sched_ext: Add lockless peek operation for DSQs

The builtin DSQ queue data structures are meant to be used by a wide
range of different sched_ext schedulers with different demands on these
data structures. They might be per-cpu with low-contention, or
high-contention shared queues. Unfortunately, DSQs have a coarse-grained
lock around the whole data structure. Without going all the way to a
lock-free, more scalable implementation, a small step we can take to
reduce lock contention is to allow a lockless, small-fixed-cost peek at
the head of the queue.

This change allows certain custom SCX schedulers to cheaply peek at
queues, e.g. during load balancing, before locking them. But it
represents a few extra memory operations to update the pointer each
time the DSQ is modified, including a memory barrier on ARM so the write
appears correctly ordered.

This commit adds a first_task pointer field which is updated
atomically when the DSQ is modified, and allows any thread to peek at
the head of the queue without holding the lock.

Signed-off-by: Ryan Newton <newton@meta.com>
Reviewed-by: Andrea Righi <arighi@nvidia.com>
Reviewed-by: Christian Loehle <christian.loehle@arm.com>
Signed-off-by: Tejun Heo <tj@kernel.org>

diff --git a/include/linux/sched/ext.h b/include/linux/sched/ext.h
index 9848aeab278649277c56d614403eb39ab73adca9..4713f374acc0e05165ecc9b0c870977f0a6a82bf 100644 (file)
--- a/include/linux/sched/ext.h
+++ b/include/linux/sched/ext.h
@@ -58,6 +58,7 @@ enum scx_dsq_id_flags {
  */
 struct scx_dispatch_q {
        raw_spinlock_t          lock;
+       struct task_struct __rcu *first_task; /* lockless peek at head */
        struct list_head        list;   /* tasks in dispatch order */
        struct rb_root          priq;   /* used to order by p->scx.dsq_vtime */
        u32                     nr;

diff --git a/kernel/sched/ext.c b/kernel/sched/ext.c
index 430749ce46ab37320caca408aef5822394a244f3..f9c0888ef2796da7c2d642e95ac2455f668fb3f1 100644 (file)
--- a/kernel/sched/ext.c
+++ b/kernel/sched/ext.c
@@ -965,8 +965,11 @@ static void dispatch_enqueue(struct scx_sched *sch, struct scx_dispatch_q *dsq,
                                container_of(rbp, struct task_struct,
                                             scx.dsq_priq);
                        list_add(&p->scx.dsq_list.node, &prev->scx.dsq_list.node);
+                       /* first task unchanged - no update needed */
                } else {
                        list_add(&p->scx.dsq_list.node, &dsq->list);
+                       /* not builtin and new task is at head - use fastpath */
+                       rcu_assign_pointer(dsq->first_task, p);
                }
        } else {
                /* a FIFO DSQ shouldn't be using PRIQ enqueuing */
@@ -974,10 +977,19 @@ static void dispatch_enqueue(struct scx_sched *sch, struct scx_dispatch_q *dsq,
                        scx_error(sch, "DSQ ID 0x%016llx already had PRIQ-enqueued tasks",
                                  dsq->id);
 
-               if (enq_flags & (SCX_ENQ_HEAD | SCX_ENQ_PREEMPT))
+               if (enq_flags & (SCX_ENQ_HEAD | SCX_ENQ_PREEMPT)) {
                        list_add(&p->scx.dsq_list.node, &dsq->list);
-               else
+                       /* new task inserted at head - use fastpath */
+                       if (!(dsq->id & SCX_DSQ_FLAG_BUILTIN))
+                               rcu_assign_pointer(dsq->first_task, p);
+               } else {
+                       bool was_empty;
+
+                       was_empty = list_empty(&dsq->list);
                        list_add_tail(&p->scx.dsq_list.node, &dsq->list);
+                       if (was_empty && !(dsq->id & SCX_DSQ_FLAG_BUILTIN))
+                               rcu_assign_pointer(dsq->first_task, p);
+               }
        }
 
        /* seq records the order tasks are queued, used by BPF DSQ iterator */
@@ -1032,6 +1044,13 @@ static void task_unlink_from_dsq(struct task_struct *p,
                p->scx.dsq_flags &= ~SCX_TASK_DSQ_ON_PRIQ;
        }
 
+       if (!(dsq->id & SCX_DSQ_FLAG_BUILTIN) && dsq->first_task == p) {
+               struct task_struct *first_task;
+
+               first_task = nldsq_next_task(dsq, NULL, false);
+               rcu_assign_pointer(dsq->first_task, first_task);
+       }
+
        list_del_init(&p->scx.dsq_list.node);
        dsq_mod_nr(dsq, -1);
 }
@@ -6292,6 +6311,40 @@ __bpf_kfunc void bpf_iter_scx_dsq_destroy(struct bpf_iter_scx_dsq *it)
        kit->dsq = NULL;
 }
 
+/**
+ * scx_bpf_dsq_peek - Lockless peek at the first element.
+ * @dsq_id: DSQ to examine.
+ *
+ * Read the first element in the DSQ. This is semantically equivalent to using
+ * the DSQ iterator, but is lockfree. Of course, like any lockless operation,
+ * this provides only a point-in-time snapshot, and the contents may change
+ * by the time any subsequent locking operation reads the queue.
+ *
+ * Returns the pointer, or NULL indicates an empty queue OR internal error.
+ */
+__bpf_kfunc struct task_struct *scx_bpf_dsq_peek(u64 dsq_id)
+{
+       struct scx_sched *sch;
+       struct scx_dispatch_q *dsq;
+
+       sch = rcu_dereference(scx_root);
+       if (unlikely(!sch))
+               return NULL;
+
+       if (unlikely(dsq_id & SCX_DSQ_FLAG_BUILTIN)) {
+               scx_error(sch, "peek disallowed on builtin DSQ 0x%llx", dsq_id);
+               return NULL;
+       }
+
+       dsq = find_user_dsq(sch, dsq_id);
+       if (unlikely(!dsq)) {
+               scx_error(sch, "peek on non-existent DSQ 0x%llx", dsq_id);
+               return NULL;
+       }
+
+       return rcu_dereference(dsq->first_task);
+}
+
 __bpf_kfunc_end_defs();
 
 static s32 __bstr_format(struct scx_sched *sch, u64 *data_buf, char *line_buf,
@@ -6851,6 +6904,7 @@ BTF_ID_FLAGS(func, scx_bpf_task_set_dsq_vtime, KF_RCU);
 BTF_ID_FLAGS(func, scx_bpf_kick_cpu)
 BTF_ID_FLAGS(func, scx_bpf_dsq_nr_queued)
 BTF_ID_FLAGS(func, scx_bpf_destroy_dsq)
+BTF_ID_FLAGS(func, scx_bpf_dsq_peek, KF_RCU_PROTECTED | KF_RET_NULL)
 BTF_ID_FLAGS(func, bpf_iter_scx_dsq_new, KF_ITER_NEW | KF_RCU_PROTECTED)
 BTF_ID_FLAGS(func, bpf_iter_scx_dsq_next, KF_ITER_NEXT | KF_RET_NULL)
 BTF_ID_FLAGS(func, bpf_iter_scx_dsq_destroy, KF_ITER_DESTROY)

diff --git a/tools/sched_ext/include/scx/common.bpf.h b/tools/sched_ext/include/scx/common.bpf.h
index eb3c99445cb3e8ed90880484b0917718731ac173..e65b1eb668ea5ed0c381d670e0a69426db6dd814 100644 (file)
--- a/tools/sched_ext/include/scx/common.bpf.h
+++ b/tools/sched_ext/include/scx/common.bpf.h
@@ -74,6 +74,7 @@ u32 scx_bpf_reenqueue_local(void) __ksym;
 void scx_bpf_kick_cpu(s32 cpu, u64 flags) __ksym;
 s32 scx_bpf_dsq_nr_queued(u64 dsq_id) __ksym;
 void scx_bpf_destroy_dsq(u64 dsq_id) __ksym;
+struct task_struct *scx_bpf_dsq_peek(u64 dsq_id) __ksym __weak;
 int bpf_iter_scx_dsq_new(struct bpf_iter_scx_dsq *it, u64 dsq_id, u64 flags) __ksym __weak;
 struct task_struct *bpf_iter_scx_dsq_next(struct bpf_iter_scx_dsq *it) __ksym __weak;
 void bpf_iter_scx_dsq_destroy(struct bpf_iter_scx_dsq *it) __ksym __weak;

diff --git a/tools/sched_ext/include/scx/compat.bpf.h b/tools/sched_ext/include/scx/compat.bpf.h
index e487c10b5e07f584b60d16119857d006892f3ddf..619a16f0d39acc92a42cbe7a91d54571772389f7 100644 (file)
--- a/tools/sched_ext/include/scx/compat.bpf.h
+++ b/tools/sched_ext/include/scx/compat.bpf.h
@@ -26,6 +26,24 @@ int bpf_cpumask_populate(struct cpumask *dst, void *src, size_t src__sz) __ksym
        (bpf_ksym_exists(bpf_cpumask_populate) ?                        \
         (bpf_cpumask_populate(cpumask, src, size__sz)) : -EOPNOTSUPP)
 
+/*
+ * v6.19: Introduce lockless peek API for user DSQs.
+ *
+ * Preserve the following macro until v6.21.
+ */
+static inline struct task_struct *__COMPAT_scx_bpf_dsq_peek(u64 dsq_id)
+{
+       struct task_struct *p = NULL;
+       struct bpf_iter_scx_dsq it;
+
+       if (bpf_ksym_exists(scx_bpf_dsq_peek))
+               return scx_bpf_dsq_peek(dsq_id);
+       if (!bpf_iter_scx_dsq_new(&it, dsq_id, 0))
+               p = bpf_iter_scx_dsq_next(&it);
+       bpf_iter_scx_dsq_destroy(&it);
+       return p;
+}
+
 /**
  * __COMPAT_is_enq_cpu_selected - Test if SCX_ENQ_CPU_SELECTED is on
  * in a compatible way. We will preserve this __COMPAT helper until v6.16.