On 12/01/2017 11:49 PM, Michael S. Tsirkin wrote:
On Wed, Nov 29, 2017 at 09:55:26PM +0800, Wei Wang wrote:
The guest free pages should not be discarded by the live migration thread
when page poisoning is enabled with PAGE_POISONING_NO_SANITY=n, because
skipping the transfer of such poisoned free pages will trigger false
positive when new pages are allocated and checked on the destination.
This patch skips the reporting of free pages in the above case.
Reported-by: Michael S. Tsirkin <mst@xxxxxxxxxx>
Signed-off-by: Wei Wang <wei.w.wang@xxxxxxxxx>
Cc: Michal Hocko <mhocko@xxxxxxxx>
---
drivers/virtio/virtio_balloon.c | 4 +++-
1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/drivers/virtio/virtio_balloon.c b/drivers/virtio/virtio_balloon.c
index 035bd3a..6ac4cff 100644
--- a/drivers/virtio/virtio_balloon.c
+++ b/drivers/virtio/virtio_balloon.c
@@ -652,7 +652,9 @@ static void report_free_page(struct work_struct *work)
/* Start by sending the obtained cmd id to the host with an outbuf */
send_one_desc(vb, vb->free_page_vq, virt_to_phys(&vb->start_cmd_id),
sizeof(uint32_t), false, true, false);
- walk_free_mem_block(vb, 0, &virtio_balloon_send_free_pages);
+ if (!(page_poisoning_enabled() &&
+ !IS_ENABLED(CONFIG_PAGE_POISONING_NO_SANITY)))
+ walk_free_mem_block(vb, 0, &virtio_balloon_send_free_pages);
/*
* End by sending the stop id to the host with an outbuf. Use the
* non-batching mode here to trigger a kick after adding the stop id.
PAGE_POISONING_ZERO is actually OK.
I think the 0-filled pages still need to be sent. If the host on the
destination doesn't use PAGE_POISONING_ZERO, then the pages it offers to
the guest on the destination may have non-0 values.
But I really would prefer it that we still send pages to host,
otherwise debugging becomes much harder.
And it does not have to be completely useless, even though
you can not discard them as they would be zero-filled then.
How about a config field telling host what should be there in the free
pages? This way even though host can not discard them, host can send
them out without reading them, still a win.
OK, but I think we would need two 32-bit config registers:
__u32 page_poison_val; // stores the PAGE_POISON VALUE, but it couldn't
indicate if page poisoning is in use
__u32 special_features; // set bit 0 to indicate page poisoning is in use
#define VIRTIO_BALLOON_SF_PAGE_POISON (1 << 0)
Best,
Wei
_______________________________________________
Virtualization mailing list
Virtualization@xxxxxxxxxxxxxxxxxxxxxxxxxx
https://lists.linuxfoundation.org/mailman/listinfo/virtualization
