Re: [PATCH] IDE: don't offer IDE_GENERIC on ia64

[Bjorn Helgaas]

On Tuesday 16 August 2005 3:38 am, Bartlomiej Zolnierkiewicz wrote:
> On 8/15/05, Bjorn Helgaas <bjorn.helgaas@xxxxxx> wrote:
> > On Friday 12 August 2005 2:40 am, Alan Cox wrote:
> Agreed but I have few comments:
> * is this change OK w.r.t. IA64_HP_SIM?

Kernels for the HP simulator (ski) use a SCSI model, not IDE,
so they should be OK.

The SGI simulator *does* use IDE, and this will break that.  But
my friends at SGI are looking at exposing the simulated IDE device
via PCI, which should fix it.

> * removing IDE_ARCH_OBSOLETE_INIT define has some implications,
>   * non-functional ide-cs driver (but there is no PCMCIA on IA64?)

PCMCIA might be of theoretical interest, but not important enough
by itself to keep IDE_ARCH_OBSOLETE_INIT, IMHO.  And it sounds like
ide-cs can be fixed fairly easily to not depend on it.

>   * ordering change for ide-pnp interfaces in case of no IDE devices
>     on default IDE PCI ports, (but there aren't any ide-pnp devices on IA64?)

I'm not aware of any ide-pnp devices on IA64, so I don't think this
is a problem for us.  The SGI simulator could use ide-pnp via
PNPACPI, but ordering shouldn't be a problem there.

But it is interesting that PNP IDE devices are discovered *after*
PCI devices.  For serial, we do PNP before PCI, on the theory that
PNP devices are usually built-in and shouldn't move if you just plug
in a new PCI card.

>   * non-functional HDIO_REGISTER_HWIF ioctl (ain't really working either)

HDIO_SCAN_HWIF - I don't know about this one.  How are we supposed to
follow the "new ports shouldn't define IDE_ARCH_OBSOLETE_INIT" injunction
if we lose all this functionality without it?  ia64 is about as close to
a new port as you get :-)

> This changes default default MAX_HWIFS value from
> 10 (6 in case of !CONFIG_PCI) to 4 which is not desirable.
> 
> IMO the best thing for now is to leave MAX_HWIFS alone.

IDE on ia64 is little-used, so I'm OK with leaving it alone, but
I do think it's wrong for an architecture with no real restriction
to specify MAX_HWIFS in ide.h.  Better to have the config option,
and make the default there larger if necessary.

> > I noticed that on a box with no devices on ide1, we probed
> > ide1 twice -- once via ide_setup_pci_device() and again via
> > ide_generic_init().  This isn't fatal, since the generic probe
> > uses the I/O ports setup by PCI IDE, but it's unnecessary
> > and a bit ugly.
> > 
> > I stuck in a "hwif->noprobe = 1" at the point where probe_hwif()
> > decides the interface isn't present, which prevents the second
> > probe by ide_generic_init().
> 
> Please separate this change out - it is a nice fix but it affects other
> archs/drivers and somebody needs to audit all IDE host drivers that
> they set hwif->noprobe = 0 correctly before probing in case of re-using
> hwif that was already owned/probed by other driver.

OK, I'll split this and look into it a bit more.

> Please make IDE_GENERIC depend on !IA_64.

Huh?  The first patch I posted did exactly that, nothing more.  But
Alan pointed out that I should instead clean up asm-ia64/ide.h, which
I think leads to a nicer solution.

Thanks for the detailed comments!
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/