Re: [PATCH] mm/page_alloc: skip debug_check_no_{obj,locks}_freed with FPI_TRYLOCK

[Harry Yoo <harry.yoo@oracle.com>]

On Fri, Feb 06, 2026 at 06:34:25PM +0100, Vlastimil Babka wrote:
> On 2/6/26 18:20, Harry Yoo wrote:
> > On Fri, Feb 06, 2026 at 12:08:04PM -0500, Zi Yan wrote:
> >> On 6 Feb 2026, at 11:58, Harry Yoo wrote:
> >> 
> >> > When CONFIG_DEBUG_OBJECTS_FREE is enabled,
> >> > debug_check_no_{obj,locks}_freed() functions are called.
> >> >
> >> > Since both of them spin on a lock, they are not safe to be called
> >> > if the FPI_TRYLOCK flag is specified. This leads to a lockdep splat:
> >> >
> >> >   ================================
> >> >   WARNING: inconsistent lock state
> >> >   6.19.0-rc5-slab-for-next+ #326 Tainted: G                 N
> >> >   --------------------------------
> >> >   inconsistent {INITIAL USE} -> {IN-NMI} usage.
> >> >   kunit_try_catch/9046 [HC2[2]:SC0[0]:HE0:SE1] takes:
> >> >   ffffffff84ed6bf8 (&obj_hash[i].lock){-.-.}-{2:2}, at: __debug_check_no_obj_freed+0xe0/0x300
> >> >   {INITIAL USE} state was registered at:
> >> >     lock_acquire+0xd9/0x2f0
> >> >     _raw_spin_lock_irqsave+0x4c/0x80
> >> >     __debug_object_init+0x9d/0x1f0
> >> >     debug_object_init+0x34/0x50
> >> >     __init_work+0x28/0x40
> >> >     init_cgroup_housekeeping+0x151/0x210
> >> >     init_cgroup_root+0x3d/0x140
> >> >     cgroup_init_early+0x30/0x240
> >> >     start_kernel+0x3e/0xcd0
> >> >     x86_64_start_reservations+0x18/0x30
> >> >     x86_64_start_kernel+0xf3/0x140
> >> >     common_startup_64+0x13e/0x148
> >> >   irq event stamp: 2998
> >> >   hardirqs last  enabled at (2997): [<ffffffff8298b77a>] exc_nmi+0x11a/0x240
> >> >   hardirqs last disabled at (2998): [<ffffffff8298b991>] sysvec_irq_work+0x11/0x110
> >> >   softirqs last  enabled at (1416): [<ffffffff813c1f72>] __irq_exit_rcu+0x132/0x1c0
> >> >   softirqs last disabled at (1303): [<ffffffff813c1f72>] __irq_exit_rcu+0x132/0x1c0
> >> >
> >> >   other info that might help us debug this:
> >> >    Possible unsafe locking scenario:
> >> >
> >> >          CPU0
> >> >          ----
> >> >     lock(&obj_hash[i].lock);
> >> >     <Interrupt>
> >> >       lock(&obj_hash[i].lock);
> >> >
> >> >    *** DEADLOCK ***
> >> >
> >> > Fix this by adding an fpi_t parameter to free_pages_prepare() and
> >> > skipping those checks if FPI_TRYLOCK is set. Since mm/compaction.c
> >> > calls free_pages_prepare(), move the fpi_t definition to mm/internal.h.
> >> >
> >> > Fixes: 8c57b687e833 ("mm, bpf: Introduce free_pages_nolock()")
> >> > Cc: <stable@vger.kernel.org>
> >> > Signed-off-by: Harry Yoo <harry.yoo@oracle.com>
> >> > ---
> >> >  mm/compaction.c |  2 +-
> >> >  mm/internal.h   | 35 ++++++++++++++++++++++++++++++++++-
> >> >  mm/page_alloc.c | 42 ++++++------------------------------------
> >> >  3 files changed, 41 insertions(+), 38 deletions(-)
> >> >
> >> > diff --git a/mm/compaction.c b/mm/compaction.c
> >> > index 1e8f8eca318c..9ffeb7c6d2b0 100644
> >> > --- a/mm/compaction.c
> >> > +++ b/mm/compaction.c
> >> > @@ -1859,7 +1859,7 @@ static void compaction_free(struct folio *dst, unsigned long data)
> >> >  	struct page *page = &dst->page;
> >> >
> >> >  	if (folio_put_testzero(dst)) {
> >> > -		free_pages_prepare(page, order);
> >> > +		free_pages_prepare(page, order, FPI_NONE);
> >> 
> >> Is it OK to add something like free_pages_prepare_fpi_none() for this one
> >> to avoid the FPI flag move?
> > 
> > Yeah, moving FPI flag definition isn't great :)
> > 
> > I'm totally fine with your suggestion,
> > as long as page allocator/compaction folks are fine with it!
> 
> Maybe even like this?
> free_pages_prepare() which calls __free_pages_prepare(FPI_NONE)

Sounds good and will respin v2 soon, thanks!

-- 
Cheers,
Harry / Hyeonggon