> On Sep 6, 2023, at 17:33, Kefeng Wang <wangkefeng.wang@xxxxxxxxxx> wrote: > > > > On 2023/9/6 11:25, Muchun Song wrote: >>> On Sep 6, 2023, at 11:13, Kefeng Wang <wangkefeng.wang@xxxxxxxxxx> wrote: >>> >>> >>> >>> On 2023/9/6 10:47, Matthew Wilcox wrote: >>>> On Tue, Sep 05, 2023 at 06:35:08PM +0800, Kefeng Wang wrote: >>>>> It is needed 4095 pages(1G) or 7 pages(2M) to be allocated once in >>>>> alloc_vmemmap_page_list(), so let's add a bulk allocator varietas >>>>> alloc_pages_bulk_list_node() and switch alloc_vmemmap_page_list() >>>>> to use it to accelerate page allocation. >>>> Argh, no, please don't do this. >>>> Iterating a linked list is _expensive_. It is about 10x quicker to >>>> iterate an array than a linked list. Adding the list_head option >>>> to __alloc_pages_bulk() was a colossal mistake. Don't perpetuate it. >>>> These pages are going into an array anyway. Don't put them on a list >>>> first. >>> >>> struct vmemmap_remap_walk - walk vmemmap page table >>> >>> * @vmemmap_pages: the list head of the vmemmap pages that can be freed >>> * or is mapped from. >>> >>> At present, the struct vmemmap_remap_walk use a list for vmemmap page table walk, so do you mean we need change vmemmap_pages from a list to a array firstly and then use array bulk api, even kill list bulk api ? >> It'll be a little complex for hugetlb_vmemmap. Should it be reasonable to >> directly use __alloc_pages_bulk in hugetlb_vmemmap itself? > > > We could use alloc_pages_bulk_array_node() here without introduce a new > alloc_pages_bulk_list_node(), only focus on accelerate page allocation > for now. > No. Using alloc_pages_bulk_array_node() will add more complexity (you need to allocate an array fist) for hugetlb_vmemap and this path that you optimized is only a control path and this optimization is at the millisecond level. So I don't think it is a great value to do this. Thanks.
