On 10/05/2023 09.32, Claudio Imbrenda wrote:
On Wed, 10 May 2023 08:47:08 +0200
Thomas Huth <th...@redhat.com> wrote:
On 09/05/2023 18.27, Claudio Imbrenda wrote:
When rebooting a small VM using asynchronous teardown, a spurious
warning is emitted when the KVM_PV_ASYNC_CLEANUP_PREPARE ioctl fails.
Why does the _PREPARE fail in that case? Why 4GiB and not more or less? This
because of kernel commit 292a7d6fca33df70ca4b8e9b0d0e74adf87582dc, which
fixes problems in case the VM is small (<2GiB)
sounds racy... what if you have a faster or slower machine?
why racy?
2 or 4GiB is still very fast, and at some point you have to draw a line.
I could make it 2GiB, which is the limit at which _PREPARE will fail,
but since I'm touching this code, I would like to avoid unnecessary
overhead, instead of "just fixing"
I can put the limit to 2GiB if you think it's more clean
Ok, so this is not due to some race (which I first suspected), but due to
some change in recent kernels. Please put a link to the related kernel
commit in the patch description, and yes, I'd prefer to use the same
boundary as the kernel here (i.e. 2GiB instead of 4GiB), just that it is
clear that we're not using an arbitrary magic value here.
Thanks,
Thomas
