Re: [PATCH 0/2] use DEFINE_DEBUGFS_ATTRIBUTE instead of DEFINE_SIMPLE_ATTRIBUTE

[Rohit Sarkar]

On Sun, Mar 29, 2020 at 03:46:17PM +0200, Lars-Peter Clausen wrote:
> On 3/29/20 1:34 PM, Rohit Sarkar wrote:
> > On Sun, Mar 29, 2020 at 10:38:18AM +0100, Jonathan Cameron wrote:
> > > On Sat, 28 Mar 2020 12:04:53 +0530
> > > Rohit Sarkar <rohitsarkar5398@xxxxxxxxx> wrote:
> > > 
> > > > The debugfs_create_file_unsafe method does not protect the fops given to
> > > > it from file removal. It must be used with DEFINE_DEBUGFS_ATTRIBUTE
> > > > which makes the fops aware of the file lifetime.
> > > > 
> > > > Further using DEFINE_DEBUGFS_ATTRIBUTE along with
> > > > debugfs_create_file_unsafe significantly reduces the overhead introduced by
> > > > debugfs_create_file which creates a lifetime managing proxy around each
> > > > fops handed in. Refer [1] for more on this.
> > > > 
> > > > Fixes the following warnings reported by coccinelle:
> > > > drivers/iio/imu//adis16460.c:126:0-23: WARNING: adis16460_flash_count_fops should be defined with DEFINE_DEBUGFS_ATTRIBUTE
> > > > drivers/iio/imu//adis16460.c:108:0-23: WARNING: adis16460_product_id_fops should be defined with DEFINE_DEBUGFS_ATTRIBUTE
> > > > drivers/iio/imu//adis16460.c:90:0-23: WARNING: adis16460_serial_number_fops should be defined with DEFINE_DEBUGFS_ATTRIBUTE
> > > > drivers/iio/imu//adis16400.c:278:0-23: WARNING: adis16400_flash_count_fops should be defined with DEFINE_DEBUGFS_ATTRIBUTE
> > > > drivers/iio/imu//adis16400.c:261:0-23: WARNING: adis16400_product_id_fops should be defined with DEFINE_DEBUGFS_ATTRIBUTE
> > > > 
> > > > [1]: https://lists.gt.net/linux/kernel/2369498
> > > > 
> > > > Rohit Sarkar (2):
> > > >    iio: imu: adis16400: use DEFINE_DEBUGFS_ATTRIBUTE instead of
> > > >      DEFINE_SIMPLE_ATTRIBUTE
> > > >    iio: imu: adis16460: use DEFINE_DEBUGFS_ATTRIBUTE instead of
> > > >      DEFINE_SIMPLE_ATTRIBUTE
> > > > 
> > > >   drivers/iio/imu/adis16400.c | 4 ++--
> > > >   drivers/iio/imu/adis16460.c | 6 +++---
> > > >   2 files changed, 5 insertions(+), 5 deletions(-)
> > > > 
> > > Hi Rohit,
> > Hey,
> > > You've opened a can of worms with this one.  There as a previous series
> > > posted doing exactly this change back in 2019 by Zhong Jiang (cc'd)
> > > 
> > > At the time I did a bit of looking into why this had been universally taken
> > > up cross tree and turned out there are some potential issues.
> > > 
> > > Alexandru added it to the list of things to test, but I guess it got
> > > buried under other work and is still outstanding.
> > > 
> > > https://lkml.org/lkml/2019/10/30/144
> > Acc. to the patch by Zhong this change kind of comes off as a cosmetic
> > change as in the commit message he mentions "it is more clear".
> > 
> > But there is certainly more to it than that:
> > In the current scenario since we are using debugfs_create_file_unsafe
> > the file has no protection whatsoever against removal.
> 
> The drivers you are patching all use debugfs_create_file() as far as I can
> see.
Ah, You are right. I dont know why I assumed that
debugfs_create_file_unsafe was being used. Was probably sleepy when I
sent this out :/
> The way I understand it using DEFINE_DEBUGFS_ATTRIBUTE without switching to
> debugfs_create_file_unsafe() will not make a difference. There will only be
> more overhead since the files are protected twice.
That's right.
it should either be ...unsafe() with DEFINE_DEBUGFS_ATTRIBUTE or what it
is currently.

In the current scenario the file is protected against removal but there
is extra overhead because debugfs_create_file creates a lifetime
managing proxy AFAIK.

I can send a v2 changing the debugfs_create_file function to unsafe if
someone can test it out on h/w as Jonathan suggested.
> - Lars
> 
Thanks,
Rohit