Re: [PATCH] KVM: stats: Add VM dirty_pages stats for the number of dirty pages

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Wed, Aug 11, 2021 at 4:19 PM Peter Feiner <pfeiner@xxxxxxxxxx> wrote:
>
> On Wed, Aug 11, 2021 at 3:53 PM Jing Zhang <jingzhangos@xxxxxxxxxx> wrote:
> >
> > Hi Peter,
> >
> > On Tue, Aug 10, 2021 at 3:45 PM Peter Feiner <pfeiner@xxxxxxxxxx> wrote:
> > >
> > > On Tue, Aug 10, 2021 at 3:32 PM Jing Zhang <jingzhangos@xxxxxxxxxx> wrote:
> > > >
> > > > Add a generic VM stats dirty_pages to record the number of dirty pages
> > > > reflected in dirty_bitmap at the moment.
> > > >
> > > > Original-by: Peter Feiner <pfeiner@xxxxxxxxxx>
> > > > Signed-off-by: Jing Zhang <jingzhangos@xxxxxxxxxx>
> > > > ---
> > > >  arch/powerpc/kvm/book3s_64_mmu_hv.c    |  8 ++++++--
> > > >  arch/powerpc/kvm/book3s_64_mmu_radix.c |  1 +
> > > >  arch/powerpc/kvm/book3s_hv_rm_mmu.c    |  1 +
> > > >  include/linux/kvm_host.h               |  3 ++-
> > > >  include/linux/kvm_types.h              |  1 +
> > > >  virt/kvm/kvm_main.c                    | 26 +++++++++++++++++++++++---
> > > >  6 files changed, 34 insertions(+), 6 deletions(-)
> > > >
> > > > diff --git a/arch/powerpc/kvm/book3s_64_mmu_hv.c b/arch/powerpc/kvm/book3s_64_mmu_hv.c
> > > > index c63e263312a4..e4aafa10efa1 100644
> > > > --- a/arch/powerpc/kvm/book3s_64_mmu_hv.c
> > > > +++ b/arch/powerpc/kvm/book3s_64_mmu_hv.c
> > > > @@ -1122,8 +1122,10 @@ long kvmppc_hv_get_dirty_log_hpt(struct kvm *kvm,
> > > >                  * since we always put huge-page HPTEs in the rmap chain
> > > >                  * corresponding to their page base address.
> > > >                  */
> > > > -               if (npages)
> > > > +               if (npages) {
> > > >                         set_dirty_bits(map, i, npages);
> > > > +                       kvm->stat.generic.dirty_pages += npages;
> > > > +               }
> > > >                 ++rmapp;
> > > >         }
> > > >         preempt_enable();
> > > > @@ -1178,8 +1180,10 @@ void kvmppc_unpin_guest_page(struct kvm *kvm, void *va, unsigned long gpa,
> > > >         gfn = gpa >> PAGE_SHIFT;
> > > >         srcu_idx = srcu_read_lock(&kvm->srcu);
> > > >         memslot = gfn_to_memslot(kvm, gfn);
> > > > -       if (memslot && memslot->dirty_bitmap)
> > > > +       if (memslot && memslot->dirty_bitmap) {
> > > >                 set_bit_le(gfn - memslot->base_gfn, memslot->dirty_bitmap);
> > > > +               ++kvm->stat.generic.dirty_pages;
> > > > +       }
> > > >         srcu_read_unlock(&kvm->srcu, srcu_idx);
> > > >  }
> > > >
> > > > diff --git a/arch/powerpc/kvm/book3s_64_mmu_radix.c b/arch/powerpc/kvm/book3s_64_mmu_radix.c
> > > > index b5905ae4377c..3a6cb3854a44 100644
> > > > --- a/arch/powerpc/kvm/book3s_64_mmu_radix.c
> > > > +++ b/arch/powerpc/kvm/book3s_64_mmu_radix.c
> > > > @@ -1150,6 +1150,7 @@ long kvmppc_hv_get_dirty_log_radix(struct kvm *kvm,
> > > >                 j = i + 1;
> > > >                 if (npages) {
> > > >                         set_dirty_bits(map, i, npages);
> > > > +                       kvm->stat.generic.dirty_pages += npages;
> > > >                         j = i + npages;
> > > >                 }
> > > >         }
> > > > diff --git a/arch/powerpc/kvm/book3s_hv_rm_mmu.c b/arch/powerpc/kvm/book3s_hv_rm_mmu.c
> > > > index 632b2545072b..16806bc473fa 100644
> > > > --- a/arch/powerpc/kvm/book3s_hv_rm_mmu.c
> > > > +++ b/arch/powerpc/kvm/book3s_hv_rm_mmu.c
> > > > @@ -109,6 +109,7 @@ void kvmppc_update_dirty_map(const struct kvm_memory_slot *memslot,
> > > >         npages = (psize + PAGE_SIZE - 1) / PAGE_SIZE;
> > > >         gfn -= memslot->base_gfn;
> > > >         set_dirty_bits_atomic(memslot->dirty_bitmap, gfn, npages);
> > > > +       kvm->stat.generic.dirty_pages += npages;
> > > >  }
> > > >  EXPORT_SYMBOL_GPL(kvmppc_update_dirty_map);
> > > >
> > > > diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
> > > > index f50bfcf225f0..1e8e66fb915b 100644
> > > > --- a/include/linux/kvm_host.h
> > > > +++ b/include/linux/kvm_host.h
> > > > @@ -1421,7 +1421,8 @@ struct _kvm_stats_desc {
> > > >                 KVM_STATS_BASE_POW10, -9)
> > > >
> > > >  #define KVM_GENERIC_VM_STATS()                                                \
> > > > -       STATS_DESC_COUNTER(VM_GENERIC, remote_tlb_flush)
> > > > +       STATS_DESC_COUNTER(VM_GENERIC, remote_tlb_flush),                      \
> > > > +       STATS_DESC_COUNTER(VM_GENERIC, dirty_pages)
> > > >
> > > >  #define KVM_GENERIC_VCPU_STATS()                                              \
> > > >         STATS_DESC_COUNTER(VCPU_GENERIC, halt_successful_poll),                \
> > > > diff --git a/include/linux/kvm_types.h b/include/linux/kvm_types.h
> > > > index ed6a985c5680..6c05df00aebf 100644
> > > > --- a/include/linux/kvm_types.h
> > > > +++ b/include/linux/kvm_types.h
> > > > @@ -78,6 +78,7 @@ struct kvm_mmu_memory_cache {
> > > >
> > > >  struct kvm_vm_stat_generic {
> > > >         u64 remote_tlb_flush;
> > > > +       u64 dirty_pages;
> > > >  };
> > > >
> > > >  struct kvm_vcpu_stat_generic {
> > > > diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
> > > > index a438a7a3774a..93f0ca2ea326 100644
> > > > --- a/virt/kvm/kvm_main.c
> > > > +++ b/virt/kvm/kvm_main.c
> > > > @@ -1228,6 +1228,19 @@ static int kvm_alloc_dirty_bitmap(struct kvm_memory_slot *memslot)
> > > >         return 0;
> > > >  }
> > > >
> > > > +static inline unsigned long hweight_dirty_bitmap(
> > > > +                                               struct kvm_memory_slot *memslot)
> > > > +{
> > > > +       unsigned long i;
> > > > +       unsigned long count = 0;
> > > > +       unsigned long n = kvm_dirty_bitmap_bytes(memslot);
> > > > +
> > > > +       for (i = 0; i < n / sizeof(long); ++i)
> > > > +               count += hweight_long(memslot->dirty_bitmap[i]);
> > > > +
> > > > +       return count;
> > > > +}
> > > > +
> > > >  /*
> > > >   * Delete a memslot by decrementing the number of used slots and shifting all
> > > >   * other entries in the array forward one spot.
> > > > @@ -1612,6 +1625,7 @@ static int kvm_delete_memslot(struct kvm *kvm,
> > > >         if (r)
> > > >                 return r;
> > > >
> > > > +       kvm->stat.generic.dirty_pages -= hweight_dirty_bitmap(old);
> > > >         kvm_free_memslot(kvm, old);
> > > >         return 0;
> > > >  }
> > > > @@ -1733,8 +1747,10 @@ int __kvm_set_memory_region(struct kvm *kvm,
> > > >         if (r)
> > > >                 goto out_bitmap;
> > > >
> > > > -       if (old.dirty_bitmap && !new.dirty_bitmap)
> > > > +       if (old.dirty_bitmap && !new.dirty_bitmap) {
> > > > +               kvm->stat.generic.dirty_pages -= hweight_dirty_bitmap(&old);
> > > >                 kvm_destroy_dirty_bitmap(&old);
> > > > +       }
> > > >         return 0;
> > > >
> > > >  out_bitmap:
> > > > @@ -1895,6 +1911,7 @@ static int kvm_get_dirty_log_protect(struct kvm *kvm, struct kvm_dirty_log *log)
> > > >                         offset = i * BITS_PER_LONG;
> > > >                         kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
> > > >                                                                 offset, mask);
> > > > +                       kvm->stat.generic.dirty_pages -= hweight_long(mask);
> > > >                 }
> > > >                 KVM_MMU_UNLOCK(kvm);
> > > >         }
> > > > @@ -2012,6 +2029,7 @@ static int kvm_clear_dirty_log_protect(struct kvm *kvm,
> > > >                         flush = true;
> > > >                         kvm_arch_mmu_enable_log_dirty_pt_masked(kvm, memslot,
> > > >                                                                 offset, mask);
> > > > +                       kvm->stat.generic.dirty_pages -= hweight_long(mask);
> > > >                 }
> > > >         }
> > > >         KVM_MMU_UNLOCK(kvm);
> > > > @@ -3062,11 +3080,13 @@ void mark_page_dirty_in_slot(struct kvm *kvm,
> > > >                 unsigned long rel_gfn = gfn - memslot->base_gfn;
> > > >                 u32 slot = (memslot->as_id << 16) | memslot->id;
> > > >
> > > > -               if (kvm->dirty_ring_size)
> > > > +               if (kvm->dirty_ring_size) {
> > > >                         kvm_dirty_ring_push(kvm_dirty_ring_get(kvm),
> > > >                                             slot, rel_gfn);
> > > > -               else
> > > > +               } else {
> > > >                         set_bit_le(rel_gfn, memslot->dirty_bitmap);
> > > > +                       ++kvm->stat.generic.dirty_pages;
> > >
> > > Couple of problems here:
> > >
> > > - Calls to mark_page_dirty_in_slot aren't serialized by the mmu_lock,
> > > so these updates will race.
> > > - The page might already be dirty in the bitmap, so you're potentially
> > > double counting here
> > >
> > > You can fix both of these problems by changing set_bit_le to a
> > > test_and_set_bit_le (might not be the function name -- but you get the
> > > idea) and conditionally incrementing dirty_pages.
> > >
> > >
> > Since we need to consider the dirty ring situation and it is not
> > trivial to check if the page is already in the dirty ring, I'll not
> > change set_bit_le. Anyway, we are going to define the stat
> > "dirty_pages" as dirtied pages in the life cycle of a VM, we only care
> > about its growth rate for your use case.
>
> To be clear, I don't have a use case anymore. It was years ago that I
> came up with the idea of doing a kind of light memory dirty tracking
> to predict how expensive pre-copy migration was going to be. However,
> I didn't follow through on the idea; so if this never saw the light of
> day in an upstream kernel, I'd be no worse for wear.
>
> With that said, the use case I imagined did require knowing how many
> pages were in the dirty bitmap at any moment. The number of pages
> dirtied across the lifetime of the VM wouldn't have served that
> purpose. So unless you have a use case for this new stat that's been
> thought through end-to-end, I would suggest just dropping this patch
> here (and from Google's kernel fork as well).
Thanks for the clarification, Peter.
I will send out another patch anyway according to Oliver's idea. We'll
see how we can proceed from there.
>
>
>
> > I'll send out another version of the patch and we'll see if it is good enough.
> > > > +               }
> > > >         }
> > > >  }
> > > >  EXPORT_SYMBOL_GPL(mark_page_dirty_in_slot);
> > > >
> > > > base-commit: d0732b0f8884d9cc0eca0082bbaef043f3fef7fb
> > > > --
> > > > 2.32.0.605.g8dce9f2422-goog
> > > >
> >
> > Thanks,
> > Jing



[Index of Archives]     [KVM ARM]     [KVM ia64]     [KVM ppc]     [Virtualization Tools]     [Spice Development]     [Libvirt]     [Libvirt Users]     [Linux USB Devel]     [Linux Audio Users]     [Yosemite Questions]     [Linux Kernel]     [Linux SCSI]     [XFree86]

  Powered by Linux