[PATCH -next v5 0/3] some optimization for cpuset

[PATCH -next v5 0/3] some optimization for cpuset

[Chen Ridong]

From: Chen Ridong <chenridong@huawei.com>

This patch series contains several cpuset improvements:

1. Decouple cpuset and tmpmasks allocation/freeing.
2. Add cpuset_full_[un]lock helpers.

---
v5:
 - add return tag comment for patch 2.

v4:
 - update the commit message for patch 3.

v3:
 - fix typos and comment errors.
 - rename cpus_read_cpuset_[un]lock to cpuset_full_[un]lock

v2:
 - dropped guard helper approach, nusing new helper instead.
 - added patches for decoupling cpuset/tmpmasks allocation


Chen Ridong (3):
  cpuset: decouple tmpmasks and cpumasks freeing in cgroup
  cpuset: separate tmpmasks and cpuset allocation logic
  cpuset: add helpers for cpus read and cpuset_mutex locks

 kernel/cgroup/cpuset-internal.h |   2 +
 kernel/cgroup/cpuset-v1.c       |  12 +-
 kernel/cgroup/cpuset.c          | 219 +++++++++++++++++---------------
 3 files changed, 122 insertions(+), 111 deletions(-)

-- 
2.34.1

[PATCH -next v5 1/3] cpuset: decouple tmpmasks and cpumasks freeing in cgroup

[Chen Ridong]

From: Chen Ridong <chenridong@huawei.com>

Currently, free_cpumasks() can free both tmpmasks and cpumasks of a cpuset
(cs). However, these two operations are not logically coupled. To improve
code clarity:
1. Move cpumask freeing to free_cpuset()
2. Rename free_cpumasks() to free_tmpmasks()

This change enforces the single responsibility principle.

Signed-off-by: Chen Ridong <chenridong@huawei.com>
Reviewed-by: Waiman Long <longman@redhat.com>
---
 kernel/cgroup/cpuset.c | 32 +++++++++++++-------------------
 1 file changed, 13 insertions(+), 19 deletions(-)

diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
index 3466ebbf1016..aebda14cc67f 100644
--- a/kernel/cgroup/cpuset.c
+++ b/kernel/cgroup/cpuset.c
@@ -459,23 +459,14 @@ static inline int alloc_cpumasks(struct cpuset *cs, struct tmpmasks *tmp)
 }
 
 /**
- * free_cpumasks - free cpumasks in a tmpmasks structure
- * @cs:  the cpuset that have cpumasks to be free.
+ * free_tmpmasks - free cpumasks in a tmpmasks structure
  * @tmp: the tmpmasks structure pointer
  */
-static inline void free_cpumasks(struct cpuset *cs, struct tmpmasks *tmp)
+static inline void free_tmpmasks(struct tmpmasks *tmp)
 {
-	if (cs) {
-		free_cpumask_var(cs->cpus_allowed);
-		free_cpumask_var(cs->effective_cpus);
-		free_cpumask_var(cs->effective_xcpus);
-		free_cpumask_var(cs->exclusive_cpus);
-	}
-	if (tmp) {
-		free_cpumask_var(tmp->new_cpus);
-		free_cpumask_var(tmp->addmask);
-		free_cpumask_var(tmp->delmask);
-	}
+	free_cpumask_var(tmp->new_cpus);
+	free_cpumask_var(tmp->addmask);
+	free_cpumask_var(tmp->delmask);
 }
 
 /**
@@ -508,7 +499,10 @@ static struct cpuset *alloc_trial_cpuset(struct cpuset *cs)
  */
 static inline void free_cpuset(struct cpuset *cs)
 {
-	free_cpumasks(cs, NULL);
+	free_cpumask_var(cs->cpus_allowed);
+	free_cpumask_var(cs->effective_cpus);
+	free_cpumask_var(cs->effective_xcpus);
+	free_cpumask_var(cs->exclusive_cpus);
 	kfree(cs);
 }
 
@@ -2427,7 +2421,7 @@ static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
 	if (cs->partition_root_state)
 		update_partition_sd_lb(cs, old_prs);
 out_free:
-	free_cpumasks(NULL, &tmp);
+	free_tmpmasks(&tmp);
 	return retval;
 }
 
@@ -2530,7 +2524,7 @@ static int update_exclusive_cpumask(struct cpuset *cs, struct cpuset *trialcs,
 	if (cs->partition_root_state)
 		update_partition_sd_lb(cs, old_prs);
 
-	free_cpumasks(NULL, &tmp);
+	free_tmpmasks(&tmp);
 	return 0;
 }
 
@@ -2983,7 +2977,7 @@ static int update_prstate(struct cpuset *cs, int new_prs)
 	notify_partition_change(cs, old_prs);
 	if (force_sd_rebuild)
 		rebuild_sched_domains_locked();
-	free_cpumasks(NULL, &tmpmask);
+	free_tmpmasks(&tmpmask);
 	return 0;
 }
 
@@ -4006,7 +4000,7 @@ static void cpuset_handle_hotplug(void)
 	if (force_sd_rebuild)
 		rebuild_sched_domains_cpuslocked();
 
-	free_cpumasks(NULL, ptmp);
+	free_tmpmasks(ptmp);
 }
 
 void cpuset_update_active_cpus(void)
-- 
2.34.1

[PATCH -next v5 2/3] cpuset: separate tmpmasks and cpuset allocation logic

[Chen Ridong]

From: Chen Ridong <chenridong@huawei.com>

The original alloc_cpumasks() served dual purposes: allocating cpumasks
for both temporary masks (tmpmasks) and cpuset structures. This patch:

1. Decouples these allocation paths for better code clarity
2. Introduces dedicated alloc_tmpmasks() and dup_or_alloc_cpuset()
   functions
3. Maintains symmetric pairing:
   - alloc_tmpmasks() ↔ free_tmpmasks()
   - dup_or_alloc_cpuset() ↔ free_cpuset()

Signed-off-by: Chen Ridong <chenridong@huawei.com>
---
 kernel/cgroup/cpuset.c | 127 ++++++++++++++++++++++-------------------
 1 file changed, 69 insertions(+), 58 deletions(-)

diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
index aebda14cc67f..7b0b81c835bf 100644
--- a/kernel/cgroup/cpuset.c
+++ b/kernel/cgroup/cpuset.c
@@ -411,51 +411,47 @@ static void guarantee_online_mems(struct cpuset *cs, nodemask_t *pmask)
 }
 
 /**
- * alloc_cpumasks - allocate three cpumasks for cpuset
- * @cs:  the cpuset that have cpumasks to be allocated.
- * @tmp: the tmpmasks structure pointer
+ * alloc_cpumasks - Allocate an array of cpumask variables
+ * @pmasks: Pointer to array of cpumask_var_t pointers
+ * @size: Number of cpumasks to allocate
  * Return: 0 if successful, -ENOMEM otherwise.
  *
- * Only one of the two input arguments should be non-NULL.
+ * Allocates @size cpumasks and initializes them to empty. Returns 0 on
+ * success, -ENOMEM on allocation failure. On failure, any previously
+ * allocated cpumasks are freed.
  */
-static inline int alloc_cpumasks(struct cpuset *cs, struct tmpmasks *tmp)
+static inline int alloc_cpumasks(cpumask_var_t *pmasks[], u32 size)
 {
-	cpumask_var_t *pmask1, *pmask2, *pmask3, *pmask4;
+	int i;
 
-	if (cs) {
-		pmask1 = &cs->cpus_allowed;
-		pmask2 = &cs->effective_cpus;
-		pmask3 = &cs->effective_xcpus;
-		pmask4 = &cs->exclusive_cpus;
-	} else {
-		pmask1 = &tmp->new_cpus;
-		pmask2 = &tmp->addmask;
-		pmask3 = &tmp->delmask;
-		pmask4 = NULL;
+	for (i = 0; i < size; i++) {
+		if (!zalloc_cpumask_var(pmasks[i], GFP_KERNEL)) {
+			while (--i >= 0)
+				free_cpumask_var(*pmasks[i]);
+			return -ENOMEM;
+		}
 	}
-
-	if (!zalloc_cpumask_var(pmask1, GFP_KERNEL))
-		return -ENOMEM;
-
-	if (!zalloc_cpumask_var(pmask2, GFP_KERNEL))
-		goto free_one;
-
-	if (!zalloc_cpumask_var(pmask3, GFP_KERNEL))
-		goto free_two;
-
-	if (pmask4 && !zalloc_cpumask_var(pmask4, GFP_KERNEL))
-		goto free_three;
-
-
 	return 0;
+}
 
-free_three:
-	free_cpumask_var(*pmask3);
-free_two:
-	free_cpumask_var(*pmask2);
-free_one:
-	free_cpumask_var(*pmask1);
-	return -ENOMEM;
+/**
+ * alloc_tmpmasks - Allocate temporary cpumasks for cpuset operations.
+ * @tmp: Pointer to tmpmasks structure to populate
+ * Return: 0 on success, -ENOMEM on allocation failure
+ */
+static inline int alloc_tmpmasks(struct tmpmasks *tmp)
+{
+	/*
+	 * Array of pointers to the three cpumask_var_t fields in tmpmasks.
+	 * Note: Array size must match actual number of masks (3)
+	 */
+	cpumask_var_t *pmask[3] = {
+		&tmp->new_cpus,
+		&tmp->addmask,
+		&tmp->delmask
+	};
+
+	return alloc_cpumasks(pmask, ARRAY_SIZE(pmask));
 }
 
 /**
@@ -470,26 +466,46 @@ static inline void free_tmpmasks(struct tmpmasks *tmp)
 }
 
 /**
- * alloc_trial_cpuset - allocate a trial cpuset
- * @cs: the cpuset that the trial cpuset duplicates
+ * dup_or_alloc_cpuset - Duplicate or allocate a new cpuset
+ * @cs: Source cpuset to duplicate (NULL for a fresh allocation)
+ *
+ * Creates a new cpuset by either:
+ * 1. Duplicating an existing cpuset (if @cs is non-NULL), or
+ * 2. Allocating a fresh cpuset with zero-initialized masks (if @cs is NULL)
+ *
+ * Return: Pointer to newly allocated cpuset on success, NULL on failure
  */
-static struct cpuset *alloc_trial_cpuset(struct cpuset *cs)
+static struct cpuset *dup_or_alloc_cpuset(struct cpuset *cs)
 {
 	struct cpuset *trial;
 
-	trial = kmemdup(cs, sizeof(*cs), GFP_KERNEL);
+	/* Allocate base structure */
+	trial = cs ? kmemdup(cs, sizeof(*cs), GFP_KERNEL) :
+		     kzalloc(sizeof(*cs), GFP_KERNEL);
 	if (!trial)
 		return NULL;
 
-	if (alloc_cpumasks(trial, NULL)) {
+	/* Setup cpumask pointer array */
+	cpumask_var_t *pmask[4] = {
+		&trial->cpus_allowed,
+		&trial->effective_cpus,
+		&trial->effective_xcpus,
+		&trial->exclusive_cpus
+	};
+
+	if (alloc_cpumasks(pmask, ARRAY_SIZE(pmask))) {
 		kfree(trial);
 		return NULL;
 	}
 
-	cpumask_copy(trial->cpus_allowed, cs->cpus_allowed);
-	cpumask_copy(trial->effective_cpus, cs->effective_cpus);
-	cpumask_copy(trial->effective_xcpus, cs->effective_xcpus);
-	cpumask_copy(trial->exclusive_cpus, cs->exclusive_cpus);
+	/* Copy masks if duplicating */
+	if (cs) {
+		cpumask_copy(trial->cpus_allowed, cs->cpus_allowed);
+		cpumask_copy(trial->effective_cpus, cs->effective_cpus);
+		cpumask_copy(trial->effective_xcpus, cs->effective_xcpus);
+		cpumask_copy(trial->exclusive_cpus, cs->exclusive_cpus);
+	}
+
 	return trial;
 }
 
@@ -2332,7 +2348,7 @@ static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
 	if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed))
 		return 0;
 
-	if (alloc_cpumasks(NULL, &tmp))
+	if (alloc_tmpmasks(&tmp))
 		return -ENOMEM;
 
 	if (old_prs) {
@@ -2476,7 +2492,7 @@ static int update_exclusive_cpumask(struct cpuset *cs, struct cpuset *trialcs,
 	if (retval)
 		return retval;
 
-	if (alloc_cpumasks(NULL, &tmp))
+	if (alloc_tmpmasks(&tmp))
 		return -ENOMEM;
 
 	if (old_prs) {
@@ -2820,7 +2836,7 @@ int cpuset_update_flag(cpuset_flagbits_t bit, struct cpuset *cs,
 	int spread_flag_changed;
 	int err;
 
-	trialcs = alloc_trial_cpuset(cs);
+	trialcs = dup_or_alloc_cpuset(cs);
 	if (!trialcs)
 		return -ENOMEM;
 
@@ -2881,7 +2897,7 @@ static int update_prstate(struct cpuset *cs, int new_prs)
 	if (new_prs && is_prs_invalid(old_prs))
 		old_prs = PRS_MEMBER;
 
-	if (alloc_cpumasks(NULL, &tmpmask))
+	if (alloc_tmpmasks(&tmpmask))
 		return -ENOMEM;
 
 	err = update_partition_exclusive_flag(cs, new_prs);
@@ -3223,7 +3239,7 @@ ssize_t cpuset_write_resmask(struct kernfs_open_file *of,
 	if (!is_cpuset_online(cs))
 		goto out_unlock;
 
-	trialcs = alloc_trial_cpuset(cs);
+	trialcs = dup_or_alloc_cpuset(cs);
 	if (!trialcs) {
 		retval = -ENOMEM;
 		goto out_unlock;
@@ -3456,15 +3472,10 @@ cpuset_css_alloc(struct cgroup_subsys_state *parent_css)
 	if (!parent_css)
 		return &top_cpuset.css;
 
-	cs = kzalloc(sizeof(*cs), GFP_KERNEL);
+	cs = dup_or_alloc_cpuset(NULL);
 	if (!cs)
 		return ERR_PTR(-ENOMEM);
 
-	if (alloc_cpumasks(cs, NULL)) {
-		kfree(cs);
-		return ERR_PTR(-ENOMEM);
-	}
-
 	__set_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
 	fmeter_init(&cs->fmeter);
 	cs->relax_domain_level = -1;
@@ -3920,7 +3931,7 @@ static void cpuset_handle_hotplug(void)
 	bool on_dfl = is_in_v2_mode();
 	struct tmpmasks tmp, *ptmp = NULL;
 
-	if (on_dfl && !alloc_cpumasks(NULL, &tmp))
+	if (on_dfl && !alloc_tmpmasks(&tmp))
 		ptmp = &tmp;
 
 	lockdep_assert_cpus_held();
-- 
2.34.1

Re: [PATCH -next v5 2/3] cpuset: separate tmpmasks and cpuset allocation logic

[Waiman Long]

On 8/24/25 11:23 PM, Chen Ridong wrote:
> From: Chen Ridong <chenridong@huawei.com>
>
> The original alloc_cpumasks() served dual purposes: allocating cpumasks
> for both temporary masks (tmpmasks) and cpuset structures. This patch:
>
> 1. Decouples these allocation paths for better code clarity
> 2. Introduces dedicated alloc_tmpmasks() and dup_or_alloc_cpuset()
>     functions
> 3. Maintains symmetric pairing:
>     - alloc_tmpmasks() ↔ free_tmpmasks()
>     - dup_or_alloc_cpuset() ↔ free_cpuset()
>
> Signed-off-by: Chen Ridong <chenridong@huawei.com>
> ---
>   kernel/cgroup/cpuset.c | 127 ++++++++++++++++++++++-------------------
>   1 file changed, 69 insertions(+), 58 deletions(-)
>
> diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
> index aebda14cc67f..7b0b81c835bf 100644
> --- a/kernel/cgroup/cpuset.c
> +++ b/kernel/cgroup/cpuset.c
> @@ -411,51 +411,47 @@ static void guarantee_online_mems(struct cpuset *cs, nodemask_t *pmask)
>   }
>   
>   /**
> - * alloc_cpumasks - allocate three cpumasks for cpuset
> - * @cs:  the cpuset that have cpumasks to be allocated.
> - * @tmp: the tmpmasks structure pointer
> + * alloc_cpumasks - Allocate an array of cpumask variables
> + * @pmasks: Pointer to array of cpumask_var_t pointers
> + * @size: Number of cpumasks to allocate
>    * Return: 0 if successful, -ENOMEM otherwise.
>    *
> - * Only one of the two input arguments should be non-NULL.
> + * Allocates @size cpumasks and initializes them to empty. Returns 0 on
> + * success, -ENOMEM on allocation failure. On failure, any previously
> + * allocated cpumasks are freed.
>    */
> -static inline int alloc_cpumasks(struct cpuset *cs, struct tmpmasks *tmp)
> +static inline int alloc_cpumasks(cpumask_var_t *pmasks[], u32 size)
>   {
> -	cpumask_var_t *pmask1, *pmask2, *pmask3, *pmask4;
> +	int i;
>   
> -	if (cs) {
> -		pmask1 = &cs->cpus_allowed;
> -		pmask2 = &cs->effective_cpus;
> -		pmask3 = &cs->effective_xcpus;
> -		pmask4 = &cs->exclusive_cpus;
> -	} else {
> -		pmask1 = &tmp->new_cpus;
> -		pmask2 = &tmp->addmask;
> -		pmask3 = &tmp->delmask;
> -		pmask4 = NULL;
> +	for (i = 0; i < size; i++) {
> +		if (!zalloc_cpumask_var(pmasks[i], GFP_KERNEL)) {
> +			while (--i >= 0)
> +				free_cpumask_var(*pmasks[i]);
> +			return -ENOMEM;
> +		}
>   	}
> -
> -	if (!zalloc_cpumask_var(pmask1, GFP_KERNEL))
> -		return -ENOMEM;
> -
> -	if (!zalloc_cpumask_var(pmask2, GFP_KERNEL))
> -		goto free_one;
> -
> -	if (!zalloc_cpumask_var(pmask3, GFP_KERNEL))
> -		goto free_two;
> -
> -	if (pmask4 && !zalloc_cpumask_var(pmask4, GFP_KERNEL))
> -		goto free_three;
> -
> -
>   	return 0;
> +}
>   
> -free_three:
> -	free_cpumask_var(*pmask3);
> -free_two:
> -	free_cpumask_var(*pmask2);
> -free_one:
> -	free_cpumask_var(*pmask1);
> -	return -ENOMEM;
> +/**
> + * alloc_tmpmasks - Allocate temporary cpumasks for cpuset operations.
> + * @tmp: Pointer to tmpmasks structure to populate
> + * Return: 0 on success, -ENOMEM on allocation failure
> + */
> +static inline int alloc_tmpmasks(struct tmpmasks *tmp)
> +{
> +	/*
> +	 * Array of pointers to the three cpumask_var_t fields in tmpmasks.
> +	 * Note: Array size must match actual number of masks (3)
> +	 */
> +	cpumask_var_t *pmask[3] = {
> +		&tmp->new_cpus,
> +		&tmp->addmask,
> +		&tmp->delmask
> +	};
> +
> +	return alloc_cpumasks(pmask, ARRAY_SIZE(pmask));
>   }
>   
>   /**
> @@ -470,26 +466,46 @@ static inline void free_tmpmasks(struct tmpmasks *tmp)
>   }
>   
>   /**
> - * alloc_trial_cpuset - allocate a trial cpuset
> - * @cs: the cpuset that the trial cpuset duplicates
> + * dup_or_alloc_cpuset - Duplicate or allocate a new cpuset
> + * @cs: Source cpuset to duplicate (NULL for a fresh allocation)
> + *
> + * Creates a new cpuset by either:
> + * 1. Duplicating an existing cpuset (if @cs is non-NULL), or
> + * 2. Allocating a fresh cpuset with zero-initialized masks (if @cs is NULL)
> + *
> + * Return: Pointer to newly allocated cpuset on success, NULL on failure
>    */
> -static struct cpuset *alloc_trial_cpuset(struct cpuset *cs)
> +static struct cpuset *dup_or_alloc_cpuset(struct cpuset *cs)
>   {
>   	struct cpuset *trial;
>   
> -	trial = kmemdup(cs, sizeof(*cs), GFP_KERNEL);
> +	/* Allocate base structure */
> +	trial = cs ? kmemdup(cs, sizeof(*cs), GFP_KERNEL) :
> +		     kzalloc(sizeof(*cs), GFP_KERNEL);
>   	if (!trial)
>   		return NULL;
>   
> -	if (alloc_cpumasks(trial, NULL)) {
> +	/* Setup cpumask pointer array */
> +	cpumask_var_t *pmask[4] = {
> +		&trial->cpus_allowed,
> +		&trial->effective_cpus,
> +		&trial->effective_xcpus,
> +		&trial->exclusive_cpus
> +	};
> +
> +	if (alloc_cpumasks(pmask, ARRAY_SIZE(pmask))) {
>   		kfree(trial);
>   		return NULL;
>   	}
>   
> -	cpumask_copy(trial->cpus_allowed, cs->cpus_allowed);
> -	cpumask_copy(trial->effective_cpus, cs->effective_cpus);
> -	cpumask_copy(trial->effective_xcpus, cs->effective_xcpus);
> -	cpumask_copy(trial->exclusive_cpus, cs->exclusive_cpus);
> +	/* Copy masks if duplicating */
> +	if (cs) {
> +		cpumask_copy(trial->cpus_allowed, cs->cpus_allowed);
> +		cpumask_copy(trial->effective_cpus, cs->effective_cpus);
> +		cpumask_copy(trial->effective_xcpus, cs->effective_xcpus);
> +		cpumask_copy(trial->exclusive_cpus, cs->exclusive_cpus);
> +	}
> +
>   	return trial;
>   }
>   
> @@ -2332,7 +2348,7 @@ static int update_cpumask(struct cpuset *cs, struct cpuset *trialcs,
>   	if (cpumask_equal(cs->cpus_allowed, trialcs->cpus_allowed))
>   		return 0;
>   
> -	if (alloc_cpumasks(NULL, &tmp))
> +	if (alloc_tmpmasks(&tmp))
>   		return -ENOMEM;
>   
>   	if (old_prs) {
> @@ -2476,7 +2492,7 @@ static int update_exclusive_cpumask(struct cpuset *cs, struct cpuset *trialcs,
>   	if (retval)
>   		return retval;
>   
> -	if (alloc_cpumasks(NULL, &tmp))
> +	if (alloc_tmpmasks(&tmp))
>   		return -ENOMEM;
>   
>   	if (old_prs) {
> @@ -2820,7 +2836,7 @@ int cpuset_update_flag(cpuset_flagbits_t bit, struct cpuset *cs,
>   	int spread_flag_changed;
>   	int err;
>   
> -	trialcs = alloc_trial_cpuset(cs);
> +	trialcs = dup_or_alloc_cpuset(cs);
>   	if (!trialcs)
>   		return -ENOMEM;
>   
> @@ -2881,7 +2897,7 @@ static int update_prstate(struct cpuset *cs, int new_prs)
>   	if (new_prs && is_prs_invalid(old_prs))
>   		old_prs = PRS_MEMBER;
>   
> -	if (alloc_cpumasks(NULL, &tmpmask))
> +	if (alloc_tmpmasks(&tmpmask))
>   		return -ENOMEM;
>   
>   	err = update_partition_exclusive_flag(cs, new_prs);
> @@ -3223,7 +3239,7 @@ ssize_t cpuset_write_resmask(struct kernfs_open_file *of,
>   	if (!is_cpuset_online(cs))
>   		goto out_unlock;
>   
> -	trialcs = alloc_trial_cpuset(cs);
> +	trialcs = dup_or_alloc_cpuset(cs);
>   	if (!trialcs) {
>   		retval = -ENOMEM;
>   		goto out_unlock;
> @@ -3456,15 +3472,10 @@ cpuset_css_alloc(struct cgroup_subsys_state *parent_css)
>   	if (!parent_css)
>   		return &top_cpuset.css;
>   
> -	cs = kzalloc(sizeof(*cs), GFP_KERNEL);
> +	cs = dup_or_alloc_cpuset(NULL);
>   	if (!cs)
>   		return ERR_PTR(-ENOMEM);
>   
> -	if (alloc_cpumasks(cs, NULL)) {
> -		kfree(cs);
> -		return ERR_PTR(-ENOMEM);
> -	}
> -
>   	__set_bit(CS_SCHED_LOAD_BALANCE, &cs->flags);
>   	fmeter_init(&cs->fmeter);
>   	cs->relax_domain_level = -1;
> @@ -3920,7 +3931,7 @@ static void cpuset_handle_hotplug(void)
>   	bool on_dfl = is_in_v2_mode();
>   	struct tmpmasks tmp, *ptmp = NULL;
>   
> -	if (on_dfl && !alloc_cpumasks(NULL, &tmp))
> +	if (on_dfl && !alloc_tmpmasks(&tmp))
>   		ptmp = &tmp;
>   
>   	lockdep_assert_cpus_held();
Reviewed-by: Waiman Long <longman@redhat.com>

[PATCH -next v5 3/3] cpuset: add helpers for cpus read and cpuset_mutex locks

[Chen Ridong]

From: Chen Ridong <chenridong@huawei.com>

cpuset: add helpers for cpus_read_lock and cpuset_mutex locks.

Replace repetitive locking patterns with new helpers:
- cpuset_full_lock()
- cpuset_full_unlock()

This makes the code cleaner and ensures consistent lock ordering.

Signed-off-by: Chen Ridong <chenridong@huawei.com>
Reviewed-by: Waiman Long <longman@redhat.com>
---
 kernel/cgroup/cpuset-internal.h |  2 ++
 kernel/cgroup/cpuset-v1.c       | 12 +++----
 kernel/cgroup/cpuset.c          | 60 +++++++++++++++++++--------------
 3 files changed, 40 insertions(+), 34 deletions(-)

diff --git a/kernel/cgroup/cpuset-internal.h b/kernel/cgroup/cpuset-internal.h
index 75b3aef39231..337608f408ce 100644
--- a/kernel/cgroup/cpuset-internal.h
+++ b/kernel/cgroup/cpuset-internal.h
@@ -276,6 +276,8 @@ int cpuset_update_flag(cpuset_flagbits_t bit, struct cpuset *cs, int turning_on)
 ssize_t cpuset_write_resmask(struct kernfs_open_file *of,
 				    char *buf, size_t nbytes, loff_t off);
 int cpuset_common_seq_show(struct seq_file *sf, void *v);
+void cpuset_full_lock(void);
+void cpuset_full_unlock(void);
 
 /*
  * cpuset-v1.c
diff --git a/kernel/cgroup/cpuset-v1.c b/kernel/cgroup/cpuset-v1.c
index b69a7db67090..12e76774c75b 100644
--- a/kernel/cgroup/cpuset-v1.c
+++ b/kernel/cgroup/cpuset-v1.c
@@ -169,8 +169,7 @@ static int cpuset_write_s64(struct cgroup_subsys_state *css, struct cftype *cft,
 	cpuset_filetype_t type = cft->private;
 	int retval = -ENODEV;
 
-	cpus_read_lock();
-	cpuset_lock();
+	cpuset_full_lock();
 	if (!is_cpuset_online(cs))
 		goto out_unlock;
 
@@ -184,8 +183,7 @@ static int cpuset_write_s64(struct cgroup_subsys_state *css, struct cftype *cft,
 		break;
 	}
 out_unlock:
-	cpuset_unlock();
-	cpus_read_unlock();
+	cpuset_full_unlock();
 	return retval;
 }
 
@@ -454,8 +452,7 @@ static int cpuset_write_u64(struct cgroup_subsys_state *css, struct cftype *cft,
 	cpuset_filetype_t type = cft->private;
 	int retval = 0;
 
-	cpus_read_lock();
-	cpuset_lock();
+	cpuset_full_lock();
 	if (!is_cpuset_online(cs)) {
 		retval = -ENODEV;
 		goto out_unlock;
@@ -498,8 +495,7 @@ static int cpuset_write_u64(struct cgroup_subsys_state *css, struct cftype *cft,
 		break;
 	}
 out_unlock:
-	cpuset_unlock();
-	cpus_read_unlock();
+	cpuset_full_unlock();
 	return retval;
 }
 
diff --git a/kernel/cgroup/cpuset.c b/kernel/cgroup/cpuset.c
index 7b0b81c835bf..a78ccd11ce9b 100644
--- a/kernel/cgroup/cpuset.c
+++ b/kernel/cgroup/cpuset.c
@@ -250,6 +250,12 @@ static struct cpuset top_cpuset = {
 
 static DEFINE_MUTEX(cpuset_mutex);
 
+/**
+ * cpuset_lock - Acquire the global cpuset mutex
+ *
+ * This locks the global cpuset mutex to prevent modifications to cpuset
+ * hierarchy and configurations. This helper is not enough to make modification.
+ */
 void cpuset_lock(void)
 {
 	mutex_lock(&cpuset_mutex);
@@ -260,6 +266,24 @@ void cpuset_unlock(void)
 	mutex_unlock(&cpuset_mutex);
 }
 
+/**
+ * cpuset_full_lock - Acquire full protection for cpuset modification
+ *
+ * Takes both CPU hotplug read lock (cpus_read_lock()) and cpuset mutex
+ * to safely modify cpuset data.
+ */
+void cpuset_full_lock(void)
+{
+	cpus_read_lock();
+	mutex_lock(&cpuset_mutex);
+}
+
+void cpuset_full_unlock(void)
+{
+	mutex_unlock(&cpuset_mutex);
+	cpus_read_unlock();
+}
+
 static DEFINE_SPINLOCK(callback_lock);
 
 void cpuset_callback_lock_irq(void)
@@ -3234,8 +3258,7 @@ ssize_t cpuset_write_resmask(struct kernfs_open_file *of,
 	int retval = -ENODEV;
 
 	buf = strstrip(buf);
-	cpus_read_lock();
-	mutex_lock(&cpuset_mutex);
+	cpuset_full_lock();
 	if (!is_cpuset_online(cs))
 		goto out_unlock;
 
@@ -3264,8 +3287,7 @@ ssize_t cpuset_write_resmask(struct kernfs_open_file *of,
 	if (force_sd_rebuild)
 		rebuild_sched_domains_locked();
 out_unlock:
-	mutex_unlock(&cpuset_mutex);
-	cpus_read_unlock();
+	cpuset_full_unlock();
 	flush_workqueue(cpuset_migrate_mm_wq);
 	return retval ?: nbytes;
 }
@@ -3368,12 +3390,10 @@ static ssize_t cpuset_partition_write(struct kernfs_open_file *of, char *buf,
 	else
 		return -EINVAL;
 
-	cpus_read_lock();
-	mutex_lock(&cpuset_mutex);
+	cpuset_full_lock();
 	if (is_cpuset_online(cs))
 		retval = update_prstate(cs, val);
-	mutex_unlock(&cpuset_mutex);
-	cpus_read_unlock();
+	cpuset_full_unlock();
 	return retval ?: nbytes;
 }
 
@@ -3498,9 +3518,7 @@ static int cpuset_css_online(struct cgroup_subsys_state *css)
 	if (!parent)
 		return 0;
 
-	cpus_read_lock();
-	mutex_lock(&cpuset_mutex);
-
+	cpuset_full_lock();
 	if (is_spread_page(parent))
 		set_bit(CS_SPREAD_PAGE, &cs->flags);
 	if (is_spread_slab(parent))
@@ -3552,8 +3570,7 @@ static int cpuset_css_online(struct cgroup_subsys_state *css)
 	cpumask_copy(cs->effective_cpus, parent->cpus_allowed);
 	spin_unlock_irq(&callback_lock);
 out_unlock:
-	mutex_unlock(&cpuset_mutex);
-	cpus_read_unlock();
+	cpuset_full_unlock();
 	return 0;
 }
 
@@ -3568,16 +3585,12 @@ static void cpuset_css_offline(struct cgroup_subsys_state *css)
 {
 	struct cpuset *cs = css_cs(css);
 
-	cpus_read_lock();
-	mutex_lock(&cpuset_mutex);
-
+	cpuset_full_lock();
 	if (!cpuset_v2() && is_sched_load_balance(cs))
 		cpuset_update_flag(CS_SCHED_LOAD_BALANCE, cs, 0);
 
 	cpuset_dec();
-
-	mutex_unlock(&cpuset_mutex);
-	cpus_read_unlock();
+	cpuset_full_unlock();
 }
 
 /*
@@ -3589,16 +3602,11 @@ static void cpuset_css_killed(struct cgroup_subsys_state *css)
 {
 	struct cpuset *cs = css_cs(css);
 
-	cpus_read_lock();
-	mutex_lock(&cpuset_mutex);
-
+	cpuset_full_lock();
 	/* Reset valid partition back to member */
 	if (is_partition_valid(cs))
 		update_prstate(cs, PRS_MEMBER);
-
-	mutex_unlock(&cpuset_mutex);
-	cpus_read_unlock();
-
+	cpuset_full_unlock();
 }
 
 static void cpuset_css_free(struct cgroup_subsys_state *css)
-- 
2.34.1

Re: [PATCH -next v5 3/3] cpuset: add helpers for cpus read and cpuset_mutex locks

[Michal Koutný]

[-- Attachment #1: Type: text/plain, Size: 1083 bytes --]

(I wrote this yesterday before merging but I'm still sending it to give
my opinion ;-))

On Mon, Aug 25, 2025 at 03:23:52AM +0000, Chen Ridong <chenridong@huaweicloud.com> wrote:
> From: Chen Ridong <chenridong@huawei.com>
> 
> cpuset: add helpers for cpus_read_lock and cpuset_mutex locks.
> 
> Replace repetitive locking patterns with new helpers:
> - cpuset_full_lock()
> - cpuset_full_unlock()

I don't see many precedents elsewhere in the kernel for such naming
(like _lock and _full_lock()). Wouldn't it be more illustrative to have
cpuset_read_lock() and cpuset_write_lock()? (As I'm looking at current
users and your accompanying comments which could be substituted with
the more conventional naming.)

(Also if you decide going this direction, please mention commit
111cd11bbc548 ("sched/cpuset: Bring back cpuset_mutex") in the message
so that it doesn't tempt to do further changes.)


> This makes the code cleaner and ensures consistent lock ordering.

Lock guards anyone? (When you're touching this and seeking clean code.)

Thanks,
Michal

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 265 bytes --]

Re: [PATCH -next v5 3/3] cpuset: add helpers for cpus read and cpuset_mutex locks

[Waiman Long]

On 8/26/25 10:23 AM, Michal Koutný wrote:
> (I wrote this yesterday before merging but I'm still sending it to give
> my opinion ;-))
>
> On Mon, Aug 25, 2025 at 03:23:52AM +0000, Chen Ridong <chenridong@huaweicloud.com> wrote:
>> From: Chen Ridong <chenridong@huawei.com>
>>
>> cpuset: add helpers for cpus_read_lock and cpuset_mutex locks.
>>
>> Replace repetitive locking patterns with new helpers:
>> - cpuset_full_lock()
>> - cpuset_full_unlock()
> I don't see many precedents elsewhere in the kernel for such naming
> (like _lock and _full_lock()). Wouldn't it be more illustrative to have
> cpuset_read_lock() and cpuset_write_lock()? (As I'm looking at current
> users and your accompanying comments which could be substituted with
> the more conventional naming.)
Good naming is always an issue. Using cpuset_read_lock/cpuset_write_lock 
will be more confusing as the current locking scheme is exclusive.
> (Also if you decide going this direction, please mention commit
> 111cd11bbc548 ("sched/cpuset: Bring back cpuset_mutex") in the message
> so that it doesn't tempt to do further changes.)
>
>
>> This makes the code cleaner and ensures consistent lock ordering.
> Lock guards anyone? (When you're touching this and seeking clean code.)

Yes, I guess we can use lock guards here. You are welcome to send a 
patch to do that.

Cheers,
Longman


>
> Thanks,
> Michal

Re: [PATCH -next v5 3/3] cpuset: add helpers for cpus read and cpuset_mutex locks

[Chen Ridong]

On 2025/8/26 22:43, Waiman Long wrote:
> 
> On 8/26/25 10:23 AM, Michal Koutný wrote:
>> (I wrote this yesterday before merging but I'm still sending it to give
>> my opinion ;-))
>>
>> On Mon, Aug 25, 2025 at 03:23:52AM +0000, Chen Ridong <chenridong@huaweicloud.com> wrote:
>>> From: Chen Ridong <chenridong@huawei.com>
>>>
>>> cpuset: add helpers for cpus_read_lock and cpuset_mutex locks.
>>>
>>> Replace repetitive locking patterns with new helpers:
>>> - cpuset_full_lock()
>>> - cpuset_full_unlock()
>> I don't see many precedents elsewhere in the kernel for such naming
>> (like _lock and _full_lock()). Wouldn't it be more illustrative to have
>> cpuset_read_lock() and cpuset_write_lock()? (As I'm looking at current
>> users and your accompanying comments which could be substituted with
>> the more conventional naming.)
> Good naming is always an issue. Using cpuset_read_lock/cpuset_write_lock will be more confusing as
> the current locking scheme is exclusive.
>> (Also if you decide going this direction, please mention commit
>> 111cd11bbc548 ("sched/cpuset: Bring back cpuset_mutex") in the message
>> so that it doesn't tempt to do further changes.)
>>
>>
>>> This makes the code cleaner and ensures consistent lock ordering.
>> Lock guards anyone? (When you're touching this and seeking clean code.)
> 
> Yes, I guess we can use lock guards here. You are welcome to send a patch to do that.
> 

I attempted to define the cpuset_full_lock() macro, but the initial implementation was inconsistent
with our coding conventions.
Initial version:

#define cpuset_full_lock() \
  guard(cpus_read_lock)(); \
  guard(mutex)(&cpuset_mutex);

It was suggested to use a do-while construct for proper scoping. but it could not work if we define as:

#define cpuset_full_lock() \
 do { 			   \
  guard(cpus_read_lock)(); \
  guard(mutex)(&cpuset_mutex); \
 } while(0)

So I sent this patch version.

-- 
Best regards,
Ridong

Re: [PATCH -next v5 3/3] cpuset: add helpers for cpus read and cpuset_mutex locks

[Michal Koutný]

[-- Attachment #1: Type: text/plain, Size: 364 bytes --]

On Wed, Aug 27, 2025 at 02:23:17PM +0800, Chen Ridong <chenridong@huaweicloud.com> wrote:

> It was suggested to use a do-while construct for proper scoping. but it could not work if we define as:

Perhaps like this:
DEFINE_LOCK_GUARD_0(cpuset_full, cpuset_full_lock(), cpuset_full_unlock())

> So I sent this patch version.

No probs, it's a minor issue.

Michal

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 265 bytes --]

Re: [PATCH -next v5 0/3] some optimization for cpuset

[Tejun Heo]

On Mon, Aug 25, 2025 at 03:23:49AM +0000, Chen Ridong wrote:
> From: Chen Ridong <chenridong@huawei.com>
> 
> This patch series contains several cpuset improvements:
> 
> 1. Decouple cpuset and tmpmasks allocation/freeing.
> 2. Add cpuset_full_[un]lock helpers.

Applied 1-3 to cgroup/for-6.18.

Thanks.

-- 
tejun