virtio-blk-pci: default num_queues to -smp N

Automatically size the number of virtio-blk-pci request virtqueues to
match the number of vCPUs.  Other transports continue to default to 1
request virtqueue.

A 1:1 virtqueue:vCPU mapping ensures that completion interrupts are
handled on the same vCPU that submitted the request.  No IPI is
necessary to complete an I/O request and performance is improved.  The
maximum number of MSI-X vectors and virtqueues limit are respected.

Performance improves from 78k to 104k IOPS on a 32 vCPU guest with 101
virtio-blk-pci devices (ioengine=libaio, iodepth=1, bs=4k, rw=randread
with NVMe storage).

Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
Reviewed-by: Cornelia Huck <cohuck@redhat.com>
Reviewed-by: Pankaj Gupta <pankaj.gupta.linux@gmail.com>
Message-Id: <20200818143348.310613-7-stefanha@redhat.com>
Reviewed-by: Michael S. Tsirkin <mst@redhat.com>
Signed-off-by: Michael S. Tsirkin <mst@redhat.com>

4 files changed, 14 insertions, 2 deletions

diff --git a/hw/block/virtio-blk.c b/hw/block/virtio-blk.c
index 413783693c..2204ba149e 100644
--- a/hw/block/virtio-blk.c
+++ b/hw/block/virtio-blk.c
@@ -1147,6 +1147,9 @@ static void virtio_blk_device_realize(DeviceState *dev, Error **errp)
         error_setg(errp, "Device needs media, but drive is empty");
         return;
     }
+    if (conf->num_queues == VIRTIO_BLK_AUTO_NUM_QUEUES) {
+        conf->num_queues = 1;
+    }
     if (!conf->num_queues) {
         error_setg(errp, "num-queues property must be larger than 0");
         return;
@@ -1281,7 +1284,8 @@ static Property virtio_blk_properties[] = {
 #endif
     DEFINE_PROP_BIT("request-merging", VirtIOBlock, conf.request_merging, 0,
                     true),
-    DEFINE_PROP_UINT16("num-queues", VirtIOBlock, conf.num_queues, 1),
+    DEFINE_PROP_UINT16("num-queues", VirtIOBlock, conf.num_queues,
+                       VIRTIO_BLK_AUTO_NUM_QUEUES),
     DEFINE_PROP_UINT16("queue-size", VirtIOBlock, conf.queue_size, 256),
     DEFINE_PROP_BOOL("seg-max-adjust", VirtIOBlock, conf.seg_max_adjust, true),
     DEFINE_PROP_LINK("iothread", VirtIOBlock, conf.iothread, TYPE_IOTHREAD,
diff --git a/hw/core/machine.c b/hw/core/machine.c
index 9ee2aa0f7b..7f65fa8743 100644
--- a/hw/core/machine.c
+++ b/hw/core/machine.c
@@ -31,6 +31,7 @@
 GlobalProperty hw_compat_5_1[] = {
     { "vhost-scsi", "num_queues", "1"},
     { "vhost-user-scsi", "num_queues", "1"},
+    { "virtio-blk-device", "num-queues", "1"},
     { "virtio-scsi-device", "num_queues", "1"},
 };
 const size_t hw_compat_5_1_len = G_N_ELEMENTS(hw_compat_5_1);
diff --git a/hw/virtio/virtio-blk-pci.c b/hw/virtio/virtio-blk-pci.c
index 849cc7dfd8..37c6e0aeb4 100644
--- a/hw/virtio/virtio-blk-pci.c
+++ b/hw/virtio/virtio-blk-pci.c
@@ -50,9 +50,14 @@ static void virtio_blk_pci_realize(VirtIOPCIProxy *vpci_dev, Error **errp)
 {
     VirtIOBlkPCI *dev = VIRTIO_BLK_PCI(vpci_dev);
     DeviceState *vdev = DEVICE(&dev->vdev);
+    VirtIOBlkConf *conf = &dev->vdev.conf;
+
+    if (conf->num_queues == VIRTIO_BLK_AUTO_NUM_QUEUES) {
+        conf->num_queues = virtio_pci_optimal_num_queues(0);
+    }
 
     if (vpci_dev->nvectors == DEV_NVECTORS_UNSPECIFIED) {
-        vpci_dev->nvectors = dev->vdev.conf.num_queues + 1;
+        vpci_dev->nvectors = conf->num_queues + 1;
     }
 
     qdev_realize(vdev, BUS(&vpci_dev->bus), errp);
diff --git a/include/hw/virtio/virtio-blk.h b/include/hw/virtio/virtio-blk.h
index b1334c3904..7539c2b848 100644
--- a/include/hw/virtio/virtio-blk.h
+++ b/include/hw/virtio/virtio-blk.h
@@ -30,6 +30,8 @@ struct virtio_blk_inhdr
     unsigned char status;
 };
 
+#define VIRTIO_BLK_AUTO_NUM_QUEUES UINT16_MAX
+
 struct VirtIOBlkConf
 {
     BlockConf conf;