Re: [mm: meminit] WARNING: CPU: 1 PID: 15 at kernel/locking/lockdep.c:3382 lock_release()

[Peter Zijlstra]

On Wed, Jul 08, 2015 at 11:32:13AM +0100, Mel Gorman wrote:
> From: Nicolai Stange <nicstange@xxxxxxxxx>
> Subject: Re: [PATCH] mm/page_alloc: deferred meminit: replace rwsem with completion
> 
> Commit 0e1cc95b4cc7
>   ("mm: meminit: finish initialisation of struct pages before basic setup")
> introduced a rwsem to signal completion of the initialization workers.
> 
> Lockdep complains about possible recursive locking:
>   =============================================
>   [ INFO: possible recursive locking detected ]
>   4.1.0-12802-g1dc51b8 #3 Not tainted
>   ---------------------------------------------
>   swapper/0/1 is trying to acquire lock:
>   (pgdat_init_rwsem){++++.+},
>     at: [<ffffffff8424c7fb>] page_alloc_init_late+0xc7/0xe6
> 
>   but task is already holding lock:
>   (pgdat_init_rwsem){++++.+},
>     at: [<ffffffff8424c772>] page_alloc_init_late+0x3e/0xe6
> 
> Replace the rwsem by a completion together with an atomic
> "outstanding work counter".
> 
> Signed-off-by: Nicolai Stange <nicstange@xxxxxxxxx>
> Acked-by: Mel Gorman <mgorman@xxxxxxx>
> ---
>  mm/page_alloc.c | 34 +++++++++++++++++++++++++++-------
>  1 file changed, 27 insertions(+), 7 deletions(-)
> 
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index 506eac8..3886e66 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -18,7 +18,9 @@
>  #include <linux/mm.h>
>  #include <linux/swap.h>
>  #include <linux/interrupt.h>
> -#include <linux/rwsem.h>
> +#include <linux/completion.h>
> +#include <linux/atomic.h>
> +#include <asm/barrier.h>
>  #include <linux/pagemap.h>
>  #include <linux/jiffies.h>
>  #include <linux/bootmem.h>
> @@ -1062,7 +1064,20 @@ static void __init deferred_free_range(struct page *page,
>  		__free_pages_boot_core(page, pfn, 0);
>  }
>  
> -static __initdata DECLARE_RWSEM(pgdat_init_rwsem);
> +/* counter and completion tracking outstanding deferred_init_memmap()
> +   threads */

Wrong comment style.

> +static atomic_t pgdat_init_n_undone __initdata;
> +static __initdata DECLARE_COMPLETION(pgdat_init_all_done_comp);
> +
> +static inline void __init pgdat_init_report_one_done(void)
> +{
> +	/* Write barrier is paired with read barrier in
> +	   page_alloc_init_late(). It makes all writes visible to
> +	   readers seeing our decrement on pgdat_init_n_undone. */

Wrong comment style.

> +	smp_wmb();

Pointless barrier, because

> +	if (atomic_dec_and_test(&pgdat_init_n_undone))

implies a full memory barrier, furthermore see below.

> +		complete(&pgdat_init_all_done_comp);
> +}
>  
>  /* Initialise remaining memory on a node */
>  static int __init deferred_init_memmap(void *data)


> @@ -1187,14 +1203,18 @@ void __init page_alloc_init_late(void)
>  {
>  	int nid;
>  
> +	/* There will be num_node_state(N_MEMORY) threads */
> +	atomic_set(&pgdat_init_n_undone, num_node_state(N_MEMORY));
>  	for_each_node_state(nid, N_MEMORY) {
> -		down_read(&pgdat_init_rwsem);
>  		kthread_run(deferred_init_memmap, NODE_DATA(nid), "pgdatinit%d", nid);
>  	}
>  
>  	/* Block until all are initialised */
> -	down_write(&pgdat_init_rwsem);
> -	up_write(&pgdat_init_rwsem);
> +	wait_for_completion(&pgdat_init_all_done_comp);
> +
> +	/* Paired with write barrier in deferred_init_memmap(),
> +	   ensures a consistent view of all its writes. */

Wrong comment style

> +	smp_rmb();

Wrong barrier, IF you want a barrier it should be before
wait_for_completion, such that if you observe complete, you then must
also observe whatever happened prior to the completion.

But I would argue a completion had better imply that anyway.

>  }
>  #endif /* CONFIG_DEFERRED_STRUCT_PAGE_INIT */
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/