From 1f6394e382f137f07136182c591ca8a7eeba6a06 Mon Sep 17 00:00:00 2001 From: Sowmini Varadhan Date: Tue, 12 Aug 2014 10:27:24 -0400 Subject: sunvnet: Do not ask for an ACK for every dring transmit No need to ask for an ack with every vnet_start_xmit()- the single ACK with DRING_STOPPED is sufficient for the protocol, and we free the sk_buff in vnet_start_xmit itself, so we dont need an ACK back. Signed-off-by: Sowmini Varadhan Acked-by: Raghuram Kothakota Signed-off-by: David S. Miller --- drivers/net/ethernet/sun/sunvnet.c | 10 +++++++++- 1 file changed, 9 insertions(+), 1 deletion(-) (limited to 'drivers/net/ethernet/sun/sunvnet.c') diff --git a/drivers/net/ethernet/sun/sunvnet.c b/drivers/net/ethernet/sun/sunvnet.c index d813bfb1a847..238434dcfe02 100644 --- a/drivers/net/ethernet/sun/sunvnet.c +++ b/drivers/net/ethernet/sun/sunvnet.c @@ -691,7 +691,15 @@ static int vnet_start_xmit(struct sk_buff *skb, struct net_device *dev) memset(tx_buf+VNET_PACKET_SKIP+skb->len, 0, len - skb->len); } - d->hdr.ack = VIO_ACK_ENABLE; + /* We don't rely on the ACKs to free the skb in vnet_start_xmit(), + * thus it is safe to not set VIO_ACK_ENABLE for each transmission: + * the protocol itself does not require it as long as the peer + * sends a VIO_SUBTYPE_ACK for VIO_DRING_STOPPED. + * + * An ACK for every packet in the ring is expensive as the + * sending of LDC messages is slow and affects performance. + */ + d->hdr.ack = VIO_ACK_DISABLE; d->size = len; d->ncookies = port->tx_bufs[dr->prod].ncookies; for (i = 0; i < d->ncookies; i++) -- cgit v1.2.3 From adddc32d6fde82156c779997f928865100542e55 Mon Sep 17 00:00:00 2001 From: Sowmini Varadhan Date: Tue, 12 Aug 2014 10:33:10 -0400 Subject: sunvnet: Do not spin in an infinite loop when vio_ldc_send() returns EAGAIN ldc_rx -> vnet_rx -> .. -> vnet_walk_rx->vnet_send_ack should not spin into an infinite loop waiting EAGAIN to lift. The sender could have sent us a burst, and gone to lunch without doing any more ldc_read()'s. That should not cause the receiver to loop infinitely till soft-lockup kicks in. Similarly __vnet_tx_trigger should only loop on EAGAIN a finite number of times. The caller (vnet_start_xmit()) already has code to reset the dring state and bail on errors from __vnet_tx_trigger Signed-off-by: Sowmini Varadhan Acked-by: Raghuram Kothakota Signed-off-by: David S. Miller --- drivers/net/ethernet/sun/sunvnet.c | 16 ++++++++++++++++ 1 file changed, 16 insertions(+) (limited to 'drivers/net/ethernet/sun/sunvnet.c') diff --git a/drivers/net/ethernet/sun/sunvnet.c b/drivers/net/ethernet/sun/sunvnet.c index 238434dcfe02..a5871791e452 100644 --- a/drivers/net/ethernet/sun/sunvnet.c +++ b/drivers/net/ethernet/sun/sunvnet.c @@ -32,6 +32,11 @@ MODULE_DESCRIPTION("Sun LDOM virtual network driver"); MODULE_LICENSE("GPL"); MODULE_VERSION(DRV_MODULE_VERSION); +/* Heuristic for the number of times to exponentially backoff and + * retry sending an LDC trigger when EAGAIN is encountered + */ +#define VNET_MAX_RETRIES 10 + /* Ordered from largest major to lowest */ static struct vio_version vnet_versions[] = { { .major = 1, .minor = 0 }, @@ -260,6 +265,7 @@ static int vnet_send_ack(struct vnet_port *port, struct vio_dring_state *dr, .state = vio_dring_state, }; int err, delay; + int retries = 0; hdr.seq = dr->snd_nxt; delay = 1; @@ -272,6 +278,13 @@ static int vnet_send_ack(struct vnet_port *port, struct vio_dring_state *dr, udelay(delay); if ((delay <<= 1) > 128) delay = 128; + if (retries++ > VNET_MAX_RETRIES) { + pr_info("ECONNRESET %x:%x:%x:%x:%x:%x\n", + port->raddr[0], port->raddr[1], + port->raddr[2], port->raddr[3], + port->raddr[4], port->raddr[5]); + err = -ECONNRESET; + } } while (err == -EAGAIN); return err; @@ -593,6 +606,7 @@ static int __vnet_tx_trigger(struct vnet_port *port) .end_idx = (u32) -1, }; int err, delay; + int retries = 0; hdr.seq = dr->snd_nxt; delay = 1; @@ -605,6 +619,8 @@ static int __vnet_tx_trigger(struct vnet_port *port) udelay(delay); if ((delay <<= 1) > 128) delay = 128; + if (retries++ > VNET_MAX_RETRIES) + break; } while (err == -EAGAIN); return err; -- cgit v1.2.3 From 1d311ad2f9fe5d75cf228372174170d0f435e3ef Mon Sep 17 00:00:00 2001 From: Sowmini Varadhan Date: Wed, 13 Aug 2014 10:29:41 -0400 Subject: sunvnet: Schedule maybe_tx_wakeup() as a tasklet from ldc_rx path At the tail of vnet_event(), if we hit the maybe_tx_wakeup() condition, we try to take the netif_tx_lock() in the recv-interrupt-context and can deadlock with dev_watchdog(). vnet_event() should schedule maybe_tx_wakeup() as a tasklet to avoid this deadlock Signed-off-by: Sowmini Varadhan Signed-off-by: David S. Miller --- drivers/net/ethernet/sun/sunvnet.c | 12 ++++++++++-- 1 file changed, 10 insertions(+), 2 deletions(-) (limited to 'drivers/net/ethernet/sun/sunvnet.c') diff --git a/drivers/net/ethernet/sun/sunvnet.c b/drivers/net/ethernet/sun/sunvnet.c index a5871791e452..23c89ab5a6ad 100644 --- a/drivers/net/ethernet/sun/sunvnet.c +++ b/drivers/net/ethernet/sun/sunvnet.c @@ -488,8 +488,9 @@ static int handle_mcast(struct vnet_port *port, void *msgbuf) return 0; } -static void maybe_tx_wakeup(struct vnet *vp) +static void maybe_tx_wakeup(unsigned long param) { + struct vnet *vp = (struct vnet *)param; struct net_device *dev = vp->dev; netif_tx_lock(dev); @@ -586,8 +587,13 @@ static void vnet_event(void *arg, int event) break; } spin_unlock(&vio->lock); + /* Kick off a tasklet to wake the queue. We cannot call + * maybe_tx_wakeup directly here because we could deadlock on + * netif_tx_lock() with dev_watchdog() + */ if (unlikely(tx_wakeup && err != -ECONNRESET)) - maybe_tx_wakeup(port->vp); + tasklet_schedule(&port->vp->vnet_tx_wakeup); + local_irq_restore(flags); } @@ -1070,6 +1076,7 @@ static struct vnet *vnet_new(const u64 *local_mac) vp = netdev_priv(dev); spin_lock_init(&vp->lock); + tasklet_init(&vp->vnet_tx_wakeup, maybe_tx_wakeup, (unsigned long)vp); vp->dev = dev; INIT_LIST_HEAD(&vp->port_list); @@ -1129,6 +1136,7 @@ static void vnet_cleanup(void) vp = list_first_entry(&vnet_list, struct vnet, list); list_del(&vp->list); dev = vp->dev; + tasklet_kill(&vp->vnet_tx_wakeup); /* vio_unregister_driver() should have cleaned up port_list */ BUG_ON(!list_empty(&vp->port_list)); unregister_netdev(dev); -- cgit v1.2.3 From 78dcff7b73ad1c9052f0b5999f9aaab7648ba2ef Mon Sep 17 00:00:00 2001 From: David L Stevens Date: Mon, 8 Sep 2014 16:23:01 -0400 Subject: sunvnet - add missing rmb() for sunvnet driver The sunvnet driver does not have an rmb() in the ring consumer corresponding to the wmb() in the producer. According to Documentation/memory-barriers.txt: "When dealing with CPU-CPU interactions, certain types of memory barrier should always be paired. A lack of appropriate pairing is almost certainly an error." In cases where an rmb() is not a no-op and a consumer is removing data from the ring while a producer is adding new entries, a load reorder would allow CPU1 CPU2 ---- ---- LOAD desc.size [e.g] STORE desc.size set desc.hdr.state = VIO_DESC_READY LOAD desc.hdr.state [because VIO_DESC_READY, use old desc.size, already loaded out of order] [CPU2 has reordered apparently unrelated LOADs] To ensure other desc fields are not loaded before checking VIO_DESC_READY, we need an rmb() between the check and desc data accesses. I've also moved the viodbg() call to after the rmb() so that it, too, has current descriptor data even with reordering, which has the side effect that it won't print anything for descriptors that are not VIO_DESC_READY as before. That's a) probably a good thing, since the fields are not necessarily set and, b) better than adding another rmb() just for viodbg(). This would not be possible if strict-ordering is enforced, but then the memory barriers should be no-ops in that case. Signed-off-by: David L Stevens Signed-off-by: David S. Miller --- drivers/net/ethernet/sun/sunvnet.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-) (limited to 'drivers/net/ethernet/sun/sunvnet.c') diff --git a/drivers/net/ethernet/sun/sunvnet.c b/drivers/net/ethernet/sun/sunvnet.c index 23c89ab5a6ad..f67539650c38 100644 --- a/drivers/net/ethernet/sun/sunvnet.c +++ b/drivers/net/ethernet/sun/sunvnet.c @@ -350,14 +350,17 @@ static int vnet_walk_rx_one(struct vnet_port *port, if (IS_ERR(desc)) return PTR_ERR(desc); + if (desc->hdr.state != VIO_DESC_READY) + return 1; + + rmb(); + viodbg(DATA, "vio_walk_rx_one desc[%02x:%02x:%08x:%08x:%llx:%llx]\n", desc->hdr.state, desc->hdr.ack, desc->size, desc->ncookies, desc->cookies[0].cookie_addr, desc->cookies[0].cookie_size); - if (desc->hdr.state != VIO_DESC_READY) - return 1; err = vnet_rx_one(port, desc->size, desc->cookies, desc->ncookies); if (err == -ECONNRESET) return err; -- cgit v1.2.3