u8 interface_type[8];
 	union {
 		struct {
-			u16 base_address;
-			u16 reserved1;
-			u32 reserved2;
+			__u16 base_address;
+			__u16 reserved1;
+			__u32 reserved2;
 		} __attribute__ ((packed)) isa;
 		struct {
-			u8 bus;
-			u8 slot;
-			u8 function;
-			u8 channel;
-			u32 reserved;
+			__u8 bus;
+			__u8 slot;
+			__u8 function;
+			__u8 channel;
+			__u32 reserved;
 		} __attribute__ ((packed)) pci;
 		/* pcix is same as pci */
 		struct {
-			u64 reserved;
+			__u64 reserved;
 		} __attribute__ ((packed)) ibnd;
 		struct {
-			u64 reserved;
+			__u64 reserved;
 		} __attribute__ ((packed)) xprs;
 		struct {
-			u64 reserved;
+			__u64 reserved;
 		} __attribute__ ((packed)) htpt;
 		struct {
-			u64 reserved;
+			__u64 reserved;
 		} __attribute__ ((packed)) unknown;
 	} interface_path;
 	union {
 		struct {
-			u8 device;
-			u8 reserved1;
-			u16 reserved2;
-			u32 reserved3;
-			u64 reserved4;
+			__u8 device;
+			__u8 reserved1;
+			__u16 reserved2;
+			__u32 reserved3;
+			__u64 reserved4;
 		} __attribute__ ((packed)) ata;
 		struct {
-			u8 device;
-			u8 lun;
-			u8 reserved1;
-			u8 reserved2;
-			u32 reserved3;
-			u64 reserved4;
+			__u8 device;
+			__u8 lun;
+			__u8 reserved1;
+			__u8 reserved2;
+			__u32 reserved3;
+			__u64 reserved4;
 		} __attribute__ ((packed)) atapi;
 		struct {
-			u16 id;
-			u64 lun;
-			u16 reserved1;
-			u32 reserved2;
+			__u16 id;
+			__u64 lun;
+			__u16 reserved1;
+			__u32 reserved2;
 		} __attribute__ ((packed)) scsi;
 		struct {
-			u64 serial_number;
-			u64 reserved;
+			__u64 serial_number;
+			__u64 reserved;
 		} __attribute__ ((packed)) usb;
 		struct {
-			u64 eui;
-			u64 reserved;
+			__u64 eui;
+			__u64 reserved;
 		} __attribute__ ((packed)) i1394;
 		struct {
-			u64 wwid;
-			u64 lun;
+			__u64 wwid;
+			__u64 lun;
 		} __attribute__ ((packed)) fibre;
 		struct {
-			u64 identity_tag;
-			u64 reserved;
+			__u64 identity_tag;
+			__u64 reserved;
 		} __attribute__ ((packed)) i2o;
 		struct {
-			u32 array_number;
-			u32 reserved1;
-			u64 reserved2;
+			__u32 array_number;
+			__u32 reserved1;
+			__u64 reserved2;
 		} __attribute__ ((packed)) raid;
 		struct {
-			u8 device;
-			u8 reserved1;
-			u16 reserved2;
-			u32 reserved3;
-			u64 reserved4;
+			__u8 device;
+			__u8 reserved1;
+			__u16 reserved2;
+			__u32 reserved3;
+			__u64 reserved4;
 		} __attribute__ ((packed)) sata;
 		struct {
-			u64 reserved1;
-			u64 reserved2;
+			__u64 reserved1;
+			__u64 reserved2;
 		} __attribute__ ((packed)) unknown;
 	} device_path;
-	u8 reserved4;
-	u8 checksum;
+	__u8 reserved4;
+	__u8 checksum;
 } __attribute__ ((packed));
 
 struct edd_info {
-	u8 device;
-	u8 version;
-	u16 interface_support;
-	u16 legacy_max_cylinder;
-	u8 legacy_max_head;
-	u8 legacy_sectors_per_track;
+	__u8 device;
+	__u8 version;
+	__u16 interface_support;
+	__u16 legacy_max_cylinder;
+	__u8 legacy_max_head;
+	__u8 legacy_sectors_per_track;
 	struct edd_device_params params;
 } __attribute__ ((packed));
 
@@ -184,8 +184,9 @@ struct edd {
 	unsigned char edd_info_nr;
 };
 
+#ifdef __KERNEL__
 extern struct edd edd;
-
+#endif /* __KERNEL__ */
 #endif				/*!__ASSEMBLY__ */
 
 #endif				/* _LINUX_EDD_H */

include/linux/lguest.h

@@ -1,76 +1,16 @@
 /* Things the lguest guest needs to know.  Note: like all lguest interfaces,
  * this is subject to wild and random change between versions. */
-#ifndef _ASM_LGUEST_H
-#define _ASM_LGUEST_H
+#ifndef _LINUX_LGUEST_H
+#define _LINUX_LGUEST_H
 
 #ifndef __ASSEMBLY__
+#include <linux/time.h>
 #include <asm/irq.h>
-
-#define LHCALL_FLUSH_ASYNC	0
-#define LHCALL_LGUEST_INIT	1
-#define LHCALL_CRASH		2
-#define LHCALL_LOAD_GDT		3
-#define LHCALL_NEW_PGTABLE	4
-#define LHCALL_FLUSH_TLB	5
-#define LHCALL_LOAD_IDT_ENTRY	6
-#define LHCALL_SET_STACK	7
-#define LHCALL_TS		8
-#define LHCALL_SET_CLOCKEVENT	9
-#define LHCALL_HALT		10
-#define LHCALL_BIND_DMA		12
-#define LHCALL_SEND_DMA		13
-#define LHCALL_SET_PTE		14
-#define LHCALL_SET_PMD		15
-#define LHCALL_LOAD_TLS		16
+#include <asm/lguest_hcall.h>
 
 #define LG_CLOCK_MIN_DELTA	100UL
 #define LG_CLOCK_MAX_DELTA	ULONG_MAX
 
-/*G:031 First, how does our Guest contact the Host to ask for privileged
- * operations?  There are two ways: the direct way is to make a "hypercall",
- * to make requests of the Host Itself.
- *
- * Our hypercall mechanism uses the highest unused trap code (traps 32 and
- * above are used by real hardware interrupts).  Seventeen hypercalls are
- * available: the hypercall number is put in the %eax register, and the
- * arguments (when required) are placed in %edx, %ebx and %ecx.  If a return
- * value makes sense, it's returned in %eax.
- *
- * Grossly invalid calls result in Sudden Death at the hands of the vengeful
- * Host, rather than returning failure.  This reflects Winston Churchill's
- * definition of a gentleman: "someone who is only rude intentionally". */
-#define LGUEST_TRAP_ENTRY 0x1F
-
-static inline unsigned long
-hcall(unsigned long call,
-      unsigned long arg1, unsigned long arg2, unsigned long arg3)
-{
-	/* "int" is the Intel instruction to trigger a trap. */
-	asm volatile("int $" __stringify(LGUEST_TRAP_ENTRY)
-		       /* The call is in %eax (aka "a"), and can be replaced */
-		     : "=a"(call)
-		       /* The other arguments are in %eax, %edx, %ebx & %ecx */
-		     : "a"(call), "d"(arg1), "b"(arg2), "c"(arg3)
-		       /* "memory" means this might write somewhere in memory.
-			* This isn't true for all calls, but it's safe to tell
-			* gcc that it might happen so it doesn't get clever. */
-		     : "memory");
-	return call;
-}
-/*:*/
-
-void async_hcall(unsigned long call,
-		 unsigned long arg1, unsigned long arg2, unsigned long arg3);
-
-/* Can't use our min() macro here: needs to be a constant */
-#define LGUEST_IRQS (NR_IRQS < 32 ? NR_IRQS: 32)
-
-#define LHCALL_RING_SIZE 64
-struct hcall_ring
-{
-	u32 eax, edx, ebx, ecx;
-};
-
 /*G:032 The second method of communicating with the Host is to via "struct
  * lguest_data".  The Guest's very first hypercall is to tell the Host where
  * this is, and then the Guest and Host both publish information in it. :*/
@@ -97,20 +37,24 @@ struct lguest_data
 	/* 0xFF == done (set by Host), 0 == pending (set by Guest). */
 	u8 hcall_status[LHCALL_RING_SIZE];
 	/* The actual registers for the hypercalls. */
-	struct hcall_ring hcalls[LHCALL_RING_SIZE];
+	struct hcall_args hcalls[LHCALL_RING_SIZE];
 
 /* Fields initialized by the Host at boot: */
 	/* Memory not to try to access */
 	unsigned long reserve_mem;
-	/* ID of this Guest (used by network driver to set ethernet address) */
-	u16 guestid;
 	/* KHz for the TSC clock. */
 	u32 tsc_khz;
+	/* Page where the top-level pagetable is */
+	unsigned long pgdir;
 
 /* Fields initialized by the Guest at boot: */
 	/* Instruction range to suppress interrupts even if enabled */
 	unsigned long noirq_start, noirq_end;
+	/* Address above which page tables are all identical. */
+	unsigned long kernel_address;
+	/* The vector to try to use for system calls (0x40 or 0x80). */
+	unsigned int syscall_vec;
 };
 extern struct lguest_data lguest_data;
 #endif /* __ASSEMBLY__ */
-#endif	/* _ASM_LGUEST_H */
+#endif	/* _LINUX_LGUEST_H */

include/linux/lguest_bus.h

@@ -1,51 +0,0 @@
-#ifndef _ASM_LGUEST_DEVICE_H
-#define _ASM_LGUEST_DEVICE_H
-/* Everything you need to know about lguest devices. */
-#include <linux/device.h>
-#include <linux/lguest.h>
-#include <linux/lguest_launcher.h>
-
-struct lguest_device {
-	/* Unique busid, and index into lguest_page->devices[] */
-	unsigned int index;
-
-	struct device dev;
-
-	/* Driver can hang data off here. */
-	void *private;
-};
-
-/*D:380 Since interrupt numbers are arbitrary, we use a convention: each device
- * can use the interrupt number corresponding to its index.  The +1 is because
- * interrupt 0 is not usable (it's actually the timer interrupt). */
-static inline int lgdev_irq(const struct lguest_device *dev)
-{
-	return dev->index + 1;
-}
-/*:*/
-
-/* dma args must not be vmalloced! */
-void lguest_send_dma(unsigned long key, struct lguest_dma *dma);
-int lguest_bind_dma(unsigned long key, struct lguest_dma *dmas,
-		    unsigned int num, u8 irq);
-void lguest_unbind_dma(unsigned long key, struct lguest_dma *dmas);
-
-/* Map the virtual device space */
-void *lguest_map(unsigned long phys_addr, unsigned long pages);
-void lguest_unmap(void *);
-
-struct lguest_driver {
-	const char *name;
-	struct module *owner;
-	u16 device_type;
-	int (*probe)(struct lguest_device *dev);
-	void (*remove)(struct lguest_device *dev);
-
-	struct device_driver drv;
-};
-
-extern int register_lguest_driver(struct lguest_driver *drv);
-extern void unregister_lguest_driver(struct lguest_driver *drv);
-
-extern struct lguest_device_desc *lguest_devices; /* Just past max_pfn */
-#endif /* _ASM_LGUEST_DEVICE_H */

include/linux/lguest_launcher.h

@@ -1,6 +1,7 @@
 #ifndef _ASM_LGUEST_USER
 #define _ASM_LGUEST_USER
 /* Everything the "lguest" userspace program needs to know. */
+#include <linux/types.h>
 /* They can register up to 32 arrays of lguest_dma. */
 #define LGUEST_MAX_DMA		32
 /* At most we can dma 16 lguest_dma in one op. */
@@ -9,66 +10,6 @@
 /* How many devices?  Assume each one wants up to two dma arrays per device. */
 #define LGUEST_MAX_DEVICES (LGUEST_MAX_DMA/2)
 
-/*D:200
- * Lguest I/O
- *
- * The lguest I/O mechanism is the only way Guests can talk to devices.  There
- * are two hypercalls involved: SEND_DMA for output and BIND_DMA for input.  In
- * each case, "struct lguest_dma" describes the buffer: this contains 16
- * addr/len pairs, and if there are fewer buffer elements the len array is
- * terminated with a 0.
- *
- * I/O is organized by keys: BIND_DMA attaches buffers to a particular key, and
- * SEND_DMA transfers to buffers bound to particular key.  By convention, keys
- * correspond to a physical address within the device's page.  This means that
- * devices will never accidentally end up with the same keys, and allows the
- * Host use The Futex Trick (as we'll see later in our journey).
- *
- * SEND_DMA simply indicates a key to send to, and the physical address of the
- * "struct lguest_dma" to send.  The Host will write the number of bytes
- * transferred into the "struct lguest_dma"'s used_len member.
- *
- * BIND_DMA indicates a key to bind to, a pointer to an array of "struct
- * lguest_dma"s ready for receiving, the size of that array, and an interrupt
- * to trigger when data is received.  The Host will only allow transfers into
- * buffers with a used_len of zero: it then sets used_len to the number of
- * bytes transferred and triggers the interrupt for the Guest to process the
- * new input. */
-struct lguest_dma
-{
-	/* 0 if free to be used, filled by the Host. */
- 	u32 used_len;
-	unsigned long addr[LGUEST_MAX_DMA_SECTIONS];
-	u16 len[LGUEST_MAX_DMA_SECTIONS];
-};
-/*:*/
-
-/*D:460 This is the layout of a block device memory page.  The Launcher sets up
- * the num_sectors initially to tell the Guest the size of the disk.  The Guest
- * puts the type, sector and length of the request in the first three fields,
- * then DMAs to the Host.  The Host processes the request, sets up the result,
- * then DMAs back to the Guest. */
-struct lguest_block_page
-{
-	/* 0 is a read, 1 is a write. */
-	int type;
-	u32 sector; 	/* Offset in device = sector * 512. */
-	u32 bytes;	/* Length expected to be read/written in bytes */
-	/* 0 = pending, 1 = done, 2 = done, error */
-	int result;
-	u32 num_sectors; /* Disk length = num_sectors * 512 */
-};
-
-/*D:520 The network device is basically a memory page where all the Guests on
- * the network publish their MAC (ethernet) addresses: it's an array of "struct
- * lguest_net": */
-struct lguest_net
-{
-	/* Simply the mac address (with multicast bit meaning promisc). */
-	unsigned char mac[6];
-};
-/*:*/
-
 /* Where the Host expects the Guest to SEND_DMA console output to. */
 #define LGUEST_CONSOLE_DMA_KEY 0
 
@@ -81,38 +22,29 @@ struct lguest_net
  * complex burden for the Host and suboptimal for the Guest, so we have our own
  * "lguest" bus and simple drivers.
  *
- * Devices are described by an array of LGUEST_MAX_DEVICES of these structs,
- * placed by the Launcher just above the top of physical memory:
+ * Devices are described by a simplified ID, a status byte, and some "config"
+ * bytes which describe this device's configuration.  This is placed by the
+ * Launcher just above the top of physical memory:
  */
 struct lguest_device_desc {
-	/* The device type: console, network, disk etc. */
-	u16 type;
-#define LGUEST_DEVICE_T_CONSOLE	1
-#define LGUEST_DEVICE_T_NET	2
-#define LGUEST_DEVICE_T_BLOCK	3
-
-	/* The specific features of this device: these depends on device type
-	 * except for LGUEST_DEVICE_F_RANDOMNESS. */
-	u16 features;
-#define LGUEST_NET_F_NOCSUM		0x4000 /* Don't bother checksumming */
-#define LGUEST_DEVICE_F_RANDOMNESS	0x8000 /* IRQ is fairly random */
-
-	/* This is how the Guest reports status of the device: the Host can set
-	 * LGUEST_DEVICE_S_REMOVED to indicate removal, but the rest are only
-	 * ever manipulated by the Guest, and only ever set. */
-	u16 status;
-/* 256 and above are device specific. */
-#define LGUEST_DEVICE_S_ACKNOWLEDGE	1 /* We have seen device. */
-#define LGUEST_DEVICE_S_DRIVER		2 /* We have found a driver */
-#define LGUEST_DEVICE_S_DRIVER_OK	4 /* Driver says OK! */
-#define LGUEST_DEVICE_S_REMOVED		8 /* Device has gone away. */
-#define LGUEST_DEVICE_S_REMOVED_ACK	16 /* Driver has been told. */
-#define LGUEST_DEVICE_S_FAILED		128 /* Something actually failed */
+	/* The device type: console, network, disk etc.  Type 0 terminates. */
+	__u8 type;
+	/* The number of bytes of the config array. */
+	__u8 config_len;
+	/* A status byte, written by the Guest. */
+	__u8 status;
+	__u8 config[0];
+};
 
-	/* Each device exists somewhere in Guest physical memory, over some
-	 * number of pages. */
-	u16 num_pages;
-	u32 pfn;
+/*D:135 This is how we expect the device configuration field for a virtqueue
+ * (type VIRTIO_CONFIG_F_VIRTQUEUE) to be laid out: */
+struct lguest_vqconfig {
+	/* The number of entries in the virtio_ring */
+	__u16 num;
+	/* The interrupt we get when something happens. */
+	__u16 irq;
+	/* The page number of the virtio ring for this device. */
+	__u32 pfn;
 };
 /*:*/
 
@@ -120,7 +52,7 @@ struct lguest_device_desc {
 enum lguest_req
 {
 	LHREQ_INITIALIZE, /* + pfnlimit, pgdir, start, pageoffset */
-	LHREQ_GETDMA, /* + addr (returns &lguest_dma, irq in ->used_len) */
+	LHREQ_GETDMA, /* No longer used */
 	LHREQ_IRQ, /* + irq */
 	LHREQ_BREAK, /* + on/off flag (on blocks until someone does off) */
 };

include/linux/mod_devicetable.h

@@ -361,4 +361,10 @@ struct ssb_device_id {
 #define SSB_ANY_ID		0xFFFF
 #define SSB_ANY_REV		0xFF
 
+struct virtio_device_id {
+	__u32 device;
+	__u32 vendor;
+};
+#define VIRTIO_DEV_ANY_ID	0xffffffff
+
 #endif /* LINUX_MOD_DEVICETABLE_H */

include/linux/screen_info.h

@@ -8,45 +8,43 @@
  */
 
 struct screen_info {
-	u8  orig_x;		/* 0x00 */
-	u8  orig_y;		/* 0x01 */
-	u16 ext_mem_k;		/* 0x02 */
-	u16 orig_video_page;	/* 0x04 */
-	u8  orig_video_mode;	/* 0x06 */
-	u8  orig_video_cols;	/* 0x07 */
-	u16 unused2;		/* 0x08 */
-	u16 orig_video_ega_bx;	/* 0x0a */
-	u16 unused3;		/* 0x0c */
-	u8  orig_video_lines;	/* 0x0e */
-	u8  orig_video_isVGA;	/* 0x0f */
-	u16 orig_video_points;	/* 0x10 */
+	__u8  orig_x;		/* 0x00 */
+	__u8  orig_y;		/* 0x01 */
+	__u16 ext_mem_k;	/* 0x02 */
+	__u16 orig_video_page;	/* 0x04 */
+	__u8  orig_video_mode;	/* 0x06 */
+	__u8  orig_video_cols;	/* 0x07 */
+	__u16 unused2;		/* 0x08 */
+	__u16 orig_video_ega_bx;/* 0x0a */
+	__u16 unused3;		/* 0x0c */
+	__u8  orig_video_lines;	/* 0x0e */
+	__u8  orig_video_isVGA;	/* 0x0f */
+	__u16 orig_video_points;/* 0x10 */
 
 	/* VESA graphic mode -- linear frame buffer */
-	u16 lfb_width;		/* 0x12 */
-	u16 lfb_height;		/* 0x14 */
-	u16 lfb_depth;		/* 0x16 */
-	u32 lfb_base;		/* 0x18 */
-	u32 lfb_size;		/* 0x1c */
-	u16 cl_magic, cl_offset; /* 0x20 */
-	u16 lfb_linelength;	/* 0x24 */
-	u8  red_size;		/* 0x26 */
-	u8  red_pos;		/* 0x27 */
-	u8  green_size;		/* 0x28 */
-	u8  green_pos;		/* 0x29 */
-	u8  blue_size;		/* 0x2a */
-	u8  blue_pos;		/* 0x2b */
-	u8  rsvd_size;		/* 0x2c */
-	u8  rsvd_pos;		/* 0x2d */
-	u16 vesapm_seg;		/* 0x2e */
-	u16 vesapm_off;		/* 0x30 */
-	u16 pages;		/* 0x32 */
-	u16 vesa_attributes;	/* 0x34 */
-	u32 capabilities;       /* 0x36 */
-	u8  _reserved[6];	/* 0x3a */
+	__u16 lfb_width;	/* 0x12 */
+	__u16 lfb_height;	/* 0x14 */
+	__u16 lfb_depth;	/* 0x16 */
+	__u32 lfb_base;		/* 0x18 */
+	__u32 lfb_size;		/* 0x1c */
+	__u16 cl_magic, cl_offset; /* 0x20 */
+	__u16 lfb_linelength;	/* 0x24 */
+	__u8  red_size;		/* 0x26 */
+	__u8  red_pos;		/* 0x27 */
+	__u8  green_size;	/* 0x28 */
+	__u8  green_pos;	/* 0x29 */
+	__u8  blue_size;	/* 0x2a */
+	__u8  blue_pos;		/* 0x2b */
+	__u8  rsvd_size;	/* 0x2c */
+	__u8  rsvd_pos;		/* 0x2d */
+	__u16 vesapm_seg;	/* 0x2e */
+	__u16 vesapm_off;	/* 0x30 */
+	__u16 pages;		/* 0x32 */
+	__u16 vesa_attributes;	/* 0x34 */
+	__u32 capabilities;     /* 0x36 */
+	__u8  _reserved[6];	/* 0x3a */
 } __attribute__((packed));
 
-extern struct screen_info screen_info;
-
 #define VIDEO_TYPE_MDA		0x10	/* Monochrome Text Display	*/
 #define VIDEO_TYPE_CGA		0x11	/* CGA Display 			*/
 #define VIDEO_TYPE_EGAM		0x20	/* EGA/VGA in Monochrome Mode	*/
@@ -65,4 +63,17 @@ extern struct screen_info screen_info;
 
 #define VIDEO_TYPE_PMAC		0x60	/* PowerMacintosh frame buffer. */
 
+#ifdef __KERNEL__
+extern struct screen_info screen_info;
+
+#define ORIG_X			(screen_info.orig_x)
+#define ORIG_Y			(screen_info.orig_y)
+#define ORIG_VIDEO_MODE		(screen_info.orig_video_mode)
+#define ORIG_VIDEO_COLS 	(screen_info.orig_video_cols)
+#define ORIG_VIDEO_EGA_BX	(screen_info.orig_video_ega_bx)
+#define ORIG_VIDEO_LINES	(screen_info.orig_video_lines)
+#define ORIG_VIDEO_ISVGA	(screen_info.orig_video_isVGA)
+#define ORIG_VIDEO_POINTS       (screen_info.orig_video_points)
+#endif /* __KERNEL__ */
+
 #endif /* _SCREEN_INFO_H */

include/linux/virtio.h

@@ -0,0 +1,110 @@
+#ifndef _LINUX_VIRTIO_H
+#define _LINUX_VIRTIO_H
+/* Everything a virtio driver needs to work with any particular virtio
+ * implementation. */
+#include <linux/types.h>
+#include <linux/scatterlist.h>
+#include <linux/spinlock.h>
+#include <linux/device.h>
+#include <linux/mod_devicetable.h>
+
+/**
+ * virtqueue - a queue to register buffers for sending or receiving.
+ * @callback: the function to call when buffers are consumed (can be NULL).
+ *    If this returns false, callbacks are suppressed until vq_ops->restart
+ *    is called.
+ * @vdev: the virtio device this queue was created for.
+ * @vq_ops: the operations for this virtqueue (see below).
+ * @priv: a pointer for the virtqueue implementation to use.
+ */
+struct virtqueue
+{
+	bool (*callback)(struct virtqueue *vq);
+	struct virtio_device *vdev;
+	struct virtqueue_ops *vq_ops;
+	void *priv;
+};
+
+/**
+ * virtqueue_ops - operations for virtqueue abstraction layer
+ * @add_buf: expose buffer to other end
+ *	vq: the struct virtqueue we're talking about.
+ *	sg: the description of the buffer(s).
+ *	out_num: the number of sg readable by other side
+ *	in_num: the number of sg which are writable (after readable ones)
+ *	data: the token identifying the buffer.
+ *      Returns 0 or an error.
+ * @kick: update after add_buf
+ *	vq: the struct virtqueue
+ *	After one or more add_buf calls, invoke this to kick the other side.
+ * @get_buf: get the next used buffer
+ *	vq: the struct virtqueue we're talking about.
+ *	len: the length written into the buffer
+ *	Returns NULL or the "data" token handed to add_buf.
+ * @restart: restart callbacks after callback returned false.
+ *	vq: the struct virtqueue we're talking about.
+ *	This returns "false" (and doesn't re-enable) if there are pending
+ *	buffers in the queue, to avoid a race.
+ * @shutdown: "unadd" all buffers.
+ *	vq: the struct virtqueue we're talking about.
+ *	Remove everything from the queue.
+ *
+ * Locking rules are straightforward: the driver is responsible for
+ * locking.  No two operations may be invoked simultaneously.
+ *
+ * All operations can be called in any context.
+ */
+struct virtqueue_ops {
+	int (*add_buf)(struct virtqueue *vq,
+		       struct scatterlist sg[],
+		       unsigned int out_num,
+		       unsigned int in_num,
+		       void *data);
+
+	void (*kick)(struct virtqueue *vq);
+
+	void *(*get_buf)(struct virtqueue *vq, unsigned int *len);
+
+	bool (*restart)(struct virtqueue *vq);
+
+	void (*shutdown)(struct virtqueue *vq);
+};
+
+/**
+ * virtio_device - representation of a device using virtio
+ * @index: unique position on the virtio bus
+ * @dev: underlying device.
+ * @id: the device type identification (used to match it with a driver).
+ * @config: the configuration ops for this device.
+ * @priv: private pointer for the driver's use.
+ */
+struct virtio_device
+{
+	int index;
+	struct device dev;
+	struct virtio_device_id id;
+	struct virtio_config_ops *config;
+	void *priv;
+};
+
+int register_virtio_device(struct virtio_device *dev);
+void unregister_virtio_device(struct virtio_device *dev);
+
+/**
+ * virtio_driver - operations for a virtio I/O driver
+ * @driver: underlying device driver (populate name and owner).
+ * @id_table: the ids serviced by this driver.
+ * @probe: the function to call when a device is found.  Returns a token for
+ *    remove, or PTR_ERR().
+ * @remove: the function when a device is removed.
+ */
+struct virtio_driver {
+	struct device_driver driver;
+	const struct virtio_device_id *id_table;
+	int (*probe)(struct virtio_device *dev);
+	void (*remove)(struct virtio_device *dev);
+};
+
+int register_virtio_driver(struct virtio_driver *drv);
+void unregister_virtio_driver(struct virtio_driver *drv);
+#endif /* _LINUX_VIRTIO_H */

include/linux/virtio_blk.h

@@ -0,0 +1,51 @@
+#ifndef _LINUX_VIRTIO_BLK_H
+#define _LINUX_VIRTIO_BLK_H
+#include <linux/virtio_config.h>
+
+/* The ID for virtio_block */
+#define VIRTIO_ID_BLOCK	2
+
+/* Feature bits */
+#define VIRTIO_CONFIG_BLK_F	0x40
+#define VIRTIO_BLK_F_BARRIER	1	/* Does host support barriers? */
+
+/* The capacity (in 512-byte sectors). */
+#define VIRTIO_CONFIG_BLK_F_CAPACITY	0x41
+/* The maximum segment size. */
+#define VIRTIO_CONFIG_BLK_F_SIZE_MAX	0x42
+/* The maximum number of segments. */
+#define VIRTIO_CONFIG_BLK_F_SEG_MAX	0x43
+
+/* These two define direction. */
+#define VIRTIO_BLK_T_IN		0
+#define VIRTIO_BLK_T_OUT	1
+
+/* This bit says it's a scsi command, not an actual read or write. */
+#define VIRTIO_BLK_T_SCSI_CMD	2
+
+/* Barrier before this op. */
+#define VIRTIO_BLK_T_BARRIER	0x80000000
+
+/* This is the first element of the read scatter-gather list. */
+struct virtio_blk_outhdr
+{
+	/* VIRTIO_BLK_T* */
+	__u32 type;
+	/* io priority. */
+	__u32 ioprio;
+	/* Sector (ie. 512 byte offset) */
+	__u64 sector;
+	/* Where to put reply. */
+	__u64 id;
+};
+
+#define VIRTIO_BLK_S_OK		0
+#define VIRTIO_BLK_S_IOERR	1
+#define VIRTIO_BLK_S_UNSUPP	2
+
+/* This is the first element of the write scatter-gather list */
+struct virtio_blk_inhdr
+{
+	unsigned char status;
+};
+#endif /* _LINUX_VIRTIO_BLK_H */

include/linux/virtio_config.h

@@ -0,0 +1,111 @@
+#ifndef _LINUX_VIRTIO_CONFIG_H
+#define _LINUX_VIRTIO_CONFIG_H
+/* Virtio devices use a standardized configuration space to define their
+ * features and pass configuration information, but each implementation can
+ * store and access that space differently. */
+#include <linux/types.h>
+
+/* Status byte for guest to report progress, and synchronize config. */
+/* We have seen device and processed generic fields (VIRTIO_CONFIG_F_VIRTIO) */
+#define VIRTIO_CONFIG_S_ACKNOWLEDGE	1
+/* We have found a driver for the device. */
+#define VIRTIO_CONFIG_S_DRIVER		2
+/* Driver has used its parts of the config, and is happy */
+#define VIRTIO_CONFIG_S_DRIVER_OK	4
+/* We've given up on this device. */
+#define VIRTIO_CONFIG_S_FAILED		0x80
+
+/* Feature byte (actually 7 bits availabe): */
+/* Requirements/features of the virtio implementation. */
+#define VIRTIO_CONFIG_F_VIRTIO 1
+/* Requirements/features of the virtqueue (may have more than one). */
+#define VIRTIO_CONFIG_F_VIRTQUEUE 2
+
+#ifdef __KERNEL__
+struct virtio_device;
+
+/**
+ * virtio_config_ops - operations for configuring a virtio device
+ * @find: search for the next configuration field of the given type.
+ *	vdev: the virtio_device
+ *	type: the feature type
+ *	len: the (returned) length of the field if found.
+ *	Returns a token if found, or NULL.  Never returnes the same field twice
+ *	(ie. it's used up).
+ * @get: read the value of a configuration field after find().
+ *	vdev: the virtio_device
+ *	token: the token returned from find().
+ *	buf: the buffer to write the field value into.
+ *	len: the length of the buffer (given by find()).
+ *	Note that contents are conventionally little-endian.
+ * @set: write the value of a configuration field after find().
+ *	vdev: the virtio_device
+ *	token: the token returned from find().
+ *	buf: the buffer to read the field value from.
+ *	len: the length of the buffer (given by find()).
+ *	Note that contents are conventionally little-endian.
+ * @get_status: read the status byte
+ *	vdev: the virtio_device
+ *	Returns the status byte
+ * @set_status: write the status byte
+ *	vdev: the virtio_device
+ *	status: the new status byte
+ * @find_vq: find the first VIRTIO_CONFIG_F_VIRTQUEUE and create a virtqueue.
+ *	vdev: the virtio_device
+ *	callback: the virqtueue callback
+ *	Returns the new virtqueue or ERR_PTR().
+ * @del_vq: free a virtqueue found by find_vq().
+ */
+struct virtio_config_ops
+{
+	void *(*find)(struct virtio_device *vdev, u8 type, unsigned *len);
+	void (*get)(struct virtio_device *vdev, void *token,
+		    void *buf, unsigned len);
+	void (*set)(struct virtio_device *vdev, void *token,
+		    const void *buf, unsigned len);
+	u8 (*get_status)(struct virtio_device *vdev);
+	void (*set_status)(struct virtio_device *vdev, u8 status);
+	struct virtqueue *(*find_vq)(struct virtio_device *vdev,
+				     bool (*callback)(struct virtqueue *));
+	void (*del_vq)(struct virtqueue *vq);
+};
+
+/**
+ * virtio_config_val - get a single virtio config and mark it used.
+ * @config: the virtio config space
+ * @type: the type to search for.
+ * @val: a pointer to the value to fill in.
+ *
+ * Once used, the config type is marked with VIRTIO_CONFIG_F_USED so it can't
+ * be found again.  This version does endian conversion. */
+#define virtio_config_val(vdev, type, v) ({				\
+	int _err = __virtio_config_val((vdev),(type),(v),sizeof(*(v))); \
+									\
+	BUILD_BUG_ON(sizeof(*(v)) != 1 && sizeof(*(v)) != 2		\
+		     && sizeof(*(v)) != 4 && sizeof(*(v)) != 8);	\
+	if (!_err) {							\
+		switch (sizeof(*(v))) {					\
+		case 2: le16_to_cpus((__u16 *) v); break;		\
+		case 4: le32_to_cpus((__u32 *) v); break;		\
+		case 8: le64_to_cpus((__u64 *) v); break;		\
+		}							\
+	}								\
+	_err;								\
+})
+
+int __virtio_config_val(struct virtio_device *dev,
+			u8 type, void *val, size_t size);
+
+/**
+ * virtio_use_bit - helper to use a feature bit in a bitfield value.
+ * @dev: the virtio device
+ * @token: the token as returned from vdev->config->find().
+ * @len: the length of the field.
+ * @bitnum: the bit to test.
+ *
+ * If handed a NULL token, it returns false, otherwise returns bit status.
+ * If it's one, it sets the mirroring acknowledgement bit. */
+int virtio_use_bit(struct virtio_device *vdev,
+		   void *token, unsigned int len, unsigned int bitnum);
+#endif /* __KERNEL__ */
+#endif /* _LINUX_VIRTIO_CONFIG_H */

include/linux/virtio_console.h

@@ -0,0 +1,12 @@
+#ifndef _LINUX_VIRTIO_CONSOLE_H
+#define _LINUX_VIRTIO_CONSOLE_H
+#include <linux/virtio_config.h>
+
+/* The ID for virtio console */
+#define VIRTIO_ID_CONSOLE	3
+
+#ifdef __KERNEL__
+int __init virtio_cons_early_init(int (*put_chars)(u32, const char *, int));
+#endif /* __KERNEL__ */
+
+#endif /* _LINUX_VIRTIO_CONSOLE_H */

include/linux/virtio_net.h

@@ -0,0 +1,36 @@
+#ifndef _LINUX_VIRTIO_NET_H
+#define _LINUX_VIRTIO_NET_H
+#include <linux/virtio_config.h>
+
+/* The ID for virtio_net */
+#define VIRTIO_ID_NET	1
+
+/* The bitmap of config for virtio net */
+#define VIRTIO_CONFIG_NET_F	0x40
+#define VIRTIO_NET_F_NO_CSUM	0
+#define VIRTIO_NET_F_TSO4	1
+#define VIRTIO_NET_F_UFO	2
+#define VIRTIO_NET_F_TSO4_ECN	3
+#define VIRTIO_NET_F_TSO6	4
+
+/* The config defining mac address. */
+#define VIRTIO_CONFIG_NET_MAC_F	0x41
+
+/* This is the first element of the scatter-gather list.  If you don't
+ * specify GSO or CSUM features, you can simply ignore the header. */
+struct virtio_net_hdr
+{
+#define VIRTIO_NET_HDR_F_NEEDS_CSUM	1	// Use csum_start, csum_offset
+      __u8 flags;
+#define VIRTIO_NET_HDR_GSO_NONE		0	// Not a GSO frame
+#define VIRTIO_NET_HDR_GSO_TCPV4	1	// GSO frame, IPv4 TCP (TSO)
+/* FIXME: Do we need this?  If they said they can handle ECN, do they care? */
+#define VIRTIO_NET_HDR_GSO_TCPV4_ECN	2	// GSO frame, IPv4 TCP w/ ECN
+#define VIRTIO_NET_HDR_GSO_UDP		3	// GSO frame, IPv4 UDP (UFO)
+#define VIRTIO_NET_HDR_GSO_TCPV6	4	// GSO frame, IPv6 TCP
+      __u8 gso_type;
+      __u16 gso_size;
+      __u16 csum_start;
+      __u16 csum_offset;
+};
+#endif /* _LINUX_VIRTIO_NET_H */

include/linux/virtio_ring.h

@@ -0,0 +1,119 @@
+#ifndef _LINUX_VIRTIO_RING_H
+#define _LINUX_VIRTIO_RING_H
+/* An interface for efficient virtio implementation, currently for use by KVM
+ * and lguest, but hopefully others soon.  Do NOT change this since it will
+ * break existing servers and clients.
+ *
+ * This header is BSD licensed so anyone can use the definitions to implement
+ * compatible drivers/servers.
+ *
+ * Copyright Rusty Russell IBM Corporation 2007. */
+#include <linux/types.h>
+
+/* This marks a buffer as continuing via the next field. */
+#define VRING_DESC_F_NEXT	1
+/* This marks a buffer as write-only (otherwise read-only). */
+#define VRING_DESC_F_WRITE	2
+
+/* This means don't notify other side when buffer added. */
+#define VRING_USED_F_NO_NOTIFY	1
+/* This means don't interrupt guest when buffer consumed. */
+#define VRING_AVAIL_F_NO_INTERRUPT	1
+
+/* Virtio ring descriptors: 16 bytes.  These can chain together via "next". */
+struct vring_desc
+{
+	/* Address (guest-physical). */
+	__u64 addr;
+	/* Length. */
+	__u32 len;
+	/* The flags as indicated above. */
+	__u16 flags;
+	/* We chain unused descriptors via this, too */
+	__u16 next;
+};
+
+struct vring_avail
+{
+	__u16 flags;
+	__u16 idx;
+	__u16 ring[];
+};
+
+/* u32 is used here for ids for padding reasons. */
+struct vring_used_elem
+{
+	/* Index of start of used descriptor chain. */
+	__u32 id;
+	/* Total length of the descriptor chain which was used (written to) */
+	__u32 len;
+};
+
+struct vring_used
+{
+	__u16 flags;
+	__u16 idx;
+	struct vring_used_elem ring[];
+};
+
+struct vring {
+	unsigned int num;
+
+	struct vring_desc *desc;
+
+	struct vring_avail *avail;
+
+	struct vring_used *used;
+};
+
+/* The standard layout for the ring is a continuous chunk of memory which looks
+ * like this.  The used fields will be aligned to a "num+1" boundary.
+ *
+ * struct vring
+ * {
+ *	// The actual descriptors (16 bytes each)
+ *	struct vring_desc desc[num];
+ *
+ *	// A ring of available descriptor heads with free-running index.
+ *	__u16 avail_flags;
+ *	__u16 avail_idx;
+ *	__u16 available[num];
+ *
+ *	// Padding so a correctly-chosen num value will cache-align used_idx.
+ *	char pad[sizeof(struct vring_desc) - sizeof(avail_flags)];
+ *
+ *	// A ring of used descriptor heads with free-running index.
+ *	__u16 used_flags;
+ *	__u16 used_idx;
+ *	struct vring_used_elem used[num];
+ * };
+ */
+static inline void vring_init(struct vring *vr, unsigned int num, void *p)
+{
+	vr->num = num;
+	vr->desc = p;
+	vr->avail = p + num*sizeof(struct vring);
+	vr->used = p + (num+1)*(sizeof(struct vring) + sizeof(__u16));
+}
+
+static inline unsigned vring_size(unsigned int num)
+{
+	return (num + 1) * (sizeof(struct vring_desc) + sizeof(__u16))
+		+ sizeof(__u32) + num * sizeof(struct vring_used_elem);
+}
+
+#ifdef __KERNEL__
+#include <linux/irqreturn.h>
+struct virtio_device;
+struct virtqueue;
+
+struct virtqueue *vring_new_virtqueue(unsigned int num,
+				      struct virtio_device *vdev,
+				      void *pages,
+				      void (*notify)(struct virtqueue *vq),
+				      bool (*callback)(struct virtqueue *vq));
+void vring_del_virtqueue(struct virtqueue *vq);
+
+irqreturn_t vring_interrupt(int irq, void *_vq);
+#endif /* __KERNEL__ */
+#endif /* _LINUX_VIRTIO_RING_H */

include/video/Kbuild

@@ -1 +1,2 @@
 unifdef-y += sisfb.h uvesafb.h
+unifdef-y += edid.h

include/video/edid.h

@@ -1,17 +1,16 @@
 #ifndef __linux_video_edid_h__
 #define __linux_video_edid_h__
 
-#ifdef __KERNEL__
-
+#if !defined(__KERNEL__) || defined(CONFIG_X86)
 
-#ifdef CONFIG_X86
 struct edid_info {
 	unsigned char dummy[128];
 };
 
+#ifdef __KERNEL__
 extern struct edid_info edid_info;
-#endif /* CONFIG_X86 */
-
 #endif /* __KERNEL__ */
 
+#endif
+
 #endif /* __linux_video_edid_h__ */

scripts/mod/file2alias.c

@@ -525,6 +525,20 @@ static int do_ssb_entry(const char *filename,
 	return 1;
 }
 
+/* Looks like: virtio:dNvN */
+static int do_virtio_entry(const char *filename, struct virtio_device_id *id,
+			   char *alias)
+{
+	id->device = TO_NATIVE(id->device);
+	id->vendor = TO_NATIVE(id->vendor);
+
+	strcpy(alias, "virtio:");
+	ADD(alias, "d", 1, id->device);
+	ADD(alias, "v", id->vendor != VIRTIO_DEV_ANY_ID, id->vendor);
+
+	return 1;
+}
+
 /* Ignore any prefix, eg. v850 prepends _ */
 static inline int sym_is(const char *symbol, const char *name)
 {
@@ -651,6 +665,10 @@ void handle_moddevtable(struct module *mod, struct elf_info *info,
 		do_table(symval, sym->st_size,
 			 sizeof(struct ssb_device_id), "ssb",
 			 do_ssb_entry, mod);
+	else if (sym_is(symname, "__mod_virtio_device_table"))
+		do_table(symval, sym->st_size,
+			 sizeof(struct virtio_device_id), "virtio",
+			 do_virtio_entry, mod);
 	free(zeros);
 }