Intercept the user32!SystemParametersInfoW API in the win32k user gdi lockdown mode

Intercept the user32!SystemParametersInfoW API in the win32k user gdi lockdown mode

This API fails for the font related values queried by webkit like SPI_GETFONTSMOOTHING and
SPI_GETFONTSMOOTHINGTYPE. This eventually causes skia to default to no antialiasing and no cleartype
which causes fonts to appear blurry and distorted.

Our solution for now is to fake the return values for SPI_GETFONTSMOOTHING to smooth and
SPI_GETFONTSMOOTHINGTYPE to cleartype.

I added a TODO in there to look into an IPC based approach in the near future if this causes problems.

BUG=365160

[jschuh, 2014-06-13 13:09:03]

Awesome job tracking this down, but I don't think the call hooking is necessary.
Given that there are so few in-renderer callers, how about I just land some CLs
to the change the default behavior on SystemParametersInfo failure? That way, in
win32k lockdown we'll just fall back to behaving as if font smoothing with
cleartype is enabled.

Here's the caller list, for reference:
https://code.google.com/p/chromium/codesearch#search/&q=%5CbSPI_GETFONTSMOOTH...

[ananta, 2014-06-13 14:57:11]

On 2014/06/13 13:09:03, Justin Schuh wrote:
> Awesome job tracking this down, but I don't think the call hooking is
necessary.
> Given that there are so few in-renderer callers, how about I just land some
CLs
> to the change the default behavior on SystemParametersInfo failure? That way,
in
> win32k lockdown we'll just fall back to behaving as if font smoothing with
> cleartype is enabled.
> 
> Here's the caller list, for reference:
>
https://code.google.com/p/chromium/codesearch#search/&q=%5CbSPI_GETFONTSMOOTH...

I could help with that. I am worried about dwrite using this api and failing in
the future. It does not today. But who knows?

[rvargas (doing something else), 2014-06-13 18:18:49]

On 2014/06/13 14:57:11, ananta wrote:
> On 2014/06/13 13:09:03, Justin Schuh wrote:
> > Awesome job tracking this down, but I don't think the call hooking is
> necessary.
> > Given that there are so few in-renderer callers, how about I just land some
> CLs
> > to the change the default behavior on SystemParametersInfo failure? That
way,
> in
> > win32k lockdown we'll just fall back to behaving as if font smoothing with
> > cleartype is enabled.
> > 
> > Here's the caller list, for reference:
> >
>
https://code.google.com/p/chromium/codesearch#search/&q=%5CbSPI_GETFONTSMOOTH...
> 
> I could help with that. I am worried about dwrite using this api and failing
in
> the future. It does not today. But who knows?

As a general rule, if the code that we want to intercept is under our control
(we call it), we should modify the call instead of adding an interception. We
should probably start there anyway and see if that is enough or not.

[jschuh, 2014-06-13 19:13:34]

I have a CL up here for the alternative change in blink:
https://codereview.chromium.org/331843004/

Broadly speaking, I'm not worried about MS introducing new win32k dependencies
because they're both actively moving and messaging in the opposite direction.

[jschuh, 2014-06-13 20:04:20]

Okay, the blink CL is reviewed and in the CQ. So, that should resolve this.