xen/sched: use new idle scheduler for free cpus
v2 cover xen/sched: use new idle scheduler for free cpus
These three patches have been carved out from my core scheduling series as they are sufficiently independent to be applied even without the big series.
Without this little series there are messages like the following to be seen on the console when booting with smt=0:
(XEN) Adding cpu 1 to runqueue 0
(XEN) CPU 1 still not dead...
(XEN) CPU 1 still not dead...
(XEN) CPU 1 still not dead...
(XEN) CPU 1 still not dead...
By assigning cpus to Cpupool-0 only after all cpus are up and by not using the more complicated credit2 scheduler for cpus not in any cpupool this situation can simply no longer happen, as parking the not to be started threads is done before.
Changes in V2:
- commit messages updated
- renamed functions (Patch 2)
Juergen Gross (3):
xen/sched: populate cpupool0 only after all cpus are up xen/sched: remove cpu from pool0 before removing it xen/sched: add minimalistic idle scheduler for free cpus
xen/common/cpupool.c | 194 ++++++++++++++++++++++++++++----------------- xen/common/sched_credit.c | 9 --- xen/common/sched_null.c | 7 -- xen/common/schedule.c | 172 +++++++++++++++++++++------------------- xen/include/xen/sched-if.h | 2 + 5 files changed, 213 insertions(+), 171 deletions(-)
Signature · 7 lines
2.16.4 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v2 1/3 xen/sched: populate cpupool0 only after all cpus are up
Simplify cpupool initialization by populating cpupool0 with cpus only after all cpus are up. This avoids having to call the cpu notifier directly for cpu 0.
With that in place there is no need to create cpupool0 earlier, so do that just before assigning the cpus. Initialize free cpus with all online cpus at that time in order to be able to add the cpu notifier late, too.
---
V1: new patch
--- xen/common/cpupool.c | 18 ++++++++++++++---- 1 file changed, 14 insertions(+), 4 deletions(-)
xen/common/cpupool.c +14 -4
@@ -762,18 +762,28 @@ static struct notifier_block cpu_nfb = { .notifier_call = cpu_callback }; -static int __init cpupool_presmp_init (void ) +static int __init cpupool_init (void ) { +unsigned int cpu;int err; -void *cpu = (void *)(long )smp_processor_id (); + cpupool0 =cpupool_create (0 ,0 , &err);BUG_ON (cpupool0 ==NULL );cpupool_put (cpupool0); -cpu_callback (&cpu_nfb,CPU_ONLINE , cpu);register_cpu_notifier (&cpu_nfb); + +spin_lock (&cpupool_lock); + +cpumask_copy (&cpupool_free_cpus, &cpu_online_map); + + for_each_cpu ( cpu, &cpupool_free_cpus ) +cpupool_assign_cpu_locked (cpupool0, cpu); + +spin_unlock (&cpupool_lock); +return 0 ; } -presmp_initcall (cpupool_presmp_init); +__initcall (cpupool_init);/* * Local variables:
Signature · 7 lines
2.16.4 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v2 2/3 xen/sched: remove cpu from pool0 before removing it
In reply to Juergen GrossToday a cpu which is removed from the system is taken directly from Pool0 to the offline state. This will conflict with the new idle scheduler, so remove it from Pool0 first. Additionally accept removing a free cpu instead of requiring it to be in Pool0.
For the resume failed case we need to call the scheduler code for that situation after the cpupool handling, so move the scheduler code into a function and call it from cpupool_cpu_remove_forced() and remove the CPU_RESUME_FAILED case from cpu_schedule_callback().
Note that we are calling now schedule_cpu_switch() in stop_machine context so we need to switch from spinlock_irq to spinlock_irqsave.
---
V2: rename cpupool_unassign_cpu_[epi|pro]logue() (Dario Faggioli)
--- xen/common/cpupool.c | 176 +++++++++++++++++++++++++++------------------ xen/common/schedule.c | 27 ++++--- xen/include/xen/sched-if.h | 2 + 3 files changed, 128 insertions(+), 77 deletions(-)
xen/common/cpupool.c +108 -68
@@ -282,22 +282,14 @@ static int cpupool_assign_cpu_locked(struct cpupool *c, unsigned int cpu)return 0 ; } -static long cpupool_unassign_cpu_helper (void *info) +static int cpupool_unassign_cpu_finish (struct cpupool *c) {int cpu = cpupool_moving_cpu; -struct cpupool *c = info;struct domain *d; -long ret; - -cpupool_dprintk ("cpupool_unassign_cpu(pool=%d,cpu=%d)\n" , - cpupool_cpu_moving->cpupool_id, cpu); +int ret; -spin_lock (&cpupool_lock);if ( c != cpupool_cpu_moving ) - { - ret = -EADDRNOTAVAIL ; - goto out; - } +return -EADDRNOTAVAIL ;/* * We need this for scanning the domain list, both in
@@ -332,39 +324,19 @@ static long cpupool_unassign_cpu_helper(void *info)domain_update_node_affinity (d); }rcu_read_unlock (&domlist_read_lock); -out: -spin_unlock (&cpupool_lock); -cpupool_dprintk ("cpupool_unassign_cpu ret=%ld\n" , ret); +return ret; } -/* -* unassign a specific cpu from a cpupool -* we must be sure not to run on the cpu to be unassigned! to achieve this -* the main functionality is performed via continue_hypercall_on_cpu on a -* specific cpu. -* if the cpu to be removed is the last one of the cpupool no active domain -* must be bound to the cpupool. dying domains are moved to cpupool0 as they -* might be zombies. -* possible failures: -* - last cpu and still active domains in cpupool -* - cpu just being unplugged -*/ -static int cpupool_unassign_cpu (struct cpupool *c,unsigned int cpu) +static int cpupool_unassign_cpu_start (struct cpupool *c,unsigned int cpu) { -int work_cpu;int ret;struct domain *d; -cpupool_dprintk ("cpupool_unassign_cpu(pool=%d,cpu=%d)\n" , - c->cpupool_id, cpu); -spin_lock (&cpupool_lock); ret = -EADDRNOTAVAIL ;if ( (cpupool_moving_cpu !=-1 ) && (cpu != cpupool_moving_cpu) ) goto out; -if (cpumask_test_cpu (cpu, &cpupool_locked_cpus) ) - goto out; ret =0 ;if ( !cpumask_test_cpu (cpu, c->cpu_valid) && (cpu != cpupool_moving_cpu) )
@@ -376,7 +348,7 @@ static int cpupool_unassign_cpu(struct cpupool *c, unsigned int cpu)rcu_read_lock (&domlist_read_lock);for_each_domain_in_cpupool (d, c) { -if ( !d->is_dying ) +if ( !d->is_dying && system_state == SYS_STATE_active ) { ret = -EBUSY ;break ;
@@ -393,8 +365,58 @@ static int cpupool_unassign_cpu(struct cpupool *c, unsigned int cpu)atomic_inc (&c->refcnt); cpupool_cpu_moving = c;cpumask_clear_cpu (cpu, c->cpu_valid); + +out:spin_unlock (&cpupool_lock); +return ret; +} + +static long cpupool_unassign_cpu_helper (void *info) +{ +struct cpupool *c = info; +long ret; + +cpupool_dprintk ("cpupool_unassign_cpu(pool=%d,cpu=%d)\n" , + cpupool_cpu_moving->cpupool_id, cpu); +spin_lock (&cpupool_lock); + + ret =cpupool_unassign_cpu_finish (c); + +spin_unlock (&cpupool_lock); +cpupool_dprintk ("cpupool_unassign_cpu ret=%ld\n" , ret); + +return ret; +} + +/* +* unassign a specific cpu from a cpupool +* we must be sure not to run on the cpu to be unassigned! to achieve this +* the main functionality is performed via continue_hypercall_on_cpu on a +* specific cpu. +* if the cpu to be removed is the last one of the cpupool no active domain +* must be bound to the cpupool. dying domains are moved to cpupool0 as they +* might be zombies. +* possible failures: +* - last cpu and still active domains in cpupool +* - cpu just being unplugged +*/ +static int cpupool_unassign_cpu (struct cpupool *c,unsigned int cpu) +{ +int work_cpu; +int ret; + +cpupool_dprintk ("cpupool_unassign_cpu(pool=%d,cpu=%d)\n" , + c->cpupool_id, cpu); + + ret =cpupool_unassign_cpu_start (c, cpu); +if ( ret ) + { +cpupool_dprintk ("cpupool_unassign_cpu(pool=%d,cpu=%d) ret %d\n" , + c->cpupool_id, cpu, ret); +return ret; + } + work_cpu =smp_processor_id ();if ( work_cpu == cpu ) {
@@ -403,12 +425,6 @@ static int cpupool_unassign_cpu(struct cpupool *c, unsigned int cpu) work_cpu =cpumask_next (cpu, cpupool0->cpu_valid); }return continue_hypercall_on_cpu (work_cpu, cpupool_unassign_cpu_helper, c); - -out: -spin_unlock (&cpupool_lock); -cpupool_dprintk ("cpupool_unassign_cpu(pool=%d,cpu=%d) ret %d\n" , - c->cpupool_id, cpu, ret); -return ret; }/*
@@ -492,30 +508,54 @@ static int cpupool_cpu_add(unsigned int cpu) }/* -* Called to remove a CPU from a pool. The CPU is locked, to forbid removing -* it from pool0. In fact, if we want to hot-unplug a CPU, it must belong to -* pool0, or we fail. +* This function is called in stop_machine context, so we can be sure no +* non-idle vcpu is active on the system. */ -static int cpupool_cpu_remove (unsigned int cpu) +static void cpupool_cpu_remove (unsigned int cpu) { -int ret = -ENODEV ; +int ret; -spin_lock (&cpupool_lock); +ASSERT (is_idle_vcpu (current)); -if (cpumask_test_cpu (cpu, cpupool0->cpu_valid) ) +if ( !cpumask_test_cpu (cpu, &cpupool_free_cpus) ) { -/* -* If we are not suspending, we are hot-unplugging cpu, and that is -* allowed only for CPUs in pool0. -*/ -cpumask_clear_cpu (cpu, cpupool0->cpu_valid); - ret =0 ; + ret =cpupool_unassign_cpu_finish (cpupool0); +BUG_ON (ret); } +} -if ( !ret ) +/* +* Called before a CPU is being removed from the system. +* Removing a CPU is allowed for free CPUs or CPUs in Pool-0 (those are moved +* to free cpus actually before removing them). +* The CPU is locked, to forbid adding it again to another cpupool. +*/ +static int cpupool_cpu_remove_prologue (unsigned int cpu) +{ +int ret =0 ; + +spin_lock (&cpupool_lock); + +if (cpumask_test_cpu (cpu, &cpupool_locked_cpus) ) + ret = -EBUSY ; +else cpumask_set_cpu (cpu, &cpupool_locked_cpus); +spin_unlock (&cpupool_lock); +if ( ret ) +return ret; + +if (cpumask_test_cpu (cpu, cpupool0->cpu_valid) ) + { +/* Cpupool0 is populated only after all cpus are up. */ +ASSERT (system_state == SYS_STATE_active); + + ret =cpupool_unassign_cpu_start (cpupool0, cpu); + } +else if ( !cpumask_test_cpu (cpu, &cpupool_free_cpus) ) + ret = -ENODEV ; +return ret; }
@@ -523,13 +563,13 @@ static int cpupool_cpu_remove(unsigned int cpu)* Called during resume for all cpus which didn't come up again. The cpu must * be removed from the cpupool it is assigned to. In case a cpupool will be * left without cpu we move all domains of that cpupool to cpupool0. +* As we are called with all domains still frozen there is no need to take the +* cpupool lock here. */ static void cpupool_cpu_remove_forced (unsigned int cpu) {struct cpupool **c; -struct domain *d; - -spin_lock (&cpupool_lock); +int ret;if (cpumask_test_cpu (cpu, &cpupool_free_cpus) )cpumask_clear_cpu (cpu, &cpupool_free_cpus);
@@ -539,19 +579,13 @@ static void cpupool_cpu_remove_forced(unsigned int cpu) {if (cpumask_test_cpu (cpu, (*c)->cpu_valid) ) { -cpumask_clear_cpu (cpu, (*c)->cpu_valid); -if (cpumask_weight ((*c)->cpu_valid) ==0 ) - { -if ( *c == cpupool0 ) -panic ("No cpu left in cpupool0\n" ); -for_each_domain_in_cpupool (d, *c) -cpupool_move_domain_locked (d, cpupool0); - } + ret =cpupool_unassign_cpu (*c, cpu); +BUG_ON (ret); } } } -spin_unlock (&cpupool_lock); +sched_rm_cpu (cpu); }/*
@@ -619,7 +653,8 @@ int cpupool_do_sysctl(struct xen_sysctl_cpupool_op *op)if ( cpu >= nr_cpu_ids ) goto addcpu_out; ret = -ENODEV ; -if ( !cpumask_test_cpu (cpu, &cpupool_free_cpus) ) +if ( !cpumask_test_cpu (cpu, &cpupool_free_cpus) || +cpumask_test_cpu (cpu, &cpupool_locked_cpus) ) goto addcpu_out; c =cpupool_find_by_id (op->cpupool_id); ret = -ENOENT ;
@@ -746,7 +781,12 @@ static int cpu_callback(case CPU_DOWN_PREPARE :/* Suspend/Resume don't change assignments of cpus to cpupools. */ if ( system_state <= SYS_STATE_active ) - rc =cpupool_cpu_remove (cpu); + rc =cpupool_cpu_remove_prologue (cpu); +break ; +case CPU_DYING : +/* Suspend/Resume don't change assignments of cpus to cpupools. */ +if ( system_state <= SYS_STATE_active ) +cpupool_cpu_remove (cpu);break ;case CPU_RESUME_FAILED :cpupool_cpu_remove_forced (cpu);
xen/common/schedule.c +18 -9
@@ -1654,6 +1654,20 @@ static void cpu_schedule_down(unsigned int cpu)kill_timer (&sd->s_timer); } +void sched_rm_cpu (unsigned int cpu) +{ +int rc; +struct schedule_data *sd = &per_cpu (schedule_data, cpu); +struct scheduler *sched =per_cpu (scheduler, cpu); + +rcu_read_lock (&domlist_read_lock); + rc =cpu_disable_scheduler (cpu); +BUG_ON (rc); +rcu_read_unlock (&domlist_read_lock); +sched_deinit_pdata (sched, sd->sched_priv, cpu); +cpu_schedule_down (cpu); +} +static int cpu_schedule_callback (struct notifier_block *nfb,unsigned long action,void *hcpu) {
@@ -1709,16 +1723,10 @@ static int cpu_schedule_callback( rc =cpu_disable_scheduler_check (cpu);rcu_read_unlock (&domlist_read_lock);break ; -case CPU_RESUME_FAILED :case CPU_DEAD :if ( system_state == SYS_STATE_suspend )break ; -rcu_read_lock (&domlist_read_lock); - rc =cpu_disable_scheduler (cpu); -BUG_ON (rc); -rcu_read_unlock (&domlist_read_lock); -sched_deinit_pdata (sched, sd->sched_priv, cpu); -cpu_schedule_down (cpu); +sched_rm_cpu (cpu);break ;case CPU_UP_CANCELED :if ( system_state != SYS_STATE_resume )
@@ -1841,6 +1849,7 @@ int schedule_cpu_switch(unsigned int cpu, struct cpupool *c)struct cpupool *old_pool =per_cpu (cpupool, cpu);struct schedule_data *sd = &per_cpu (schedule_data, cpu);spinlock_t *old_lock, *new_lock; +unsigned long flags;/* * pCPUs only move from a valid cpupool to free (i.e., out of any pool),
@@ -1895,7 +1904,7 @@ int schedule_cpu_switch(unsigned int cpu, struct cpupool *c)* that the lock itself changed, and retry acquiring the new one (which * will be the correct, remapped one, at that point). */ - old_lock =pcpu_schedule_lock_irq (cpu); + old_lock =pcpu_schedule_lock_irqsave (cpu, &flags); vpriv_old = idle->sched_priv; ppriv_old = sd->sched_priv;
@@ -1913,7 +1922,7 @@ int schedule_cpu_switch(unsigned int cpu, struct cpupool *c) sd->schedule_lock = new_lock;/* _Not_ pcpu_schedule_unlock(): schedule_lock may have changed! */ -spin_unlock_irq (old_lock); +spin_unlock_irqrestore (old_lock, flags);sched_do_tick_resume (new_ops, cpu);
xen/include/xen/sched-if.h +2 -0
@@ -437,4 +437,6 @@ affinity_balance_cpumask(const struct vcpu *v, int step, cpumask_t *mask) cpumask_copy(mask, v->cpu_hard_affinity); } +void sched_rm_cpu (unsigned int cpu); +#endif /* __XEN_SCHED_IF_H__ */
Signature · 7 lines
2.16.4 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
PATCH v2 3/3 xen/sched: add minimalistic idle scheduler for free cpus
In reply to Juergen GrossInstead of having a full blown scheduler running for the free cpus add a very minimalistic scheduler for that purpose only ever scheduling the related idle vcpu. This has the big advantage of not needing any per-cpu, per-domain or per-scheduling unit data for free cpus and in turn simplifying moving cpus to and from cpupools a lot.
Right now, CPUs that are not in any pool, still belong to Pool-0's scheduler. This forces us to make, within the scheduler, extra effort to avoid actually running vCPUs on those.
In the case of Credit1, this also cause issue to weights (re)distribution, as the number of CPUs available to the scheduler is wrong.
This is described in the changelog of commit e7191920261d ("xen:
credit2: never consider CPUs outside of our cpupool").
This new scheduler will just use a common lock for all free cpus.
As this new scheduler is not user selectable don't register it as an official scheduler, but just include it in schedule.c.
--- xen/common/sched_credit.c | 9 --- xen/common/sched_null.c | 7 --- xen/common/schedule.c | 153 +++++++++++++++++++++++----------------------- 3 files changed, 75 insertions(+), 94 deletions(-)
xen/common/sched_credit.c +0 -9
@@ -617,15 +617,6 @@ csched_init_pdata(const struct scheduler *ops, void *pdata, int cpu) {unsigned long flags;struct csched_private *prv =CSCHED_PRIV (ops); -struct schedule_data *sd = &per_cpu (schedule_data, cpu); - -/* -* This is called either during during boot, resume or hotplug, in -* case Credit1 is the scheduler chosen at boot. In such cases, the -* scheduler lock for cpu is already pointing to the default per-cpu -* spinlock, as Credit1 needs it, so there is no remapping to be done. -*/ -ASSERT (sd->schedule_lock == &sd->_lock && !spin_is_locked (&sd->_lock));spin_lock_irqsave (&prv->lock, flags);init_pdata (prv, pdata, cpu);
xen/common/sched_null.c +0 -7
@@ -167,17 +167,10 @@ static void init_pdata(struct null_private *prv, unsigned int cpu)static void null_init_pdata (const struct scheduler *ops,void *pdata,int cpu) {struct null_private *prv =null_priv (ops); -struct schedule_data *sd = &per_cpu (schedule_data, cpu);/* alloc_pdata is not implemented, so we want this to be NULL. */ ASSERT (!pdata); -/* -* The scheduler lock points already to the default per-cpu spinlock, -* so there is no remapping to be done. -*/ -ASSERT (sd->schedule_lock == &sd->_lock && !spin_is_locked (&sd->_lock)); -init_pdata (prv, cpu); }
xen/common/schedule.c +75 -78
256 lines not on this page · raw message
Signature · 7 lines
2.16.4 _______________________________________________ Xen-devel mailing list [email protected] https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v2 3/3] xen/sched: add minimalistic idle scheduler for free cpus
xen/common/schedule.c extern const struct scheduler *__start_schedulers_array[], *__end_schedulers_arr · line 100
Quoted 6 lines · Juergen Gross
xen/common/schedule.c +6
-0+static void * +sched_idle_alloc_vdata (const struct scheduler *ops,struct vcpu *v, +void *dd) +{ +/* Any non-NULL pointer is fine here. */ +return (void *)1UL ;
As an observation, the vdata interface (and others, if applicable) could do with being updated to use ERR_PTR(), just as done in c/s 340edc390
Passing 1 back here is rather dangerous.
~Andrew
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v2 3/3] xen/sched: add minimalistic idle scheduler for free cpus
Quoted 10 lines · Andrew Cooper
xen/common/schedule.c extern const struct scheduler *__start_schedulers_array[], *__end_schedulers_arr · line 100
Quoted 6 lines · Juergen Gross
xen/common/schedule.c +6
-0+static void * +sched_idle_alloc_vdata (const struct scheduler *ops,struct vcpu *v, +void *dd) +{ +/* Any non-NULL pointer is fine here. */ +return (void *)1UL ;As an observation, the vdata interface (and others, if applicable) could do with being updated to use
ERR_PTR(), just as done in c/s340edc390
One of the items for my scheduler cleanup patches.
Quoted 1 line
Passing 1 back here is rather dangerous.
Not really. vdata is scheduler specific, and the idle-scheduler doesn't use it.
Juergen
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v2 3/3] xen/sched: add minimalistic idle scheduler for free cpus
Quoted 12 lines · Juergen Gross
Quoted 10 lines · Andrew Cooper
xen/common/schedule.c extern const struct scheduler *__start_schedulers_array[], *__end_schedulers_arr · line 100
Quoted 6 lines · Juergen Gross
xen/common/schedule.c +6
-0+static void * +sched_idle_alloc_vdata (const struct scheduler *ops,struct vcpu *v, +void *dd) +{ +/* Any non-NULL pointer is fine here. */ +return (void *)1UL ;As an observation, the vdata interface (and others, if applicable) could do with being updated to use
ERR_PTR(), just as done in c/s340edc390
Quoted 6 lines
One of the items for my scheduler cleanup patches.
Quoted 1 line
Passing 1 back here is rather dangerous.
Not really. vdata is scheduler specific, and the idle-scheduler doesn't use it.
But maybe handing back e.g. ZERO_BLOCK_PTR would still be better? Not the least because of avoiding an open-coded cast?
Jan
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel
Re: [Xen-devel] [PATCH v2 3/3] xen/sched: add minimalistic idle scheduler for free cpus
Quoted 15 lines · Jan Beulich
Quoted 14 lines · Juergen Gross
Quoted 10 lines · Andrew Cooper
xen/common/schedule.c extern const struct scheduler *__start_schedulers_array[], *__end_schedulers_arr · line 100
Quoted 6 lines · Juergen Gross
xen/common/schedule.c +6
-0+static void * +sched_idle_alloc_vdata (const struct scheduler *ops,struct vcpu *v, +void *dd) +{ +/* Any non-NULL pointer is fine here. */ +return (void *)1UL ;As an observation, the vdata interface (and others, if applicable) could do with being updated to use
ERR_PTR(), just as done in c/s340edc390One of the items for my scheduler cleanup patches.
Quoted 7 lines
Quoted 4 lines
Quoted 1 line
Passing 1 back here is rather dangerous.
Not really. vdata is scheduler specific, and the idle-scheduler doesn't use it.
But maybe handing back e.g.
ZERO_BLOCK_PTRwould still be better? Not the least because of avoiding an open-coded cast?
Yes. I wasn't aware of its existance.
Juergen
_______________________________________________
Xen-devel mailing list
[email protected]
https://lists.xenproject.org/mailman/listinfo/xen-devel