RCU stall when using function_graph
Subsystems (2)
Quoted 5 lines
Hi Paul,
I have been trying to set the
function_graphtracer for ftrace and each time I get a CPU stall.
Quoted 8 lines
How to reproduce:
-----------------echo
function_graph>/sys/kernel/debug/tracing/current_tracerThis error appears with v4.13-rc3 and v4.12-rc6.
Is it something already reported ?
I have seen this sort of thing, but only when actually dumping the trace out, and I though those got fixed. You are seeing this just accumulating the trace?
These RCU CPU stall warnings usually occur when something grabs hold of a CPU for too long, as in 21 seconds or so. One way that they can happen is excessive lock contention, another is having the kernel run through too much data at one shot.
Adding Steven Rostedt on CC for his thoughts.
Thanx, Paul
Quoted 82 lines
[ ... ]
[ 472.934986] INFO: rcu_preempt detected stalls on CPUs/tasks: [ 472.941302] 0-...: (3 GPs behind) idle=d1a/2/0 softirq=2556/2557 fqs=3195 [ 472.948982] (detected by 3, t=6502 jiffies, g=549, c=548, q=10) [ 472.955180] Task dump for CPU 0: [ 472.958513] swapper/0 R running task 0 0 0 0x00000002 [ 550.954786] INFO: rcu_preempt detected stalls on CPUs/tasks: [ 550.960657] 0-...: (3 GPs behind) idle=d1a/2/0 softirq=2556/2557 fqs=12926 [ 550.967897] (detected by 3, t=26007 jiffies, g=549, c=548, q=27) [ 550.974181] Task dump for CPU 0: [ 550.977513] swapper/0 R running task 0 0 0 0x00000002 [ 628.974788] INFO: rcu_preempt detected stalls on CPUs/tasks: [ 628.980659] 0-...: (3 GPs behind) idle=d1a/2/0 softirq=2556/2557 fqs=22666 [ 628.987900] (detected by 3, t=45512 jiffies, g=549, c=548, q=43) [ 628.994184] Task dump for CPU 0: [ 628.997518] swapper/0 R running task 0 0 0 0x00000002 [ 706.994787] INFO: rcu_preempt detected stalls on CPUs/tasks: [ 707.000658] 0-...: (3 GPs behind) idle=d1a/2/0 softirq=2556/2557 fqs=31785 [ 707.007899] (detected by 3, t=65017 jiffies, g=549, c=548, q=74) [ 707.014183] Task dump for CPU 0: [ 707.017516] swapper/0 R running task 0 0 0 0x00000002 [ 707.024902] rcu_preempt kthread starved for 1261 jiffies! g549 c548 f0x0 RCU_GP_WAIT_FQS(3) ->state=0x1 [ 707.034548] rcu_preempt S 0 8 2 0x00000000 [ 726.049583] INFO: task jbd2/mmcblk1p2-:1120 blocked for more than 120 seconds. [ 726.061772] Not tainted 4.13.0-rc3-ktest-linaro-hikey #202 [ 726.074294] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 726.089027] jbd2/mmcblk1p2- D 0 1120 2 0x00000000 [ 726.126737] INFO: task rs:main Q:Reg:2272 blocked for more than 120 seconds. [ 726.140484] Not tainted 4.13.0-rc3-ktest-linaro-hikey #202 [ 726.154515] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 726.170140] rs:main Q:Reg D 0 2272 1 0x00000000 [ 726.205626] [ 726.205626] Showing all locks held in the system: [ 726.272952] [ 726.278194] ============================================= [ 726.278194] [ 785.014779] INFO: rcu_preempt detected stalls on CPUs/tasks: [ 785.020658] 0-...: (3 GPs behind) idle=d1a/2/0 softirq=2556/2557 fqs=41387 [ 785.027905] (detected by 3, t=84524 jiffies, g=549, c=548, q=94) [ 785.034194] Task dump for CPU 0: [ 785.037532] swapper/0 R running task 0 0 0 0x00000002 [ 846.879753] INFO: task jbd2/mmcblk1p2-:1120 blocked for more than 120 seconds. [ 846.892966] Not tainted 4.13.0-rc3-ktest-linaro-hikey #202 [ 846.909597] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 846.926726] jbd2/mmcblk1p2- D 0 1120 2 0x00000000 [ 846.981356] INFO: task rs:main Q:Reg:2272 blocked for more than 120 seconds. [ 847.001593] Not tainted 4.13.0-rc3-ktest-linaro-hikey #202 [ 847.017642] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 847.034135] rs:main Q:Reg D 0 2272 1 0x00000000 [ 847.077209] [ 847.077209] Showing all locks held in the system: [ 847.180736] [ 847.186753] ============================================= [ 847.186753] [ 863.034778] INFO: rcu_preempt detected stalls on CPUs/tasks: [ 863.040660] 0-...: (3 GPs behind) idle=d1a/2/0 softirq=2556/2557 fqs=50881 [ 863.047907] (detected by 4, t=104030 jiffies, g=549, c=548, q=114) [ 863.054373] Task dump for CPU 0: [ 863.057711] swapper/0 R running task 0 0 0 0x00000002 [ 941.054780] INFO: rcu_preempt detected stalls on CPUs/tasks: [ 941.060659] 0-...: (3 GPs behind) idle=d1a/2/0 softirq=2556/2557 fqs=59726 [ 941.067903] (detected by 7, t=123535 jiffies, g=549, c=548, q=303) [ 941.074370] Task dump for CPU 0: [ 941.077708] swapper/0 R running task 0 0 0 0x00000002 [ 967.711682] INFO: task jbd2/mmcblk1p2-:1120 blocked for more than 120 seconds. [ 967.725731] Not tainted 4.13.0-rc3-ktest-linaro-hikey #202 [ 967.743347] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 967.765658] jbd2/mmcblk1p2- D 0 1120 2 0x00000000 [ 967.810595] INFO: task rs:main Q:Reg:2272 blocked for more than 120 seconds. [ 967.825360] Not tainted 4.13.0-rc3-ktest-linaro-hikey #202 [ 967.838595] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 967.854460] rs:main Q:Reg D 0 2272 1 0x00000000 [ 967.897804] [ 967.897804] Showing all locks held in the system: [ 967.997067] [ 968.002570] =============================================[ ... ]
etc ...
Quoted 10 lines · Paul E. McKenney
Quoted 10 lines
Hi Paul,
I have been trying to set the
function_graphtracer for ftrace and each time I get a CPU stall.How to reproduce:
-----------------echo
function_graph>/sys/kernel/debug/tracing/current_tracer
Quoted 7 lines
Quoted 3 lines
This error appears with v4.13-rc3 and v4.12-rc6.
Is it something already reported ?
I have seen this sort of thing, but only when actually dumping the trace out, and I though those got fixed. You are seeing this just accumulating the trace?
No, just by changing the tracer. It is the first operation I do after rebooting and it is reproducible each time. That happens on an ARM64 platform.
Quoted 92 lines
These RCU CPU stall warnings usually occur when something grabs hold of a CPU for too long, as in 21 seconds or so. One way that they can happen is excessive lock contention, another is having the kernel run through too much data at one shot.
Adding Steven Rostedt on CC for his thoughts.
Thanx, Paul
Quoted 82 lines
[ ... ]
[ 472.934986] INFO: rcu_preempt detected stalls on CPUs/tasks: [ 472.941302] 0-...: (3 GPs behind) idle=d1a/2/0 softirq=2556/2557 fqs=3195 [ 472.948982] (detected by 3, t=6502 jiffies, g=549, c=548, q=10) [ 472.955180] Task dump for CPU 0: [ 472.958513] swapper/0 R running task 0 0 0 0x00000002 [ 550.954786] INFO: rcu_preempt detected stalls on CPUs/tasks: [ 550.960657] 0-...: (3 GPs behind) idle=d1a/2/0 softirq=2556/2557 fqs=12926 [ 550.967897] (detected by 3, t=26007 jiffies, g=549, c=548, q=27) [ 550.974181] Task dump for CPU 0: [ 550.977513] swapper/0 R running task 0 0 0 0x00000002 [ 628.974788] INFO: rcu_preempt detected stalls on CPUs/tasks: [ 628.980659] 0-...: (3 GPs behind) idle=d1a/2/0 softirq=2556/2557 fqs=22666 [ 628.987900] (detected by 3, t=45512 jiffies, g=549, c=548, q=43) [ 628.994184] Task dump for CPU 0: [ 628.997518] swapper/0 R running task 0 0 0 0x00000002 [ 706.994787] INFO: rcu_preempt detected stalls on CPUs/tasks: [ 707.000658] 0-...: (3 GPs behind) idle=d1a/2/0 softirq=2556/2557 fqs=31785 [ 707.007899] (detected by 3, t=65017 jiffies, g=549, c=548, q=74) [ 707.014183] Task dump for CPU 0: [ 707.017516] swapper/0 R running task 0 0 0 0x00000002 [ 707.024902] rcu_preempt kthread starved for 1261 jiffies! g549 c548 f0x0 RCU_GP_WAIT_FQS(3) ->state=0x1 [ 707.034548] rcu_preempt S 0 8 2 0x00000000 [ 726.049583] INFO: task jbd2/mmcblk1p2-:1120 blocked for more than 120 seconds. [ 726.061772] Not tainted 4.13.0-rc3-ktest-linaro-hikey #202 [ 726.074294] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 726.089027] jbd2/mmcblk1p2- D 0 1120 2 0x00000000 [ 726.126737] INFO: task rs:main Q:Reg:2272 blocked for more than 120 seconds. [ 726.140484] Not tainted 4.13.0-rc3-ktest-linaro-hikey #202 [ 726.154515] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 726.170140] rs:main Q:Reg D 0 2272 1 0x00000000 [ 726.205626] [ 726.205626] Showing all locks held in the system: [ 726.272952] [ 726.278194] ============================================= [ 726.278194] [ 785.014779] INFO: rcu_preempt detected stalls on CPUs/tasks: [ 785.020658] 0-...: (3 GPs behind) idle=d1a/2/0 softirq=2556/2557 fqs=41387 [ 785.027905] (detected by 3, t=84524 jiffies, g=549, c=548, q=94) [ 785.034194] Task dump for CPU 0: [ 785.037532] swapper/0 R running task 0 0 0 0x00000002 [ 846.879753] INFO: task jbd2/mmcblk1p2-:1120 blocked for more than 120 seconds. [ 846.892966] Not tainted 4.13.0-rc3-ktest-linaro-hikey #202 [ 846.909597] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 846.926726] jbd2/mmcblk1p2- D 0 1120 2 0x00000000 [ 846.981356] INFO: task rs:main Q:Reg:2272 blocked for more than 120 seconds. [ 847.001593] Not tainted 4.13.0-rc3-ktest-linaro-hikey #202 [ 847.017642] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 847.034135] rs:main Q:Reg D 0 2272 1 0x00000000 [ 847.077209] [ 847.077209] Showing all locks held in the system: [ 847.180736] [ 847.186753] ============================================= [ 847.186753] [ 863.034778] INFO: rcu_preempt detected stalls on CPUs/tasks: [ 863.040660] 0-...: (3 GPs behind) idle=d1a/2/0 softirq=2556/2557 fqs=50881 [ 863.047907] (detected by 4, t=104030 jiffies, g=549, c=548, q=114) [ 863.054373] Task dump for CPU 0: [ 863.057711] swapper/0 R running task 0 0 0 0x00000002 [ 941.054780] INFO: rcu_preempt detected stalls on CPUs/tasks: [ 941.060659] 0-...: (3 GPs behind) idle=d1a/2/0 softirq=2556/2557 fqs=59726 [ 941.067903] (detected by 7, t=123535 jiffies, g=549, c=548, q=303) [ 941.074370] Task dump for CPU 0: [ 941.077708] swapper/0 R running task 0 0 0 0x00000002 [ 967.711682] INFO: task jbd2/mmcblk1p2-:1120 blocked for more than 120 seconds. [ 967.725731] Not tainted 4.13.0-rc3-ktest-linaro-hikey #202 [ 967.743347] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 967.765658] jbd2/mmcblk1p2- D 0 1120 2 0x00000000 [ 967.810595] INFO: task rs:main Q:Reg:2272 blocked for more than 120 seconds. [ 967.825360] Not tainted 4.13.0-rc3-ktest-linaro-hikey #202 [ 967.838595] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 967.854460] rs:main Q:Reg D 0 2272 1 0x00000000 [ 967.897804] [ 967.897804] Showing all locks held in the system: [ 967.997067] [ 968.002570] =============================================[ ... ]
etc ...
Signature · 5 lines
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs Follow Linaro: <http://www.facebook.com/pages/Linaro> Facebook | <http://twitter.com/#!/linaroorg> Twitter | <http://www.linaro.org/linaro-blog/> Blog
Steven Rostedt
reviewerQuoted 6 lines · On Wed, 2 Aug 2017 00:15:44 +0200 Daniel Lezcano <[email protected]>…
Quoted 5 lines · Paul E. McKenney
Quoted 5 lines
Hi Paul,
I have been trying to set the
function_graphtracer for ftrace and each time I get a CPU stall.
Quoted 6 lines
Quoted 6 lines
Quoted 6 lines
How to reproduce:
-----------------echo
function_graph>/sys/kernel/debug/tracing/current_tracerThis error appears with v4.13-rc3 and v4.12-rc6.
Can you bisect this? It may be due to this commit:
0598e4f08 ("ftrace: Add use of synchronize_rcu_tasks() with dynamic trampolines")
-- Steve
Quoted 19 lines
Quoted 6 lines
Quoted 2 lines
Is it something already reported ?
I have seen this sort of thing, but only when actually dumping the trace out, and I though those got fixed. You are seeing this just accumulating the trace?
No, just by changing the tracer. It is the first operation I do after rebooting and it is reproducible each time. That happens on an ARM64 platform.
Quoted 7 lines
These RCU CPU stall warnings usually occur when something grabs hold of a CPU for too long, as in 21 seconds or so. One way that they can happen is excessive lock contention, another is having the kernel run through too much data at one shot.
Adding Steven Rostedt on CC for his thoughts.
Quoted 12 lines · Steven Rostedt
Quoted 9 lines · On Wed, 2 Aug 2017 00:15:44 +0200 Daniel Lezcano <[email protected]>…
Quoted 8 lines · Paul E. McKenney
Quoted 8 lines
Hi Paul,
I have been trying to set the
function_graphtracer for ftrace and each time I get a CPU stall.How to reproduce:
-----------------
Quoted 7 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
echo
function_graph>/sys/kernel/debug/tracing/current_tracerThis error appears with v4.13-rc3 and v4.12-rc6.
Can you bisect this? It may be due to this commit:
0598e4f08("ftrace: Add use ofsynchronize_rcu_tasks()with dynamic trampolines")
Hi Steve,
I git bisected but each time the issue occured. I went through the different version down to v4.4 where the board was not fully supported and it ended up to have the same issue.
Finally, I had the intuition it could be related to the wall time (there is no RTC clock with battery on the board and the wall time is Jan 1st, 1970).
Setting up the with ntpdate solved the problem.
Even if it is rarely the case to have the time not set, is it normal to have a RCU cpu stall ?
Quoted 19 lines
Quoted 19 lines
Quoted 6 lines
Quoted 2 lines
Is it something already reported ?
I have seen this sort of thing, but only when actually dumping the trace out, and I though those got fixed. You are seeing this just accumulating the trace?
No, just by changing the tracer. It is the first operation I do after rebooting and it is reproducible each time. That happens on an ARM64 platform.
Quoted 7 lines
These RCU CPU stall warnings usually occur when something grabs hold of a CPU for too long, as in 21 seconds or so. One way that they can happen is excessive lock contention, another is having the kernel run through too much data at one shot.
Adding Steven Rostedt on CC for his thoughts.
Signature · 5 lines
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs Follow Linaro: <http://www.facebook.com/pages/Linaro> Facebook | <http://twitter.com/#!/linaroorg> Twitter | <http://www.linaro.org/linaro-blog/> Blog
Steven Rostedt
reviewerQuoted 27 lines · On Wed, 2 Aug 2017 14:42:39 +0200 Daniel Lezcano <[email protected]>…
Quoted 12 lines · Steven Rostedt
Quoted 9 lines · On Wed, 2 Aug 2017 00:15:44 +0200 Daniel Lezcano <[email protected]>…
Quoted 8 lines · Paul E. McKenney
Quoted 8 lines
Hi Paul,
I have been trying to set the
function_graphtracer for ftrace and each time I get a CPU stall.How to reproduce:
-----------------Quoted 7 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
echo
function_graph>/sys/kernel/debug/tracing/current_tracerThis error appears with v4.13-rc3 and v4.12-rc6.
Can you bisect this? It may be due to this commit:
0598e4f08("ftrace: Add use ofsynchronize_rcu_tasks()with dynamic trampolines")Hi Steve,
I git bisected but each time the issue occured. I went through the different version down to v4.4 where the board was not fully supported and it ended up to have the same issue.
Quoted 9 lines
Finally, I had the intuition it could be related to the wall time (there is no RTC clock with battery on the board and the wall time is Jan 1st, 1970).
Setting up the with ntpdate solved the problem.
Even if it is rarely the case to have the time not set, is it normal to have a RCU cpu stall ?
BTW, function_graph tracer is the most invasive of the tracers. It's 4x slower than function tracer. I'm wondering if the tracer isn't the cause, but just slows things down enough to cause a some other race condition that triggers the bug.
-- Steve
Quoted 40 lines · Steven Rostedt
Quoted 27 lines · On Wed, 2 Aug 2017 14:42:39 +0200 Daniel Lezcano <[email protected]>…
Quoted 12 lines · Steven Rostedt
Quoted 9 lines · On Wed, 2 Aug 2017 00:15:44 +0200 Daniel Lezcano <[email protected]>…
Quoted 8 lines · Paul E. McKenney
Quoted 8 lines
Hi Paul,
I have been trying to set the
function_graphtracer for ftrace and each time I get a CPU stall.How to reproduce:
-----------------Quoted 7 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
echo
function_graph>/sys/kernel/debug/tracing/current_tracerThis error appears with v4.13-rc3 and v4.12-rc6.
Can you bisect this? It may be due to this commit:
0598e4f08("ftrace: Add use ofsynchronize_rcu_tasks()with dynamic trampolines")Hi Steve,
I git bisected but each time the issue occured. I went through the different version down to v4.4 where the board was not fully supported and it ended up to have the same issue.
Quoted 9 lines
Finally, I had the intuition it could be related to the wall time (there is no RTC clock with battery on the board and the wall time is Jan 1st, 1970).
Setting up the with ntpdate solved the problem.
Even if it is rarely the case to have the time not set, is it normal to have a RCU cpu stall ?
Quoted 4 lines
BTW,
function_graphtracer is the most invasive of the tracers. It's 4x slower than function tracer. I'm wondering if the tracer isn't the cause, but just slows things down enough to cause a some other race condition that triggers the bug.
Easy to check! Use the rcupdate.rcu_cpu_stall_timeout kernel boot parameter to increase this timeout by a factor of four. Mainline default is 21 seconds, but many distros set it to 60 seconds. You can always check sysfs to find the value for your system, or CONFIG_RCU_CPU_STALL_TIMEOUT in your .config file.
Thanx, Paul
Quoted 26 lines · Steven Rostedt
Quoted 23 lines · On Wed, 2 Aug 2017 14:42:39 +0200 Daniel Lezcano <[email protected]>…
Quoted 12 lines · Steven Rostedt
Quoted 9 lines · On Wed, 2 Aug 2017 00:15:44 +0200 Daniel Lezcano <[email protected]>…
Quoted 8 lines · Paul E. McKenney
Quoted 8 lines
Hi Paul,
I have been trying to set the
function_graphtracer for ftrace and each time I get a CPU stall.How to reproduce:
-----------------Quoted 7 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
echo
function_graph>/sys/kernel/debug/tracing/current_tracerThis error appears with v4.13-rc3 and v4.12-rc6.
Can you bisect this? It may be due to this commit:
0598e4f08("ftrace: Add use ofsynchronize_rcu_tasks()with dynamic trampolines")Hi Steve,
Quoted 8 lines
Quoted 8 lines
I git bisected but each time the issue occured. I went through the different version down to v4.4 where the board was not fully supported and it ended up to have the same issue.
Finally, I had the intuition it could be related to the wall time (there is no RTC clock with battery on the board and the wall time is Jan 1st, 1970).
Setting up the with ntpdate solved the problem.
Actually, it did not solve the problem. The function_graph trace is set, I can use the system but after awhile (no tracing enabled at anytime), the stall appears.
Quoted 9 lines
Quoted 4 lines
Even if it is rarely the case to have the time not set, is it normal to have a RCU cpu stall ?
BTW,
function_graphtracer is the most invasive of the tracers. It's 4x slower than function tracer. I'm wondering if the tracer isn't the cause, but just slows things down enough to cause a some other race condition that triggers the bug.
Yes, that could be true.
I tried the following scenario:
- cpufreq governor => userspace + max_freq (1.2GHz)
- function_graph set ==> OK
- cpufreq governor => userspace + min_freq (200MHz)
- function_graph set ==> RCU stall
Beside that, I realize the board is constantly processing SOF interrupts every 124us, so that adds more overhead.
Removing the USB support, thus the associated processing for the SOF interrupts, I don't see anymore the RCU stall.
Is it the expected behavior to have the system hang after a RCU stall raises ?
Signature · 5 lines
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs Follow Linaro: <http://www.facebook.com/pages/Linaro> Facebook | <http://twitter.com/#!/linaroorg> Twitter | <http://www.linaro.org/linaro-blog/> Blog
Quoted 57 lines · Daniel Lezcano
Quoted 26 lines · Steven Rostedt
Quoted 23 lines · On Wed, 2 Aug 2017 14:42:39 +0200 Daniel Lezcano <[email protected]>…
Quoted 12 lines · Steven Rostedt
Quoted 9 lines · On Wed, 2 Aug 2017 00:15:44 +0200 Daniel Lezcano <[email protected]>…
Quoted 8 lines · Paul E. McKenney
Quoted 8 lines
Hi Paul,
I have been trying to set the
function_graphtracer for ftrace and each time I get a CPU stall.How to reproduce:
-----------------Quoted 7 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
echo
function_graph>/sys/kernel/debug/tracing/current_tracerThis error appears with v4.13-rc3 and v4.12-rc6.
Can you bisect this? It may be due to this commit:
0598e4f08("ftrace: Add use ofsynchronize_rcu_tasks()with dynamic trampolines")Hi Steve,
Quoted 8 lines
Quoted 8 lines
I git bisected but each time the issue occured. I went through the different version down to v4.4 where the board was not fully supported and it ended up to have the same issue.
Finally, I had the intuition it could be related to the wall time (there is no RTC clock with battery on the board and the wall time is Jan 1st, 1970).
Setting up the with ntpdate solved the problem.
Actually, it did not solve the problem. The
function_graphtrace is set, I can use the system but after awhile (no tracing enabled at anytime), the stall appears.Quoted 9 lines
Quoted 4 lines
Even if it is rarely the case to have the time not set, is it normal to have a RCU cpu stall ?
BTW,
function_graphtracer is the most invasive of the tracers. It's 4x slower than function tracer. I'm wondering if the tracer isn't the cause, but just slows things down enough to cause a some other race condition that triggers the bug.Yes, that could be true.
I tried the following scenario:
- cpufreq governor => userspace +
max_freq(1.2GHz)
-function_graphset ==> OK
Quoted 8 lines
- cpufreq governor => userspace +
min_freq(200MHz)
-function_graphset ==> RCU stallBeside that, I realize the board is constantly processing SOF interrupts every 124us, so that adds more overhead.
Removing the USB support, thus the associated processing for the SOF interrupts, I don't see anymore the RCU stall.
Looks like Steve called this one! ;-)
Quoted 2 lines
Is it the expected behavior to have the system hang after a RCU stall raises ?
No, but if NMI stack traces are enabled and there are any NMI problems, bad things can happen. In addition, the bulk of output can cause problems if you have a slow console connection.
Thanx, Paul
On Thu, Aug 03, 2017 at 05:44:21AM -0700, Paul E. McKenney wrote:
[ ... ]
Quoted 15 lines
Quoted 15 lines
Quoted 4 lines
BTW,
function_graphtracer is the most invasive of the tracers. It's 4x slower than function tracer. I'm wondering if the tracer isn't the cause, but just slows things down enough to cause a some other race condition that triggers the bug.Yes, that could be true.
I tried the following scenario:
- cpufreq governor => userspace +
max_freq(1.2GHz)
-function_graphset ==> OK- cpufreq governor => userspace +
min_freq(200MHz)
-function_graphset ==> RCU stall
Quoted 7 lines
Quoted 5 lines
Beside that, I realize the board is constantly processing SOF interrupts every 124us, so that adds more overhead.
Removing the USB support, thus the associated processing for the SOF interrupts, I don't see anymore the RCU stall.
Looks like Steve called this one! ;-)
Yep :)
Quoted 6 lines
Quoted 2 lines
Is it the expected behavior to have the system hang after a RCU stall raises ?
No, but if NMI stack traces are enabled and there are any NMI problems, bad things can happen. In addition, the bulk of output can cause problems if you have a slow console connection.
Ok, thanks.
-- Daniel
Signature · 5 lines
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs Follow Linaro: <http://www.facebook.com/pages/Linaro> Facebook | <http://twitter.com/#!/linaroorg> Twitter | <http://www.linaro.org/linaro-blog/> Blog
A message that is not in the archive (sent off-list or not yet received): CAOoBcBXo-=VYy2+TYEp=8+WSkOpDBr1x6uY=-r_GnTFKctXndQ@mail.gmail.com
A message that is not in the archive (sent off-list or not yet received): CAOoBcBVKpQkAVXji5qQu8r8GErqxpy9Ae9N97NhGpOQPgXudZg@mail.gmail.com
A message that is not in the archive (sent off-list or not yet received): CAOoBcBU00VRXmrNNEOjJHgXf9BimxKYOorJC0d3766mNdda=Bg@mail.gmail.com
Quoted 9 lines · 김동현
Dear All
As for me, after configuring
function_graphas below, crash disappears. "echo 0 > d/tracing/tracing_on"
"sleep 1""echo
function_graph> d/tracing/current_tracer"
"sleep 1"
Quoted 8 lines
"echo
smp_call_function_single> d/tracing/set_ftrace_filter" adb shell "sleep 1""echo 1 > d/tracing/tracing_on"
adb shell "sleep 1"Right after
function_graphis enabled, too many logs are traced upon IRQ transaction which many times eventually causes stall.
That would do it!
Hmmm...
Steven, would it be helpful if RCU were to inform tracing (say) halfway through the RCU CPU stall interval, allowing the tracer to do something like cond_resched_rcu_qs()? I can imagine all sorts of reasons why this wouldn't work, for example, if all the tracing was with irqs disabled or some such, but figured I should ask.
Does Guillermo's approach work for others?
Thanx, Paul
Quoted 53 lines
BR,
Guillermo Austin Kim2017. 8. 3. 오후 11:38에 "Daniel Lezcano" <[email protected]>님이 작성:
On Thu, Aug 03, 2017 at 05:44:21AM -0700, Paul E. McKenney wrote:
[ ... ]
Quoted 1 line
Quoted 1 line
Quoted 1 line
BTW,
function_graphtracer is the most invasive of the tracers. It's4x
Quoted 14 lines
Quoted 14 lines
Quoted 3 lines
slower than function tracer. I'm wondering if the tracer isn't the cause, but just slows things down enough to cause a some other race condition that triggers the bug.
Yes, that could be true.
I tried the following scenario:
- cpufreq governor => userspace +
max_freq(1.2GHz)
-function_graphset ==> OK- cpufreq governor => userspace +
min_freq(200MHz)
-function_graphset ==> RCU stallQuoted 7 lines
Quoted 5 lines
Beside that, I realize the board is constantly processing SOF interrupts every 124us, so that adds more overhead.
Removing the USB support, thus the associated processing for the SOF interrupts, I don't see anymore the RCU stall.
Looks like Steve called this one! ;-)
Yep :)
Quoted 6 lines
Quoted 2 lines
Is it the expected behavior to have the system hang after a RCU stall raises ?
No, but if NMI stack traces are enabled and there are any NMI problems, bad things can happen. In addition, the bulk of output can cause problems if you have a slow console connection.
Ok, thanks.
-- Daniel
--
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs
Follow Linaro: <http://www.facebook.com/pages/Linaro> Facebook |
<http://twitter.com/#!/linaroorg> Twitter |
<http://www.linaro.org/linaro-blog/> Blog
Quoted 11 lines · Paul E. McKenney
Quoted 10 lines · 김동현
Dear All
As for me, after configuring
function_graphas below, crash disappears. "echo 0 > d/tracing/tracing_on"
"sleep 1""echo
function_graph> d/tracing/current_tracer"
"sleep 1""echo
smp_call_function_single> d/tracing/set_ftrace_filter"
It will limit trace output to only for the filtered function (smp_call_function_single).
Quoted 12 lines
Quoted 7 lines
adb shell "sleep 1"
"echo 1 > d/tracing/tracing_on"
adb shell "sleep 1"Right after
function_graphis enabled, too many logs are traced upon IRQ transaction which many times eventually causes stall.That would do it!
Hmmm...
Quoted 7 lines
Steven, would it be helpful if RCU were to inform tracing (say) halfway through the RCU CPU stall interval, allowing the tracer to do something like
cond_resched_rcu_qs()? I can imagine all sorts of reasons why this wouldn't work, for example, if all the tracing was with irqs disabled or some such, but figured I should ask.Does Guillermo's approach work for others?
Limited output with a couple of filtered function will definitely not cause RCU schedule stall. But the question is whether we should expect a full function graph trace working on every platform or not (specially the one which generates high interrupts)?
Signature · 2 lines
Regards Pratyush
Quoted 36 lines · Pratyush Anand
Quoted 11 lines · Paul E. McKenney
Quoted 10 lines · 김동현
Dear All
As for me, after configuring
function_graphas below, crash disappears. "echo 0 > d/tracing/tracing_on"
"sleep 1""echo
function_graph> d/tracing/current_tracer"
"sleep 1""echo
smp_call_function_single> d/tracing/set_ftrace_filter"It will limit trace output to only for the filtered function (
smp_call_function_single).Quoted 18 lines
Quoted 7 lines
adb shell "sleep 1"
"echo 1 > d/tracing/tracing_on"
adb shell "sleep 1"Right after
function_graphis enabled, too many logs are traced upon IRQ transaction which many times eventually causes stall.That would do it!
Hmmm...
Steven, would it be helpful if RCU were to inform tracing (say) halfway through the RCU CPU stall interval, allowing the tracer to do something like
cond_resched_rcu_qs()? I can imagine all sorts of reasons why this wouldn't work, for example, if all the tracing was with irqs disabled or some such, but figured I should ask.
Quoted 6 lines
Quoted 1 line
Does Guillermo's approach work for others?
Limited output with a couple of filtered function will definitely not cause RCU schedule stall. But the question is whether we should expect a full function graph trace working on every platform or not (specially the one which generates high interrupts)?
It might well be that the user must disable RCU CPU stall warnings via the rcu_cpu_stall_suppress sysfs entry (or increase their timeout via the rcu_cpu_stall_timeout sysfs entry) before doing something that greatly increases overhead. Like enabling large quantities of tracing. ;-)
It -might- be possible to do this automatically, but reliable automation would require that tracing understand how often each function was called, which sounds to me to be a bit of a stretch.
Thoughts?
Thanx, Paul
Quoted 49 lines · Paul E. McKenney
Quoted 36 lines · Pratyush Anand
Quoted 11 lines · Paul E. McKenney
Quoted 10 lines · 김동현
Dear All
As for me, after configuring
function_graphas below, crash disappears. "echo 0 > d/tracing/tracing_on"
"sleep 1""echo
function_graph> d/tracing/current_tracer"
"sleep 1""echo
smp_call_function_single> d/tracing/set_ftrace_filter"It will limit trace output to only for the filtered function (
smp_call_function_single).Quoted 18 lines
Quoted 7 lines
adb shell "sleep 1"
"echo 1 > d/tracing/tracing_on"
adb shell "sleep 1"Right after
function_graphis enabled, too many logs are traced upon IRQ transaction which many times eventually causes stall.That would do it!
Hmmm...
Steven, would it be helpful if RCU were to inform tracing (say) halfway through the RCU CPU stall interval, allowing the tracer to do something like
cond_resched_rcu_qs()? I can imagine all sorts of reasons why this wouldn't work, for example, if all the tracing was with irqs disabled or some such, but figured I should ask.Quoted 6 lines
Quoted 1 line
Does Guillermo's approach work for others?
Limited output with a couple of filtered function will definitely not cause RCU schedule stall. But the question is whether we should expect a full function graph trace working on every platform or not (specially the one which generates high interrupts)?
It might well be that the user must disable RCU CPU stall warnings via the
rcu_cpu_stall_suppresssysfs entry (or increase their timeout via thrcu_cpu_stall_timeoutsysfs entry) before doing something that greatly increases overhead. Like enabling large quantities of tracing. ;-)
Quoted 5 lines
It -might- be possible to do this automatically, but reliable automation would require that tracing understand how often each function was called, which sounds to me to be a bit of a stretch.
Thoughts?
A random thought:
Is it possible to have a mid-timeout happening and store some information like the instruction pointer, so when the timeout happen we can compare if there was some progress, if yes, very likely, system performance collapsed and we are not fast enough.
Signature · 5 lines
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs Follow Linaro: <http://www.facebook.com/pages/Linaro> Facebook | <http://twitter.com/#!/linaroorg> Twitter | <http://www.linaro.org/linaro-blog/> Blog
Quoted 54 lines · Daniel Lezcano
Quoted 53 lines · Paul E. McKenney
Quoted 36 lines · Pratyush Anand
Quoted 11 lines · Paul E. McKenney
Quoted 10 lines · 김동현
Dear All
As for me, after configuring
function_graphas below, crash disappears. "echo 0 > d/tracing/tracing_on"
"sleep 1""echo
function_graph> d/tracing/current_tracer"
"sleep 1""echo
smp_call_function_single> d/tracing/set_ftrace_filter"It will limit trace output to only for the filtered function (
smp_call_function_single).Quoted 18 lines
Quoted 7 lines
adb shell "sleep 1"
"echo 1 > d/tracing/tracing_on"
adb shell "sleep 1"Right after
function_graphis enabled, too many logs are traced upon IRQ transaction which many times eventually causes stall.That would do it!
Hmmm...
Steven, would it be helpful if RCU were to inform tracing (say) halfway through the RCU CPU stall interval, allowing the tracer to do something like
cond_resched_rcu_qs()? I can imagine all sorts of reasons why this wouldn't work, for example, if all the tracing was with irqs disabled or some such, but figured I should ask.Quoted 6 lines
Quoted 1 line
Does Guillermo's approach work for others?
Limited output with a couple of filtered function will definitely not cause RCU schedule stall. But the question is whether we should expect a full function graph trace working on every platform or not (specially the one which generates high interrupts)?
It might well be that the user must disable RCU CPU stall warnings via the
rcu_cpu_stall_suppresssysfs entry (or increase their timeout via thrcu_cpu_stall_timeoutsysfs entry) before doing something that greatly increases overhead. Like enabling large quantities of tracing. ;-)It -might- be possible to do this automatically, but reliable automation would require that tracing understand how often each function was called, which sounds to me to be a bit of a stretch.
Quoted 8 lines
Quoted 1 line
Thoughts?
A random thought:
Is it possible to have a mid-timeout happening and store some information like the instruction pointer, so when the timeout happen we can compare if there was some progress, if yes, very likely, system performance collapsed and we are not fast enough.
RCU already does take various actions for an impending stall, so something could be done. But in most slowdowns, the instruction pointer will be changing rapidly, just not as rapidly as it would normally. So exactly how would the forward-progress comparison be carried out?
It would be easy to set up a notifier, so that if any notifier in the chain returned an error, stall warnings would be suppressed. It would be harder to figure out when to re-enable them, though I suppose that they could be suppressed only for the duration of the current grace period or some such.
But what exactly would you use such a notifier for?
Or am I misunderstanding your suggestion?
Thanx, Paul
Quoted 75 lines · Paul E. McKenney
Quoted 54 lines · Daniel Lezcano
Quoted 53 lines · Paul E. McKenney
Quoted 36 lines · Pratyush Anand
Quoted 11 lines · Paul E. McKenney
Quoted 10 lines · 김동현
Dear All
As for me, after configuring
function_graphas below, crash disappears. "echo 0 > d/tracing/tracing_on"
"sleep 1""echo
function_graph> d/tracing/current_tracer"
"sleep 1""echo
smp_call_function_single> d/tracing/set_ftrace_filter"It will limit trace output to only for the filtered function (
smp_call_function_single).Quoted 18 lines
Quoted 7 lines
adb shell "sleep 1"
"echo 1 > d/tracing/tracing_on"
adb shell "sleep 1"Right after
function_graphis enabled, too many logs are traced upon IRQ transaction which many times eventually causes stall.That would do it!
Hmmm...
Steven, would it be helpful if RCU were to inform tracing (say) halfway through the RCU CPU stall interval, allowing the tracer to do something like
cond_resched_rcu_qs()? I can imagine all sorts of reasons why this wouldn't work, for example, if all the tracing was with irqs disabled or some such, but figured I should ask.Quoted 6 lines
Quoted 1 line
Does Guillermo's approach work for others?
Limited output with a couple of filtered function will definitely not cause RCU schedule stall. But the question is whether we should expect a full function graph trace working on every platform or not (specially the one which generates high interrupts)?
It might well be that the user must disable RCU CPU stall warnings via the
rcu_cpu_stall_suppresssysfs entry (or increase their timeout via thrcu_cpu_stall_timeoutsysfs entry) before doing something that greatly increases overhead. Like enabling large quantities of tracing. ;-)It -might- be possible to do this automatically, but reliable automation would require that tracing understand how often each function was called, which sounds to me to be a bit of a stretch.
Quoted 8 lines
Quoted 1 line
Thoughts?
A random thought:
Is it possible to have a mid-timeout happening and store some information like the instruction pointer, so when the timeout happen we can compare if there was some progress, if yes, very likely, system performance collapsed and we are not fast enough.
RCU already does take various actions for an impending stall, so something could be done. But in most slowdowns, the instruction pointer will be changing rapidly, just not as rapidly as it would normally. So exactly how would the forward-progress comparison be carried out?
It would be easy to set up a notifier, so that if any notifier in the chain returned an error, stall warnings would be suppressed. It would be harder to figure out when to re-enable them, though I suppose that they could be suppressed only for the duration of the current grace period or some such.
Quoted 3 lines
But what exactly would you use such a notifier for?
Or am I misunderstanding your suggestion?
Well, may be the instruction pointer thing is not a good idea.
I learnt from this experience, an overloaded kernel with a lot of interrupts can hang the console and issue RCU stall.
However, someone else can face the same situation. Even if he reads the RCU/stallwarn.txt documentation, it will be hard to figure out the issue.
A message telling the grace period can't be reached because we are too busy processing interrupts would have helped but I understand it is not easy to implement.
Perhaps, adding a new bullet in the documentation can help:
"If the interrupt processing time is longer than the interval between each interrupt, the CPU will keep processing the interrupts without allowing the RCU's grace period kthread. This situation can happen if there is a highly rated number of interrupts and the function_graph tracer is enabled".
Signature · 5 lines
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs Follow Linaro: <http://www.facebook.com/pages/Linaro> Facebook | <http://twitter.com/#!/linaroorg> Twitter | <http://www.linaro.org/linaro-blog/> Blog
Quoted 92 lines · Daniel Lezcano
Quoted 75 lines · Paul E. McKenney
Quoted 54 lines · Daniel Lezcano
Quoted 53 lines · Paul E. McKenney
Quoted 36 lines · Pratyush Anand
Quoted 11 lines · Paul E. McKenney
Quoted 10 lines · 김동현
Dear All
As for me, after configuring
function_graphas below, crash disappears. "echo 0 > d/tracing/tracing_on"
"sleep 1""echo
function_graph> d/tracing/current_tracer"
"sleep 1""echo
smp_call_function_single> d/tracing/set_ftrace_filter"It will limit trace output to only for the filtered function (
smp_call_function_single).Quoted 18 lines
Quoted 7 lines
adb shell "sleep 1"
"echo 1 > d/tracing/tracing_on"
adb shell "sleep 1"Right after
function_graphis enabled, too many logs are traced upon IRQ transaction which many times eventually causes stall.That would do it!
Hmmm...
Steven, would it be helpful if RCU were to inform tracing (say) halfway through the RCU CPU stall interval, allowing the tracer to do something like
cond_resched_rcu_qs()? I can imagine all sorts of reasons why this wouldn't work, for example, if all the tracing was with irqs disabled or some such, but figured I should ask.Quoted 6 lines
Quoted 1 line
Does Guillermo's approach work for others?
Limited output with a couple of filtered function will definitely not cause RCU schedule stall. But the question is whether we should expect a full function graph trace working on every platform or not (specially the one which generates high interrupts)?
It might well be that the user must disable RCU CPU stall warnings via the
rcu_cpu_stall_suppresssysfs entry (or increase their timeout via thrcu_cpu_stall_timeoutsysfs entry) before doing something that greatly increases overhead. Like enabling large quantities of tracing. ;-)It -might- be possible to do this automatically, but reliable automation would require that tracing understand how often each function was called, which sounds to me to be a bit of a stretch.
Quoted 8 lines
Quoted 1 line
Thoughts?
A random thought:
Is it possible to have a mid-timeout happening and store some information like the instruction pointer, so when the timeout happen we can compare if there was some progress, if yes, very likely, system performance collapsed and we are not fast enough.
RCU already does take various actions for an impending stall, so something could be done. But in most slowdowns, the instruction pointer will be changing rapidly, just not as rapidly as it would normally. So exactly how would the forward-progress comparison be carried out?
It would be easy to set up a notifier, so that if any notifier in the chain returned an error, stall warnings would be suppressed. It would be harder to figure out when to re-enable them, though I suppose that they could be suppressed only for the duration of the current grace period or some such.
Quoted 3 lines
But what exactly would you use such a notifier for?
Or am I misunderstanding your suggestion?
Well, may be the instruction pointer thing is not a good idea.
I learnt from this experience, an overloaded kernel with a lot of interrupts can hang the console and issue RCU stall.
However, someone else can face the same situation. Even if he reads the
RCU/stallwarn.txtdocumentation, it will be hard to figure out the issue.A message telling the grace period can't be reached because we are too busy processing interrupts would have helped but I understand it is not easy to implement.
Quoted 7 lines
Perhaps, adding a new bullet in the documentation can help:
"If the interrupt processing time is longer than the interval between each interrupt, the CPU will keep processing the interrupts without allowing the RCU's grace period kthread. This situation can happen if there is a highly rated number of interrupts and the
function_graphtracer is enabled".
How about this?
Any other debugging options that should be called out? I bet that the function_graph tracer isn't the only way to make this happen.
Thanx, Paul
------------------------------------------------------------------------
commit 8b12d9919f59fe6855429e2eacc1c2e03ecdfe96
Author: Paul E. McKenney <[email protected]> Date: Wed Aug 9 10:16:29 2017 -0700
documentation: Long-running irq handlers can stall RCU grace periods
If a periodic interrupt's handler takes longer to execute than the period between successive interrupts, RCU's kthreads and softirq handlers can be prevented from executing, resulting in otherwise inexplicable RCU CPU stall warnings. This commit therefore calls out this possibility in Documentation/RCU/stallwarn.txt.
Documentation/RCU/stallwarn.txt +11 -1
@@ -40,7 +40,9 @@ o Booting Linux using a console connection that is too slow to o Anything that prevents RCU's grace-period kthreads from running. This can result in the "All QSes seen" console-log message. This message will include information on when the kthread last - ran and how often it should be expected to run. + ran and how often it should be expected to run. It can also + result in the "rcu_.*kthread starved for" console-log message, + which will include additional debugging information. o A CPU-bound real-time task in a CONFIG_PREEMPT kernel, which might happen to preempt a low-priority task in the middle of an RCU
@@ -60,6 +62,14 @@ o A CPU-bound real-time task in a CONFIG_PREEMPT_RT kernel that CONFIG_PREEMPT_RCU case, you might see stall-warning messages. +o A periodic interrupt whose handler takes longer than the time + interval between successive pairs of interrupts. This can + prevent RCU's kthreads and softirq handlers from running. + Note that certain high-overhead debugging options, for example + the function_graph tracer, can result in interrupt handler taking + considerably longer than normal, which can in turn result in + RCU CPU stall warnings. + o A hardware or software issue shuts off the scheduler-clock interrupt on a CPU that is not in dyntick-idle mode. This problem really has happened, and seems to be most likely to
Quoted 95 lines · Paul E. McKenney
Quoted 94 lines · Daniel Lezcano
Quoted 75 lines · Paul E. McKenney
Quoted 54 lines · Daniel Lezcano
Quoted 53 lines · Paul E. McKenney
Quoted 36 lines · Pratyush Anand
Quoted 11 lines · Paul E. McKenney
Quoted 10 lines · 김동현
Dear All
As for me, after configuring
function_graphas below, crash disappears. "echo 0 > d/tracing/tracing_on"
"sleep 1""echo
function_graph> d/tracing/current_tracer"
"sleep 1""echo
smp_call_function_single> d/tracing/set_ftrace_filter"It will limit trace output to only for the filtered function (
smp_call_function_single).Quoted 18 lines
Quoted 7 lines
adb shell "sleep 1"
"echo 1 > d/tracing/tracing_on"
adb shell "sleep 1"Right after
function_graphis enabled, too many logs are traced upon IRQ transaction which many times eventually causes stall.That would do it!
Hmmm...
Steven, would it be helpful if RCU were to inform tracing (say) halfway through the RCU CPU stall interval, allowing the tracer to do something like
cond_resched_rcu_qs()? I can imagine all sorts of reasons why this wouldn't work, for example, if all the tracing was with irqs disabled or some such, but figured I should ask.Quoted 6 lines
Quoted 1 line
Does Guillermo's approach work for others?
Limited output with a couple of filtered function will definitely not cause RCU schedule stall. But the question is whether we should expect a full function graph trace working on every platform or not (specially the one which generates high interrupts)?
It might well be that the user must disable RCU CPU stall warnings via the
rcu_cpu_stall_suppresssysfs entry (or increase their timeout via thrcu_cpu_stall_timeoutsysfs entry) before doing something that greatly increases overhead. Like enabling large quantities of tracing. ;-)It -might- be possible to do this automatically, but reliable automation would require that tracing understand how often each function was called, which sounds to me to be a bit of a stretch.
Quoted 8 lines
Quoted 1 line
Thoughts?
A random thought:
Is it possible to have a mid-timeout happening and store some information like the instruction pointer, so when the timeout happen we can compare if there was some progress, if yes, very likely, system performance collapsed and we are not fast enough.
RCU already does take various actions for an impending stall, so something could be done. But in most slowdowns, the instruction pointer will be changing rapidly, just not as rapidly as it would normally. So exactly how would the forward-progress comparison be carried out?
It would be easy to set up a notifier, so that if any notifier in the chain returned an error, stall warnings would be suppressed. It would be harder to figure out when to re-enable them, though I suppose that they could be suppressed only for the duration of the current grace period or some such.
Quoted 3 lines
But what exactly would you use such a notifier for?
Or am I misunderstanding your suggestion?
Well, may be the instruction pointer thing is not a good idea.
I learnt from this experience, an overloaded kernel with a lot of interrupts can hang the console and issue RCU stall.
However, someone else can face the same situation. Even if he reads the
RCU/stallwarn.txtdocumentation, it will be hard to figure out the issue.A message telling the grace period can't be reached because we are too busy processing interrupts would have helped but I understand it is not easy to implement.
Perhaps, adding a new bullet in the documentation can help:
Quoted 7 lines
Quoted 5 lines
"If the interrupt processing time is longer than the interval between each interrupt, the CPU will keep processing the interrupts without allowing the RCU's grace period kthread. This situation can happen if there is a highly rated number of interrupts and the
function_graphtracer is enabled".How about this?
Yes, it is clear. Thanks for reformulating it.
Quoted 2 lines
Any other debugging options that should be called out? I bet that the
function_graphtracer isn't the only way to make this happen.
Nothing coming in mind but may be worth to mention the slowness of the CPU is the aggravating factor. In particular I was able to reproduce the issue by setting to the min CPU frequency. With the ondemand governor, we can have the frequency high (hence enough CPU power) at the moment we set the function_graph because another CPU is loaded (and both CPUs are sharing the same clock line). The system became stuck at the moment the other CPU went idle with the lowest frequency. That introduced randomness in the issue and made hard to figure out why the RCU stall was happening.
-- Daniel
Quoted 48 lines
------------------------------------------------------------------------
commit
8b12d9919f59fe6855429e2eacc1c2e03ecdfe96Author: Paul E. McKenney <[email protected]> Date: Wed Aug 9 10:16:29 2017 -0700documentation: Long-running irq handlers can stall RCU grace periods
If a periodic interrupt's handler takes longer to execute than the period between successive interrupts, RCU's kthreads and softirq handlers can be prevented from executing, resulting in otherwise inexplicable RCU CPU stall warnings. This commit therefore calls out this possibility in
Documentation/RCU/stallwarn.txt.Reported-by Daniel Lezcano <[email protected]>Signed-off-by Paul E. McKenney <[email protected]>Documentation/RCU/stallwarn.txt +11
-1@@ -40,7 +40,9 @@ o Booting Linux using a console connection that is too slow to o Anything that prevents RCU's grace-period kthreads from running. This can result in the "All QSes seen" console-log message. This message will include information on when the kthread last - ran and how often it should be expected to run. + ran and how often it should be expected to run. It can also + result in the "rcu_.*kthread starved for" console-log message, + which will include additional debugging information. o A CPU-bound real-time task in a CONFIG_PREEMPT kernel, which might happen to preempt a low-priority task in the middle of an RCU@@ -60,6 +62,14 @@ o A CPU-bound real-time task in a CONFIG_PREEMPT_RT kernel that CONFIG_PREEMPT_RCU case, you might see stall-warning messages. +o A periodic interrupt whose handler takes longer than the time + interval between successive pairs of interrupts. This can + prevent RCU's kthreads and softirq handlers from running. + Note that certain high-overhead debugging options, for example + the function_graph tracer, can result in interrupt handler taking + considerably longer than normal, which can in turn result in + RCU CPU stall warnings. + o A hardware or software issue shuts off the scheduler-clock interrupt on a CPU that is not in dyntick-idle mode. This problem really has happened, and seems to be most likely to
Signature · 5 lines
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs Follow Linaro: <http://www.facebook.com/pages/Linaro> Facebook | <http://twitter.com/#!/linaroorg> Twitter | <http://www.linaro.org/linaro-blog/> Blog
Quoted 109 lines · Daniel Lezcano
Quoted 95 lines · Paul E. McKenney
Quoted 94 lines · Daniel Lezcano
Quoted 75 lines · Paul E. McKenney
Quoted 54 lines · Daniel Lezcano
Quoted 53 lines · Paul E. McKenney
Quoted 36 lines · Pratyush Anand
Quoted 11 lines · Paul E. McKenney
Quoted 10 lines · 김동현
Dear All
As for me, after configuring
function_graphas below, crash disappears. "echo 0 > d/tracing/tracing_on"
"sleep 1""echo
function_graph> d/tracing/current_tracer"
"sleep 1""echo
smp_call_function_single> d/tracing/set_ftrace_filter"It will limit trace output to only for the filtered function (
smp_call_function_single).Quoted 18 lines
Quoted 7 lines
adb shell "sleep 1"
"echo 1 > d/tracing/tracing_on"
adb shell "sleep 1"Right after
function_graphis enabled, too many logs are traced upon IRQ transaction which many times eventually causes stall.That would do it!
Hmmm...
Steven, would it be helpful if RCU were to inform tracing (say) halfway through the RCU CPU stall interval, allowing the tracer to do something like
cond_resched_rcu_qs()? I can imagine all sorts of reasons why this wouldn't work, for example, if all the tracing was with irqs disabled or some such, but figured I should ask.Quoted 6 lines
Quoted 1 line
Does Guillermo's approach work for others?
Limited output with a couple of filtered function will definitely not cause RCU schedule stall. But the question is whether we should expect a full function graph trace working on every platform or not (specially the one which generates high interrupts)?
It might well be that the user must disable RCU CPU stall warnings via the
rcu_cpu_stall_suppresssysfs entry (or increase their timeout via thrcu_cpu_stall_timeoutsysfs entry) before doing something that greatly increases overhead. Like enabling large quantities of tracing. ;-)It -might- be possible to do this automatically, but reliable automation would require that tracing understand how often each function was called, which sounds to me to be a bit of a stretch.
Quoted 8 lines
Quoted 1 line
Thoughts?
A random thought:
Is it possible to have a mid-timeout happening and store some information like the instruction pointer, so when the timeout happen we can compare if there was some progress, if yes, very likely, system performance collapsed and we are not fast enough.
RCU already does take various actions for an impending stall, so something could be done. But in most slowdowns, the instruction pointer will be changing rapidly, just not as rapidly as it would normally. So exactly how would the forward-progress comparison be carried out?
It would be easy to set up a notifier, so that if any notifier in the chain returned an error, stall warnings would be suppressed. It would be harder to figure out when to re-enable them, though I suppose that they could be suppressed only for the duration of the current grace period or some such.
Quoted 3 lines
But what exactly would you use such a notifier for?
Or am I misunderstanding your suggestion?
Well, may be the instruction pointer thing is not a good idea.
I learnt from this experience, an overloaded kernel with a lot of interrupts can hang the console and issue RCU stall.
However, someone else can face the same situation. Even if he reads the
RCU/stallwarn.txtdocumentation, it will be hard to figure out the issue.A message telling the grace period can't be reached because we are too busy processing interrupts would have helped but I understand it is not easy to implement.
Perhaps, adding a new bullet in the documentation can help:
Quoted 7 lines
Quoted 5 lines
"If the interrupt processing time is longer than the interval between each interrupt, the CPU will keep processing the interrupts without allowing the RCU's grace period kthread. This situation can happen if there is a highly rated number of interrupts and the
function_graphtracer is enabled".How about this?
Yes, it is clear. Thanks for reformulating it.
Quoted 2 lines
Any other debugging options that should be called out? I bet that the
function_graphtracer isn't the only way to make this happen.
Quoted 9 lines
Nothing coming in mind but may be worth to mention the slowness of the CPU is the aggravating factor. In particular I was able to reproduce the issue by setting to the min CPU frequency. With the ondemand governor, we can have the frequency high (hence enough CPU power) at the moment we set the
function_graphbecause another CPU is loaded (and both CPUs are sharing the same clock line). The system became stuck at the moment the other CPU went idle with the lowest frequency. That introduced randomness in the issue and made hard to figure out why the RCU stall was happening.
Adding this, then?
Thanx, Paul
------------------------------------------------------------------------
commit f7d9ce95064f76be583c775fac32076fa59f1617
Author: Paul E. McKenney <[email protected]> Date: Thu Aug 10 14:33:17 2017 -0700
documentation: Slow systems can stall RCU grace periods
If a fast system has a worst-case grace-period duration of (say) ten seconds, then running the same workload on a system ten times as slow will get you an RCU CPU stall warning given default stall-warning timeout settings. This commit therefore adds this possibility to stallwarn.txt.
Documentation/RCU/stallwarn.txt +6 -0
@@ -70,6 +70,12 @@ o A periodic interrupt whose handler takes longer than the time considerably longer than normal, which can in turn result in RCU CPU stall warnings. +o Testing a workload on a fast system, tuning the stall-warning + timeout down to just barely avoid RCU CPU stall warnings, and then + running the same workload with the same stall-warning timeout on a + slow system. Note that thermal throttling and on-demand governors + can cause a single system to be sometimes fast and sometimes slow! + o A hardware or software issue shuts off the scheduler-clock interrupt on a CPU that is not in dyntick-idle mode. This problem really has happened, and seems to be most likely to
Quoted 1 line · Paul E. McKenney
On Thu, Aug 10, 2017 at 11:45:09AM +0200, Daniel Lezcano wrote:
[ ... ]
Quoted 11 lines
Quoted 9 lines
Nothing coming in mind but may be worth to mention the slowness of the CPU is the aggravating factor. In particular I was able to reproduce the issue by setting to the min CPU frequency. With the ondemand governor, we can have the frequency high (hence enough CPU power) at the moment we set the
function_graphbecause another CPU is loaded (and both CPUs are sharing the same clock line). The system became stuck at the moment the other CPU went idle with the lowest frequency. That introduced randomness in the issue and made hard to figure out why the RCU stall was happening.Adding this, then?
Yes, sure.
Thanks Paul.
-- Daniel
Quoted 35 lines
------------------------------------------------------------------------
commit
f7d9ce95064f76be583c775fac32076fa59f1617Author: Paul E. McKenney <[email protected]> Date: Thu Aug 10 14:33:17 2017 -0700documentation: Slow systems can stall RCU grace periods
If a fast system has a worst-case grace-period duration of (say) ten seconds, then running the same workload on a system ten times as slow will get you an RCU CPU stall warning given default stall-warning timeout settings. This commit therefore adds this possibility to stallwarn.txt.
Reported-by Daniel Lezcano <[email protected]>Signed-off-by Paul E. McKenney <[email protected]>Documentation/RCU/stallwarn.txt +6
-0@@ -70,6 +70,12 @@ o A periodic interrupt whose handler takes longer than the time considerably longer than normal, which can in turn result in RCU CPU stall warnings. +o Testing a workload on a fast system, tuning the stall-warning + timeout down to just barely avoid RCU CPU stall warnings, and then + running the same workload with the same stall-warning timeout on a + slow system. Note that thermal throttling and on-demand governors + can cause a single system to be sometimes fast and sometimes slow! + o A hardware or software issue shuts off the scheduler-clock interrupt on a CPU that is not in dyntick-idle mode. This problem really has happened, and seems to be most likely to
Signature · 5 lines
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs Follow Linaro: <http://www.facebook.com/pages/Linaro> Facebook | <http://twitter.com/#!/linaroorg> Twitter | <http://www.linaro.org/linaro-blog/> Blog
Steven Rostedt
reviewer[ I'm back from vacation! ]
Quoted 11 lines · On Wed, 9 Aug 2017 17:51:33 +0200 Daniel Lezcano <[email protected]>…
Well, may be the instruction pointer thing is not a good idea.
I learnt from this experience, an overloaded kernel with a lot of interrupts can hang the console and issue RCU stall.
However, someone else can face the same situation. Even if he reads the
RCU/stallwarn.txtdocumentation, it will be hard to figure out the issue.A message telling the grace period can't be reached because we are too busy processing interrupts would have helped but I understand it is not easy to implement.
What if the stall code triggered an irqwork first? The irqwork would trigger as soon as interrupts were enabled again (or at the next tick, depending on the arch), and then it would know that RCU stalled due to an irq storm if the irqwork is being hit.
-- Steve
Quoted 12 lines
Perhaps, adding a new bullet in the documentation can help:
"If the interrupt processing time is longer than the interval between each interrupt, the CPU will keep processing the interrupts without allowing the RCU's grace period kthread. This situation can happen if there is a highly rated number of interrupts and the
function_graphtracer is enabled".
Hi Steven,
Quoted 2 lines · Steven Rostedt
[ I'm back from vacation! ]
Did you get the tapes? :)
Quoted 11 lines
Quoted 8 lines · On Wed, 9 Aug 2017 17:51:33 +0200 Daniel Lezcano <[email protected]>…
Well, may be the instruction pointer thing is not a good idea.
I learnt from this experience, an overloaded kernel with a lot of interrupts can hang the console and issue RCU stall.
However, someone else can face the same situation. Even if he reads the
RCU/stallwarn.txtdocumentation, it will be hard to figure out the issue.
Quoted 8 lines
Quoted 3 lines
A message telling the grace period can't be reached because we are too busy processing interrupts would have helped but I understand it is not easy to implement.
What if the stall code triggered an irqwork first? The irqwork would trigger as soon as interrupts were enabled again (or at the next tick, depending on the arch), and then it would know that RCU stalled due to an irq storm if the irqwork is being hit.
Is that condition enough to tell the CPU is over utilized by the interrupts handling?
And I'm wondering if it wouldn't make sense to have this detection in the irq code. With or without the RCU stall warning kernel option set, the irq framework will be warning about this situation. If the RCU stall option is set, that will issue a second message. It will be easy to do the connection between the first message and the second one, no ?
Signature · 5 lines
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs Follow Linaro: <http://www.facebook.com/pages/Linaro> Facebook | <http://twitter.com/#!/linaroorg> Twitter | <http://www.linaro.org/linaro-blog/> Blog
Steven Rostedt
reviewerQuoted 8 lines · On Wed, 16 Aug 2017 10:42:15 +0200 Daniel Lezcano <[email protected]>…
Hi Steven,
Quoted 2 lines · Steven Rostedt
[ I'm back from vacation! ]
Did you get the tapes? :)
Yes, but nothing in them would cause the reputation of the POTUS to become any worse than it already is.
Quoted 21 lines
Quoted 12 lines
Quoted 8 lines · On Wed, 9 Aug 2017 17:51:33 +0200 Daniel Lezcano <[email protected]>…
Well, may be the instruction pointer thing is not a good idea.
I learnt from this experience, an overloaded kernel with a lot of interrupts can hang the console and issue RCU stall.
However, someone else can face the same situation. Even if he reads the
RCU/stallwarn.txtdocumentation, it will be hard to figure out the issue.Quoted 8 lines
Quoted 3 lines
A message telling the grace period can't be reached because we are too busy processing interrupts would have helped but I understand it is not easy to implement.
What if the stall code triggered an irqwork first? The irqwork would trigger as soon as interrupts were enabled again (or at the next tick, depending on the arch), and then it would know that RCU stalled due to an irq storm if the irqwork is being hit.
Quoted 8 lines
Is that condition enough to tell the CPU is over utilized by the interrupts handling?
And I'm wondering if it wouldn't make sense to have this detection in the irq code. With or without the RCU stall warning kernel option set, the irq framework will be warning about this situation. If the RCU stall option is set, that will issue a second message. It will be easy to do the connection between the first message and the second one, no ?
The thing is, the RCU code keeps track of the state of progress, I don't believe the interrupt code does. It just worries about handling interrupts. I'm not excited about adding infrastructure to the interrupt code to do accounting of IRQ storms.
On the other hand, the RCU code already does this. If it notices a stall, it can trigger a irq_work and wait a little more. If the irq_work doesn't fire, then it can do the normal RCU stall message. But if the irq_work does fire, and the RCU progress still hasn't moved forward, then it would be able to say this is due to an IRQ storm and produce a better error message.
-- Steve
Quoted 50 lines · Steven Rostedt
Quoted 8 lines · On Wed, 16 Aug 2017 10:42:15 +0200 Daniel Lezcano <[email protected]>…
Hi Steven,
Quoted 2 lines · Steven Rostedt
[ I'm back from vacation! ]
Did you get the tapes? :)
Yes, but nothing in them would cause the reputation of the POTUS to become any worse than it already is.
Quoted 21 lines
Quoted 12 lines
Quoted 8 lines · On Wed, 9 Aug 2017 17:51:33 +0200 Daniel Lezcano <[email protected]>…
Well, may be the instruction pointer thing is not a good idea.
I learnt from this experience, an overloaded kernel with a lot of interrupts can hang the console and issue RCU stall.
However, someone else can face the same situation. Even if he reads the
RCU/stallwarn.txtdocumentation, it will be hard to figure out the issue.Quoted 8 lines
Quoted 3 lines
A message telling the grace period can't be reached because we are too busy processing interrupts would have helped but I understand it is not easy to implement.
What if the stall code triggered an irqwork first? The irqwork would trigger as soon as interrupts were enabled again (or at the next tick, depending on the arch), and then it would know that RCU stalled due to an irq storm if the irqwork is being hit.
Quoted 8 lines
Is that condition enough to tell the CPU is over utilized by the interrupts handling?
And I'm wondering if it wouldn't make sense to have this detection in the irq code. With or without the RCU stall warning kernel option set, the irq framework will be warning about this situation. If the RCU stall option is set, that will issue a second message. It will be easy to do the connection between the first message and the second one, no ?
The thing is, the RCU code keeps track of the state of progress, I don't believe the interrupt code does. It just worries about handling interrupts. I'm not excited about adding infrastructure to the interrupt code to do accounting of IRQ storms.
Quoted 6 lines
On the other hand, the RCU code already does this. If it notices a stall, it can trigger a
irq_workand wait a little more. If theirq_workdoesn't fire, then it can do the normal RCU stall message. But if theirq_workdoes fire, and the RCU progress still hasn't moved forward, then it would be able to say this is due to an IRQ storm and produce a better error message.
Let me see if I understand you... About halfway to the stall limit, RCU triggers an irq_work (on each CPU that has not yet passed through a quiescent state, IPIing them in turn?), and if the irq_work has not completed by the end of the stall limit, RCU adds that to its stall-warning message.
Or am I missing something here?
Thanx, Paul
Steven Rostedt
reviewerQuoted 5 lines · On Wed, 16 Aug 2017 09:32:28 -0700 "Paul E. McKenney"…
Let me see if I understand you... About halfway to the stall limit, RCU triggers an
irq_work(on each CPU that has not yet passed through a quiescent state, IPIing them in turn?), and if theirq_workhas not completed by the end of the stall limit, RCU adds that to its stall-warning message.
Doesn't even have to be half way through. It could be done at the limit, and then wait a little more.
Quoted 2 lines
Or am I missing something here?
No, I think that's what I was suggesting.
-- Steve
Quoted 11 lines · Steven Rostedt
Quoted 5 lines · On Wed, 16 Aug 2017 09:32:28 -0700 "Paul E. McKenney"…
Let me see if I understand you... About halfway to the stall limit, RCU triggers an
irq_work(on each CPU that has not yet passed through a quiescent state, IPIing them in turn?), and if theirq_workhas not completed by the end of the stall limit, RCU adds that to its stall-warning message.Doesn't even have to be half way through. It could be done at the limit, and then wait a little more.
Agreed. However, I take other evasive action halfway through, so there is a nice place to put the code.
Quoted 3 lines
Quoted 1 line
Or am I missing something here?
No, I think that's what I was suggesting.
I don't see a way of canceling an irq_work.
Ah, but there is an irq_work_queue_on() that fails if still pending. I should be able to use that instead of cancel. Plus, -I- don't have to send the IPIs, I can just let irq_work_queue_on() do my dirty work!
This should provide sufficient entertainment for a bit! ;-)
Thanx, Paul
Quoted 20 lines · Paul E. McKenney
Quoted 11 lines · Steven Rostedt
Quoted 5 lines · On Wed, 16 Aug 2017 09:32:28 -0700 "Paul E. McKenney"…
Let me see if I understand you... About halfway to the stall limit, RCU triggers an
irq_work(on each CPU that has not yet passed through a quiescent state, IPIing them in turn?), and if theirq_workhas not completed by the end of the stall limit, RCU adds that to its stall-warning message.Doesn't even have to be half way through. It could be done at the limit, and then wait a little more.
Agreed. However, I take other evasive action halfway through, so there is a nice place to put the code.
Quoted 3 lines
Quoted 1 line
Or am I missing something here?
No, I think that's what I was suggesting.
Quoted 7 lines
I don't see a way of canceling an
irq_work.Ah, but there is an
irq_work_queue_on()that fails if still pending. I should be able to use that instead of cancel. Plus, -I- don't have to send the IPIs, I can just letirq_work_queue_on()do my dirty work!This should provide sufficient entertainment for a bit! ;-)
Finally got to this -- how does the patch below look to you?
It adds a digit to the stall warning saying for how many full grace periods interrupts have been disabled, "." if interrupts were enabled recently, and "!" if RCU was in too much trouble to tell one way or the other. (This last can happen if RCU's grace-period kthread has been prevented from executing for most of the grace period.)
Thanx, Paul
------------------------------------------------------------------------
commit 7b32575c82ed5c891b93a4a5961e608a97cdb192
Author: Paul E. McKenney <[email protected]> Date: Thu Aug 17 17:05:59 2017 -0700
rcu: Make RCU CPU stall warnings check for irq-disabled CPUs
One common question upon seeing an RCU CPU stall warning is "did the stalled CPUs have interrupts disabled?" However, the current stall warnings are silent on this point. This commit therefore uses irq_work to check whether stalled CPUs still respond to IPIs, and flags this state in the RCU CPU stall warning console messages.
tree.c | 103 +++++++++++++++++++++++++++++++++++++++++++++++++++------- tree.h | 5 ++ tree_plugin.h | 5 ++ 3 files changed, 100 insertions(+), 13 deletions(-)
kernel/rcu/tree.c +91 -12
@@ -1197,6 +1197,22 @@ static int rcu_is_cpu_rrupt_from_idle(void) }/* +* We are reporting a quiescent state on behalf of some other CPU, so +* it is our responsibility to check for and handle potential overflow +* of the rcu_node ->gpnum counter with respect to the rcu_data counters. +* After all, the CPU might be in deep idle state, and thus executing no +* code whatsoever. +*/ +static void rcu_gpnum_ovf (struct rcu_node *rnp,struct rcu_data *rdp) +{ +lockdep_assert_held (&rnp->lock); +if (ULONG_CMP_LT (READ_ONCE (rdp->gpnum) +ULONG_MAX /4 , rnp->gpnum)) +WRITE_ONCE (rdp->gpwrap, true); +if (ULONG_CMP_LT (rdp->rcu_iw_gpnum +ULONG_MAX /4 , rnp->gpnum)) + rdp->rcu_iw_gpnum = rnp->gpnum +ULONG_MAX /4 ; +} + +/* * Snapshot the specified CPU's dynticks counter so that we can later * credit them with an implicit quiescent state. Return 1 if this CPU * is in dynticks idle mode, which is an extended quiescent state.
@@ -1206,15 +1222,34 @@ static int dyntick_save_progress_counter(struct rcu_data *rdp) rdp->dynticks_snap =rcu_dynticks_snap (rdp->dynticks);if (rcu_dynticks_in_eqs (rdp->dynticks_snap)) {trace_rcu_fqs (rdp->rsp->name, rdp->gpnum, rdp->cpu,TPS ("dti" )); -if (ULONG_CMP_LT (READ_ONCE (rdp->gpnum) +ULONG_MAX /4 , - rdp->mynode->gpnum)) -WRITE_ONCE (rdp->gpwrap, true); +rcu_gpnum_ovf (rdp->mynode, rdp);return 1 ; }return 0 ; }/* +* Handler for the irq_work request posted when a grace period has +* gone on for too long, but not yet long enough for an RCU CPU +* stall warning. Set state appropriately, but just complain if +* there is unexpected state on entry. +*/ +static void rcu_iw_handler (struct irq_work *iwp) +{ +struct rcu_data *rdp; +struct rcu_node *rnp; + + rdp = container_of(iwp, struct rcu_data, rcu_iw); + rnp = rdp->mynode; +raw_spin_lock_rcu_node (rnp); +if (!WARN_ON_ONCE (!rdp->rcu_iw_pending)) { + rdp->rcu_iw_gpnum = rnp->gpnum; + rdp->rcu_iw_pending = false; + } +raw_spin_unlock_rcu_node (rnp); +} + +/* * Return true if the specified CPU has passed through a quiescent * state by virtue of being in or having passed through an dynticks * idle state since the last call to dyntick_save_progress_counter()
@@ -1225,7 +1260,7 @@ static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)unsigned long jtsq;bool *rnhqp;bool *ruqp; -struct rcu_node *rnp; +struct rcu_node *rnp = rdp->mynode;/* * If the CPU passed through or entered a dynticks idle phase with
@@ -1238,6 +1273,7 @@ static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)if (rcu_dynticks_in_eqs_since (rdp->dynticks, rdp->dynticks_snap)) {trace_rcu_fqs (rdp->rsp->name, rdp->gpnum, rdp->cpu,TPS ("dti" )); rdp->dynticks_fqs++; +rcu_gpnum_ovf (rnp, rdp);return 1 ; }
@@ -1248,12 +1284,12 @@ static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)* might not be the case for nohz_full CPUs looping in the kernel. */ jtsq = jiffies_till_sched_qs; - rnp = rdp->mynode; ruqp =per_cpu_ptr (&rcu_dynticks.rcu_urgent_qs, rdp->cpu);if (time_after (jiffies, rdp->rsp->gp_start + jtsq) &&READ_ONCE (rdp->rcu_qs_ctr_snap) !=per_cpu (rcu_dynticks.rcu_qs_ctr, rdp->cpu) &&READ_ONCE (rdp->gpnum) == rnp->gpnum && !rdp->gpwrap) {trace_rcu_fqs (rdp->rsp->name, rdp->gpnum, rdp->cpu,TPS ("rqc" )); +rcu_gpnum_ovf (rnp, rdp);return 1 ; }else if (time_after (jiffies, rdp->rsp->gp_start + jtsq)) {/* Load rcu_qs_ctr before store to rcu_urgent_qs. */
@@ -1264,6 +1300,7 @@ static int rcu_implicit_dynticks_qs(struct rcu_data *rdp)if (!(rdp->grpmask &rcu_rnp_online_cpus (rnp))) {trace_rcu_fqs (rdp->rsp->name, rdp->gpnum, rdp->cpu,TPS ("ofl" )); rdp->offline_fqs++; +rcu_gpnum_ovf (rnp, rdp);return 1 ; }
@@ -1295,11 +1332,21 @@ static int rcu_implicit_dynticks_qs(struct rcu_data *rdp) }/* -* If more than halfway to RCU CPU stall-warning time, do -* a resched_cpu() to try to loosen things up a bit. +* If more than halfway to RCU CPU stall-warning time, do a +* resched_cpu() to try to loosen things up a bit. Also check to +* see if the CPU is getting hammered with interrupts, but only +* once per grace period, just to keep the IPIs down to a dull roar. */ -if (jiffies - rdp->rsp->gp_start >rcu_jiffies_till_stall_check () /2 ) +if (jiffies - rdp->rsp->gp_start >rcu_jiffies_till_stall_check () /2 ) {resched_cpu (rdp->cpu); +if (!rdp->rcu_iw_pending && rdp->rcu_iw_gpnum != rnp->gpnum && + (rnp->ffmask & rdp->grpmask)) { +init_irq_work (&rdp->rcu_iw, rcu_iw_handler); + rdp->rcu_iw_pending = true; + rdp->rcu_iw_gpnum = rnp->gpnum; +irq_work_queue_on (&rdp->rcu_iw, rdp->cpu); + } + }return 0 ; }
@@ -1488,6 +1535,7 @@ static void print_cpu_stall(struct rcu_state *rsp) {int cpu;unsigned long flags; +struct rcu_data *rdp =this_cpu_ptr (rsp->rda);struct rcu_node *rnp =rcu_get_root (rsp);long totqlen =0 ;
@@ -1503,7 +1551,9 @@ static void print_cpu_stall(struct rcu_state *rsp)*/ pr_err ("INFO: %s self-detected stall on CPU" , rsp->name);print_cpu_stall_info_begin (); +raw_spin_lock_irqsave_rcu_node (rdp->mynode, flags);print_cpu_stall_info (rsp,smp_processor_id ()); +raw_spin_unlock_irqrestore_rcu_node (rdp->mynode, flags);print_cpu_stall_info_end ();for_each_possible_cpu (cpu) totqlen += rcu_segcblist_n_cbs(&per_cpu_ptr(rsp->rda,
@@ -1897,6 +1947,7 @@ static bool __note_gp_changes(struct rcu_state *rsp, struct rcu_node *rnp, rdp->core_needs_qs = need_gp;zero_cpu_stall_ticks (rdp);WRITE_ONCE (rdp->gpwrap, false); +rcu_gpnum_ovf (rnp, rdp); }return ret; }
@@ -3675,6 +3726,8 @@ rcu_init_percpu_data(int cpu, struct rcu_state *rsp) rdp->cpu_no_qs.b.norm = true; rdp->rcu_qs_ctr_snap =per_cpu (rcu_dynticks.rcu_qs_ctr, cpu); rdp->core_needs_qs = false; + rdp->rcu_iw_pending = false; + rdp->rcu_iw_gpnum = rnp->gpnum -1 ;trace_rcu_grace_period (rsp->name, rdp->gpnum,TPS ("cpuonl" ));raw_spin_unlock_irqrestore_rcu_node (rnp, flags); }
@@ -3712,10 +3765,24 @@ static void rcutree_affinity_setting(unsigned int cpu, int outgoing)*/ int rcutree_online_cpu (unsigned int cpu) { -sync_sched_exp_online_cleanup (cpu); -rcutree_affinity_setting (cpu,-1 ); +unsigned long flags; +struct rcu_data *rdp; +struct rcu_node *rnp; +struct rcu_state *rsp; + +for_each_rcu_flavor (rsp) { + rdp =per_cpu_ptr (rsp->rda, cpu); + rnp = rdp->mynode; +raw_spin_lock_irqsave_rcu_node (rnp, flags); + rnp->ffmask |= rdp->grpmask; +raw_spin_unlock_irqrestore_rcu_node (rnp, flags); + }if (IS_ENABLED (CONFIG_TREE_SRCU ))srcu_online_cpu (cpu); +if (rcu_scheduler_active ==RCU_SCHEDULER_INACTIVE ) +return 0 ;/* Too early in boot for scheduler work. */ +sync_sched_exp_online_cleanup (cpu); +rcutree_affinity_setting (cpu,-1 );return 0 ; }
@@ -3725,6 +3792,19 @@ int rcutree_online_cpu(unsigned int cpu)*/ int rcutree_offline_cpu (unsigned int cpu) { +unsigned long flags; +struct rcu_data *rdp; +struct rcu_node *rnp; +struct rcu_state *rsp; + +for_each_rcu_flavor (rsp) { + rdp =per_cpu_ptr (rsp->rda, cpu); + rnp = rdp->mynode; +raw_spin_lock_irqsave_rcu_node (rnp, flags); + rnp->ffmask &= ~rdp->grpmask; +raw_spin_unlock_irqrestore_rcu_node (rnp, flags); + } +rcutree_affinity_setting (cpu, cpu);if (IS_ENABLED (CONFIG_TREE_SRCU ))srcu_offline_cpu (cpu);
@@ -4173,8 +4253,7 @@ void __init rcu_init(void)for_each_online_cpu (cpu) {rcutree_prepare_cpu (cpu);rcu_cpu_starting (cpu); -if (IS_ENABLED (CONFIG_TREE_SRCU )) -srcu_online_cpu (cpu); +rcutree_online_cpu (cpu); } }
kernel/rcu/tree.h +5 -0
@@ -103,6 +103,7 @@ struct rcu_node {/* Online CPUs for next expedited GP. */ /* Any CPU that has ever been online will */ /* have its bit set. */ +unsigned long ffmask;/* Fully functional CPUs. */ unsigned long grpmask;/* Mask to apply to parent qsmask. */ /* Only one bit will be set in this mask. */ int grplo;/* lowest-numbered CPU or group here. */
@@ -285,6 +286,10 @@ struct rcu_data {/* 8) RCU CPU stall data. */ unsigned int softirq_snap;/* Snapshot of softirq activity. */ +/* ->rcu_iw* fields protected by leaf rcu_node ->lock. */ +struct irq_work rcu_iw;/* Check for non-irq activity. */ +bool rcu_iw_pending;/* Is ->rcu_iw pending? */ +unsigned long rcu_iw_gpnum;/* ->gpnum associated with ->rcu_iw. */ int cpu;struct rcu_state *rsp;
kernel/rcu/tree_plugin.h +4 -1
@@ -1671,6 +1671,7 @@ static void print_cpu_stall_info_begin(void)*/ static void print_cpu_stall_info (struct rcu_state *rsp,int cpu) { +unsigned long delta;char fast_no_hz[72 ];struct rcu_data *rdp =per_cpu_ptr (rsp->rda, cpu);struct rcu_dynticks *rdtp = rdp->dynticks;
@@ -1685,11 +1686,13 @@ static void print_cpu_stall_info(struct rcu_state *rsp, int cpu) ticks_value = rsp->gpnum - rdp->gpnum; }print_cpu_stall_fast_no_hz (fast_no_hz, cpu); - pr_err("\t%d-%c%c%c: (%lu %s) idle=%03x/%llx/%d softirq=%u/%u fqs=%ld %s\n", + delta = rdp->mynode->gpnum - rdp->rcu_iw_gpnum; + pr_err("\t%d-%c%c%c%c: (%lu %s) idle=%03x/%llx/%d softirq=%u/%u fqs=%ld %s\n", cpu, "O."[!!cpu_online(cpu)], "o."[!!(rdp->grpmask & rdp->mynode->qsmaskinit)], "N."[!!(rdp->grpmask & rdp->mynode->qsmaskinitnext)], + rdp->rcu_iw_pending ? (int)min(delta, 9UL) + '0' : "!."[!delta], ticks_value, ticks_title, rcu_dynticks_snap(rdtp) & 0xfff, rdtp->dynticks_nesting, rdtp->dynticks_nmi_nesting,
Quoted 27 lines · Daniel Lezcano
Quoted 12 lines · Steven Rostedt
Quoted 9 lines · On Wed, 2 Aug 2017 00:15:44 +0200 Daniel Lezcano <[email protected]>…
Quoted 8 lines · Paul E. McKenney
Quoted 8 lines
Hi Paul,
I have been trying to set the
function_graphtracer for ftrace and each time I get a CPU stall.How to reproduce:
-----------------Quoted 7 lines
Quoted 3 lines
Quoted 3 lines
Quoted 3 lines
echo
function_graph>/sys/kernel/debug/tracing/current_tracerThis error appears with v4.13-rc3 and v4.12-rc6.
Can you bisect this? It may be due to this commit:
0598e4f08("ftrace: Add use ofsynchronize_rcu_tasks()with dynamic trampolines")Hi Steve,
I git bisected but each time the issue occured. I went through the different version down to v4.4 where the board was not fully supported and it ended up to have the same issue.
Quoted 7 lines
Finally, I had the intuition it could be related to the wall time (there is no RTC clock with battery on the board and the wall time is Jan 1st, 1970).
Setting up the with ntpdate solved the problem.
Even if it is rarely the case to have the time not set, is it normal to have a RCU cpu stall ?
If the system is sufficiently confused about the time, you can indeed get RCU CPU stall warnings.
In one memorable case, a pair of CPUs had a multi-minute disagreement as to the current time, which meant that when one of the started an RCU grace period, the other would immediately issue an RCU CPU stall warning.
Thanx, Paul
Quoted 28 lines
Quoted 19 lines
Quoted 19 lines
Quoted 6 lines
Quoted 2 lines
Is it something already reported ?
I have seen this sort of thing, but only when actually dumping the trace out, and I though those got fixed. You are seeing this just accumulating the trace?
No, just by changing the tracer. It is the first operation I do after rebooting and it is reproducible each time. That happens on an ARM64 platform.
Quoted 7 lines
These RCU CPU stall warnings usually occur when something grabs hold of a CPU for too long, as in 21 seconds or so. One way that they can happen is excessive lock contention, another is having the kernel run through too much data at one shot.
Adding Steven Rostedt on CC for his thoughts.
Signature · 5 lines
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs Follow Linaro: <http://www.facebook.com/pages/Linaro> Facebook | <http://twitter.com/#!/linaroorg> Twitter | <http://www.linaro.org/linaro-blog/> Blog
Quoted 18 lines · Paul E. McKenney
Quoted 5 lines
Hi Paul,
I have been trying to set the
function_graphtracer for ftrace and each time I get a CPU stall.Quoted 8 lines
How to reproduce:
-----------------echo
function_graph>/sys/kernel/debug/tracing/current_tracerThis error appears with v4.13-rc3 and v4.12-rc6.
Is it something already reported ?
I have seen this sort of thing, but only when actually dumping the trace out, and I though those got fixed. You are seeing this just accumulating the trace?
Quoted 6 lines
These RCU CPU stall warnings usually occur when something grabs hold of a CPU for too long, as in 21 seconds or so. One way that they can happen is excessive lock contention, another is having the kernel run through too much data at one shot.
Adding Steven Rostedt on CC for his thoughts.
And Peter Zijlstra suggested this in response to a similar splat:
Thanx, Paul
Quoted 82 lines
Quoted 82 lines
[ ... ]
[ 472.934986] INFO: rcu_preempt detected stalls on CPUs/tasks: [ 472.941302] 0-...: (3 GPs behind) idle=d1a/2/0 softirq=2556/2557 fqs=3195 [ 472.948982] (detected by 3, t=6502 jiffies, g=549, c=548, q=10) [ 472.955180] Task dump for CPU 0: [ 472.958513] swapper/0 R running task 0 0 0 0x00000002 [ 550.954786] INFO: rcu_preempt detected stalls on CPUs/tasks: [ 550.960657] 0-...: (3 GPs behind) idle=d1a/2/0 softirq=2556/2557 fqs=12926 [ 550.967897] (detected by 3, t=26007 jiffies, g=549, c=548, q=27) [ 550.974181] Task dump for CPU 0: [ 550.977513] swapper/0 R running task 0 0 0 0x00000002 [ 628.974788] INFO: rcu_preempt detected stalls on CPUs/tasks: [ 628.980659] 0-...: (3 GPs behind) idle=d1a/2/0 softirq=2556/2557 fqs=22666 [ 628.987900] (detected by 3, t=45512 jiffies, g=549, c=548, q=43) [ 628.994184] Task dump for CPU 0: [ 628.997518] swapper/0 R running task 0 0 0 0x00000002 [ 706.994787] INFO: rcu_preempt detected stalls on CPUs/tasks: [ 707.000658] 0-...: (3 GPs behind) idle=d1a/2/0 softirq=2556/2557 fqs=31785 [ 707.007899] (detected by 3, t=65017 jiffies, g=549, c=548, q=74) [ 707.014183] Task dump for CPU 0: [ 707.017516] swapper/0 R running task 0 0 0 0x00000002 [ 707.024902] rcu_preempt kthread starved for 1261 jiffies! g549 c548 f0x0 RCU_GP_WAIT_FQS(3) ->state=0x1 [ 707.034548] rcu_preempt S 0 8 2 0x00000000 [ 726.049583] INFO: task jbd2/mmcblk1p2-:1120 blocked for more than 120 seconds. [ 726.061772] Not tainted 4.13.0-rc3-ktest-linaro-hikey #202 [ 726.074294] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 726.089027] jbd2/mmcblk1p2- D 0 1120 2 0x00000000 [ 726.126737] INFO: task rs:main Q:Reg:2272 blocked for more than 120 seconds. [ 726.140484] Not tainted 4.13.0-rc3-ktest-linaro-hikey #202 [ 726.154515] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 726.170140] rs:main Q:Reg D 0 2272 1 0x00000000 [ 726.205626] [ 726.205626] Showing all locks held in the system: [ 726.272952] [ 726.278194] ============================================= [ 726.278194] [ 785.014779] INFO: rcu_preempt detected stalls on CPUs/tasks: [ 785.020658] 0-...: (3 GPs behind) idle=d1a/2/0 softirq=2556/2557 fqs=41387 [ 785.027905] (detected by 3, t=84524 jiffies, g=549, c=548, q=94) [ 785.034194] Task dump for CPU 0: [ 785.037532] swapper/0 R running task 0 0 0 0x00000002 [ 846.879753] INFO: task jbd2/mmcblk1p2-:1120 blocked for more than 120 seconds. [ 846.892966] Not tainted 4.13.0-rc3-ktest-linaro-hikey #202 [ 846.909597] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 846.926726] jbd2/mmcblk1p2- D 0 1120 2 0x00000000 [ 846.981356] INFO: task rs:main Q:Reg:2272 blocked for more than 120 seconds. [ 847.001593] Not tainted 4.13.0-rc3-ktest-linaro-hikey #202 [ 847.017642] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 847.034135] rs:main Q:Reg D 0 2272 1 0x00000000 [ 847.077209] [ 847.077209] Showing all locks held in the system: [ 847.180736] [ 847.186753] ============================================= [ 847.186753] [ 863.034778] INFO: rcu_preempt detected stalls on CPUs/tasks: [ 863.040660] 0-...: (3 GPs behind) idle=d1a/2/0 softirq=2556/2557 fqs=50881 [ 863.047907] (detected by 4, t=104030 jiffies, g=549, c=548, q=114) [ 863.054373] Task dump for CPU 0: [ 863.057711] swapper/0 R running task 0 0 0 0x00000002 [ 941.054780] INFO: rcu_preempt detected stalls on CPUs/tasks: [ 941.060659] 0-...: (3 GPs behind) idle=d1a/2/0 softirq=2556/2557 fqs=59726 [ 941.067903] (detected by 7, t=123535 jiffies, g=549, c=548, q=303) [ 941.074370] Task dump for CPU 0: [ 941.077708] swapper/0 R running task 0 0 0 0x00000002 [ 967.711682] INFO: task jbd2/mmcblk1p2-:1120 blocked for more than 120 seconds. [ 967.725731] Not tainted 4.13.0-rc3-ktest-linaro-hikey #202 [ 967.743347] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 967.765658] jbd2/mmcblk1p2- D 0 1120 2 0x00000000 [ 967.810595] INFO: task rs:main Q:Reg:2272 blocked for more than 120 seconds. [ 967.825360] Not tainted 4.13.0-rc3-ktest-linaro-hikey #202 [ 967.838595] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message. [ 967.854460] rs:main Q:Reg D 0 2272 1 0x00000000 [ 967.897804] [ 967.897804] Showing all locks held in the system: [ 967.997067] [ 968.002570] =============================================[ ... ]
etc ...