Merge branch 'xen-netback-next'

Zoltan Kiss says:

====================
xen-netback: TX grant mapping with SKBTX_DEV_ZEROCOPY instead of copy

A long known problem of the upstream netback implementation that on the TX
path (from guest to Dom0) it copies the whole packet from guest memory into
Dom0. That simply became a bottleneck with 10Gb NICs, and generally it's a
huge perfomance penalty. The classic kernel version of netback used grant
mapping, and to get notified when the page can be unmapped, it used page
destructors. Unfortunately that destructor is not an upstreamable solution.
Ian Campbell's skb fragment destructor patch series [1] tried to solve this
problem, however it seems to be very invasive on the network stack's code,
and therefore haven't progressed very well.
This patch series use SKBTX_DEV_ZEROCOPY flags to tell the stack it needs to
know when the skb is freed up. That is the way KVM solved the same problem,
and based on my initial tests it can do the same for us. Avoiding the extra
copy boosted up TX throughput from 6.8 Gbps to 7.9 (I used a slower AMD
Interlagos box, both Dom0 and guest on upstream kernel, on the same NUMA node,
running iperf 2.0.5, and the remote end was a bare metal box on the same 10Gb
switch)
Based on my investigations the packet get only copied if it is delivered to
Dom0 IP stack through deliver_skb, which is due to this [2] patch. This affects
DomU->Dom0 IP traffic and when Dom0 does routing/NAT for the guest. That's a bit
unfortunate, but luckily it doesn't cause a major regression for this usecase.
In the future we should try to eliminate that copy somehow.
There are a few spinoff tasks which will be addressed in separate patches:
- grant copy the header directly instead of map and memcpy. This should help
  us avoiding TLB flushing
- use something else than ballooned pages
- fix grant map to use page->index properly
I've tried to broke it down to smaller patches, with mixed results, so I
welcome suggestions on that part as well:
1: Use skb->cb to store pending_idx
2: Some refactoring
3: Change RX path for mapped SKB fragments (moved here to keep bisectability,
review it after #4)
4: Introduce TX grant mapping
5: Remove old TX grant copy definitons and fix indentations
6: Add stat counters for zerocopy
7: Handle guests with too many frags
8: Timeout packets in RX path
9: Aggregate TX unmap operations

v2: I've fixed some smaller things, see the individual patches. I've added a
few new stat counters, and handling the important use case when an older guest
sends lots of slots. Instead of delayed copy now we timeout packets on the RX
path, based on the assumption that otherwise packets should get stucked
anywhere else. Finally some unmap batching to avoid too much TLB flush

v3: Apart from fixing a few things mentioned in responses the important change
is the use the hypercall directly for grant [un]mapping, therefore we can
avoid m2p override.

v4: Now we are using a new grant mapping API to avoid m2p_override. The RX queue
timeout logic changed also.

v5: Only minor fixes based on Wei's comments

v6: Important bugfixes for xenvif_poll exit path and zerocopy callback, see
first 2 patches. Also rework of handling packets with too many slots, and
reorder the series a bit.

v7: Small fixes in comments/log messages/error paths, and merging the frag
overflow stats patch into its parent.

[1] http://lwn.net/Articles/491522/
[2] https://lkml.org/lkml/2012/7/20/363
====================

Signed-off-by: Zoltan Kiss <zoltan.kiss@citrix.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

1 files changed, 121 insertions, 7 deletions

diff --git a/drivers/net/xen-netback/interface.c b/drivers/net/xen-netback/interface.c
index 7669d49a67e2..83a71ac5b93a 100644
--- a/drivers/net/xen-netback/interface.c
+++ b/drivers/net/xen-netback/interface.c
@@ -38,6 +38,7 @@
 
 #include <xen/events.h>
 #include <asm/xen/hypercall.h>
+#include <xen/balloon.h>
 
 #define XENVIF_QUEUE_LENGTH 32
 #define XENVIF_NAPI_WEIGHT  64
@@ -87,7 +88,8 @@ static int xenvif_poll(struct napi_struct *napi, int budget)
 		local_irq_save(flags);
 
 		RING_FINAL_CHECK_FOR_REQUESTS(&vif->tx, more_to_do);
-		if (!more_to_do)
+		if (!(more_to_do &&
+		      xenvif_tx_pending_slots_available(vif)))
 			__napi_complete(napi);
 
 		local_irq_restore(flags);
@@ -113,6 +115,18 @@ static irqreturn_t xenvif_interrupt(int irq, void *dev_id)
 	return IRQ_HANDLED;
 }
 
+static void xenvif_wake_queue(unsigned long data)
+{
+	struct xenvif *vif = (struct xenvif *)data;
+
+	if (netif_queue_stopped(vif->dev)) {
+		netdev_err(vif->dev, "draining TX queue\n");
+		vif->rx_queue_purge = true;
+		xenvif_kick_thread(vif);
+		netif_wake_queue(vif->dev);
+	}
+}
+
 static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
 {
 	struct xenvif *vif = netdev_priv(dev);
@@ -121,7 +135,9 @@ static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
 	BUG_ON(skb->dev != dev);
 
 	/* Drop the packet if vif is not ready */
-	if (vif->task == NULL || !xenvif_schedulable(vif))
+	if (vif->task == NULL ||
+	    vif->dealloc_task == NULL ||
+	    !xenvif_schedulable(vif))
 		goto drop;
 
 	/* At best we'll need one slot for the header and one for each
@@ -140,8 +156,13 @@ static int xenvif_start_xmit(struct sk_buff *skb, struct net_device *dev)
 	 * then turn off the queue to give the ring a chance to
 	 * drain.
 	 */
-	if (!xenvif_rx_ring_slots_available(vif, min_slots_needed))
+	if (!xenvif_rx_ring_slots_available(vif, min_slots_needed)) {
+		vif->wake_queue.function = xenvif_wake_queue;
+		vif->wake_queue.data = (unsigned long)vif;
 		xenvif_stop_queue(vif);
+		mod_timer(&vif->wake_queue,
+			jiffies + rx_drain_timeout_jiffies);
+	}
 
 	skb_queue_tail(&vif->rx_queue, skb);
 	xenvif_kick_thread(vif);
@@ -234,6 +255,28 @@ static const struct xenvif_stat {
 		"rx_gso_checksum_fixup",
 		offsetof(struct xenvif, rx_gso_checksum_fixup)
 	},
+	/* If (sent != success + fail), there are probably packets never
+	 * freed up properly!
+	 */
+	{
+		"tx_zerocopy_sent",
+		offsetof(struct xenvif, tx_zerocopy_sent),
+	},
+	{
+		"tx_zerocopy_success",
+		offsetof(struct xenvif, tx_zerocopy_success),
+	},
+	{
+		"tx_zerocopy_fail",
+		offsetof(struct xenvif, tx_zerocopy_fail)
+	},
+	/* Number of packets exceeding MAX_SKB_FRAG slots. You should use
+	 * a guest with the same MAX_SKB_FRAG
+	 */
+	{
+		"tx_frag_overflow",
+		offsetof(struct xenvif, tx_frag_overflow)
+	},
 };
 
 static int xenvif_get_sset_count(struct net_device *dev, int string_set)
@@ -327,6 +370,8 @@ struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
 	init_timer(&vif->credit_timeout);
 	vif->credit_window_start = get_jiffies_64();
 
+	init_timer(&vif->wake_queue);
+
 	dev->netdev_ops	= &xenvif_netdev_ops;
 	dev->hw_features = NETIF_F_SG |
 		NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
@@ -343,8 +388,27 @@ struct xenvif *xenvif_alloc(struct device *parent, domid_t domid,
 	vif->pending_prod = MAX_PENDING_REQS;
 	for (i = 0; i < MAX_PENDING_REQS; i++)
 		vif->pending_ring[i] = i;
-	for (i = 0; i < MAX_PENDING_REQS; i++)
-		vif->mmap_pages[i] = NULL;
+	spin_lock_init(&vif->callback_lock);
+	spin_lock_init(&vif->response_lock);
+	/* If ballooning is disabled, this will consume real memory, so you
+	 * better enable it. The long term solution would be to use just a
+	 * bunch of valid page descriptors, without dependency on ballooning
+	 */
+	err = alloc_xenballooned_pages(MAX_PENDING_REQS,
+				       vif->mmap_pages,
+				       false);
+	if (err) {
+		netdev_err(dev, "Could not reserve mmap_pages\n");
+		return ERR_PTR(-ENOMEM);
+	}
+	for (i = 0; i < MAX_PENDING_REQS; i++) {
+		vif->pending_tx_info[i].callback_struct = (struct ubuf_info)
+			{ .callback = xenvif_zerocopy_callback,
+			  .ctx = NULL,
+			  .desc = i };
+		vif->grant_tx_handle[i] = NETBACK_INVALID_HANDLE;
+	}
+	init_timer(&vif->dealloc_delay);
 
 	/*
 	 * Initialise a dummy MAC address. We choose the numerically
@@ -382,12 +446,14 @@ int xenvif_connect(struct xenvif *vif, unsigned long tx_ring_ref,
 
 	BUG_ON(vif->tx_irq);
 	BUG_ON(vif->task);
+	BUG_ON(vif->dealloc_task);
 
 	err = xenvif_map_frontend_rings(vif, tx_ring_ref, rx_ring_ref);
 	if (err < 0)
 		goto err;
 
 	init_waitqueue_head(&vif->wq);
+	init_waitqueue_head(&vif->dealloc_wq);
 
 	if (tx_evtchn == rx_evtchn) {
 		/* feature-split-event-channels == 0 */
@@ -421,8 +487,8 @@ int xenvif_connect(struct xenvif *vif, unsigned long tx_ring_ref,
 		disable_irq(vif->rx_irq);
 	}
 
-	task = kthread_create(xenvif_kthread,
-			      (void *)vif, "%s", vif->dev->name);
+	task = kthread_create(xenvif_kthread_guest_rx,
+			      (void *)vif, "%s-guest-rx", vif->dev->name);
 	if (IS_ERR(task)) {
 		pr_warn("Could not allocate kthread for %s\n", vif->dev->name);
 		err = PTR_ERR(task);
@@ -431,6 +497,16 @@ int xenvif_connect(struct xenvif *vif, unsigned long tx_ring_ref,
 
 	vif->task = task;
 
+	task = kthread_create(xenvif_dealloc_kthread,
+			      (void *)vif, "%s-dealloc", vif->dev->name);
+	if (IS_ERR(task)) {
+		pr_warn("Could not allocate kthread for %s\n", vif->dev->name);
+		err = PTR_ERR(task);
+		goto err_rx_unbind;
+	}
+
+	vif->dealloc_task = task;
+
 	rtnl_lock();
 	if (!vif->can_sg && vif->dev->mtu > ETH_DATA_LEN)
 		dev_set_mtu(vif->dev, ETH_DATA_LEN);
@@ -441,6 +517,7 @@ int xenvif_connect(struct xenvif *vif, unsigned long tx_ring_ref,
 	rtnl_unlock();
 
 	wake_up_process(vif->task);
+	wake_up_process(vif->dealloc_task);
 
 	return 0;
 
@@ -474,10 +551,17 @@ void xenvif_disconnect(struct xenvif *vif)
 		xenvif_carrier_off(vif);
 
 	if (vif->task) {
+		del_timer_sync(&vif->wake_queue);
 		kthread_stop(vif->task);
 		vif->task = NULL;
 	}
 
+	if (vif->dealloc_task) {
+		del_timer_sync(&vif->dealloc_delay);
+		kthread_stop(vif->dealloc_task);
+		vif->dealloc_task = NULL;
+	}
+
 	if (vif->tx_irq) {
 		if (vif->tx_irq == vif->rx_irq)
 			unbind_from_irqhandler(vif->tx_irq, vif);
@@ -493,6 +577,36 @@ void xenvif_disconnect(struct xenvif *vif)
 
 void xenvif_free(struct xenvif *vif)
 {
+	int i, unmap_timeout = 0;
+	/* Here we want to avoid timeout messages if an skb can be legitimatly
+	 * stucked somewhere else. Realisticly this could be an another vif's
+	 * internal or QDisc queue. That another vif also has this
+	 * rx_drain_timeout_msecs timeout, but the timer only ditches the
+	 * internal queue. After that, the QDisc queue can put in worst case
+	 * XEN_NETIF_RX_RING_SIZE / MAX_SKB_FRAGS skbs into that another vif's
+	 * internal queue, so we need several rounds of such timeouts until we
+	 * can be sure that no another vif should have skb's from us. We are
+	 * not sending more skb's, so newly stucked packets are not interesting
+	 * for us here.
+	 */
+	unsigned int worst_case_skb_lifetime = (rx_drain_timeout_msecs/1000) *
+		DIV_ROUND_UP(XENVIF_QUEUE_LENGTH, (XEN_NETIF_RX_RING_SIZE / MAX_SKB_FRAGS));
+
+	for (i = 0; i < MAX_PENDING_REQS; ++i) {
+		if (vif->grant_tx_handle[i] != NETBACK_INVALID_HANDLE) {
+			unmap_timeout++;
+			schedule_timeout(msecs_to_jiffies(1000));
+			if (unmap_timeout > worst_case_skb_lifetime &&
+			    net_ratelimit())
+				netdev_err(vif->dev,
+					   "Page still granted! Index: %x\n",
+					   i);
+			i = -1;
+		}
+	}
+
+	free_xenballooned_pages(MAX_PENDING_REQS, vif->mmap_pages);
+
 	netif_napi_del(&vif->napi);
 
 	unregister_netdev(vif->dev);