virtio: harsher barriers for rpmsg.

We were cheating with our barriers; using the smp ones rather than the
real device ones.  That was fine, until rpmsg came along, which is
used to talk to a real device (a non-SMP CPU).

Unfortunately, just putting back the real barriers (reverting
d57ed95d) causes a performance regression on virtio-pci.  In
particular, Amos reports netbench's TCP_RR over virtio_net CPU
utilization increased up to 35% while throughput went down by up to
14%.

By comparison, this branch is in the noise.

Reference: https://lkml.org/lkml/2011/12/11/22

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>
diff --git a/drivers/lguest/lguest_device.c b/drivers/lguest/lguest_device.c
index 595d731..6a1d644 100644
--- a/drivers/lguest/lguest_device.c
+++ b/drivers/lguest/lguest_device.c
@@ -292,10 +292,12 @@
 
 	/*
 	 * OK, tell virtio_ring.c to set up a virtqueue now we know its size
-	 * and we've got a pointer to its pages.
+	 * and we've got a pointer to its pages.  Note that we set weak_barriers
+	 * to 'true': the host just a(nother) SMP CPU, so we only need inter-cpu
+	 * barriers.
 	 */
-	vq = vring_new_virtqueue(lvq->config.num, LGUEST_VRING_ALIGN,
-				 vdev, lvq->pages, lg_notify, callback, name);
+	vq = vring_new_virtqueue(lvq->config.num, LGUEST_VRING_ALIGN, vdev,
+				 true, lvq->pages, lg_notify, callback, name);
 	if (!vq) {
 		err = -ENOMEM;
 		goto unmap;
diff --git a/drivers/s390/kvm/kvm_virtio.c b/drivers/s390/kvm/kvm_virtio.c
index 8af868b..7bc1955 100644
--- a/drivers/s390/kvm/kvm_virtio.c
+++ b/drivers/s390/kvm/kvm_virtio.c
@@ -198,7 +198,7 @@
 		goto out;
 
 	vq = vring_new_virtqueue(config->num, KVM_S390_VIRTIO_RING_ALIGN,
-				 vdev, (void *) config->address,
+				 vdev, true, (void *) config->address,
 				 kvm_notify, callback, name);
 	if (!vq) {
 		err = -ENOMEM;
diff --git a/drivers/virtio/virtio_mmio.c b/drivers/virtio/virtio_mmio.c
index 0269717..01d6dc2 100644
--- a/drivers/virtio/virtio_mmio.c
+++ b/drivers/virtio/virtio_mmio.c
@@ -310,8 +310,8 @@
 			vm_dev->base + VIRTIO_MMIO_QUEUE_PFN);
 
 	/* Create the vring */
-	vq = vring_new_virtqueue(info->num, VIRTIO_MMIO_VRING_ALIGN,
-				 vdev, info->queue, vm_notify, callback, name);
+	vq = vring_new_virtqueue(info->num, VIRTIO_MMIO_VRING_ALIGN, vdev,
+				 true, info->queue, vm_notify, callback, name);
 	if (!vq) {
 		err = -ENOMEM;
 		goto error_new_virtqueue;
diff --git a/drivers/virtio/virtio_pci.c b/drivers/virtio/virtio_pci.c
index baabb79..688b42d 100644
--- a/drivers/virtio/virtio_pci.c
+++ b/drivers/virtio/virtio_pci.c
@@ -414,8 +414,8 @@
 		  vp_dev->ioaddr + VIRTIO_PCI_QUEUE_PFN);
 
 	/* create the vring */
-	vq = vring_new_virtqueue(info->num, VIRTIO_PCI_VRING_ALIGN,
-				 vdev, info->queue, vp_notify, callback, name);
+	vq = vring_new_virtqueue(info->num, VIRTIO_PCI_VRING_ALIGN, vdev,
+				 true, info->queue, vp_notify, callback, name);
 	if (!vq) {
 		err = -ENOMEM;
 		goto out_activate_queue;
diff --git a/drivers/virtio/virtio_ring.c b/drivers/virtio/virtio_ring.c
index c7a2c20..50da920 100644
--- a/drivers/virtio/virtio_ring.c
+++ b/drivers/virtio/virtio_ring.c
@@ -28,17 +28,20 @@
 #ifdef CONFIG_SMP
 /* Where possible, use SMP barriers which are more lightweight than mandatory
  * barriers, because mandatory barriers control MMIO effects on accesses
- * through relaxed memory I/O windows (which virtio does not use). */
-#define virtio_mb() smp_mb()
-#define virtio_rmb() smp_rmb()
-#define virtio_wmb() smp_wmb()
+ * through relaxed memory I/O windows (which virtio-pci does not use). */
+#define virtio_mb(vq) \
+	do { if ((vq)->weak_barriers) smp_mb(); else mb(); } while(0)
+#define virtio_rmb(vq) \
+	do { if ((vq)->weak_barriers) smp_rmb(); else rmb(); } while(0)
+#define virtio_wmb(vq) \
+	do { if ((vq)->weak_barriers) smp_rmb(); else rmb(); } while(0)
 #else
 /* We must force memory ordering even if guest is UP since host could be
  * running on another CPU, but SMP barriers are defined to barrier() in that
  * configuration. So fall back to mandatory barriers instead. */
-#define virtio_mb() mb()
-#define virtio_rmb() rmb()
-#define virtio_wmb() wmb()
+#define virtio_mb(vq) mb()
+#define virtio_rmb(vq) rmb()
+#define virtio_wmb(vq) wmb()
 #endif
 
 #ifdef DEBUG
@@ -77,6 +80,9 @@
 	/* Actual memory layout for this queue */
 	struct vring vring;
 
+	/* Can we use weak barriers? */
+	bool weak_barriers;
+
 	/* Other side has made a mess, don't try any more. */
 	bool broken;
 
@@ -245,14 +251,14 @@
 	START_USE(vq);
 	/* Descriptors and available array need to be set before we expose the
 	 * new available array entries. */
-	virtio_wmb();
+	virtio_wmb(vq);
 
 	old = vq->vring.avail->idx;
 	new = vq->vring.avail->idx = old + vq->num_added;
 	vq->num_added = 0;
 
 	/* Need to update avail index before checking if we should notify */
-	virtio_mb();
+	virtio_mb(vq);
 
 	if (vq->event ?
 	    vring_need_event(vring_avail_event(&vq->vring), new, old) :
@@ -314,7 +320,7 @@
 	}
 
 	/* Only get used array entries after they have been exposed by host. */
-	virtio_rmb();
+	virtio_rmb(vq);
 
 	i = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].id;
 	*len = vq->vring.used->ring[vq->last_used_idx%vq->vring.num].len;
@@ -337,7 +343,7 @@
 	 * the read in the next get_buf call. */
 	if (!(vq->vring.avail->flags & VRING_AVAIL_F_NO_INTERRUPT)) {
 		vring_used_event(&vq->vring) = vq->last_used_idx;
-		virtio_mb();
+		virtio_mb(vq);
 	}
 
 	END_USE(vq);
@@ -366,7 +372,7 @@
 	 * entry. Always do both to keep code simple. */
 	vq->vring.avail->flags &= ~VRING_AVAIL_F_NO_INTERRUPT;
 	vring_used_event(&vq->vring) = vq->last_used_idx;
-	virtio_mb();
+	virtio_mb(vq);
 	if (unlikely(more_used(vq))) {
 		END_USE(vq);
 		return false;
@@ -393,7 +399,7 @@
 	/* TODO: tune this threshold */
 	bufs = (u16)(vq->vring.avail->idx - vq->last_used_idx) * 3 / 4;
 	vring_used_event(&vq->vring) = vq->last_used_idx + bufs;
-	virtio_mb();
+	virtio_mb(vq);
 	if (unlikely((u16)(vq->vring.used->idx - vq->last_used_idx) > bufs)) {
 		END_USE(vq);
 		return false;
@@ -453,6 +459,7 @@
 struct virtqueue *vring_new_virtqueue(unsigned int num,
 				      unsigned int vring_align,
 				      struct virtio_device *vdev,
+				      bool weak_barriers,
 				      void *pages,
 				      void (*notify)(struct virtqueue *),
 				      void (*callback)(struct virtqueue *),
@@ -476,6 +483,7 @@
 	vq->vq.vdev = vdev;
 	vq->vq.name = name;
 	vq->notify = notify;
+	vq->weak_barriers = weak_barriers;
 	vq->broken = false;
 	vq->last_used_idx = 0;
 	vq->num_added = 0;
diff --git a/include/linux/virtio_ring.h b/include/linux/virtio_ring.h
index 36be0f6..e338730 100644
--- a/include/linux/virtio_ring.h
+++ b/include/linux/virtio_ring.h
@@ -168,6 +168,7 @@
 struct virtqueue *vring_new_virtqueue(unsigned int num,
 				      unsigned int vring_align,
 				      struct virtio_device *vdev,
+				      bool weak_barriers,
 				      void *pages,
 				      void (*notify)(struct virtqueue *vq),
 				      void (*callback)(struct virtqueue *vq),
diff --git a/tools/virtio/linux/virtio.h b/tools/virtio/linux/virtio.h
index 669bcdd..953db2a 100644
--- a/tools/virtio/linux/virtio.h
+++ b/tools/virtio/linux/virtio.h
@@ -214,6 +214,7 @@
 struct virtqueue *vring_new_virtqueue(unsigned int num,
 				      unsigned int vring_align,
 				      struct virtio_device *vdev,
+				      bool weak_barriers,
 				      void *pages,
 				      void (*notify)(struct virtqueue *vq),
 				      void (*callback)(struct virtqueue *vq),
diff --git a/tools/virtio/virtio_test.c b/tools/virtio/virtio_test.c
index 74d3331..0740284 100644
--- a/tools/virtio/virtio_test.c
+++ b/tools/virtio/virtio_test.c
@@ -92,7 +92,8 @@
 	assert(r >= 0);
 	memset(info->ring, 0, vring_size(num, 4096));
 	vring_init(&info->vring, num, info->ring, 4096);
-	info->vq = vring_new_virtqueue(info->vring.num, 4096, &dev->vdev, info->ring,
+	info->vq = vring_new_virtqueue(info->vring.num, 4096, &dev->vdev,
+				       true, info->ring,
 				       vq_notify, vq_callback, "test");
 	assert(info->vq);
 	info->vq->priv = info;