Re: [1/9] wined3d: Make the card selection for binary drivers depend only on gl_renderer like it already is for the mesa drivers and remove the gl_info parameter from the select_card callback. (try2)

[Alexandre Julliard]

Andrei Slăvoiu  writes:
> dlls/wined3d/directx.c | 568 +++++++++++++++++++++++--------------------------
>  1 file changed, 269 insertions(+), 299 deletions(-)

It doesn't work here:

../../../tools/runtest -q -P wine -T ../../.. -M d3d8.dll -p d3d8_test.exe.so device && touch device.ok
err:d3d:resource_init Out of adapter memory
wine: Unhandled page fault on read access to 0x69766574 at address 0x6862ab1c (thread 0009), starting debugger...
Unhandled exception: page fault on read access to 0x69766574 in 32-bit code (0x6862ab1c).
Register dump:
 CS:0023 SS:002b DS:002b ES:002b FS:0063 GS:006b
 EIP:6862ab1c ESP:0032fa50 EBP:0032fb08 EFLAGS:00010202(  R- --  I   - - - )
 EAX:68667f05 EBX:68678e50 ECX:00000000 EDX:69766564
 ESI:68667f05 EDI:0032fab8
Stack dump:
0x0032fa50:  68667f05 00000000 00000000 0032fab8
0x0032fa60:  00000000 00000000 00000280 000001e0
0x0032fa70:  00000000 00000000 00000000 00000000
0x0032fa80:  00000000 00000000 00060048 68667229
0x0032fa90:  00139670 68668b04 00139688 0032fab8
0x0032faa0:  00000000 00132818 0032fad8 00134650
Backtrace:
=>0 0x6862ab1c test_swapchain+0x28c() [/home/julliard/wine/wine/dlls/d3d8/tests/device.c:261] in d3d8_test (0x0032fb08)
  1 0x68644593 func_device+0x832() [/home/julliard/wine/wine/dlls/d3d8/tests/device.c:6461] in d3d8_test (0x0032fd48)
  2 0x6862a22f main+0x37e(argc=<is not available>, argv=<is not available>) [/home/julliard/wine/wine/dlls/d3d8/tests/../../../include/wine/test.h:584] in d3d8_test (0x0032fe18)
  3 0x68667e20 __wine_spec_exe_entry+0x7f(peb=<couldn't compute location>) [/home/julliard/wine/wine/dlls/winecrt0/exe_entry.c:36] in d3d8_test (0x0032fe58)
  4 0x7b85f65c call_process_entry+0xb() in kernel32 (0x0032fe78)
  5 0x7b8606a3 start_process+0x62(peb=<couldn't compute location>) [/home/julliard/wine/wine/dlls/kernel32/process.c:1097] in kernel32 (0x0032feb8)
  6 0x7bc7fac0 call_thread_func_wrapper+0xb() in ntdll (0x0032fed8)
  7 0x7bc82a6d call_thread_func+0x7c(entry=0x7b860640, arg=0x7ffdf000, frame=0x32ffc8) [/home/julliard/wine/wine/dlls/ntdll/signal_i386.c:2630] in ntdll (0x0032ffa8)
  8 0x7bc7fa9e call_thread_entry_point+0x11() in ntdll (0x0032ffc8)
  9 0x7bc53f0e start_process+0x1d(kernel_start=0x7b860640) [/home/julliard/wine/wine/dlls/ntdll/loader.c:2856] in ntdll (0x0032ffe8)
  10 0x680316dd wine_call_on_stack+0x1c() in libwine.so.1 (0x00000000)
  11 0x6803179b wine_switch_to_stack+0x2a(func=0x7bc53ef0, arg=0x7b860640, stack=0x330000) [/home/julliard/wine/wine/libs/wine/port.c:59] in libwine.so.1 (0xff81e868)
  12 0x7bc59a09 LdrInitializeThunk+0x238(kernel_start=<couldn't compute location>, unknown2=<couldn't compute location>, unknown3=<couldn't compute location>, unknown4=<couldn't compute location>) [/home/julliard/wine/wine/dlls/ntdll/loader.c:2910] in ntdll (0xff81e8a8)
  13 0x7b866ea3 __wine_kernel_init+0xa02() [/home/julliard/wine/wine/dlls/kernel32/process.c:1269] in kernel32 (0xff81f7a8)
  14 0x7bc5a903 __wine_process_init+0x192() [/home/julliard/wine/wine/dlls/ntdll/loader.c:3119] in ntdll (0xff81f838)
  15 0x6802ede8 wine_init+0x2c7(argc=0x3, argv=0xff81fd74, error="", error_size=0x400) [/home/julliard/wine/wine/libs/wine/loader.c:952] in libwine.so.1 (0xff81f898)
  16 0x7bf00d6c main+0x7b(argc=<is not available>, argv=<is not available>) [/home/julliard/wine/wine/loader/main.c:237] in <wine-loader> (0xff81fcd8)
  17 0x682518c5 __libc_start_main+0xf4() in libc.so.6 (0x00000000)
0x6862ab1c test_swapchain+0x28c [/home/julliard/wine/wine/dlls/d3d8/tests/device.c:261] in d3d8_test: call	*0x10(%edx)
261	        hr = IDirect3DSwapChain8_GetBackBuffer(swapchain3, 0, 0, &backbuffer);

[1/9] wined3d: Make the card selection for binary drivers depend only on gl_renderer like it already is for the mesa drivers and remove the gl_info parameter from the select_card callback. (try2)

[Andrei Slavoiu]

On Monday 14 July 2014 17:20:38 Alexandre Julliard wrote:
> Andrei Slăvoiu  writes:
>
> It doesn't work here:
>
> ../../../tools/runtest -q -P wine -T ../../.. -M d3d8.dll -p
> d3d8_test.exe.so device && touch device.ok err:d3d:resource_init Out of
> adapter memory
> wine: Unhandled page fault on read access to 0x69766574 at address
> 0x6862ab1c (thread 0009), starting debugger... Unhandled exception: page
> fault on read access to 0x69766574 in 32-bit code (0x6862ab1c). Register
> dump:

Does that test pass without my changes? I'm asking because for me it crashes 
with vanilla wine. I logged a bug for mesa about a month ago: 
https://bugs.freedesktop.org/show_bug.cgi?id=79575

[1/9] wined3d: Make the card selection for binary drivers depend only on gl_renderer like it already is for the mesa drivers and remove the gl_info parameter from the select_card callback. (try2)

[Alexandre Julliard]

Andrei Slavoiu  writes:
> On Monday 14 July 2014 17:20:38 Alexandre Julliard wrote:
> > Andrei Slăvoiu  writes:
> >
> > It doesn't work here:
> >
> > ../../../tools/runtest -q -P wine -T ../../.. -M d3d8.dll -p
> > d3d8_test.exe.so device && touch device.ok err:d3d:resource_init Out of
> > adapter memory
> > wine: Unhandled page fault on read access to 0x69766574 at address
> > 0x6862ab1c (thread 0009), starting debugger... Unhandled exception: page
> > fault on read access to 0x69766574 in 32-bit code (0x6862ab1c). Register
> > dump:
>
> Does that test pass without my changes? I'm asking because for me it crashes 
> with vanilla wine. I logged a bug for mesa about a month ago: 
> https://bugs.freedesktop.org/show_bug.cgi?id=79575

Yes, of course it passes on my box, otherwise it wouldn't have been
committed ;-)

[1/9] wined3d: Make the card selection for binary drivers depend only on gl_renderer like it already is for the mesa drivers and remove the gl_info parameter from the select_card callback. (try2)

[Henri Verbeet]

What happens is that the renderer string is "GeForce 210/PCIe/SSE2",
which now ends up getting matched against "GeForce", and the rest
mostly just follows from there.

[1/9] wined3d: Make the card selection for binary drivers depend only on gl_renderer like it already is for the mesa drivers and remove the gl_info parameter from the select_card callback. (try2)

[Andrei Slăvoiu]

În ziua de Lun 14 Iul 2014, la 18:38:09, Henri Verbeet a scris:
> What happens is that the renderer string is "GeForce 210/PCIe/SSE2",
> which now ends up getting matched against "GeForce", and the rest
> mostly just follows from there.

I see on Nvidia's site that the oldest cards supported by the legacy drivers 
on Linux are Geforce 5 FX series. So is it ok if I just remove all the D3D8 
and lower cards completely?

[1/9] wined3d: Make the card selection for binary drivers depend only on gl_renderer like it already is for the mesa drivers and remove the gl_info parameter from the select_card callback. (try2)

[Stefan Dösinger]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Am 2014-07-14 21:22, schrieb Andrei Slăvoiu:
> În ziua de Lun 14 Iul 2014, la 18:38:09, Henri Verbeet a scris:
> > What happens is that the renderer string is "GeForce
> > 210/PCIe/SSE2", which now ends up getting matched against
> > "GeForce", and the rest mostly just follows from there.
>
> I see on Nvidia's site that the oldest cards supported by the
> legacy drivers on Linux are Geforce 5 FX series. So is it ok if I
> just remove all the D3D8 and lower cards completely?

Aren't there different legacy drivers for even older cards? I think we
still have some (>= 1) users that use GeForce 4 MX cards(*), which are
actually just GeForce 2 GPUs.

(*): http://bugs.winehq.org/show_bug.cgi?id=34574
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v2
Comment: Using GnuPG with Thunderbird - http://www.enigmail.net/
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=tVi6
-----END PGP SIGNATURE-----

[1/9] wined3d: Make the card selection for binary drivers depend only on gl_renderer like it already is for the mesa drivers and remove the gl_info parameter from the select_card callback. (try2)

[Andrei Slăvoiu]

În ziua de Lun 14 Iul 2014, la 21:30:43, Stefan Dösinger a scris:
> Am 2014-07-14 21:22, schrieb Andrei Slăvoiu:
> > În ziua de Lun 14 Iul 2014, la 18:38:09, Henri Verbeet a scris:
> > > What happens is that the renderer string is "GeForce
> > > 210/PCIe/SSE2", which now ends up getting matched against
> > > "GeForce", and the rest mostly just follows from there.
> >
> > I see on Nvidia's site that the oldest cards supported by the
> > legacy drivers on Linux are Geforce 5 FX series. So is it ok if I
> > just remove all the D3D8 and lower cards completely?
>
> Aren't there different legacy drivers for even older cards? I think we
> still have some (>= 1) users that use GeForce 4 MX cards(*), which are
> actually just GeForce 2 GPUs.
>
> (*): http://bugs.winehq.org/show_bug.cgi?id=34574

So leave those supported by driver 96.43.23, that would be all D3D8 cards + 
Geforce4 MX. Previous to that was driver 71.86.15 from August 2011. I hope 
nobody uses that with a recent wine version any more.

[1/9] wined3d: Make the card selection for binary drivers depend only on gl_renderer like it already is for the mesa drivers and remove the gl_info parameter from the select_card callback. (try2)

[Henri Verbeet]

On 14 July 2014 21:22, Andrei Slăvoiu  wrote:
> I see on Nvidia's site that the oldest cards supported by the legacy drivers
> on Linux are Geforce 5 FX series. So is it ok if I just remove all the D3D8
> and lower cards completely?

Possibly, but it's a bit besides the point. You can't match against
something generic like "GeForce", and returning CARD_NVIDIA_RIVA_TNT
instead of PCI_DEVICE_NONE breaks the fallback mechanism.

[1/9] wined3d: Make the card selection for binary drivers depend only on gl_renderer like it already is for the mesa drivers and remove the gl_info parameter from the select_card callback. (try2)

[Andrei Slăvoiu]

În ziua de Lun 14 Iul 2014, la 22:58:40, Henri Verbeet a scris:
> On 14 July 2014 21:22, Andrei Slăvoiu  wrote:
> > I see on Nvidia's site that the oldest cards supported by the legacy
> > drivers on Linux are Geforce 5 FX series. So is it ok if I just remove
> > all the D3D8 and lower cards completely?
>
> Possibly, but it's a bit besides the point. You can't match against
> something generic like "GeForce", and returning CARD_NVIDIA_RIVA_TNT
> instead of PCI_DEVICE_NONE breaks the fallback mechanism.

Yes, I realize now this was completely broken, and patch nr. 3 was just a 
partial fix for the damage I done here. But how do I get the exact renderer 
string for those old cards? I guess from the comments that the original 
Geforce cards or either identified as "Geforce 256" or "Geforce DDR". But what 
about the Quadro cards? How do I make those less generic? So the easy fix is to 
just get rid of those old cards identified by such generic strings.

Now the question is how should I update the series, the current patch 3 no 
longer makes sense, and removing the old Nvidia cards should have been a 
change of its own before this one. Is it ok to renumber patches 1 and 2 as 2 
and 3 and make a new patch 1 or should I keep the original numbering, scrap 
patch 3 and remove the old cards as part of the gl_info removal?

[1/9] wined3d: Make the card selection for binary drivers depend only on gl_renderer like it already is for the mesa drivers and remove the gl_info parameter from the select_card callback. (try2)

[Henri Verbeet]

On 15 July 2014 01:37, Andrei Slăvoiu  wrote:
> Yes, I realize now this was completely broken, and patch nr. 3 was just a
> partial fix for the damage I done here. But how do I get the exact renderer
> string for those old cards? I guess from the comments that the original
> Geforce cards or either identified as "Geforce 256" or "Geforce DDR". But what
> about the Quadro cards? How do I make those less generic? So the easy fix is to
> just get rid of those old cards identified by such generic strings.

I'd probably let the cards that currently aren't matched against a
specific renderer string just hit the fallback code, since that's
effectively what was happening before. I probably wouldn't bother
removing existing entries, since it doesn't really gain us anything.
For what it's worth, you may also want to consider splitting this by
e.g. first getting rid of d3d_level_from_gl_info() in
select_card_nvidia_binary(), then in select_card_amd_binary(), etc.,
and try to avoid large series. I.e., first work on getting the first 5
or so patches comitted, then work on the next couple, etc.

[1/9] wined3d: Make the card selection for binary drivers depend only on gl_renderer like it already is for the mesa drivers and remove the gl_info parameter from the select_card callback. (try2)

[Andrei Slăvoiu]

În ziua de Mar 15 Iul 2014, la 12:52:25, Henri Verbeet a scris:
> I'd probably let the cards that currently aren't matched against a
> specific renderer string just hit the fallback code, since that's
> effectively what was happening before.

So remove the matchers that I added (GeForce3 and Geforce) and leave the 
existing ones. BTW, GeForce2 was a already there, but because of the OpenGL 
level check it was not matched for newer cards. Now it might cause problems if 
Nvidia decides to use GeForce2000 for a future card.
> For what it's worth, you may also want to consider splitting this by
> e.g. first getting rid of d3d_level_from_gl_info() in
> select_card_nvidia_binary(), then in select_card_amd_binary(), etc.,
> and try to avoid large series. I.e., first work on getting the first 5
> or so patches comitted, then work on the next couple, etc.

So drop this series completely and start over with smaller ones?