[Xorg-driver-geode] Geode LX DDC freeze in emulator when executing OUTW 0x20

[Jordan Crouse]

On 24/12/07 18:37 +0200, Martin-Éric Racine wrote:
> On 12/24/07, Jordan Crouse <jordan.crouse at amd.com> wrote:
> > On 24/12/07 14:31 +0200, Martin-Éric Racine wrote:
> > > Issues that have been reported by end-users include:
> > >
> > > 1. Failure to switch back to virtual console using CTRL+ALT+Fx.
> >
> > This is a new failure that has come along on the heals of Bart's
> > investigations.  I don't believe that anybody has had trouble with this
> > prior to the last  few weeks.  Anyway, this is a well known issue related
> > to VBE/int10, or rather, the lack of it.  Like I explained before, when
> > switching back to a virtual VGA text console, we need to issue an int10 to
> > put the screen back into the VGA mode it was in when we entered.  A
> > crippled or broken int10 will act exactly like you have seen.
> >
> > Since we know that the VBE and/or the int10 is broken on a small subset
> > of hardware and BIOS, then it follows that once Bart cured the hang
> > for the DDC, that you would still have int10 breakage elsewhere.
> >
> > I want to be clear that this is not universal breakage across all Geodes -
> > it seems to be localized to a small subset.
> 
> I want to be clear that it in fact does break across the board on
> General Software BIOS and on LinuxBIOS - on both the GX and LX - as
> verified by a number of people with a wide variety of Geode-based
> products to test against.

It breaks on LinuxBIOS with VGA BIOS attached, but not with vanilla
LinuxBIOS without VGA.  It also works on platforms with 
Insyde BIOS and Award BIOS.   So out of the four and a half BIOSes
that we might encounter, it only breaks on one and a half of them.

GSW is on a number of different platforms, that is true, but its not
like every Geode everywhere just started to break after 1.1 - that
is an incorrect characterization of the problem.

> 
> > > 2. Failure to auto-configure on X core 1.3 and newer, shown as either:
> > >   a) failed DDC falls back to 640x480 @ 4-bit.
> >
> > Again - this has been explained - no DDC only gives us a very small range
> > of available modes, by design in the X core.  Autoconfigure by its very
> > nature depends on DDC.  If you have no DDC then it doesn't help you very
> > much.   Complaining to the X core team isn't likely to change this - the
> > values they have chosen in lieu of DDC are sane given the real possibility
> > of damaging hardware.
> 
> That's assuming that we don't have DDC. The point is moot since the
> same AMD X.org breakage is noticed on hardware that has DDC - not just
> on the ThinCan.

I have seen some logs of these failures, and every single one shows
DDC failing.  If you can show me a log of DDC being successful, yet
the mode reverting to something very low, then please, send them on.

The logs are very descriptive, sometimes overly so.  But it is really
easy to piece together the series of events that lead to a failure -
it is clear when DDC fails, when it succeeds, what modes were selected,
and why.  Thats why I always plead for X logs, because in most cases,
we can figure out what happened in just a few seconds.

Jordan

-- 
Jordan Crouse
Systems Software Development Engineer 
Advanced Micro Devices, Inc.