lkml.org 
[lkml]   [2019]   [Sep]   [6]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH 4/4] genirq: use irq's affinity for threaded irq with IRQF_RESCUE_THREAD
From
Date
On 27/08/2019 09:53, Ming Lei wrote:
> In case of IRQF_RESCUE_THREAD, the threaded handler is only used to
> handle interrupt when IRQ flood comes, use irq's affinity for this thread
> so that scheduler may select other not too busy CPUs for handling the
> interrupt.
>
> Cc: Long Li <longli@microsoft.com>
> Cc: Ingo Molnar <mingo@redhat.com>,
> Cc: Peter Zijlstra <peterz@infradead.org>
> Cc: Keith Busch <keith.busch@intel.com>
> Cc: Jens Axboe <axboe@fb.com>
> Cc: Christoph Hellwig <hch@lst.de>
> Cc: Sagi Grimberg <sagi@grimberg.me>
> Cc: John Garry <john.garry@huawei.com>
> Cc: Thomas Gleixner <tglx@linutronix.de>
> Cc: Hannes Reinecke <hare@suse.com>
> Cc: linux-nvme@lists.infradead.org
> Cc: linux-scsi@vger.kernel.org
> Signed-off-by: Ming Lei <ming.lei@redhat.com>


> ---
> kernel/irq/manage.c | 13 ++++++++++++-
> 1 file changed, 12 insertions(+), 1 deletion(-)
>
> diff --git a/kernel/irq/manage.c b/kernel/irq/manage.c
> index 1566abbf50e8..03bc041348b7 100644
> --- a/kernel/irq/manage.c
> +++ b/kernel/irq/manage.c
> @@ -968,7 +968,18 @@ irq_thread_check_affinity(struct irq_desc *desc, struct irqaction *action)
> if (cpumask_available(desc->irq_common_data.affinity)) {
> const struct cpumask *m;
>
> - m = irq_data_get_effective_affinity_mask(&desc->irq_data);
> + /*
> + * Managed IRQ's affinity is setup gracefull on MUNA locality,

gracefully

> + * also if IRQF_RESCUE_THREAD is set, interrupt flood has been
> + * triggered, so ask scheduler to run the thread on CPUs
> + * specified by this interrupt's affinity.
> + */

Hi Ming,

> + if ((action->flags & IRQF_RESCUE_THREAD) &&
> + irqd_affinity_is_managed(&desc->irq_data))

This doesn't look to solve the other issue I reported - that being that
we handle the interrupt in a threaded handler natively, and the hard
irq+threaded handler fully occupies the cpu, limiting throughput.

So can we expand the scope to cover that scenario also? I don't think
that it’s right to solve that separately. So if we're continuing this
approach, can we add separate judgment for spreading the cpumask for the
threaded part?

Thanks,
John

> + m = desc->irq_common_data.affinity;
> + else
> + m = irq_data_get_effective_affinity_mask(
> + &desc->irq_data);
> cpumask_copy(mask, m);
> } else {
> valid = false;
>


\
 
 \ /
  Last update: 2019-09-06 10:51    [W:0.181 / U:46.176 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site