Currently, cpudl.free_cpus contains all cpus during init(see cpudl_init()),
so when calling cpudl_find() we have to add rd->span cpumask(cpus_allowed is
undependable when performing clustered scheduling using the cpuset) to avoid
selecting the cpu outside current root domain, see find_later_rq().
This patch adds cpudl_set_freecpu() to initialize cpudl.free_cpus when doing
rq_attach_root(), so we can avoid the extra rd->span operation when calling
cpudl_find().
Signed-off-by: pang.xunlei <[email protected]>
---
kernel/sched/core.c | 2 ++
kernel/sched/cpudeadline.c | 18 ++++++++++++++----
kernel/sched/cpudeadline.h | 1 +
kernel/sched/deadline.c | 3 ---
4 files changed, 17 insertions(+), 7 deletions(-)
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 240157c..1b417de 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -5584,6 +5584,8 @@ static void rq_attach_root(struct rq *rq, struct root_domain *rd)
rq->rd = rd;
cpumask_set_cpu(rq->cpu, rd->span);
+ cpudl_set_freecpu(&rd->cpudl, rq->cpu);
+
if (cpumask_test_cpu(rq->cpu, cpu_active_mask))
set_rq_online(rq);
diff --git a/kernel/sched/cpudeadline.c b/kernel/sched/cpudeadline.c
index 539ca3c..c79f0d7 100644
--- a/kernel/sched/cpudeadline.c
+++ b/kernel/sched/cpudeadline.c
@@ -107,7 +107,9 @@ int cpudl_find(struct cpudl *cp, struct task_struct *p,
int best_cpu = -1;
const struct sched_dl_entity *dl_se = &p->dl;
- if (later_mask && cpumask_and(later_mask, later_mask, cp->free_cpus)) {
+ if (later_mask &&
+ cpumask_and(later_mask, cp->free_cpus, &p->cpus_allowed) &&
+ cpumask_and(later_mask, later_mask, cpu_active_mask)) {
best_cpu = cpumask_any(later_mask);
goto out;
} else if (cpumask_test_cpu(cpudl_maximum(cp), &p->cpus_allowed) &&
@@ -186,6 +188,16 @@ out:
}
/*
+ * cpudl_set_freecpu - Set the cpudl.free_cpus
+ * @cp: the cpudl max-heap context
+ * @cpu: rd attached cpu
+ */
+void cpudl_set_freecpu(struct cpudl *cp, int cpu)
+{
+ cpumask_set_cpu(cpu, cp->free_cpus);
+}
+
+/*
* cpudl_init - initialize the cpudl structure
* @cp: the cpudl max-heap context
*/
@@ -203,7 +215,7 @@ int cpudl_init(struct cpudl *cp)
if (!cp->elements)
return -ENOMEM;
- if (!alloc_cpumask_var(&cp->free_cpus, GFP_KERNEL)) {
+ if (!zalloc_cpumask_var(&cp->free_cpus, GFP_KERNEL)) {
kfree(cp->elements);
return -ENOMEM;
}
@@ -211,8 +223,6 @@ int cpudl_init(struct cpudl *cp)
for_each_possible_cpu(i)
cp->elements[i].idx = IDX_INVALID;
- cpumask_setall(cp->free_cpus);
-
return 0;
}
diff --git a/kernel/sched/cpudeadline.h b/kernel/sched/cpudeadline.h
index 020039b..4a10a65 100644
--- a/kernel/sched/cpudeadline.h
+++ b/kernel/sched/cpudeadline.h
@@ -24,6 +24,7 @@ int cpudl_find(struct cpudl *cp, struct task_struct *p,
struct cpumask *later_mask);
void cpudl_set(struct cpudl *cp, int cpu, u64 dl, int is_valid);
int cpudl_init(struct cpudl *cp);
+void cpudl_set_freecpu(struct cpudl *cp, int cpu);
void cpudl_cleanup(struct cpudl *cp);
#endif /* CONFIG_SMP */
diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
index 5285332..bd83272 100644
--- a/kernel/sched/deadline.c
+++ b/kernel/sched/deadline.c
@@ -1187,9 +1187,6 @@ static int find_later_rq(struct task_struct *task)
* We have to consider system topology and task affinity
* first, then we can look for a suitable cpu.
*/
- cpumask_copy(later_mask, task_rq(task)->rd->span);
- cpumask_and(later_mask, later_mask, cpu_active_mask);
- cpumask_and(later_mask, later_mask, &task->cpus_allowed);
best_cpu = cpudl_find(&task_rq(task)->rd->cpudl,
task, later_mask);
if (best_cpu == -1)
--
1.7.9.5
In check_preempt_equal_dl(), cpudl_find() is called with a NULL later_mask,
thus cpudl_find() here doesn't check cpudl.free_cpus at all.
This patch takles this issue by always passing a non-NULL cpumask to cpudl_find(),
and assigns later_mask in this function.
Signed-off-by: pang.xunlei <[email protected]>
---
kernel/sched/cpudeadline.c | 10 ++++------
kernel/sched/deadline.c | 10 ++++++----
2 files changed, 10 insertions(+), 10 deletions(-)
diff --git a/kernel/sched/cpudeadline.c b/kernel/sched/cpudeadline.c
index c79f0d7..c01b3aa 100644
--- a/kernel/sched/cpudeadline.c
+++ b/kernel/sched/cpudeadline.c
@@ -97,7 +97,7 @@ static inline int cpudl_maximum(struct cpudl *cp)
* cpudl_find - find the best (later-dl) CPU in the system
* @cp: the cpudl max-heap context
* @p: the task
- * @later_mask: a mask to fill in with the selected CPUs (or NULL)
+ * @later_mask: a mask to fill in with the selected CPUs (not NULL)
*
* Returns: int - best CPU (heap maximum if suitable)
*/
@@ -107,16 +107,14 @@ int cpudl_find(struct cpudl *cp, struct task_struct *p,
int best_cpu = -1;
const struct sched_dl_entity *dl_se = &p->dl;
- if (later_mask &&
- cpumask_and(later_mask, cp->free_cpus, &p->cpus_allowed) &&
- cpumask_and(later_mask, later_mask, cpu_active_mask)) {
+ cpumask_and(later_mask, cpu_active_mask, &p->cpus_allowed);
+ if (cpumask_and(later_mask, later_mask, cp->free_cpus)) {
best_cpu = cpumask_any(later_mask);
goto out;
} else if (cpumask_test_cpu(cpudl_maximum(cp), &p->cpus_allowed) &&
dl_time_before(dl_se->deadline, cp->elements[0].dl)) {
best_cpu = cpudl_maximum(cp);
- if (later_mask)
- cpumask_set_cpu(best_cpu, later_mask);
+ cpumask_set_cpu(best_cpu, later_mask);
}
out:
diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
index bd83272..3ecf838 100644
--- a/kernel/sched/deadline.c
+++ b/kernel/sched/deadline.c
@@ -965,14 +965,18 @@ out:
return cpu;
}
+static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask_dl);
+
static void check_preempt_equal_dl(struct rq *rq, struct task_struct *p)
{
+ struct cpumask *later_mask = this_cpu_cpumask_var_ptr(local_cpu_mask_dl);
+
/*
* Current can't be migrated, useless to reschedule,
* let's hope p can move out.
*/
if (rq->curr->nr_cpus_allowed == 1 ||
- cpudl_find(&rq->rd->cpudl, rq->curr, NULL) == -1)
+ cpudl_find(&rq->rd->cpudl, rq->curr, later_mask) == -1)
return;
/*
@@ -980,7 +984,7 @@ static void check_preempt_equal_dl(struct rq *rq, struct task_struct *p)
* see if it is pushed or pulled somewhere else.
*/
if (p->nr_cpus_allowed != 1 &&
- cpudl_find(&rq->rd->cpudl, p, NULL) != -1)
+ cpudl_find(&rq->rd->cpudl, p, later_mask) != -1)
return;
resched_curr(rq);
@@ -1167,8 +1171,6 @@ next_node:
return NULL;
}
-static DEFINE_PER_CPU(cpumask_var_t, local_cpu_mask_dl);
-
static int find_later_rq(struct task_struct *task)
{
struct sched_domain *sd;
--
1.7.9.5
cpudl_find() is not a good place to select the best cpu, so leave
this role to its call site, currently it is find_later_rq() where
we can do the election of the best cpu according to sd topology.
Signed-off-by: pang.xunlei <[email protected]>
---
kernel/sched/cpudeadline.c | 15 +++++----------
kernel/sched/deadline.c | 17 ++++++-----------
2 files changed, 11 insertions(+), 21 deletions(-)
diff --git a/kernel/sched/cpudeadline.c b/kernel/sched/cpudeadline.c
index 3047846..41d3578 100644
--- a/kernel/sched/cpudeadline.c
+++ b/kernel/sched/cpudeadline.c
@@ -100,29 +100,24 @@ static inline int cpudl_maximum(struct cpudl *cp)
* @later_mask: a mask to fill in with the selected CPUs (not NULL)
* @set_flag: indicate if later_mask should be set
*
- * Returns: int - best CPU (heap maximum if suitable)
+ * Return: (int)bool - CPUs were found
*/
int cpudl_find(struct cpudl *cp, struct task_struct *p,
struct cpumask *later_mask, int set_flag)
{
- int best_cpu = -1;
const struct sched_dl_entity *dl_se = &p->dl;
cpumask_and(later_mask, cpu_active_mask, &p->cpus_allowed);
if (cpumask_and(later_mask, later_mask, cp->free_cpus)) {
- best_cpu = cpumask_any(later_mask);
- goto out;
+ return 1;
} else if (cpumask_test_cpu(cpudl_maximum(cp), &p->cpus_allowed) &&
dl_time_before(dl_se->deadline, cp->elements[0].dl)) {
- best_cpu = cpudl_maximum(cp);
if (set_flag)
- cpumask_set_cpu(best_cpu, later_mask);
+ cpumask_set_cpu(cpudl_maximum(cp), later_mask);
+ return 1;
}
-out:
- WARN_ON(best_cpu != -1 && !cpu_present(best_cpu));
-
- return best_cpu;
+ return 0;
}
/*
diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
index e8208d0..3e82cf3 100644
--- a/kernel/sched/deadline.c
+++ b/kernel/sched/deadline.c
@@ -976,7 +976,7 @@ static void check_preempt_equal_dl(struct rq *rq, struct task_struct *p)
* let's hope p can move out.
*/
if (rq->curr->nr_cpus_allowed == 1 ||
- cpudl_find(&rq->rd->cpudl, rq->curr, later_mask, 0) == -1)
+ !cpudl_find(&rq->rd->cpudl, rq->curr, later_mask, 0))
return;
/*
@@ -984,7 +984,7 @@ static void check_preempt_equal_dl(struct rq *rq, struct task_struct *p)
* see if it is pushed or pulled somewhere else.
*/
if (p->nr_cpus_allowed != 1 &&
- cpudl_find(&rq->rd->cpudl, p, later_mask, 0) != -1)
+ cpudl_find(&rq->rd->cpudl, p, later_mask, 0))
return;
resched_curr(rq);
@@ -1189,9 +1189,7 @@ static int find_later_rq(struct task_struct *task)
* We have to consider system topology and task affinity
* first, then we can look for a suitable cpu.
*/
- best_cpu = cpudl_find(&task_rq(task)->rd->cpudl,
- task, later_mask, 1);
- if (best_cpu == -1)
+ if (!cpudl_find(&task_rq(task)->rd->cpudl, task, later_mask, 1))
return -1;
/*
@@ -1230,12 +1228,9 @@ static int find_later_rq(struct task_struct *task)
return this_cpu;
}
- /*
- * Last chance: if best_cpu is valid and is
- * in the mask, that becomes our choice.
- */
- if (best_cpu < nr_cpu_ids &&
- cpumask_test_cpu(best_cpu, sched_domain_span(sd))) {
+ best_cpu = cpumask_first_and(lowest_mask,
+ sched_domain_span(sd));
+ if (best_cpu < nr_cpu_ids) {
rcu_read_unlock();
return best_cpu;
}
--
1.7.9.5
The call site of cpudl_find() in check_preempt_equal_dl() doesn't
use later_mask, so add this extra argument to distinquish the case.
Signed-off-by: pang.xunlei <[email protected]>
---
kernel/sched/cpudeadline.c | 6 ++++--
kernel/sched/cpudeadline.h | 2 +-
kernel/sched/deadline.c | 6 +++---
3 files changed, 8 insertions(+), 6 deletions(-)
diff --git a/kernel/sched/cpudeadline.c b/kernel/sched/cpudeadline.c
index c01b3aa..3047846 100644
--- a/kernel/sched/cpudeadline.c
+++ b/kernel/sched/cpudeadline.c
@@ -98,11 +98,12 @@ static inline int cpudl_maximum(struct cpudl *cp)
* @cp: the cpudl max-heap context
* @p: the task
* @later_mask: a mask to fill in with the selected CPUs (not NULL)
+ * @set_flag: indicate if later_mask should be set
*
* Returns: int - best CPU (heap maximum if suitable)
*/
int cpudl_find(struct cpudl *cp, struct task_struct *p,
- struct cpumask *later_mask)
+ struct cpumask *later_mask, int set_flag)
{
int best_cpu = -1;
const struct sched_dl_entity *dl_se = &p->dl;
@@ -114,7 +115,8 @@ int cpudl_find(struct cpudl *cp, struct task_struct *p,
} else if (cpumask_test_cpu(cpudl_maximum(cp), &p->cpus_allowed) &&
dl_time_before(dl_se->deadline, cp->elements[0].dl)) {
best_cpu = cpudl_maximum(cp);
- cpumask_set_cpu(best_cpu, later_mask);
+ if (set_flag)
+ cpumask_set_cpu(best_cpu, later_mask);
}
out:
diff --git a/kernel/sched/cpudeadline.h b/kernel/sched/cpudeadline.h
index 4a10a65..a3a85e8 100644
--- a/kernel/sched/cpudeadline.h
+++ b/kernel/sched/cpudeadline.h
@@ -22,7 +22,7 @@ struct cpudl {
#ifdef CONFIG_SMP
int cpudl_find(struct cpudl *cp, struct task_struct *p,
struct cpumask *later_mask);
-void cpudl_set(struct cpudl *cp, int cpu, u64 dl, int is_valid);
+void cpudl_set(struct cpudl *cp, int cpu, u64 dl, int is_valid, int set_flag);
int cpudl_init(struct cpudl *cp);
void cpudl_set_freecpu(struct cpudl *cp, int cpu);
void cpudl_cleanup(struct cpudl *cp);
diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
index 3ecf838..e8208d0 100644
--- a/kernel/sched/deadline.c
+++ b/kernel/sched/deadline.c
@@ -976,7 +976,7 @@ static void check_preempt_equal_dl(struct rq *rq, struct task_struct *p)
* let's hope p can move out.
*/
if (rq->curr->nr_cpus_allowed == 1 ||
- cpudl_find(&rq->rd->cpudl, rq->curr, later_mask) == -1)
+ cpudl_find(&rq->rd->cpudl, rq->curr, later_mask, 0) == -1)
return;
/*
@@ -984,7 +984,7 @@ static void check_preempt_equal_dl(struct rq *rq, struct task_struct *p)
* see if it is pushed or pulled somewhere else.
*/
if (p->nr_cpus_allowed != 1 &&
- cpudl_find(&rq->rd->cpudl, p, later_mask) != -1)
+ cpudl_find(&rq->rd->cpudl, p, later_mask, 0) != -1)
return;
resched_curr(rq);
@@ -1190,7 +1190,7 @@ static int find_later_rq(struct task_struct *task)
* first, then we can look for a suitable cpu.
*/
best_cpu = cpudl_find(&task_rq(task)->rd->cpudl,
- task, later_mask);
+ task, later_mask, 1);
if (best_cpu == -1)
return -1;
--
1.7.9.5
On Wed, 19 Nov 2014 23:46:21 +0800
"pang.xunlei" <[email protected]> wrote:
> The call site of cpudl_find() in check_preempt_equal_dl() doesn't
> use later_mask, so add this extra argument to distinquish the case.
>
> Signed-off-by: pang.xunlei <[email protected]>
> ---
> kernel/sched/cpudeadline.c | 6 ++++--
> kernel/sched/cpudeadline.h | 2 +-
> kernel/sched/deadline.c | 6 +++---
> 3 files changed, 8 insertions(+), 6 deletions(-)
>
> diff --git a/kernel/sched/cpudeadline.c b/kernel/sched/cpudeadline.c
> index c01b3aa..3047846 100644
> --- a/kernel/sched/cpudeadline.c
> +++ b/kernel/sched/cpudeadline.c
> @@ -98,11 +98,12 @@ static inline int cpudl_maximum(struct cpudl *cp)
> * @cp: the cpudl max-heap context
> * @p: the task
> * @later_mask: a mask to fill in with the selected CPUs (not NULL)
> + * @set_flag: indicate if later_mask should be set
> *
> * Returns: int - best CPU (heap maximum if suitable)
> */
> int cpudl_find(struct cpudl *cp, struct task_struct *p,
> - struct cpumask *later_mask)
> + struct cpumask *later_mask, int set_flag)
set_flag should be a bool type.
> {
> int best_cpu = -1;
> const struct sched_dl_entity *dl_se = &p->dl;
> @@ -114,7 +115,8 @@ int cpudl_find(struct cpudl *cp, struct task_struct *p,
> } else if (cpumask_test_cpu(cpudl_maximum(cp), &p->cpus_allowed) &&
> dl_time_before(dl_se->deadline, cp->elements[0].dl)) {
> best_cpu = cpudl_maximum(cp);
> - cpumask_set_cpu(best_cpu, later_mask);
> + if (set_flag)
> + cpumask_set_cpu(best_cpu, later_mask);
I'm not sure this is worth it. cpumask_set_cpu() is rather efficient.
> }
>
> out:
On Wed, 19 Nov 2014 23:46:22 +0800
"pang.xunlei" <[email protected]> wrote:
> cpudl_find() is not a good place to select the best cpu, so leave
> this role to its call site, currently it is find_later_rq() where
> we can do the election of the best cpu according to sd topology.
>
> Signed-off-by: pang.xunlei <[email protected]>
> ---
> kernel/sched/cpudeadline.c | 15 +++++----------
> kernel/sched/deadline.c | 17 ++++++-----------
> 2 files changed, 11 insertions(+), 21 deletions(-)
>
> diff --git a/kernel/sched/cpudeadline.c b/kernel/sched/cpudeadline.c
> index 3047846..41d3578 100644
> --- a/kernel/sched/cpudeadline.c
> +++ b/kernel/sched/cpudeadline.c
> @@ -100,29 +100,24 @@ static inline int cpudl_maximum(struct cpudl *cp)
> * @later_mask: a mask to fill in with the selected CPUs (not NULL)
> * @set_flag: indicate if later_mask should be set
> *
> - * Returns: int - best CPU (heap maximum if suitable)
> + * Return: (int)bool - CPUs were found
> */
> int cpudl_find(struct cpudl *cp, struct task_struct *p,
> struct cpumask *later_mask, int set_flag)
> {
> - int best_cpu = -1;
> const struct sched_dl_entity *dl_se = &p->dl;
>
> cpumask_and(later_mask, cpu_active_mask, &p->cpus_allowed);
> if (cpumask_and(later_mask, later_mask, cp->free_cpus)) {
> - best_cpu = cpumask_any(later_mask);
> - goto out;
> + return 1;
> } else if (cpumask_test_cpu(cpudl_maximum(cp), &p->cpus_allowed) &&
> dl_time_before(dl_se->deadline, cp->elements[0].dl)) {
> - best_cpu = cpudl_maximum(cp);
> if (set_flag)
> - cpumask_set_cpu(best_cpu, later_mask);
> + cpumask_set_cpu(cpudl_maximum(cp), later_mask);
> + return 1;
> }
>
> -out:
> - WARN_ON(best_cpu != -1 && !cpu_present(best_cpu));
You lost this warning. It should be moved too.
> -
> - return best_cpu;
> + return 0;
> }
>
> /*
> diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
> index e8208d0..3e82cf3 100644
> --- a/kernel/sched/deadline.c
> +++ b/kernel/sched/deadline.c
> @@ -976,7 +976,7 @@ static void check_preempt_equal_dl(struct rq *rq, struct task_struct *p)
> * let's hope p can move out.
> */
> if (rq->curr->nr_cpus_allowed == 1 ||
> - cpudl_find(&rq->rd->cpudl, rq->curr, later_mask, 0) == -1)
> + !cpudl_find(&rq->rd->cpudl, rq->curr, later_mask, 0))
> return;
>
> /*
> @@ -984,7 +984,7 @@ static void check_preempt_equal_dl(struct rq *rq, struct task_struct *p)
> * see if it is pushed or pulled somewhere else.
> */
> if (p->nr_cpus_allowed != 1 &&
> - cpudl_find(&rq->rd->cpudl, p, later_mask, 0) != -1)
> + cpudl_find(&rq->rd->cpudl, p, later_mask, 0))
> return;
>
> resched_curr(rq);
> @@ -1189,9 +1189,7 @@ static int find_later_rq(struct task_struct *task)
> * We have to consider system topology and task affinity
> * first, then we can look for a suitable cpu.
> */
> - best_cpu = cpudl_find(&task_rq(task)->rd->cpudl,
> - task, later_mask, 1);
> - if (best_cpu == -1)
> + if (!cpudl_find(&task_rq(task)->rd->cpudl, task, later_mask, 1))
> return -1;
>
> /*
> @@ -1230,12 +1228,9 @@ static int find_later_rq(struct task_struct *task)
> return this_cpu;
> }
>
> - /*
> - * Last chance: if best_cpu is valid and is
> - * in the mask, that becomes our choice.
> - */
> - if (best_cpu < nr_cpu_ids &&
> - cpumask_test_cpu(best_cpu, sched_domain_span(sd))) {
> + best_cpu = cpumask_first_and(lowest_mask,
> + sched_domain_span(sd));
Sometimes that 80 character rule isn't the best for readability. But
that is Peter's or Juri's call.
> + if (best_cpu < nr_cpu_ids) {
The warning should probably go here.
-- Steve
> rcu_read_unlock();
> return best_cpu;
> }
Hi,
On 11/19/14, 11:46 PM, pang.xunlei wrote:
> Currently, cpudl.free_cpus contains all cpus during init(see cpudl_init()),
> so when calling cpudl_find() we have to add rd->span cpumask(cpus_allowed is
> undependable when performing clustered scheduling using the cpuset) to avoid
> selecting the cpu outside current root domain, see find_later_rq().
>
> This patch adds cpudl_set_freecpu() to initialize cpudl.free_cpus when doing
> rq_attach_root(), so we can avoid the extra rd->span operation when calling
> cpudl_find().
>
> Signed-off-by: pang.xunlei <[email protected]>
Reviewed-by: Wanpeng Li <[email protected]>
I think this patch solve one problem I meet in my progress to handle dl
task migration during cpu hotplug.
https://www.mail-archive.com/[email protected]/msg770579.html
Regards,
Wanpeng Li
> ---
> kernel/sched/core.c | 2 ++
> kernel/sched/cpudeadline.c | 18 ++++++++++++++----
> kernel/sched/cpudeadline.h | 1 +
> kernel/sched/deadline.c | 3 ---
> 4 files changed, 17 insertions(+), 7 deletions(-)
>
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index 240157c..1b417de 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -5584,6 +5584,8 @@ static void rq_attach_root(struct rq *rq, struct root_domain *rd)
> rq->rd = rd;
>
> cpumask_set_cpu(rq->cpu, rd->span);
> + cpudl_set_freecpu(&rd->cpudl, rq->cpu);
> +
> if (cpumask_test_cpu(rq->cpu, cpu_active_mask))
> set_rq_online(rq);
>
> diff --git a/kernel/sched/cpudeadline.c b/kernel/sched/cpudeadline.c
> index 539ca3c..c79f0d7 100644
> --- a/kernel/sched/cpudeadline.c
> +++ b/kernel/sched/cpudeadline.c
> @@ -107,7 +107,9 @@ int cpudl_find(struct cpudl *cp, struct task_struct *p,
> int best_cpu = -1;
> const struct sched_dl_entity *dl_se = &p->dl;
>
> - if (later_mask && cpumask_and(later_mask, later_mask, cp->free_cpus)) {
> + if (later_mask &&
> + cpumask_and(later_mask, cp->free_cpus, &p->cpus_allowed) &&
> + cpumask_and(later_mask, later_mask, cpu_active_mask)) {
> best_cpu = cpumask_any(later_mask);
> goto out;
> } else if (cpumask_test_cpu(cpudl_maximum(cp), &p->cpus_allowed) &&
> @@ -186,6 +188,16 @@ out:
> }
>
> /*
> + * cpudl_set_freecpu - Set the cpudl.free_cpus
> + * @cp: the cpudl max-heap context
> + * @cpu: rd attached cpu
> + */
> +void cpudl_set_freecpu(struct cpudl *cp, int cpu)
> +{
> + cpumask_set_cpu(cpu, cp->free_cpus);
> +}
> +
> +/*
> * cpudl_init - initialize the cpudl structure
> * @cp: the cpudl max-heap context
> */
> @@ -203,7 +215,7 @@ int cpudl_init(struct cpudl *cp)
> if (!cp->elements)
> return -ENOMEM;
>
> - if (!alloc_cpumask_var(&cp->free_cpus, GFP_KERNEL)) {
> + if (!zalloc_cpumask_var(&cp->free_cpus, GFP_KERNEL)) {
> kfree(cp->elements);
> return -ENOMEM;
> }
> @@ -211,8 +223,6 @@ int cpudl_init(struct cpudl *cp)
> for_each_possible_cpu(i)
> cp->elements[i].idx = IDX_INVALID;
>
> - cpumask_setall(cp->free_cpus);
> -
> return 0;
> }
>
> diff --git a/kernel/sched/cpudeadline.h b/kernel/sched/cpudeadline.h
> index 020039b..4a10a65 100644
> --- a/kernel/sched/cpudeadline.h
> +++ b/kernel/sched/cpudeadline.h
> @@ -24,6 +24,7 @@ int cpudl_find(struct cpudl *cp, struct task_struct *p,
> struct cpumask *later_mask);
> void cpudl_set(struct cpudl *cp, int cpu, u64 dl, int is_valid);
> int cpudl_init(struct cpudl *cp);
> +void cpudl_set_freecpu(struct cpudl *cp, int cpu);
> void cpudl_cleanup(struct cpudl *cp);
> #endif /* CONFIG_SMP */
>
> diff --git a/kernel/sched/deadline.c b/kernel/sched/deadline.c
> index 5285332..bd83272 100644
> --- a/kernel/sched/deadline.c
> +++ b/kernel/sched/deadline.c
> @@ -1187,9 +1187,6 @@ static int find_later_rq(struct task_struct *task)
> * We have to consider system topology and task affinity
> * first, then we can look for a suitable cpu.
> */
> - cpumask_copy(later_mask, task_rq(task)->rd->span);
> - cpumask_and(later_mask, later_mask, cpu_active_mask);
> - cpumask_and(later_mask, later_mask, &task->cpus_allowed);
> best_cpu = cpudl_find(&task_rq(task)->rd->cpudl,
> task, later_mask);
> if (best_cpu == -1)
On 20 November 2014 00:24, Steven Rostedt <[email protected]> wrote:
> On Wed, 19 Nov 2014 23:46:21 +0800
> "pang.xunlei" <[email protected]> wrote:
>
>> The call site of cpudl_find() in check_preempt_equal_dl() doesn't
>> use later_mask, so add this extra argument to distinquish the case.
>>
>> Signed-off-by: pang.xunlei <[email protected]>
>> ---
>> kernel/sched/cpudeadline.c | 6 ++++--
>> kernel/sched/cpudeadline.h | 2 +-
>> kernel/sched/deadline.c | 6 +++---
>> 3 files changed, 8 insertions(+), 6 deletions(-)
>>
>> diff --git a/kernel/sched/cpudeadline.c b/kernel/sched/cpudeadline.c
>> index c01b3aa..3047846 100644
>> --- a/kernel/sched/cpudeadline.c
>> +++ b/kernel/sched/cpudeadline.c
>> @@ -98,11 +98,12 @@ static inline int cpudl_maximum(struct cpudl *cp)
>> * @cp: the cpudl max-heap context
>> * @p: the task
>> * @later_mask: a mask to fill in with the selected CPUs (not NULL)
>> + * @set_flag: indicate if later_mask should be set
>> *
>> * Returns: int - best CPU (heap maximum if suitable)
>> */
>> int cpudl_find(struct cpudl *cp, struct task_struct *p,
>> - struct cpumask *later_mask)
>> + struct cpumask *later_mask, int set_flag)
>
> set_flag should be a bool type.
>
>> {
>> int best_cpu = -1;
>> const struct sched_dl_entity *dl_se = &p->dl;
>> @@ -114,7 +115,8 @@ int cpudl_find(struct cpudl *cp, struct task_struct *p,
>> } else if (cpumask_test_cpu(cpudl_maximum(cp), &p->cpus_allowed) &&
>> dl_time_before(dl_se->deadline, cp->elements[0].dl)) {
>> best_cpu = cpudl_maximum(cp);
>> - cpumask_set_cpu(best_cpu, later_mask);
>> + if (set_flag)
>> + cpumask_set_cpu(best_cpu, later_mask);
>
> I'm not sure this is worth it. cpumask_set_cpu() is rather efficient.
HI Steve,
Thanks for your commenting, I've rethinked this a bit.
We can do a little trick with its return value, then could avoid this
extra cpumask_set_cpu() without this extra set_flag:
1) define macros for the return values of cpudl_find(), like:
#define CPUDL_FIND_NONE -2 /* no available cpus */
#define CPUDL_FIND_CPUMASK -1 /* available cpus in later_mask */
then, with the return value >=0, means it returns the only one available cpu.
2) In the leg of "if", it can just return CPUDL_FIND_CPUMASK, as we
want to select the best_cpu in find_later_rq().
In the leg of "else if", just returns cpudl_maximum(cp), apparently
there is no need to set the later_mask, since we will definitely
select this cpu as the best_cpu in find_later_rq() .
int cpudl_find(struct cpudl *cp, struct task_struct *p,
struct cpumask *later_mask)
{
const struct sched_dl_entity *dl_se = &p->dl;
cpumask_and(later_mask, &p->cpus_allowed, &p->cpus_allowed);
if (cpumask_and(later_mask, later_mask, cp->free_cpus)) {
return CPUDL_FIND_CPUMASK;
} else if (cpumask_test_cpu(cpudl_maximum(cp), &p->cpus_allowed) &&
dl_time_before(dl_se->deadline, cp->elements[0].dl))
int cpu;
cpu = cpudl_maximum(cp);
WARN_ON(!cpu_present(cpu));
return cpu;
}
out:
return CPUDL_FIND_NONE;
}
Thus, in find_later_rq() we can change the call site code like:
best_cpu = cpudl_find(&task_rq(task)->rd->cpudl, task,
later_mask);
if (best_cpu == CPUDL_FIND_NONE)
return -1;
if (best_cpu != CPUDL_FIND_CPUMASK)
return best_cpu;
/* adjust the following code as that in RT find_lowest_rq(), omit here... */
What's your view about this?
Thanks,
Xunlei
>
>> }
>>
>> out:
>
On 20 November 2014 22:58, pang.xunlei <[email protected]> wrote:
> On 20 November 2014 00:24, Steven Rostedt <[email protected]> wrote:
>> On Wed, 19 Nov 2014 23:46:21 +0800
>> "pang.xunlei" <[email protected]> wrote:
>>
>>> The call site of cpudl_find() in check_preempt_equal_dl() doesn't
>>> use later_mask, so add this extra argument to distinquish the case.
>>>
>>> Signed-off-by: pang.xunlei <[email protected]>
>>> ---
>>> kernel/sched/cpudeadline.c | 6 ++++--
>>> kernel/sched/cpudeadline.h | 2 +-
>>> kernel/sched/deadline.c | 6 +++---
>>> 3 files changed, 8 insertions(+), 6 deletions(-)
>>>
>>> diff --git a/kernel/sched/cpudeadline.c b/kernel/sched/cpudeadline.c
>>> index c01b3aa..3047846 100644
>>> --- a/kernel/sched/cpudeadline.c
>>> +++ b/kernel/sched/cpudeadline.c
>>> @@ -98,11 +98,12 @@ static inline int cpudl_maximum(struct cpudl *cp)
>>> * @cp: the cpudl max-heap context
>>> * @p: the task
>>> * @later_mask: a mask to fill in with the selected CPUs (not NULL)
>>> + * @set_flag: indicate if later_mask should be set
>>> *
>>> * Returns: int - best CPU (heap maximum if suitable)
>>> */
>>> int cpudl_find(struct cpudl *cp, struct task_struct *p,
>>> - struct cpumask *later_mask)
>>> + struct cpumask *later_mask, int set_flag)
>>
>> set_flag should be a bool type.
>>
>>> {
>>> int best_cpu = -1;
>>> const struct sched_dl_entity *dl_se = &p->dl;
>>> @@ -114,7 +115,8 @@ int cpudl_find(struct cpudl *cp, struct task_struct *p,
>>> } else if (cpumask_test_cpu(cpudl_maximum(cp), &p->cpus_allowed) &&
>>> dl_time_before(dl_se->deadline, cp->elements[0].dl)) {
>>> best_cpu = cpudl_maximum(cp);
>>> - cpumask_set_cpu(best_cpu, later_mask);
>>> + if (set_flag)
>>> + cpumask_set_cpu(best_cpu, later_mask);
>>
>> I'm not sure this is worth it. cpumask_set_cpu() is rather efficient.
> HI Steve,
>
> Thanks for your commenting, I've rethinked this a bit.
> We can do a little trick with its return value, then could avoid this
> extra cpumask_set_cpu() without this extra set_flag:
> 1) define macros for the return values of cpudl_find(), like:
> #define CPUDL_FIND_NONE -2 /* no available cpus */
> #define CPUDL_FIND_CPUMASK -1 /* available cpus in later_mask */
>
> then, with the return value >=0, means it returns the only one available cpu.
>
> 2) In the leg of "if", it can just return CPUDL_FIND_CPUMASK, as we
> want to select the best_cpu in find_later_rq().
> In the leg of "else if", just returns cpudl_maximum(cp), apparently
> there is no need to set the later_mask, since we will definitely
> select this cpu as the best_cpu in find_later_rq() .
>
> int cpudl_find(struct cpudl *cp, struct task_struct *p,
> struct cpumask *later_mask)
> {
> const struct sched_dl_entity *dl_se = &p->dl;
>
> cpumask_and(later_mask, &p->cpus_allowed, &p->cpus_allowed);
Apologies for this typo, it should be:
cpumask_and(later_mask, cpu_active_mask, &p->cpus_allowed);
> if (cpumask_and(later_mask, later_mask, cp->free_cpus)) {
> return CPUDL_FIND_CPUMASK;
> } else if (cpumask_test_cpu(cpudl_maximum(cp), &p->cpus_allowed) &&
> dl_time_before(dl_se->deadline, cp->elements[0].dl))
> int cpu;
>
> cpu = cpudl_maximum(cp);
> WARN_ON(!cpu_present(cpu));
> return cpu;
> }
>
> out:
also delete this lable.
>
> return CPUDL_FIND_NONE;
> }
>
> Thus, in find_later_rq() we can change the call site code like:
> best_cpu = cpudl_find(&task_rq(task)->rd->cpudl, task,
> later_mask);
> if (best_cpu == CPUDL_FIND_NONE)
> return -1;
> if (best_cpu != CPUDL_FIND_CPUMASK)
> return best_cpu;
>
> /* adjust the following code as that in RT find_lowest_rq(), omit here... */
>
> What's your view about this?
>
> Thanks,
> Xunlei
>>
>>> }
>>>
>>> out:
>>