lkml.org 
[lkml]   [2021]   [Feb]   [26]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH v3 2/2] mm: Make alloc_contig_range handle in-use hugetlb pages
On Fri, Feb 26, 2021 at 09:46:57AM +0100, Michal Hocko wrote:
> On Mon 22-02-21 14:51:37, Oscar Salvador wrote:
> [...]
> > @@ -2394,9 +2397,19 @@ bool isolate_or_dissolve_huge_page(struct page *page)
> > */
> > if (hstate_is_gigantic(h))
> > return ret;
> > -
> > - if (!page_count(head) && alloc_and_dissolve_huge_page(h, head))
> > +retry:
> > + if (page_count(head) && isolate_huge_page(head, list)) {
> > ret = true;
> > + } else if (!page_count(head)) {
>
> This is rather head spinning. Do we need to test page_count in the else
> branch? Do you want to optimize for a case where the page cannot be
> isolated because of page_huge_active?

Well, I wanted to explictly call out both cases.
We either 1) have an in-use page and we try to issolate it or 2) we have a free
page (count == 0).

If the page could not be dissolved due to page_huge_active, this would either
mean that page is about to be freed, or that someone has already issolated the
page.
Being the former case, one could say that falling-through alloc_and_dissolve is
ok.

But no, I did not really want to optimize anything here, just wanted to be explicit
about what we are checking and why.

> > +
> > + if (!err) {
> > + ret = true;
> > + } else if (err == -EBUSY && try_again) {
> > + try_again = false;
> > + goto retry;
> > + }
>
> Is this retry once logic really needed? Does it really give us any real
> benefit? alloc_and_dissolve_huge_page already retries when the page is
> being freed.

alloc_and_dissolve_huge_page retries when the page is about to be freed.
Here we retry in case alloc_and_dissolve_huge_page() noticed that someone grabbed
the page (refcount 0 -> 1), which would possibly mean userspace started using this
page. If that is the case, we give isolate_huge_page another chance to see if we
can succeed and we can migrate the page.

Chances this to happen? Honestly, as any race, quite low.
So, the answer to your question would be, no, it is not really needed, I just felt
we could play "clever" here.

--
Oscar Salvador
SUSE L3

\
 
 \ /
  Last update: 2021-02-26 11:25    [W:0.117 / U:0.148 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site