From: David Jander <david.jander@protonic.nl>
To: Dimitris Papastamos <dp@opensource.wolfsonmicro.com>
Cc: Mark Brown <broonie@kernel.org>, linux-kernel@vger.kernel.org
Subject: Re: [PATCH] drivers: regmap: bugfix in regcache-rbtree.c
Date: Wed, 21 Aug 2013 15:14:23 +0200 [thread overview]
Message-ID: <20130821151423.70c38fb3@archvile> (raw)
In-Reply-To: <1377090155-3443-1-git-send-email-david@protonic.nl>
Hi Dimitris,
On Wed, 21 Aug 2013 15:02:35 +0200
David Jander <david@protonic.nl> wrote:
> The functionality of rbtree_ctx->cached_rbnode is broken. Remove it to
> avoid hitting the wrong rbnode when locating a register.
> rbnode register ranges can overlap, which is not a problem as long as
> every lookup for a register returns the same rbnode. Therefor we need
> to start searching from the top of the rb-tree _always_.
Some clarification to this patch. I want to be careful in "attacking" such a
core function that is used in quite a lot of places around the kernel, and it
strikes me that this has been gone unnoticed so far.
Here's the explanation:
1. If a driver initializes a regmap with a RB-tree cache, and starts writing to
registers in some arbitrary order, you might get overlapping rbnodes:
Suppose I have this rb-tree (a real case I happen to have):
# cat /sys/kernel/debug/regmap/1-000a/rbtree
2-19 (24)
4-1b (24)
20-37 (24)
22-39 (24)
3c-53 (24)
100-117 (24)
104-11c (25)
11e-135 (24)
8 nodes, 193 registers, average 24 registers, used 626 bytes
2. Suppose you just added node 20-37 by writing to previously uncached register
20. Now this node would be in rbtree_ctx->cached_rbnode, and would get search
first by the next access. Let's say that next access is register 30, then
rbtree_ctx->cached_rbnode would be a hit on the node 20-37.
Otherwise the hit would be in node 22-39 which is higher up the tree and
should be searched first. Two different possible hits for register 30 with
different contents! That cannot be right, right?
Therefor I dare to declare the rbtree_ctx->cached_rbnode functionality broken,
and removed it.
I stumbled upon this problem while debugging a non-functioning audio codec
(SGTL5000). This patch fixes the problem. Using a flat-cache in sgtl5000.c also
fixes it, so at least if I didn't fix this bug the right way, there sure is a
bug that needs to be fixed in regcache-rbtree.c.
Best regards,
--
David Jander
Protonic Holland.
next prev parent reply other threads:[~2013-08-21 13:41 UTC|newest]
Thread overview: 12+ messages / expand[flat|nested] mbox.gz Atom feed top
2013-08-21 13:02 [PATCH] drivers: regmap: bugfix in regcache-rbtree.c David Jander
2013-08-21 13:14 ` David Jander [this message]
2013-08-21 14:08 ` Mark Brown
2013-08-21 14:24 ` David Jander
2013-08-21 14:41 ` David Jander
2013-08-21 13:32 ` Mark Brown
2013-08-21 14:21 ` David Jander
2013-08-21 14:44 ` Mark Brown
2013-08-21 15:08 ` David Jander
2013-08-21 15:28 ` Mark Brown
2013-08-21 17:03 ` Dimitris Papastamos
2013-08-22 10:04 ` Mark Brown
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20130821151423.70c38fb3@archvile \
--to=david.jander@protonic.nl \
--cc=broonie@kernel.org \
--cc=dp@opensource.wolfsonmicro.com \
--cc=linux-kernel@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox