From: Tim Pepper <lnxninja@linux.vnet.ibm.com>
To: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Al Viro <viro@ftp.linux.org.uk>,
Tim Pepper <lnxninja@linux.vnet.ibm.com>,
Ingo Molnar <mingo@redhat.com>,
linux-kernel@vger.kernel.org
Subject: Re: [PATCH] lockdep: Avoid /proc/lockdep & lock_stat infinite output
Date: Tue, 9 Oct 2007 15:10:32 -0700 [thread overview]
Message-ID: <20071009221031.GA28980@tpepper-t42p.dolavim.us> (raw)
In-Reply-To: <1191928489.6848.4.camel@twins>
On Tue 09 Oct at 13:14:49 +0200 a.p.zijlstra@chello.nl said:
> FWIW I had to do Tim's bits too. Just moving all output from the start
> into the show method didn't fix it.
Yes. The way the original lockdep_proc.c code was doing its pointers
around its seq data was definitely wrong, regardless of the output
outside of _show().
> -static void *l_start(struct seq_file *m, loff_t *pos)
> +static void *l_next(struct seq_file *m, void *v, loff_t *pos)
> {
> - struct lock_class *class = m->private;
> -
> - if (&class->lock_entry == all_lock_classes.next)
> - seq_printf(m, "all lock classes:\n");
> -
> - return class;
> + (*pos)++;
> + return l_start(m, pos);
> }
Isn't this is going to make l_start/l_next() approach O(n^2) when they can
be more like O(n)? It appears to be the case that _next() will always
get a valid *v, so you can just step immediately to the next element.
The _start() seems to be the only place where you'd actually need to
search based on *pos.
The below moves the headers out of the _start() functions, but by using
SEQ_START_TOKEN (as appears to be the trend in other seq_file users)
to differentiate. This means *pos==0 then is the header and *pos==1..n
are the lock class elements 0..(n-1), which again appears to be what
others are doing.
================================================================
Both /proc/lockdep and /proc/lock_stat output may loop infinitely.
When a read() requests an amount of data smaller than the amount of data
that the seq_file's foo_show() outputs, the output starts looping and
outputs the "stuck" element's data infinitely. There may be multiple
sequential calls to foo_start(), foo_next()/foo_show(), and foo_stop()
for a single open with sequential read of the file. The _start() does not
have to start with the 0th element and _show() might be called multiple
times in a row for the same element for a given open/read of the seq_file.
Also header output should not be happening in _start(). All output should
be in _show(), which SEQ_START_TOKEN is meant to help. Having output in
_start() may also negatively impact seq_file's seq_read() and traverse()
accounting.
Signed-off-by: Tim Pepper <lnxninja@linux.vnet.ibm.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Ingo Molnar <mingo@elte.hu>
Cc: Al Viro <viro@ftp.linux.org.uk>
---
Compared to my previous version this now also has output only happening
in _show(). Compared to Peter's version with output only in _show(),
this is more efficient in its _next().
--- linux-2.6.23-rc9.orig/kernel/lockdep_proc.c
+++ linux-2.6.23-rc9/kernel/lockdep_proc.c
@@ -25,28 +25,38 @@
static void *l_next(struct seq_file *m, void *v, loff_t *pos)
{
- struct lock_class *class = v;
+ struct lock_class *class;
(*pos)++;
+ if (v == SEQ_START_TOKEN)
+ class = m->private;
+ else {
+ class = v;
+
+ if (class->lock_entry.next != &all_lock_classes)
+ class = list_entry(class->lock_entry.next,
+ struct lock_class, lock_entry);
+ else
+ class = NULL;
+ }
- if (class->lock_entry.next != &all_lock_classes)
- class = list_entry(class->lock_entry.next, struct lock_class,
- lock_entry);
- else
- class = NULL;
- m->private = class;
return class;
}
static void *l_start(struct seq_file *m, loff_t *pos)
{
- struct lock_class *class = m->private;
+ struct lock_class *class;
+ loff_t i = 0;
- if (&class->lock_entry == all_lock_classes.next)
- seq_printf(m, "all lock classes:\n");
+ if (*pos == 0)
+ return SEQ_START_TOKEN;
+ list_for_each_entry(class, &all_lock_classes, lock_entry) {
+ if (++i == *pos)
+ return class;
+ }
+ return NULL;
- return class;
}
static void l_stop(struct seq_file *m, void *v)
@@ -101,10 +111,15 @@ static void print_name(struct seq_file *
static int l_show(struct seq_file *m, void *v)
{
unsigned long nr_forward_deps, nr_backward_deps;
- struct lock_class *class = m->private;
+ struct lock_class *class = v;
struct lock_list *entry;
char c1, c2, c3, c4;
+ if (v == SEQ_START_TOKEN) {
+ seq_printf(m, "all lock classes:\n");
+ return 0;
+ }
+
seq_printf(m, "%p", class->key);
#ifdef CONFIG_DEBUG_LOCKDEP
seq_printf(m, " OPS:%8ld", class->ops);
@@ -523,10 +538,11 @@ static void *ls_start(struct seq_file *m
{
struct lock_stat_seq *data = m->private;
- if (data->iter == data->stats)
- seq_header(m);
+ if (*pos == 0)
+ return SEQ_START_TOKEN;
- if (data->iter == data->iter_end)
+ data->iter = data->stats + *pos;
+ if (data->iter >= data->iter_end)
data->iter = NULL;
return data->iter;
@@ -538,8 +554,13 @@ static void *ls_next(struct seq_file *m,
(*pos)++;
+ if (v == SEQ_START_TOKEN)
+ data->iter = data->stats;
+ else {
+ data->iter = v;
+ data->iter++;
+ }
+
- data->iter = v;
- data->iter++;
if (data->iter == data->iter_end)
data->iter = NULL;
@@ -552,9 +573,11 @@ static void ls_stop(struct seq_file *m,
static int ls_show(struct seq_file *m, void *v)
{
- struct lock_stat_seq *data = m->private;
+ if (v == SEQ_START_TOKEN)
+ seq_header(m);
+ else
+ seq_stats(m, v);
- seq_stats(m, data->iter);
return 0;
}
prev parent reply other threads:[~2007-10-09 22:10 UTC|newest]
Thread overview: 5+ messages / expand[flat|nested] mbox.gz Atom feed top
2007-10-09 1:15 Tim Pepper
2007-10-09 1:30 ` Al Viro
2007-10-09 4:04 ` Tim Pepper
2007-10-09 11:14 ` Peter Zijlstra
2007-10-09 22:10 ` Tim Pepper [this message]
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=20071009221031.GA28980@tpepper-t42p.dolavim.us \
--to=lnxninja@linux.vnet.ibm.com \
--cc=a.p.zijlstra@chello.nl \
--cc=linux-kernel@vger.kernel.org \
--cc=mingo@redhat.com \
--cc=viro@ftp.linux.org.uk \
/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
all inboxes | Powered by JetHome®