Merge branch 'irq/core-v6' of git://git.kernel.org/pub/scm/linux/kernel/git/frederic/linux-dynticks into irq/core

Pull hardirq and softirq nesting updates from Frederic Weisbecker,
which fix nesting related stack overruns such as:

  http://lkml.kernel.org/r/1378330796.4321.50.camel%40pasglop

Beyond being a fix, this series also optimizes and reorganizes arch
hardirq/softirq stack processing to be faster and more robust.

Signed-off-by: Ingo Molnar <mingo@kernel.org>

arch/Kconfig

@@ -390,6 +390,16 @@ config HAVE_UNDERSCORE_SYMBOL_PREFIX
 	  Some architectures generate an _ in front of C symbols; things like
 	  module loading and assembly files need to know about this.
 
+config HAVE_IRQ_EXIT_ON_IRQ_STACK
+	bool
+	help
+	  Architecture doesn't only execute the irq handler on the irq stack
+	  but also irq_exit(). This way we can process softirqs on this irq
+	  stack instead of switching to a new one when we call __do_softirq()
+	  in the end of an hardirq.
+	  This spares a stack switch and improves cache usage on softirq
+	  processing.
+
 #
 # ABI hall of shame
 #

arch/metag/kernel/irq.c

@@ -159,44 +159,30 @@ void irq_ctx_exit(int cpu)
 
 extern asmlinkage void __do_softirq(void);
 
-asmlinkage void do_softirq(void)
+void do_softirq_own_stack(void)
 {
-	unsigned long flags;
 	struct thread_info *curctx;
 	union irq_ctx *irqctx;
 	u32 *isp;
 
-	if (in_interrupt())
-		return;
-
-	local_irq_save(flags);
-
-	if (local_softirq_pending()) {
-		curctx = current_thread_info();
-		irqctx = softirq_ctx[smp_processor_id()];
-		irqctx->tinfo.task = curctx->task;
-
-		/* build the stack frame on the softirq stack */
-		isp = (u32 *) ((char *)irqctx + sizeof(struct thread_info));
-
-		asm volatile (
-			"MOV   D0.5,%0\n"
-			"SWAP  A0StP,D0.5\n"
-			"CALLR D1RtP,___do_softirq\n"
-			"MOV   A0StP,D0.5\n"
-			:
-			: "r" (isp)
-			: "memory", "cc", "D1Ar1", "D0Ar2", "D1Ar3", "D0Ar4",
-			  "D1Ar5", "D0Ar6", "D0Re0", "D1Re0", "D0.4", "D1RtP",
-			  "D0.5"
-			);
-		/*
-		 * Shouldn't happen, we returned above if in_interrupt():
-		 */
-		WARN_ON_ONCE(softirq_count());
-	}
-
-	local_irq_restore(flags);
+	curctx = current_thread_info();
+	irqctx = softirq_ctx[smp_processor_id()];
+	irqctx->tinfo.task = curctx->task;
+
+	/* build the stack frame on the softirq stack */
+	isp = (u32 *) ((char *)irqctx + sizeof(struct thread_info));
+
+	asm volatile (
+		"MOV   D0.5,%0\n"
+		"SWAP  A0StP,D0.5\n"
+		"CALLR D1RtP,___do_softirq\n"
+		"MOV   A0StP,D0.5\n"
+		:
+		: "r" (isp)
+		: "memory", "cc", "D1Ar1", "D0Ar2", "D1Ar3", "D0Ar4",
+		  "D1Ar5", "D0Ar6", "D0Re0", "D1Re0", "D0.4", "D1RtP",
+		  "D0.5"
+		);
 }
 #endif
 

arch/parisc/kernel/irq.c

@@ -499,22 +499,9 @@ static void execute_on_irq_stack(void *func, unsigned long param1)
 	*irq_stack_in_use = 1;
 }
 
-asmlinkage void do_softirq(void)
+void do_softirq_own_stack(void)
 {
-	__u32 pending;
-	unsigned long flags;
-
-	if (in_interrupt())
-		return;
-
-	local_irq_save(flags);
-
-	pending = local_softirq_pending();
-
-	if (pending)
-		execute_on_irq_stack(__do_softirq, 0);
-
-	local_irq_restore(flags);
+	execute_on_irq_stack(__do_softirq, 0);
 }
 #endif /* CONFIG_IRQSTACKS */
 

arch/powerpc/Kconfig

@@ -138,6 +138,7 @@ config PPC
 	select OLD_SIGSUSPEND
 	select OLD_SIGACTION if PPC32
 	select HAVE_DEBUG_STACKOVERFLOW
+	select HAVE_IRQ_EXIT_ON_IRQ_STACK
 
 config EARLY_PRINTK
 	bool

arch/powerpc/kernel/irq.c

@@ -593,7 +593,7 @@ void irq_ctx_init(void)
 	}
 }
 
-static inline void do_softirq_onstack(void)
+void do_softirq_own_stack(void)
 {
 	struct thread_info *curtp, *irqtp;
 
@@ -611,21 +611,6 @@ static inline void do_softirq_onstack(void)
 		set_bits(irqtp->flags, &curtp->flags);
 }
 
-void do_softirq(void)
-{
-	unsigned long flags;
-
-	if (in_interrupt())
-		return;
-
-	local_irq_save(flags);
-
-	if (local_softirq_pending())
-		do_softirq_onstack();
-
-	local_irq_restore(flags);
-}
-
 irq_hw_number_t virq_to_hw(unsigned int virq)
 {
 	struct irq_data *irq_data = irq_get_irq_data(virq);

arch/s390/kernel/irq.c

@@ -157,39 +157,29 @@ int arch_show_interrupts(struct seq_file *p, int prec)
 /*
  * Switch to the asynchronous interrupt stack for softirq execution.
  */
-asmlinkage void do_softirq(void)
+void do_softirq_own_stack(void)
 {
-	unsigned long flags, old, new;
-
-	if (in_interrupt())
-		return;
-
-	local_irq_save(flags);
-
-	if (local_softirq_pending()) {
-		/* Get current stack pointer. */
-		asm volatile("la %0,0(15)" : "=a" (old));
-		/* Check against async. stack address range. */
-		new = S390_lowcore.async_stack;
-		if (((new - old) >> (PAGE_SHIFT + THREAD_ORDER)) != 0) {
-			/* Need to switch to the async. stack. */
-			new -= STACK_FRAME_OVERHEAD;
-			((struct stack_frame *) new)->back_chain = old;
-
-			asm volatile("   la    15,0(%0)\n"
-				     "   basr  14,%2\n"
-				     "   la    15,0(%1)\n"
-				     : : "a" (new), "a" (old),
-				         "a" (__do_softirq)
-				     : "0", "1", "2", "3", "4", "5", "14",
-				       "cc", "memory" );
-		} else {
-			/* We are already on the async stack. */
-			__do_softirq();
-		}
+	unsigned long old, new;
+
+	/* Get current stack pointer. */
+	asm volatile("la %0,0(15)" : "=a" (old));
+	/* Check against async. stack address range. */
+	new = S390_lowcore.async_stack;
+	if (((new - old) >> (PAGE_SHIFT + THREAD_ORDER)) != 0) {
+		/* Need to switch to the async. stack. */
+		new -= STACK_FRAME_OVERHEAD;
+		((struct stack_frame *) new)->back_chain = old;
+		asm volatile("   la    15,0(%0)\n"
+			     "   basr  14,%2\n"
+			     "   la    15,0(%1)\n"
+			     : : "a" (new), "a" (old),
+			         "a" (__do_softirq)
+			     : "0", "1", "2", "3", "4", "5", "14",
+			       "cc", "memory" );
+	} else {
+		/* We are already on the async stack. */
+		__do_softirq();
 	}
-
-	local_irq_restore(flags);
 }
 
 /*

arch/sh/kernel/irq.c

@@ -149,47 +149,32 @@ void irq_ctx_exit(int cpu)
 	hardirq_ctx[cpu] = NULL;
 }
 
-asmlinkage void do_softirq(void)
+void do_softirq_own_stack(void)
 {
-	unsigned long flags;
 	struct thread_info *curctx;
 	union irq_ctx *irqctx;
 	u32 *isp;
 
-	if (in_interrupt())
-		return;
-
-	local_irq_save(flags);
-
-	if (local_softirq_pending()) {
-		curctx = current_thread_info();
-		irqctx = softirq_ctx[smp_processor_id()];
-		irqctx->tinfo.task = curctx->task;
-		irqctx->tinfo.previous_sp = current_stack_pointer;
-
-		/* build the stack frame on the softirq stack */
-		isp = (u32 *)((char *)irqctx + sizeof(*irqctx));
-
-		__asm__ __volatile__ (
-			"mov	r15, r9		\n"
-			"jsr	@%0		\n"
-			/* switch to the softirq stack */
-			" mov	%1, r15		\n"
-			/* restore the thread stack */
-			"mov	r9, r15		\n"
-			: /* no outputs */
-			: "r" (__do_softirq), "r" (isp)
-			: "memory", "r0", "r1", "r2", "r3", "r4",
-			  "r5", "r6", "r7", "r8", "r9", "r15", "t", "pr"
-		);
-
-		/*
-		 * Shouldn't happen, we returned above if in_interrupt():
-		 */
-		WARN_ON_ONCE(softirq_count());
-	}
-
-	local_irq_restore(flags);
+	curctx = current_thread_info();
+	irqctx = softirq_ctx[smp_processor_id()];
+	irqctx->tinfo.task = curctx->task;
+	irqctx->tinfo.previous_sp = current_stack_pointer;
+
+	/* build the stack frame on the softirq stack */
+	isp = (u32 *)((char *)irqctx + sizeof(*irqctx));
+
+	__asm__ __volatile__ (
+		"mov	r15, r9		\n"
+		"jsr	@%0		\n"
+		/* switch to the softirq stack */
+		" mov	%1, r15		\n"
+		/* restore the thread stack */
+		"mov	r9, r15		\n"
+		: /* no outputs */
+		: "r" (__do_softirq), "r" (isp)
+		: "memory", "r0", "r1", "r2", "r3", "r4",
+		  "r5", "r6", "r7", "r8", "r9", "r15", "t", "pr"
+	);
 }
 #else
 static inline void handle_one_irq(unsigned int irq)

arch/sparc/kernel/irq_64.c

@@ -698,30 +698,19 @@ void __irq_entry handler_irq(int pil, struct pt_regs *regs)
 	set_irq_regs(old_regs);
 }
 
-void do_softirq(void)
+void do_softirq_own_stack(void)
 {
-	unsigned long flags;
-
-	if (in_interrupt())
-		return;
-
-	local_irq_save(flags);
+	void *orig_sp, *sp = softirq_stack[smp_processor_id()];
 
-	if (local_softirq_pending()) {
-		void *orig_sp, *sp = softirq_stack[smp_processor_id()];
-
-		sp += THREAD_SIZE - 192 - STACK_BIAS;
-
-		__asm__ __volatile__("mov %%sp, %0\n\t"
-				     "mov %1, %%sp"
-				     : "=&r" (orig_sp)
-				     : "r" (sp));
-		__do_softirq();
-		__asm__ __volatile__("mov %0, %%sp"
-				     : : "r" (orig_sp));
-	}
+	sp += THREAD_SIZE - 192 - STACK_BIAS;
 
-	local_irq_restore(flags);
+	__asm__ __volatile__("mov %%sp, %0\n\t"
+			     "mov %1, %%sp"
+			     : "=&r" (orig_sp)
+			     : "r" (sp));
+	__do_softirq();
+	__asm__ __volatile__("mov %0, %%sp"
+			     : : "r" (orig_sp));
 }
 
 #ifdef CONFIG_HOTPLUG_CPU

arch/x86/Kconfig

@@ -123,6 +123,7 @@ config X86
 	select COMPAT_OLD_SIGACTION if IA32_EMULATION
 	select RTC_LIB
 	select HAVE_DEBUG_STACKOVERFLOW
+	select HAVE_IRQ_EXIT_ON_IRQ_STACK if X86_64
 
 config INSTRUCTION_DECODER
 	def_bool y

arch/x86/kernel/entry_64.S

@@ -1342,7 +1342,7 @@ bad_gs:
 	.previous
 
 /* Call softirq on interrupt stack. Interrupts are off. */
-ENTRY(call_softirq)
+ENTRY(do_softirq_own_stack)
 	CFI_STARTPROC
 	pushq_cfi %rbp
 	CFI_REL_OFFSET rbp,0
@@ -1359,7 +1359,7 @@ ENTRY(call_softirq)
 	decl PER_CPU_VAR(irq_count)
 	ret
 	CFI_ENDPROC
-END(call_softirq)
+END(do_softirq_own_stack)
 
 #ifdef CONFIG_XEN
 zeroentry xen_hypervisor_callback xen_do_hypervisor_callback

arch/x86/kernel/irq_32.c

@@ -149,35 +149,21 @@ void irq_ctx_init(int cpu)
 	       cpu, per_cpu(hardirq_ctx, cpu),  per_cpu(softirq_ctx, cpu));
 }
 
-asmlinkage void do_softirq(void)
+void do_softirq_own_stack(void)
 {
-	unsigned long flags;
 	struct thread_info *curctx;
 	union irq_ctx *irqctx;
 	u32 *isp;
 
-	if (in_interrupt())
-		return;
-
-	local_irq_save(flags);
-
-	if (local_softirq_pending()) {
-		curctx = current_thread_info();
-		irqctx = __this_cpu_read(softirq_ctx);
-		irqctx->tinfo.task = curctx->task;
-		irqctx->tinfo.previous_esp = current_stack_pointer;
-
-		/* build the stack frame on the softirq stack */
-		isp = (u32 *) ((char *)irqctx + sizeof(*irqctx));
+	curctx = current_thread_info();
+	irqctx = __this_cpu_read(softirq_ctx);
+	irqctx->tinfo.task = curctx->task;
+	irqctx->tinfo.previous_esp = current_stack_pointer;
 
-		call_on_stack(__do_softirq, isp);
-		/*
-		 * Shouldn't happen, we returned above if in_interrupt():
-		 */
-		WARN_ON_ONCE(softirq_count());
-	}
+	/* build the stack frame on the softirq stack */
+	isp = (u32 *) ((char *)irqctx + sizeof(*irqctx));
 
-	local_irq_restore(flags);
+	call_on_stack(__do_softirq, isp);
 }
 
 bool handle_irq(unsigned irq, struct pt_regs *regs)

arch/x86/kernel/irq_64.c

@@ -87,24 +87,3 @@ bool handle_irq(unsigned irq, struct pt_regs *regs)
 	generic_handle_irq_desc(irq, desc);
 	return true;
 }
-
-
-extern void call_softirq(void);
-
-asmlinkage void do_softirq(void)
-{
-	__u32 pending;
-	unsigned long flags;
-
-	if (in_interrupt())
-		return;
-
-	local_irq_save(flags);
-	pending = local_softirq_pending();
-	/* Switch to interrupt stack */
-	if (pending) {
-		call_softirq();
-		WARN_ON_ONCE(softirq_count());
-	}
-	local_irq_restore(flags);
-}

include/linux/interrupt.h

@@ -19,6 +19,7 @@
 
 #include <linux/atomic.h>
 #include <asm/ptrace.h>
+#include <asm/irq.h>
 
 /*
  * These correspond to the IORESOURCE_IRQ_* defines in
@@ -374,6 +375,16 @@ struct softirq_action
 
 asmlinkage void do_softirq(void);
 asmlinkage void __do_softirq(void);
+
+#ifdef __ARCH_HAS_DO_SOFTIRQ
+void do_softirq_own_stack(void);
+#else
+static inline void do_softirq_own_stack(void)
+{
+	__do_softirq();
+}
+#endif
+
 extern void open_softirq(int nr, void (*action)(struct softirq_action *));
 extern void softirq_init(void);
 extern void __raise_softirq_irqoff(unsigned int nr);

kernel/softirq.c

@@ -29,7 +29,6 @@
 #define CREATE_TRACE_POINTS
 #include <trace/events/irq.h>
 
-#include <asm/irq.h>
 /*
    - No shared variables, all the data are CPU local.
    - If a softirq needs serialization, let it serialize itself
@@ -134,7 +133,6 @@ EXPORT_SYMBOL(local_bh_disable);
 
 static void __local_bh_enable(unsigned int cnt)
 {
-	WARN_ON_ONCE(in_irq());
 	WARN_ON_ONCE(!irqs_disabled());
 
 	if (softirq_count() == cnt)
@@ -149,6 +147,7 @@ static void __local_bh_enable(unsigned int cnt)
  */
 void _local_bh_enable(void)
 {
+	WARN_ON_ONCE(in_irq());
 	__local_bh_enable(SOFTIRQ_DISABLE_OFFSET);
 }
 
@@ -171,8 +170,13 @@ static inline void _local_bh_enable_ip(unsigned long ip)
  	 */
 	sub_preempt_count(SOFTIRQ_DISABLE_OFFSET - 1);
 
-	if (unlikely(!in_interrupt() && local_softirq_pending()))
+	if (unlikely(!in_interrupt() && local_softirq_pending())) {
+		/*
+		 * Run softirq if any pending. And do it in its own stack
+		 * as we may be calling this deep in a task call stack already.
+		 */
 		do_softirq();
+	}
 
 	dec_preempt_count();
 #ifdef CONFIG_TRACE_IRQFLAGS
@@ -280,10 +284,11 @@ restart:
 
 	account_irq_exit_time(current);
 	__local_bh_enable(SOFTIRQ_OFFSET);
+	WARN_ON_ONCE(in_interrupt());
 	tsk_restore_flags(current, old_flags, PF_MEMALLOC);
 }
 
-#ifndef __ARCH_HAS_DO_SOFTIRQ
+
 
 asmlinkage void do_softirq(void)
 {
@@ -298,13 +303,11 @@ asmlinkage void do_softirq(void)
 	pending = local_softirq_pending();
 
 	if (pending)
-		__do_softirq();
+		do_softirq_own_stack();
 
 	local_irq_restore(flags);
 }
 
-#endif
-
 /*
  * Enter an interrupt context.
  */
@@ -328,10 +331,25 @@ void irq_enter(void)
 
 static inline void invoke_softirq(void)
 {
-	if (!force_irqthreads)
+	if (!force_irqthreads) {
+#ifdef CONFIG_HAVE_IRQ_EXIT_ON_IRQ_STACK
+		/*
+		 * We can safely execute softirq on the current stack if
+		 * it is the irq stack, because it should be near empty
+		 * at this stage.
+		 */
 		__do_softirq();
-	else
+#else
+		/*
+		 * Otherwise, irq_exit() is called on the task stack that can
+		 * be potentially deep already. So call softirq in its own stack
+		 * to prevent from any overrun.
+		 */
+		do_softirq_own_stack();
+#endif
+	} else {
 		wakeup_softirqd();
+	}
 }
 
 static inline void tick_irq_exit(void)
@@ -762,6 +780,10 @@ static void run_ksoftirqd(unsigned int cpu)
 {
 	local_irq_disable();
 	if (local_softirq_pending()) {
+		/*
+		 * We can safely run softirq on inline stack, as we are not deep
+		 * in the task stack here.
+		 */
 		__do_softirq();
 		rcu_note_context_switch(cpu);
 		local_irq_enable();