diff mbox series

[0.5/3] livepatch: Convert stack entries array to percpu

Message ID 20230313233346.kayh4t2lpicjkpsv@treble
State New
Headers
Series livepatch,sched: Add livepatch task switching to cond_resched() |

Commit Message

Josh Poimboeuf March 13, 2023, 11:33 p.m. UTC 
  On Fri, Mar 03, 2023 at 03:00:13PM +0100, Petr Mladek wrote:
> > MAX_STACK_ENTRIES is 100, which seems excessive.  If we halved that, the
> > array would be "only" 400 bytes, which is *almost* reasonable to
> > allocate on the stack?
> 
> It is just for the stack in the process context. Right?
> 
> I think that I have never seen a stack with over 50 entries. And in
> the worst case, a bigger amount of entries would "just" result in
> a non-reliable stack which might be acceptable.
> 
> It looks acceptable to me.
> 
> > Alternatively we could have a percpu entries array... :-/
> 
> That said, percpu entries would be fine as well. It sounds like
> a good price for the livepatching feature. I think that livepatching
> is used on big systems anyway.
> 
> I slightly prefer the per-cpu solution.

Booting a kernel with PREEMPT+LOCKDEP gave me a high-water mark of 60+
stack entries, seen when probing a device.  I decided not to mess with
MAX_STACK_ENTRIES, and instead just convert the entries to percpu.  This
patch could be inserted at the beginning of the set.

---8<---

Subject: [PATCH 0.5/3] livepatch: Convert stack entries array to percpu

The entries array in klp_check_stack() is static local because it's too
big to be reasonably allocated on the stack.  Serialized access is
enforced by the klp_mutex.

In preparation for calling klp_check_stack() without the mutex (from
cond_resched), convert it to a percpu variable.

Signed-off-by: Josh Poimboeuf <jpoimboe@kernel.org>
---
 kernel/livepatch/transition.c | 9 +++++++--
 1 file changed, 7 insertions(+), 2 deletions(-)

Comments

Petr Mladek March 14, 2023, 10:50 a.m. UTC | #1 
On Mon 2023-03-13 16:33:46, Josh Poimboeuf wrote:
> On Fri, Mar 03, 2023 at 03:00:13PM +0100, Petr Mladek wrote:
> > > MAX_STACK_ENTRIES is 100, which seems excessive.  If we halved that, the
> > > array would be "only" 400 bytes, which is *almost* reasonable to
> > > allocate on the stack?
> > 
> > It is just for the stack in the process context. Right?
> > 
> > I think that I have never seen a stack with over 50 entries. And in
> > the worst case, a bigger amount of entries would "just" result in
> > a non-reliable stack which might be acceptable.
> > 
> > It looks acceptable to me.
> > 
> > > Alternatively we could have a percpu entries array... :-/
> > 
> > That said, percpu entries would be fine as well. It sounds like
> > a good price for the livepatching feature. I think that livepatching
> > is used on big systems anyway.
> > 
> > I slightly prefer the per-cpu solution.
> 
> Booting a kernel with PREEMPT+LOCKDEP gave me a high-water mark of 60+
> stack entries, seen when probing a device.  I decided not to mess with
> MAX_STACK_ENTRIES, and instead just convert the entries to percpu.  This
> patch could be inserted at the beginning of the set.

Good to know.

> 
> ---8<---
> 
> Subject: [PATCH 0.5/3] livepatch: Convert stack entries array to percpu
> 
> --- a/kernel/livepatch/transition.c
> +++ b/kernel/livepatch/transition.c
> @@ -240,12 +242,15 @@ static int klp_check_stack_func(struct klp_func *func, unsigned long *entries,
>   */
>  static int klp_check_stack(struct task_struct *task, const char **oldname)
>  {
> -	static unsigned long entries[MAX_STACK_ENTRIES];
> +	unsigned long *entries = this_cpu_ptr(klp_stack_entries);
>  	struct klp_object *obj;
>  	struct klp_func *func;
>  	int ret, nr_entries;
>  
> -	ret = stack_trace_save_tsk_reliable(task, entries, ARRAY_SIZE(entries));
> +	/* Protect 'klp_stack_entries' */
> +	lockdep_assert_preemption_disabled();

I think about adding:

	/*
	 * Stay on the safe side even when cond_resched() is called from
	 * an IRQ context by mistake.
	 */
	if (!in_task())
		return -EINVAL;

Or is this prevented another way, please?

> +
> +	ret = stack_trace_save_tsk_reliable(task, entries, MAX_STACK_ENTRIES);
>  	if (ret < 0)
>  		return -EINVAL;
>  	nr_entries = ret;

Otherwise, it looks good to me.

Best Regards,
Petr
Josh Poimboeuf March 15, 2023, 12:26 a.m. UTC | #2 
On Tue, Mar 14, 2023 at 11:50:21AM +0100, Petr Mladek wrote:
> >  static int klp_check_stack(struct task_struct *task, const char **oldname)
> >  {
> > -	static unsigned long entries[MAX_STACK_ENTRIES];
> > +	unsigned long *entries = this_cpu_ptr(klp_stack_entries);
> >  	struct klp_object *obj;
> >  	struct klp_func *func;
> >  	int ret, nr_entries;
> >  
> > -	ret = stack_trace_save_tsk_reliable(task, entries, ARRAY_SIZE(entries));
> > +	/* Protect 'klp_stack_entries' */
> > +	lockdep_assert_preemption_disabled();
> 
> I think about adding:
> 
> 	/*
> 	 * Stay on the safe side even when cond_resched() is called from
> 	 * an IRQ context by mistake.
> 	 */
> 	if (!in_task())
> 		return -EINVAL;
> 
> Or is this prevented another way, please?

An IRQ handler trying to schedule would be a pretty major bug, no?

I think there are already several different checks for that.  For
example, the __might_resched() call in cond_resched().
diff mbox series

Patch

diff --git a/kernel/livepatch/transition.c b/kernel/livepatch/transition.c
index f1b25ec581e0..135fc73e2e5d 100644
--- a/kernel/livepatch/transition.c
+++ b/kernel/livepatch/transition.c
@@ -14,6 +14,8 @@ 
 #include "transition.h"
 
 #define MAX_STACK_ENTRIES  100
+DEFINE_PER_CPU(unsigned long[MAX_STACK_ENTRIES], klp_stack_entries);
+
 #define STACK_ERR_BUF_SIZE 128
 
 #define SIGNALS_TIMEOUT 15
@@ -240,12 +242,15 @@  static int klp_check_stack_func(struct klp_func *func, unsigned long *entries,
  */
 static int klp_check_stack(struct task_struct *task, const char **oldname)
 {
-	static unsigned long entries[MAX_STACK_ENTRIES];
+	unsigned long *entries = this_cpu_ptr(klp_stack_entries);
 	struct klp_object *obj;
 	struct klp_func *func;
 	int ret, nr_entries;
 
-	ret = stack_trace_save_tsk_reliable(task, entries, ARRAY_SIZE(entries));
+	/* Protect 'klp_stack_entries' */
+	lockdep_assert_preemption_disabled();
+
+	ret = stack_trace_save_tsk_reliable(task, entries, MAX_STACK_ENTRIES);
 	if (ret < 0)
 		return -EINVAL;
 	nr_entries = ret;