Hi Alex, Valentin,
On Sun, Feb 11, 2024 at 12:37 AM <al...@kernel.org> wrote:
>
> From: Alex Shi <al...@kernel.org>
>
> SD_CLUSTER shares the CPU resources like llc tags or l2 cache, that's
> easy confuse with SD_SHARE_PKG_RESOURCES. So let's specifical point
> what the latter shares: LLC. That would reduce some confusing.
On neither JACOBSVILLE nor kunpeng920, it seems CLUSTER isn't LLC.
on Jacobsville, cluster is L2-cache while Jacobsville has L3; on kunpeng920,
cluster is L3-tag. On kunpeng920, actually 24 cpus or 32cpus share one LLC,
the whole L3. cluster is kind of like middle-level caches.
So I feel this patch isn't precise.
>
> Suggested-by: Valentin Schneider <vschn...@redhat.com>
> Signed-off-by: Alex Shi <al...@kernel.org>
> Cc: linux-ker...@vger.kernel.org
> Cc: linuxppc-dev@lists.ozlabs.org
> Cc: Miaohe Lin <linmia...@huawei.com>
> Cc: Barry Song <song.bao....@hisilicon.com>
> Cc: Mark Rutland <mark.rutl...@arm.com>
> Cc: Frederic Weisbecker <frede...@kernel.org>
> Cc: Daniel Bristot de Oliveira <bris...@redhat.com>
> Cc: Ben Segall <bseg...@google.com>
> Cc: Steven Rostedt <rost...@goodmis.org>
> Cc: Dietmar Eggemann <dietmar.eggem...@arm.com>
> Cc: Juri Lelli <juri.le...@redhat.com>
> Cc: Ingo Molnar <mi...@redhat.com>
> Cc: "Naveen N. Rao" <naveen.n....@linux.ibm.com>
> Cc: "Aneesh Kumar K.V" <aneesh.ku...@kernel.org>
> Cc: Christophe Leroy <christophe.le...@csgroup.eu>
> Cc: "Gautham R. Shenoy" <gautham.she...@amd.com>
> Cc: Yicong Yang <yangyic...@hisilicon.com>
> Cc: Ricardo Neri <ricardo.neri-calde...@linux.intel.com>
> Cc: Josh Poimboeuf <jpoim...@kernel.org>
> Cc: Srikar Dronamraju <sri...@linux.vnet.ibm.com>
> Cc: Valentin Schneider <vschn...@redhat.com>
> Cc: Nicholas Piggin <npig...@gmail.com>
> Cc: Michael Ellerman <m...@ellerman.id.au>
> Reviewed-by: Valentin Schneider <vschn...@redhat.com>
> Reviewed-by: Ricardo Neri <ricardo.neri-calde...@linux.intel.com>
> ---
> arch/powerpc/kernel/smp.c | 6 +++---
> include/linux/sched/sd_flags.h | 4 ++--
> include/linux/sched/topology.h | 6 +++---
> kernel/sched/fair.c | 2 +-
> kernel/sched/topology.c | 28 ++++++++++++++--------------
> 5 files changed, 23 insertions(+), 23 deletions(-)
>
> diff --git a/arch/powerpc/kernel/smp.c b/arch/powerpc/kernel/smp.c
> index 693334c20d07..a60e4139214b 100644
> --- a/arch/powerpc/kernel/smp.c
> +++ b/arch/powerpc/kernel/smp.c
> @@ -984,7 +984,7 @@ static bool shared_caches __ro_after_init;
> /* cpumask of CPUs with asymmetric SMT dependency */
> static int powerpc_smt_flags(void)
> {
> - int flags = SD_SHARE_CPUCAPACITY | SD_SHARE_PKG_RESOURCES;
> + int flags = SD_SHARE_CPUCAPACITY | SD_SHARE_LLC;
>
> if (cpu_has_feature(CPU_FTR_ASYM_SMT)) {
> printk_once(KERN_INFO "Enabling Asymmetric SMT scheduling\n");
> @@ -1010,9 +1010,9 @@ static __ro_after_init
> DEFINE_STATIC_KEY_FALSE(splpar_asym_pack);
> static int powerpc_shared_cache_flags(void)
> {
> if (static_branch_unlikely(&splpar_asym_pack))
> - return SD_SHARE_PKG_RESOURCES | SD_ASYM_PACKING;
> + return SD_SHARE_LLC | SD_ASYM_PACKING;
>
> - return SD_SHARE_PKG_RESOURCES;
> + return SD_SHARE_LLC;
> }
>
> static int powerpc_shared_proc_flags(void)
> diff --git a/include/linux/sched/sd_flags.h b/include/linux/sched/sd_flags.h
> index a8b28647aafc..b04a5d04dee9 100644
> --- a/include/linux/sched/sd_flags.h
> +++ b/include/linux/sched/sd_flags.h
> @@ -117,13 +117,13 @@ SD_FLAG(SD_SHARE_CPUCAPACITY, SDF_SHARED_CHILD |
> SDF_NEEDS_GROUPS)
> SD_FLAG(SD_CLUSTER, SDF_NEEDS_GROUPS)
>
> /*
> - * Domain members share CPU package resources (i.e. caches)
> + * Domain members share CPU Last Level Caches
> *
> * SHARED_CHILD: Set from the base domain up until spanned CPUs no longer
> share
> * the same cache(s).
> * NEEDS_GROUPS: Caches are shared between groups.
> */
> -SD_FLAG(SD_SHARE_PKG_RESOURCES, SDF_SHARED_CHILD | SDF_NEEDS_GROUPS)
> +SD_FLAG(SD_SHARE_LLC, SDF_SHARED_CHILD | SDF_NEEDS_GROUPS)
>
> /*
> * Only a single load balancing instance
> diff --git a/include/linux/sched/topology.h b/include/linux/sched/topology.h
> index a6e04b4a21d7..191b122158fb 100644
> --- a/include/linux/sched/topology.h
> +++ b/include/linux/sched/topology.h
> @@ -38,21 +38,21 @@ extern const struct sd_flag_debug sd_flag_debug[];
> #ifdef CONFIG_SCHED_SMT
> static inline int cpu_smt_flags(void)
> {
> - return SD_SHARE_CPUCAPACITY | SD_SHARE_PKG_RESOURCES;
> + return SD_SHARE_CPUCAPACITY | SD_SHARE_LLC;
> }
> #endif
>
> #ifdef CONFIG_SCHED_CLUSTER
> static inline int cpu_cluster_flags(void)
> {
> - return SD_CLUSTER | SD_SHARE_PKG_RESOURCES;
> + return SD_CLUSTER | SD_SHARE_LLC;
> }
> #endif
>
> #ifdef CONFIG_SCHED_MC
> static inline int cpu_core_flags(void)
> {
> - return SD_SHARE_PKG_RESOURCES;
> + return SD_SHARE_LLC;
> }
> #endif
>
> diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
> index cd1ec57c0b7b..da6c77d05d07 100644
> --- a/kernel/sched/fair.c
> +++ b/kernel/sched/fair.c
> @@ -10687,7 +10687,7 @@ static inline void calculate_imbalance(struct lb_env
> *env, struct sd_lb_stats *s
> */
> if (local->group_type == group_has_spare) {
> if ((busiest->group_type > group_fully_busy) &&
> - !(env->sd->flags & SD_SHARE_PKG_RESOURCES)) {
> + !(env->sd->flags & SD_SHARE_LLC)) {
> /*
> * If busiest is overloaded, try to fill spare
> * capacity. This might end up creating spare capacity
> diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c
> index 0b33f7b05d21..99ea5986038c 100644
> --- a/kernel/sched/topology.c
> +++ b/kernel/sched/topology.c
> @@ -657,13 +657,13 @@ static void destroy_sched_domains(struct sched_domain
> *sd)
> }
>
> /*
> - * Keep a special pointer to the highest sched_domain that has
> - * SD_SHARE_PKG_RESOURCE set (Last Level Cache Domain) for this
> - * allows us to avoid some pointer chasing select_idle_sibling().
> + * Keep a special pointer to the highest sched_domain that has SD_SHARE_LLC
> set
> + * (Last Level Cache Domain) for this allows us to avoid some pointer chasing
> + * select_idle_sibling().
> *
> - * Also keep a unique ID per domain (we use the first CPU number in
> - * the cpumask of the domain), this allows us to quickly tell if
> - * two CPUs are in the same cache domain, see cpus_share_cache().
> + * Also keep a unique ID per domain (we use the first CPU number in the
> cpumask
> + * of the domain), this allows us to quickly tell if two CPUs are in the same
> + * cache domain, see cpus_share_cache().
> */
> DEFINE_PER_CPU(struct sched_domain __rcu *, sd_llc);
> DEFINE_PER_CPU(int, sd_llc_size);
> @@ -684,7 +684,7 @@ static void update_top_cache_domain(int cpu)
> int id = cpu;
> int size = 1;
>
> - sd = highest_flag_domain(cpu, SD_SHARE_PKG_RESOURCES);
> + sd = highest_flag_domain(cpu, SD_SHARE_LLC);
> if (sd) {
> id = cpumask_first(sched_domain_span(sd));
> size = cpumask_weight(sched_domain_span(sd));
> @@ -1554,7 +1554,7 @@ static struct cpumask
> ***sched_domains_numa_masks;
> * function. For details, see include/linux/sched/sd_flags.h.
> *
> * SD_SHARE_CPUCAPACITY
> - * SD_SHARE_PKG_RESOURCES
> + * SD_SHARE_LLC
> * SD_CLUSTER
> * SD_NUMA
> *
> @@ -1566,7 +1566,7 @@ static struct cpumask
> ***sched_domains_numa_masks;
> #define TOPOLOGY_SD_FLAGS \
> (SD_SHARE_CPUCAPACITY | \
> SD_CLUSTER | \
> - SD_SHARE_PKG_RESOURCES | \
> + SD_SHARE_LLC | \
> SD_NUMA | \
> SD_ASYM_PACKING)
>
> @@ -1609,7 +1609,7 @@ sd_init(struct sched_domain_topology_level *tl,
> | 0*SD_BALANCE_WAKE
> | 1*SD_WAKE_AFFINE
> | 0*SD_SHARE_CPUCAPACITY
> - | 0*SD_SHARE_PKG_RESOURCES
> + | 0*SD_SHARE_LLC
> | 0*SD_SERIALIZE
> | 1*SD_PREFER_SIBLING
> | 0*SD_NUMA
> @@ -1646,7 +1646,7 @@ sd_init(struct sched_domain_topology_level *tl,
> if (sd->flags & SD_SHARE_CPUCAPACITY) {
> sd->imbalance_pct = 110;
>
> - } else if (sd->flags & SD_SHARE_PKG_RESOURCES) {
> + } else if (sd->flags & SD_SHARE_LLC) {
> sd->imbalance_pct = 117;
> sd->cache_nice_tries = 1;
>
> @@ -1671,7 +1671,7 @@ sd_init(struct sched_domain_topology_level *tl,
> * For all levels sharing cache; connect a sched_domain_shared
> * instance.
> */
> - if (sd->flags & SD_SHARE_PKG_RESOURCES) {
> + if (sd->flags & SD_SHARE_LLC) {
> sd->shared = *per_cpu_ptr(sdd->sds, sd_id);
> atomic_inc(&sd->shared->ref);
> atomic_set(&sd->shared->nr_busy_cpus, sd_weight);
> @@ -2446,8 +2446,8 @@ build_sched_domains(const struct cpumask *cpu_map,
> struct sched_domain_attr *att
> for (sd = *per_cpu_ptr(d.sd, i); sd; sd = sd->parent) {
> struct sched_domain *child = sd->child;
>
> - if (!(sd->flags & SD_SHARE_PKG_RESOURCES) && child &&
> - (child->flags & SD_SHARE_PKG_RESOURCES)) {
> + if (!(sd->flags & SD_SHARE_LLC) && child &&
> + (child->flags & SD_SHARE_LLC)) {
> struct sched_domain __rcu *top_p;
> unsigned int nr_llcs;
>
> --
> 2.43.0
>
>
Thanks
Barry
