LKML Archive on lore.kernel.org
help / color / mirror / Atom feed
* [PATCH 1/2] xenbus_client: extend interface to suppurt multi-page ring
@ 2015-01-23 10:14 Bob Liu
  2015-01-23 10:14 ` [PATCH 2/2] drivers: xen/block: add multi-page ring support Bob Liu
  2015-02-06 10:57 ` [PATCH 1/2] xenbus_client: extend interface to suppurt multi-page ring Wei Liu
  0 siblings, 2 replies; 9+ messages in thread
From: Bob Liu @ 2015-01-23 10:14 UTC (permalink / raw)
  To: xen-devel
  Cc: david.vrabel, linux-kernel, roger.pau, konrad.wilk,
	boris.ostrovsky, wei.liu2, Bob Liu

From: Wei Liu <wei.liu2@citrix.com>

Originally Xen PV drivers only use single-page ring to pass along
information. This might limit the throughput between frontend and
backend.

The patch extends Xenbus driver to support multi-page ring, which in
general should improve throughput if ring is the bottleneck. Changes to
various frontend / backend to adapt to the new interface are also
included.

Affected Xen drivers:
* blkfront/back
* netfront/back
* pcifront/back
* tpmfront

The interface is documented, as before, in xenbus_client.c.

Change in V2:
* allow ring has arbitrary number of pages <= XENBUS_MAX_RING_PAGES

Change in V3:
* update function prototypes
* carefully deal with types of different sizes

Change in V4:
* use PAGE_KERNEL instead of PAGE_KERNEL_IO to avoid breakage on Arm

Change in V5:
* fix off-by-one error and other minor glitches spotted by Mathew Daley

Signed-off-by: Wei Liu <wei.liu2@citrix.com>
Signed-off-by: Bob Liu <bob.liu@oracle.com>
---
 drivers/block/xen-blkback/xenbus.c |    5 +-
 drivers/block/xen-blkfront.c       |    5 +-
 drivers/char/tpm/xen-tpmfront.c    |    5 +-
 drivers/net/xen-netback/netback.c  |    4 +-
 drivers/net/xen-netfront.c         |    9 +-
 drivers/pci/xen-pcifront.c         |    5 +-
 drivers/xen/xen-pciback/xenbus.c   |    2 +-
 drivers/xen/xenbus/xenbus_client.c |  390 ++++++++++++++++++++++++++----------
 include/xen/xenbus.h               |   20 +-
 9 files changed, 318 insertions(+), 127 deletions(-)

diff --git a/drivers/block/xen-blkback/xenbus.c b/drivers/block/xen-blkback/xenbus.c
index 630a489..cbd13c9 100644
--- a/drivers/block/xen-blkback/xenbus.c
+++ b/drivers/block/xen-blkback/xenbus.c
@@ -193,7 +193,7 @@ fail:
 	return ERR_PTR(-ENOMEM);
 }
 
-static int xen_blkif_map(struct xen_blkif *blkif, unsigned long shared_page,
+static int xen_blkif_map(struct xen_blkif *blkif, grant_ref_t gref,
 			 unsigned int evtchn)
 {
 	int err;
@@ -202,7 +202,8 @@ static int xen_blkif_map(struct xen_blkif *blkif, unsigned long shared_page,
 	if (blkif->irq)
 		return 0;
 
-	err = xenbus_map_ring_valloc(blkif->be->dev, shared_page, &blkif->blk_ring);
+	err = xenbus_map_ring_valloc(blkif->be->dev, &gref, 1,
+			&blkif->blk_ring);
 	if (err < 0)
 		return err;
 
diff --git a/drivers/block/xen-blkfront.c b/drivers/block/xen-blkfront.c
index 2236c6f..2fcf75d 100644
--- a/drivers/block/xen-blkfront.c
+++ b/drivers/block/xen-blkfront.c
@@ -1246,6 +1246,7 @@ static int setup_blkring(struct xenbus_device *dev,
 {
 	struct blkif_sring *sring;
 	int err;
+	grant_ref_t gref;
 
 	info->ring_ref = GRANT_INVALID_REF;
 
@@ -1257,13 +1258,13 @@ static int setup_blkring(struct xenbus_device *dev,
 	SHARED_RING_INIT(sring);
 	FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
 
-	err = xenbus_grant_ring(dev, virt_to_mfn(info->ring.sring));
+	err = xenbus_grant_ring(dev, info->ring.sring, 1, &gref);
 	if (err < 0) {
 		free_page((unsigned long)sring);
 		info->ring.sring = NULL;
 		goto fail;
 	}
-	info->ring_ref = err;
+	info->ring_ref = gref;
 
 	err = xenbus_alloc_evtchn(dev, &info->evtchn);
 	if (err)
diff --git a/drivers/char/tpm/xen-tpmfront.c b/drivers/char/tpm/xen-tpmfront.c
index 441b44e..d3df997 100644
--- a/drivers/char/tpm/xen-tpmfront.c
+++ b/drivers/char/tpm/xen-tpmfront.c
@@ -193,6 +193,7 @@ static int setup_ring(struct xenbus_device *dev, struct tpm_private *priv)
 	struct xenbus_transaction xbt;
 	const char *message = NULL;
 	int rv;
+	grant_ref_t gref;
 
 	priv->shr = (void *)__get_free_page(GFP_KERNEL|__GFP_ZERO);
 	if (!priv->shr) {
@@ -200,11 +201,11 @@ static int setup_ring(struct xenbus_device *dev, struct tpm_private *priv)
 		return -ENOMEM;
 	}
 
-	rv = xenbus_grant_ring(dev, virt_to_mfn(priv->shr));
+	rv = xenbus_grant_ring(dev, &priv->shr, 1, &gref);
 	if (rv < 0)
 		return rv;
 
-	priv->ring_ref = rv;
+	priv->ring_ref = gref;
 
 	rv = xenbus_alloc_evtchn(dev, &priv->evtchn);
 	if (rv)
diff --git a/drivers/net/xen-netback/netback.c b/drivers/net/xen-netback/netback.c
index 908e65e..8513764 100644
--- a/drivers/net/xen-netback/netback.c
+++ b/drivers/net/xen-netback/netback.c
@@ -1955,7 +1955,7 @@ int xenvif_map_frontend_rings(struct xenvif_queue *queue,
 	int err = -ENOMEM;
 
 	err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif),
-				     tx_ring_ref, &addr);
+				&tx_ring_ref, 1, &addr);
 	if (err)
 		goto err;
 
@@ -1963,7 +1963,7 @@ int xenvif_map_frontend_rings(struct xenvif_queue *queue,
 	BACK_RING_INIT(&queue->tx, txs, PAGE_SIZE);
 
 	err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif),
-				     rx_ring_ref, &addr);
+				&rx_ring_ref, 1, &addr);
 	if (err)
 		goto err;
 
diff --git a/drivers/net/xen-netfront.c b/drivers/net/xen-netfront.c
index 22bcb4e..b97d26a 100644
--- a/drivers/net/xen-netfront.c
+++ b/drivers/net/xen-netfront.c
@@ -1531,6 +1531,7 @@ static int setup_netfront(struct xenbus_device *dev,
 	struct xen_netif_tx_sring *txs;
 	struct xen_netif_rx_sring *rxs;
 	int err;
+	grant_ref_t gref;
 
 	queue->tx_ring_ref = GRANT_INVALID_REF;
 	queue->rx_ring_ref = GRANT_INVALID_REF;
@@ -1546,10 +1547,10 @@ static int setup_netfront(struct xenbus_device *dev,
 	SHARED_RING_INIT(txs);
 	FRONT_RING_INIT(&queue->tx, txs, PAGE_SIZE);
 
-	err = xenbus_grant_ring(dev, virt_to_mfn(txs));
+	err = xenbus_grant_ring(dev, txs, 1, &gref);
 	if (err < 0)
 		goto grant_tx_ring_fail;
-	queue->tx_ring_ref = err;
+	queue->tx_ring_ref = gref;
 
 	rxs = (struct xen_netif_rx_sring *)get_zeroed_page(GFP_NOIO | __GFP_HIGH);
 	if (!rxs) {
@@ -1560,10 +1561,10 @@ static int setup_netfront(struct xenbus_device *dev,
 	SHARED_RING_INIT(rxs);
 	FRONT_RING_INIT(&queue->rx, rxs, PAGE_SIZE);
 
-	err = xenbus_grant_ring(dev, virt_to_mfn(rxs));
+	err = xenbus_grant_ring(dev, rxs, 1, &gref);
 	if (err < 0)
 		goto grant_rx_ring_fail;
-	queue->rx_ring_ref = err;
+	queue->rx_ring_ref = gref;
 
 	if (feature_split_evtchn)
 		err = setup_netfront_split(queue);
diff --git a/drivers/pci/xen-pcifront.c b/drivers/pci/xen-pcifront.c
index b1ffebe..7cfd2db 100644
--- a/drivers/pci/xen-pcifront.c
+++ b/drivers/pci/xen-pcifront.c
@@ -777,12 +777,13 @@ static int pcifront_publish_info(struct pcifront_device *pdev)
 {
 	int err = 0;
 	struct xenbus_transaction trans;
+	grant_ref_t gref;
 
-	err = xenbus_grant_ring(pdev->xdev, virt_to_mfn(pdev->sh_info));
+	err = xenbus_grant_ring(pdev->xdev, pdev->sh_info, 1, &gref);
 	if (err < 0)
 		goto out;
 
-	pdev->gnt_ref = err;
+	pdev->gnt_ref = gref;
 
 	err = xenbus_alloc_evtchn(pdev->xdev, &pdev->evtchn);
 	if (err)
diff --git a/drivers/xen/xen-pciback/xenbus.c b/drivers/xen/xen-pciback/xenbus.c
index fe17c80..98bc345 100644
--- a/drivers/xen/xen-pciback/xenbus.c
+++ b/drivers/xen/xen-pciback/xenbus.c
@@ -113,7 +113,7 @@ static int xen_pcibk_do_attach(struct xen_pcibk_device *pdev, int gnt_ref,
 		"Attaching to frontend resources - gnt_ref=%d evtchn=%d\n",
 		gnt_ref, remote_evtchn);
 
-	err = xenbus_map_ring_valloc(pdev->xdev, gnt_ref, &vaddr);
+	err = xenbus_map_ring_valloc(pdev->xdev, &gnt_ref, 1, &vaddr);
 	if (err < 0) {
 		xenbus_dev_fatal(pdev->xdev, err,
 				"Error mapping other domain page in ours.");
diff --git a/drivers/xen/xenbus/xenbus_client.c b/drivers/xen/xenbus/xenbus_client.c
index ca74410..7ef161a 100644
--- a/drivers/xen/xenbus/xenbus_client.c
+++ b/drivers/xen/xenbus/xenbus_client.c
@@ -52,17 +52,24 @@
 struct xenbus_map_node {
 	struct list_head next;
 	union {
-		struct vm_struct *area; /* PV */
-		struct page *page;     /* HVM */
+		struct {
+			struct vm_struct *area;
+		} pv;
+		struct {
+			struct page *pages[XENBUS_MAX_RING_PAGES];
+			void *addr;
+		} hvm;
 	};
-	grant_handle_t handle;
+	grant_handle_t handles[XENBUS_MAX_RING_PAGES];
+	unsigned int   nr_handles;
 };
 
 static DEFINE_SPINLOCK(xenbus_valloc_lock);
 static LIST_HEAD(xenbus_valloc_pages);
 
 struct xenbus_ring_ops {
-	int (*map)(struct xenbus_device *dev, int gnt, void **vaddr);
+	int (*map)(struct xenbus_device *dev, grant_ref_t *gnt_refs,
+			unsigned int nr_grefs, void **vaddr);
 	int (*unmap)(struct xenbus_device *dev, void *vaddr);
 };
 
@@ -355,17 +362,39 @@ static void xenbus_switch_fatal(struct xenbus_device *dev, int depth, int err,
 /**
  * xenbus_grant_ring
  * @dev: xenbus device
- * @ring_mfn: mfn of ring to grant
-
- * Grant access to the given @ring_mfn to the peer of the given device.  Return
- * a grant reference on success, or -errno on error. On error, the device will
- * switch to XenbusStateClosing, and the error will be saved in the store.
+ * @vaddr: starting virtual address of the ring
+ * @nr_pages: number of pages to be granted
+ * @grefs: grant reference array to be filled in
+ *
+ * Grant access to the given @vaddr to the peer of the given device.
+ * Then fill in @grefs with grant references.  Return a grant reference on
+ * success, or -errno on error.  On error, the device will switch to
+ * XenbusStateClosing, and the error will be saved in the store.
  */
-int xenbus_grant_ring(struct xenbus_device *dev, unsigned long ring_mfn)
+int xenbus_grant_ring(struct xenbus_device *dev, void *vaddr,
+		unsigned int nr_pages, grant_ref_t *grefs)
 {
-	int err = gnttab_grant_foreign_access(dev->otherend_id, ring_mfn, 0);
-	if (err < 0)
-		xenbus_dev_fatal(dev, err, "granting access to ring page");
+	int err;
+	int i, j;
+
+	for (i = 0; i < nr_pages; i++) {
+		unsigned long addr = (unsigned long)vaddr +
+			(PAGE_SIZE * i);
+		err = gnttab_grant_foreign_access(dev->otherend_id,
+				virt_to_mfn(addr), 0);
+		if (err < 0) {
+			xenbus_dev_fatal(dev, err,
+					"granting access to ring page");
+			goto fail;
+		}
+		grefs[i] = err;
+	}
+
+	return 0;
+
+fail:
+	for (j = 0; j < i; j++)
+		gnttab_end_foreign_access_ref(grefs[j], 0);
 	return err;
 }
 EXPORT_SYMBOL_GPL(xenbus_grant_ring);
@@ -419,62 +448,129 @@ EXPORT_SYMBOL_GPL(xenbus_free_evtchn);
 /**
  * xenbus_map_ring_valloc
  * @dev: xenbus device
- * @gnt_ref: grant reference
+ * @gnt_refs: grant reference array
+ * @nr_grefs: number of grant references
  * @vaddr: pointer to address to be filled out by mapping
  *
- * Based on Rusty Russell's skeleton driver's map_page.
- * Map a page of memory into this domain from another domain's grant table.
- * xenbus_map_ring_valloc allocates a page of virtual address space, maps the
- * page to that address, and sets *vaddr to that address.
- * Returns 0 on success, and GNTST_* (see xen/include/interface/grant_table.h)
- * or -ENOMEM on error. If an error is returned, device will switch to
+ * Map @nr_grefs pages of memory into this domain from another
+ * domain's grant table.  xenbus_map_ring_valloc allocates @nr_grefs
+ * pages of virtual address space, maps the pages to that address, and
+ * sets *vaddr to that address.  Returns 0 on success, and GNTST_*
+ * (see xen/include/interface/grant_table.h) or -ENOMEM / -EINVAL on
+ * error. If an error is returned, device will switch to
  * XenbusStateClosing and the error message will be saved in XenStore.
  */
-int xenbus_map_ring_valloc(struct xenbus_device *dev, int gnt_ref, void **vaddr)
+int xenbus_map_ring_valloc(struct xenbus_device *dev, grant_ref_t *gnt_refs,
+		unsigned int nr_grefs, void **vaddr)
 {
-	return ring_ops->map(dev, gnt_ref, vaddr);
+	return ring_ops->map(dev, gnt_refs, nr_grefs, vaddr);
 }
 EXPORT_SYMBOL_GPL(xenbus_map_ring_valloc);
 
+/* N.B. sizeof(phys_addr_t) doesn't always equal to sizeof(unsigned
+ * long), e.g. 32-on-64.  Caller is responsible for preparing the
+ * right array to feed into this function */
+static int __xenbus_map_ring(struct xenbus_device *dev,
+		grant_ref_t *gnt_refs,
+		unsigned int nr_grefs,
+		grant_handle_t *handles,
+		phys_addr_t *addrs,
+		unsigned int flags,
+		bool *leaked)
+{
+	struct gnttab_map_grant_ref map[XENBUS_MAX_RING_PAGES];
+	struct gnttab_unmap_grant_ref unmap[XENBUS_MAX_RING_PAGES];
+	int i, j;
+	int err = GNTST_okay;
+
+	if (nr_grefs > XENBUS_MAX_RING_PAGES)
+		return -EINVAL;
+
+	for (i = 0; i < nr_grefs; i++) {
+		memset(&map[i], 0, sizeof(map[i]));
+		gnttab_set_map_op(&map[i], addrs[i], flags, gnt_refs[i],
+				dev->otherend_id);
+		handles[i] = INVALID_GRANT_HANDLE;
+	}
+
+	gnttab_batch_map(map, i);
+
+	for (i = 0; i < nr_grefs; i++) {
+		if (map[i].status != GNTST_okay) {
+			err = map[i].status;
+			xenbus_dev_fatal(dev, map[i].status,
+					"mapping in shared page %d from domain %d",
+					gnt_refs[i], dev->otherend_id);
+			goto fail;
+		} else
+			handles[i] = map[i].handle;
+	}
+
+	return GNTST_okay;
+
+fail:
+	for (i = j = 0; i < nr_grefs; i++) {
+		if (handles[i] != INVALID_GRANT_HANDLE) {
+			memset(&unmap[j], 0, sizeof(unmap[j]));
+			gnttab_set_unmap_op(&unmap[j], (phys_addr_t)addrs[i],
+					GNTMAP_host_map, handles[i]);
+			j++;
+		}
+	}
+
+	if (HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref, unmap, j))
+		BUG();
+
+	*leaked = false;
+	for (i = 0; i < j; i++) {
+		if (unmap[i].status != GNTST_okay) {
+			*leaked = true;
+			break;
+		}
+	}
+
+	return err;
+}
+
 static int xenbus_map_ring_valloc_pv(struct xenbus_device *dev,
-				     int gnt_ref, void **vaddr)
+		grant_ref_t *gnt_refs, unsigned int nr_grefs, void **vaddr)
 {
-	struct gnttab_map_grant_ref op = {
-		.flags = GNTMAP_host_map | GNTMAP_contains_pte,
-		.ref   = gnt_ref,
-		.dom   = dev->otherend_id,
-	};
 	struct xenbus_map_node *node;
 	struct vm_struct *area;
-	pte_t *pte;
+	pte_t *ptes[XENBUS_MAX_RING_PAGES];
+	phys_addr_t phys_addrs[XENBUS_MAX_RING_PAGES];
+	int err = GNTST_okay;
+	int i;
+	bool leaked;
 
 	*vaddr = NULL;
 
+	if (nr_grefs > XENBUS_MAX_RING_PAGES)
+		return -EINVAL;
+
+
 	node = kzalloc(sizeof(*node), GFP_KERNEL);
 	if (!node)
 		return -ENOMEM;
 
-	area = alloc_vm_area(PAGE_SIZE, &pte);
+	area = alloc_vm_area(PAGE_SIZE * nr_grefs, ptes);
 	if (!area) {
 		kfree(node);
 		return -ENOMEM;
 	}
 
-	op.host_addr = arbitrary_virt_to_machine(pte).maddr;
-
-	gnttab_batch_map(&op, 1);
+	for (i = 0; i < nr_grefs; i++)
+		phys_addrs[i] = arbitrary_virt_to_machine(ptes[i]).maddr;
 
-	if (op.status != GNTST_okay) {
-		free_vm_area(area);
-		kfree(node);
-		xenbus_dev_fatal(dev, op.status,
-				 "mapping in shared page %d from domain %d",
-				 gnt_ref, dev->otherend_id);
-		return op.status;
-	}
+	err = __xenbus_map_ring(dev, gnt_refs, nr_grefs, node->handles,
+			phys_addrs,
+			GNTMAP_host_map | GNTMAP_contains_pte,
+			&leaked);
+	if (err)
+		goto failed;
 
-	node->handle = op.handle;
-	node->area = area;
+	node->nr_handles = nr_grefs;
+	node->pv.area = area;
 
 	spin_lock(&xenbus_valloc_lock);
 	list_add(&node->next, &xenbus_valloc_pages);
@@ -482,14 +578,30 @@ static int xenbus_map_ring_valloc_pv(struct xenbus_device *dev,
 
 	*vaddr = area->addr;
 	return 0;
+
+failed:
+	if (!leaked)
+		free_vm_area(area);
+	else
+		pr_alert("leaking VM area %p size %u page(s)", area, nr_grefs);
+
+	kfree(node);
+	return err;
 }
 
 static int xenbus_map_ring_valloc_hvm(struct xenbus_device *dev,
-				      int gnt_ref, void **vaddr)
+		grant_ref_t *gnt_ref, unsigned int nr_grefs, void **vaddr)
 {
 	struct xenbus_map_node *node;
-	int err;
+	int err, i;
 	void *addr;
+	bool leaked = false;
+	/* Why do we need two arrays? See comment of __xenbus_map_ring */
+	phys_addr_t phys_addrs[XENBUS_MAX_RING_PAGES];
+	unsigned long addrs[XENBUS_MAX_RING_PAGES];
+
+	if (nr_grefs > XENBUS_MAX_RING_PAGES)
+		return -EINVAL;
 
 	*vaddr = NULL;
 
@@ -497,15 +609,30 @@ static int xenbus_map_ring_valloc_hvm(struct xenbus_device *dev,
 	if (!node)
 		return -ENOMEM;
 
-	err = alloc_xenballooned_pages(1, &node->page, false /* lowmem */);
+	err = alloc_xenballooned_pages(nr_grefs, node->hvm.pages, false /* lowmem */);
 	if (err)
 		goto out_err;
 
-	addr = pfn_to_kaddr(page_to_pfn(node->page));
+	for (i = 0; i < nr_grefs; i++) {
+		unsigned long pfn = page_to_pfn(node->hvm.pages[i]);
+		phys_addrs[i] = (unsigned long)pfn_to_kaddr(pfn);
+		addrs[i] = (unsigned long)pfn_to_kaddr(pfn);
+	}
 
-	err = xenbus_map_ring(dev, gnt_ref, &node->handle, addr);
+	err = __xenbus_map_ring(dev, gnt_ref, nr_grefs, node->handles,
+			phys_addrs, GNTMAP_host_map, &leaked);
+	node->nr_handles = nr_grefs;
 	if (err)
-		goto out_err_free_ballooned_pages;
+		goto out_free_ballooned_pages;
+
+	addr = vmap(node->hvm.pages, nr_grefs, VM_MAP | VM_IOREMAP,
+			PAGE_KERNEL);
+	if (!addr) {
+		err = -ENOMEM;
+		goto out_xenbus_unmap_ring;
+	}
+
+	node->hvm.addr = addr;
 
 	spin_lock(&xenbus_valloc_lock);
 	list_add(&node->next, &xenbus_valloc_pages);
@@ -514,9 +641,17 @@ static int xenbus_map_ring_valloc_hvm(struct xenbus_device *dev,
 	*vaddr = addr;
 	return 0;
 
- out_err_free_ballooned_pages:
-	free_xenballooned_pages(1, &node->page);
- out_err:
+out_xenbus_unmap_ring:
+	if (!leaked)
+		xenbus_unmap_ring(dev, node->handles, node->nr_handles,
+				addrs);
+	else
+		pr_alert("leaking %p size %u page(s)",
+				addr, nr_grefs);
+out_free_ballooned_pages:
+	if (!leaked)
+		free_xenballooned_pages(nr_grefs, node->hvm.pages);
+out_err:
 	kfree(node);
 	return err;
 }
@@ -525,35 +660,37 @@ static int xenbus_map_ring_valloc_hvm(struct xenbus_device *dev,
 /**
  * xenbus_map_ring
  * @dev: xenbus device
- * @gnt_ref: grant reference
- * @handle: pointer to grant handle to be filled
- * @vaddr: address to be mapped to
+ * @gnt_refs: grant reference array
+ * @nr_grefs: number of grant reference
+ * @handles: pointer to grant handle to be filled
+ * @vaddrs: addresses to be mapped to
+ * @leaked: fail to clean up a failed map, caller should not free vaddr
  *
- * Map a page of memory into this domain from another domain's grant table.
+ * Map pages of memory into this domain from another domain's grant table.
  * xenbus_map_ring does not allocate the virtual address space (you must do
- * this yourself!). It only maps in the page to the specified address.
+ * this yourself!). It only maps in the pages to the specified address.
  * Returns 0 on success, and GNTST_* (see xen/include/interface/grant_table.h)
- * or -ENOMEM on error. If an error is returned, device will switch to
- * XenbusStateClosing and the error message will be saved in XenStore.
+ * or -ENOMEM / -EINVAL on error. If an error is returned, device will switch to
+ * XenbusStateClosing and the first error message will be saved in XenStore.
+ * Further more if we fail to map the ring, caller should check @leaked.
+ * If @leaked is not zero it means xenbus_map_ring fails to clean up, caller
+ * should not free the address space of @vaddr.
  */
-int xenbus_map_ring(struct xenbus_device *dev, int gnt_ref,
-		    grant_handle_t *handle, void *vaddr)
+int xenbus_map_ring(struct xenbus_device *dev, grant_ref_t *gnt_refs,
+		unsigned int nr_grefs, grant_handle_t *handles,
+		unsigned long *vaddrs, bool *leaked)
 {
-	struct gnttab_map_grant_ref op;
-
-	gnttab_set_map_op(&op, (unsigned long)vaddr, GNTMAP_host_map, gnt_ref,
-			  dev->otherend_id);
+	phys_addr_t phys_addrs[XENBUS_MAX_RING_PAGES];
+	int i;
 
-	gnttab_batch_map(&op, 1);
+	if (nr_grefs > XENBUS_MAX_RING_PAGES)
+		return -EINVAL;
 
-	if (op.status != GNTST_okay) {
-		xenbus_dev_fatal(dev, op.status,
-				 "mapping in shared page %d from domain %d",
-				 gnt_ref, dev->otherend_id);
-	} else
-		*handle = op.handle;
+	for (i = 0; i < nr_grefs; i++)
+		phys_addrs[i] = (unsigned long)vaddrs[i];
 
-	return op.status;
+	return __xenbus_map_ring(dev, gnt_refs, nr_grefs, handles,
+			phys_addrs, GNTMAP_host_map, leaked);
 }
 EXPORT_SYMBOL_GPL(xenbus_map_ring);
 
@@ -563,8 +700,7 @@ EXPORT_SYMBOL_GPL(xenbus_map_ring);
  * @dev: xenbus device
  * @vaddr: addr to unmap
  *
- * Based on Rusty Russell's skeleton driver's unmap_page.
- * Unmap a page of memory in this domain that was imported from another domain.
+ * Unmap memory in this domain that was imported from another domain.
  * Use xenbus_unmap_ring_vfree if you mapped in your memory with
  * xenbus_map_ring_valloc (it will free the virtual address space).
  * Returns 0 on success and returns GNTST_* on error
@@ -579,20 +715,21 @@ EXPORT_SYMBOL_GPL(xenbus_unmap_ring_vfree);
 static int xenbus_unmap_ring_vfree_pv(struct xenbus_device *dev, void *vaddr)
 {
 	struct xenbus_map_node *node;
-	struct gnttab_unmap_grant_ref op = {
-		.host_addr = (unsigned long)vaddr,
-	};
+	struct gnttab_unmap_grant_ref unmap[XENBUS_MAX_RING_PAGES];
 	unsigned int level;
+	int i;
+	bool leaked = false;
+	int err;
 
 	spin_lock(&xenbus_valloc_lock);
 	list_for_each_entry(node, &xenbus_valloc_pages, next) {
-		if (node->area->addr == vaddr) {
+		if (node->pv.area->addr == vaddr) {
 			list_del(&node->next);
 			goto found;
 		}
 	}
 	node = NULL;
- found:
+found:
 	spin_unlock(&xenbus_valloc_lock);
 
 	if (!node) {
@@ -601,22 +738,42 @@ static int xenbus_unmap_ring_vfree_pv(struct xenbus_device *dev, void *vaddr)
 		return GNTST_bad_virt_addr;
 	}
 
-	op.handle = node->handle;
-	op.host_addr = arbitrary_virt_to_machine(
-		lookup_address((unsigned long)vaddr, &level)).maddr;
+	for (i = 0; i < node->nr_handles; i++) {
+		unsigned long addr;
 
-	if (HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref, &op, 1))
+		memset(&unmap[i], 0, sizeof(unmap[i]));
+		addr = (unsigned long)vaddr + (PAGE_SIZE * i);
+		unmap[i].host_addr = arbitrary_virt_to_machine(
+				lookup_address(addr, &level)).maddr;
+		unmap[i].dev_bus_addr = 0;
+		unmap[i].handle = node->handles[i];
+	}
+
+	if (HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref, unmap, i))
 		BUG();
 
-	if (op.status == GNTST_okay)
-		free_vm_area(node->area);
+	err = GNTST_okay;
+	leaked = false;
+	for (i = 0; i < node->nr_handles; i++) {
+		if (unmap[i].status != GNTST_okay) {
+			leaked = true;
+			xenbus_dev_error(dev, unmap[i].status,
+					"unmapping page at handle %d error %d",
+					node->handles[i], unmap[i].status);
+			err = unmap[i].status;
+			break;
+		}
+	}
+
+	if (!leaked)
+		free_vm_area(node->pv.area);
+
 	else
-		xenbus_dev_error(dev, op.status,
-				 "unmapping page at handle %d error %d",
-				 node->handle, op.status);
+		pr_alert("leaking VM area %p size %u page(s)",
+				node->pv.area, node->nr_handles);
 
 	kfree(node);
-	return op.status;
+	return err;
 }
 
 static int xenbus_unmap_ring_vfree_hvm(struct xenbus_device *dev, void *vaddr)
@@ -624,10 +781,12 @@ static int xenbus_unmap_ring_vfree_hvm(struct xenbus_device *dev, void *vaddr)
 	int rv;
 	struct xenbus_map_node *node;
 	void *addr;
+	unsigned long addrs[XENBUS_MAX_RING_PAGES];
+	int i;
 
 	spin_lock(&xenbus_valloc_lock);
 	list_for_each_entry(node, &xenbus_valloc_pages, next) {
-		addr = pfn_to_kaddr(page_to_pfn(node->page));
+		addr = node->hvm.addr;
 		if (addr == vaddr) {
 			list_del(&node->next);
 			goto found;
@@ -643,12 +802,16 @@ static int xenbus_unmap_ring_vfree_hvm(struct xenbus_device *dev, void *vaddr)
 		return GNTST_bad_virt_addr;
 	}
 
-	rv = xenbus_unmap_ring(dev, node->handle, addr);
+	for (i = 0; i < node->nr_handles; i++)
+		addrs[i] = (unsigned long)pfn_to_kaddr(page_to_pfn(node->hvm.pages[i]));
 
+	rv = xenbus_unmap_ring(dev, node->handles, node->nr_handles,
+			addrs);
 	if (!rv)
-		free_xenballooned_pages(1, &node->page);
+		vunmap(vaddr);
 	else
-		WARN(1, "Leaking %p\n", vaddr);
+		WARN(1, "Leaking %p, size %u page(s)\n", vaddr,
+				node->nr_handles);
 
 	kfree(node);
 	return rv;
@@ -657,29 +820,44 @@ static int xenbus_unmap_ring_vfree_hvm(struct xenbus_device *dev, void *vaddr)
 /**
  * xenbus_unmap_ring
  * @dev: xenbus device
- * @handle: grant handle
- * @vaddr: addr to unmap
+ * @handles: grant handle array
+ * @nr_handles: number of handles in the array
+ * @vaddrs: addresses to unmap
  *
- * Unmap a page of memory in this domain that was imported from another domain.
+ * Unmap memory in this domain that was imported from another domain.
  * Returns 0 on success and returns GNTST_* on error
  * (see xen/include/interface/grant_table.h).
  */
 int xenbus_unmap_ring(struct xenbus_device *dev,
-		      grant_handle_t handle, void *vaddr)
+		grant_handle_t *handles, unsigned int nr_handles,
+		unsigned long *vaddrs)
 {
-	struct gnttab_unmap_grant_ref op;
+	struct gnttab_unmap_grant_ref unmap[XENBUS_MAX_RING_PAGES];
+	int i;
+	int err;
 
-	gnttab_set_unmap_op(&op, (unsigned long)vaddr, GNTMAP_host_map, handle);
+	if (nr_handles > XENBUS_MAX_RING_PAGES)
+		return -EINVAL;
 
-	if (HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref, &op, 1))
+	for (i = 0; i < nr_handles; i++)
+		gnttab_set_unmap_op(&unmap[i], vaddrs[i],
+				GNTMAP_host_map, handles[i]);
+
+	if (HYPERVISOR_grant_table_op(GNTTABOP_unmap_grant_ref, unmap, i))
 		BUG();
 
-	if (op.status != GNTST_okay)
-		xenbus_dev_error(dev, op.status,
-				 "unmapping page at handle %d error %d",
-				 handle, op.status);
+	err = GNTST_okay;
+	for (i = 0; i < nr_handles; i++) {
+		if (unmap[i].status != GNTST_okay) {
+			xenbus_dev_error(dev, unmap[i].status,
+					"unmapping page at handle %d error %d",
+					handles[i], unmap[i].status);
+			err = unmap[i].status;
+			break;
+		}
+	}
 
-	return op.status;
+	return err;
 }
 EXPORT_SYMBOL_GPL(xenbus_unmap_ring);
 
diff --git a/include/xen/xenbus.h b/include/xen/xenbus.h
index b78f21c..84cd1e4 100644
--- a/include/xen/xenbus.h
+++ b/include/xen/xenbus.h
@@ -46,6 +46,10 @@
 #include <xen/interface/io/xenbus.h>
 #include <xen/interface/io/xs_wire.h>
 
+#define XENBUS_MAX_RING_PAGE_ORDER 4
+#define XENBUS_MAX_RING_PAGES      (1U << XENBUS_MAX_RING_PAGE_ORDER)
+#define INVALID_GRANT_HANDLE       (~0U)
+
 /* Register callback to watch this node. */
 struct xenbus_watch
 {
@@ -199,15 +203,19 @@ int xenbus_watch_pathfmt(struct xenbus_device *dev, struct xenbus_watch *watch,
 			 const char *pathfmt, ...);
 
 int xenbus_switch_state(struct xenbus_device *dev, enum xenbus_state new_state);
-int xenbus_grant_ring(struct xenbus_device *dev, unsigned long ring_mfn);
-int xenbus_map_ring_valloc(struct xenbus_device *dev,
-			   int gnt_ref, void **vaddr);
-int xenbus_map_ring(struct xenbus_device *dev, int gnt_ref,
-			   grant_handle_t *handle, void *vaddr);
+int xenbus_grant_ring(struct xenbus_device *dev, void *vaddr,
+		unsigned int nr_pages, grant_ref_t *grefs);
+int xenbus_map_ring_valloc(struct xenbus_device *dev, grant_ref_t *gnt_refs,
+		unsigned int nr_grefs, void **vaddr);
+int xenbus_map_ring(struct xenbus_device *dev,
+		grant_ref_t *gnt_refs, unsigned int nr_grefs,
+		grant_handle_t *handles, unsigned long *vaddrs,
+		bool *leaked);
 
 int xenbus_unmap_ring_vfree(struct xenbus_device *dev, void *vaddr);
 int xenbus_unmap_ring(struct xenbus_device *dev,
-		      grant_handle_t handle, void *vaddr);
+		grant_handle_t *handles, unsigned int nr_handles,
+		unsigned long *vaddrs);
 
 int xenbus_alloc_evtchn(struct xenbus_device *dev, int *port);
 int xenbus_free_evtchn(struct xenbus_device *dev, int port);
-- 
1.7.10.4


^ permalink raw reply	[flat|nested] 9+ messages in thread

* [PATCH 2/2] drivers: xen/block: add multi-page ring support
  2015-01-23 10:14 [PATCH 1/2] xenbus_client: extend interface to suppurt multi-page ring Bob Liu
@ 2015-01-23 10:14 ` Bob Liu
  2015-02-02 10:43   ` Roger Pau Monné
  2015-02-05 17:41   ` [Xen-devel] " Felipe Franciosi
  2015-02-06 10:57 ` [PATCH 1/2] xenbus_client: extend interface to suppurt multi-page ring Wei Liu
  1 sibling, 2 replies; 9+ messages in thread
From: Bob Liu @ 2015-01-23 10:14 UTC (permalink / raw)
  To: xen-devel
  Cc: david.vrabel, linux-kernel, roger.pau, konrad.wilk,
	boris.ostrovsky, wei.liu2, Bob Liu

Extend xen/block to support multi-page ring.
 * xen-blkback notify blkfront with feature-multi-ring-pages
 * xen-blkfront write to xenstore about how many pages are used as the ring

If using 4 pages as the ring, inflight requests inscreased from 32 to 128 and
IOPS improved nearly 400% on our system.

Signed-off-by: Bob Liu <bob.liu@oracle.com>
---
 drivers/block/xen-blkback/xenbus.c |   86 +++++++++++++++++++++++++--------
 drivers/block/xen-blkfront.c       |   94 ++++++++++++++++++++++++++----------
 2 files changed, 133 insertions(+), 47 deletions(-)

diff --git a/drivers/block/xen-blkback/xenbus.c b/drivers/block/xen-blkback/xenbus.c
index cbd13c9..a5c9c62 100644
--- a/drivers/block/xen-blkback/xenbus.c
+++ b/drivers/block/xen-blkback/xenbus.c
@@ -193,8 +193,8 @@ fail:
 	return ERR_PTR(-ENOMEM);
 }
 
-static int xen_blkif_map(struct xen_blkif *blkif, grant_ref_t gref,
-			 unsigned int evtchn)
+static int xen_blkif_map(struct xen_blkif *blkif, grant_ref_t *gref,
+			unsigned int nr_grefs, unsigned int evtchn)
 {
 	int err;
 
@@ -202,7 +202,7 @@ static int xen_blkif_map(struct xen_blkif *blkif, grant_ref_t gref,
 	if (blkif->irq)
 		return 0;
 
-	err = xenbus_map_ring_valloc(blkif->be->dev, &gref, 1,
+	err = xenbus_map_ring_valloc(blkif->be->dev, gref, nr_grefs,
 			&blkif->blk_ring);
 	if (err < 0)
 		return err;
@@ -212,21 +212,21 @@ static int xen_blkif_map(struct xen_blkif *blkif, grant_ref_t gref,
 	{
 		struct blkif_sring *sring;
 		sring = (struct blkif_sring *)blkif->blk_ring;
-		BACK_RING_INIT(&blkif->blk_rings.native, sring, PAGE_SIZE);
+		BACK_RING_INIT(&blkif->blk_rings.native, sring, PAGE_SIZE * nr_grefs);
 		break;
 	}
 	case BLKIF_PROTOCOL_X86_32:
 	{
 		struct blkif_x86_32_sring *sring_x86_32;
 		sring_x86_32 = (struct blkif_x86_32_sring *)blkif->blk_ring;
-		BACK_RING_INIT(&blkif->blk_rings.x86_32, sring_x86_32, PAGE_SIZE);
+		BACK_RING_INIT(&blkif->blk_rings.x86_32, sring_x86_32, PAGE_SIZE * nr_grefs);
 		break;
 	}
 	case BLKIF_PROTOCOL_X86_64:
 	{
 		struct blkif_x86_64_sring *sring_x86_64;
 		sring_x86_64 = (struct blkif_x86_64_sring *)blkif->blk_ring;
-		BACK_RING_INIT(&blkif->blk_rings.x86_64, sring_x86_64, PAGE_SIZE);
+		BACK_RING_INIT(&blkif->blk_rings.x86_64, sring_x86_64, PAGE_SIZE * nr_grefs);
 		break;
 	}
 	default:
@@ -588,6 +588,13 @@ static int xen_blkbk_probe(struct xenbus_device *dev,
 	if (err)
 		goto fail;
 
+	err = xenbus_printf(XBT_NIL, dev->nodename,
+			"feature-multi-ring-pages", "%u", 1);
+	if (err) {
+		pr_debug("Error writing feature-multi-ring-pages\n");
+		goto fail;
+	}
+
 	err = xenbus_switch_state(dev, XenbusStateInitWait);
 	if (err)
 		goto fail;
@@ -852,23 +859,61 @@ again:
 static int connect_ring(struct backend_info *be)
 {
 	struct xenbus_device *dev = be->dev;
-	unsigned long ring_ref;
-	unsigned int evtchn;
+	unsigned int ring_ref[XENBUS_MAX_RING_PAGES];
+	unsigned int evtchn, nr_grefs;
 	unsigned int pers_grants;
 	char protocol[64] = "";
 	int err;
 
 	DPRINTK("%s", dev->otherend);
-
-	err = xenbus_gather(XBT_NIL, dev->otherend, "ring-ref", "%lu",
-			    &ring_ref, "event-channel", "%u", &evtchn, NULL);
-	if (err) {
-		xenbus_dev_fatal(dev, err,
-				 "reading %s/ring-ref and event-channel",
-				 dev->otherend);
+	err = xenbus_scanf(XBT_NIL, dev->otherend, "event-channel", "%u",
+			&evtchn);
+	if (err != 1) {
+		err = -EINVAL;
+		xenbus_dev_fatal(dev, err, "reading %s/event-channel",
+				dev->otherend);
 		return err;
 	}
 
+	err = xenbus_scanf(XBT_NIL, dev->otherend, "max-ring-pages", "%u",
+			&nr_grefs);
+	if (err != 1) {
+		err = xenbus_scanf(XBT_NIL, dev->otherend, "ring-ref",
+				"%u", &ring_ref[0]);
+		if (err != 1) {
+			err = -EINVAL;
+			xenbus_dev_fatal(dev, err, "reading %s/ring-ref",
+					dev->otherend);
+			return err;
+		}
+		nr_grefs = 1;
+		pr_debug("%s:using one-page ring with ref: %d\n",
+				dev->otherend, ring_ref[0]);
+	} else {
+		unsigned int i;
+
+		if (nr_grefs > XENBUS_MAX_RING_PAGES) {
+			err = -EINVAL;
+			xenbus_dev_fatal(dev, err,
+				"%s/ring-pages:%d too big", dev->otherend, nr_grefs);
+			return err;
+		}
+		for (i = 0; i < nr_grefs; i++) {
+			char ring_ref_name[15];
+			snprintf(ring_ref_name, sizeof(ring_ref_name),
+					"ring-ref%u", i);
+			err = xenbus_scanf(XBT_NIL, dev->otherend,
+					ring_ref_name, "%d", &ring_ref[i]);
+			if (err != 1) {
+				err = -EINVAL;
+				xenbus_dev_fatal(dev, err, "reading %s/%s",
+						dev->otherend, ring_ref_name);
+				return err;
+			}
+			pr_debug("blkback: ring-ref%u: %d\n", i, ring_ref[i]);
+		}
+	}
+
 	be->blkif->blk_protocol = BLKIF_PROTOCOL_NATIVE;
 	err = xenbus_gather(XBT_NIL, dev->otherend, "protocol",
 			    "%63s", protocol, NULL);
@@ -893,15 +938,14 @@ static int connect_ring(struct backend_info *be)
 	be->blkif->vbd.feature_gnt_persistent = pers_grants;
 	be->blkif->vbd.overflow_max_grants = 0;
 
-	pr_info(DRV_PFX "ring-ref %ld, event-channel %d, protocol %d (%s) %s\n",
-		ring_ref, evtchn, be->blkif->blk_protocol, protocol,
-		pers_grants ? "persistent grants" : "");
+	pr_info(DRV_PFX "ring-pages:%d, event-channel %d, protocol %d (%s) %s\n",
+			nr_grefs, evtchn, be->blkif->blk_protocol, protocol,
+			pers_grants ? "persistent grants" : "");
 
 	/* Map the shared frame, irq etc. */
-	err = xen_blkif_map(be->blkif, ring_ref, evtchn);
+	err = xen_blkif_map(be->blkif, ring_ref, nr_grefs, evtchn);
 	if (err) {
-		xenbus_dev_fatal(dev, err, "mapping ring-ref %lu port %u",
-				 ring_ref, evtchn);
+		xenbus_dev_fatal(dev, err, "mapping ring-ref port %u", evtchn);
 		return err;
 	}
 
diff --git a/drivers/block/xen-blkfront.c b/drivers/block/xen-blkfront.c
index 2fcf75d..dec9fe7 100644
--- a/drivers/block/xen-blkfront.c
+++ b/drivers/block/xen-blkfront.c
@@ -98,7 +98,12 @@ static unsigned int xen_blkif_max_segments = 32;
 module_param_named(max, xen_blkif_max_segments, int, S_IRUGO);
 MODULE_PARM_DESC(max, "Maximum amount of segments in indirect requests (default is 32)");
 
-#define BLK_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE)
+static unsigned int xen_blkif_max_ring_pages = 1;
+module_param_named(max_ring_pages, xen_blkif_max_ring_pages, int, 0);
+MODULE_PARM_DESC(max_ring_pages, "Maximum amount of pages to be used as the ring");
+
+#define BLK_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE * info->nr_ring_pages)
+#define BLK_MAX_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE * XENBUS_MAX_RING_PAGES)
 
 /*
  * We have one of these per vbd, whether ide, scsi or 'other'.  They
@@ -114,13 +119,14 @@ struct blkfront_info
 	int vdevice;
 	blkif_vdev_t handle;
 	enum blkif_state connected;
-	int ring_ref;
+	int ring_ref[XENBUS_MAX_RING_PAGES];
+	int nr_ring_pages;
 	struct blkif_front_ring ring;
 	unsigned int evtchn, irq;
 	struct request_queue *rq;
 	struct work_struct work;
 	struct gnttab_free_callback callback;
-	struct blk_shadow shadow[BLK_RING_SIZE];
+	struct blk_shadow shadow[BLK_MAX_RING_SIZE];
 	struct list_head grants;
 	struct list_head indirect_pages;
 	unsigned int persistent_gnts_c;
@@ -139,8 +145,6 @@ static unsigned int nr_minors;
 static unsigned long *minors;
 static DEFINE_SPINLOCK(minor_lock);
 
-#define MAXIMUM_OUTSTANDING_BLOCK_REQS \
-	(BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
 #define GRANT_INVALID_REF	0
 
 #define PARTS_PER_DISK		16
@@ -1033,12 +1037,15 @@ free_shadow:
 	flush_work(&info->work);
 
 	/* Free resources associated with old device channel. */
-	if (info->ring_ref != GRANT_INVALID_REF) {
-		gnttab_end_foreign_access(info->ring_ref, 0,
-					  (unsigned long)info->ring.sring);
-		info->ring_ref = GRANT_INVALID_REF;
-		info->ring.sring = NULL;
+	for (i = 0; i < info->nr_ring_pages; i++) {
+		if (info->ring_ref[i] != GRANT_INVALID_REF) {
+			gnttab_end_foreign_access(info->ring_ref[i], 0, 0);
+			info->ring_ref[i] = GRANT_INVALID_REF;
+		}
 	}
+	free_pages((unsigned long)info->ring.sring, get_order(info->nr_ring_pages * PAGE_SIZE));
+	info->ring.sring = NULL;
+
 	if (info->irq)
 		unbind_from_irqhandler(info->irq, info);
 	info->evtchn = info->irq = 0;
@@ -1245,26 +1252,30 @@ static int setup_blkring(struct xenbus_device *dev,
 			 struct blkfront_info *info)
 {
 	struct blkif_sring *sring;
-	int err;
-	grant_ref_t gref;
+	int err, i;
+	unsigned long ring_size = info->nr_ring_pages * PAGE_SIZE;
+	grant_ref_t gref[XENBUS_MAX_RING_PAGES];
 
-	info->ring_ref = GRANT_INVALID_REF;
+	for (i = 0; i < XENBUS_MAX_RING_PAGES; i++)
+		info->ring_ref[i] = GRANT_INVALID_REF;
 
-	sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
+	sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
+						get_order(ring_size));
 	if (!sring) {
 		xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
 		return -ENOMEM;
 	}
 	SHARED_RING_INIT(sring);
-	FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
+	FRONT_RING_INIT(&info->ring, sring, ring_size);
 
-	err = xenbus_grant_ring(dev, info->ring.sring, 1, &gref);
+	err = xenbus_grant_ring(dev, info->ring.sring, info->nr_ring_pages, gref);
 	if (err < 0) {
 		free_page((unsigned long)sring);
 		info->ring.sring = NULL;
 		goto fail;
 	}
-	info->ring_ref = gref;
+	for (i = 0; i < info->nr_ring_pages; i++)
+		info->ring_ref[i] = gref[i];
 
 	err = xenbus_alloc_evtchn(dev, &info->evtchn);
 	if (err)
@@ -1292,7 +1303,14 @@ static int talk_to_blkback(struct xenbus_device *dev,
 {
 	const char *message = NULL;
 	struct xenbus_transaction xbt;
-	int err;
+	int err, i, multi_ring_pages = 0;
+
+	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
+			   "feature-multi-ring-pages", "%u", &multi_ring_pages);
+	if (err != 1)
+		info->nr_ring_pages = 1;
+	else
+		info->nr_ring_pages = xen_blkif_max_ring_pages;
 
 	/* Create shared ring, alloc event channel. */
 	err = setup_blkring(dev, info);
@@ -1306,12 +1324,33 @@ again:
 		goto destroy_blkring;
 	}
 
-	err = xenbus_printf(xbt, dev->nodename,
-			    "ring-ref", "%u", info->ring_ref);
-	if (err) {
-		message = "writing ring-ref";
-		goto abort_transaction;
+	if (info->nr_ring_pages == 1) {
+		err = xenbus_printf(xbt, dev->nodename,
+				"ring-ref", "%u", info->ring_ref[0]);
+		if (err) {
+			message = "writing ring-ref";
+			goto abort_transaction;
+		}
+	} else {
+		err = xenbus_printf(xbt, dev->nodename,
+				"max-ring-pages", "%u", info->nr_ring_pages);
+		if (err) {
+			message = "writing max-ring-pages";
+			goto abort_transaction;
+		}
+
+		for (i = 0; i < info->nr_ring_pages; i++) {
+			char ring_ref_name[15];
+			sprintf(ring_ref_name, "ring-ref%d", i);
+			err = xenbus_printf(xbt, dev->nodename,
+					ring_ref_name, "%d", info->ring_ref[i]);
+			if (err) {
+				message = "writing ring-ref";
+				goto abort_transaction;
+			}
+		}
 	}
+
 	err = xenbus_printf(xbt, dev->nodename,
 			    "event-channel", "%u", info->evtchn);
 	if (err) {
@@ -1422,9 +1461,8 @@ static int blkfront_probe(struct xenbus_device *dev,
 	info->connected = BLKIF_STATE_DISCONNECTED;
 	INIT_WORK(&info->work, blkif_restart_queue);
 
-	for (i = 0; i < BLK_RING_SIZE; i++)
+	for (i = 0; i < BLK_MAX_RING_SIZE; i++)
 		info->shadow[i].req.u.rw.id = i+1;
-	info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
 
 	/* Front end dir is a number, which is used as the id. */
 	info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0);
@@ -1436,6 +1474,7 @@ static int blkfront_probe(struct xenbus_device *dev,
 		dev_set_drvdata(&dev->dev, NULL);
 		return err;
 	}
+	info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
 
 	return 0;
 }
@@ -1476,7 +1515,7 @@ static int blkif_recover(struct blkfront_info *info)
 
 	/* Stage 2: Set up free list. */
 	memset(&info->shadow, 0, sizeof(info->shadow));
-	for (i = 0; i < BLK_RING_SIZE; i++)
+	for (i = 0; i < BLK_MAX_RING_SIZE; i++)
 		info->shadow[i].req.u.rw.id = i+1;
 	info->shadow_free = info->ring.req_prod_pvt;
 	info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
@@ -2088,6 +2127,9 @@ static int __init xlblk_init(void)
 {
 	int ret;
 
+	if (xen_blkif_max_ring_pages > XENBUS_MAX_RING_PAGES)
+		return -EINVAL;
+
 	if (!xen_domain())
 		return -ENODEV;
 
-- 
1.7.10.4


^ permalink raw reply	[flat|nested] 9+ messages in thread

* Re: [PATCH 2/2] drivers: xen/block: add multi-page ring support
  2015-01-23 10:14 ` [PATCH 2/2] drivers: xen/block: add multi-page ring support Bob Liu
@ 2015-02-02 10:43   ` Roger Pau Monné
  2015-02-06 10:47     ` Bob Liu
  2015-02-05 17:41   ` [Xen-devel] " Felipe Franciosi
  1 sibling, 1 reply; 9+ messages in thread
From: Roger Pau Monné @ 2015-02-02 10:43 UTC (permalink / raw)
  To: Bob Liu, xen-devel
  Cc: david.vrabel, linux-kernel, konrad.wilk, boris.ostrovsky, wei.liu2

El 23/01/15 a les 10.14, Bob Liu ha escrit:
> Extend xen/block to support multi-page ring.
>  * xen-blkback notify blkfront with feature-multi-ring-pages
>  * xen-blkfront write to xenstore about how many pages are used as the ring
> 
> If using 4 pages as the ring, inflight requests inscreased from 32 to 128 and
> IOPS improved nearly 400% on our system.
> 
> Signed-off-by: Bob Liu <bob.liu@oracle.com>
> ---
>  drivers/block/xen-blkback/xenbus.c |   86 +++++++++++++++++++++++++--------
>  drivers/block/xen-blkfront.c       |   94 ++++++++++++++++++++++++++----------

This requires a patch to xen/include/public/io/blkif.h in the Xen
repository describing how this protocol extension is going to work
before reviewing the Linux implementation.

Roger.


^ permalink raw reply	[flat|nested] 9+ messages in thread

* RE: [Xen-devel] [PATCH 2/2] drivers: xen/block: add multi-page ring support
  2015-01-23 10:14 ` [PATCH 2/2] drivers: xen/block: add multi-page ring support Bob Liu
  2015-02-02 10:43   ` Roger Pau Monné
@ 2015-02-05 17:41   ` Felipe Franciosi
  2015-02-06 10:43     ` Bob Liu
  1 sibling, 1 reply; 9+ messages in thread
From: Felipe Franciosi @ 2015-02-05 17:41 UTC (permalink / raw)
  To: 'Bob Liu', xen-devel
  Cc: Wei Liu, linux-kernel, David Vrabel, boris.ostrovsky, Roger Pau Monne

Hi Bob,

Can you elaborate on the environment where you measured such an improvement?

I'm particularly interested in:
What workload were you issuing? (e.g. 4K seq reads?)
What backend were you using? (e.g. null driver? what parameters? some specific disk/array?)
What was the host configuration for the test?
What was the VM configuration for the test?

Thanks,
Felipe


-----Original Message-----
From: xen-devel-bounces@lists.xen.org [mailto:xen-devel-bounces@lists.xen.org] On Behalf Of Bob Liu
Sent: 23 January 2015 10:15
To: xen-devel@lists.xen.org
Cc: Wei Liu; linux-kernel@vger.kernel.org; Bob Liu; David Vrabel; boris.ostrovsky@oracle.com; Roger Pau Monne
Subject: [Xen-devel] [PATCH 2/2] drivers: xen/block: add multi-page ring support

Extend xen/block to support multi-page ring.
 * xen-blkback notify blkfront with feature-multi-ring-pages
 * xen-blkfront write to xenstore about how many pages are used as the ring

If using 4 pages as the ring, inflight requests inscreased from 32 to 128 and IOPS improved nearly 400% on our system.

Signed-off-by: Bob Liu <bob.liu@oracle.com>
---
 drivers/block/xen-blkback/xenbus.c |   86 +++++++++++++++++++++++++--------
 drivers/block/xen-blkfront.c       |   94 ++++++++++++++++++++++++++----------
 2 files changed, 133 insertions(+), 47 deletions(-)

diff --git a/drivers/block/xen-blkback/xenbus.c b/drivers/block/xen-blkback/xenbus.c
index cbd13c9..a5c9c62 100644
--- a/drivers/block/xen-blkback/xenbus.c
+++ b/drivers/block/xen-blkback/xenbus.c
@@ -193,8 +193,8 @@ fail:
 	return ERR_PTR(-ENOMEM);
 }
 
-static int xen_blkif_map(struct xen_blkif *blkif, grant_ref_t gref,
-			 unsigned int evtchn)
+static int xen_blkif_map(struct xen_blkif *blkif, grant_ref_t *gref,
+			unsigned int nr_grefs, unsigned int evtchn)
 {
 	int err;
 
@@ -202,7 +202,7 @@ static int xen_blkif_map(struct xen_blkif *blkif, grant_ref_t gref,
 	if (blkif->irq)
 		return 0;
 
-	err = xenbus_map_ring_valloc(blkif->be->dev, &gref, 1,
+	err = xenbus_map_ring_valloc(blkif->be->dev, gref, nr_grefs,
 			&blkif->blk_ring);
 	if (err < 0)
 		return err;
@@ -212,21 +212,21 @@ static int xen_blkif_map(struct xen_blkif *blkif, grant_ref_t gref,
 	{
 		struct blkif_sring *sring;
 		sring = (struct blkif_sring *)blkif->blk_ring;
-		BACK_RING_INIT(&blkif->blk_rings.native, sring, PAGE_SIZE);
+		BACK_RING_INIT(&blkif->blk_rings.native, sring, PAGE_SIZE * 
+nr_grefs);
 		break;
 	}
 	case BLKIF_PROTOCOL_X86_32:
 	{
 		struct blkif_x86_32_sring *sring_x86_32;
 		sring_x86_32 = (struct blkif_x86_32_sring *)blkif->blk_ring;
-		BACK_RING_INIT(&blkif->blk_rings.x86_32, sring_x86_32, PAGE_SIZE);
+		BACK_RING_INIT(&blkif->blk_rings.x86_32, sring_x86_32, PAGE_SIZE * 
+nr_grefs);
 		break;
 	}
 	case BLKIF_PROTOCOL_X86_64:
 	{
 		struct blkif_x86_64_sring *sring_x86_64;
 		sring_x86_64 = (struct blkif_x86_64_sring *)blkif->blk_ring;
-		BACK_RING_INIT(&blkif->blk_rings.x86_64, sring_x86_64, PAGE_SIZE);
+		BACK_RING_INIT(&blkif->blk_rings.x86_64, sring_x86_64, PAGE_SIZE * 
+nr_grefs);
 		break;
 	}
 	default:
@@ -588,6 +588,13 @@ static int xen_blkbk_probe(struct xenbus_device *dev,
 	if (err)
 		goto fail;
 
+	err = xenbus_printf(XBT_NIL, dev->nodename,
+			"feature-multi-ring-pages", "%u", 1);
+	if (err) {
+		pr_debug("Error writing feature-multi-ring-pages\n");
+		goto fail;
+	}
+
 	err = xenbus_switch_state(dev, XenbusStateInitWait);
 	if (err)
 		goto fail;
@@ -852,23 +859,61 @@ again:
 static int connect_ring(struct backend_info *be)  {
 	struct xenbus_device *dev = be->dev;
-	unsigned long ring_ref;
-	unsigned int evtchn;
+	unsigned int ring_ref[XENBUS_MAX_RING_PAGES];
+	unsigned int evtchn, nr_grefs;
 	unsigned int pers_grants;
 	char protocol[64] = "";
 	int err;
 
 	DPRINTK("%s", dev->otherend);
-
-	err = xenbus_gather(XBT_NIL, dev->otherend, "ring-ref", "%lu",
-			    &ring_ref, "event-channel", "%u", &evtchn, NULL);
-	if (err) {
-		xenbus_dev_fatal(dev, err,
-				 "reading %s/ring-ref and event-channel",
-				 dev->otherend);
+	err = xenbus_scanf(XBT_NIL, dev->otherend, "event-channel", "%u",
+			&evtchn);
+	if (err != 1) {
+		err = -EINVAL;
+		xenbus_dev_fatal(dev, err, "reading %s/event-channel",
+				dev->otherend);
 		return err;
 	}
 
+	err = xenbus_scanf(XBT_NIL, dev->otherend, "max-ring-pages", "%u",
+			&nr_grefs);
+	if (err != 1) {
+		err = xenbus_scanf(XBT_NIL, dev->otherend, "ring-ref",
+				"%u", &ring_ref[0]);
+		if (err != 1) {
+			err = -EINVAL;
+			xenbus_dev_fatal(dev, err, "reading %s/ring-ref",
+					dev->otherend);
+			return err;
+		}
+		nr_grefs = 1;
+		pr_debug("%s:using one-page ring with ref: %d\n",
+				dev->otherend, ring_ref[0]);
+	} else {
+		unsigned int i;
+
+		if (nr_grefs > XENBUS_MAX_RING_PAGES) {
+			err = -EINVAL;
+			xenbus_dev_fatal(dev, err,
+				"%s/ring-pages:%d too big", dev->otherend, nr_grefs);
+			return err;
+		}
+		for (i = 0; i < nr_grefs; i++) {
+			char ring_ref_name[15];
+			snprintf(ring_ref_name, sizeof(ring_ref_name),
+					"ring-ref%u", i);
+			err = xenbus_scanf(XBT_NIL, dev->otherend,
+					ring_ref_name, "%d", &ring_ref[i]);
+			if (err != 1) {
+				err = -EINVAL;
+				xenbus_dev_fatal(dev, err, "reading %s/%s",
+						dev->otherend, ring_ref_name);
+				return err;
+			}
+			pr_debug("blkback: ring-ref%u: %d\n", i, ring_ref[i]);
+		}
+	}
+
 	be->blkif->blk_protocol = BLKIF_PROTOCOL_NATIVE;
 	err = xenbus_gather(XBT_NIL, dev->otherend, "protocol",
 			    "%63s", protocol, NULL);
@@ -893,15 +938,14 @@ static int connect_ring(struct backend_info *be)
 	be->blkif->vbd.feature_gnt_persistent = pers_grants;
 	be->blkif->vbd.overflow_max_grants = 0;
 
-	pr_info(DRV_PFX "ring-ref %ld, event-channel %d, protocol %d (%s) %s\n",
-		ring_ref, evtchn, be->blkif->blk_protocol, protocol,
-		pers_grants ? "persistent grants" : "");
+	pr_info(DRV_PFX "ring-pages:%d, event-channel %d, protocol %d (%s) %s\n",
+			nr_grefs, evtchn, be->blkif->blk_protocol, protocol,
+			pers_grants ? "persistent grants" : "");
 
 	/* Map the shared frame, irq etc. */
-	err = xen_blkif_map(be->blkif, ring_ref, evtchn);
+	err = xen_blkif_map(be->blkif, ring_ref, nr_grefs, evtchn);
 	if (err) {
-		xenbus_dev_fatal(dev, err, "mapping ring-ref %lu port %u",
-				 ring_ref, evtchn);
+		xenbus_dev_fatal(dev, err, "mapping ring-ref port %u", evtchn);
 		return err;
 	}
 
diff --git a/drivers/block/xen-blkfront.c b/drivers/block/xen-blkfront.c index 2fcf75d..dec9fe7 100644
--- a/drivers/block/xen-blkfront.c
+++ b/drivers/block/xen-blkfront.c
@@ -98,7 +98,12 @@ static unsigned int xen_blkif_max_segments = 32;  module_param_named(max, xen_blkif_max_segments, int, S_IRUGO);  MODULE_PARM_DESC(max, "Maximum amount of segments in indirect requests (default is 32)");
 
-#define BLK_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE)
+static unsigned int xen_blkif_max_ring_pages = 1; 
+module_param_named(max_ring_pages, xen_blkif_max_ring_pages, int, 0); 
+MODULE_PARM_DESC(max_ring_pages, "Maximum amount of pages to be used as 
+the ring");
+
+#define BLK_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE * 
+info->nr_ring_pages) #define BLK_MAX_RING_SIZE __CONST_RING_SIZE(blkif, 
+PAGE_SIZE * XENBUS_MAX_RING_PAGES)
 
 /*
  * We have one of these per vbd, whether ide, scsi or 'other'.  They @@ -114,13 +119,14 @@ struct blkfront_info
 	int vdevice;
 	blkif_vdev_t handle;
 	enum blkif_state connected;
-	int ring_ref;
+	int ring_ref[XENBUS_MAX_RING_PAGES];
+	int nr_ring_pages;
 	struct blkif_front_ring ring;
 	unsigned int evtchn, irq;
 	struct request_queue *rq;
 	struct work_struct work;
 	struct gnttab_free_callback callback;
-	struct blk_shadow shadow[BLK_RING_SIZE];
+	struct blk_shadow shadow[BLK_MAX_RING_SIZE];
 	struct list_head grants;
 	struct list_head indirect_pages;
 	unsigned int persistent_gnts_c;
@@ -139,8 +145,6 @@ static unsigned int nr_minors;  static unsigned long *minors;  static DEFINE_SPINLOCK(minor_lock);
 
-#define MAXIMUM_OUTSTANDING_BLOCK_REQS \
-	(BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
 #define GRANT_INVALID_REF	0
 
 #define PARTS_PER_DISK		16
@@ -1033,12 +1037,15 @@ free_shadow:
 	flush_work(&info->work);
 
 	/* Free resources associated with old device channel. */
-	if (info->ring_ref != GRANT_INVALID_REF) {
-		gnttab_end_foreign_access(info->ring_ref, 0,
-					  (unsigned long)info->ring.sring);
-		info->ring_ref = GRANT_INVALID_REF;
-		info->ring.sring = NULL;
+	for (i = 0; i < info->nr_ring_pages; i++) {
+		if (info->ring_ref[i] != GRANT_INVALID_REF) {
+			gnttab_end_foreign_access(info->ring_ref[i], 0, 0);
+			info->ring_ref[i] = GRANT_INVALID_REF;
+		}
 	}
+	free_pages((unsigned long)info->ring.sring, get_order(info->nr_ring_pages * PAGE_SIZE));
+	info->ring.sring = NULL;
+
 	if (info->irq)
 		unbind_from_irqhandler(info->irq, info);
 	info->evtchn = info->irq = 0;
@@ -1245,26 +1252,30 @@ static int setup_blkring(struct xenbus_device *dev,
 			 struct blkfront_info *info)
 {
 	struct blkif_sring *sring;
-	int err;
-	grant_ref_t gref;
+	int err, i;
+	unsigned long ring_size = info->nr_ring_pages * PAGE_SIZE;
+	grant_ref_t gref[XENBUS_MAX_RING_PAGES];
 
-	info->ring_ref = GRANT_INVALID_REF;
+	for (i = 0; i < XENBUS_MAX_RING_PAGES; i++)
+		info->ring_ref[i] = GRANT_INVALID_REF;
 
-	sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
+	sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
+						get_order(ring_size));
 	if (!sring) {
 		xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
 		return -ENOMEM;
 	}
 	SHARED_RING_INIT(sring);
-	FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
+	FRONT_RING_INIT(&info->ring, sring, ring_size);
 
-	err = xenbus_grant_ring(dev, info->ring.sring, 1, &gref);
+	err = xenbus_grant_ring(dev, info->ring.sring, info->nr_ring_pages, 
+gref);
 	if (err < 0) {
 		free_page((unsigned long)sring);
 		info->ring.sring = NULL;
 		goto fail;
 	}
-	info->ring_ref = gref;
+	for (i = 0; i < info->nr_ring_pages; i++)
+		info->ring_ref[i] = gref[i];
 
 	err = xenbus_alloc_evtchn(dev, &info->evtchn);
 	if (err)
@@ -1292,7 +1303,14 @@ static int talk_to_blkback(struct xenbus_device *dev,  {
 	const char *message = NULL;
 	struct xenbus_transaction xbt;
-	int err;
+	int err, i, multi_ring_pages = 0;
+
+	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
+			   "feature-multi-ring-pages", "%u", &multi_ring_pages);
+	if (err != 1)
+		info->nr_ring_pages = 1;
+	else
+		info->nr_ring_pages = xen_blkif_max_ring_pages;
 
 	/* Create shared ring, alloc event channel. */
 	err = setup_blkring(dev, info);
@@ -1306,12 +1324,33 @@ again:
 		goto destroy_blkring;
 	}
 
-	err = xenbus_printf(xbt, dev->nodename,
-			    "ring-ref", "%u", info->ring_ref);
-	if (err) {
-		message = "writing ring-ref";
-		goto abort_transaction;
+	if (info->nr_ring_pages == 1) {
+		err = xenbus_printf(xbt, dev->nodename,
+				"ring-ref", "%u", info->ring_ref[0]);
+		if (err) {
+			message = "writing ring-ref";
+			goto abort_transaction;
+		}
+	} else {
+		err = xenbus_printf(xbt, dev->nodename,
+				"max-ring-pages", "%u", info->nr_ring_pages);
+		if (err) {
+			message = "writing max-ring-pages";
+			goto abort_transaction;
+		}
+
+		for (i = 0; i < info->nr_ring_pages; i++) {
+			char ring_ref_name[15];
+			sprintf(ring_ref_name, "ring-ref%d", i);
+			err = xenbus_printf(xbt, dev->nodename,
+					ring_ref_name, "%d", info->ring_ref[i]);
+			if (err) {
+				message = "writing ring-ref";
+				goto abort_transaction;
+			}
+		}
 	}
+
 	err = xenbus_printf(xbt, dev->nodename,
 			    "event-channel", "%u", info->evtchn);
 	if (err) {
@@ -1422,9 +1461,8 @@ static int blkfront_probe(struct xenbus_device *dev,
 	info->connected = BLKIF_STATE_DISCONNECTED;
 	INIT_WORK(&info->work, blkif_restart_queue);
 
-	for (i = 0; i < BLK_RING_SIZE; i++)
+	for (i = 0; i < BLK_MAX_RING_SIZE; i++)
 		info->shadow[i].req.u.rw.id = i+1;
-	info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
 
 	/* Front end dir is a number, which is used as the id. */
 	info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0); @@ -1436,6 +1474,7 @@ static int blkfront_probe(struct xenbus_device *dev,
 		dev_set_drvdata(&dev->dev, NULL);
 		return err;
 	}
+	info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
 
 	return 0;
 }
@@ -1476,7 +1515,7 @@ static int blkif_recover(struct blkfront_info *info)
 
 	/* Stage 2: Set up free list. */
 	memset(&info->shadow, 0, sizeof(info->shadow));
-	for (i = 0; i < BLK_RING_SIZE; i++)
+	for (i = 0; i < BLK_MAX_RING_SIZE; i++)
 		info->shadow[i].req.u.rw.id = i+1;
 	info->shadow_free = info->ring.req_prod_pvt;
 	info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff; @@ -2088,6 +2127,9 @@ static int __init xlblk_init(void)  {
 	int ret;
 
+	if (xen_blkif_max_ring_pages > XENBUS_MAX_RING_PAGES)
+		return -EINVAL;
+
 	if (!xen_domain())
 		return -ENODEV;
 
--
1.7.10.4


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

^ permalink raw reply	[flat|nested] 9+ messages in thread

* Re: [Xen-devel] [PATCH 2/2] drivers: xen/block: add multi-page ring support
  2015-02-05 17:41   ` [Xen-devel] " Felipe Franciosi
@ 2015-02-06 10:43     ` Bob Liu
  0 siblings, 0 replies; 9+ messages in thread
From: Bob Liu @ 2015-02-06 10:43 UTC (permalink / raw)
  To: Felipe Franciosi
  Cc: xen-devel, Wei Liu, linux-kernel, David Vrabel, boris.ostrovsky,
	Roger Pau Monne


On 02/06/2015 01:41 AM, Felipe Franciosi wrote:
> Hi Bob,
> 
> Can you elaborate on the environment where you measured such an improvement?
> 
> I'm particularly interested in:
> What workload were you issuing? (e.g. 4K seq reads?)

8k writes.

> What backend were you using? (e.g. null driver? what parameters? some specific disk/array?)

Iscsi lun spread across about 100 physical drives, so the limit
outstanding requests(32) is really not enough to keep them busy.

With more inflight requests from DomU, we can see IOPS improved
significantly. E.g after using 4 pages as the ring, there could be 128
inflight requests instead of 32.

Thanks,
-Bob

> What was the host configuration for the test?
> What was the VM configuration for the test?
> 
> Thanks,
> Felipe
> 
> 
> -----Original Message-----
> From: xen-devel-bounces@lists.xen.org [mailto:xen-devel-bounces@lists.xen.org] On Behalf Of Bob Liu
> Sent: 23 January 2015 10:15
> To: xen-devel@lists.xen.org
> Cc: Wei Liu; linux-kernel@vger.kernel.org; Bob Liu; David Vrabel; boris.ostrovsky@oracle.com; Roger Pau Monne
> Subject: [Xen-devel] [PATCH 2/2] drivers: xen/block: add multi-page ring support
> 
> Extend xen/block to support multi-page ring.
>  * xen-blkback notify blkfront with feature-multi-ring-pages
>  * xen-blkfront write to xenstore about how many pages are used as the ring
> 
> If using 4 pages as the ring, inflight requests inscreased from 32 to 128 and IOPS improved nearly 400% on our system.
> 
> Signed-off-by: Bob Liu <bob.liu@oracle.com>
> ---
>  drivers/block/xen-blkback/xenbus.c |   86 +++++++++++++++++++++++++--------
>  drivers/block/xen-blkfront.c       |   94 ++++++++++++++++++++++++++----------
>  2 files changed, 133 insertions(+), 47 deletions(-)
> 
> diff --git a/drivers/block/xen-blkback/xenbus.c b/drivers/block/xen-blkback/xenbus.c
> index cbd13c9..a5c9c62 100644
> --- a/drivers/block/xen-blkback/xenbus.c
> +++ b/drivers/block/xen-blkback/xenbus.c
> @@ -193,8 +193,8 @@ fail:
>  	return ERR_PTR(-ENOMEM);
>  }
>  
> -static int xen_blkif_map(struct xen_blkif *blkif, grant_ref_t gref,
> -			 unsigned int evtchn)
> +static int xen_blkif_map(struct xen_blkif *blkif, grant_ref_t *gref,
> +			unsigned int nr_grefs, unsigned int evtchn)
>  {
>  	int err;
>  
> @@ -202,7 +202,7 @@ static int xen_blkif_map(struct xen_blkif *blkif, grant_ref_t gref,
>  	if (blkif->irq)
>  		return 0;
>  
> -	err = xenbus_map_ring_valloc(blkif->be->dev, &gref, 1,
> +	err = xenbus_map_ring_valloc(blkif->be->dev, gref, nr_grefs,
>  			&blkif->blk_ring);
>  	if (err < 0)
>  		return err;
> @@ -212,21 +212,21 @@ static int xen_blkif_map(struct xen_blkif *blkif, grant_ref_t gref,
>  	{
>  		struct blkif_sring *sring;
>  		sring = (struct blkif_sring *)blkif->blk_ring;
> -		BACK_RING_INIT(&blkif->blk_rings.native, sring, PAGE_SIZE);
> +		BACK_RING_INIT(&blkif->blk_rings.native, sring, PAGE_SIZE * 
> +nr_grefs);
>  		break;
>  	}
>  	case BLKIF_PROTOCOL_X86_32:
>  	{
>  		struct blkif_x86_32_sring *sring_x86_32;
>  		sring_x86_32 = (struct blkif_x86_32_sring *)blkif->blk_ring;
> -		BACK_RING_INIT(&blkif->blk_rings.x86_32, sring_x86_32, PAGE_SIZE);
> +		BACK_RING_INIT(&blkif->blk_rings.x86_32, sring_x86_32, PAGE_SIZE * 
> +nr_grefs);
>  		break;
>  	}
>  	case BLKIF_PROTOCOL_X86_64:
>  	{
>  		struct blkif_x86_64_sring *sring_x86_64;
>  		sring_x86_64 = (struct blkif_x86_64_sring *)blkif->blk_ring;
> -		BACK_RING_INIT(&blkif->blk_rings.x86_64, sring_x86_64, PAGE_SIZE);
> +		BACK_RING_INIT(&blkif->blk_rings.x86_64, sring_x86_64, PAGE_SIZE * 
> +nr_grefs);
>  		break;
>  	}
>  	default:
> @@ -588,6 +588,13 @@ static int xen_blkbk_probe(struct xenbus_device *dev,
>  	if (err)
>  		goto fail;
>  
> +	err = xenbus_printf(XBT_NIL, dev->nodename,
> +			"feature-multi-ring-pages", "%u", 1);
> +	if (err) {
> +		pr_debug("Error writing feature-multi-ring-pages\n");
> +		goto fail;
> +	}
> +
>  	err = xenbus_switch_state(dev, XenbusStateInitWait);
>  	if (err)
>  		goto fail;
> @@ -852,23 +859,61 @@ again:
>  static int connect_ring(struct backend_info *be)  {
>  	struct xenbus_device *dev = be->dev;
> -	unsigned long ring_ref;
> -	unsigned int evtchn;
> +	unsigned int ring_ref[XENBUS_MAX_RING_PAGES];
> +	unsigned int evtchn, nr_grefs;
>  	unsigned int pers_grants;
>  	char protocol[64] = "";
>  	int err;
>  
>  	DPRINTK("%s", dev->otherend);
> -
> -	err = xenbus_gather(XBT_NIL, dev->otherend, "ring-ref", "%lu",
> -			    &ring_ref, "event-channel", "%u", &evtchn, NULL);
> -	if (err) {
> -		xenbus_dev_fatal(dev, err,
> -				 "reading %s/ring-ref and event-channel",
> -				 dev->otherend);
> +	err = xenbus_scanf(XBT_NIL, dev->otherend, "event-channel", "%u",
> +			&evtchn);
> +	if (err != 1) {
> +		err = -EINVAL;
> +		xenbus_dev_fatal(dev, err, "reading %s/event-channel",
> +				dev->otherend);
>  		return err;
>  	}
>  
> +	err = xenbus_scanf(XBT_NIL, dev->otherend, "max-ring-pages", "%u",
> +			&nr_grefs);
> +	if (err != 1) {
> +		err = xenbus_scanf(XBT_NIL, dev->otherend, "ring-ref",
> +				"%u", &ring_ref[0]);
> +		if (err != 1) {
> +			err = -EINVAL;
> +			xenbus_dev_fatal(dev, err, "reading %s/ring-ref",
> +					dev->otherend);
> +			return err;
> +		}
> +		nr_grefs = 1;
> +		pr_debug("%s:using one-page ring with ref: %d\n",
> +				dev->otherend, ring_ref[0]);
> +	} else {
> +		unsigned int i;
> +
> +		if (nr_grefs > XENBUS_MAX_RING_PAGES) {
> +			err = -EINVAL;
> +			xenbus_dev_fatal(dev, err,
> +				"%s/ring-pages:%d too big", dev->otherend, nr_grefs);
> +			return err;
> +		}
> +		for (i = 0; i < nr_grefs; i++) {
> +			char ring_ref_name[15];
> +			snprintf(ring_ref_name, sizeof(ring_ref_name),
> +					"ring-ref%u", i);
> +			err = xenbus_scanf(XBT_NIL, dev->otherend,
> +					ring_ref_name, "%d", &ring_ref[i]);
> +			if (err != 1) {
> +				err = -EINVAL;
> +				xenbus_dev_fatal(dev, err, "reading %s/%s",
> +						dev->otherend, ring_ref_name);
> +				return err;
> +			}
> +			pr_debug("blkback: ring-ref%u: %d\n", i, ring_ref[i]);
> +		}
> +	}
> +
>  	be->blkif->blk_protocol = BLKIF_PROTOCOL_NATIVE;
>  	err = xenbus_gather(XBT_NIL, dev->otherend, "protocol",
>  			    "%63s", protocol, NULL);
> @@ -893,15 +938,14 @@ static int connect_ring(struct backend_info *be)
>  	be->blkif->vbd.feature_gnt_persistent = pers_grants;
>  	be->blkif->vbd.overflow_max_grants = 0;
>  
> -	pr_info(DRV_PFX "ring-ref %ld, event-channel %d, protocol %d (%s) %s\n",
> -		ring_ref, evtchn, be->blkif->blk_protocol, protocol,
> -		pers_grants ? "persistent grants" : "");
> +	pr_info(DRV_PFX "ring-pages:%d, event-channel %d, protocol %d (%s) %s\n",
> +			nr_grefs, evtchn, be->blkif->blk_protocol, protocol,
> +			pers_grants ? "persistent grants" : "");
>  
>  	/* Map the shared frame, irq etc. */
> -	err = xen_blkif_map(be->blkif, ring_ref, evtchn);
> +	err = xen_blkif_map(be->blkif, ring_ref, nr_grefs, evtchn);
>  	if (err) {
> -		xenbus_dev_fatal(dev, err, "mapping ring-ref %lu port %u",
> -				 ring_ref, evtchn);
> +		xenbus_dev_fatal(dev, err, "mapping ring-ref port %u", evtchn);
>  		return err;
>  	}
>  
> diff --git a/drivers/block/xen-blkfront.c b/drivers/block/xen-blkfront.c index 2fcf75d..dec9fe7 100644
> --- a/drivers/block/xen-blkfront.c
> +++ b/drivers/block/xen-blkfront.c
> @@ -98,7 +98,12 @@ static unsigned int xen_blkif_max_segments = 32;  module_param_named(max, xen_blkif_max_segments, int, S_IRUGO);  MODULE_PARM_DESC(max, "Maximum amount of segments in indirect requests (default is 32)");
>  
> -#define BLK_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE)
> +static unsigned int xen_blkif_max_ring_pages = 1; 
> +module_param_named(max_ring_pages, xen_blkif_max_ring_pages, int, 0); 
> +MODULE_PARM_DESC(max_ring_pages, "Maximum amount of pages to be used as 
> +the ring");
> +
> +#define BLK_RING_SIZE __CONST_RING_SIZE(blkif, PAGE_SIZE * 
> +info->nr_ring_pages) #define BLK_MAX_RING_SIZE __CONST_RING_SIZE(blkif, 
> +PAGE_SIZE * XENBUS_MAX_RING_PAGES)
>  
>  /*
>   * We have one of these per vbd, whether ide, scsi or 'other'.  They @@ -114,13 +119,14 @@ struct blkfront_info
>  	int vdevice;
>  	blkif_vdev_t handle;
>  	enum blkif_state connected;
> -	int ring_ref;
> +	int ring_ref[XENBUS_MAX_RING_PAGES];
> +	int nr_ring_pages;
>  	struct blkif_front_ring ring;
>  	unsigned int evtchn, irq;
>  	struct request_queue *rq;
>  	struct work_struct work;
>  	struct gnttab_free_callback callback;
> -	struct blk_shadow shadow[BLK_RING_SIZE];
> +	struct blk_shadow shadow[BLK_MAX_RING_SIZE];
>  	struct list_head grants;
>  	struct list_head indirect_pages;
>  	unsigned int persistent_gnts_c;
> @@ -139,8 +145,6 @@ static unsigned int nr_minors;  static unsigned long *minors;  static DEFINE_SPINLOCK(minor_lock);
>  
> -#define MAXIMUM_OUTSTANDING_BLOCK_REQS \
> -	(BLKIF_MAX_SEGMENTS_PER_REQUEST * BLK_RING_SIZE)
>  #define GRANT_INVALID_REF	0
>  
>  #define PARTS_PER_DISK		16
> @@ -1033,12 +1037,15 @@ free_shadow:
>  	flush_work(&info->work);
>  
>  	/* Free resources associated with old device channel. */
> -	if (info->ring_ref != GRANT_INVALID_REF) {
> -		gnttab_end_foreign_access(info->ring_ref, 0,
> -					  (unsigned long)info->ring.sring);
> -		info->ring_ref = GRANT_INVALID_REF;
> -		info->ring.sring = NULL;
> +	for (i = 0; i < info->nr_ring_pages; i++) {
> +		if (info->ring_ref[i] != GRANT_INVALID_REF) {
> +			gnttab_end_foreign_access(info->ring_ref[i], 0, 0);
> +			info->ring_ref[i] = GRANT_INVALID_REF;
> +		}
>  	}
> +	free_pages((unsigned long)info->ring.sring, get_order(info->nr_ring_pages * PAGE_SIZE));
> +	info->ring.sring = NULL;
> +
>  	if (info->irq)
>  		unbind_from_irqhandler(info->irq, info);
>  	info->evtchn = info->irq = 0;
> @@ -1245,26 +1252,30 @@ static int setup_blkring(struct xenbus_device *dev,
>  			 struct blkfront_info *info)
>  {
>  	struct blkif_sring *sring;
> -	int err;
> -	grant_ref_t gref;
> +	int err, i;
> +	unsigned long ring_size = info->nr_ring_pages * PAGE_SIZE;
> +	grant_ref_t gref[XENBUS_MAX_RING_PAGES];
>  
> -	info->ring_ref = GRANT_INVALID_REF;
> +	for (i = 0; i < XENBUS_MAX_RING_PAGES; i++)
> +		info->ring_ref[i] = GRANT_INVALID_REF;
>  
> -	sring = (struct blkif_sring *)__get_free_page(GFP_NOIO | __GFP_HIGH);
> +	sring = (struct blkif_sring *)__get_free_pages(GFP_NOIO | __GFP_HIGH,
> +						get_order(ring_size));
>  	if (!sring) {
>  		xenbus_dev_fatal(dev, -ENOMEM, "allocating shared ring");
>  		return -ENOMEM;
>  	}
>  	SHARED_RING_INIT(sring);
> -	FRONT_RING_INIT(&info->ring, sring, PAGE_SIZE);
> +	FRONT_RING_INIT(&info->ring, sring, ring_size);
>  
> -	err = xenbus_grant_ring(dev, info->ring.sring, 1, &gref);
> +	err = xenbus_grant_ring(dev, info->ring.sring, info->nr_ring_pages, 
> +gref);
>  	if (err < 0) {
>  		free_page((unsigned long)sring);
>  		info->ring.sring = NULL;
>  		goto fail;
>  	}
> -	info->ring_ref = gref;
> +	for (i = 0; i < info->nr_ring_pages; i++)
> +		info->ring_ref[i] = gref[i];
>  
>  	err = xenbus_alloc_evtchn(dev, &info->evtchn);
>  	if (err)
> @@ -1292,7 +1303,14 @@ static int talk_to_blkback(struct xenbus_device *dev,  {
>  	const char *message = NULL;
>  	struct xenbus_transaction xbt;
> -	int err;
> +	int err, i, multi_ring_pages = 0;
> +
> +	err = xenbus_scanf(XBT_NIL, info->xbdev->otherend,
> +			   "feature-multi-ring-pages", "%u", &multi_ring_pages);
> +	if (err != 1)
> +		info->nr_ring_pages = 1;
> +	else
> +		info->nr_ring_pages = xen_blkif_max_ring_pages;
>  
>  	/* Create shared ring, alloc event channel. */
>  	err = setup_blkring(dev, info);
> @@ -1306,12 +1324,33 @@ again:
>  		goto destroy_blkring;
>  	}
>  
> -	err = xenbus_printf(xbt, dev->nodename,
> -			    "ring-ref", "%u", info->ring_ref);
> -	if (err) {
> -		message = "writing ring-ref";
> -		goto abort_transaction;
> +	if (info->nr_ring_pages == 1) {
> +		err = xenbus_printf(xbt, dev->nodename,
> +				"ring-ref", "%u", info->ring_ref[0]);
> +		if (err) {
> +			message = "writing ring-ref";
> +			goto abort_transaction;
> +		}
> +	} else {
> +		err = xenbus_printf(xbt, dev->nodename,
> +				"max-ring-pages", "%u", info->nr_ring_pages);
> +		if (err) {
> +			message = "writing max-ring-pages";
> +			goto abort_transaction;
> +		}
> +
> +		for (i = 0; i < info->nr_ring_pages; i++) {
> +			char ring_ref_name[15];
> +			sprintf(ring_ref_name, "ring-ref%d", i);
> +			err = xenbus_printf(xbt, dev->nodename,
> +					ring_ref_name, "%d", info->ring_ref[i]);
> +			if (err) {
> +				message = "writing ring-ref";
> +				goto abort_transaction;
> +			}
> +		}
>  	}
> +
>  	err = xenbus_printf(xbt, dev->nodename,
>  			    "event-channel", "%u", info->evtchn);
>  	if (err) {
> @@ -1422,9 +1461,8 @@ static int blkfront_probe(struct xenbus_device *dev,
>  	info->connected = BLKIF_STATE_DISCONNECTED;
>  	INIT_WORK(&info->work, blkif_restart_queue);
>  
> -	for (i = 0; i < BLK_RING_SIZE; i++)
> +	for (i = 0; i < BLK_MAX_RING_SIZE; i++)
>  		info->shadow[i].req.u.rw.id = i+1;
> -	info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
>  
>  	/* Front end dir is a number, which is used as the id. */
>  	info->handle = simple_strtoul(strrchr(dev->nodename, '/')+1, NULL, 0); @@ -1436,6 +1474,7 @@ static int blkfront_probe(struct xenbus_device *dev,
>  		dev_set_drvdata(&dev->dev, NULL);
>  		return err;
>  	}
> +	info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff;
>  
>  	return 0;
>  }
> @@ -1476,7 +1515,7 @@ static int blkif_recover(struct blkfront_info *info)
>  
>  	/* Stage 2: Set up free list. */
>  	memset(&info->shadow, 0, sizeof(info->shadow));
> -	for (i = 0; i < BLK_RING_SIZE; i++)
> +	for (i = 0; i < BLK_MAX_RING_SIZE; i++)
>  		info->shadow[i].req.u.rw.id = i+1;
>  	info->shadow_free = info->ring.req_prod_pvt;
>  	info->shadow[BLK_RING_SIZE-1].req.u.rw.id = 0x0fffffff; @@ -2088,6 +2127,9 @@ static int __init xlblk_init(void)  {
>  	int ret;
>  
> +	if (xen_blkif_max_ring_pages > XENBUS_MAX_RING_PAGES)
> +		return -EINVAL;
> +
>  	if (!xen_domain())
>  		return -ENODEV;
>  
> --
> 1.7.10.4
> 
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xen.org
> http://lists.xen.org/xen-devel
> 

^ permalink raw reply	[flat|nested] 9+ messages in thread

* Re: [PATCH 2/2] drivers: xen/block: add multi-page ring support
  2015-02-02 10:43   ` Roger Pau Monné
@ 2015-02-06 10:47     ` Bob Liu
  2015-02-06 11:01       ` Wei Liu
  0 siblings, 1 reply; 9+ messages in thread
From: Bob Liu @ 2015-02-06 10:47 UTC (permalink / raw)
  To: Roger Pau Monné
  Cc: xen-devel, david.vrabel, linux-kernel, konrad.wilk,
	boris.ostrovsky, wei.liu2


On 02/02/2015 06:43 PM, Roger Pau Monné wrote:
> El 23/01/15 a les 10.14, Bob Liu ha escrit:
>> Extend xen/block to support multi-page ring.
>>  * xen-blkback notify blkfront with feature-multi-ring-pages
>>  * xen-blkfront write to xenstore about how many pages are used as the ring
>>
>> If using 4 pages as the ring, inflight requests inscreased from 32 to 128 and
>> IOPS improved nearly 400% on our system.
>>
>> Signed-off-by: Bob Liu <bob.liu@oracle.com>
>> ---
>>  drivers/block/xen-blkback/xenbus.c |   86 +++++++++++++++++++++++++--------
>>  drivers/block/xen-blkfront.c       |   94 ++++++++++++++++++++++++++----------
> 
> This requires a patch to xen/include/public/io/blkif.h in the Xen
> repository describing how this protocol extension is going to work
> before reviewing the Linux implementation.
> 

But there is no protocol change, just add an parameter "max_ring_pages"
to control how many pages can be used as the ring. Now with one page we
can only have 32 requests.

-- 
Regards,
-Bob

^ permalink raw reply	[flat|nested] 9+ messages in thread

* Re: [PATCH 1/2] xenbus_client: extend interface to suppurt multi-page ring
  2015-01-23 10:14 [PATCH 1/2] xenbus_client: extend interface to suppurt multi-page ring Bob Liu
  2015-01-23 10:14 ` [PATCH 2/2] drivers: xen/block: add multi-page ring support Bob Liu
@ 2015-02-06 10:57 ` Wei Liu
  1 sibling, 0 replies; 9+ messages in thread
From: Wei Liu @ 2015-02-06 10:57 UTC (permalink / raw)
  To: Bob Liu
  Cc: xen-devel, david.vrabel, linux-kernel, roger.pau, konrad.wilk,
	boris.ostrovsky, wei.liu2

On Fri, Jan 23, 2015 at 06:14:30PM +0800, Bob Liu wrote:
[...]
>  	rv = xenbus_alloc_evtchn(dev, &priv->evtchn);
>  	if (rv)
> diff --git a/drivers/net/xen-netback/netback.c b/drivers/net/xen-netback/netback.c
> index 908e65e..8513764 100644
> --- a/drivers/net/xen-netback/netback.c
> +++ b/drivers/net/xen-netback/netback.c
> @@ -1955,7 +1955,7 @@ int xenvif_map_frontend_rings(struct xenvif_queue *queue,
>  	int err = -ENOMEM;
>  
>  	err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif),
> -				     tx_ring_ref, &addr);
> +				&tx_ring_ref, 1, &addr);
>  	if (err)
>  		goto err;
>  
> @@ -1963,7 +1963,7 @@ int xenvif_map_frontend_rings(struct xenvif_queue *queue,
>  	BACK_RING_INIT(&queue->tx, txs, PAGE_SIZE);
>  
>  	err = xenbus_map_ring_valloc(xenvif_to_xenbus_device(queue->vif),
> -				     rx_ring_ref, &addr);
> +				&rx_ring_ref, 1, &addr);

With my netback maintainer hat on, indentation here looks wrong. Not
sure if this is my fault when writing the original patch, but it would
be good if you fix them in next posting. :-)

Thanks
Wei.

^ permalink raw reply	[flat|nested] 9+ messages in thread

* Re: [PATCH 2/2] drivers: xen/block: add multi-page ring support
  2015-02-06 10:47     ` Bob Liu
@ 2015-02-06 11:01       ` Wei Liu
  2015-02-06 12:52         ` Bob Liu
  0 siblings, 1 reply; 9+ messages in thread
From: Wei Liu @ 2015-02-06 11:01 UTC (permalink / raw)
  To: Bob Liu
  Cc: Roger Pau Monné,
	xen-devel, david.vrabel, linux-kernel, konrad.wilk,
	boris.ostrovsky, wei.liu2

On Fri, Feb 06, 2015 at 06:47:17PM +0800, Bob Liu wrote:
> 
> On 02/02/2015 06:43 PM, Roger Pau Monné wrote:
> > El 23/01/15 a les 10.14, Bob Liu ha escrit:
> >> Extend xen/block to support multi-page ring.
> >>  * xen-blkback notify blkfront with feature-multi-ring-pages
> >>  * xen-blkfront write to xenstore about how many pages are used as the ring
> >>
> >> If using 4 pages as the ring, inflight requests inscreased from 32 to 128 and
> >> IOPS improved nearly 400% on our system.
> >>
> >> Signed-off-by: Bob Liu <bob.liu@oracle.com>
> >> ---
> >>  drivers/block/xen-blkback/xenbus.c |   86 +++++++++++++++++++++++++--------
> >>  drivers/block/xen-blkfront.c       |   94 ++++++++++++++++++++++++++----------
> > 
> > This requires a patch to xen/include/public/io/blkif.h in the Xen
> > repository describing how this protocol extension is going to work
> > before reviewing the Linux implementation.
> > 
> 
> But there is no protocol change, just add an parameter "max_ring_pages"
> to control how many pages can be used as the ring. Now with one page we
> can only have 32 requests.
> 

I think Roger meant the xenbus protocol used to negotiate this feature,
not the data protocol regarding how request and response are
interpreted. Blkif.h (and netif.h FWIW) has section to describe xenbus
protocol.

Wei.

> -- 
> Regards,
> -Bob

^ permalink raw reply	[flat|nested] 9+ messages in thread

* Re: [PATCH 2/2] drivers: xen/block: add multi-page ring support
  2015-02-06 11:01       ` Wei Liu
@ 2015-02-06 12:52         ` Bob Liu
  0 siblings, 0 replies; 9+ messages in thread
From: Bob Liu @ 2015-02-06 12:52 UTC (permalink / raw)
  To: Wei Liu
  Cc: Roger Pau Monné,
	xen-devel, david.vrabel, linux-kernel, konrad.wilk,
	boris.ostrovsky


On 02/06/2015 07:01 PM, Wei Liu wrote:
> On Fri, Feb 06, 2015 at 06:47:17PM +0800, Bob Liu wrote:
>>
>> On 02/02/2015 06:43 PM, Roger Pau Monné wrote:
>>> El 23/01/15 a les 10.14, Bob Liu ha escrit:
>>>> Extend xen/block to support multi-page ring.
>>>>  * xen-blkback notify blkfront with feature-multi-ring-pages
>>>>  * xen-blkfront write to xenstore about how many pages are used as the ring
>>>>
>>>> If using 4 pages as the ring, inflight requests inscreased from 32 to 128 and
>>>> IOPS improved nearly 400% on our system.
>>>>
>>>> Signed-off-by: Bob Liu <bob.liu@oracle.com>
>>>> ---
>>>>  drivers/block/xen-blkback/xenbus.c |   86 +++++++++++++++++++++++++--------
>>>>  drivers/block/xen-blkfront.c       |   94 ++++++++++++++++++++++++++----------
>>>
>>> This requires a patch to xen/include/public/io/blkif.h in the Xen
>>> repository describing how this protocol extension is going to work
>>> before reviewing the Linux implementation.
>>>
>>
>> But there is no protocol change, just add an parameter "max_ring_pages"
>> to control how many pages can be used as the ring. Now with one page we
>> can only have 32 requests.
>>
> 
> I think Roger meant the xenbus protocol used to negotiate this feature,
> not the data protocol regarding how request and response are
> interpreted. Blkif.h (and netif.h FWIW) has section to describe xenbus
> protocol.
> 

Oh, I see. Sorry for the misunderstood.
Also will fix the indentation issue of the first patch.

Thanks,
-Bob


^ permalink raw reply	[flat|nested] 9+ messages in thread

end of thread, other threads:[~2015-02-06 12:52 UTC | newest]

Thread overview: 9+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2015-01-23 10:14 [PATCH 1/2] xenbus_client: extend interface to suppurt multi-page ring Bob Liu
2015-01-23 10:14 ` [PATCH 2/2] drivers: xen/block: add multi-page ring support Bob Liu
2015-02-02 10:43   ` Roger Pau Monné
2015-02-06 10:47     ` Bob Liu
2015-02-06 11:01       ` Wei Liu
2015-02-06 12:52         ` Bob Liu
2015-02-05 17:41   ` [Xen-devel] " Felipe Franciosi
2015-02-06 10:43     ` Bob Liu
2015-02-06 10:57 ` [PATCH 1/2] xenbus_client: extend interface to suppurt multi-page ring Wei Liu

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).