Re: [PATCH 1/6] usb:gadget:mass-storage: Improve the signature of SCSI handler function

On Sun, Jun 27, 2021 at 06:32:03PM +0300, i.kononenko wrote:

Good morning, Alan!

First of all, thank you for your time to review my first patchset for 
the Linux Kernel and valuable advice on the right way of patchwriting!

On 27.06.2021 17:18, Alan Stern wrote:

quoted

On Sun, Jun 27, 2021 at 12:18:14AM +0300, Igor Kononenko wrote:

quoted

SCSI command handlers currently have an ambiguous return value. This

(I dislike very much this way of writing patch descriptions.  Unless
the reader has already looked at the email subject line and remembers
that this patch affects the mass-storage gadget, he will think the
sentence above is talking about command handlers in the SCSI core -- a
completely different part of the kernel.  When writing patch
descriptions, please do not assume that the reader already knows what
the patch is about.)

quoted

return value may indicate the length of the data written to the response
buffer and the command's processing status. Thus, the understanding of
command handling may be implicit.

First of all, thank you for your time to review my first patchset for the
Linux Kernel and valuable advice on the right way of patchwriting!

I noticed that the status/datasize return value pattern is pervasive for 
Linux and used through many subsystems. But for the f_mass_storage.c,
such approach use case is not documented anywhere, and implementation has 
too many magic-constant, e.g.

static int do_inquiry(struct fsg_common *common, struct fsg_buffhd *bh)
{
   ....
   return 36;
}

IMHO, this way is not giving the developer an explicit understanding of 
'what is the 36' and its origin.
If moving to the suggested way is unwanted, I'd keep the implementation 
as is with additional documentation for each function where uses this 
approach.

Since every one of the command handler functions uses this convention, 
it would be wasteful to have separate documentation of the return value 
for each function.  A single documentation comment that covers all the 
command handlers would be acceptable.

Additionally, I guess, define clarify macros of return value instead of 
magic numbers is required.

If you want, okay.  That should go in a separate patch from the 
documentation patch.

Also, since the return values are different for each command handler, I 
suggest that the macro definitions be placed along with the handler 
functions and not in a separate header file.  Having a separate file for 
these macros would not make any sense, because the values do not need to 
be shared across multiple functions or source files.

quoted

The return value is _not_ ambiguous.  If the value is >= 0 then it is
a data length, otherwise it is a status.  Yes, this is implicit, but it
is a very common pattern used throughout the kernel and everyone
understands it.

quoted

After this patch, the output buffer's size will be set in the
'data_size_to_handle' field of 'struct fsg_common', and the command
handler's return value indicates only the processing status.

What is the reason for making this change?  Does it fix any problems
or prepare the way for any future patches?  It seems like this is
completely unnecessary.

Yes, the patch uses as part of the incoming implementation of refactoring
'usb:gadget:mass-storage:scsi' command handling.

That incoming implementation uses the refactored command handling but 
doesn't depend on the refactoring.  It could just as easily use the 
existing command handling.

I believed the suggested improvement would be useful for the community as 
an improvement of code.

Unless you can provide a convincing reason for this change, it doesn't 
seem like an improvement to me.  It's no easier to read or understand, 
and it doesn't improve execution speed on a critical pathway.  It just 
seems like pointless code churn.

Alan Stern