Refactor ublk_thread to be a thread-local variable instead of storing it in ublk_dev: - Remove pthread_t thread field from struct ublk_thread and move it to struct ublk_thread_info - Remove struct ublk_thread array from struct ublk_dev, reducing memory footprint - Define struct ublk_thread as local variable in __ublk_io_handler_fn() instead of accessing it from dev->threads[] - Extract main IO handling logic into __ublk_io_handler_fn() which is marked as noinline - Move CPU affinity setup to ublk_io_handler_fn() before calling __ublk_io_handler_fn() - Update ublk_thread_set_sched_affinity() to take struct ublk_thread_info * instead of struct ublk_thread *, and use pthread_setaffinity_np() instead of sched_setaffinity() - Reorder struct ublk_thread fields to group related state together This change makes each thread's ublk_thread structure truly local to the thread, improving cache locality and reducing memory usage. Signed-off-by: Ming Lei --- tools/testing/selftests/ublk/kublk.c | 76 +++++++++++++++------------- tools/testing/selftests/ublk/kublk.h | 9 ++-- 2 files changed, 45 insertions(+), 40 deletions(-) diff --git a/tools/testing/selftests/ublk/kublk.c b/tools/testing/selftests/ublk/kublk.c index 062537ab8976..f8fa102a627f 100644 --- a/tools/testing/selftests/ublk/kublk.c +++ b/tools/testing/selftests/ublk/kublk.c @@ -836,62 +836,70 @@ static int ublk_process_io(struct ublk_thread *t) return reapped; } -static void ublk_thread_set_sched_affinity(const struct ublk_thread *t, - cpu_set_t *cpuset) -{ - if (pthread_setaffinity_np(pthread_self(), sizeof(*cpuset), cpuset) < 0) - ublk_err("ublk dev %u thread %u set affinity failed", - t->dev->dev_info.dev_id, t->idx); -} - struct ublk_thread_info { struct ublk_dev *dev; + pthread_t thread; unsigned idx; sem_t *ready; cpu_set_t *affinity; unsigned long long extra_flags; }; -static void *ublk_io_handler_fn(void *data) +static void ublk_thread_set_sched_affinity(const struct ublk_thread_info *info) { - struct ublk_thread_info *info = data; - struct ublk_thread *t = &info->dev->threads[info->idx]; + if (pthread_setaffinity_np(pthread_self(), sizeof(*info->affinity), info->affinity) < 0) + ublk_err("ublk dev %u thread %u set affinity failed", + info->dev->dev_info.dev_id, info->idx); +} + +static __attribute__((noinline)) int __ublk_io_handler_fn(struct ublk_thread_info *info) +{ + struct ublk_thread t = { + .dev = info->dev, + .idx = info->idx, + }; int dev_id = info->dev->dev_info.dev_id; int ret; - t->dev = info->dev; - t->idx = info->idx; - - /* - * IO perf is sensitive with queue pthread affinity on NUMA machine - * - * Set sched_affinity at beginning, so following allocated memory/pages - * could be CPU/NUMA aware. - */ - if (info->affinity) - ublk_thread_set_sched_affinity(t, info->affinity); - - ret = ublk_thread_init(t, info->extra_flags); + ret = ublk_thread_init(&t, info->extra_flags); if (ret) { ublk_err("ublk dev %d thread %u init failed\n", - dev_id, t->idx); - return NULL; + dev_id, t.idx); + return ret; } sem_post(info->ready); ublk_dbg(UBLK_DBG_THREAD, "tid %d: ublk dev %d thread %u started\n", - gettid(), dev_id, t->idx); + gettid(), dev_id, t.idx); /* submit all io commands to ublk driver */ - ublk_submit_fetch_commands(t); + ublk_submit_fetch_commands(&t); do { - if (ublk_process_io(t) < 0) + if (ublk_process_io(&t) < 0) break; } while (1); ublk_dbg(UBLK_DBG_THREAD, "tid %d: ublk dev %d thread %d exiting\n", - gettid(), dev_id, t->idx); - ublk_thread_deinit(t); + gettid(), dev_id, t.idx); + ublk_thread_deinit(&t); + return 0; +} + +static void *ublk_io_handler_fn(void *data) +{ + struct ublk_thread_info *info = data; + + /* + * IO perf is sensitive with queue pthread affinity on NUMA machine + * + * Set sched_affinity at beginning, so following allocated memory/pages + * could be CPU/NUMA aware. + */ + if (info->affinity) + ublk_thread_set_sched_affinity(info); + + __ublk_io_handler_fn(info); + return NULL; } @@ -989,14 +997,13 @@ static int ublk_start_daemon(const struct dev_ctx *ctx, struct ublk_dev *dev) */ if (dev->nthreads == dinfo->nr_hw_queues) tinfo[i].affinity = &affinity_buf[i]; - pthread_create(&dev->threads[i].thread, NULL, + pthread_create(&tinfo[i].thread, NULL, ublk_io_handler_fn, &tinfo[i]); } for (i = 0; i < dev->nthreads; i++) sem_wait(&ready); - free(tinfo); free(affinity_buf); /* everything is fine now, start us */ @@ -1019,7 +1026,8 @@ static int ublk_start_daemon(const struct dev_ctx *ctx, struct ublk_dev *dev) /* wait until we are terminated */ for (i = 0; i < dev->nthreads; i++) - pthread_join(dev->threads[i].thread, &thread_ret); + pthread_join(tinfo[i].thread, &thread_ret); + free(tinfo); fail: for (i = 0; i < dinfo->nr_hw_queues; i++) ublk_queue_deinit(&dev->q[i]); diff --git a/tools/testing/selftests/ublk/kublk.h b/tools/testing/selftests/ublk/kublk.h index 5e55484fb0aa..fe42705c6d42 100644 --- a/tools/testing/selftests/ublk/kublk.h +++ b/tools/testing/selftests/ublk/kublk.h @@ -175,23 +175,20 @@ struct ublk_queue { struct ublk_thread { struct ublk_dev *dev; - struct io_uring ring; - unsigned int cmd_inflight; - unsigned int io_inflight; - - pthread_t thread; unsigned idx; #define UBLKS_T_STOPPING (1U << 0) #define UBLKS_T_IDLE (1U << 1) unsigned state; + unsigned int cmd_inflight; + unsigned int io_inflight; + struct io_uring ring; }; struct ublk_dev { struct ublk_tgt tgt; struct ublksrv_ctrl_dev_info dev_info; struct ublk_queue q[UBLK_MAX_QUEUES]; - struct ublk_thread threads[UBLK_MAX_THREADS]; unsigned nthreads; unsigned per_io_tasks; -- 2.47.0