Re: [SeaBIOS] [PATCH] vgabios: Reorder video modes to work around a Windows bug

[Kevin O'Connor]

On Wed, Oct 19, 2016 at 11:32:41AM +0200, Ladi Prosek wrote:
> On Wed, Oct 19, 2016 at 10:06 AM, Ladi Prosek  wrote:
> > On Wed, Oct 19, 2016 at 1:29 AM, Kevin O'Connor  wrote:
> > > On Mon, Oct 17, 2016 at 05:53:04PM +0200, Ladi Prosek wrote:
> > > > Windows Server 2016 and Windows 10 RS1 come with a bug in its blue screen
> > > > of death rendering logic which prevents it from generating crash dumps.
> > > >
> > > > The bug does not manifest if Windows sees a suitable 32 bpp video mode
> > > > before a suitable 24 bpp video mode in the list of modes returned from
> > > > vgabios. This commit moves all 32 bpp modes to the front of the list to
> > > > make sure that this is always the case.
> > > > Signed-off-by: Ladi Prosek
> > > >
> > > > vgasrc/bochsvga.c | 39 ++++++++++++++++++++-------------------
> > > >  1 file changed, 20 insertions(+), 19 deletions(-)
> > > >
> > > > diff --git a/vgasrc/bochsvga.c b/vgasrc/bochsvga.c
> > > > index ec5d101..c5d1511 100644
> > > > --- a/vgasrc/bochsvga.c
> > > > +++ b/vgasrc/bochsvga.c
> > > > @@ -28,6 +28,25 @@ static struct bochsvga_mode
> > > >      u16 mode;
> > > >      struct vgamode_s info;
> > > >  } bochsvga_modes[] VAR16 = {
> > > > /* 32 bpp BOCHS modes */
> > > > { 0x140, { MM_DIRECT, 320,  200,  32, 8, 16, SEG_GRAPH } },
> > > > { 0x141, { MM_DIRECT, 640,  400,  32, 8, 16, SEG_GRAPH } },
> > > > { 0x142, { MM_DIRECT, 640,  480,  32, 8, 16, SEG_GRAPH } },
> > > > { 0x143, { MM_DIRECT, 800,  600,  32, 8, 16, SEG_GRAPH } },
> > > > { 0x144, { MM_DIRECT, 1024, 768,  32, 8, 16, SEG_GRAPH } },
> > > > { 0x145, { MM_DIRECT, 1280, 1024, 32, 8, 16, SEG_GRAPH } },
> > > > { 0x147, { MM_DIRECT, 1600, 1200, 32, 8, 16, SEG_GRAPH } },
> > > > { 0x14c, { MM_DIRECT, 1152, 864,  32, 8, 16, SEG_GRAPH } },
> > > > { 0x177, { MM_DIRECT, 1280, 768,  32, 8, 16, SEG_GRAPH } },
> > > > { 0x17a, { MM_DIRECT, 1280, 800,  32, 8, 16, SEG_GRAPH } },
> > > > { 0x17d, { MM_DIRECT, 1280, 960,  32, 8, 16, SEG_GRAPH } },
> > > > { 0x180, { MM_DIRECT, 1440, 900,  32, 8, 16, SEG_GRAPH } },
> > > > { 0x183, { MM_DIRECT, 1400, 1050, 32, 8, 16, SEG_GRAPH } },
> > > > { 0x186, { MM_DIRECT, 1680, 1050, 32, 8, 16, SEG_GRAPH } },
> > > > { 0x189, { MM_DIRECT, 1920, 1200, 32, 8, 16, SEG_GRAPH } },
> > > > { 0x18c, { MM_DIRECT, 2560, 1600, 32, 8, 16, SEG_GRAPH } },
> > > > { 0x18f, { MM_DIRECT, 1280, 720,  32, 8, 16, SEG_GRAPH } },
> > > > { 0x192, { MM_DIRECT, 1920, 1080, 32, 8, 16, SEG_GRAPH } },
> > > > /* standard modes */
> > > > { 0x100, { MM_PACKED, 640,  400,  8,  8, 16, SEG_GRAPH } },
> > > > { 0x101, { MM_PACKED, 640,  480,  8,  8, 16, SEG_GRAPH } },
> > > >
> > > > @@ -56,50 +75,32 @@ static struct bochsvga_mode
> > > >      { 0x11D, { MM_DIRECT, 1600, 1200, 15, 8, 16, SEG_GRAPH } },
> > > >      { 0x11E, { MM_DIRECT, 1600, 1200, 16, 8, 16, SEG_GRAPH } },
> > > >      { 0x11F, { MM_DIRECT, 1600, 1200, 24, 8, 16, SEG_GRAPH } },
> > > > /* BOCHS modes */
> > > > { 0x140, { MM_DIRECT, 320,  200,  32, 8, 16, SEG_GRAPH } },
> > > > { 0x141, { MM_DIRECT, 640,  400,  32, 8, 16, SEG_GRAPH } },
> > > > { 0x142, { MM_DIRECT, 640,  480,  32, 8, 16, SEG_GRAPH } },
> > > > { 0x143, { MM_DIRECT, 800,  600,  32, 8, 16, SEG_GRAPH } },
> > > > { 0x144, { MM_DIRECT, 1024, 768,  32, 8, 16, SEG_GRAPH } },
> > > > { 0x145, { MM_DIRECT, 1280, 1024, 32, 8, 16, SEG_GRAPH } },
> > > > /* 8, 15, 16, and 24 bpp BOCHS modes */
> > > > { 0x146, { MM_PACKED, 320,  200,  8,  8, 16, SEG_GRAPH } },
> > > > { 0x147, { MM_DIRECT, 1600, 1200, 32, 8, 16, SEG_GRAPH } },
> > >
> > > Thanks.  Instead of moving all of the 32bit modes to the top of the
> > > list, can the 32bit modes be kept next to the other modes with the
> > > same resolution, just above the 24bpp mode?
> >
> > Yes, that would work too. As long as 32 bpp come before 24 bpp for the
> > same resolution, it will be fine.
> > > Also, all of the mode numbers above 0x11b are arbitrary, so it would
> > > be preferable to renumber the mode numbers as lines are moved.
> >
> > Will do.
> > > We were about to make a SeaBIOS release - does this need to go into
> > > that release?
> >
> > It would be highly desired as the problematic Windows builds have
> > already RTM'ed.
> >
> > I'll send a v2 with the changes as suggested above.
>
> Actually, I spoke too soon. The problematic 24 bpp mode that Windows
> picks right now is 118 and that's within the standard range so I can't
> move 144 above it and renumber. What would be considered safer,
> keeping the same mode numbers or partially renumbering to keep them
> monotonically increasing except for the standard range?

I wouldn't renumber anything at 0x11B or below - those are defined in
the vbe3 spec.  Ordering and the numbering of all the other modes can
be changed.  Despite the comment in the source, modes 0x11C-0x11F do
not appear to be standard modes.

Thinking on this further, there's a good chance other systems (in
addition to Windows) choose the first valid mode found.  So,
reordering the 32bit modes above the 24bit modes could cause changes
in other systems as well.  So, I think it may be too risky to push
this into the pending SeaBIOS release.

-Kevin

[PATCH] vgabios: Reorder video modes to work around a Windows bug

[Ladi Prosek]

On Wed, Oct 19, 2016 at 2:23 PM, Kevin O'Connor  wrote:
> On Wed, Oct 19, 2016 at 11:32:41AM +0200, Ladi Prosek wrote:
> > On Wed, Oct 19, 2016 at 10:06 AM, Ladi Prosek  wrote:
> > > On Wed, Oct 19, 2016 at 1:29 AM, Kevin O'Connor  wrote:
> > > > On Mon, Oct 17, 2016 at 05:53:04PM +0200, Ladi Prosek wrote:
> > > > > Windows Server 2016 and Windows 10 RS1 come with a bug in its blue screen
> > > > > of death rendering logic which prevents it from generating crash dumps.
> > > > >
> > > > > The bug does not manifest if Windows sees a suitable 32 bpp video mode
> > > > > before a suitable 24 bpp video mode in the list of modes returned from
> > > > > vgabios. This commit moves all 32 bpp modes to the front of the list to
> > > > > make sure that this is always the case.
> > > > > Signed-off-by: Ladi Prosek
> > > > >
> > > > > vgasrc/bochsvga.c | 39 ++++++++++++++++++++-------------------
> > > > >  1 file changed, 20 insertions(+), 19 deletions(-)
> > > > >
> > > > > diff --git a/vgasrc/bochsvga.c b/vgasrc/bochsvga.c
> > > > > index ec5d101..c5d1511 100644
> > > > > --- a/vgasrc/bochsvga.c
> > > > > +++ b/vgasrc/bochsvga.c
> > > > > @@ -28,6 +28,25 @@ static struct bochsvga_mode
> > > > >      u16 mode;
> > > > >      struct vgamode_s info;
> > > > >  } bochsvga_modes[] VAR16 = {
> > > > > /* 32 bpp BOCHS modes */
> > > > > { 0x140, { MM_DIRECT, 320,  200,  32, 8, 16, SEG_GRAPH } },
> > > > > { 0x141, { MM_DIRECT, 640,  400,  32, 8, 16, SEG_GRAPH } },
> > > > > { 0x142, { MM_DIRECT, 640,  480,  32, 8, 16, SEG_GRAPH } },
> > > > > { 0x143, { MM_DIRECT, 800,  600,  32, 8, 16, SEG_GRAPH } },
> > > > > { 0x144, { MM_DIRECT, 1024, 768,  32, 8, 16, SEG_GRAPH } },
> > > > > { 0x145, { MM_DIRECT, 1280, 1024, 32, 8, 16, SEG_GRAPH } },
> > > > > { 0x147, { MM_DIRECT, 1600, 1200, 32, 8, 16, SEG_GRAPH } },
> > > > > { 0x14c, { MM_DIRECT, 1152, 864,  32, 8, 16, SEG_GRAPH } },
> > > > > { 0x177, { MM_DIRECT, 1280, 768,  32, 8, 16, SEG_GRAPH } },
> > > > > { 0x17a, { MM_DIRECT, 1280, 800,  32, 8, 16, SEG_GRAPH } },
> > > > > { 0x17d, { MM_DIRECT, 1280, 960,  32, 8, 16, SEG_GRAPH } },
> > > > > { 0x180, { MM_DIRECT, 1440, 900,  32, 8, 16, SEG_GRAPH } },
> > > > > { 0x183, { MM_DIRECT, 1400, 1050, 32, 8, 16, SEG_GRAPH } },
> > > > > { 0x186, { MM_DIRECT, 1680, 1050, 32, 8, 16, SEG_GRAPH } },
> > > > > { 0x189, { MM_DIRECT, 1920, 1200, 32, 8, 16, SEG_GRAPH } },
> > > > > { 0x18c, { MM_DIRECT, 2560, 1600, 32, 8, 16, SEG_GRAPH } },
> > > > > { 0x18f, { MM_DIRECT, 1280, 720,  32, 8, 16, SEG_GRAPH } },
> > > > > { 0x192, { MM_DIRECT, 1920, 1080, 32, 8, 16, SEG_GRAPH } },
> > > > > /* standard modes */
> > > > > { 0x100, { MM_PACKED, 640,  400,  8,  8, 16, SEG_GRAPH } },
> > > > > { 0x101, { MM_PACKED, 640,  480,  8,  8, 16, SEG_GRAPH } },
> > > > >
> > > > > @@ -56,50 +75,32 @@ static struct bochsvga_mode
> > > > >      { 0x11D, { MM_DIRECT, 1600, 1200, 15, 8, 16, SEG_GRAPH } },
> > > > >      { 0x11E, { MM_DIRECT, 1600, 1200, 16, 8, 16, SEG_GRAPH } },
> > > > >      { 0x11F, { MM_DIRECT, 1600, 1200, 24, 8, 16, SEG_GRAPH } },
> > > > > /* BOCHS modes */
> > > > > { 0x140, { MM_DIRECT, 320,  200,  32, 8, 16, SEG_GRAPH } },
> > > > > { 0x141, { MM_DIRECT, 640,  400,  32, 8, 16, SEG_GRAPH } },
> > > > > { 0x142, { MM_DIRECT, 640,  480,  32, 8, 16, SEG_GRAPH } },
> > > > > { 0x143, { MM_DIRECT, 800,  600,  32, 8, 16, SEG_GRAPH } },
> > > > > { 0x144, { MM_DIRECT, 1024, 768,  32, 8, 16, SEG_GRAPH } },
> > > > > { 0x145, { MM_DIRECT, 1280, 1024, 32, 8, 16, SEG_GRAPH } },
> > > > > /* 8, 15, 16, and 24 bpp BOCHS modes */
> > > > > { 0x146, { MM_PACKED, 320,  200,  8,  8, 16, SEG_GRAPH } },
> > > > > { 0x147, { MM_DIRECT, 1600, 1200, 32, 8, 16, SEG_GRAPH } },
> > > >
> > > > Thanks.  Instead of moving all of the 32bit modes to the top of the
> > > > list, can the 32bit modes be kept next to the other modes with the
> > > > same resolution, just above the 24bpp mode?
> > >
> > > Yes, that would work too. As long as 32 bpp come before 24 bpp for the
> > > same resolution, it will be fine.
> > > > Also, all of the mode numbers above 0x11b are arbitrary, so it would
> > > > be preferable to renumber the mode numbers as lines are moved.
> > >
> > > Will do.
> > > > We were about to make a SeaBIOS release - does this need to go into
> > > > that release?
> > >
> > > It would be highly desired as the problematic Windows builds have
> > > already RTM'ed.
> > >
> > > I'll send a v2 with the changes as suggested above.
> >
> > Actually, I spoke too soon. The problematic 24 bpp mode that Windows
> > picks right now is 118 and that's within the standard range so I can't
> > move 144 above it and renumber. What would be considered safer,
> > keeping the same mode numbers or partially renumbering to keep them
> > monotonically increasing except for the standard range?
>
> I wouldn't renumber anything at 0x11B or below - those are defined in
> the vbe3 spec.  Ordering and the numbering of all the other modes can
> be changed.  Despite the comment in the source, modes 0x11C-0x11F do
> not appear to be standard modes.

Got it. My point is that given how the standard modes are fixed, it's
not possible to renumber so all modes are in an increasing order. So I
wonder if renumbering is worth it at all. In fact, the issue would be
solved just by changing the order of the two 1024x768 modes mentioned
above: 118 and 144.
> Thinking on this further, there's a good chance other systems (in
> addition to Windows) choose the first valid mode found.  So,
> reordering the 32bit modes above the 24bit modes could cause changes
> in other systems as well.  So, I think it may be too risky to push
> this into the pending SeaBIOS release.

Understood. Would be ok to get it into the next release (assuming this
is how it gets some test exposure - apologies for not being familiar
with the SeaBIOS release process) or would you be opposed to the
change at all?

Another hypothetical alternative would be to try to find an API
invocation pattern reasonably unique to Windows that would trigger the
new behavior, to minimize the risk.
> -Kevin

[PATCH] vgabios: Reorder video modes to work around a Windows bug

[Gerd Hoffmann]

Hi,
> > Actually, I spoke too soon. The problematic 24 bpp mode that Windows
> > picks right now is 118 and that's within the standard range so I can't
> > move 144 above it and renumber. What would be considered safer,
> > keeping the same mode numbers or partially renumbering to keep them
> > monotonically increasing except for the standard range?
>
> I wouldn't renumber anything at 0x11B or below - those are defined in
> the vbe3 spec.  Ordering and the numbering of all the other modes can
> be changed.  Despite the comment in the source, modes 0x11C-0x11F do
> not appear to be standard modes.
>
> Thinking on this further, there's a good chance other systems (in
> addition to Windows) choose the first valid mode found.  So,
> reordering the 32bit modes above the 24bit modes could cause changes
> in other systems as well.

Indeed, there is a risk that this reordering uncovers hidden bugs.

I think at least for modern software the risk is pretty small though.
Real hardware apparently already lists 32bpp before 24bpp (or stopped
supporting 24bpp altogether), otherwise this windows bug wouldn't have
managed to hide long enough to make it to the RTM media ...

For oldish 90ies DOS games the risk might be higher.

It might also make use software use 32bpp instead of 24bpp (without
breaking stuff), which would be a useful change as qemu uses 32bpp
internally so it'll avoid a conversion step and give a small performance
benefit.
> So, I think it may be too risky to push
> this into the pending SeaBIOS release.

I expect in case we uncover bugs with this we will probably not notice
before we update the seabios version shipped with qemu ...

cheers,
  Gerd