| Message ID | 20211205183220.818872-5-longman@redhat.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | cgroup/cpuset: Add new cpuset partition type & empty effecitve cpus | expand |
On Sun, Dec 05, 2021 at 01:32:17PM -0500, Waiman Long wrote: > Cpuset v1 uses the sched_load_balance control file to determine if load > balancing should be enabled. Cpuset v2 gets rid of sched_load_balance > as its use may require disabling load balancing at cgroup root. > > For workloads that require very low latency like DPDK, the latency > jitters caused by periodic load balancing may exceed the desired > latency limit. > > When cpuset v2 is in use, the only way to avoid this latency cost is to > use the "isolcpus=" kernel boot option to isolate a set of CPUs. After > the kernel boot, however, there is no way to add or remove CPUs from > this isolated set. For workloads that are more dynamic in nature, that > means users have to provision enough CPUs for the worst case situation > resulting in excess idle CPUs. > > To address this issue for cpuset v2, a new cpuset.cpus.partition type > "isolated" is added which allows the creation of a cpuset partition > without load balancing. This will allow system administrators to > dynamically adjust the size of isolated partition to the current need > of the workload without rebooting the system. you can, ofcourse, create lots of 1 cpu partitions, which is effectively what you're doing, except there was a problem with that which you also forgot to mention.
On 1/12/22 10:21, Peter Zijlstra wrote: > On Sun, Dec 05, 2021 at 01:32:17PM -0500, Waiman Long wrote: >> Cpuset v1 uses the sched_load_balance control file to determine if load >> balancing should be enabled. Cpuset v2 gets rid of sched_load_balance >> as its use may require disabling load balancing at cgroup root. >> >> For workloads that require very low latency like DPDK, the latency >> jitters caused by periodic load balancing may exceed the desired >> latency limit. >> >> When cpuset v2 is in use, the only way to avoid this latency cost is to >> use the "isolcpus=" kernel boot option to isolate a set of CPUs. After >> the kernel boot, however, there is no way to add or remove CPUs from >> this isolated set. For workloads that are more dynamic in nature, that >> means users have to provision enough CPUs for the worst case situation >> resulting in excess idle CPUs. >> >> To address this issue for cpuset v2, a new cpuset.cpus.partition type >> "isolated" is added which allows the creation of a cpuset partition >> without load balancing. This will allow system administrators to >> dynamically adjust the size of isolated partition to the current need >> of the workload without rebooting the system. > you can, ofcourse, create lots of 1 cpu partitions, which is effectively > what you're doing, except there was a problem with that which you also > forgot to mention. Yes, that is a possible workaround. However, it makes cgroup management much harder especially in the cgroup v2 environment where multiple controllers are likely to be enabled in the same cgroup. Cheers, Longman
On Wed, Jan 12, 2022 at 10:40:01AM -0500, Waiman Long wrote: > Yes, that is a possible workaround. However, it makes cgroup management much > harder especially in the cgroup v2 environment where multiple controllers > are likely to be enabled in the same cgroup. In most cases, cgroup2 shouldn't be a problem here given that controllers can be enabled selectively and this 1-cpu cgroups will most likely be leaves. But yeah, not super convenient. Thanks.
diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c index c877eecf4c5e..cfab10911682 100644 --- a/kernel/cgroup/cpuset.c +++ b/kernel/cgroup/cpuset.c @@ -179,6 +179,8 @@ struct cpuset { * * 1 - partition root * + * 2 - partition root without load balancing (isolated) + * * -1 - invalid partition root * None of the cpus in cpus_allowed can be put into the parent's * subparts_cpus. In this case, the cpuset is not a real partition @@ -188,6 +190,7 @@ struct cpuset { */ #define PRS_DISABLED 0 #define PRS_ENABLED 1 +#define PRS_ISOLATED 2 #define PRS_ERROR -1 /* @@ -1316,17 +1319,22 @@ static int update_parent_subparts_cpumask(struct cpuset *cpuset, int cmd, if (cmd == partcmd_update) { /* - * Check for possible transition between PRS_ENABLED - * and PRS_ERROR. + * Check for possible transition between PRS_ERROR and + * PRS_ENABLED/PRS_ISOLATED. */ switch (cpuset->partition_root_state) { case PRS_ENABLED: + case PRS_ISOLATED: if (part_error) new_prs = PRS_ERROR; break; case PRS_ERROR: - if (!part_error) + if (part_error) + break; + if (is_sched_load_balance(cpuset)) new_prs = PRS_ENABLED; + else + new_prs = PRS_ISOLATED; break; } } @@ -1451,6 +1459,7 @@ static void update_cpumasks_hier(struct cpuset *cs, struct tmpmasks *tmp, if ((cp != cs) && old_prs) { switch (parent->partition_root_state) { case PRS_ENABLED: + case PRS_ISOLATED: update_parent = true; break; @@ -2023,6 +2032,7 @@ static int update_flag(cpuset_flagbits_t bit, struct cpuset *cs, static int update_prstate(struct cpuset *cs, int new_prs) { int err, old_prs = cs->partition_root_state; + bool sched_domain_rebuilt = false; struct cpuset *parent = parent_cs(cs); struct tmpmasks tmpmask; @@ -2059,6 +2069,22 @@ static int update_prstate(struct cpuset *cs, int new_prs) update_flag(CS_CPU_EXCLUSIVE, cs, 0); goto out; } + + if (new_prs == PRS_ISOLATED) { + /* + * Disable the load balance flag should not return an + * error unless the system is running out of memory. + */ + update_flag(CS_SCHED_LOAD_BALANCE, cs, 0); + sched_domain_rebuilt = true; + } + } else if (old_prs && new_prs) { + /* + * A change in load balance state only, no change in cpumasks. + */ + update_flag(CS_SCHED_LOAD_BALANCE, cs, (new_prs != PRS_ISOLATED)); + err = 0; + goto out; /* Sched domain is rebuilt in update_flag() */ } else { /* * Switch back to member is always allowed even if it @@ -2080,6 +2106,12 @@ static int update_prstate(struct cpuset *cs, int new_prs) /* Turning off CS_CPU_EXCLUSIVE will not return error */ update_flag(CS_CPU_EXCLUSIVE, cs, 0); + + if (!is_sched_load_balance(cs)) { + /* Make sure load balance is on */ + update_flag(CS_SCHED_LOAD_BALANCE, cs, 1); + sched_domain_rebuilt = true; + } } /* @@ -2092,7 +2124,8 @@ static int update_prstate(struct cpuset *cs, int new_prs) if (parent->child_ecpus_count) update_sibling_cpumasks(parent, cs, &tmpmask); - rebuild_sched_domains_locked(); + if (!sched_domain_rebuilt) + rebuild_sched_domains_locked(); out: if (!err) { spin_lock_irq(&callback_lock); @@ -2599,16 +2632,21 @@ static s64 cpuset_read_s64(struct cgroup_subsys_state *css, struct cftype *cft) static int sched_partition_show(struct seq_file *seq, void *v) { struct cpuset *cs = css_cs(seq_css(seq)); + const char *type; switch (cs->partition_root_state) { case PRS_ENABLED: seq_puts(seq, "root\n"); break; + case PRS_ISOLATED: + seq_puts(seq, "isolated\n"); + break; case PRS_DISABLED: seq_puts(seq, "member\n"); break; case PRS_ERROR: - seq_puts(seq, "root invalid\n"); + type = is_sched_load_balance(cs) ? "root" : "isolated"; + seq_printf(seq, "%s invalid\n", type); break; } return 0; @@ -2630,6 +2668,8 @@ static ssize_t sched_partition_write(struct kernfs_open_file *of, char *buf, val = PRS_ENABLED; else if (!strcmp(buf, "member")) val = PRS_DISABLED; + else if (!strcmp(buf, "isolated")) + val = PRS_ISOLATED; else return -EINVAL;
Cpuset v1 uses the sched_load_balance control file to determine if load balancing should be enabled. Cpuset v2 gets rid of sched_load_balance as its use may require disabling load balancing at cgroup root. For workloads that require very low latency like DPDK, the latency jitters caused by periodic load balancing may exceed the desired latency limit. When cpuset v2 is in use, the only way to avoid this latency cost is to use the "isolcpus=" kernel boot option to isolate a set of CPUs. After the kernel boot, however, there is no way to add or remove CPUs from this isolated set. For workloads that are more dynamic in nature, that means users have to provision enough CPUs for the worst case situation resulting in excess idle CPUs. To address this issue for cpuset v2, a new cpuset.cpus.partition type "isolated" is added which allows the creation of a cpuset partition without load balancing. This will allow system administrators to dynamically adjust the size of isolated partition to the current need of the workload without rebooting the system. Signed-off-by: Waiman Long <longman@redhat.com> --- kernel/cgroup/cpuset.c | 50 +++++++++++++++++++++++++++++++++++++----- 1 file changed, 45 insertions(+), 5 deletions(-)