Re: [PATCH v2 05/13] usb: musb: create debugfs directory under usb root

[Bin Liu]

On Fri, Nov 15, 2019 at 11:21:53AM +0800, Greg Kroah-Hartman wrote:
> On Thu, Nov 14, 2019 at 08:02:34AM -0600, Bin Liu wrote:
> > On Thu, Nov 14, 2019 at 11:20:18AM +0800, Greg Kroah-Hartman wrote:
> > > On Tue, Nov 12, 2019 at 09:28:57AM -0600, Bin Liu wrote:
> > > > Hi,
> > > > 
> > > > On Tue, Nov 12, 2019 at 02:51:51PM +0800, Chunfeng Yun wrote:
> > > > > Now the USB gadget subsystem can use the USB debugfs root directory,
> > > > > so move musb's directory from the root of the debugfs filesystem into
> > > > > the root of usb
> > > > 
> > > > My opinion is this move is unnecessary. I breaks existing debug tools or
> > > > documentation which is already published on Internet. 
> > > 
> > > Having a "root" directory for a single random driver seems like you are
> > > making your driver a "very important" thing in the overall scheme of the
> > > kernel, right?  What's wrong with using the usb subdirectory like all
> > 
> > Agree, it wasn't the right thing to do at the first place. But now
> > changing it adds support burden, because people very often refer to the
> > old information on the internet which no longer matches to the new
> > location. Basically, it is a cost of ABI change.
> 
> What information says that /sys/kernel/debug/mdev/ is the location for
> this?  Is it in-kernel?

No, they are not in-kernel, but many places mainly on https://e2e.ti.com
and http://processors.wiki.ti.com. It basically says to get musb
regdump, cat /sys/kernel/debug/musb-hdrc.{0,1}/regdump, or to enter test
mode, do echo _testmode_ > /sys/kernel/debug/musb-hdrc.{0,1}/testmode...

> > > other USB drivers use (after this patch series is merged)?  That feels
> > > like a much more "sane" way to handle the wide-open debugfs namespace.
> > 
> > Though I commented on this musb patch, my opinion is for this whole
> > series, either drop the whole series or apply the whole series.
> 
> I've applied all but this one and 2 others that did not build properly.

Okay.

> > > Yes, there are no rules when it comes to debugfs file names and
> > > locations, but let's try to be sane please.
> > 
> > Fine with me. I can still support questions such as "can't open
> > /sys/kernel/debug/musb-hdrc.0/testmode: No such file or directory".
> 
> What tool looks for that?

I wrote a usb diagnosis tool called chkusb.sh, which basically checks
usb related information in
  - /proc/config.gz
  - /lib/modules/`uname -r`/kernel/drivers/
  - /proc/device-tree/
  - /sys/device/
  - /sys/kernel/debug/

So whenever I got a report on e2e.ti.com saying that usb doesn't work, I
just post this script and get a log which would give a clue if there was
any fundamental mistake in kernel configuration or deployment.

But not a big issue here for this tool regarding this patch set, I can
update the script to check both locations - /sys/kernel/debug/ and
/sys/kernel/debug/usb/.

-Bin.