sched/mmcid: Implement deferred mode change

When affinity changes cause an increase of the number of CPUs allowed for
tasks which are related to a MM, that might results in a situation where
the ownership mode can go back from per CPU mode to per task mode.

As affinity changes happen with runqueue lock held there is no way to do
the actual mode change and required fixup right there.

Add the infrastructure to defer it to a workqueue. The scheduled work can
race with a fork() or exit(). Whatever happens first takes care of it.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Link: https://patch.msgid.link/20251119172550.216484739@linutronix.de

diff --git a/include/linux/rseq_types.h b/include/linux/rseq_types.h
index a3a4f3f10862460758e868cec1fa34658ddc5531..81fbb8885e8d7bdf163c0a1f2f8985a0a12cce6f 100644 (file)
--- a/include/linux/rseq_types.h
+++ b/include/linux/rseq_types.h
@@ -2,7 +2,9 @@
 #ifndef _LINUX_RSEQ_TYPES_H
 #define _LINUX_RSEQ_TYPES_H
 
+#include <linux/irq_work_types.h>
 #include <linux/types.h>
+#include <linux/workqueue_types.h>
 
 #ifdef CONFIG_RSEQ
 struct rseq;
@@ -122,6 +124,8 @@ struct mm_cid_pcpu {
  * @percpu:            Set, when CIDs are in per CPU mode
  * @transit:           Set to MM_CID_TRANSIT during a mode change transition phase
  * @max_cids:          The exclusive maximum CID value for allocation and convergence
+ * @irq_work:          irq_work to handle the affinity mode change case
+ * @work:              Regular work to handle the affinity mode change case
  * @lock:              Spinlock to protect against affinity setting which can't take @mutex
  * @mutex:             Mutex to serialize forks and exits related to this mm
  * @nr_cpus_allowed:   The number of CPUs in the per MM allowed CPUs map. The map
@@ -139,6 +143,10 @@ struct mm_mm_cid {
        unsigned int            transit;
        unsigned int            max_cids;
 
+       /* Rarely used. Moves @lock and @mutex into the second cacheline */
+       struct irq_work         irq_work;
+       struct work_struct      work;
+
        raw_spinlock_t          lock;
        struct mutex            mutex;
 

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index eb0d59df8acc32b54b8d5d7239dd18835f362b65..cbb543a6efda1beb3463738c544b5c7b542f4088 100644 (file)
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -10539,8 +10539,17 @@ static inline void mm_update_cpus_allowed(struct mm_struct *mm, const struct cpu
 
        /* Adjust the threshold to the wider set */
        mc->pcpu_thrs = mm_cid_calc_pcpu_thrs(mc);
+       /* Switch back to per task mode? */
+       if (mc->users >= mc->pcpu_thrs)
+               return;
+
+       /* Don't queue twice */
+       if (mc->update_deferred)
+               return;
 
-       /* Scheduling of deferred mode switch goes here */
+       /* Queue the irq work, which schedules the real work */
+       mc->update_deferred = true;
+       irq_work_queue(&mc->irq_work);
 }
 
 static inline void mm_cid_transit_to_task(struct task_struct *t, struct mm_cid_pcpu *pcp)
@@ -10553,7 +10562,7 @@ static inline void mm_cid_transit_to_task(struct task_struct *t, struct mm_cid_p
        }
 }
 
-static void __maybe_unused mm_cid_fixup_cpus_to_tasks(struct mm_struct *mm)
+static void mm_cid_fixup_cpus_to_tasks(struct mm_struct *mm)
 {
        unsigned int cpu;
 
@@ -10714,14 +10723,47 @@ void sched_mm_cid_after_execve(struct task_struct *t)
        mm_cid_select(t);
 }
 
-void mm_init_cid(struct mm_struct *mm, struct task_struct *p)
+static void mm_cid_work_fn(struct work_struct *work)
 {
-       struct mm_cid_pcpu __percpu *pcpu = mm->mm_cid.pcpu;
-       int cpu;
+       struct mm_struct *mm = container_of(work, struct mm_struct, mm_cid.work);
 
-       for_each_possible_cpu(cpu)
-               per_cpu_ptr(pcpu, cpu)->cid = MM_CID_UNSET;
+       /* Make it compile, but not functional yet */
+       if (!IS_ENABLED(CONFIG_NEW_MM_CID))
+               return;
+
+       guard(mutex)(&mm->mm_cid.mutex);
+       /* Did the last user task exit already? */
+       if (!mm->mm_cid.users)
+               return;
+
+       scoped_guard(raw_spinlock_irq, &mm->mm_cid.lock) {
+               /* Have fork() or exit() handled it already? */
+               if (!mm->mm_cid.update_deferred)
+                       return;
+               /* This clears mm_cid::update_deferred */
+               if (!mm_update_max_cids(mm))
+                       return;
+               /* Affinity changes can only switch back to task mode */
+               if (WARN_ON_ONCE(mm->mm_cid.percpu))
+                       return;
+       }
+       mm_cid_fixup_cpus_to_tasks(mm);
+}
+
+static void mm_cid_irq_work(struct irq_work *work)
+{
+       struct mm_struct *mm = container_of(work, struct mm_struct, mm_cid.irq_work);
 
+       /*
+        * Needs to be unconditional because mm_cid::lock cannot be held
+        * when scheduling work as mm_update_cpus_allowed() nests inside
+        * rq::lock and schedule_work() might end up in wakeup...
+        */
+       schedule_work(&mm->mm_cid.work);
+}
+
+void mm_init_cid(struct mm_struct *mm, struct task_struct *p)
+{
        mm->mm_cid.max_cids = 0;
        mm->mm_cid.percpu = 0;
        mm->mm_cid.transit = 0;
@@ -10731,6 +10773,8 @@ void mm_init_cid(struct mm_struct *mm, struct task_struct *p)
        mm->mm_cid.update_deferred = 0;
        raw_spin_lock_init(&mm->mm_cid.lock);
        mutex_init(&mm->mm_cid.mutex);
+       mm->mm_cid.irq_work = IRQ_WORK_INIT_HARD(mm_cid_irq_work);
+       INIT_WORK(&mm->mm_cid.work, mm_cid_work_fn);
        cpumask_copy(mm_cpus_allowed(mm), &p->cpus_mask);
        bitmap_zero(mm_cidmask(mm), num_possible_cpus());
 }