timers: Exclude isolated cpus from timer migration
Subsystems (7)
v15 cover timers: Exclude isolated cpus from timer migration
The timer migration mechanism allows active CPUs to pull timers from idle ones to improve the overall idle time. This is however undesired when CPU intensive workloads run on isolated cores, as the algorithm would move the timers from housekeeping to isolated cores, negatively affecting the isolation.
Exclude isolated cores from the timer migration algorithm, extend the concept of unavailable cores, currently used for offline ones, to isolated ones:
* A core is unavailable if isolated or offline;
* A core is available if non isolated and online;
A core is considered unavailable as isolated if it belongs to:
* the isolcpus (domain) list
* an isolated cpuset
Except if it is:
* in the nohz_full list (already idle for the hierarchy)
* the nohz timekeeper core (must be available to handle global timers)
CPUs are added to the hierarchy during late boot, excluding isolated ones, the hierarchy is also adapted when the cpuset isolation changes.
Due to how the timer migration algorithm works, any CPU part of the hierarchy can have their global timers pulled by remote CPUs and have to pull remote timers, only skipping pulling remote timers would break the logic.
For this reason, prevent isolated CPUs from pulling remote global timers, but also the other way around: any global timer started on an isolated CPU will run there. This does not break the concept of isolation (global timers don't come from outside the CPU) and, if considered inappropriate, can usually be mitigated with other isolation techniques (e.g. IRQ pinning).
This effect was noticed on a 128 cores machine running oslat on the isolated cores (1-31,33-63,65-95,97-127). The tool monopolises CPUs, and the CPU with lowest count in a timer migration hierarchy (here 1 and 65) appears as always active and continuously pulls global timers, from the housekeeping CPUs. This ends up moving driver work (e.g. delayed work) to isolated CPUs and causes latency spikes:
before the change:
# oslat -c 1-31,33-63,65-95,97-127 -D 62s ... Maximum: 1203 10 3 4 ... 5 (us)
after the change:
# oslat -c 1-31,33-63,65-95,97-127 -D 62s ... Maximum: 10 4 3 4 3 ... 5 (us)
The same behaviour was observed on a machine with as few as 20 cores / 40 threads with isocpus set to: 1-9,11-39 with rtla-osnoise-top.
The first 4 patches are preparatory work to change the concept of online/offline to available/unavailable, keep track of those in a separate cpumask cleanup the setting/clearing functions and change a function name in cpuset code.
Patch 5 adapt isolation to prevent domain isolated and nohz_full from covering all CPUs not leaving any housekeeping one. This can lead to problems with the changes introduced in this series because no CPU would remain to handle global timers.
(The corresponding change for cpuset was removed from this version of the series and is present in [2]).
Patch 7 extends the unavailable status to domain isolated CPUs, which is the main contribution of the series.
Changes since v14:
* Rebase on tip/timers/core, rename one more ->online field
* Mark the static key as static
* Share code between tmigr_init_isolation and tmigr_isolated_exclude_cpumask
Changes since v13:
* Remove tmigr late initialisation and restore late isolation (as in v8 [1])
* Use workqueues in initialisation just like tmigr_available_cpumask()
* Use static key for tmigr_exclude_isolated
* Remove cpuset patch checking for HK conflict (included in [2])
* Rename cpuset helper to update_isolation_cpumasks as in [2]
Changes since v12:
* Pick and adapt patch by Yury Norov to initialise cpumasks
* Reorganise accesses to tmigr_available_cpumask to avoid races
Changes since v11:
* Rename isolcpus_nohz_conflict() to isolated_cpus_can_update()
* Move tick_nohz_cpu_hotpluggable() check to tmigr_is_isolated()
* Use workqueues in tmigr_isolated_exclude_cpumask() to avoid sleeping while atomic
* Add cpumask initialiser to safely use cpumask cleanup helpers
Changes since v10:
* Simplify housekeeping conflict condition
* Reword commit (Frederic Weisbecker)
Changes since v9:
* Fix total housekeeping enforcement to focus only on nohz and domain
* Avoid out of bound access in the housekeeping array if no flag is set
* Consider isolated_cpus while checking for nohz conflicts in cpuset
* Improve comment about why nohz CPUs are not excluded by tmigr
Changes since v8 [1]:
* Postpone hotplug registration to late initcall (Frederic Weisbecker)
* Move main activation logic in _tmigr_set_cpu_available() and call it after checking for isolation on hotplug and cpusets changes
* Call _tmigr_set_cpu_available directly to force enable tick CPU if required (this saves checking for that on every hotplug change).
Changes since v7:
* Move tmigr_available_cpumask out of tmc lock and specify conditions.
* Initialise tmigr isolation despite the state of isolcpus.
* Move tick CPU check to condition to run SMP call.
* Fix descriptions.
Changes since v6 [3]:
* Prevent isolation checks from running during early boot
* Prevent double (de)activation while setting cpus (un)available
* Use synchronous smp calls from the isolation path
* General cleanup
Changes since v5:
* Remove fallback if no housekeeping is left by isolcpus and nohz_full
* Adjust condition not to activate CPUs in the migration hierarchy
* Always force the nohz tick CPU active in the hierarchy
Changes since v4 [4]:
* use on_each_cpu_mask() with changes on isolated CPUs to avoid races
* keep nohz_full CPUs included in the timer migration hierarchy
* prevent domain isolated and nohz_full to cover all CPUs
Changes since v3:
* add parameter to function documentation
* split into multiple straightforward patches
Changes since v2:
* improve comments about handling CPUs isolated at boot
* minor cleanup
Changes since v1 [5]:
* split into smaller patches
* use available mask instead of unavailable
* simplification and cleanup
[1] - https://lore.kernel.org/lkml/[email protected]
[2] - https://lore.kernel.org/lkml/[email protected]
[3] - https://lore.kernel.org/lkml/[email protected]
[4] - https://lore.kernel.org/lkml/[email protected]
[5] - https://lore.kernel.org/lkml/[email protected]
Gabriele Monaco (6):
timers: Rename tmigr 'online' bit to 'available'
timers: Add the available mask in timer migration
timers: Use scoped_guard when setting/clearing the tmigr available
flag
cgroup/cpuset: Rename update_unbound_workqueue_cpumask() to
update_isolation_cpumasks()
sched/isolation: Force housekeeping if isolcpus and nohz_full don't
leave any
timers: Exclude isolated cpus from timer migration
Yury Norov (1):
cpumask: Add initialiser to use cleanup helpers
include/linux/cpumask.h | 2 + include/linux/timer.h | 9 ++ include/trace/events/timer_migration.h | 4 +- kernel/cgroup/cpuset.c | 15 +- kernel/sched/isolation.c | 23 +++ kernel/time/timer_migration.c | 213 +++++++++++++++++++++---- kernel/time/timer_migration.h | 2 +- 7 files changed, 232 insertions(+), 36 deletions(-)
base-commit: ba14500e4bfcab5e841fbf8d7fcbbc80e98d6b9e
Signature · 1 line
2.51.1
PATCH v15 1/7 timers: Rename tmigr 'online' bit to 'available'
The timer migration hierarchy excludes offline CPUs via the tmigr_is_not_available function, which is essentially checking the online bit for the CPU.
Rename the online bit to available and all references in function names and tracepoint to generalise the concept of available CPUs.
--- include/trace/events/timer_migration.h | 4 ++-- kernel/time/timer_migration.c | 24 ++++++++++++------------ kernel/time/timer_migration.h | 2 +- 3 files changed, 15 insertions(+), 15 deletions(-)
include/trace/events/timer_migration.h +2 -2
@@ -173,14 +173,14 @@ DEFINE_EVENT(tmigr_cpugroup, tmigr_cpu_active,TP_ARGS (tmc) ); -DEFINE_EVENT (tmigr_cpugroup, tmigr_cpu_online, +DEFINE_EVENT (tmigr_cpugroup, tmigr_cpu_available, TP_PROTO(struct tmigr_cpu *tmc),TP_ARGS (tmc) ); -DEFINE_EVENT(tmigr_cpugroup, tmigr_cpu_offline, +DEFINE_EVENT(tmigr_cpugroup, tmigr_cpu_unavailable, TP_PROTO(struct tmigr_cpu *tmc),
kernel/time/timer_migration.c +12 -12
@@ -429,7 +429,7 @@ static DEFINE_PER_CPU(struct tmigr_cpu, tmigr_cpu);static inline bool tmigr_is_not_available (struct tmigr_cpu *tmc) { -return !(tmc->tmgroup && tmc->online); +return !(tmc->tmgroup && tmc->available); }/*
@@ -926,7 +926,7 @@ static void tmigr_handle_remote_cpu(unsigned int cpu, u64 now,* updated the event takes care when hierarchy is completely * idle. Otherwise the migrator does it as the event is enqueued. */ -if (!tmc->online || tmc->remote || tmc->cpuevt.ignore || +if (!tmc->available || tmc->remote || tmc->cpuevt.ignore || now < tmc->cpuevt.nextevt.expires) {raw_spin_unlock_irq (&tmc->lock);return ;
@@ -973,7 +973,7 @@ static void tmigr_handle_remote_cpu(unsigned int cpu, u64 now,* (See also section "Required event and timerqueue update after a * remote expiry" in the documentation at the top) */ -if (!tmc->online || !tmc->idle) { +if (!tmc->available || !tmc->idle) {timer_unlock_remote_bases (cpu); goto unlock; }
@@ -1422,19 +1422,19 @@ static long tmigr_trigger_active(void *unused) {struct tmigr_cpu *tmc =this_cpu_ptr (&tmigr_cpu); -WARN_ON_ONCE (!tmc->online || tmc->idle); +WARN_ON_ONCE (!tmc->available || tmc->idle);return 0 ; } -static int tmigr_cpu_offline (unsigned int cpu) +static int tmigr_clear_cpu_available (unsigned int cpu) {struct tmigr_cpu *tmc =this_cpu_ptr (&tmigr_cpu);int migrator;u64 firstexp;raw_spin_lock_irq (&tmc->lock); - tmc->online = false; + tmc->available = false;WRITE_ONCE (tmc->wakeup,KTIME_MAX );/*
@@ -1442,7 +1442,7 @@ static int tmigr_cpu_offline(unsigned int cpu)* offline; Therefore nextevt value is set to KTIME_MAX */ firstexp =__tmigr_cpu_deactivate (tmc,KTIME_MAX ); -trace_tmigr_cpu_offline (tmc); +trace_tmigr_cpu_unavailable (tmc);raw_spin_unlock_irq (&tmc->lock);if (firstexp !=KTIME_MAX ) {
@@ -1453,7 +1453,7 @@ static int tmigr_cpu_offline(unsigned int cpu)return 0 ; } -static int tmigr_cpu_online (unsigned int cpu) +static int tmigr_set_cpu_available (unsigned int cpu) {struct tmigr_cpu *tmc =this_cpu_ptr (&tmigr_cpu);
@@ -1462,11 +1462,11 @@ static int tmigr_cpu_online(unsigned int cpu)return -EINVAL ;raw_spin_lock_irq (&tmc->lock); -trace_tmigr_cpu_online (tmc); +trace_tmigr_cpu_available (tmc); tmc->idle =timer_base_is_idle ();if (!tmc->idle)__tmigr_cpu_activate (tmc); - tmc->online = true; + tmc->available = true;raw_spin_unlock_irq (&tmc->lock);return 0 ; }
@@ -1758,7 +1758,7 @@ static int tmigr_add_cpu(unsigned int cpu)* The (likely) current CPU is expected to be online in the hierarchy, * otherwise the old root may not be active as expected. */ -WARN_ON_ONCE (!per_cpu_ptr (&tmigr_cpu,raw_smp_processor_id ())->online); +WARN_ON_ONCE (!per_cpu_ptr (&tmigr_cpu,raw_smp_processor_id ())->available); ret =tmigr_setup_groups (-1 , old_root->numa_node, old_root, true); }
@@ -1854,7 +1854,7 @@ static int __init tmigr_init(void) goto err; ret =cpuhp_setup_state (CPUHP_AP_TMIGR_ONLINE ,"tmigr:online" , - tmigr_cpu_online, tmigr_cpu_offline); + tmigr_set_cpu_available, tmigr_clear_cpu_available);if (ret) goto err;
kernel/time/timer_migration.h +1 -1
@@ -97,7 +97,7 @@ struct tmigr_group {*/ struct tmigr_cpu {raw_spinlock_t lock; -bool online; +bool available;bool idle;bool remote;struct tmigr_group *tmgroup;
Signature · 1 line
2.51.1
Re: [PATCH v15 1/7] timers: Rename tmigr 'online' bit to 'available'
Quoted 9 lines · Gabriele Monaco
The timer migration hierarchy excludes offline CPUs via the
tmigr_is_not_availablefunction, which is essentially checking the online bit for the CPU.Rename the online bit to available and all references in function names and tracepoint to generalise the concept of available CPUs.
Reviewed-by Frederic Weisbecker <[email protected]>Signed-off-by Gabriele Monaco <[email protected]>
PATCH v15 2/7 timers: Add the available mask in timer migration
In reply to Gabriele MonacoKeep track of the CPUs available for timer migration in a cpumask. This prepares the ground to generalise the concept of unavailable CPUs.
--- kernel/time/timer_migration.c | 15 ++++++++++++++- 1 file changed, 14 insertions(+), 1 deletion(-)
kernel/time/timer_migration.c +14 -1
@@ -424,6 +424,12 @@ static struct tmigr_group *tmigr_root; static DEFINE_PER_CPU(struct tmigr_cpu, tmigr_cpu); +/* +* CPUs available for timer migration. +* Protected by cpuset_mutex (with cpus_read_lock held) or cpus_write_lock. +*/ +static cpumask_var_t tmigr_available_cpumask; +#define TMIGR_NONE 0xFF#define BIT_CNT 8
@@ -1433,6 +1439,7 @@ static int tmigr_clear_cpu_available(unsigned int cpu)int migrator;u64 firstexp; +cpumask_clear_cpu (cpu, tmigr_available_cpumask);raw_spin_lock_irq (&tmc->lock); tmc->available = false;WRITE_ONCE (tmc->wakeup,KTIME_MAX );
@@ -1446,7 +1453,7 @@ static int tmigr_clear_cpu_available(unsigned int cpu)raw_spin_unlock_irq (&tmc->lock);if (firstexp !=KTIME_MAX ) { - migrator =cpumask_any_but (cpu_online_mask, cpu); + migrator =cpumask_any (tmigr_available_cpumask);work_on_cpu (migrator, tmigr_trigger_active,NULL ); }
@@ -1461,6 +1468,7 @@ static int tmigr_set_cpu_available(unsigned int cpu)if (WARN_ON_ONCE (!tmc->tmgroup))return -EINVAL ; +cpumask_set_cpu (cpu, tmigr_available_cpumask);raw_spin_lock_irq (&tmc->lock);trace_tmigr_cpu_available (tmc); tmc->idle =timer_base_is_idle ();
@@ -1805,6 +1813,11 @@ static int __init tmigr_init(void)if (ncpus ==1 )return 0 ; +if (!zalloc_cpumask_var (&tmigr_available_cpumask,GFP_KERNEL )) { + ret = -ENOMEM ; + goto err; + } +/* * Calculate the required hierarchy levels. Unfortunately there is no * reliable information available, unless all possible CPUs have been
Signature · 1 line
2.51.1
Re: [PATCH v15 2/7] timers: Add the available mask in timer migration
On Thu, Nov 13 2025 at 09:33, Gabriele Monaco wrote:
Can we please consistently have 'timers/migration:' as prefix like the other existing commits for this series?
That allows to write this subject line in a coherent way:
timers/migration: Provide and use cpumask for available CPUs
or something like that.
Why did I waste time for documenting all of this?
https://www.kernel.org/doc/html/latest/process/maintainer-tip.html#patch-subject
"‘git log path/to/file’ should give you a reasonable hint in most cases."
Is it really that hard?
Quoted 5 lines
Keep track of the CPUs available for timer migration in a cpumask. This prepares the ground to generalise the concept of unavailable CPUs.
Reviewed-by Frederic Weisbecker <[email protected]>Signed-off-by Gabriele Monaco <[email protected]>
Aside of the nit above:
PATCH v15 3/7 timers: Use scoped_guard when setting/clearing the tmigr available flag
In reply to Gabriele MonacoCleanup tmigr_clear_cpu_available() and tmigr_set_cpu_available() to prepare for easier checks on the available flag.
--- kernel/time/timer_migration.c | 34 +++++++++++++++++----------------- 1 file changed, 17 insertions(+), 17 deletions(-)
kernel/time/timer_migration.c +17 -17
@@ -1440,17 +1440,17 @@ static int tmigr_clear_cpu_available(unsigned int cpu)u64 firstexp;cpumask_clear_cpu (cpu, tmigr_available_cpumask); -raw_spin_lock_irq (&tmc->lock); - tmc->available = false; -WRITE_ONCE (tmc->wakeup,KTIME_MAX ); + scoped_guard(raw_spinlock_irq, &tmc->lock) { + tmc->available = false; +WRITE_ONCE (tmc->wakeup,KTIME_MAX ); -/* -* CPU has to handle the local events on his own, when on the way to -* offline; Therefore nextevt value is set to KTIME_MAX -*/ - firstexp =__tmigr_cpu_deactivate (tmc,KTIME_MAX ); -trace_tmigr_cpu_unavailable (tmc); -raw_spin_unlock_irq (&tmc->lock); +/* +* CPU has to handle the local events on his own, when on the way to +* offline; Therefore nextevt value is set to KTIME_MAX +*/ + firstexp =__tmigr_cpu_deactivate (tmc,KTIME_MAX ); +trace_tmigr_cpu_unavailable (tmc); + }if (firstexp !=KTIME_MAX ) { migrator =cpumask_any (tmigr_available_cpumask);
@@ -1469,13 +1469,13 @@ static int tmigr_set_cpu_available(unsigned int cpu)return -EINVAL ;cpumask_set_cpu (cpu, tmigr_available_cpumask); -raw_spin_lock_irq (&tmc->lock); -trace_tmigr_cpu_available (tmc); - tmc->idle =timer_base_is_idle (); -if (!tmc->idle) -__tmigr_cpu_activate (tmc); - tmc->available = true; -raw_spin_unlock_irq (&tmc->lock); + scoped_guard(raw_spinlock_irq, &tmc->lock) { +trace_tmigr_cpu_available (tmc); + tmc->idle =timer_base_is_idle (); +if (!tmc->idle) +__tmigr_cpu_activate (tmc); + tmc->available = true; + }return 0 ; }
Signature · 1 line
2.51.1
Re: [PATCH v15 3/7] timers: Use scoped_guard when setting/clearing the tmigr available flag
Quoted 5 lines · Gabriele Monaco
Cleanup
tmigr_clear_cpu_available()andtmigr_set_cpu_available()to prepare for easier checks on the available flag.Reviewed-by Frederic Weisbecker <[email protected]>Signed-off-by Gabriele Monaco <[email protected]>
PATCH v15 4/7 cgroup/cpuset: Rename update_unbound_workqueue_cpumask() to update_isolation_cpumasks()
In reply to Gabriele Monacoupdate_unbound_workqueue_cpumask() updates unbound workqueues settings when there's a change in isolated CPUs, but it can be used for other subsystems requiring updated when isolated CPUs change.
Generalise the name to update_isolation_cpumasks() to prepare for other functions unrelated to workqueues to be called in that spot.
[longman: Change the function name to update_isolation_cpumasks()]
--- kernel/cgroup/cpuset.c | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-)
kernel/cgroup/cpuset.c +6 -6
@@ -1339,7 +1339,7 @@ static bool partition_xcpus_del(int old_prs, struct cpuset *parent,return isolcpus_updated; } -static void update_unbound_workqueue_cpumask (bool isolcpus_updated) +static void update_isolation_cpumasks (bool isolcpus_updated) {int ret;
@@ -1470,7 +1470,7 @@ static int remote_partition_enable(struct cpuset *cs, int new_prs,list_add (&cs->remote_sibling, &remote_children);cpumask_copy (cs->effective_xcpus, tmp->new_cpus);spin_unlock_irq (&callback_lock); -update_unbound_workqueue_cpumask (isolcpus_updated); +update_isolation_cpumasks (isolcpus_updated);cpuset_force_rebuild (); cs->prs_err =0 ;
@@ -1511,7 +1511,7 @@ static void remote_partition_disable(struct cpuset *cs, struct tmpmasks *tmp)compute_effective_exclusive_cpumask (cs,NULL ,NULL );reset_partition_data (cs);spin_unlock_irq (&callback_lock); -update_unbound_workqueue_cpumask (isolcpus_updated); +update_isolation_cpumasks (isolcpus_updated);cpuset_force_rebuild ();/*
@@ -1580,7 +1580,7 @@ static void remote_cpus_update(struct cpuset *cs, struct cpumask *xcpus,if (xcpus)cpumask_copy (cs->exclusive_cpus, xcpus);spin_unlock_irq (&callback_lock); -update_unbound_workqueue_cpumask (isolcpus_updated); +update_isolation_cpumasks (isolcpus_updated);if (adding || deleting)cpuset_force_rebuild ();
@@ -1943,7 +1943,7 @@ static int update_parent_effective_cpumask(struct cpuset *cs, int cmd,WARN_ON_ONCE (parent->nr_subparts <0 ); }spin_unlock_irq (&callback_lock); -update_unbound_workqueue_cpumask (isolcpus_updated); +update_isolation_cpumasks (isolcpus_updated);if ((old_prs != new_prs) && (cmd == partcmd_update))update_partition_exclusive_flag (cs, new_prs);
@@ -2968,7 +2968,7 @@ static int update_prstate(struct cpuset *cs, int new_prs)else if (isolcpus_updated)isolated_cpus_update (old_prs, new_prs, cs->effective_xcpus);spin_unlock_irq (&callback_lock); -update_unbound_workqueue_cpumask (isolcpus_updated); +update_isolation_cpumasks (isolcpus_updated);/* Force update if switching back to member & update effective_xcpus */ update_cpumasks_hier (cs, &tmpmask, !new_prs);
Signature · 1 line
2.51.1
PATCH v15 5/7 sched/isolation: Force housekeeping if isolcpus and nohz_full don't leave any
In reply to Gabriele MonacoCurrently the user can set up isolcpus and nohz_full in such a way that leaves no housekeeping CPU (i.e. no CPU that is neither domain isolated nor nohz full). This can be a problem for other subsystems (e.g. the timer wheel imgration).
Prevent this configuration by invalidating the last setting in case the union of isolcpus (domain) and nohz_full covers all CPUs.
--- kernel/sched/isolation.c | 23 +++++++++++++++++++++++ 1 file changed, 23 insertions(+)
kernel/sched/isolation.c +23 -0
@@ -167,6 +167,29 @@ static int __init housekeeping_setup(char *str, unsigned long flags) } } +/* +* Check the combination of nohz_full and isolcpus=domain, +* necessary to avoid problems with the timer migration +* hierarchy. managed_irq is ignored by this check since it +* isn't considered in the timer migration logic. +*/ + iter_flags = housekeeping.flags & (HK_FLAG_KERNEL_NOISE |HK_FLAG_DOMAIN ); + type =find_first_bit (&iter_flags,HK_TYPE_MAX ); +/* +* Pass the check if none of these flags were previously set or +* are not in the current selection. +*/ + iter_flags = flags & (HK_FLAG_KERNEL_NOISE |HK_FLAG_DOMAIN ); + first_cpu = (type ==HK_TYPE_MAX || !iter_flags) ?0 : +cpumask_first_and_and (cpu_present_mask, + housekeeping_staging, housekeeping.cpumasks[type]); +if (first_cpu >=min (nr_cpu_ids, setup_max_cpus)) { +pr_warn ("Housekeeping: must include one present CPU " +"neither in nohz_full= nor in isolcpus=domain, " +"ignoring setting %s\n" , str); + goto free_housekeeping_staging; + } + iter_flags = flags & ~housekeeping.flags;for_each_set_bit (type, &iter_flags,HK_TYPE_MAX )
Signature · 1 line
2.51.1
PATCH v15 6/7 cpumask: Add initialiser to use cleanup helpers
In reply to Gabriele MonacoFrom: Yury Norov <[email protected]>
Now we can simplify a code that allocates cpumasks for local needs.
Automatic variables have to be initialized at declaration, or at least before any possibility for the logic to return, so that compiler wouldn't try to call an associate destructor function on a random stack number.
Because cpumask_var_t, depending on the CPUMASK_OFFSTACK config, is either a pointer or an array, we have to have a macro for initialization.
So define a CPUMASK_VAR_NULL macro, which allows to init struct cpumask pointer with NULL when CPUMASK_OFFSTACK is enabled, and effectively a no-op when CPUMASK_OFFSTACK is disabled (initialisation optimised out with -O2).
--- include/linux/cpumask.h | 2 ++ 1 file changed, 2 insertions(+)
include/linux/cpumask.h +2 -0
@@ -1005,6 +1005,7 @@ static __always_inline unsigned int cpumask_size(void)#define this_cpu_cpumask_var_ptr (x) this_cpu_read(x)#define __cpumask_var_read_mostly __read_mostly +#define CPUMASK_VAR_NULL NULLbool alloc_cpumask_var_node (cpumask_var_t *mask,gfp_t flags,int node);
@@ -1051,6 +1052,7 @@ static __always_inline bool cpumask_available(cpumask_var_t mask)#define this_cpu_cpumask_var_ptr (x) this_cpu_ptr(x)#define __cpumask_var_read_mostly +#define CPUMASK_VAR_NULL {} static __always_inline bool alloc_cpumask_var(cpumask_var_t *mask, gfp_t flags) {
Signature · 1 line
2.51.1
PATCH v15 7/7 timers: Exclude isolated cpus from timer migration
In reply to Gabriele MonacoThe timer migration mechanism allows active CPUs to pull timers from idle ones to improve the overall idle time. This is however undesired when CPU intensive workloads run on isolated cores, as the algorithm would move the timers from housekeeping to isolated cores, negatively affecting the isolation.
Exclude isolated cores from the timer migration algorithm, extend the concept of unavailable cores, currently used for offline ones, to isolated ones:
* A core is unavailable if isolated or offline;
* A core is available if non isolated and online;
A core is considered unavailable as isolated if it belongs to:
* the isolcpus (domain) list
* an isolated cpuset
Except if it is:
* in the nohz_full list (already idle for the hierarchy)
* the nohz timekeeper core (must be available to handle global timers)
CPUs are added to the hierarchy during late boot, excluding isolated ones, the hierarchy is also adapted when the cpuset isolation changes.
Due to how the timer migration algorithm works, any CPU part of the hierarchy can have their global timers pulled by remote CPUs and have to pull remote timers, only skipping pulling remote timers would break the logic.
For this reason, prevent isolated CPUs from pulling remote global timers, but also the other way around: any global timer started on an isolated CPU will run there. This does not break the concept of isolation (global timers don't come from outside the CPU) and, if considered inappropriate, can usually be mitigated with other isolation techniques (e.g. IRQ pinning).
This effect was noticed on a 128 cores machine running oslat on the isolated cores (1-31,33-63,65-95,97-127). The tool monopolises CPUs, and the CPU with lowest count in a timer migration hierarchy (here 1 and 65) appears as always active and continuously pulls global timers, from the housekeeping CPUs. This ends up moving driver work (e.g. delayed work) to isolated CPUs and causes latency spikes:
before the change:
# oslat -c 1-31,33-63,65-95,97-127 -D 62s ... Maximum: 1203 10 3 4 ... 5 (us)
after the change:
# oslat -c 1-31,33-63,65-95,97-127 -D 62s ... Maximum: 10 4 3 4 3 ... 5 (us)
The same behaviour was observed on a machine with as few as 20 cores / 40 threads with isocpus set to: 1-9,11-39 with rtla-osnoise-top.
--- include/linux/timer.h | 9 ++ kernel/cgroup/cpuset.c | 3 + kernel/time/timer_migration.c | 156 ++++++++++++++++++++++++++++++++-- 3 files changed, 163 insertions(+), 5 deletions(-)
include/linux/timer.h +9 -0
@@ -188,4 +188,13 @@ int timers_dead_cpu(unsigned int cpu);#define timers_dead_cpu NULL#endif +#if defined(CONFIG_SMP ) && defined(CONFIG_NO_HZ_COMMON ) +extern int tmigr_isolated_exclude_cpumask (struct cpumask *exclude_cpumask); +#else +static inline int tmigr_isolated_exclude_cpumask (struct cpumask *exclude_cpumask) +{ +return 0 ; +} +#endif +#endif
kernel/cgroup/cpuset.c +3 -0
@@ -1350,6 +1350,9 @@ static void update_isolation_cpumasks(bool isolcpus_updated) ret =workqueue_unbound_exclude_cpumask (isolated_cpus);WARN_ON_ONCE (ret <0 ); + + ret =tmigr_isolated_exclude_cpumask (isolated_cpus); +WARN_ON_ONCE (ret <0 ); }/**
kernel/time/timer_migration.c +151 -5
215 lines not on this page · raw message
Signature · 1 line
2.51.1
Quoted 16 lines · Gabriele Monaco
kernel/time/timer_migration.c +9
-0+/* Enabled during late initcall */ +static DEFINE_STATIC_KEY_FALSE (tmigr_exclude_isolated); +#define TMIGR_NONE 0xFF#define BIT_CNT 8@@ -438,6 +442,33 @@ static inline bool tmigr_is_not_available(struct tmigr_cpu *tmc)return !(tmc->tmgroup && tmc->available); } +/* +* Returns true if @cpu should be excluded from the hierarchy as isolated. +* Domain isolated CPUs don't participate in timer migration, nohz_full CPUs +* are still part of the hierarchy but become idle (from a tick and timer +* migration perspective) when they stop their tick. This lets the timekeeping +* CPU handle their global timers. Marking also isolated CPUs as idle would be
kernel/time/timer_migration.c static inline bool tmigr_is_not_available(struct tmigr_cpu *tmc) · line 460
Quoted 10 lines
kernel/time/timer_migration.c +10
-0+* too costly, hence they are completely excluded from the hierarchy. +* This check is necessary, for instance, to prevent offline isolated CPUs from +* being incorrectly marked as available once getting back online. +* +* This function returns false during early boot and the isolation logic is +* enabled only after isolated CPUs are marked as unavailable at late boot. +* The tick CPU can be isolated at boot, however we cannot mark it as +* unavailable to avoid having no global migrator for the nohz_full CPUs. This +* should be ensured by the callers of this function: implicitly from hotplug +* callbacs and explicitly in tmigr_init_isolation and
callbacks tmigr_init_isolation()
kernel/time/timer_migration.c static inline bool tmigr_is_not_available(struct tmigr_cpu *tmc) · line 461
Quoted 1 line
kernel/time/timer_migration.c +1
-0+* tmigr_isolated_exclude_cpumask.
tmigr_isolated_exclude_cpumask() It's documented how functions should be denoted in comments and change logs, no?
kernel/time/timer_migration.c static inline bool tmigr_is_not_available(struct tmigr_cpu *tmc) · line 468
Quoted 7 lines
kernel/time/timer_migration.c +7
-0+*/ +static inline bool tmigr_is_isolated (int cpu) +{ +if (static_branch_unlikely (&tmigr_exclude_isolated)) +return (!housekeeping_cpu (cpu,HK_TYPE_DOMAIN ) || +cpuset_cpu_is_isolated (cpu)) && +housekeeping_cpu (cpu,HK_TYPE_KERNEL_NOISE );
Lacks brackets on the if ()
https://www.kernel.org/doc/html/latest/process/maintainer-tip.html#bracket-rules
Also you can make this way more readable by inverting the condition:
if (!static_branch_unlikely(&tmigr_exclude_isolated))
return false;
return .....;No?
kernel/time/timer_migration.c static int tmigr_clear_cpu_available(unsigned int cpu) · line 1442
Quoted 11 lines
kernel/time/timer_migration.c +3
-1+return false; +} +/* * Returns true, when @childmask corresponds to the group migrator or when the * group is not active - so no migrator is set. @@ -1439,8 +1470,9 @@ static int tmigr_clear_cpu_available(unsigned int cpu)int migrator;u64 firstexp; -cpumask_clear_cpu (cpu, tmigr_available_cpumask);
By removing this the function name does not make any sense any more. Splitting the cpumask_clear_set() out, renaming the function
kernel/time/timer_migration.c static int tmigr_clear_cpu_available(unsigned int cpu) · line 1475
Quoted 3 lines
kernel/time/timer_migration.c +2
-0scoped_guard(raw_spinlock_irq, &tmc->lock) { +if (!tmc->available) +return 0 ;
and adding this
kernel/time/timer_migration.c static int tmigr_clear_cpu_available(unsigned int cpu) · line 1488
Quoted 9 lines
kernel/time/timer_migration.c +1
-1tmc->available = false;WRITE_ONCE (tmc->wakeup,KTIME_MAX );@@ -1453,11 +1485,11 @@ static int tmigr_clear_cpu_available(unsigned int cpu) }if (firstexp !=KTIME_MAX ) { - migrator =cpumask_any (tmigr_available_cpumask); + migrator =cpumask_any_but (tmigr_available_cpumask, cpu);
and this should be done in a preparatory patch along with a reasonable explanation in the change log.
kernel/time/timer_migration.c static int tmigr_clear_cpu_available(unsigned int cpu) · line 1492
Quoted 5 lines
kernel/time/timer_migration.c +1
-1work_on_cpu (migrator, tmigr_trigger_active,NULL ); } -return 0 ; +return 1 ;
But thinking more about it. What's the actual point of moving this 'clear' out instead of just moving it further down?
It does not matter at all whether the isol/unisol muck clears an already cleared bit or not. But it would keep the function name comprehensible and avoid all this online/offline wrapper nonsense.
kernel/time/timer_migration.c static int tmigr_set_cpu_available(unsigned int cpu) · line 1471
Quoted 8 lines
kernel/time/timer_migration.c +0
-1}static int tmigr_set_cpu_available (unsigned int cpu)@@ -1468,17 +1500,130 @@ static int tmigr_set_cpu_available(unsigned int cpu)if (WARN_ON_ONCE (!tmc->tmgroup))return -EINVAL ; -cpumask_set_cpu (cpu, tmigr_available_cpumask);
Ditto.
Quoted 4 lines
kernel/time/timer_migration.c +4
-0+static int __tmigr_isolated_exclude_cpumask (struct cpumask *exclude_cpumask) +{ + struct work_struct __percpu *works __free(free_percpu) = + alloc_percpu(struct work_struct);
kernel/time/timer_migration.c static int tmigr_set_cpu_available(unsigned int cpu) · line 1555
Quoted 10 lines
kernel/time/timer_migration.c +10
-0+ cpumask_var_t cpumask_unisol __free(free_cpumask_var) = CPUMASK_VAR_NULL; + cpumask_var_t cpumask_isol __free(free_cpumask_var) = CPUMASK_VAR_NULL; +int cpu; + +if (!alloc_cpumask_var (&cpumask_isol,GFP_KERNEL )) +return -ENOMEM ; +if (!alloc_cpumask_var (&cpumask_unisol,GFP_KERNEL )) +return -ENOMEM ; +if (!works) +return -ENOMEM ;
Checking the first allocation after trying to allocate other stuff makes a lot of sense - NOT.
kernel/time/timer_migration.c static int tmigr_set_cpu_available(unsigned int cpu) · line 1560
Quoted 4 lines
kernel/time/timer_migration.c +4
-0+cpumask_andnot (cpumask_unisol, cpu_online_mask, exclude_cpumask); +cpumask_andnot (cpumask_unisol, cpumask_unisol, tmigr_available_cpumask); +/* Set up the mask earlier to avoid races with the migrator CPU */ +cpumask_or (tmigr_available_cpumask, tmigr_available_cpumask, cpumask_unisol);
Your new line key is broken. This comment is barely noticeable. What's worse is that it completely fails to explain what the actual race is.
Quoted 13 lines
kernel/time/timer_migration.c +13
-0+for_each_cpu (cpu, cpumask_unisol) { +struct work_struct *work =per_cpu_ptr (works, cpu); + +INIT_WORK (work, tmigr_cpu_unisolate); +schedule_work_on (cpu, work); + } + +cpumask_and (cpumask_isol, exclude_cpumask, tmigr_available_cpumask); +cpumask_and (cpumask_isol, cpumask_isol,housekeeping_cpumask (HK_TYPE_KERNEL_NOISE )); +/* +* Handle this here and not in the cpuset code because exclude_cpumask +* might include also the tick CPU if included in isolcpus. +*/
kernel/time/timer_migration.c static int tmigr_set_cpu_available(unsigned int cpu) · line 1583
Quoted 10 lines
kernel/time/timer_migration.c +10
-0+for_each_cpu (cpu, cpumask_isol) { +if (!tick_nohz_cpu_hotpluggable (cpu)) { +cpumask_clear_cpu (cpu, cpumask_isol); +break ; + } + } +/* Set up the mask earlier to avoid races with the migrator CPU */ +cpumask_andnot (tmigr_available_cpumask, tmigr_available_cpumask, cpumask_isol); +for_each_cpu (cpu, cpumask_isol) { +struct work_struct *work =per_cpu_ptr (works, cpu);
This lacks a comment explaining that cpumask_unisol and _isol are not overlapping. I had to stare at this five times to convince myself that it's correct.
Quoted 15 lines
kernel/time/timer_migration.c +12
-0+ +INIT_WORK (work, tmigr_cpu_isolate); +schedule_work_on (cpu, work); + } + +for_each_cpu_or (cpu, cpumask_isol, cpumask_unisol) +flush_work (per_cpu_ptr (works, cpu)); +return 0 ; } +/** +* tmigr_isolated_exclude_cpumask - Exclude given CPUs from hierarchy +* @exclude_cpumask: the cpumask to be excluded from timer migration hierarchy +*
kernel/time/timer_migration.c static int tmigr_set_cpu_available(unsigned int cpu) · line 1608
Quoted 10 lines
kernel/time/timer_migration.c +10
-0+* This function can be called from cpuset code to provide the new set of +* isolated CPUs that should be excluded from the hierarchy. +* Online CPUs not present in exclude_cpumask but already excluded are brought +* back to the hierarchy. +* Functions to isolate/unisolate need to be called locally and can sleep. +*/ +int tmigr_isolated_exclude_cpumask (struct cpumask *exclude_cpumask) +{ +lockdep_assert_cpus_held (); +return __tmigr_isolated_exclude_cpumask (exclude_cpumask);
This wrapper is required because...
Quoted 4 lines
kernel/time/timer_migration.c +4
-0+} + +static int __init tmigr_init_isolation (void ) +{
kernel/time/timer_migration.c static int tmigr_set_cpu_available(unsigned int cpu) · line 1622
Quoted 10 lines
kernel/time/timer_migration.c +10
-0+ cpumask_var_t cpumask __free(free_cpumask_var) = CPUMASK_VAR_NULL; + +static_branch_enable (&tmigr_exclude_isolated); + +if (!housekeeping_enabled (HK_TYPE_DOMAIN )) +return 0 ; +if (!alloc_cpumask_var (&cpumask,GFP_KERNEL )) +return -ENOMEM ; + +cpumask_andnot (cpumask, cpu_possible_mask,housekeeping_cpumask (HK_TYPE_DOMAIN ));
... it would be too sensible to guard this with:
guard(cpus_read_lock)();
for consistency sake _AND_ what's more important to protect it against the RCU torture test code which plays with CPU hotplug starting in device_initcall(), which runs before
kernel/time/timer_migration.c static int __init tmigr_init(void) · line 2026
Quoted 1 line
kernel/time/timer_migration.c +1
-0+late_initcall (tmigr_init_isolation);
Thanks,
tglx
Frederic Weisbecker
maintainerQuoted 34 lines · Thomas Gleixner
kernel/time/timer_migration.c static int tmigr_clear_cpu_available(unsigned int cpu) · line 1442
Quoted 1 line · Gabriele Monaco
kernel/time/timer_migration.c +0
-1-cpumask_clear_cpu (cpu, tmigr_available_cpumask);By removing this the function name does not make any sense any more. Splitting the
cpumask_clear_set()out, renaming the functionkernel/time/timer_migration.c static int tmigr_clear_cpu_available(unsigned int cpu) · line 1475
Quoted 3 lines
kernel/time/timer_migration.c +2
-0scoped_guard(raw_spinlock_irq, &tmc->lock) { +if (!tmc->available) +return 0 ;and adding this
kernel/time/timer_migration.c static int tmigr_clear_cpu_available(unsigned int cpu) · line 1488
Quoted 9 lines
kernel/time/timer_migration.c +1
-1tmc->available = false;WRITE_ONCE (tmc->wakeup,KTIME_MAX );@@ -1453,11 +1485,11 @@ static int tmigr_clear_cpu_available(unsigned int cpu) }if (firstexp !=KTIME_MAX ) { - migrator =cpumask_any (tmigr_available_cpumask); + migrator =cpumask_any_but (tmigr_available_cpumask, cpu);and this should be done in a preparatory patch along with a reasonable explanation in the change log.
kernel/time/timer_migration.c static int tmigr_clear_cpu_available(unsigned int cpu) · line 1492
Quoted 5 lines
kernel/time/timer_migration.c +1
-1work_on_cpu (migrator, tmigr_trigger_active,NULL ); } -return 0 ; +return 1 ;But thinking more about it. What's the actual point of moving this 'clear' out instead of just moving it further down?
Quoted 3 lines
It does not matter at all whether the isol/unisol muck clears an already cleared bit or not. But it would keep the function name comprehensible and avoid all this online/offline wrapper nonsense.
That was my suggestion.
It's because tmigr_clear_cpu_available() and tmigr_set_cpu_available() can now all be called concurrently through the workqueues and race and mess up the cpumask if they all try to clear/set at the same time...
And I couldn't find a saner way to order things...
Thanks.
Signature · 2 lines
Frederic Weisbecker SUSE Labs
Quoted 8 lines · Frederic Weisbecker
Quoted 6 lines · Thomas Gleixner
But thinking more about it. What's the actual point of moving this 'clear' out instead of just moving it further down?
It does not matter at all whether the isol/unisol muck clears an already cleared bit or not. But it would keep the function name comprehensible and avoid all this online/offline wrapper nonsense.
Quoted 5 lines
That was my suggestion.
It's because
tmigr_clear_cpu_available()andtmigr_set_cpu_available()can now all be called concurrently through the workqueues and race and mess up the cpumask if they all try to clear/set at the same time...
Huch?
cpumask_set_cpu() uses set_bit() and cpumask_clear_cpu() uses clear_bit(). Both are atomic and nothing gets messed up.
The only undefined case would be if you end up setting/clearing the same bit, which would require that the unisol and isol maps overlap. But that would be a bug on it's own, no?
Thanks,
tglx
Frederic Weisbecker
maintainerQuoted 11 lines · Thomas Gleixner
Quoted 10 lines · Frederic Weisbecker
Quoted 6 lines · Thomas Gleixner
But thinking more about it. What's the actual point of moving this 'clear' out instead of just moving it further down?
It does not matter at all whether the isol/unisol muck clears an already cleared bit or not. But it would keep the function name comprehensible and avoid all this online/offline wrapper nonsense.
That was my suggestion.
Quoted 8 lines
Quoted 3 lines
It's because
tmigr_clear_cpu_available()andtmigr_set_cpu_available()can now all be called concurrently through the workqueues and race and mess up the cpumask if they all try to clear/set at the same time...Huch?
cpumask_set_cpu()usesset_bit()andcpumask_clear_cpu()usesclear_bit(). Both are atomic and nothing gets messed up.
Urgh, right...
Quoted 4 lines
The only undefined case would be if you end up setting/clearing the same bit, which would require that the unisol and isol maps overlap. But that would be a bug on it's own, no?
Ok.
But then the "unisolate" works must be flushed before this line:
+ /* Set up the mask earlier to avoid races with the migrator CPU */
+ cpumask_andnot(tmigr_available_cpumask, tmigr_available_cpumask, cpumask_isol);
Because that is non-atomic and can race with the cpumask_set_cpu() from the works, right?
Thanks.
Signature · 2 lines
Frederic Weisbecker SUSE Labs
Quoted 35 lines · Frederic Weisbecker
Quoted 11 lines · Thomas Gleixner
Quoted 10 lines · Frederic Weisbecker
Quoted 6 lines · Thomas Gleixner
But thinking more about it. What's the actual point of moving this 'clear' out instead of just moving it further down?
It does not matter at all whether the isol/unisol muck clears an already cleared bit or not. But it would keep the function name comprehensible and avoid all this online/offline wrapper nonsense.
That was my suggestion.
Quoted 8 lines
Quoted 3 lines
It's because
tmigr_clear_cpu_available()andtmigr_set_cpu_available()can now all be called concurrently through the workqueues and race and mess up the cpumask if they all try to clear/set at the same time...Huch?
cpumask_set_cpu()usesset_bit()andcpumask_clear_cpu()usesclear_bit(). Both are atomic and nothing gets messed up.Urgh, right...
Quoted 4 lines
The only undefined case would be if you end up setting/clearing the same bit, which would require that the unisol and isol maps overlap. But that would be a bug on it's own, no?
Ok.
But then the "unisolate" works must be flushed before this line:
(quoted diff) +2
-0+ /* Set up the mask earlier to avoid races with the migrator CPU */ + cpumask_andnot(tmigr_available_cpumask, tmigr_available_cpumask, cpumask_isol);
Quoted 2 lines
Because that is non-atomic and can race with the
cpumask_set_cpu()from the works, right?
Correct, but I still have to understand why this has to happen upfront. As I said before this comment is useless:
Quoted 1 line
+ /* Set up the mask earlier to avoid races with the migrator CPU */
which decodes to:
Set up the mask earlier than something unspecified to avoid unspecified races with the migrator CPU.
Seriously?
But I finally understood what this is about after staring at it with less disgust again for five times in a row:
tmigr_clear_cpu_available() requires a stable cpumask to find a migrator.
So what this is about is to avoid touching the cpumask in those worker functions so that tmigr_isolated_exclude_cpumask() can fiddle with them asynchronously.
Right?
That's voodoo programming which nobody - even those involved - will understand anymore three months down the road.
The idea that updating the masks upfront will provide stable state is flawed to begin with. Let's assume the following scenario:
1) The isol/unisol mask is just flipped around
2) The newly available CPUs are marked in the mask
3) The work is scheduled on those CPUs
4) The newly unavailable CPUs are cleared in the mask
5) The work is scheduled on those CPUs
6) The newly available CPU workers are delayed so that the newly unavailable workers get there first. They find a stable CPU mask, but none of these now "available" CPUs are actually brought into active state yet.
That's just a perfect recipe for some completely undecodable bug to happen anytime soon even it it supposed to "work" by chance.
The way how this was designed in the first place is that changing the "availability" is fully serialized by the CPU hotplug machinery. Which means zero surprises.
Now you create a side channel, which lacks these serialization guarantees for absolutely no reason. Changing this isolation muck at run-time is not a hotpath operation and trying to optimize it for no good reason is just stupid.
The most trivial solution is to schedule the works one by one walking through the newly available and then through the unavailable maps.
If you want to be a bit smarter, then you can just use a global mutex, which is taken inside the set/clear_available() functions which serializes the related functionality and bulk schedule/flush the unisol (make available) first and then proceed to the isol (make unavailable).
Then nothing has to change vs. the set/clear operations and everything just works.
That mutex does not do any harm in the CPU hotplug case and the serialization vs. the workers is not going to be the end of the world.
I'm willing to bet that no real-world use-case will ever notice the existance of this mutex. The microbenchmark which shows off the "I'm so smart" metric is completely irrelevant especially when the result is fragile, incomprehensible and therefore unmaintainable.
Keep it correct and simple is still the most important engineering principle. Premature optimization is a guaranteed path to failure.
If there is a compelling use case which justifies the resulting complexity, then it can be built on top. I'm not holding my breath. See above...
Thanks,
tglx
Frederic Weisbecker
maintainerQuoted 79 lines · Thomas Gleixner
Quoted 35 lines · Frederic Weisbecker
Quoted 11 lines · Thomas Gleixner
Quoted 10 lines · Frederic Weisbecker
Quoted 6 lines · Thomas Gleixner
But thinking more about it. What's the actual point of moving this 'clear' out instead of just moving it further down?
It does not matter at all whether the isol/unisol muck clears an already cleared bit or not. But it would keep the function name comprehensible and avoid all this online/offline wrapper nonsense.
That was my suggestion.
Quoted 8 lines
Quoted 3 lines
It's because
tmigr_clear_cpu_available()andtmigr_set_cpu_available()can now all be called concurrently through the workqueues and race and mess up the cpumask if they all try to clear/set at the same time...Huch?
cpumask_set_cpu()usesset_bit()andcpumask_clear_cpu()usesclear_bit(). Both are atomic and nothing gets messed up.Urgh, right...
Quoted 4 lines
The only undefined case would be if you end up setting/clearing the same bit, which would require that the unisol and isol maps overlap. But that would be a bug on it's own, no?
Ok.
But then the "unisolate" works must be flushed before this line:
(quoted diff) +2
-0+ /* Set up the mask earlier to avoid races with the migrator CPU */ + cpumask_andnot(tmigr_available_cpumask, tmigr_available_cpumask, cpumask_isol);Quoted 2 lines
Because that is non-atomic and can race with the
cpumask_set_cpu()from the works, right?Correct, but I still have to understand why this has to happen upfront. As I said before this comment is useless:
Quoted 1 line
+ /* Set up the mask earlier to avoid races with the migrator CPU */
which decodes to:
Set up the mask earlier than something unspecified to avoid unspecified races with the migrator CPU.
Seriously?
But I finally understood what this is about after staring at it with less disgust again for five times in a row:
tmigr_clear_cpu_available()requires a stable cpumask to find a migrator.So what this is about is to avoid touching the cpumask in those worker functions so that
tmigr_isolated_exclude_cpumask()can fiddle with them asynchronously.Right?
That's voodoo programming which nobody - even those involved - will understand anymore three months down the road.
The idea that updating the masks upfront will provide stable state is flawed to begin with. Let's assume the following scenario:
1) The isol/unisol mask is just flipped around
2) The newly available CPUs are marked in the mask
3) The work is scheduled on those CPUs
4) The newly unavailable CPUs are cleared in the mask
5) The work is scheduled on those CPUs
Quoted 7 lines
6) The newly available CPU workers are delayed so that the newly unavailable workers get there first. They find a stable CPU mask, but none of these now "available" CPUs are actually brought into active state yet.
That's just a perfect recipe for some completely undecodable bug to happen anytime soon even it it supposed to "work" by chance.
Urgh, indeed I completely missed that...
Quoted 5 lines
The way how this was designed in the first place is that changing the "availability" is fully serialized by the CPU hotplug machinery. Which means zero surprises.
Quoted 7 lines
Now you create a side channel, which lacks these serialization guarantees for absolutely no reason. Changing this isolation muck at run-time is not a hotpath operation and trying to optimize it for no good reason is just stupid.
The most trivial solution is to schedule the works one by one walking through the newly available and then through the unavailable maps.
Right.
Quoted 17 lines
If you want to be a bit smarter, then you can just use a global mutex, which is taken inside the set/clear_available() functions which serializes the related functionality and bulk schedule/flush the unisol (make available) first and then proceed to the isol (make unavailable).
Then nothing has to change vs. the set/clear operations and everything just works.
That mutex does not do any harm in the CPU hotplug case and the serialization vs. the workers is not going to be the end of the world.
I'm willing to bet that no real-world use-case will ever notice the existance of this mutex. The microbenchmark which shows off the "I'm so smart" metric is completely irrelevant especially when the result is fragile, incomprehensible and therefore unmaintainable.
Quoted 6 lines
Keep it correct and simple is still the most important engineering principle. Premature optimization is a guaranteed path to failure.
If there is a compelling use case which justifies the resulting complexity, then it can be built on top. I'm not holding my breath. See above...
Perhaps the only thing that worries me is if an isolated partition is inverted. Say 0-3 is non isolated and 4-7 is isolated. And then cpuset is overwritten so that the reverse is applied: 0-3 is isolated and 4-7 is not isolated. If all isol works reach before unisol works, then tmigr_clear_cpu_available() -> cpumask_any(tmigr_available_mask) won't find any CPU left on the last call.
Now last time I tried to invert an isolated cpumask in cpuset, I got -EINVAL so something must be preventing from that. Just in case let's have a WARN_ON_ONCE(cpumask_any(tmigr_available_mask) >= nr_cpu_ids).
But other than this detail, your solution looks good!
Thanks.
Signature · 2 lines
Frederic Weisbecker SUSE Labs
Quoted 24 lines · Frederic Weisbecker
Quoted 16 lines · Thomas Gleixner
If you want to be a bit smarter, then you can just use a global mutex, which is taken inside the set/clear_available() functions which serializes the related functionality and bulk schedule/flush the unisol (make available) first and then proceed to the isol (make unavailable).
Then nothing has to change vs. the set/clear operations and everything just works.
That mutex does not do any harm in the CPU hotplug case and the serialization vs. the workers is not going to be the end of the world.
I'm willing to bet that no real-world use-case will ever notice the existance of this mutex. The microbenchmark which shows off the "I'm so smart" metric is completely irrelevant especially when the result is fragile, incomprehensible and therefore unmaintainable.
Quoted 6 lines
Keep it correct and simple is still the most important engineering principle. Premature optimization is a guaranteed path to failure.
If there is a compelling use case which justifies the resulting complexity, then it can be built on top. I'm not holding my breath. See above...
Quoted 6 lines
Perhaps the only thing that worries me is if an isolated partition is inverted. Say 0-3 is non isolated and 4-7 is isolated. And then cpuset is overwritten so that the reverse is applied: 0-3 is isolated and 4-7 is not isolated. If all isol works reach before unisol works, then
tmigr_clear_cpu_available()->cpumask_any(tmigr_available_mask)won't find any CPU left on the last call.
schedule the newly available (now unisolated) ones first and flush that work. After that you can safely mark the others unavailable, no?
Thanks
tglx
Frederic Weisbecker
maintainerQuoted 32 lines · Thomas Gleixner
Quoted 24 lines · Frederic Weisbecker
Quoted 16 lines · Thomas Gleixner
If you want to be a bit smarter, then you can just use a global mutex, which is taken inside the set/clear_available() functions which serializes the related functionality and bulk schedule/flush the unisol (make available) first and then proceed to the isol (make unavailable).
Then nothing has to change vs. the set/clear operations and everything just works.
That mutex does not do any harm in the CPU hotplug case and the serialization vs. the workers is not going to be the end of the world.
I'm willing to bet that no real-world use-case will ever notice the existance of this mutex. The microbenchmark which shows off the "I'm so smart" metric is completely irrelevant especially when the result is fragile, incomprehensible and therefore unmaintainable.
Quoted 6 lines
Keep it correct and simple is still the most important engineering principle. Premature optimization is a guaranteed path to failure.
If there is a compelling use case which justifies the resulting complexity, then it can be built on top. I'm not holding my breath. See above...
Quoted 6 lines
Perhaps the only thing that worries me is if an isolated partition is inverted. Say 0-3 is non isolated and 4-7 is isolated. And then cpuset is overwritten so that the reverse is applied: 0-3 is isolated and 4-7 is not isolated. If all isol works reach before unisol works, then
tmigr_clear_cpu_available()->cpumask_any(tmigr_available_mask)won't find any CPU left on the last call.
Quoted 2 lines
schedule the newly available (now unisolated) ones first and flush that work. After that you can safely mark the others unavailable, no?
Yes!
Signature · 2 lines
Frederic Weisbecker SUSE Labs
Quoted 219 lines · Gabriele Monaco
The timer migration mechanism allows active CPUs to pull timers from idle ones to improve the overall idle time. This is however undesired when CPU intensive workloads run on isolated cores, as the algorithm would move the timers from housekeeping to isolated cores, negatively affecting the isolation.
Exclude isolated cores from the timer migration algorithm, extend the concept of unavailable cores, currently used for offline ones, to isolated ones:
* A core is unavailable if isolated or offline;
* A core is available if non isolated and online;A core is considered unavailable as isolated if it belongs to:
* the isolcpus (domain) list
* an isolated cpuset
Except if it is:
* in thenohz_fulllist (already idle for the hierarchy)
* the nohz timekeeper core (must be available to handle global timers)CPUs are added to the hierarchy during late boot, excluding isolated ones, the hierarchy is also adapted when the cpuset isolation changes.
Due to how the timer migration algorithm works, any CPU part of the hierarchy can have their global timers pulled by remote CPUs and have to pull remote timers, only skipping pulling remote timers would break the logic.
For this reason, prevent isolated CPUs from pulling remote global timers, but also the other way around: any global timer started on an isolated CPU will run there. This does not break the concept of isolation (global timers don't come from outside the CPU) and, if considered inappropriate, can usually be mitigated with other isolation techniques (e.g. IRQ pinning).This effect was noticed on a 128 cores machine running oslat on the isolated cores (1-31,33-63,65-95,97-127). The tool monopolises CPUs, and the CPU with lowest count in a timer migration hierarchy (here 1 and 65) appears as always active and continuously pulls global timers, from the housekeeping CPUs. This ends up moving driver work (e.g. delayed work) to isolated CPUs and causes latency spikes:
before the change:
# oslat -c 1-31,33-63,65-95,97-127 -D 62s ... Maximum: 1203 10 3 4 ... 5 (us)after the change:
# oslat -c 1-31,33-63,65-95,97-127 -D 62s ... Maximum: 10 4 3 4 3 ... 5 (us)The same behaviour was observed on a machine with as few as 20 cores / 40 threads with isocpus set to: 1-9,11-39 with rtla-osnoise-top.
Tested-by John B. Wyatt IV <[email protected]>Tested-by John B. Wyatt IV <[email protected]>Reviewed-by Frederic Weisbecker <[email protected]>Signed-off-by Gabriele Monaco <[email protected]>--- include/linux/timer.h | 9 ++ kernel/cgroup/cpuset.c | 3 + kernel/time/timer_migration.c | 156 ++++++++++++++++++++++++++++++++-- 3 files changed, 163 insertions(+), 5 deletions(-)include/linux/timer.h +9
-0@@ -188,4 +188,13 @@ int timers_dead_cpu(unsigned int cpu);#define timers_dead_cpu NULL#endif +#if defined(CONFIG_SMP ) && defined(CONFIG_NO_HZ_COMMON ) +extern int tmigr_isolated_exclude_cpumask (struct cpumask *exclude_cpumask); +#else +static inline int tmigr_isolated_exclude_cpumask (struct cpumask *exclude_cpumask) +{ +return 0 ; +} +#endif +#endif kernel/cgroup/cpuset.c +3
-0@@ -1350,6 +1350,9 @@ static void update_isolation_cpumasks(bool isolcpus_updated) ret =workqueue_unbound_exclude_cpumask (isolated_cpus);WARN_ON_ONCE (ret <0 ); + + ret =tmigr_isolated_exclude_cpumask (isolated_cpus); +WARN_ON_ONCE (ret <0 ); }/** kernel/time/timer_migration.c +64
-4@@ -10,6 +10,7 @@#include <linux/spinlock.h> #include <linux/timerqueue.h> #include <trace/events/ipi.h> +#include <linux/sched/isolation.h> #include "timer_migration.h" #include "tick-internal.h" @@ -430,6 +431,9 @@ static DEFINE_PER_CPU(struct tmigr_cpu, tmigr_cpu);*/ static cpumask_var_t tmigr_available_cpumask; +/* Enabled during late initcall */ +static DEFINE_STATIC_KEY_FALSE (tmigr_exclude_isolated); +#define TMIGR_NONE 0xFF#define BIT_CNT 8@@ -438,6 +442,33 @@ static inline bool tmigr_is_not_available(struct tmigr_cpu *tmc)return !(tmc->tmgroup && tmc->available); } +/* +* Returns true if @cpu should be excluded from the hierarchy as isolated. +* Domain isolated CPUs don't participate in timer migration, nohz_full CPUs +* are still part of the hierarchy but become idle (from a tick and timer +* migration perspective) when they stop their tick. This lets the timekeeping +* CPU handle their global timers. Marking also isolated CPUs as idle would be +* too costly, hence they are completely excluded from the hierarchy. +* This check is necessary, for instance, to prevent offline isolated CPUs from +* being incorrectly marked as available once getting back online. +* +* This function returns false during early boot and the isolation logic is +* enabled only after isolated CPUs are marked as unavailable at late boot. +* The tick CPU can be isolated at boot, however we cannot mark it as +* unavailable to avoid having no global migrator for the nohz_full CPUs. This +* should be ensured by the callers of this function: implicitly from hotplug +* callbacs and explicitly in tmigr_init_isolation and +* tmigr_isolated_exclude_cpumask. +*/ +static inline bool tmigr_is_isolated (int cpu) +{ +if (static_branch_unlikely (&tmigr_exclude_isolated)) +return (!housekeeping_cpu (cpu,HK_TYPE_DOMAIN ) || +cpuset_cpu_is_isolated (cpu)) && +housekeeping_cpu (cpu,HK_TYPE_KERNEL_NOISE ); +return false; +} +/* * Returns true, when @childmask corresponds to the group migrator or when the * group is not active - so no migrator is set. @@ -1439,8 +1470,9 @@ static int tmigr_clear_cpu_available(unsigned int cpu)int migrator;u64 firstexp; -cpumask_clear_cpu (cpu, tmigr_available_cpumask); scoped_guard(raw_spinlock_irq, &tmc->lock) { +if (!tmc->available) +return 0 ; tmc->available = false;WRITE_ONCE (tmc->wakeup,KTIME_MAX );@@ -1453,11 +1485,11 @@ static int tmigr_clear_cpu_available(unsigned int cpu) }if (firstexp !=KTIME_MAX ) { - migrator =cpumask_any (tmigr_available_cpumask); + migrator =cpumask_any_but (tmigr_available_cpumask, cpu);work_on_cpu (migrator, tmigr_trigger_active,NULL ); } -return 0 ; +return 1 ; }static int tmigr_set_cpu_available (unsigned int cpu)@@ -1468,17 +1500,130 @@ static int tmigr_set_cpu_available(unsigned int cpu)if (WARN_ON_ONCE (!tmc->tmgroup))return -EINVAL ; -cpumask_set_cpu (cpu, tmigr_available_cpumask); +if (tmigr_is_isolated (cpu)) +return 0 ; + scoped_guard(raw_spinlock_irq, &tmc->lock) { +if (tmc->available) +return 0 ;trace_tmigr_cpu_available (tmc); tmc->idle =timer_base_is_idle ();if (!tmc->idle)__tmigr_cpu_activate (tmc); tmc->available = true; } +return 1 ; +} + +static int tmigr_online_cpu (unsigned int cpu) +{ +if (tmigr_set_cpu_available (cpu) >0 ) +cpumask_set_cpu (cpu, tmigr_available_cpumask); +return 0 ; +} + +static int tmigr_offline_cpu (unsigned int cpu) +{ +if (tmigr_clear_cpu_available (cpu) >0 ) +cpumask_clear_cpu (cpu, tmigr_available_cpumask); +return 0 ; +} + +static void tmigr_cpu_isolate (struct work_struct *ignored) +{ +tmigr_clear_cpu_available (smp_processor_id ()); +} + +static void tmigr_cpu_unisolate (struct work_struct *ignored) +{
kernel/time/timer_migration.c static int tmigr_set_cpu_available(unsigned int cpu) · line 1548
Quoted 10 lines
kernel/time/timer_migration.c +10
-0+tmigr_set_cpu_available (smp_processor_id ()); +} + +static int __tmigr_isolated_exclude_cpumask (struct cpumask *exclude_cpumask) +{ + struct work_struct __percpu *works __free(free_percpu) = + alloc_percpu(struct work_struct); + cpumask_var_t cpumask_unisol __free(free_cpumask_var) = CPUMASK_VAR_NULL; + cpumask_var_t cpumask_isol __free(free_cpumask_var) = CPUMASK_VAR_NULL; +int cpu;
There are currently only 2 callers for this function - from late_init call and from cpuset. Concurrent call is not possible. Maybe we can just pre-allocate these cpumask_var_t and percpu work structures once and reuse it instead of doing an allocation and free each time it is called. The pre-allocation can be done in tmigr_init_isolation().
Quoted 67 lines
kernel/time/timer_migration.c +64
-0+ +if (!alloc_cpumask_var (&cpumask_isol,GFP_KERNEL )) +return -ENOMEM ; +if (!alloc_cpumask_var (&cpumask_unisol,GFP_KERNEL )) +return -ENOMEM ; +if (!works) +return -ENOMEM ; + +cpumask_andnot (cpumask_unisol, cpu_online_mask, exclude_cpumask); +cpumask_andnot (cpumask_unisol, cpumask_unisol, tmigr_available_cpumask); +/* Set up the mask earlier to avoid races with the migrator CPU */ +cpumask_or (tmigr_available_cpumask, tmigr_available_cpumask, cpumask_unisol); +for_each_cpu (cpu, cpumask_unisol) { +struct work_struct *work =per_cpu_ptr (works, cpu); + +INIT_WORK (work, tmigr_cpu_unisolate); +schedule_work_on (cpu, work); + } + +cpumask_and (cpumask_isol, exclude_cpumask, tmigr_available_cpumask); +cpumask_and (cpumask_isol, cpumask_isol,housekeeping_cpumask (HK_TYPE_KERNEL_NOISE )); +/* +* Handle this here and not in the cpuset code because exclude_cpumask +* might include also the tick CPU if included in isolcpus. +*/ +for_each_cpu (cpu, cpumask_isol) { +if (!tick_nohz_cpu_hotpluggable (cpu)) { +cpumask_clear_cpu (cpu, cpumask_isol); +break ; + } + } +/* Set up the mask earlier to avoid races with the migrator CPU */ +cpumask_andnot (tmigr_available_cpumask, tmigr_available_cpumask, cpumask_isol); +for_each_cpu (cpu, cpumask_isol) { +struct work_struct *work =per_cpu_ptr (works, cpu); + +INIT_WORK (work, tmigr_cpu_isolate); +schedule_work_on (cpu, work); + } + +for_each_cpu_or (cpu, cpumask_isol, cpumask_unisol) +flush_work (per_cpu_ptr (works, cpu)); +return 0 ; } +/** +* tmigr_isolated_exclude_cpumask - Exclude given CPUs from hierarchy +* @exclude_cpumask: the cpumask to be excluded from timer migration hierarchy +* +* This function can be called from cpuset code to provide the new set of +* isolated CPUs that should be excluded from the hierarchy. +* Online CPUs not present in exclude_cpumask but already excluded are brought +* back to the hierarchy. +* Functions to isolate/unisolate need to be called locally and can sleep. +*/ +int tmigr_isolated_exclude_cpumask (struct cpumask *exclude_cpumask) +{ +lockdep_assert_cpus_held (); +return __tmigr_isolated_exclude_cpumask (exclude_cpumask); +} + +static int __init tmigr_init_isolation (void ) +{ + cpumask_var_t cpumask __free(free_cpumask_var) = CPUMASK_VAR_NULL; + +static_branch_enable (&tmigr_exclude_isolated);
kernel/time/timer_migration.c static int tmigr_set_cpu_available(unsigned int cpu) · line 1625
Quoted 10 lines
kernel/time/timer_migration.c +10
-0+ +if (!housekeeping_enabled (HK_TYPE_DOMAIN )) +return 0 ; +if (!alloc_cpumask_var (&cpumask,GFP_KERNEL )) +return -ENOMEM ; + +cpumask_andnot (cpumask, cpu_possible_mask,housekeeping_cpumask (HK_TYPE_DOMAIN )); + +return __tmigr_isolated_exclude_cpumask (cpumask); +}
Should we put all these functions under "#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)" like in the timer.h header file?
Cheers,
Longman
Quoted 18 lines
kernel/time/timer_migration.c +3
-1+static void tmigr_init_group (struct tmigr_group *group,unsigned int lvl,int node) {@@ -1867,7 +2012,7 @@ static int __init tmigr_init(void) goto err; ret =cpuhp_setup_state (CPUHP_AP_TMIGR_ONLINE ,"tmigr:online" , - tmigr_set_cpu_available, tmigr_clear_cpu_available); + tmigr_online_cpu, tmigr_offline_cpu);if (ret) goto err;@@ -1878,3 +2023,4 @@ static int __init tmigr_init(void)return ret; }early_initcall (tmigr_init); +late_initcall (tmigr_init_isolation);
Quoted 19 lines · Waiman Long
kernel/time/timer_migration.c static int tmigr_set_cpu_available(unsigned int cpu) · line 1548
Quoted 11 lines · Gabriele Monaco
+static int
__tmigr_isolated_exclude_cpumask(struct cpumask *exclude_cpumask)kernel/time/timer_migration.c +4
-0+{ + struct work_struct __percpu *works __free(free_percpu) = + alloc_percpu(struct work_struct); + cpumask_var_t cpumask_unisol __free(free_cpumask_var) =
CPUMASK_VAR_NULL;(quoted diff) +1
-0+ cpumask_var_t cpumask_isol __free(free_cpumask_var) =
CPUMASK_VAR_NULL;kernel/time/timer_migration.c +1
-0+int cpu;There are currently only 2 callers for this function - from
late_initcall and from cpuset. Concurrent call is not possible. Maybe we can just pre-allocate thesecpumask_var_tand percpu work structures once and reuse it instead of doing an allocation and free each time it is called. The pre-allocation can be done intmigr_init_isolation().
I have no strong opinion on this, but after changes suggested by Thomas it gets superfluous to allocate 2 cpumasks (after flushing what is now cpumask_unisol it's no longer needed and we can re-use it).
Considering this only runs at boot and every time a cpuset changes isolation, is it worth the extra steps to pre-allocate?
Quoted 14 lines
Quoted 14 lines
+/**
+ *tmigr_isolated_exclude_cpumask- Exclude given CPUs from hierarchy
+ * @exclude_cpumask: the cpumask to be excluded from timer migration hierarchy
+ *
+ * This function can be called from cpuset code to provide the new set of
+ * isolated CPUs that should be excluded from the hierarchy.
+ * Online CPUs not present inexclude_cpumaskbut already excluded are brought
+ * back to the hierarchy.
+ * Functions to isolate/unisolate need to be called locally and can sleep.
+ */
+inttmigr_isolated_exclude_cpumask(struct cpumask *exclude_cpumask)
Quoted 8 lines
kernel/time/timer_migration.c static int tmigr_set_cpu_available(unsigned int cpu) · line 1609
Quoted 5 lines
kernel/time/timer_migration.c +4
-0+{ +lockdep_assert_cpus_held (); +return __tmigr_isolated_exclude_cpumask (exclude_cpumask); +}Should we put all these functions under "#if
defined(CONFIG_SMP)&&defined(CONFIG_NO_HZ_COMMON)" like in thetimer.hheader file?
I think that's implied in the build condition of timer_migration.o
https://elixir.bootlin.com/linux/v6.17.8/source/kernel/time/Makefile#L27
At least I got these ifdefs from timer_migration.h and none of those functions are ifdeffed in timer_migration.c
Thanks,
Gabriele
Quoted 18 lines · Gabriele Monaco
Quoted 17 lines · Waiman Long
kernel/time/timer_migration.c static int tmigr_set_cpu_available(unsigned int cpu) · line 1548
Quoted 10 lines · Gabriele Monaco
+static int
__tmigr_isolated_exclude_cpumask(struct cpumask *exclude_cpumask)kernel/time/timer_migration.c +4
-0+{ + struct work_struct __percpu *works __free(free_percpu) = + alloc_percpu(struct work_struct); + cpumask_var_t cpumask_unisol __free(free_cpumask_var) =
CPUMASK_VAR_NULL;(quoted diff) +1
-0+ cpumask_var_t cpumask_isol __free(free_cpumask_var) =
CPUMASK_VAR_NULL;kernel/time/timer_migration.c +1
-0+int cpu;There are currently only 2 callers for this function - from
late_initcall and from cpuset. Concurrent call is not possible. Maybe we can just pre-allocate thesecpumask_var_tand percpu work structures once and reuse it instead of doing an allocation and free each time it is called. The pre-allocation can be done intmigr_init_isolation().
Quoted 6 lines
I have no strong opinion on this, but after changes suggested by Thomas it gets superfluous to allocate 2 cpumasks (after flushing what is now
cpumask_unisolit's no longer needed and we can re-use it).Considering this only runs at boot and every time a cpuset changes isolation, is it worth the extra steps to pre-allocate?
It is just a suggestion. We can see how it goes and decide if this change is needed or not.
Quoted 22 lines
Quoted 14 lines
Quoted 14 lines
+/**
+ *tmigr_isolated_exclude_cpumask- Exclude given CPUs from hierarchy
+ * @exclude_cpumask: the cpumask to be excluded from timer migration hierarchy
+ *
+ * This function can be called from cpuset code to provide the new set of
+ * isolated CPUs that should be excluded from the hierarchy.
+ * Online CPUs not present inexclude_cpumaskbut already excluded are brought
+ * back to the hierarchy.
+ * Functions to isolate/unisolate need to be called locally and can sleep.
+ */Quoted 8 lines
kernel/time/timer_migration.c static int tmigr_set_cpu_available(unsigned int cpu) · line 1609
Quoted 6 lines
kernel/time/timer_migration.c +5
-0+int tmigr_isolated_exclude_cpumask (struct cpumask *exclude_cpumask) +{ +lockdep_assert_cpus_held (); +return __tmigr_isolated_exclude_cpumask (exclude_cpumask); +}Should we put all these functions under "#if
defined(CONFIG_SMP)&&defined(CONFIG_NO_HZ_COMMON)" like in thetimer.hheader file?
Quoted 5 lines
I think that's implied in the build condition of timer_migration.o
https://elixir.bootlin.com/linux/v6.17.8/source/kernel/time/Makefile#L27At least I got these ifdefs from
timer_migration.hand none of those functions are ifdeffed intimer_migration.c
You are right. I haven't check the condition for building timer_migration.o.
Cheers,
Longman
Frederic Weisbecker
maintainerQuoted 175 lines · Gabriele Monaco
The timer migration mechanism allows active CPUs to pull timers from idle ones to improve the overall idle time. This is however undesired when CPU intensive workloads run on isolated cores, as the algorithm would move the timers from housekeeping to isolated cores, negatively affecting the isolation.
Exclude isolated cores from the timer migration algorithm, extend the concept of unavailable cores, currently used for offline ones, to isolated ones:
* A core is unavailable if isolated or offline;
* A core is available if non isolated and online;A core is considered unavailable as isolated if it belongs to:
* the isolcpus (domain) list
* an isolated cpuset
Except if it is:
* in thenohz_fulllist (already idle for the hierarchy)
* the nohz timekeeper core (must be available to handle global timers)CPUs are added to the hierarchy during late boot, excluding isolated ones, the hierarchy is also adapted when the cpuset isolation changes.
Due to how the timer migration algorithm works, any CPU part of the hierarchy can have their global timers pulled by remote CPUs and have to pull remote timers, only skipping pulling remote timers would break the logic.
For this reason, prevent isolated CPUs from pulling remote global timers, but also the other way around: any global timer started on an isolated CPU will run there. This does not break the concept of isolation (global timers don't come from outside the CPU) and, if considered inappropriate, can usually be mitigated with other isolation techniques (e.g. IRQ pinning).This effect was noticed on a 128 cores machine running oslat on the isolated cores (1-31,33-63,65-95,97-127). The tool monopolises CPUs, and the CPU with lowest count in a timer migration hierarchy (here 1 and 65) appears as always active and continuously pulls global timers, from the housekeeping CPUs. This ends up moving driver work (e.g. delayed work) to isolated CPUs and causes latency spikes:
before the change:
# oslat -c 1-31,33-63,65-95,97-127 -D 62s ... Maximum: 1203 10 3 4 ... 5 (us)after the change:
# oslat -c 1-31,33-63,65-95,97-127 -D 62s ... Maximum: 10 4 3 4 3 ... 5 (us)The same behaviour was observed on a machine with as few as 20 cores / 40 threads with isocpus set to: 1-9,11-39 with rtla-osnoise-top.
The first 4 patches are preparatory work to change the concept of online/offline to available/unavailable, keep track of those in a separate cpumask cleanup the setting/clearing functions and change a function name in cpuset code.
Patch 5 adapt isolation to prevent domain isolated and
nohz_fullfrom covering all CPUs not leaving any housekeeping one. This can lead to problems with the changes introduced in this series because no CPU would remain to handle global timers.
(The corresponding change for cpuset was removed from this version of the series and is present in [2]).Patch 7 extends the unavailable status to domain isolated CPUs, which is the main contribution of the series.
Changes since v14:
* Rebase on tip/timers/core, rename one more ->online field
* Mark the static key as static
* Share code betweentmigr_init_isolationandtmigr_isolated_exclude_cpumaskChanges since v13:
* Remove tmigr late initialisation and restore late isolation (as in v8 [1])
* Use workqueues in initialisation just liketmigr_available_cpumask()
* Use static key fortmigr_exclude_isolated
* Remove cpuset patch checking for HK conflict (included in [2])
* Rename cpuset helper toupdate_isolation_cpumasksas in [2]Changes since v12:
* Pick and adapt patch by Yury Norov to initialise cpumasks
* Reorganise accesses totmigr_available_cpumaskto avoid racesChanges since v11:
* Renameisolcpus_nohz_conflict()toisolated_cpus_can_update()
* Movetick_nohz_cpu_hotpluggable()check totmigr_is_isolated()
* Use workqueues intmigr_isolated_exclude_cpumask()to avoid sleeping while atomic
* Add cpumask initialiser to safely use cpumask cleanup helpersChanges since v10:
* Simplify housekeeping conflict condition
* Reword commit (Frederic Weisbecker)Changes since v9:
* Fix total housekeeping enforcement to focus only on nohz and domain
* Avoid out of bound access in the housekeeping array if no flag is set
* Considerisolated_cpuswhile checking for nohz conflicts in cpuset
* Improve comment about why nohz CPUs are not excluded by tmigrChanges since v8 [1]:
* Postpone hotplug registration to late initcall (Frederic Weisbecker)
* Move main activation logic in_tmigr_set_cpu_available()and call it after checking for isolation on hotplug and cpusets changes
* Call_tmigr_set_cpu_availabledirectly to force enable tick CPU if required (this saves checking for that on every hotplug change).Changes since v7:
* Movetmigr_available_cpumaskout of tmc lock and specify conditions.
* Initialise tmigr isolation despite the state of isolcpus.
* Move tick CPU check to condition to run SMP call.
* Fix descriptions.Changes since v6 [3]:
* Prevent isolation checks from running during early boot
* Prevent double (de)activation while setting cpus (un)available
* Use synchronous smp calls from the isolation path
* General cleanupChanges since v5:
* Remove fallback if no housekeeping is left by isolcpus andnohz_full
* Adjust condition not to activate CPUs in the migration hierarchy
* Always force the nohz tick CPU active in the hierarchyChanges since v4 [4]:
* useon_each_cpu_mask()with changes on isolated CPUs to avoid races
* keepnohz_fullCPUs included in the timer migration hierarchy
* prevent domain isolated andnohz_fullto cover all CPUsChanges since v3:
* add parameter to function documentation
* split into multiple straightforward patchesChanges since v2:
* improve comments about handling CPUs isolated at boot
* minor cleanupChanges since v1 [5]:
* split into smaller patches
* use available mask instead of unavailable
* simplification and cleanup[1] - https://lore.kernel.org/lkml/[email protected]
[2] - https://lore.kernel.org/lkml/[email protected]
[3] - https://lore.kernel.org/lkml/[email protected]
[4] - https://lore.kernel.org/lkml/[email protected]
[5] - https://lore.kernel.org/lkml/[email protected]Gabriele Monaco (6):
timers: Rename tmigr 'online' bit to 'available'
timers: Add the available mask in timer migration
timers: Usescoped_guardwhen setting/clearing the tmigr available
flag
cgroup/cpuset: Renameupdate_unbound_workqueue_cpumask()to
update_isolation_cpumasks()
sched/isolation: Force housekeeping if isolcpus andnohz_fulldon't
leave any
timers: Exclude isolated cpus from timer migrationYury Norov (1):
cpumask: Add initialiser to use cleanup helpersinclude/linux/cpumask.h | 2 + include/linux/timer.h | 9 ++ include/trace/events/timer_migration.h | 4 +- kernel/cgroup/cpuset.c | 15 +- kernel/sched/isolation.c | 23 +++ kernel/time/timer_migration.c | 213 +++++++++++++++++++++---- kernel/time/timer_migration.h | 2 +- 7 files changed, 232 insertions(+), 36 deletions(-)
Quoted 1 line
base-commit:
ba14500e4bfcab5e841fbf8d7fcbbc80e98d6b9e
Looks good to me now, thanks!
Signature · 2 lines
Frederic Weisbecker SUSE Labs
Quoted 8 lines · Frederic Weisbecker
Quoted 5 lines · Gabriele Monaco
The timer migration mechanism allows active CPUs to pull timers from idle ones to improve the overall idle time. This is however undesired when CPU intensive workloads run on isolated cores, as the algorithm would move the timers from housekeeping to isolated cores, negatively affecting the isolation.
Looks good to me now, thanks!
Thank you for all the comments! Will this series make it for this merge window?
Thanks,
Gabriele