When restoring a pending interrupt, we are setting the Q bit to force
a retrigger in xive_finish_unmask(). But we also need to force an EOI
in this case to reach the same initial state : P=1, Q=0.
This can be done by not setting 'old_p' for pending interrupts which
will inform xive_finish_unmask() that an EOI needs to be sent.
Suggested-by: Benjamin Herrenschmidt <b...@kernel.crashing.org>
Signed-off-by: Cédric Le Goater <c...@kaod.org>
---
Tested with a guest running iozone.
arch/powerpc/kvm/book3s_xive.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/arch/powerpc/kvm/book3s_xive.c b/arch/powerpc/kvm/book3s_xive.c
index bf457843e032..b5e6d227a034 100644
--- a/arch/powerpc/kvm/book3s_xive.c
+++ b/arch/powerpc/kvm/book3s_xive.c
@@ -1558,7 +1558,7 @@ static int xive_set_source(struct kvmppc_xive *xive, long
irq, u64 addr)
/*
* Restore P and Q. If the interrupt was pending, we
- * force both P and Q, which will trigger a resend.
+ * force Q and !P, which will trigger a resend.
*
* That means that a guest that had both an interrupt
* pending (queued) and Q set will restore with only
@@ -1566,7 +1566,7 @@ static int xive_set_source(struct kvmppc_xive *xive, long
irq, u64 addr)
* is perfectly fine as coalescing interrupts that haven't
* been presented yet is always allowed.
*/
- if (val & KVM_XICS_PRESENTED || val & KVM_XICS_PENDING)
+ if (val & KVM_XICS_PRESENTED && !(val & KVM_XICS_PENDING))
state->old_p = true;
if (val & KVM_XICS_QUEUED || val & KVM_XICS_PENDING)
state->old_q = true;
--
2.13.6
