Implement basic support for allocating memory from a specific numa node in the LinuxKPI.

[• hselasky]

Summary

Implement basic support for allocating memory from a specific numa node in the LinuxKPI.

MFC after: 1 week
Sponsored by: Mellanox Technologies // NVIDIA Networking

[kib]

sys/compat/linuxkpi/common/src/linux_domain.c

47

Are you sure that FIXED policy is right there? What is the Linux' semantic?

From my experience, it almost never is. You typically want a graceful fallback to to other domains if requested domain is full. AFAIR in our terminology this means _PREF().

sys/kern/kern_malloc.c

808 ↗

(On Diff #85195)

This line looks too long.

kern_malloc.c/malloc.h bits should be committed separately.

[• hselasky]

Strip the malloc.h bits from this commit and push them separately.

[• hselasky]

Update patch as per suggestions from kib@

[kib]

Looks fine to me. I did not looked at the Linux KPI to see if this is a proper implementation.

I suggest to wait some time for Mark' comments.

[markj]

Looks ok to me.

sys/compat/linuxkpi/common/include/linux/device.h

557–559

Why is it variadic?

[• hselasky]

sys/compat/linuxkpi/common/include/linux/device.h

557–559

Just to get a better compile error, if the argument passed is bad.
Do you think I should change it?

[markj]

sys/compat/linuxkpi/common/include/linux/device.h

557–559

I think there should at least be a comment explaining it, otherwise it looks strange. And, why is it especially likely to get compile errors with this function?

[• hselasky]

sys/compat/linuxkpi/common/include/linux/device.h

557–559

OK, I'll fix.

[• hselasky]

Fix comment by @markj .

[• hselasky]

Added build-fix for DRM KMOD.