iommu/vt-d: Don't free parent pagetable of the PTE we're adding

When adding a large scatterlist entry that covers more than the L3
superpage size (1GB) but has an alignment such that we must use L2
superpages (2MB) , we give dma_pte_free_level() a range that causes it
to free the L3 pagetable we're about to populate. We fix this by telling
dma_pte_free_pagetable() about the pagetable level we're about to populate
to prevent freeing it.

For example, mapping a scatterlist with entry lengths 854MB and 1194MB
at IOVA 0xffff80000000 would, when processing the 2MB-aligned second
entry, cause pfn_to_dma_pte() to create a L3 directory to hold L2
superpages for the mapping at IOVA 0xffffc0000000. We would previously
call dma_pte_free_pagetable(domain, 0xffffc0000, 0xfffffffff), which
would free the L3 directory pfn_to_dma_pte() just created for IO PFN
0xffffc0000. Telling dma_pte_free_pagetable() to retain the L3
directories while using L2 superpages avoids the erroneous free.

Signed-off-by: David Dillow <dillow@google.com>
Signed-off-by: Joerg Roedel <jroedel@suse.de>

diff --git a/drivers/iommu/intel-iommu.c b/drivers/iommu/intel-iommu.c
index 1a79a4ec6f0900ed8dbbb8f072ee5b997072d579..bffc880f3fef68653a37d6ac84950237cb612a74 100644 (file)
--- a/drivers/iommu/intel-iommu.c
+++ b/drivers/iommu/intel-iommu.c
@@ -1137,8 +1137,9 @@ static void dma_pte_clear_range(struct dmar_domain *domain,
 }
 
 static void dma_pte_free_level(struct dmar_domain *domain, int level,
-                              struct dma_pte *pte, unsigned long pfn,
-                              unsigned long start_pfn, unsigned long last_pfn)
+                              int retain_level, struct dma_pte *pte,
+                              unsigned long pfn, unsigned long start_pfn,
+                              unsigned long last_pfn)
 {
        pfn = max(start_pfn, pfn);
        pte = &pte[pfn_level_offset(pfn, level)];
@@ -1153,12 +1154,17 @@ static void dma_pte_free_level(struct dmar_domain *domain, int level,
                level_pfn = pfn & level_mask(level);
                level_pte = phys_to_virt(dma_pte_addr(pte));
 
-               if (level > 2)
-                       dma_pte_free_level(domain, level - 1, level_pte,
-                                          level_pfn, start_pfn, last_pfn);
+               if (level > 2) {
+                       dma_pte_free_level(domain, level - 1, retain_level,
+                                          level_pte, level_pfn, start_pfn,
+                                          last_pfn);
+               }
 
-               /* If range covers entire pagetable, free it */
-               if (!(start_pfn > level_pfn ||
+               /*
+                * Free the page table if we're below the level we want to
+                * retain and the range covers the entire table.
+                */
+               if (level < retain_level && !(start_pfn > level_pfn ||
                      last_pfn < level_pfn + level_size(level) - 1)) {
                        dma_clear_pte(pte);
                        domain_flush_cache(domain, pte, sizeof(*pte));
@@ -1169,10 +1175,14 @@ next:
        } while (!first_pte_in_page(++pte) && pfn <= last_pfn);
 }
 
-/* clear last level (leaf) ptes and free page table pages. */
+/*
+ * clear last level (leaf) ptes and free page table pages below the
+ * level we wish to keep intact.
+ */
 static void dma_pte_free_pagetable(struct dmar_domain *domain,
                                   unsigned long start_pfn,
-                                  unsigned long last_pfn)
+                                  unsigned long last_pfn,
+                                  int retain_level)
 {
        BUG_ON(!domain_pfn_supported(domain, start_pfn));
        BUG_ON(!domain_pfn_supported(domain, last_pfn));
@@ -1181,7 +1191,7 @@ static void dma_pte_free_pagetable(struct dmar_domain *domain,
        dma_pte_clear_range(domain, start_pfn, last_pfn);
 
        /* We don't need lock here; nobody else touches the iova range */
-       dma_pte_free_level(domain, agaw_to_level(domain->agaw),
+       dma_pte_free_level(domain, agaw_to_level(domain->agaw), retain_level,
                           domain->pgd, 0, start_pfn, last_pfn);
 
        /* free pgd */
@@ -2274,8 +2284,11 @@ static int __domain_mapping(struct dmar_domain *domain, unsigned long iov_pfn,
                                /*
                                 * Ensure that old small page tables are
                                 * removed to make room for superpage(s).
+                                * We're adding new large pages, so make sure
+                                * we don't remove their parent tables.
                                 */
-                               dma_pte_free_pagetable(domain, iov_pfn, end_pfn);
+                               dma_pte_free_pagetable(domain, iov_pfn, end_pfn,
+                                                      largepage_lvl + 1);
                        } else {
                                pteval &= ~(uint64_t)DMA_PTE_LARGE_PAGE;
                        }
@@ -3935,7 +3948,8 @@ static int intel_map_sg(struct device *dev, struct scatterlist *sglist, int nele
        ret = domain_sg_mapping(domain, start_vpfn, sglist, size, prot);
        if (unlikely(ret)) {
                dma_pte_free_pagetable(domain, start_vpfn,
-                                      start_vpfn + size - 1);
+                                      start_vpfn + size - 1,
+                                      agaw_to_level(domain->agaw) + 1);
                free_iova_fast(&domain->iovad, iova_pfn, dma_to_mm_pfn(size));
                return 0;
        }