On Sun, 2011-05-15 at 17:26 +0200, Jan Kiszka wrote:
> On 2011-05-15 17:21, Philippe Gerum wrote:
> > On Sat, 2011-05-14 at 19:25 +0200, Gilles Chanteperdrix wrote:
> >> On 05/14/2011 07:21 PM, Gilles Chanteperdrix wrote:
> >>> On 05/14/2011 12:39 AM, Thomas Schaefer wrote:
> >>>>
> >>>> Hello,
> >>>>
> >>>> ----------------------------------------
> >>>>> Subject: Re: [Xenomai-help] Problems with rt pipes after upgrade
> >>>>> From: [email protected]
> >>>>> To: [email protected]
> >>>>> CC: [email protected]
> >>>>> Date: Fri, 13 May 2011 12:26:26 +0200
> >>>>>
> >>>>> On Wed, 2011-05-11 at 07:46 -0700, Thomas Schaefer wrote:
> >>>>>>
> >>>>>> Hi,
> >>>>>>
> >>>>>> we are currently running an earlier version of Xenomai (2.5.3) on a
> >>>>>> Quad core Xeon with kernel 2.6.31.8(X86_64).
> >>>>>> I was looking into upgrading to the latest Xenomai version from the
> >>>>>> git head and kernel 2.6.37.6.
> >>>>>> Everything seems to work well except creating rt_pipes in the kernel
> >>>>>> driver.
> >>>>>> First I thought this is a problem with our PCIe driver but I see the
> >>>>>> same problem when loading xeno_klat.
> >>>>>>
> >>>>>> # modprobe xeno_klat
> >>>>>> [ 70.511784] ------------[ cut here ]------------
> >>>>>> [ 70.516422] WARNING: at fs/proc/generic.c:860
> >>>>>> remove_proc_entry+0x25b/0x260()
> >>>>>> [ 70.523556] Hardware name: D-Mitri
> >>>>>> [ 70.526960] remove_proc_entry: removing non-empty directory
> >>>>>> 'native/pipes', leaking at least 'klat_pipe'
> >>>>>> [ 70.536433] Modules linked in: xeno_klat bonding [last unloaded:
> >>>>>> scsi_wait_scan]
> >>>>>> [ 70.545910] Pid: 296, comm: kworker/1:1 Not tainted 2.6.37.6 #4
> >>>>>> [ 70.551834] Call Trace:
> >>>>>> [ 70.554291] [] ? warn_slowpath_common+0x7b/0xc0
> >>>>>> [ 70.560484] [] ? warn_slowpath_fmt+0x45/0x50
> >>>>>> [ 70.566402] [] ? __xlate_proc_name+0x35/0xd0
> >>>>>> [ 70.572330] [] ? remove_proc_entry+0x25b/0x260
> >>>>>> [ 70.578434] [] ? registry_proc_callback+0x4b5/0x5c0
> >>>>>> [ 70.584963] [] ? registry_proc_callback+0x0/0x5c0
> >>>>>> [ 70.591330] [] ? process_one_work+0x107/0x3c0
> >>>>>> [ 70.597351] [] ? worker_thread+0x14c/0x410
> >>>>>> [ 70.603106] [] ? worker_thread+0x0/0x410
> >>>>>> [ 70.608678] [] ? kthread+0x95/0xa0
> >>>>>> [ 70.613758] [] ? kernel_thread_helper+0x4/0x10
> >>>>>> [ 70.619857] [] ? kthread+0x0/0xa0
> >>>>>> [ 70.624840] [] ? kernel_thread_helper+0x0/0x10
> >>>>>> [ 70.630941] ---[ end trace 793ec26c5b485748 ]---
> >>>>>>
> >>>>>>
> >>>>>> I would appreciate if someone could point me in the right direction of
> >>>>>> the possible cause.
> >>>>>
> >>>>> This very much looks like a race in the registry support. We have to
> >>>>> resync registry object deletion done from primary mode with the linux
> >>>>> kernel actually doing the work for us (via /procfs), and somehow, we
> >>>>> don't do this right. This is not a critical issue and will likely not
> >>>>> break your platform, but this is still quite ugly, and requires a fix.
> >>>>>
> >>>>> Could you give us some hints about how to reproduce this easily? TIA,
> >>>>>
> >>>>
> >>>> I used VBox and installed a very basic Debian squeeze with the
> >>>> netinst.iso.
> >>>> Installed our kernel into it from the debian package I created and
> >>>> copied xeno_klat.ko to it.
> >>>> Running insmod xeno_klat.ko shows the same symptom.
> >>>
> >>> xeno_klat insmods here without any problem. You probably have some
> >>> debugging option enabled. Could your post the .config of the kernel you
> >>> use?
> >>>
> >>
> >> I enabled approximately all debugging options. Now I reproduce this bug.
> >>
> >>
> >
> > Ok, I introduced this one. Two bugs in one actually, here are the fixes:
> >
> > diff --git a/ksrc/nucleus/vfile.c b/ksrc/nucleus/vfile.c
> > index d8ac398..6aa42b3 100644
> > --- a/ksrc/nucleus/vfile.c
> > +++ b/ksrc/nucleus/vfile.c
> > @@ -752,8 +752,10 @@ int xnvfile_init_link(const char *from,
> >
> > ppde = parent->entry.pde;
> > pde = proc_symlink(from, ppde, to);
> > - if (vlink->entry.pde == NULL)
> > + if (pde == NULL) {
> > + remove_proc_entry(pde->name, ppde);
>
> Sure? :)
This one should be better.
diff --git a/ksrc/nucleus/vfile.c b/ksrc/nucleus/vfile.c
index d8ac398..4898e86 100644
--- a/ksrc/nucleus/vfile.c
+++ b/ksrc/nucleus/vfile.c
@@ -752,7 +752,7 @@ int xnvfile_init_link(const char *from,
ppde = parent->entry.pde;
pde = proc_symlink(from, ppde, to);
- if (vlink->entry.pde == NULL)
+ if (pde == NULL)
return -ENOMEM;
vlink->entry.parent = parent;
>
> Jan
>
> > return -ENOMEM;
> > + }
> >
> > vlink->entry.parent = parent;
> > vlink->entry.pde = pde;
> >
>
--
Philippe.
_______________________________________________
Xenomai-help mailing list
[email protected]
https://mail.gna.org/listinfo/xenomai-help
