virtio-mem: Fix the bitmap index of the section offset
PATCH 1/1 virtio-mem: Fix the bitmap index of the section offset
vmem->bitmap indexes the memory region of the virtio-mem backend at a granularity of block_size. To calculate the index of target section offset, the block_size should be divided instead of the bitmap_size.
--- hw/virtio/virtio-mem.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
hw/virtio/virtio-mem.c +2 -2
@@ -235,7 +235,7 @@ static int virtio_mem_for_each_plugged_section(const VirtIOMEM *vmem,uint64_t offset, size;int ret =0 ; - first_bit = s->offset_within_region / vmem->bitmap_size; + first_bit = s->offset_within_region / vmem->block_size; first_bit =find_next_bit (vmem->bitmap, vmem->bitmap_size, first_bit);while (first_bit < vmem->bitmap_size) {MemoryRegionSection tmp = *s;
@@ -267,7 +267,7 @@ static int virtio_mem_for_each_unplugged_section(const VirtIOMEM *vmem,uint64_t offset, size;int ret =0 ; - first_bit = s->offset_within_region / vmem->bitmap_size; + first_bit = s->offset_within_region / vmem->block_size; first_bit =find_next_zero_bit (vmem->bitmap, vmem->bitmap_size, first_bit);while (first_bit < vmem->bitmap_size) {MemoryRegionSection tmp = *s;
Signature · 1 line
2.17.1
Quoted 3 lines · Chenyi Qiang
vmem->bitmapindexes the memory region of the virtio-mem backend at a granularity ofblock_size. To calculate the index of target section offset, theblock_sizeshould be divided instead of thebitmap_size.
I'm curious, what's the user-visible effect and how did you identify this issue?
IIUC, we could end up our search for a plugged/unplugged block "too late", such that we miss to process blocks.
That would be the case if the bitmap_size < block_size, which should effectively always happen ...
unplug_all and migration would be affected, which is why a simple test case without a guest reboot/migration wouldn't run into it.
Quoted 21 lines
Fixes 2044969f0b ("virtio-mem: Implement RamDiscardManager interface")Signed-off-by Chenyi Qiang <[email protected]>--- hw/virtio/virtio-mem.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)hw/virtio/virtio-mem.c +1
-1@@ -235,7 +235,7 @@ static int virtio_mem_for_each_plugged_section(const VirtIOMEM *vmem,uint64_t offset, size;int ret =0 ; - first_bit = s->offset_within_region / vmem->bitmap_size; + first_bit = s->offset_within_region / vmem->block_size; first_bit =find_next_bit (vmem->bitmap, vmem->bitmap_size, first_bit);while (first_bit < vmem->bitmap_size) {MemoryRegionSection tmp = *s;
hw/virtio/virtio-mem.c static int virtio_mem_for_each_unplugged_section(const VirtIOMEM *vmem, · line 273
Quoted 8 lines
hw/virtio/virtio-mem.c +1
-1@@ -267,7 +267,7 @@ static int virtio_mem_for_each_unplugged_section(const VirtIOMEM *vmem,uint64_t offset, size;int ret =0 ; - first_bit = s->offset_within_region / vmem->bitmap_size; + first_bit = s->offset_within_region / vmem->block_size; first_bit =find_next_zero_bit (vmem->bitmap, vmem->bitmap_size, first_bit);while (first_bit < vmem->bitmap_size) {MemoryRegionSection tmp = *s;
Looks correct to me
Thanks!
Signature · 3 lines
Thanks, David / dhildenb
Quoted 11 lines · David Hildenbrand
Quoted 3 lines · Chenyi Qiang
vmem->bitmapindexes the memory region of the virtio-mem backend at a granularity ofblock_size. To calculate the index of target section offset, theblock_sizeshould be divided instead of thebitmap_size.I'm curious, what's the user-visible effect and how did you identify this issue?
IIUC, we could end up our search for a plugged/unplugged block "too late", such that we miss to process blocks.
Quoted 6 lines
That would be the case if the
bitmap_size<block_size, which should effectively always happen ...
unplug_alland migration would be affected, which is why a simple test case without a guest reboot/migration wouldn't run into it.
I just realized that unplug_all is fine because only vfio implements the ram_discard_listener so far and always sets double_discard_supported=true. So migration should be the issue (and IIRC migration with VFIO is still shaky).
Signature · 3 lines
Thanks, David / dhildenb
Quoted 17 lines · David Hildenbrand
Quoted 16 lines · David Hildenbrand
Quoted 4 lines · Chenyi Qiang
vmem->bitmapindexes the memory region of the virtio-mem backend at a granularity ofblock_size. To calculate the index of target section offset,
theblock_sizeshould be divided instead of thebitmap_size.I'm curious, what's the user-visible effect and how did you identify this issue?
IIUC, we could end up our search for a plugged/unplugged block "too late", such that we miss to process blocks.
That would be the case if the
bitmap_size<block_size, which should effectively always happen ...
Quoted 7 lines
Quoted 2 lines
unplug_alland migration would be affected, which is why a simple test case without a guest reboot/migration wouldn't run into it.I just realized that
unplug_allis fine because only vfio implements theram_discard_listenerso far and always sets double_discard_supported=true. So migration should be the issue (and IIRC migration with VFIO is still shaky).
Yes, actually, no obvious visible effect on my side. I was just learning the RamDiscardManager interface and found this issue. :)
Quoted 1 line
Quoted 23 lines · Chenyi Qiang
Quoted 20 lines · David Hildenbrand
Quoted 18 lines · David Hildenbrand
Quoted 4 lines · Chenyi Qiang
vmem->bitmapindexes the memory region of the virtio-mem backend at a granularity ofblock_size. To calculate the index of target section offset,
theblock_sizeshould be divided instead of thebitmap_size.I'm curious, what's the user-visible effect and how did you identify this issue?
IIUC, we could end up our search for a plugged/unplugged block "too late", such that we miss to process blocks.
That would be the case if the
bitmap_size<block_size, which should effectively always happen ...
unplug_alland migration would be affected, which is why a simple test case without a guest reboot/migration wouldn't run into it.
Quoted 7 lines
Quoted 4 lines
I just realized that
unplug_allis fine because only vfio implements theram_discard_listenerso far and always sets double_discard_supported=true. So migration should be the issue (and IIRC migration with VFIO is still shaky).Yes, actually, no obvious visible effect on my side. I was just learning the RamDiscardManager interface and found this issue. :)
Good, thanks.
Queuing this to
https://github.com/davidhildenbrand/qemu.git mem-next
Signature · 3 lines
Thanks, David / dhildenb
Quoted 6 lines · Chenyi Qiang
vmem->bitmapindexes the memory region of the virtio-mem backend at a granularity ofblock_size. To calculate the index of target section offset, theblock_sizeshould be divided instead of thebitmap_size.Fixes 2044969f0b ("virtio-mem: Implement RamDiscardManager interface")Signed-off-by Chenyi Qiang <[email protected]>
I see David's queueing this.
Quoted 28 lines
--- hw/virtio/virtio-mem.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)hw/virtio/virtio-mem.c +2
-2@@ -235,7 +235,7 @@ static int virtio_mem_for_each_plugged_section(const VirtIOMEM *vmem,uint64_t offset, size;int ret =0 ; - first_bit = s->offset_within_region / vmem->bitmap_size; + first_bit = s->offset_within_region / vmem->block_size; first_bit =find_next_bit (vmem->bitmap, vmem->bitmap_size, first_bit);while (first_bit < vmem->bitmap_size) {MemoryRegionSection tmp = *s;@@ -267,7 +267,7 @@ static int virtio_mem_for_each_unplugged_section(const VirtIOMEM *vmem,uint64_t offset, size;int ret =0 ; - first_bit = s->offset_within_region / vmem->bitmap_size; + first_bit = s->offset_within_region / vmem->block_size; first_bit =find_next_zero_bit (vmem->bitmap, vmem->bitmap_size, first_bit);while (first_bit < vmem->bitmap_size) {MemoryRegionSection tmp = *s;Signature · 1 line
2.17.1