LinuxKPI: Factor out vmf_insert_pfn_prot() routine

[wulf]

Summary

from GEM and TTM page fault handlers and move it in to base system. This
code is tightly integrated with LKPI mmap support to belong to drm-kmod.

As this routine requires associated vm_object to be locked, it got
additional _locked suffix.

[• hselasky]

Looks good. Was this code tested?

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

205

<tab> character after define?

[wulf]

In D32068#723767, @hselasky wrote:

Looks good. Was this code tested?

The code itself existed for a long time [1][2]

But results of refactoring were tested on GEM(i915) only yet.

[1] https://github.com/freebsd/drm-kmod/blob/5.5-stable/drivers/gpu/drm/i915/intel_freebsd.c#L167
[2] https://github.com/freebsd/drm-kmod/blob/5.5-stable/drivers/gpu/drm/ttm/ttm_bo_vm.c#L316

[markj]

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

190

I don't really understand the namespacing rules for the linuxkpi. Is lkpi_* for functions that are not compatible with Linux but nonetheless are useful in ported code?

206

It would be useful to have some comment explaining why this is here, IMHO. I presume it's to avoid races around vm_pfn_count?

[wulf]

Clarify _Static_assert reason

[wulf]

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

206

It would be useful to have some comment explaining why this is here, IMHO. I presume it's to avoid races around vm_pfn_count?

It is done by performance reasons. Description is updated.

[wulf]

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

190

IMO linux_* belongs to linuxolator. And new LKPI code should use lkpi_

[markj]

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

207

or "Consider use of the locked version"