On 10/20/21 18:12, Dan Williams wrote:
> On Wed, Oct 20, 2021 at 10:09 AM Joao Martins <joao.m.mart...@oracle.com>
> wrote:
>> On 10/19/21 20:21, Dan Williams wrote:
>>> On Tue, Oct 19, 2021 at 9:02 AM Jason Gunthorpe <j...@ziepe.ca> wrote:
>>>> On Tue, Oct 19, 2021 at 04:13:34PM +0100, Joao Martins wrote:
>>>>> On 10/19/21 00:06, Jason Gunthorpe wrote:
>>>>>> On Mon, Oct 18, 2021 at 12:37:30PM -0700, Dan Williams wrote:
>>>>> Whats the benefit between preventing longterm at start
>>>>> versus only after mounting the filesystem? Or is the intended future
>>>>> purpose
>>>>> to pass more context into an holder potential future callback e.g. nack
>>>>> longterm
>>>>> pins on a page basis?
>>>>
>>>> I understood Dan's remark that the device-dax path allows
>>>> FOLL_LONGTERM and the FSDAX path does not ?
>>>>
>>>> Which, IIRC, today is signaled basd on vma properties and in all cases
>>>> fast-gup is denied.
>>>
>>> Yeah, I forgot that 7af75561e171 eliminated any possibility of
>>> longterm-gup-fast for device-dax, let's not disturb that status quo.
>>>
>> I am slightly confused by this comment -- the status quo is what we are
>> questioning here -- And we talked about changing that in the past too
>> (thread below), that longterm-gup-fast was an oversight that that commit
>> was only applicable to fsdax. [Maybe this is just my english confusion]
>
> No, you have it correct. However that "regression" has gone unnoticed,
> so unless there is data showing that gup-fast on device-dax is
> critical for longterm page pinning workflows I'm ok for longterm to
> continue to trigger gup-slow.
>
To be fair, it's not surprising that nobody noticed -- my general intent
was just to special-case less for device-dax. Not many places use
pin_user_pages_fast(FOLL_LONGTERM). This is only exposed on those
few cases that do try to use it (e.g. RDMA/IB, vDPA), and specifically
when the page fault occurs (that requires fallback-ing to gup-slow) at a
higher level than the amount you're pinning e.g. pinning in 2M extents on a
device-dax of 1G pagesize. Pin size is limited to a 2M extent at a time by the
users I mentioned above -- regardless of the total size of the extent you will
be pinning (i.e. 512 struct pages pointers fit one page). But even with all
this,
this [FOLL_LONGTERM on pin-fast] would still go unnoticed because gup-fast
on devmap is just as slow as gup :)
