initio.h header user confusion

initio.h header user confusion

[Srdjan Todorovic]

Hi,

I'm working on a set of patches for the initio scsi driver and I found
some strangeness with its header file, initio.h. I noticed that
sun3_scsi.c and sun3_scsi_vme.c include the header, and it uses defines
such as SCSI_ABORT_SNOOZE (defined in initio.h).

I did some grepping and am now confused. Grepping the entire kernel tree
like so:

  grep -r SCSI_ABORT_ ~/kernelcoding/linux-2.6-rc-latest/*

gives me a handful of scsi drivers that use these defines.

Grepping the kernel source for these defines like so:

   grep -r SCSI_ABORT_ ~/kernelcoding/linux-2.6-rc-latest/* | grep define

only gives me drivers/scsi/initio.h, so they are only defined in
initio.h

Grepping the whole kernel source for drivers/headers that include
initio.h like so:
   grep -r "initio.h" ~/kernelcoding/linux-2.6-rc-latest/*

only gives me 3 results - initio.c, sun3_scsi.c and sun3_scsi_vme.c.

So how is it possible that other drivers such as aha1542.c use these defines?

Can anyone please explain this as it's really confused me, and I want
to make sure that my patches don't break other drivers when I post
them here.

Many thanks,

Srdjan

Re: initio.h header user confusion

[Matthew Wilcox]

On Wed, Feb 21, 2007 at 04:12:06PM +0000, Srdjan Todorovic wrote:
> So how is it possible that other drivers such as aha1542.c use these defines?

Possibly, they don't.  I just looked at the ncr53c8xx driver, and the
only uses of SCSI_ABORT_* are inside an #if 0 block.  The same is true
for aha1542.

Re: initio.h header user confusion

[Srdjan Todorovic]

On Wed, 21 Feb 2007, Matthew Wilcox wrote:

> On Wed, Feb 21, 2007 at 04:12:06PM +0000, Srdjan Todorovic wrote:
> > So how is it possible that other drivers such as aha1542.c use these defines?
> 
> Possibly, they don't.  I just looked at the ncr53c8xx driver, and the
> only uses of SCSI_ABORT_* are inside an #if 0 block.  The same is true
> for aha1542.

Those pesky #if 0 blocks are annoying.
I just looked at aha1542.c and saw layered #if 0 blocks. :-|

I'll recheck the other drivers then. It would seem the only users then
are initio and sun3*. What's the best thing to do about it?
Should I keep things as they are with SCSI_ABORT_* or should I
move them to include/scsi/ or something entirely different?

Thanks,

Srdjan

Re: initio.h header user confusion

[Matthew Wilcox]

On Wed, Feb 21, 2007 at 04:36:59PM +0000, Srdjan Todorovic wrote:
> I'll recheck the other drivers then. It would seem the only users then
> are initio and sun3*. What's the best thing to do about it?
> Should I keep things as they are with SCSI_ABORT_* or should I
> move them to include/scsi/ or something entirely different?

These drivers ought to be moved away from using SCSI_ABORT_*.  If you
don't have the expertise and hardware to do this, best to leave them
alone.

Re: initio.h header user confusion

[James Bottomley]

On Wed, 2007-02-21 at 16:12 +0000, Srdjan Todorovic wrote:
> I'm working on a set of patches for the initio scsi driver and I found
> some strangeness with its header file, initio.h. I noticed that
> sun3_scsi.c and sun3_scsi_vme.c include the header, and it uses defines
> such as SCSI_ABORT_SNOOZE (defined in initio.h).

It's a remnant of the old 2.4 error handler and should be excised.

In early 2.4 or perhaps even 2.2 the error handler ran in interrupt
context, so there was no way of waiting for an abort to complete, hence
this return value (it told the eh to wait a while for the abort).

In 2.6 all eh routines run in the context of the host eh thread, so you
can use standard msleep() etc. calls to wait for aborts to timeout or
complete.

James

Re: initio.h header user confusion

[Srdjan Todorovic]

On Wed, 21 Feb 2007, James Bottomley wrote:

> On Wed, 2007-02-21 at 16:12 +0000, Srdjan Todorovic wrote:
> > I'm working on a set of patches for the initio scsi driver and I found
> > some strangeness with its header file, initio.h. I noticed that
> > sun3_scsi.c and sun3_scsi_vme.c include the header, and it uses defines
> > such as SCSI_ABORT_SNOOZE (defined in initio.h).
> 
> It's a remnant of the old 2.4 error handler and should be excised.
> 
> In early 2.4 or perhaps even 2.2 the error handler ran in interrupt
> context, so there was no way of waiting for an abort to complete, hence
> this return value (it told the eh to wait a while for the abort).
> 
> In 2.6 all eh routines run in the context of the host eh thread, so you
> can use standard msleep() etc. calls to wait for aborts to timeout or
> complete.
> 
> James

Thanks for a good short explanation on the scsi error handling.

I poked around a bit more and all but one cases of SCSI_RESET_* are
in #if 0 blocks. The one case is tul_reset_scsi() which returns a
SCSI_RESET_SUCCESS value, which is discarded by the caller anyway.

SCSI_ABORT_* is only used in tul_abort_srb(). This doesn't use the
new error handler threads.

As Matthew Wilcox suggested in a previous email in this thread,
I don't have the expertise to fix this (though I do have the hardware).
I'd like to read up on scsi eh just the same, and might have a go
at fixing this at some point.

Thanks

Srdjan