On Tue, Dec 18, 2012 at 01:32:52PM +0100, Paolo Bonzini wrote:
> This patch adds queue steering to virtio-scsi. When a target is sent
> multiple requests, we always drive them to the same queue so that FIFO
> processing order is kept. However, if a target was idle, we can choose
> a queue arbitrarily. In this case the queue is chosen according to the
> current VCPU, so the driver expects the number of request queues to be
> equal to the number of VCPUs. This makes it easy and fast to select
> the queue, and also lets the driver optimize the IRQ affinity for the
> virtqueues (each virtqueue's affinity is set to the CPU that "owns"
> the queue).
>
> The speedup comes from improving cache locality and giving CPU affinity
> to the virtqueues, which is why this scheme was selected. Assuming that
> the thread that is sending requests to the device is I/O-bound, it is
> likely to be sleeping at the time the ISR is executed, and thus executing
> the ISR on the same processor that sent the requests is cheap.
>
> However, the kernel will not execute the ISR on the "best" processor
> unless you explicitly set the affinity. This is because in practice
> you will have many such I/O-bound processes and thus many otherwise
> idle processors. Then the kernel will execute the ISR on a random
> processor, rather than the one that is sending requests to the device.
>
> The alternative to per-CPU virtqueues is per-target virtqueues. To
> achieve the same locality, we could dynamically choose the virtqueue's
> affinity based on the CPU of the last task that sent a request. This
> is less appealing because we do not set the affinity directly---we only
> provide a hint to the irqbalanced running in userspace. Dynamically
> changing the affinity only works if the userspace applies the hint
> fast enough.
>
> Signed-off-by: Paolo Bonzini <pbonzini@xxxxxxxxxx>
> ---
> v1->v2: improved comments and commit messages, added memory barriers
>
> drivers/scsi/virtio_scsi.c | 234 +++++++++++++++++++++++++++++++++++++------
> 1 files changed, 201 insertions(+), 33 deletions(-)
>
> diff --git a/drivers/scsi/virtio_scsi.c b/drivers/scsi/virtio_scsi.c
> index 4f6c6a3..ca9d29d 100644
> --- a/drivers/scsi/virtio_scsi.c
> +++ b/drivers/scsi/virtio_scsi.c
> @@ -26,6 +26,7 @@
>
> #define VIRTIO_SCSI_MEMPOOL_SZ 64
> #define VIRTIO_SCSI_EVENT_LEN 8
> +#define VIRTIO_SCSI_VQ_BASE 2
>
> /* Command queue element */
> struct virtio_scsi_cmd {
> @@ -57,24 +58,57 @@ struct virtio_scsi_vq {
> struct virtqueue *vq;
> };
>
> -/* Per-target queue state */
> +/*
> + * Per-target queue state.
> + *
> + * This struct holds the data needed by the queue steering policy. When a
> + * target is sent multiple requests, we need to drive them to the same queue so
> + * that FIFO processing order is kept. However, if a target was idle, we can
> + * choose a queue arbitrarily. In this case the queue is chosen according to
> + * the current VCPU, so the driver expects the number of request queues to be
> + * equal to the number of VCPUs. This makes it easy and fast to select the
> + * queue, and also lets the driver optimize the IRQ affinity for the virtqueues
> + * (each virtqueue's affinity is set to the CPU that "owns" the queue).
> + *
> + * An interesting effect of this policy is that only writes to req_vq need to
> + * take the tgt_lock. Read can be done outside the lock because:
> + *
> + * - writes of req_vq only occur when atomic_inc_return(&tgt->reqs) returns 1.
> + * In that case, no other CPU is reading req_vq: even if they were in
> + * virtscsi_queuecommand_multi, they would be spinning on tgt_lock.
> + *
> + * - reads of req_vq only occur when the target is not idle (reqs != 0).
> + * A CPU that enters virtscsi_queuecommand_multi will not modify req_vq.
> + *
> + * Similarly, decrements of reqs are never concurrent with writes of req_vq.
> + * Thus they can happen outside the tgt_lock, provided of course we make reqs
> + * an atomic_t.
> + */
> struct virtio_scsi_target_state {
> - /* Never held at the same time as vq_lock. */
> + /* This spinlock ever held at the same time as vq_lock. */
> spinlock_t tgt_lock;
> +
> + /* Count of outstanding requests. */
> + atomic_t reqs;
> +
> + /* Currently active virtqueue for requests sent to this target. */
> + struct virtio_scsi_vq *req_vq;
> };
>
> /* Driver instance state */
> struct virtio_scsi {
> struct virtio_device *vdev;
>
> - struct virtio_scsi_vq ctrl_vq;
> - struct virtio_scsi_vq event_vq;
> - struct virtio_scsi_vq req_vq;
> -
> /* Get some buffers ready for event vq */
> struct virtio_scsi_event_node event_list[VIRTIO_SCSI_EVENT_LEN];
>
> struct virtio_scsi_target_state *tgt;
> +
> + u32 num_queues;
> +
> + struct virtio_scsi_vq ctrl_vq;
> + struct virtio_scsi_vq event_vq;
> + struct virtio_scsi_vq req_vqs[];
> };
>
> static struct kmem_cache *virtscsi_cmd_cache;
> @@ -109,6 +143,7 @@ static void virtscsi_complete_cmd(struct virtio_scsi *vscsi, void *buf)
> struct virtio_scsi_cmd *cmd = buf;
> struct scsi_cmnd *sc = cmd->sc;
> struct virtio_scsi_cmd_resp *resp = &cmd->resp.cmd;
> + struct virtio_scsi_target_state *tgt = &vscsi->tgt[sc->device->id];
>
> dev_dbg(&sc->device->sdev_gendev,
> "cmd %p response %u status %#02x sense_len %u\n",
> @@ -163,6 +198,8 @@ static void virtscsi_complete_cmd(struct virtio_scsi *vscsi, void *buf)
>
> mempool_free(cmd, virtscsi_cmd_pool);
> sc->scsi_done(sc);
> +
> + atomic_dec(&tgt->reqs);
> }
>
> static void virtscsi_vq_done(struct virtio_scsi *vscsi, struct virtqueue *vq,
> @@ -182,11 +219,45 @@ static void virtscsi_req_done(struct virtqueue *vq)
> {
> struct Scsi_Host *sh = virtio_scsi_host(vq->vdev);
> struct virtio_scsi *vscsi = shost_priv(sh);
> + int index = vq->index - VIRTIO_SCSI_VQ_BASE;
> + struct virtio_scsi_vq *req_vq = &vscsi->req_vqs[index];
> unsigned long flags;
>
> - spin_lock_irqsave(&vscsi->req_vq.vq_lock, flags);
> + /*
> + * Read req_vq before decrementing the reqs field in
> + * virtscsi_complete_cmd.
> + *
> + * With barriers:
> + *
> + * CPU #0 virtscsi_queuecommand_multi (CPU #1)
> + * ------------------------------------------------------------
> + * lock vq_lock
> + * read req_vq
> + * read reqs (reqs = 1)
> + * write reqs (reqs = 0)
> + * increment reqs (reqs = 1)
> + * write req_vq
> + *
> + * Possible reordering without barriers:
> + *
> + * CPU #0 virtscsi_queuecommand_multi (CPU #1)
> + * ------------------------------------------------------------
> + * lock vq_lock
> + * read reqs (reqs = 1)
> + * write reqs (reqs = 0)
> + * increment reqs (reqs = 1)
> + * write req_vq
> + * read (wrong) req_vq
> + *
> + * We do not need a full smp_rmb, because req_vq is required to get
> + * to tgt->reqs: tgt is &vscsi->tgt[sc->device->id], where sc is stored
> + * in the virtqueue as the user token.
> + */
> + smp_read_barrier_depends();
> +
> + spin_lock_irqsave(&req_vq->vq_lock, flags);
> virtscsi_vq_done(vscsi, vq, virtscsi_complete_cmd);
> - spin_unlock_irqrestore(&vscsi->req_vq.vq_lock, flags);
> + spin_unlock_irqrestore(&req_vq->vq_lock, flags);
> };
>
> static void virtscsi_complete_free(struct virtio_scsi *vscsi, void *buf)
> @@ -424,11 +495,12 @@ static int virtscsi_kick_cmd(struct virtio_scsi_target_state *tgt,
> return ret;
> }
>
> -static int virtscsi_queuecommand(struct Scsi_Host *sh, struct scsi_cmnd *sc)
> +static int virtscsi_queuecommand(struct virtio_scsi *vscsi,
> + struct virtio_scsi_target_state *tgt,
> + struct scsi_cmnd *sc)
> {
> - struct virtio_scsi *vscsi = shost_priv(sh);
> - struct virtio_scsi_target_state *tgt = &vscsi->tgt[sc->device->id];
> struct virtio_scsi_cmd *cmd;
> + struct virtio_scsi_vq *req_vq;
> int ret;
>
> struct Scsi_Host *shost = virtio_scsi_host(vscsi->vdev);
> @@ -461,7 +533,8 @@ static int virtscsi_queuecommand(struct Scsi_Host *sh, struct scsi_cmnd *sc)
> BUG_ON(sc->cmd_len > VIRTIO_SCSI_CDB_SIZE);
> memcpy(cmd->req.cmd.cdb, sc->cmnd, sc->cmd_len);
>
> - if (virtscsi_kick_cmd(tgt, &vscsi->req_vq, cmd,
> + req_vq = ACCESS_ONCE(tgt->req_vq);
This ACCESS_ONCE without a barrier looks strange to me.
Can req_vq change? Needs a comment.
> + if (virtscsi_kick_cmd(tgt, req_vq, cmd,
> sizeof cmd->req.cmd, sizeof cmd->resp.cmd,
> GFP_ATOMIC) == 0)
> ret = 0;
> @@ -472,6 +545,48 @@ out:
> return ret;
> }
>
> +static int virtscsi_queuecommand_single(struct Scsi_Host *sh,
> + struct scsi_cmnd *sc)
> +{
> + struct virtio_scsi *vscsi = shost_priv(sh);
> + struct virtio_scsi_target_state *tgt = &vscsi->tgt[sc->device->id];
> +
> + atomic_inc(&tgt->reqs);
And here we don't have barrier after atomic? Why? Needs a comment.
> + return virtscsi_queuecommand(vscsi, tgt, sc);
> +}
> +
> +static int virtscsi_queuecommand_multi(struct Scsi_Host *sh,
> + struct scsi_cmnd *sc)
> +{
> + struct virtio_scsi *vscsi = shost_priv(sh);
> + struct virtio_scsi_target_state *tgt = &vscsi->tgt[sc->device->id];
> + unsigned long flags;
> + u32 queue_num;
> +
> + /*
> + * Using an atomic_t for tgt->reqs lets the virtqueue handler
> + * decrement it without taking the spinlock.
> + *
> + * We still need a critical section to prevent concurrent submissions
> + * from picking two different req_vqs.
> + */
> + spin_lock_irqsave(&tgt->tgt_lock, flags);
> + if (atomic_inc_return(&tgt->reqs) == 1) {
> + queue_num = smp_processor_id();
> + while (unlikely(queue_num >= vscsi->num_queues))
> + queue_num -= vscsi->num_queues;
> +
> + /*
> + * Write reqs before writing req_vq, matching the
> + * smp_read_barrier_depends() in virtscsi_req_done.
> + */
> + smp_wmb();
> + tgt->req_vq = &vscsi->req_vqs[queue_num];
> + }
> + spin_unlock_irqrestore(&tgt->tgt_lock, flags);
> + return virtscsi_queuecommand(vscsi, tgt, sc);
> +}
> +
> static int virtscsi_tmf(struct virtio_scsi *vscsi, struct virtio_scsi_cmd *cmd)
> {
> DECLARE_COMPLETION_ONSTACK(comp);
> @@ -541,12 +656,26 @@ static int virtscsi_abort(struct scsi_cmnd *sc)
> return virtscsi_tmf(vscsi, cmd);
> }
>
> -static struct scsi_host_template virtscsi_host_template = {
> +static struct scsi_host_template virtscsi_host_template_single = {
> .module = THIS_MODULE,
> .name = "Virtio SCSI HBA",
> .proc_name = "virtio_scsi",
> - .queuecommand = virtscsi_queuecommand,
> .this_id = -1,
> + .queuecommand = virtscsi_queuecommand_single,
> + .eh_abort_handler = virtscsi_abort,
> + .eh_device_reset_handler = virtscsi_device_reset,
> +
> + .can_queue = 1024,
> + .dma_boundary = UINT_MAX,
> + .use_clustering = ENABLE_CLUSTERING,
> +};
> +
> +static struct scsi_host_template virtscsi_host_template_multi = {
> + .module = THIS_MODULE,
> + .name = "Virtio SCSI HBA",
> + .proc_name = "virtio_scsi",
> + .this_id = -1,
> + .queuecommand = virtscsi_queuecommand_multi,
> .eh_abort_handler = virtscsi_abort,
> .eh_device_reset_handler = virtscsi_device_reset,
>
> @@ -572,16 +701,27 @@ static struct scsi_host_template virtscsi_host_template = {
> &__val, sizeof(__val)); \
> })
>
> +
> static void virtscsi_init_vq(struct virtio_scsi_vq *virtscsi_vq,
> - struct virtqueue *vq)
> + struct virtqueue *vq, bool affinity)
> {
> spin_lock_init(&virtscsi_vq->vq_lock);
> virtscsi_vq->vq = vq;
> + if (affinity)
> + virtqueue_set_affinity(vq, vq->index - VIRTIO_SCSI_VQ_BASE);
I've been thinking about how set_affinity
interacts with online/offline CPUs.
Any idea?
> }
>
> -static void virtscsi_init_tgt(struct virtio_scsi_target_state *tgt)
> +static void virtscsi_init_tgt(struct virtio_scsi *vscsi, int i)
> {
> + struct virtio_scsi_target_state *tgt = &vscsi->tgt[i];
> spin_lock_init(&tgt->tgt_lock);
> + atomic_set(&tgt->reqs, 0);
> +
> + /*
> + * The default is unused for multiqueue, but with a single queue
> + * or target we use it in virtscsi_queuecommand.
> + */
> + tgt->req_vq = &vscsi->req_vqs[0];
> }
>
> static void virtscsi_scan(struct virtio_device *vdev)
> @@ -609,28 +749,41 @@ static int virtscsi_init(struct virtio_device *vdev,
> struct virtio_scsi *vscsi, int num_targets)
> {
> int err;
> - struct virtqueue *vqs[3];
> u32 i, sg_elems;
> + u32 num_vqs;
> + vq_callback_t **callbacks;
> + const char **names;
> + struct virtqueue **vqs;
>
> - vq_callback_t *callbacks[] = {
> - virtscsi_ctrl_done,
> - virtscsi_event_done,
> - virtscsi_req_done
> - };
> - const char *names[] = {
> - "control",
> - "event",
> - "request"
> - };
> + num_vqs = vscsi->num_queues + VIRTIO_SCSI_VQ_BASE;
> + vqs = kmalloc(num_vqs * sizeof(struct virtqueue *), GFP_KERNEL);
> + callbacks = kmalloc(num_vqs * sizeof(vq_callback_t *), GFP_KERNEL);
> + names = kmalloc(num_vqs * sizeof(char *), GFP_KERNEL);
> +
> + if (!callbacks || !vqs || !names) {
> + err = -ENOMEM;
> + goto out;
> + }
> +
> + callbacks[0] = virtscsi_ctrl_done;
> + callbacks[1] = virtscsi_event_done;
> + names[0] = "control";
> + names[1] = "event";
> + for (i = VIRTIO_SCSI_VQ_BASE; i < num_vqs; i++) {
> + callbacks[i] = virtscsi_req_done;
> + names[i] = "request";
> + }
>
> /* Discover virtqueues and write information to configuration. */
> - err = vdev->config->find_vqs(vdev, 3, vqs, callbacks, names);
> + err = vdev->config->find_vqs(vdev, num_vqs, vqs, callbacks, names);
> if (err)
> return err;
>
> - virtscsi_init_vq(&vscsi->ctrl_vq, vqs[0]);
> - virtscsi_init_vq(&vscsi->event_vq, vqs[1]);
> - virtscsi_init_vq(&vscsi->req_vq, vqs[2]);
> + virtscsi_init_vq(&vscsi->ctrl_vq, vqs[0], false);
> + virtscsi_init_vq(&vscsi->event_vq, vqs[1], false);
> + for (i = VIRTIO_SCSI_VQ_BASE; i < num_vqs; i++)
> + virtscsi_init_vq(&vscsi->req_vqs[i - VIRTIO_SCSI_VQ_BASE],
> + vqs[i], vscsi->num_queues > 1);
So affinity is true if >1 vq? I am guessing this is not
going to do the right thing unless you have at least
as many vqs as CPUs.
>
> virtscsi_config_set(vdev, cdb_size, VIRTIO_SCSI_CDB_SIZE);
> virtscsi_config_set(vdev, sense_size, VIRTIO_SCSI_SENSE_SIZE);
> @@ -647,11 +800,14 @@ static int virtscsi_init(struct virtio_device *vdev,
> goto out;
> }
> for (i = 0; i < num_targets; i++)
> - virtscsi_init_tgt(&vscsi->tgt[i]);
> + virtscsi_init_tgt(vscsi, i);
>
> err = 0;
>
> out:
> + kfree(names);
> + kfree(callbacks);
> + kfree(vqs);
> if (err)
> virtscsi_remove_vqs(vdev);
> return err;
> @@ -664,11 +820,22 @@ static int __devinit virtscsi_probe(struct virtio_device *vdev)
> int err;
> u32 sg_elems, num_targets;
> u32 cmd_per_lun;
> + u32 num_queues;
> + struct scsi_host_template *hostt;
> +
> + /* We need to know how many queues before we allocate. */
> + num_queues = virtscsi_config_get(vdev, num_queues) ?: 1;
>
> /* Allocate memory and link the structs together. */
> num_targets = virtscsi_config_get(vdev, max_target) + 1;
> - shost = scsi_host_alloc(&virtscsi_host_template, sizeof(*vscsi));
>
> + if (num_queues == 1)
> + hostt = &virtscsi_host_template_single;
> + else
> + hostt = &virtscsi_host_template_multi;
> +
> + shost = scsi_host_alloc(hostt,
> + sizeof(*vscsi) + sizeof(vscsi->req_vqs[0]) * num_queues);
> if (!shost)
> return -ENOMEM;
>
> @@ -676,6 +843,7 @@ static int __devinit virtscsi_probe(struct virtio_device *vdev)
> shost->sg_tablesize = sg_elems;
> vscsi = shost_priv(shost);
> vscsi->vdev = vdev;
> + vscsi->num_queues = num_queues;
> vdev->priv = shost;
>
> err = virtscsi_init(vdev, vscsi, num_targets);
> --
> 1.7.1
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
