Re: [PATCH 2/2] tracing/events: block: dev_t via driver core for plug and unplug events

[Bart Van Assche]

On Thu, 2018-04-19 at 12:24 -0700, Omar Sandoval wrote:
> We quiesce and freeze the queue before tearing it down, so it won't be
> NULL while we're still doing I/O. Cc'ing Bart in case I'm lying to you,
> though ;)

blk_cleanup_queue() waits until all I/O requests have finished. Since the
block layer tracing code is triggered from inside the code that processes a
request it is safe to access the request pointer from inside the tracing code.
But I think the question was about the parent of the request queue kobj
instead of about the request queue pointer, the device structure that is
embedded in struct gendisk? I think that parent can disappear at any time.
Most block drivers call del_gendisk() before they call blk_cleanup_queue().
Unless I'm overlooking something I think the request queue will need to
obtain a reference to the gendisk device from inside blk_register_queue()
and drop that reference from inside blk_cleanup_queue() to make it safe for
the tracing code to access struct gendisk.

Bart.