L_PTE_MT_BUFFERABLE / device ordered memory

[Prof. Michael Taylor]

Bumping this. Thanks!

M


On Tue, Jan 3, 2023 at 10:49 AM Prof. Michael Taylor
<prof.taylor at gmail.com> wrote:
>
> Hi Russell,
>
> Thanks for your reply! This is very helpful.
>
> So it sounds like what I would like to do is to leave most of the
> operation of the kernel the same, but have a select region of address
> space marked as L_PTE_MT_DEV_SHARED instead of L_PTE_MT_DEV_UNCACHED.
> Would that be the correct one to use for device ordered memory?
>
> And assuming yes, would it be suffice to modify phys_mem_access_prot
> so that it checks (pfn << PAGE_SIZE) to see if it is in the target
> range of physical memory, and then return pgprot_device() rather than
> pgprot_noncached() for that region?
>
> Is there a better way for me to do this than this brute force way (I
> noticed you alluded to this in this post:
> https://www.spinics.net/lists/arm-kernel/msg75913.html)?
>
>
> M