Re: [RFC PATCH 0/2] remove SWAP_MAP_SHMEM

[Nhat Pham <nphamcs@gmail.com>]

On Wed, Sep 25, 2024 at 8:59 PM Barry Song <baohua@kernel.org> wrote:
>
> On Thu, Sep 26, 2024 at 2:37 AM Nhat Pham <nphamcs@gmail.com> wrote:
>
> I’m not entirely clear on your point. If your proposal is to support the
> case where usage == 1 and nr > 1 only when we don’t require
> CONTINUED, and to issue a warning once we determine that
> CONTINUED is needed, then I’m completely on board with that
> approach.
>
> It seems that your intention is to simply relocate the existing warning
> to the scenario where CONTINUED is actually required, rather than
> maintaining a warning for the case where usage == 1 and nr > 1 at
> all times?

Ohhh yeah we definitely agreed on intentions, but I think I
misunderstood your request :) The code below was an attempt to satisfy
that request...

Please ignore it. I'll submit an actual patch taking into account our
discussions :) Hopefully I won't forget to actually test with thp
swaps this time...

>
> I wasn't actually suggesting a rollback as you posted:
>      err = __swap_duplicate(entry, 1, nr);
>      if (err == -ENOMEM) {
>          /* fallback to non-batched version */
>          for (i = 0; i < nr; i++) {
>              cur_entry = (swp_entry_t){entry.val + i};
>              if (swap_duplicate(cur_entry)) {
>                  /* rollback */
>                  while (--i >= 0) {
>                       cur_entry = (swp_entry_t){entry.val + i};
>                       swap_free(cur_entry);
>                  }
>