lkml.org 
[lkml]   [2018]   [Oct]   [17]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
Patch in this message
/
Date
From
SubjectRe: [RFC PATCH for 4.21 03/16] sched: Implement push_task_to_cpu (v2)
----- On Oct 17, 2018, at 2:51 AM, Srikar Dronamraju srikar@linux.vnet.ibm.com wrote:

> Hi Mathieu,
>
>> +int push_task_to_cpu(struct task_struct *p, unsigned int dest_cpu)
>> +{
>
> In your use case, is the task going to be current?
> If yes, we should simply be using migrate_task_to.
>
>> + struct rq_flags rf;
>> + struct rq *rq;
>> + int ret = 0;
>> +
>> + rq = task_rq_lock(p, &rf);
>> + update_rq_clock(rq);
>> +
>> + if (!cpumask_test_cpu(dest_cpu, &p->cpus_allowed)) {
>> + ret = -EINVAL;
>> + goto out;
>> + }
>
> Ideally we should have checked cpus_allowed/cpu_active_mask before taking
> the lock. This would help reduce the contention on the rqlock when the
> passed parameter is not correct.
>
>> +
>> + if (!cpumask_test_cpu(dest_cpu, cpu_active_mask)) {
>> + ret = -EBUSY;
>> + goto out;
>> + }
>> +
>> + if (task_cpu(p) == dest_cpu)
>> + goto out;
>
> Same as above.
>
>> +
>> + if (task_running(rq, p) || p->state == TASK_WAKING) {
>
> Why are we using migration thread to move a task in TASK_WAKING state?
>
>> + struct migration_arg arg = { p, dest_cpu };
>> + /* Need help from migration thread: drop lock and wait. */
>> + task_rq_unlock(rq, p, &rf);
>> + stop_one_cpu(cpu_of(rq), migration_cpu_stop, &arg);
>> + tlb_migrate_finish(p->mm);
>> + return 0;
>
> Why cant we use migrate_task_to instead?

I could do that be moving migrate_task_to outside of NUMA-specific #ifdef,
but I think we can do much, much simpler than that, see below.

>
>> + } else if (task_on_rq_queued(p)) {
>> diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
>> index 455fa330de04..27ad25780204 100644
>> --- a/kernel/sched/sched.h
>> +++ b/kernel/sched/sched.h
>> @@ -1340,6 +1340,15 @@ static inline void __set_task_cpu(struct task_struct *p,
>> unsigned int cpu)
>> #endif
>> }
>>
>> +#ifdef CONFIG_SMP
>> +int push_task_to_cpu(struct task_struct *p, unsigned int dest_cpu);
>> +#else
>> +static inline int push_task_to_cpu(struct task_struct *p, unsigned int
>> dest_cpu)
>> +{
>> + return 0;
>> +}
>> +#endif
>> +
>
> Your usecase is outside kernel/sched. So I am not sure if this is the right
> place for the declaration.

Actually, now that I think of it, we may not need to migrate the task at all.
Now that cpu_opv implementation takes a temporary vmap() of the user-space pages,
we can touch that virtual address range from interrupt context from another CPU.

So cpu_opv can simply execute the vector of operations in IPI context rather than
do all this silly dance with migration.

Thoughts ?

Thanks,

Mathieu

--
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com

\
 
 \ /
  Last update: 2018-10-17 17:09    [W:0.060 / U:4.648 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site