CONFIG_HSA_AMD unset in recent kernels

  • Done
  • quality assurance status badge
Details
One participant
  • John Kehayias
Owner
unassigned
Submitted by
John Kehayias
Severity
normal
J
J
John Kehayias wrote on 25 Apr 2022 17:59
(name . bug-guix@gnu.org)(address . bug-guix@gnu.org)
FP123ZvGi0I1vIwZ5HF2zzp0T2QdNEyN7uxlnYBRNc8Vc4qbIqc8Fd-SKn8eVktcSfIy8stjzBJCFFseqUqqdr0S-SWTOQ_BrNK8wL7AONU=@protonmail.com
Hi Guix,

In trying to update AMD's OpenCL stuff (ROCm) in Guix to the latest version, I found that CONFIG_HSA_AMD is not set in our kernel config which I think might be needed (for HSA, most visibly having a /dev/kfd if I'm understanding). I see it was set to 'm' in previous versions, like 4.19, but at some point has been unset.

Does anyone know why this was changed or have any reason not to set it in our kernel config?

I've tried building the kernel with this configuration change and it went fine, but haven't been able to properly test it just yet.

John
J
J
John Kehayias wrote on 26 Apr 2022 19:07
(name . 55111@debbugs.gnu.org)(address . 55111@debbugs.gnu.org)
mThHVnW01NGjs0VzOu83MxcTRl6DjJu8Dx0IIiRaoTg3ad5Ch7ArkA_QWxmSZCxJOIjVBvmueDqiYCGyu9uIkMhG6lv-haZtMIr3RMwIAfg=@protonmail.com
I can confirm setting CONFIG_HSA_AMD=y (setting it to 'm' didn't seem to work for me) made /dev/kfd available and the OpenCL ROCm work, including making OpenCL available to Darktable. Note that I tested this on a non linux-libre kernel (what I use) which may or may not be required for OpenCL on some hardware. This is on x86-64.

Are there any objections to enabling this option in our kernel config?
J
J
John Kehayias wrote on 16 Jun 2022 18:02
(name . 55111@debbugs.gnu.org)(address . 55111@debbugs.gnu.org)
Fkp-M4LxN835rhhBk1OrsfYYmLiAtxtRW4obX-A741guOURv5lUz6xdI9ziED4i4YsxhO68NjHFfM1fXdbxkyksV0RxOXnQfOCjjYPc2Nsg=@protonmail.com
Hello,

------- Original Message -------
On Tuesday, April 26th, 2022 at 1:07 PM, John Kehayias <john.kehayias@protonmail.com> wrote:
Toggle quote (5 lines)
>
> I can confirm setting CONFIG_HSA_AMD=y (setting it to 'm' didn't seem to work for me) made /dev/kfd available and the OpenCL ROCm work, including making OpenCL available to Darktable. Note that I tested this on a non linux-libre kernel (what I use) which may or may not be required for OpenCL on some hardware. This is on x86-64.
>
> Are there any objections to enabling this option in our kernel config?

Now that ROCm has been updated in #55948, I tested again if I needed this kernel config change. Indeed, without CONFIG_HSA_AMD I do not get OpenCL with ROCm (darktable-cltest fails, as does rocminfo). I again set this configuration option and it works.

Any obstacles to (re)enabling this configuration option?

Thanks!
John
J
J
John Kehayias wrote on 29 Jun 2022 21:54
(name . 55111-done@debbugs.gnu.org)(address . 55111-done@debbugs.gnu.org)
HtXxh9X8rm6pzUuQGZEAvwVQf5tYc-vV3qZKSs8gpjEZQW82Nrw6r4YInlcWnN6gTZZbjkYGiA5jhkPns5nfevH0EWHfVOxBVQePPqNIyqE=@protonmail.com
Addressed in #56200, pushed by Leo as 44f2c7c2ca5d0a275c61d5bfd9dde1205590504b
Closed
?
Your comment

This issue is archived.

To comment on this conversation send an email to 55111@debbugs.gnu.org

To respond to this issue using the mumi CLI, first switch to it
mumi current 55111
Then, you may apply the latest patchset in this issue (with sign off)
mumi am -- -s
Or, compose a reply to this issue
mumi compose
Or, send patches to this issue
mumi send-email *.patch