| Message ID | 20230211014626.3659152-5-vipinsh@google.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | Optimize clear dirty log | expand |
On Fri, Feb 10, 2023 at 05:46:23PM -0800, Vipin Sharma wrote: > No need to check all of the conditions in __handle_changed_spte(). Aging > a gfn range implies resetting access bit or marking spte for access > tracking. nit: State what the patch does first. > > Use atomic operation to only reset those bits. This avoids checking many > conditions in __handle_changed_spte() API. Also, clean up code by > removing dead code and API parameters. > > Signed-off-by: Vipin Sharma <vipinsh@google.com> nits aside, Reviewed-by: David Matlack <dmatlack@google.com> > --- > arch/x86/kvm/mmu/tdp_mmu.c | 36 +++++++++++++++++++----------------- > 1 file changed, 19 insertions(+), 17 deletions(-) > > diff --git a/arch/x86/kvm/mmu/tdp_mmu.c b/arch/x86/kvm/mmu/tdp_mmu.c > index c895560244de..5d6e77554797 100644 > --- a/arch/x86/kvm/mmu/tdp_mmu.c > +++ b/arch/x86/kvm/mmu/tdp_mmu.c > @@ -758,13 +758,6 @@ static inline void tdp_mmu_set_spte(struct kvm *kvm, struct tdp_iter *iter, > _tdp_mmu_set_spte(kvm, iter, new_spte, true); > } > > -static inline void tdp_mmu_set_spte_no_acc_track(struct kvm *kvm, > - struct tdp_iter *iter, > - u64 new_spte) > -{ > - _tdp_mmu_set_spte(kvm, iter, new_spte, false); > -} > - > #define tdp_root_for_each_pte(_iter, _root, _start, _end) \ > for_each_tdp_pte(_iter, _root, _start, _end) > > @@ -1251,32 +1244,41 @@ static __always_inline bool kvm_tdp_mmu_handle_gfn(struct kvm *kvm, > /* > * Mark the SPTEs range of GFNs [start, end) unaccessed and return non-zero > * if any of the GFNs in the range have been accessed. > + * > + * No need to mark corresponding PFN as accessed as this call is coming from > + * the clear_young() or clear_flush_young() notifier, which uses the return > + * value to determine if the page has been accessed. > */ > static bool age_gfn_range(struct kvm *kvm, struct tdp_iter *iter, > struct kvm_gfn_range *range) > { > - u64 new_spte = 0; > + u64 new_spte; > > /* If we have a non-accessed entry we don't need to change the pte. */ > if (!is_accessed_spte(iter->old_spte)) > return false; > > - new_spte = iter->old_spte; > - > - if (spte_ad_enabled(new_spte)) { > - new_spte &= ~shadow_accessed_mask; > + if (spte_ad_enabled(iter->old_spte)) { > + iter->old_spte = tdp_mmu_clear_spte_bits(iter->sptep, nit: Extra space before =
On Fri, Feb 10, 2023, Vipin Sharma wrote: > } else { > + new_spte = mark_spte_for_access_track(iter->old_spte); > + iter->old_spte = kvm_tdp_mmu_write_spte(iter->sptep, > + iter->old_spte, new_spte, > + iter->level); > /* > * Capture the dirty status of the page, so that it doesn't get > * lost when the SPTE is marked for access tracking. > */ > - if (is_writable_pte(new_spte)) > - kvm_set_pfn_dirty(spte_to_pfn(new_spte)); > - > - new_spte = mark_spte_for_access_track(new_spte); > + if (is_writable_pte(iter->old_spte)) > + kvm_set_pfn_dirty(spte_to_pfn(iter->old_spte)); Moving this block below kvm_tdp_mmu_write_spte() is an unrelated change. Much to my chagrin, I discovered that past me gave you this code. I still think the change is correct, but I dropped it for now, mostly because the legacy/shadow MMU has the same pattern (marks the PFN dirty before setting the SPTE). I think this might actually be a bug fix, e.g. if the XCHG races with a fast page fault fix and drops the Writable bit, the CPU could insert writable entry into the TLB without KVM invoking kvm_set_pfn_dirty(). But I'm not 100% confident that I'm not missing something, and _if_ there's a bug then mmu_spte_age() needs the same fix, so for now, I dropped it.
diff --git a/arch/x86/kvm/mmu/tdp_mmu.c b/arch/x86/kvm/mmu/tdp_mmu.c index c895560244de..5d6e77554797 100644 --- a/arch/x86/kvm/mmu/tdp_mmu.c +++ b/arch/x86/kvm/mmu/tdp_mmu.c @@ -758,13 +758,6 @@ static inline void tdp_mmu_set_spte(struct kvm *kvm, struct tdp_iter *iter, _tdp_mmu_set_spte(kvm, iter, new_spte, true); } -static inline void tdp_mmu_set_spte_no_acc_track(struct kvm *kvm, - struct tdp_iter *iter, - u64 new_spte) -{ - _tdp_mmu_set_spte(kvm, iter, new_spte, false); -} - #define tdp_root_for_each_pte(_iter, _root, _start, _end) \ for_each_tdp_pte(_iter, _root, _start, _end) @@ -1251,32 +1244,41 @@ static __always_inline bool kvm_tdp_mmu_handle_gfn(struct kvm *kvm, /* * Mark the SPTEs range of GFNs [start, end) unaccessed and return non-zero * if any of the GFNs in the range have been accessed. + * + * No need to mark corresponding PFN as accessed as this call is coming from + * the clear_young() or clear_flush_young() notifier, which uses the return + * value to determine if the page has been accessed. */ static bool age_gfn_range(struct kvm *kvm, struct tdp_iter *iter, struct kvm_gfn_range *range) { - u64 new_spte = 0; + u64 new_spte; /* If we have a non-accessed entry we don't need to change the pte. */ if (!is_accessed_spte(iter->old_spte)) return false; - new_spte = iter->old_spte; - - if (spte_ad_enabled(new_spte)) { - new_spte &= ~shadow_accessed_mask; + if (spte_ad_enabled(iter->old_spte)) { + iter->old_spte = tdp_mmu_clear_spte_bits(iter->sptep, + iter->old_spte, + shadow_accessed_mask, + iter->level); + new_spte = iter->old_spte & ~shadow_accessed_mask; } else { + new_spte = mark_spte_for_access_track(iter->old_spte); + iter->old_spte = kvm_tdp_mmu_write_spte(iter->sptep, + iter->old_spte, new_spte, + iter->level); /* * Capture the dirty status of the page, so that it doesn't get * lost when the SPTE is marked for access tracking. */ - if (is_writable_pte(new_spte)) - kvm_set_pfn_dirty(spte_to_pfn(new_spte)); - - new_spte = mark_spte_for_access_track(new_spte); + if (is_writable_pte(iter->old_spte)) + kvm_set_pfn_dirty(spte_to_pfn(iter->old_spte)); } - tdp_mmu_set_spte_no_acc_track(kvm, iter, new_spte); + trace_kvm_tdp_mmu_spte_changed(iter->as_id, iter->gfn, iter->level, + iter->old_spte, new_spte); return true; }
No need to check all of the conditions in __handle_changed_spte(). Aging a gfn range implies resetting access bit or marking spte for access tracking. Use atomic operation to only reset those bits. This avoids checking many conditions in __handle_changed_spte() API. Also, clean up code by removing dead code and API parameters. Signed-off-by: Vipin Sharma <vipinsh@google.com> --- arch/x86/kvm/mmu/tdp_mmu.c | 36 +++++++++++++++++++----------------- 1 file changed, 19 insertions(+), 17 deletions(-)