linux 2.6.0-test8 reaim tests fail to exit
I've observed in the last two days, linux 2.6.0-test8 and I think linux 2.6.0-test8-mm1. The reaim test workload fails to exit. All of the reaim tasks are blocked in sys_wait4(). But non of them seem to have any obvious child processes.
There are also lots of sync() processes. Many of those seem to be blocked somewhere scheduling IO. These kernels were all with the as-isoched IO scheduler. I may retry with deadline scheduler, just to rule out IO scheduler. Is there any link between the sync() processes and the reaim waiting for children?
Could there be a problem with IO not completing for the sync() tasks that causes the reaim tasks to not complete?
Attached is a console output from a system hung in this state. Included (towards the bottom) is a sysrq t output.
I'm hoping to investigate this more closely over the week end.
Re: [BUG] linux 2.6.0-test8 reaim tests (attachment URL)
My attachment was too big to send directly to lkml. You can find it at:
http://developer.osdl.org/dmo/REAIM/typescript_stp8-002
Quoted 19 lines · Dave Olien
I've observed in the last two days, linux 2.6.0-test8 and I think linux 2.6.0-test8-mm1. The reaim test workload fails to exit. All of the reaim tasks are blocked in
sys_wait4(). But non of them seem to have any obvious child processes.There are also lots of
sync()processes. Many of those seem to be blocked somewhere scheduling IO. These kernels were all with the as-isoched IO scheduler. I may retry with deadline scheduler, just to rule out IO scheduler. Is there any link between thesync()processes and the reaim waiting for children?Could there be a problem with IO not completing for the
sync()tasks that causes the reaim tasks to not complete?Attached is a console output from a system hung in this state. Included (towards the bottom) is a sysrq t output.
I'm hoping to investigate this more closely over the week end.
Quoted 11 lines · Dave Olien
I've observed in the last two days, linux 2.6.0-test8 and I think linux 2.6.0-test8-mm1. The reaim test workload fails to exit. All of the reaim tasks are blocked in
sys_wait4(). But non of them seem to have any obvious child processes.There are also lots of
sync()processes. Many of those seem to be blocked somewhere scheduling IO. These kernels were all with the as-isoched IO scheduler. I may retry with deadline scheduler, just to rule out IO scheduler. Is there any link between thesync()processes and the reaim waiting for children?
Quoted 8 lines
Could there be a problem with IO not completing for the
sync()tasks that causes the reaim tasks to not complete?Attached is a console output from a system hung in this state. Included (towards the bottom) is a sysrq t output.
I'm hoping to investigate this more closely over the week end.
Looks like IO scheduler. Testing deadline would be good. Thanks.