]> git.karo-electronics.de Git - karo-tx-linux.git/commitdiff
xen/blkback: pseudo support for multi hardware queues/rings
authorKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Fri, 11 Dec 2015 17:08:48 +0000 (12:08 -0500)
committerKonrad Rzeszutek Wilk <konrad.wilk@oracle.com>
Mon, 4 Jan 2016 17:21:05 +0000 (12:21 -0500)
Preparatory patch for multiple hardware queues (rings). The number of
rings is unconditionally set to 1, larger number will be enabled in
"xen/blkback: get the number of hardware queues/rings from blkfront".

Signed-off-by: Arianna Avanzini <avanzini.arianna@gmail.com>
Signed-off-by: Bob Liu <bob.liu@oracle.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
---
v2: Align variables in the structures.

drivers/block/xen-blkback/common.h
drivers/block/xen-blkback/xenbus.c

index dbdf4164c83f06182a953e273d77b34d0ba3a8db..310eff3cf43f8bb0c6230fe83fed6c74d89499ef 100644 (file)
@@ -340,7 +340,8 @@ struct xen_blkif {
        struct work_struct      free_work;
        unsigned int            nr_ring_pages;
        /* All rings for this device. */
-       struct xen_blkif_ring   ring;
+       struct xen_blkif_ring   *rings;
+       unsigned int            nr_rings;
 };
 
 struct seg_buf {
index e4bfc928035dbe02c847c29679afb46d0f4467a2..f5bfedd0e948eacc4b2cfe423b8e93dc5e8151b1 100644 (file)
@@ -86,9 +86,11 @@ static void xen_update_blkif_status(struct xen_blkif *blkif)
 {
        int err;
        char name[BLKBACK_NAME_LEN];
+       struct xen_blkif_ring *ring;
+       int i;
 
        /* Not ready to connect? */
-       if (!blkif->ring.irq || !blkif->vbd.bdev)
+       if (!blkif->rings || !blkif->rings[0].irq || !blkif->vbd.bdev)
                return;
 
        /* Already connected? */
@@ -113,19 +115,55 @@ static void xen_update_blkif_status(struct xen_blkif *blkif)
        }
        invalidate_inode_pages2(blkif->vbd.bdev->bd_inode->i_mapping);
 
-       blkif->ring.xenblkd = kthread_run(xen_blkif_schedule, &blkif->ring, "%s", name);
-       if (IS_ERR(blkif->ring.xenblkd)) {
-               err = PTR_ERR(blkif->ring.xenblkd);
-               blkif->ring.xenblkd = NULL;
-               xenbus_dev_error(blkif->be->dev, err, "start xenblkd");
-               return;
+       for (i = 0; i < blkif->nr_rings; i++) {
+               ring = &blkif->rings[i];
+               ring->xenblkd = kthread_run(xen_blkif_schedule, ring, "%s-%d", name, i);
+               if (IS_ERR(ring->xenblkd)) {
+                       err = PTR_ERR(ring->xenblkd);
+                       ring->xenblkd = NULL;
+                       xenbus_dev_fatal(blkif->be->dev, err,
+                                       "start %s-%d xenblkd", name, i);
+                       goto out;
+               }
+       }
+       return;
+
+out:
+       while (--i >= 0) {
+               ring = &blkif->rings[i];
+               kthread_stop(ring->xenblkd);
        }
+       return;
+}
+
+static int xen_blkif_alloc_rings(struct xen_blkif *blkif)
+{
+       unsigned int r;
+
+       blkif->rings = kzalloc(blkif->nr_rings * sizeof(struct xen_blkif_ring), GFP_KERNEL);
+       if (!blkif->rings)
+               return -ENOMEM;
+
+       for (r = 0; r < blkif->nr_rings; r++) {
+               struct xen_blkif_ring *ring = &blkif->rings[r];
+
+               spin_lock_init(&ring->blk_ring_lock);
+               init_waitqueue_head(&ring->wq);
+               INIT_LIST_HEAD(&ring->pending_free);
+
+               spin_lock_init(&ring->pending_free_lock);
+               init_waitqueue_head(&ring->pending_free_wq);
+               init_waitqueue_head(&ring->shutdown_wq);
+               ring->blkif = blkif;
+               xen_blkif_get(blkif);
+       }
+
+       return 0;
 }
 
 static struct xen_blkif *xen_blkif_alloc(domid_t domid)
 {
        struct xen_blkif *blkif;
-       struct xen_blkif_ring *ring;
 
        BUILD_BUG_ON(MAX_INDIRECT_PAGES > BLKIF_MAX_INDIRECT_PAGES_PER_REQUEST);
 
@@ -143,15 +181,11 @@ static struct xen_blkif *xen_blkif_alloc(domid_t domid)
        blkif->st_print = jiffies;
        INIT_WORK(&blkif->persistent_purge_work, xen_blkbk_unmap_purged_grants);
 
-       ring = &blkif->ring;
-       ring->blkif = blkif;
-       spin_lock_init(&ring->blk_ring_lock);
-       init_waitqueue_head(&ring->wq);
-
-       INIT_LIST_HEAD(&ring->pending_free);
-       spin_lock_init(&ring->pending_free_lock);
-       init_waitqueue_head(&ring->pending_free_wq);
-       init_waitqueue_head(&ring->shutdown_wq);
+       blkif->nr_rings = 1;
+       if (xen_blkif_alloc_rings(blkif)) {
+               kmem_cache_free(xen_blkif_cachep, blkif);
+               return ERR_PTR(-ENOMEM);
+       }
 
        return blkif;
 }
@@ -216,50 +250,54 @@ static int xen_blkif_map(struct xen_blkif_ring *ring, grant_ref_t *gref,
 static int xen_blkif_disconnect(struct xen_blkif *blkif)
 {
        struct pending_req *req, *n;
-       int i = 0, j;
-       struct xen_blkif_ring *ring = &blkif->ring;
+       unsigned int j, r;
 
-       if (ring->xenblkd) {
-               kthread_stop(ring->xenblkd);
-               wake_up(&ring->shutdown_wq);
-               ring->xenblkd = NULL;
-       }
+       for (r = 0; r < blkif->nr_rings; r++) {
+               struct xen_blkif_ring *ring = &blkif->rings[r];
+               unsigned int i = 0;
 
-       /* The above kthread_stop() guarantees that at this point we
-        * don't have any discard_io or other_io requests. So, checking
-        * for inflight IO is enough.
-        */
-       if (atomic_read(&ring->inflight) > 0)
-               return -EBUSY;
+               if (ring->xenblkd) {
+                       kthread_stop(ring->xenblkd);
+                       wake_up(&ring->shutdown_wq);
+                       ring->xenblkd = NULL;
+               }
 
-       if (ring->irq) {
-               unbind_from_irqhandler(ring->irq, ring);
-               ring->irq = 0;
-       }
+               /* The above kthread_stop() guarantees that at this point we
+                * don't have any discard_io or other_io requests. So, checking
+                * for inflight IO is enough.
+                */
+               if (atomic_read(&ring->inflight) > 0)
+                       return -EBUSY;
 
-       if (ring->blk_rings.common.sring) {
-               xenbus_unmap_ring_vfree(blkif->be->dev, ring->blk_ring);
-               ring->blk_rings.common.sring = NULL;
-       }
+               if (ring->irq) {
+                       unbind_from_irqhandler(ring->irq, ring);
+                       ring->irq = 0;
+               }
 
-       /* Remove all persistent grants and the cache of ballooned pages. */
-       xen_blkbk_free_caches(ring);
+               if (ring->blk_rings.common.sring) {
+                       xenbus_unmap_ring_vfree(blkif->be->dev, ring->blk_ring);
+                       ring->blk_rings.common.sring = NULL;
+               }
 
-       /* Check that there is no request in use */
-       list_for_each_entry_safe(req, n, &ring->pending_free, free_list) {
-               list_del(&req->free_list);
+               /* Remove all persistent grants and the cache of ballooned pages. */
+               xen_blkbk_free_caches(ring);
 
-               for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++)
-                       kfree(req->segments[j]);
+               /* Check that there is no request in use */
+               list_for_each_entry_safe(req, n, &ring->pending_free, free_list) {
+                       list_del(&req->free_list);
 
-               for (j = 0; j < MAX_INDIRECT_PAGES; j++)
-                       kfree(req->indirect_pages[j]);
+                       for (j = 0; j < MAX_INDIRECT_SEGMENTS; j++)
+                               kfree(req->segments[j]);
 
-               kfree(req);
-               i++;
-       }
+                       for (j = 0; j < MAX_INDIRECT_PAGES; j++)
+                               kfree(req->indirect_pages[j]);
 
-       WARN_ON(i != (XEN_BLKIF_REQS_PER_PAGE * blkif->nr_ring_pages));
+                       kfree(req);
+                       i++;
+               }
+
+               WARN_ON(i != (XEN_BLKIF_REQS_PER_PAGE * blkif->nr_ring_pages));
+       }
        blkif->nr_ring_pages = 0;
 
        return 0;
@@ -279,6 +317,7 @@ static void xen_blkif_free(struct xen_blkif *blkif)
        BUG_ON(!list_empty(&blkif->free_pages));
        BUG_ON(!RB_EMPTY_ROOT(&blkif->persistent_gnts));
 
+       kfree(blkif->rings);
        kmem_cache_free(xen_blkif_cachep, blkif);
 }
 
@@ -427,6 +466,7 @@ static int xen_vbd_create(struct xen_blkif *blkif, blkif_vdev_t handle,
 static int xen_blkbk_remove(struct xenbus_device *dev)
 {
        struct backend_info *be = dev_get_drvdata(&dev->dev);
+       unsigned int i;
 
        pr_debug("%s %p %d\n", __func__, dev, dev->otherend_id);
 
@@ -443,7 +483,8 @@ static int xen_blkbk_remove(struct xenbus_device *dev)
 
        if (be->blkif) {
                xen_blkif_disconnect(be->blkif);
-               xen_blkif_put(be->blkif);
+               for (i = 0; i < be->blkif->nr_rings; i++)
+                       xen_blkif_put(be->blkif);
        }
 
        kfree(be->mode);
@@ -826,51 +867,43 @@ again:
        xenbus_transaction_end(xbt, 1);
 }
 
-
-static int connect_ring(struct backend_info *be)
+/*
+ * Each ring may have multi pages, depends on "ring-page-order".
+ */
+static int read_per_ring_refs(struct xen_blkif_ring *ring, const char *dir)
 {
-       struct xenbus_device *dev = be->dev;
        unsigned int ring_ref[XENBUS_MAX_RING_GRANTS];
-       unsigned int evtchn, nr_grefs, ring_page_order;
-       unsigned int pers_grants;
-       char protocol[64] = "";
        struct pending_req *req, *n;
        int err, i, j;
-       struct xen_blkif_ring *ring = &be->blkif->ring;
-
-       pr_debug("%s %s\n", __func__, dev->otherend);
+       struct xen_blkif *blkif = ring->blkif;
+       struct xenbus_device *dev = blkif->be->dev;
+       unsigned int ring_page_order, nr_grefs, evtchn;
 
-       err = xenbus_scanf(XBT_NIL, dev->otherend, "event-channel", "%u",
+       err = xenbus_scanf(XBT_NIL, dir, "event-channel", "%u",
                          &evtchn);
        if (err != 1) {
                err = -EINVAL;
-               xenbus_dev_fatal(dev, err, "reading %s/event-channel",
-                                dev->otherend);
+               xenbus_dev_fatal(dev, err, "reading %s/event-channel", dir);
                return err;
        }
-       pr_info("event-channel %u\n", evtchn);
 
        err = xenbus_scanf(XBT_NIL, dev->otherend, "ring-page-order", "%u",
                          &ring_page_order);
        if (err != 1) {
-               err = xenbus_scanf(XBT_NIL, dev->otherend, "ring-ref",
-                                 "%u", &ring_ref[0]);
+               err = xenbus_scanf(XBT_NIL, dir, "ring-ref", "%u", &ring_ref[0]);
                if (err != 1) {
                        err = -EINVAL;
-                       xenbus_dev_fatal(dev, err, "reading %s/ring-ref",
-                                        dev->otherend);
+                       xenbus_dev_fatal(dev, err, "reading %s/ring-ref", dir);
                        return err;
                }
                nr_grefs = 1;
-               pr_info("%s:using single page: ring-ref %d\n", dev->otherend,
-                       ring_ref[0]);
        } else {
                unsigned int i;
 
                if (ring_page_order > xen_blkif_max_ring_order) {
                        err = -EINVAL;
                        xenbus_dev_fatal(dev, err, "%s/request %d ring page order exceed max:%d",
-                                        dev->otherend, ring_page_order,
+                                        dir, ring_page_order,
                                         xen_blkif_max_ring_order);
                        return err;
                }
@@ -880,46 +913,17 @@ static int connect_ring(struct backend_info *be)
                        char ring_ref_name[RINGREF_NAME_LEN];
 
                        snprintf(ring_ref_name, RINGREF_NAME_LEN, "ring-ref%u", i);
-                       err = xenbus_scanf(XBT_NIL, dev->otherend, ring_ref_name,
+                       err = xenbus_scanf(XBT_NIL, dir, ring_ref_name,
                                           "%u", &ring_ref[i]);
                        if (err != 1) {
                                err = -EINVAL;
                                xenbus_dev_fatal(dev, err, "reading %s/%s",
-                                                dev->otherend, ring_ref_name);
+                                                dir, ring_ref_name);
                                return err;
                        }
-                       pr_info("ring-ref%u: %u\n", i, ring_ref[i]);
                }
        }
-
-       be->blkif->blk_protocol = BLKIF_PROTOCOL_DEFAULT;
-       err = xenbus_gather(XBT_NIL, dev->otherend, "protocol",
-                           "%63s", protocol, NULL);
-       if (err)
-               strcpy(protocol, "unspecified, assuming default");
-       else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_NATIVE))
-               be->blkif->blk_protocol = BLKIF_PROTOCOL_NATIVE;
-       else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_32))
-               be->blkif->blk_protocol = BLKIF_PROTOCOL_X86_32;
-       else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_64))
-               be->blkif->blk_protocol = BLKIF_PROTOCOL_X86_64;
-       else {
-               xenbus_dev_fatal(dev, err, "unknown fe protocol %s", protocol);
-               return -1;
-       }
-       err = xenbus_gather(XBT_NIL, dev->otherend,
-                           "feature-persistent", "%u",
-                           &pers_grants, NULL);
-       if (err)
-               pers_grants = 0;
-
-       be->blkif->vbd.feature_gnt_persistent = pers_grants;
-       be->blkif->vbd.overflow_max_grants = 0;
-       be->blkif->nr_ring_pages = nr_grefs;
-
-       pr_info("ring-pages:%d, event-channel %d, protocol %d (%s) %s\n",
-               nr_grefs, evtchn, be->blkif->blk_protocol, protocol,
-               pers_grants ? "persistent grants" : "");
+       blkif->nr_ring_pages = nr_grefs;
 
        for (i = 0; i < nr_grefs * XEN_BLKIF_REQS_PER_PAGE; i++) {
                req = kzalloc(sizeof(*req), GFP_KERNEL);
@@ -964,6 +968,71 @@ fail:
                kfree(req);
        }
        return -ENOMEM;
+
+}
+
+static int connect_ring(struct backend_info *be)
+{
+       struct xenbus_device *dev = be->dev;
+       unsigned int pers_grants;
+       char protocol[64] = "";
+       int err, i;
+       char *xspath;
+       size_t xspathsize;
+       const size_t xenstore_path_ext_size = 11; /* sufficient for "/queue-NNN" */
+
+       pr_debug("%s %s\n", __func__, dev->otherend);
+
+       be->blkif->blk_protocol = BLKIF_PROTOCOL_DEFAULT;
+       err = xenbus_gather(XBT_NIL, dev->otherend, "protocol",
+                           "%63s", protocol, NULL);
+       if (err)
+               strcpy(protocol, "unspecified, assuming default");
+       else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_NATIVE))
+               be->blkif->blk_protocol = BLKIF_PROTOCOL_NATIVE;
+       else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_32))
+               be->blkif->blk_protocol = BLKIF_PROTOCOL_X86_32;
+       else if (0 == strcmp(protocol, XEN_IO_PROTO_ABI_X86_64))
+               be->blkif->blk_protocol = BLKIF_PROTOCOL_X86_64;
+       else {
+               xenbus_dev_fatal(dev, err, "unknown fe protocol %s", protocol);
+               return -1;
+       }
+       err = xenbus_gather(XBT_NIL, dev->otherend,
+                           "feature-persistent", "%u",
+                           &pers_grants, NULL);
+       if (err)
+               pers_grants = 0;
+
+       be->blkif->vbd.feature_gnt_persistent = pers_grants;
+       be->blkif->vbd.overflow_max_grants = 0;
+
+       pr_info("%s: using %d queues, protocol %d (%s) %s\n", dev->nodename,
+                be->blkif->nr_rings, be->blkif->blk_protocol, protocol,
+                pers_grants ? "persistent grants" : "");
+
+       if (be->blkif->nr_rings == 1)
+               return read_per_ring_refs(&be->blkif->rings[0], dev->otherend);
+       else {
+               xspathsize = strlen(dev->otherend) + xenstore_path_ext_size;
+               xspath = kmalloc(xspathsize, GFP_KERNEL);
+               if (!xspath) {
+                       xenbus_dev_fatal(dev, -ENOMEM, "reading ring references");
+                       return -ENOMEM;
+               }
+
+               for (i = 0; i < be->blkif->nr_rings; i++) {
+                       memset(xspath, 0, xspathsize);
+                       snprintf(xspath, xspathsize, "%s/queue-%u", dev->otherend, i);
+                       err = read_per_ring_refs(&be->blkif->rings[i], xspath);
+                       if (err) {
+                               kfree(xspath);
+                               return err;
+                       }
+               }
+               kfree(xspath);
+       }
+       return 0;
 }
 
 static const struct xenbus_device_id xen_blkbk_ids[] = {