Re: [PATCH v4 2/2] kho: Introduce preserve/restore APIs for high-order pages

[Pratyush Yadav <pratyush@kernel.org>]

On Wed, Aug 12 2026, Pranjal Shrivastava wrote:

> On Wed, Aug 12, 2026 at 01:10:23PM +0200, Pratyush Yadav wrote:
>> On Mon, Aug 03 2026, Pranjal Shrivastava wrote:
>> 
>> > The current KHO page preservation APIs (e.g. kho_preserve_pages) assume
>> > that multi-page blocks are split into independent 4KB pages during
>> > restoration. This is incompatible with high-order non-compound pages,
>> > such as DMA buffers, which must be restored with tail pages having a
>> > zero reference count.
>> >
>> > Introduce explicit preserve and restore APIs for high-order pages,
>> > which preserve and restore a high-order page block as a single unit,
>> > applying a refcount of 1 to the head page while leaving tail pages at 0.
>> > Rename the existing internal helper to __kho_restore_page() and
>> > consolidate the common restoration code into it.
>> >
>> > Signed-off-by: Pranjal Shrivastava <praan@google.com>
>> 
>> The code here looks very convoluted TBH. I think it will be simpler to
>> make kho_restore_page() only return non-compound pages. That is, it
>> returns 0 or higher order non-compound page.
>> 
>> Then kho_restore_pages() can call kho_restore_page() and then do
>> split_page() on the page it got to turn it into 0-order pages.
>> kho_restore_folio() can call kho_restore_page() and then do
>> prep_compound_page() on the page it got.
>> 
>> And then you expose kho_restore_page() to be used by DMA APIs to get
>> non-compound high order pages directly.
>> 
>> How does that sound?
>> 
>
> I like the suggestion to use a single base primitive and mold it via
> split_page() and prep_compound_page().
>
> However, To implement this, we'll need __kho_restore_page(phys, &order)
> as the internal base primitive because callers (like kho_restore_folio)
> need to retrieve the order from the KHO metadata to pass into 
> prep_compound_page(), and a raw non-compound struct page doesn't store 
> its order.

Oh, right. I didn't think of that.

>
> By having __kho_restore_page() initialize the head_ref = 1 & all_tails
> =0 refcount pattern by default, all three public APIs can just wrap it
> and call the appropriate MM subsystem helpers. I'll spin v5 with this

Makes sense I think.

-- 
Regards,
Pratyush Yadav