On 2024-07-09 13:17 UTC, Pavel Begunkov wrote:
> On 6/28/24 09:44, Chenliang Li wrote:
>> - if (folio) {
>> - bvec_set_page(&imu->bvec[0], pages[0], size, off);
>> - goto done;
>> - }
>> for (i = 0; i < nr_pages; i++) {
>> size_t vec_len;
>>
>> - vec_len = min_t(size_t, size, PAGE_SIZE - off);
>> + if (coalesced) {
>> + size_t seg_size = i ? data.folio_size :
>> + PAGE_SIZE * data.nr_pages_head;
>
> When you're compacting the page array, instead of taking a middle
> page for the first folio, you can set it to the first page in the
> folio and fix up the offset. Kind of:
>
> new_array[0] = compound_head(old_array[0]);
> off += folio_page_idx(folio, old_array[0]) << PAGE_SHIFT;
>
>
> With that change you should be able to treat it in a uniform way
> without branching.
>
> off = (unsigned long) iov->iov_base & ~folio_mask;
> vec_len = min_t(size_t, size, folio_size - off);
That's brilliant. Will change it this way.
Thanks,
Chenliang Li
