Optimization issue in RISC-V BSP
Denis Obrezkov
denisobrezkov at gmail.com
Sun Jul 30 00:03:13 UTC 2017
2017-07-30 1:35 GMT+02:00 Denis Obrezkov <denisobrezkov at gmail.com>:
> 2017-07-29 19:14 GMT+02:00 Joel Sherrill <joel at rtems.org>:
>
>>
>>
>> On Jul 29, 2017 4:04 AM, "Denis Obrezkov" <denisobrezkov at gmail.com>
>> wrote:
>>
>> 2017-07-29 3:45 GMT+02:00 Joel Sherrill <joel at rtems.org>:
>>
>>>
>>>
>>> On Jul 28, 2017 7:11 PM, "Denis Obrezkov" <denisobrezkov at gmail.com>
>>> wrote:
>>>
>>> 2017-07-29 1:41 GMT+02:00 Joel Sherrill <joel at rtems.org>:
>>>
>>>>
>>>>
>>>> On Jul 28, 2017 6:39 PM, "Denis Obrezkov" <denisobrezkov at gmail.com>
>>>> wrote:
>>>>
>>>> 2017-07-29 1:28 GMT+02:00 Joel Sherrill <joel at rtems.org>:
>>>>
>>>>>
>>>>>
>>>>> On Jul 28, 2017 6:14 PM, "Denis Obrezkov" <denisobrezkov at gmail.com>
>>>>> wrote:
>>>>>
>>>>> 2017-07-29 0:57 GMT+02:00 Joel Sherrill <joel at rtems.org>:
>>>>>
>>>>>>
>>>>>>
>>>>>> On Jul 28, 2017 5:55 PM, "Denis Obrezkov" <denisobrezkov at gmail.com>
>>>>>> wrote:
>>>>>>
>>>>>> 2017-07-28 22:36 GMT+02:00 Joel Sherrill <joel at rtems.org>:
>>>>>>
>>>>>>> Can you check the memory immediately after a download'?
>>>>>>>
>>>>>>> Then after the loop that copies initialized data into place?
>>>>>>>
>>>>>>> I suspect something off there. Could be a linker script issue or the
>>>>>>> copy gone crazy.
>>>>>>>
>>>>>>> --joel
>>>>>>>
>>>>>>> On Fri, Jul 28, 2017 at 3:20 PM, Denis Obrezkov <
>>>>>>> denisobrezkov at gmail.com> wrote:
>>>>>>>
>>>>>>>> 2017-07-28 22:16 GMT+02:00 Joel Sherrill <joel at rtems.org>:
>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> On Fri, Jul 28, 2017 at 2:50 PM, Denis Obrezkov <
>>>>>>>>> denisobrezkov at gmail.com> wrote:
>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>>> I can see that during task initialization I have a call:
>>>>>>>>>>>> _Thread_Initialize_information (information=information at entry=0x80000ad4
>>>>>>>>>>>> <_RTEMS_tasks_Information>, the_api=the_api at entry=OBJECTS_CLASSIC_API,
>>>>>>>>>>>> the_class=the_class at entry=1, maximum=124,
>>>>>>>>>>>> is_string=is_string at entry=false,
>>>>>>>>>>>> maximum_name_length=maximum_name_length at entry=4)
>>>>>>>>>>>>
>>>>>>>>>>>> And maximum is 124, but I have a configuration parameter:
>>>>>>>>>>>> #define CONFIGURE_MAXIMUM_TASKS 4
>>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> I can't imagine any standard RTEMS test configuring that many
>>>>>>>>>>> tasks.
>>>>>>>>>>> Is there a data corruption issue?
>>>>>>>>>>>
>>>>>>>>>>> 124 = 0x7c which doesn't ring any bells for me on odd memory
>>>>>>>>>>> issues.
>>>>>>>>>>>
>>>>>>>>>>> What is the contents of "Configuration_RTEMS_API"?
>>>>>>>>>>>
>>>>>>>>>> Oh, I change my configuration options a bit, they are:
>>>>>>>>>> #define CONFIGURE_APPLICATION_NEEDS_CLOCK_DRIVER
>>>>>>>>>> #define CONFIGURE_APPLICATION_DISABLE_FILESYSTEM
>>>>>>>>>> #define CONFIGURE_DISABLE_NEWLIB_REENTRANCY
>>>>>>>>>> #define CONFIGURE_TERMIOS_DISABLED
>>>>>>>>>> #define CONFIGURE_LIBIO_MAXIMUM_FILE_DESCRIPTORS 0
>>>>>>>>>> #define CONFIGURE_MINIMUM_TASK_STACK_SIZE 512
>>>>>>>>>> #define CONFIGURE_MAXIMUM_PRIORITY 3
>>>>>>>>>> #define CONFIGURE_DISABLE_CLASSIC_API_NOTEPADS
>>>>>>>>>> #define CONFIGURE_IDLE_TASK_BODY Init
>>>>>>>>>> #define CONFIGURE_IDLE_TASK_INITIALIZES_APPLICATION
>>>>>>>>>> #define CONFIGURE_TASKS 4
>>>>>>>>>>
>>>>>>>>>> #define CONFIGURE_MAXIMUM_TASKS 4
>>>>>>>>>>
>>>>>>>>>> #define CONFIGURE_UNIFIED_WORK_AREAS
>>>>>>>>>>
>>>>>>>>>> Also it is the test from a lower ticker example.
>>>>>>>>>> Configuration_RTEMS_API with -O0 option:
>>>>>>>>>> {maximum_tasks = 5, maximum_timers = 0, maximum_semaphores = 7,
>>>>>>>>>> maximum_message_queues = 0, maximum_partitions = 0, maximum_regions = 0,
>>>>>>>>>> maximum_ports = 0, maximum_periods = 0,
>>>>>>>>>> maximum_barriers = 0, number_of_initialization_tasks = 0,
>>>>>>>>>> User_initialization_tasks_table = 0x0}
>>>>>>>>>>
>>>>>>>>>> with -Os option:
>>>>>>>>>> {maximum_tasks = 124, maximum_timers = 0, maximum_semaphores = 7,
>>>>>>>>>> maximum_message_queues = 0, maximum_partitions = 0, maximum_regions = 0,
>>>>>>>>>> maximum_ports = 0, maximum_periods = 0,
>>>>>>>>>> maximum_barriers = 0, number_of_initialization_tasks = 0,
>>>>>>>>>> User_initialization_tasks_table = 0x0}
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>> Hmmm.. If you look at this structure in gdb without attaching to
>>>>>>>>> the target, what
>>>>>>>>> is maximum_tasks?
>>>>>>>>>
>>>>>>>>> --joel
>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>> It seems that other tasks are LIBBLOCK tasks.
>>>>>>>>>>>>
>>>>>>>>>>>> Also, this is my Configuration during run:
>>>>>>>>>>>> (gdb) p Configuration.stack_space_size
>>>>>>>>>>>> $1 = 2648
>>>>>>>>>>>> (gdb) p Configuration.work_space_size
>>>>>>>>>>>> $2 = 4216
>>>>>>>>>>>> (gdb) p Configuration.interrupt_stack_size
>>>>>>>>>>>> $3 = 512
>>>>>>>>>>>> (gdb) p Configuration.idle_task_stack_size
>>>>>>>>>>>> $4 = 512
>>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> That looks reasonable. Add CONFIGURE_MAXIMUM_PRIORITY and set it
>>>>>>>>>>> to 4. That should
>>>>>>>>>>> reduce the workspace.
>>>>>>>>>>>
>>>>>>>>>>> long term, we might want to consider lowering it permanently
>>>>>>>>>>> like one of the Coldfires
>>>>>>>>>>> had to. Or change the default scheduler to the Simple one to
>>>>>>>>>>> save memory.
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>> I haven't dealt with the Scheduler option yet.
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>> --
>>>>>>>>>> Regards, Denis Obrezkov
>>>>>>>>>>
>>>>>>>>>
>>>>>>>>> maximum_tasks = 4
>>>>>>>> So, is it a linker file issue?
>>>>>>>>
>>>>>>>> This is it:
>>>>>>>> https://github.com/embeddedden/rtems-riscv/blob/hifive1/c/sr
>>>>>>>> c/lib/libbsp/riscv32/hifive1/startup/linkcmds
>>>>>>>>
>>>>>>>> --
>>>>>>>> Regards, Denis Obrezkov
>>>>>>>>
>>>>>>>
>>>>>>> After download:
>>>>>> {maximum_tasks = 938162044, maximum_timers = 1270834941,
>>>>>> maximum_semaphores = 2534801264 <(253)%20480-1264>,
>>>>>> maximum_message_queues = 425684620, maximum_partitions = 1496738036,
>>>>>> maximum_regions = 3085560870 <(308)%20556-0870>, maximum_ports =
>>>>>> 4269782132, maximum_periods = 2362012542 <(236)%20201-2542>,
>>>>>> maximum_barriers = 1138223297, number_of_initialization_tasks = 4224313421,
>>>>>> User_initialization_tasks_table = 0x43bd1bd3}
>>>>>>
>>>>>> right after data copying:
>>>>>> {maximum_tasks = 124, maximum_timers = 0, maximum_semaphores = 1,
>>>>>> maximum_message_queues = 0, maximum_partitions = 0, maximum_regions = 0,
>>>>>> maximum_ports = 0, maximum_periods = 0,
>>>>>> maximum_barriers = 0, number_of_initialization_tasks = 0,
>>>>>> User_initialization_tasks_table = 0x0}
>>>>>>
>>>>>> But I found the mistake - I made it in .data initialization code
>>>>>> (https://github.com/embeddedden/rtems-riscv/blob/hifive1/c/s
>>>>>> rc/lib/libbsp/riscv32/hifive1/start/start.S#L116 - first byte in the
>>>>>> loop was uninitialized)
>>>>>>
>>>>>>
>>>>>> Awesome! Does that mean it is running?
>>>>>>
>>>>>>
>>>>>>
>>>>>> --
>>>>>> Regards, Denis Obrezkov
>>>>>>
>>>>>>
>>>>>> Yes, it is running now. Not far, but running.
>>>>> Now I am having an exception during atexit( Clock_exit )
>>>>>
>>>>>
>>>>> Does it get to bsp_cleanup and bsp_reset? Are you seeing the Terminate?
>>>>>
>>>>> I think those are the names. Basically some BSPs deliberately throw an
>>>>> exception as the way to end.
>>>>>
>>>>>
>>>>>
>>>>>
>>>>>
>>>>> --
>>>>> Regards, Denis Obrezkov
>>>>>
>>>>>
>>>>> Unfortunately, I have an exception in the beginning during clock
>>>> driver initialization around this line:
>>>> 0x204053a0 80 in ../../../../../gcc-7.1.0/newli
>>>> b/libc/stdlib/__atexit.c
>>>>
>>>>
>>>> No obvious suggestions from me right now except to debug.
>>>>
>>>>
>>>>
>>>>
>>>>
>>>> --
>>>> Regards, Denis Obrezkov
>>>>
>>>>
>>>> Is it possible that newlib has a wrong linker file? That a variable
>>> placed out of the rtems-linkcmd-defined bounds?
>>>
>>>
>>> I don't think so. Look at the atexit source and see what it assume is
>>> initialized
>>>
>>>
>>>
>>>
>>> --
>>> Regards, Denis Obrezkov
>>>
>>>
>>> I found out that the error occurs in gcc-7.1.0/newlib/libc/stdlib/_
>> _atexit.c:80:
>> p = _GLOBAL_ATEXIT;
>>
>> p defined like:
>> register struct _atexit *p;
>>
>> on the one hand the value of p is:
>> (gdb) print p
>> $56 = <optimized out>
>>
>> on the other hand:
>> _GLOBAL_ATEXIT:
>>
>> #ifdef _REENT_GLOBAL_ATEXIT
>> extern struct _atexit *_global_atexit; /* points to head of LIFO stack */
>> # define _GLOBAL_ATEXIT _global_atexit
>> #else
>> # define _GLOBAL_ATEXIT (_GLOBAL_REENT->_atexit)
>> #endif
>>
>> and _REENT_GLOBAL_ATEXIT should be defined due to
>> (newlib/libc/include/sys):
>> #if defined(__rtems__)
>> #define __FILENAME_MAX__ 255
>> #define _READ_WRITE_RETURN_TYPE _ssize_t
>> #define __DYNAMIC_REENT__
>> #define _REENT_GLOBAL_ATEXIT
>> #endif
>>
>> but _global_atexit located at random locations outside of my memory
>> regions.
>>
>>
>> Does it show up at a valid address when you look at the nm output?
>>
>> Does it behave differently when you drop optimization?
>>
>>
>>
>> --
>> Regards, Denis Obrezkov
>>
>>
> Oh, the pointer itself is located in a proper location, but the value it
> stores is out of memory.
> Anyway, it fails before it is able to utilize the wrong value.
> It behaves the same way without optimization.
>
> --
> Regards, Denis Obrezkov
>
I've also found that my bad address (caught in mbadaddr register)
is 0x1a913cb0:
(gdb) x /3i 0x1a913cb0-4
0x1a913cac: unimp
0x1a913cae: unimp
0x1a913cb0: unimp
At the same time:
(gdb) p _global_atexit
$5 = (struct _atexit *) 0x1a913cac
Located very close to the wrong address.
--
Regards, Denis Obrezkov
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rtems.org/pipermail/devel/attachments/20170730/662edf11/attachment-0001.html>
More information about the devel
mailing list