On 2017/7/29 4:34, Nitesh Narayan Lal wrote:
Hi,
I am working on the guest page hinting support with Rik.
As per my implementation when pages are freed, they are added to a
cpu-local list. Once that list fills up, the pages need to be passed to
the hypervisor. I am currently using a seqlock to ensure page allocation
does not allocate pages while they are being freed by the hypervisor.
This list is passed to the hypervisor using virtio_balloon
infrastructure. As per the current flow virtqueue_add() requires a
memory allocation, which leads to a deadlock condition.
What should be the right way to get around this?
Looking forward for the suggestions.
Hi Nitesh,
I am interesting on the idea. Could you give more details of the
background? What page hinting can help and what's the expected usage model?
--
Yang
Alibaba Cloud Computing
