Hi Prasad,
On Tue, May 22, 2018 at 12:40:05PM -0700, Sodagudi Prasad wrote:
> When following test is executed on 4.14.41 stable kernel, observed that one
> of the core is waiting for tasklist_lock for long time with IRQs disabled.
> ./stress-ng-64 --get 8 -t 3h --times --metrics-brief
>
> Every time when device is crashed, I observed that one the task stuck at
> fork system call and waiting for tasklist_lock as writer with irq disabled.
> https://git.kernel.org/pub/scm/linux/kernel/git/stable/linux-stable.git/tree/kernel/fork.c?h=linux-4.14.y#n1843
Please use a newer kernel. We've addressed this in mainline by moving
arm64 over to the qrwlock implementation which (after some other changes)
guarantees forward progress for well-behaved readers and writers.
To give an example from locktorture with 2 writers and 8 readers, after
a few seconds I see:
rwlock:
Writes: Total: 6725 Max/Min: 0/0 Fail: 0
Reads: Total: 5103727 Max/Min: 0/0 Fail: 0
qrwlock:
Writes: Total: 155284 Max/Min: 0/0 Fail: 0
Reads: Total: 767703 Max/Min: 0/0 Fail: 0
so the ratio is closer to ~6:1 rather than ~191:1 for this case. The
total locking throughput has dropped, but that's expected for this type
of lock where maximum throughput would be achieved by favouring readers.
Will
