or1k test was .. Re: [PATCH] or1k: New cache manager.
Hesham Moustafa
heshamelmatary at gmail.com
Wed Sep 17 17:44:12 UTC 2014
On Tue, Sep 16, 2014 at 11:08 PM, Joel Sherrill <joel.sherrill at oarcorp.com>
wrote:
> Gedare.. cc'ed you for help in spotting an empty rbtree
> in gdb. See below.
> On 9/16/2014 2:45 PM, Hesham Moustafa wrote:
> > Breakpoint 2, 0x00000600 in _unalign ()
> > (gdb) bt
> > #0 0x00000600 in _unalign ()
> > #1 0x0002ec4c in _RBTree_Next (
> > node=0x40890, dir=RBT_RIGHT)
> > at
> ../../../../../../rtems/c/src/../../cpukit/score/src/rbtreenext.c:35
> > #2 0x0002e2f4 in _RBTree_Successor (
> > node=0x40890)
> > at ../../cpukit/../../../or1ksim/lib/include/rtems/score/rbtree.h:512
> > #3 0x0002e8c0 in _RBTree_Extract (
> > the_rbtree=0x4198c,
> > the_node=0x40890)
> > at
> ../../../../../../rtems/c/src/../../cpukit/score/src/rbtreeextract.c:106
> > #4 0x00021524 in _RBTree_Get (
> > the_rbtree=0x4198c, dir=RBT_LEFT)
> > at ../../cpukit/../../../or1ksim/lib/include/rtems/score/rbtree.h:540
> > #5 0x000215c8 in _Thread_queue_Dequeue
> > (the_thread_queue=0x4198c)
> > ---Type <return> to continue, or q <return> to quit---
> > at
> ../../../../../../rtems/c/src/../../cpukit/score/src/threadqdequeue.c:51
> > #6 0x00017c14 in _CORE_semaphore_Surrender (the_semaphore=0x4198c,
> > id=436273153,
> > api_semaphore_mp_support=0x0)
> > at
> ../../../../../../rtems/c/src/../../cpukit/score/src/coresemsurrender.c:37
> > #7 0x00014868 in rtems_semaphore_release (id=436273153)
> > at
> ../../../../../../rtems/c/src/../../cpukit/rtems/src/semrelease.c:102
> > #8 0x00026cfc in rtems_libio_unlock ()
> > at ../../cpukit/../../../or1ksim/lib/include/rtems/libio_.h:253
> > #9 0x00026d5c in rtems_filesystem_default_unlock (mt_entry=0x49ce0)
> > at
> ../../../../../../rtems/c/src/../../cpukit/libfs/src/defaults/default_loc---Type
> > <return> to continue, or q <return> to quit---
> > k_and_unlock.c:39
> > #10 0x0002920c in rtems_filesystem_instance_unlock (loc=0x49c5c)
> > at ../../cpukit/../../../or1ksim/lib/include/rtems/libio_.h:292
> > #11 0x00029268 in rtems_filesystem_location_free (loc=0x49c5c)
> > at
> ../../../../../../rtems/c/src/../../cpukit/libcsupport/src/freenode.c:29
> > #12 0x00029734 in rtems_libio_free (
> > iop=0x49c50)
> > at
> ../../../../../../rtems/c/src/../../cpukit/libcsupport/src/libio.c:136
> > #13 0x0002912c in close (fd=0)
> > at
> ../../../../../../rtems/c/src/../../cpukit/libcsupport/src/close.c:38
> > #14 0x000064b0 in rtems_libio_exit ()
> > at
> ../../../../../../rtems/c/src/../../cpukit/libcsupport/src/libio_exit.c:31
> > ---Type <return> to continue, or q <return> to quit---
> > #15 0x0003b058 in _exit (status=0)
> > at
> ../../../../../../rtems/c/src/../../cpukit/libcsupport/src/newlibc_exit.c:46
> > #16 0x00034798 in exit (code=0)
> > at ../../../../../gcc-4.8.3/newlib/libc/stdlib/exit.c:70
> > #17 0x00002e3c in Test_task (unused=1)
> > at
> ../../../../../../../rtems/c/src/../../testsuites/samples/ticker/tasks.c:41
> > #18 0x000340f0 in _Thread_Handler ()
> > at
> ../../../../../../rtems/c/src/../../cpukit/score/src/threadhandler.c:192
> > #19 0x00034078 in _User_extensions_Thread_exitted (executing=0x40890)
> > at
> ../../cpukit/../../../or1ksim/lib/include/rtems/score/userextimpl.h:243
> > Backtrace stopped: frame did not save the PC
> > (gdb)
> >
> >
> > It breaks at _RBTree_Next specifically at the following line:
> > while ( ( current = current->child[ opp_dir ] ) != NULL )
> >
> > (gdb) p current->child[ opp_dir ]
> > Cannot access memory at address 0xa010006
> > (gdb) p current
> > $1 = (RBTree_Node *) 0xa010002
> These look like object ids.
> > This address is invalid, the current memory length should be only 32
> > MB (0x2000000)
> >
> >
> http://git.rtems.org/rtems/tree/c/src/lib/libbsp/or1k/or1ksim/startup/linkcmds#n20
> >
> > So I guest current->child is overwritten somehow?
> Yep. Two approaches.
>
> + Set a watchpoint in gdb if it is supported. But even if supported,
> it will likely slow the run tremendously.
>
There is no HW watchpoint supported.
> + Break selectively and more or less binary search for where it is
> overwritten. I would break at the first call to _ISR_Dispatch
> (or whatever you called it) and see if it gets clobbered.
>
> That could be clobbered VERY early in the program. It could be
> a blown stack. But it could just be a stray write. Check the value
> of that semaphore's rbtree when you get to Init and just
> break periodically and see where it is corrupt.
>
> That's what I did. As you assumed, it's clobbered very early.
Breakpoint 1, _Objects_Extend_information (
information=0x3e26c <_RTEMS_tasks_Information>)
at
../../../../../../rtems/c/src/../../cpukit/score/src/objectextendinformation.c:67
67 do_extend = true;
(gdb) bt
#0 _Objects_Extend_information (
information=0x3e26c <_RTEMS_tasks_Information>)
at
../../../../../../rtems/c/src/../../cpukit/score/src/objectextendinformation.c:67
#1 0x0001b554 in _Objects_Initialize_information (
information=0x3e26c <_RTEMS_tasks_Information>,
the_api=OBJECTS_CLASSIC_API, the_class=1, maximum=4,
size=1424, is_string=false, maximum_name_length=4)
at
../../../../../../rtems/c/src/../../cpukit/score/src/objectinitializeinformation.c:126
#2 0x0002c688 in _RTEMS_tasks_Manager_initialization ()
at ../../../../../../rtems/c/src/../../cpukit/rtems/src/tasks.c:197
#3 0x00015bd4 in _RTEMS_API_Initialize ()
at ../../../../../../rtems/c/src/../../cpukit/sapi/src/rtemsapi.c:59
#4 0x0001590c in rtems_initialize_data_structures ()
at ../../../../../../rtems/c/src/../../cpukit/sapi/src/exinit.c:140
#5 0x0000333c in boot_card (cmdline=0x0)
---Type <return> to continue, or q <return> to quit---
at
../../../../../../../../rtems/c/src/lib/libbsp/or1k/or1ksim/../../shared/bootcard.c:92
#6 0x00000000 in ?? ()
(gdb)
Specifically, here
http://git.rtems.org/rtems/tree/cpukit/score/src/objectextendinformation.c#n261
I cc'ed Gedare because I don't know how to spot that the rbtree
> is empty in gdb.
>
> You need to see where that memory is overwritten.
>
> Again running all tests with the simulator clock tick could
> eliminate the ISR code as the culprit. :)
> > On Tue, Sep 16, 2014 at 9:21 PM, Joel Sherrill
> > <joel.sherrill at oarcorp.com> wrote:
> >> On 9/16/2014 2:17 PM, Hesham Moustafa wrote:
> >>> On Tue, Sep 16, 2014 at 8:42 PM, Joel Sherrill
> >>> <joel.sherrill at oarcorp.com> wrote:
> >>>>
> >>>> On 9/16/2014 1:34 PM, Hesham Moustafa wrote:
> >>>>> On Tue, Sep 16, 2014 at 8:15 PM, Joel Sherrill
> >>>>> <joel.sherrill at oarcorp.com> wrote:
> >>>>>> On 9/16/2014 12:54 PM, Hesham Moustafa wrote:
> >>>>>>> Hi
> >>>>>>>
> >>>>>>> On Tue, Sep 16, 2014 at 7:47 PM, Joel Sherrill
> >>>>>>> <joel.sherrill at oarcorp.com> wrote:
> >>>>>>>> I don't understand this but I got it applied.
> >>>>>>>>
> >>>>>>>> I manually edited the saved email to delete the preinstall.am
> >>>>>>>> changes. I committed the rest. Then I ran bootstrap -p myself
> >>>>>>>> and folded that into the rest of your patch.
> >>>>>>>>
> >>>>>>>> It should all be committed now.
> >>>>>>>>
> >>>>>>> Thanks for doing this, me too do not know what's wrong. BTW,
> commits
> >>>>>>> are not mirrored on github since 4 days ago.
> >>>>>>>
> >>>>>>>> How about some new test results. :)
> >>>>>>>>
> >>>>>>> I did run one last night, no big progress since previous results
> :( Is
> >>>>>>> there any tool, script, utility program or whatever that I can use
> to
> >>>>>>> detect wrong memory access (i.e, stack overwrite, heap corruption,
> >>>>>>> access to another task context)? I tried to add
> -fstack-protector-all
> >>>>>>> to gcc, but QEMU did not get anything or core-dump, ticker just
> hangs.
> >>>>>> I haven't checked into how gcc does its stack overwrite protection.
> >>>>>>
> >>>>>> The tests by themselves don't have these problems. The first
> >>>>>> possible source is incorrect layout of sections to memory by
> >>>>>> the linker script. There is some debug code in boot
> >>>>>>
> >>>>>> There used to be debug printk's in bspgetworkarea.c so you
> >>>>>> could check if areas overlapped. That usually causes bad issues
> >>>>>> though. Let's go through some basics:
> >>>>>>
> >>>>>> + Does hello world run and exit cleanly?
> >>>>>>
> >>>>> The output of Hello World is:
> >>>>>
> >>>>> *** BEGIN OF TEST HELLO WORLD ***
> >>>>> Hello World
> >>>>> *** END OF TEST HELLO WORLD ***
> >>>>> Fatal Error 5.0 Halted
> >>>>>
> >>>>> From GDB:
> >>>>>
> >>>>> Breakpoint 1, _Terminate (
> >>>>> the_source=RTEMS_FATAL_SOURCE_EXIT, is_internal=false,
> >>>>> the_error=0)
> >>>>> at
> >>>>> ../../../../../../rtems/c/src/../../cpukit/score/src/interr.c:39
> >>>>> 39 _ISR_Disable_without_giant( level );
> >>>>> (gdb) bt
> >>>>> #0 _Terminate (
> >>>>> the_source=RTEMS_FATAL_SOURCE_EXIT, is_internal=false,
> >>>>> the_error=0)
> >>>>> at
> >>>>> ../../../../../../rtems/c/src/../../cpukit/score/src/interr.c:39
> >>>>> #1 0x0003b5f8 in rtems_shutdown_executive (result=0)
> >>>>> at
> >>>>> ../../../../../../rtems/c/src/../../cpukit/sapi/src/exshutdown.c:21
> >>>>> #2 0x0003b350 in _exit (status=0)
> >>>>> at
> >>>>>
> >>>>>
> ../../../../../../rtems/c/src/../../cpukit/libcsupport/src/newlibc_exit.c:47
> >>>>> #3 0x0002cc30 in exit (code=0)
> >>>>> at ../../../../../gcc-4.8.3/newlib/libc/stdlib/exit.c:70
> >>>>> #4 0x00002184 in Init (ignored=253816)
> >>>>> at
> >>>>>
> >>>>>
> ../../../../../../../rtems/c/src/../../testsuites/samples/hello/init.c:33
> >>>>> ---Type <return> to continue, or q <return> to quit---
> >>>>> #5 0x0002c5b8 in _Thread_Handler ()
> >>>>> at
> >>>>>
> ../../../../../../rtems/c/src/../../cpukit/score/src/threadhandler.c:192
> >>>>> #6 0x0002c540 in _User_extensions_Thread_exitted (executing=0x40080)
> >>>>> at
> >>>>>
> ../../cpukit/../../../or1ksim/lib/include/rtems/score/userextimpl.h:243
> >>>> This is normal and OK. Look at the arguments to _Terminate.
> >>>>>> + How far does ticker get?
> >>>>>>
> >>>>> Ticker goes to the end:
> >>>>>
> >>>>> *** BEGIN OF TEST CLOCK TICK ***
> >>>>> TA1 - rtems_clock_get_tod - 09:00:00 12/31/1988
> >>>>> TA2 - rtems_clock_get_tod - 09:00:00 12/31/1988
> >>>>> TA3 - rtems_clock_get_tod - 09:00:00 12/31/1988
> >>>>> TA1 - rtems_clock_get_tod - 09:00:05 12/31/1988
> >>>>> TA2 - rtems_clock_get_tod - 09:00:10 12/31/1988
> >>>>> TA1 - rtems_clock_get_tod - 09:00:10 12/31/1988
> >>>>> TA3 - rtems_clock_get_tod - 09:00:15 12/31/1988
> >>>>> TA1 - rtems_clock_get_tod - 09:00:15 12/31/1988
> >>>>> TA2 - rtems_clock_get_tod - 09:00:20 12/31/1988
> >>>>> TA1 - rtems_clock_get_tod - 09:00:20 12/31/1988
> >>>>> TA1 - rtems_clock_get_tod - 09:00:25 12/31/1988
> >>>>> TA3 - rtems_clock_get_tod - 09:00:30 12/31/1988
> >>>>> TA2 - rtems_clock_get_tod - 09:00:30 12/31/1988
> >>>>> TA1 - rtems_clock_get_tod - 09:00:30 12/31/1988
> >>>>> *** END OF TEST CLOCK TICK ***
> >>>>> Fatal Error 9.276564 Halted
> >>>>>
> >>>>> From GDB:
> >>>>>
> >>>>> (gdb) break _Terminate
> >>>>> Breakpoint 1 at 0x19a68: file
> >>>>> ../../../../../../rtems/c/src/../../cpukit/score/src/interr.c, line
> >>>>> 39.
> >>>>> (gdb) break _OR1K_Exception_default
> >>>>> Breakpoint 2 at 0x2686c: file
> >>>>>
> >>>>>
> >>>>>
> ../../../../../../../../rtems/c/src/../../cpukit/score/cpu/or1k/or1k-exception-default.c,
> >>>>> line 22.
> >>>>> (gdb) c
> >>>>> The program is not being run.
> >>>>> (gdb) target remote :50001
> >>>>> Remote debugging using :50001
> >>>>> 0x00000100 in _reset ()
> >>>>> (gdb) c
> >>>>> Continuing.
> >>>>>
> >>>>> Breakpoint 2, _OR1K_Exception_default (vector=6, frame=0x43854) at
> >>>>>
> >>>>>
> >>>>>
> ../../../../../../../../rtems/c/src/../../cpukit/score/cpu/or1k/or1k-exception-default.c:22
> >>>>> 22 rtems_fatal( RTEMS_FATAL_SOURCE_EXCEPTION, (rtems_fatal_code)
> frame
> >>>>> );
> >>>>> (gdb) bt
> >>>>> #0 _OR1K_Exception_default (vector=6, frame=0x43854) at
> >>>>>
> >>>>>
> >>>>>
> ../../../../../../../../rtems/c/src/../../cpukit/score/cpu/or1k/or1k-exception-default.c:22
> >>>>> #1 0x00026980 in jump_to_c_handler ()
> >>>>> Backtrace stopped: frame did not save the PC
> >>>>>
> >>>>> vector 6 is _unalign exception.
> >>>> Set a break point at exit() (I think) and rtems_shutdown_executive().
> You
> >>>> could start in the task which calls whatever kicks off the shutdown
> >>>> sequence.
> >>>> It looks like something in the shutdown procedure trips over
> something.
> >>>> This might be easy to debug.
> >>>>
> >>> I did add just a function call to _CPU_Exception_frame_print(frame);
> >>> from _OR1K_Exception_default(uint32_t vector, CPU_Exception_frame
> >>> *frame)
> >>> And ticker exits normally without even entering
> >>> _OR1K_Exception_defaul as it did before. This is very weird. Does this
> >>> mean that some areas of the code are overlapped from the linker
> >>> script?
> >> I doubt it. I suspect something unitialized or not aligned properly.
> >>
> >> Set a breakpoint at
> >> http://git.rtems.org/rtems/tree/testsuites/samples/ticker/tasks.c#n40
> >> next over the print and then step through rtems_test_exit() and see
> >> where it faults.
> >>>> If the fault address is in the exception data, you can map that back
> to
> >>>> the
> >>>> nm file and see what file that was in, then that might help.
> >>>>>> + Have you tried the trick I suggested earlier to disable the
> >>>>>> real clock tick driver, use the simulator idle tick code, and
> >>>>>> disable all the tests that are known to fail that way. This
> >>>>>> will eliminate the ISR code as an issue because you won't
> >>>>>> have any (if console output if polled). See h8sim for
> >>>>>> an example. Should be a Makefile.am change, adding
> >>>>>> an include to the testsuite configuration file, building
> >>>>>> and running.
> >>>>>>
> >> --
> >> Joel Sherrill, Ph.D. Director of Research & Development
> >> joel.sherrill at OARcorp.com On-Line Applications Research
> >> Ask me about RTEMS: a free RTOS Huntsville AL 35805
> >> Support Available (256) 722-9985
> >>
>
> --
> Joel Sherrill, Ph.D. Director of Research & Development
> joel.sherrill at OARcorp.com On-Line Applications Research
> Ask me about RTEMS: a free RTOS Huntsville AL 35805
> Support Available (256) 722-9985
>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rtems.org/pipermail/devel/attachments/20140917/4c420d64/attachment-0002.html>
More information about the devel
mailing list