Re: [PATCH] mm/gup: restore the ability to pin more than 2GB at a time

[John Hubbard <jhubbard@nvidia.com>]

On 10/30/24 10:42 AM, Vlastimil Babka wrote:
> On 10/30/24 18:29, John Hubbard wrote:
>> On 10/30/24 4:03 AM, Vlastimil Babka wrote:
>>> On 10/30/24 05:39, John Hubbard wrote:
>>>> On 10/29/24 9:33 PM, Christoph Hellwig wrote:
>>>>> On Tue, Oct 29, 2024 at 09:30:41PM -0700, John Hubbard wrote:
>> ...
>>> It might be a regression even if you don't try to pin over 2GB. high-order
>>> (>costly order) allocations can fail and/or cause disruptive
>>> reclaim/compaction cycles even below MAX_PAGE_ORDER and it's better to use
>>> kvmalloc if physical contiguity is not needed, it will attempt the physical
>>> kmalloc() allocation with __GFP_NORETRY (little disruption) and fallback to
>>> vmalloc() quickly.
>>>
>>> Of course if there's a way to avoid the allocation completely, even beter.
>>
>> Why not both? I'm going to ask our driver team to batch the pinning calls,
>> as recommended nearby, just to be sure that we are following best
>> practices.
>>
>> But it also seems good to use kvmalloc() here, and avoid any other
>> regressions. That's also a best practice.
> 
> By "avoid the allocation completely" I meant David's proof of concept
> elsewhere in this thread, that seems to replace that kmalloc_array() with no
> allocation :)
> 

aha, OK let me look into that then.

thanks,
-- 
John Hubbard