[Bug 51210] winediag:nodrv_CreateWindow Unknown error (998)

[WineHQ Bugzilla]

https://bugs.winehq.org/show_bug.cgi?id=51210

Vasily Galkin <galkin-vv at ya.ru> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |galkin-vv at ya.ru

--- Comment #11 from Vasily Galkin <galkin-vv at ya.ru> ---
Created attachment 71120
  --> https://bugs.winehq.org/attachment.cgi?id=71120
Minimal crash repoducer

TLDR: if wrong vulkan ICD is loaded pvkGetRandROutputDisplayEXT may crash.

I ran into the same problem in a multiseat system with one seat on NVidia gpu
and the other seat on AMD gpu (both sessions X11).

Trying to run wine or even winedbg on the X11 DISPLAY on the amd gots this
error.

thread log with a part near the crash:

> 172488.042:0078:007c:trace:vulkan:X11DRV_vkCreateInstance create_info 0x6ace50, allocator (nil), instance 0x6ace40
> 172488.042:0078:007c:trace:heap:RtlAllocateHeap (0000000000020000,7000006a,00000030): returning 0000000000070BF0
> 172488.097:0078:007c:trace:heap:RtlFreeHeap (0000000000020000,70000062,0000000000070BF0): returning TRUE
> 172488.097:0078:007c:trace:vulkan:X11DRV_vkGetInstanceProcAddr 0x7e61ed10, "vkEnumeratePhysicalDevices"
> 172488.097:0078:007c:trace:vulkan:X11DRV_vkGetInstanceProcAddr 0x7e61ed10, "vkGetPhysicalDeviceProperties2KHR"
> 172488.097:0078:007c:trace:vulkan:X11DRV_vkGetInstanceProcAddr 0x7e61ed10, "vkGetRandROutputDisplayEXT"
> 172488.099:0078:007c:trace:heap:RtlAllocateHeap (0000000000020000,7000006a,00000018): returning 0000000000070C40
> 172488.115:0078:007c:trace:seh:dispatch_exception code=c0000005 flags=0 addr=0000000000000000 ip=0000000000000000 tid=007c
> 172488.115:0078:007c:trace:unwind:dispatch_exception  info[0]=0000000000000008
> 172488.115:0078:007c:trace:unwind:dispatch_exception  info[1]=0000000000000000
> 172488.115:0078:007c:warn:seh:dispatch_exception EXCEPTION_ACCESS_VIOLATION exception (code=c0000005) raised
> 172488.115:0078:007c:trace:seh:dispatch_exception  rax=00007f24aff77340 rbx=00000000006ace48 rcx=00000000006acd60 rdx=00000000006acd2c
> 172488.115:0078:007c:trace:seh:dispatch_exception  rsi=0000000000000055 rdi=000000007e5ebb40 rbp=000000007e8b4420 rsp=00000000006acd18
> 172488.115:0078:007c:trace:seh:dispatch_exception   r8=00000000006ace48  r9=0000000000000004 r10=00007ffdd5f46080 r11=0000000000000246
> 172488.115:0078:007c:trace:seh:dispatch_exception  r12=0000000000000055 r13=000000007e5ebb40 r14=00000000006ace48 r15=00007f24bb1a3870
> 172488.115:0078:007c:trace:unwind:dwarf_virtual_unwind function 7f24ae851ca0 base 0x7f24ae851c70 cie 0x7f24af0744f8 len 14 id 0 version 3 aug 'zR' code_align 1 data_align -8 retaddr %rip

I investigated it further and found the reason: for DISPLAY containing all
AMD-related
the NVidia vulkan ICD is loaded, and pvkGetRandROutputDisplayEXT calls into
libnvidia-glcore.so.470.86 that crashes for that AMD display.

Attached file is a minimal C source that reproduces the problem.

It runs fine with nvidia DISPLAY=:0 and segfaults with amd DISPLAY=:1 

>  % DISPLAY=:0  ./wine-vk-randr-crash
> 0x7fadf49c4870 0x7fadf49e3750 0x7fadf49e4210
> device_count 3 
> pvkGetRandROutputDisplayEXT 0x55ce42baf750
> pvkGetRandROutputDisplayEXT 0x55ce42baad20
> pvkGetRandROutputDisplayEXT 0x55ce42baf5b0
>  % DISPLAY=:1  ./wine-vk-randr-crash
> 0x7fb9ee250870 0x7fb9ee26f750 0x7fb9ee270210
> device_count 3 
> pvkGetRandROutputDisplayEXT 0x564206a0bd00
> zsh: segmentation fault 

Dircet specifying VK_ICD_FILENAMES helps
> % VK_ICD_FILENAMES=/usr/share/vulkan/icd.d/radeon_icd.x86_64.json DISPLAY=:1  > ./wine-vk-randr-crash 
> 0x7fc07c50c870 0x7fc07c52b750 0x7fc07c52c210
> device_count 1 
> pvkGetRandROutputDisplayEXT 0x556edbb1c320
> %

So this minimal representer has behaviour identical to wine.

running DISPLAY=:1  ./wine-vk-randr-crash under gdb gives the stacktrace into
libnvidia-glcore.so.470.86 ending with NULL instruction pointer:

> pvkGetRandROutputDisplayEXT 0x557c6aba4cf0
> 
> Thread 1 "wine-vk-randr-c" received signal SIGSEGV, Segmentation fault.
> 0x0000000000000000 in ?? ()
> (gdb) bt
> #0  0x0000000000000000 in ?? ()
> #1  0x00007f86799ddca0 in ?? () from /usr/lib/x86_64-linux-gnu/libnvidia-glcore.so.470.86
> #2  0x00007f86799d8afd in ?? () from /usr/lib/x86_64-linux-gnu/libnvidia-glcore.so.470.86
> #3  0x0000557c68a124d8 in main () at wine-vk-randr-crash.c:83

If anyone know how vkCreateInstance can be called in a way that returns ICD
that is compatible with specific X11 disaply, this maybe can be used to fix the
issue.

-- 
Do not reply to this email, post in Bugzilla using the
above URL to reply.
You are receiving this mail because:
You are watching all bug changes.