1156567 - Blacklisting code handles intel version numbers incorrectly

Status: NEW

(Core :: Graphics, defect, P3)

Description

[Matt Woodrow (:mattwoodrow)]

http://www.intel.com/support/graphics/sb/CS-020667.htm

According to the above document, the first two values in the intel (and NVIDIA I believe) driver numbers are related to the windows and DirectX versions, and are entirely orthogonal to the actual driver build number.

We currently treat the entire driver version string as a single number, so blacklisting a version on a newer operating system will blacklist *all* driver versions on older OS's.

In particular, we have a blacklist entry for < 8,17,11,8265 (Driver 182.65 on windows 7), that will blacklist all drivers versions for vista and below.

I think we can fix this fairly easily by blacklisting using the build number instead of the full string, but this will likely unblacklist a decent number of machines and we risk some fallout from doing that.

Comment 1

[Jeff Muizelaar [:jrmuizel]]

Isn't blacklisting usually OS specific?

Comment 2

[Matt Woodrow (:mattwoodrow)]

(In reply to Jeff Muizelaar [:jrmuizel] from comment #1)
> Isn't blacklisting usually OS specific?

Usually, but not always.

A few concrete issues:

We blacklist NVIDIA 8.17.11.8265 for Windows 7. Anthony has a Windows 7 machine running 8.15.11.8593 which gets blacklisted due to this rule even though 185.93 > 182.65. Amusingly about:support also recommends he 'upgrade' to 182.65.

[1] is attempting to blacklist a Windows 7 specific driver on Vista, so won't be doing anything useful.

[1] http://mxr.mozilla.org/mozilla-central/source/widget/windows/GfxInfo.cpp#883

Comment 3

[Anthony Jones (:ajones, :kentuckyfriedtakahe, :k17e)]

There are two parts to the driver version. One is the OS binding and the other is the actual driver itself. In that specific case the issue is that _.15._._ refers to D3D9 and _.17._._ refers to D3D11. This driver is known as version 185 on Linux.

This rule blocks all HD3000 drivers on Windows XP and Vista for this hardware because they have 6._._._ and 7._._._ version numbers.

  APPEND_TO_DRIVER_BLOCKLIST( DRIVER_OS_ALL,
        (nsAString&) GfxDriverInfo::GetDeviceVendor(VendorIntel), (GfxDeviceFamily*) GfxDriverInfo::GetDeviceFamily(IntelHD3000),
      GfxDriverInfo::allFeatures, nsIGfxInfo::FEATURE_BLOCKED_DRIVER_VERSION,
      DRIVER_LESS_THAN_OR_EQUAL, V(8,15,10,2321), "8.15.10.2342" );

Comment 4

[Jeff Muizelaar [:jrmuizel]]

(In reply to Anthony Jones (:kentuckyfriedtakahe, :k17e) from comment #3)
> There are two parts to the driver version. One is the OS binding and the
> other is the actual driver itself. In that specific case the issue is that
> _.15._._ refers to D3D9 and _.17._._ refers to D3D11. This driver is known
> as version 185 on Linux.
> 
> This rule blocks all HD3000 drivers on Windows XP and Vista for this
> hardware because they have 6._._._ and 7._._._ version numbers.
> 
>   APPEND_TO_DRIVER_BLOCKLIST( DRIVER_OS_ALL,
>         (nsAString&) GfxDriverInfo::GetDeviceVendor(VendorIntel),
> (GfxDeviceFamily*) GfxDriverInfo::GetDeviceFamily(IntelHD3000),
>       GfxDriverInfo::allFeatures, nsIGfxInfo::FEATURE_BLOCKED_DRIVER_VERSION,
>       DRIVER_LESS_THAN_OR_EQUAL, V(8,15,10,2321), "8.15.10.2342" );

This rule should probably be fixed at a minimum.

Comment 5

[Anthony Jones (:ajones, :kentuckyfriedtakahe, :k17e)]

DRIVER_OS_ALL is incompatible with DRIVER_LESS_THAN_OR_EQUAL unless we ignore the first two fields in Intel/nVidia version numbers.

Comment 6

[Jeff Muizelaar [:jrmuizel]]

(In reply to Anthony Jones (:kentuckyfriedtakahe, :k17e) from comment #5)
> DRIVER_OS_ALL is incompatible with DRIVER_LESS_THAN_OR_EQUAL unless we
> ignore the first two fields in Intel/nVidia version numbers.

I think that's fine or at least ok.

Comment 7

[Anthony Jones (:ajones, :kentuckyfriedtakahe, :k17e)]

Can we remove the rules and associated macros that exhibit this problem? Should we rely on the runtime testing instead for these Intel cases?