io_uring/sqpoll: fix io-wq affinity when IORING_SETUP_SQPOLL is used

commit ebdfefc09c6de7897962769bd3e63a2ff443ebf5 upstream.

If we setup the ring with SQPOLL, then that polling thread has its
own io-wq setup. This means that if the application uses
IORING_REGISTER_IOWQ_AFF to set the io-wq affinity, we should not be
setting it for the invoking task, but rather the sqpoll task.

Add an sqpoll helper that parks the thread and updates the affinity,
and use that one if we're using SQPOLL.

Fixes: fe76421d1da1 ("io_uring: allow user configurable IO thread CPU affinity")
Cc: stable@vger.kernel.org # 5.10+
Link: https://github.com/axboe/liburing/discussions/884
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>

diff --git a/io_uring/io-wq.c b/io_uring/io-wq.c
index 399e9a15c38d6f152db1bca71ff04f66706ec7be..140f72f51f8af74f8631bb9b6037b0b7b824d8be 100644 (file)
--- a/io_uring/io-wq.c
+++ b/io_uring/io-wq.c
@@ -1285,13 +1285,16 @@ static int io_wq_cpu_offline(unsigned int cpu, struct hlist_node *node)
        return __io_wq_cpu_online(wq, cpu, false);
 }
 
-int io_wq_cpu_affinity(struct io_wq *wq, cpumask_var_t mask)
+int io_wq_cpu_affinity(struct io_uring_task *tctx, cpumask_var_t mask)
 {
+       if (!tctx || !tctx->io_wq)
+               return -EINVAL;
+
        rcu_read_lock();
        if (mask)
-               cpumask_copy(wq->cpu_mask, mask);
+               cpumask_copy(tctx->io_wq->cpu_mask, mask);
        else
-               cpumask_copy(wq->cpu_mask, cpu_possible_mask);
+               cpumask_copy(tctx->io_wq->cpu_mask, cpu_possible_mask);
        rcu_read_unlock();
 
        return 0;

diff --git a/io_uring/io-wq.h b/io_uring/io-wq.h
index 31228426d1924169bb2b94975643fc5d4997f92d..06d9ca90c5771d3ba379d0f96bf8d4cb7da1f01a 100644 (file)
--- a/io_uring/io-wq.h
+++ b/io_uring/io-wq.h
@@ -50,7 +50,7 @@ void io_wq_put_and_exit(struct io_wq *wq);
 void io_wq_enqueue(struct io_wq *wq, struct io_wq_work *work);
 void io_wq_hash_work(struct io_wq_work *work, void *val);
 
-int io_wq_cpu_affinity(struct io_wq *wq, cpumask_var_t mask);
+int io_wq_cpu_affinity(struct io_uring_task *tctx, cpumask_var_t mask);
 int io_wq_max_workers(struct io_wq *wq, int *new_count);
 
 static inline bool io_wq_is_hashed(struct io_wq_work *work)

diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
index 775e44888602d3b464d1532414fd4a230cf1f71e..471f827039799f145fac09351b52eea13a3cb9dc 100644 (file)
--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c
@@ -4177,16 +4177,28 @@ static int io_register_enable_rings(struct io_ring_ctx *ctx)
        return 0;
 }
 
+static __cold int __io_register_iowq_aff(struct io_ring_ctx *ctx,
+                                        cpumask_var_t new_mask)
+{
+       int ret;
+
+       if (!(ctx->flags & IORING_SETUP_SQPOLL)) {
+               ret = io_wq_cpu_affinity(current->io_uring, new_mask);
+       } else {
+               mutex_unlock(&ctx->uring_lock);
+               ret = io_sqpoll_wq_cpu_affinity(ctx, new_mask);
+               mutex_lock(&ctx->uring_lock);
+       }
+
+       return ret;
+}
+
 static __cold int io_register_iowq_aff(struct io_ring_ctx *ctx,
                                       void __user *arg, unsigned len)
 {
-       struct io_uring_task *tctx = current->io_uring;
        cpumask_var_t new_mask;
        int ret;
 
-       if (!tctx || !tctx->io_wq)
-               return -EINVAL;
-
        if (!alloc_cpumask_var(&new_mask, GFP_KERNEL))
                return -ENOMEM;
 
@@ -4207,19 +4219,14 @@ static __cold int io_register_iowq_aff(struct io_ring_ctx *ctx,
                return -EFAULT;
        }
 
-       ret = io_wq_cpu_affinity(tctx->io_wq, new_mask);
+       ret = __io_register_iowq_aff(ctx, new_mask);
        free_cpumask_var(new_mask);
        return ret;
 }
 
 static __cold int io_unregister_iowq_aff(struct io_ring_ctx *ctx)
 {
-       struct io_uring_task *tctx = current->io_uring;
-
-       if (!tctx || !tctx->io_wq)
-               return -EINVAL;
-
-       return io_wq_cpu_affinity(tctx->io_wq, NULL);
+       return __io_register_iowq_aff(ctx, NULL);
 }
 
 static __cold int io_register_iowq_max_workers(struct io_ring_ctx *ctx,

diff --git a/io_uring/sqpoll.c b/io_uring/sqpoll.c
index 5e329e3cd4706196fb6ce643ce35fbf9ad097360..ee2d2c687fdaa954952eb0b1917d3d11b559da90 100644 (file)
--- a/io_uring/sqpoll.c
+++ b/io_uring/sqpoll.c
@@ -421,3 +421,18 @@ err:
        io_sq_thread_finish(ctx);
        return ret;
 }
+
+__cold int io_sqpoll_wq_cpu_affinity(struct io_ring_ctx *ctx,
+                                    cpumask_var_t mask)
+{
+       struct io_sq_data *sqd = ctx->sq_data;
+       int ret = -EINVAL;
+
+       if (sqd) {
+               io_sq_thread_park(sqd);
+               ret = io_wq_cpu_affinity(sqd->thread->io_uring, mask);
+               io_sq_thread_unpark(sqd);
+       }
+
+       return ret;
+}

diff --git a/io_uring/sqpoll.h b/io_uring/sqpoll.h
index e1b8d508d22d155957f2754f21d2a57a252adcbc..8df37e8c914936d777b9d0495796c236f41e5189 100644 (file)
--- a/io_uring/sqpoll.h
+++ b/io_uring/sqpoll.h
@@ -27,3 +27,4 @@ void io_sq_thread_park(struct io_sq_data *sqd);
 void io_sq_thread_unpark(struct io_sq_data *sqd);
 void io_put_sq_data(struct io_sq_data *sqd);
 void io_sqpoll_wait_sq(struct io_ring_ctx *ctx);
+int io_sqpoll_wq_cpu_affinity(struct io_ring_ctx *ctx, cpumask_var_t mask);