Skip to contentSkip to conversation

iommu/vt-d: Clean up and fix page table clear/free behaviour

RFC PATCH In review David Woodhouse 2 participants 7 messages Mar 6, 2014 – Mar 20, 2014 3 messages not in the archive IOMMU

lore
Share & tools
lore
#0 lore raw

RFC PATCH 1/1 iommu/vt-d: Clean up and fix page table clear/free behaviour

There is a race condition between the existing clear/free code and the hardware. The IOMMU is actually permitted to cache the intermediate levels of the page tables, and doesn't need to walk the table from the very top of the PGD each time. So the existing back-to-back calls to dma_pte_clear_range() and dma_pte_free_pagetable() can lead to a use-after-free where the IOMMU reads from a freed page table.

When freeing page tables we need to do things in the following order:
- First unlink the pages from the higher-level page tables.
- Then flush the IOTLB (with the 'invalidation hint' bit clear).
- Finally, free the page table pages which are no longer in use.

So in the rewritten domain_unmap() we just return a list of pages (using pg->freelist to make a list of them), and then the caller is expected to do the appropriate IOTLB flush (or tear down the domain completely, whatever), before finally calling dma_free_pagelist() to free the pages.

As an added bonus, we no longer need to flush the CPU's data cache for pages which are about to be *removed* from the page table hierarchy anyway, in the non-cache-coherent case. This drastically improves the performance of large unmaps.

As a side-effect of all these changes, this also fixes the fact that intel_iommu_unmap() was neglecting to free the page tables for the range in question after clearing them.

---
Potential issues:
- Is it OK to use page->freelist like this?

- I've made intel_iommu_unmap() unmap and return the size it was asked to unmap, and assume that it *won't* be asked to do "partial" unmaps, unmapping a smaller size than was originally requested with the corresponding map() call. So it's assuming it'll never be asked to break superpages apart.

The existing API is that *if* it's asked to break superpages apart, it will just unmap the whole superpage and return a 'size' result indicating that it's done so. Does anything really *use* that insanity?

 drivers/iommu/intel-iommu.c | 234 +++++++++++++++++++++++++++++++++++---------
 1 file changed, 188 insertions(+), 46 deletions(-)
drivers/iommu/intel-iommu.c +188 -46
index 484d669..6740942 100644
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006, Intel Corporation.
+ * Copyright © 2006-2014 Intel Corporation.
  *
  * This program is free software; you can redistribute it and/or modify it
  * under the terms and conditions of the GNU General Public License,
@@ -10,15 +10,11 @@
  * 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.
- *
- * Copyright (C) 2006-2008 Intel Corporation
- * Author: Ashok Raj <[email protected]>
- * Author: Shaohua Li <[email protected]>
- * Author: Anil S Keshavamurthy <[email protected]>
- * Author: Fenghua Yu <[email protected]>
+ * Authors: David Woodhouse <[email protected]>,
+ *          Ashok Raj <[email protected]>,
+ *          Shaohua Li <[email protected]>,
+ *          Anil S Keshavamurthy <[email protected]>,
+ *          Fenghua Yu <[email protected]>
  */
 
 #include <linux/init.h>
@@ -413,6 +409,7 @@ struct deferred_flush_tables {
 	int next;
 	struct iova *iova[HIGH_WATER_MARK];
 	struct dmar_domain *domain[HIGH_WATER_MARK];
+	struct page *freelist[HIGH_WATER_MARK];
 };
 
 static struct deferred_flush_tables *deferred_flush;
@@ -957,6 +954,123 @@ static void dma_pte_free_pagetable(struct dmar_domain *domain,
 	}
 }
 
+/* When a page at a given level is being unlinked from its parent, we don't
+   need to *modify* it at all. All we need to do is make a list of all the
+   pages which can be freed just as soon as we've flushed the IOTLB and we
+   know the hardware page-walk will no longer touch them.
+   The 'pte' argument is the *parent* PTE, pointing to the page that is to
+   be freed. */
+static struct page *dma_pte_list_pagetables(struct dmar_domain *domain,
+					    int level, struct dma_pte *pte,
+					    struct page *freelist)
+{
+	struct page *pg;
+
+	pg = pfn_to_page(dma_pte_addr(pte) >> PAGE_SHIFT);
+	pg->freelist = freelist;
+	freelist = pg;
+
+	if (level == 1)
+		return freelist;
+
+	for (pte = page_address(pg); !first_pte_in_page(pte); pte++) {
+		if (dma_pte_present(pte) && !dma_pte_superpage(pte))
+			freelist = dma_pte_list_pagetables(domain, level - 1,
+							   pte, freelist);
+	}
+
+	return freelist;
+}
+
+static struct page *dma_pte_clear_level(struct dmar_domain *domain, int level,
+					struct dma_pte *pte, unsigned long pfn,
+					unsigned long start_pfn,
+					unsigned long last_pfn,
+					struct page *freelist)
+{
+	struct dma_pte *first_pte = NULL, *last_pte = NULL;
+
+	pfn = max(start_pfn, pfn);
+	pte = &pte[pfn_level_offset(pfn, level)];
+
+	do {
+		unsigned long level_pfn;
+
+		if (!dma_pte_present(pte))
+			goto next;
+
+		level_pfn = pfn & level_mask(level);
+
+		/* If range covers entire pagetable, free it */
+		if (start_pfn <= level_pfn &&
+		    last_pfn >= level_pfn + level_size(level) - 1) {
+			/* These suborbinate page tables are going away entirely. Don't
+			   bother to clear them; we're just going to *free* them. */
+			if (level > 1 && !dma_pte_superpage(pte))
+				freelist = dma_pte_list_pagetables(domain, level - 1, pte, freelist);
+
+			dma_clear_pte(pte);
+			if (!first_pte)
+				first_pte = pte;
+			last_pte = pte;
+		} else if (level > 1) {
+			/* Recurse down into a level that isn't *entirely* obsolete */
+			freelist = dma_pte_clear_level(domain, level - 1,
+						       phys_to_virt(dma_pte_addr(pte)),
+						       level_pfn, start_pfn, last_pfn,
+						       freelist);
+		}
+next:
+		pfn += level_size(level);
+	} while (!first_pte_in_page(++pte) && pfn <= last_pfn);
+
+	if (first_pte)
+		domain_flush_cache(domain, first_pte,
+				   (void *)++last_pte - (void *)first_pte);
+
+	return freelist;
+}
+
+/* We can't just free the pages because the IOMMU may still be walking
+   the page tables, and may have cached the intermediate levels. The
+   pages can only be freed after the IOTLB flush has been done. */
+struct page *domain_unmap(struct dmar_domain *domain,
+			  unsigned long start_pfn,
+			  unsigned long last_pfn)
+{
+	int addr_width = agaw_to_width(domain->agaw) - VTD_PAGE_SHIFT;
+	struct page *freelist = NULL;
+
+	BUG_ON(addr_width < BITS_PER_LONG && start_pfn >> addr_width);
+	BUG_ON(addr_width < BITS_PER_LONG && last_pfn >> addr_width);
+	BUG_ON(start_pfn > last_pfn);
+
+	/* we don't need lock here; nobody else touches the iova range */
+	freelist = dma_pte_clear_level(domain, agaw_to_level(domain->agaw),
+				       domain->pgd, 0, start_pfn, last_pfn, NULL);
+
+	/* free pgd */
+	if (start_pfn == 0 && last_pfn == DOMAIN_MAX_PFN(domain->gaw)) {
+		struct page *pgd_page = virt_to_page(domain->pgd);
+		pgd_page->freelist = freelist;
+		freelist = pgd_page;
+
+		domain->pgd = NULL;
+	}
+
+	return freelist;
+}
+
+void dma_free_pagelist(struct page *freelist)
+{
+	struct page *pg;
+
+	while ((pg = freelist)) {
+		freelist = pg->freelist;
+		free_pgtable_page(page_address(pg));
+	}
+}
+
 /* iommu handling */
 static int iommu_alloc_root_entry(struct intel_iommu *iommu)
 {
@@ -1066,7 +1180,7 @@ static void __iommu_flush_iotlb(struct intel_iommu *iommu, u16 did,
 		break;
 	case DMA_TLB_PSI_FLUSH:
 		val = DMA_TLB_PSI_FLUSH|DMA_TLB_IVT|DMA_TLB_DID(did);
-		/* Note: always flush non-leaf currently */
+		/* IH bit is passed in as part of address */
 		val_iva = size_order | addr;
 		break;
 	default:
@@ -1177,13 +1291,15 @@ static void iommu_flush_dev_iotlb(struct dmar_domain *domain,
 }
 
 static void iommu_flush_iotlb_psi(struct intel_iommu *iommu, u16 did,
-				  unsigned long pfn, unsigned int pages, int map)
+				  unsigned long pfn, unsigned int pages, int ih, int map)
 {
 	unsigned int mask = ilog2(__roundup_pow_of_two(pages));
 	uint64_t addr = (uint64_t)pfn << VTD_PAGE_SHIFT;
 
 	BUG_ON(pages == 0);
 
+	if (ih)
+		ih = 1 << 6;
 	/*
 	 * Fallback to domain selective flush if no PSI support or the size is
 	 * too big.
@@ -1194,7 +1310,7 @@ static void iommu_flush_iotlb_psi(struct intel_iommu *iommu, u16 did,
 		iommu->flush.flush_iotlb(iommu, did, 0, 0,
 						DMA_TLB_DSI_FLUSH);
 	else
-		iommu->flush.flush_iotlb(iommu, did, addr, mask,
+		iommu->flush.flush_iotlb(iommu, did, addr | ih, mask,
 						DMA_TLB_PSI_FLUSH);
 
 	/*
@@ -1519,6 +1635,7 @@ static void domain_exit(struct dmar_domain *domain)
 {
 	struct dmar_drhd_unit *drhd;
 	struct intel_iommu *iommu;
+	struct page *freelist = NULL;
 
 	/* Domain 0 is reserved, so dont process it */
 	if (!domain)
@@ -1534,11 +1651,7 @@ static void domain_exit(struct dmar_domain *domain)
 	/* destroy iovas */
 	put_iova_domain(&domain->iovad);
 
-	/* clear ptes */
-	dma_pte_clear_range(domain, 0, DOMAIN_MAX_PFN(domain->gaw));
-
-	/* free page tables */
-	dma_pte_free_pagetable(domain, 0, DOMAIN_MAX_PFN(domain->gaw));
+	freelist = domain_unmap(domain, 0, DOMAIN_MAX_PFN(domain->gaw));
 
 	/* clear attached or cached domains */
 	rcu_read_lock();
@@ -1548,6 +1661,8 @@ static void domain_exit(struct dmar_domain *domain)
 			iommu_detach_domain(domain, iommu);
 	rcu_read_unlock();
 
+	dma_free_pagelist(freelist);
+
 	free_domain_mem(domain);
 }
 
@@ -2847,7 +2962,7 @@ static dma_addr_t __intel_map_single(struct device *hwdev, phys_addr_t paddr,
 
 	/* it's a non-present to present mapping. Only flush if caching mode */
 	if (cap_caching_mode(iommu->cap))
-		iommu_flush_iotlb_psi(iommu, domain->id, mm_to_dma_pfn(iova->pfn_lo), size, 1);
+		iommu_flush_iotlb_psi(iommu, domain->id, mm_to_dma_pfn(iova->pfn_lo), size, 0, 1);
 	else
 		iommu_flush_write_buffer(iommu);
 
@@ -2899,13 +3014,16 @@ static void flush_unmaps(void)
 			/* On real hardware multiple invalidations are expensive */
 			if (cap_caching_mode(iommu->cap))
 				iommu_flush_iotlb_psi(iommu, domain->id,
-				iova->pfn_lo, iova->pfn_hi - iova->pfn_lo + 1, 0);
+					iova->pfn_lo, iova->pfn_hi - iova->pfn_lo + 1,
+					!deferred_flush[i].freelist[j], 0);
 			else {
 				mask = ilog2(mm_to_dma_pfn(iova->pfn_hi - iova->pfn_lo + 1));
 				iommu_flush_dev_iotlb(deferred_flush[i].domain[j],
 						(uint64_t)iova->pfn_lo << PAGE_SHIFT, mask);
 			}
 			__free_iova(&deferred_flush[i].domain[j]->iovad, iova);
+			if (deferred_flush[i].freelist[j])
+				dma_free_pagelist(deferred_flush[i].freelist[j]);
 		}
 		deferred_flush[i].next = 0;
 	}
@@ -2922,7 +3040,7 @@ static void flush_unmaps_timeout(unsigned long data)
 	spin_unlock_irqrestore(&async_umap_flush_lock, flags);
 }
 
-static void add_unmap(struct dmar_domain *dom, struct iova *iova)
+static void add_unmap(struct dmar_domain *dom, struct iova *iova, struct page *freelist)
 {
 	unsigned long flags;
 	int next, iommu_id;
@@ -2938,6 +3056,7 @@ static void add_unmap(struct dmar_domain *dom, struct iova *iova)
 	next = deferred_flush[iommu_id].next;
 	deferred_flush[iommu_id].domain[next] = dom;
 	deferred_flush[iommu_id].iova[next] = iova;
+	deferred_flush[iommu_id].freelist[next] = freelist;
 	deferred_flush[iommu_id].next++;
 
 	if (!timer_on) {
@@ -2957,6 +3076,7 @@ static void intel_unmap_page(struct device *dev, dma_addr_t dev_addr,
 	unsigned long start_pfn, last_pfn;
 	struct iova *iova;
 	struct intel_iommu *iommu;
+	struct page *freelist;
 
 	if (iommu_no_mapping(dev))
 		return;
@@ -2977,19 +3097,16 @@ static void intel_unmap_page(struct device *dev, dma_addr_t dev_addr,
 	pr_debug("Device %s unmapping: pfn %lx-%lx\n",
 		 pci_name(pdev), start_pfn, last_pfn);
 
-	/*  clear the whole page */
-	dma_pte_clear_range(domain, start_pfn, last_pfn);
-
-	/* free page tables */
-	dma_pte_free_pagetable(domain, start_pfn, last_pfn);
+	freelist = domain_unmap(domain, start_pfn, last_pfn);
 
 	if (intel_iommu_strict) {
 		iommu_flush_iotlb_psi(iommu, domain->id, start_pfn,
-				      last_pfn - start_pfn + 1, 0);
+				      last_pfn - start_pfn + 1, !freelist, 0);
 		/* free iova */
 		__free_iova(&domain->iovad, iova);
+		dma_free_pagelist(freelist);
 	} else {
-		add_unmap(domain, iova);
+		add_unmap(domain, iova, freelist);
 		/*
 		 * queue up the release of the unmap to save the 1/6th of the
 		 * cpu used up by the iotlb flush operation...
@@ -3051,6 +3168,7 @@ static void intel_unmap_sg(struct device *hwdev, struct scatterlist *sglist,
 	unsigned long start_pfn, last_pfn;
 	struct iova *iova;
 	struct intel_iommu *iommu;
+	struct page *freelist;
 
 	if (iommu_no_mapping(hwdev))
 		return;
@@ -3068,19 +3186,16 @@ static void intel_unmap_sg(struct device *hwdev, struct scatterlist *sglist,
 	start_pfn = mm_to_dma_pfn(iova->pfn_lo);
 	last_pfn = mm_to_dma_pfn(iova->pfn_hi + 1) - 1;
 
-	/*  clear the whole page */
-	dma_pte_clear_range(domain, start_pfn, last_pfn);
-
-	/* free page tables */
-	dma_pte_free_pagetable(domain, start_pfn, last_pfn);
+	freelist = domain_unmap(domain, start_pfn, last_pfn);
 
 	if (intel_iommu_strict) {
 		iommu_flush_iotlb_psi(iommu, domain->id, start_pfn,
-				      last_pfn - start_pfn + 1, 0);
+				      last_pfn - start_pfn + 1, !freelist, 0);
 		/* free iova */
 		__free_iova(&domain->iovad, iova);
+		dma_free_pagelist(freelist);
 	} else {
-		add_unmap(domain, iova);
+		add_unmap(domain, iova, freelist);
 		/*
 		 * queue up the release of the unmap to save the 1/6th of the
 		 * cpu used up by the iotlb flush operation...
@@ -3163,7 +3278,7 @@ static int intel_map_sg(struct device *hwdev, struct scatterlist *sglist, int ne
 
 	/* it's a non-present to present mapping. Only flush if caching mode */
 	if (cap_caching_mode(iommu->cap))
-		iommu_flush_iotlb_psi(iommu, domain->id, start_vpfn, size, 1);
+		iommu_flush_iotlb_psi(iommu, domain->id, start_vpfn, size, 0, 1);
 	else
 		iommu_flush_write_buffer(iommu);
 
@@ -3710,6 +3825,7 @@ static int intel_iommu_memory_notifier(struct notifier_block *nb,
 			struct iova *iova;
 			struct dmar_drhd_unit *drhd;
 			struct intel_iommu *iommu;
+			struct page *freelist;
 
 			iova = find_iova(&si_domain->iovad, start_vpfn);
 			if (iova == NULL) {
@@ -3726,16 +3842,17 @@ static int intel_iommu_memory_notifier(struct notifier_block *nb,
 				return NOTIFY_BAD;
 			}
 
+			freelist = domain_unmap(si_domain, iova->pfn_lo,
+					       iova->pfn_hi);
+
 			rcu_read_lock();
 			for_each_active_iommu(iommu, drhd)
 				iommu_flush_iotlb_psi(iommu, si_domain->id,
 					iova->pfn_lo,
-					iova->pfn_hi - iova->pfn_lo + 1, 0);
+					iova->pfn_hi - iova->pfn_lo + 1,
+					!freelist, 0);
 			rcu_read_unlock();
-			dma_pte_clear_range(si_domain, iova->pfn_lo,
-					    iova->pfn_hi);
-			dma_pte_free_pagetable(si_domain, iova->pfn_lo,
-					       iova->pfn_hi);
+			dma_free_pagelist(freelist);
 
 			start_vpfn = iova->pfn_hi + 1;
 			free_iova_mem(iova);
@@ -4096,18 +4213,43 @@ static int intel_iommu_map(struct iommu_domain *domain,
 }
 
 static size_t intel_iommu_unmap(struct iommu_domain *domain,
-			     unsigned long iova, size_t size)
+				unsigned long iova, size_t size)
 {
 	struct dmar_domain *dmar_domain = domain->priv;
-	int order;
+	struct page *freelist = NULL;
+	struct intel_iommu *iommu;
+	unsigned long start_pfn, last_pfn;
+	unsigned int npages;
+	int iommu_id, num, ndomains;
+
+	start_pfn = iova >> VTD_PAGE_SHIFT;
+	last_pfn = (iova + size - 1) >> VTD_PAGE_SHIFT;
+
+	freelist = domain_unmap(dmar_domain, start_pfn, last_pfn);
+
+	npages = last_pfn - start_pfn + 1;
+
+	for_each_set_bit(iommu_id, dmar_domain->iommu_bmp, g_num_of_iommus) {
+               iommu = g_iommus[iommu_id];
+
+               /*
+                * find bit position of dmar_domain
+                */
+               ndomains = cap_ndoms(iommu->cap);
+               for_each_set_bit(num, iommu->domain_ids, ndomains) {
+                       if (iommu->domains[num] == dmar_domain)
+                               iommu_flush_iotlb_psi(iommu, num, start_pfn,
+						     npages, !freelist, 0);
+	       }
+
+	}
 
-	order = dma_pte_clear_range(dmar_domain, iova >> VTD_PAGE_SHIFT,
-			    (iova + size - 1) >> VTD_PAGE_SHIFT);
+	dma_free_pagelist(freelist);
 
 	if (dmar_domain->max_addr == iova + size)
 		dmar_domain->max_addr = iova;
 
-	return PAGE_SIZE << order;
+	return size;
 }
 
 static phys_addr_t intel_iommu_iova_to_phys(struct iommu_domain *domain,
Signature · 6 lines
1.8.5.3


-- 
David Woodhouse                            Open Source Technology Centre
[email protected]                              Intel Corporation

A message that is not in the archive (sent off-list or not yet received): 1394125043.9994.13.camel-W2I5cNIroUsVm/[email protected]

#2 lore raw
Quoted 37 lines · David Woodhouse

There is a race condition between the existing clear/free code and the hardware. The IOMMU is actually permitted to cache the intermediate levels of the page tables, and doesn't need to walk the table from the very top of the PGD each time. So the existing back-to-back calls to dma_pte_clear_range() and dma_pte_free_pagetable() can lead to a use-after-free where the IOMMU reads from a freed page table.

When freeing page tables we need to do things in the following order:
- First unlink the pages from the higher-level page tables.
- Then flush the IOTLB (with the 'invalidation hint' bit clear).
- Finally, free the page table pages which are no longer in use.

So in the rewritten domain_unmap() we just return a list of pages (using pg->freelist to make a list of them), and then the caller is expected to do the appropriate IOTLB flush (or tear down the domain completely, whatever), before finally calling dma_free_pagelist() to free the pages.

As an added bonus, we no longer need to flush the CPU's data cache for pages which are about to be *removed* from the page table hierarchy anyway, in the non-cache-coherent case. This drastically improves the performance of large unmaps.

As a side-effect of all these changes, this also fixes the fact that intel_iommu_unmap() was neglecting to free the page tables for the range in question after clearing them.

---
Potential issues:
- Is it OK to use page->freelist like this?

- I've made intel_iommu_unmap() unmap and return the size it was asked to unmap, and assume that it *won't* be asked to do "partial" unmaps, unmapping a smaller size than was originally requested with the corresponding map() call. So it's assuming it'll never be asked to break superpages apart.

Quoted 4 lines

The existing API is that *if* it's asked to break superpages apart, it will just unmap the whole superpage and return a 'size' result indicating that it's done so. Does anything really *use* that insanity?

I think you've likely already figured out, but it should be stated in this thread, yes interfaces do rely on the insane "tell me if you unmapped more than I asked for" API. Specifically KVM and VFIO use this because they get told about an IOVA to virtual address range to map. The IOMMU API maps IOVA to physical addresses, so we may have multiple sets of contiguous physical ranges we can map within the virtual address range. We don't particularly want to create our own page tables for tracking these mappings, so we rely on the IOMMU to do it for us. I don't think anybody is terribly happy with this interface, but if an IOMMU driver were to ignore this quirk, it would break superpage support for KVM & VFIO. Thanks,

Alex

Quoted 437 lines
 drivers/iommu/intel-iommu.c | 234 +++++++++++++++++++++++++++++++++++---------
 1 file changed, 188 insertions(+), 46 deletions(-)
drivers/iommu/intel-iommu.c +188 -46
index 484d669..6740942 100644
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006, Intel Corporation.
+ * Copyright © 2006-2014 Intel Corporation.
  *
  * This program is free software; you can redistribute it and/or modify it
  * under the terms and conditions of the GNU General Public License,
@@ -10,15 +10,11 @@
  * 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.
- *
- * Copyright (C) 2006-2008 Intel Corporation
- * Author: Ashok Raj <[email protected]>
- * Author: Shaohua Li <[email protected]>
- * Author: Anil S Keshavamurthy <[email protected]>
- * Author: Fenghua Yu <[email protected]>
+ * Authors: David Woodhouse <[email protected]>,
+ *          Ashok Raj <[email protected]>,
+ *          Shaohua Li <[email protected]>,
+ *          Anil S Keshavamurthy <[email protected]>,
+ *          Fenghua Yu <[email protected]>
  */
 
 #include <linux/init.h>
@@ -413,6 +409,7 @@ struct deferred_flush_tables {
 	int next;
 	struct iova *iova[HIGH_WATER_MARK];
 	struct dmar_domain *domain[HIGH_WATER_MARK];
+	struct page *freelist[HIGH_WATER_MARK];
 };
 
 static struct deferred_flush_tables *deferred_flush;
@@ -957,6 +954,123 @@ static void dma_pte_free_pagetable(struct dmar_domain *domain,
 	}
 }
 
+/* When a page at a given level is being unlinked from its parent, we don't
+   need to *modify* it at all. All we need to do is make a list of all the
+   pages which can be freed just as soon as we've flushed the IOTLB and we
+   know the hardware page-walk will no longer touch them.
+   The 'pte' argument is the *parent* PTE, pointing to the page that is to
+   be freed. */
+static struct page *dma_pte_list_pagetables(struct dmar_domain *domain,
+					    int level, struct dma_pte *pte,
+					    struct page *freelist)
+{
+	struct page *pg;
+
+	pg = pfn_to_page(dma_pte_addr(pte) >> PAGE_SHIFT);
+	pg->freelist = freelist;
+	freelist = pg;
+
+	if (level == 1)
+		return freelist;
+
+	for (pte = page_address(pg); !first_pte_in_page(pte); pte++) {
+		if (dma_pte_present(pte) && !dma_pte_superpage(pte))
+			freelist = dma_pte_list_pagetables(domain, level - 1,
+							   pte, freelist);
+	}
+
+	return freelist;
+}
+
+static struct page *dma_pte_clear_level(struct dmar_domain *domain, int level,
+					struct dma_pte *pte, unsigned long pfn,
+					unsigned long start_pfn,
+					unsigned long last_pfn,
+					struct page *freelist)
+{
+	struct dma_pte *first_pte = NULL, *last_pte = NULL;
+
+	pfn = max(start_pfn, pfn);
+	pte = &pte[pfn_level_offset(pfn, level)];
+
+	do {
+		unsigned long level_pfn;
+
+		if (!dma_pte_present(pte))
+			goto next;
+
+		level_pfn = pfn & level_mask(level);
+
+		/* If range covers entire pagetable, free it */
+		if (start_pfn <= level_pfn &&
+		    last_pfn >= level_pfn + level_size(level) - 1) {
+			/* These suborbinate page tables are going away entirely. Don't
+			   bother to clear them; we're just going to *free* them. */
+			if (level > 1 && !dma_pte_superpage(pte))
+				freelist = dma_pte_list_pagetables(domain, level - 1, pte, freelist);
+
+			dma_clear_pte(pte);
+			if (!first_pte)
+				first_pte = pte;
+			last_pte = pte;
+		} else if (level > 1) {
+			/* Recurse down into a level that isn't *entirely* obsolete */
+			freelist = dma_pte_clear_level(domain, level - 1,
+						       phys_to_virt(dma_pte_addr(pte)),
+						       level_pfn, start_pfn, last_pfn,
+						       freelist);
+		}
+next:
+		pfn += level_size(level);
+	} while (!first_pte_in_page(++pte) && pfn <= last_pfn);
+
+	if (first_pte)
+		domain_flush_cache(domain, first_pte,
+				   (void *)++last_pte - (void *)first_pte);
+
+	return freelist;
+}
+
+/* We can't just free the pages because the IOMMU may still be walking
+   the page tables, and may have cached the intermediate levels. The
+   pages can only be freed after the IOTLB flush has been done. */
+struct page *domain_unmap(struct dmar_domain *domain,
+			  unsigned long start_pfn,
+			  unsigned long last_pfn)
+{
+	int addr_width = agaw_to_width(domain->agaw) - VTD_PAGE_SHIFT;
+	struct page *freelist = NULL;
+
+	BUG_ON(addr_width < BITS_PER_LONG && start_pfn >> addr_width);
+	BUG_ON(addr_width < BITS_PER_LONG && last_pfn >> addr_width);
+	BUG_ON(start_pfn > last_pfn);
+
+	/* we don't need lock here; nobody else touches the iova range */
+	freelist = dma_pte_clear_level(domain, agaw_to_level(domain->agaw),
+				       domain->pgd, 0, start_pfn, last_pfn, NULL);
+
+	/* free pgd */
+	if (start_pfn == 0 && last_pfn == DOMAIN_MAX_PFN(domain->gaw)) {
+		struct page *pgd_page = virt_to_page(domain->pgd);
+		pgd_page->freelist = freelist;
+		freelist = pgd_page;
+
+		domain->pgd = NULL;
+	}
+
+	return freelist;
+}
+
+void dma_free_pagelist(struct page *freelist)
+{
+	struct page *pg;
+
+	while ((pg = freelist)) {
+		freelist = pg->freelist;
+		free_pgtable_page(page_address(pg));
+	}
+}
+
 /* iommu handling */
 static int iommu_alloc_root_entry(struct intel_iommu *iommu)
 {
@@ -1066,7 +1180,7 @@ static void __iommu_flush_iotlb(struct intel_iommu *iommu, u16 did,
 		break;
 	case DMA_TLB_PSI_FLUSH:
 		val = DMA_TLB_PSI_FLUSH|DMA_TLB_IVT|DMA_TLB_DID(did);
-		/* Note: always flush non-leaf currently */
+		/* IH bit is passed in as part of address */
 		val_iva = size_order | addr;
 		break;
 	default:
@@ -1177,13 +1291,15 @@ static void iommu_flush_dev_iotlb(struct dmar_domain *domain,
 }
 
 static void iommu_flush_iotlb_psi(struct intel_iommu *iommu, u16 did,
-				  unsigned long pfn, unsigned int pages, int map)
+				  unsigned long pfn, unsigned int pages, int ih, int map)
 {
 	unsigned int mask = ilog2(__roundup_pow_of_two(pages));
 	uint64_t addr = (uint64_t)pfn << VTD_PAGE_SHIFT;
 
 	BUG_ON(pages == 0);
 
+	if (ih)
+		ih = 1 << 6;
 	/*
 	 * Fallback to domain selective flush if no PSI support or the size is
 	 * too big.
@@ -1194,7 +1310,7 @@ static void iommu_flush_iotlb_psi(struct intel_iommu *iommu, u16 did,
 		iommu->flush.flush_iotlb(iommu, did, 0, 0,
 						DMA_TLB_DSI_FLUSH);
 	else
-		iommu->flush.flush_iotlb(iommu, did, addr, mask,
+		iommu->flush.flush_iotlb(iommu, did, addr | ih, mask,
 						DMA_TLB_PSI_FLUSH);
 
 	/*
@@ -1519,6 +1635,7 @@ static void domain_exit(struct dmar_domain *domain)
 {
 	struct dmar_drhd_unit *drhd;
 	struct intel_iommu *iommu;
+	struct page *freelist = NULL;
 
 	/* Domain 0 is reserved, so dont process it */
 	if (!domain)
@@ -1534,11 +1651,7 @@ static void domain_exit(struct dmar_domain *domain)
 	/* destroy iovas */
 	put_iova_domain(&domain->iovad);
 
-	/* clear ptes */
-	dma_pte_clear_range(domain, 0, DOMAIN_MAX_PFN(domain->gaw));
-
-	/* free page tables */
-	dma_pte_free_pagetable(domain, 0, DOMAIN_MAX_PFN(domain->gaw));
+	freelist = domain_unmap(domain, 0, DOMAIN_MAX_PFN(domain->gaw));
 
 	/* clear attached or cached domains */
 	rcu_read_lock();
@@ -1548,6 +1661,8 @@ static void domain_exit(struct dmar_domain *domain)
 			iommu_detach_domain(domain, iommu);
 	rcu_read_unlock();
 
+	dma_free_pagelist(freelist);
+
 	free_domain_mem(domain);
 }
 
@@ -2847,7 +2962,7 @@ static dma_addr_t __intel_map_single(struct device *hwdev, phys_addr_t paddr,
 
 	/* it's a non-present to present mapping. Only flush if caching mode */
 	if (cap_caching_mode(iommu->cap))
-		iommu_flush_iotlb_psi(iommu, domain->id, mm_to_dma_pfn(iova->pfn_lo), size, 1);
+		iommu_flush_iotlb_psi(iommu, domain->id, mm_to_dma_pfn(iova->pfn_lo), size, 0, 1);
 	else
 		iommu_flush_write_buffer(iommu);
 
@@ -2899,13 +3014,16 @@ static void flush_unmaps(void)
 			/* On real hardware multiple invalidations are expensive */
 			if (cap_caching_mode(iommu->cap))
 				iommu_flush_iotlb_psi(iommu, domain->id,
-				iova->pfn_lo, iova->pfn_hi - iova->pfn_lo + 1, 0);
+					iova->pfn_lo, iova->pfn_hi - iova->pfn_lo + 1,
+					!deferred_flush[i].freelist[j], 0);
 			else {
 				mask = ilog2(mm_to_dma_pfn(iova->pfn_hi - iova->pfn_lo + 1));
 				iommu_flush_dev_iotlb(deferred_flush[i].domain[j],
 						(uint64_t)iova->pfn_lo << PAGE_SHIFT, mask);
 			}
 			__free_iova(&deferred_flush[i].domain[j]->iovad, iova);
+			if (deferred_flush[i].freelist[j])
+				dma_free_pagelist(deferred_flush[i].freelist[j]);
 		}
 		deferred_flush[i].next = 0;
 	}
@@ -2922,7 +3040,7 @@ static void flush_unmaps_timeout(unsigned long data)
 	spin_unlock_irqrestore(&async_umap_flush_lock, flags);
 }
 
-static void add_unmap(struct dmar_domain *dom, struct iova *iova)
+static void add_unmap(struct dmar_domain *dom, struct iova *iova, struct page *freelist)
 {
 	unsigned long flags;
 	int next, iommu_id;
@@ -2938,6 +3056,7 @@ static void add_unmap(struct dmar_domain *dom, struct iova *iova)
 	next = deferred_flush[iommu_id].next;
 	deferred_flush[iommu_id].domain[next] = dom;
 	deferred_flush[iommu_id].iova[next] = iova;
+	deferred_flush[iommu_id].freelist[next] = freelist;
 	deferred_flush[iommu_id].next++;
 
 	if (!timer_on) {
@@ -2957,6 +3076,7 @@ static void intel_unmap_page(struct device *dev, dma_addr_t dev_addr,
 	unsigned long start_pfn, last_pfn;
 	struct iova *iova;
 	struct intel_iommu *iommu;
+	struct page *freelist;
 
 	if (iommu_no_mapping(dev))
 		return;
@@ -2977,19 +3097,16 @@ static void intel_unmap_page(struct device *dev, dma_addr_t dev_addr,
 	pr_debug("Device %s unmapping: pfn %lx-%lx\n",
 		 pci_name(pdev), start_pfn, last_pfn);
 
-	/*  clear the whole page */
-	dma_pte_clear_range(domain, start_pfn, last_pfn);
-
-	/* free page tables */
-	dma_pte_free_pagetable(domain, start_pfn, last_pfn);
+	freelist = domain_unmap(domain, start_pfn, last_pfn);
 
 	if (intel_iommu_strict) {
 		iommu_flush_iotlb_psi(iommu, domain->id, start_pfn,
-				      last_pfn - start_pfn + 1, 0);
+				      last_pfn - start_pfn + 1, !freelist, 0);
 		/* free iova */
 		__free_iova(&domain->iovad, iova);
+		dma_free_pagelist(freelist);
 	} else {
-		add_unmap(domain, iova);
+		add_unmap(domain, iova, freelist);
 		/*
 		 * queue up the release of the unmap to save the 1/6th of the
 		 * cpu used up by the iotlb flush operation...
@@ -3051,6 +3168,7 @@ static void intel_unmap_sg(struct device *hwdev, struct scatterlist *sglist,
 	unsigned long start_pfn, last_pfn;
 	struct iova *iova;
 	struct intel_iommu *iommu;
+	struct page *freelist;
 
 	if (iommu_no_mapping(hwdev))
 		return;
@@ -3068,19 +3186,16 @@ static void intel_unmap_sg(struct device *hwdev, struct scatterlist *sglist,
 	start_pfn = mm_to_dma_pfn(iova->pfn_lo);
 	last_pfn = mm_to_dma_pfn(iova->pfn_hi + 1) - 1;
 
-	/*  clear the whole page */
-	dma_pte_clear_range(domain, start_pfn, last_pfn);
-
-	/* free page tables */
-	dma_pte_free_pagetable(domain, start_pfn, last_pfn);
+	freelist = domain_unmap(domain, start_pfn, last_pfn);
 
 	if (intel_iommu_strict) {
 		iommu_flush_iotlb_psi(iommu, domain->id, start_pfn,
-				      last_pfn - start_pfn + 1, 0);
+				      last_pfn - start_pfn + 1, !freelist, 0);
 		/* free iova */
 		__free_iova(&domain->iovad, iova);
+		dma_free_pagelist(freelist);
 	} else {
-		add_unmap(domain, iova);
+		add_unmap(domain, iova, freelist);
 		/*
 		 * queue up the release of the unmap to save the 1/6th of the
 		 * cpu used up by the iotlb flush operation...
@@ -3163,7 +3278,7 @@ static int intel_map_sg(struct device *hwdev, struct scatterlist *sglist, int ne
 
 	/* it's a non-present to present mapping. Only flush if caching mode */
 	if (cap_caching_mode(iommu->cap))
-		iommu_flush_iotlb_psi(iommu, domain->id, start_vpfn, size, 1);
+		iommu_flush_iotlb_psi(iommu, domain->id, start_vpfn, size, 0, 1);
 	else
 		iommu_flush_write_buffer(iommu);
 
@@ -3710,6 +3825,7 @@ static int intel_iommu_memory_notifier(struct notifier_block *nb,
 			struct iova *iova;
 			struct dmar_drhd_unit *drhd;
 			struct intel_iommu *iommu;
+			struct page *freelist;
 
 			iova = find_iova(&si_domain->iovad, start_vpfn);
 			if (iova == NULL) {
@@ -3726,16 +3842,17 @@ static int intel_iommu_memory_notifier(struct notifier_block *nb,
 				return NOTIFY_BAD;
 			}
 
+			freelist = domain_unmap(si_domain, iova->pfn_lo,
+					       iova->pfn_hi);
+
 			rcu_read_lock();
 			for_each_active_iommu(iommu, drhd)
 				iommu_flush_iotlb_psi(iommu, si_domain->id,
 					iova->pfn_lo,
-					iova->pfn_hi - iova->pfn_lo + 1, 0);
+					iova->pfn_hi - iova->pfn_lo + 1,
+					!freelist, 0);
 			rcu_read_unlock();
-			dma_pte_clear_range(si_domain, iova->pfn_lo,
-					    iova->pfn_hi);
-			dma_pte_free_pagetable(si_domain, iova->pfn_lo,
-					       iova->pfn_hi);
+			dma_free_pagelist(freelist);
 
 			start_vpfn = iova->pfn_hi + 1;
 			free_iova_mem(iova);
@@ -4096,18 +4213,43 @@ static int intel_iommu_map(struct iommu_domain *domain,
 }
 
 static size_t intel_iommu_unmap(struct iommu_domain *domain,
-			     unsigned long iova, size_t size)
+				unsigned long iova, size_t size)
 {
 	struct dmar_domain *dmar_domain = domain->priv;
-	int order;
+	struct page *freelist = NULL;
+	struct intel_iommu *iommu;
+	unsigned long start_pfn, last_pfn;
+	unsigned int npages;
+	int iommu_id, num, ndomains;
+
+	start_pfn = iova >> VTD_PAGE_SHIFT;
+	last_pfn = (iova + size - 1) >> VTD_PAGE_SHIFT;
+
+	freelist = domain_unmap(dmar_domain, start_pfn, last_pfn);
+
+	npages = last_pfn - start_pfn + 1;
+
+	for_each_set_bit(iommu_id, dmar_domain->iommu_bmp, g_num_of_iommus) {
+               iommu = g_iommus[iommu_id];
+
+               /*
+                * find bit position of dmar_domain
+                */
+               ndomains = cap_ndoms(iommu->cap);
+               for_each_set_bit(num, iommu->domain_ids, ndomains) {
+                       if (iommu->domains[num] == dmar_domain)
+                               iommu_flush_iotlb_psi(iommu, num, start_pfn,
+						     npages, !freelist, 0);
+	       }
+
+	}
 
-	order = dma_pte_clear_range(dmar_domain, iova >> VTD_PAGE_SHIFT,
-			    (iova + size - 1) >> VTD_PAGE_SHIFT);
+	dma_free_pagelist(freelist);
 
 	if (dmar_domain->max_addr == iova + size)
 		dmar_domain->max_addr = iova;
 
-	return PAGE_SIZE << order;
+	return size;
 }
 
 static phys_addr_t intel_iommu_iova_to_phys(struct iommu_domain *domain,
Signature · 7 lines
1.8.5.3


_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu

_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu

A message that is not in the archive (sent off-list or not yet received): 1395070209.8201.8.camel-85EaTFmN5p//[email protected]

#4 lore raw
Quoted 4 lines · Alex Williamson

I think you've likely already figured out, but it should be stated in this thread, yes interfaces do rely on the insane "tell me if you unmapped more than I asked for" API.

I'm going to decouple this patch from that question by adding a simple hack at the start of iommu_unmap() to look at the page tables and see the size of page that happens to be mapped, then increase the size parameter accordingly. Then assume sanity from that point onwards.

That's basically the same as my other suggestion of augmenting iommu_iova_to_phys() to return the size information, but as a private implementation detail.

Signature · 1 line
dwmw2

A message that is not in the archive (sent off-list or not yet received): 1395073392.4588.51.camel-W2I5cNIroUsVm/[email protected]

#6 lore raw

PATCH 1/4 iommu/vt-d: Clean up size handling for intel_iommu_unmap()

We have this horrid API where iommu_unmap() can unmap more than it's asked to, if the IOVA in question happens to be mapped with a large page.

Instead of propagating this nonsense to the point where we end up returning the page order from dma_pte_clear_range(), let's just do it once and adjust the 'size' parameter accordingly.

Augment pfn_to_dma_pte() to return the level at which the PTE was found, which will also be useful later if we end up changing the API for iommu_iova_to_phys() to behave the same way as is being discussed upstream.

---
 drivers/iommu/intel-iommu.c | 37 +++++++++++++++++++++++++------------
 1 file changed, 25 insertions(+), 12 deletions(-)
drivers/iommu/intel-iommu.c +25 -12
index 484d669..6472bf1 100644
@@ -784,7 +784,7 @@ out:
 }
 
 static struct dma_pte *pfn_to_dma_pte(struct dmar_domain *domain,
-				      unsigned long pfn, int target_level)
+				      unsigned long pfn, int *target_level)
 {
 	int addr_width = agaw_to_width(domain->agaw) - VTD_PAGE_SHIFT;
 	struct dma_pte *parent, *pte = NULL;
@@ -799,14 +799,14 @@ static struct dma_pte *pfn_to_dma_pte(struct dmar_domain *domain,
 
 	parent = domain->pgd;
 
-	while (level > 0) {
+	while (1) {
 		void *tmp_page;
 
 		offset = pfn_level_offset(pfn, level);
 		pte = &parent[offset];
-		if (!target_level && (dma_pte_superpage(pte) || !dma_pte_present(pte)))
+		if (!*target_level && (dma_pte_superpage(pte) || !dma_pte_present(pte)))
 			break;
-		if (level == target_level)
+		if (level == *target_level)
 			break;
 
 		if (!dma_pte_present(pte)) {
@@ -827,10 +827,16 @@ static struct dma_pte *pfn_to_dma_pte(struct dmar_domain *domain,
 				domain_flush_cache(domain, pte, sizeof(*pte));
 			}
 		}
+		if (level == 1)
+			break;
+
 		parent = phys_to_virt(dma_pte_addr(pte));
 		level--;
 	}
 
+	if (!*target_level)
+		*target_level = level;
+
 	return pte;
 }
 
@@ -868,7 +874,7 @@ static struct dma_pte *dma_pfn_level_pte(struct dmar_domain *domain,
 }
 
 /* clear last level pte, a tlb flush should be followed */
-static int dma_pte_clear_range(struct dmar_domain *domain,
+static void dma_pte_clear_range(struct dmar_domain *domain,
 				unsigned long start_pfn,
 				unsigned long last_pfn)
 {
@@ -898,8 +904,6 @@ static int dma_pte_clear_range(struct dmar_domain *domain,
 				   (void *)pte - (void *)first_pte);
 
 	} while (start_pfn && start_pfn <= last_pfn);
-
-	return min_t(int, (large_page - 1) * 9, MAX_AGAW_PFN_WIDTH);
 }
 
 static void dma_pte_free_level(struct dmar_domain *domain, int level,
@@ -1832,7 +1836,7 @@ static int __domain_mapping(struct dmar_domain *domain, unsigned long iov_pfn,
 		if (!pte) {
 			largepage_lvl = hardware_largepage_caps(domain, iov_pfn, phys_pfn, sg_res);
 
-			first_pte = pte = pfn_to_dma_pte(domain, iov_pfn, largepage_lvl);
+			first_pte = pte = pfn_to_dma_pte(domain, iov_pfn, &largepage_lvl);
 			if (!pte)
 				return -ENOMEM;
 			/* It is large page*/
@@ -4099,15 +4103,23 @@ static size_t intel_iommu_unmap(struct iommu_domain *domain,
 			     unsigned long iova, size_t size)
 {
 	struct dmar_domain *dmar_domain = domain->priv;
-	int order;
+	int level = 0;
+
+	/* Cope with horrid API which requires us to unmap more than the
+	   size argument if it happens to be a large-page mapping. */
+	if (!pfn_to_dma_pte(dmar_domain, iova >> VTD_PAGE_SHIFT, &level))
+		BUG();
+
+	if (size < VTD_PAGE_SIZE << level_to_offset_bits(level))
+		size = VTD_PAGE_SIZE << level_to_offset_bits(level);
 
-	order = dma_pte_clear_range(dmar_domain, iova >> VTD_PAGE_SHIFT,
+	dma_pte_clear_range(dmar_domain, iova >> VTD_PAGE_SHIFT,
 			    (iova + size - 1) >> VTD_PAGE_SHIFT);
 
 	if (dmar_domain->max_addr == iova + size)
 		dmar_domain->max_addr = iova;
 
-	return PAGE_SIZE << order;
+	return size;
 }
 
 static phys_addr_t intel_iommu_iova_to_phys(struct iommu_domain *domain,
@@ -4115,9 +4127,10 @@ static phys_addr_t intel_iommu_iova_to_phys(struct iommu_domain *domain,
 {
 	struct dmar_domain *dmar_domain = domain->priv;
 	struct dma_pte *pte;
+	int level = 0;
 	u64 phys = 0;
 
-	pte = pfn_to_dma_pte(dmar_domain, iova >> VTD_PAGE_SHIFT, 0);
+	pte = pfn_to_dma_pte(dmar_domain, iova >> VTD_PAGE_SHIFT, &level);
 	if (pte)
 		phys = dma_pte_addr(pte);
 
Signature · 5 lines
1.8.5.3


-- 
dwmw2
#7 lore raw

PATCH 2/4 iommu/vt-d: Clean up and fix page table clear/free behaviour

In reply to a message not in the archive

There is a race condition between the existing clear/free code and the hardware. The IOMMU is actually permitted to cache the intermediate levels of the page tables, and doesn't need to walk the table from the very top of the PGD each time. So the existing back-to-back calls to dma_pte_clear_range() and dma_pte_free_pagetable() can lead to a use-after-free where the IOMMU reads from a freed page table.

When freeing page tables we actually need to do the IOTLB flush, with the 'invalidation hint' bit clear to indicate that it's not just a leaf-node flush, after unlinking each page table page from the next level up but before actually freeing it.

So in the rewritten domain_unmap() we just return a list of pages (using pg->freelist to make a list of them), and then the caller is expected to do the appropriate IOTLB flush (or tear down the domain completely, whatever), before finally calling dma_free_pagelist() to free the pages.

As an added bonus, we no longer need to flush the CPU's data cache for pages which are about to be *removed* from the page table hierarchy anyway, in the non-cache-coherent case. This drastically improves the performance of large unmaps.

As a side-effect of all these changes, this also fixes the fact that intel_iommu_unmap() was neglecting to free the page tables for the range in question after clearing them.

---
 drivers/iommu/intel-iommu.c | 232 +++++++++++++++++++++++++++++++++++---------
 1 file changed, 187 insertions(+), 45 deletions(-)
drivers/iommu/intel-iommu.c +187 -45
index 6472bf1..f5934fc 100644
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006, Intel Corporation.
+ * Copyright © 2006-2014 Intel Corporation.
  *
  * This program is free software; you can redistribute it and/or modify it
  * under the terms and conditions of the GNU General Public License,
@@ -10,15 +10,11 @@
  * 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.
- *
- * Copyright (C) 2006-2008 Intel Corporation
- * Author: Ashok Raj <[email protected]>
- * Author: Shaohua Li <[email protected]>
- * Author: Anil S Keshavamurthy <[email protected]>
- * Author: Fenghua Yu <[email protected]>
+ * Authors: David Woodhouse <[email protected]>,
+ *          Ashok Raj <[email protected]>,
+ *          Shaohua Li <[email protected]>,
+ *          Anil S Keshavamurthy <[email protected]>,
+ *          Fenghua Yu <[email protected]>
  */
 
 #include <linux/init.h>
@@ -413,6 +409,7 @@ struct deferred_flush_tables {
 	int next;
 	struct iova *iova[HIGH_WATER_MARK];
 	struct dmar_domain *domain[HIGH_WATER_MARK];
+	struct page *freelist[HIGH_WATER_MARK];
 };
 
 static struct deferred_flush_tables *deferred_flush;
@@ -961,6 +958,123 @@ static void dma_pte_free_pagetable(struct dmar_domain *domain,
 	}
 }
 
+/* When a page at a given level is being unlinked from its parent, we don't
+   need to *modify* it at all. All we need to do is make a list of all the
+   pages which can be freed just as soon as we've flushed the IOTLB and we
+   know the hardware page-walk will no longer touch them.
+   The 'pte' argument is the *parent* PTE, pointing to the page that is to
+   be freed. */
+static struct page *dma_pte_list_pagetables(struct dmar_domain *domain,
+					    int level, struct dma_pte *pte,
+					    struct page *freelist)
+{
+	struct page *pg;
+
+	pg = pfn_to_page(dma_pte_addr(pte) >> PAGE_SHIFT);
+	pg->freelist = freelist;
+	freelist = pg;
+
+	if (level == 1)
+		return freelist;
+
+	for (pte = page_address(pg); !first_pte_in_page(pte); pte++) {
+		if (dma_pte_present(pte) && !dma_pte_superpage(pte))
+			freelist = dma_pte_list_pagetables(domain, level - 1,
+							   pte, freelist);
+	}
+
+	return freelist;
+}
+
+static struct page *dma_pte_clear_level(struct dmar_domain *domain, int level,
+					struct dma_pte *pte, unsigned long pfn,
+					unsigned long start_pfn,
+					unsigned long last_pfn,
+					struct page *freelist)
+{
+	struct dma_pte *first_pte = NULL, *last_pte = NULL;
+
+	pfn = max(start_pfn, pfn);
+	pte = &pte[pfn_level_offset(pfn, level)];
+
+	do {
+		unsigned long level_pfn;
+
+		if (!dma_pte_present(pte))
+			goto next;
+
+		level_pfn = pfn & level_mask(level);
+
+		/* If range covers entire pagetable, free it */
+		if (start_pfn <= level_pfn &&
+		    last_pfn >= level_pfn + level_size(level) - 1) {
+			/* These suborbinate page tables are going away entirely. Don't
+			   bother to clear them; we're just going to *free* them. */
+			if (level > 1 && !dma_pte_superpage(pte))
+				freelist = dma_pte_list_pagetables(domain, level - 1, pte, freelist);
+
+			dma_clear_pte(pte);
+			if (!first_pte)
+				first_pte = pte;
+			last_pte = pte;
+		} else if (level > 1) {
+			/* Recurse down into a level that isn't *entirely* obsolete */
+			freelist = dma_pte_clear_level(domain, level - 1,
+						       phys_to_virt(dma_pte_addr(pte)),
+						       level_pfn, start_pfn, last_pfn,
+						       freelist);
+		}
+next:
+		pfn += level_size(level);
+	} while (!first_pte_in_page(++pte) && pfn <= last_pfn);
+
+	if (first_pte)
+		domain_flush_cache(domain, first_pte,
+				   (void *)++last_pte - (void *)first_pte);
+
+	return freelist;
+}
+
+/* We can't just free the pages because the IOMMU may still be walking
+   the page tables, and may have cached the intermediate levels. The
+   pages can only be freed after the IOTLB flush has been done. */
+struct page *domain_unmap(struct dmar_domain *domain,
+			  unsigned long start_pfn,
+			  unsigned long last_pfn)
+{
+	int addr_width = agaw_to_width(domain->agaw) - VTD_PAGE_SHIFT;
+	struct page *freelist = NULL;
+
+	BUG_ON(addr_width < BITS_PER_LONG && start_pfn >> addr_width);
+	BUG_ON(addr_width < BITS_PER_LONG && last_pfn >> addr_width);
+	BUG_ON(start_pfn > last_pfn);
+
+	/* we don't need lock here; nobody else touches the iova range */
+	freelist = dma_pte_clear_level(domain, agaw_to_level(domain->agaw),
+				       domain->pgd, 0, start_pfn, last_pfn, NULL);
+
+	/* free pgd */
+	if (start_pfn == 0 && last_pfn == DOMAIN_MAX_PFN(domain->gaw)) {
+		struct page *pgd_page = virt_to_page(domain->pgd);
+		pgd_page->freelist = freelist;
+		freelist = pgd_page;
+
+		domain->pgd = NULL;
+	}
+
+	return freelist;
+}
+
+void dma_free_pagelist(struct page *freelist)
+{
+	struct page *pg;
+
+	while ((pg = freelist)) {
+		freelist = pg->freelist;
+		free_pgtable_page(page_address(pg));
+	}
+}
+
 /* iommu handling */
 static int iommu_alloc_root_entry(struct intel_iommu *iommu)
 {
@@ -1070,7 +1184,7 @@ static void __iommu_flush_iotlb(struct intel_iommu *iommu, u16 did,
 		break;
 	case DMA_TLB_PSI_FLUSH:
 		val = DMA_TLB_PSI_FLUSH|DMA_TLB_IVT|DMA_TLB_DID(did);
-		/* Note: always flush non-leaf currently */
+		/* IH bit is passed in as part of address */
 		val_iva = size_order | addr;
 		break;
 	default:
@@ -1181,13 +1295,15 @@ static void iommu_flush_dev_iotlb(struct dmar_domain *domain,
 }
 
 static void iommu_flush_iotlb_psi(struct intel_iommu *iommu, u16 did,
-				  unsigned long pfn, unsigned int pages, int map)
+				  unsigned long pfn, unsigned int pages, int ih, int map)
 {
 	unsigned int mask = ilog2(__roundup_pow_of_two(pages));
 	uint64_t addr = (uint64_t)pfn << VTD_PAGE_SHIFT;
 
 	BUG_ON(pages == 0);
 
+	if (ih)
+		ih = 1 << 6;
 	/*
 	 * Fallback to domain selective flush if no PSI support or the size is
 	 * too big.
@@ -1198,7 +1314,7 @@ static void iommu_flush_iotlb_psi(struct intel_iommu *iommu, u16 did,
 		iommu->flush.flush_iotlb(iommu, did, 0, 0,
 						DMA_TLB_DSI_FLUSH);
 	else
-		iommu->flush.flush_iotlb(iommu, did, addr, mask,
+		iommu->flush.flush_iotlb(iommu, did, addr | ih, mask,
 						DMA_TLB_PSI_FLUSH);
 
 	/*
@@ -1523,6 +1639,7 @@ static void domain_exit(struct dmar_domain *domain)
 {
 	struct dmar_drhd_unit *drhd;
 	struct intel_iommu *iommu;
+	struct page *freelist = NULL;
 
 	/* Domain 0 is reserved, so dont process it */
 	if (!domain)
@@ -1538,11 +1655,7 @@ static void domain_exit(struct dmar_domain *domain)
 	/* destroy iovas */
 	put_iova_domain(&domain->iovad);
 
-	/* clear ptes */
-	dma_pte_clear_range(domain, 0, DOMAIN_MAX_PFN(domain->gaw));
-
-	/* free page tables */
-	dma_pte_free_pagetable(domain, 0, DOMAIN_MAX_PFN(domain->gaw));
+	freelist = domain_unmap(domain, 0, DOMAIN_MAX_PFN(domain->gaw));
 
 	/* clear attached or cached domains */
 	rcu_read_lock();
@@ -1552,6 +1665,8 @@ static void domain_exit(struct dmar_domain *domain)
 			iommu_detach_domain(domain, iommu);
 	rcu_read_unlock();
 
+	dma_free_pagelist(freelist);
+
 	free_domain_mem(domain);
 }
 
@@ -2851,7 +2966,7 @@ static dma_addr_t __intel_map_single(struct device *hwdev, phys_addr_t paddr,
 
 	/* it's a non-present to present mapping. Only flush if caching mode */
 	if (cap_caching_mode(iommu->cap))
-		iommu_flush_iotlb_psi(iommu, domain->id, mm_to_dma_pfn(iova->pfn_lo), size, 1);
+		iommu_flush_iotlb_psi(iommu, domain->id, mm_to_dma_pfn(iova->pfn_lo), size, 0, 1);
 	else
 		iommu_flush_write_buffer(iommu);
 
@@ -2903,13 +3018,16 @@ static void flush_unmaps(void)
 			/* On real hardware multiple invalidations are expensive */
 			if (cap_caching_mode(iommu->cap))
 				iommu_flush_iotlb_psi(iommu, domain->id,
-				iova->pfn_lo, iova->pfn_hi - iova->pfn_lo + 1, 0);
+					iova->pfn_lo, iova->pfn_hi - iova->pfn_lo + 1,
+					!deferred_flush[i].freelist[j], 0);
 			else {
 				mask = ilog2(mm_to_dma_pfn(iova->pfn_hi - iova->pfn_lo + 1));
 				iommu_flush_dev_iotlb(deferred_flush[i].domain[j],
 						(uint64_t)iova->pfn_lo << PAGE_SHIFT, mask);
 			}
 			__free_iova(&deferred_flush[i].domain[j]->iovad, iova);
+			if (deferred_flush[i].freelist[j])
+				dma_free_pagelist(deferred_flush[i].freelist[j]);
 		}
 		deferred_flush[i].next = 0;
 	}
@@ -2926,7 +3044,7 @@ static void flush_unmaps_timeout(unsigned long data)
 	spin_unlock_irqrestore(&async_umap_flush_lock, flags);
 }
 
-static void add_unmap(struct dmar_domain *dom, struct iova *iova)
+static void add_unmap(struct dmar_domain *dom, struct iova *iova, struct page *freelist)
 {
 	unsigned long flags;
 	int next, iommu_id;
@@ -2942,6 +3060,7 @@ static void add_unmap(struct dmar_domain *dom, struct iova *iova)
 	next = deferred_flush[iommu_id].next;
 	deferred_flush[iommu_id].domain[next] = dom;
 	deferred_flush[iommu_id].iova[next] = iova;
+	deferred_flush[iommu_id].freelist[next] = freelist;
 	deferred_flush[iommu_id].next++;
 
 	if (!timer_on) {
@@ -2961,6 +3080,7 @@ static void intel_unmap_page(struct device *dev, dma_addr_t dev_addr,
 	unsigned long start_pfn, last_pfn;
 	struct iova *iova;
 	struct intel_iommu *iommu;
+	struct page *freelist;
 
 	if (iommu_no_mapping(dev))
 		return;
@@ -2981,19 +3101,16 @@ static void intel_unmap_page(struct device *dev, dma_addr_t dev_addr,
 	pr_debug("Device %s unmapping: pfn %lx-%lx\n",
 		 pci_name(pdev), start_pfn, last_pfn);
 
-	/*  clear the whole page */
-	dma_pte_clear_range(domain, start_pfn, last_pfn);
-
-	/* free page tables */
-	dma_pte_free_pagetable(domain, start_pfn, last_pfn);
+	freelist = domain_unmap(domain, start_pfn, last_pfn);
 
 	if (intel_iommu_strict) {
 		iommu_flush_iotlb_psi(iommu, domain->id, start_pfn,
-				      last_pfn - start_pfn + 1, 0);
+				      last_pfn - start_pfn + 1, !freelist, 0);
 		/* free iova */
 		__free_iova(&domain->iovad, iova);
+		dma_free_pagelist(freelist);
 	} else {
-		add_unmap(domain, iova);
+		add_unmap(domain, iova, freelist);
 		/*
 		 * queue up the release of the unmap to save the 1/6th of the
 		 * cpu used up by the iotlb flush operation...
@@ -3055,6 +3172,7 @@ static void intel_unmap_sg(struct device *hwdev, struct scatterlist *sglist,
 	unsigned long start_pfn, last_pfn;
 	struct iova *iova;
 	struct intel_iommu *iommu;
+	struct page *freelist;
 
 	if (iommu_no_mapping(hwdev))
 		return;
@@ -3072,19 +3190,16 @@ static void intel_unmap_sg(struct device *hwdev, struct scatterlist *sglist,
 	start_pfn = mm_to_dma_pfn(iova->pfn_lo);
 	last_pfn = mm_to_dma_pfn(iova->pfn_hi + 1) - 1;
 
-	/*  clear the whole page */
-	dma_pte_clear_range(domain, start_pfn, last_pfn);
-
-	/* free page tables */
-	dma_pte_free_pagetable(domain, start_pfn, last_pfn);
+	freelist = domain_unmap(domain, start_pfn, last_pfn);
 
 	if (intel_iommu_strict) {
 		iommu_flush_iotlb_psi(iommu, domain->id, start_pfn,
-				      last_pfn - start_pfn + 1, 0);
+				      last_pfn - start_pfn + 1, !freelist, 0);
 		/* free iova */
 		__free_iova(&domain->iovad, iova);
+		dma_free_pagelist(freelist);
 	} else {
-		add_unmap(domain, iova);
+		add_unmap(domain, iova, freelist);
 		/*
 		 * queue up the release of the unmap to save the 1/6th of the
 		 * cpu used up by the iotlb flush operation...
@@ -3167,7 +3282,7 @@ static int intel_map_sg(struct device *hwdev, struct scatterlist *sglist, int ne
 
 	/* it's a non-present to present mapping. Only flush if caching mode */
 	if (cap_caching_mode(iommu->cap))
-		iommu_flush_iotlb_psi(iommu, domain->id, start_vpfn, size, 1);
+		iommu_flush_iotlb_psi(iommu, domain->id, start_vpfn, size, 0, 1);
 	else
 		iommu_flush_write_buffer(iommu);
 
@@ -3714,6 +3829,7 @@ static int intel_iommu_memory_notifier(struct notifier_block *nb,
 			struct iova *iova;
 			struct dmar_drhd_unit *drhd;
 			struct intel_iommu *iommu;
+			struct page *freelist;
 
 			iova = find_iova(&si_domain->iovad, start_vpfn);
 			if (iova == NULL) {
@@ -3730,16 +3846,17 @@ static int intel_iommu_memory_notifier(struct notifier_block *nb,
 				return NOTIFY_BAD;
 			}
 
+			freelist = domain_unmap(si_domain, iova->pfn_lo,
+					       iova->pfn_hi);
+
 			rcu_read_lock();
 			for_each_active_iommu(iommu, drhd)
 				iommu_flush_iotlb_psi(iommu, si_domain->id,
 					iova->pfn_lo,
-					iova->pfn_hi - iova->pfn_lo + 1, 0);
+					iova->pfn_hi - iova->pfn_lo + 1,
+					!freelist, 0);
 			rcu_read_unlock();
-			dma_pte_clear_range(si_domain, iova->pfn_lo,
-					    iova->pfn_hi);
-			dma_pte_free_pagetable(si_domain, iova->pfn_lo,
-					       iova->pfn_hi);
+			dma_free_pagelist(freelist);
 
 			start_vpfn = iova->pfn_hi + 1;
 			free_iova_mem(iova);
@@ -4100,10 +4217,14 @@ static int intel_iommu_map(struct iommu_domain *domain,
 }
 
 static size_t intel_iommu_unmap(struct iommu_domain *domain,
-			     unsigned long iova, size_t size)
+				unsigned long iova, size_t size)
 {
 	struct dmar_domain *dmar_domain = domain->priv;
-	int level = 0;
+	struct page *freelist = NULL;
+	struct intel_iommu *iommu;
+	unsigned long start_pfn, last_pfn;
+	unsigned int npages;
+	int iommu_id, num, ndomains, level = 0;
 
 	/* Cope with horrid API which requires us to unmap more than the
 	   size argument if it happens to be a large-page mapping. */
@@ -4113,8 +4234,29 @@ static size_t intel_iommu_unmap(struct iommu_domain *domain,
 	if (size < VTD_PAGE_SIZE << level_to_offset_bits(level))
 		size = VTD_PAGE_SIZE << level_to_offset_bits(level);
 
-	dma_pte_clear_range(dmar_domain, iova >> VTD_PAGE_SHIFT,
-			    (iova + size - 1) >> VTD_PAGE_SHIFT);
+	start_pfn = iova >> VTD_PAGE_SHIFT;
+	last_pfn = (iova + size - 1) >> VTD_PAGE_SHIFT;
+
+	freelist = domain_unmap(dmar_domain, start_pfn, last_pfn);
+
+	npages = last_pfn - start_pfn + 1;
+
+	for_each_set_bit(iommu_id, dmar_domain->iommu_bmp, g_num_of_iommus) {
+               iommu = g_iommus[iommu_id];
+
+               /*
+                * find bit position of dmar_domain
+                */
+               ndomains = cap_ndoms(iommu->cap);
+               for_each_set_bit(num, iommu->domain_ids, ndomains) {
+                       if (iommu->domains[num] == dmar_domain)
+                               iommu_flush_iotlb_psi(iommu, num, start_pfn,
+						     npages, !freelist, 0);
+	       }
+
+	}
+
+	dma_free_pagelist(freelist);
 
 	if (dmar_domain->max_addr == iova + size)
 		dmar_domain->max_addr = iova;
Signature · 5 lines
1.8.5.3


-- 
dwmw2
#8 lore raw

PATCH 3/4 iommu/vt-d: Honour intel_iommu=sp_off for non-VMM domains

In reply to a message not in the archive
---
 drivers/iommu/intel-iommu.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
drivers/iommu/intel-iommu.c +5 -1
index f5934fc..c3d4bc9 100644
@@ -1624,7 +1624,11 @@ static int domain_init(struct dmar_domain *domain, int guest_width)
 	else
 		domain->iommu_snooping = 0;
 
-	domain->iommu_superpage = fls(cap_super_page_val(iommu->cap));
+	if (intel_iommu_superpage)
+		domain->iommu_superpage = fls(cap_super_page_val(iommu->cap));
+	else
+		domain->iommu_superpage = 0;
+
 	domain->nid = iommu->node;
 
 	/* always allocate the top pgd */
Signature · 5 lines
1.8.5.3


-- 
dwmw2
#9 lore raw

PATCH 4/4 iommu/vt-d: Be less pessimistic about domain coherency where possible

In reply to a message not in the archive

In commit 2e12bc29 ("intel-iommu: Default to non-coherent for domains unattached to iommus") we decided to err on the side of caution and always assume that it's possible that a device will be attached which is behind a non-coherent IOMMU.

In some cases, however, that just *cannot* happen. If there *are* no IOMMUs in the system which are non-coherent, then we don't need to do it. And flushing the dcache is a *significant* performance hit.

---
 drivers/iommu/intel-iommu.c | 21 +++++++++++++++++----
 1 file changed, 17 insertions(+), 4 deletions(-)
drivers/iommu/intel-iommu.c +17 -4
index c3d4bc9..1599cb1 100644
@@ -592,18 +592,31 @@ static struct intel_iommu *domain_get_iommu(struct dmar_domain *domain)
 
 static void domain_update_iommu_coherency(struct dmar_domain *domain)
 {
-	int i;
-
-	i = find_first_bit(domain->iommu_bmp, g_num_of_iommus);
+	struct dmar_drhd_unit *drhd;
+	struct intel_iommu *iommu;
+	int i, found = 0;
 
-	domain->iommu_coherency = i < g_num_of_iommus ? 1 : 0;
+	domain->iommu_coherency = 1;
 
 	for_each_set_bit(i, domain->iommu_bmp, g_num_of_iommus) {
+		found = 1;
 		if (!ecap_coherent(g_iommus[i]->ecap)) {
 			domain->iommu_coherency = 0;
 			break;
 		}
 	}
+	if (found)
+		return;
+
+	/* No hardware attached; use lowest common denominator */
+	rcu_read_lock();
+	for_each_active_iommu(iommu, drhd) {
+		if (!ecap_coherent(iommu->ecap)) {
+			domain->iommu_coherency = 0;
+			break;
+		}
+	}
+	rcu_read_unlock();
 }
 
 static void domain_update_iommu_snooping(struct dmar_domain *domain)
Signature · 5 lines
1.8.5.3


-- 
dwmw2