[SOLVED] opencl in AMDGPU : clinfo get SIGSEGV , libreoffice not run

maziar · 2024-08-27 14:17:22

clinfo -v : Segmentation fault (core dumped)

amdgpu 0000:01:00.0: amdgpu: bo 00000000c4e66a7a va 0x0000000208-0x000000020f conflict with 0x0000000200-0x0000000400
amdgpu: Failed to map VA 0x208000 in vm. ret -22
amdgpu: Failed to map bo to gpuvm
amdgpu 0000:01:00.0: amdgpu: Failed to map peer:0000:01:00.0 mem_domain:2
amdgpu 0000:01:00.0: amdgpu: bo 000000009dab45f2 va 0x0000000208-0x000000020f conflict with 0x0000000200-0x0000000400
amdgpu: Failed to map VA 0x208000 in vm. ret -22
amdgpu: Failed to map bo to gpuvm
amdgpu 0000:01:00.0: amdgpu: Failed to map peer:0000:01:00.0 mem_domain:2
amdgpu 0000:01:00.0: amdgpu: bo 000000009dab45f2 va 0x0000000201-0x0000000201 conflict with 0x0000000200-0x0000000400
amdgpu: Failed to map VA 0x201000 in vm. ret -22
amdgpu: Failed to map bo to gpuvm
amdgpu 0000:01:00.0: amdgpu: Failed to map peer:0000:01:00.0 mem_domain:2
clinfo[47723]: segfault at 0 ip 0000707045811049 sp 00007ffe4ac45598 error 6 in libc.so.6[16d049,7070456c8000+171000] likely on CPU 63 (core 31, socket 0)
Code: 20 c5 fe 7f 07 c5 fe 7f 44 17 e0 c5 f8 77 c3 66 90 c5 fe 7f 47 c0 c5 fe 7f 47 e0 c5 f8 77 c3 66 90 48 3b 15 c9 a1 07 00 77 77 <c5> fe 7f 07 c5 fe 7f 47 20 48 01 d7 48 81 fa 80 00 00 00 76 d2 c5
Process 47723 (clinfo) of user 1000 terminated abnormally with signal 11/SEGV, processing...
Started Process Core Dump (PID 47730/UID 0).
Started Pass systemd-coredump journal entries to relevant user for potential DrKonqi handling.
Process 47723 (clinfo) of user 1000 dumped core.

Module libnxegl.so without build-id.
Stack trace of thread 47723:
#0  0x0000707045811049 n/a (libc.so.6 + 0x16d049)
#1  0x0000707035e3529f n/a (libhsa-runtime64.so.1 + 0x3529f)
#2  0x0000707035e355ad n/a (libhsa-runtime64.so.1 + 0x355ad)
#3  0x0000707035e39e1d n/a (libhsa-runtime64.so.1 + 0x39e1d)
#4  0x0000707035e78d6d n/a (libhsa-runtime64.so.1 + 0x78d6d)
#5  0x0000707035e78f5c n/a (libhsa-runtime64.so.1 + 0x78f5c)
#6  0x0000707035e4bf4f n/a (libhsa-runtime64.so.1 + 0x4bf4f)
#7  0x0000707036fa361c n/a (libamdocl64.so + 0x12861c)
#8  0x0000707036ef22cd n/a (libamdocl64.so + 0x772cd)
#9  0x000070704573d8fb n/a (libc.so.6 + 0x998fb)
#10 0x000070704573d979 pthread_once (libc.so.6 + 0x99979)
#11 0x0000707036ef2100 clIcdGetPlatformIDsKHR (libamdocl64.so + 0x77100)
#12 0x00007070458c64d4 n/a (libOpenCL.so.1 + 0x34d4)
#13 0x00007070458c6980 n/a (libOpenCL.so.1 + 0x3980)
#14 0x000070704573d8fb n/a (libc.so.6 + 0x998fb)
#15 0x000070704573d979 pthread_once (libc.so.6 + 0x99979)
#16 0x00007070458c6a45 clGetPlatformIDs (libOpenCL.so.1 + 0x3a45)
#17 0x000057c56519c1c8 n/a (clinfo + 0xd1c8)
#18 0x000057c565191b7d n/a (clinfo + 0x2b7d)
#19 0x00007070456c9e08 n/a (libc.so.6 + 0x25e08)
#20 0x00007070456c9ecc __libc_start_main (libc.so.6 + 0x25ecc)
#21 0x000057c565198275 n/a (clinfo + 0x9275)

Stack trace of thread 47727:
#0  0x0000707045734a19 n/a (libc.so.6 + 0x90a19)
#1  0x0000707045737479 pthread_cond_wait (libc.so.6 + 0x93479)
#2  0x000070703635679e n/a (pipe_radeonsi.so + 0x15679e)
#3  0x00007070363394ec n/a (pipe_radeonsi.so + 0x1394ec)
#4  0x00007070363566fd n/a (pipe_radeonsi.so + 0x1566fd)
#5  0x000070704573839d n/a (libc.so.6 + 0x9439d)
#6  0x00007070457bd49c n/a (libc.so.6 + 0x11949c)

Stack trace of thread 47725:
#0  0x0000707045734a19 n/a (libc.so.6 + 0x90a19)
#1  0x0000707045737479 pthread_cond_wait (libc.so.6 + 0x93479)
#2  0x000070703635679e n/a (pipe_radeonsi.so + 0x15679e)
#3  0x00007070363394ec n/a (pipe_radeonsi.so + 0x1394ec)
#4  0x00007070363566fd n/a (pipe_radeonsi.so + 0x1566fd)
#5  0x000070704573839d n/a (libc.so.6 + 0x9439d)
#6  0x00007070457bd49c n/a (libc.so.6 + 0x11949c)

Stack trace of thread 47724:
#0  0x0000707045734a19 n/a (libc.so.6 + 0x90a19)
#1  0x0000707045737479 pthread_cond_wait (libc.so.6 + 0x93479)
#2  0x000070703635679e n/a (pipe_radeonsi.so + 0x15679e)
#3  0x00007070363394ec n/a (pipe_radeonsi.so + 0x1394ec)
#4  0x00007070363566fd n/a (pipe_radeonsi.so + 0x1566fd)
#5  0x000070704573839d n/a (libc.so.6 + 0x9439d)
#6  0x00007070457bd49c n/a (libc.so.6 + 0x11949c)

Stack trace of thread 47726:
#0  0x0000707045734a19 n/a (libc.so.6 + 0x90a19)
#1  0x0000707045737479 pthread_cond_wait (libc.so.6 + 0x93479)
#2  0x000070703635679e n/a (pipe_radeonsi.so + 0x15679e)
#3  0x00007070363394ec n/a (pipe_radeonsi.so + 0x1394ec)
#4  0x00007070363566fd n/a (pipe_radeonsi.so + 0x1566fd)
#5  0x000070704573839d n/a (libc.so.6 + 0x9439d)
#6  0x00007070457bd49c n/a (libc.so.6 + 0x11949c)

Stack trace of thread 47728:
#0  0x0000707045734a19 n/a (libc.so.6 + 0x90a19)
#1  0x0000707045737479 pthread_cond_wait (libc.so.6 + 0x93479)
#2  0x000070703635679e n/a (pipe_radeonsi.so + 0x15679e)
#3  0x00007070363394ec n/a (pipe_radeonsi.so + 0x1394ec)
#4  0x00007070363566fd n/a (pipe_radeonsi.so + 0x1566fd)
#5  0x000070704573839d n/a (libc.so.6 + 0x9439d)
#6  0x00007070457bd49c n/a (libc.so.6 + 0x11949c)

Stack trace of thread 47729:
#0  0x0000707045734a19 n/a (libc.so.6 + 0x90a19)
#1  0x0000707045737479 pthread_cond_wait (libc.so.6 + 0x93479)
#2  0x000070703635679e n/a (pipe_radeonsi.so + 0x15679e)
#3  0x00007070363394ec n/a (pipe_radeonsi.so + 0x1394ec)
#4  0x00007070363566fd n/a (pipe_radeonsi.so + 0x1566fd)
#5  0x000070704573839d n/a (libc.so.6 + 0x9439d)
#6  0x00007070457bd49c n/a (libc.so.6 + 0x11949c)
ELF object binary architecture: AMD x86-64

move the : mv /etc/OpenCL/vendors/amdocl64.icd /etc/OpenCL/vendors/amdocl64.icd.bak
fix it & libreoffice run
gpu : Radeon Pro WX 7100
last kernel : 6.10.6-arch1-1 all update

Last edited by maziar (2024-09-03 15:26:34)

cryptearth · 2024-08-27 14:30:48

took me a bit to find it https://wiki.archlinux.org/title/AMDGPU_PRO

AMDGPU PRO OpenCL - used because Mesa OpenCL is not fully complete. Proprietary component only for Polaris GPUs. The onward GPUs use the open ROCm OpenCL.

do you have https://aur.archlinux.org/packages/open … amdgpu-pro installed?

Last edited by cryptearth (2024-08-27 14:32:01)

maziar · 2024-08-27 15:12:13

the opencl-legacy-amdgpu-pro is out of date and as comment only have advantage of 32 bit :
"A Package providing non-free legacy ("PAL") based OpenCL ICDs , Just like the opencl-amd (https://aur.archlinux.org/packages/opencl-amd) Package , but the with the advantage of a having a 32-bit version (https://aur.archlinux.org/packages/lib3 … amdgpu-pro) , and not including ROCm parts which allows the use of more up-to-date & "for now" polaris compatible standalone ROCm opencl stack (https://aur.archlinux.org/packages/rocm-opencl-runtime) and other parts of the standalone ROCm Stack."

not nice but openocl-amd also out of date !
"

Note for anyone who has a Polaris GPU (Radeon RX 5xx) debugging issues with this package; Packages that use OpenCL like clinfo or davinci-resolve-studio will need you to downgrade opencl-amd to 1:5.7.1-1 as well as amdgpu-pro-oglp to 23.10_1620044-1 to avoid coredumps & segfaults.

DVR would not open unless these 2 packages were downgraded (along with their dependencies). Had to figure it out the hard way after hours using valgrind and rebooting over and over. Hopefully someone else will not have to pull their hair out trying to resolve their issue."

cryptearth · 2024-08-27 18:25:29

well - the more reasonable option is to upgrade the hardware - as the software dropped support for it

Luciddream · 2024-08-28 06:24:04

maziar wrote:

not nice but openocl-amd also out of date !

opencl-amd is not out of date. It's the only ROCm package that has been constantly updated for the past 4 years.

maziar · 2024-08-28 11:43:26

cryptearth wrote:

well - the more reasonable option is to upgrade the hardware - as the software dropped support for it

Radeon pro wx 7100 not old gpu is realase 2018 & sales 2019 to now !

Lone_Wolf · 2024-08-28 12:32:51

Maziar, have you tried using opencl-rusticl-mesa ?

It does provide a lot more funtionality then opencl-mesa ever did and does work with darktable & Davinci Resolve .

seth · 2024-08-28 12:36:52

Also https://bbs.archlinux.org/viewtopic.php?id=290926 - the less invasive measure is to disable HW acceleration in LO or try to "export OCL_ICD_FILENAMES=/etc/OpenCL/vendors/mesa.icd" to LO.
There's also https://gitlab.freedesktop.org/drm/amd/-/issues/3101 and a bunch of duplicates.

Do you also get any kind of kernel / amdgpu module crashes?

cryptearth · 2024-08-29 09:21:02

maziar wrote:

cryptearth wrote:
well - the more reasonable option is to upgrade the hardware - as the software dropped support for it
Radeon pro wx 7100 not old gpu is realase 2018 & sales 2019 to now !

I agree - but the devs decided to drop support for pre-vega anyway - same as amd dropped support for rx500 series - because they decided these are end of life and it's thier right to drop support in future releases

TLDR: if you want to keep usig old hardware you have to keep using old software - don't expect devs to keep support for 5+ year old hardware in current releases

aside from that there's quite a jump between GCN (up to vega) and RDNA (starting from rx5000) - gcn is EOL and hence more and more devs drop support for it

seth · 2024-08-29 12:37:14

rocm-opencl-runtime will at least require you to "export ROC_ENABLE_PRE_VEGA=1" and I'd not be surprised if that doesn't work for a palaris 10 chip (I'd bank on polaris 22 support)
Why do you install that at all?

maziar · 2024-08-30 10:47:47

in next

Last edited by maziar (2024-08-30 11:05:03)

maziar · 2024-08-30 10:57:11

cryptearth wrote:

maziar wrote:
cryptearth wrote:
well - the more reasonable option is to upgrade the hardware - as the software dropped support for it
Radeon pro wx 7100 not old gpu is realase 2018 & sales 2019 to now !
I agree - but the devs decided to drop support for pre-vega anyway - same as amd dropped support for rx500 series - because they decided these are end of life and it's thier right to drop support in future releases
TLDR: if you want to keep usig old hardware you have to keep using old software - don't expect devs to keep support for 5+ year old hardware in current releases
aside from that there's quite a jump between GCN (up to vega) and RDNA (starting from rx5000) - gcn is EOL and hence more and more devs drop support for it

Yes, for the traditional line of GPUs.

In AMD and NVIDIA models, users can access the gaming model of the chip after six mounts. Furthermore, after the release of the base gaming chip, such as Polaris, users have the option to purchase a similar pro version within a period of 1 to 2 years in comparison with the same chip in the gaming model.

Then this really changes, especially the name PRO, and note that this happens on Linux.
More wonderfully, the Win Pro version works like a charm to now !!.

Lone_Wolf · 2024-08-30 11:03:13

Then this really changes, especially the name PRO, and note that this happens on Linux.
More wonderfully, the Win Pro version works like a charm to now !!.

That's probably because the PRO versions are intended to be used on supported distributions .
Typically only distros targeting enterprises are on the support list.

maziar · 2024-08-30 11:10:09

Lone_Wolf wrote:

Then this really changes, especially the name PRO, and note that this happens on Linux.
More wonderfully, the Win Pro version works like a charm to now !!.
That's probably because the PRO versions are intended to be used on supported distributions.
Typically, only distros targeting enterprises are on the support list.

I sure my vm in Ubuntu 22.04.4 has a similar problem. In next, similar last.
As solo user must challenge with AMD support., same as last get false support.

seth · 2024-08-30 14:09:14

=> https://deepl.com
Can you please explain the goal of this thread?

Your options are https://aur.archlinux.org/packages/open … amdgpu-pro or to try your luck w/ https://archlinux.org/packages/?name=ro … cl-runtime (which doesn't really support your GPU and in any even requires you to export ROC_ENABLE_PRE_VEGA=1) or try https://wiki.archlinux.org/title/GPGPU#Rusticl

Right now you installed rocm-opencl-runtime figured that it doesn't work and so broke the package by an unsustainable (will rendered void w/ the next update) file rename.

maziar · 2024-09-01 11:06:34

seth wrote:

=> https://deepl.com
Can you please explain the goal of this thread?
Your options are https://aur.archlinux.org/packages/open … amdgpu-pro or to try your luck w/ https://archlinux.org/packages/?name=ro … cl-runtime (which doesn't really support your GPU and in any even requires you to export ROC_ENABLE_PRE_VEGA=1) or try https://wiki.archlinux.org/title/GPGPU#Rusticl
Currently, you installed rocm-opencl-runtime figured that it doesn't work and so broke the package by an unsustainable (will render void w/ the next update) file rename.

1.the opencl-legacy-amdgpu-pro last update 2022 see pinned message : “for now” polaris compatible standalone ROCm opencl stack (https://aur.archlinux.org/packages/rocm-opencl-runtime) and other parts of the standalone ROCm Stack. "Not for now, not is now in 2024 !

2:rocm-opencl-runtime : flag out of date! : test same problem with new mesa after update, need after this downgrade mesa all time! Possible?

3.Rusticl : check talk : rocm-opencl-runtime does not support “GFX8 and later cards (Fiji, Polaris, Vega)” anymore. It's only supporting: CDNA3, CDNA2, CDNA, RDNA3, RDNA2.

I wonder if you think, the goal of the thread is unknown.

Last edited by maziar (2024-09-01 11:08:52)

seth · 2024-09-01 11:40:26

Because there's No question in the OP.
You record a Crash in clinfo and that you Work around IT by breaking rocm.

I'm Not Sure what your concerns ist reg. The amdgpu pro Implementation. The age ist irrelevant, your GPU wasn't produced Yesterday either. The only question ist whether it works (unlike rocm)

Your commenta about mesa are unintelligible, there's another thread reg. rocm and polaris, suggesting this ist a Regression in latest rocm releases and you can Stall the Update of that package as Long as the older Version still works until and in Case the Bug gets addressed upstream.

Edit: the wiki Talk suggests that Polaris AND vega are No longer supported by rocm

Last edited by seth (2024-09-01 11:42:43)

maziar · 2024-09-01 11:59:03

seth wrote:

Because there's No question in the OP.
You record a Crash in clinfo and that you Work around IT by breaking rocm.
I'm Not Sure what your concerns ist reg. The amdgpu pro Implementation. The age ist irrelevant, your GPU wasn't produced Yesterday either. The only question ist whether it works (unlike rocm)
Your commenta about mesa are unintelligible, there's another thread reg. rocm and polaris, suggesting this ist a Regression in latest rocm releases and you can Stall the Update of that package as Long as the older Version still works until and in Case the Bug gets addressed upstream.
Edit: the wiki Talk suggests that Polaris AND vega are No longer supported by rocm

And not only clinfo, applications need open-cl such as LibreOffice not run!

Lone_Wolf · 2024-09-01 12:14:15

The method with the greatest chance of success to solve that is :

remove all rocm & amdgpu pro stuff .

install opencl-rusticl-mesa
run RUSTICL_ENAB:E=radeonsi clinfo and post the output .

seth · 2024-09-01 14:08:45

And not only clinfo, applications need open-cl such as LibreOffice not run!

Yeah, no shit…

I assume you will benefit greatly from

seth wrote:

=> https://deepl.com

maziar · 2024-09-03 14:43:45

Lone_Wolf wrote:

The method with the greatest chance of success to solve that is :
remove all rocm & amdgpu pro stuff .
install opencl-rusticl-mesa
run RUSTICL_ENAB:E=radeonsi clinfo and post the output .

Are you sure about the parameter RUSTICL_ENAB:E=radeonsi?

seth · 2024-09-03 14:46:36

Obvious fat-finger - try "RUSTICL_ENABLE", also see https://bbs.archlinux.org/viewtopic.php … 6#p2193616

maziar · 2024-09-03 15:26:08

thx
RUSTICL_ENABLE=radeonsi with opencl-rusticl-mesa

My problem was solved with ten radeon-pro users.

Arch Linux

#1 2024-08-27 14:17:22

[SOLVED] opencl in AMDGPU : clinfo get SIGSEGV , libreoffice not run

#2 2024-08-27 14:30:48

Re: [SOLVED] opencl in AMDGPU : clinfo get SIGSEGV , libreoffice not run

#3 2024-08-27 15:12:13

Re: [SOLVED] opencl in AMDGPU : clinfo get SIGSEGV , libreoffice not run

#4 2024-08-27 18:25:29

Re: [SOLVED] opencl in AMDGPU : clinfo get SIGSEGV , libreoffice not run

#5 2024-08-28 06:24:04

Re: [SOLVED] opencl in AMDGPU : clinfo get SIGSEGV , libreoffice not run

#6 2024-08-28 11:43:26

Re: [SOLVED] opencl in AMDGPU : clinfo get SIGSEGV , libreoffice not run

#7 2024-08-28 12:32:51

Re: [SOLVED] opencl in AMDGPU : clinfo get SIGSEGV , libreoffice not run

#8 2024-08-28 12:36:52

Re: [SOLVED] opencl in AMDGPU : clinfo get SIGSEGV , libreoffice not run

#9 2024-08-29 09:21:02

Re: [SOLVED] opencl in AMDGPU : clinfo get SIGSEGV , libreoffice not run

#10 2024-08-29 12:37:14

Re: [SOLVED] opencl in AMDGPU : clinfo get SIGSEGV , libreoffice not run

#11 2024-08-30 10:47:47

Re: [SOLVED] opencl in AMDGPU : clinfo get SIGSEGV , libreoffice not run

#12 2024-08-30 10:57:11

Re: [SOLVED] opencl in AMDGPU : clinfo get SIGSEGV , libreoffice not run

#13 2024-08-30 11:03:13

Re: [SOLVED] opencl in AMDGPU : clinfo get SIGSEGV , libreoffice not run

#14 2024-08-30 11:10:09

Re: [SOLVED] opencl in AMDGPU : clinfo get SIGSEGV , libreoffice not run

#15 2024-08-30 14:09:14

Re: [SOLVED] opencl in AMDGPU : clinfo get SIGSEGV , libreoffice not run

#16 2024-09-01 11:06:34

Re: [SOLVED] opencl in AMDGPU : clinfo get SIGSEGV , libreoffice not run

#17 2024-09-01 11:40:26

Re: [SOLVED] opencl in AMDGPU : clinfo get SIGSEGV , libreoffice not run

#18 2024-09-01 11:59:03

Re: [SOLVED] opencl in AMDGPU : clinfo get SIGSEGV , libreoffice not run

#19 2024-09-01 12:14:15

Re: [SOLVED] opencl in AMDGPU : clinfo get SIGSEGV , libreoffice not run

#20 2024-09-01 14:08:45

Re: [SOLVED] opencl in AMDGPU : clinfo get SIGSEGV , libreoffice not run

#21 2024-09-03 14:43:45

Re: [SOLVED] opencl in AMDGPU : clinfo get SIGSEGV , libreoffice not run

#22 2024-09-03 14:46:36

Re: [SOLVED] opencl in AMDGPU : clinfo get SIGSEGV , libreoffice not run

#23 2024-09-03 15:26:08

Re: [SOLVED] opencl in AMDGPU : clinfo get SIGSEGV , libreoffice not run

Board footer