From: Andrzej Hajda <andrzej.hajda@intel.com>
To: Eric Dumazet <edumazet@google.com>
Cc: LKML <linux-kernel@vger.kernel.org>,
<intel-gfx@lists.freedesktop.org>,
<dri-devel@lists.freedesktop.org>,
netdev <netdev@vger.kernel.org>,
"Jani Nikula" <jani.nikula@linux.intel.com>,
Daniel Vetter <daniel@ffwll.ch>,
"Lucas De Marchi" <lucas.demarchi@intel.com>,
Chris Wilson <chris.p.wilson@intel.com>,
Dmitry Vyukov <dvyukov@google.com>,
"Jakub Kicinski" <kuba@kernel.org>
Subject: Re: [PATCH 2/9] lib/ref_tracker: compact stacktraces before printing
Date: Fri, 18 Feb 2022 11:54:54 +0100 [thread overview]
Message-ID: <199aebfb-f364-cd9b-5d2b-dbe42b779a41@intel.com> (raw)
In-Reply-To: <CANn89iKgzztLA3y6V+vw3RiyoScC3K=1Z1_gajj8H56wGWDw6A@mail.gmail.com>
On 17.02.2022 16:23, Eric Dumazet wrote:
> On Thu, Feb 17, 2022 at 6:05 AM Andrzej Hajda <andrzej.hajda@intel.com> wrote:
>> In cases references are taken alternately on multiple exec paths leak
>> report can grow substantially, sorting and grouping leaks by stack_handle
>> allows to compact it.
>>
>> Signed-off-by: Andrzej Hajda <andrzej.hajda@intel.com>
>> Reviewed-by: Chris Wilson <chris.p.wilson@intel.com>
>> ---
>> lib/ref_tracker.c | 35 +++++++++++++++++++++++++++--------
>> 1 file changed, 27 insertions(+), 8 deletions(-)
>>
>> diff --git a/lib/ref_tracker.c b/lib/ref_tracker.c
>> index 1b0c6d645d64a..0e9c7d2828ccb 100644
>> --- a/lib/ref_tracker.c
>> +++ b/lib/ref_tracker.c
>> @@ -1,5 +1,6 @@
>> // SPDX-License-Identifier: GPL-2.0-or-later
>> #include <linux/export.h>
>> +#include <linux/list_sort.h>
>> #include <linux/ref_tracker.h>
>> #include <linux/slab.h>
>> #include <linux/stacktrace.h>
>> @@ -14,23 +15,41 @@ struct ref_tracker {
>> depot_stack_handle_t free_stack_handle;
>> };
>>
>> +static int ref_tracker_cmp(void *priv, const struct list_head *a, const struct list_head *b)
>> +{
>> + const struct ref_tracker *ta = list_entry(a, const struct ref_tracker, head);
>> + const struct ref_tracker *tb = list_entry(b, const struct ref_tracker, head);
>> +
>> + return ta->alloc_stack_handle - tb->alloc_stack_handle;
>> +}
>> +
>> void __ref_tracker_dir_print(struct ref_tracker_dir *dir,
>> unsigned int display_limit)
>> {
>> + unsigned int i = 0, count = 0;
>> struct ref_tracker *tracker;
>> - unsigned int i = 0;
>> + depot_stack_handle_t stack;
>>
>> lockdep_assert_held(&dir->lock);
>>
>> + if (list_empty(&dir->list))
>> + return;
>> +
>> + list_sort(NULL, &dir->list, ref_tracker_cmp);
> What is going to be the cost of sorting a list with 1,000,000 items in it ?
Do we really have such cases?
>
> I just want to make sure we do not trade printing at most ~10 references
> (from netdev_wait_allrefs()) to a soft lockup :/ with no useful info
> if something went terribly wrong.
>
> I suggest that you do not sort a potential big list, and instead
> attempt to allocate an array of @display_limits 'struct stack_counts'
>
> I suspect @display_limits will always be kept to a reasonable value
> (less than 100 ?)
I though rather about 16 :)
In theory everything is possible, but do we have real case examples
which could lead to 100 stack traces?
Maybe some frameworks used by multiple consumers (drivers) ???
>
> struct stack_counts {
> depot_stack_handle_t stack_handle;
> unsigned int count;
> }
>
> Then, iterating the list and update the array (that you can keep
> sorted by ->stack_handle)
>
> Then after iterating, print the (at_most) @display_limits handles
> found in the temp array.
OK, could be faster and less invasive.
Other solution would be keeping the array in dir and update in every
tracker alloc/free, this way we avoid iteration over potentially big
list, but it would cost memory and since printing is rather rare I am
not sure if it is worth.
I will try your proposition.
Regards
Andrzej
>
>> +
>> list_for_each_entry(tracker, &dir->list, head) {
>> - if (i < display_limit) {
>> - pr_err("leaked reference.\n");
>> - if (tracker->alloc_stack_handle)
>> - stack_depot_print(tracker->alloc_stack_handle);
>> - i++;
>> - } else {
>> + if (i++ >= display_limit)
>> break;
>> - }
>> + if (!count++)
>> + stack = tracker->alloc_stack_handle;
>> + if (stack == tracker->alloc_stack_handle &&
>> + !list_is_last(&tracker->head, &dir->list))
>> + continue;
>> +
>> + pr_err("leaked %d references.\n", count);
>> + if (stack)
>> + stack_depot_print(stack);
>> + count = 0;
>> }
>> }
>> EXPORT_SYMBOL(__ref_tracker_dir_print);
>> --
>> 2.25.1
>>
next prev parent reply other threads:[~2022-02-18 10:55 UTC|newest]
Thread overview: 21+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-02-17 14:04 [PATCH 0/9] drm/i915: use ref_tracker library for tracking wakerefs Andrzej Hajda
2022-02-17 14:04 ` [PATCH 1/9] lib/ref_tracker: add unlocked leak print helper Andrzej Hajda
2022-02-17 14:04 ` [PATCH 2/9] lib/ref_tracker: compact stacktraces before printing Andrzej Hajda
2022-02-17 15:23 ` Eric Dumazet
2022-02-17 15:25 ` Eric Dumazet
2022-02-18 10:54 ` Andrzej Hajda [this message]
2022-02-18 13:01 ` Eric Dumazet
2022-02-17 14:04 ` [PATCH 3/9] lib/ref_tracker: __ref_tracker_dir_print improve printing Andrzej Hajda
2022-02-17 15:38 ` Eric Dumazet
2022-02-18 10:11 ` Andrzej Hajda
2022-02-17 14:04 ` [PATCH 4/9] lib/ref_tracker: add printing to memory buffer Andrzej Hajda
2022-02-17 14:04 ` [PATCH 5/9] lib/ref_tracker: improve allocation flags Andrzej Hajda
2022-02-17 15:13 ` Eric Dumazet
2022-02-18 10:28 ` Andrzej Hajda
2022-02-18 13:05 ` Eric Dumazet
2022-02-17 14:04 ` [PATCH 6/9] drm/i915: Separate wakeref tracking Andrzej Hajda
2022-02-17 14:48 ` [Intel-gfx] " Ville Syrjälä
2022-02-18 10:32 ` Andrzej Hajda
2022-02-17 14:04 ` [PATCH 7/9] drm/i915: Track leaked gt->wakerefs Andrzej Hajda
2022-02-17 14:04 ` [PATCH 8/9] drm/i915: Correct type of wakeref variable Andrzej Hajda
2022-02-17 14:04 ` [PATCH 9/9] drm/i915: replace Intel internal tracker with kernel core ref_tracker Andrzej Hajda
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=199aebfb-f364-cd9b-5d2b-dbe42b779a41@intel.com \
--to=andrzej.hajda@intel.com \
--cc=chris.p.wilson@intel.com \
--cc=daniel@ffwll.ch \
--cc=dri-devel@lists.freedesktop.org \
--cc=dvyukov@google.com \
--cc=edumazet@google.com \
--cc=intel-gfx@lists.freedesktop.org \
--cc=jani.nikula@linux.intel.com \
--cc=kuba@kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=lucas.demarchi@intel.com \
--cc=netdev@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