Re: [patch 03/50] sched: Prepare for RT sleeping spin/rwlocks

[Valentin Schneider]

On 15/07/21 11:27, Peter Zijlstra wrote:
> On Thu, Jul 15, 2021 at 12:20:28AM +0100, Valentin Schneider wrote:
>> Hi,
>>
>> On 13/07/21 17:10, Thomas Gleixner wrote:
>> > From: Thomas Gleixner <tglx@xxxxxxxxxxxxx>
>> >
>> > Waiting for spinlocks and rwlocks on non RT enabled kernels is task::state
>> > preserving. Any wakeup which matches the state is valid.
>> >
>> > RT enabled kernels substitutes them with 'sleeping' spinlocks. This creates
>> > an issue vs. task::state.
>> >
>> > In order to block on the lock the task has to overwrite task::state and a
>> > consecutive wakeup issued by the unlocker sets the state back to
>> > TASK_RUNNING. As a consequence the task loses the state which was set
>> > before the lock acquire and also any regular wakeup targeted at the task
>> > while it is blocked on the lock.
>> >
>>
>> I'm not sure I get this for spinlocks - p->__state != TASK_RUNNING means
>> task is stopped (or about to be), IMO that doesn't go with spinning. I was
>> thinking perhaps ptrace could be an issue, but I don't have a clear picture
>> on that either. What am I missing?
>
> spinlocks will become rtmutex. They're going to clobber __state by
> virtue of a nested block.

I wasn't expecting there to be any task taking spinlocks with state !=
TASK_RUNNING, but I just didn't know where to look.

For instance do_wait() sets current to TASK_INTERRUPTIBLE and can then faff
around with some sighand locks before eventually calling into schedule(),
so clearly that one would be affected by the clobbering.

The more you know...