Re: [patch V2 00/18] mm/highmem: Preemptible variant of kmap_atomic & friends

[Thomas Gleixner]

On Sat, Oct 31 2020 at 00:26, Thomas Gleixner wrote:

> On Fri, Oct 30 2020 at 15:46, Linus Torvalds wrote:
>> On Fri, Oct 30, 2020 at 3:26 PM Thomas Gleixner <tglx@xxxxxxxxxxxxx> wrote:
>> To me, your patch series has two big advantages:
>>
>>  - more common code
>>
>>  - kmap_local() becomes more of a no-op
>>
>> and the last thing we want is to expand on kmap.
>
> Happy to go with that.
>
> While trying to document the mess, I just stumbled over the abuse of
> kmap_atomic_prot() in
>
> drivers/gpu/drm/ttm/ttm_bo_util.c:      dst = kmap_atomic_prot(d, prot);
> drivers/gpu/drm/ttm/ttm_bo_util.c:      src = kmap_atomic_prot(s, prot);
> drivers/gpu/drm/vmwgfx/vmwgfx_blit.c:   kmap_atomic_prot(d->dst_pages[dst_page],
> drivers/gpu/drm/vmwgfx/vmwgfx_blit.c:   kmap_atomic_prot(d->src_pages[src_page],
>
> For !HIGHMEM kmap_atomic_prot() just ignores the 'prot' argument and
> returns the page address. 
>
> Moar patches to be written ... Sigh!

Or not. This is actually correct by some definition of correct. For
the non highmem case pgprot is set via the set_memory_*() functions and
this just handles the highmem case.

Comments are overrrated...