Merge branches 'amd-iommu/fixes', 'amd-iommu/debug', 'amd-iommu/suspend-resume' and 'amd-iommu/extended-allocator' into amd-iommu/2.6.31

Conflicts:
	arch/x86/kernel/amd_iommu.c
	arch/x86/kernel/amd_iommu_init.c

Documentation/kernel-parameters.txt

@@ -329,11 +329,6 @@ and is between 256 and 4096 characters. It is defined in the file
 				    flushed before they will be reused, which
 				    is a lot of faster
 
-	amd_iommu_size= [HW,X86-64]
-			Define the size of the aperture for the AMD IOMMU
-			driver. Possible values are:
-			'32M', '64M' (default), '128M', '256M', '512M', '1G'
-
 	amijoy.map=	[HW,JOY] Amiga joystick support
 			Map of devices attached to JOY0DAT and JOY1DAT
 			Format: <a>,<b>

arch/x86/Kconfig.debug

@@ -159,6 +159,14 @@ config IOMMU_DEBUG
 	  options. See Documentation/x86_64/boot-options.txt for more
 	  details.
 
+config IOMMU_STRESS
+	bool "Enable IOMMU stress-test mode"
+	---help---
+	  This option disables various optimizations in IOMMU related
+	  code to do real stress testing of the IOMMU code. This option
+	  will cause a performance drop and should only be enabled for
+	  testing.
+
 config IOMMU_LEAK
 	bool "IOMMU leak tracing"
 	depends on IOMMU_DEBUG && DMA_API_DEBUG

arch/x86/include/asm/amd_iommu.h

@@ -27,6 +27,8 @@ extern int amd_iommu_init(void);
 extern int amd_iommu_init_dma_ops(void);
 extern void amd_iommu_detect(void);
 extern irqreturn_t amd_iommu_int_handler(int irq, void *data);
+extern void amd_iommu_flush_all_domains(void);
+extern void amd_iommu_flush_all_devices(void);
 #else
 static inline int amd_iommu_init(void) { return -ENODEV; }
 static inline void amd_iommu_detect(void) { }

arch/x86/include/asm/amd_iommu_types.h

@@ -194,6 +194,27 @@
 #define PD_DMA_OPS_MASK		(1UL << 0) /* domain used for dma_ops */
 #define PD_DEFAULT_MASK		(1UL << 1) /* domain is a default dma_ops
 					      domain for an IOMMU */
+extern bool amd_iommu_dump;
+#define DUMP_printk(format, arg...)					\
+	do {								\
+		if (amd_iommu_dump)						\
+			printk(KERN_INFO "AMD IOMMU: " format, ## arg);	\
+	} while(0);
+
+/*
+ * Make iterating over all IOMMUs easier
+ */
+#define for_each_iommu(iommu) \
+	list_for_each_entry((iommu), &amd_iommu_list, list)
+#define for_each_iommu_safe(iommu, next) \
+	list_for_each_entry_safe((iommu), (next), &amd_iommu_list, list)
+
+#define APERTURE_RANGE_SHIFT	27	/* 128 MB */
+#define APERTURE_RANGE_SIZE	(1ULL << APERTURE_RANGE_SHIFT)
+#define APERTURE_RANGE_PAGES	(APERTURE_RANGE_SIZE >> PAGE_SHIFT)
+#define APERTURE_MAX_RANGES	32	/* allows 4GB of DMA address space */
+#define APERTURE_RANGE_INDEX(a)	((a) >> APERTURE_RANGE_SHIFT)
+#define APERTURE_PAGE_INDEX(a)	(((a) >> 21) & 0x3fULL)
 
 /*
  * This structure contains generic data for  IOMMU protection domains
@@ -209,6 +230,26 @@ struct protection_domain {
 	void *priv;		/* private data */
 };
 
+/*
+ * For dynamic growth the aperture size is split into ranges of 128MB of
+ * DMA address space each. This struct represents one such range.
+ */
+struct aperture_range {
+
+	/* address allocation bitmap */
+	unsigned long *bitmap;
+
+	/*
+	 * Array of PTE pages for the aperture. In this array we save all the
+	 * leaf pages of the domain page table used for the aperture. This way
+	 * we don't need to walk the page table to find a specific PTE. We can
+	 * just calculate its address in constant time.
+	 */
+	u64 *pte_pages[64];
+
+	unsigned long offset;
+};
+
 /*
  * Data container for a dma_ops specific protection domain
  */
@@ -222,18 +263,10 @@ struct dma_ops_domain {
 	unsigned long aperture_size;
 
 	/* address we start to search for free addresses */
-	unsigned long next_bit;
-
-	/* address allocation bitmap */
-	unsigned long *bitmap;
+	unsigned long next_address;
 
-	/*
-	 * Array of PTE pages for the aperture. In this array we save all the
-	 * leaf pages of the domain page table used for the aperture. This way
-	 * we don't need to walk the page table to find a specific PTE. We can
-	 * just calculate its address in constant time.
-	 */
-	u64 **pte_pages;
+	/* address space relevant data */
+	struct aperture_range *aperture[APERTURE_MAX_RANGES];
 
 	/* This will be set to true when TLB needs to be flushed */
 	bool need_flush;

arch/x86/kernel/amd_iommu.c

@@ -55,7 +55,12 @@ struct iommu_cmd {
 static int dma_ops_unity_map(struct dma_ops_domain *dma_dom,
 			     struct unity_map_entry *e);
 static struct dma_ops_domain *find_protection_domain(u16 devid);
-
+static u64* alloc_pte(struct protection_domain *dom,
+		      unsigned long address, u64
+		      **pte_page, gfp_t gfp);
+static void dma_ops_reserve_addresses(struct dma_ops_domain *dom,
+				      unsigned long start_page,
+				      unsigned int pages);
 
 #ifndef BUS_NOTIFY_UNBOUND_DRIVER
 #define BUS_NOTIFY_UNBOUND_DRIVER 0x0005
@@ -217,7 +222,7 @@ irqreturn_t amd_iommu_int_handler(int irq, void *data)
 {
 	struct amd_iommu *iommu;
 
-	list_for_each_entry(iommu, &amd_iommu_list, list)
+	for_each_iommu(iommu)
 		iommu_poll_events(iommu);
 
 	return IRQ_HANDLED;
@@ -444,7 +449,7 @@ static void iommu_flush_domain(u16 domid)
 	__iommu_build_inv_iommu_pages(&cmd, CMD_INV_IOMMU_ALL_PAGES_ADDRESS,
 				      domid, 1, 1);
 
-	list_for_each_entry(iommu, &amd_iommu_list, list) {
+	for_each_iommu(iommu) {
 		spin_lock_irqsave(&iommu->lock, flags);
 		__iommu_queue_command(iommu, &cmd);
 		__iommu_completion_wait(iommu);
@@ -453,6 +458,35 @@ static void iommu_flush_domain(u16 domid)
 	}
 }
 
+void amd_iommu_flush_all_domains(void)
+{
+	int i;
+
+	for (i = 1; i < MAX_DOMAIN_ID; ++i) {
+		if (!test_bit(i, amd_iommu_pd_alloc_bitmap))
+			continue;
+		iommu_flush_domain(i);
+	}
+}
+
+void amd_iommu_flush_all_devices(void)
+{
+	struct amd_iommu *iommu;
+	int i;
+
+	for (i = 0; i <= amd_iommu_last_bdf; ++i) {
+		if (amd_iommu_pd_table[i] == NULL)
+			continue;
+
+		iommu = amd_iommu_rlookup_table[i];
+		if (!iommu)
+			continue;
+
+		iommu_queue_inv_dev_entry(iommu, i);
+		iommu_completion_wait(iommu);
+	}
+}
+
 /****************************************************************************
  *
  * The functions below are used the create the page table mappings for
@@ -472,7 +506,7 @@ static int iommu_map_page(struct protection_domain *dom,
 			  unsigned long phys_addr,
 			  int prot)
 {
-	u64 __pte, *pte, *page;
+	u64 __pte, *pte;
 
 	bus_addr  = PAGE_ALIGN(bus_addr);
 	phys_addr = PAGE_ALIGN(phys_addr);
@@ -481,27 +515,7 @@ static int iommu_map_page(struct protection_domain *dom,
 	if (bus_addr > IOMMU_MAP_SIZE_L3 || !(prot & IOMMU_PROT_MASK))
 		return -EINVAL;
 
-	pte = &dom->pt_root[IOMMU_PTE_L2_INDEX(bus_addr)];
-
-	if (!IOMMU_PTE_PRESENT(*pte)) {
-		page = (u64 *)get_zeroed_page(GFP_KERNEL);
-		if (!page)
-			return -ENOMEM;
-		*pte = IOMMU_L2_PDE(virt_to_phys(page));
-	}
-
-	pte = IOMMU_PTE_PAGE(*pte);
-	pte = &pte[IOMMU_PTE_L1_INDEX(bus_addr)];
-
-	if (!IOMMU_PTE_PRESENT(*pte)) {
-		page = (u64 *)get_zeroed_page(GFP_KERNEL);
-		if (!page)
-			return -ENOMEM;
-		*pte = IOMMU_L1_PDE(virt_to_phys(page));
-	}
-
-	pte = IOMMU_PTE_PAGE(*pte);
-	pte = &pte[IOMMU_PTE_L0_INDEX(bus_addr)];
+	pte = alloc_pte(dom, bus_addr, NULL, GFP_KERNEL);
 
 	if (IOMMU_PTE_PRESENT(*pte))
 		return -EBUSY;
@@ -599,7 +613,8 @@ static int dma_ops_unity_map(struct dma_ops_domain *dma_dom,
 		 * as allocated in the aperture
 		 */
 		if (addr < dma_dom->aperture_size)
-			__set_bit(addr >> PAGE_SHIFT, dma_dom->bitmap);
+			__set_bit(addr >> PAGE_SHIFT,
+				  dma_dom->aperture[0]->bitmap);
 	}
 
 	return 0;
@@ -636,42 +651,191 @@ static int init_unity_mappings_for_device(struct dma_ops_domain *dma_dom,
  ****************************************************************************/
 
 /*
- * The address allocator core function.
+ * The address allocator core functions.
  *
  * called with domain->lock held
  */
+
+/*
+ * This function checks if there is a PTE for a given dma address. If
+ * there is one, it returns the pointer to it.
+ */
+static u64* fetch_pte(struct protection_domain *domain,
+		      unsigned long address)
+{
+	u64 *pte;
+
+	pte = &domain->pt_root[IOMMU_PTE_L2_INDEX(address)];
+
+	if (!IOMMU_PTE_PRESENT(*pte))
+		return NULL;
+
+	pte = IOMMU_PTE_PAGE(*pte);
+	pte = &pte[IOMMU_PTE_L1_INDEX(address)];
+
+	if (!IOMMU_PTE_PRESENT(*pte))
+		return NULL;
+
+	pte = IOMMU_PTE_PAGE(*pte);
+	pte = &pte[IOMMU_PTE_L0_INDEX(address)];
+
+	return pte;
+}
+
+/*
+ * This function is used to add a new aperture range to an existing
+ * aperture in case of dma_ops domain allocation or address allocation
+ * failure.
+ */
+static int alloc_new_range(struct amd_iommu *iommu,
+			   struct dma_ops_domain *dma_dom,
+			   bool populate, gfp_t gfp)
+{
+	int index = dma_dom->aperture_size >> APERTURE_RANGE_SHIFT;
+	int i;
+
+#ifdef CONFIG_IOMMU_STRESS
+	populate = false;
+#endif
+
+	if (index >= APERTURE_MAX_RANGES)
+		return -ENOMEM;
+
+	dma_dom->aperture[index] = kzalloc(sizeof(struct aperture_range), gfp);
+	if (!dma_dom->aperture[index])
+		return -ENOMEM;
+
+	dma_dom->aperture[index]->bitmap = (void *)get_zeroed_page(gfp);
+	if (!dma_dom->aperture[index]->bitmap)
+		goto out_free;
+
+	dma_dom->aperture[index]->offset = dma_dom->aperture_size;
+
+	if (populate) {
+		unsigned long address = dma_dom->aperture_size;
+		int i, num_ptes = APERTURE_RANGE_PAGES / 512;
+		u64 *pte, *pte_page;
+
+		for (i = 0; i < num_ptes; ++i) {
+			pte = alloc_pte(&dma_dom->domain, address,
+					&pte_page, gfp);
+			if (!pte)
+				goto out_free;
+
+			dma_dom->aperture[index]->pte_pages[i] = pte_page;
+
+			address += APERTURE_RANGE_SIZE / 64;
+		}
+	}
+
+	dma_dom->aperture_size += APERTURE_RANGE_SIZE;
+
+	/* Intialize the exclusion range if necessary */
+	if (iommu->exclusion_start &&
+	    iommu->exclusion_start >= dma_dom->aperture[index]->offset &&
+	    iommu->exclusion_start < dma_dom->aperture_size) {
+		unsigned long startpage = iommu->exclusion_start >> PAGE_SHIFT;
+		int pages = iommu_num_pages(iommu->exclusion_start,
+					    iommu->exclusion_length,
+					    PAGE_SIZE);
+		dma_ops_reserve_addresses(dma_dom, startpage, pages);
+	}
+
+	/*
+	 * Check for areas already mapped as present in the new aperture
+	 * range and mark those pages as reserved in the allocator. Such
+	 * mappings may already exist as a result of requested unity
+	 * mappings for devices.
+	 */
+	for (i = dma_dom->aperture[index]->offset;
+	     i < dma_dom->aperture_size;
+	     i += PAGE_SIZE) {
+		u64 *pte = fetch_pte(&dma_dom->domain, i);
+		if (!pte || !IOMMU_PTE_PRESENT(*pte))
+			continue;
+
+		dma_ops_reserve_addresses(dma_dom, i << PAGE_SHIFT, 1);
+	}
+
+	return 0;
+
+out_free:
+	free_page((unsigned long)dma_dom->aperture[index]->bitmap);
+
+	kfree(dma_dom->aperture[index]);
+	dma_dom->aperture[index] = NULL;
+
+	return -ENOMEM;
+}
+
+static unsigned long dma_ops_area_alloc(struct device *dev,
+					struct dma_ops_domain *dom,
+					unsigned int pages,
+					unsigned long align_mask,
+					u64 dma_mask,
+					unsigned long start)
+{
+	unsigned long next_bit = dom->next_address % APERTURE_RANGE_SIZE;
+	int max_index = dom->aperture_size >> APERTURE_RANGE_SHIFT;
+	int i = start >> APERTURE_RANGE_SHIFT;
+	unsigned long boundary_size;
+	unsigned long address = -1;
+	unsigned long limit;
+
+	next_bit >>= PAGE_SHIFT;
+
+	boundary_size = ALIGN(dma_get_seg_boundary(dev) + 1,
+			PAGE_SIZE) >> PAGE_SHIFT;
+
+	for (;i < max_index; ++i) {
+		unsigned long offset = dom->aperture[i]->offset >> PAGE_SHIFT;
+
+		if (dom->aperture[i]->offset >= dma_mask)
+			break;
+
+		limit = iommu_device_max_index(APERTURE_RANGE_PAGES, offset,
+					       dma_mask >> PAGE_SHIFT);
+
+		address = iommu_area_alloc(dom->aperture[i]->bitmap,
+					   limit, next_bit, pages, 0,
+					    boundary_size, align_mask);
+		if (address != -1) {
+			address = dom->aperture[i]->offset +
+				  (address << PAGE_SHIFT);
+			dom->next_address = address + (pages << PAGE_SHIFT);
+			break;
+		}
+
+		next_bit = 0;
+	}
+
+	return address;
+}
+
 static unsigned long dma_ops_alloc_addresses(struct device *dev,
 					     struct dma_ops_domain *dom,
 					     unsigned int pages,
 					     unsigned long align_mask,
 					     u64 dma_mask)
 {
-	unsigned long limit;
 	unsigned long address;
-	unsigned long boundary_size;
 
-	boundary_size = ALIGN(dma_get_seg_boundary(dev) + 1,
-			PAGE_SIZE) >> PAGE_SHIFT;
-	limit = iommu_device_max_index(dom->aperture_size >> PAGE_SHIFT, 0,
-				       dma_mask >> PAGE_SHIFT);
+#ifdef CONFIG_IOMMU_STRESS
+	dom->next_address = 0;
+	dom->need_flush = true;
+#endif
 
-	if (dom->next_bit >= limit) {
-		dom->next_bit = 0;
-		dom->need_flush = true;
-	}
+	address = dma_ops_area_alloc(dev, dom, pages, align_mask,
+				     dma_mask, dom->next_address);
 
-	address = iommu_area_alloc(dom->bitmap, limit, dom->next_bit, pages,
-				   0 , boundary_size, align_mask);
 	if (address == -1) {
-		address = iommu_area_alloc(dom->bitmap, limit, 0, pages,
-				0, boundary_size, align_mask);
+		dom->next_address = 0;
+		address = dma_ops_area_alloc(dev, dom, pages, align_mask,
+					     dma_mask, 0);
 		dom->need_flush = true;
 	}
 
-	if (likely(address != -1)) {
-		dom->next_bit = address + pages;
-		address <<= PAGE_SHIFT;
-	} else
+	if (unlikely(address == -1))
 		address = bad_dma_address;
 
 	WARN_ON((address + (PAGE_SIZE*pages)) > dom->aperture_size);
@@ -688,11 +852,23 @@ static void dma_ops_free_addresses(struct dma_ops_domain *dom,
 				   unsigned long address,
 				   unsigned int pages)
 {
-	address >>= PAGE_SHIFT;
-	iommu_area_free(dom->bitmap, address, pages);
+	unsigned i = address >> APERTURE_RANGE_SHIFT;
+	struct aperture_range *range = dom->aperture[i];
+
+	BUG_ON(i >= APERTURE_MAX_RANGES || range == NULL);
+
+#ifdef CONFIG_IOMMU_STRESS
+	if (i < 4)
+		return;
+#endif
 
-	if (address >= dom->next_bit)
+	if (address >= dom->next_address)
 		dom->need_flush = true;
+
+	address = (address % APERTURE_RANGE_SIZE) >> PAGE_SHIFT;
+
+	iommu_area_free(range->bitmap, address, pages);
+
 }
 
 /****************************************************************************
@@ -740,12 +916,16 @@ static void dma_ops_reserve_addresses(struct dma_ops_domain *dom,
 				      unsigned long start_page,
 				      unsigned int pages)
 {
-	unsigned int last_page = dom->aperture_size >> PAGE_SHIFT;
+	unsigned int i, last_page = dom->aperture_size >> PAGE_SHIFT;
 
 	if (start_page + pages > last_page)
 		pages = last_page - start_page;
 
-	iommu_area_reserve(dom->bitmap, start_page, pages);
+	for (i = start_page; i < start_page + pages; ++i) {
+		int index = i / APERTURE_RANGE_PAGES;
+		int page  = i % APERTURE_RANGE_PAGES;
+		__set_bit(page, dom->aperture[index]->bitmap);
+	}
 }
 
 static void free_pagetable(struct protection_domain *domain)
@@ -784,14 +964,19 @@ static void free_pagetable(struct protection_domain *domain)
  */
 static void dma_ops_domain_free(struct dma_ops_domain *dom)
 {
+	int i;
+
 	if (!dom)
 		return;
 
 	free_pagetable(&dom->domain);
 
-	kfree(dom->pte_pages);
-
-	kfree(dom->bitmap);
+	for (i = 0; i < APERTURE_MAX_RANGES; ++i) {
+		if (!dom->aperture[i])
+			continue;
+		free_page((unsigned long)dom->aperture[i]->bitmap);
+		kfree(dom->aperture[i]);
+	}
 
 	kfree(dom);
 }
@@ -801,19 +986,9 @@ static void dma_ops_domain_free(struct dma_ops_domain *dom)
  * It also intializes the page table and the address allocator data
  * structures required for the dma_ops interface
  */
-static struct dma_ops_domain *dma_ops_domain_alloc(struct amd_iommu *iommu,
-						   unsigned order)
+static struct dma_ops_domain *dma_ops_domain_alloc(struct amd_iommu *iommu)
 {
 	struct dma_ops_domain *dma_dom;
-	unsigned i, num_pte_pages;
-	u64 *l2_pde;
-	u64 address;
-
-	/*
-	 * Currently the DMA aperture must be between 32 MB and 1GB in size
-	 */
-	if ((order < 25) || (order > 30))
-		return NULL;
 
 	dma_dom = kzalloc(sizeof(struct dma_ops_domain), GFP_KERNEL);
 	if (!dma_dom)
@@ -830,55 +1005,20 @@ static struct dma_ops_domain *dma_ops_domain_alloc(struct amd_iommu *iommu,
 	dma_dom->domain.priv = dma_dom;
 	if (!dma_dom->domain.pt_root)
 		goto free_dma_dom;
-	dma_dom->aperture_size = (1ULL << order);
-	dma_dom->bitmap = kzalloc(dma_dom->aperture_size / (PAGE_SIZE * 8),
-				  GFP_KERNEL);
-	if (!dma_dom->bitmap)
-		goto free_dma_dom;
-	/*
-	 * mark the first page as allocated so we never return 0 as
-	 * a valid dma-address. So we can use 0 as error value
-	 */
-	dma_dom->bitmap[0] = 1;
-	dma_dom->next_bit = 0;
 
 	dma_dom->need_flush = false;
 	dma_dom->target_dev = 0xffff;
 
-	/* Intialize the exclusion range if necessary */
-	if (iommu->exclusion_start &&
-	    iommu->exclusion_start < dma_dom->aperture_size) {
-		unsigned long startpage = iommu->exclusion_start >> PAGE_SHIFT;
-		int pages = iommu_num_pages(iommu->exclusion_start,
-					    iommu->exclusion_length,
-					    PAGE_SIZE);
-		dma_ops_reserve_addresses(dma_dom, startpage, pages);
-	}
+	if (alloc_new_range(iommu, dma_dom, true, GFP_KERNEL))
+		goto free_dma_dom;
 
 	/*
-	 * At the last step, build the page tables so we don't need to
-	 * allocate page table pages in the dma_ops mapping/unmapping
-	 * path.
+	 * mark the first page as allocated so we never return 0 as
+	 * a valid dma-address. So we can use 0 as error value
 	 */
-	num_pte_pages = dma_dom->aperture_size / (PAGE_SIZE * 512);
-	dma_dom->pte_pages = kzalloc(num_pte_pages * sizeof(void *),
-			GFP_KERNEL);
-	if (!dma_dom->pte_pages)
-		goto free_dma_dom;
-
-	l2_pde = (u64 *)get_zeroed_page(GFP_KERNEL);
-	if (l2_pde == NULL)
-		goto free_dma_dom;
+	dma_dom->aperture[0]->bitmap[0] = 1;
+	dma_dom->next_address = 0;
 
-	dma_dom->domain.pt_root[0] = IOMMU_L2_PDE(virt_to_phys(l2_pde));
-
-	for (i = 0; i < num_pte_pages; ++i) {
-		dma_dom->pte_pages[i] = (u64 *)get_zeroed_page(GFP_KERNEL);
-		if (!dma_dom->pte_pages[i])
-			goto free_dma_dom;
-		address = virt_to_phys(dma_dom->pte_pages[i]);
-		l2_pde[i] = IOMMU_L1_PDE(address);
-	}
 
 	return dma_dom;
 
@@ -987,7 +1127,6 @@ static int device_change_notifier(struct notifier_block *nb,
 	struct protection_domain *domain;
 	struct dma_ops_domain *dma_domain;
 	struct amd_iommu *iommu;
-	int order = amd_iommu_aperture_order;
 	unsigned long flags;
 
 	if (devid > amd_iommu_last_bdf)
@@ -1013,8 +1152,9 @@ static int device_change_notifier(struct notifier_block *nb,
 		if (!dma_domain)
 			dma_domain = iommu->default_dom;
 		attach_device(iommu, &dma_domain->domain, devid);
-		printk(KERN_INFO "AMD IOMMU: Using protection domain %d for "
-		       "device %s\n", dma_domain->domain.id, dev_name(dev));
+		DUMP_printk(KERN_INFO "AMD IOMMU: Using protection domain "
+			    "%d for device %s\n",
+			    dma_domain->domain.id, dev_name(dev));
 		break;
 	case BUS_NOTIFY_UNBOUND_DRIVER:
 		if (!domain)
@@ -1026,7 +1166,7 @@ static int device_change_notifier(struct notifier_block *nb,
 		dma_domain = find_protection_domain(devid);
 		if (dma_domain)
 			goto out;
-		dma_domain = dma_ops_domain_alloc(iommu, order);
+		dma_domain = dma_ops_domain_alloc(iommu);
 		if (!dma_domain)
 			goto out;
 		dma_domain->target_dev = devid;
@@ -1137,8 +1277,9 @@ static int get_device_resources(struct device *dev,
 			dma_dom = (*iommu)->default_dom;
 		*domain = &dma_dom->domain;
 		attach_device(*iommu, *domain, *bdf);
-		printk(KERN_INFO "AMD IOMMU: Using protection domain %d for "
-				"device %s\n", (*domain)->id, dev_name(dev));
+		DUMP_printk(KERN_INFO "AMD IOMMU: Using protection domain "
+				"%d for device %s\n",
+				(*domain)->id, dev_name(dev));
 	}
 
 	if (domain_for_device(_bdf) == NULL)
@@ -1147,6 +1288,66 @@ static int get_device_resources(struct device *dev,
 	return 1;
 }
 
+/*
+ * If the pte_page is not yet allocated this function is called
+ */
+static u64* alloc_pte(struct protection_domain *dom,
+		      unsigned long address, u64 **pte_page, gfp_t gfp)
+{
+	u64 *pte, *page;
+
+	pte = &dom->pt_root[IOMMU_PTE_L2_INDEX(address)];
+
+	if (!IOMMU_PTE_PRESENT(*pte)) {
+		page = (u64 *)get_zeroed_page(gfp);
+		if (!page)
+			return NULL;
+		*pte = IOMMU_L2_PDE(virt_to_phys(page));
+	}
+
+	pte = IOMMU_PTE_PAGE(*pte);
+	pte = &pte[IOMMU_PTE_L1_INDEX(address)];
+
+	if (!IOMMU_PTE_PRESENT(*pte)) {
+		page = (u64 *)get_zeroed_page(gfp);
+		if (!page)
+			return NULL;
+		*pte = IOMMU_L1_PDE(virt_to_phys(page));
+	}
+
+	pte = IOMMU_PTE_PAGE(*pte);
+
+	if (pte_page)
+		*pte_page = pte;
+
+	pte = &pte[IOMMU_PTE_L0_INDEX(address)];
+
+	return pte;
+}
+
+/*
+ * This function fetches the PTE for a given address in the aperture
+ */
+static u64* dma_ops_get_pte(struct dma_ops_domain *dom,
+			    unsigned long address)
+{
+	struct aperture_range *aperture;
+	u64 *pte, *pte_page;
+
+	aperture = dom->aperture[APERTURE_RANGE_INDEX(address)];
+	if (!aperture)
+		return NULL;
+
+	pte = aperture->pte_pages[APERTURE_PAGE_INDEX(address)];
+	if (!pte) {
+		pte = alloc_pte(&dom->domain, address, &pte_page, GFP_ATOMIC);
+		aperture->pte_pages[APERTURE_PAGE_INDEX(address)] = pte_page;
+	} else
+		pte += IOMMU_PTE_L0_INDEX(address);
+
+	return pte;
+}
+
 /*
  * This is the generic map function. It maps one 4kb page at paddr to
  * the given address in the DMA address space for the domain.
@@ -1163,8 +1364,9 @@ static dma_addr_t dma_ops_domain_map(struct amd_iommu *iommu,
 
 	paddr &= PAGE_MASK;
 
-	pte  = dom->pte_pages[IOMMU_PTE_L1_INDEX(address)];
-	pte += IOMMU_PTE_L0_INDEX(address);
+	pte  = dma_ops_get_pte(dom, address);
+	if (!pte)
+		return bad_dma_address;
 
 	__pte = paddr | IOMMU_PTE_P | IOMMU_PTE_FC;
 
@@ -1189,14 +1391,20 @@ static void dma_ops_domain_unmap(struct amd_iommu *iommu,
 				 struct dma_ops_domain *dom,
 				 unsigned long address)
 {
+	struct aperture_range *aperture;
 	u64 *pte;
 
 	if (address >= dom->aperture_size)
 		return;
 
-	WARN_ON(address & ~PAGE_MASK || address >= dom->aperture_size);
+	aperture = dom->aperture[APERTURE_RANGE_INDEX(address)];
+	if (!aperture)
+		return;
+
+	pte  = aperture->pte_pages[APERTURE_PAGE_INDEX(address)];
+	if (!pte)
+		return;
 
-	pte  = dom->pte_pages[IOMMU_PTE_L1_INDEX(address)];
 	pte += IOMMU_PTE_L0_INDEX(address);
 
 	WARN_ON(!*pte);
@@ -1220,7 +1428,7 @@ static dma_addr_t __map_single(struct device *dev,
 			       u64 dma_mask)
 {
 	dma_addr_t offset = paddr & ~PAGE_MASK;
-	dma_addr_t address, start;
+	dma_addr_t address, start, ret;
 	unsigned int pages;
 	unsigned long align_mask = 0;
 	int i;
@@ -1236,14 +1444,33 @@ static dma_addr_t __map_single(struct device *dev,
 	if (align)
 		align_mask = (1UL << get_order(size)) - 1;
 
+retry:
 	address = dma_ops_alloc_addresses(dev, dma_dom, pages, align_mask,
 					  dma_mask);
-	if (unlikely(address == bad_dma_address))
-		goto out;
+	if (unlikely(address == bad_dma_address)) {
+		/*
+		 * setting next_address here will let the address
+		 * allocator only scan the new allocated range in the
+		 * first run. This is a small optimization.
+		 */
+		dma_dom->next_address = dma_dom->aperture_size;
+
+		if (alloc_new_range(iommu, dma_dom, false, GFP_ATOMIC))
+			goto out;
+
+		/*
+		 * aperture was sucessfully enlarged by 128 MB, try
+		 * allocation again
+		 */
+		goto retry;
+	}
 
 	start = address;
 	for (i = 0; i < pages; ++i) {
-		dma_ops_domain_map(iommu, dma_dom, start, paddr, dir);
+		ret = dma_ops_domain_map(iommu, dma_dom, start, paddr, dir);
+		if (ret == bad_dma_address)
+			goto out_unmap;
+
 		paddr += PAGE_SIZE;
 		start += PAGE_SIZE;
 	}
@@ -1259,6 +1486,17 @@ static dma_addr_t __map_single(struct device *dev,
 
 out:
 	return address;
+
+out_unmap:
+
+	for (--i; i >= 0; --i) {
+		start -= PAGE_SIZE;
+		dma_ops_domain_unmap(iommu, dma_dom, start);
+	}
+
+	dma_ops_free_addresses(dma_dom, address, pages);
+
+	return bad_dma_address;
 }
 
 /*
@@ -1629,7 +1867,6 @@ static void prealloc_protection_domains(void)
 	struct pci_dev *dev = NULL;
 	struct dma_ops_domain *dma_dom;
 	struct amd_iommu *iommu;
-	int order = amd_iommu_aperture_order;
 	u16 devid;
 
 	while ((dev = pci_get_device(PCI_ANY_ID, PCI_ANY_ID, dev)) != NULL) {
@@ -1642,7 +1879,7 @@ static void prealloc_protection_domains(void)
 		iommu = amd_iommu_rlookup_table[devid];
 		if (!iommu)
 			continue;
-		dma_dom = dma_ops_domain_alloc(iommu, order);
+		dma_dom = dma_ops_domain_alloc(iommu);
 		if (!dma_dom)
 			continue;
 		init_unity_mappings_for_device(dma_dom, devid);
@@ -1668,7 +1905,6 @@ static struct dma_map_ops amd_iommu_dma_ops = {
 int __init amd_iommu_init_dma_ops(void)
 {
 	struct amd_iommu *iommu;
-	int order = amd_iommu_aperture_order;
 	int ret;
 
 	/*
@@ -1676,8 +1912,8 @@ int __init amd_iommu_init_dma_ops(void)
 	 * found in the system. Devices not assigned to any other
 	 * protection domain will be assigned to the default one.
 	 */
-	list_for_each_entry(iommu, &amd_iommu_list, list) {
-		iommu->default_dom = dma_ops_domain_alloc(iommu, order);
+	for_each_iommu(iommu) {
+		iommu->default_dom = dma_ops_domain_alloc(iommu);
 		if (iommu->default_dom == NULL)
 			return -ENOMEM;
 		iommu->default_dom->domain.flags |= PD_DEFAULT_MASK;
@@ -1714,7 +1950,7 @@ int __init amd_iommu_init_dma_ops(void)
 
 free_domains:
 
-	list_for_each_entry(iommu, &amd_iommu_list, list) {
+	for_each_iommu(iommu) {
 		if (iommu->default_dom)
 			dma_ops_domain_free(iommu->default_dom);
 	}

arch/x86/kernel/amd_iommu_init.c

@@ -115,15 +115,21 @@ struct ivmd_header {
 	u64 range_length;
 } __attribute__((packed));
 
+bool amd_iommu_dump;
+
 static int __initdata amd_iommu_detected;
 
 u16 amd_iommu_last_bdf;			/* largest PCI device id we have
 					   to handle */
 LIST_HEAD(amd_iommu_unity_map);		/* a list of required unity mappings
 					   we find in ACPI */
-unsigned amd_iommu_aperture_order = 26; /* size of aperture in power of 2 */
+#ifdef CONFIG_IOMMU_STRESS
+bool amd_iommu_isolate = false;
+#else
 bool amd_iommu_isolate = true;		/* if true, device isolation is
 					   enabled */
+#endif
+
 bool amd_iommu_unmap_flush;		/* if true, flush on every unmap */
 
 LIST_HEAD(amd_iommu_list);		/* list of all AMD IOMMUs in the
@@ -193,7 +199,7 @@ static inline unsigned long tbl_size(int entry_size)
  * This function set the exclusion range in the IOMMU. DMA accesses to the
  * exclusion range are passed through untranslated
  */
-static void __init iommu_set_exclusion_range(struct amd_iommu *iommu)
+static void iommu_set_exclusion_range(struct amd_iommu *iommu)
 {
 	u64 start = iommu->exclusion_start & PAGE_MASK;
 	u64 limit = (start + iommu->exclusion_length) & PAGE_MASK;
@@ -225,7 +231,7 @@ static void __init iommu_set_device_table(struct amd_iommu *iommu)
 }
 
 /* Generic functions to enable/disable certain features of the IOMMU. */
-static void __init iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
+static void iommu_feature_enable(struct amd_iommu *iommu, u8 bit)
 {
 	u32 ctrl;
 
@@ -244,7 +250,7 @@ static void __init iommu_feature_disable(struct amd_iommu *iommu, u8 bit)
 }
 
 /* Function to enable the hardware */
-static void __init iommu_enable(struct amd_iommu *iommu)
+static void iommu_enable(struct amd_iommu *iommu)
 {
 	printk(KERN_INFO "AMD IOMMU: Enabling IOMMU at %s cap 0x%hx\n",
 	       dev_name(&iommu->dev->dev), iommu->cap_ptr);
@@ -252,11 +258,9 @@ static void __init iommu_enable(struct amd_iommu *iommu)
 	iommu_feature_enable(iommu, CONTROL_IOMMU_EN);
 }
 
-/* Function to enable IOMMU event logging and event interrupts */
-static void __init iommu_enable_event_logging(struct amd_iommu *iommu)
+static void iommu_disable(struct amd_iommu *iommu)
 {
-	iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
-	iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
+	iommu_feature_disable(iommu, CONTROL_IOMMU_EN);
 }
 
 /*
@@ -413,25 +417,36 @@ static u8 * __init alloc_command_buffer(struct amd_iommu *iommu)
 {
 	u8 *cmd_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
 			get_order(CMD_BUFFER_SIZE));
-	u64 entry;
 
 	if (cmd_buf == NULL)
 		return NULL;
 
 	iommu->cmd_buf_size = CMD_BUFFER_SIZE;
 
-	entry = (u64)virt_to_phys(cmd_buf);
+	return cmd_buf;
+}
+
+/*
+ * This function writes the command buffer address to the hardware and
+ * enables it.
+ */
+static void iommu_enable_command_buffer(struct amd_iommu *iommu)
+{
+	u64 entry;
+
+	BUG_ON(iommu->cmd_buf == NULL);
+
+	entry = (u64)virt_to_phys(iommu->cmd_buf);
 	entry |= MMIO_CMD_SIZE_512;
+
 	memcpy_toio(iommu->mmio_base + MMIO_CMD_BUF_OFFSET,
-			&entry, sizeof(entry));
+		    &entry, sizeof(entry));
 
 	/* set head and tail to zero manually */
 	writel(0x00, iommu->mmio_base + MMIO_CMD_HEAD_OFFSET);
 	writel(0x00, iommu->mmio_base + MMIO_CMD_TAIL_OFFSET);
 
 	iommu_feature_enable(iommu, CONTROL_CMDBUF_EN);
-
-	return cmd_buf;
 }
 
 static void __init free_command_buffer(struct amd_iommu *iommu)
@@ -443,20 +458,27 @@ static void __init free_command_buffer(struct amd_iommu *iommu)
 /* allocates the memory where the IOMMU will log its events to */
 static u8 * __init alloc_event_buffer(struct amd_iommu *iommu)
 {
-	u64 entry;
 	iommu->evt_buf = (u8 *)__get_free_pages(GFP_KERNEL | __GFP_ZERO,
 						get_order(EVT_BUFFER_SIZE));
 
 	if (iommu->evt_buf == NULL)
 		return NULL;
 
+	return iommu->evt_buf;
+}
+
+static void iommu_enable_event_buffer(struct amd_iommu *iommu)
+{
+	u64 entry;
+
+	BUG_ON(iommu->evt_buf == NULL);
+
 	entry = (u64)virt_to_phys(iommu->evt_buf) | EVT_LEN_MASK;
+
 	memcpy_toio(iommu->mmio_base + MMIO_EVT_BUF_OFFSET,
 		    &entry, sizeof(entry));
 
-	iommu->evt_buf_size = EVT_BUFFER_SIZE;
-
-	return iommu->evt_buf;
+	iommu_feature_enable(iommu, CONTROL_EVT_LOG_EN);
 }
 
 static void __init free_event_buffer(struct amd_iommu *iommu)
@@ -596,32 +618,83 @@ static void __init init_iommu_from_acpi(struct amd_iommu *iommu,
 	p += sizeof(struct ivhd_header);
 	end += h->length;
 
+
 	while (p < end) {
 		e = (struct ivhd_entry *)p;
 		switch (e->type) {
 		case IVHD_DEV_ALL:
+
+			DUMP_printk("  DEV_ALL\t\t\t first devid: %02x:%02x.%x"
+				    " last device %02x:%02x.%x flags: %02x\n",
+				    PCI_BUS(iommu->first_device),
+				    PCI_SLOT(iommu->first_device),
+				    PCI_FUNC(iommu->first_device),
+				    PCI_BUS(iommu->last_device),
+				    PCI_SLOT(iommu->last_device),
+				    PCI_FUNC(iommu->last_device),
+				    e->flags);
+
 			for (dev_i = iommu->first_device;
 					dev_i <= iommu->last_device; ++dev_i)
 				set_dev_entry_from_acpi(iommu, dev_i,
 							e->flags, 0);
 			break;
 		case IVHD_DEV_SELECT:
+
+			DUMP_printk("  DEV_SELECT\t\t\t devid: %02x:%02x.%x "
+				    "flags: %02x\n",
+				    PCI_BUS(e->devid),
+				    PCI_SLOT(e->devid),
+				    PCI_FUNC(e->devid),
+				    e->flags);
+
 			devid = e->devid;
 			set_dev_entry_from_acpi(iommu, devid, e->flags, 0);
 			break;
 		case IVHD_DEV_SELECT_RANGE_START:
+
+			DUMP_printk("  DEV_SELECT_RANGE_START\t "
+				    "devid: %02x:%02x.%x flags: %02x\n",
+				    PCI_BUS(e->devid),
+				    PCI_SLOT(e->devid),
+				    PCI_FUNC(e->devid),
+				    e->flags);
+
 			devid_start = e->devid;
 			flags = e->flags;
 			ext_flags = 0;
 			alias = false;
 			break;
 		case IVHD_DEV_ALIAS:
+
+			DUMP_printk("  DEV_ALIAS\t\t\t devid: %02x:%02x.%x "
+				    "flags: %02x devid_to: %02x:%02x.%x\n",
+				    PCI_BUS(e->devid),
+				    PCI_SLOT(e->devid),
+				    PCI_FUNC(e->devid),
+				    e->flags,
+				    PCI_BUS(e->ext >> 8),
+				    PCI_SLOT(e->ext >> 8),
+				    PCI_FUNC(e->ext >> 8));
+
 			devid = e->devid;
 			devid_to = e->ext >> 8;
 			set_dev_entry_from_acpi(iommu, devid_to, e->flags, 0);
 			amd_iommu_alias_table[devid] = devid_to;
 			break;
 		case IVHD_DEV_ALIAS_RANGE:
+
+			DUMP_printk("  DEV_ALIAS_RANGE\t\t "
+				    "devid: %02x:%02x.%x flags: %02x "
+				    "devid_to: %02x:%02x.%x\n",
+				    PCI_BUS(e->devid),
+				    PCI_SLOT(e->devid),
+				    PCI_FUNC(e->devid),
+				    e->flags,
+				    PCI_BUS(e->ext >> 8),
+				    PCI_SLOT(e->ext >> 8),
+				    PCI_FUNC(e->ext >> 8));
+
 			devid_start = e->devid;
 			flags = e->flags;
 			devid_to = e->ext >> 8;
@@ -629,17 +702,39 @@ static void __init init_iommu_from_acpi(struct amd_iommu *iommu,
 			alias = true;
 			break;
 		case IVHD_DEV_EXT_SELECT:
+
+			DUMP_printk("  DEV_EXT_SELECT\t\t devid: %02x:%02x.%x "
+				    "flags: %02x ext: %08x\n",
+				    PCI_BUS(e->devid),
+				    PCI_SLOT(e->devid),
+				    PCI_FUNC(e->devid),
+				    e->flags, e->ext);
+
 			devid = e->devid;
 			set_dev_entry_from_acpi(iommu, devid, e->flags,
 						e->ext);
 			break;
 		case IVHD_DEV_EXT_SELECT_RANGE:
+
+			DUMP_printk("  DEV_EXT_SELECT_RANGE\t devid: "
+				    "%02x:%02x.%x flags: %02x ext: %08x\n",
+				    PCI_BUS(e->devid),
+				    PCI_SLOT(e->devid),
+				    PCI_FUNC(e->devid),
+				    e->flags, e->ext);
+
 			devid_start = e->devid;
 			flags = e->flags;
 			ext_flags = e->ext;
 			alias = false;
 			break;
 		case IVHD_DEV_RANGE_END:
+
+			DUMP_printk("  DEV_RANGE_END\t\t devid: %02x:%02x.%x\n",
+				    PCI_BUS(e->devid),
+				    PCI_SLOT(e->devid),
+				    PCI_FUNC(e->devid));
+
 			devid = e->devid;
 			for (dev_i = devid_start; dev_i <= devid; ++dev_i) {
 				if (alias)
@@ -679,7 +774,7 @@ static void __init free_iommu_all(void)
 {
 	struct amd_iommu *iommu, *next;
 
-	list_for_each_entry_safe(iommu, next, &amd_iommu_list, list) {
+	for_each_iommu_safe(iommu, next) {
 		list_del(&iommu->list);
 		free_iommu_one(iommu);
 		kfree(iommu);
@@ -710,7 +805,6 @@ static int __init init_iommu_one(struct amd_iommu *iommu, struct ivhd_header *h)
 	if (!iommu->mmio_base)
 		return -ENOMEM;
 
-	iommu_set_device_table(iommu);
 	iommu->cmd_buf = alloc_command_buffer(iommu);
 	if (!iommu->cmd_buf)
 		return -ENOMEM;
@@ -746,6 +840,15 @@ static int __init init_iommu_all(struct acpi_table_header *table)
 		h = (struct ivhd_header *)p;
 		switch (*p) {
 		case ACPI_IVHD_TYPE:
+
+			DUMP_printk("IOMMU: device: %02x:%02x.%01x cap: %04x "
+				    "seg: %d flags: %01x info %04x\n",
+				    PCI_BUS(h->devid), PCI_SLOT(h->devid),
+				    PCI_FUNC(h->devid), h->cap_ptr,
+				    h->pci_seg, h->flags, h->info);
+			DUMP_printk("       mmio-addr: %016llx\n",
+				    h->mmio_phys);
+
 			iommu = kzalloc(sizeof(struct amd_iommu), GFP_KERNEL);
 			if (iommu == NULL)
 				return -ENOMEM;
@@ -773,56 +876,9 @@ static int __init init_iommu_all(struct acpi_table_header *table)
  *
  ****************************************************************************/
 
-static int __init iommu_setup_msix(struct amd_iommu *iommu)
-{
-	struct amd_iommu *curr;
-	struct msix_entry entries[32]; /* only 32 supported by AMD IOMMU */
-	int nvec = 0, i;
-
-	list_for_each_entry(curr, &amd_iommu_list, list) {
-		if (curr->dev == iommu->dev) {
-			entries[nvec].entry = curr->evt_msi_num;
-			entries[nvec].vector = 0;
-			curr->int_enabled = true;
-			nvec++;
-		}
-	}
-
-	if (pci_enable_msix(iommu->dev, entries, nvec)) {
-		pci_disable_msix(iommu->dev);
-		return 1;
-	}
-
-	for (i = 0; i < nvec; ++i) {
-		int r = request_irq(entries->vector, amd_iommu_int_handler,
-				    IRQF_SAMPLE_RANDOM,
-				    "AMD IOMMU",
-				    NULL);
-		if (r)
-			goto out_free;
-	}
-
-	return 0;
-
-out_free:
-	for (i -= 1; i >= 0; --i)
-		free_irq(entries->vector, NULL);
-
-	pci_disable_msix(iommu->dev);
-
-	return 1;
-}
-
 static int __init iommu_setup_msi(struct amd_iommu *iommu)
 {
 	int r;
-	struct amd_iommu *curr;
-
-	list_for_each_entry(curr, &amd_iommu_list, list) {
-		if (curr->dev == iommu->dev)
-			curr->int_enabled = true;
-	}
-
 
 	if (pci_enable_msi(iommu->dev))
 		return 1;
@@ -837,17 +893,18 @@ static int __init iommu_setup_msi(struct amd_iommu *iommu)
 		return 1;
 	}
 
+	iommu->int_enabled = true;
+	iommu_feature_enable(iommu, CONTROL_EVT_INT_EN);
+
 	return 0;
 }
 
-static int __init iommu_init_msi(struct amd_iommu *iommu)
+static int iommu_init_msi(struct amd_iommu *iommu)
 {
 	if (iommu->int_enabled)
 		return 0;
 
-	if (pci_find_capability(iommu->dev, PCI_CAP_ID_MSIX))
-		return iommu_setup_msix(iommu);
-	else if (pci_find_capability(iommu->dev, PCI_CAP_ID_MSI))
+	if (pci_find_capability(iommu->dev, PCI_CAP_ID_MSI))
 		return iommu_setup_msi(iommu);
 
 	return 1;
@@ -899,6 +956,7 @@ static int __init init_exclusion_range(struct ivmd_header *m)
 static int __init init_unity_map_range(struct ivmd_header *m)
 {
 	struct unity_map_entry *e = 0;
+	char *s;
 
 	e = kzalloc(sizeof(*e), GFP_KERNEL);
 	if (e == NULL)
@@ -909,13 +967,16 @@ static int __init init_unity_map_range(struct ivmd_header *m)
 		kfree(e);
 		return 0;
 	case ACPI_IVMD_TYPE:
+		s = "IVMD_TYPEi\t\t\t";
 		e->devid_start = e->devid_end = m->devid;
 		break;
 	case ACPI_IVMD_TYPE_ALL:
+		s = "IVMD_TYPE_ALL\t\t";
 		e->devid_start = 0;
 		e->devid_end = amd_iommu_last_bdf;
 		break;
 	case ACPI_IVMD_TYPE_RANGE:
+		s = "IVMD_TYPE_RANGE\t\t";
 		e->devid_start = m->devid;
 		e->devid_end = m->aux;
 		break;
@@ -924,6 +985,13 @@ static int __init init_unity_map_range(struct ivmd_header *m)
 	e->address_end = e->address_start + PAGE_ALIGN(m->range_length);
 	e->prot = m->flags >> 1;
 
+	DUMP_printk("%s devid_start: %02x:%02x.%x devid_end: %02x:%02x.%x"
+		    " range_start: %016llx range_end: %016llx flags: %x\n", s,
+		    PCI_BUS(e->devid_start), PCI_SLOT(e->devid_start),
+		    PCI_FUNC(e->devid_start), PCI_BUS(e->devid_end),
+		    PCI_SLOT(e->devid_end), PCI_FUNC(e->devid_end),
+		    e->address_start, e->address_end, m->flags);
+
 	list_add_tail(&e->list, &amd_iommu_unity_map);
 
 	return 0;
@@ -969,18 +1037,28 @@ static void init_device_table(void)
  * This function finally enables all IOMMUs found in the system after
  * they have been initialized
  */
-static void __init enable_iommus(void)
+static void enable_iommus(void)
 {
 	struct amd_iommu *iommu;
 
-	list_for_each_entry(iommu, &amd_iommu_list, list) {
+	for_each_iommu(iommu) {
+		iommu_set_device_table(iommu);
+		iommu_enable_command_buffer(iommu);
+		iommu_enable_event_buffer(iommu);
 		iommu_set_exclusion_range(iommu);
 		iommu_init_msi(iommu);
-		iommu_enable_event_logging(iommu);
 		iommu_enable(iommu);
 	}
 }
 
+static void disable_iommus(void)
+{
+	struct amd_iommu *iommu;
+
+	for_each_iommu(iommu)
+		iommu_disable(iommu);
+}
+
 /*
  * Suspend/Resume support
  * disable suspend until real resume implemented
@@ -988,12 +1066,31 @@ static void __init enable_iommus(void)
 
 static int amd_iommu_resume(struct sys_device *dev)
 {
+	/*
+	 * Disable IOMMUs before reprogramming the hardware registers.
+	 * IOMMU is still enabled from the resume kernel.
+	 */
+	disable_iommus();
+
+	/* re-load the hardware */
+	enable_iommus();
+
+	/*
+	 * we have to flush after the IOMMUs are enabled because a
+	 * disabled IOMMU will never execute the commands we send
+	 */
+	amd_iommu_flush_all_domains();
+	amd_iommu_flush_all_devices();
+
 	return 0;
 }
 
 static int amd_iommu_suspend(struct sys_device *dev, pm_message_t state)
 {
-	return -EINVAL;
+	/* disable IOMMUs to go out of the way for BIOS */
+	disable_iommus();
+
+	return 0;
 }
 
 static struct sysdev_class amd_iommu_sysdev_class = {
@@ -1139,9 +1236,6 @@ int __init amd_iommu_init(void)
 
 	enable_iommus();
 
-	printk(KERN_INFO "AMD IOMMU: aperture size is %d MB\n",
-			(1 << (amd_iommu_aperture_order-20)));
-
 	printk(KERN_INFO "AMD IOMMU: device isolation ");
 	if (amd_iommu_isolate)
 		printk("enabled\n");
@@ -1213,6 +1307,13 @@ void __init amd_iommu_detect(void)
  *
  ****************************************************************************/
 
+static int __init parse_amd_iommu_dump(char *str)
+{
+	amd_iommu_dump = true;
+
+	return 1;
+}
+
 static int __init parse_amd_iommu_options(char *str)
 {
 	for (; *str; ++str) {
@@ -1227,15 +1328,5 @@ static int __init parse_amd_iommu_options(char *str)
 	return 1;
 }
 
-static int __init parse_amd_iommu_size_options(char *str)
-{
-	unsigned order = PAGE_SHIFT + get_order(memparse(str, &str));
-
-	if ((order > 24) && (order < 31))
-		amd_iommu_aperture_order = order;
-
-	return 1;
-}
-
+__setup("amd_iommu_dump", parse_amd_iommu_dump);
 __setup("amd_iommu=", parse_amd_iommu_options);
-__setup("amd_iommu_size=", parse_amd_iommu_size_options);