* [PATCH] mm/page_owner: make page_owner stack depth runtime configurable
@ 2026-10-10 9:38 Kanishka Solanki
2026-10-10 11:42 ` Zi Yan
0 siblings, 1 reply; 3+ messages in thread
From: Kanishka Solanki @ 2026-10-10 9:38 UTC (permalink / raw)
To: akpm, vbabka
Cc: surenb, mhocko, brendan.jackman, hannes, ziy, linux-mm,
linux-kernel, Kanishka Solanki
The stack depth for page_owner allocations was previously fixed at
PAGE_OWNER_STACK_DEPTH (16).
Replace the fixed stack depth constant with a compile-time max ceiling
(PAGE_OWNER_STACK_DEPTH_MAX = 64) and expose a module parameter
'stack_depth' via module_param_named. This allows users to configure
the stack recording depth between 1 and 64 at boot or runtime via sysfs,
while preventing unbounded stack usage.
Signed-off-by: Kanishka <kanishkasolanki456s@gmail.com>
Signed-off-by: Kanishka Solanki <kanishkasolanki456s@gmail.com>
---
mm/page_owner.c | 15 ++++++++-------
1 file changed, 8 insertions(+), 7 deletions(-)
diff --git a/mm/page_owner.c b/mm/page_owner.c
index fbbda7ba914b..fcb584070f1d 100644
--- a/mm/page_owner.c
+++ b/mm/page_owner.c
@@ -15,11 +15,10 @@
#include "page_alloc.h"
-/*
- * TODO: teach PAGE_OWNER_STACK_DEPTH (__dump_page_owner and save_stack)
- * to use off stack temporal storage
- */
-#define PAGE_OWNER_STACK_DEPTH (16)
+#define PAGE_OWNER_STACK_DEPTH_MAX (64)
+
+static unsigned int page_owner_stack_depth = 16;
+module_param_named(stack_depth, page_owner_stack_depth, uint, 0644);
struct page_owner {
unsigned short order;
@@ -172,15 +171,17 @@ static inline struct page_owner *get_page_owner(struct page_ext *page_ext)
static noinline depot_stack_handle_t save_stack(gfp_t flags)
{
- unsigned long entries[PAGE_OWNER_STACK_DEPTH];
+ unsigned long entries[PAGE_OWNER_STACK_DEPTH_MAX];
depot_stack_handle_t handle;
unsigned int nr_entries;
+ unsigned int max_entries;
if (current->in_page_owner)
return dummy_handle;
set_current_in_page_owner();
- nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 2);
+ max_entries = min_t(unsigned int, page_owner_stack_depth, ARRAY_SIZE(entries));
+ nr_entries = stack_trace_save(entries, max_entries, 2);
handle = stack_depot_save(entries, nr_entries, flags);
if (!handle)
handle = failure_handle;
--
2.43.0
^ permalink raw reply [flat|nested] 3+ messages in thread* Re: [PATCH] mm/page_owner: make page_owner stack depth runtime configurable
2026-10-10 9:38 [PATCH] mm/page_owner: make page_owner stack depth runtime configurable Kanishka Solanki
@ 2026-10-10 11:42 ` Zi Yan
2026-10-12 4:15 ` Kanishka Solanki
0 siblings, 1 reply; 3+ messages in thread
From: Zi Yan @ 2026-10-10 11:42 UTC (permalink / raw)
To: Kanishka Solanki, akpm, vbabka
Cc: surenb, mhocko, brendan.jackman, hannes, linux-mm, linux-kernel
On Sat Oct 10, 2026 at 5:38 AM EDT, Kanishka Solanki wrote:
> The stack depth for page_owner allocations was previously fixed at
> PAGE_OWNER_STACK_DEPTH (16).
>
> Replace the fixed stack depth constant with a compile-time max ceiling
> (PAGE_OWNER_STACK_DEPTH_MAX = 64) and expose a module parameter
> 'stack_depth' via module_param_named. This allows users to configure
> the stack recording depth between 1 and 64 at boot or runtime via sysfs,
> while preventing unbounded stack usage.
What is your motivation? If you see page owner stack gets cutoff due to
insufficient depth, explain your use cases first.
>
> Signed-off-by: Kanishka <kanishkasolanki456s@gmail.com>
> Signed-off-by: Kanishka Solanki <kanishkasolanki456s@gmail.com>
> ---
> mm/page_owner.c | 15 ++++++++-------
> 1 file changed, 8 insertions(+), 7 deletions(-)
>
> diff --git a/mm/page_owner.c b/mm/page_owner.c
> index fbbda7ba914b..fcb584070f1d 100644
> --- a/mm/page_owner.c
> +++ b/mm/page_owner.c
> @@ -15,11 +15,10 @@
>
> #include "page_alloc.h"
>
> -/*
> - * TODO: teach PAGE_OWNER_STACK_DEPTH (__dump_page_owner and save_stack)
> - * to use off stack temporal storage
> - */
> -#define PAGE_OWNER_STACK_DEPTH (16)
> +#define PAGE_OWNER_STACK_DEPTH_MAX (64)
This uses more stack. Do not do this.
> +
> +static unsigned int page_owner_stack_depth = 16;
> +module_param_named(stack_depth, page_owner_stack_depth, uint, 0644);
>
> struct page_owner {
> unsigned short order;
> @@ -172,15 +171,17 @@ static inline struct page_owner *get_page_owner(struct page_ext *page_ext)
>
> static noinline depot_stack_handle_t save_stack(gfp_t flags)
> {
> - unsigned long entries[PAGE_OWNER_STACK_DEPTH];
> + unsigned long entries[PAGE_OWNER_STACK_DEPTH_MAX];
> depot_stack_handle_t handle;
> unsigned int nr_entries;
> + unsigned int max_entries;
>
> if (current->in_page_owner)
> return dummy_handle;
>
> set_current_in_page_owner();
> - nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 2);
> + max_entries = min_t(unsigned int, page_owner_stack_depth, ARRAY_SIZE(entries));
> + nr_entries = stack_trace_save(entries, max_entries, 2);
> handle = stack_depot_save(entries, nr_entries, flags);
> if (!handle)
> handle = failure_handle;
--
Best Regards,
Yan, Zi
^ permalink raw reply [flat|nested] 3+ messages in thread* Re: [PATCH] mm/page_owner: make page_owner stack depth runtime configurable
2026-10-10 11:42 ` Zi Yan
@ 2026-10-12 4:15 ` Kanishka Solanki
0 siblings, 0 replies; 3+ messages in thread
From: Kanishka Solanki @ 2026-10-12 4:15 UTC (permalink / raw)
To: Zi Yan
Cc: akpm, vbabka, surenb, mhocko, brendan.jackman, hannes, linux-mm,
linux-kernel
On Sat, Oct 10, 2026 at 5:12 PM Zi Yan <ziy@nvidia.com> wrote:
>
> On Sat Oct 10, 2026 at 5:38 AM EDT, Kanishka Solanki wrote:
> > The stack depth for page_owner allocations was previously fixed at
> > PAGE_OWNER_STACK_DEPTH (16).
> >
> > Replace the fixed stack depth constant with a compile-time max ceiling
> > (PAGE_OWNER_STACK_DEPTH_MAX = 64) and expose a module parameter
> > 'stack_depth' via module_param_named. This allows users to configure
> > the stack recording depth between 1 and 64 at boot or runtime via sysfs,
> > while preventing unbounded stack usage.
>
> What is your motivation? If you see page owner stack gets cutoff due to
> insufficient depth, explain your use cases first.
>
> >
> > Signed-off-by: Kanishka <kanishkasolanki456s@gmail.com>
> > Signed-off-by: Kanishka Solanki <kanishkasolanki456s@gmail.com>
> > ---
> > mm/page_owner.c | 15 ++++++++-------
> > 1 file changed, 8 insertions(+), 7 deletions(-)
> >
> > diff --git a/mm/page_owner.c b/mm/page_owner.c
> > index fbbda7ba914b..fcb584070f1d 100644
> > --- a/mm/page_owner.c
> > +++ b/mm/page_owner.c
> > @@ -15,11 +15,10 @@
> >
> > #include "page_alloc.h"
> >
> > -/*
> > - * TODO: teach PAGE_OWNER_STACK_DEPTH (__dump_page_owner and save_stack)
> > - * to use off stack temporal storage
> > - */
> > -#define PAGE_OWNER_STACK_DEPTH (16)
> > +#define PAGE_OWNER_STACK_DEPTH_MAX (64)
>
> This uses more stack. Do not do this.
>
> > +
> > +static unsigned int page_owner_stack_depth = 16;
> > +module_param_named(stack_depth, page_owner_stack_depth, uint, 0644);
> >
> > struct page_owner {
> > unsigned short order;
> > @@ -172,15 +171,17 @@ static inline struct page_owner *get_page_owner(struct page_ext *page_ext)
> >
> > static noinline depot_stack_handle_t save_stack(gfp_t flags)
> > {
> > - unsigned long entries[PAGE_OWNER_STACK_DEPTH];
> > + unsigned long entries[PAGE_OWNER_STACK_DEPTH_MAX];
> > depot_stack_handle_t handle;
> > unsigned int nr_entries;
> > + unsigned int max_entries;
> >
> > if (current->in_page_owner)
> > return dummy_handle;
> >
> > set_current_in_page_owner();
> > - nr_entries = stack_trace_save(entries, ARRAY_SIZE(entries), 2);
> > + max_entries = min_t(unsigned int, page_owner_stack_depth, ARRAY_SIZE(entries));
> > + nr_entries = stack_trace_save(entries, max_entries, 2);
> > handle = stack_depot_save(entries, nr_entries, flags);
> > if (!handle)
> > handle = failure_handle;
>
>
>
>
> --
> Best Regards,
> Yan, Zi
>
Hi Zi,
Thank you for the feedback.
I misunderstood the original TODO comment in mm/page_owner.c:
/*
- TODO: teach PAGE_OWNER_STACK_DEPTH (__dump_page_owner and save_stack)
- to use off stack temporal storage
*/
You are right. I will drop this patch.
Thank you
Kanishka
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2026-10-12 4:16 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2026-10-10 9:38 [PATCH] mm/page_owner: make page_owner stack depth runtime configurable Kanishka Solanki
2026-10-10 11:42 ` Zi Yan
2026-10-12 4:15 ` Kanishka Solanki
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox
all inboxes | Powered by JetHome®