[I have only briefly looked at patches so I might have missed some
details.]
On Thu 16-11-17 12:14:25, Byungchul Park wrote:
> Although lock_page() and its family can cause deadlock, lockdep have not
> worked with them, because unlock_page() might be called in a different
> context from the acquire context, which violated lockdep's assumption.
>
> Now CONFIG_LOCKDEP_CROSSRELEASE has been introduced, lockdep can work
> with page locks.
I definitely agree that debugging page_lock deadlocks is a major PITA
but your implementation seems prohibitively too expensive.
[...]
> @@ -218,6 +222,10 @@ struct page {
> #ifdef LAST_CPUPID_NOT_IN_PAGE_FLAGS
> int _last_cpupid;
> #endif
> +
> +#ifdef CONFIG_LOCKDEP_PAGELOCK
> + struct lockdep_map_cross map;
> +#endif
> }
now you are adding
struct lockdep_map_cross {
struct lockdep_map map; /* 0 40 */
struct cross_lock xlock; /* 40 56 */
/* --- cacheline 1 boundary (64 bytes) was 32 bytes ago --- */
/* size: 96, cachelines: 2, members: 2 */
/* last cacheline: 32 bytes */
};
for each struct page. So you are doubling the size. Who is going to
enable this config option? You are moving this to page_ext in a later
patch which is a good step but it doesn't go far enough because this
still consumes those resources. Is there any problem to make this
kernel command line controllable? Something we do for page_owner for
example?
Also it would be really great if you could give us some measures about
the runtime overhead. I do not expect it to be very large but this is
something people are usually interested in when enabling debugging
features.
--
Michal Hocko
SUSE Labs
