[PATCH v2 0/6] sched/cache: Fixes for cache aware scheduling

[PATCH v2 0/6] sched/cache: Fixes for cache aware scheduling

[Tim Chen]

Hi all,

This is an update of the patches to fix cache aware scheduling issues
found in v7.2.

We collect the fixes in this series so it is easier to track. We have
added two new fixes for issues found since v1 of this series.

Patch 1: alb_break_llc() compares nr_pref_llc_running with
cfs.h_nr_runnable, but those count different sets - one follows queued
tasks, the other drops delay-dequeued ones.  With DELAY_DEQUEUE the
equality stops holding and active balance pulls a task off its preferred
LLC.  So fix the counter.  Reported by Zhan Xusheng:
https://lore.kernel.org/lkml/20260827135000.735138-1-zhanxusheng@xiaomi.com/

v1->v2: Minor code rearrangements in set_delayed().

Patch 2 (Lu Wang): the stopper doing active load balance builds a fresh
lb_env that doesn't inherit migration_type, so can_migrate_task() can
move a task *out* of its preferred LLC.  A new LBF_ACTIVE_LB_LLC flag
and picking the stopper callback at kick time keep the intent; passing
migration_type through the stopper would muddy delayed dequeue.  v4:
https://lore.kernel.org/lkml/20260903020656.3793626-1-wanglu.priv@gmail.com/

v1->v2: No change.

Patches 3-4 are for the use after free Hyunwoo Kim caught with KASAN:
https://lore.kernel.org/lkml/apPb-Dr4nPYuHQOK@v4bel/
account_mm_sched() reaches the stats via p->mm->sc_stat, but a task can
be switching mm on one CPU while another is inside account_mm_sched(),
so the mm and the stats inside it can go away underneath.  Locking the
rq in the mm free path felt like the wrong trade, so patch 4 pulls
sched_cache_stat out of mm_struct into a refcounted, RCU freed
sched_cache_group - just moving code - and patch 4 does the real fix:
each task takes its own reference (copy_mm(), exec_mmap(), dropped in
exit_mm()), and leverage call_rcu() to to protect
against UAF in account_mm_sched() so the group outlives any mm switch.
Zehnghui Yu also independentaly found this issue with memory poison.
https://lore.kernel.org/all/343a7e07-7fad-4979-9c9b-82ec038c293c@linux.dev/

@Hyunwoo and @Zhenhui, will appreciated you can test these patches
and add your Tested-by

Nice side effect: the group no longer follows the address space,
so a user defined group, or cgroup or numa_group could own it later.
These are also the grouping by prctl RFC's first two patches, sent here
so the fix isn't held up by that discussion.

v1->v2: Put cache aware related code in process exit/fork/copy in
its own functions

Patch 5 This patch makes sure kernel threads are excluded from cache
aware scheduling consideration.

v1->v2: Split from previous patch 3 as suggested by Peter Z.

Patch 6 is a new patch to fix an issue of undercomputing the LLC size
during CPU hot plug events.  The LLC size was used for estimating
if a process's memory footprint will fit a LLC. 
https://lore.kernel.org/all/20260916134432.11767-1-davichazbh@gmail.com/

BTW, there are two other issues in discussion currently and need
a bit more work:
1. Incorrect donor context being passed to task_tick_cache().  
https://lore.kernel.org/lkml/20260909092901.2989564-1-sh_def@163.com/
It is currently under discussion and is not included in this series.
2. Cache aware scheduling interfering with ITMT.
https://lore.kernel.org/lkml/20260810033742.1688718-1-yu.c.chen@intel.com/
https://lore.kernel.org/lkml/2fe2c681-b748-41fa-8b56-1169c86cefbc@intel.com/ 

Applies on sched/urgent branch. 

Tim Chen and Chen Yu

Chen Yu (1):
  sched/cache: Skip kernel thread for cache aware scheduling

Davi Chaves Azevedo (1):
  sched/cache: Refresh LLC capacity across CPU hotplug

Lu Wang (1):
  sched/cache: Honor migrate_llc_task semantics in active load balance

Tim Chen (3):
  sched/cache: Keep nr_pref_llc_running in the runnable domain
  sched/cache: Decouple sched_cache_group from mm
  sched/cache: Introduce task_struct->sched_cache_grp

 drivers/base/cacheinfo.c       |  11 +-
 fs/exec.c                      |   1 +
 include/linux/mm_types.h       |  15 +-
 include/linux/sched.h          |  21 ++-
 include/linux/sched/topology.h |   4 +-
 kernel/exit.c                  |  28 +--
 kernel/fork.c                  |   2 +
 kernel/sched/build_utility.c   |   4 +
 kernel/sched/cache_sched.c     | 106 +++++++++++
 kernel/sched/fair.c            | 310 ++++++++++++++++++++++++---------
 kernel/sched/sched.h           |   3 +
 kernel/sched/topology.c        |  22 ++-
 12 files changed, 394 insertions(+), 133 deletions(-)
 create mode 100644 kernel/sched/cache_sched.c

[PATCH v2 1/6] sched/cache: Keep nr_pref_llc_running in the runnable domain

[Tim Chen]

alb_break_llc() decides whether to break LLC preference during active
load balance. It does so by testing that every runnable fair task on the
source rq prefers its LLC:

	env->src_rq->nr_pref_llc_running == env->src_rq->cfs.h_nr_runnable

But the two counters cover different sets. nr_pref_llc_running is updated
in account_llc_enqueue()/account_llc_dequeue(), next to cfs_rq->nr_queued,
so it follows queued tasks. h_nr_runnable is updated in set_delayed()/
clear_delayed() and drops delay-dequeued tasks.

So under DELAY_DEQUEUE, a preferring task that goes to sleep stays counted
in nr_pref_llc_running while h_nr_runnable falls. The equality then breaks,
alb_break_llc() returns false, and active balance is free to pull a task
off its preferred LLC. Active balance only moves runnable tasks, and this
is the only LLC check it consults: once the stopper runs, LBF_ACTIVE_LB
skips the per-task test in can_migrate_task(). The runnable set is the one
we want.

Fix it on the counter side. A task should be counted in
nr_pref_llc_running exactly while it is both queued on its preferred LLC
(pref_llc_queued) and runnable (!sched_delayed). Define that membership
once in task_pref_llc_runnable(), and adjust the counter only through
pref_llc_running_inc()/pref_llc_running_dec() from the four sites that
change either input: account_llc_enqueue(), account_llc_dequeue(),
set_delayed() and clear_delayed(). Gating every update on the same
predicate keeps the delay, wake and dequeue paths from double-counting
or underflowing; see the comments at those sites for the ordering.

nr_llc_running and sd->llc_counts are not touched and stay on queued
semantics.

Fixes: 714059f79ff0 ("sched/cache: Handle moving single tasks to/from their preferred LLC")
Reported-by: Zhan Xusheng <zhanxusheng@xiaomi.com>
Closes: https://lore.kernel.org/lkml/20260827135000.735138-1-zhanxusheng@xiaomi.com/
Suggested-by: Chen Yu <yu.c.chen@intel.com>
Reviewed-by: Kayra Cizmeci <kayracizmeci@gmail.com>
Cc: stable@kernel.org #7.2.x
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
 kernel/sched/fair.c | 63 +++++++++++++++++++++++++++++++++++++++++----
 1 file changed, 58 insertions(+), 5 deletions(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index d5989b53adef..19765ee1af83 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -1551,6 +1551,28 @@ static bool invalid_llc_nr(struct mm_struct *mm, struct task_struct *p,
 			(scale * per_cpu(sd_llc_size, cpu)));
 }
 
+/*
+ * A task counts in nr_pref_llc_running while it is queued on its preferred
+ * LLC (pref_llc_queued) and runnable (!sched_delayed), keeping the counter in
+ * the runnable domain so alb_break_llc() can compare it with h_nr_runnable.
+ */
+static bool task_pref_llc_runnable(struct task_struct *p)
+{
+	return p->pref_llc_queued && !p->se.sched_delayed;
+}
+
+static void pref_llc_running_inc(struct rq *rq, struct task_struct *p)
+{
+	if (task_pref_llc_runnable(p))
+		rq->nr_pref_llc_running++;
+}
+
+static void pref_llc_running_dec(struct rq *rq, struct task_struct *p)
+{
+	if (task_pref_llc_runnable(p))
+		rq->nr_pref_llc_running--;
+}
+
 static void account_llc_enqueue(struct rq *rq, struct task_struct *p)
 {
 	int pref_llc, pref_llc_queued;
@@ -1562,7 +1584,6 @@ static void account_llc_enqueue(struct rq *rq, struct task_struct *p)
 
 	pref_llc_queued = (pref_llc == task_llc(p));
 	rq->nr_llc_running++;
-	rq->nr_pref_llc_running += pref_llc_queued;
 
 	/*
 	 * Record whether p is enqueued on its preferred
@@ -1580,6 +1601,9 @@ static void account_llc_enqueue(struct rq *rq, struct task_struct *p)
 	 */
 	p->pref_llc_queued = pref_llc_queued;
 
+	/* Skipped while delayed; clear_delayed() adds it back on wake. */
+	pref_llc_running_inc(rq, p);
+
 	sd = rcu_dereference_all(rq->sd);
 	if (sd && (unsigned int)pref_llc < sd->llc_max)
 		sd->llc_counts[pref_llc]++;
@@ -1596,7 +1620,12 @@ static void account_llc_dequeue(struct rq *rq, struct task_struct *p)
 
 	rq->nr_llc_running--;
 	if (p->pref_llc_queued) {
-		rq->nr_pref_llc_running--;
+		/*
+		 * Skipped if still delayed (set_delayed() already removed it);
+		 * clearing pref_llc_queued below also stops clear_delayed()
+		 * from re-adding it.
+		 */
+		pref_llc_running_dec(rq, p);
 		/*
 		 * Update the status in case
 		 * other logic might query
@@ -2000,6 +2029,7 @@ void init_sched_mm(struct task_struct *p)
 	 * polluting account_llc_enqueue().
 	 */
 	p->preferred_llc = -1;
+	p->pref_llc_queued = 0;
 }
 
 #else /* CONFIG_SCHED_CACHE */
@@ -2021,6 +2051,10 @@ static void account_llc_enqueue(struct rq *rq, struct task_struct *p) {}
 
 static void account_llc_dequeue(struct rq *rq, struct task_struct *p) {}
 
+static void pref_llc_running_inc(struct rq *rq, struct task_struct *p) {}
+
+static void pref_llc_running_dec(struct rq *rq, struct task_struct *p) {}
+
 #endif /* CONFIG_SCHED_CACHE */
 
 /*
@@ -6395,15 +6429,27 @@ static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq);
 
 static void set_delayed(struct sched_entity *se)
 {
-	se->sched_delayed = 1;
-
 	/*
 	 * Delayed se of cfs_rq have no tasks queued on them.
 	 * Do not adjust h_nr_runnable since __dequeue_task()
 	 * will account it for blocked tasks.
+	 *
+	 * This check can be removed because when flat pick
+	 * patches get merged as only task can get delayed,
+	 * same for clear_delayed().
 	 */
-	if (!entity_is_task(se))
+	if (!entity_is_task(se)) {
+		se->sched_delayed = 1;
 		return;
+	}
+
+	/*
+	 * Drop a task leaving the runnable set.
+	 * Needs to be called before sched_delayed is set.
+	 * clear_delayed() mirrors this after clearing the flag.
+	 */
+	pref_llc_running_dec(rq_of(cfs_rq_of(se)), task_of(se));
+	se->sched_delayed = 1;
 
 	for_each_sched_entity(se) {
 		struct cfs_rq *cfs_rq = cfs_rq_of(se);
@@ -6425,6 +6471,13 @@ static void clear_delayed(struct sched_entity *se)
 	if (!entity_is_task(se))
 		return;
 
+	/*
+	 * Re-add on wake, after sched_delayed is cleared. On a final delayed
+	 * dequeue account_llc_dequeue() already cleared pref_llc_queued, so
+	 * this does nothing.
+	 */
+	pref_llc_running_inc(rq_of(cfs_rq_of(se)), task_of(se));
+
 	for_each_sched_entity(se) {
 		struct cfs_rq *cfs_rq = cfs_rq_of(se);
 
-- 
2.32.0

Re: [PATCH v2 1/6] sched/cache: Keep nr_pref_llc_running in the runnable domain

[Peter Zijlstra]

On Mon, Sep 21, 2026 at 05:37:22PM -0700, Tim Chen wrote:
> @@ -6395,15 +6429,27 @@ static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq);
>  
>  static void set_delayed(struct sched_entity *se)
>  {
> -	se->sched_delayed = 1;
> -
>  	/*
>  	 * Delayed se of cfs_rq have no tasks queued on them.
>  	 * Do not adjust h_nr_runnable since __dequeue_task()
>  	 * will account it for blocked tasks.
> +	 *
> +	 * This check can be removed because when flat pick
> +	 * patches get merged as only task can get delayed,
> +	 * same for clear_delayed().
>  	 */
> -	if (!entity_is_task(se))
> +	if (!entity_is_task(se)) {
> +		se->sched_delayed = 1;
>  		return;
> +	}

This is dead code. In Linus' tree, where this will be applied, this can
never happen. A possible backport of this fix however will need to take
care.

> +
> +	/*
> +	 * Drop a task leaving the runnable set.
> +	 * Needs to be called before sched_delayed is set.
> +	 * clear_delayed() mirrors this after clearing the flag.
> +	 */
> +	pref_llc_running_dec(rq_of(cfs_rq_of(se)), task_of(se));
> +	se->sched_delayed = 1;
>  
>  	for_each_sched_entity(se) {
>  		struct cfs_rq *cfs_rq = cfs_rq_of(se);

Re: [PATCH v2 1/6] sched/cache: Keep nr_pref_llc_running in the runnable domain

[Chen, Yu C]

Hi Peter,

On 9/22/2026 3:22 PM, Peter Zijlstra wrote:
> On Mon, Sep 21, 2026 at 05:37:22PM -0700, Tim Chen wrote:
>> @@ -6395,15 +6429,27 @@ static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq);
>>   
>>   static void set_delayed(struct sched_entity *se)
>>   {
>> -	se->sched_delayed = 1;
>> -
>>   	/*
>>   	 * Delayed se of cfs_rq have no tasks queued on them.
>>   	 * Do not adjust h_nr_runnable since __dequeue_task()
>>   	 * will account it for blocked tasks.
>> +	 *
>> +	 * This check can be removed because when flat pick
>> +	 * patches get merged as only task can get delayed,
>> +	 * same for clear_delayed().
>>   	 */
>> -	if (!entity_is_task(se))
>> +	if (!entity_is_task(se)) {
>> +		se->sched_delayed = 1;
>>   		return;
>> +	}
> 
> This is dead code. In Linus' tree, where this will be applied, this can
> never happen. A possible backport of this fix however will need to take
> care.
> 

Would you prefer that we create a separate patch to remove entity_is_task()
for Linus's tree (so that the current patch can be backported, since 
cache-aware
scheduling was merged before flat task-pickup)?

thanks,
Chenyu

Re: [PATCH v2 1/6] sched/cache: Keep nr_pref_llc_running in the runnable domain

[Peter Zijlstra]

On Tue, Sep 22, 2026 at 04:25:14PM +0800, Chen, Yu C wrote:
> Hi Peter,
> 
> On 9/22/2026 3:22 PM, Peter Zijlstra wrote:
> > On Mon, Sep 21, 2026 at 05:37:22PM -0700, Tim Chen wrote:
> > > @@ -6395,15 +6429,27 @@ static __always_inline void return_cfs_rq_runtime(struct cfs_rq *cfs_rq);
> > >   static void set_delayed(struct sched_entity *se)
> > >   {
> > > -	se->sched_delayed = 1;
> > > -
> > >   	/*
> > >   	 * Delayed se of cfs_rq have no tasks queued on them.
> > >   	 * Do not adjust h_nr_runnable since __dequeue_task()
> > >   	 * will account it for blocked tasks.
> > > +	 *
> > > +	 * This check can be removed because when flat pick
> > > +	 * patches get merged as only task can get delayed,
> > > +	 * same for clear_delayed().
> > >   	 */
> > > -	if (!entity_is_task(se))
> > > +	if (!entity_is_task(se)) {
> > > +		se->sched_delayed = 1;
> > >   		return;
> > > +	}
> > 
> > This is dead code. In Linus' tree, where this will be applied, this can
> > never happen. A possible backport of this fix however will need to take
> > care.
> > 
> 
> Would you prefer that we create a separate patch to remove entity_is_task()
> for Linus's tree (so that the current patch can be backported, since
> cache-aware
> scheduling was merged before flat task-pickup)?

Nah, I munged it in with a few changes (I removed that cache-sched.c
file) and it seems to build. See queue.git/sched/urgent.

I'll queue up a patch for sched/core that removes some of that -- if I
don't forget and all that.

[PATCH v2 2/6] sched/cache: Honor migrate_llc_task semantics in active load balance

[Tim Chen]

From: Lu Wang <wanglu.priv@gmail.com>

CAS introduced the migrate_llc_task migration type to direct tasks
toward their preferred LLC, but its semantics can be lost when passive
load balance falls back to active load balance. This may allow ALB to
select a candidate whose preferred LLC does not match the destination,
moving it away from its preferred LLC.

Example scenario:

src_rq has two runnable tasks, p1 and p2. p1 prefers dst_rq (dst_llc),
while p2 prefers src_rq (src_llc). In this case, migrate_llc_task is
set because src_rq has at least one task, p1, that wants to migrate to
dst_rq. In ALB, can_migrate_task() finds p2 and returns true for it,
thus moving p2 out of its preferred LLC.

Solution:

The CPU stopper in ALB constructs a fresh lb_env that does not inherit
migration_type from the passive load-balance pass. Two approaches are
possible:

(a) Add a new member to struct rq so ALB can inherit migrate_llc_task
    from the passive LB that triggered it.

(b) Define a new flag LBF_ACTIVE_LB_LLC and select the stopper callback
    at kick time to preserve the migration semantics across the
    asynchronous boundary.

We choose (b) because it avoids passing migration_type through the
stopper, which would affect the meaning of migration_type for
delayed-dequeue tasks.

Fixes: e4c9a4cb244a ("sched/cache: Add migrate_llc_task migration type for cache-aware balancing")
Suggested-by: Chen Yu <yu.c.chen@intel.com>
Signed-off-by: Lu Wang <wanglu.priv@gmail.com>
Reviewed-by: Tim Chen <tim.c.chen@linux.intel.com>
Reviewed-by: Chen Yu <yu.c.chen@intel.com>
Cc: stable@kernel.org #7.2.x
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
 kernel/sched/fair.c | 57 ++++++++++++++++++++++++++++++++++++++++-----
 1 file changed, 51 insertions(+), 6 deletions(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 19765ee1af83..6f1939d17e9e 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -10453,6 +10453,7 @@ enum migration_type {
 #define LBF_SOME_PINNED	0x08
 #define LBF_ACTIVE_LB	0x10
 #define LBF_LLC_PINNED	0x20
+#define LBF_ACTIVE_LB_LLC	0x40
 
 struct lb_env {
 	struct sched_domain	*sd;
@@ -10871,6 +10872,21 @@ alb_break_llc(struct lb_env *env)
 	return false;
 }
 
+/*
+ * Returns true if p's preferred LLC does not match the destination CPU
+ * under migrate_llc_task semantics. Passive LB passes migrate_llc_task
+ * in env->migration_type, while active LB carries LBF_ACTIVE_LB_LLC in
+ * env->flags to avoid overwriting env->migration_type.
+ */
+static inline bool
+migrate_llc_task_wrong_dst(struct task_struct *p, struct lb_env *env)
+{
+	return sched_cache_enabled() &&
+	       (env->migration_type == migrate_llc_task ||
+		env->flags & LBF_ACTIVE_LB_LLC) &&
+	       READ_ONCE(p->preferred_llc) != llc_id(env->dst_cpu);
+}
+
 /*
  * Check if migrating task p from env->src_cpu to
  * env->dst_cpu breaks LLC localiy.
@@ -10899,8 +10915,7 @@ static bool migrate_degrades_llc(struct task_struct *p, struct lb_env *env)
 	 * run on env->dst_cpu, skip the tasks do not prefer
 	 * env->dst_cpu, and find the one that prefers.
 	 */
-	if (env->migration_type == migrate_llc_task &&
-	    READ_ONCE(p->preferred_llc) != llc_id(env->dst_cpu))
+	if (migrate_llc_task_wrong_dst(p, env))
 		return true;
 
 	if (can_migrate_llc_task(env, p) != mig_forbid)
@@ -10922,6 +10937,12 @@ alb_break_llc(struct lb_env *env)
 	return false;
 }
 
+static inline bool
+migrate_llc_task_wrong_dst(struct task_struct *p, struct lb_env *env)
+{
+	return false;
+}
+
 static inline bool
 migrate_degrades_llc(struct task_struct *p, struct lb_env *env)
 {
@@ -11021,7 +11042,7 @@ int can_migrate_task(struct task_struct *p, struct lb_env *env)
 	 * 4) too many balance attempts have failed.
 	 */
 	if (env->flags & LBF_ACTIVE_LB)
-		return 1;
+		return !migrate_llc_task_wrong_dst(p, env);
 
 	degrades = migrate_degrades_locality(p, env);
 	if (!degrades) {
@@ -13420,6 +13441,20 @@ static int need_active_balance(struct lb_env *env)
 }
 
 static int active_load_balance_cpu_stop(void *data);
+static int active_load_balance_llc_cpu_stop(void *data);
+
+/*
+ * migration_type is checked elsewhere to decide migration policy, so
+ * it shouldn't be repurposed just to flag an LLC-directed active
+ * balance across the stopper. Pick the callback here instead.
+ */
+static inline cpu_stop_fn_t alb_stop_fn(struct lb_env *env)
+{
+	if (env->migration_type == migrate_llc_task)
+		return active_load_balance_llc_cpu_stop;
+
+	return active_load_balance_cpu_stop;
+}
 
 static int should_we_balance(struct lb_env *env)
 {
@@ -13765,7 +13800,7 @@ static int sched_balance_rq(int this_cpu, struct rq *this_rq,
 	}
 	if (active_balance) {
 		stop_one_cpu_nowait(cpu_of(busiest),
-				    active_load_balance_cpu_stop, busiest,
+				    alb_stop_fn(&env), busiest,
 				    &busiest->active_balance_work);
 	}
 	preempt_enable();
@@ -13870,7 +13905,7 @@ update_next_balance(struct sched_domain *sd, unsigned long *next_balance)
  * least 1 task to be running on each physical CPU where possible, and
  * avoids physical / logical imbalances.
  */
-static int active_load_balance_cpu_stop(void *data)
+static int __active_load_balance_cpu_stop(void *data, unsigned int lb_flags)
 {
 	struct rq *busiest_rq = data;
 	int busiest_cpu = cpu_of(busiest_rq);
@@ -13920,7 +13955,7 @@ static int active_load_balance_cpu_stop(void *data)
 			.src_cpu	= busiest_rq->cpu,
 			.src_rq		= busiest_rq,
 			.idle		= CPU_IDLE,
-			.flags		= LBF_ACTIVE_LB,
+			.flags		= LBF_ACTIVE_LB | lb_flags,
 		};
 
 		schedstat_inc(sd->alb_count);
@@ -13948,6 +13983,16 @@ static int active_load_balance_cpu_stop(void *data)
 	return 0;
 }
 
+static int active_load_balance_cpu_stop(void *data)
+{
+	return __active_load_balance_cpu_stop(data, 0);
+}
+
+static int active_load_balance_llc_cpu_stop(void *data)
+{
+	return __active_load_balance_cpu_stop(data, LBF_ACTIVE_LB_LLC);
+}
+
 /*
  * Scale the max sched_balance_rq interval with the number of CPUs in the system.
  * This trades load-balance latency on larger machines for less cross talk.
-- 
2.32.0

Re: [PATCH v2 2/6] sched/cache: Honor migrate_llc_task semantics in active load balance

[Peter Zijlstra]

On Mon, Sep 21, 2026 at 05:37:23PM -0700, Tim Chen wrote:
> From: Lu Wang <wanglu.priv@gmail.com>
> 
> CAS introduced the migrate_llc_task migration type to direct tasks

FWIW, every time I read this I get confused, CAS maps to
Compare-And-Swap, aka cmpxchg on x86.

Re: [PATCH v2 2/6] sched/cache: Honor migrate_llc_task semantics in active load balance

[Chen, Yu C]

On 9/22/2026 3:23 PM, Peter Zijlstra wrote:
> FWIW, every time I read this I get confused, CAS maps to
> Compare-And-Swap, aka cmpxchg on x86.

OK, we will change it to Cache-Aware-Scheduling wherever it is mentioned.

thanks,
Chenyu

[PATCH v2 3/6] sched/cache: Decouple sched_cache_group from mm

[Tim Chen]

Currently the sched cache grouping is by mm and the scheduling statistics
sched_cache_stat lives in the mm structure.  This ties the life cycle
of scheduling stats with mm.

In account_mm_sched(), the scheduling stats are accessed by
task->mm->sc_stat.  However, a task may be switching mm on one CPU when
another CPU is running account_mm_sched(), and possibly accessing the
old mm that was freed.  This problem was found when running tests with
KASAN by Hyunwoo.  https://lore.kernel.org/lkml/apPb-Dr4nPYuHQOK@v4bel/

Instead of serializing the mm access by introducing extra acquisition of
rq lock in the mm free path, extract sched_cache_stat from mm_struct,
rename it as sched_cache_group and manage its life cycle apart from
mm_struct with its own ref counting.  This allows us in the next patch
access sched_cache_group directly from task, and add a refcount
on sched_cache_group when a task links to it. This prevents the use
after free issue when accessing stale and released old mm and its
sched cache stat a task switches to a new mm while account_mm_sched()
is done elsewhere.

The other benefit of this restructure is in the future, the grouping of
tasks to a LLC would have the flexibility to be associated with a user
defined grouping, or cgroup, cookie group, numa_group or others instead
of just with a single mm address space.

Rename sched_cache_stat to sched_cache_group and turn it into a refcounted
object allocated from mm_struct.  The mm_struct now holds a pointer
(sched_cache_grp) to this object instead of embedding it.

Introduce kernel/sched/cache_sched.c to host the cache aware scheduling
helpers and define sched_cache_group_put() there.

Fixes: df0d98475954 ("sched/cache: Introduce infrastructure for cache-aware load balancing")
Reported-by: Hyunwoo Kim <imv4bel@gmail.com>
Closes: https://lore.kernel.org/lkml/apPb-Dr4nPYuHQOK@v4bel/
Reported-by: Zenghui Yu (Huawei) <zenghui.yu@linux.dev>
Closes: https://lore.kernel.org/all/343a7e07-7fad-4979-9c9b-82ec038c293c@linux.dev/
Cc: stable@kernel.org #7.2.x
Co-developed-by: Chen Yu <yu.c.chen@intel.com>
Signed-off-by: Chen Yu <yu.c.chen@intel.com>
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
 include/linux/mm_types.h     |  15 ++--
 include/linux/sched.h        |   8 +-
 kernel/exit.c                |  11 ++-
 kernel/sched/build_utility.c |   4 +
 kernel/sched/cache_sched.c   |  19 +++++
 kernel/sched/fair.c          | 156 ++++++++++++++++++++++++-----------
 6 files changed, 152 insertions(+), 61 deletions(-)
 create mode 100644 kernel/sched/cache_sched.c

diff --git a/include/linux/mm_types.h b/include/linux/mm_types.h
index 6d815f6440c9..f3e5a2fadbe5 100644
--- a/include/linux/mm_types.h
+++ b/include/linux/mm_types.h
@@ -1226,7 +1226,7 @@ struct mm_struct {
 		struct mm_mm_cid mm_cid;
 
 		/* sched_cache related statistics */
-		struct sched_cache_stat sc_stat;
+		struct sched_cache_group *sched_cache_grp;
 #ifdef CONFIG_MMU
 		atomic_long_t pgtables_bytes;	/* size of all page tables */
 #endif
@@ -1624,8 +1624,9 @@ static inline unsigned int mm_cid_size(void)
 #endif /* CONFIG_SCHED_MM_CID */
 
 #ifdef CONFIG_SCHED_CACHE
-void mm_init_sched(struct mm_struct *mm,
-		   struct sched_cache_time __percpu *pcpu_sched);
+int mm_init_sched(struct mm_struct *mm,
+		  struct sched_cache_time __percpu *pcpu_sched);
+void mm_destroy_sched(struct mm_struct *mm);
 
 static inline int mm_alloc_sched_noprof(struct mm_struct *mm)
 {
@@ -1635,17 +1636,11 @@ static inline int mm_alloc_sched_noprof(struct mm_struct *mm)
 	if (!pcpu_sched)
 		return -ENOMEM;
 
-	mm_init_sched(mm, pcpu_sched);
-	return 0;
+	return mm_init_sched(mm, pcpu_sched);
 }
 
 #define mm_alloc_sched(...)	alloc_hooks(mm_alloc_sched_noprof(__VA_ARGS__))
 
-static inline void mm_destroy_sched(struct mm_struct *mm)
-{
-	free_percpu(mm->sc_stat.pcpu_sched);
-	mm->sc_stat.pcpu_sched = NULL;
-}
 #else /* !CONFIG_SCHED_CACHE */
 
 static inline int mm_alloc_sched(struct mm_struct *mm) { return 0; }
diff --git a/include/linux/sched.h b/include/linux/sched.h
index 8b3d47a325cc..1f254364f216 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -2405,7 +2405,7 @@ struct sched_cache_time {
 	unsigned long epoch;
 };
 
-struct sched_cache_stat {
+struct sched_cache_group {
 	struct sched_cache_time __percpu *pcpu_sched;
 	raw_spinlock_t lock;
 	unsigned long epoch;
@@ -2413,11 +2413,15 @@ struct sched_cache_stat {
 	unsigned long next_scan;
 	unsigned long footprint;
 	int cpu;
+	refcount_t refcnt;
+	struct rcu_head rcu;
 } ____cacheline_aligned_in_smp;
 
+void sched_cache_group_put(struct sched_cache_group *grp);
+
 #else
 
-struct sched_cache_stat { };
+struct sched_cache_group { };
 
 #endif
 
diff --git a/kernel/exit.c b/kernel/exit.c
index 97686af89501..16abbe1cf682 100644
--- a/kernel/exit.c
+++ b/kernel/exit.c
@@ -554,18 +554,23 @@ void mm_update_next_owner(struct mm_struct *mm)
  */
 static void exit_mm_sched_cache(struct mm_struct *mm)
 {
+	struct sched_cache_group *grp;
 	unsigned long fp, sub;
 
 	if (!current->total_numa_faults)
 		return;
 	/*
 	 * No lock protection due to performance considerations.
-	 * Make sure mm->sc_stat.footprint does not become
+	 * Make sure the group footprint does not become
 	 * negative.
 	 */
-	fp = READ_ONCE(mm->sc_stat.footprint);
+	grp = READ_ONCE(mm->sched_cache_grp);
+	if (!grp)
+		return;
+
+	fp = READ_ONCE(grp->footprint);
 	sub = min(fp, current->total_numa_faults);
-	WRITE_ONCE(mm->sc_stat.footprint, fp - sub);
+	WRITE_ONCE(grp->footprint, fp - sub);
 }
 #else
 static inline void exit_mm_sched_cache(struct mm_struct *mm)
diff --git a/kernel/sched/build_utility.c b/kernel/sched/build_utility.c
index e2cf3b08d4e9..24202893b262 100644
--- a/kernel/sched/build_utility.c
+++ b/kernel/sched/build_utility.c
@@ -89,6 +89,10 @@
 # include "core_sched.c"
 #endif
 
+#ifdef CONFIG_SCHED_CACHE
+# include "cache_sched.c"
+#endif
+
 #ifdef CONFIG_PSI
 # include "psi.c"
 #endif
diff --git a/kernel/sched/cache_sched.c b/kernel/sched/cache_sched.c
new file mode 100644
index 000000000000..ff3d9538e7a4
--- /dev/null
+++ b/kernel/sched/cache_sched.c
@@ -0,0 +1,19 @@
+// SPDX-License-Identifier: GPL-2.0-only
+#include "sched.h"
+
+static void sched_cache_group_free_rcu(struct rcu_head *rcu)
+{
+	struct sched_cache_group *grp =
+		container_of(rcu, struct sched_cache_group, rcu);
+
+	free_percpu(grp->pcpu_sched);
+	kfree(grp);
+}
+
+void sched_cache_group_put(struct sched_cache_group *grp)
+{
+	if (!grp || !refcount_dec_and_test(&grp->refcnt))
+		return;
+
+	call_rcu(&grp->rcu, sched_cache_group_free_rcu);
+}
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 6f1939d17e9e..6e939807dff2 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -1497,12 +1497,17 @@ static bool exceed_llc_capacity(struct mm_struct *mm, int cpu)
 		return true;
 
 	if (static_branch_likely(&sched_numa_balancing)) {
+		struct sched_cache_group *grp = READ_ONCE(mm->sched_cache_grp);
+
+		if (!grp)
+			return true;
+
 		/*
 		 * TBD: RDT exclusive LLC ways reserved should be
 		 * excluded.
 		 */
 		llc = sd->llc_bytes;
-		footprint = READ_ONCE(mm->sc_stat.footprint);
+		footprint = READ_ONCE(grp->footprint);
 
 		/*
 		 * Scale the LLC size by 256*llc_aggr_tolerance
@@ -1534,6 +1539,7 @@ static bool exceed_llc_capacity(struct mm_struct *mm, int cpu)
 static bool invalid_llc_nr(struct mm_struct *mm, struct task_struct *p,
 			   int cpu)
 {
+	struct sched_cache_group *grp;
 	int scale;
 
 	if (get_nr_threads(p) <= 1)
@@ -1547,7 +1553,11 @@ static bool invalid_llc_nr(struct mm_struct *mm, struct task_struct *p,
 	if (scale == INT_MAX)
 		return false;
 
-	return !fits_capacity((mm->sc_stat.nr_running_avg * cpu_smt_num_threads),
+	grp = READ_ONCE(mm->sched_cache_grp);
+	if (!grp)
+		return true;
+
+	return !fits_capacity((READ_ONCE(grp->nr_running_avg) * cpu_smt_num_threads),
 			(scale * per_cpu(sd_llc_size, cpu)));
 }
 
@@ -1653,12 +1663,20 @@ static void account_llc_dequeue(struct rq *rq, struct task_struct *p)
 	}
 }
 
-void mm_init_sched(struct mm_struct *mm,
-		   struct sched_cache_time __percpu *_pcpu_sched)
+int mm_init_sched(struct mm_struct *mm,
+		  struct sched_cache_time __percpu *_pcpu_sched)
 {
+	struct sched_cache_group *grp;
 	unsigned long epoch = 0;
 	int i;
 
+	grp = kzalloc_obj(*grp);
+	if (!grp) {
+		free_percpu(_pcpu_sched);
+		mm->sched_cache_grp = NULL;
+		return -ENOMEM;
+	}
+
 	for_each_possible_cpu(i) {
 		struct sched_cache_time *pcpu_sched = per_cpu_ptr(_pcpu_sched, i);
 		struct rq *rq = cpu_rq(i);
@@ -1669,18 +1687,34 @@ void mm_init_sched(struct mm_struct *mm,
 		epoch = rq->cpu_epoch;
 	}
 
-	raw_spin_lock_init(&mm->sc_stat.lock);
-	mm->sc_stat.epoch = epoch;
-	mm->sc_stat.cpu = -1;
-	mm->sc_stat.next_scan = jiffies;
-	mm->sc_stat.nr_running_avg = 0;
-	mm->sc_stat.footprint = 0;
+	raw_spin_lock_init(&grp->lock);
+	grp->epoch = epoch;
+	grp->cpu = -1;
+	grp->next_scan = jiffies;
+	grp->nr_running_avg = 0;
+	grp->footprint = 0;
+	refcount_set(&grp->refcnt, 1);
 	/*
-	 * The update to mm->sc_stat should not be reordered
-	 * before initialization to mm's other fields, in case
+	 * The update to grp->pcpu_sched should not be reordered
+	 * before initialization to grp's other fields, in case
 	 * the readers may get invalid mm_sched_epoch, etc.
 	 */
-	smp_store_release(&mm->sc_stat.pcpu_sched, _pcpu_sched);
+	smp_store_release(&grp->pcpu_sched, _pcpu_sched);
+	/*
+	 * Publish the group last.  Not every reader qualifies it by
+	 * grp->pcpu_sched - can_migrate_llc_task() only checks that the
+	 * pointer is non-NULL before reading grp->footprint and
+	 * grp->nr_running_avg - so a reachable group must already be
+	 * fully initialized.
+	 */
+	smp_store_release(&mm->sched_cache_grp, grp);
+	return 0;
+}
+
+void mm_destroy_sched(struct mm_struct *mm)
+{
+	sched_cache_group_put(mm->sched_cache_grp);
+	mm->sched_cache_grp = NULL;
 }
 
 /* because why would C be fully specified */
@@ -1734,11 +1768,16 @@ static unsigned long fraction_mm_sched(struct rq *rq,
 static int get_pref_llc(struct task_struct *p, struct mm_struct *mm)
 {
 	int mm_sched_llc = -1, mm_sched_cpu;
+	struct sched_cache_group *grp;
 
 	if (!mm)
 		return -1;
 
-	mm_sched_cpu = READ_ONCE(mm->sc_stat.cpu);
+	grp = READ_ONCE(mm->sched_cache_grp);
+	if (!grp)
+		return -1;
+
+	mm_sched_cpu = READ_ONCE(grp->cpu);
 	if (mm_sched_cpu != -1) {
 		mm_sched_llc = llc_id(mm_sched_cpu);
 
@@ -1769,6 +1808,7 @@ static inline
 void account_mm_sched(struct rq *rq, struct task_struct *p, s64 delta_exec)
 {
 	struct sched_cache_time *pcpu_sched;
+	struct sched_cache_group *grp;
 	struct mm_struct *mm = p->mm;
 	int mm_sched_llc = -1;
 	unsigned long epoch;
@@ -1782,10 +1822,14 @@ void account_mm_sched(struct rq *rq, struct task_struct *p, s64 delta_exec)
 	 * init_task, kthreads and user thread created
 	 * by user_mode_thread() don't have mm.
 	 */
-	if (!mm || !mm->sc_stat.pcpu_sched)
+	if (!mm)
+		return;
+
+	grp = READ_ONCE(mm->sched_cache_grp);
+	if (!grp || !grp->pcpu_sched)
 		return;
 
-	pcpu_sched = per_cpu_ptr(mm->sc_stat.pcpu_sched, cpu_of(rq));
+	pcpu_sched = per_cpu_ptr(grp->pcpu_sched, cpu_of(rq));
 
 	scoped_guard (raw_spinlock, &rq->cpu_epoch_lock) {
 		__update_mm_sched(rq, pcpu_sched);
@@ -1798,11 +1842,11 @@ void account_mm_sched(struct rq *rq, struct task_struct *p, s64 delta_exec)
 	 * If this process hasn't hit task_cache_work() for a while invalidate
 	 * its preferred state.
 	 */
-	if ((long)(epoch - READ_ONCE(mm->sc_stat.epoch)) > llc_epoch_affinity_timeout ||
+	if ((long)(epoch - READ_ONCE(grp->epoch)) > llc_epoch_affinity_timeout ||
 	    invalid_llc_nr(mm, p, cpu_of(rq)) ||
 	    exceed_llc_capacity(mm, cpu_of(rq))) {
-		if (READ_ONCE(mm->sc_stat.cpu) != -1)
-			WRITE_ONCE(mm->sc_stat.cpu, -1);
+		if (READ_ONCE(grp->cpu) != -1)
+			WRITE_ONCE(grp->cpu, -1);
 	}
 
 	mm_sched_llc = get_pref_llc(p, mm);
@@ -1819,30 +1863,35 @@ void account_mm_sched(struct rq *rq, struct task_struct *p, s64 delta_exec)
 static void task_tick_cache(struct rq *rq, struct task_struct *p)
 {
 	struct callback_head *work = &p->cache_work;
+	struct sched_cache_group *grp;
 	struct mm_struct *mm = p->mm;
 	unsigned long epoch;
 
 	if (!sched_cache_enabled())
 		return;
 
-	if (!mm || p->flags & PF_KTHREAD ||
-	    !mm->sc_stat.pcpu_sched)
+	if (!mm || p->flags & PF_KTHREAD)
+		return;
+
+	grp = READ_ONCE(mm->sched_cache_grp);
+	if (!grp || !grp->pcpu_sched)
 		return;
 
 	epoch = rq->cpu_epoch;
 	/* avoid moving backwards */
-	if (time_after_eq(mm->sc_stat.epoch, epoch))
+	if (time_after_eq(grp->epoch, epoch))
 		return;
 
-	guard(raw_spinlock)(&mm->sc_stat.lock);
+	guard(raw_spinlock)(&grp->lock);
 
 	if (work->next == work) {
 		task_work_add(p, work, TWA_RESUME);
-		WRITE_ONCE(mm->sc_stat.epoch, epoch);
+		WRITE_ONCE(grp->epoch, epoch);
 	}
 }
 
-static void get_scan_cpumasks(cpumask_var_t cpus, struct task_struct *p)
+static void get_scan_cpumasks(cpumask_var_t cpus, struct task_struct *p,
+			      struct sched_cache_group *grp)
 {
 #ifdef CONFIG_NUMA_BALANCING
 	int cpu, curr_cpu, nid, pref_nid;
@@ -1850,7 +1899,7 @@ static void get_scan_cpumasks(cpumask_var_t cpus, struct task_struct *p)
 	if (!static_branch_likely(&sched_numa_balancing))
 		goto out;
 
-	cpu = READ_ONCE(p->mm->sc_stat.cpu);
+	cpu = READ_ONCE(grp->cpu);
 	if (cpu != -1)
 		nid = cpu_to_node(cpu);
 	curr_cpu = task_cpu(p);
@@ -1911,6 +1960,7 @@ static void task_cache_work(struct callback_head *work)
 	unsigned long next_scan, now = jiffies;
 	struct task_struct *p = current, *cur;
 	unsigned long curr_m_a_occ = 0;
+	struct sched_cache_group *grp;
 	struct mm_struct *mm = p->mm;
 	unsigned long m_a_occ = 0;
 	cpumask_var_t cpus;
@@ -1922,12 +1972,16 @@ static void task_cache_work(struct callback_head *work)
 	if (p->flags & PF_EXITING)
 		return;
 
-	next_scan = READ_ONCE(mm->sc_stat.next_scan);
+	grp = READ_ONCE(mm->sched_cache_grp);
+	if (!grp)
+		return;
+
+	next_scan = READ_ONCE(grp->next_scan);
 	if (time_before(now, next_scan))
 		return;
 
 	/* only 1 thread is allowed to scan */
-	if (!try_cmpxchg(&mm->sc_stat.next_scan, &next_scan,
+	if (!try_cmpxchg(&grp->next_scan, &next_scan,
 			 now + max_t(unsigned long,
 				     READ_ONCE(llc_epoch_period), 1)))
 		return;
@@ -1935,8 +1989,8 @@ static void task_cache_work(struct callback_head *work)
 	curr_cpu = task_cpu(p);
 	if (invalid_llc_nr(mm, p, curr_cpu) ||
 	    exceed_llc_capacity(mm, curr_cpu)) {
-		if (READ_ONCE(mm->sc_stat.cpu) != -1)
-			WRITE_ONCE(mm->sc_stat.cpu, -1);
+		if (READ_ONCE(grp->cpu) != -1)
+			WRITE_ONCE(grp->cpu, -1);
 
 		return;
 	}
@@ -1947,7 +2001,7 @@ static void task_cache_work(struct callback_head *work)
 	scoped_guard (cpus_read_lock) {
 		guard(rcu)();
 
-		get_scan_cpumasks(cpus, p);
+		get_scan_cpumasks(cpus, p, grp);
 
 		for_each_cpu(cpu, cpus) {
 			/* XXX sched_cluster_active */
@@ -1960,7 +2014,7 @@ static void task_cache_work(struct callback_head *work)
 
 			for_each_cpu(i, sched_domain_span(sd)) {
 				occ = fraction_mm_sched(cpu_rq(i),
-							per_cpu_ptr(mm->sc_stat.pcpu_sched, i));
+							per_cpu_ptr(grp->pcpu_sched, i));
 				a_occ += occ;
 				if (occ > m_occ) {
 					m_occ = occ;
@@ -1993,7 +2047,7 @@ static void task_cache_work(struct callback_head *work)
 				m_a_cpu = m_cpu;
 			}
 
-			if (llc_id(cpu) == llc_id(READ_ONCE(mm->sc_stat.cpu)))
+			if (llc_id(cpu) == llc_id(READ_ONCE(grp->cpu)))
 				curr_m_a_occ = a_occ;
 
 			cpumask_andnot(cpus, cpus, sched_domain_span(sd));
@@ -2002,7 +2056,7 @@ static void task_cache_work(struct callback_head *work)
 
 	if (m_a_occ > (2 * curr_m_a_occ)) {
 		/*
-		 * Avoid switching sc_stat.cpu too fast.
+		 * Avoid switching sched_cache_grp->cpu too fast.
 		 * The reason to choose 2X is because:
 		 * 1. It is better to keep the preferred LLC stable,
 		 *    rather than changing it frequently and cause migrations
@@ -2011,10 +2065,10 @@ static void task_cache_work(struct callback_head *work)
 		 * 3. 2X is chosen based on test results, as it delivers
 		 *    the optimal performance gain so far.
 		 */
-		WRITE_ONCE(mm->sc_stat.cpu, m_a_cpu);
+		WRITE_ONCE(grp->cpu, m_a_cpu);
 	}
 
-	update_avg_scale(&mm->sc_stat.nr_running_avg, nr_running);
+	update_avg_scale(&grp->nr_running_avg, nr_running);
 	free_cpumask_var(cpus);
 }
 
@@ -3731,6 +3785,7 @@ static int preferred_group_nid(struct task_struct *p, int nid)
 static void task_numa_placement(struct task_struct *p)
 	__context_unsafe(/* conditional locking */)
 {
+	struct sched_cache_group __maybe_unused *grp;
 	int seq, nid, max_nid = NUMA_NO_NODE;
 	unsigned long max_faults = 0;
 	unsigned long fault_types[2] = { 0, 0 };
@@ -3823,19 +3878,23 @@ static void task_numa_placement(struct task_struct *p)
 			 * heuristic and occasional lost updates are tolerable.
 			 *
 			 * If a task exits, its corresponding footprint must
-			 * be subtracted from the mm->sc_stat.footprint, otherwise
-			 * the mm->sc_stat.footprint will not converge:
-			 * the exiting thread's footprint remains unchanged/undecayed
-			 * in mm->sc_stat.footprint. See exit_mm().
+			 * be subtracted from the mm->sched_cache_grp->footprint,
+			 * otherwise the mm->sched_cache_grp->footprint will not
+			 * converge: the exiting thread's footprint remains
+			 * unchanged/undecayed in mm->sched_cache_grp->footprint.
+			 * See exit_mm().
 			 *
 			 * Lost updates and unsynchronized subtraction
 			 * in exit_mm() can cause footprint + diff to
 			 * go negative. Clamp to zero to prevent the
 			 * unsigned footprint from wrapping.
 			 */
-			new_fp = (long)READ_ONCE(p->mm->sc_stat.footprint) + diff;
-			WRITE_ONCE(p->mm->sc_stat.footprint,
-				   max(new_fp, 0L));
+			grp = READ_ONCE(p->mm->sched_cache_grp);
+			if (!grp)
+				continue;
+
+			new_fp = (long)READ_ONCE(grp->footprint) + diff;
+			WRITE_ONCE(grp->footprint, max(new_fp, 0L));
 #endif
 		}
 
@@ -10783,6 +10842,7 @@ static inline bool task_misfits_asym_cpu(struct lb_env *env, struct task_struct
 static enum llc_mig can_migrate_llc_task(struct lb_env *env,
 					 struct task_struct *p)
 {
+	struct sched_cache_group *grp;
 	struct mm_struct *mm;
 	bool to_pref;
 	int cpu, src_cpu, dst_cpu;
@@ -10796,15 +10856,19 @@ static enum llc_mig can_migrate_llc_task(struct lb_env *env,
 	if (!mm)
 		return mig_unrestricted;
 
-	cpu = READ_ONCE(mm->sc_stat.cpu);
+	grp = READ_ONCE(mm->sched_cache_grp);
+	if (!grp)
+		return mig_unrestricted;
+
+	cpu = READ_ONCE(grp->cpu);
 	if (cpu < 0 || cpus_share_cache(src_cpu, dst_cpu))
 		return mig_unrestricted;
 
 	/* skip cache aware load balance for too many threads */
 	if (invalid_llc_nr(mm, p, dst_cpu) ||
 	    exceed_llc_capacity(mm, dst_cpu)) {
-		if (READ_ONCE(mm->sc_stat.cpu) != -1)
-			WRITE_ONCE(mm->sc_stat.cpu, -1);
+		if (READ_ONCE(grp->cpu) != -1)
+			WRITE_ONCE(grp->cpu, -1);
 		return mig_unrestricted;
 	}
 
-- 
2.32.0

[PATCH v2 4/6] sched/cache: Introduce task_struct->sched_cache_grp

[Tim Chen]

Add a sched_cache_grp pointer to task_struct so that scheduler code
can access the cache group directly via the task, without going
through mm->sched_cache_grp.  This decouples the scheduler's hot-path
accesses from the mm_struct.

Each task holds its own refcount on the sched_cache_group, separate
from the reference held by its mm_struct.  The reference is acquired
in copy_mm() (fork) and exec_mmap() (exec), and released in exit_mm().
This fixes the use-after-free when account_mm_sched() reaches the group
through a task whose mm is being switched, as reported by Hyunwoo:
https://lore.kernel.org/lkml/apPb-Dr4nPYuHQOK@v4bel/

Convert all scheduler code in fair.c and exit.c to use
p->sched_cache_grp instead of p->mm->sched_cache_grp.

Keep the fork/exec/exit reference management out of the generic mm
paths: add sched_cache_fork(), sched_cache_fork_cleanup(),
sched_cache_exec_mmap() and sched_cache_exit_mm() in
kernel/sched/cache_sched.c (with empty stubs for !CONFIG_SCHED_CACHE),
so fs/exec.c, kernel/fork.c and kernel/exit.c each call one helper
instead of open-coding the refcounting under #ifdef.  Also add
sched_cache_group_get() and task_cache_group_get().

Fixes: df0d98475954 ("sched/cache: Introduce infrastructure for cache-aware load balancing")
Reported-by: Hyunwoo Kim <imv4bel@gmail.com>
Closes: https://lore.kernel.org/lkml/apPb-Dr4nPYuHQOK@v4bel/
Reported-by: Zenghui Yu (Huawei) <zenghui.yu@linux.dev>
Closes: https://lore.kernel.org/all/343a7e07-7fad-4979-9c9b-82ec038c293c@linux.dev/
Cc: stable@kernel.org #7.2.x
Co-developed-by: Chen Yu <yu.c.chen@intel.com>
Signed-off-by: Chen Yu <yu.c.chen@intel.com>
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
 fs/exec.c                  |   1 +
 include/linux/sched.h      |  13 +++++
 kernel/exit.c              |  33 +-----------
 kernel/fork.c              |   2 +
 kernel/sched/cache_sched.c |  87 ++++++++++++++++++++++++++++++
 kernel/sched/fair.c        | 106 ++++++++++++++++---------------------
 kernel/sched/sched.h       |   3 ++
 7 files changed, 153 insertions(+), 92 deletions(-)

diff --git a/fs/exec.c b/fs/exec.c
index 745f6eb5279e..6400bae97a32 100644
--- a/fs/exec.c
+++ b/fs/exec.c
@@ -882,6 +882,7 @@ static int exec_mmap(struct linux_binprm *bprm)
 	active_mm = tsk->active_mm;
 	tsk->active_mm = mm;
 	tsk->mm = mm;
+	sched_cache_exec_mmap(tsk, mm);
 	mm_init_cid(mm, tsk);
 	exec_state = task_exec_state_replace(tsk, exec_state);
 	/*
diff --git a/include/linux/sched.h b/include/linux/sched.h
index 1f254364f216..1aa81cb6637b 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -1434,6 +1434,7 @@ struct task_struct {
 #ifdef CONFIG_SCHED_CACHE
 	struct callback_head		cache_work;
 	int				preferred_llc;
+	struct sched_cache_group __rcu	*sched_cache_grp;
 	/* 1: task was enqueued to its preferred LLC, 0 otherwise */
 	int				pref_llc_queued;
 #endif
@@ -2418,11 +2419,23 @@ struct sched_cache_group {
 } ____cacheline_aligned_in_smp;
 
 void sched_cache_group_put(struct sched_cache_group *grp);
+struct sched_cache_group *sched_cache_group_get(struct sched_cache_group *grp);
+struct sched_cache_group *task_cache_group_get(struct task_struct *p);
+
+void sched_cache_fork(struct task_struct *p);
+void sched_cache_fork_cleanup(struct task_struct *p);
+void sched_cache_exec_mmap(struct task_struct *p, struct mm_struct *mm);
+void sched_cache_exit_mm(struct task_struct *p);
 
 #else
 
 struct sched_cache_group { };
 
+static inline void sched_cache_fork(struct task_struct *p) { }
+static inline void sched_cache_fork_cleanup(struct task_struct *p) { }
+static inline void sched_cache_exec_mmap(struct task_struct *p, struct mm_struct *mm) { }
+static inline void sched_cache_exit_mm(struct task_struct *p) { }
+
 #endif
 
 #ifndef MODULE
diff --git a/kernel/exit.c b/kernel/exit.c
index 16abbe1cf682..35afa1d2d251 100644
--- a/kernel/exit.c
+++ b/kernel/exit.c
@@ -547,37 +547,6 @@ void mm_update_next_owner(struct mm_struct *mm)
 }
 #endif /* CONFIG_MEMCG */
 
-#if defined(CONFIG_SCHED_CACHE) && defined(CONFIG_NUMA_BALANCING)
-/*
- * Subtract the memory footprint of the current task from
- * mm.
- */
-static void exit_mm_sched_cache(struct mm_struct *mm)
-{
-	struct sched_cache_group *grp;
-	unsigned long fp, sub;
-
-	if (!current->total_numa_faults)
-		return;
-	/*
-	 * No lock protection due to performance considerations.
-	 * Make sure the group footprint does not become
-	 * negative.
-	 */
-	grp = READ_ONCE(mm->sched_cache_grp);
-	if (!grp)
-		return;
-
-	fp = READ_ONCE(grp->footprint);
-	sub = min(fp, current->total_numa_faults);
-	WRITE_ONCE(grp->footprint, fp - sub);
-}
-#else
-static inline void exit_mm_sched_cache(struct mm_struct *mm)
-{
-}
-#endif /* CONFIG_SCHED_CACHE CONFIG_NUMA_BALANCING */
-
 /*
  * Turn us into a lazy TLB process if we
  * aren't already..
@@ -590,7 +559,7 @@ static void exit_mm(void)
 	if (!mm)
 		return;
 
-	exit_mm_sched_cache(mm);
+	sched_cache_exit_mm(current);
 
 	mmap_read_lock(mm);
 	mmgrab_lazy_tlb(mm);
diff --git a/kernel/fork.c b/kernel/fork.c
index 416758c8a3d4..d9b263a32471 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -1599,6 +1599,7 @@ static int copy_mm(u64 clone_flags, struct task_struct *tsk)
 
 	tsk->mm = mm;
 	tsk->active_mm = mm;
+	sched_cache_fork(tsk);
 	return 0;
 }
 
@@ -2599,6 +2600,7 @@ __latent_entropy struct task_struct *copy_process(
 bad_fork_cleanup_namespaces:
 	exit_nsproxy_namespaces(p);
 bad_fork_cleanup_mm:
+	sched_cache_fork_cleanup(p);
 	if (p->mm) {
 		mm_clear_owner(p->mm, p);
 		mmput(p->mm);
diff --git a/kernel/sched/cache_sched.c b/kernel/sched/cache_sched.c
index ff3d9538e7a4..7c5d23a1e09c 100644
--- a/kernel/sched/cache_sched.c
+++ b/kernel/sched/cache_sched.c
@@ -1,6 +1,93 @@
 // SPDX-License-Identifier: GPL-2.0-only
 #include "sched.h"
 
+#define rcu_deref_sched_cache_grp(tsk) \
+	rcu_dereference_check((tsk)->sched_cache_grp, (tsk) == current)
+
+static struct sched_cache_group *sched_cache_replace_grp(struct task_struct *p,
+							 struct sched_cache_group *new)
+{
+	struct sched_cache_group *old;
+
+	old = rcu_deref_sched_cache_grp(p);
+	rcu_assign_pointer(p->sched_cache_grp, new);
+
+	return old;
+}
+
+struct sched_cache_group *sched_cache_group_get(struct sched_cache_group *grp)
+{
+	/*
+	 * refcount_inc_not_zero() is the acquire primitive for lockless
+	 * (RCU) lookups; plain refcount_inc() would scribble the count if
+	 * it already reached zero. Return NULL in that case.
+	 */
+	if (grp && !refcount_inc_not_zero(&grp->refcnt))
+		grp = NULL;
+
+	return grp;
+}
+
+struct sched_cache_group *task_cache_group_get(struct task_struct *p)
+{
+	guard(rcu)();
+	return sched_cache_group_get(rcu_dereference(p->sched_cache_grp));
+}
+
+void sched_cache_fork(struct task_struct *p)
+{
+	/*
+	 * The child takes its own reference on the mm's cache group, separate
+	 * from the reference held by the mm. @p is not yet visible to readers,
+	 * so a plain initializing store is enough.
+	 */
+	RCU_INIT_POINTER(p->sched_cache_grp,
+			 sched_cache_group_get(p->mm->sched_cache_grp));
+}
+
+void sched_cache_fork_cleanup(struct task_struct *p)
+{
+	/*
+	 * A fork that fails after sched_cache_fork() never reaches exit_mm(),
+	 * so drop the reference here. @p never became visible, so there are no
+	 * concurrent readers and the reference we hold keeps the group alive.
+	 */
+	sched_cache_group_put(rcu_access_pointer(p->sched_cache_grp));
+	RCU_INIT_POINTER(p->sched_cache_grp, NULL);
+}
+
+void sched_cache_exec_mmap(struct task_struct *p, struct mm_struct *mm)
+{
+	struct sched_cache_group *old;
+
+	/*
+	 * Acquire the new reference before publishing the pointer, then drop
+	 * the old one. @p is current and the only writer of its own pointer.
+	 */
+	old = sched_cache_replace_grp(p, sched_cache_group_get(mm->sched_cache_grp));
+	sched_cache_group_put(old);
+}
+
+void sched_cache_exit_mm(struct task_struct *p)
+{
+	struct sched_cache_group *grp = sched_cache_replace_grp(p, NULL);
+
+#ifdef CONFIG_NUMA_BALANCING
+	/*
+	 * Subtract this task's footprint from the group before dropping the
+	 * reference, so the group footprint converges as its threads exit.
+	 * Unlocked for performance; clamp to avoid underflow.
+	 */
+	if (grp && p->total_numa_faults) {
+		unsigned long fp = READ_ONCE(grp->footprint);
+		unsigned long sub = min(fp, p->total_numa_faults);
+
+		WRITE_ONCE(grp->footprint, fp - sub);
+	}
+#endif
+	sched_cache_group_put(grp);
+}
+
 static void sched_cache_group_free_rcu(struct rcu_head *rcu)
 {
 	struct sched_cache_group *grp =
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 6e939807dff2..3e2d236be2a3 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -1483,7 +1483,7 @@ static inline int get_sched_cache_scale(int mul)
 	return (1 + (tol - 1) * mul);
 }
 
-static bool exceed_llc_capacity(struct mm_struct *mm, int cpu)
+static bool exceed_llc_capacity(struct sched_cache_group *grp, int cpu)
 {
 #ifdef CONFIG_NUMA_BALANCING
 	unsigned long llc, footprint;
@@ -1497,11 +1497,6 @@ static bool exceed_llc_capacity(struct mm_struct *mm, int cpu)
 		return true;
 
 	if (static_branch_likely(&sched_numa_balancing)) {
-		struct sched_cache_group *grp = READ_ONCE(mm->sched_cache_grp);
-
-		if (!grp)
-			return true;
-
 		/*
 		 * TBD: RDT exclusive LLC ways reserved should be
 		 * excluded.
@@ -1536,10 +1531,9 @@ static bool exceed_llc_capacity(struct mm_struct *mm, int cpu)
 	return false;
 }
 
-static bool invalid_llc_nr(struct mm_struct *mm, struct task_struct *p,
+static bool invalid_llc_nr(struct sched_cache_group *grp, struct task_struct *p,
 			   int cpu)
 {
-	struct sched_cache_group *grp;
 	int scale;
 
 	if (get_nr_threads(p) <= 1)
@@ -1553,10 +1547,6 @@ static bool invalid_llc_nr(struct mm_struct *mm, struct task_struct *p,
 	if (scale == INT_MAX)
 		return false;
 
-	grp = READ_ONCE(mm->sched_cache_grp);
-	if (!grp)
-		return true;
-
 	return !fits_capacity((READ_ONCE(grp->nr_running_avg) * cpu_smt_num_threads),
 			(scale * per_cpu(sd_llc_size, cpu)));
 }
@@ -1765,15 +1755,10 @@ static unsigned long fraction_mm_sched(struct rq *rq,
 	return div64_u64(NICE_0_LOAD * pcpu_sched->runtime, rq->cpu_runtime + 1);
 }
 
-static int get_pref_llc(struct task_struct *p, struct mm_struct *mm)
+static int get_pref_llc(struct task_struct *p, struct sched_cache_group *grp)
 {
 	int mm_sched_llc = -1, mm_sched_cpu;
-	struct sched_cache_group *grp;
 
-	if (!mm)
-		return -1;
-
-	grp = READ_ONCE(mm->sched_cache_grp);
 	if (!grp)
 		return -1;
 
@@ -1807,9 +1792,8 @@ static unsigned int task_running_on_cpu(int cpu, struct task_struct *p);
 static inline
 void account_mm_sched(struct rq *rq, struct task_struct *p, s64 delta_exec)
 {
+	struct sched_cache_group *grp = rcu_dereference_all(p->sched_cache_grp);
 	struct sched_cache_time *pcpu_sched;
-	struct sched_cache_group *grp;
-	struct mm_struct *mm = p->mm;
 	int mm_sched_llc = -1;
 	unsigned long epoch;
 
@@ -1820,12 +1804,8 @@ void account_mm_sched(struct rq *rq, struct task_struct *p, s64 delta_exec)
 		return;
 	/*
 	 * init_task, kthreads and user thread created
-	 * by user_mode_thread() don't have mm.
+	 * by user_mode_thread() don't have a cache group.
 	 */
-	if (!mm)
-		return;
-
-	grp = READ_ONCE(mm->sched_cache_grp);
 	if (!grp || !grp->pcpu_sched)
 		return;
 
@@ -1843,13 +1823,13 @@ void account_mm_sched(struct rq *rq, struct task_struct *p, s64 delta_exec)
 	 * its preferred state.
 	 */
 	if ((long)(epoch - READ_ONCE(grp->epoch)) > llc_epoch_affinity_timeout ||
-	    invalid_llc_nr(mm, p, cpu_of(rq)) ||
-	    exceed_llc_capacity(mm, cpu_of(rq))) {
+	    invalid_llc_nr(grp, p, cpu_of(rq)) ||
+	    exceed_llc_capacity(grp, cpu_of(rq))) {
 		if (READ_ONCE(grp->cpu) != -1)
 			WRITE_ONCE(grp->cpu, -1);
 	}
 
-	mm_sched_llc = get_pref_llc(p, mm);
+	mm_sched_llc = get_pref_llc(p, grp);
 
 	/* task not on rq accounted later in account_entity_enqueue() */
 	if (task_running_on_cpu(rq->cpu, p) &&
@@ -1862,19 +1842,15 @@ void account_mm_sched(struct rq *rq, struct task_struct *p, s64 delta_exec)
 
 static void task_tick_cache(struct rq *rq, struct task_struct *p)
 {
+	struct sched_cache_group *grp = rcu_dereference_all(p->sched_cache_grp);
 	struct callback_head *work = &p->cache_work;
-	struct sched_cache_group *grp;
-	struct mm_struct *mm = p->mm;
 	unsigned long epoch;
 
 	if (!sched_cache_enabled())
 		return;
 
-	if (!mm || p->flags & PF_KTHREAD)
-		return;
-
-	grp = READ_ONCE(mm->sched_cache_grp);
-	if (!grp || !grp->pcpu_sched)
+	if (!grp || p->flags & PF_KTHREAD ||
+	    !grp->pcpu_sched)
 		return;
 
 	epoch = rq->cpu_epoch;
@@ -1956,14 +1932,13 @@ static inline void update_avg_scale(u64 *avg, u64 sample)
 
 static void task_cache_work(struct callback_head *work)
 {
+	struct sched_cache_group *grp __free(sched_cache_group_put) = NULL;
+	cpumask_var_t cpus __free(free_cpumask_var) = CPUMASK_VAR_NULL;
 	int cpu, m_a_cpu = -1, nr_running = 0, curr_cpu;
 	unsigned long next_scan, now = jiffies;
 	struct task_struct *p = current, *cur;
 	unsigned long curr_m_a_occ = 0;
-	struct sched_cache_group *grp;
-	struct mm_struct *mm = p->mm;
 	unsigned long m_a_occ = 0;
-	cpumask_var_t cpus;
 
 	WARN_ON_ONCE(work != &p->cache_work);
 
@@ -1972,7 +1947,12 @@ static void task_cache_work(struct callback_head *work)
 	if (p->flags & PF_EXITING)
 		return;
 
-	grp = READ_ONCE(mm->sched_cache_grp);
+	/*
+	 * A reference makes sure grp is not released by others. The rcu
+	 * lock can not be held till after zalloc_cpumask_var() below,
+	 * because the latter might sleep.
+	 */
+	grp = task_cache_group_get(p);
 	if (!grp)
 		return;
 
@@ -1987,8 +1967,8 @@ static void task_cache_work(struct callback_head *work)
 		return;
 
 	curr_cpu = task_cpu(p);
-	if (invalid_llc_nr(mm, p, curr_cpu) ||
-	    exceed_llc_capacity(mm, curr_cpu)) {
+	if (invalid_llc_nr(grp, p, curr_cpu) ||
+	    exceed_llc_capacity(grp, curr_cpu)) {
 		if (READ_ONCE(grp->cpu) != -1)
 			WRITE_ONCE(grp->cpu, -1);
 
@@ -2021,9 +2001,13 @@ static void task_cache_work(struct callback_head *work)
 					m_cpu = i;
 				}
 
+				/*
+				 * rcu_access_pointer() is used because the
+				 * pointer is only compared, never dereferenced.
+				 */
 				cur = rcu_dereference_all(cpu_rq(i)->curr);
 				if (cur && !(cur->flags & (PF_EXITING | PF_KTHREAD)) &&
-				    cur->mm == mm)
+				    rcu_access_pointer(cur->sched_cache_grp) == grp)
 					nr_running++;
 			}
 
@@ -2069,7 +2053,6 @@ static void task_cache_work(struct callback_head *work)
 	}
 
 	update_avg_scale(&grp->nr_running_avg, nr_running);
-	free_cpumask_var(cpus);
 }
 
 void init_sched_mm(struct task_struct *p)
@@ -2078,6 +2061,13 @@ void init_sched_mm(struct task_struct *p)
 
 	init_task_work(work, task_cache_work);
 	work->next = work;
+	/*
+	 * dup_task_struct() copies the parent's task_struct, including its
+	 * sched_cache_grp, for which the child holds no reference.  Clear it
+	 * here - before copy_mm() runs - so the child never carries a
+	 * borrowed pointer that the fork error path would put.
+	 */
+	RCU_INIT_POINTER(p->sched_cache_grp, NULL);
 	/*
 	 * Reset new task's preference to avoid
 	 * polluting account_llc_enqueue().
@@ -3878,10 +3868,9 @@ static void task_numa_placement(struct task_struct *p)
 			 * heuristic and occasional lost updates are tolerable.
 			 *
 			 * If a task exits, its corresponding footprint must
-			 * be subtracted from the mm->sched_cache_grp->footprint,
-			 * otherwise the mm->sched_cache_grp->footprint will not
-			 * converge: the exiting thread's footprint remains
-			 * unchanged/undecayed in mm->sched_cache_grp->footprint.
+			 * be subtracted from p->sched_cache_grp->footprint,
+			 * otherwise the footprint will not converge: the
+			 * exiting thread's footprint remains unchanged/undecayed.
 			 * See exit_mm().
 			 *
 			 * Lost updates and unsynchronized subtraction
@@ -3889,12 +3878,14 @@ static void task_numa_placement(struct task_struct *p)
 			 * go negative. Clamp to zero to prevent the
 			 * unsigned footprint from wrapping.
 			 */
-			grp = READ_ONCE(p->mm->sched_cache_grp);
-			if (!grp)
-				continue;
+			scoped_guard(rcu) {
+				grp = rcu_dereference(p->sched_cache_grp);
 
-			new_fp = (long)READ_ONCE(grp->footprint) + diff;
-			WRITE_ONCE(grp->footprint, max(new_fp, 0L));
+				if (grp) {
+					new_fp = (long)READ_ONCE(grp->footprint) + diff;
+					WRITE_ONCE(grp->footprint, max(new_fp, 0L));
+				}
+			}
 #endif
 		}
 
@@ -10843,7 +10834,6 @@ static enum llc_mig can_migrate_llc_task(struct lb_env *env,
 					 struct task_struct *p)
 {
 	struct sched_cache_group *grp;
-	struct mm_struct *mm;
 	bool to_pref;
 	int cpu, src_cpu, dst_cpu;
 
@@ -10852,11 +10842,7 @@ static enum llc_mig can_migrate_llc_task(struct lb_env *env,
 
 	src_cpu = env->src_cpu;
 	dst_cpu = env->dst_cpu;
-	mm = p->mm;
-	if (!mm)
-		return mig_unrestricted;
-
-	grp = READ_ONCE(mm->sched_cache_grp);
+	grp = rcu_dereference_all(p->sched_cache_grp);
 	if (!grp)
 		return mig_unrestricted;
 
@@ -10865,8 +10851,8 @@ static enum llc_mig can_migrate_llc_task(struct lb_env *env,
 		return mig_unrestricted;
 
 	/* skip cache aware load balance for too many threads */
-	if (invalid_llc_nr(mm, p, dst_cpu) ||
-	    exceed_llc_capacity(mm, dst_cpu)) {
+	if (invalid_llc_nr(grp, p, dst_cpu) ||
+	    exceed_llc_capacity(grp, dst_cpu)) {
 		if (READ_ONCE(grp->cpu) != -1)
 			WRITE_ONCE(grp->cpu, -1);
 		return mig_unrestricted;
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index e656c7059bf8..8b67af28a471 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -4145,6 +4145,9 @@ static inline bool sched_cache_enabled(void)
 	return static_branch_unlikely(&sched_cache_active);
 }
 
+DEFINE_FREE(sched_cache_group_put, struct sched_cache_group *,
+	    sched_cache_group_put(_T));
+
 extern void sched_cache_active_set(void);
 
 #endif
-- 
2.32.0

[PATCH v2 5/6] sched/cache: Skip kernel thread for cache aware scheduling

[Tim Chen]

From: Chen Yu <yu.c.chen@intel.com>

Kernel thread should not be covered by cache aware scheduling as
it borrows the statistics from the user space thread. Filter the
kernel thread in account_mm_sched().

In theory a kernel thread does not have any valid
cache group, so !grp should gate the kernel thread.
Add the PF_KTHREAD check explicitly here for safety
reasons, to guard against future modifications and
to pair with task_tick_cache().

Fixes: df0d98475954 ("sched/cache: Introduce infrastructure for cache-aware load balancing")
Cc: stable@kernel.org #7.2.x
Signed-off-by: Chen Yu <yu.c.chen@intel.com>
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
 kernel/sched/fair.c | 10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 3e2d236be2a3..341d2f9ed72b 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -1805,8 +1805,14 @@ void account_mm_sched(struct rq *rq, struct task_struct *p, s64 delta_exec)
 	/*
 	 * init_task, kthreads and user thread created
 	 * by user_mode_thread() don't have a cache group.
-	 */
-	if (!grp || !grp->pcpu_sched)
+	 * In theory a kernel thread does not have any valid
+	 * cache group, because sched_cache_fork() is not
+	 * invoked for a kernel thread - !grp should gate the
+	 * kernel thread. Add the PF_KTHREAD check explicitly
+	 * here for safety reasons, to guard against future
+	 * modifications and to pair with task_tick_cache().
+	 */
+	if (!grp || p->flags & PF_KTHREAD || !grp->pcpu_sched)
 		return;
 
 	pcpu_sched = per_cpu_ptr(grp->pcpu_sched, cpu_of(rq));
-- 
2.32.0

[PATCH v2 6/6] sched/cache: Refresh LLC capacity across CPU hotplug

[Tim Chen]

From: Davi Chaves Azevedo <davichazbh@gmail.com>

The scheduler scales LLC capacity by the fraction of cache-sharing CPUs
covered by a domain:

  llc_bytes = cache_size * span_weight / shared_weight

During CPU teardown, sched_cpu_deactivate() rebuilds scheduler domains
before cacheinfo_cpu_pre_down() removes the CPU from shared_cpu_map. The
new domains therefore use the old sharing weight. The later call to
sched_update_llc_bytes() looks up the departing CPU's sd_llc, which has
already been detached, and returns without correcting the surviving CPUs.

On a Ryzen 5 7535U with twelve logical CPUs sharing a 16 MiB LLC,
offlining one SMT sibling left the remaining CPUs with:

  llc_bytes = floor(16777216 * 11 / 12) = 15379114 bytes

The correct capacity is still 16777216 bytes. On systems with active
cache-aware scheduling, an underestimated capacity can cause
exceed_llc_capacity() to reject aggregation for a process whose footprint
would fit. Unchanged cpuset partitions sharing the physical cache can
also retain stale capacity when a CPU comes online in another partition.

Pass the cache-sharing mask already retained by cacheinfo to the
scheduler update. Refresh every surviving CPU using its own LLC domain
so that each partition receives the correct share. This also preserves
the correction needed as cache-sharing maps grow during boot.

Keep the existing CPU-hotplug and scheduler-domain synchronization. The
update remains on the hotplug path; no steady-state scheduling operation
or persistent allocation is added.

Fixes: 7030513a0877 ("sched/cache: Calculate the LLC size and store it in sched_domain")
Signed-off-by: Davi Chaves Azevedo <davichazbh@gmail.com>
Reviewed-by: Chen Yu <yu.c.chen@intel.com>
Tested-by: Chen Yu <yu.c.chen@intel.com>
Reviewed-by: Tim Chen <tim.c.chen@linux.intel.com>
Reviewed-by: K Prateek Nayak <kprateek.nayak@amd.com>
Tested-by: K Prateek Nayak <kprateek.nayak@amd.com>
Cc: stable@kernel.org #7.2.x
Signed-off-by: Tim Chen <tim.c.chen@linux.intel.com>
---
 drivers/base/cacheinfo.c       | 11 ++++++-----
 include/linux/sched/topology.h |  4 ++--
 kernel/sched/topology.c        | 22 +++++++++++++---------
 3 files changed, 21 insertions(+), 16 deletions(-)

diff --git a/drivers/base/cacheinfo.c b/drivers/base/cacheinfo.c
index 9f9c72727a05..7a47a392568a 100644
--- a/drivers/base/cacheinfo.c
+++ b/drivers/base/cacheinfo.c
@@ -1040,9 +1040,10 @@ static int cacheinfo_cpu_online(unsigned int cpu)
 	rc = cache_add_dev(cpu);
 	if (rc)
 		goto err;
-	if (cpu_map_shared_cache(true, cpu, &cpu_map))
+	if (cpu_map_shared_cache(true, cpu, &cpu_map)) {
 		update_per_cpu_data_slice_size(true, cpu, cpu_map);
-	sched_update_llc_bytes(cpu);
+		sched_update_llc_bytes(cpu_map);
+	}
 	return 0;
 err:
 	free_cache_attributes(cpu);
@@ -1059,10 +1060,10 @@ static int cacheinfo_cpu_pre_down(unsigned int cpu)
 		cpu_cache_sysfs_exit(cpu);
 
 	free_cache_attributes(cpu);
-	if (nr_shared > 1)
+	if (nr_shared > 1) {
 		update_per_cpu_data_slice_size(false, cpu, cpu_map);
-
-	sched_update_llc_bytes(cpu);
+		sched_update_llc_bytes(cpu_map);
+	}
 
 	return 0;
 }
diff --git a/include/linux/sched/topology.h b/include/linux/sched/topology.h
index b5d9d7c2b8ad..f96812d71c51 100644
--- a/include/linux/sched/topology.h
+++ b/include/linux/sched/topology.h
@@ -281,9 +281,9 @@ static inline int task_node(const struct task_struct *p)
 }
 
 #ifdef CONFIG_SCHED_CACHE
-extern void sched_update_llc_bytes(unsigned int cpu);
+extern void sched_update_llc_bytes(const struct cpumask *cpus);
 #else
-static inline void sched_update_llc_bytes(unsigned int cpu) { }
+static inline void sched_update_llc_bytes(const struct cpumask *cpus) { }
 #endif
 
 #endif /* _LINUX_SCHED_TOPOLOGY_H */
diff --git a/kernel/sched/topology.c b/kernel/sched/topology.c
index 0248227d983a..3dab0253976f 100644
--- a/kernel/sched/topology.c
+++ b/kernel/sched/topology.c
@@ -985,8 +985,8 @@ void sched_cache_active_set(void)
 }
 
 /*
- * Update the bottom sched_domain's llc_bytes for @cpu and all its
- * LLC siblings. Called from cacheinfo_cpu_online() or
+ * Update the bottom sched_domain's llc_bytes for @cpus sharing a physical
+ * LLC. Called from cacheinfo_cpu_online() or
  * cacheinfo_cpu_pre_down() with cpu hotplug lock held.
  *
  * Note: get_effective_llc_bytes() returns 0 on PowerPC.
@@ -996,17 +996,13 @@ void sched_cache_active_set(void)
  * and does not populates the per-CPU struct cpu_cacheinfo array
  * that get_cpu_cacheinfo_llc() reads.
  */
-void sched_update_llc_bytes(unsigned int cpu)
+void sched_update_llc_bytes(const struct cpumask *cpus)
 {
 	struct sched_domain *sd, *sdp;
 	unsigned int i;
 
 	sched_domains_mutex_lock();
 
-	sdp = rcu_dereference_sched_domain(per_cpu(sd_llc, cpu));
-	if (!sdp)
-		goto unlock;
-
 	/*
 	 * ci->shared_cpu_map is built incrementally as CPUs come
 	 * online, so the first CPU in an LLC initially sees
@@ -1014,14 +1010,22 @@ void sched_update_llc_bytes(unsigned int cpu)
 	 * get_effective_llc_bytes().  Re-evaluating every LLC
 	 * sibling on each online event corrects this once the full
 	 * shared_cpu_map is known.
+	 *
+	 * The departing CPU's domains have already been detached when
+	 * cacheinfo removes it. Use the surviving cache siblings instead.
+	 * They may belong to different cpuset partitions, so use each CPU's
+	 * own LLC domain to scale its share of the physical cache.
 	 */
-	for_each_cpu(i, sched_domain_span(sdp)) {
+	for_each_cpu(i, cpus) {
+		sdp = rcu_dereference_sched_domain(per_cpu(sd_llc, i));
+		if (!sdp)
+			continue;
+
 		sd = rcu_dereference_sched_domain(cpu_rq(i)->sd);
 		if (sd)
 			sd->llc_bytes = get_effective_llc_bytes(i, sdp);
 	}
 
-unlock:
 	sched_domains_mutex_unlock();
 }
 
-- 
2.32.0