When MSIx interrupts are enabled, each CQ is associated with
a vector which is expected to be called upon ready data
in the CQ buffer. After guest reads from the buffer, it updates
the processed buffer position using a CQ dedicated DB.
Upon reading only partial data from the CQ buffer, device
fails to re-send a new MSI message stating buffer still has
unread data. The fix ensures a new message will be sent.
Change-Id: I10190be127b8dcbd89732cfb95ea37faf8c5779b
Signed-off-by: Shimi Gersner <gers...@gmail.com>
---
hw/block/nvme.c | 4 ++++
1 file changed, 4 insertions(+)
diff --git a/hw/block/nvme.c b/hw/block/nvme.c
index f639d7ae73..e2dca6e57f 100644
--- a/hw/block/nvme.c
+++ b/hw/block/nvme.c
@@ -1152,8 +1152,12 @@ static void nvme_process_db(NvmeCtrl *n, hwaddr addr,
int val)
}
}
+ // When we have more messages, we should make sure irq is asserted.
When MSIx is used
+ // this will make sure another notification is sent to the guest.
if (cq->tail == cq->head) {
nvme_irq_deassert(n, cq);
+ } else {
+ nvme_irq_assert(n, cq);
}
} else {
/* Submission queue doorbell write */
--
2.17.1
