xen/blkback: Persistent grant maps for xen blk drivers

This patch implements persistent grants for the xen-blk{front,back}
mechanism. The effect of this change is to reduce the number of unmap
operations performed, since they cause a (costly) TLB shootdown. This
allows the I/O performance to scale better when a large number of VMs
are performing I/O.

Previously, the blkfront driver was supplied a bvec[] from the request
queue. This was granted to dom0; dom0 performed the I/O and wrote
directly into the grant-mapped memory and unmapped it; blkfront then
removed foreign access for that grant. The cost of unmapping scales
badly with the number of CPUs in Dom0. An experiment showed that when
Dom0 has 24 VCPUs, and guests are performing parallel I/O to a
ramdisk, the IPIs from performing unmap's is a bottleneck at 5 guests
(at which point 650,000 IOPS are being performed in total). If more
than 5 guests are used, the performance declines. By 10 guests, only
400,000 IOPS are being performed.

This patch improves performance by only unmapping when the connection
between blkfront and back is broken.

On startup blkfront notifies blkback that it is using persistent
grants, and blkback will do the same. If blkback is not capable of
persistent mapping, blkfront will still use the same grants, since it
is compatible with the previous protocol, and simplifies the code
complexity in blkfront.

To perform a read, in persistent mode, blkfront uses a separate pool
of pages that it maps to dom0. When a request comes in, blkfront
transmutes the request so that blkback will write into one of these
free pages. Blkback keeps note of which grefs it has already
mapped. When a new ring request comes to blkback, it looks to see if
it has already mapped that page. If so, it will not map it again. If
the page hasn't been previously mapped, it is mapped now, and a record
is kept of this mapping. Blkback proceeds as usual. When blkfront is
notified that blkback has completed a request, it memcpy's from the
shared memory, into the bvec supplied. A record that the {gref, page}
tuple is mapped, and not inflight is kept.

Writes are similar, except that the memcpy is peformed from the
supplied bvecs, into the shared pages, before the request is put onto
the ring.

Blkback stores a mapping of grefs=>{page mapped to by gref} in
a red-black tree. As the grefs are not known apriori, and provide no
guarantees on their ordering, we have to perform a search
through this tree to find the page, for every gref we receive. This
operation takes O(log n) time in the worst case. In blkfront grants
are stored using a single linked list.

The maximum number of grants that blkback will persistenly map is
currently set to RING_SIZE * BLKIF_MAX_SEGMENTS_PER_REQUEST, to
prevent a malicios guest from attempting a DoS, by supplying fresh
grefs, causing the Dom0 kernel to map excessively. If a guest
is using persistent grants and exceeds the maximum number of grants to
map persistenly the newly passed grefs will be mapped and unmaped.
Using this approach, we can have requests that mix persistent and
non-persistent grants, and we need to handle them correctly.
This allows us to set the maximum number of persistent grants to a
lower value than RING_SIZE * BLKIF_MAX_SEGMENTS_PER_REQUEST, although
setting it will lead to unpredictable performance.

In writing this patch, the question arrises as to if the additional
cost of performing memcpys in the guest (to/from the pool of granted
pages) outweigh the gains of not performing TLB shootdowns. The answer
to that question is `no'. There appears to be very little, if any
additional cost to the guest of using persistent grants. There is
perhaps a small saving, from the reduced number of hypercalls
performed in granting, and ending foreign access.

Signed-off-by: Oliver Chick <oliver.chick@citrix.com>
Signed-off-by: Roger Pau Monne <roger.pau@citrix.com>
Signed-off-by: Konrad Rzeszutek Wilk <konrad.wilk@oracle.com>
[v1: Fixed up the misuse of bool as int]

1 files changed, 169 insertions, 28 deletions

diff --git a/drivers/block/xen-blkfront.c b/drivers/block/xen-blkfront.c
index 007db89..911d733 100644
--- a/drivers/block/xen-blkfront.c
+++ b/drivers/block/xen-blkfront.c
@@ -44,6 +44,7 @@
 #include <linux/mutex.h>
 #include <linux/scatterlist.h>
 #include <linux/bitmap.h>
+#include <linux/llist.h>
 
 #include <xen/xen.h>
 #include <xen/xenbus.h>
@@ -64,10 +65,17 @@ enum blkif_state {
 	BLKIF_STATE_SUSPENDED,
 };
 
+struct grant {
+	grant_ref_t gref;
+	unsigned long pfn;
+	struct llist_node node;
+};
+
 struct blk_shadow {
 	struct blkif_request req;
 	struct request *request;
 	unsigned long frame[BLKIF_MAX_SEGMENTS_PER_REQUEST];
+	struct grant *grants_used[BLKIF_MAX_SEGMENTS_PER_REQUEST];
 };
 
 static DEFINE_MUTEX(blkfront_mutex);
@@ -97,6 +105,8 @@ struct blkfront_info
 	struct work_struct work;
 	struct gnttab_free_callback callback;
 	struct blk_shadow shadow[BLK_RING_SIZE];
+	struct llist_head persistent_gnts;
+	unsigned int persistent_gnts_c;
 	unsigned long shadow_free;
 	unsigned int feature_flush;
 	unsigned int flush_op;
@@ -104,6 +114,7 @@ struct blkfront_info
 	unsigned int feature_secdiscard:1;
 	unsigned int discard_granularity;
 	unsigned int discard_alignment;
+	unsigned int feature_persistent:1;
 	int is_ready;
 };
 
@@ -287,21 +298,36 @@ static int blkif_queue_request(struct request *req)
 	unsigned long id;
 	unsigned int fsect, lsect;
 	int i, ref;
+
+	/*
+	 * Used to store if we are able to queue the request by just using
+	 * existing persistent grants, or if we have to get new grants,
+	 * as there are not sufficiently many free.
+	 */
+	bool new_persistent_gnts;
 	grant_ref_t gref_head;
+	struct page *granted_page;
+	struct grant *gnt_list_entry = NULL;
 	struct scatterlist *sg;
 
 	if (unlikely(info->connected != BLKIF_STATE_CONNECTED))
 		return 1;
 
-	if (gnttab_alloc_grant_references(
-		BLKIF_MAX_SEGMENTS_PER_REQUEST, &gref_head) < 0) {
-		gnttab_request_free_callback(
-			&info->callback,
-			blkif_restart_queue_callback,
-			info,
-			BLKIF_MAX_SEGMENTS_PER_REQUEST);
-		return 1;
-	}
+	/* Check if we have enought grants to allocate a requests */
+	if (info->persistent_gnts_c < BLKIF_MAX_SEGMENTS_PER_REQUEST) {
+		new_persistent_gnts = 1;
+		if (gnttab_alloc_grant_references(
+		    BLKIF_MAX_SEGMENTS_PER_REQUEST - info->persistent_gnts_c,
+		    &gref_head) < 0) {
+			gnttab_request_free_callback(
+				&info->callback,
+				blkif_restart_queue_callback,
+				info,
+				BLKIF_MAX_SEGMENTS_PER_REQUEST);
+			return 1;
+		}
+	} else
+		new_persistent_gnts = 0;
 
 	/* Fill out a communications ring structure. */
 	ring_req = RING_GET_REQUEST(&info->ring, info->ring.req_prod_pvt);
@@ -341,18 +367,73 @@ static int blkif_queue_request(struct request *req)
 		       BLKIF_MAX_SEGMENTS_PER_REQUEST);
 
 		for_each_sg(info->sg, sg, ring_req->u.rw.nr_segments, i) {
-			buffer_mfn = pfn_to_mfn(page_to_pfn(sg_page(sg)));
 			fsect = sg->offset >> 9;
 			lsect = fsect + (sg->length >> 9) - 1;
-			/* install a grant reference. */
-			ref = gnttab_claim_grant_reference(&gref_head);
-			BUG_ON(ref == -ENOSPC);
 
-			gnttab_grant_foreign_access_ref(
-					ref,
+			if (info->persistent_gnts_c) {
+				BUG_ON(llist_empty(&info->persistent_gnts));
+				gnt_list_entry = llist_entry(
+					llist_del_first(&info->persistent_gnts),
+					struct grant, node);
+
+				ref = gnt_list_entry->gref;
+				buffer_mfn = pfn_to_mfn(gnt_list_entry->pfn);
+				info->persistent_gnts_c--;
+			} else {
+				ref = gnttab_claim_grant_reference(&gref_head);
+				BUG_ON(ref == -ENOSPC);
+
+				gnt_list_entry =
+					kmalloc(sizeof(struct grant),
+							 GFP_ATOMIC);
+				if (!gnt_list_entry)
+					return -ENOMEM;
+
+				granted_page = alloc_page(GFP_ATOMIC);
+				if (!granted_page) {
+					kfree(gnt_list_entry);
+					return -ENOMEM;
+				}
+
+				gnt_list_entry->pfn =
+					page_to_pfn(granted_page);
+				gnt_list_entry->gref = ref;
+
+				buffer_mfn = pfn_to_mfn(page_to_pfn(
+								granted_page));
+				gnttab_grant_foreign_access_ref(ref,
 					info->xbdev->otherend_id,
-					buffer_mfn,
-					rq_data_dir(req));
+					buffer_mfn, 0);
+			}
+
+			info->shadow[id].grants_used[i] = gnt_list_entry;
+
+			if (rq_data_dir(req)) {
+				char *bvec_data;
+				void *shared_data;
+
+				BUG_ON(sg->offset + sg->length > PAGE_SIZE);
+
+				shared_data = kmap_atomic(
+					pfn_to_page(gnt_list_entry->pfn));
+				bvec_data = kmap_atomic(sg_page(sg));
+
+				/*
+				 * this does not wipe data stored outside the
+				 * range sg->offset..sg->offset+sg->length.
+				 * Therefore, blkback *could* see data from
+				 * previous requests. This is OK as long as
+				 * persistent grants are shared with just one
+				 * domain. It may need refactoring if this
+				 * changes
+				 */
+				memcpy(shared_data + sg->offset,
+				       bvec_data   + sg->offset,
+				       sg->length);
+
+				kunmap_atomic(bvec_data);
+				kunmap_atomic(shared_data);
+			}
 
 			info->shadow[id].frame[i] = mfn_to_pfn(buffer_mfn);
 			ring_req->u.rw.seg[i] =
@@ -368,7 +449,8 @@ static int blkif_queue_request(struct request *req)
 	/* Keep a private copy so we can reissue requests when recovering. */
 	info->shadow[id].req = *ring_req;
 
-	gnttab_free_grant_references(gref_head);
+	if (new_persistent_gnts)
+		gnttab_free_grant_references(gref_head);
 
 	return 0;
 }
@@ -480,12 +562,13 @@ static int xlvbd_init_blk_queue(struct gendisk *gd, u16 sector_size)
 static void xlvbd_flush(struct blkfront_info *info)
 {
 	blk_queue_flush(info->rq, info->feature_flush);
-	printk(KERN_INFO "blkfront: %s: %s: %s\n",
+	printk(KERN_INFO "blkfront: %s: %s: %s %s\n",
 	       info->gd->disk_name,
 	       info->flush_op == BLKIF_OP_WRITE_BARRIER ?
 		"barrier" : (info->flush_op == BLKIF_OP_FLUSH_DISKCACHE ?
 		"flush diskcache" : "barrier or flush"),
-	       info->feature_flush ? "enabled" : "disabled");
+	       info->feature_flush ? "enabled" : "disabled",
+	       info->feature_persistent ? "using persistent grants" : "");
 }
 
 static int xen_translate_vdev(int vdevice, int *minor, unsigned int *offset)
@@ -707,6 +790,9 @@ static void blkif_restart_queue(struct work_struct *work)
 
 static void blkif_free(struct blkfront_info *info, int suspend)
 {
+	struct llist_node *all_gnts;
+	struct grant *persistent_gnt;
+
 	/* Prevent new requests being issued until we fix things up. */
 	spin_lock_irq(&info->io_lock);
 	info->connected = suspend ?
@@ -714,6 +800,17 @@ static void blkif_free(struct blkfront_info *info, int suspend)
 	/* No more blkif_request(). */
 	if (info->rq)
 		blk_stop_queue(info->rq);
+
+	/* Remove all persistent grants */
+	if (info->persistent_gnts_c) {
+		all_gnts = llist_del_all(&info->persistent_gnts);
+		llist_for_each_entry(persistent_gnt, all_gnts, node) {
+			gnttab_end_foreign_access(persistent_gnt->gref, 0, 0UL);
+			kfree(persistent_gnt);
+		}
+		info->persistent_gnts_c = 0;
+	}
+
 	/* No more gnttab callback work. */
 	gnttab_cancel_free_callback(&info->callback);
 	spin_unlock_irq(&info->io_lock);
@@ -734,13 +831,42 @@ static void blkif_free(struct blkfront_info *info, int suspend)
 
 }
 
-static void blkif_completion(struct blk_shadow *s)
+static void blkif_completion(struct blk_shadow *s, struct blkfront_info *info,
+			     struct blkif_response *bret)
 {
 	int i;
-	/* Do not let BLKIF_OP_DISCARD as nr_segment is in the same place
-	 * flag. */
-	for (i = 0; i < s->req.u.rw.nr_segments; i++)
-		gnttab_end_foreign_access(s->req.u.rw.seg[i].gref, 0, 0UL);
+	struct bio_vec *bvec;
+	struct req_iterator iter;
+	unsigned long flags;
+	char *bvec_data;
+	void *shared_data;
+	unsigned int offset = 0;
+
+	if (bret->operation == BLKIF_OP_READ) {
+		/*
+		 * Copy the data received from the backend into the bvec.
+		 * Since bv_offset can be different than 0, and bv_len different
+		 * than PAGE_SIZE, we have to keep track of the current offset,
+		 * to be sure we are copying the data from the right shared page.
+		 */
+		rq_for_each_segment(bvec, s->request, iter) {
+			BUG_ON((bvec->bv_offset + bvec->bv_len) > PAGE_SIZE);
+			i = offset >> PAGE_SHIFT;
+			shared_data = kmap_atomic(
+				pfn_to_page(s->grants_used[i]->pfn));
+			bvec_data = bvec_kmap_irq(bvec, &flags);
+			memcpy(bvec_data, shared_data + bvec->bv_offset,
+				bvec->bv_len);
+			bvec_kunmap_irq(bvec_data, &flags);
+			kunmap_atomic(shared_data);
+			offset += bvec->bv_len;
+		}
+	}
+	/* Add the persistent grant into the list of free grants */
+	for (i = 0; i < s->req.u.rw.nr_segments; i++) {
+		llist_add(&s->grants_used[i]->node, &info->persistent_gnts);
+		info->persistent_gnts_c++;
+	}
 }
 
 static irqreturn_t blkif_interrupt(int irq, void *dev_id)
@@ -783,7 +909,7 @@ static irqreturn_t blkif_interrupt(int irq, void *dev_id)
 		req  = info->shadow[id].request;
 
 		if (bret->operation != BLKIF_OP_DISCARD)
-			blkif_completion(&info->shadow[id]);
+			blkif_completion(&info->shadow[id], info, bret);
 
 		if (add_id_to_freelist(info, id)) {
 			WARN(1, "%s: response to %s (id %ld) couldn't be recycled!\n",
@@ -942,6 +1068,11 @@ again:
 		message = "writing protocol";
 		goto abort_transaction;
 	}
+	err = xenbus_printf(xbt, dev->nodename,
+			    "feature-persistent-grants", "%u", 1);
+	if (err)
+		dev_warn(&dev->dev,
+			 "writing persistent grants feature to xenbus");
 
 	err = xenbus_transaction_end(xbt, 0);
 	if (err) {
@@ -1029,6 +1160,8 @@ static int blkfront_probe(struct xenbus_device *dev,
 	spin_lock_init(&info->io_lock);
 	info->xbdev = dev;
 	info->vdevice = vdevice;
+	init_llist_head(&info->persistent_gnts);
+	info->persistent_gnts_c = 0;
 	info->connected = BLKIF_STATE_DISCONNECTED;
 	INIT_WORK(&info->work, blkif_restart_queue);
 
@@ -1093,7 +1226,7 @@ static int blkif_recover(struct blkfront_info *info)
 					req->u.rw.seg[j].gref,
 					info->xbdev->otherend_id,
 					pfn_to_mfn(info->shadow[req->u.rw.id].frame[j]),
-					rq_data_dir(info->shadow[req->u.rw.id].request));
+					0);
 		}
 		info->shadow[req->u.rw.id].req = *req;
 
@@ -1225,7 +1358,7 @@ static void blkfront_connect(struct blkfront_info *info)
 	unsigned long sector_size;
 	unsigned int binfo;
 	int err;
-	int barrier, flush, discard;
+	int barrier, flush, discard, persistent;
 
 	switch (info->connected) {
 	case BLKIF_STATE_CONNECTED:
@@ -1303,6 +1436,14 @@ static void blkfront_connect(struct blkfront_info *info)
 	if (!err && discard)
 		blkfront_setup_discard(info);
 
+	err = xenbus_gather(XBT_NIL, info->xbdev->otherend,
+			    "feature-persistent", "%u", &persistent,
+			    NULL);
+	if (err)
+		info->feature_persistent = 0;
+	else
+		info->feature_persistent = persistent;
+
 	err = xlvbd_alloc_gendisk(sectors, info, binfo, sector_size);
 	if (err) {
 		xenbus_dev_fatal(info->xbdev, err, "xlvbd_add at %s",