On 2011-05-24 06:31, Gilles Chanteperdrix wrote:
> On 05/23/2011 03:53 PM, Jan Kiszka wrote:
>> The following changes since commit aec30a2543afa18fa7832deee85e187b0faeb1f0:
>>
>> xeno-test: fix reference to @XENO_TEST_DIR@ (2011-05-15 21:20:41 +0200)
>>
>> are available in the git repository at:
>> git://git.xenomai.org/xenomai-jki.git for-upstream
>>
>> Jan Kiszka (1):
>> native: Fix msendq fastlock leakage
>>
>> include/native/task.h | 5 +++++
>> ksrc/skins/native/task.c | 13 ++++++-------
>> 2 files changed, 11 insertions(+), 7 deletions(-)
>>
>> ------8<------
>>
>> When a native task terminates without going through rt_task_delete, we
>> leaked the fastlock so far. Fix it by moving the release into the delete
>> hook. As the ppd is already invalid at that point, we have to save the
>> heap address in the task data structure.
>
> I Jan, I once worked on a patch to reverse the ppd cleanup order, in order
> to fix bugs of this kind. Here it comes. I do not remember why I did not
> commit it, but I guess it was not working well. Could we restart working
> from this patch?
>
> From 038ecf08cd66b3112e0fe277d71d294b8eb83bcc Mon Sep 17 00:00:00 2001
> From: Gilles Chanteperdrix <gilles.chanteperd...@xenomai.org>
> Date: Sun, 29 Aug 2010 14:52:08 +0200
> Subject: [PATCH] nucleus: reverse ppd cleanup order
>
> ---
> ksrc/nucleus/shadow.c | 11 ++++++-----
> 1 files changed, 6 insertions(+), 5 deletions(-)
>
> diff --git a/ksrc/nucleus/shadow.c b/ksrc/nucleus/shadow.c
> index b2d4326..725ae43 100644
> --- a/ksrc/nucleus/shadow.c
> +++ b/ksrc/nucleus/shadow.c
> @@ -556,7 +556,7 @@ static unsigned ppd_lookup_inner(xnqueue_t **pq,
> }
> while (holder &&
> (ppd->key.mm < pkey->mm ||
> - (ppd->key.mm == pkey->mm && ppd->key.muxid < pkey->muxid)));
> + (ppd->key.mm == pkey->mm && ppd->key.muxid > pkey->muxid)));
>
> if (ppd->key.mm == pkey->mm && ppd->key.muxid == pkey->muxid) {
> /* found it, return it. */
> @@ -566,7 +566,7 @@ static unsigned ppd_lookup_inner(xnqueue_t **pq,
>
> /* not found, return successor for insertion. */
> if (ppd->key.mm < pkey->mm ||
> - (ppd->key.mm == pkey->mm && ppd->key.muxid < pkey->muxid))
> + (ppd->key.mm == pkey->mm && ppd->key.muxid > pkey->muxid))
> *pholder = holder ? link2ppd(holder) : NULL;
> else
> *pholder = ppd;
> @@ -589,10 +589,11 @@ static int ppd_insert(xnshadow_ppd_t * holder)
> }
>
> inith(&holder->link);
> - if (next)
> + if (next) {
> insertq(q, &next->link, &holder->link);
> - else
> + } else {
> appendq(q, &holder->link);
> + }
> xnlock_put_irqrestore(&nklock, s);
>
> return 0;
> @@ -640,7 +641,7 @@ static inline void ppd_remove_mm(struct mm_struct *mm,
> xnqueue_t *q;
> spl_t s;
>
> - key.muxid = 0;
> + key.muxid = ~0UL;
> key.mm = mm;
> xnlock_get_irqsave(&nklock, s);
> ppd_lookup_inner(&q, &ppd, &key);
Unfortunately, that won't help. I think we are forced to clear
xnshadow_thrptd before calling into xnpod_delete_thread, but we would
need that for xnshadow_ppd_get (=>xnpod_userspace_p()).
Jan
--
Siemens AG, Corporate Technology, CT T DE IT 1
Corporate Competence Center Embedded Linux
_______________________________________________
Xenomai-core mailing list
Xenomai-core@gna.org
https://mail.gna.org/listinfo/xenomai-core
