Re: [RFC PATCH v2] mm/zswap: store <PAGE_SIZE compression failed page as-is

[David Hildenbrand <david@redhat.com>]

On 07.08.25 01:48, Shakeel Butt wrote:
> On Wed, Aug 06, 2025 at 10:14:39PM +0200, David Hildenbrand wrote:
>>
>> But yeah, whether these "as is" pages are movable or not is a good question
>> as well -- in particular when zsmalloc supports page migration and the "as
>> is" pages would not.
> 
> By "as is" page, do you mean the page which the reclaim code is trying
> to reclaim or the page within zsmalloc on which the content of original
> pages are copied as is?

I mean whatever the "dst" is here.

+	if (zswap_save_as_is(comp_ret, dlen, page)) {
+		comp_ret = 0;
+		dlen = PAGE_SIZE;
+		memcpy_from_page(dst, page, 0, dlen);

IIUC SJ correctly, in case of zsmalloc "dst" is just the same page that 
would have stored encrypted data.

If that is the case, nothing should change, really.

Thanks for clarifying, all!

> Most probably you meant the page which the reclaim
> code is trying to reclaim. This page is on its way to get freed after
> [z]swapout is completed and this patch is not changing any behavior for
> that path.

Yeah, that's the "page" in the hunk above I guess.

-- 
Cheers,

David / dhildenb