On 11/30/22 20:51, David Woodhouse wrote:
On Wed, 2022-11-30 at 17:03 +0100, Paolo Bonzini wrote:
On 11/27/22 13:22, David Woodhouse wrote:
Clean the update code up a little bit by unifying the fast and slow
paths as discussed, and make the update flag conditional to avoid
confusing older guests that don't ask for it.
On top of kvm/queue as of today at commit da5f28e10aa7d.
(This is identical to what I sent a couple of minutes ago, except that
this time I didn't forget to Cc the list)
Merged, thanks.
Thanks. I've rebased the remaining GPC fixes on top and pushed them out
(along with Metin's SCHEDOP_poll 32-bit compat support) to
https://git.infradead.org/users/dwmw2/linux.git/shortlog/refs/heads/gpc-fixes
Oh, so we do pull requests now too? I'm all for it, but please use
signed tags. :)
I still haven't reinstated the last of those patches to make gpc->len
immutable, although I think we concluded it's fine just to make the
runstate code claim gpc->len=1 and manage its own destiny, right?
Yeah, I'm not super keen on that either, but I guess it can work with
any of len == 1 or len == PAGE_SIZE - offset.
Related to this, for 6.3 I will send a cleanup of the API to put
together lock and check.
Paolo
In reviewing the merge/squash I spotted a minor cosmetic nit in my
'Allow XEN_RUNSTATE_UPDATE flag behaviour to be configured' commit.
It'd be slightly prettier like this, although the compiler really ought
to emit identical code.
diff --git a/arch/x86/kvm/xen.c b/arch/x86/kvm/xen.c
index 5208e05ca9a6..76b7fc6d543a 100644
--- a/arch/x86/kvm/xen.c
+++ b/arch/x86/kvm/xen.c
@@ -382,7 +382,7 @@ static void kvm_xen_update_runstate_guest(struct kvm_vcpu *v, bool atomic)
entry_time = vx->runstate_entry_time;
if (update_bit) {
entry_time |= XEN_RUNSTATE_UPDATE;
- *update_bit = (vx->runstate_entry_time | XEN_RUNSTATE_UPDATE) >> 56;
+ *update_bit = entry_time >> 56;
smp_wmb();
}
