Re: [RFC][PATCH] sched,livepatch: Untangle cond_resched() and live-patching

[Petr Mladek <pmladek@suse.com>]

On Mon 2025-03-24 14:49:09, Peter Zijlstra wrote:
> 
> With the goal of deprecating / removing VOLUNTARY preempt, live-patch
> needs to stop relying on cond_resched() to make forward progress.
> 
> Instead, rely on schedule() with TASK_FREEZABLE set. Just like
> live-patching, the freezer needs to be able to stop tasks in a safe /
> known state.

> Compile tested only.
> 
> Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
> ---
>  include/linux/livepatch_sched.h | 15 +++++--------
>  include/linux/sched.h           |  6 -----
>  kernel/livepatch/transition.c   | 30 ++++++-------------------
>  kernel/sched/core.c             | 50 +++++++----------------------------------
>  4 files changed, 21 insertions(+), 80 deletions(-)
> 
> diff --git a/include/linux/livepatch_sched.h b/include/linux/livepatch_sched.h
> index 013794fb5da0..7e8171226dd7 100644
> --- a/include/linux/livepatch_sched.h
> +++ b/include/linux/livepatch_sched.h
> @@ -3,27 +3,24 @@
>  #define _LINUX_LIVEPATCH_SCHED_H_
>  
>  #include <linux/jump_label.h>
> -#include <linux/static_call_types.h>
> +#include <linux/sched.h>
> +
>  
>  #ifdef CONFIG_LIVEPATCH
>  
>  void __klp_sched_try_switch(void);
>  
> -#if !defined(CONFIG_PREEMPT_DYNAMIC) || !defined(CONFIG_HAVE_PREEMPT_DYNAMIC_CALL)
> -
>  DECLARE_STATIC_KEY_FALSE(klp_sched_try_switch_key);
>  
> -static __always_inline void klp_sched_try_switch(void)
> +static __always_inline void klp_sched_try_switch(struct task_struct *curr)
>  {
> -	if (static_branch_unlikely(&klp_sched_try_switch_key))
> +	if (static_branch_unlikely(&klp_sched_try_switch_key) &&
> +	    READ_ONCE(curr->__state) & TASK_FREEZABLE)
>  		__klp_sched_try_switch();
>  }

Do we really need to check the TASK_FREEZABLE state, please?

My understanding is that TASK_FREEZABLE is set when kernel kthreads go into
a "freezable" sleep, e.g. wait_event_freezable().

But __klp_sched_try_switch() should be safe when the task is not
running and the stack is reliable. IMHO, it should be safe anytime
it is being scheduled out.

Note that wait_event_freezable() is a good location. It is usually called in
the main loop of the kthread where the stack is small. So that the chance
that it is not running a livepatched function is higher than on
another random schedulable location.

But we actually wanted to have it in cond_resched() because
it might take a long time to reach the main loop, and sleep there.

Best Regards,
Petr