2023-08-15 17:25:40

by Yicong Yang

Subject: [PATCH v2] drivers/perf: hisi: Schedule perf session according to locality

From: Yicong Yang <[email protected]>

The PCIe PMUs locate on different NUMA node but currently we don't
consider it and likely stack all the sessions on the same CPU:

[root@localhost tmp]# cat /sys/devices/hisi_pcie*/cpumask
0
0
0
0
0
0

This can be optimize a bit to use a local CPU for the PMU.

Signed-off-by: Yicong Yang <[email protected]>
Change since v2:
- Make interrupt affinity consistent with CPU in online handler
Link: https://lore.kernel.org/all/[email protected]/

---
drivers/perf/hisilicon/hisi_pcie_pmu.c | 17 +++++++++++++----
1 file changed, 13 insertions(+), 4 deletions(-)

diff --git a/drivers/perf/hisilicon/hisi_pcie_pmu.c b/drivers/perf/hisilicon/hisi_pcie_pmu.c
index e10fc7cb9493..5a00adb2de8c 100644
--- a/drivers/perf/hisilicon/hisi_pcie_pmu.c
+++ b/drivers/perf/hisilicon/hisi_pcie_pmu.c
@@ -665,8 +665,8 @@ static int hisi_pcie_pmu_online_cpu(unsigned int cpu, struct hlist_node *node)
struct hisi_pcie_pmu *pcie_pmu = hlist_entry_safe(node, struct hisi_pcie_pmu, node);

if (pcie_pmu->on_cpu == -1) {
- pcie_pmu->on_cpu = cpu;
- WARN_ON(irq_set_affinity(pcie_pmu->irq, cpumask_of(cpu)));
+ pcie_pmu->on_cpu = cpumask_local_spread(0, dev_to_node(&pcie_pmu->pdev->dev));
+ WARN_ON(irq_set_affinity(pcie_pmu->irq, cpumask_of(pcie_pmu->on_cpu)));
}

return 0;
@@ -676,14 +676,23 @@ static int hisi_pcie_pmu_offline_cpu(unsigned int cpu, struct hlist_node *node)
{
struct hisi_pcie_pmu *pcie_pmu = hlist_entry_safe(node, struct hisi_pcie_pmu, node);
unsigned int target;
+ cpumask_t mask;
+ int numa_node;

/* Nothing to do if this CPU doesn't own the PMU */
if (pcie_pmu->on_cpu != cpu)
return 0;

pcie_pmu->on_cpu = -1;
- /* Choose a new CPU from all online cpus. */
- target = cpumask_any_but(cpu_online_mask, cpu);
+
+ /* Choose a local CPU from all online cpus. */
+ numa_node = dev_to_node(&pcie_pmu->pdev->dev);
+ if (cpumask_and(&mask, cpumask_of_node(numa_node), cpu_online_mask) &&
+ cpumask_andnot(&mask, &mask, cpumask_of(cpu)))
+ target = cpumask_any(&mask);
+ else
+ target = cpumask_any_but(cpu_online_mask, cpu);
+
if (target >= nr_cpu_ids) {
pci_err(pcie_pmu->pdev, "There is no CPU to set\n");
return 0;
--
2.24.0

2023-08-16 10:40:01

by Yicong Yang

Subject: Re: [PATCH v2] drivers/perf: hisi: Schedule perf session according to locality

Hi,

Please ignore this due to the wrong format of the changelog. Has resend. Sorry for the noise.

On 2023/8/15 21:06, Yicong Yang wrote:
> From: Yicong Yang <[email protected]>
>
> The PCIe PMUs locate on different NUMA node but currently we don't
> consider it and likely stack all the sessions on the same CPU:
>
> [root@localhost tmp]# cat /sys/devices/hisi_pcie*/cpumask
> 0
> 0
> 0
> 0
> 0
> 0
>
> This can be optimize a bit to use a local CPU for the PMU.
>
> Signed-off-by: Yicong Yang <[email protected]>
> Change since v2:
> - Make interrupt affinity consistent with CPU in online handler
> Link: https://lore.kernel.org/all/[email protected]/
>
> ---
> drivers/perf/hisilicon/hisi_pcie_pmu.c | 17 +++++++++++++----
> 1 file changed, 13 insertions(+), 4 deletions(-)
>
> diff --git a/drivers/perf/hisilicon/hisi_pcie_pmu.c b/drivers/perf/hisilicon/hisi_pcie_pmu.c
> index e10fc7cb9493..5a00adb2de8c 100644
> --- a/drivers/perf/hisilicon/hisi_pcie_pmu.c
> +++ b/drivers/perf/hisilicon/hisi_pcie_pmu.c
> @@ -665,8 +665,8 @@ static int hisi_pcie_pmu_online_cpu(unsigned int cpu, struct hlist_node *node)
> struct hisi_pcie_pmu *pcie_pmu = hlist_entry_safe(node, struct hisi_pcie_pmu, node);
>
> if (pcie_pmu->on_cpu == -1) {
> - pcie_pmu->on_cpu = cpu;
> - WARN_ON(irq_set_affinity(pcie_pmu->irq, cpumask_of(cpu)));
> + pcie_pmu->on_cpu = cpumask_local_spread(0, dev_to_node(&pcie_pmu->pdev->dev));
> + WARN_ON(irq_set_affinity(pcie_pmu->irq, cpumask_of(pcie_pmu->on_cpu)));
> }
>
> return 0;
> @@ -676,14 +676,23 @@ static int hisi_pcie_pmu_offline_cpu(unsigned int cpu, struct hlist_node *node)
> {
> struct hisi_pcie_pmu *pcie_pmu = hlist_entry_safe(node, struct hisi_pcie_pmu, node);
> unsigned int target;
> + cpumask_t mask;
> + int numa_node;
>
> /* Nothing to do if this CPU doesn't own the PMU */
> if (pcie_pmu->on_cpu != cpu)
> return 0;
>
> pcie_pmu->on_cpu = -1;
> - /* Choose a new CPU from all online cpus. */
> - target = cpumask_any_but(cpu_online_mask, cpu);
> +
> + /* Choose a local CPU from all online cpus. */
> + numa_node = dev_to_node(&pcie_pmu->pdev->dev);
> + if (cpumask_and(&mask, cpumask_of_node(numa_node), cpu_online_mask) &&
> + cpumask_andnot(&mask, &mask, cpumask_of(cpu)))
> + target = cpumask_any(&mask);
> + else
> + target = cpumask_any_but(cpu_online_mask, cpu);
> +
> if (target >= nr_cpu_ids) {
> pci_err(pcie_pmu->pdev, "There is no CPU to set\n");
> return 0;
>