Commit 1b7f390e authored by Asias He's avatar Asias He Committed by Nicholas Bellinger

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: default avatarAsias He <asias@redhat.com>
Signed-off-by: default avatarNicholas Bellinger <nab@linux-iscsi.org>
parent 67e18cf9
...@@ -48,6 +48,7 @@ ...@@ -48,6 +48,7 @@
#include <linux/virtio_net.h> /* TODO vhost.h currently depends on this */ #include <linux/virtio_net.h> /* TODO vhost.h currently depends on this */
#include <linux/virtio_scsi.h> #include <linux/virtio_scsi.h>
#include <linux/llist.h> #include <linux/llist.h>
#include <linux/bitmap.h>
#include "vhost.c" #include "vhost.c"
#include "vhost.h" #include "vhost.h"
...@@ -59,7 +60,8 @@ enum { ...@@ -59,7 +60,8 @@ enum {
VHOST_SCSI_VQ_IO = 2, 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 { struct vhost_scsi {
/* Protected by vhost_scsi->dev.mutex */ /* Protected by vhost_scsi->dev.mutex */
...@@ -68,7 +70,7 @@ struct vhost_scsi { ...@@ -68,7 +70,7 @@ struct vhost_scsi {
bool vs_endpoint; bool vs_endpoint;
struct vhost_dev dev; 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 vhost_work vs_completion_work; /* cmd completion work item */
struct llist_head vs_completion_list; /* cmd completion queue */ struct llist_head vs_completion_list; /* cmd completion queue */
...@@ -366,12 +368,14 @@ static void vhost_scsi_complete_cmd_work(struct vhost_work *work) ...@@ -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, struct vhost_scsi *vs = container_of(work, struct vhost_scsi,
vs_completion_work); vs_completion_work);
DECLARE_BITMAP(signal, VHOST_SCSI_MAX_VQ);
struct virtio_scsi_cmd_resp v_rsp; struct virtio_scsi_cmd_resp v_rsp;
struct tcm_vhost_cmd *tv_cmd; struct tcm_vhost_cmd *tv_cmd;
struct llist_node *llnode; struct llist_node *llnode;
struct se_cmd *se_cmd; 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); llnode = llist_del_all(&vs->vs_completion_list);
while (llnode) { while (llnode) {
tv_cmd = llist_entry(llnode, struct tcm_vhost_cmd, tv_cmd = llist_entry(llnode, struct tcm_vhost_cmd,
...@@ -390,15 +394,20 @@ static void vhost_scsi_complete_cmd_work(struct vhost_work *work) ...@@ -390,15 +394,20 @@ static void vhost_scsi_complete_cmd_work(struct vhost_work *work)
memcpy(v_rsp.sense, tv_cmd->tvc_sense_buf, memcpy(v_rsp.sense, tv_cmd->tvc_sense_buf,
v_rsp.sense_len); v_rsp.sense_len);
ret = copy_to_user(tv_cmd->tvc_resp, &v_rsp, sizeof(v_rsp)); ret = copy_to_user(tv_cmd->tvc_resp, &v_rsp, sizeof(v_rsp));
if (likely(ret == 0)) if (likely(ret == 0)) {
vhost_add_used(&vs->vqs[2], tv_cmd->tvc_vq_desc, 0); vhost_add_used(tv_cmd->tvc_vq, tv_cmd->tvc_vq_desc, 0);
else vq = tv_cmd->tvc_vq - vs->vqs;
__set_bit(vq, signal);
} else
pr_err("Faulted on virtio_scsi_cmd_resp\n"); pr_err("Faulted on virtio_scsi_cmd_resp\n");
vhost_scsi_free_cmd(tv_cmd); 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( static struct tcm_vhost_cmd *vhost_scsi_allocate_cmd(
...@@ -561,9 +570,9 @@ static void tcm_vhost_submission_work(struct work_struct *work) ...@@ -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 virtio_scsi_cmd_req v_req;
struct tcm_vhost_tpg *tv_tpg; struct tcm_vhost_tpg *tv_tpg;
struct tcm_vhost_cmd *tv_cmd; struct tcm_vhost_cmd *tv_cmd;
...@@ -656,7 +665,7 @@ static void vhost_scsi_handle_vq(struct vhost_scsi *vs) ...@@ -656,7 +665,7 @@ static void vhost_scsi_handle_vq(struct vhost_scsi *vs)
ret = __copy_to_user(resp, &rsp, sizeof(rsp)); ret = __copy_to_user(resp, &rsp, sizeof(rsp));
if (!ret) if (!ret)
vhost_add_used_and_signal(&vs->dev, vhost_add_used_and_signal(&vs->dev,
&vs->vqs[2], head, 0); vq, head, 0);
else else
pr_err("Faulted on virtio_scsi_cmd_resp\n"); pr_err("Faulted on virtio_scsi_cmd_resp\n");
...@@ -678,6 +687,7 @@ static void vhost_scsi_handle_vq(struct vhost_scsi *vs) ...@@ -678,6 +687,7 @@ static void vhost_scsi_handle_vq(struct vhost_scsi *vs)
": %d\n", tv_cmd, exp_data_len, data_direction); ": %d\n", tv_cmd, exp_data_len, data_direction);
tv_cmd->tvc_vhost = vs; tv_cmd->tvc_vhost = vs;
tv_cmd->tvc_vq = vq;
if (unlikely(vq->iov[out].iov_len != if (unlikely(vq->iov[out].iov_len !=
sizeof(struct virtio_scsi_cmd_resp))) { sizeof(struct virtio_scsi_cmd_resp))) {
...@@ -758,7 +768,7 @@ static void vhost_scsi_handle_kick(struct vhost_work *work) ...@@ -758,7 +768,7 @@ static void vhost_scsi_handle_kick(struct vhost_work *work)
poll.work); poll.work);
struct vhost_scsi *vs = container_of(vq->dev, struct vhost_scsi, dev); 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 @@ static int vhost_scsi_clear_endpoint( ...@@ -879,7 +889,7 @@ static int vhost_scsi_clear_endpoint(
static int vhost_scsi_open(struct inode *inode, struct file *f) static int vhost_scsi_open(struct inode *inode, struct file *f)
{ {
struct vhost_scsi *s; struct vhost_scsi *s;
int r; int r, i;
s = kzalloc(sizeof(*s), GFP_KERNEL); s = kzalloc(sizeof(*s), GFP_KERNEL);
if (!s) if (!s)
...@@ -889,8 +899,9 @@ static int vhost_scsi_open(struct inode *inode, struct file *f) ...@@ -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_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_EVT].handle_kick = vhost_scsi_evt_handle_kick;
s->vqs[VHOST_SCSI_VQ_IO].handle_kick = vhost_scsi_handle_kick; for (i = VHOST_SCSI_VQ_IO; i < VHOST_SCSI_MAX_VQ; i++)
r = vhost_dev_init(&s->dev, s->vqs, 3); s->vqs[i].handle_kick = vhost_scsi_handle_kick;
r = vhost_dev_init(&s->dev, s->vqs, VHOST_SCSI_MAX_VQ);
if (r < 0) { if (r < 0) {
kfree(s); kfree(s);
return r; return r;
...@@ -922,9 +933,10 @@ static void vhost_scsi_flush_vq(struct vhost_scsi *vs, int index) ...@@ -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) static void vhost_scsi_flush(struct vhost_scsi *vs)
{ {
vhost_scsi_flush_vq(vs, VHOST_SCSI_VQ_CTL); int i;
vhost_scsi_flush_vq(vs, VHOST_SCSI_VQ_EVT);
vhost_scsi_flush_vq(vs, VHOST_SCSI_VQ_IO); 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) static int vhost_scsi_set_features(struct vhost_scsi *vs, u64 features)
......
...@@ -23,6 +23,8 @@ struct tcm_vhost_cmd { ...@@ -23,6 +23,8 @@ struct tcm_vhost_cmd {
struct virtio_scsi_cmd_resp __user *tvc_resp; struct virtio_scsi_cmd_resp __user *tvc_resp;
/* Pointer to vhost_scsi for our device */ /* Pointer to vhost_scsi for our device */
struct vhost_scsi *tvc_vhost; struct vhost_scsi *tvc_vhost;
/* Pointer to vhost_virtqueue for the cmd */
struct vhost_virtqueue *tvc_vq;
/* Pointer to vhost nexus memory */ /* Pointer to vhost nexus memory */
struct tcm_vhost_nexus *tvc_nexus; struct tcm_vhost_nexus *tvc_nexus;
/* The TCM I/O descriptor that is accessed via container_of() */ /* The TCM I/O descriptor that is accessed via container_of() */
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment