On 2019/1/30 下午1:48, Yongji Xie wrote:
On Wed, 30 Jan 2019 at 10:32, Jason Wang <jasow...@redhat.com> wrote:
On 2019/1/22 下午4:31, elohi...@gmail.com wrote:
+static int
+vu_queue_inflight_get(VuDev *dev, VuVirtq *vq, int desc_idx)
+{
+ if (!has_feature(dev->protocol_features,
+ VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)) {
+ return 0;
+ }
+
+ if (unlikely(!vq->inflight)) {
+ return -1;
+ }
+
+ vq->inflight->desc[desc_idx].inuse = 1;
+
+ vq->inflight->desc[desc_idx].avail_idx = vq->last_avail_idx;
+
+ return 0;
+}
+
+static int
+vu_queue_inflight_pre_put(VuDev *dev, VuVirtq *vq, int desc_idx)
+{
+ if (!has_feature(dev->protocol_features,
+ VHOST_USER_PROTOCOL_F_INFLIGHT_SHMFD)) {
+ return 0;
+ }
+
+ if (unlikely(!vq->inflight)) {
+ return -1;
+ }
+
+ vq->inflight->desc[desc_idx].used_idx = vq->used_idx;
+
+ barrier();
+
+ vq->inflight->desc[desc_idx].version++;
+
+ return 0;
+}
You probably need WRITE_ONCE() semantic (e.g volatile) to make sure the
value reach memory.
The cache line should have been flushed during crash. So we can see
the correct value when backend reconnecting. If so, compile barrier
should be enough here, right?
Maybe I worry too much but it's not about flushing cache, but about
whether or not compiler can generate mov to memory instead of mov to
registers.
Thanks
Thanks,
Yongji
