On 2025-02-27 08:50, David Hildenbrand wrote:
On 25.02.25 23:30, Luiz Capitulino wrote:
On 2025-02-25 11:44, David Hildenbrand wrote:
On 24.02.25 22:59, Luiz Capitulino wrote:
The page_ext_next() function assumes that page extension objects for a
page order allocation always reside in the same memory section, which
may not be true and could lead to crashes. Use the new page_ext
iteration API instead.
Fixes: cf54f310d0d3 ("mm/hugetlb: use __GFP_COMP for gigantic folios")
Signed-off-by: Luiz Capitulino <luizcap@xxxxxxxxxx>
---
mm/page_owner.c | 61 +++++++++++++++++++++++--------------------------
1 file changed, 29 insertions(+), 32 deletions(-)
[...]
void __reset_page_owner(struct page *page, unsigned short order)
@@ -293,11 +297,11 @@ void __reset_page_owner(struct page *page, unsigned short order)
page_owner = get_page_owner(page_ext);
alloc_handle = page_owner->handle;
+ page_ext_put(page_ext);
handle = save_stack(GFP_NOWAIT | __GFP_NOWARN);
- __update_page_owner_free_handle(page_ext, handle, order, current->pid,
+ __update_page_owner_free_handle(page, handle, order, current->pid,
current->tgid, free_ts_nsec);
- page_ext_put(page_ext);
I assume moving that is fine ...
but I'll not that ...
- for (i = 0; i < (1 << new_page_owner->order); i++) {
+ rcu_read_lock();
+ for_each_page_ext(&old->page, 1 << new_page_owner->order, page_ext, iter) {
+ old_page_owner = get_page_owner(page_ext);
old_page_owner->handle = migrate_handle;
- old_ext = page_ext_next(old_ext);
- old_page_owner = get_page_owner(old_ext);
}
+ rcu_read_unlock();
page_ext_put(new_ext);
page_ext_put(old_ext);
... here you are not moving it?
In general, LGTM, only the remaining page_ext_put() are a bit confusing.
Which part you found confusing: the fact that I'm not moving them up or that
we still make use of them?
How we are deferring page_ext_put() when not actually working on these
values anymore. The page_owner itself should not go away here unless we
have a serious bug.
To be precise, can't we simply do the following on top?
Yes, that looks good and I like how the new API allows for simpler code.
My only concern is that if the user is not familiar with the page_ext
internals, it might not be clear what page_ext_put() is actually
protecting in which case it looks wrong that we're using a reference
returned by get_page_owner() after releasing the lock. If you think
that that's not an issue then I can apply this change on top.
diff --git a/mm/page_owner.c b/mm/page_owner.c
index c9d2c688eb981..12044340adf89 100644
--- a/mm/page_owner.c
+++ b/mm/page_owner.c
@@ -356,26 +356,24 @@ void __split_page_owner(struct page *page, int old_order, int new_order)
void __folio_copy_owner(struct folio *newfolio, struct folio *old)
{
- struct page_ext *old_ext;
- struct page_ext *new_ext;
struct page_ext *page_ext;
struct page_ext_iter iter;
struct page_owner *old_page_owner;
struct page_owner *new_page_owner;
depot_stack_handle_t migrate_handle;
- old_ext = page_ext_get(&old->page);
- if (unlikely(!old_ext))
+ page_ext = page_ext_get(&old->page);
+ if (unlikely(!page_ext))
return;
+ old_page_owner = get_page_owner(page_ext);
+ page_ext_put(page_ext);
- new_ext = page_ext_get(&newfolio->page);
- if (unlikely(!new_ext)) {
- page_ext_put(old_ext);
+ page_ext = page_ext_get(&newfolio->page);
+ if (unlikely(!page_ext))
return;
- }
+ new_page_owner = get_page_owner(page_ext);
+ page_ext_put(page_ext);
- old_page_owner = get_page_owner(old_ext);
- new_page_owner = get_page_owner(new_ext);
migrate_handle = new_page_owner->handle;
__update_page_owner_handle(&newfolio->page, old_page_owner->handle,
old_page_owner->order, old_page_owner->gfp_mask,
@@ -402,9 +400,6 @@ void __folio_copy_owner(struct folio *newfolio, struct folio *old)
old_page_owner->handle = migrate_handle;
}
rcu_read_unlock();
-
- page_ext_put(new_ext);
- page_ext_put(old_ext);
}
void pagetypeinfo_showmixedcount_print(struct seq_file *m,
