RE: [PATCH v1 1/2] power: fix power library with --lcores

["Tummala, Sivaprasad" <Sivaprasad.Tummala@amd.com>]

[AMD Official Use Only - AMD Internal Distribution Only]

Hi Lihuisong,

> -----Original Message-----
> From: lihuisong (C) <lihuisong@huawei.com>
> Sent: Thursday, October 17, 2024 5:35 PM
> To: Tummala, Sivaprasad <Sivaprasad.Tummala@amd.com>;
> david.hunt@intel.com; anatoly.burakov@intel.com; jerinj@marvell.com;
> radu.nicolau@intel.com; gakhil@marvell.com; cristian.dumitrescu@intel.com; Yigit,
> Ferruh <Ferruh.Yigit@amd.com>; konstantin.ananyev@huawei.com
> Cc: dev@dpdk.org
> Subject: Re: [PATCH v1 1/2] power: fix power library with --lcores
>
> Caution: This message originated from an External Source. Use proper caution
> when opening attachments, clicking links, or responding.
>
>
> Good job.
> With belows to change,
> Acked-by: Huisong Li <lihuisong@huawei.com>
>
>
> 在 2024/10/17 19:02, Sivaprasad Tummala 写道:
> > This commit fixes an issue in the power library related to using
> > lcores mapped to different physical cores (--lcores option in EAL).
> >
> > Previously, the power library incorrectly accessed CPU sysfs
> > attributes for power management, treating lcore IDs as CPU IDs.
> > e.g. with --lcores '1@128', lcore_id '1' was interpreted as CPU_id
> > instead of '128'.
> >
> > This patch corrects the cpu_id based on lcore and CPU mappings. It
> > also constraints power management support for lcores mapped to
> > multiple physical cores/threads.
> >
> > When multiple lcores are mapped to the same physical core, invoking
> > frequency scaling APIs on any lcore will apply the changes
> > effectively.
> >
> > Signed-off-by: Sivaprasad Tummala <sivaprasad.tummala@amd.com>
> > ---
> >   lib/power/power_acpi_cpufreq.c       |  7 ++++++-
> >   lib/power/power_amd_pstate_cpufreq.c |  7 ++++++-
> >   lib/power/power_common.c             | 23 +++++++++++++++++++++++
> >   lib/power/power_common.h             |  2 +-
> >   lib/power/power_cppc_cpufreq.c       |  7 ++++++-
> >   lib/power/power_pstate_cpufreq.c     |  7 ++++++-
> >   6 files changed, 48 insertions(+), 5 deletions(-)
> >
> > diff --git a/lib/power/power_acpi_cpufreq.c
> > b/lib/power/power_acpi_cpufreq.c index 81996e1c13..259bc8a263 100644
> > --- a/lib/power/power_acpi_cpufreq.c
> > +++ b/lib/power/power_acpi_cpufreq.c
> > @@ -258,7 +258,12 @@ power_acpi_cpufreq_init(unsigned int lcore_id)
> >               return -1;
> >       }
> >
> > -     pi->lcore_id = lcore_id;
> > +     if (check_lcore_and_set_cpu(lcore_id, &pi->lcore_id) < 0) {
> > +             POWER_LOG(ERR,
> > +                     "Cannot get cpu id mapped for lcore %u", lcore_id);
> > +             return -1;
> > +     }
> > +
> >       /* Check and set the governor */
> >       if (power_set_governor_userspace(pi) < 0) {
> >               POWER_LOG(ERR, "Cannot set governor of lcore %u to "
> > diff --git a/lib/power/power_amd_pstate_cpufreq.c
> > b/lib/power/power_amd_pstate_cpufreq.c
> > index 090a0d96cb..9bfb271bb7 100644
> > --- a/lib/power/power_amd_pstate_cpufreq.c
> > +++ b/lib/power/power_amd_pstate_cpufreq.c
> > @@ -376,7 +376,12 @@ power_amd_pstate_cpufreq_init(unsigned int lcore_id)
> >               return -1;
> >       }
> >
> > -     pi->lcore_id = lcore_id;
> > +     if (check_lcore_and_set_cpu(lcore_id, &pi->lcore_id) < 0) {
> > +             POWER_LOG(ERR,
> > +                     "Cannot get cpu id mapped for lcore %u", lcore_id);
> > +             return -1;
> > +     }
> > +
> >       /* Check and set the governor */
> >       if (power_set_governor_userspace(pi) < 0) {
> >               POWER_LOG(ERR, "Cannot set governor of lcore %u to "
> > diff --git a/lib/power/power_common.c b/lib/power/power_common.c index
> > 590986d5ef..a8d5cd1c50 100644
> > --- a/lib/power/power_common.c
> > +++ b/lib/power/power_common.c
> > @@ -9,6 +9,7 @@
> >
> >   #include <rte_log.h>
> >   #include <rte_string_fns.h>
> > +#include <rte_lcore.h>
> >
> >   #include "power_common.h"
> >
> > @@ -204,3 +205,25 @@ power_set_governor(unsigned int lcore_id, const
> > char *new_governor,
> >
> >       return ret;
> >   }
> > +
> > +int check_lcore_and_set_cpu(uint32_t lcore_id, uint32_t *cpu_id)
> How about the following function name? It is probably more common for other power
> library like PM QoS which is upstreaming.
> -->
> int power_get_lcore_mapped_cpu_id(lcore_id)
> negative on failure
ACK!

>  >=0 on success
> > +{
> > +     rte_cpuset_t lcore_cpus;
> > +     uint32_t cpu;
> > +
> > +     lcore_cpus = rte_lcore_cpuset(lcore_id);
> > +     if (CPU_COUNT(&lcore_cpus) != 1) {
> > +             POWER_LOG(ERR,
> > +                     "Power library does not support lcore %u mapping to %u cpus",
> > +                     lcore_id, CPU_COUNT(&lcore_cpus));
> > +             return -1;
> > +     }
> > +
> > +     for (cpu = 0; cpu < CPU_SETSIZE; cpu++) {
> > +             if (CPU_ISSET(cpu, &lcore_cpus))
> > +                     break;
> > +     }
> > +     *cpu_id = cpu;
> > +
> > +     return 0;
> > +}
> <...>