Re: [BUG] [ tip/sched/core ] System unresponsive after booting

[Peter Zijlstra]

On Thu, Jan 16, 2014 at 02:48:51PM +0100, Daniel Lezcano wrote:
> 3570  sched_getparam(3570, { 0 })       = 0
> 3570  sched_getscheduler(3570)          = 0 (SCHED_OTHER)
> 3570  sched_get_priority_min(SCHED_OTHER) = 0
> 3570  sched_get_priority_max(SCHED_OTHER) = 0
> 3571  sched_get_priority_min(SCHED_OTHER) = 0
> 3571  sched_get_priority_max(SCHED_OTHER) = 0
> 3571  sched_get_priority_min(SCHED_OTHER) = 0
> 3571  sched_get_priority_max(SCHED_OTHER) = 0
> 3571  sched_setscheduler(3572, SCHED_OTHER, { 0 } <unfinished ...>
> 3571  <... sched_setscheduler resumed> ) = 0
> 3571  sched_get_priority_min(SCHED_OTHER <unfinished ...>
> 3571  <... sched_get_priority_min resumed> ) = 0
> 3571  sched_get_priority_max(SCHED_OTHER <unfinished ...>
> 3571  <... sched_get_priority_max resumed> ) = 0
> 3571  sched_setscheduler(3573, SCHED_OTHER, { 0 } <unfinished ...>
> 3571  <... sched_setscheduler resumed> ) = -1 EPERM (Operation not
> permitted)
> 3571  sched_get_priority_min(SCHED_OTHER <unfinished ...>
> 3571  <... sched_get_priority_min resumed> ) = 0
> 3571  sched_get_priority_max(SCHED_OTHER <unfinished ...>
> 3571  <... sched_get_priority_max resumed> ) = 0
> 3571  sched_setscheduler(3574, SCHED_OTHER, { 0 } <unfinished ...>
> 3571  <... sched_setscheduler resumed> ) = -1 EPERM (Operation not
> permitted)
> 
> The same strace but on a kernel which does not hang. The calls to
> sched_setscheduler do not fail.
> 
> 3292  sched_getparam(3292, { 0 })       = 0
> 3292  sched_getscheduler(3292)          = 0 (SCHED_OTHER)
> 3292  sched_get_priority_min(SCHED_OTHER) = 0
> 3292  sched_get_priority_max(SCHED_OTHER) = 0
> 3293  sched_get_priority_min(SCHED_OTHER) = 0
> 3293  sched_get_priority_max(SCHED_OTHER) = 0
> 3293  sched_get_priority_min(SCHED_OTHER) = 0
> 3293  sched_get_priority_max(SCHED_OTHER) = 0
> 3293  sched_setscheduler(3294, SCHED_OTHER, { 0 } <unfinished ...>
> 3293  <... sched_setscheduler resumed> ) = 0
> 3293  sched_get_priority_min(SCHED_OTHER <unfinished ...>
> 3293  <... sched_get_priority_min resumed> ) = 0
> 3293  sched_get_priority_max(SCHED_OTHER <unfinished ...>
> 3293  <... sched_get_priority_max resumed> ) = 0
> 3293  sched_setscheduler(3295, SCHED_OTHER, { 0 } <unfinished ...>
> 3293  <... sched_setscheduler resumed> ) = 0
> 3293  sched_get_priority_min(SCHED_OTHER <unfinished ...>
> 3293  <... sched_get_priority_min resumed> ) = 0
> 3293  sched_get_priority_max(SCHED_OTHER <unfinished ...>
> 3293  <... sched_get_priority_max resumed> ) = 0
> 3293  sched_setscheduler(3296, SCHED_OTHER, { 0 } <unfinished ...>
> 3293  <... sched_setscheduler resumed> ) = 0
> 
> The EPERM error comes from kernel/sched/core.c:3303
> 
> ...
> 		if (fair_policy(policy)) {
> 			if (!can_nice(p, attr->sched_nice))
> 				return -EPERM;
> 		}
> ...
> 
> 
> But I don't know why this is leading to block a process or making rsyslogd
> being not woken up by a packet coming in the af_unix socket.

Could you test with a fresh tip/master, Ingo just pushed out a stack of
fixes, in particularly:

e3de300d1212b ("sched: Preserve the nice level over sched_setscheduler() and sched_setparam() calls")
39fd8fd22b322 ("sched: Fix up scheduler syscall LTP fails")

Could have affected things.

Meanwhile I'll try and better read what the above says.

Thanks!
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/