[PATCH v2] memory: Optimize replay of guest mapping

[PATCH v2] memory: Optimize replay of guest mapping

[Zhenzhong Duan]

On x86, there are two notifiers registered due to vtd-ir memory region
splitting the whole address space. During replay of the address space
for each notifier, the whole address space is scanned which is
unnecessory.

We only need to scan the space belong to notifier montiored space.

Assert when notifier is used to monitor beyond iommu memory region's
address space.

Signed-off-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
---
v2: Add an assertion per Peter Xu
Tested only on x86 with a net card passed to guest(kvm/tcg), ping/ssh pass.

 hw/i386/intel_iommu.c | 2 +-
 softmmu/memory.c      | 4 ++--
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/hw/i386/intel_iommu.c b/hw/i386/intel_iommu.c
index 98a5c304a7d7..6b1de80e8573 100644
--- a/hw/i386/intel_iommu.c
+++ b/hw/i386/intel_iommu.c
@@ -3831,7 +3831,7 @@ static void vtd_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n)
                 .domain_id = vtd_get_domain_id(s, &ce, vtd_as->pasid),
             };
 
-            vtd_page_walk(s, &ce, 0, ~0ULL, &info, vtd_as->pasid);
+            vtd_page_walk(s, &ce, n->start, n->end, &info, vtd_as->pasid);
         }
     } else {
         trace_vtd_replay_ce_invalid(bus_n, PCI_SLOT(vtd_as->devfn),
diff --git a/softmmu/memory.c b/softmmu/memory.c
index 9d64efca269b..da7d84661972 100644
--- a/softmmu/memory.c
+++ b/softmmu/memory.c
@@ -1900,6 +1900,7 @@ int memory_region_register_iommu_notifier(MemoryRegion *mr,
     iommu_mr = IOMMU_MEMORY_REGION(mr);
     assert(n->notifier_flags != IOMMU_NOTIFIER_NONE);
     assert(n->start <= n->end);
+    assert(n->end <= memory_region_size(mr));
     assert(n->iommu_idx >= 0 &&
            n->iommu_idx < memory_region_iommu_num_indexes(iommu_mr));
 
@@ -1923,7 +1924,6 @@ uint64_t memory_region_iommu_get_min_page_size(IOMMUMemoryRegion *iommu_mr)
 
 void memory_region_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n)
 {
-    MemoryRegion *mr = MEMORY_REGION(iommu_mr);
     IOMMUMemoryRegionClass *imrc = IOMMU_MEMORY_REGION_GET_CLASS(iommu_mr);
     hwaddr addr, granularity;
     IOMMUTLBEntry iotlb;
@@ -1936,7 +1936,7 @@ void memory_region_iommu_replay(IOMMUMemoryRegion *iommu_mr, IOMMUNotifier *n)
 
     granularity = memory_region_iommu_get_min_page_size(iommu_mr);
 
-    for (addr = 0; addr < memory_region_size(mr); addr += granularity) {
+    for (addr = n->start; addr < n->end; addr += granularity) {
         iotlb = imrc->translate(iommu_mr, addr, IOMMU_NONE, n->iommu_idx);
         if (iotlb.perm != IOMMU_NONE) {
             n->notify(n, &iotlb);
-- 
2.25.1

Re: [PATCH v2] memory: Optimize replay of guest mapping

[Peter Xu]

On Wed, Feb 15, 2023 at 02:52:38PM +0800, Zhenzhong Duan wrote:
> On x86, there are two notifiers registered due to vtd-ir memory region
> splitting the whole address space. During replay of the address space
> for each notifier, the whole address space is scanned which is
> unnecessory.
> 
> We only need to scan the space belong to notifier montiored space.
> 
> Assert when notifier is used to monitor beyond iommu memory region's
> address space.
> 
> Signed-off-by: Zhenzhong Duan <zhenzhong.duan@intel.com>

Acked-by: Peter Xu <peterx@redhat.com>

-- 
Peter Xu

Re: [PATCH v2] memory: Optimize replay of guest mapping

[Jason Wang]

On Wed, Feb 15, 2023 at 10:46 PM Peter Xu <peterx@redhat.com> wrote:
>
> On Wed, Feb 15, 2023 at 02:52:38PM +0800, Zhenzhong Duan wrote:
> > On x86, there are two notifiers registered due to vtd-ir memory region
> > splitting the whole address space. During replay of the address space
> > for each notifier, the whole address space is scanned which is
> > unnecessory.
> >
> > We only need to scan the space belong to notifier montiored space.
> >
> > Assert when notifier is used to monitor beyond iommu memory region's
> > address space.
> >
> > Signed-off-by: Zhenzhong Duan <zhenzhong.duan@intel.com>
>
> Acked-by: Peter Xu <peterx@redhat.com>

Acked-by: Jason Wang <jasowang@redhat.com>

Thanks

>
> --
> Peter Xu
>