mbox series

[v3,0/5] cpumask: improve on cpumask_local_spread() locality

Message ID 20221208183101.1162006-1-yury.norov@gmail.com
Headers
Series cpumask: improve on cpumask_local_spread() locality |

Message

Yury Norov Dec. 8, 2022, 6:30 p.m. UTC 
  cpumask_local_spread() currently checks local node for presence of i'th
CPU, and then if it finds nothing makes a flat search among all non-local
CPUs. We can do it better by checking CPUs per NUMA hops.

This series is inspired by Tariq Toukan and Valentin Schneider's
"net/mlx5e: Improve remote NUMA preferences used for the IRQ affinity
hints"

https://patchwork.kernel.org/project/netdevbpf/patch/20220728191203.4055-3-tariqt@nvidia.com/

According to their measurements, for mlx5e:

        Bottleneck in RX side is released, reached linerate (~1.8x speedup).
        ~30% less cpu util on TX.

This patch makes cpumask_local_spread() traversing CPUs based on NUMA
distance, just as well, and I expect comparable improvement for its
users, as in case of mlx5e.

I tested new behavior on my VM with the following NUMA configuration:

root@debian:~# numactl -H
available: 4 nodes (0-3)
node 0 cpus: 0 1 2 3
node 0 size: 3869 MB
node 0 free: 3740 MB
node 1 cpus: 4 5
node 1 size: 1969 MB
node 1 free: 1937 MB
node 2 cpus: 6 7
node 2 size: 1967 MB
node 2 free: 1873 MB
node 3 cpus: 8 9 10 11 12 13 14 15
node 3 size: 7842 MB
node 3 free: 7723 MB
node distances:
node   0   1   2   3
  0:  10  50  30  70
  1:  50  10  70  30
  2:  30  70  10  50
  3:  70  30  50  10

And the cpumask_local_spread() for each node and offset traversing looks
like this:

node 0:   0   1   2   3   6   7   4   5   8   9  10  11  12  13  14  15
node 1:   4   5   8   9  10  11  12  13  14  15   0   1   2   3   6   7
node 2:   6   7   0   1   2   3   8   9  10  11  12  13  14  15   4   5
node 3:   8   9  10  11  12  13  14  15   4   5   6   7   0   1   2   3

v1: https://lore.kernel.org/lkml/20221111040027.621646-5-yury.norov@gmail.com/T/
v2: https://lore.kernel.org/all/20221112190946.728270-3-yury.norov@gmail.com/T/
v3:
 - fix typo in find_nth_and_andnot_bit();
 - add 5th patch that simplifies cpumask_local_spread();
 - address various coding style nits.

Yury Norov (5):
  lib/find: introduce find_nth_and_andnot_bit
  cpumask: introduce cpumask_nth_and_andnot
  sched: add sched_numa_find_nth_cpu()
  cpumask: improve on cpumask_local_spread() locality
  lib/cpumask: reorganize cpumask_local_spread() logic

 include/linux/cpumask.h  | 20 ++++++++++++++
 include/linux/find.h     | 33 +++++++++++++++++++++++
 include/linux/topology.h |  8 ++++++
 kernel/sched/topology.c  | 57 ++++++++++++++++++++++++++++++++++++++++
 lib/cpumask.c            | 26 +++++-------------
 lib/find_bit.c           |  9 +++++++
 6 files changed, 134 insertions(+), 19 deletions(-)

Comments

Jacob Keller Dec. 8, 2022, 6:45 p.m. UTC | #1 
On 12/8/2022 10:30 AM, Yury Norov wrote:
> cpumask_local_spread() currently checks local node for presence of i'th
> CPU, and then if it finds nothing makes a flat search among all non-local
> CPUs. We can do it better by checking CPUs per NUMA hops.
> 
> This series is inspired by Tariq Toukan and Valentin Schneider's
> "net/mlx5e: Improve remote NUMA preferences used for the IRQ affinity
> hints"
> 
> https://patchwork.kernel.org/project/netdevbpf/patch/20220728191203.4055-3-tariqt@nvidia.com/
> 
> According to their measurements, for mlx5e:
> 
>          Bottleneck in RX side is released, reached linerate (~1.8x speedup).
>          ~30% less cpu util on TX.
> 
> This patch makes cpumask_local_spread() traversing CPUs based on NUMA
> distance, just as well, and I expect comparable improvement for its
> users, as in case of mlx5e.
> 
> I tested new behavior on my VM with the following NUMA configuration:
> 
> root@debian:~# numactl -H
> available: 4 nodes (0-3)
> node 0 cpus: 0 1 2 3
> node 0 size: 3869 MB
> node 0 free: 3740 MB
> node 1 cpus: 4 5
> node 1 size: 1969 MB
> node 1 free: 1937 MB
> node 2 cpus: 6 7
> node 2 size: 1967 MB
> node 2 free: 1873 MB
> node 3 cpus: 8 9 10 11 12 13 14 15
> node 3 size: 7842 MB
> node 3 free: 7723 MB
> node distances:
> node   0   1   2   3
>    0:  10  50  30  70
>    1:  50  10  70  30
>    2:  30  70  10  50
>    3:  70  30  50  10
> 
> And the cpumask_local_spread() for each node and offset traversing looks
> like this:
> 
> node 0:   0   1   2   3   6   7   4   5   8   9  10  11  12  13  14  15
> node 1:   4   5   8   9  10  11  12  13  14  15   0   1   2   3   6   7
> node 2:   6   7   0   1   2   3   8   9  10  11  12  13  14  15   4   5
> node 3:   8   9  10  11  12  13  14  15   4   5   6   7   0   1   2   3
> 
> v1: https://lore.kernel.org/lkml/20221111040027.621646-5-yury.norov@gmail.com/T/
> v2: https://lore.kernel.org/all/20221112190946.728270-3-yury.norov@gmail.com/T/
> v3:
>   - fix typo in find_nth_and_andnot_bit();
>   - add 5th patch that simplifies cpumask_local_spread();
>   - address various coding style nits.
> 

The whole series look reasonable to me!

Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Tariq Toukan Dec. 8, 2022, 8:22 p.m. UTC | #2 
On 12/8/2022 8:30 PM, Yury Norov wrote:
> cpumask_local_spread() currently checks local node for presence of i'th
> CPU, and then if it finds nothing makes a flat search among all non-local
> CPUs. We can do it better by checking CPUs per NUMA hops.
> 
> This series is inspired by Tariq Toukan and Valentin Schneider's
> "net/mlx5e: Improve remote NUMA preferences used for the IRQ affinity
> hints"
> 
> https://patchwork.kernel.org/project/netdevbpf/patch/20220728191203.4055-3-tariqt@nvidia.com/
> 
> According to their measurements, for mlx5e:
> 
>          Bottleneck in RX side is released, reached linerate (~1.8x speedup).
>          ~30% less cpu util on TX.
> 
> This patch makes cpumask_local_spread() traversing CPUs based on NUMA
> distance, just as well, and I expect comparable improvement for its
> users, as in case of mlx5e.
> 
> I tested new behavior on my VM with the following NUMA configuration:
> 
> root@debian:~# numactl -H
> available: 4 nodes (0-3)
> node 0 cpus: 0 1 2 3
> node 0 size: 3869 MB
> node 0 free: 3740 MB
> node 1 cpus: 4 5
> node 1 size: 1969 MB
> node 1 free: 1937 MB
> node 2 cpus: 6 7
> node 2 size: 1967 MB
> node 2 free: 1873 MB
> node 3 cpus: 8 9 10 11 12 13 14 15
> node 3 size: 7842 MB
> node 3 free: 7723 MB
> node distances:
> node   0   1   2   3
>    0:  10  50  30  70
>    1:  50  10  70  30
>    2:  30  70  10  50
>    3:  70  30  50  10
> 
> And the cpumask_local_spread() for each node and offset traversing looks
> like this:
> 
> node 0:   0   1   2   3   6   7   4   5   8   9  10  11  12  13  14  15
> node 1:   4   5   8   9  10  11  12  13  14  15   0   1   2   3   6   7
> node 2:   6   7   0   1   2   3   8   9  10  11  12  13  14  15   4   5
> node 3:   8   9  10  11  12  13  14  15   4   5   6   7   0   1   2   3
> 
> v1: https://lore.kernel.org/lkml/20221111040027.621646-5-yury.norov@gmail.com/T/
> v2: https://lore.kernel.org/all/20221112190946.728270-3-yury.norov@gmail.com/T/
> v3:
>   - fix typo in find_nth_and_andnot_bit();
>   - add 5th patch that simplifies cpumask_local_spread();
>   - address various coding style nits.
> 
> Yury Norov (5):
>    lib/find: introduce find_nth_and_andnot_bit
>    cpumask: introduce cpumask_nth_and_andnot
>    sched: add sched_numa_find_nth_cpu()
>    cpumask: improve on cpumask_local_spread() locality
>    lib/cpumask: reorganize cpumask_local_spread() logic
> 
>   include/linux/cpumask.h  | 20 ++++++++++++++
>   include/linux/find.h     | 33 +++++++++++++++++++++++
>   include/linux/topology.h |  8 ++++++
>   kernel/sched/topology.c  | 57 ++++++++++++++++++++++++++++++++++++++++
>   lib/cpumask.c            | 26 +++++-------------
>   lib/find_bit.c           |  9 +++++++
>   6 files changed, 134 insertions(+), 19 deletions(-)
> 

Acked-by: Tariq Toukan <tariqt@nvidia.com>
Yury Norov Dec. 8, 2022, 8:51 p.m. UTC | #3 
On Thu, Dec 08, 2022 at 10:22:22PM +0200, Tariq Toukan wrote:
> 
> 
> On 12/8/2022 8:30 PM, Yury Norov wrote:
> > cpumask_local_spread() currently checks local node for presence of i'th
> > CPU, and then if it finds nothing makes a flat search among all non-local
> > CPUs. We can do it better by checking CPUs per NUMA hops.
> > 
> > This series is inspired by Tariq Toukan and Valentin Schneider's
> > "net/mlx5e: Improve remote NUMA preferences used for the IRQ affinity
> > hints"
> > 
> > https://patchwork.kernel.org/project/netdevbpf/patch/20220728191203.4055-3-tariqt@nvidia.com/
> > 
> > According to their measurements, for mlx5e:
> > 
> >          Bottleneck in RX side is released, reached linerate (~1.8x speedup).
> >          ~30% less cpu util on TX.
> > 
> > This patch makes cpumask_local_spread() traversing CPUs based on NUMA
> > distance, just as well, and I expect comparable improvement for its
> > users, as in case of mlx5e.
> > 
> > I tested new behavior on my VM with the following NUMA configuration:
> > 
> > root@debian:~# numactl -H
> > available: 4 nodes (0-3)
> > node 0 cpus: 0 1 2 3
> > node 0 size: 3869 MB
> > node 0 free: 3740 MB
> > node 1 cpus: 4 5
> > node 1 size: 1969 MB
> > node 1 free: 1937 MB
> > node 2 cpus: 6 7
> > node 2 size: 1967 MB
> > node 2 free: 1873 MB
> > node 3 cpus: 8 9 10 11 12 13 14 15
> > node 3 size: 7842 MB
> > node 3 free: 7723 MB
> > node distances:
> > node   0   1   2   3
> >    0:  10  50  30  70
> >    1:  50  10  70  30
> >    2:  30  70  10  50
> >    3:  70  30  50  10
> > 
> > And the cpumask_local_spread() for each node and offset traversing looks
> > like this:
> > 
> > node 0:   0   1   2   3   6   7   4   5   8   9  10  11  12  13  14  15
> > node 1:   4   5   8   9  10  11  12  13  14  15   0   1   2   3   6   7
> > node 2:   6   7   0   1   2   3   8   9  10  11  12  13  14  15   4   5
> > node 3:   8   9  10  11  12  13  14  15   4   5   6   7   0   1   2   3
> > 
> > v1: https://lore.kernel.org/lkml/20221111040027.621646-5-yury.norov@gmail.com/T/
> > v2: https://lore.kernel.org/all/20221112190946.728270-3-yury.norov@gmail.com/T/
> > v3:
> >   - fix typo in find_nth_and_andnot_bit();
> >   - add 5th patch that simplifies cpumask_local_spread();
> >   - address various coding style nits.
> > 
> > Yury Norov (5):
> >    lib/find: introduce find_nth_and_andnot_bit
> >    cpumask: introduce cpumask_nth_and_andnot
> >    sched: add sched_numa_find_nth_cpu()
> >    cpumask: improve on cpumask_local_spread() locality
> >    lib/cpumask: reorganize cpumask_local_spread() logic
> > 
> >   include/linux/cpumask.h  | 20 ++++++++++++++
> >   include/linux/find.h     | 33 +++++++++++++++++++++++
> >   include/linux/topology.h |  8 ++++++
> >   kernel/sched/topology.c  | 57 ++++++++++++++++++++++++++++++++++++++++
> >   lib/cpumask.c            | 26 +++++-------------
> >   lib/find_bit.c           |  9 +++++++
> >   6 files changed, 134 insertions(+), 19 deletions(-)
> > 
> 
> Acked-by: Tariq Toukan <tariqt@nvidia.com>

Thanks Tariq, Jacob and Peter for review. I'll add the series in
bitmap-for-next for testing. Still, I think that sched/numa branches
would be more suitable.

Thanks,
Yury