lkml.org 
[lkml]   [2008]   [Dec]   [1]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 01/15] kmemleak: Add the base support
Hi Catalin,

On Sat, Nov 29, 2008 at 12:43 PM, Catalin Marinas
<catalin.marinas@arm.com> wrote:
> +/*
> + * Insert a pointer into the pointer hash table.
> + */
> +static inline void create_object(unsigned long ptr, size_t size, int ref_count)
> +{

[...]

> + if (ptr < min_addr)
> + min_addr = ptr;
> + if (ptr + size > max_addr)
> + max_addr = ptr + size;
> + /*
> + * Update the boundaries before inserting the object in the
> + * prio search tree.
> + */
> + smp_mb();

I'm not sure I understand the purpose of this memory barrier. As soon
as some other CPU acquires object_tree_lock, updates to the boundaries
will be visible due to the implicit memory barriers in locking
functions (see Documentation/memory-barrier.txt for details).

However, I'm wondering why this isn't a smp_wmb() and..

> +/*
> + * Scan a block of memory (exclusive range) for pointers and move
> + * those found to the gray list.
> + */
> +static void scan_block(void *_start, void *_end, struct memleak_object *scanned)
> +{
> + unsigned long *ptr;
> + unsigned long *start = PTR_ALIGN(_start, BYTES_PER_WORD);
> + unsigned long *end = _end - (BYTES_PER_WORD - 1);
> +
> + for (ptr = start; ptr < end; ptr++) {

...why don't we have the pairing smp_rmb() here before we read
min_addr and max_addr?

> +
> + /*
> + * The boundaries check doesn't need to be precise
> + * (hence no locking) since orphan objects need to
> + * pass a scanning threshold before being reported.
> + */
> + if (pointer < min_addr || pointer >= max_addr)
> + continue;

Pekka


\
 
 \ /
  Last update: 2008-12-01 08:17    [from the cache]
©2003-2011 Jasper Spaans