[PATCH v2 bpf-next 07/13] bpf: Change bpf_mem_cache draining process.

[Alexei Starovoitov <alexei.starovoitov@xxxxxxxxx>]

From: Alexei Starovoitov <ast@xxxxxxxxxx>

The next patch will introduce cross-cpu llist access and existing
irq_work_sync() + drain_mem_cache() + rcu_barrier_tasks_trace() mechanism will
not be enough, since irq_work_sync() + drain_mem_cache() on cpu A won't
guarantee that llist on cpu A are empty. The free_bulk() on cpu B might add
objects back to llist of cpu A. Add 'bool draining' flag and set it all cpus
before proceeding with irq_work_sync.
The modified sequence looks like:
for_each_cpu:
  WRITE_ONCE(c->draining, true); // make RCU callback a nop
  irq_work_sync(); // wait for irq_work callback (free_bulk) to finish
for_each_cpu:
  drain_mem_cache(); // free all objects
rcu_barrier_tasks_trace(); // wait for RCU callbacks to execute as a nop

Signed-off-by: Alexei Starovoitov <ast@xxxxxxxxxx>
---
 kernel/bpf/memalloc.c | 24 +++++++++++++++---------
 1 file changed, 15 insertions(+), 9 deletions(-)

diff --git a/kernel/bpf/memalloc.c b/kernel/bpf/memalloc.c
index 4fd79bd51f5a..d68a854f45ee 100644
--- a/kernel/bpf/memalloc.c
+++ b/kernel/bpf/memalloc.c
@@ -98,6 +98,7 @@ struct bpf_mem_cache {
 	int free_cnt;
 	int low_watermark, high_watermark, batch;
 	int percpu_size;
+	bool draining;
 
 	/* list of objects to be freed after RCU tasks trace GP */
 	struct llist_head free_by_rcu_ttrace;
@@ -252,7 +253,10 @@ static void __free_rcu(struct rcu_head *head)
 {
 	struct bpf_mem_cache *c = container_of(head, struct bpf_mem_cache, rcu_ttrace);
 
+	if (unlikely(READ_ONCE(c->draining)))
+		goto out;
 	free_all(llist_del_all(&c->waiting_for_gp_ttrace), !!c->percpu_size);
+out:
 	atomic_set(&c->call_rcu_ttrace_in_progress, 0);
 }
 
@@ -542,16 +546,11 @@ void bpf_mem_alloc_destroy(struct bpf_mem_alloc *ma)
 		rcu_in_progress = 0;
 		for_each_possible_cpu(cpu) {
 			c = per_cpu_ptr(ma->cache, cpu);
-			/*
-			 * refill_work may be unfinished for PREEMPT_RT kernel
-			 * in which irq work is invoked in a per-CPU RT thread.
-			 * It is also possible for kernel with
-			 * arch_irq_work_has_interrupt() being false and irq
-			 * work is invoked in timer interrupt. So waiting for
-			 * the completion of irq work to ease the handling of
-			 * concurrency.
-			 */
+			WRITE_ONCE(c->draining, true);
 			irq_work_sync(&c->refill_work);
+		}
+		for_each_possible_cpu(cpu) {
+			c = per_cpu_ptr(ma->cache, cpu);
 			drain_mem_cache(c);
 			rcu_in_progress += atomic_read(&c->call_rcu_ttrace_in_progress);
 		}
@@ -566,7 +565,14 @@ void bpf_mem_alloc_destroy(struct bpf_mem_alloc *ma)
 			cc = per_cpu_ptr(ma->caches, cpu);
 			for (i = 0; i < NUM_CACHES; i++) {
 				c = &cc->cache[i];
+				WRITE_ONCE(c->draining, true);
 				irq_work_sync(&c->refill_work);
+			}
+		}
+		for_each_possible_cpu(cpu) {
+			cc = per_cpu_ptr(ma->caches, cpu);
+			for (i = 0; i < NUM_CACHES; i++) {
+				c = &cc->cache[i];
 				drain_mem_cache(c);
 				rcu_in_progress += atomic_read(&c->call_rcu_ttrace_in_progress);
 			}
-- 
2.34.1