[PATCH RESEND v3 3/4] virtio-blk: default num_queues to -smp N

[Stefan Hajnoczi <stefanha@redhat.com>]

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.

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>
---
 hw/block/virtio-blk.c          | 6 +++++-
 hw/core/machine.c              | 1 +
 hw/virtio/virtio-blk-pci.c     | 9 ++++++++-
 include/hw/virtio/virtio-blk.h | 2 ++
 4 files changed, 16 insertions(+), 2 deletions(-)

diff --git a/hw/block/virtio-blk.c b/hw/block/virtio-blk.c
index 142863a3b2..ed5df71779 100644
--- a/hw/block/virtio-blk.c
+++ b/hw/block/virtio-blk.c
@@ -1135,6 +1135,9 @@ 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;
@@ -1271,7 +1274,8 @@ 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 4bbcec8fbd..c993b8f489 100644
--- a/hw/core/machine.c
+++ b/hw/core/machine.c
@@ -31,6 +31,7 @@
 GlobalProperty hw_compat_4_2[] = {
     { "virtio-blk-device", "queue-size", "128"},
     { "virtio-scsi-device", "virtqueue_size", "128"},
+    { "virtio-blk-device", "num-queues", "1"},
     { "virtio-blk-device", "x-enable-wce-if-config-wce", "off" },
     { "virtio-blk-device", "seg-max-adjust", "off"},
     { "virtio-scsi-device", "num_queues", "1"},
diff --git a/hw/virtio/virtio-blk-pci.c b/hw/virtio/virtio-blk-pci.c
index efb2c22a1d..bf628de675 100644
--- a/hw/virtio/virtio-blk-pci.c
+++ b/hw/virtio/virtio-blk-pci.c
@@ -17,6 +17,7 @@
 
 #include "qemu/osdep.h"
 
+#include "hw/boards.h"
 #include "hw/qdev-properties.h"
 #include "hw/virtio/virtio-blk.h"
 #include "virtio-pci.h"
@@ -50,9 +51,15 @@ 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;
+
+    /* 1:1 vq to vcpu mapping is ideal because it avoids IPIs */
+    if (conf->num_queues == VIRTIO_BLK_AUTO_NUM_QUEUES) {
+        conf->num_queues = current_machine->smp.cpus;
+    }
 
     if (vpci_dev->nvectors == DEV_NVECTORS_UNSPECIFIED) {
-        vpci_dev->nvectors = dev->vdev.conf.num_queues + 1;
+        vpci_dev->nvectors = conf->num_queues + 1;
     }
 
     qdev_set_parent_bus(vdev, BUS(&vpci_dev->bus));
diff --git a/include/hw/virtio/virtio-blk.h b/include/hw/virtio/virtio-blk.h
index 1e62f869b2..4e5e903f4a 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;
-- 
2.24.1