2023-01-09 22:01:10

by Raghavendra Rao Ananta

Subject: [RFC PATCH 4/6] KVM: arm64: Optimize TLBIs in the dirty logging path

Currently the dirty-logging paths, including
kvm_arch_flush_remote_tlbs_memslot() and kvm_mmu_wp_memory_region()
ivalidates the entire VM's TLB entries using kvm_flush_remote_tlbs().
As the range of IPAs is provided by these functions, this is highly
inefficient on the systems which support FEAT_TLBIRANGE. Hence,
use kvm_flush_remote_tlbs_range() to flush the TLBs instead.

Signed-off-by: Raghavendra Rao Ananta <[email protected]>
---
arch/arm64/kvm/arm.c | 7 ++++++-
arch/arm64/kvm/mmu.c | 2 +-
2 files changed, 7 insertions(+), 2 deletions(-)

diff --git a/arch/arm64/kvm/arm.c b/arch/arm64/kvm/arm.c
index 00da570ed72bd..179520888c697 100644
--- a/arch/arm64/kvm/arm.c
+++ b/arch/arm64/kvm/arm.c
@@ -1433,7 +1433,12 @@ void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
const struct kvm_memory_slot *memslot)
{
- kvm_flush_remote_tlbs(kvm);
+ phys_addr_t start, end;
+
+ start = memslot->base_gfn << PAGE_SHIFT;
+ end = (memslot->base_gfn + memslot->npages) << PAGE_SHIFT;
+
+ kvm_flush_remote_tlbs_range(kvm, start, end);
}

static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
diff --git a/arch/arm64/kvm/mmu.c b/arch/arm64/kvm/mmu.c
index 70f76bc909c5d..e34b81f5922ce 100644
--- a/arch/arm64/kvm/mmu.c
+++ b/arch/arm64/kvm/mmu.c
@@ -976,7 +976,7 @@ static void kvm_mmu_wp_memory_region(struct kvm *kvm, int slot)
write_lock(&kvm->mmu_lock);
stage2_wp_range(&kvm->arch.mmu, start, end);
write_unlock(&kvm->mmu_lock);
- kvm_flush_remote_tlbs(kvm);
+ kvm_flush_remote_tlbs_range(kvm, start, end);
}

/**
--
2.39.0.314.g84b9a713c41-goog

2023-01-24 22:54:22

by Oliver Upton

Subject: Re: [RFC PATCH 4/6] KVM: arm64: Optimize TLBIs in the dirty logging path

Hi Raghavendra,

I find the commit title rather ambiguous. May I suggest:

KVM: arm64: Use range-based TLBIs for write protection

On Mon, Jan 09, 2023 at 09:53:45PM +0000, Raghavendra Rao Ananta wrote:
> Currently the dirty-logging paths, including
> kvm_arch_flush_remote_tlbs_memslot() and kvm_mmu_wp_memory_region()
> ivalidates the entire VM's TLB entries using kvm_flush_remote_tlbs().
> As the range of IPAs is provided by these functions, this is highly
> inefficient on the systems which support FEAT_TLBIRANGE. Hence,
> use kvm_flush_remote_tlbs_range() to flush the TLBs instead.

This commit message gives a rather mechanical description of the commit.
Instead of describing the change, could you describe _why_ this is an
improvement over the VM-wide invalidation?

--
Thanks,
Oliver

> Signed-off-by: Raghavendra Rao Ananta <[email protected]>
> ---
> arch/arm64/kvm/arm.c | 7 ++++++-
> arch/arm64/kvm/mmu.c | 2 +-
> 2 files changed, 7 insertions(+), 2 deletions(-)
>
> diff --git a/arch/arm64/kvm/arm.c b/arch/arm64/kvm/arm.c
> index 00da570ed72bd..179520888c697 100644
> --- a/arch/arm64/kvm/arm.c
> +++ b/arch/arm64/kvm/arm.c
> @@ -1433,7 +1433,12 @@ void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
> void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
> const struct kvm_memory_slot *memslot)
> {
> - kvm_flush_remote_tlbs(kvm);
> + phys_addr_t start, end;
> +
> + start = memslot->base_gfn << PAGE_SHIFT;
> + end = (memslot->base_gfn + memslot->npages) << PAGE_SHIFT;
> +
> + kvm_flush_remote_tlbs_range(kvm, start, end);
> }
>
> static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
> diff --git a/arch/arm64/kvm/mmu.c b/arch/arm64/kvm/mmu.c
> index 70f76bc909c5d..e34b81f5922ce 100644
> --- a/arch/arm64/kvm/mmu.c
> +++ b/arch/arm64/kvm/mmu.c
> @@ -976,7 +976,7 @@ static void kvm_mmu_wp_memory_region(struct kvm *kvm, int slot)
> write_lock(&kvm->mmu_lock);
> stage2_wp_range(&kvm->arch.mmu, start, end);
> write_unlock(&kvm->mmu_lock);
> - kvm_flush_remote_tlbs(kvm);
> + kvm_flush_remote_tlbs_range(kvm, start, end);
> }
>
> /**
> --
> 2.39.0.314.g84b9a713c41-goog
>
>

2023-01-25 21:53:14

by Raghavendra Rao Ananta

Subject: Re: [RFC PATCH 4/6] KVM: arm64: Optimize TLBIs in the dirty logging path

Hi Oliver,

On Tue, Jan 24, 2023 at 2:54 PM Oliver Upton <[email protected]> wrote:
>
> Hi Raghavendra,
>
> I find the commit title rather ambiguous. May I suggest:
>
> KVM: arm64: Use range-based TLBIs for write protection
>
> On Mon, Jan 09, 2023 at 09:53:45PM +0000, Raghavendra Rao Ananta wrote:
> > Currently the dirty-logging paths, including
> > kvm_arch_flush_remote_tlbs_memslot() and kvm_mmu_wp_memory_region()
> > ivalidates the entire VM's TLB entries using kvm_flush_remote_tlbs().
> > As the range of IPAs is provided by these functions, this is highly
> > inefficient on the systems which support FEAT_TLBIRANGE. Hence,
> > use kvm_flush_remote_tlbs_range() to flush the TLBs instead.
>
> This commit message gives a rather mechanical description of the commit.
> Instead of describing the change, could you describe _why_ this is an
> improvement over the VM-wide invalidation?
>
Of course. I assumed the optimization would be obvious, but sure,
it'll be better to describe it.
FYI, thanks to David's common code for range-based TLBI, this patch
shrunk to just one line, and would impact only the flush after
write-protect.

Thanks,
Raghavendra
> --
> Thanks,
> Oliver
>
> > Signed-off-by: Raghavendra Rao Ananta <[email protected]>
> > ---
> > arch/arm64/kvm/arm.c | 7 ++++++-
> > arch/arm64/kvm/mmu.c | 2 +-
> > 2 files changed, 7 insertions(+), 2 deletions(-)
> >
> > diff --git a/arch/arm64/kvm/arm.c b/arch/arm64/kvm/arm.c
> > index 00da570ed72bd..179520888c697 100644
> > --- a/arch/arm64/kvm/arm.c
> > +++ b/arch/arm64/kvm/arm.c
> > @@ -1433,7 +1433,12 @@ void kvm_arch_sync_dirty_log(struct kvm *kvm, struct kvm_memory_slot *memslot)
> > void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
> > const struct kvm_memory_slot *memslot)
> > {
> > - kvm_flush_remote_tlbs(kvm);
> > + phys_addr_t start, end;
> > +
> > + start = memslot->base_gfn << PAGE_SHIFT;
> > + end = (memslot->base_gfn + memslot->npages) << PAGE_SHIFT;
> > +
> > + kvm_flush_remote_tlbs_range(kvm, start, end);
> > }
> >
> > static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
> > diff --git a/arch/arm64/kvm/mmu.c b/arch/arm64/kvm/mmu.c
> > index 70f76bc909c5d..e34b81f5922ce 100644
> > --- a/arch/arm64/kvm/mmu.c
> > +++ b/arch/arm64/kvm/mmu.c
> > @@ -976,7 +976,7 @@ static void kvm_mmu_wp_memory_region(struct kvm *kvm, int slot)
> > write_lock(&kvm->mmu_lock);
> > stage2_wp_range(&kvm->arch.mmu, start, end);
> > write_unlock(&kvm->mmu_lock);
> > - kvm_flush_remote_tlbs(kvm);
> > + kvm_flush_remote_tlbs_range(kvm, start, end);
> > }
> >
> > /**
> > --
> > 2.39.0.314.g84b9a713c41-goog
> >
> >