[PATCH v3 21/23] AMD/IOMMU: replace all-contiguous page tables by superpage mappings

[Jan Beulich <jbeulich@xxxxxxxx>]

When a page table ends up with all contiguous entries (including all
identical attributes), it can be replaced by a superpage entry at the
next higher level. The page table itself can then be scheduled for
freeing.

Signed-off-by: Jan Beulich <jbeulich@xxxxxxxx>
---
Unlike the freeing of all-empty page tables, this causes quite a bit of
back and forth for PV domains, due to their mapping/unmapping of pages
when they get converted to/from being page tables. It may therefore be
worth considering to delay re-coalescing a little, to avoid doing so
when the superpage would otherwise get split again pretty soon. But I
think this would better be the subject of a separate change anyway.

Of course this could also be helped by more "aware" kernel side
behavior: They could avoid immediately mapping freed page tables
writable again, in anticipation of re-using that same page for another
page table elsewhere.
---
v3: New.

--- a/xen/drivers/passthrough/amd/iommu_map.c
+++ b/xen/drivers/passthrough/amd/iommu_map.c
@@ -81,7 +81,8 @@ static union amd_iommu_pte set_iommu_pte
                                                  unsigned long dfn,
                                                  unsigned long next_mfn,
                                                  unsigned int level,
-                                                 bool iw, bool ir)
+                                                 bool iw, bool ir,
+                                                 bool *contig)
 {
     union amd_iommu_pte *table, *pde, old;
 
@@ -94,11 +95,15 @@ static union amd_iommu_pte set_iommu_pte
          old.iw != iw || old.ir != ir )
     {
         set_iommu_pde_present(pde, next_mfn, 0, iw, ir);
-        pt_update_contig_markers(&table->raw, pfn_to_pde_idx(dfn, level),
-                                 level, PTE_kind_leaf);
+        *contig = pt_update_contig_markers(&table->raw,
+                                           pfn_to_pde_idx(dfn, level),
+                                           level, PTE_kind_leaf);
     }
     else
+    {
         old.pr = false; /* signal "no change" to the caller */
+        *contig = false;
+    }
 
     unmap_domain_page(table);
 
@@ -346,6 +351,7 @@ int amd_iommu_map_page(struct domain *d,
 {
     struct domain_iommu *hd = dom_iommu(d);
     unsigned int level = (IOMMUF_order(flags) / PTE_PER_TABLE_SHIFT) + 1;
+    bool contig;
     int rc;
     unsigned long pt_mfn = 0;
     union amd_iommu_pte old;
@@ -386,8 +392,26 @@ int amd_iommu_map_page(struct domain *d,
 
     /* Install mapping */
     old = set_iommu_pte_present(pt_mfn, dfn_x(dfn), mfn_x(mfn), level,
-                                (flags & IOMMUF_writable),
-                                (flags & IOMMUF_readable));
+                                flags & IOMMUF_writable,
+                                flags & IOMMUF_readable, &contig);
+
+    while ( unlikely(contig) && ++level < hd->arch.amd.paging_mode )
+    {
+        struct page_info *pg = mfn_to_page(_mfn(pt_mfn));
+        unsigned long next_mfn;
+
+        if ( iommu_pde_from_dfn(d, dfn_x(dfn), level, &pt_mfn, flush_flags,
+                                false) )
+            BUG();
+        BUG_ON(!pt_mfn);
+
+        next_mfn = mfn_x(mfn) & (~0UL << (PTE_PER_TABLE_SHIFT * (level - 1)));
+        set_iommu_pte_present(pt_mfn, dfn_x(dfn), next_mfn, level,
+                              flags & IOMMUF_writable,
+                              flags & IOMMUF_readable, &contig);
+        *flush_flags |= IOMMU_FLUSHF_modified | IOMMU_FLUSHF_all;
+        iommu_queue_free_pgtable(d, pg);
+    }
 
     spin_unlock(&hd->arch.mapping_lock);