On 12 Jan 2026, at 18:53, Jason Gunthorpe wrote:

> On Mon, Jan 12, 2026 at 06:34:06PM -0500, Zi Yan wrote:
>> page[1].flags.f &= ~PAGE_FLAGS_SECOND. It clears folio->order.
>>
>> free_tail_page_prepare() clears ->mapping, which is TAIL_MAPPING, and
>> compound_head at the end.
>>
>> page->flags.f &= ~PAGE_FLAGS_CHECK_AT_PREP. It clears PG_head for compound
>> pages.
>>
>> These three parts undo prep_compound_page().
>
> Well, mm doesn't clear all things on alloc..
>
>> In current nouveau code, ->free_folios is used holding the freed folio.
>> In nouveau_dmem_page_alloc_locked(), the freed folio is passed to
>> zone_device_folio_init(). If the allocated folio order is different
>> from the freed folio order, I do not know how you are going to keep
>> track of the rest of the freed folio. Of course you can implement a
>> buddy allocator there.
>
> nouveau doesn't support high order folios.
>
> A simple linked list is not really a suitable data structure to ever
> support high order folios with.. If it were to use such a thing, and
> did want to take a high order folio off the list, and reduce its
> order, then it would have to put the remainder back on the list with a
> revised order value. That's all, nothing hard.
>
> Again if the driver needs to store information in the struct page to
> manage its free list mechanism (ie linked pointers, order, whatever)
> then it should be doing that directly.
>
> When it takes the memory range off the free list it should call
> zone_device_page_init() to make it ready to be used again. I think it
> is a poor argument to say that zone_device_page_init() should rely on
> values already in the struct page to work properly :\
>
> The usable space within the struct page, and what values must be fixed
> for correct system function, should exactly mirror what frozen pages
> require. After free it is effectively now a frozen page owned by the
> device driver.
>
> I haven't seen any documentation on that, but I suspect Matthew and
> David have some ideas..
>
> If there is a reason for order, flags and mapping to be something
> particular then it should flow from the definition of frozen pages,
> and be documented, IMHO.

Thank you for the explanation.

It seems that I do not have enough knowledge to comment on device private
pages. I will refrain myself from doing so from now on

Best Regards,
Yan, Zi

Reply via email to