With that patch I now get:
[ 40.620619] virbr0: port 1(virbr0-nic) entered disabled state
[ 47.418592] run queue from wrong CPU 3, hctx inactive
[ 47.418602] CPU: 3 PID: 2153 Comm: kworker/3:1H Tainted: G W 4.16.0-rc7+ #27
[ 47.418604] Hardware name: IBM 2964 NC9 704 (LPAR)
[ 47.418613] Workqueue: kblockd blk_mq_run_work_fn
[ 47.418615] Call Trace:
[ 47.418621] ([<0000000000113b86>] show_stack+0x56/0x80)
[ 47.418626] [<0000000000a5cd9a>] dump_stack+0x82/0xb0
[ 47.418627] [<000000000069c4be>] __blk_mq_run_hw_queue+0x136/0x160
[ 47.418631] [<0000000000163906>] process_one_work+0x1be/0x420
[ 47.418633] [<0000000000163bc0>] worker_thread+0x58/0x458
[ 47.418635] [<000000000016a9d0>] kthread+0x148/0x160
[ 47.418639] [<0000000000a7bf3a>] kernel_thread_starter+0x6/0xc
[ 47.418640] [<0000000000a7bf34>] kernel_thread_starter+0x0/0xc
[ 77.670407] run queue from wrong CPU 4, hctx inactive
[ 77.670416] CPU: 4 PID: 2155 Comm: kworker/4:1H Tainted: G W 4.16.0-rc7+ #27
[ 77.670418] Hardware name: IBM 2964 NC9 704 (LPAR)
[ 77.670428] Workqueue: kblockd blk_mq_run_work_fn
[ 77.670430] Call Trace:
[ 77.670436] ([<0000000000113b86>] show_stack+0x56/0x80)
[ 77.670441] [<0000000000a5cd9a>] dump_stack+0x82/0xb0
[ 77.670442] [<000000000069c4be>] __blk_mq_run_hw_queue+0x136/0x160
[ 77.670446] [<0000000000163906>] process_one_work+0x1be/0x420
[ 77.670448] [<0000000000163bc0>] worker_thread+0x58/0x458
[ 77.670450] [<000000000016a9d0>] kthread+0x148/0x160
[ 77.670454] [<0000000000a7bf3a>] kernel_thread_starter+0x6/0xc
[ 77.670455] [<0000000000a7bf34>] kernel_thread_starter+0x0/0xc
On 03/28/2018 04:53 PM, Jens Axboe wrote:
> On 3/28/18 8:38 AM, Jens Axboe wrote:
>> On 3/28/18 1:45 AM, Christian Borntraeger wrote:
>>> FWIW, this patch does not fix the issue for me:
>>
>> Looks like I didn't do the delayed path. How about the below?
>
> OK, final version... This is more in line with what I originally
> suggested.
>
> diff --git a/block/blk-mq.c b/block/blk-mq.c
> index 16e83e6df404..c90016c36a70 100644
> --- a/block/blk-mq.c
> +++ b/block/blk-mq.c
> @@ -1306,6 +1306,10 @@ static void __blk_mq_run_hw_queue(struct blk_mq_hw_ctx *hctx)
> {
> int srcu_idx;
>
> + if (!blk_mq_hw_queue_mapped(hctx) &&
> + !WARN_ON_ONCE(blk_mq_hctx_has_pending(hctx)))
> + return;
> +
> /*
> * We should be running this queue from one of the CPUs that
> * are mapped to it.
> @@ -1399,9 +1403,6 @@ static int blk_mq_hctx_next_cpu(struct blk_mq_hw_ctx *hctx)
> static void __blk_mq_delay_run_hw_queue(struct blk_mq_hw_ctx *hctx, bool async,
> unsigned long msecs)
> {
> - if (WARN_ON_ONCE(!blk_mq_hw_queue_mapped(hctx)))
> - return;
> -
> if (unlikely(blk_mq_hctx_stopped(hctx)))
> return;
>
> @@ -1586,9 +1587,6 @@ static void blk_mq_run_work_fn(struct work_struct *work)
>
> void blk_mq_delay_queue(struct blk_mq_hw_ctx *hctx, unsigned long msecs)
> {
> - if (WARN_ON_ONCE(!blk_mq_hw_queue_mapped(hctx)))
> - return;
> -
> /*
> * Stop the hw queue, then modify currently delayed work.
> * This should prevent us from running the queue prematurely.
>
