Re: Struct page proposal

From: Kent Overstreet
Date: Mon Sep 27 2021 - 16:14:13 EST

On Mon, Sep 27, 2021 at 09:07:21PM +0200, Vlastimil Babka wrote:
> On 9/27/2021 8:09 PM, Kent Overstreet wrote:
> > On Mon, Sep 27, 2021 at 07:05:26PM +0100, Matthew Wilcox wrote:
> >> On Mon, Sep 27, 2021 at 07:48:15PM +0200, Vlastimil Babka wrote:
> >>> Won't be easy to cram all that into two unsigned long's, or even a single
> >>> one. We should avoid storing anything in the free page itself. Allocating
> >>> some external structures to track free pages is going to have funny
> >>> bootstrap problems. Probably a major redesign would be needed...
> >>
> >> Wait, why do we want to avoid using the memory that we're allocating?
> >
> > The issue is where to stick the state for free pages. If that doesn't fit in two
> > ulongs, then we'd need a separate allocation, which means slab needs to be up
> > and running before free pages are initialized.
> So that's what I meant by the funny bootstrap problems - slab allocates pages
> from the buddy allocator. And well, not just bootstrap, imagine free memory
> becomes low, we need to reclaim pages, and in order to turn full pages to free
> buddy pages we need to allocate these slab structures, and the slab is full too
> and needs to allocate more backing pages...
> By "major redesign" I meant e.g. something along - bitmaps of free pages per
> each order? (instead of the free lists) Hm but I guess no, the worst case times
> searching for free pages would just suck...

We can have arrays of pointers to free pages - then within struct page we'd keep
the index of that page in the freelist array. To take a page off the middle of
the freelist we'd just swap it with the one at the end. And instead of using a
literal array, we'd want to use a simple radix tree.