On 02/06/2013 04:39 PM, Nicholas A. Bellinger wrote: > On Wed, 2013-02-06 at 15:09 +0800, Asias He wrote: >> On 02/06/2013 02:45 PM, Nicholas A. Bellinger wrote: >>> On Wed, 2013-02-06 at 13:20 +0800, Asias He wrote: >>>> This adds virtio-scsi multi-queue support to tcm_vhost. >>>> >>>> Guest side virtio-scsi multi-queue support can be found here: >>>> >>>> https://lkml.org/lkml/2012/12/18/166 >>>> >>>> Some initial perf numbers: >>>> 1 queue, 4 targets, 1 lun per target >>>> 4K request size, 50% randread + 50% randwrite: 127K/127k IOPS >>>> >>>> 4 queues, 4 targets, 1 lun per target >>>> 4K request size, 50% randread + 50% randwrite: 181K/181k IOPS >>>> >>> >>> Nice single LUN small block random I/O improvement here with 4x vqueues. >>> >>> Curious to see how virtio-scsi small block performance looks with >>> SCSI-core to multi-LUN tcm_vhost endpoints as well.. 8-) >> >> Do you mean something like this? >> >> 1 queue, 2 targets, 2 lun per target >> 4 queue, 2 targets, 2 lun per target >> >>> Btw, this does not apply atop current target-pending.git/for-next with >>> your other pending vhost patch series, and AFAICT this patch is supposed >>> to apply on top of your last PATCH-v3, no..? >> >> Ah, this applies on top of mst's 'tcm_vhost: fix pr_err on early kick >> patch.' plus my last v3 of 'tcm_vhost: Multi-target support'. >> > > In that case, applying this patch + PATCH-v3 to auto-next for testing > for the moment, and will respin for-next against upstream w/ MST's patch > shortly. Okay. Looking forward to more perf numbers. > Also, please include a proper changelog for this second patch. :) Sure. ----> tcm_vhost: Multi-queue support This adds virtio-scsi multi-queue support to tcm_vhost. In order to use multi-queue, guest side multi-queue support is need. It can be found here: https://lkml.org/lkml/2012/12/18/166 Currently, only one thread is created by vhost core code for each vhost_scsi instance. Even if there are multi-queues, all the handling of guest kick (vhost_scsi_handle_kick) are processed in one thread. This is not optimal. Luckily, most of the work is offloaded to the tcm_vhost workqueue. Some initial perf numbers: 1 queue, 4 targets, 1 lun per target 4K request size, 50% randread + 50% randwrite: 127K/127k IOPS 4 queues, 4 targets, 1 lun per target 4K request size, 50% randread + 50% randwrite: 181K/181k IOPS Signed-off-by: Asias He <asias@xxxxxxxxxx> <---- > Thank you! > > --nab > > > >>> --nab >>> >>>> Signed-off-by: Asias He <asias@xxxxxxxxxx> >>>> --- >>>> drivers/vhost/tcm_vhost.c | 46 +++++++++++++++++++++++++++++----------------- >>>> drivers/vhost/tcm_vhost.h | 2 ++ >>>> 2 files changed, 31 insertions(+), 17 deletions(-) >>>> >>>> diff --git a/drivers/vhost/tcm_vhost.c b/drivers/vhost/tcm_vhost.c >>>> index 81ecda5..9951297 100644 >>>> --- a/drivers/vhost/tcm_vhost.c >>>> +++ b/drivers/vhost/tcm_vhost.c >>>> @@ -48,6 +48,7 @@ >>>> #include <linux/virtio_net.h> /* TODO vhost.h currently depends on this */ >>>> #include <linux/virtio_scsi.h> >>>> #include <linux/llist.h> >>>> +#include <linux/bitmap.h> >>>> >>>> #include "vhost.c" >>>> #include "vhost.h" >>>> @@ -59,7 +60,8 @@ enum { >>>> VHOST_SCSI_VQ_IO = 2, >>>> }; >>>> >>>> -#define VHOST_SCSI_MAX_TARGET 256 >>>> +#define VHOST_SCSI_MAX_TARGET 256 >>>> +#define VHOST_SCSI_MAX_VQ 128 >>>> >>>> struct vhost_scsi { >>>> /* Protected by vhost_scsi->dev.mutex */ >>>> @@ -68,7 +70,7 @@ struct vhost_scsi { >>>> bool vs_endpoint; >>>> >>>> struct vhost_dev dev; >>>> - struct vhost_virtqueue vqs[3]; >>>> + struct vhost_virtqueue vqs[VHOST_SCSI_MAX_VQ]; >>>> >>>> struct vhost_work vs_completion_work; /* cmd completion work item */ >>>> struct llist_head vs_completion_list; /* cmd completion queue */ >>>> @@ -366,12 +368,14 @@ static void vhost_scsi_complete_cmd_work(struct vhost_work *work) >>>> { >>>> struct vhost_scsi *vs = container_of(work, struct vhost_scsi, >>>> vs_completion_work); >>>> + DECLARE_BITMAP(signal, VHOST_SCSI_MAX_VQ); >>>> struct virtio_scsi_cmd_resp v_rsp; >>>> struct tcm_vhost_cmd *tv_cmd; >>>> struct llist_node *llnode; >>>> struct se_cmd *se_cmd; >>>> - int ret; >>>> + int ret, vq; >>>> >>>> + bitmap_zero(signal, VHOST_SCSI_MAX_VQ); >>>> llnode = llist_del_all(&vs->vs_completion_list); >>>> while (llnode) { >>>> tv_cmd = llist_entry(llnode, struct tcm_vhost_cmd, >>>> @@ -390,15 +394,20 @@ static void vhost_scsi_complete_cmd_work(struct vhost_work *work) >>>> memcpy(v_rsp.sense, tv_cmd->tvc_sense_buf, >>>> v_rsp.sense_len); >>>> ret = copy_to_user(tv_cmd->tvc_resp, &v_rsp, sizeof(v_rsp)); >>>> - if (likely(ret == 0)) >>>> - vhost_add_used(&vs->vqs[2], tv_cmd->tvc_vq_desc, 0); >>>> - else >>>> + if (likely(ret == 0)) { >>>> + vhost_add_used(tv_cmd->tvc_vq, tv_cmd->tvc_vq_desc, 0); >>>> + vq = tv_cmd->tvc_vq - vs->vqs; >>>> + __set_bit(vq, signal); >>>> + } else >>>> pr_err("Faulted on virtio_scsi_cmd_resp\n"); >>>> >>>> vhost_scsi_free_cmd(tv_cmd); >>>> } >>>> >>>> - vhost_signal(&vs->dev, &vs->vqs[2]); >>>> + vq = -1; >>>> + while ((vq = find_next_bit(signal, VHOST_SCSI_MAX_VQ, vq + 1)) >>>> + < VHOST_SCSI_MAX_VQ) >>>> + vhost_signal(&vs->dev, &vs->vqs[vq]); >>>> } >>>> >>>> static struct tcm_vhost_cmd *vhost_scsi_allocate_cmd( >>>> @@ -561,9 +570,9 @@ static void tcm_vhost_submission_work(struct work_struct *work) >>>> } >>>> } >>>> >>>> -static void vhost_scsi_handle_vq(struct vhost_scsi *vs) >>>> +static void vhost_scsi_handle_vq(struct vhost_scsi *vs, >>>> + struct vhost_virtqueue *vq) >>>> { >>>> - struct vhost_virtqueue *vq = &vs->vqs[2]; >>>> struct virtio_scsi_cmd_req v_req; >>>> struct tcm_vhost_tpg *tv_tpg; >>>> struct tcm_vhost_cmd *tv_cmd; >>>> @@ -656,7 +665,7 @@ static void vhost_scsi_handle_vq(struct vhost_scsi *vs) >>>> ret = __copy_to_user(resp, &rsp, sizeof(rsp)); >>>> if (!ret) >>>> vhost_add_used_and_signal(&vs->dev, >>>> - &vs->vqs[2], head, 0); >>>> + vq, head, 0); >>>> else >>>> pr_err("Faulted on virtio_scsi_cmd_resp\n"); >>>> >>>> @@ -678,6 +687,7 @@ static void vhost_scsi_handle_vq(struct vhost_scsi *vs) >>>> ": %d\n", tv_cmd, exp_data_len, data_direction); >>>> >>>> tv_cmd->tvc_vhost = vs; >>>> + tv_cmd->tvc_vq = vq; >>>> >>>> if (unlikely(vq->iov[out].iov_len != >>>> sizeof(struct virtio_scsi_cmd_resp))) { >>>> @@ -758,7 +768,7 @@ static void vhost_scsi_handle_kick(struct vhost_work *work) >>>> poll.work); >>>> struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev); >>>> >>>> - vhost_scsi_handle_vq(vs); >>>> + vhost_scsi_handle_vq(vs, vq); >>>> } >>>> >>>> /* >>>> @@ -879,7 +889,7 @@ err: >>>> static int vhost_scsi_open(struct inode *inode, struct file *f) >>>> { >>>> struct vhost_scsi *s; >>>> - int r; >>>> + int r, i; >>>> >>>> s = kzalloc(sizeof(*s), GFP_KERNEL); >>>> if (!s) >>>> @@ -889,8 +899,9 @@ static int vhost_scsi_open(struct inode *inode, struct file *f) >>>> >>>> s->vqs[VHOST_SCSI_VQ_CTL].handle_kick = vhost_scsi_ctl_handle_kick; >>>> s->vqs[VHOST_SCSI_VQ_EVT].handle_kick = vhost_scsi_evt_handle_kick; >>>> - s->vqs[VHOST_SCSI_VQ_IO].handle_kick = vhost_scsi_handle_kick; >>>> - r = vhost_dev_init(&s->dev, s->vqs, 3); >>>> + for (i = VHOST_SCSI_VQ_IO; i < VHOST_SCSI_MAX_VQ; i++) >>>> + s->vqs[i].handle_kick = vhost_scsi_handle_kick; >>>> + r = vhost_dev_init(&s->dev, s->vqs, VHOST_SCSI_MAX_VQ); >>>> if (r < 0) { >>>> kfree(s); >>>> return r; >>>> @@ -922,9 +933,10 @@ static void vhost_scsi_flush_vq(struct vhost_scsi *vs, int index) >>>> >>>> static void vhost_scsi_flush(struct vhost_scsi *vs) >>>> { >>>> - vhost_scsi_flush_vq(vs, VHOST_SCSI_VQ_CTL); >>>> - vhost_scsi_flush_vq(vs, VHOST_SCSI_VQ_EVT); >>>> - vhost_scsi_flush_vq(vs, VHOST_SCSI_VQ_IO); >>>> + int i; >>>> + >>>> + for (i = 0; i < VHOST_SCSI_MAX_VQ; i++) >>>> + vhost_scsi_flush_vq(vs, i); >>>> } >>>> >>>> static int vhost_scsi_set_features(struct vhost_scsi *vs, u64 features) >>>> diff --git a/drivers/vhost/tcm_vhost.h b/drivers/vhost/tcm_vhost.h >>>> index 519a550..1d2ae7a 100644 >>>> --- a/drivers/vhost/tcm_vhost.h >>>> +++ b/drivers/vhost/tcm_vhost.h >>>> @@ -23,6 +23,8 @@ struct tcm_vhost_cmd { >>>> struct virtio_scsi_cmd_resp __user *tvc_resp; >>>> /* Pointer to vhost_scsi for our device */ >>>> struct vhost_scsi *tvc_vhost; >>>> + /* Pointer to vhost_virtqueue for the cmd */ >>>> + struct vhost_virtqueue *tvc_vq; >>>> /* Pointer to vhost nexus memory */ >>>> struct tcm_vhost_nexus *tvc_nexus; >>>> /* The TCM I/O descriptor that is accessed via container_of() */ >>> >>> >> >> > > -- Asias -- 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
