Re: [PATCH][RT] Fix pushable_tasks list corruption

[Gregory Haskins <ghaskins@novell.com>]

[-- Attachment #1: Type: text/plain, Size: 1598 bytes --]

Hi Gilles

Gilles Carry wrote:
> From: gilles.carry <gilles.carry@bull.net>
>
> Symptoms:
> System hang (endless loop in plist_check_list) or BUG because
> of faulty prev/next pointers in pushable_task node.
>
>
> When push_rt_task successes finding a task to push away, it
> performs a double lock on the runqueues (local and target) but
> before getting both locks, it releases the local rq lock letting
> other cpus grab the task in between. (eg. pull_rt_task, timers...)
> When push_rt_task calls deactivate_task (which calls
> dequeue_pushable_task) the task may have already been removed
> from the pushable_tasks list by another cpu.
> Removing the node again corrupts the list.
>   

Hmm,  I was looking at this same area of the code earlier this week. 
The problem with your assessment is that find_lock_lowest_rq() already
accounts for the dropped-lock-migration and will return NULL if the task
was moved in the interim.  I suppose there could be some weird
circumstance where the task is moved away, and then moved back, but even
so plist_del() is supposed to be idempotent, so I dont see why an extra
dequeue_pushable itself would be a problem.

At this point I don't really *love* your patch because it seems to just
be plastering over the problem that the list is corrupted.  I do
appreciate that you are looking at this problem, however!  So thank you
for that and please keep it up.

I am on vacation every thursday+friday for a while, so I will not be
responsive until Monday.  Ill catch up with you guys then.  Have a good
weekend.

-Greg


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 257 bytes --]