2013-03-12 17:26:52

by Dimitris Papastamos

[permalink] [raw]
Subject: [PATCH v2] regmap: Expose total memory consumption in the rbtree debugfs entry

Provide a feel of how much overhead the rbtree cache adds to
the game.

Signed-off-by: Dimitris Papastamos <[email protected]>
---
Print the size in bytes instead of kB.

drivers/base/regmap/regcache-rbtree.c | 9 +++++++--
1 file changed, 7 insertions(+), 2 deletions(-)

diff --git a/drivers/base/regmap/regcache-rbtree.c b/drivers/base/regmap/regcache-rbtree.c
index 461cff8..11011ec 100644
--- a/drivers/base/regmap/regcache-rbtree.c
+++ b/drivers/base/regmap/regcache-rbtree.c
@@ -138,15 +138,20 @@ static int rbtree_show(struct seq_file *s, void *ignored)
struct regcache_rbtree_node *n;
struct rb_node *node;
unsigned int base, top;
+ size_t mem_size;
int nodes = 0;
int registers = 0;
int this_registers, average;

map->lock(map);

+ mem_size = sizeof(*rbtree_ctx);
+
for (node = rb_first(&rbtree_ctx->root); node != NULL;
node = rb_next(node)) {
n = container_of(node, struct regcache_rbtree_node, node);
+ mem_size += sizeof(*n);
+ mem_size += (n->blklen * map->cache_word_size);

regcache_rbtree_get_base_top_reg(map, n, &base, &top);
this_registers = ((top - base) / map->reg_stride) + 1;
@@ -161,8 +166,8 @@ static int rbtree_show(struct seq_file *s, void *ignored)
else
average = 0;

- seq_printf(s, "%d nodes, %d registers, average %d registers\n",
- nodes, registers, average);
+ seq_printf(s, "%d nodes, %d registers, average %d registers, memory overhead %zuB\n",
+ nodes, registers, average, mem_size);

map->unlock(map);

--
1.8.1.5


2013-03-12 18:12:51

by Mark Brown

[permalink] [raw]
Subject: Re: [PATCH v2] regmap: Expose total memory consumption in the rbtree debugfs entry

On Tue, Mar 12, 2013 at 05:26:49PM +0000, Dimitris Papastamos wrote:

> + mem_size = sizeof(*rbtree_ctx);
> +
> for (node = rb_first(&rbtree_ctx->root); node != NULL;
> node = rb_next(node)) {
> n = container_of(node, struct regcache_rbtree_node, node);
> + mem_size += sizeof(*n);
> + mem_size += (n->blklen * map->cache_word_size);

This appears to ignore the size of the node structure and only have the
root context and the data.


Attachments:
(No filename) (448.00 B)
signature.asc (836.00 B)
Digital signature
Download all attachments

2013-03-12 20:08:22

by Dimitris Papastamos

[permalink] [raw]
Subject: Re: [PATCH v2] regmap: Expose total memory consumption in the rbtree debugfs entry

On Tue, Mar 12, 2013 at 06:12:47PM +0000, Mark Brown wrote:
> On Tue, Mar 12, 2013 at 05:26:49PM +0000, Dimitris Papastamos wrote:
>
> > + mem_size = sizeof(*rbtree_ctx);
> > +
> > for (node = rb_first(&rbtree_ctx->root); node != NULL;
> > node = rb_next(node)) {
> > n = container_of(node, struct regcache_rbtree_node, node);
> > + mem_size += sizeof(*n);
> > + mem_size += (n->blklen * map->cache_word_size);
>
> This appears to ignore the size of the node structure and only have the
> root context and the data.

I've got mem_size += sizeof(*n) which is the size of the rbnode. That
contains the underlying rbtree node links.

Thanks,
Dimitris

2013-03-13 11:13:29

by Mark Brown

[permalink] [raw]
Subject: Re: [PATCH v2] regmap: Expose total memory consumption in the rbtree debugfs entry

On Tue, Mar 12, 2013 at 05:26:49PM +0000, Dimitris Papastamos wrote:
> Provide a feel of how much overhead the rbtree cache adds to
> the game.

Applied, with a tweak to the log to say "used X bytes" for clarity (it's
arguable if the cached data itself is overhead, depends on what you're
comparing to).


Attachments:
(No filename) (304.00 B)
signature.asc (836.00 B)
Digital signature
Download all attachments

2013-03-13 11:54:34

by Dimitris Papastamos

[permalink] [raw]
Subject: Re: [PATCH v2] regmap: Expose total memory consumption in the rbtree debugfs entry

On Wed, Mar 13, 2013 at 11:13:26AM +0000, Mark Brown wrote:
> On Tue, Mar 12, 2013 at 05:26:49PM +0000, Dimitris Papastamos wrote:
> > Provide a feel of how much overhead the rbtree cache adds to
> > the game.
>
> Applied, with a tweak to the log to say "used X bytes" for clarity (it's
> arguable if the cached data itself is overhead, depends on what you're
> comparing to).

Aha yes, it makes sense.

Thanks,
Dimitris