| Message ID | 20200622032037.31112-1-zhukeqian1@huawei.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [v3] migration: Count new_dirty instead of real_dirty | expand |
* Keqian Zhu (zhukeqian1@huawei.com) wrote: > real_dirty_pages becomes equal to total ram size after dirty log sync > in ram_init_bitmaps, the reason is that the bitmap of ramblock is > initialized to be all set, so old path counts them as "real dirty" at > beginning. > > This causes wrong dirty rate and false positive throttling. > > Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com> OK, Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com> and queued. you might still want to look at migration_trigger_thrtottle and see if you can stop the throttling if in the RAM bulk stage. > --- > Changelog: > > v3: > - Address Dave's comments. > > v2: > - Use new_dirty_pages instead of accu_dirty_pages. > - Adjust commit messages. > --- > include/exec/ram_addr.h | 5 +---- > migration/ram.c | 8 +++++--- > 2 files changed, 6 insertions(+), 7 deletions(-) > > diff --git a/include/exec/ram_addr.h b/include/exec/ram_addr.h > index 7b5c24e928..3ef729a23c 100644 > --- a/include/exec/ram_addr.h > +++ b/include/exec/ram_addr.h > @@ -442,8 +442,7 @@ static inline void cpu_physical_memory_clear_dirty_range(ram_addr_t start, > static inline > uint64_t cpu_physical_memory_sync_dirty_bitmap(RAMBlock *rb, > ram_addr_t start, > - ram_addr_t length, > - uint64_t *real_dirty_pages) > + ram_addr_t length) > { > ram_addr_t addr; > unsigned long word = BIT_WORD((start + rb->offset) >> TARGET_PAGE_BITS); > @@ -469,7 +468,6 @@ uint64_t cpu_physical_memory_sync_dirty_bitmap(RAMBlock *rb, > if (src[idx][offset]) { > unsigned long bits = atomic_xchg(&src[idx][offset], 0); > unsigned long new_dirty; > - *real_dirty_pages += ctpopl(bits); > new_dirty = ~dest[k]; > dest[k] |= bits; > new_dirty &= bits; > @@ -502,7 +500,6 @@ uint64_t cpu_physical_memory_sync_dirty_bitmap(RAMBlock *rb, > start + addr + offset, > TARGET_PAGE_SIZE, > DIRTY_MEMORY_MIGRATION)) { > - *real_dirty_pages += 1; > long k = (start + addr) >> TARGET_PAGE_BITS; > if (!test_and_set_bit(k, dest)) { > num_dirty++; > diff --git a/migration/ram.c b/migration/ram.c > index 069b6e30bc..5554a7d2d8 100644 > --- a/migration/ram.c > +++ b/migration/ram.c > @@ -859,9 +859,11 @@ static inline bool migration_bitmap_clear_dirty(RAMState *rs, > /* Called with RCU critical section */ > static void ramblock_sync_dirty_bitmap(RAMState *rs, RAMBlock *rb) > { > - rs->migration_dirty_pages += > - cpu_physical_memory_sync_dirty_bitmap(rb, 0, rb->used_length, > - &rs->num_dirty_pages_period); > + uint64_t new_dirty_pages = > + cpu_physical_memory_sync_dirty_bitmap(rb, 0, rb->used_length); > + > + rs->migration_dirty_pages += new_dirty_pages; > + rs->num_dirty_pages_period += new_dirty_pages; > } > > /** > -- > 2.19.1 > > -- Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK
Hi Dave, On 2020/7/3 22:20, Dr. David Alan Gilbert wrote: > * Keqian Zhu (zhukeqian1@huawei.com) wrote: >> real_dirty_pages becomes equal to total ram size after dirty log sync >> in ram_init_bitmaps, the reason is that the bitmap of ramblock is >> initialized to be all set, so old path counts them as "real dirty" at >> beginning. >> >> This causes wrong dirty rate and false positive throttling. >> >> Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com> > > OK, > > Reviewed-by: Dr. David Alan Gilbert <dgilbert@redhat.com> > > and queued. > > you might still want to look at migration_trigger_thrtottle and see if > you can stop the throttling if in the RAM bulk stage. Yes, I tested it and it worked well. Thanks, Keqian
diff --git a/include/exec/ram_addr.h b/include/exec/ram_addr.h index 7b5c24e928..3ef729a23c 100644 --- a/include/exec/ram_addr.h +++ b/include/exec/ram_addr.h @@ -442,8 +442,7 @@ static inline void cpu_physical_memory_clear_dirty_range(ram_addr_t start, static inline uint64_t cpu_physical_memory_sync_dirty_bitmap(RAMBlock *rb, ram_addr_t start, - ram_addr_t length, - uint64_t *real_dirty_pages) + ram_addr_t length) { ram_addr_t addr; unsigned long word = BIT_WORD((start + rb->offset) >> TARGET_PAGE_BITS); @@ -469,7 +468,6 @@ uint64_t cpu_physical_memory_sync_dirty_bitmap(RAMBlock *rb, if (src[idx][offset]) { unsigned long bits = atomic_xchg(&src[idx][offset], 0); unsigned long new_dirty; - *real_dirty_pages += ctpopl(bits); new_dirty = ~dest[k]; dest[k] |= bits; new_dirty &= bits; @@ -502,7 +500,6 @@ uint64_t cpu_physical_memory_sync_dirty_bitmap(RAMBlock *rb, start + addr + offset, TARGET_PAGE_SIZE, DIRTY_MEMORY_MIGRATION)) { - *real_dirty_pages += 1; long k = (start + addr) >> TARGET_PAGE_BITS; if (!test_and_set_bit(k, dest)) { num_dirty++; diff --git a/migration/ram.c b/migration/ram.c index 069b6e30bc..5554a7d2d8 100644 --- a/migration/ram.c +++ b/migration/ram.c @@ -859,9 +859,11 @@ static inline bool migration_bitmap_clear_dirty(RAMState *rs, /* Called with RCU critical section */ static void ramblock_sync_dirty_bitmap(RAMState *rs, RAMBlock *rb) { - rs->migration_dirty_pages += - cpu_physical_memory_sync_dirty_bitmap(rb, 0, rb->used_length, - &rs->num_dirty_pages_period); + uint64_t new_dirty_pages = + cpu_physical_memory_sync_dirty_bitmap(rb, 0, rb->used_length); + + rs->migration_dirty_pages += new_dirty_pages; + rs->num_dirty_pages_period += new_dirty_pages; } /**
real_dirty_pages becomes equal to total ram size after dirty log sync in ram_init_bitmaps, the reason is that the bitmap of ramblock is initialized to be all set, so old path counts them as "real dirty" at beginning. This causes wrong dirty rate and false positive throttling. Signed-off-by: Keqian Zhu <zhukeqian1@huawei.com> --- Changelog: v3: - Address Dave's comments. v2: - Use new_dirty_pages instead of accu_dirty_pages. - Adjust commit messages. --- include/exec/ram_addr.h | 5 +---- migration/ram.c | 8 +++++--- 2 files changed, 6 insertions(+), 7 deletions(-)