Currently pfifo_fast has both TCQ_F_CAN_BYPASS and TCQ_F_NOLOCK
flag set, but queue discipline by-pass does not work for lockless
qdisc because skb is always enqueued to qdisc even when the qdisc
is empty, see __dev_xmit_skb().
This patch calls sch_direct_xmit() to transmit the skb directly
to the driver for empty lockless qdisc, which aviod enqueuing
and dequeuing operation.
As qdisc->empty is not reliable to indicate a empty qdisc because
there is a time window between enqueuing and setting qdisc->empty.
So we use the MISSED state added in commit a90c57f2ce ("net:
sched: fix packet stuck problem for lockless qdisc"), which
indicate there is lock contention, suggesting that it is better
not to do the qdisc bypass in order to avoid packet out of order
problem.
In order to make MISSED state reliable to indicate a empty qdisc,
we need to ensure that testing and clearing of MISSED state is
within the protection of qdisc->seqlock, only setting MISSED state
can be done without the protection of qdisc->seqlock. A MISSED
state testing is added without the protection of qdisc->seqlock to
aviod doing unnecessary spin_trylock() for contention case.
As the enqueuing is not within the protection of qdisc->seqlock,
there is still a potential data race as mentioned by Jakub [1]:
thread1 thread2 thread3
qdisc_run_begin() # true
qdisc_run_begin(q)
set(MISSED)
pfifo_fast_dequeue
clear(MISSED)
# recheck the queue
qdisc_run_end()
enqueue skb1
qdisc empty # true
qdisc_run_begin() # true
sch_direct_xmit() # skb2
qdisc_run_begin()
set(MISSED)
When above happens, skb1 enqueued by thread2 is transmited after
skb2 is transmited by thread3 because MISSED state setting and
enqueuing is not under the qdisc->seqlock. If qdisc bypass is
disabled, skb1 has better chance to be transmited quicker than
skb2.
This patch does not take care of the above data race, because we
view this as similar as below:
Even at the same time CPU1 and CPU2 write the skb to two socket
which both heading to the same qdisc, there is no guarantee that
which skb will hit the qdisc first, because there is a lot of
factor like interrupt/softirq/cache miss/scheduling afffecting
that.
There are below cases that need special handling:
1. When MISSED state is cleared before another round of dequeuing
in pfifo_fast_dequeue(), and __qdisc_run() might not be able to
dequeue all skb in one round and call __netif_schedule(), which
might result in a non-empty qdisc without MISSED set. In order
to avoid this, the MISSED state is set for lockless qdisc and
__netif_schedule() will be called at the end of qdisc_run_end.
2. The MISSED state also need to be set for lockless qdisc instead
of calling __netif_schedule() directly when requeuing a skb for
a similar reason.
3. For netdev queue stopped case, the MISSED case need clearing
while the netdev queue is stopped, otherwise there may be
unnecessary __netif_schedule() calling. So a new DRAINING state
is added to indicate this case, which also indicate a non-empty
qdisc.
4. As there is already netif_xmit_frozen_or_stopped() checking in
dequeue_skb() and sch_direct_xmit(), which are both within the
protection of qdisc->seqlock, but the same checking in
__dev_xmit_skb() is without the protection, which might cause
empty indication of a lockless qdisc to be not reliable. So
remove the checking in __dev_xmit_skb(), and the checking in
the protection of qdisc->seqlock seems enough to avoid the cpu
consumption problem for netdev queue stopped case.
1. https://lkml.org/lkml/2021/5/29/215
Acked-by: Jakub Kicinski <kuba@kernel.org>
Tested-by: Vladimir Oltean <vladimir.oltean@nxp.com> # flexcan
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
mlx5 devices were observed generating MLX5_PORT_CHANGE_SUBTYPE_ACTIVE
events without an intervening MLX5_PORT_CHANGE_SUBTYPE_DOWN. This
breaks link flap detection based on Linux carrier state transition
count as netif_carrier_on() does nothing if carrier is already on.
Make sure we count such events.
netif_carrier_event() increments the counters and fires the linkwatch
events. The latter is not necessary for the use case but seems like
the right thing to do.
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Saeed Mahameed <saeedm@nvidia.com>
The netdev qeueue might be stopped when byte queue limit has
reached or tx hw ring is full, net_tx_action() may still be
rescheduled if STATE_MISSED is set, which consumes unnecessary
cpu without dequeuing and transmiting any skb because the
netdev queue is stopped, see qdisc_run_end().
This patch fixes it by checking the netdev queue state before
calling qdisc_run() and clearing STATE_MISSED if netdev queue is
stopped during qdisc_run(), the net_tx_action() is rescheduled
again when netdev qeueue is restarted, see netif_tx_wake_queue().
As there is time window between netif_xmit_frozen_or_stopped()
checking and STATE_MISSED clearing, between which STATE_MISSED
may set by net_tx_action() scheduled by netif_tx_wake_queue(),
so set the STATE_MISSED again if netdev queue is restarted.
Fixes: 6b3ba9146f ("net: sched: allow qdiscs to handle locking")
Reported-by: Michal Kubecek <mkubecek@suse.cz>
Acked-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently qdisc_run() checks the STATE_DEACTIVATED of lockless
qdisc before calling __qdisc_run(), which ultimately clear the
STATE_MISSED when all the skb is dequeued. If STATE_DEACTIVATED
is set before clearing STATE_MISSED, there may be rescheduling
of net_tx_action() at the end of qdisc_run_end(), see below:
CPU0(net_tx_atcion) CPU1(__dev_xmit_skb) CPU2(dev_deactivate)
. . .
. set STATE_MISSED .
. __netif_schedule() .
. . set STATE_DEACTIVATED
. . qdisc_reset()
. . .
.<--------------- . synchronize_net()
clear __QDISC_STATE_SCHED | . .
. | . .
. | . some_qdisc_is_busy()
. | . return *false*
. | . .
test STATE_DEACTIVATED | . .
__qdisc_run() *not* called | . .
. | . .
test STATE_MISS | . .
__netif_schedule()--------| . .
. . .
. . .
__qdisc_run() is not called by net_tx_atcion() in CPU0 because
CPU2 has set STATE_DEACTIVATED flag during dev_deactivate(), and
STATE_MISSED is only cleared in __qdisc_run(), __netif_schedule
is called at the end of qdisc_run_end(), causing tx action
rescheduling problem.
qdisc_run() called by net_tx_action() runs in the softirq context,
which should has the same semantic as the qdisc_run() called by
__dev_xmit_skb() protected by rcu_read_lock_bh(). And there is a
synchronize_net() between STATE_DEACTIVATED flag being set and
qdisc_reset()/some_qdisc_is_busy in dev_deactivate(), we can safely
bail out for the deactived lockless qdisc in net_tx_action(), and
qdisc_reset() will reset all skb not dequeued yet.
So add the rcu_read_lock() explicitly to protect the qdisc_run()
and do the STATE_DEACTIVATED checking in net_tx_action() before
calling qdisc_run_begin(). Another option is to do the checking in
the qdisc_run_end(), but it will add unnecessary overhead for
non-tx_action case, because __dev_queue_xmit() will not see qdisc
with STATE_DEACTIVATED after synchronize_net(), the qdisc with
STATE_DEACTIVATED can only be seen by net_tx_action() because of
__netif_schedule().
The STATE_DEACTIVATED checking in qdisc_run() is to avoid race
between net_tx_action() and qdisc_reset(), see:
commit d518d2ed86 ("net/sched: fix race between deactivation
and dequeue for NOLOCK qdisc"). As the bailout added above for
deactived lockless qdisc in net_tx_action() provides better
protection for the race without calling qdisc_run() at all, so
remove the STATE_DEACTIVATED checking in qdisc_run().
After qdisc_reset(), there is no skb in qdisc to be dequeued, so
clear the STATE_MISSED in dev_reset_queue() too.
Fixes: 6b3ba9146f ("net: sched: allow qdiscs to handle locking")
Acked-by: Jakub Kicinski <kuba@kernel.org>
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
V8: Clearing STATE_MISSED before calling __netif_schedule() has
avoid the endless rescheduling problem, but there may still
be a unnecessary rescheduling, so adjust the commit log.
Signed-off-by: David S. Miller <davem@davemloft.net>
Lockless qdisc has below concurrent problem:
cpu0 cpu1
. .
q->enqueue .
. .
qdisc_run_begin() .
. .
dequeue_skb() .
. .
sch_direct_xmit() .
. .
. q->enqueue
. qdisc_run_begin()
. return and do nothing
. .
qdisc_run_end() .
cpu1 enqueue a skb without calling __qdisc_run() because cpu0
has not released the lock yet and spin_trylock() return false
for cpu1 in qdisc_run_begin(), and cpu0 do not see the skb
enqueued by cpu1 when calling dequeue_skb() because cpu1 may
enqueue the skb after cpu0 calling dequeue_skb() and before
cpu0 calling qdisc_run_end().
Lockless qdisc has below another concurrent problem when
tx_action is involved:
cpu0(serving tx_action) cpu1 cpu2
. . .
. q->enqueue .
. qdisc_run_begin() .
. dequeue_skb() .
. . q->enqueue
. . .
. sch_direct_xmit() .
. . qdisc_run_begin()
. . return and do nothing
. . .
clear __QDISC_STATE_SCHED . .
qdisc_run_begin() . .
return and do nothing . .
. . .
. qdisc_run_end() .
This patch fixes the above data race by:
1. If the first spin_trylock() return false and STATE_MISSED is
not set, set STATE_MISSED and retry another spin_trylock() in
case other CPU may not see STATE_MISSED after it releases the
lock.
2. reschedule if STATE_MISSED is set after the lock is released
at the end of qdisc_run_end().
For tx_action case, STATE_MISSED is also set when cpu1 is at the
end if qdisc_run_end(), so tx_action will be rescheduled again
to dequeue the skb enqueued by cpu2.
Clear STATE_MISSED before retrying a dequeuing when dequeuing
returns NULL in order to reduce the overhead of the second
spin_trylock() and __netif_schedule() calling.
Also clear the STATE_MISSED before calling __netif_schedule()
at the end of qdisc_run_end() to avoid doing another round of
dequeuing in the pfifo_fast_dequeue().
The performance impact of this patch, tested using pktgen and
dummy netdev with pfifo_fast qdisc attached:
threads without+this_patch with+this_patch delta
1 2.61Mpps 2.60Mpps -0.3%
2 3.97Mpps 3.82Mpps -3.7%
4 5.62Mpps 5.59Mpps -0.5%
8 2.78Mpps 2.77Mpps -0.3%
16 2.22Mpps 2.22Mpps -0.0%
Fixes: 6b3ba9146f ("net: sched: allow qdiscs to handle locking")
Acked-by: Jakub Kicinski <kuba@kernel.org>
Tested-by: Juergen Gross <jgross@suse.com>
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Allow a policer action to enforce a rate-limit based on packets-per-second,
configurable using a packet-per-second rate and burst parameters.
e.g.
tc filter add dev tap1 parent ffff: u32 match \
u32 0 0 police pkts_rate 3000 pkts_burst 1000
Testing was unable to uncover a performance impact of this change on
existing features.
Signed-off-by: Baowen Zheng <baowen.zheng@corigine.com>
Signed-off-by: Simon Horman <simon.horman@netronome.com>
Signed-off-by: Louis Peens <louis.peens@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
kmalloc() of sufficiently big portion of memory is cache-aligned
in regular conditions. If some debugging options are used,
there is no reason qdisc structures would need 64-byte alignment
if most other kernel structures are not aligned.
This get rid of QDISC_ALIGN and QDISC_ALIGNTO.
Addition of privdata field will help implementing
the reverse of qdisc_priv() and documents where
the private data is.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Allen Pais <allen.lkml@gmail.com>
Acked-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Currently there is concurrent reset and enqueue operation for the
same lockless qdisc when there is no lock to synchronize the
q->enqueue() in __dev_xmit_skb() with the qdisc reset operation in
qdisc_deactivate() called by dev_deactivate_queue(), which may cause
out-of-bounds access for priv->ring[] in hns3 driver if user has
requested a smaller queue num when __dev_xmit_skb() still enqueue a
skb with a larger queue_mapping after the corresponding qdisc is
reset, and call hns3_nic_net_xmit() with that skb later.
Reused the existing synchronize_net() in dev_deactivate_many() to
make sure skb with larger queue_mapping enqueued to old qdisc(which
is saved in dev_queue->qdisc_sleeping) will always be reset when
dev_reset_queue() is called.
Fixes: 6b3ba9146f ("net: sched: allow qdiscs to handle locking")
Signed-off-by: Yunsheng Lin <linyunsheng@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since the quiesce/activate rework, __netdev_watchdog_up() is directly
called in the ucc_geth driver.
Unfortunately, this function is not available for modules and thus
ucc_geth cannot be built as a module anymore. Fix it by exporting
__netdev_watchdog_up().
Since the commit introducing the regression was backported to stable
branches, this one should ideally be as well.
Fixes: 79dde73cf9 ("net/ethernet/freescale: rework quiesce/activate for ucc_geth")
Signed-off-by: Valentin Longchamp <valentin@longchamp.me>
Signed-off-by: David S. Miller <davem@davemloft.net>
Resetting old qdisc on dev_queue->qdisc_sleeping in
dev_qdisc_reset() is redundant, because this qdisc,
even if not same with dev_queue->qdisc, is reset via
qdisc_put() right after calling dev_graft_qdisc() when
hitting refcnt 0.
This is very easy to observe with qdisc_reset() tracepoint
and stack traces.
Reported-by: Václav Zindulka <vaclav.zindulka@tlapnet.cz>
Tested-by: Václav Zindulka <vaclav.zindulka@tlapnet.cz>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Except for sch_mq and sch_mqprio, each dev queue points to the
same root qdisc, so when we reset the dev queues with
netdev_for_each_tx_queue() we end up resetting the same instance
of the root qdisc for multiple times.
Avoid this by checking the __QDISC_STATE_DEACTIVATED bit in
each iteration, so for sch_mq/sch_mqprio, we still reset all
of them like before, for the rest, we only reset it once.
Reported-by: Václav Zindulka <vaclav.zindulka@tlapnet.cz>
Tested-by: Václav Zindulka <vaclav.zindulka@tlapnet.cz>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
qdisc_destroy() calls ops->reset() and cleans up qdisc->gso_skb
and qdisc->skb_bad_txq, these are nearly same with qdisc_reset(),
so just call it directly, and cosolidate the code for the next
patch.
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This patch reverts the folowing commits:
commit 064ff66e2b
"bonding: add missing netdev_update_lockdep_key()"
commit 53d374979e
"net: avoid updating qdisc_xmit_lock_key in netdev_update_lockdep_key()"
commit 1f26c0d3d2
"net: fix kernel-doc warning in <linux/netdevice.h>"
commit ab92d68fc2
"net: core: add generic lockdep keys"
but keeps the addr_list_lock_key because we still lock
addr_list_lock nestedly on stack devices, unlikely xmit_lock
this is safe because we don't take addr_list_lock on any fast
path.
Reported-and-tested-by: syzbot+aaa6fa4949cc5d9b7b25@syzkaller.appspotmail.com
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently if the default qdisc setup/init fails, the device ends up with
qdisc "noop", which causes all TX packets to get dropped.
With the introduction of sysctl net/core/default_qdisc it is possible
to change the default qdisc to be more advanced, which opens for the
possibility that Qdisc_ops->init() can fail.
This patch detect these kind of failures, and choose to fallback to
qdisc "noqueue", which is so simple that its init call will not fail.
This allows the interface to continue functioning.
V2:
As this also captures memory failures, which are transient, the
device is not kept in IFF_NO_QUEUE state. This allows the net_device
to retry to default qdisc assignment.
Signed-off-by: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
On ingress and cls_act qdiscs init, save the block on ingress
mini_Qdisc and and pass it on to ingress classification, so it
can be used for the looking up a specified chain index.
Co-developed-by: Vlad Buslov <vladbu@mellanox.com>
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Signed-off-by: Paul Blakey <paulb@mellanox.com>
Reviewed-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Saeed Mahameed <saeedm@mellanox.com>
The only slightly tricky merge conflict was the netdevsim because the
mutex locking fix overlapped a lot of driver reload reorganization.
The rest were (relatively) trivial in nature.
Signed-off-by: David S. Miller <davem@davemloft.net>
There is networking hardware that isn't based on Ethernet for layers 1 and 2.
For example CAN.
CAN is a multi-master serial bus standard for connecting Electronic Control
Units [ECUs] also known as nodes. A frame on the CAN bus carries up to 8 bytes
of payload. Frame corruption is detected by a CRC. However frame loss due to
corruption is possible, but a quite unusual phenomenon.
While fq_codel works great for TCP/IP, it doesn't for CAN. There are a lot of
legacy protocols on top of CAN, which are not build with flow control or high
CAN frame drop rates in mind.
When using fq_codel, as soon as the queue reaches a certain delay based length,
skbs from the head of the queue are silently dropped. Silently meaning that the
user space using a send() or similar syscall doesn't get an error. However
TCP's flow control algorithm will detect dropped packages and adjust the
bandwidth accordingly.
When using fq_codel and sending raw frames over CAN, which is the common use
case, the user space thinks the package has been sent without problems, because
send() returned without an error. pfifo_fast will drop skbs, if the queue
length exceeds the maximum. But with this scheduler the skbs at the tail are
dropped, an error (-ENOBUFS) is propagated to user space. So that the user
space can slow down the package generation.
On distributions, where fq_codel is made default via CONFIG_DEFAULT_NET_SCH
during compile time, or set default during runtime with sysctl
net.core.default_qdisc (see [1]), we get a bad user experience. In my test case
with pfifo_fast, I can transfer thousands of million CAN frames without a frame
drop. On the other hand with fq_codel there is more then one lost CAN frame per
thousand frames.
As pointed out fq_codel is not suited for CAN hardware, so this patch changes
attach_one_default_qdisc() to use pfifo_fast for "ARPHRD_CAN" network devices.
During transition of a netdev from down to up state the default queuing
discipline is attached by attach_default_qdiscs() with the help of
attach_one_default_qdisc(). This patch modifies attach_one_default_qdisc() to
attach the pfifo_fast (pfifo_fast_ops) if the network device type is
"ARPHRD_CAN".
[1] https://github.com/systemd/systemd/issues/9194
Suggested-by: Marc Kleine-Budde <mkl@pengutronix.de>
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
Signed-off-by: Vincent Prince <vincent.prince.fr@gmail.com>
Acked-by: Dave Taht <dave.taht@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
There is networking hardware that isn't based on Ethernet for layers 1 and 2.
For example CAN.
CAN is a multi-master serial bus standard for connecting Electronic Control
Units [ECUs] also known as nodes. A frame on the CAN bus carries up to 8 bytes
of payload. Frame corruption is detected by a CRC. However frame loss due to
corruption is possible, but a quite unusual phenomenon.
While fq_codel works great for TCP/IP, it doesn't for CAN. There are a lot of
legacy protocols on top of CAN, which are not build with flow control or high
CAN frame drop rates in mind.
When using fq_codel, as soon as the queue reaches a certain delay based length,
skbs from the head of the queue are silently dropped. Silently meaning that the
user space using a send() or similar syscall doesn't get an error. However
TCP's flow control algorithm will detect dropped packages and adjust the
bandwidth accordingly.
When using fq_codel and sending raw frames over CAN, which is the common use
case, the user space thinks the package has been sent without problems, because
send() returned without an error. pfifo_fast will drop skbs, if the queue
length exceeds the maximum. But with this scheduler the skbs at the tail are
dropped, an error (-ENOBUFS) is propagated to user space. So that the user
space can slow down the package generation.
On distributions, where fq_codel is made default via CONFIG_DEFAULT_NET_SCH
during compile time, or set default during runtime with sysctl
net.core.default_qdisc (see [1]), we get a bad user experience. In my test case
with pfifo_fast, I can transfer thousands of million CAN frames without a frame
drop. On the other hand with fq_codel there is more then one lost CAN frame per
thousand frames.
As pointed out fq_codel is not suited for CAN hardware, so this patch changes
attach_one_default_qdisc() to use pfifo_fast for "ARPHRD_CAN" network devices.
During transition of a netdev from down to up state the default queuing
discipline is attached by attach_default_qdiscs() with the help of
attach_one_default_qdisc(). This patch modifies attach_one_default_qdisc() to
attach the pfifo_fast (pfifo_fast_ops) if the network device type is
"ARPHRD_CAN".
[1] https://github.com/systemd/systemd/issues/9194
Signed-off-by: Vincent Prince <vincent.prince.fr@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Some interface types could be nested.
(VLAN, BONDING, TEAM, MACSEC, MACVLAN, IPVLAN, VIRT_WIFI, VXLAN, etc..)
These interface types should set lockdep class because, without lockdep
class key, lockdep always warn about unexisting circular locking.
In the current code, these interfaces have their own lockdep class keys and
these manage itself. So that there are so many duplicate code around the
/driver/net and /net/.
This patch adds new generic lockdep keys and some helper functions for it.
This patch does below changes.
a) Add lockdep class keys in struct net_device
- qdisc_running, xmit, addr_list, qdisc_busylock
- these keys are used as dynamic lockdep key.
b) When net_device is being allocated, lockdep keys are registered.
- alloc_netdev_mqs()
c) When net_device is being free'd llockdep keys are unregistered.
- free_netdev()
d) Add generic lockdep key helper function
- netdev_register_lockdep_key()
- netdev_unregister_lockdep_key()
- netdev_update_lockdep_key()
e) Remove unnecessary generic lockdep macro and functions
f) Remove unnecessary lockdep code of each interfaces.
After this patch, each interface modules don't need to maintain
their lockdep keys.
Signed-off-by: Taehee Yoo <ap420073@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
With threaded interrupts enabled, the interrupt thread runs as SCHED_RR
with priority 50. If a user application with a higher priority preempts
the interrupt thread and tries to shutdown the network interface then it
will loop forever. The kernel will spin in the loop waiting for the
device to become idle and the scheduler will never consider the
interrupt thread because its priority is lower.
Avoid the problem by sleeping for a jiffy giving other tasks,
including the interrupt thread, a chance to run and make progress.
In the original thread it has been suggested to use wait_event() and
properly waiting for the state to occur. DaveM explained that this would
require to add expensive checks in the fast paths of packet processing.
Link: https://lkml.kernel.org/r/1393976987-23555-1-git-send-email-mkl@pengutronix.de
Signed-off-by: Marc Kleine-Budde <mkl@pengutronix.de>
[bigeasy: Rewrite commit message, add comment, use
schedule_timeout_uninterruptible()]
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Acked-by: Cong Wang <xiyou.wangcong@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The introduction of this schedule point was done in commit
2ba2506ca7 ("[NET]: Add preemption point in qdisc_run")
at a time the loop was not bounded.
Then later in commit d5b8aa1d24 ("net_sched: fix dequeuer fairness")
we added a limit on the number of packets.
Now is the time to remove the schedule point, since the default
limit of 64 packets matches the number of packets a typical NAPI
poll can process in a row.
This solves a latency problem for most TCP receivers under moderate load :
1) host receives a packet.
NET_RX_SOFTIRQ is raised by NIC hard IRQ handler
2) __do_softirq() does its first loop, handling NET_RX_SOFTIRQ
and calling the driver napi->loop() function
3) TCP stores the skb in socket receive queue:
4) TCP calls sk->sk_data_ready() and wakeups a user thread
waiting for EPOLLIN (as a result, need_resched() might now be true)
5) TCP cooks an ACK and sends it.
6) qdisc_run() processes one packet from qdisc, and sees need_resched(),
this raises NET_TX_SOFTIRQ (even if there are no more packets in
the qdisc)
Then we go back to the __do_softirq() in 2), and we see that new
softirqs were raised. Since need_resched() is true, we end up waking
ksoftirqd in this path :
if (pending) {
if (time_before(jiffies, end) && !need_resched() &&
--max_restart)
goto restart;
wakeup_softirqd();
}
So we have many wakeups of ksoftirqd kernel threads,
and more calls to qdisc_run() with associated lock overhead.
Note that another way to solve the issue would be to change TCP
to first send the ACK packet, then signal the EPOLLIN,
but this changes P99 latencies, as sending the ACK packet
can add a long delay.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When tcf_block_get() fails in sfb_init(), q->qdisc is still a NULL
pointer which leads to a crash in sfb_destroy(). Similar for
sch_dsmark.
Instead of fixing each separately, Linus suggested to just accept
NULL pointer in qdisc_put(), which would make callers easier.
(For sch_dsmark, the bug probably exists long before commit
6529eaba33f0.)
Fixes: 6529eaba33 ("net: sched: introduce tcf block infractructure")
Reported-by: syzbot+d5870a903591faaca4ae@syzkaller.appspotmail.com
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Whenever MQ is not used on a multiqueue device, we experience
serious reordering problems. Bisection found the cited
commit.
The issue can be described this way :
- A single qdisc hierarchy is shared by all transmit queues.
(eg : tc qdisc replace dev eth0 root fq_codel)
- When/if try_bulk_dequeue_skb_slow() dequeues a packet targetting
a different transmit queue than the one used to build a packet train,
we stop building the current list and save the 'bad' skb (P1) in a
special queue. (bad_txq)
- When dequeue_skb() calls qdisc_dequeue_skb_bad_txq() and finds this
skb (P1), it checks if the associated transmit queues is still in frozen
state. If the queue is still blocked (by BQL or NIC tx ring full),
we leave the skb in bad_txq and return NULL.
- dequeue_skb() calls q->dequeue() to get another packet (P2)
The other packet can target the problematic queue (that we found
in frozen state for the bad_txq packet), but another cpu just ran
TX completion and made room in the txq that is now ready to accept
new packets.
- Packet P2 is sent while P1 is still held in bad_txq, P1 might be sent
at next round. In practice P2 is the lead of a big packet train
(P2,P3,P4 ...) filling the BQL budget and delaying P1 by many packets :/
To solve this problem, we have to block the dequeue process as long
as the first packet in bad_txq can not be sent. Reordering issues
disappear and no side effects have been seen.
Fixes: a53851e2c3 ("net: sched: explicit locking in gso_cpu fallback")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: John Fastabend <john.fastabend@gmail.com>
Acked-by: John Fastabend <john.fastabend@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that 'TCQ_F_CPUSTATS' bit can be cleared, depending on the value of
'TCQ_F_NOLOCK' bit in the parent qdisc, we need to be sure that per-cpu
counters are present when 'reset()' is called for pfifo_fast qdiscs.
Otherwise, the following script:
# tc q a dev lo handle 1: root htb default 100
# tc c a dev lo parent 1: classid 1:100 htb \
> rate 95Mbit ceil 100Mbit burst 64k
[...]
# tc f a dev lo parent 1: protocol arp basic classid 1:100
[...]
# tc q a dev lo parent 1:100 handle 100: pfifo_fast
[...]
# tc q d dev lo root
can generate the following splat:
Unable to handle kernel paging request at virtual address dfff2c01bd148000
Mem abort info:
ESR = 0x96000004
Exception class = DABT (current EL), IL = 32 bits
SET = 0, FnV = 0
EA = 0, S1PTW = 0
Data abort info:
ISV = 0, ISS = 0x00000004
CM = 0, WnR = 0
[dfff2c01bd148000] address between user and kernel address ranges
Internal error: Oops: 96000004 [#1] SMP
[...]
pstate: 80000005 (Nzcv daif -PAN -UAO)
pc : pfifo_fast_reset+0x280/0x4d8
lr : pfifo_fast_reset+0x21c/0x4d8
sp : ffff800d09676fa0
x29: ffff800d09676fa0 x28: ffff200012ee22e4
x27: dfff200000000000 x26: 0000000000000000
x25: ffff800ca0799958 x24: ffff1001940f332b
x23: 0000000000000007 x22: ffff200012ee1ab8
x21: 0000600de8a40000 x20: 0000000000000000
x19: ffff800ca0799900 x18: 0000000000000000
x17: 0000000000000002 x16: 0000000000000000
x15: 0000000000000000 x14: 0000000000000000
x13: 0000000000000000 x12: ffff1001b922e6e2
x11: 1ffff001b922e6e1 x10: 0000000000000000
x9 : 1ffff001b922e6e1 x8 : dfff200000000000
x7 : 0000000000000000 x6 : 0000000000000000
x5 : 1fffe400025dc45c x4 : 1fffe400025dc357
x3 : 00000c01bd148000 x2 : 0000600de8a40000
x1 : 0000000000000007 x0 : 0000600de8a40004
Call trace:
pfifo_fast_reset+0x280/0x4d8
qdisc_reset+0x6c/0x370
htb_reset+0x150/0x3b8 [sch_htb]
qdisc_reset+0x6c/0x370
dev_deactivate_queue.constprop.5+0xe0/0x1a8
dev_deactivate_many+0xd8/0x908
dev_deactivate+0xe4/0x190
qdisc_graft+0x88c/0xbd0
tc_get_qdisc+0x418/0x8a8
rtnetlink_rcv_msg+0x3a8/0xa78
netlink_rcv_skb+0x18c/0x328
rtnetlink_rcv+0x28/0x38
netlink_unicast+0x3c4/0x538
netlink_sendmsg+0x538/0x9a0
sock_sendmsg+0xac/0xf8
___sys_sendmsg+0x53c/0x658
__sys_sendmsg+0xc8/0x140
__arm64_sys_sendmsg+0x74/0xa8
el0_svc_handler+0x164/0x468
el0_svc+0x10/0x14
Code: 910012a0 92400801 d343fc03 11000c21 (38fb6863)
Fix this by testing the value of 'TCQ_F_CPUSTATS' bit in 'qdisc->flags',
before dereferencing 'qdisc->cpu_qstats'.
Changes since v1:
- coding style improvements, thanks to Stefano Brivio
Fixes: 8a53e616de ("net: sched: when clearing NOLOCK, clear TCQ_F_CPUSTATS, too")
CC: Paolo Abeni <pabeni@redhat.com>
Reported-by: Li Shuang <shuali@redhat.com>
Signed-off-by: Davide Caratti <dcaratti@redhat.com>
Acked-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Stefano Brivio <sbrivio@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Based on 1 normalized pattern(s):
this program is free software you can redistribute it and or modify
it under the terms of the gnu general public license as published by
the free software foundation either version 2 of the license or at
your option any later version
extracted by the scancode license scanner the SPDX license identifier
GPL-2.0-or-later
has been chosen to replace the boilerplate/reference in 3029 file(s).
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Allison Randal <allison@lohutok.net>
Cc: linux-spdx@vger.kernel.org
Link: https://lkml.kernel.org/r/20190527070032.746973796@linutronix.de
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Although devlink health report does a nice job on reporting TX
timeout and other NIC errors, unfortunately it requires drivers
to support it but currently only mlx5 has implemented it.
Before other drivers could catch up, it is useful to have a
generic tracepoint to monitor this kind of TX timeout. We have
been suffering TX timeout with different drivers, we plan to
start to monitor it with rasdaemon which just needs a new tracepoint.
Sample output:
ksoftirqd/1-16 [001] ..s2 144.043173: net_dev_xmit_timeout: dev=ens3 driver=e1000 queue=0
Cc: Eran Ben Elisha <eranbe@mellanox.com>
Cc: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: Cong Wang <xiyou.wangcong@gmail.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Reviewed-by: Eran Ben Elisha <eranbe@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
This revert commit 46b1c18f9d ("net: sched: put back q.qlen into
a single location").
After the previous patch, when a NOLOCK qdisc is enslaved to a
locking qdisc it switches to global stats accounting. As a consequence,
when a classful qdisc accesses directly a child qdisc's qlen, such
qdisc is not doing per CPU accounting and qlen value is consistent.
In the control path nobody uses directly qlen since commit
e5f0e8f8e4 ("net: sched: introduce and use qdisc tree flush/purge
helpers"), so we can remove the contented atomic ops from the
datapath.
v1 -> v2:
- complete the qdisc_qstats_atomic_qlen_dec() ->
qdisc_qstats_cpu_qlen_dec() replacement, fix build issue
- more descriptive commit message
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Since stats updating is always consistent with TCQ_F_CPUSTATS flag,
we can disable it at qdisc creation time flipping such bit.
In my experiments, if the NOLOCK flag is cleared, per CPU stats
accounting does not give any measurable performance gain, but it
waste some memory.
Let's clear TCQ_F_CPUSTATS together with NOLOCK, when enslaving
a NOLOCK qdisc to 'lock' one.
Use stats update helper inside pfifo_fast, to cope correctly with
TCQ_F_CPUSTATS flag change.
As a side effect, q.qlen value for any child qdiscs is always
consistent for all lock classfull qdiscs.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The core sched implementation checks independently for NOLOCK flag
to acquire/release the root spin lock and for qdisc_is_percpu_stats()
to account per CPU values in many places.
This change update the last few places checking the TCQ_F_NOLOCK to
do per CPU stats accounting according to qdisc_is_percpu_stats()
value.
The above allows to clean dev_requeue_skb() implementation a bit
and makes stats update always consistent with a single flag.
v1 -> v2:
- do not move qdisc_is_empty definition, fix build issue
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The queue is marked not empty after acquiring the seqlock,
and it's up to the NOLOCK qdisc clearing such flag on dequeue.
Since the empty status lays on the same cache-line of the
seqlock, it's always hot on cache during the updates.
This makes the empty flag update a little bit loosy. Given
the lack of synchronization between enqueue and dequeue, this
is unavoidable.
v2 -> v3:
- qdisc_is_empty() has a const argument (Eric)
v1 -> v2:
- use really an 'empty' flag instead of 'not_empty', as
suggested by Eric
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Reviewed-by: Eric Dumazet <edumazet@google.com>
Reviewed-by: Ivan Vecera <ivecera@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In the series fc8b81a598 ("Merge branch 'lockless-qdisc-series'")
John made the assumption that the data path had no need to read
the qdisc qlen (number of packets in the qdisc).
It is true when pfifo_fast is used as the root qdisc, or as direct MQ/MQPRIO
children.
But pfifo_fast can be used as leaf in class full qdiscs, and existing
logic needs to access the child qlen in an efficient way.
HTB breaks badly, since it uses cl->leaf.q->q.qlen in :
htb_activate() -> WARN_ON()
htb_dequeue_tree() to decide if a class can be htb_deactivated
when it has no more packets.
HFSC, DRR, CBQ, QFQ have similar issues, and some calls to
qdisc_tree_reduce_backlog() also read q.qlen directly.
Using qdisc_qlen_sum() (which iterates over all possible cpus)
in the data path is a non starter.
It seems we have to put back qlen in a central location,
at least for stable kernels.
For all qdisc but pfifo_fast, qlen is guarded by the qdisc lock,
so the existing q.qlen{++|--} are correct.
For 'lockless' qdisc (pfifo_fast so far), we need to use atomic_{inc|dec}()
because the spinlock might be not held (for example from
pfifo_fast_enqueue() and pfifo_fast_dequeue())
This patch adds atomic_qlen (in the same location than qlen)
and renames the following helpers, since we want to express
they can be used without qdisc lock, and that qlen is no longer percpu.
- qdisc_qstats_cpu_qlen_dec -> qdisc_qstats_atomic_qlen_dec()
- qdisc_qstats_cpu_qlen_inc -> qdisc_qstats_atomic_qlen_inc()
Later (net-next) we might revert this patch by tracking all these
qlen uses and replace them by a more efficient method (not having
to access a precise qlen, but an empty/non_empty status that might
be less expensive to maintain/track).
Another possibility is to have a legacy pfifo_fast version that would
be used when used a a child qdisc, since the parent qdisc needs
a spinlock anyway. But then, future lockless qdiscs would also
have the same problem.
Fixes: 7e66016f2c ("net: sched: helpers to sum qlen and qlen for per cpu logic")
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Cong Wang <xiyou.wangcong@gmail.com>
Cc: Jiri Pirko <jiri@resnulli.us>
Signed-off-by: David S. Miller <davem@davemloft.net>
This pointer is RCU protected, so proper primitives should be used.
Signed-off-by: Zhang Yu <zhangyu31@baidu.com>
Signed-off-by: Li RongQing <lirongqing@baidu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The netfilter conflicts were rather simple overlapping
changes.
However, the cls_tcindex.c stuff was a bit more complex.
On the 'net' side, Cong is fixing several races and memory
leaks. Whilst on the 'net-next' side we have Vlad adding
the rtnl-ness support.
What I've decided to do, in order to resolve this, is revert the
conversion over to using a workqueue that Cong did, bringing us back
to pure RCU. I did it this way because I believe that either Cong's
races don't apply with have Vlad did things, or Cong will have to
implement the race fix slightly differently.
Signed-off-by: David S. Miller <davem@davemloft.net>
Extend tcf_chain with new filter_chain_lock mutex. Always lock the chain
when accessing filter_chain list, instead of relying on rtnl lock.
Dereference filter_chain with tcf_chain_dereference() lockdep macro to
verify that all users of chain_list have the lock taken.
Rearrange tp insert/remove code in tc_new_tfilter/tc_del_tfilter to execute
all necessary code while holding chain lock in order to prevent
invalidation of chain_info structure by potential concurrent change. This
also serializes calls to tcf_chain0_head_change(), which allows head change
callbacks to rely on filter_chain_lock for synchronization instead of rtnl
mutex.
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Netlink has moved from bitmasks to group numbers long ago.
Signed-off-by: Jouke Witteveen <j.witteveen@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Now that call_rcu()'s callback is not invoked until after bh-disable
regions of code have completed (in addition to explicitly marked
RCU read-side critical sections), call_rcu() can be used in place
of call_rcu_bh(). Similarly, rcu_barrier() can be used in place o
frcu_barrier_bh(). This commit therefore makes these changes.
Signed-off-by: Paul E. McKenney <paulmck@linux.ibm.com>
Cc: Jamal Hadi Salim <jhs@mojatatu.com>
Cc: Cong Wang <xiyou.wangcong@gmail.com>
Cc: Jiri Pirko <jiri@resnulli.us>
Cc: "David S. Miller" <davem@davemloft.net>
Cc: <netdev@vger.kernel.org>
While noop_qdisc.gso_skb and noop_qdisc.skb_bad_txq are not used
in other places, it seems not correct to overwrite their fields
in dev_init_scheduler_queue().
noop_qdisc is essentially a shared and read-only object, even if
it is not marked as const because of some implementation detail.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Fixes the following sparse warning:
net/sched/sch_generic.c:944:6: warning:
symbol 'qdisc_free_cb' was not declared. Should it be static?
Fixes: 3a7d0d07a3 ("net: sched: extend Qdisc with rcu")
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Currently, Qdisc API functions assume that users have rtnl lock taken. To
implement rtnl unlocked classifiers update interface, Qdisc API must be
extended with functions that do not require rtnl lock.
Extend Qdisc structure with rcu. Implement special version of put function
qdisc_put_unlocked() that is called without rtnl lock taken. This function
only takes rtnl lock if Qdisc reference counter reached zero and is
intended to be used as optimization.
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Current implementation of qdisc_destroy() decrements Qdisc reference
counter and only actually destroy Qdisc if reference counter value reached
zero. Rename qdisc_destroy() to qdisc_put() in order for it to better
describe the way in which this function currently implemented and used.
Extract code that deallocates Qdisc into new private qdisc_destroy()
function. It is intended to be shared between regular qdisc_put() and its
unlocked version that is introduced in next patch in this series.
Signed-off-by: Vlad Buslov <vladbu@mellanox.com>
Acked-by: Jiri Pirko <jiri@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
An SKB is not on a list if skb->next is NULL.
Codify this convention into a helper function and use it
where we are dequeueing an SKB and need to mark it as such.
Signed-off-by: David S. Miller <davem@davemloft.net>
Checking netif_xmit_frozen_or_stopped() at the end of sch_direct_xmit()
is being bypassed. This is because "ret" from sch_direct_xmit() will be
either NETDEV_TX_OK or NETDEV_TX_BUSY, and only ret == NETDEV_TX_OK == 0
will reach the condition:
if (ret && netif_xmit_frozen_or_stopped(txq))
return false;
This patch cleans up the code by removing the whole condition.
For more discussion about this, please refer to
https://marc.info/?t=152727195700008
Signed-off-by: Song Liu <songliubraving@fb.com>
Cc: John Fastabend <john.fastabend@gmail.com>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: David S. Miller <davem@davemloft.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
After the previous patch, for NOLOCK qdiscs, q->seqlock is
always held when the dequeue() is invoked, we can drop
any additional locking to protect such operation.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
So that we can use lockdep on it.
The newly introduced sequence lock has the same scope of busylock,
so it shares the same lockdep annotation, but it's only used for
NOLOCK qdiscs.
With this changeset we acquire such lock in the control path around
flushing operation (qdisc reset), to allow more NOLOCK qdisc perf
improvement in the next patch.
Signed-off-by: Paolo Abeni <pabeni@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
