Re: [PATCH] x86/efi: Fix multiple GOP device support

[Ingo Molnar <mingo-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>]

* Ingo Molnar <mingo-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org> wrote:

> * Matt Fleming <matt-mF/unelCI9GS6iBeEJttW/XRex20P6io@public.gmane.org> wrote:
> 
> > From: Kővágó, Zoltán <dirty.ice.hu-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
> > 
> > When multiple GOP devices exists, but none of them implements ConOut,
> > the code should just choose the first GOP (according to the comments).
> > But currently fb_base will refer to the last GOP, while other parameters
> > to the first GOP, which will likely result in a garbled display.
> > 
> > I can reliably reproduce this bug using my ASRock Z87M Extreme4
> > motherboard with CSM and integrated GPU disabled, and two PCIe video
> > cards (NVidia GT640 and GTX980), booting from efi-stub (booting from
> > grub works fine).  On the primary display the asrock logo remains and on
> > the secondary screen is garbled up completely.
> > 
> > Signed-off-by: Kővágó, Zoltán <DirtY.iCE.hu@gmail.com>
> > Cc: Matthew Garrett <mjg59-1xO5oi07KQx4cg9Nei1l7Q@public.gmane.org>
> > Cc: <stable-u79uwXL29TY76Z2rM5mHXA@public.gmane.org>
> > Signed-off-by: Matt Fleming <matt.fleming-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>
> > ---
> >  arch/x86/boot/compressed/eboot.c | 8 ++++++--
> >  1 file changed, 6 insertions(+), 2 deletions(-)
> > 
> > diff --git a/arch/x86/boot/compressed/eboot.c b/arch/x86/boot/compressed/eboot.c
> > index ee1b6d346b98..db51c1f27446 100644
> > --- a/arch/x86/boot/compressed/eboot.c
> > +++ b/arch/x86/boot/compressed/eboot.c
> > @@ -667,6 +667,7 @@ setup_gop32(struct screen_info *si, efi_guid_t *proto,
> >  		bool conout_found = false;
> >  		void *dummy = NULL;
> >  		u32 h = handles[i];
> > +		u32 current_fb_base;
> 
> Sigh, fb_base is u64...
> 
> > @@ -770,6 +772,7 @@ setup_gop64(struct screen_info *si, efi_guid_t *proto,
> >  		bool conout_found = false;
> >  		void *dummy = NULL;
> >  		u64 h = handles[i];
> > +		u32 current_fb_base;
> 
> Ditto.
> 
> So I've applied it with that obvious bug fixed, but could you guys please double 
> check how on earth this patch could possibly have worked fine in testing, without 
> crashing 64-bit kernels?

Ah, I see, this is a subtle semantic conflict with pending v4.4 EFI changes in 
tip:core/efi, which changed fb_base from u32 to u64:

  ae2ee627dc87 ("efifb: Add support for 64-bit frame buffer addresses")

(Interestingly there was no textual conflict between this patch and that commit.)

So the fix patch is fine as-is for v4.3, but needs a conflict resolution for the 
pending v4.4 commit.

I've applied it that way.

Thanks,

	Ingo