blk-mq: use percpu csd to remote complete instead of per-rq csd
authorChengming Zhou <zhouchengming@bytedance.com>
Mon, 17 Jul 2023 04:00:55 +0000 (12:00 +0800)
committerJens Axboe <axboe@kernel.dk>
Mon, 17 Jul 2023 14:18:21 +0000 (08:18 -0600)
If request need to be completed remotely, we insert it into percpu llist,
and smp_call_function_single_async() if llist is empty previously.

We don't need to use per-rq csd, percpu csd is enough. And the size of
struct request is decreased by 24 bytes.

This way is cleaner, and looks correct, given block softirq is guaranteed
to be scheduled to consume the list if one new request is added to this
percpu list, either smp_call_function_single_async() returns -EBUSY or 0.

Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20230717040058.3993930-2-chengming.zhou@linux.dev
Signed-off-by: Jens Axboe <axboe@kernel.dk>
block/blk-mq.c
include/linux/blk-mq.h

index d50b1d62a3d923fc050c53802d87fb08cd9aab49..d9865486961510547eaaa92a281641b4ca8a61cb 100644 (file)
@@ -43,6 +43,7 @@
 #include "blk-ioprio.h"
 
 static DEFINE_PER_CPU(struct llist_head, blk_cpu_done);
+static DEFINE_PER_CPU(call_single_data_t, blk_cpu_csd);
 
 static void blk_mq_insert_request(struct request *rq, blk_insert_t flags);
 static void blk_mq_request_bypass_insert(struct request *rq,
@@ -1157,15 +1158,11 @@ static inline bool blk_mq_complete_need_ipi(struct request *rq)
 
 static void blk_mq_complete_send_ipi(struct request *rq)
 {
-       struct llist_head *list;
        unsigned int cpu;
 
        cpu = rq->mq_ctx->cpu;
-       list = &per_cpu(blk_cpu_done, cpu);
-       if (llist_add(&rq->ipi_list, list)) {
-               INIT_CSD(&rq->csd, __blk_mq_complete_request_remote, rq);
-               smp_call_function_single_async(cpu, &rq->csd);
-       }
+       if (llist_add(&rq->ipi_list, &per_cpu(blk_cpu_done, cpu)))
+               smp_call_function_single_async(cpu, &per_cpu(blk_cpu_csd, cpu));
 }
 
 static void blk_mq_raise_softirq(struct request *rq)
@@ -4829,6 +4826,9 @@ static int __init blk_mq_init(void)
 
        for_each_possible_cpu(i)
                init_llist_head(&per_cpu(blk_cpu_done, i));
+       for_each_possible_cpu(i)
+               INIT_CSD(&per_cpu(blk_cpu_csd, i),
+                        __blk_mq_complete_request_remote, NULL);
        open_softirq(BLOCK_SOFTIRQ, blk_done_softirq);
 
        cpuhp_setup_state_nocalls(CPUHP_BLOCK_SOFTIRQ_DEAD,
index b96e00499f9eee3eaa8bc2757bd7c096ce4d86a0..67f810857634071e6d96120fa16bfd76dfbc21ac 100644 (file)
@@ -182,10 +182,7 @@ struct request {
                rq_end_io_fn            *saved_end_io;
        } flush;
 
-       union {
-               struct __call_single_data csd;
-               u64 fifo_time;
-       };
+       u64 fifo_time;
 
        /*
         * completion callback.