Re: [PATCH 1/2] srcu: Fix broken node geometry after early ssp init

[Paul E. McKenney]

On Wed, Apr 14, 2021 at 03:24:12PM +0200, Frederic Weisbecker wrote:
> An ssp initialized before rcu_init_geometry() will have its snp hierarchy
> based on CONFIG_NR_CPUS.
> 
> Once rcu_init_geometry() is called, the nodes distribution is shrinked
> and optimized toward meeting the actual possible number of CPUs detected
> on boot.
> 
> Later on, the ssp that was initialized before rcu_init_geometry() is
> confused and sometimes refers to its initial CONFIG_NR_CPUS based node
> hierarchy, sometimes to the new num_possible_cpus() based one instead.
> For example each of its sdp->mynode remain backward and refer to the
> early node leaves that may not exist anymore. On the other hand the
> srcu_for_each_node_breadth_first() refers to the new node hierarchy.
> 
> There are at least two bad possible outcomes to this:
> 
> 1) a) A callback enqueued early on an sdp is recorded pending on
>       sdp->mynode->srcu_data_have_cbs in srcu_funnel_gp_start() with
>       sdp->mynode pointing to a deep leaf (say 3 levels).
> 
>    b) The grace period ends after rcu_init_geometry() which shrinks the
>       nodes level to a single one. srcu_gp_end() walks through the new
>       snp hierarchy without ever reaching the old leaves so the callback
>       is never executed.
> 
>    This is easily reproduced on an 8 CPUs machine with
>    CONFIG_NR_CPUS >= 32 and "rcupdate.rcu_self_test=1". The
>    srcu_barrier() after early tests verification never completes and
>    the boot hangs:
> 
> 	[ 5413.141029] INFO: task swapper/0:1 blocked for more than 4915 seconds.
> 	[ 5413.147564]       Not tainted 5.12.0-rc4+ #28
> 	[ 5413.151927] "echo 0 > /proc/sys/kernel/hung_task_timeout_secs" disables this message.
> 	[ 5413.159753] task:swapper/0       state:D stack:    0 pid:    1 ppid:     0 flags:0x00004000
> 	[ 5413.168099] Call Trace:
> 	[ 5413.170555]  __schedule+0x36c/0x930
> 	[ 5413.174057]  ? wait_for_completion+0x88/0x110
> 	[ 5413.178423]  schedule+0x46/0xf0
> 	[ 5413.181575]  schedule_timeout+0x284/0x380
> 	[ 5413.185591]  ? wait_for_completion+0x88/0x110
> 	[ 5413.189957]  ? mark_held_locks+0x61/0x80
> 	[ 5413.193882]  ? mark_held_locks+0x61/0x80
> 	[ 5413.197809]  ? _raw_spin_unlock_irq+0x24/0x50
> 	[ 5413.202173]  ? wait_for_completion+0x88/0x110
> 	[ 5413.206535]  wait_for_completion+0xb4/0x110
> 	[ 5413.210724]  ? srcu_torture_stats_print+0x110/0x110
> 	[ 5413.215610]  srcu_barrier+0x187/0x200
> 	[ 5413.219277]  ? rcu_tasks_verify_self_tests+0x50/0x50
> 	[ 5413.224244]  ? rdinit_setup+0x2b/0x2b
> 	[ 5413.227907]  rcu_verify_early_boot_tests+0x2d/0x40
> 	[ 5413.232700]  do_one_initcall+0x63/0x310
> 	[ 5413.236541]  ? rdinit_setup+0x2b/0x2b
> 	[ 5413.240207]  ? rcu_read_lock_sched_held+0x52/0x80
> 	[ 5413.244912]  kernel_init_freeable+0x253/0x28f
> 	[ 5413.249273]  ? rest_init+0x250/0x250
> 	[ 5413.252846]  kernel_init+0xa/0x110
> 	[ 5413.256257]  ret_from_fork+0x22/0x30
> 
> 2) An ssp that gets initialized before rcu_init_geometry() and used
>    afterward will always have stale rdp->mynode references, resulting in
>    callbacks to be missed in srcu_gp_end(), just like in the previous
>    scenario.
> 
> Solve this with calling rcu_init_geometry() whenever an struct srcu_state
> happens to be initialized before rcu_init(). This way we make sure the
> RCU nodes hierarchy is properly built and distributed before the nodes
> of a struct srcu_state are allocated.
> 
> Suggested-by: Paul E. McKenney <paulmck@xxxxxxxxxx>
> Signed-off-by: Frederic Weisbecker <frederic@xxxxxxxxxx>

Much much nicer, thank you!  Comments and questions interspersed.

							Thanx, Paul

> Cc: Boqun Feng <boqun.feng@xxxxxxxxx>
> Cc: Lai Jiangshan <jiangshanlai@xxxxxxxxx>
> Cc: Neeraj Upadhyay <neeraju@xxxxxxxxxxxxxx>
> Cc: Josh Triplett <josh@xxxxxxxxxxxxxxxx>
> Cc: Joel Fernandes <joel@xxxxxxxxxxxxxxxxx>
> Cc: Uladzislau Rezki <urezki@xxxxxxxxx>
> ---
>  kernel/rcu/rcu.h      |  3 +++
>  kernel/rcu/srcutree.c |  2 ++
>  kernel/rcu/tree.c     | 18 +++++++++++++++++-
>  3 files changed, 22 insertions(+), 1 deletion(-)
> 
> diff --git a/kernel/rcu/rcu.h b/kernel/rcu/rcu.h
> index 75ed367d5b60..24db97cbf76b 100644
> --- a/kernel/rcu/rcu.h
> +++ b/kernel/rcu/rcu.h
> @@ -278,6 +278,7 @@ extern void resched_cpu(int cpu);
>  extern int rcu_num_lvls;
>  extern int num_rcu_lvl[];
>  extern int rcu_num_nodes;
> +extern bool rcu_geometry_initialized;

Can this be a static local variable inside rcu_init_geometry()?

After all, init_srcu_struct() isn't called all that often, and its overhead
is such that an extra function call and check is going to hurt it.  This
of course requires removing __init from rcu_init_geometry(), but it is not
all that large, so why not just remove the __init?

But if we really are worried about reclaiming rcu_init_geometry()'s
instructions (maybe we are?), then rcu_init_geometry() can be split
into a function that just does the check (which is not __init) and the
remainder of the function, which could remain __init.

>  static bool rcu_fanout_exact;
>  static int rcu_fanout_leaf;
>  
> @@ -308,6 +309,8 @@ static inline void rcu_init_levelspread(int *levelspread, const int *levelcnt)
>  	}
>  }
>  
> +extern void rcu_init_geometry(void);
> +
>  /* Returns a pointer to the first leaf rcu_node structure. */
>  #define rcu_first_leaf_node() (rcu_state.level[rcu_num_lvls - 1])
>  
> diff --git a/kernel/rcu/srcutree.c b/kernel/rcu/srcutree.c
> index 108f9ca06047..05ca3c275af1 100644
> --- a/kernel/rcu/srcutree.c
> +++ b/kernel/rcu/srcutree.c
> @@ -171,6 +171,8 @@ static int init_srcu_struct_fields(struct srcu_struct *ssp, bool is_static)
>  		ssp->sda = alloc_percpu(struct srcu_data);
>  	if (!ssp->sda)
>  		return -ENOMEM;
> +	if (!rcu_geometry_initialized)
> +		rcu_init_geometry();

With the suggested change above, this just becomes an unconditional call
to rcu_init_geometry().

>  	init_srcu_struct_nodes(ssp);
>  	ssp->srcu_gp_seq_needed_exp = 0;
>  	ssp->srcu_last_gp_end = ktime_get_mono_fast_ns();
> diff --git a/kernel/rcu/tree.c b/kernel/rcu/tree.c
> index 740f5cd34459..b1d6e60e08d1 100644
> --- a/kernel/rcu/tree.c
> +++ b/kernel/rcu/tree.c
> @@ -4556,17 +4556,33 @@ static void __init rcu_init_one(void)
>  	}
>  }
>  
> +bool rcu_geometry_initialized __read_mostly;

This could then go away.

> +
>  /*
>   * Compute the rcu_node tree geometry from kernel parameters.  This cannot
>   * replace the definitions in tree.h because those are needed to size
>   * the ->node array in the rcu_state structure.
>   */
> -static void __init rcu_init_geometry(void)
> +void rcu_init_geometry(void)
>  {
>  	ulong d;
>  	int i;
> +	static unsigned long old_nr_cpu_ids;
>  	int rcu_capacity[RCU_NUM_LVLS];

And then rcu_geometry_initialized is declared static here.

Or am I missing something?

> +	if (rcu_geometry_initialized) {
> +		/*
> +		 * Arrange for warning if rcu_init_geometry() was called before
> +		 * setup_nr_cpu_ids(). We may miss cases when
> +		 * nr_cpus_ids == NR_CPUS but that shouldn't matter too much.
> +		 */
> +		WARN_ON_ONCE(old_nr_cpu_ids != nr_cpu_ids);
> +		return;
> +	}
> +
> +	old_nr_cpu_ids = nr_cpu_ids;
> +	rcu_geometry_initialized = true;
> +
>  	/*
>  	 * Initialize any unspecified boot parameters.
>  	 * The default values of jiffies_till_first_fqs and
> -- 
> 2.25.1
>