On Tue, Feb 08, 2005 at 10:03:47AM +0100, Andi Kleen wrote: > On Mon, Feb 07, 2005 at 04:02:44PM -0800, Matthew Dharm wrote: > > On Tue, Feb 08, 2005 at 12:19:19AM +0100, Andi Kleen wrote: > > > Remove useless AMD 8111 EHCI USB 2.0 errata check. The BIOS disable it > > > anyways > > > when needed, and when not then the hardware works. This cleans up > > > the EHCI driver a bit which shouldn't need to know about such Erratas. > > > > > > Signed-off-by: Andi Kleen <[EMAIL PROTECTED]> > > > > This is not a valid assertion. The IBM 970FX Evaluation Platform (aka > > Maple-D) has an 8111 but does _not_ have an x86 BIOS. So the device tree > > is built from a normal PCI-scan (unlike x86 BIOS which passes an explicit > > tree). That device tree therefore shows the device. > > That sounds like a firmware bug then. Perhaps it would be better > to fix the firmware than to add these workarounds to Linux code?
We've tried. AMD tells us that there is no way to physically disable the
controller. So, Linux will see it when it does a PCI scan. There's
nothing the firmware can do about this.
> > The patch was actually constructed to account for this case and all
> > derivative designs of the 970FX technology.
>
> Just apparently nobody tested it because it didn't work at all and
> caused a oops at boot. Therefore this code cannot be too useful,
> otherwise all these 970FX users would have noticed the failure.
When the patch was created, it worked just fine. These OOPSes only started
happening well after that patch was merged. That patch is quite old now
-- it was created in the 2.6.9 timeframe.
And, as I understand it, the problem isn't specific to that patch -- that
patch just forces us down an error-handling code-path by returing -EIO, and
you can get to that code paths in other ways. So the proper thing to do is
fix the usage of the driver core (which, as I seem to recall, a patch has
been available to do for weeks).
> BTW EHCI works, you just shouldn't use any USB 2.0 devices.
> But there are valid reasons in some cases to want to have EHCI
> even on a 8111, e.g. if you want to use the debug port. It's annoying
> to have to fight against such dumb code in the kernel then.
The official errata from AMD says "don't use EHCI". There's no indication
that it works under any circumstances.
Matt
--
Matthew Dharm Home: [EMAIL PROTECTED]
Maintainer, Linux USB Mass Storage Driver
C: Like the Furby?
DP: He gives me the creeps. Think the SPCA will take him?
-- Cobb and Dust Puppy
User Friendly, 1/2/1999
pgpWlhiMSfSWu.pgp
Description: PGP signature
