linux/kernel/softirq.c

765 lines
19 KiB
C
Raw Normal View History

/*
* linux/kernel/softirq.c
*
* Copyright (C) 1992 Linus Torvalds
*
* Distribute under GPLv2.
*
* Rewritten. Old one was good in 2.2, but in 2.3 it was immoral. --ANK (990903)
*/
#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
#include <linux/export.h>
#include <linux/kernel_stat.h>
#include <linux/interrupt.h>
#include <linux/init.h>
#include <linux/mm.h>
#include <linux/notifier.h>
#include <linux/percpu.h>
#include <linux/cpu.h>
#include <linux/freezer.h>
#include <linux/kthread.h>
#include <linux/rcupdate.h>
#include <linux/ftrace.h>
#include <linux/smp.h>
#include <linux/smpboot.h>
#include <linux/tick.h>
#include <linux/irq.h>
#define CREATE_TRACE_POINTS
#include <trace/events/irq.h>
/*
- No shared variables, all the data are CPU local.
- If a softirq needs serialization, let it serialize itself
by its own spinlocks.
- Even if softirq is serialized, only local cpu is marked for
execution. Hence, we get something sort of weak cpu binding.
Though it is still not clear, will it result in better locality
or will not.
Examples:
- NET RX softirq. It is multithreaded and does not require
any global serialization.
- NET TX softirq. It kicks software netdevice queues, hence
it is logically serialized per device, but this serialization
is invisible to common code.
- Tasklets: serialized wrt itself.
*/
#ifndef __ARCH_IRQ_STAT
DEFINE_PER_CPU_ALIGNED(irq_cpustat_t, irq_stat);
EXPORT_PER_CPU_SYMBOL(irq_stat);
#endif
static struct softirq_action softirq_vec[NR_SOFTIRQS] __cacheline_aligned_in_smp;
DEFINE_PER_CPU(struct task_struct *, ksoftirqd);
const char * const softirq_to_name[NR_SOFTIRQS] = {
"HI", "TIMER", "NET_TX", "NET_RX", "BLOCK", "IRQ_POLL",
rcu: Use softirq to address performance regression Commit a26ac2455ffcf3(rcu: move TREE_RCU from softirq to kthread) introduced performance regression. In an AIM7 test, this commit degraded performance by about 40%. The commit runs rcu callbacks in a kthread instead of softirq. We observed high rate of context switch which is caused by this. Out test system has 64 CPUs and HZ is 1000, so we saw more than 64k context switch per second which is caused by RCU's per-CPU kthread. A trace showed that most of the time the RCU per-CPU kthread doesn't actually handle any callbacks, but instead just does a very small amount of work handling grace periods. This means that RCU's per-CPU kthreads are making the scheduler do quite a bit of work in order to allow a very small amount of RCU-related processing to be done. Alex Shi's analysis determined that this slowdown is due to lock contention within the scheduler. Unfortunately, as Peter Zijlstra points out, the scheduler's real-time semantics require global action, which means that this contention is inherent in real-time scheduling. (Yes, perhaps someone will come up with a workaround -- otherwise, -rt is not going to do well on large SMP systems -- but this patch will work around this issue in the meantime. And "the meantime" might well be forever.) This patch therefore re-introduces softirq processing to RCU, but only for core RCU work. RCU callbacks are still executed in kthread context, so that only a small amount of RCU work runs in softirq context in the common case. This should minimize ksoftirqd execution, allowing us to skip boosting of ksoftirqd for CONFIG_RCU_BOOST=y kernels. Signed-off-by: Shaohua Li <shaohua.li@intel.com> Tested-by: "Alex,Shi" <alex.shi@intel.com> Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
2011-06-14 05:26:25 +00:00
"TASKLET", "SCHED", "HRTIMER", "RCU"
};
/*
* we cannot loop indefinitely here to avoid userspace starvation,
* but we also don't want to introduce a worst case 1/HZ latency
* to the pending events, so lets the scheduler to balance
* the softirq load for us.
*/
static void wakeup_softirqd(void)
{
/* Interrupts are disabled: no need to stop preemption */
struct task_struct *tsk = __this_cpu_read(ksoftirqd);
if (tsk && tsk->state != TASK_RUNNING)
wake_up_process(tsk);
}
softirq: Let ksoftirqd do its job A while back, Paolo and Hannes sent an RFC patch adding threaded-able napi poll loop support : (https://patchwork.ozlabs.org/patch/620657/) The problem seems to be that softirqs are very aggressive and are often handled by the current process, even if we are under stress and that ksoftirqd was scheduled, so that innocent threads would have more chance to make progress. This patch makes sure that if ksoftirq is running, we let it perform the softirq work. Jonathan Corbet summarized the issue in https://lwn.net/Articles/687617/ Tested: - NIC receiving traffic handled by CPU 0 - UDP receiver running on CPU 0, using a single UDP socket. - Incoming flood of UDP packets targeting the UDP socket. Before the patch, the UDP receiver could almost never get CPU cycles and could only receive ~2,000 packets per second. After the patch, CPU cycles are split 50/50 between user application and ksoftirqd/0, and we can effectively read ~900,000 packets per second, a huge improvement in DOS situation. (Note that more packets are now dropped by the NIC itself, since the BH handlers get less CPU cycles to drain RX ring buffer) Since the load runs in well identified threads context, an admin can more easily tune process scheduling parameters if needed. Reported-by: Paolo Abeni <pabeni@redhat.com> Reported-by: Hannes Frederic Sowa <hannes@stressinduktion.org> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: David Miller <davem@davemloft.net> Cc: Hannes Frederic Sowa <hannes@redhat.com> Cc: Jesper Dangaard Brouer <jbrouer@redhat.com> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Rik van Riel <riel@redhat.com> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/r/1472665349.14381.356.camel@edumazet-glaptop3.roam.corp.google.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-08-31 17:42:29 +00:00
/*
* If ksoftirqd is scheduled, we do not want to process pending softirqs
Mark HI and TASKLET softirq synchronous Way back in 4.9, we committed 4cd13c21b207 ("softirq: Let ksoftirqd do its job"), and ever since we've had small nagging issues with it. For example, we've had: 1ff688209e2e ("watchdog: core: make sure the watchdog_worker is not deferred") 8d5755b3f77b ("watchdog: softdog: fire watchdog even if softirqs do not get to run") 217f69743681 ("net: busy-poll: allow preemption in sk_busy_loop()") all of which worked around some of the effects of that commit. The DVB people have also complained that the commit causes excessive USB URB latencies, which seems to be due to the USB code using tasklets to schedule USB traffic. This seems to be an issue mainly when already living on the edge, but waiting for ksoftirqd to handle it really does seem to cause excessive latencies. Now Hanna Hawa reports that this issue isn't just limited to USB URB and DVB, but also causes timeout problems for the Marvell SoC team: "I'm facing kernel panic issue while running raid 5 on sata disks connected to Macchiatobin (Marvell community board with Armada-8040 SoC with 4 ARMv8 cores of CA72) Raid 5 built with Marvell DMA engine and async_tx mechanism (ASYNC_TX_DMA [=y]); the DMA driver (mv_xor_v2) uses a tasklet to clean the done descriptors from the queue" The latency problem causes a panic: mv_xor_v2 f0400000.xor: dma_sync_wait: timeout! Kernel panic - not syncing: async_tx_quiesce: DMA error waiting for transaction We've discussed simply just reverting the original commit entirely, and also much more involved solutions (with per-softirq threads etc). This patch is intentionally stupid and fairly limited, because the issue still remains, and the other solutions either got sidetracked or had other issues. We should probably also consider the timer softirqs to be synchronous and not be delayed to ksoftirqd (since they were the issue with the earlier watchdog problems), but that should be done as a separate patch. This does only the tasklet cases. Reported-and-tested-by: Hanna Hawa <hannah@marvell.com> Reported-and-tested-by: Josef Griebichler <griebichler.josef@gmx.at> Reported-by: Mauro Carvalho Chehab <mchehab@s-opensource.com> Cc: Alan Stern <stern@rowland.harvard.edu> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Eric Dumazet <edumazet@google.com> Cc: Ingo Molnar <mingo@kernel.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-01-08 19:51:04 +00:00
* right now. Let ksoftirqd handle this at its own rate, to get fairness,
* unless we're doing some of the synchronous softirqs.
softirq: Let ksoftirqd do its job A while back, Paolo and Hannes sent an RFC patch adding threaded-able napi poll loop support : (https://patchwork.ozlabs.org/patch/620657/) The problem seems to be that softirqs are very aggressive and are often handled by the current process, even if we are under stress and that ksoftirqd was scheduled, so that innocent threads would have more chance to make progress. This patch makes sure that if ksoftirq is running, we let it perform the softirq work. Jonathan Corbet summarized the issue in https://lwn.net/Articles/687617/ Tested: - NIC receiving traffic handled by CPU 0 - UDP receiver running on CPU 0, using a single UDP socket. - Incoming flood of UDP packets targeting the UDP socket. Before the patch, the UDP receiver could almost never get CPU cycles and could only receive ~2,000 packets per second. After the patch, CPU cycles are split 50/50 between user application and ksoftirqd/0, and we can effectively read ~900,000 packets per second, a huge improvement in DOS situation. (Note that more packets are now dropped by the NIC itself, since the BH handlers get less CPU cycles to drain RX ring buffer) Since the load runs in well identified threads context, an admin can more easily tune process scheduling parameters if needed. Reported-by: Paolo Abeni <pabeni@redhat.com> Reported-by: Hannes Frederic Sowa <hannes@stressinduktion.org> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: David Miller <davem@davemloft.net> Cc: Hannes Frederic Sowa <hannes@redhat.com> Cc: Jesper Dangaard Brouer <jbrouer@redhat.com> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Rik van Riel <riel@redhat.com> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/r/1472665349.14381.356.camel@edumazet-glaptop3.roam.corp.google.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-08-31 17:42:29 +00:00
*/
Mark HI and TASKLET softirq synchronous Way back in 4.9, we committed 4cd13c21b207 ("softirq: Let ksoftirqd do its job"), and ever since we've had small nagging issues with it. For example, we've had: 1ff688209e2e ("watchdog: core: make sure the watchdog_worker is not deferred") 8d5755b3f77b ("watchdog: softdog: fire watchdog even if softirqs do not get to run") 217f69743681 ("net: busy-poll: allow preemption in sk_busy_loop()") all of which worked around some of the effects of that commit. The DVB people have also complained that the commit causes excessive USB URB latencies, which seems to be due to the USB code using tasklets to schedule USB traffic. This seems to be an issue mainly when already living on the edge, but waiting for ksoftirqd to handle it really does seem to cause excessive latencies. Now Hanna Hawa reports that this issue isn't just limited to USB URB and DVB, but also causes timeout problems for the Marvell SoC team: "I'm facing kernel panic issue while running raid 5 on sata disks connected to Macchiatobin (Marvell community board with Armada-8040 SoC with 4 ARMv8 cores of CA72) Raid 5 built with Marvell DMA engine and async_tx mechanism (ASYNC_TX_DMA [=y]); the DMA driver (mv_xor_v2) uses a tasklet to clean the done descriptors from the queue" The latency problem causes a panic: mv_xor_v2 f0400000.xor: dma_sync_wait: timeout! Kernel panic - not syncing: async_tx_quiesce: DMA error waiting for transaction We've discussed simply just reverting the original commit entirely, and also much more involved solutions (with per-softirq threads etc). This patch is intentionally stupid and fairly limited, because the issue still remains, and the other solutions either got sidetracked or had other issues. We should probably also consider the timer softirqs to be synchronous and not be delayed to ksoftirqd (since they were the issue with the earlier watchdog problems), but that should be done as a separate patch. This does only the tasklet cases. Reported-and-tested-by: Hanna Hawa <hannah@marvell.com> Reported-and-tested-by: Josef Griebichler <griebichler.josef@gmx.at> Reported-by: Mauro Carvalho Chehab <mchehab@s-opensource.com> Cc: Alan Stern <stern@rowland.harvard.edu> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Eric Dumazet <edumazet@google.com> Cc: Ingo Molnar <mingo@kernel.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-01-08 19:51:04 +00:00
#define SOFTIRQ_NOW_MASK ((1 << HI_SOFTIRQ) | (1 << TASKLET_SOFTIRQ))
static bool ksoftirqd_running(unsigned long pending)
softirq: Let ksoftirqd do its job A while back, Paolo and Hannes sent an RFC patch adding threaded-able napi poll loop support : (https://patchwork.ozlabs.org/patch/620657/) The problem seems to be that softirqs are very aggressive and are often handled by the current process, even if we are under stress and that ksoftirqd was scheduled, so that innocent threads would have more chance to make progress. This patch makes sure that if ksoftirq is running, we let it perform the softirq work. Jonathan Corbet summarized the issue in https://lwn.net/Articles/687617/ Tested: - NIC receiving traffic handled by CPU 0 - UDP receiver running on CPU 0, using a single UDP socket. - Incoming flood of UDP packets targeting the UDP socket. Before the patch, the UDP receiver could almost never get CPU cycles and could only receive ~2,000 packets per second. After the patch, CPU cycles are split 50/50 between user application and ksoftirqd/0, and we can effectively read ~900,000 packets per second, a huge improvement in DOS situation. (Note that more packets are now dropped by the NIC itself, since the BH handlers get less CPU cycles to drain RX ring buffer) Since the load runs in well identified threads context, an admin can more easily tune process scheduling parameters if needed. Reported-by: Paolo Abeni <pabeni@redhat.com> Reported-by: Hannes Frederic Sowa <hannes@stressinduktion.org> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: David Miller <davem@davemloft.net> Cc: Hannes Frederic Sowa <hannes@redhat.com> Cc: Jesper Dangaard Brouer <jbrouer@redhat.com> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Rik van Riel <riel@redhat.com> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/r/1472665349.14381.356.camel@edumazet-glaptop3.roam.corp.google.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-08-31 17:42:29 +00:00
{
struct task_struct *tsk = __this_cpu_read(ksoftirqd);
Mark HI and TASKLET softirq synchronous Way back in 4.9, we committed 4cd13c21b207 ("softirq: Let ksoftirqd do its job"), and ever since we've had small nagging issues with it. For example, we've had: 1ff688209e2e ("watchdog: core: make sure the watchdog_worker is not deferred") 8d5755b3f77b ("watchdog: softdog: fire watchdog even if softirqs do not get to run") 217f69743681 ("net: busy-poll: allow preemption in sk_busy_loop()") all of which worked around some of the effects of that commit. The DVB people have also complained that the commit causes excessive USB URB latencies, which seems to be due to the USB code using tasklets to schedule USB traffic. This seems to be an issue mainly when already living on the edge, but waiting for ksoftirqd to handle it really does seem to cause excessive latencies. Now Hanna Hawa reports that this issue isn't just limited to USB URB and DVB, but also causes timeout problems for the Marvell SoC team: "I'm facing kernel panic issue while running raid 5 on sata disks connected to Macchiatobin (Marvell community board with Armada-8040 SoC with 4 ARMv8 cores of CA72) Raid 5 built with Marvell DMA engine and async_tx mechanism (ASYNC_TX_DMA [=y]); the DMA driver (mv_xor_v2) uses a tasklet to clean the done descriptors from the queue" The latency problem causes a panic: mv_xor_v2 f0400000.xor: dma_sync_wait: timeout! Kernel panic - not syncing: async_tx_quiesce: DMA error waiting for transaction We've discussed simply just reverting the original commit entirely, and also much more involved solutions (with per-softirq threads etc). This patch is intentionally stupid and fairly limited, because the issue still remains, and the other solutions either got sidetracked or had other issues. We should probably also consider the timer softirqs to be synchronous and not be delayed to ksoftirqd (since they were the issue with the earlier watchdog problems), but that should be done as a separate patch. This does only the tasklet cases. Reported-and-tested-by: Hanna Hawa <hannah@marvell.com> Reported-and-tested-by: Josef Griebichler <griebichler.josef@gmx.at> Reported-by: Mauro Carvalho Chehab <mchehab@s-opensource.com> Cc: Alan Stern <stern@rowland.harvard.edu> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Eric Dumazet <edumazet@google.com> Cc: Ingo Molnar <mingo@kernel.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-01-08 19:51:04 +00:00
if (pending & SOFTIRQ_NOW_MASK)
return false;
softirq: Let ksoftirqd do its job A while back, Paolo and Hannes sent an RFC patch adding threaded-able napi poll loop support : (https://patchwork.ozlabs.org/patch/620657/) The problem seems to be that softirqs are very aggressive and are often handled by the current process, even if we are under stress and that ksoftirqd was scheduled, so that innocent threads would have more chance to make progress. This patch makes sure that if ksoftirq is running, we let it perform the softirq work. Jonathan Corbet summarized the issue in https://lwn.net/Articles/687617/ Tested: - NIC receiving traffic handled by CPU 0 - UDP receiver running on CPU 0, using a single UDP socket. - Incoming flood of UDP packets targeting the UDP socket. Before the patch, the UDP receiver could almost never get CPU cycles and could only receive ~2,000 packets per second. After the patch, CPU cycles are split 50/50 between user application and ksoftirqd/0, and we can effectively read ~900,000 packets per second, a huge improvement in DOS situation. (Note that more packets are now dropped by the NIC itself, since the BH handlers get less CPU cycles to drain RX ring buffer) Since the load runs in well identified threads context, an admin can more easily tune process scheduling parameters if needed. Reported-by: Paolo Abeni <pabeni@redhat.com> Reported-by: Hannes Frederic Sowa <hannes@stressinduktion.org> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: David Miller <davem@davemloft.net> Cc: Hannes Frederic Sowa <hannes@redhat.com> Cc: Jesper Dangaard Brouer <jbrouer@redhat.com> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Rik van Riel <riel@redhat.com> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/r/1472665349.14381.356.camel@edumazet-glaptop3.roam.corp.google.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-08-31 17:42:29 +00:00
return tsk && (tsk->state == TASK_RUNNING);
}
/*
* preempt_count and SOFTIRQ_OFFSET usage:
* - preempt_count is changed by SOFTIRQ_OFFSET on entering or leaving
* softirq processing.
* - preempt_count is changed by SOFTIRQ_DISABLE_OFFSET (= 2 * SOFTIRQ_OFFSET)
* on local_bh_disable or local_bh_enable.
* This lets us distinguish between whether we are currently processing
* softirq and whether we just have bh disabled.
*/
/*
* This one is for softirq.c-internal use,
* where hardirqs are disabled legitimately:
*/
[PATCH] Reducing local_bh_enable/disable overhead in irqtrace The recent changes from irqtrace feature has added overheads to local_bh_disable and local_bh_enable that reduces UDP performance across x86_64 and IA64, even though IA64 does not support the irqtrace feature. Patch in question is [PATCH]lockdep: irqtrace subsystem, core http://www.kernel.org/git/?p=linux/kernel/git/torvalds/linux-2.6.git;a=c ommit;h=de30a2b355ea85350ca2f58f3b9bf4e5bc007986 Prior to this patch, local_bh_disable was a short macro. Now it is a function which calls __local_bh_disable with added irq flags save and restore. The irq flags save and restore were also added to local_bh_enable, probably for injecting the trace irqs code. This overhead is on the generic code path across all architectures. On a IA_64 test machine (Itanium-2 1.6 GHz) running a benchmark like netperf's UDP streaming test, the added overhead results in a drop of 3% in throughput, as udp_sendmsg calls the local_bh_enable/disable several times. Other workloads that have heavy usages of local_bh_enable/disable could also be affected. The patch ideally should not have affected IA-64 performance as it does not have IRQ tracing support. A significant portion of the overhead is in the added irq flags save and restore, which I think is not needed if IRQ tracing is unused. A suggested patch is attached below that recovers the lost performance. However, the "ifdef"s in the patch are a bit ugly. Signed-off-by: Tim Chen <tim.c.chen@intel.com> Acked-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-07-30 10:04:02 +00:00
#ifdef CONFIG_TRACE_IRQFLAGS
void __local_bh_disable_ip(unsigned long ip, unsigned int cnt)
{
unsigned long flags;
WARN_ON_ONCE(in_irq());
raw_local_irq_save(flags);
/*
* The preempt tracer hooks into preempt_count_add and will break
* lockdep because it calls back into lockdep after SOFTIRQ_OFFSET
* is set and before current->softirq_enabled is cleared.
* We must manually increment preempt_count here and manually
* call the trace_preempt_off later.
*/
__preempt_count_add(cnt);
/*
* Were softirqs turned off above:
*/
if (softirq_count() == (cnt & SOFTIRQ_MASK))
trace_softirqs_off(ip);
raw_local_irq_restore(flags);
if (preempt_count() == cnt) {
#ifdef CONFIG_DEBUG_PREEMPT
current->preempt_disable_ip = get_lock_parent_ip();
#endif
trace_preempt_off(CALLER_ADDR0, get_lock_parent_ip());
}
}
EXPORT_SYMBOL(__local_bh_disable_ip);
[PATCH] Reducing local_bh_enable/disable overhead in irqtrace The recent changes from irqtrace feature has added overheads to local_bh_disable and local_bh_enable that reduces UDP performance across x86_64 and IA64, even though IA64 does not support the irqtrace feature. Patch in question is [PATCH]lockdep: irqtrace subsystem, core http://www.kernel.org/git/?p=linux/kernel/git/torvalds/linux-2.6.git;a=c ommit;h=de30a2b355ea85350ca2f58f3b9bf4e5bc007986 Prior to this patch, local_bh_disable was a short macro. Now it is a function which calls __local_bh_disable with added irq flags save and restore. The irq flags save and restore were also added to local_bh_enable, probably for injecting the trace irqs code. This overhead is on the generic code path across all architectures. On a IA_64 test machine (Itanium-2 1.6 GHz) running a benchmark like netperf's UDP streaming test, the added overhead results in a drop of 3% in throughput, as udp_sendmsg calls the local_bh_enable/disable several times. Other workloads that have heavy usages of local_bh_enable/disable could also be affected. The patch ideally should not have affected IA-64 performance as it does not have IRQ tracing support. A significant portion of the overhead is in the added irq flags save and restore, which I think is not needed if IRQ tracing is unused. A suggested patch is attached below that recovers the lost performance. However, the "ifdef"s in the patch are a bit ugly. Signed-off-by: Tim Chen <tim.c.chen@intel.com> Acked-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-07-30 10:04:02 +00:00
#endif /* CONFIG_TRACE_IRQFLAGS */
static void __local_bh_enable(unsigned int cnt)
{
lockdep_assert_irqs_disabled();
softirq: Reorder trace_softirqs_on to prevent lockdep splat I'm able to reproduce a lockdep splat with config options: CONFIG_PROVE_LOCKING=y, CONFIG_DEBUG_LOCK_ALLOC=y and CONFIG_PREEMPTIRQ_EVENTS=y $ echo 1 > /d/tracing/events/preemptirq/preempt_enable/enable [ 26.112609] DEBUG_LOCKS_WARN_ON(current->softirqs_enabled) [ 26.112636] WARNING: CPU: 0 PID: 118 at kernel/locking/lockdep.c:3854 [...] [ 26.144229] Call Trace: [ 26.144926] <IRQ> [ 26.145506] lock_acquire+0x55/0x1b0 [ 26.146499] ? __do_softirq+0x46f/0x4d9 [ 26.147571] ? __do_softirq+0x46f/0x4d9 [ 26.148646] trace_preempt_on+0x8f/0x240 [ 26.149744] ? trace_preempt_on+0x4d/0x240 [ 26.150862] ? __do_softirq+0x46f/0x4d9 [ 26.151930] preempt_count_sub+0x18a/0x1a0 [ 26.152985] __do_softirq+0x46f/0x4d9 [ 26.153937] irq_exit+0x68/0xe0 [ 26.154755] smp_apic_timer_interrupt+0x271/0x280 [ 26.156056] apic_timer_interrupt+0xf/0x20 [ 26.157105] </IRQ> The issue was this: preempt_count = 1 << SOFTIRQ_SHIFT __local_bh_enable(cnt = 1 << SOFTIRQ_SHIFT) { if (softirq_count() == (cnt && SOFTIRQ_MASK)) { trace_softirqs_on() { current->softirqs_enabled = 1; } } preempt_count_sub(cnt) { trace_preempt_on() { tracepoint() { rcu_read_lock_sched() { // jumps into lockdep Where preempt_count still has softirqs disabled, but current->softirqs_enabled is true, and we get a splat. Link: http://lkml.kernel.org/r/20180607201143.247775-1-joel@joelfernandes.org Cc: Peter Zijlstra <peterz@infradead.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com> Cc: Tom Zanussi <tom.zanussi@linux.intel.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Thomas Glexiner <tglx@linutronix.de> Cc: Boqun Feng <boqun.feng@gmail.com> Cc: Paul McKenney <paulmck@linux.vnet.ibm.com> Cc: Masami Hiramatsu <mhiramat@kernel.org> Cc: Todd Kjos <tkjos@google.com> Cc: Erick Reyes <erickreyes@google.com> Cc: Julia Cartwright <julia@ni.com> Cc: Byungchul Park <byungchul.park@lge.com> Cc: stable@vger.kernel.org Reviewed-by: Steven Rostedt (VMware) <rostedt@goodmis.org> Fixes: d59158162e032 ("tracing: Add support for preempt and irq enable/disable events") Signed-off-by: Joel Fernandes (Google) <joel@joelfernandes.org> Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
2018-06-07 20:11:43 +00:00
if (preempt_count() == cnt)
trace_preempt_on(CALLER_ADDR0, get_lock_parent_ip());
if (softirq_count() == (cnt & SOFTIRQ_MASK))
trace_softirqs_on(_RET_IP_);
softirq: Reorder trace_softirqs_on to prevent lockdep splat I'm able to reproduce a lockdep splat with config options: CONFIG_PROVE_LOCKING=y, CONFIG_DEBUG_LOCK_ALLOC=y and CONFIG_PREEMPTIRQ_EVENTS=y $ echo 1 > /d/tracing/events/preemptirq/preempt_enable/enable [ 26.112609] DEBUG_LOCKS_WARN_ON(current->softirqs_enabled) [ 26.112636] WARNING: CPU: 0 PID: 118 at kernel/locking/lockdep.c:3854 [...] [ 26.144229] Call Trace: [ 26.144926] <IRQ> [ 26.145506] lock_acquire+0x55/0x1b0 [ 26.146499] ? __do_softirq+0x46f/0x4d9 [ 26.147571] ? __do_softirq+0x46f/0x4d9 [ 26.148646] trace_preempt_on+0x8f/0x240 [ 26.149744] ? trace_preempt_on+0x4d/0x240 [ 26.150862] ? __do_softirq+0x46f/0x4d9 [ 26.151930] preempt_count_sub+0x18a/0x1a0 [ 26.152985] __do_softirq+0x46f/0x4d9 [ 26.153937] irq_exit+0x68/0xe0 [ 26.154755] smp_apic_timer_interrupt+0x271/0x280 [ 26.156056] apic_timer_interrupt+0xf/0x20 [ 26.157105] </IRQ> The issue was this: preempt_count = 1 << SOFTIRQ_SHIFT __local_bh_enable(cnt = 1 << SOFTIRQ_SHIFT) { if (softirq_count() == (cnt && SOFTIRQ_MASK)) { trace_softirqs_on() { current->softirqs_enabled = 1; } } preempt_count_sub(cnt) { trace_preempt_on() { tracepoint() { rcu_read_lock_sched() { // jumps into lockdep Where preempt_count still has softirqs disabled, but current->softirqs_enabled is true, and we get a splat. Link: http://lkml.kernel.org/r/20180607201143.247775-1-joel@joelfernandes.org Cc: Peter Zijlstra <peterz@infradead.org> Cc: Ingo Molnar <mingo@redhat.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com> Cc: Tom Zanussi <tom.zanussi@linux.intel.com> Cc: Namhyung Kim <namhyung@kernel.org> Cc: Thomas Glexiner <tglx@linutronix.de> Cc: Boqun Feng <boqun.feng@gmail.com> Cc: Paul McKenney <paulmck@linux.vnet.ibm.com> Cc: Masami Hiramatsu <mhiramat@kernel.org> Cc: Todd Kjos <tkjos@google.com> Cc: Erick Reyes <erickreyes@google.com> Cc: Julia Cartwright <julia@ni.com> Cc: Byungchul Park <byungchul.park@lge.com> Cc: stable@vger.kernel.org Reviewed-by: Steven Rostedt (VMware) <rostedt@goodmis.org> Fixes: d59158162e032 ("tracing: Add support for preempt and irq enable/disable events") Signed-off-by: Joel Fernandes (Google) <joel@joelfernandes.org> Signed-off-by: Steven Rostedt (VMware) <rostedt@goodmis.org>
2018-06-07 20:11:43 +00:00
__preempt_count_sub(cnt);
}
/*
* Special-case - softirqs can safely be enabled by __do_softirq(),
* without processing still-pending softirqs:
*/
void _local_bh_enable(void)
{
WARN_ON_ONCE(in_irq());
__local_bh_enable(SOFTIRQ_DISABLE_OFFSET);
}
EXPORT_SYMBOL(_local_bh_enable);
void __local_bh_enable_ip(unsigned long ip, unsigned int cnt)
{
WARN_ON_ONCE(in_irq());
lockdep_assert_irqs_enabled();
[PATCH] Reducing local_bh_enable/disable overhead in irqtrace The recent changes from irqtrace feature has added overheads to local_bh_disable and local_bh_enable that reduces UDP performance across x86_64 and IA64, even though IA64 does not support the irqtrace feature. Patch in question is [PATCH]lockdep: irqtrace subsystem, core http://www.kernel.org/git/?p=linux/kernel/git/torvalds/linux-2.6.git;a=c ommit;h=de30a2b355ea85350ca2f58f3b9bf4e5bc007986 Prior to this patch, local_bh_disable was a short macro. Now it is a function which calls __local_bh_disable with added irq flags save and restore. The irq flags save and restore were also added to local_bh_enable, probably for injecting the trace irqs code. This overhead is on the generic code path across all architectures. On a IA_64 test machine (Itanium-2 1.6 GHz) running a benchmark like netperf's UDP streaming test, the added overhead results in a drop of 3% in throughput, as udp_sendmsg calls the local_bh_enable/disable several times. Other workloads that have heavy usages of local_bh_enable/disable could also be affected. The patch ideally should not have affected IA-64 performance as it does not have IRQ tracing support. A significant portion of the overhead is in the added irq flags save and restore, which I think is not needed if IRQ tracing is unused. A suggested patch is attached below that recovers the lost performance. However, the "ifdef"s in the patch are a bit ugly. Signed-off-by: Tim Chen <tim.c.chen@intel.com> Acked-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-07-30 10:04:02 +00:00
#ifdef CONFIG_TRACE_IRQFLAGS
local_irq_disable();
[PATCH] Reducing local_bh_enable/disable overhead in irqtrace The recent changes from irqtrace feature has added overheads to local_bh_disable and local_bh_enable that reduces UDP performance across x86_64 and IA64, even though IA64 does not support the irqtrace feature. Patch in question is [PATCH]lockdep: irqtrace subsystem, core http://www.kernel.org/git/?p=linux/kernel/git/torvalds/linux-2.6.git;a=c ommit;h=de30a2b355ea85350ca2f58f3b9bf4e5bc007986 Prior to this patch, local_bh_disable was a short macro. Now it is a function which calls __local_bh_disable with added irq flags save and restore. The irq flags save and restore were also added to local_bh_enable, probably for injecting the trace irqs code. This overhead is on the generic code path across all architectures. On a IA_64 test machine (Itanium-2 1.6 GHz) running a benchmark like netperf's UDP streaming test, the added overhead results in a drop of 3% in throughput, as udp_sendmsg calls the local_bh_enable/disable several times. Other workloads that have heavy usages of local_bh_enable/disable could also be affected. The patch ideally should not have affected IA-64 performance as it does not have IRQ tracing support. A significant portion of the overhead is in the added irq flags save and restore, which I think is not needed if IRQ tracing is unused. A suggested patch is attached below that recovers the lost performance. However, the "ifdef"s in the patch are a bit ugly. Signed-off-by: Tim Chen <tim.c.chen@intel.com> Acked-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-07-30 10:04:02 +00:00
#endif
/*
* Are softirqs going to be turned on now:
*/
if (softirq_count() == SOFTIRQ_DISABLE_OFFSET)
trace_softirqs_on(ip);
/*
* Keep preemption disabled until we are done with
* softirq processing:
*/
preempt_count_sub(cnt - 1);
if (unlikely(!in_interrupt() && local_softirq_pending())) {
/*
* Run softirq if any pending. And do it in its own stack
* as we may be calling this deep in a task call stack already.
*/
do_softirq();
}
preempt_count_dec();
[PATCH] Reducing local_bh_enable/disable overhead in irqtrace The recent changes from irqtrace feature has added overheads to local_bh_disable and local_bh_enable that reduces UDP performance across x86_64 and IA64, even though IA64 does not support the irqtrace feature. Patch in question is [PATCH]lockdep: irqtrace subsystem, core http://www.kernel.org/git/?p=linux/kernel/git/torvalds/linux-2.6.git;a=c ommit;h=de30a2b355ea85350ca2f58f3b9bf4e5bc007986 Prior to this patch, local_bh_disable was a short macro. Now it is a function which calls __local_bh_disable with added irq flags save and restore. The irq flags save and restore were also added to local_bh_enable, probably for injecting the trace irqs code. This overhead is on the generic code path across all architectures. On a IA_64 test machine (Itanium-2 1.6 GHz) running a benchmark like netperf's UDP streaming test, the added overhead results in a drop of 3% in throughput, as udp_sendmsg calls the local_bh_enable/disable several times. Other workloads that have heavy usages of local_bh_enable/disable could also be affected. The patch ideally should not have affected IA-64 performance as it does not have IRQ tracing support. A significant portion of the overhead is in the added irq flags save and restore, which I think is not needed if IRQ tracing is unused. A suggested patch is attached below that recovers the lost performance. However, the "ifdef"s in the patch are a bit ugly. Signed-off-by: Tim Chen <tim.c.chen@intel.com> Acked-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-07-30 10:04:02 +00:00
#ifdef CONFIG_TRACE_IRQFLAGS
local_irq_enable();
[PATCH] Reducing local_bh_enable/disable overhead in irqtrace The recent changes from irqtrace feature has added overheads to local_bh_disable and local_bh_enable that reduces UDP performance across x86_64 and IA64, even though IA64 does not support the irqtrace feature. Patch in question is [PATCH]lockdep: irqtrace subsystem, core http://www.kernel.org/git/?p=linux/kernel/git/torvalds/linux-2.6.git;a=c ommit;h=de30a2b355ea85350ca2f58f3b9bf4e5bc007986 Prior to this patch, local_bh_disable was a short macro. Now it is a function which calls __local_bh_disable with added irq flags save and restore. The irq flags save and restore were also added to local_bh_enable, probably for injecting the trace irqs code. This overhead is on the generic code path across all architectures. On a IA_64 test machine (Itanium-2 1.6 GHz) running a benchmark like netperf's UDP streaming test, the added overhead results in a drop of 3% in throughput, as udp_sendmsg calls the local_bh_enable/disable several times. Other workloads that have heavy usages of local_bh_enable/disable could also be affected. The patch ideally should not have affected IA-64 performance as it does not have IRQ tracing support. A significant portion of the overhead is in the added irq flags save and restore, which I think is not needed if IRQ tracing is unused. A suggested patch is attached below that recovers the lost performance. However, the "ifdef"s in the patch are a bit ugly. Signed-off-by: Tim Chen <tim.c.chen@intel.com> Acked-by: Ingo Molnar <mingo@elte.hu> Signed-off-by: Andrew Morton <akpm@osdl.org> Signed-off-by: Linus Torvalds <torvalds@osdl.org>
2006-07-30 10:04:02 +00:00
#endif
preempt_check_resched();
}
EXPORT_SYMBOL(__local_bh_enable_ip);
/*
Fix lockup related to stop_machine being stuck in __do_softirq. The stop machine logic can lock up if all but one of the migration threads make it through the disable-irq step and the one remaining thread gets stuck in __do_softirq. The reason __do_softirq can hang is that it has a bail-out based on jiffies timeout, but in the lockup case, jiffies itself is not incremented. To work around this, re-add the max_restart counter in __do_irq and stop processing irqs after 10 restarts. Thanks to Tejun Heo and Rusty Russell and others for helping me track this down. This was introduced in 3.9 by commit c10d73671ad3 ("softirq: reduce latencies"). It may be worth looking into ath9k to see if it has issues with its irq handler at a later date. The hang stack traces look something like this: ------------[ cut here ]------------ WARNING: at kernel/watchdog.c:245 watchdog_overflow_callback+0x9c/0xa7() Watchdog detected hard LOCKUP on cpu 2 Modules linked in: ath9k ath9k_common ath9k_hw ath mac80211 cfg80211 nfsv4 auth_rpcgss nfs fscache nf_nat_ipv4 nf_nat veth 8021q garp stp mrp llc pktgen lockd sunrpc] Pid: 23, comm: migration/2 Tainted: G C 3.9.4+ #11 Call Trace: <NMI> warn_slowpath_common+0x85/0x9f warn_slowpath_fmt+0x46/0x48 watchdog_overflow_callback+0x9c/0xa7 __perf_event_overflow+0x137/0x1cb perf_event_overflow+0x14/0x16 intel_pmu_handle_irq+0x2dc/0x359 perf_event_nmi_handler+0x19/0x1b nmi_handle+0x7f/0xc2 do_nmi+0xbc/0x304 end_repeat_nmi+0x1e/0x2e <<EOE>> cpu_stopper_thread+0xae/0x162 smpboot_thread_fn+0x258/0x260 kthread+0xc7/0xcf ret_from_fork+0x7c/0xb0 ---[ end trace 4947dfa9b0a4cec3 ]--- BUG: soft lockup - CPU#1 stuck for 22s! [migration/1:17] Modules linked in: ath9k ath9k_common ath9k_hw ath mac80211 cfg80211 nfsv4 auth_rpcgss nfs fscache nf_nat_ipv4 nf_nat veth 8021q garp stp mrp llc pktgen lockd sunrpc] irq event stamp: 835637905 hardirqs last enabled at (835637904): __do_softirq+0x9f/0x257 hardirqs last disabled at (835637905): apic_timer_interrupt+0x6d/0x80 softirqs last enabled at (5654720): __do_softirq+0x1ff/0x257 softirqs last disabled at (5654725): irq_exit+0x5f/0xbb CPU 1 Pid: 17, comm: migration/1 Tainted: G WC 3.9.4+ #11 To be filled by O.E.M. To be filled by O.E.M./To be filled by O.E.M. RIP: tasklet_hi_action+0xf0/0xf0 Process migration/1 Call Trace: <IRQ> __do_softirq+0x117/0x257 irq_exit+0x5f/0xbb smp_apic_timer_interrupt+0x8a/0x98 apic_timer_interrupt+0x72/0x80 <EOI> printk+0x4d/0x4f stop_machine_cpu_stop+0x22c/0x274 cpu_stopper_thread+0xae/0x162 smpboot_thread_fn+0x258/0x260 kthread+0xc7/0xcf ret_from_fork+0x7c/0xb0 Signed-off-by: Ben Greear <greearb@candelatech.com> Acked-by: Tejun Heo <tj@kernel.org> Acked-by: Pekka Riikonen <priikone@iki.fi> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: stable@kernel.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-06-06 21:29:49 +00:00
* We restart softirq processing for at most MAX_SOFTIRQ_RESTART times,
* but break the loop if need_resched() is set or after 2 ms.
* The MAX_SOFTIRQ_TIME provides a nice upper bound in most cases, but in
* certain cases, such as stop_machine(), jiffies may cease to
* increment and so we need the MAX_SOFTIRQ_RESTART limit as
* well to make sure we eventually return from this method.
*
softirq: reduce latencies In various network workloads, __do_softirq() latencies can be up to 20 ms if HZ=1000, and 200 ms if HZ=100. This is because we iterate 10 times in the softirq dispatcher, and some actions can consume a lot of cycles. This patch changes the fallback to ksoftirqd condition to : - A time limit of 2 ms. - need_resched() being set on current task When one of this condition is met, we wakeup ksoftirqd for further softirq processing if we still have pending softirqs. Using need_resched() as the only condition can trigger RCU stalls, as we can keep BH disabled for too long. I ran several benchmarks and got no significant difference in throughput, but a very significant reduction of latencies (one order of magnitude) : In following bench, 200 antagonist "netperf -t TCP_RR" are started in background, using all available cpus. Then we start one "netperf -t TCP_RR", bound to the cpu handling the NIC IRQ (hard+soft) Before patch : # netperf -H 7.7.7.84 -t TCP_RR -T2,2 -- -k RT_LATENCY,MIN_LATENCY,MAX_LATENCY,P50_LATENCY,P90_LATENCY,P99_LATENCY,MEAN_LATENCY,STDDEV_LATENCY MIGRATED TCP REQUEST/RESPONSE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 7.7.7.84 () port 0 AF_INET : first burst 0 : cpu bind RT_LATENCY=550110.424 MIN_LATENCY=146858 MAX_LATENCY=997109 P50_LATENCY=305000 P90_LATENCY=550000 P99_LATENCY=710000 MEAN_LATENCY=376989.12 STDDEV_LATENCY=184046.92 After patch : # netperf -H 7.7.7.84 -t TCP_RR -T2,2 -- -k RT_LATENCY,MIN_LATENCY,MAX_LATENCY,P50_LATENCY,P90_LATENCY,P99_LATENCY,MEAN_LATENCY,STDDEV_LATENCY MIGRATED TCP REQUEST/RESPONSE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 7.7.7.84 () port 0 AF_INET : first burst 0 : cpu bind RT_LATENCY=40545.492 MIN_LATENCY=9834 MAX_LATENCY=78366 P50_LATENCY=33583 P90_LATENCY=59000 P99_LATENCY=69000 MEAN_LATENCY=38364.67 STDDEV_LATENCY=12865.26 Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: David Miller <davem@davemloft.net> Cc: Tom Herbert <therbert@google.com> Cc: Ben Hutchings <bhutchings@solarflare.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-01-10 23:26:34 +00:00
* These limits have been established via experimentation.
* The two things to balance is latency against fairness -
* we want to handle softirqs as soon as possible, but they
* should not be able to lock up the box.
*/
softirq: reduce latencies In various network workloads, __do_softirq() latencies can be up to 20 ms if HZ=1000, and 200 ms if HZ=100. This is because we iterate 10 times in the softirq dispatcher, and some actions can consume a lot of cycles. This patch changes the fallback to ksoftirqd condition to : - A time limit of 2 ms. - need_resched() being set on current task When one of this condition is met, we wakeup ksoftirqd for further softirq processing if we still have pending softirqs. Using need_resched() as the only condition can trigger RCU stalls, as we can keep BH disabled for too long. I ran several benchmarks and got no significant difference in throughput, but a very significant reduction of latencies (one order of magnitude) : In following bench, 200 antagonist "netperf -t TCP_RR" are started in background, using all available cpus. Then we start one "netperf -t TCP_RR", bound to the cpu handling the NIC IRQ (hard+soft) Before patch : # netperf -H 7.7.7.84 -t TCP_RR -T2,2 -- -k RT_LATENCY,MIN_LATENCY,MAX_LATENCY,P50_LATENCY,P90_LATENCY,P99_LATENCY,MEAN_LATENCY,STDDEV_LATENCY MIGRATED TCP REQUEST/RESPONSE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 7.7.7.84 () port 0 AF_INET : first burst 0 : cpu bind RT_LATENCY=550110.424 MIN_LATENCY=146858 MAX_LATENCY=997109 P50_LATENCY=305000 P90_LATENCY=550000 P99_LATENCY=710000 MEAN_LATENCY=376989.12 STDDEV_LATENCY=184046.92 After patch : # netperf -H 7.7.7.84 -t TCP_RR -T2,2 -- -k RT_LATENCY,MIN_LATENCY,MAX_LATENCY,P50_LATENCY,P90_LATENCY,P99_LATENCY,MEAN_LATENCY,STDDEV_LATENCY MIGRATED TCP REQUEST/RESPONSE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 7.7.7.84 () port 0 AF_INET : first burst 0 : cpu bind RT_LATENCY=40545.492 MIN_LATENCY=9834 MAX_LATENCY=78366 P50_LATENCY=33583 P90_LATENCY=59000 P99_LATENCY=69000 MEAN_LATENCY=38364.67 STDDEV_LATENCY=12865.26 Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: David Miller <davem@davemloft.net> Cc: Tom Herbert <therbert@google.com> Cc: Ben Hutchings <bhutchings@solarflare.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-01-10 23:26:34 +00:00
#define MAX_SOFTIRQ_TIME msecs_to_jiffies(2)
Fix lockup related to stop_machine being stuck in __do_softirq. The stop machine logic can lock up if all but one of the migration threads make it through the disable-irq step and the one remaining thread gets stuck in __do_softirq. The reason __do_softirq can hang is that it has a bail-out based on jiffies timeout, but in the lockup case, jiffies itself is not incremented. To work around this, re-add the max_restart counter in __do_irq and stop processing irqs after 10 restarts. Thanks to Tejun Heo and Rusty Russell and others for helping me track this down. This was introduced in 3.9 by commit c10d73671ad3 ("softirq: reduce latencies"). It may be worth looking into ath9k to see if it has issues with its irq handler at a later date. The hang stack traces look something like this: ------------[ cut here ]------------ WARNING: at kernel/watchdog.c:245 watchdog_overflow_callback+0x9c/0xa7() Watchdog detected hard LOCKUP on cpu 2 Modules linked in: ath9k ath9k_common ath9k_hw ath mac80211 cfg80211 nfsv4 auth_rpcgss nfs fscache nf_nat_ipv4 nf_nat veth 8021q garp stp mrp llc pktgen lockd sunrpc] Pid: 23, comm: migration/2 Tainted: G C 3.9.4+ #11 Call Trace: <NMI> warn_slowpath_common+0x85/0x9f warn_slowpath_fmt+0x46/0x48 watchdog_overflow_callback+0x9c/0xa7 __perf_event_overflow+0x137/0x1cb perf_event_overflow+0x14/0x16 intel_pmu_handle_irq+0x2dc/0x359 perf_event_nmi_handler+0x19/0x1b nmi_handle+0x7f/0xc2 do_nmi+0xbc/0x304 end_repeat_nmi+0x1e/0x2e <<EOE>> cpu_stopper_thread+0xae/0x162 smpboot_thread_fn+0x258/0x260 kthread+0xc7/0xcf ret_from_fork+0x7c/0xb0 ---[ end trace 4947dfa9b0a4cec3 ]--- BUG: soft lockup - CPU#1 stuck for 22s! [migration/1:17] Modules linked in: ath9k ath9k_common ath9k_hw ath mac80211 cfg80211 nfsv4 auth_rpcgss nfs fscache nf_nat_ipv4 nf_nat veth 8021q garp stp mrp llc pktgen lockd sunrpc] irq event stamp: 835637905 hardirqs last enabled at (835637904): __do_softirq+0x9f/0x257 hardirqs last disabled at (835637905): apic_timer_interrupt+0x6d/0x80 softirqs last enabled at (5654720): __do_softirq+0x1ff/0x257 softirqs last disabled at (5654725): irq_exit+0x5f/0xbb CPU 1 Pid: 17, comm: migration/1 Tainted: G WC 3.9.4+ #11 To be filled by O.E.M. To be filled by O.E.M./To be filled by O.E.M. RIP: tasklet_hi_action+0xf0/0xf0 Process migration/1 Call Trace: <IRQ> __do_softirq+0x117/0x257 irq_exit+0x5f/0xbb smp_apic_timer_interrupt+0x8a/0x98 apic_timer_interrupt+0x72/0x80 <EOI> printk+0x4d/0x4f stop_machine_cpu_stop+0x22c/0x274 cpu_stopper_thread+0xae/0x162 smpboot_thread_fn+0x258/0x260 kthread+0xc7/0xcf ret_from_fork+0x7c/0xb0 Signed-off-by: Ben Greear <greearb@candelatech.com> Acked-by: Tejun Heo <tj@kernel.org> Acked-by: Pekka Riikonen <priikone@iki.fi> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: stable@kernel.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-06-06 21:29:49 +00:00
#define MAX_SOFTIRQ_RESTART 10
#ifdef CONFIG_TRACE_IRQFLAGS
/*
* When we run softirqs from irq_exit() and thus on the hardirq stack we need
* to keep the lockdep irq context tracking as tight as possible in order to
* not miss-qualify lock contexts and miss possible deadlocks.
*/
static inline bool lockdep_softirq_start(void)
{
bool in_hardirq = false;
if (trace_hardirq_context(current)) {
in_hardirq = true;
trace_hardirq_exit();
}
lockdep_softirq_enter();
return in_hardirq;
}
static inline void lockdep_softirq_end(bool in_hardirq)
{
lockdep_softirq_exit();
if (in_hardirq)
trace_hardirq_enter();
}
#else
static inline bool lockdep_softirq_start(void) { return false; }
static inline void lockdep_softirq_end(bool in_hardirq) { }
#endif
asmlinkage __visible void __softirq_entry __do_softirq(void)
{
softirq: reduce latencies In various network workloads, __do_softirq() latencies can be up to 20 ms if HZ=1000, and 200 ms if HZ=100. This is because we iterate 10 times in the softirq dispatcher, and some actions can consume a lot of cycles. This patch changes the fallback to ksoftirqd condition to : - A time limit of 2 ms. - need_resched() being set on current task When one of this condition is met, we wakeup ksoftirqd for further softirq processing if we still have pending softirqs. Using need_resched() as the only condition can trigger RCU stalls, as we can keep BH disabled for too long. I ran several benchmarks and got no significant difference in throughput, but a very significant reduction of latencies (one order of magnitude) : In following bench, 200 antagonist "netperf -t TCP_RR" are started in background, using all available cpus. Then we start one "netperf -t TCP_RR", bound to the cpu handling the NIC IRQ (hard+soft) Before patch : # netperf -H 7.7.7.84 -t TCP_RR -T2,2 -- -k RT_LATENCY,MIN_LATENCY,MAX_LATENCY,P50_LATENCY,P90_LATENCY,P99_LATENCY,MEAN_LATENCY,STDDEV_LATENCY MIGRATED TCP REQUEST/RESPONSE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 7.7.7.84 () port 0 AF_INET : first burst 0 : cpu bind RT_LATENCY=550110.424 MIN_LATENCY=146858 MAX_LATENCY=997109 P50_LATENCY=305000 P90_LATENCY=550000 P99_LATENCY=710000 MEAN_LATENCY=376989.12 STDDEV_LATENCY=184046.92 After patch : # netperf -H 7.7.7.84 -t TCP_RR -T2,2 -- -k RT_LATENCY,MIN_LATENCY,MAX_LATENCY,P50_LATENCY,P90_LATENCY,P99_LATENCY,MEAN_LATENCY,STDDEV_LATENCY MIGRATED TCP REQUEST/RESPONSE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 7.7.7.84 () port 0 AF_INET : first burst 0 : cpu bind RT_LATENCY=40545.492 MIN_LATENCY=9834 MAX_LATENCY=78366 P50_LATENCY=33583 P90_LATENCY=59000 P99_LATENCY=69000 MEAN_LATENCY=38364.67 STDDEV_LATENCY=12865.26 Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: David Miller <davem@davemloft.net> Cc: Tom Herbert <therbert@google.com> Cc: Ben Hutchings <bhutchings@solarflare.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-01-10 23:26:34 +00:00
unsigned long end = jiffies + MAX_SOFTIRQ_TIME;
mm: allow PF_MEMALLOC from softirq context This is needed to allow network softirq packet processing to make use of PF_MEMALLOC. Currently softirq context cannot use PF_MEMALLOC due to it not being associated with a task, and therefore not having task flags to fiddle with - thus the gfp to alloc flag mapping ignores the task flags when in interrupts (hard or soft) context. Allowing softirqs to make use of PF_MEMALLOC therefore requires some trickery. This patch borrows the task flags from whatever process happens to be preempted by the softirq. It then modifies the gfp to alloc flags mapping to not exclude task flags in softirq context, and modify the softirq code to save, clear and restore the PF_MEMALLOC flag. The save and clear, ensures the preempted task's PF_MEMALLOC flag doesn't leak into the softirq. The restore ensures a softirq's PF_MEMALLOC flag cannot leak back into the preempted process. This should be safe due to the following reasons Softirqs can run on multiple CPUs sure but the same task should not be executing the same softirq code. Neither should the softirq handler be preempted by any other softirq handler so the flags should not leak to an unrelated softirq. Softirqs re-enable hardware interrupts in __do_softirq() so can be preempted by hardware interrupts so PF_MEMALLOC is inherited by the hard IRQ. However, this is similar to a process in reclaim being preempted by a hardirq. While PF_MEMALLOC is set, gfp_to_alloc_flags() distinguishes between hard and soft irqs and avoids giving a hardirq the ALLOC_NO_WATERMARKS flag. If the softirq is deferred to ksoftirq then its flags may be used instead of a normal tasks but as the softirq cannot be preempted, the PF_MEMALLOC flag does not leak to other code by accident. [davem@davemloft.net: Document why PF_MEMALLOC is safe] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Mel Gorman <mgorman@suse.de> Cc: David Miller <davem@davemloft.net> Cc: Neil Brown <neilb@suse.de> Cc: Mike Christie <michaelc@cs.wisc.edu> Cc: Eric B Munson <emunson@mgebm.net> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Sebastian Andrzej Siewior <sebastian@breakpoint.cc> Cc: Mel Gorman <mgorman@suse.de> Cc: Christoph Lameter <cl@linux.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-07-31 23:44:07 +00:00
unsigned long old_flags = current->flags;
Fix lockup related to stop_machine being stuck in __do_softirq. The stop machine logic can lock up if all but one of the migration threads make it through the disable-irq step and the one remaining thread gets stuck in __do_softirq. The reason __do_softirq can hang is that it has a bail-out based on jiffies timeout, but in the lockup case, jiffies itself is not incremented. To work around this, re-add the max_restart counter in __do_irq and stop processing irqs after 10 restarts. Thanks to Tejun Heo and Rusty Russell and others for helping me track this down. This was introduced in 3.9 by commit c10d73671ad3 ("softirq: reduce latencies"). It may be worth looking into ath9k to see if it has issues with its irq handler at a later date. The hang stack traces look something like this: ------------[ cut here ]------------ WARNING: at kernel/watchdog.c:245 watchdog_overflow_callback+0x9c/0xa7() Watchdog detected hard LOCKUP on cpu 2 Modules linked in: ath9k ath9k_common ath9k_hw ath mac80211 cfg80211 nfsv4 auth_rpcgss nfs fscache nf_nat_ipv4 nf_nat veth 8021q garp stp mrp llc pktgen lockd sunrpc] Pid: 23, comm: migration/2 Tainted: G C 3.9.4+ #11 Call Trace: <NMI> warn_slowpath_common+0x85/0x9f warn_slowpath_fmt+0x46/0x48 watchdog_overflow_callback+0x9c/0xa7 __perf_event_overflow+0x137/0x1cb perf_event_overflow+0x14/0x16 intel_pmu_handle_irq+0x2dc/0x359 perf_event_nmi_handler+0x19/0x1b nmi_handle+0x7f/0xc2 do_nmi+0xbc/0x304 end_repeat_nmi+0x1e/0x2e <<EOE>> cpu_stopper_thread+0xae/0x162 smpboot_thread_fn+0x258/0x260 kthread+0xc7/0xcf ret_from_fork+0x7c/0xb0 ---[ end trace 4947dfa9b0a4cec3 ]--- BUG: soft lockup - CPU#1 stuck for 22s! [migration/1:17] Modules linked in: ath9k ath9k_common ath9k_hw ath mac80211 cfg80211 nfsv4 auth_rpcgss nfs fscache nf_nat_ipv4 nf_nat veth 8021q garp stp mrp llc pktgen lockd sunrpc] irq event stamp: 835637905 hardirqs last enabled at (835637904): __do_softirq+0x9f/0x257 hardirqs last disabled at (835637905): apic_timer_interrupt+0x6d/0x80 softirqs last enabled at (5654720): __do_softirq+0x1ff/0x257 softirqs last disabled at (5654725): irq_exit+0x5f/0xbb CPU 1 Pid: 17, comm: migration/1 Tainted: G WC 3.9.4+ #11 To be filled by O.E.M. To be filled by O.E.M./To be filled by O.E.M. RIP: tasklet_hi_action+0xf0/0xf0 Process migration/1 Call Trace: <IRQ> __do_softirq+0x117/0x257 irq_exit+0x5f/0xbb smp_apic_timer_interrupt+0x8a/0x98 apic_timer_interrupt+0x72/0x80 <EOI> printk+0x4d/0x4f stop_machine_cpu_stop+0x22c/0x274 cpu_stopper_thread+0xae/0x162 smpboot_thread_fn+0x258/0x260 kthread+0xc7/0xcf ret_from_fork+0x7c/0xb0 Signed-off-by: Ben Greear <greearb@candelatech.com> Acked-by: Tejun Heo <tj@kernel.org> Acked-by: Pekka Riikonen <priikone@iki.fi> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: stable@kernel.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-06-06 21:29:49 +00:00
int max_restart = MAX_SOFTIRQ_RESTART;
struct softirq_action *h;
bool in_hardirq;
__u32 pending;
int softirq_bit;
mm: allow PF_MEMALLOC from softirq context This is needed to allow network softirq packet processing to make use of PF_MEMALLOC. Currently softirq context cannot use PF_MEMALLOC due to it not being associated with a task, and therefore not having task flags to fiddle with - thus the gfp to alloc flag mapping ignores the task flags when in interrupts (hard or soft) context. Allowing softirqs to make use of PF_MEMALLOC therefore requires some trickery. This patch borrows the task flags from whatever process happens to be preempted by the softirq. It then modifies the gfp to alloc flags mapping to not exclude task flags in softirq context, and modify the softirq code to save, clear and restore the PF_MEMALLOC flag. The save and clear, ensures the preempted task's PF_MEMALLOC flag doesn't leak into the softirq. The restore ensures a softirq's PF_MEMALLOC flag cannot leak back into the preempted process. This should be safe due to the following reasons Softirqs can run on multiple CPUs sure but the same task should not be executing the same softirq code. Neither should the softirq handler be preempted by any other softirq handler so the flags should not leak to an unrelated softirq. Softirqs re-enable hardware interrupts in __do_softirq() so can be preempted by hardware interrupts so PF_MEMALLOC is inherited by the hard IRQ. However, this is similar to a process in reclaim being preempted by a hardirq. While PF_MEMALLOC is set, gfp_to_alloc_flags() distinguishes between hard and soft irqs and avoids giving a hardirq the ALLOC_NO_WATERMARKS flag. If the softirq is deferred to ksoftirq then its flags may be used instead of a normal tasks but as the softirq cannot be preempted, the PF_MEMALLOC flag does not leak to other code by accident. [davem@davemloft.net: Document why PF_MEMALLOC is safe] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Mel Gorman <mgorman@suse.de> Cc: David Miller <davem@davemloft.net> Cc: Neil Brown <neilb@suse.de> Cc: Mike Christie <michaelc@cs.wisc.edu> Cc: Eric B Munson <emunson@mgebm.net> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Sebastian Andrzej Siewior <sebastian@breakpoint.cc> Cc: Mel Gorman <mgorman@suse.de> Cc: Christoph Lameter <cl@linux.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-07-31 23:44:07 +00:00
/*
* Mask out PF_MEMALLOC as the current task context is borrowed for the
* softirq. A softirq handled, such as network RX, might set PF_MEMALLOC
* again if the socket is related to swapping.
mm: allow PF_MEMALLOC from softirq context This is needed to allow network softirq packet processing to make use of PF_MEMALLOC. Currently softirq context cannot use PF_MEMALLOC due to it not being associated with a task, and therefore not having task flags to fiddle with - thus the gfp to alloc flag mapping ignores the task flags when in interrupts (hard or soft) context. Allowing softirqs to make use of PF_MEMALLOC therefore requires some trickery. This patch borrows the task flags from whatever process happens to be preempted by the softirq. It then modifies the gfp to alloc flags mapping to not exclude task flags in softirq context, and modify the softirq code to save, clear and restore the PF_MEMALLOC flag. The save and clear, ensures the preempted task's PF_MEMALLOC flag doesn't leak into the softirq. The restore ensures a softirq's PF_MEMALLOC flag cannot leak back into the preempted process. This should be safe due to the following reasons Softirqs can run on multiple CPUs sure but the same task should not be executing the same softirq code. Neither should the softirq handler be preempted by any other softirq handler so the flags should not leak to an unrelated softirq. Softirqs re-enable hardware interrupts in __do_softirq() so can be preempted by hardware interrupts so PF_MEMALLOC is inherited by the hard IRQ. However, this is similar to a process in reclaim being preempted by a hardirq. While PF_MEMALLOC is set, gfp_to_alloc_flags() distinguishes between hard and soft irqs and avoids giving a hardirq the ALLOC_NO_WATERMARKS flag. If the softirq is deferred to ksoftirq then its flags may be used instead of a normal tasks but as the softirq cannot be preempted, the PF_MEMALLOC flag does not leak to other code by accident. [davem@davemloft.net: Document why PF_MEMALLOC is safe] Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Mel Gorman <mgorman@suse.de> Cc: David Miller <davem@davemloft.net> Cc: Neil Brown <neilb@suse.de> Cc: Mike Christie <michaelc@cs.wisc.edu> Cc: Eric B Munson <emunson@mgebm.net> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: Sebastian Andrzej Siewior <sebastian@breakpoint.cc> Cc: Mel Gorman <mgorman@suse.de> Cc: Christoph Lameter <cl@linux.com> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-07-31 23:44:07 +00:00
*/
current->flags &= ~PF_MEMALLOC;
pending = local_softirq_pending();
account_irq_enter_time(current);
__local_bh_disable_ip(_RET_IP_, SOFTIRQ_OFFSET);
in_hardirq = lockdep_softirq_start();
restart:
/* Reset the pending bitmask before enabling irqs */
set_softirq_pending(0);
local_irq_enable();
h = softirq_vec;
while ((softirq_bit = ffs(pending))) {
unsigned int vec_nr;
int prev_count;
h += softirq_bit - 1;
vec_nr = h - softirq_vec;
prev_count = preempt_count();
kstat_incr_softirqs_this_cpu(vec_nr);
trace_softirq_entry(vec_nr);
h->action(h);
trace_softirq_exit(vec_nr);
if (unlikely(prev_count != preempt_count())) {
pr_err("huh, entered softirq %u %s %p with preempt_count %08x, exited with %08x?\n",
vec_nr, softirq_to_name[vec_nr], h->action,
prev_count, preempt_count());
preempt_count_set(prev_count);
}
h++;
pending >>= softirq_bit;
}
if (__this_cpu_read(ksoftirqd) == current)
rcu_softirq_qs();
local_irq_disable();
pending = local_softirq_pending();
softirq: reduce latencies In various network workloads, __do_softirq() latencies can be up to 20 ms if HZ=1000, and 200 ms if HZ=100. This is because we iterate 10 times in the softirq dispatcher, and some actions can consume a lot of cycles. This patch changes the fallback to ksoftirqd condition to : - A time limit of 2 ms. - need_resched() being set on current task When one of this condition is met, we wakeup ksoftirqd for further softirq processing if we still have pending softirqs. Using need_resched() as the only condition can trigger RCU stalls, as we can keep BH disabled for too long. I ran several benchmarks and got no significant difference in throughput, but a very significant reduction of latencies (one order of magnitude) : In following bench, 200 antagonist "netperf -t TCP_RR" are started in background, using all available cpus. Then we start one "netperf -t TCP_RR", bound to the cpu handling the NIC IRQ (hard+soft) Before patch : # netperf -H 7.7.7.84 -t TCP_RR -T2,2 -- -k RT_LATENCY,MIN_LATENCY,MAX_LATENCY,P50_LATENCY,P90_LATENCY,P99_LATENCY,MEAN_LATENCY,STDDEV_LATENCY MIGRATED TCP REQUEST/RESPONSE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 7.7.7.84 () port 0 AF_INET : first burst 0 : cpu bind RT_LATENCY=550110.424 MIN_LATENCY=146858 MAX_LATENCY=997109 P50_LATENCY=305000 P90_LATENCY=550000 P99_LATENCY=710000 MEAN_LATENCY=376989.12 STDDEV_LATENCY=184046.92 After patch : # netperf -H 7.7.7.84 -t TCP_RR -T2,2 -- -k RT_LATENCY,MIN_LATENCY,MAX_LATENCY,P50_LATENCY,P90_LATENCY,P99_LATENCY,MEAN_LATENCY,STDDEV_LATENCY MIGRATED TCP REQUEST/RESPONSE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 7.7.7.84 () port 0 AF_INET : first burst 0 : cpu bind RT_LATENCY=40545.492 MIN_LATENCY=9834 MAX_LATENCY=78366 P50_LATENCY=33583 P90_LATENCY=59000 P99_LATENCY=69000 MEAN_LATENCY=38364.67 STDDEV_LATENCY=12865.26 Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: David Miller <davem@davemloft.net> Cc: Tom Herbert <therbert@google.com> Cc: Ben Hutchings <bhutchings@solarflare.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-01-10 23:26:34 +00:00
if (pending) {
Fix lockup related to stop_machine being stuck in __do_softirq. The stop machine logic can lock up if all but one of the migration threads make it through the disable-irq step and the one remaining thread gets stuck in __do_softirq. The reason __do_softirq can hang is that it has a bail-out based on jiffies timeout, but in the lockup case, jiffies itself is not incremented. To work around this, re-add the max_restart counter in __do_irq and stop processing irqs after 10 restarts. Thanks to Tejun Heo and Rusty Russell and others for helping me track this down. This was introduced in 3.9 by commit c10d73671ad3 ("softirq: reduce latencies"). It may be worth looking into ath9k to see if it has issues with its irq handler at a later date. The hang stack traces look something like this: ------------[ cut here ]------------ WARNING: at kernel/watchdog.c:245 watchdog_overflow_callback+0x9c/0xa7() Watchdog detected hard LOCKUP on cpu 2 Modules linked in: ath9k ath9k_common ath9k_hw ath mac80211 cfg80211 nfsv4 auth_rpcgss nfs fscache nf_nat_ipv4 nf_nat veth 8021q garp stp mrp llc pktgen lockd sunrpc] Pid: 23, comm: migration/2 Tainted: G C 3.9.4+ #11 Call Trace: <NMI> warn_slowpath_common+0x85/0x9f warn_slowpath_fmt+0x46/0x48 watchdog_overflow_callback+0x9c/0xa7 __perf_event_overflow+0x137/0x1cb perf_event_overflow+0x14/0x16 intel_pmu_handle_irq+0x2dc/0x359 perf_event_nmi_handler+0x19/0x1b nmi_handle+0x7f/0xc2 do_nmi+0xbc/0x304 end_repeat_nmi+0x1e/0x2e <<EOE>> cpu_stopper_thread+0xae/0x162 smpboot_thread_fn+0x258/0x260 kthread+0xc7/0xcf ret_from_fork+0x7c/0xb0 ---[ end trace 4947dfa9b0a4cec3 ]--- BUG: soft lockup - CPU#1 stuck for 22s! [migration/1:17] Modules linked in: ath9k ath9k_common ath9k_hw ath mac80211 cfg80211 nfsv4 auth_rpcgss nfs fscache nf_nat_ipv4 nf_nat veth 8021q garp stp mrp llc pktgen lockd sunrpc] irq event stamp: 835637905 hardirqs last enabled at (835637904): __do_softirq+0x9f/0x257 hardirqs last disabled at (835637905): apic_timer_interrupt+0x6d/0x80 softirqs last enabled at (5654720): __do_softirq+0x1ff/0x257 softirqs last disabled at (5654725): irq_exit+0x5f/0xbb CPU 1 Pid: 17, comm: migration/1 Tainted: G WC 3.9.4+ #11 To be filled by O.E.M. To be filled by O.E.M./To be filled by O.E.M. RIP: tasklet_hi_action+0xf0/0xf0 Process migration/1 Call Trace: <IRQ> __do_softirq+0x117/0x257 irq_exit+0x5f/0xbb smp_apic_timer_interrupt+0x8a/0x98 apic_timer_interrupt+0x72/0x80 <EOI> printk+0x4d/0x4f stop_machine_cpu_stop+0x22c/0x274 cpu_stopper_thread+0xae/0x162 smpboot_thread_fn+0x258/0x260 kthread+0xc7/0xcf ret_from_fork+0x7c/0xb0 Signed-off-by: Ben Greear <greearb@candelatech.com> Acked-by: Tejun Heo <tj@kernel.org> Acked-by: Pekka Riikonen <priikone@iki.fi> Cc: Eric Dumazet <eric.dumazet@gmail.com> Cc: stable@kernel.org Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2013-06-06 21:29:49 +00:00
if (time_before(jiffies, end) && !need_resched() &&
--max_restart)
softirq: reduce latencies In various network workloads, __do_softirq() latencies can be up to 20 ms if HZ=1000, and 200 ms if HZ=100. This is because we iterate 10 times in the softirq dispatcher, and some actions can consume a lot of cycles. This patch changes the fallback to ksoftirqd condition to : - A time limit of 2 ms. - need_resched() being set on current task When one of this condition is met, we wakeup ksoftirqd for further softirq processing if we still have pending softirqs. Using need_resched() as the only condition can trigger RCU stalls, as we can keep BH disabled for too long. I ran several benchmarks and got no significant difference in throughput, but a very significant reduction of latencies (one order of magnitude) : In following bench, 200 antagonist "netperf -t TCP_RR" are started in background, using all available cpus. Then we start one "netperf -t TCP_RR", bound to the cpu handling the NIC IRQ (hard+soft) Before patch : # netperf -H 7.7.7.84 -t TCP_RR -T2,2 -- -k RT_LATENCY,MIN_LATENCY,MAX_LATENCY,P50_LATENCY,P90_LATENCY,P99_LATENCY,MEAN_LATENCY,STDDEV_LATENCY MIGRATED TCP REQUEST/RESPONSE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 7.7.7.84 () port 0 AF_INET : first burst 0 : cpu bind RT_LATENCY=550110.424 MIN_LATENCY=146858 MAX_LATENCY=997109 P50_LATENCY=305000 P90_LATENCY=550000 P99_LATENCY=710000 MEAN_LATENCY=376989.12 STDDEV_LATENCY=184046.92 After patch : # netperf -H 7.7.7.84 -t TCP_RR -T2,2 -- -k RT_LATENCY,MIN_LATENCY,MAX_LATENCY,P50_LATENCY,P90_LATENCY,P99_LATENCY,MEAN_LATENCY,STDDEV_LATENCY MIGRATED TCP REQUEST/RESPONSE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 7.7.7.84 () port 0 AF_INET : first burst 0 : cpu bind RT_LATENCY=40545.492 MIN_LATENCY=9834 MAX_LATENCY=78366 P50_LATENCY=33583 P90_LATENCY=59000 P99_LATENCY=69000 MEAN_LATENCY=38364.67 STDDEV_LATENCY=12865.26 Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: David Miller <davem@davemloft.net> Cc: Tom Herbert <therbert@google.com> Cc: Ben Hutchings <bhutchings@solarflare.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-01-10 23:26:34 +00:00
goto restart;
wakeup_softirqd();
softirq: reduce latencies In various network workloads, __do_softirq() latencies can be up to 20 ms if HZ=1000, and 200 ms if HZ=100. This is because we iterate 10 times in the softirq dispatcher, and some actions can consume a lot of cycles. This patch changes the fallback to ksoftirqd condition to : - A time limit of 2 ms. - need_resched() being set on current task When one of this condition is met, we wakeup ksoftirqd for further softirq processing if we still have pending softirqs. Using need_resched() as the only condition can trigger RCU stalls, as we can keep BH disabled for too long. I ran several benchmarks and got no significant difference in throughput, but a very significant reduction of latencies (one order of magnitude) : In following bench, 200 antagonist "netperf -t TCP_RR" are started in background, using all available cpus. Then we start one "netperf -t TCP_RR", bound to the cpu handling the NIC IRQ (hard+soft) Before patch : # netperf -H 7.7.7.84 -t TCP_RR -T2,2 -- -k RT_LATENCY,MIN_LATENCY,MAX_LATENCY,P50_LATENCY,P90_LATENCY,P99_LATENCY,MEAN_LATENCY,STDDEV_LATENCY MIGRATED TCP REQUEST/RESPONSE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 7.7.7.84 () port 0 AF_INET : first burst 0 : cpu bind RT_LATENCY=550110.424 MIN_LATENCY=146858 MAX_LATENCY=997109 P50_LATENCY=305000 P90_LATENCY=550000 P99_LATENCY=710000 MEAN_LATENCY=376989.12 STDDEV_LATENCY=184046.92 After patch : # netperf -H 7.7.7.84 -t TCP_RR -T2,2 -- -k RT_LATENCY,MIN_LATENCY,MAX_LATENCY,P50_LATENCY,P90_LATENCY,P99_LATENCY,MEAN_LATENCY,STDDEV_LATENCY MIGRATED TCP REQUEST/RESPONSE TEST from 0.0.0.0 (0.0.0.0) port 0 AF_INET to 7.7.7.84 () port 0 AF_INET : first burst 0 : cpu bind RT_LATENCY=40545.492 MIN_LATENCY=9834 MAX_LATENCY=78366 P50_LATENCY=33583 P90_LATENCY=59000 P99_LATENCY=69000 MEAN_LATENCY=38364.67 STDDEV_LATENCY=12865.26 Signed-off-by: Eric Dumazet <edumazet@google.com> Cc: David Miller <davem@davemloft.net> Cc: Tom Herbert <therbert@google.com> Cc: Ben Hutchings <bhutchings@solarflare.com> Signed-off-by: David S. Miller <davem@davemloft.net>
2013-01-10 23:26:34 +00:00
}
lockdep_softirq_end(in_hardirq);
account_irq_exit_time(current);
__local_bh_enable(SOFTIRQ_OFFSET);
WARN_ON_ONCE(in_interrupt());
current_restore_flags(old_flags, PF_MEMALLOC);
}
asmlinkage __visible void do_softirq(void)
{
__u32 pending;
unsigned long flags;
if (in_interrupt())
return;
local_irq_save(flags);
pending = local_softirq_pending();
Mark HI and TASKLET softirq synchronous Way back in 4.9, we committed 4cd13c21b207 ("softirq: Let ksoftirqd do its job"), and ever since we've had small nagging issues with it. For example, we've had: 1ff688209e2e ("watchdog: core: make sure the watchdog_worker is not deferred") 8d5755b3f77b ("watchdog: softdog: fire watchdog even if softirqs do not get to run") 217f69743681 ("net: busy-poll: allow preemption in sk_busy_loop()") all of which worked around some of the effects of that commit. The DVB people have also complained that the commit causes excessive USB URB latencies, which seems to be due to the USB code using tasklets to schedule USB traffic. This seems to be an issue mainly when already living on the edge, but waiting for ksoftirqd to handle it really does seem to cause excessive latencies. Now Hanna Hawa reports that this issue isn't just limited to USB URB and DVB, but also causes timeout problems for the Marvell SoC team: "I'm facing kernel panic issue while running raid 5 on sata disks connected to Macchiatobin (Marvell community board with Armada-8040 SoC with 4 ARMv8 cores of CA72) Raid 5 built with Marvell DMA engine and async_tx mechanism (ASYNC_TX_DMA [=y]); the DMA driver (mv_xor_v2) uses a tasklet to clean the done descriptors from the queue" The latency problem causes a panic: mv_xor_v2 f0400000.xor: dma_sync_wait: timeout! Kernel panic - not syncing: async_tx_quiesce: DMA error waiting for transaction We've discussed simply just reverting the original commit entirely, and also much more involved solutions (with per-softirq threads etc). This patch is intentionally stupid and fairly limited, because the issue still remains, and the other solutions either got sidetracked or had other issues. We should probably also consider the timer softirqs to be synchronous and not be delayed to ksoftirqd (since they were the issue with the earlier watchdog problems), but that should be done as a separate patch. This does only the tasklet cases. Reported-and-tested-by: Hanna Hawa <hannah@marvell.com> Reported-and-tested-by: Josef Griebichler <griebichler.josef@gmx.at> Reported-by: Mauro Carvalho Chehab <mchehab@s-opensource.com> Cc: Alan Stern <stern@rowland.harvard.edu> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Eric Dumazet <edumazet@google.com> Cc: Ingo Molnar <mingo@kernel.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-01-08 19:51:04 +00:00
if (pending && !ksoftirqd_running(pending))
do_softirq_own_stack();
local_irq_restore(flags);
}
/*
* Enter an interrupt context.
*/
void irq_enter(void)
{
"Tree RCU": scalable classic RCU implementation This patch fixes a long-standing performance bug in classic RCU that results in massive internal-to-RCU lock contention on systems with more than a few hundred CPUs. Although this patch creates a separate flavor of RCU for ease of review and patch maintenance, it is intended to replace classic RCU. This patch still handles stress better than does mainline, so I am still calling it ready for inclusion. This patch is against the -tip tree. Nevertheless, experience on an actual 1000+ CPU machine would still be most welcome. Most of the changes noted below were found while creating an rcutiny (which should permit ejecting the current rcuclassic) and while doing detailed line-by-line documentation. Updates from v9 (http://lkml.org/lkml/2008/12/2/334): o Fixes from remainder of line-by-line code walkthrough, including comment spelling, initialization, undesirable narrowing due to type conversion, removing redundant memory barriers, removing redundant local-variable initialization, and removing redundant local variables. I do not believe that any of these fixes address the CPU-hotplug issues that Andi Kleen was seeing, but please do give it a whirl in case the machine is smarter than I am. A writeup from the walkthrough may be found at the following URL, in case you are suffering from terminal insomnia or masochism: http://www.kernel.org/pub/linux/kernel/people/paulmck/tmp/rcutree-walkthrough.2008.12.16a.pdf o Made rcutree tracing use seq_file, as suggested some time ago by Lai Jiangshan. o Added a .csv variant of the rcudata debugfs trace file, to allow people having thousands of CPUs to drop the data into a spreadsheet. Tested with oocalc and gnumeric. Updated documentation to suit. Updates from v8 (http://lkml.org/lkml/2008/11/15/139): o Fix a theoretical race between grace-period initialization and force_quiescent_state() that could occur if more than three jiffies were required to carry out the grace-period initialization. Which it might, if you had enough CPUs. o Apply Ingo's printk-standardization patch. o Substitute local variables for repeated accesses to global variables. o Fix comment misspellings and redundant (but harmless) increments of ->n_rcu_pending (this latter after having explicitly added it). o Apply checkpatch fixes. Updates from v7 (http://lkml.org/lkml/2008/10/10/291): o Fixed a number of problems noted by Gautham Shenoy, including the cpu-stall-detection bug that he was having difficulty convincing me was real. ;-) o Changed cpu-stall detection to wait for ten seconds rather than three in order to reduce false positive, as suggested by Ingo Molnar. o Produced a design document (http://lwn.net/Articles/305782/). The act of writing this document uncovered a number of both theoretical and "here and now" bugs as noted below. o Fix dynticks_nesting accounting confusion, simplify WARN_ON() condition, fix kerneldoc comments, and add memory barriers in dynticks interface functions. o Add more data to tracing. o Remove unused "rcu_barrier" field from rcu_data structure. o Count calls to rcu_pending() from scheduling-clock interrupt to use as a surrogate timebase should jiffies stop counting. o Fix a theoretical race between force_quiescent_state() and grace-period initialization. Yes, initialization does have to go on for some jiffies for this race to occur, but given enough CPUs... Updates from v6 (http://lkml.org/lkml/2008/9/23/448): o Fix a number of checkpatch.pl complaints. o Apply review comments from Ingo Molnar and Lai Jiangshan on the stall-detection code. o Fix several bugs in !CONFIG_SMP builds. o Fix a misspelled config-parameter name so that RCU now announces at boot time if stall detection is configured. o Run tests on numerous combinations of configurations parameters, which after the fixes above, now build and run correctly. Updates from v5 (http://lkml.org/lkml/2008/9/15/92, bad subject line): o Fix a compiler error in the !CONFIG_FANOUT_EXACT case (blew a changeset some time ago, and finally got around to retesting this option). o Fix some tracing bugs in rcupreempt that caused incorrect totals to be printed. o I now test with a more brutal random-selection online/offline script (attached). Probably more brutal than it needs to be on the people reading it as well, but so it goes. o A number of optimizations and usability improvements: o Make rcu_pending() ignore the grace-period timeout when there is no grace period in progress. o Make force_quiescent_state() avoid going for a global lock in the case where there is no grace period in progress. o Rearrange struct fields to improve struct layout. o Make call_rcu() initiate a grace period if RCU was idle, rather than waiting for the next scheduling clock interrupt. o Invoke rcu_irq_enter() and rcu_irq_exit() only when idle, as suggested by Andi Kleen. I still don't completely trust this change, and might back it out. o Make CONFIG_RCU_TRACE be the single config variable manipulated for all forms of RCU, instead of the prior confusion. o Document tracing files and formats for both rcupreempt and rcutree. Updates from v4 for those missing v5 given its bad subject line: o Separated dynticks interface so that NMIs and irqs call separate functions, greatly simplifying it. In particular, this code no longer requires a proof of correctness. ;-) o Separated dynticks state out into its own per-CPU structure, avoiding the duplicated accounting. o The case where a dynticks-idle CPU runs an irq handler that invokes call_rcu() is now correctly handled, forcing that CPU out of dynticks-idle mode. o Review comments have been applied (thank you all!!!). For but one example, fixed the dynticks-ordering issue that Manfred pointed out, saving me much debugging. ;-) o Adjusted rcuclassic and rcupreempt to handle dynticks changes. Attached is an updated patch to Classic RCU that applies a hierarchy, greatly reducing the contention on the top-level lock for large machines. This passes 10-hour concurrent rcutorture and online-offline testing on 128-CPU ppc64 without dynticks enabled, and exposes some timekeeping bugs in presence of dynticks (exciting working on a system where "sleep 1" hangs until interrupted...), which were fixed in the 2.6.27 kernel. It is getting more reliable than mainline by some measures, so the next version will be against -tip for inclusion. See also Manfred Spraul's recent patches (or his earlier work from 2004 at http://marc.info/?l=linux-kernel&m=108546384711797&w=2). We will converge onto a common patch in the fullness of time, but are currently exploring different regions of the design space. That said, I have already gratefully stolen quite a few of Manfred's ideas. This patch provides CONFIG_RCU_FANOUT, which controls the bushiness of the RCU hierarchy. Defaults to 32 on 32-bit machines and 64 on 64-bit machines. If CONFIG_NR_CPUS is less than CONFIG_RCU_FANOUT, there is no hierarchy. By default, the RCU initialization code will adjust CONFIG_RCU_FANOUT to balance the hierarchy, so strongly NUMA architectures may choose to set CONFIG_RCU_FANOUT_EXACT to disable this balancing, allowing the hierarchy to be exactly aligned to the underlying hardware. Up to two levels of hierarchy are permitted (in addition to the root node), allowing up to 16,384 CPUs on 32-bit systems and up to 262,144 CPUs on 64-bit systems. I just know that I am going to regret saying this, but this seems more than sufficient for the foreseeable future. (Some architectures might wish to set CONFIG_RCU_FANOUT=4, which would limit such architectures to 64 CPUs. If this becomes a real problem, additional levels can be added, but I doubt that it will make a significant difference on real hardware.) In the common case, a given CPU will manipulate its private rcu_data structure and the rcu_node structure that it shares with its immediate neighbors. This can reduce both lock and memory contention by multiple orders of magnitude, which should eliminate the need for the strange manipulations that are reported to be required when running Linux on very large systems. Some shortcomings: o More bugs will probably surface as a result of an ongoing line-by-line code inspection. Patches will be provided as required. o There are probably hangs, rcutorture failures, &c. Seems quite stable on a 128-CPU machine, but that is kind of small compared to 4096 CPUs. However, seems to do better than mainline. Patches will be provided as required. o The memory footprint of this version is several KB larger than rcuclassic. A separate UP-only rcutiny patch will be provided, which will reduce the memory footprint significantly, even compared to the old rcuclassic. One such patch passes light testing, and has a memory footprint smaller even than rcuclassic. Initial reaction from various embedded guys was "it is not worth it", so am putting it aside. Credits: o Manfred Spraul for ideas, review comments, and bugs spotted, as well as some good friendly competition. ;-) o Josh Triplett, Ingo Molnar, Peter Zijlstra, Mathieu Desnoyers, Lai Jiangshan, Andi Kleen, Andy Whitcroft, and Andrew Morton for reviews and comments. o Thomas Gleixner for much-needed help with some timer issues (see patches below). o Jon M. Tollefson, Tim Pepper, Andrew Theurer, Jose R. Santos, Andy Whitcroft, Darrick Wong, Nishanth Aravamudan, Anton Blanchard, Dave Kleikamp, and Nathan Lynch for keeping machines alive despite my heavy abuse^Wtesting. Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Signed-off-by: Ingo Molnar <mingo@elte.hu>
2008-12-18 20:55:32 +00:00
rcu_irq_enter();
if (is_idle_task(current) && !in_interrupt()) {
/*
* Prevent raise_softirq from needlessly waking up ksoftirqd
* here, as softirq will be serviced on return from interrupt.
*/
local_bh_disable();
tick_irq_enter();
_local_bh_enable();
}
__irq_enter();
}
genirq: Provide forced interrupt threading Add a commandline parameter "threadirqs" which forces all interrupts except those marked IRQF_NO_THREAD to run threaded. That's mostly a debug option to allow retrieving better debug data from crashing interrupt handlers. If "threadirqs" is not enabled on the kernel command line, then there is no impact in the interrupt hotpath. Architecture code needs to select CONFIG_IRQ_FORCED_THREADING after marking the interrupts which cant be threaded IRQF_NO_THREAD. All interrupts which have IRQF_TIMER set are implict marked IRQF_NO_THREAD. Also all PER_CPU interrupts are excluded. Forced threading hard interrupts also forces all soft interrupt handling into thread context. When enabled it might slow down things a bit, but for debugging problems in interrupt code it's a reasonable penalty as it does not immediately crash and burn the machine when an interrupt handler is buggy. Some test results on a Core2Duo machine: Cache cold run of: # time git grep irq_desc non-threaded threaded real 1m18.741s 1m19.061s user 0m1.874s 0m1.757s sys 0m5.843s 0m5.427s # iperf -c server non-threaded [ 3] 0.0-10.0 sec 1.09 GBytes 933 Mbits/sec [ 3] 0.0-10.0 sec 1.09 GBytes 934 Mbits/sec [ 3] 0.0-10.0 sec 1.09 GBytes 933 Mbits/sec threaded [ 3] 0.0-10.0 sec 1.09 GBytes 939 Mbits/sec [ 3] 0.0-10.0 sec 1.09 GBytes 934 Mbits/sec [ 3] 0.0-10.0 sec 1.09 GBytes 937 Mbits/sec Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <20110223234956.772668648@linutronix.de>
2011-02-23 23:52:23 +00:00
static inline void invoke_softirq(void)
{
Mark HI and TASKLET softirq synchronous Way back in 4.9, we committed 4cd13c21b207 ("softirq: Let ksoftirqd do its job"), and ever since we've had small nagging issues with it. For example, we've had: 1ff688209e2e ("watchdog: core: make sure the watchdog_worker is not deferred") 8d5755b3f77b ("watchdog: softdog: fire watchdog even if softirqs do not get to run") 217f69743681 ("net: busy-poll: allow preemption in sk_busy_loop()") all of which worked around some of the effects of that commit. The DVB people have also complained that the commit causes excessive USB URB latencies, which seems to be due to the USB code using tasklets to schedule USB traffic. This seems to be an issue mainly when already living on the edge, but waiting for ksoftirqd to handle it really does seem to cause excessive latencies. Now Hanna Hawa reports that this issue isn't just limited to USB URB and DVB, but also causes timeout problems for the Marvell SoC team: "I'm facing kernel panic issue while running raid 5 on sata disks connected to Macchiatobin (Marvell community board with Armada-8040 SoC with 4 ARMv8 cores of CA72) Raid 5 built with Marvell DMA engine and async_tx mechanism (ASYNC_TX_DMA [=y]); the DMA driver (mv_xor_v2) uses a tasklet to clean the done descriptors from the queue" The latency problem causes a panic: mv_xor_v2 f0400000.xor: dma_sync_wait: timeout! Kernel panic - not syncing: async_tx_quiesce: DMA error waiting for transaction We've discussed simply just reverting the original commit entirely, and also much more involved solutions (with per-softirq threads etc). This patch is intentionally stupid and fairly limited, because the issue still remains, and the other solutions either got sidetracked or had other issues. We should probably also consider the timer softirqs to be synchronous and not be delayed to ksoftirqd (since they were the issue with the earlier watchdog problems), but that should be done as a separate patch. This does only the tasklet cases. Reported-and-tested-by: Hanna Hawa <hannah@marvell.com> Reported-and-tested-by: Josef Griebichler <griebichler.josef@gmx.at> Reported-by: Mauro Carvalho Chehab <mchehab@s-opensource.com> Cc: Alan Stern <stern@rowland.harvard.edu> Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org> Cc: Eric Dumazet <edumazet@google.com> Cc: Ingo Molnar <mingo@kernel.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2018-01-08 19:51:04 +00:00
if (ksoftirqd_running(local_softirq_pending()))
softirq: Let ksoftirqd do its job A while back, Paolo and Hannes sent an RFC patch adding threaded-able napi poll loop support : (https://patchwork.ozlabs.org/patch/620657/) The problem seems to be that softirqs are very aggressive and are often handled by the current process, even if we are under stress and that ksoftirqd was scheduled, so that innocent threads would have more chance to make progress. This patch makes sure that if ksoftirq is running, we let it perform the softirq work. Jonathan Corbet summarized the issue in https://lwn.net/Articles/687617/ Tested: - NIC receiving traffic handled by CPU 0 - UDP receiver running on CPU 0, using a single UDP socket. - Incoming flood of UDP packets targeting the UDP socket. Before the patch, the UDP receiver could almost never get CPU cycles and could only receive ~2,000 packets per second. After the patch, CPU cycles are split 50/50 between user application and ksoftirqd/0, and we can effectively read ~900,000 packets per second, a huge improvement in DOS situation. (Note that more packets are now dropped by the NIC itself, since the BH handlers get less CPU cycles to drain RX ring buffer) Since the load runs in well identified threads context, an admin can more easily tune process scheduling parameters if needed. Reported-by: Paolo Abeni <pabeni@redhat.com> Reported-by: Hannes Frederic Sowa <hannes@stressinduktion.org> Signed-off-by: Eric Dumazet <edumazet@google.com> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org> Cc: David Miller <davem@davemloft.net> Cc: Hannes Frederic Sowa <hannes@redhat.com> Cc: Jesper Dangaard Brouer <jbrouer@redhat.com> Cc: Jonathan Corbet <corbet@lwn.net> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Rik van Riel <riel@redhat.com> Cc: Thomas Gleixner <tglx@linutronix.de> Link: http://lkml.kernel.org/r/1472665349.14381.356.camel@edumazet-glaptop3.roam.corp.google.com Signed-off-by: Ingo Molnar <mingo@kernel.org>
2016-08-31 17:42:29 +00:00
return;
irq: Force hardirq exit's softirq processing on its own stack The commit facd8b80c67a3cf64a467c4a2ac5fb31f2e6745b ("irq: Sanitize invoke_softirq") converted irq exit calls of do_softirq() to __do_softirq() on all architectures, assuming it was only used there for its irq disablement properties. But as a side effect, the softirqs processed in the end of the hardirq are always called on the inline current stack that is used by irq_exit() instead of the softirq stack provided by the archs that override do_softirq(). The result is mostly safe if the architecture runs irq_exit() on a separate irq stack because then softirqs are processed on that same stack that is near empty at this stage (assuming hardirq aren't nesting). Otherwise irq_exit() runs in the task stack and so does the softirq too. The interrupted call stack can be randomly deep already and the softirq can dig through it even further. To add insult to the injury, this softirq can be interrupted by a new hardirq, maximizing the chances for a stack overrun as reported in powerpc for example: do_IRQ: stack overflow: 1920 CPU: 0 PID: 1602 Comm: qemu-system-ppc Not tainted 3.10.4-300.1.fc19.ppc64p7 #1 Call Trace: [c0000000050a8740] .show_stack+0x130/0x200 (unreliable) [c0000000050a8810] .dump_stack+0x28/0x3c [c0000000050a8880] .do_IRQ+0x2b8/0x2c0 [c0000000050a8930] hardware_interrupt_common+0x154/0x180 --- Exception: 501 at .cp_start_xmit+0x3a4/0x820 [8139cp] LR = .cp_start_xmit+0x390/0x820 [8139cp] [c0000000050a8d40] .dev_hard_start_xmit+0x394/0x640 [c0000000050a8e00] .sch_direct_xmit+0x110/0x260 [c0000000050a8ea0] .dev_queue_xmit+0x260/0x630 [c0000000050a8f40] .br_dev_queue_push_xmit+0xc4/0x130 [bridge] [c0000000050a8fc0] .br_dev_xmit+0x198/0x270 [bridge] [c0000000050a9070] .dev_hard_start_xmit+0x394/0x640 [c0000000050a9130] .dev_queue_xmit+0x428/0x630 [c0000000050a91d0] .ip_finish_output+0x2a4/0x550 [c0000000050a9290] .ip_local_out+0x50/0x70 [c0000000050a9310] .ip_queue_xmit+0x148/0x420 [c0000000050a93b0] .tcp_transmit_skb+0x4e4/0xaf0 [c0000000050a94a0] .__tcp_ack_snd_check+0x7c/0xf0 [c0000000050a9520] .tcp_rcv_established+0x1e8/0x930 [c0000000050a95f0] .tcp_v4_do_rcv+0x21c/0x570 [c0000000050a96c0] .tcp_v4_rcv+0x734/0x930 [c0000000050a97a0] .ip_local_deliver_finish+0x184/0x360 [c0000000050a9840] .ip_rcv_finish+0x148/0x400 [c0000000050a98d0] .__netif_receive_skb_core+0x4f8/0xb00 [c0000000050a99d0] .netif_receive_skb+0x44/0x110 [c0000000050a9a70] .br_handle_frame_finish+0x2bc/0x3f0 [bridge] [c0000000050a9b20] .br_nf_pre_routing_finish+0x2ac/0x420 [bridge] [c0000000050a9bd0] .br_nf_pre_routing+0x4dc/0x7d0 [bridge] [c0000000050a9c70] .nf_iterate+0x114/0x130 [c0000000050a9d30] .nf_hook_slow+0xb4/0x1e0 [c0000000050a9e00] .br_handle_frame+0x290/0x330 [bridge] [c0000000050a9ea0] .__netif_receive_skb_core+0x34c/0xb00 [c0000000050a9fa0] .netif_receive_skb+0x44/0x110 [c0000000050aa040] .napi_gro_receive+0xe8/0x120 [c0000000050aa0c0] .cp_rx_poll+0x31c/0x590 [8139cp] [c0000000050aa1d0] .net_rx_action+0x1dc/0x310 [c0000000050aa2b0] .__do_softirq+0x158/0x330 [c0000000050aa3b0] .irq_exit+0xc8/0x110 [c0000000050aa430] .do_IRQ+0xdc/0x2c0 [c0000000050aa4e0] hardware_interrupt_common+0x154/0x180 --- Exception: 501 at .bad_range+0x1c/0x110 LR = .get_page_from_freelist+0x908/0xbb0 [c0000000050aa7d0] .list_del+0x18/0x50 (unreliable) [c0000000050aa850] .get_page_from_freelist+0x908/0xbb0 [c0000000050aa9e0] .__alloc_pages_nodemask+0x21c/0xae0 [c0000000050aaba0] .alloc_pages_vma+0xd0/0x210 [c0000000050aac60] .handle_pte_fault+0x814/0xb70 [c0000000050aad50] .__get_user_pages+0x1a4/0x640 [c0000000050aae60] .get_user_pages_fast+0xec/0x160 [c0000000050aaf10] .__gfn_to_pfn_memslot+0x3b0/0x430 [kvm] [c0000000050aafd0] .kvmppc_gfn_to_pfn+0x64/0x130 [kvm] [c0000000050ab070] .kvmppc_mmu_map_page+0x94/0x530 [kvm] [c0000000050ab190] .kvmppc_handle_pagefault+0x174/0x610 [kvm] [c0000000050ab270] .kvmppc_handle_exit_pr+0x464/0x9b0 [kvm] [c0000000050ab320] kvm_start_lightweight+0x1ec/0x1fc [kvm] [c0000000050ab4f0] .kvmppc_vcpu_run_pr+0x168/0x3b0 [kvm] [c0000000050ab9c0] .kvmppc_vcpu_run+0xc8/0xf0 [kvm] [c0000000050aba50] .kvm_arch_vcpu_ioctl_run+0x5c/0x1a0 [kvm] [c0000000050abae0] .kvm_vcpu_ioctl+0x478/0x730 [kvm] [c0000000050abc90] .do_vfs_ioctl+0x4ec/0x7c0 [c0000000050abd80] .SyS_ioctl+0xd4/0xf0 [c0000000050abe30] syscall_exit+0x0/0x98 Since this is a regression, this patch proposes a minimalistic and low-risk solution by blindly forcing the hardirq exit processing of softirqs on the softirq stack. This way we should reduce significantly the opportunities for task stack overflow dug by softirqs. Longer term solutions may involve extending the hardirq stack coverage to irq_exit(), etc... Reported-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: #3.9.. <stable@vger.kernel.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Paul Mackerras <paulus@au1.ibm.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@au1.ibm.com> Cc: James Hogan <james.hogan@imgtec.com> Cc: James E.J. Bottomley <jejb@parisc-linux.org> Cc: Helge Deller <deller@gmx.de> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: David S. Miller <davem@davemloft.net> Cc: Andrew Morton <akpm@linux-foundation.org>
2013-09-23 22:50:25 +00:00
if (!force_irqthreads) {
irq: Optimize softirq stack selection in irq exit If irq_exit() is called on the arch's specified irq stack, it should be safe to run softirqs inline under that same irq stack as it is near empty by the time we call irq_exit(). For example if we use the same stack for both hard and soft irqs here, the worst case scenario is: hardirq -> softirq -> hardirq. But then the softirq supersedes the first hardirq as the stack user since irq_exit() is called in a mostly empty stack. So the stack merge in this case looks acceptable. Stack overrun still have a chance to happen if hardirqs have more opportunities to nest, but then it's another problem to solve. So lets adapt the irq exit's softirq stack on top of a new Kconfig symbol that can be defined when irq_exit() runs on the irq stack. That way we can spare some stack switch on irq processing and all the cache issues that come along. Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Paul Mackerras <paulus@au1.ibm.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@au1.ibm.com> Cc: James Hogan <james.hogan@imgtec.com> Cc: James E.J. Bottomley <jejb@parisc-linux.org> Cc: Helge Deller <deller@gmx.de> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: David S. Miller <davem@davemloft.net> Cc: Andrew Morton <akpm@linux-foundation.org>
2013-09-24 15:17:47 +00:00
#ifdef CONFIG_HAVE_IRQ_EXIT_ON_IRQ_STACK
irq: Force hardirq exit's softirq processing on its own stack The commit facd8b80c67a3cf64a467c4a2ac5fb31f2e6745b ("irq: Sanitize invoke_softirq") converted irq exit calls of do_softirq() to __do_softirq() on all architectures, assuming it was only used there for its irq disablement properties. But as a side effect, the softirqs processed in the end of the hardirq are always called on the inline current stack that is used by irq_exit() instead of the softirq stack provided by the archs that override do_softirq(). The result is mostly safe if the architecture runs irq_exit() on a separate irq stack because then softirqs are processed on that same stack that is near empty at this stage (assuming hardirq aren't nesting). Otherwise irq_exit() runs in the task stack and so does the softirq too. The interrupted call stack can be randomly deep already and the softirq can dig through it even further. To add insult to the injury, this softirq can be interrupted by a new hardirq, maximizing the chances for a stack overrun as reported in powerpc for example: do_IRQ: stack overflow: 1920 CPU: 0 PID: 1602 Comm: qemu-system-ppc Not tainted 3.10.4-300.1.fc19.ppc64p7 #1 Call Trace: [c0000000050a8740] .show_stack+0x130/0x200 (unreliable) [c0000000050a8810] .dump_stack+0x28/0x3c [c0000000050a8880] .do_IRQ+0x2b8/0x2c0 [c0000000050a8930] hardware_interrupt_common+0x154/0x180 --- Exception: 501 at .cp_start_xmit+0x3a4/0x820 [8139cp] LR = .cp_start_xmit+0x390/0x820 [8139cp] [c0000000050a8d40] .dev_hard_start_xmit+0x394/0x640 [c0000000050a8e00] .sch_direct_xmit+0x110/0x260 [c0000000050a8ea0] .dev_queue_xmit+0x260/0x630 [c0000000050a8f40] .br_dev_queue_push_xmit+0xc4/0x130 [bridge] [c0000000050a8fc0] .br_dev_xmit+0x198/0x270 [bridge] [c0000000050a9070] .dev_hard_start_xmit+0x394/0x640 [c0000000050a9130] .dev_queue_xmit+0x428/0x630 [c0000000050a91d0] .ip_finish_output+0x2a4/0x550 [c0000000050a9290] .ip_local_out+0x50/0x70 [c0000000050a9310] .ip_queue_xmit+0x148/0x420 [c0000000050a93b0] .tcp_transmit_skb+0x4e4/0xaf0 [c0000000050a94a0] .__tcp_ack_snd_check+0x7c/0xf0 [c0000000050a9520] .tcp_rcv_established+0x1e8/0x930 [c0000000050a95f0] .tcp_v4_do_rcv+0x21c/0x570 [c0000000050a96c0] .tcp_v4_rcv+0x734/0x930 [c0000000050a97a0] .ip_local_deliver_finish+0x184/0x360 [c0000000050a9840] .ip_rcv_finish+0x148/0x400 [c0000000050a98d0] .__netif_receive_skb_core+0x4f8/0xb00 [c0000000050a99d0] .netif_receive_skb+0x44/0x110 [c0000000050a9a70] .br_handle_frame_finish+0x2bc/0x3f0 [bridge] [c0000000050a9b20] .br_nf_pre_routing_finish+0x2ac/0x420 [bridge] [c0000000050a9bd0] .br_nf_pre_routing+0x4dc/0x7d0 [bridge] [c0000000050a9c70] .nf_iterate+0x114/0x130 [c0000000050a9d30] .nf_hook_slow+0xb4/0x1e0 [c0000000050a9e00] .br_handle_frame+0x290/0x330 [bridge] [c0000000050a9ea0] .__netif_receive_skb_core+0x34c/0xb00 [c0000000050a9fa0] .netif_receive_skb+0x44/0x110 [c0000000050aa040] .napi_gro_receive+0xe8/0x120 [c0000000050aa0c0] .cp_rx_poll+0x31c/0x590 [8139cp] [c0000000050aa1d0] .net_rx_action+0x1dc/0x310 [c0000000050aa2b0] .__do_softirq+0x158/0x330 [c0000000050aa3b0] .irq_exit+0xc8/0x110 [c0000000050aa430] .do_IRQ+0xdc/0x2c0 [c0000000050aa4e0] hardware_interrupt_common+0x154/0x180 --- Exception: 501 at .bad_range+0x1c/0x110 LR = .get_page_from_freelist+0x908/0xbb0 [c0000000050aa7d0] .list_del+0x18/0x50 (unreliable) [c0000000050aa850] .get_page_from_freelist+0x908/0xbb0 [c0000000050aa9e0] .__alloc_pages_nodemask+0x21c/0xae0 [c0000000050aaba0] .alloc_pages_vma+0xd0/0x210 [c0000000050aac60] .handle_pte_fault+0x814/0xb70 [c0000000050aad50] .__get_user_pages+0x1a4/0x640 [c0000000050aae60] .get_user_pages_fast+0xec/0x160 [c0000000050aaf10] .__gfn_to_pfn_memslot+0x3b0/0x430 [kvm] [c0000000050aafd0] .kvmppc_gfn_to_pfn+0x64/0x130 [kvm] [c0000000050ab070] .kvmppc_mmu_map_page+0x94/0x530 [kvm] [c0000000050ab190] .kvmppc_handle_pagefault+0x174/0x610 [kvm] [c0000000050ab270] .kvmppc_handle_exit_pr+0x464/0x9b0 [kvm] [c0000000050ab320] kvm_start_lightweight+0x1ec/0x1fc [kvm] [c0000000050ab4f0] .kvmppc_vcpu_run_pr+0x168/0x3b0 [kvm] [c0000000050ab9c0] .kvmppc_vcpu_run+0xc8/0xf0 [kvm] [c0000000050aba50] .kvm_arch_vcpu_ioctl_run+0x5c/0x1a0 [kvm] [c0000000050abae0] .kvm_vcpu_ioctl+0x478/0x730 [kvm] [c0000000050abc90] .do_vfs_ioctl+0x4ec/0x7c0 [c0000000050abd80] .SyS_ioctl+0xd4/0xf0 [c0000000050abe30] syscall_exit+0x0/0x98 Since this is a regression, this patch proposes a minimalistic and low-risk solution by blindly forcing the hardirq exit processing of softirqs on the softirq stack. This way we should reduce significantly the opportunities for task stack overflow dug by softirqs. Longer term solutions may involve extending the hardirq stack coverage to irq_exit(), etc... Reported-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: #3.9.. <stable@vger.kernel.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Paul Mackerras <paulus@au1.ibm.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@au1.ibm.com> Cc: James Hogan <james.hogan@imgtec.com> Cc: James E.J. Bottomley <jejb@parisc-linux.org> Cc: Helge Deller <deller@gmx.de> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: David S. Miller <davem@davemloft.net> Cc: Andrew Morton <akpm@linux-foundation.org>
2013-09-23 22:50:25 +00:00
/*
* We can safely execute softirq on the current stack if
* it is the irq stack, because it should be near empty
irq: Optimize softirq stack selection in irq exit If irq_exit() is called on the arch's specified irq stack, it should be safe to run softirqs inline under that same irq stack as it is near empty by the time we call irq_exit(). For example if we use the same stack for both hard and soft irqs here, the worst case scenario is: hardirq -> softirq -> hardirq. But then the softirq supersedes the first hardirq as the stack user since irq_exit() is called in a mostly empty stack. So the stack merge in this case looks acceptable. Stack overrun still have a chance to happen if hardirqs have more opportunities to nest, but then it's another problem to solve. So lets adapt the irq exit's softirq stack on top of a new Kconfig symbol that can be defined when irq_exit() runs on the irq stack. That way we can spare some stack switch on irq processing and all the cache issues that come along. Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Paul Mackerras <paulus@au1.ibm.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@au1.ibm.com> Cc: James Hogan <james.hogan@imgtec.com> Cc: James E.J. Bottomley <jejb@parisc-linux.org> Cc: Helge Deller <deller@gmx.de> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: David S. Miller <davem@davemloft.net> Cc: Andrew Morton <akpm@linux-foundation.org>
2013-09-24 15:17:47 +00:00
* at this stage.
*/
__do_softirq();
#else
/*
* Otherwise, irq_exit() is called on the task stack that can
* be potentially deep already. So call softirq in its own stack
* to prevent from any overrun.
irq: Force hardirq exit's softirq processing on its own stack The commit facd8b80c67a3cf64a467c4a2ac5fb31f2e6745b ("irq: Sanitize invoke_softirq") converted irq exit calls of do_softirq() to __do_softirq() on all architectures, assuming it was only used there for its irq disablement properties. But as a side effect, the softirqs processed in the end of the hardirq are always called on the inline current stack that is used by irq_exit() instead of the softirq stack provided by the archs that override do_softirq(). The result is mostly safe if the architecture runs irq_exit() on a separate irq stack because then softirqs are processed on that same stack that is near empty at this stage (assuming hardirq aren't nesting). Otherwise irq_exit() runs in the task stack and so does the softirq too. The interrupted call stack can be randomly deep already and the softirq can dig through it even further. To add insult to the injury, this softirq can be interrupted by a new hardirq, maximizing the chances for a stack overrun as reported in powerpc for example: do_IRQ: stack overflow: 1920 CPU: 0 PID: 1602 Comm: qemu-system-ppc Not tainted 3.10.4-300.1.fc19.ppc64p7 #1 Call Trace: [c0000000050a8740] .show_stack+0x130/0x200 (unreliable) [c0000000050a8810] .dump_stack+0x28/0x3c [c0000000050a8880] .do_IRQ+0x2b8/0x2c0 [c0000000050a8930] hardware_interrupt_common+0x154/0x180 --- Exception: 501 at .cp_start_xmit+0x3a4/0x820 [8139cp] LR = .cp_start_xmit+0x390/0x820 [8139cp] [c0000000050a8d40] .dev_hard_start_xmit+0x394/0x640 [c0000000050a8e00] .sch_direct_xmit+0x110/0x260 [c0000000050a8ea0] .dev_queue_xmit+0x260/0x630 [c0000000050a8f40] .br_dev_queue_push_xmit+0xc4/0x130 [bridge] [c0000000050a8fc0] .br_dev_xmit+0x198/0x270 [bridge] [c0000000050a9070] .dev_hard_start_xmit+0x394/0x640 [c0000000050a9130] .dev_queue_xmit+0x428/0x630 [c0000000050a91d0] .ip_finish_output+0x2a4/0x550 [c0000000050a9290] .ip_local_out+0x50/0x70 [c0000000050a9310] .ip_queue_xmit+0x148/0x420 [c0000000050a93b0] .tcp_transmit_skb+0x4e4/0xaf0 [c0000000050a94a0] .__tcp_ack_snd_check+0x7c/0xf0 [c0000000050a9520] .tcp_rcv_established+0x1e8/0x930 [c0000000050a95f0] .tcp_v4_do_rcv+0x21c/0x570 [c0000000050a96c0] .tcp_v4_rcv+0x734/0x930 [c0000000050a97a0] .ip_local_deliver_finish+0x184/0x360 [c0000000050a9840] .ip_rcv_finish+0x148/0x400 [c0000000050a98d0] .__netif_receive_skb_core+0x4f8/0xb00 [c0000000050a99d0] .netif_receive_skb+0x44/0x110 [c0000000050a9a70] .br_handle_frame_finish+0x2bc/0x3f0 [bridge] [c0000000050a9b20] .br_nf_pre_routing_finish+0x2ac/0x420 [bridge] [c0000000050a9bd0] .br_nf_pre_routing+0x4dc/0x7d0 [bridge] [c0000000050a9c70] .nf_iterate+0x114/0x130 [c0000000050a9d30] .nf_hook_slow+0xb4/0x1e0 [c0000000050a9e00] .br_handle_frame+0x290/0x330 [bridge] [c0000000050a9ea0] .__netif_receive_skb_core+0x34c/0xb00 [c0000000050a9fa0] .netif_receive_skb+0x44/0x110 [c0000000050aa040] .napi_gro_receive+0xe8/0x120 [c0000000050aa0c0] .cp_rx_poll+0x31c/0x590 [8139cp] [c0000000050aa1d0] .net_rx_action+0x1dc/0x310 [c0000000050aa2b0] .__do_softirq+0x158/0x330 [c0000000050aa3b0] .irq_exit+0xc8/0x110 [c0000000050aa430] .do_IRQ+0xdc/0x2c0 [c0000000050aa4e0] hardware_interrupt_common+0x154/0x180 --- Exception: 501 at .bad_range+0x1c/0x110 LR = .get_page_from_freelist+0x908/0xbb0 [c0000000050aa7d0] .list_del+0x18/0x50 (unreliable) [c0000000050aa850] .get_page_from_freelist+0x908/0xbb0 [c0000000050aa9e0] .__alloc_pages_nodemask+0x21c/0xae0 [c0000000050aaba0] .alloc_pages_vma+0xd0/0x210 [c0000000050aac60] .handle_pte_fault+0x814/0xb70 [c0000000050aad50] .__get_user_pages+0x1a4/0x640 [c0000000050aae60] .get_user_pages_fast+0xec/0x160 [c0000000050aaf10] .__gfn_to_pfn_memslot+0x3b0/0x430 [kvm] [c0000000050aafd0] .kvmppc_gfn_to_pfn+0x64/0x130 [kvm] [c0000000050ab070] .kvmppc_mmu_map_page+0x94/0x530 [kvm] [c0000000050ab190] .kvmppc_handle_pagefault+0x174/0x610 [kvm] [c0000000050ab270] .kvmppc_handle_exit_pr+0x464/0x9b0 [kvm] [c0000000050ab320] kvm_start_lightweight+0x1ec/0x1fc [kvm] [c0000000050ab4f0] .kvmppc_vcpu_run_pr+0x168/0x3b0 [kvm] [c0000000050ab9c0] .kvmppc_vcpu_run+0xc8/0xf0 [kvm] [c0000000050aba50] .kvm_arch_vcpu_ioctl_run+0x5c/0x1a0 [kvm] [c0000000050abae0] .kvm_vcpu_ioctl+0x478/0x730 [kvm] [c0000000050abc90] .do_vfs_ioctl+0x4ec/0x7c0 [c0000000050abd80] .SyS_ioctl+0xd4/0xf0 [c0000000050abe30] syscall_exit+0x0/0x98 Since this is a regression, this patch proposes a minimalistic and low-risk solution by blindly forcing the hardirq exit processing of softirqs on the softirq stack. This way we should reduce significantly the opportunities for task stack overflow dug by softirqs. Longer term solutions may involve extending the hardirq stack coverage to irq_exit(), etc... Reported-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: #3.9.. <stable@vger.kernel.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Paul Mackerras <paulus@au1.ibm.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@au1.ibm.com> Cc: James Hogan <james.hogan@imgtec.com> Cc: James E.J. Bottomley <jejb@parisc-linux.org> Cc: Helge Deller <deller@gmx.de> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: David S. Miller <davem@davemloft.net> Cc: Andrew Morton <akpm@linux-foundation.org>
2013-09-23 22:50:25 +00:00
*/
do_softirq_own_stack();
irq: Optimize softirq stack selection in irq exit If irq_exit() is called on the arch's specified irq stack, it should be safe to run softirqs inline under that same irq stack as it is near empty by the time we call irq_exit(). For example if we use the same stack for both hard and soft irqs here, the worst case scenario is: hardirq -> softirq -> hardirq. But then the softirq supersedes the first hardirq as the stack user since irq_exit() is called in a mostly empty stack. So the stack merge in this case looks acceptable. Stack overrun still have a chance to happen if hardirqs have more opportunities to nest, but then it's another problem to solve. So lets adapt the irq exit's softirq stack on top of a new Kconfig symbol that can be defined when irq_exit() runs on the irq stack. That way we can spare some stack switch on irq processing and all the cache issues that come along. Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Paul Mackerras <paulus@au1.ibm.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@au1.ibm.com> Cc: James Hogan <james.hogan@imgtec.com> Cc: James E.J. Bottomley <jejb@parisc-linux.org> Cc: Helge Deller <deller@gmx.de> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: David S. Miller <davem@davemloft.net> Cc: Andrew Morton <akpm@linux-foundation.org>
2013-09-24 15:17:47 +00:00
#endif
irq: Force hardirq exit's softirq processing on its own stack The commit facd8b80c67a3cf64a467c4a2ac5fb31f2e6745b ("irq: Sanitize invoke_softirq") converted irq exit calls of do_softirq() to __do_softirq() on all architectures, assuming it was only used there for its irq disablement properties. But as a side effect, the softirqs processed in the end of the hardirq are always called on the inline current stack that is used by irq_exit() instead of the softirq stack provided by the archs that override do_softirq(). The result is mostly safe if the architecture runs irq_exit() on a separate irq stack because then softirqs are processed on that same stack that is near empty at this stage (assuming hardirq aren't nesting). Otherwise irq_exit() runs in the task stack and so does the softirq too. The interrupted call stack can be randomly deep already and the softirq can dig through it even further. To add insult to the injury, this softirq can be interrupted by a new hardirq, maximizing the chances for a stack overrun as reported in powerpc for example: do_IRQ: stack overflow: 1920 CPU: 0 PID: 1602 Comm: qemu-system-ppc Not tainted 3.10.4-300.1.fc19.ppc64p7 #1 Call Trace: [c0000000050a8740] .show_stack+0x130/0x200 (unreliable) [c0000000050a8810] .dump_stack+0x28/0x3c [c0000000050a8880] .do_IRQ+0x2b8/0x2c0 [c0000000050a8930] hardware_interrupt_common+0x154/0x180 --- Exception: 501 at .cp_start_xmit+0x3a4/0x820 [8139cp] LR = .cp_start_xmit+0x390/0x820 [8139cp] [c0000000050a8d40] .dev_hard_start_xmit+0x394/0x640 [c0000000050a8e00] .sch_direct_xmit+0x110/0x260 [c0000000050a8ea0] .dev_queue_xmit+0x260/0x630 [c0000000050a8f40] .br_dev_queue_push_xmit+0xc4/0x130 [bridge] [c0000000050a8fc0] .br_dev_xmit+0x198/0x270 [bridge] [c0000000050a9070] .dev_hard_start_xmit+0x394/0x640 [c0000000050a9130] .dev_queue_xmit+0x428/0x630 [c0000000050a91d0] .ip_finish_output+0x2a4/0x550 [c0000000050a9290] .ip_local_out+0x50/0x70 [c0000000050a9310] .ip_queue_xmit+0x148/0x420 [c0000000050a93b0] .tcp_transmit_skb+0x4e4/0xaf0 [c0000000050a94a0] .__tcp_ack_snd_check+0x7c/0xf0 [c0000000050a9520] .tcp_rcv_established+0x1e8/0x930 [c0000000050a95f0] .tcp_v4_do_rcv+0x21c/0x570 [c0000000050a96c0] .tcp_v4_rcv+0x734/0x930 [c0000000050a97a0] .ip_local_deliver_finish+0x184/0x360 [c0000000050a9840] .ip_rcv_finish+0x148/0x400 [c0000000050a98d0] .__netif_receive_skb_core+0x4f8/0xb00 [c0000000050a99d0] .netif_receive_skb+0x44/0x110 [c0000000050a9a70] .br_handle_frame_finish+0x2bc/0x3f0 [bridge] [c0000000050a9b20] .br_nf_pre_routing_finish+0x2ac/0x420 [bridge] [c0000000050a9bd0] .br_nf_pre_routing+0x4dc/0x7d0 [bridge] [c0000000050a9c70] .nf_iterate+0x114/0x130 [c0000000050a9d30] .nf_hook_slow+0xb4/0x1e0 [c0000000050a9e00] .br_handle_frame+0x290/0x330 [bridge] [c0000000050a9ea0] .__netif_receive_skb_core+0x34c/0xb00 [c0000000050a9fa0] .netif_receive_skb+0x44/0x110 [c0000000050aa040] .napi_gro_receive+0xe8/0x120 [c0000000050aa0c0] .cp_rx_poll+0x31c/0x590 [8139cp] [c0000000050aa1d0] .net_rx_action+0x1dc/0x310 [c0000000050aa2b0] .__do_softirq+0x158/0x330 [c0000000050aa3b0] .irq_exit+0xc8/0x110 [c0000000050aa430] .do_IRQ+0xdc/0x2c0 [c0000000050aa4e0] hardware_interrupt_common+0x154/0x180 --- Exception: 501 at .bad_range+0x1c/0x110 LR = .get_page_from_freelist+0x908/0xbb0 [c0000000050aa7d0] .list_del+0x18/0x50 (unreliable) [c0000000050aa850] .get_page_from_freelist+0x908/0xbb0 [c0000000050aa9e0] .__alloc_pages_nodemask+0x21c/0xae0 [c0000000050aaba0] .alloc_pages_vma+0xd0/0x210 [c0000000050aac60] .handle_pte_fault+0x814/0xb70 [c0000000050aad50] .__get_user_pages+0x1a4/0x640 [c0000000050aae60] .get_user_pages_fast+0xec/0x160 [c0000000050aaf10] .__gfn_to_pfn_memslot+0x3b0/0x430 [kvm] [c0000000050aafd0] .kvmppc_gfn_to_pfn+0x64/0x130 [kvm] [c0000000050ab070] .kvmppc_mmu_map_page+0x94/0x530 [kvm] [c0000000050ab190] .kvmppc_handle_pagefault+0x174/0x610 [kvm] [c0000000050ab270] .kvmppc_handle_exit_pr+0x464/0x9b0 [kvm] [c0000000050ab320] kvm_start_lightweight+0x1ec/0x1fc [kvm] [c0000000050ab4f0] .kvmppc_vcpu_run_pr+0x168/0x3b0 [kvm] [c0000000050ab9c0] .kvmppc_vcpu_run+0xc8/0xf0 [kvm] [c0000000050aba50] .kvm_arch_vcpu_ioctl_run+0x5c/0x1a0 [kvm] [c0000000050abae0] .kvm_vcpu_ioctl+0x478/0x730 [kvm] [c0000000050abc90] .do_vfs_ioctl+0x4ec/0x7c0 [c0000000050abd80] .SyS_ioctl+0xd4/0xf0 [c0000000050abe30] syscall_exit+0x0/0x98 Since this is a regression, this patch proposes a minimalistic and low-risk solution by blindly forcing the hardirq exit processing of softirqs on the softirq stack. This way we should reduce significantly the opportunities for task stack overflow dug by softirqs. Longer term solutions may involve extending the hardirq stack coverage to irq_exit(), etc... Reported-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: #3.9.. <stable@vger.kernel.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Paul Mackerras <paulus@au1.ibm.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@au1.ibm.com> Cc: James Hogan <james.hogan@imgtec.com> Cc: James E.J. Bottomley <jejb@parisc-linux.org> Cc: Helge Deller <deller@gmx.de> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: David S. Miller <davem@davemloft.net> Cc: Andrew Morton <akpm@linux-foundation.org>
2013-09-23 22:50:25 +00:00
} else {
genirq: Provide forced interrupt threading Add a commandline parameter "threadirqs" which forces all interrupts except those marked IRQF_NO_THREAD to run threaded. That's mostly a debug option to allow retrieving better debug data from crashing interrupt handlers. If "threadirqs" is not enabled on the kernel command line, then there is no impact in the interrupt hotpath. Architecture code needs to select CONFIG_IRQ_FORCED_THREADING after marking the interrupts which cant be threaded IRQF_NO_THREAD. All interrupts which have IRQF_TIMER set are implict marked IRQF_NO_THREAD. Also all PER_CPU interrupts are excluded. Forced threading hard interrupts also forces all soft interrupt handling into thread context. When enabled it might slow down things a bit, but for debugging problems in interrupt code it's a reasonable penalty as it does not immediately crash and burn the machine when an interrupt handler is buggy. Some test results on a Core2Duo machine: Cache cold run of: # time git grep irq_desc non-threaded threaded real 1m18.741s 1m19.061s user 0m1.874s 0m1.757s sys 0m5.843s 0m5.427s # iperf -c server non-threaded [ 3] 0.0-10.0 sec 1.09 GBytes 933 Mbits/sec [ 3] 0.0-10.0 sec 1.09 GBytes 934 Mbits/sec [ 3] 0.0-10.0 sec 1.09 GBytes 933 Mbits/sec threaded [ 3] 0.0-10.0 sec 1.09 GBytes 939 Mbits/sec [ 3] 0.0-10.0 sec 1.09 GBytes 934 Mbits/sec [ 3] 0.0-10.0 sec 1.09 GBytes 937 Mbits/sec Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <20110223234956.772668648@linutronix.de>
2011-02-23 23:52:23 +00:00
wakeup_softirqd();
irq: Force hardirq exit's softirq processing on its own stack The commit facd8b80c67a3cf64a467c4a2ac5fb31f2e6745b ("irq: Sanitize invoke_softirq") converted irq exit calls of do_softirq() to __do_softirq() on all architectures, assuming it was only used there for its irq disablement properties. But as a side effect, the softirqs processed in the end of the hardirq are always called on the inline current stack that is used by irq_exit() instead of the softirq stack provided by the archs that override do_softirq(). The result is mostly safe if the architecture runs irq_exit() on a separate irq stack because then softirqs are processed on that same stack that is near empty at this stage (assuming hardirq aren't nesting). Otherwise irq_exit() runs in the task stack and so does the softirq too. The interrupted call stack can be randomly deep already and the softirq can dig through it even further. To add insult to the injury, this softirq can be interrupted by a new hardirq, maximizing the chances for a stack overrun as reported in powerpc for example: do_IRQ: stack overflow: 1920 CPU: 0 PID: 1602 Comm: qemu-system-ppc Not tainted 3.10.4-300.1.fc19.ppc64p7 #1 Call Trace: [c0000000050a8740] .show_stack+0x130/0x200 (unreliable) [c0000000050a8810] .dump_stack+0x28/0x3c [c0000000050a8880] .do_IRQ+0x2b8/0x2c0 [c0000000050a8930] hardware_interrupt_common+0x154/0x180 --- Exception: 501 at .cp_start_xmit+0x3a4/0x820 [8139cp] LR = .cp_start_xmit+0x390/0x820 [8139cp] [c0000000050a8d40] .dev_hard_start_xmit+0x394/0x640 [c0000000050a8e00] .sch_direct_xmit+0x110/0x260 [c0000000050a8ea0] .dev_queue_xmit+0x260/0x630 [c0000000050a8f40] .br_dev_queue_push_xmit+0xc4/0x130 [bridge] [c0000000050a8fc0] .br_dev_xmit+0x198/0x270 [bridge] [c0000000050a9070] .dev_hard_start_xmit+0x394/0x640 [c0000000050a9130] .dev_queue_xmit+0x428/0x630 [c0000000050a91d0] .ip_finish_output+0x2a4/0x550 [c0000000050a9290] .ip_local_out+0x50/0x70 [c0000000050a9310] .ip_queue_xmit+0x148/0x420 [c0000000050a93b0] .tcp_transmit_skb+0x4e4/0xaf0 [c0000000050a94a0] .__tcp_ack_snd_check+0x7c/0xf0 [c0000000050a9520] .tcp_rcv_established+0x1e8/0x930 [c0000000050a95f0] .tcp_v4_do_rcv+0x21c/0x570 [c0000000050a96c0] .tcp_v4_rcv+0x734/0x930 [c0000000050a97a0] .ip_local_deliver_finish+0x184/0x360 [c0000000050a9840] .ip_rcv_finish+0x148/0x400 [c0000000050a98d0] .__netif_receive_skb_core+0x4f8/0xb00 [c0000000050a99d0] .netif_receive_skb+0x44/0x110 [c0000000050a9a70] .br_handle_frame_finish+0x2bc/0x3f0 [bridge] [c0000000050a9b20] .br_nf_pre_routing_finish+0x2ac/0x420 [bridge] [c0000000050a9bd0] .br_nf_pre_routing+0x4dc/0x7d0 [bridge] [c0000000050a9c70] .nf_iterate+0x114/0x130 [c0000000050a9d30] .nf_hook_slow+0xb4/0x1e0 [c0000000050a9e00] .br_handle_frame+0x290/0x330 [bridge] [c0000000050a9ea0] .__netif_receive_skb_core+0x34c/0xb00 [c0000000050a9fa0] .netif_receive_skb+0x44/0x110 [c0000000050aa040] .napi_gro_receive+0xe8/0x120 [c0000000050aa0c0] .cp_rx_poll+0x31c/0x590 [8139cp] [c0000000050aa1d0] .net_rx_action+0x1dc/0x310 [c0000000050aa2b0] .__do_softirq+0x158/0x330 [c0000000050aa3b0] .irq_exit+0xc8/0x110 [c0000000050aa430] .do_IRQ+0xdc/0x2c0 [c0000000050aa4e0] hardware_interrupt_common+0x154/0x180 --- Exception: 501 at .bad_range+0x1c/0x110 LR = .get_page_from_freelist+0x908/0xbb0 [c0000000050aa7d0] .list_del+0x18/0x50 (unreliable) [c0000000050aa850] .get_page_from_freelist+0x908/0xbb0 [c0000000050aa9e0] .__alloc_pages_nodemask+0x21c/0xae0 [c0000000050aaba0] .alloc_pages_vma+0xd0/0x210 [c0000000050aac60] .handle_pte_fault+0x814/0xb70 [c0000000050aad50] .__get_user_pages+0x1a4/0x640 [c0000000050aae60] .get_user_pages_fast+0xec/0x160 [c0000000050aaf10] .__gfn_to_pfn_memslot+0x3b0/0x430 [kvm] [c0000000050aafd0] .kvmppc_gfn_to_pfn+0x64/0x130 [kvm] [c0000000050ab070] .kvmppc_mmu_map_page+0x94/0x530 [kvm] [c0000000050ab190] .kvmppc_handle_pagefault+0x174/0x610 [kvm] [c0000000050ab270] .kvmppc_handle_exit_pr+0x464/0x9b0 [kvm] [c0000000050ab320] kvm_start_lightweight+0x1ec/0x1fc [kvm] [c0000000050ab4f0] .kvmppc_vcpu_run_pr+0x168/0x3b0 [kvm] [c0000000050ab9c0] .kvmppc_vcpu_run+0xc8/0xf0 [kvm] [c0000000050aba50] .kvm_arch_vcpu_ioctl_run+0x5c/0x1a0 [kvm] [c0000000050abae0] .kvm_vcpu_ioctl+0x478/0x730 [kvm] [c0000000050abc90] .do_vfs_ioctl+0x4ec/0x7c0 [c0000000050abd80] .SyS_ioctl+0xd4/0xf0 [c0000000050abe30] syscall_exit+0x0/0x98 Since this is a regression, this patch proposes a minimalistic and low-risk solution by blindly forcing the hardirq exit processing of softirqs on the softirq stack. This way we should reduce significantly the opportunities for task stack overflow dug by softirqs. Longer term solutions may involve extending the hardirq stack coverage to irq_exit(), etc... Reported-by: Benjamin Herrenschmidt <benh@kernel.crashing.org> Acked-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: #3.9.. <stable@vger.kernel.org> Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org> Cc: Paul Mackerras <paulus@au1.ibm.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: H. Peter Anvin <hpa@zytor.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Paul Mackerras <paulus@au1.ibm.com> Cc: James Hogan <james.hogan@imgtec.com> Cc: James E.J. Bottomley <jejb@parisc-linux.org> Cc: Helge Deller <deller@gmx.de> Cc: Martin Schwidefsky <schwidefsky@de.ibm.com> Cc: Heiko Carstens <heiko.carstens@de.ibm.com> Cc: David S. Miller <davem@davemloft.net> Cc: Andrew Morton <akpm@linux-foundation.org>
2013-09-23 22:50:25 +00:00
}
genirq: Provide forced interrupt threading Add a commandline parameter "threadirqs" which forces all interrupts except those marked IRQF_NO_THREAD to run threaded. That's mostly a debug option to allow retrieving better debug data from crashing interrupt handlers. If "threadirqs" is not enabled on the kernel command line, then there is no impact in the interrupt hotpath. Architecture code needs to select CONFIG_IRQ_FORCED_THREADING after marking the interrupts which cant be threaded IRQF_NO_THREAD. All interrupts which have IRQF_TIMER set are implict marked IRQF_NO_THREAD. Also all PER_CPU interrupts are excluded. Forced threading hard interrupts also forces all soft interrupt handling into thread context. When enabled it might slow down things a bit, but for debugging problems in interrupt code it's a reasonable penalty as it does not immediately crash and burn the machine when an interrupt handler is buggy. Some test results on a Core2Duo machine: Cache cold run of: # time git grep irq_desc non-threaded threaded real 1m18.741s 1m19.061s user 0m1.874s 0m1.757s sys 0m5.843s 0m5.427s # iperf -c server non-threaded [ 3] 0.0-10.0 sec 1.09 GBytes 933 Mbits/sec [ 3] 0.0-10.0 sec 1.09 GBytes 934 Mbits/sec [ 3] 0.0-10.0 sec 1.09 GBytes 933 Mbits/sec threaded [ 3] 0.0-10.0 sec 1.09 GBytes 939 Mbits/sec [ 3] 0.0-10.0 sec 1.09 GBytes 934 Mbits/sec [ 3] 0.0-10.0 sec 1.09 GBytes 937 Mbits/sec Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Cc: Peter Zijlstra <peterz@infradead.org> LKML-Reference: <20110223234956.772668648@linutronix.de>
2011-02-23 23:52:23 +00:00
}
static inline void tick_irq_exit(void)
{
#ifdef CONFIG_NO_HZ_COMMON
int cpu = smp_processor_id();
/* Make sure that timer wheel updates are propagated */
if ((idle_cpu(cpu) && !need_resched()) || tick_nohz_full_cpu(cpu)) {
nohz: Fix missing tick reprogram when interrupting an inline softirq The full nohz tick is reprogrammed in irq_exit() only if the exit is not in a nesting interrupt. This stands as an optimization: whether a hardirq or a softirq is interrupted, the tick is going to be reprogrammed when necessary at the end of the inner interrupt, with even potential new updates on the timer queue. When soft interrupts are interrupted, it's assumed that they are executing on the tail of an interrupt return. In that case tick_nohz_irq_exit() is called after softirq processing to take care of the tick reprogramming. But the assumption is wrong: softirqs can be processed inline as well, ie: outside of an interrupt, like in a call to local_bh_enable() or from ksoftirqd. Inline softirqs don't reprogram the tick once they are done, as opposed to interrupt tail softirq processing. So if a tick interrupts an inline softirq processing, the next timer will neither be reprogrammed from the interrupting tick's irq_exit() nor after the interrupted softirq processing. This situation may leave the tick unprogrammed while timers are armed. To fix this, simply keep reprogramming the tick even if a softirq has been interrupted. That can be optimized further, but for now correctness is more important. Note that new timers enqueued in nohz_full mode after a softirq gets interrupted will still be handled just fine through self-IPIs triggered by the timer code. Reported-by: Anna-Maria Gleixner <anna-maria@linutronix.de> Signed-off-by: Frederic Weisbecker <frederic@kernel.org> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Anna-Maria Gleixner <anna-maria@linutronix.de> Cc: stable@vger.kernel.org # 4.14+ Link: https://lkml.kernel.org/r/1533303094-15855-1-git-send-email-frederic@kernel.org
2018-08-03 13:31:34 +00:00
if (!in_irq())
tick_nohz_irq_exit();
}
#endif
}
/*
* Exit an interrupt context. Process softirqs if needed and possible:
*/
void irq_exit(void)
{
#ifndef __ARCH_IRQ_EXIT_IRQS_DISABLED
local_irq_disable();
#else
lockdep_assert_irqs_disabled();
#endif
account_irq_exit_time(current);
preempt_count_sub(HARDIRQ_OFFSET);
if (!in_interrupt() && local_softirq_pending())
invoke_softirq();
tick_irq_exit();
rcu: Fix early call to rcu_idle_enter() On the irq exit path, tick_nohz_irq_exit() may raise a softirq, which action leads to the wake up path and select_task_rq_fair() that makes use of rcu to iterate the domains. This is an illegal use of RCU because we may be in RCU extended quiescent state if we interrupted an RCU-idle window in the idle loop: [ 132.978883] =============================== [ 132.978883] [ INFO: suspicious RCU usage. ] [ 132.978883] ------------------------------- [ 132.978883] kernel/sched_fair.c:1707 suspicious rcu_dereference_check() usage! [ 132.978883] [ 132.978883] other info that might help us debug this: [ 132.978883] [ 132.978883] [ 132.978883] rcu_scheduler_active = 1, debug_locks = 0 [ 132.978883] RCU used illegally from extended quiescent state! [ 132.978883] 2 locks held by swapper/0: [ 132.978883] #0: (&p->pi_lock){-.-.-.}, at: [<ffffffff8105a729>] try_to_wake_up+0x39/0x2f0 [ 132.978883] #1: (rcu_read_lock){.+.+..}, at: [<ffffffff8105556a>] select_task_rq_fair+0x6a/0xec0 [ 132.978883] [ 132.978883] stack backtrace: [ 132.978883] Pid: 0, comm: swapper Tainted: G W 3.0.0+ #178 [ 132.978883] Call Trace: [ 132.978883] <IRQ> [<ffffffff810a01f6>] lockdep_rcu_suspicious+0xe6/0x100 [ 132.978883] [<ffffffff81055c49>] select_task_rq_fair+0x749/0xec0 [ 132.978883] [<ffffffff8105556a>] ? select_task_rq_fair+0x6a/0xec0 [ 132.978883] [<ffffffff812fe494>] ? do_raw_spin_lock+0x54/0x150 [ 132.978883] [<ffffffff810a1f2d>] ? trace_hardirqs_on+0xd/0x10 [ 132.978883] [<ffffffff8105a7c3>] try_to_wake_up+0xd3/0x2f0 [ 132.978883] [<ffffffff81094f98>] ? ktime_get+0x68/0xf0 [ 132.978883] [<ffffffff8105aa35>] wake_up_process+0x15/0x20 [ 132.978883] [<ffffffff81069dd5>] raise_softirq_irqoff+0x65/0x110 [ 132.978883] [<ffffffff8108eb65>] __hrtimer_start_range_ns+0x415/0x5a0 [ 132.978883] [<ffffffff812fe3ee>] ? do_raw_spin_unlock+0x5e/0xb0 [ 132.978883] [<ffffffff8108ed08>] hrtimer_start+0x18/0x20 [ 132.978883] [<ffffffff8109c9c3>] tick_nohz_stop_sched_tick+0x393/0x450 [ 132.978883] [<ffffffff810694f2>] irq_exit+0xd2/0x100 [ 132.978883] [<ffffffff81829e96>] do_IRQ+0x66/0xe0 [ 132.978883] [<ffffffff81820d53>] common_interrupt+0x13/0x13 [ 132.978883] <EOI> [<ffffffff8103434b>] ? native_safe_halt+0xb/0x10 [ 132.978883] [<ffffffff810a1f2d>] ? trace_hardirqs_on+0xd/0x10 [ 132.978883] [<ffffffff810144ea>] default_idle+0xba/0x370 [ 132.978883] [<ffffffff810147fe>] amd_e400_idle+0x5e/0x130 [ 132.978883] [<ffffffff8100a9f6>] cpu_idle+0xb6/0x120 [ 132.978883] [<ffffffff817f217f>] rest_init+0xef/0x150 [ 132.978883] [<ffffffff817f20e2>] ? rest_init+0x52/0x150 [ 132.978883] [<ffffffff81ed9cf3>] start_kernel+0x3da/0x3e5 [ 132.978883] [<ffffffff81ed9346>] x86_64_start_reservations+0x131/0x135 [ 132.978883] [<ffffffff81ed944d>] x86_64_start_kernel+0x103/0x112 Fix this by calling rcu_idle_enter() after tick_nohz_irq_exit(). Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Ingo Molnar <mingo@redhat.com> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Peter Zijlstra <a.p.zijlstra@chello.nl> Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Reviewed-by: Josh Triplett <josh@joshtriplett.org>
2011-10-07 23:31:02 +00:00
rcu_irq_exit();
trace_hardirq_exit(); /* must be last! */
}
/*
* This function must run with irqs disabled!
*/
inline void raise_softirq_irqoff(unsigned int nr)
{
__raise_softirq_irqoff(nr);
/*
* If we're in an interrupt or softirq, we're done
* (this also catches softirq-disabled code). We will
* actually run the softirq once we return from
* the irq or softirq.
*
* Otherwise we wake up ksoftirqd to make sure we
* schedule the softirq soon.
*/
if (!in_interrupt())
wakeup_softirqd();
}
void raise_softirq(unsigned int nr)
{
unsigned long flags;
local_irq_save(flags);
raise_softirq_irqoff(nr);
local_irq_restore(flags);
}
void __raise_softirq_irqoff(unsigned int nr)
{
trace_softirq_raise(nr);
or_softirq_pending(1UL << nr);
}
void open_softirq(int nr, void (*action)(struct softirq_action *))
{
softirq_vec[nr].action = action;
}
/*
* Tasklets
*/
struct tasklet_head {
struct tasklet_struct *head;
struct tasklet_struct **tail;
};
static DEFINE_PER_CPU(struct tasklet_head, tasklet_vec);
static DEFINE_PER_CPU(struct tasklet_head, tasklet_hi_vec);
static void __tasklet_schedule_common(struct tasklet_struct *t,
struct tasklet_head __percpu *headp,
unsigned int softirq_nr)
{
struct tasklet_head *head;
unsigned long flags;
local_irq_save(flags);
head = this_cpu_ptr(headp);
t->next = NULL;
*head->tail = t;
head->tail = &(t->next);
raise_softirq_irqoff(softirq_nr);
local_irq_restore(flags);
}
void __tasklet_schedule(struct tasklet_struct *t)
{
__tasklet_schedule_common(t, &tasklet_vec,
TASKLET_SOFTIRQ);
}
EXPORT_SYMBOL(__tasklet_schedule);
void __tasklet_hi_schedule(struct tasklet_struct *t)
{
__tasklet_schedule_common(t, &tasklet_hi_vec,
HI_SOFTIRQ);
}
EXPORT_SYMBOL(__tasklet_hi_schedule);
static void tasklet_action_common(struct softirq_action *a,
struct tasklet_head *tl_head,
unsigned int softirq_nr)
{
struct tasklet_struct *list;
local_irq_disable();
list = tl_head->head;
tl_head->head = NULL;
tl_head->tail = &tl_head->head;
local_irq_enable();
while (list) {
struct tasklet_struct *t = list;
list = list->next;
if (tasklet_trylock(t)) {
if (!atomic_read(&t->count)) {
if (!test_and_clear_bit(TASKLET_STATE_SCHED,
&t->state))
BUG();
t->func(t->data);
tasklet_unlock(t);
continue;
}
tasklet_unlock(t);
}
local_irq_disable();
t->next = NULL;
*tl_head->tail = t;
tl_head->tail = &t->next;
__raise_softirq_irqoff(softirq_nr);
local_irq_enable();
}
}
static __latent_entropy void tasklet_action(struct softirq_action *a)
{
tasklet_action_common(a, this_cpu_ptr(&tasklet_vec), TASKLET_SOFTIRQ);
}
static __latent_entropy void tasklet_hi_action(struct softirq_action *a)
{
tasklet_action_common(a, this_cpu_ptr(&tasklet_hi_vec), HI_SOFTIRQ);
}
void tasklet_init(struct tasklet_struct *t,
void (*func)(unsigned long), unsigned long data)
{
t->next = NULL;
t->state = 0;
atomic_set(&t->count, 0);
t->func = func;
t->data = data;
}
EXPORT_SYMBOL(tasklet_init);
void tasklet_kill(struct tasklet_struct *t)
{
if (in_interrupt())
pr_notice("Attempt to kill tasklet from interrupt\n");
while (test_and_set_bit(TASKLET_STATE_SCHED, &t->state)) {
do {
yield();
} while (test_bit(TASKLET_STATE_SCHED, &t->state));
}
tasklet_unlock_wait(t);
clear_bit(TASKLET_STATE_SCHED, &t->state);
}
EXPORT_SYMBOL(tasklet_kill);
/*
* tasklet_hrtimer
*/
/*
* The trampoline is called when the hrtimer expires. It schedules a tasklet
* to run __tasklet_hrtimer_trampoline() which in turn will call the intended
* hrtimer callback, but from softirq context.
*/
static enum hrtimer_restart __hrtimer_tasklet_trampoline(struct hrtimer *timer)
{
struct tasklet_hrtimer *ttimer =
container_of(timer, struct tasklet_hrtimer, timer);
tasklet_hi_schedule(&ttimer->tasklet);
return HRTIMER_NORESTART;
}
/*
* Helper function which calls the hrtimer callback from
* tasklet/softirq context
*/
static void __tasklet_hrtimer_trampoline(unsigned long data)
{
struct tasklet_hrtimer *ttimer = (void *)data;
enum hrtimer_restart restart;
restart = ttimer->function(&ttimer->timer);
if (restart != HRTIMER_NORESTART)
hrtimer_restart(&ttimer->timer);
}
/**
* tasklet_hrtimer_init - Init a tasklet/hrtimer combo for softirq callbacks
* @ttimer: tasklet_hrtimer which is initialized
* @function: hrtimer callback function which gets called from softirq context
* @which_clock: clock id (CLOCK_MONOTONIC/CLOCK_REALTIME)
* @mode: hrtimer mode (HRTIMER_MODE_ABS/HRTIMER_MODE_REL)
*/
void tasklet_hrtimer_init(struct tasklet_hrtimer *ttimer,
enum hrtimer_restart (*function)(struct hrtimer *),
clockid_t which_clock, enum hrtimer_mode mode)
{
hrtimer_init(&ttimer->timer, which_clock, mode);
ttimer->timer.function = __hrtimer_tasklet_trampoline;
tasklet_init(&ttimer->tasklet, __tasklet_hrtimer_trampoline,
(unsigned long)ttimer);
ttimer->function = function;
}
EXPORT_SYMBOL_GPL(tasklet_hrtimer_init);
void __init softirq_init(void)
{
int cpu;
for_each_possible_cpu(cpu) {
per_cpu(tasklet_vec, cpu).tail =
&per_cpu(tasklet_vec, cpu).head;
per_cpu(tasklet_hi_vec, cpu).tail =
&per_cpu(tasklet_hi_vec, cpu).head;
}
open_softirq(TASKLET_SOFTIRQ, tasklet_action);
open_softirq(HI_SOFTIRQ, tasklet_hi_action);
}
static int ksoftirqd_should_run(unsigned int cpu)
{
return local_softirq_pending();
}
static void run_ksoftirqd(unsigned int cpu)
{
local_irq_disable();
if (local_softirq_pending()) {
/*
* We can safely run softirq on inline stack, as we are not deep
* in the task stack here.
*/
__do_softirq();
local_irq_enable();
cond_resched();
return;
}
local_irq_enable();
}
#ifdef CONFIG_HOTPLUG_CPU
/*
* tasklet_kill_immediate is called to remove a tasklet which can already be
* scheduled for execution on @cpu.
*
* Unlike tasklet_kill, this function removes the tasklet
* _immediately_, even if the tasklet is in TASKLET_STATE_SCHED state.
*
* When this function is called, @cpu must be in the CPU_DEAD state.
*/
void tasklet_kill_immediate(struct tasklet_struct *t, unsigned int cpu)
{
struct tasklet_struct **i;
BUG_ON(cpu_online(cpu));
BUG_ON(test_bit(TASKLET_STATE_RUN, &t->state));
if (!test_bit(TASKLET_STATE_SCHED, &t->state))
return;
/* CPU is dead, so no lock needed. */
for (i = &per_cpu(tasklet_vec, cpu).head; *i; i = &(*i)->next) {
if (*i == t) {
*i = t->next;
/* If this was the tail element, move the tail ptr */
if (*i == NULL)
per_cpu(tasklet_vec, cpu).tail = i;
return;
}
}
BUG();
}
static int takeover_tasklets(unsigned int cpu)
{
/* CPU is dead, so no lock needed. */
local_irq_disable();
/* Find end, append list for that CPU. */
if (&per_cpu(tasklet_vec, cpu).head != per_cpu(tasklet_vec, cpu).tail) {
*__this_cpu_read(tasklet_vec.tail) = per_cpu(tasklet_vec, cpu).head;
this_cpu_write(tasklet_vec.tail, per_cpu(tasklet_vec, cpu).tail);
per_cpu(tasklet_vec, cpu).head = NULL;
per_cpu(tasklet_vec, cpu).tail = &per_cpu(tasklet_vec, cpu).head;
}
raise_softirq_irqoff(TASKLET_SOFTIRQ);
if (&per_cpu(tasklet_hi_vec, cpu).head != per_cpu(tasklet_hi_vec, cpu).tail) {
*__this_cpu_read(tasklet_hi_vec.tail) = per_cpu(tasklet_hi_vec, cpu).head;
__this_cpu_write(tasklet_hi_vec.tail, per_cpu(tasklet_hi_vec, cpu).tail);
per_cpu(tasklet_hi_vec, cpu).head = NULL;
per_cpu(tasklet_hi_vec, cpu).tail = &per_cpu(tasklet_hi_vec, cpu).head;
}
raise_softirq_irqoff(HI_SOFTIRQ);
local_irq_enable();
return 0;
}
#else
#define takeover_tasklets NULL
#endif /* CONFIG_HOTPLUG_CPU */
static struct smp_hotplug_thread softirq_threads = {
.store = &ksoftirqd,
.thread_should_run = ksoftirqd_should_run,
.thread_fn = run_ksoftirqd,
.thread_comm = "ksoftirqd/%u",
};
static __init int spawn_ksoftirqd(void)
{
cpuhp_setup_state_nocalls(CPUHP_SOFTIRQ_DEAD, "softirq:dead", NULL,
takeover_tasklets);
BUG_ON(smpboot_register_percpu_thread(&softirq_threads));
return 0;
}
early_initcall(spawn_ksoftirqd);
/*
* [ These __weak aliases are kept in a separate compilation unit, so that
* GCC does not inline them incorrectly. ]
*/
int __init __weak early_irq_init(void)
{
return 0;
}
int __init __weak arch_probe_nr_irqs(void)
{
return NR_IRQS_LEGACY;
}
int __init __weak arch_early_irq_init(void)
{
return 0;
}
genirq: x86: Ensure that dynamic irq allocation does not conflict On x86 the allocation of irq descriptors may allocate interrupts which are in the range of the GSI interrupts. That's wrong as those interrupts are hardwired and we don't have the irq domain translation like PPC. So one of these interrupts can be hooked up later to one of the devices which are hard wired to it and the io_apic init code for that particular interrupt line happily reuses that descriptor with a completely different configuration so hell breaks lose. Inside x86 we allocate dynamic interrupts from above nr_gsi_irqs, except for a few usage sites which have not yet blown up in our face for whatever reason. But for drivers which need an irq range, like the GPIO drivers, we have no limit in place and we don't want to expose such a detail to a driver. To cure this introduce a function which an architecture can implement to impose a lower bound on the dynamic interrupt allocations. Implement it for x86 and set the lower bound to nr_gsi_irqs, which is the end of the hardwired interrupt space, so all dynamic allocations happen above. That not only allows the GPIO driver to work sanely, it also protects the bogus callsites of create_irq_nr() in hpet, uv, irq_remapping and htirq code. They need to be cleaned up as well, but that's a separate issue. Reported-by: Jin Yao <yao.jin@linux.intel.com> Signed-off-by: Thomas Gleixner <tglx@linutronix.de> Tested-by: Mika Westerberg <mika.westerberg@linux.intel.com> Cc: Mathias Nyman <mathias.nyman@linux.intel.com> Cc: Linus Torvalds <torvalds@linux-foundation.org> Cc: Grant Likely <grant.likely@linaro.org> Cc: H. Peter Anvin <hpa@linux.intel.com> Cc: Rafael J. Wysocki <rafael.j.wysocki@intel.com> Cc: Andy Shevchenko <andriy.shevchenko@linux.intel.com> Cc: Krogerus Heikki <heikki.krogerus@intel.com> Cc: Linus Walleij <linus.walleij@linaro.org> Link: http://lkml.kernel.org/r/alpine.DEB.2.02.1404241617360.28206@ionos.tec.linutronix.de Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
2014-04-24 07:50:53 +00:00
unsigned int __weak arch_dynirq_lower_bound(unsigned int from)
{
return from;
}