Hi, 在 2026/2/13 12:43, Ming Lei 写道: > On Thu, Feb 12, 2026 at 04:01:35PM +0800, Yu Kuai wrote: >> Creating debugfs entries can trigger fs reclaim, which can enter back >> into the block layer request_queue. This can cause deadlock if the >> queue is frozen. >> >> Previously, a WARN_ON_ONCE check was used in debugfs_create_files() >> to detect this condition, but it was racy since the queue can be frozen >> from another context at any time. > debugfs lock doesn't depend on queue freeze, so the implied deadlock isn't > obvious. > > Can you explain a bit why freeze from another context may cause dead lock?
It's probably fine to freeze from another context concurrently, I mean the checking will cause the false warning like: Re: [bug report] WARNING: block/blk-mq-debugfs.c:620 at debugfs_create_files.isra.0+0x54/0x60, CPU#220: kworker/u1036:7/70421 - Shinichiro Kawasaki <https://lore.kernel.org/all/aYWQR7CtYdk3K39g@shinmob/> So checking itself is racy and should be removed, meanwhile there are two options: 1) use noio for debugfs creation; 2) move debugfs creation out of queue frozen from nvme target; BTW, I added the blk_queue_enter/exit() protection for 2 because I think it's safe, and can prevent deadlock if there are any implied deadlock related to queue frozen. > > > Thanks, > Ming > > -- Thansk, Kuai
