| Message ID | 20170628024358.29956-1-haozhong.zhang@intel.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
Haozhong Zhang <haozhong.zhang@intel.com> wrote: > In cpu_physical_memory_sync_dirty_bitmap(rb, start, ...), the 2nd > argument 'start' is relative to the start of the ramblock 'rb'. When > it's used to access the dirty memory bitmap of ram_list (i.e. > ram_list.dirty_memory[DIRTY_MEMORY_MIGRATION]->blocks[]), an offset to > the start of all RAM (i.e. rb->offset) should be added to it, which has > however been missed since c/s 6b6712efcc. For a ramblock of host memory > backend whose offset is not zero, cpu_physical_memory_sync_dirty_bitmap() > synchronizes the incorrect part of the dirty memory bitmap of ram_list > to the per ramblock dirty bitmap. As a result, a guest with host > memory backend may crash after migration. > > Fix it by adding the offset of ramblock when accessing the dirty memory > bitmap of ram_list in cpu_physical_memory_sync_dirty_bitmap(). > > Reported-by: Stefan Hajnoczi <stefanha@redhat.com> > Signed-off-by: Haozhong Zhang <haozhong.zhang@intel.com> Reviewed-by: Juan Quintela <quintela@redhat.com> As this function is only used on migration, should I integrate it on my next push, or do you want to pull it, Paolo? Later, Juan. > --- > include/exec/ram_addr.h | 8 +++++--- > 1 file changed, 5 insertions(+), 3 deletions(-) > > diff --git a/include/exec/ram_addr.h b/include/exec/ram_addr.h > index 73d1bea8b6..cbc797ed05 100644 > --- a/include/exec/ram_addr.h > +++ b/include/exec/ram_addr.h > @@ -377,6 +377,7 @@ uint64_t cpu_physical_memory_sync_dirty_bitmap(RAMBlock *rb, > uint64_t *real_dirty_pages) > { > ram_addr_t addr; > + ram_addr_t offset = rb->offset; > unsigned long page = BIT_WORD(start >> TARGET_PAGE_BITS); > uint64_t num_dirty = 0; > unsigned long *dest = rb->bmap; > @@ -386,8 +387,9 @@ uint64_t cpu_physical_memory_sync_dirty_bitmap(RAMBlock *rb, > int k; > int nr = BITS_TO_LONGS(length >> TARGET_PAGE_BITS); > unsigned long * const *src; > - unsigned long idx = (page * BITS_PER_LONG) / DIRTY_MEMORY_BLOCK_SIZE; > - unsigned long offset = BIT_WORD((page * BITS_PER_LONG) % > + unsigned long word = BIT_WORD((start + offset) >> TARGET_PAGE_BITS); > + unsigned long idx = (word * BITS_PER_LONG) / DIRTY_MEMORY_BLOCK_SIZE; > + unsigned long offset = BIT_WORD((word * BITS_PER_LONG) % > DIRTY_MEMORY_BLOCK_SIZE); > > rcu_read_lock(); > @@ -416,7 +418,7 @@ uint64_t cpu_physical_memory_sync_dirty_bitmap(RAMBlock *rb, > } else { > for (addr = 0; addr < length; addr += TARGET_PAGE_SIZE) { > if (cpu_physical_memory_test_and_clear_dirty( > - start + addr, > + start + addr + offset, > TARGET_PAGE_SIZE, > DIRTY_MEMORY_MIGRATION)) { > *real_dirty_pages += 1;
On 28/06/2017 09:30, Juan Quintela wrote: > Haozhong Zhang <haozhong.zhang@intel.com> wrote: >> In cpu_physical_memory_sync_dirty_bitmap(rb, start, ...), the 2nd >> argument 'start' is relative to the start of the ramblock 'rb'. When >> it's used to access the dirty memory bitmap of ram_list (i.e. >> ram_list.dirty_memory[DIRTY_MEMORY_MIGRATION]->blocks[]), an offset to >> the start of all RAM (i.e. rb->offset) should be added to it, which has >> however been missed since c/s 6b6712efcc. For a ramblock of host memory >> backend whose offset is not zero, cpu_physical_memory_sync_dirty_bitmap() >> synchronizes the incorrect part of the dirty memory bitmap of ram_list >> to the per ramblock dirty bitmap. As a result, a guest with host >> memory backend may crash after migration. >> >> Fix it by adding the offset of ramblock when accessing the dirty memory >> bitmap of ram_list in cpu_physical_memory_sync_dirty_bitmap(). >> >> Reported-by: Stefan Hajnoczi <stefanha@redhat.com> >> Signed-off-by: Haozhong Zhang <haozhong.zhang@intel.com> > > Reviewed-by: Juan Quintela <quintela@redhat.com> > > As this function is only used on migration, should I integrate it on my > next push, or do you want to pull it, Paolo? > > Later, Juan. > > >> --- >> include/exec/ram_addr.h | 8 +++++--- >> 1 file changed, 5 insertions(+), 3 deletions(-) >> >> diff --git a/include/exec/ram_addr.h b/include/exec/ram_addr.h >> index 73d1bea8b6..cbc797ed05 100644 >> --- a/include/exec/ram_addr.h >> +++ b/include/exec/ram_addr.h >> @@ -377,6 +377,7 @@ uint64_t cpu_physical_memory_sync_dirty_bitmap(RAMBlock *rb, >> uint64_t *real_dirty_pages) >> { >> ram_addr_t addr; >> + ram_addr_t offset = rb->offset; >> unsigned long page = BIT_WORD(start >> TARGET_PAGE_BITS); >> uint64_t num_dirty = 0; >> unsigned long *dest = rb->bmap; >> @@ -386,8 +387,9 @@ uint64_t cpu_physical_memory_sync_dirty_bitmap(RAMBlock *rb, >> int k; >> int nr = BITS_TO_LONGS(length >> TARGET_PAGE_BITS); >> unsigned long * const *src; >> - unsigned long idx = (page * BITS_PER_LONG) / DIRTY_MEMORY_BLOCK_SIZE; >> - unsigned long offset = BIT_WORD((page * BITS_PER_LONG) % >> + unsigned long word = BIT_WORD((start + offset) >> TARGET_PAGE_BITS); >> + unsigned long idx = (word * BITS_PER_LONG) / DIRTY_MEMORY_BLOCK_SIZE; >> + unsigned long offset = BIT_WORD((word * BITS_PER_LONG) % >> DIRTY_MEMORY_BLOCK_SIZE); The shadowing between the two variables named offset is a bit ugly. Please use rb->offset in the initialization of "word", and declare... >> >> rcu_read_lock(); >> @@ -416,7 +418,7 @@ uint64_t cpu_physical_memory_sync_dirty_bitmap(RAMBlock *rb, >> } else { ... "ram_addr_t offset = offset" here. Paolo >> for (addr = 0; addr < length; addr += TARGET_PAGE_SIZE) { >> if (cpu_physical_memory_test_and_clear_dirty( >> - start + addr, >> + start + addr + offset, >> TARGET_PAGE_SIZE, >> DIRTY_MEMORY_MIGRATION)) { >> *real_dirty_pages += 1;
diff --git a/include/exec/ram_addr.h b/include/exec/ram_addr.h index 73d1bea8b6..cbc797ed05 100644 --- a/include/exec/ram_addr.h +++ b/include/exec/ram_addr.h @@ -377,6 +377,7 @@ uint64_t cpu_physical_memory_sync_dirty_bitmap(RAMBlock *rb, uint64_t *real_dirty_pages) { ram_addr_t addr; + ram_addr_t offset = rb->offset; unsigned long page = BIT_WORD(start >> TARGET_PAGE_BITS); uint64_t num_dirty = 0; unsigned long *dest = rb->bmap; @@ -386,8 +387,9 @@ uint64_t cpu_physical_memory_sync_dirty_bitmap(RAMBlock *rb, int k; int nr = BITS_TO_LONGS(length >> TARGET_PAGE_BITS); unsigned long * const *src; - unsigned long idx = (page * BITS_PER_LONG) / DIRTY_MEMORY_BLOCK_SIZE; - unsigned long offset = BIT_WORD((page * BITS_PER_LONG) % + unsigned long word = BIT_WORD((start + offset) >> TARGET_PAGE_BITS); + unsigned long idx = (word * BITS_PER_LONG) / DIRTY_MEMORY_BLOCK_SIZE; + unsigned long offset = BIT_WORD((word * BITS_PER_LONG) % DIRTY_MEMORY_BLOCK_SIZE); rcu_read_lock(); @@ -416,7 +418,7 @@ uint64_t cpu_physical_memory_sync_dirty_bitmap(RAMBlock *rb, } else { for (addr = 0; addr < length; addr += TARGET_PAGE_SIZE) { if (cpu_physical_memory_test_and_clear_dirty( - start + addr, + start + addr + offset, TARGET_PAGE_SIZE, DIRTY_MEMORY_MIGRATION)) { *real_dirty_pages += 1;
In cpu_physical_memory_sync_dirty_bitmap(rb, start, ...), the 2nd argument 'start' is relative to the start of the ramblock 'rb'. When it's used to access the dirty memory bitmap of ram_list (i.e. ram_list.dirty_memory[DIRTY_MEMORY_MIGRATION]->blocks[]), an offset to the start of all RAM (i.e. rb->offset) should be added to it, which has however been missed since c/s 6b6712efcc. For a ramblock of host memory backend whose offset is not zero, cpu_physical_memory_sync_dirty_bitmap() synchronizes the incorrect part of the dirty memory bitmap of ram_list to the per ramblock dirty bitmap. As a result, a guest with host memory backend may crash after migration. Fix it by adding the offset of ramblock when accessing the dirty memory bitmap of ram_list in cpu_physical_memory_sync_dirty_bitmap(). Reported-by: Stefan Hajnoczi <stefanha@redhat.com> Signed-off-by: Haozhong Zhang <haozhong.zhang@intel.com> --- include/exec/ram_addr.h | 8 +++++--- 1 file changed, 5 insertions(+), 3 deletions(-)