[ofw] RE: Assert in checked ibbus driver?
Tzachi Dar
tzachid at mellanox.co.il
Mon Jun 23 00:54:05 PDT 2008
Checked in (with minor changes) at revision 1295.
Thanks
Tzachi
________________________________
From: ofw-bounces at lists.openfabrics.org
[mailto:ofw-bounces at lists.openfabrics.org] On Behalf Of James Yang
Sent: Friday, June 20, 2008 1:38 AM
To: Tzachi Dar; Sean Hefty; Fab Tillier;
ofw at lists.openfabrics.org
Subject: RE: [ofw] RE: Assert in checked ibbus driver?
It seems it's a false alarm because p_list is not reset when an
item is removed. One more line is needed "p_list_item->p_list = NULL;"
in function __cl_primitive_remove().
Can this line be added to the svn?
Thanks,
James
CL_INLINE void CL_API
__cl_primitive_remove(
IN cl_list_item_t* const p_list_item )
{
/* CL_ASSERT that a non-null pointer is provided. */
CL_ASSERT( p_list_item );
CL_ASSERT( p_list_item->p_next );
CL_ASSERT( p_list_item->p_prev );
/* set the back pointer */
p_list_item->p_next->p_prev= p_list_item->p_prev;
/* set the next pointer */
p_list_item->p_prev->p_next= p_list_item->p_next;
/* if we're debugging, spruce up the pointers to help find
bugs */
#if defined( _DEBUG_ )
if( p_list_item != p_list_item->p_next )
{
p_list_item->p_next = NULL;
p_list_item->p_prev = NULL;
p_list_item->p_list = NULL;
}
#endif /* defined( _DEBUG_ ) */
}
________________________________
From: Tzachi Dar [mailto:tzachid at mellanox.co.il]
Sent: Thursday, June 19, 2008 12:07 PM
To: Sean Hefty; James Yang; Fab Tillier;
ofw at lists.openfabrics.org
Subject: RE: [ofw] RE: Assert in checked ibbus driver?
From a fast reading of the file CL_qlist.h it seems that all
"remove functions" clear the p_list_item->p_list field. So I guess that
most chances that this is a real bug. Can you try and add a printing to
each time you add or remove an object from the gp_mem_tracker list?
Thanks
Tzachi
I can see item's p_list pointer being cleared when it's
removed from the list. My guess is p_list is left over there when items
are allocated from free memory. I'm not familiar with the code enough to
find all these items and clear them during initialization. It's odd to
see p_list value was there but not p_prev and p_next.
If I ignore the assertion, it will be OK. However the
assert will come once again, especially during startup and shutdown.
Thanks,
James
________________________________
From: Fab Tillier
[mailto:ftillier at windows.microsoft.com]
Sent: Thursday, June 19, 2008 9:11 AM
To: James Yang; ofw at lists.openfabrics.org
Subject: RE: Assert in checked ibbus driver?
Hi James,
If you clear the p_list pointer does the assert go away?
-Fab
From: ofw-bounces at lists.openfabrics.org
[mailto:ofw-bounces at lists.openfabrics.org] On Behalf Of James Yang
Sent: Wednesday, June 18, 2008 6:05 PM
To: ofw at lists.openfabrics.org
Subject: [ofw] Assert in checked ibbus driver?
Can someone shed some light on the following assert I
saw in checked build driver?
It happened in the function cl_qlist_insert_head() when
a new item is added to the list. And the assertion happens because it
thinks the item is already in the list. However when checking the list
item, I don't see it is linked because its previous and next pointers
are both NULL. The item is also is not in the list by manually checking
the list.
Is this because the item is not initialized properly as
it's said in the comments, or it's a true bug?
Thanks,
James
----------------------
CL_INLINE void CL_API
cl_qlist_insert_head(
IN cl_qlist_t* const p_list,
IN cl_list_item_t* const p_list_item )
{
/* CL_ASSERT that a non-null pointer is provided.
*/
CL_ASSERT( p_list );
/* CL_ASSERT that a non-null pointer is provided.
*/
CL_ASSERT( p_list_item );
/* CL_ASSERT that the list was initialized. */
CL_ASSERT( p_list->state == CL_INITIALIZED );
/*
* The list item must not already be part of the
list. Note that this
* assertion may fail if an uninitialized list
item happens to have its
* list pointer equal to the specified list. The
chances of this
* happening are acceptable in light of the value
of this check.
*/
CL_ASSERT( p_list_item->p_list != p_list );
#if defined( _DEBUG_ )
p_list_item->p_list = p_list;
#endif
/* Insert before the head. */
__cl_primitive_insert( cl_qlist_head( p_list ),
p_list_item );
p_list->count++;
}
1: kd> k
ChildEBP RetAddr
f795e9fc 80873922 nt!DbgBreakPoint
f795ece4 f71588c4 nt!RtlAssert+0xba
f795ed0c f715a5df ibbus!cl_qlist_insert_head+0xb4
[k:\windows-openib\inc\complib\cl_qlist.h @ 782]
f795ed24 f71b08da ibbus!__cl_free_trk+0x6f
[k:\windows-openib\\core\complib\cl_memory.c @ 342]
f795ed38 f7218e8d ibbus!__pnp_reg_free+0xfa
[k:\windows-openib\\core\al\kernel\al_pnp.c @ 386]
f795ed70 f715c0e4 ibbus!async_destroy_cb+0x72d
[k:\windows-openib\\core\al\al_common.c @ 685]
f795ed8c f7162534 ibbus!__cl_async_proc_worker+0x94
[k:\windows-openib\core\complib\cl_async_proc.c @ 153]
f795eda0 f71645ec ibbus!__cl_thread_pool_routine+0x54
[k:\windows-openib\core\complib\cl_threadpool.c @ 67]
f795edac 80949b7c ibbus!__thread_callback+0x2c
[k:\windows-openib\core\complib\kernel\cl_thread.c @ 49]
f795eddc 8088e062 nt!PspSystemThreadStartup+0x2e
00000000 00000000 nt!KiThreadStartup+0x16
1: kd> dt 0x863416a0 _cl_qlist
+0x000 end : _cl_list_item
+0x00c count : 0x29
+0x010 state : 2 ( CL_INITIALIZED )
1: kd> dt 0x85485268 _cl_list_item
+0x000 p_next : (null)
+0x004 p_prev : (null)
+0x008 p_list : 0x863416a0 _cl_qlist
P Please consider the environment before printing this e-mail
or its attachment(s)
Disclaimer added by CodeTwo Exchange Rules
www.codetwo.com
--
Tzachi Dar
Mellanox Technologies LTD.
SW- Windows
Phone: +972 (4) 909 7200 (ext 271)
Mobile: +972 (57) 741 1269
E-mail: tzachid at mellanox.co.il
----------------------------------------------------------------------
Emails belong on computers, trees belong in forests; if you must print
this, do it on recycled paper.
http://www.greenpeace.org/international/
----------------------------------------------------------------------
Disclaimer added by CodeTwo Exchange Rules
http://www.codetwo.com
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/ofw/attachments/20080623/82839876/attachment.html>
More information about the ofw
mailing list