[v2,0/5] sched, net: NUMA-aware CPU spreading interface

Message

Valentin Schneider Aug. 17, 2022, 5:58 p.m. UTC

Hi folks,

Tariq pointed out in [1] that drivers allocating IRQ vectors would benefit
from having smarter NUMA-awareness (cpumask_local_spread() doesn't quite cut
it).

The proposed interface involved an array of CPUs and a temporary cpumask, and
being my difficult self what I'm proposing here is an interface that doesn't
require any temporary storage other than some stack variables (at the cost of
one wild macro).

Patch 5/5 is just there to showcase how the thing would be used. If this doesn't
get hated on, I'll let Tariq pick this up and push it with his networking driver
changes (with actual changelogs).

[1]: https://lore.kernel.org/all/[e-mail]/

Revisions
=========

v1 -> v2
++++++++

o Split _find_next_bit() @invert into @invert1 and @invert2 (Yury)
o Rebase onto v6.0-rc1

Cheers,
Valentin

Valentin Schneider (5):
  bitops: Introduce find_next_andnot_bit()
  cpumask: Introduce for_each_cpu_andnot()
  sched/topology: Introduce sched_numa_hop_mask()
  sched/topology: Introduce for_each_numa_hop_cpu()
  SHOWCASE: net/mlx5e: Leverage for_each_numa_hop_cpu()

 drivers/net/ethernet/mellanox/mlx5/core/eq.c | 12 ++++-
 include/linux/cpumask.h                      | 38 ++++++++++++++++
 include/linux/find.h                         | 44 ++++++++++++++++---
 include/linux/topology.h                     | 46 ++++++++++++++++++++
 kernel/sched/topology.c                      | 28 ++++++++++++
 lib/find_bit.c                               | 23 +++++-----
 6 files changed, 172 insertions(+), 19 deletions(-)

--
2.31.1

Comments

Jesse Brandeburg Aug. 18, 2022, 4:28 p.m. UTC | #1

On 8/17/2022 10:58 AM, Valentin Schneider wrote:
> Hi folks,
> 
> Tariq pointed out in [1] that drivers allocating IRQ vectors would benefit
> from having smarter NUMA-awareness (cpumask_local_spread() doesn't quite cut
> it).
> 
> The proposed interface involved an array of CPUs and a temporary cpumask, and
> being my difficult self what I'm proposing here is an interface that doesn't
> require any temporary storage other than some stack variables (at the cost of
> one wild macro).
> 
> Patch 5/5 is just there to showcase how the thing would be used. If this doesn't
> get hated on, I'll let Tariq pick this up and push it with his networking driver
> changes (with actual changelogs).

I am interested in this work, but it seems that at least on lore and in 
my inbox, patch 3,4,5 didn't show up.

Valentin Schneider Aug. 18, 2022, 4:43 p.m. UTC | #2

On 18/08/22 09:28, Jesse Brandeburg wrote:
> On 8/17/2022 10:58 AM, Valentin Schneider wrote:
>> Hi folks,
>>
>> Tariq pointed out in [1] that drivers allocating IRQ vectors would benefit
>> from having smarter NUMA-awareness (cpumask_local_spread() doesn't quite cut
>> it).
>>
>> The proposed interface involved an array of CPUs and a temporary cpumask, and
>> being my difficult self what I'm proposing here is an interface that doesn't
>> require any temporary storage other than some stack variables (at the cost of
>> one wild macro).
>>
>> Patch 5/5 is just there to showcase how the thing would be used. If this doesn't
>> get hated on, I'll let Tariq pick this up and push it with his networking driver
>> changes (with actual changelogs).
>
> I am interested in this work, but it seems that at least on lore and in
> my inbox, patch 3,4,5 didn't show up.

I used exactly the same git send-email command for this than for v1 (which
shows up in its entirety on lore), but I can't see these either. I'm going
to assume they got lost and will resend them.

Valentin Schneider Aug. 18, 2022, 4:51 p.m. UTC | #3

On 18/08/22 17:43, Valentin Schneider wrote:
> On 18/08/22 09:28, Jesse Brandeburg wrote:
>> On 8/17/2022 10:58 AM, Valentin Schneider wrote:
>>> Hi folks,
>>>
>>> Tariq pointed out in [1] that drivers allocating IRQ vectors would benefit
>>> from having smarter NUMA-awareness (cpumask_local_spread() doesn't quite cut
>>> it).
>>>
>>> The proposed interface involved an array of CPUs and a temporary cpumask, and
>>> being my difficult self what I'm proposing here is an interface that doesn't
>>> require any temporary storage other than some stack variables (at the cost of
>>> one wild macro).
>>>
>>> Patch 5/5 is just there to showcase how the thing would be used. If this doesn't
>>> get hated on, I'll let Tariq pick this up and push it with his networking driver
>>> changes (with actual changelogs).
>>
>> I am interested in this work, but it seems that at least on lore and in
>> my inbox, patch 3,4,5 didn't show up.
>
> I used exactly the same git send-email command for this than for v1 (which
> shows up in its entirety on lore), but I can't see these either. I'm going
> to assume they got lost and will resend them.

Welp, it's there now, but clearly should've used --no-thread when resending
them :/