Hi Aaro,
>
> On Sun, Jan 26, 2014 at 10:56:38PM -0800, Dmitry Torokhov wrote:
> > On Mon, Jan 27, 2014 at 12:32:36AM +0000, Raghu Gandham wrote:
> > > > On Sat, Jan 25, 2014 at 11:01:54AM -0800, Raghu Gandham wrote:
> > > > > The standard IO regions are already reserved by the platform
> > > > > code on most MIPS devices(malta, cobalt, sni). The Commit
> > > > > 197a1e96c8be5b6005145af3a4c0e45e2d651444
> > > > > ("Input: i8042-io - fix up region handling on MIPS") introduced
> > > > > a bug on these MIPS platforms causing i8042 driver to fail when
> > > > > trying to reserve IO ports.
> > > > > Prior to the above mentioned commit request_region is skipped on
> > > > > MIPS but release_region is called.
> > > > >
> > > > > This patch reverts commit
> > > > > 197a1e96c8be5b6005145af3a4c0e45e2d651444
> > > > > and also avoids calling release_region for MIPS.
> > > >
> > > > The problem is that IO regions are reserved on _most_, but not
> > > > _all_ devices.
> > > > MIPS should figure out what they want to do with i8042 registers
> > > > and be consistent on all devices.
> > >
> > > Please examine the attached patch which went upstream in April of 2004.
> > > Since then it had been a convention not to call request_region
> > > routine in
> > > i8042 for MIPS. The attached patch had a glitch that it guarded
> > > request_region in i8042-io.h but skipped guarding release_region in
> > > i8042-io.h. I believe that the issue Aaro saw was due to this
> > > glitch. Below is the error quoted in Aaro's commit message.
> > >
> > > [ 2.112000] Trying to free nonexistent resource <0000000000000060-
> 000000000000006f>
> > >
> > > My patch reinstates the convention followed on MIPS devices along
> > > with fixing Aaro's issue.
> >
> > I assume that Aaro did test his patch and on his box request_region()
> > succeeds. That would indicate that various MIPS sub-arches still not
> > settled on the topic.
>
> request_region() succeeds on Loongson and OCTEON.
This would mean that before your patch in oct of 2012, Loongson and Octeon
were not reserving the IO space for i8042.
>
> On OCTEONs without PCI, request_region() will fail which is correct as there
> is no I/O space.
>
> I wasn't aware of that 2004 patch (it pre-dates GIT history of mainline Linux).
> Why the regions are already reserved by the platform code?
The only information I have is the comment before request_region in i8042-io.h that
touching data register on some platforms is flaky. If your patch was primarily aimed at
addressing the error message from release_region, the current patch I uploaded should
also take care of it too.
Thanks,
Raghu
