On Wed, Feb 01, 2023 at 02:01:18PM -0800, Bart Van Assche wrote:
> On 2/1/23 13:23, Luis Chamberlain wrote:
> > On Wed, Feb 01, 2023 at 12:58:00PM -0800, Luis Chamberlain wrote:
> > > On Mon, Jan 30, 2023 at 01:26:50PM -0800, Bart Van Assche wrote:
> > > > Move the code for creating the block layer debugfs root directory into
> > > > blk-mq-debugfs.c. This patch prepares for adding more debugfs
> > > > initialization code by introducing the function blk_mq_debugfs_init().
> > > >
> > > > Cc: Christoph Hellwig <hch@xxxxxx>
> > > > Cc: Ming Lei <ming.lei@xxxxxxxxxx>
> > > > Cc: Keith Busch <kbusch@xxxxxxxxxx>
> > > > Signed-off-by: Bart Van Assche <bvanassche@xxxxxxx>
> > >
> > > Reviewed-by: Luis Chamberlain <mcgrof@xxxxxxxxxx>
> >
> > Sorry but actually a neuron triggered after this to remind me of commit
> > 85e0cbbb8a ("block: create the request_queue debugfs_dir on
> > registration") and so using the terminology on that commit, wouldn't
> > this not create now the root block debugfs dir for request-based block
> > drivers?
>
> Hi Luis,
>
> This patch should not change any behavior with CONFIG_DEBUG_FS=y.
Ignore CONFIG_DEBUG_FS=y. This is about blktrace which needs it for
both types of drivers.
> As one can see in include/linux/debugfs.h, debugfs_create_dir() does not
> create a directory with CONFIG_DEBUG_FS=n:
>
> static inline struct dentry *debugfs_create_dir(const char *name,
> struct dentry *parent)
> {
> return ERR_PTR(-ENODEV);
> }
>
> I think the only behavior change introduced by this patch is that
> blk_debugfs_root remains NULL with CONFIG_DEBUG_FS=n instead of being set to
> ERR_PTR(-ENODEV).
My point was commit 85e0cbbb8a made blk_debugfs_root non-null always now
when debugfs is enabled for both request-based block drivers and for
make_request block drivers (multiqueue). My reading is that with your
patch blk_debugfs_root will not be created for request-based block
drivers.
Luis
