| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Date: Thu, 31 Jul 2008 05:29:32 -0700 (PDT)
From: David Miller <davem@...emloft.net>
To: jarkao2@...il.com
Cc: johannes@...solutions.net, netdev@...eo.de, peterz@...radead.org,
Larry.Finger@...inger.net, kaber@...sh.net,
torvalds@...ux-foundation.org, akpm@...ux-foundation.org,
netdev@...r.kernel.org, linux-kernel@...r.kernel.org,
linux-wireless@...r.kernel.org, mingo@...hat.com
Subject: Re: Kernel WARNING: at net/core/dev.c:1330
__netif_schedule+0x2c/0x98()
From: Jarek Poplawski <jarkao2@...il.com>
Date: Sun, 27 Jul 2008 22:37:57 +0200
> Looks like enough to me. (Probably it could even share space with
> the state.)
So I made some progress on this, three things:
1) I remember why I choose a to use a bit in my design, it's so that
it does not increase the costs of the checks in the fast paths.
test_bit(X) && test_bit(Y) can be combined into a single test by
the compiler.
2) We can't use the reference counting scheme, because we don't want
to let a second cpu into these protected code paths just because
another is in the middle of using a freeze too.
3) So we can simply put a top-level TX spinlock around these things.
Therefore all the hot paths:
a) grab _xmit_lock
b) check XOFF and FROZEN
c) only call ->hard_start_xmit() if both bits are clear
netif_tx_lock() does:
1) grab netdev->tx_global_lock
2) for_each_tx_queue() {
lock(txq);
set_bit(FROZEN);
unlock(txq);
}
and unlock does:
1) for_each_tx_queue() {
clear_bit(FROZEN);
if (!test_bit(XOFF))
__netif_schedule();
}
2) release netdev->tx_global_lock
And this seems to satisfy all the constraints which are:
1) Must act like a lock and protect execution of the code path
which occurs inside of "netif_tx_{lock,unlock}()"
2) Must ensure no cpus are executing inside of ->hard_start_xmit()
after netif_tx_lock() returns.
3) Must not try to grab all the TX queue locks at once.
This top-level tx_global_lock also simplifies the freezing, as
it makes sure only one cpu is initiating or finishing a freeze
at any given time.
I've also adjusted code that really and truly only wanted to
lock one queue at a time, which in particular was IFB and the
teql scheduler.
It's late here, but I'll start testing the following patch on my
multiqueue capable cards after some sleep.
diff --git a/drivers/net/ifb.c b/drivers/net/ifb.c
index 0960e69..e4fbefc 100644
--- a/drivers/net/ifb.c
+++ b/drivers/net/ifb.c
@@ -69,18 +69,20 @@ static void ri_tasklet(unsigned long dev)
struct net_device *_dev = (struct net_device *)dev;
struct ifb_private *dp = netdev_priv(_dev);
struct net_device_stats *stats = &_dev->stats;
+ struct netdev_queue *txq;
struct sk_buff *skb;
+ txq = netdev_get_tx_queue(_dev, 0);
dp->st_task_enter++;
if ((skb = skb_peek(&dp->tq)) == NULL) {
dp->st_txq_refl_try++;
- if (netif_tx_trylock(_dev)) {
+ if (__netif_tx_trylock(txq)) {
dp->st_rxq_enter++;
while ((skb = skb_dequeue(&dp->rq)) != NULL) {
skb_queue_tail(&dp->tq, skb);
dp->st_rx2tx_tran++;
}
- netif_tx_unlock(_dev);
+ __netif_tx_unlock(txq);
} else {
/* reschedule */
dp->st_rxq_notenter++;
@@ -115,7 +117,7 @@ static void ri_tasklet(unsigned long dev)
BUG();
}
- if (netif_tx_trylock(_dev)) {
+ if (__netif_tx_trylock(txq)) {
dp->st_rxq_check++;
if ((skb = skb_peek(&dp->rq)) == NULL) {
dp->tasklet_pending = 0;
@@ -123,10 +125,10 @@ static void ri_tasklet(unsigned long dev)
netif_wake_queue(_dev);
} else {
dp->st_rxq_rsch++;
- netif_tx_unlock(_dev);
+ __netif_tx_unlock(txq);
goto resched;
}
- netif_tx_unlock(_dev);
+ __netif_tx_unlock(txq);
} else {
resched:
dp->tasklet_pending = 1;
diff --git a/include/linux/netdevice.h b/include/linux/netdevice.h
index b4d056c..ee583f6 100644
--- a/include/linux/netdevice.h
+++ b/include/linux/netdevice.h
@@ -440,6 +440,7 @@ static inline void napi_synchronize(const struct napi_struct *n)
enum netdev_queue_state_t
{
__QUEUE_STATE_XOFF,
+ __QUEUE_STATE_FROZEN,
};
struct netdev_queue {
@@ -636,7 +637,7 @@ struct net_device
unsigned int real_num_tx_queues;
unsigned long tx_queue_len; /* Max frames per queue allowed */
-
+ spinlock_t tx_global_lock;
/*
* One part is mostly used on xmit path (device)
*/
@@ -1099,6 +1100,11 @@ static inline int netif_queue_stopped(const struct net_device *dev)
return netif_tx_queue_stopped(netdev_get_tx_queue(dev, 0));
}
+static inline int netif_tx_queue_frozen(const struct netdev_queue *dev_queue)
+{
+ return test_bit(__QUEUE_STATE_FROZEN, &dev_queue->state);
+}
+
/**
* netif_running - test if up
* @dev: network device
@@ -1475,6 +1481,26 @@ static inline void __netif_tx_lock_bh(struct netdev_queue *txq)
txq->xmit_lock_owner = smp_processor_id();
}
+static inline int __netif_tx_trylock(struct netdev_queue *txq)
+{
+ int ok = spin_trylock(&txq->_xmit_lock);
+ if (likely(ok))
+ txq->xmit_lock_owner = smp_processor_id();
+ return ok;
+}
+
+static inline void __netif_tx_unlock(struct netdev_queue *txq)
+{
+ txq->xmit_lock_owner = -1;
+ spin_unlock(&txq->_xmit_lock);
+}
+
+static inline void __netif_tx_unlock_bh(struct netdev_queue *txq)
+{
+ txq->xmit_lock_owner = -1;
+ spin_unlock_bh(&txq->_xmit_lock);
+}
+
/**
* netif_tx_lock - grab network device transmit lock
* @dev: network device
@@ -1484,12 +1510,23 @@ static inline void __netif_tx_lock_bh(struct netdev_queue *txq)
*/
static inline void netif_tx_lock(struct net_device *dev)
{
- int cpu = smp_processor_id();
unsigned int i;
+ int cpu;
+ spin_lock(&dev->tx_global_lock);
+ cpu = smp_processor_id();
for (i = 0; i < dev->num_tx_queues; i++) {
struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
+
+ /* We are the only thread of execution doing a
+ * freeze, but we have to grab the _xmit_lock in
+ * order to synchronize with threads which are in
+ * the ->hard_start_xmit() handler and already
+ * checked the frozen bit.
+ */
__netif_tx_lock(txq, cpu);
+ set_bit(__QUEUE_STATE_FROZEN, &txq->state);
+ __netif_tx_unlock(txq);
}
}
@@ -1499,40 +1536,22 @@ static inline void netif_tx_lock_bh(struct net_device *dev)
netif_tx_lock(dev);
}
-static inline int __netif_tx_trylock(struct netdev_queue *txq)
-{
- int ok = spin_trylock(&txq->_xmit_lock);
- if (likely(ok))
- txq->xmit_lock_owner = smp_processor_id();
- return ok;
-}
-
-static inline int netif_tx_trylock(struct net_device *dev)
-{
- return __netif_tx_trylock(netdev_get_tx_queue(dev, 0));
-}
-
-static inline void __netif_tx_unlock(struct netdev_queue *txq)
-{
- txq->xmit_lock_owner = -1;
- spin_unlock(&txq->_xmit_lock);
-}
-
-static inline void __netif_tx_unlock_bh(struct netdev_queue *txq)
-{
- txq->xmit_lock_owner = -1;
- spin_unlock_bh(&txq->_xmit_lock);
-}
-
static inline void netif_tx_unlock(struct net_device *dev)
{
unsigned int i;
for (i = 0; i < dev->num_tx_queues; i++) {
struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
- __netif_tx_unlock(txq);
- }
+ /* No need to grab the _xmit_lock here. If the
+ * queue is not stopped for another reason, we
+ * force a schedule.
+ */
+ clear_bit(__QUEUE_STATE_FROZEN, &txq->state);
+ if (!test_bit(__QUEUE_STATE_XOFF, &txq->state))
+ __netif_schedule(txq->qdisc);
+ }
+ spin_unlock(&dev->tx_global_lock);
}
static inline void netif_tx_unlock_bh(struct net_device *dev)
@@ -1556,13 +1575,18 @@ static inline void netif_tx_unlock_bh(struct net_device *dev)
static inline void netif_tx_disable(struct net_device *dev)
{
unsigned int i;
+ int cpu;
- netif_tx_lock_bh(dev);
+ local_bh_disable();
+ cpu = smp_processor_id();
for (i = 0; i < dev->num_tx_queues; i++) {
struct netdev_queue *txq = netdev_get_tx_queue(dev, i);
+
+ __netif_tx_lock(txq, cpu);
netif_tx_stop_queue(txq);
+ __netif_tx_unlock(txq);
}
- netif_tx_unlock_bh(dev);
+ local_bh_enable();
}
static inline void netif_addr_lock(struct net_device *dev)
diff --git a/net/core/dev.c b/net/core/dev.c
index 63d6bcd..69320a5 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -4200,6 +4200,7 @@ static void netdev_init_queues(struct net_device *dev)
{
netdev_init_one_queue(dev, &dev->rx_queue, NULL);
netdev_for_each_tx_queue(dev, netdev_init_one_queue, NULL);
+ spin_lock_init(&dev->tx_global_lock);
}
/**
diff --git a/net/core/netpoll.c b/net/core/netpoll.c
index c127208..6c7af39 100644
--- a/net/core/netpoll.c
+++ b/net/core/netpoll.c
@@ -70,6 +70,7 @@ static void queue_process(struct work_struct *work)
local_irq_save(flags);
__netif_tx_lock(txq, smp_processor_id());
if (netif_tx_queue_stopped(txq) ||
+ netif_tx_queue_frozen(txq) ||
dev->hard_start_xmit(skb, dev) != NETDEV_TX_OK) {
skb_queue_head(&npinfo->txq, skb);
__netif_tx_unlock(txq);
diff --git a/net/core/pktgen.c b/net/core/pktgen.c
index c7d484f..3284605 100644
--- a/net/core/pktgen.c
+++ b/net/core/pktgen.c
@@ -3305,6 +3305,7 @@ static __inline__ void pktgen_xmit(struct pktgen_dev *pkt_dev)
txq = netdev_get_tx_queue(odev, queue_map);
if (netif_tx_queue_stopped(txq) ||
+ netif_tx_queue_frozen(txq) ||
need_resched()) {
idle_start = getCurUs();
@@ -3320,7 +3321,8 @@ static __inline__ void pktgen_xmit(struct pktgen_dev *pkt_dev)
pkt_dev->idle_acc += getCurUs() - idle_start;
- if (netif_tx_queue_stopped(txq)) {
+ if (netif_tx_queue_stopped(txq) ||
+ netif_tx_queue_frozen(txq)) {
pkt_dev->next_tx_us = getCurUs(); /* TODO */
pkt_dev->next_tx_ns = 0;
goto out; /* Try the next interface */
@@ -3352,7 +3354,8 @@ static __inline__ void pktgen_xmit(struct pktgen_dev *pkt_dev)
txq = netdev_get_tx_queue(odev, queue_map);
__netif_tx_lock_bh(txq);
- if (!netif_tx_queue_stopped(txq)) {
+ if (!netif_tx_queue_stopped(txq) &&
+ !netif_tx_queue_frozen(txq)) {
atomic_inc(&(pkt_dev->skb->users));
retry_now:
diff --git a/net/sched/sch_generic.c b/net/sched/sch_generic.c
index 345838a..9c9cd4d 100644
--- a/net/sched/sch_generic.c
+++ b/net/sched/sch_generic.c
@@ -135,7 +135,8 @@ static inline int qdisc_restart(struct Qdisc *q)
txq = netdev_get_tx_queue(dev, skb_get_queue_mapping(skb));
HARD_TX_LOCK(dev, txq, smp_processor_id());
- if (!netif_subqueue_stopped(dev, skb))
+ if (!netif_tx_queue_stopped(txq) &&
+ !netif_tx_queue_frozen(txq))
ret = dev_hard_start_xmit(skb, dev, txq);
HARD_TX_UNLOCK(dev, txq);
@@ -162,7 +163,8 @@ static inline int qdisc_restart(struct Qdisc *q)
break;
}
- if (ret && netif_tx_queue_stopped(txq))
+ if (ret && (netif_tx_queue_stopped(txq) ||
+ netif_tx_queue_frozen(txq)))
ret = 0;
return ret;
diff --git a/net/sched/sch_teql.c b/net/sched/sch_teql.c
index 5372236..2c35c67 100644
--- a/net/sched/sch_teql.c
+++ b/net/sched/sch_teql.c
@@ -305,10 +305,11 @@ restart:
switch (teql_resolve(skb, skb_res, slave)) {
case 0:
- if (netif_tx_trylock(slave)) {
- if (!__netif_subqueue_stopped(slave, subq) &&
+ if (__netif_tx_trylock(slave_txq)) {
+ if (!netif_tx_queue_stopped(slave_txq) &&
+ !netif_tx_queue_frozen(slave_txq) &&
slave->hard_start_xmit(skb, slave) == 0) {
- netif_tx_unlock(slave);
+ __netif_tx_unlock(slave_txq);
master->slaves = NEXT_SLAVE(q);
netif_wake_queue(dev);
master->stats.tx_packets++;
@@ -316,7 +317,7 @@ restart:
qdisc_pkt_len(skb);
return 0;
}
- netif_tx_unlock(slave);
+ __netif_tx_unlock(slave_txq);
}
if (netif_queue_stopped(dev))
busy = 1;
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
Powered by blists - more mailing lists