Re: RFC: Compact Flash True IDE Mode Driver

[Bartlomiej Zolnierkiewicz <bzolnier@gmail.com>]

On 2/13/06, Kumar Gala <galak@kernel.crashing.org> wrote:
>
> On Feb 13, 2006, at 10:53 AM, Bartlomiej Zolnierkiewicz wrote:
>
> > On 2/13/06, Kumar Gala <galak@kernel.crashing.org> wrote:
> >
> >>>> +static void cfide_outsl(unsigned long port, void *addr, u32 count)
> >>>> +{
> >>>> +       panic("outsl unsupported");
> >>>> +}
> >>>
> >>> This will panic as soon as somebody tries to enable 32-bit I/O
> >>> using hdparm.  Please add ide_hwif_t.no_io_32bit flag and teach
> >>> ide-disk.c:ide_disk_setup() about it (separate patch).
> >>
> >> I'm not sure I follow this, can you expand.
> >
> > Do "hdparm -c 2 /dev/hdx" first and then read/write to the device
> > and you should see it. :)
> >
> > We need to make "hdparm -c 2" (and "hdparm -c 3") unsupported
> > (see how "io_32bit" setting is handled in ide_add_generic_settings()
> > and how it can be read-only or read-write setting depending on the
> > value of drive->no_io_32bit).
> >
> > To do this we need to set drive->no_io_32bit to 1 (see how
> > ide_disk_setup() handles it).  Unfortunately 32-bit I/O capability
> > is based on capabilities of both host and device so we have to
> > add new flag hwif->no_io_32bit to indicate that host doesn't
> > support 32-bit I/O.
>
> This all make sense, should I check for hwif->no_io_32bit in
> idedisk_setup() and set drive->no_io_32bit to 1 if hwif->no_io_32bit
> is 1 or do this the test in ide_add_generic_settings()?

Good question.  idedisk_setup() seems more logical but in the current
model "io_32bit" setting is still accessible without ide-disk driver through
/proc/ide/ interface so...

Moreover the current drive->no_io_32bit code in ide-disk is wrong:
* it shouldn't be overriding drive->no_io_32bit flag if it is 1
* doing this in ide-disk may be too late w.r.t. ide_add_generic_settings()

Therefore your previous suggestion is the right one - the best place
to deal with ->no_io_32bit is ide-probe.c - doing this for all drives at
the end of probe_hwif() should fix all above issues.

Thanks,
Bartlomiej