Risc-v memory issue

Wed Jun 14 05:35:26 UTC 2017

On Wed, Jun 14, 2017 at 2:58 PM, Denis Obrezkov <denisobrezkov at gmail.com> wrote:
> 2017-06-14 2:18 GMT+03:00 Hesham Almatary <heshamelmatary at gmail.com>:
>>
>>
>>
>> On Tue, Jun 13, 2017 at 10:20 PM, Joel Sherrill <joel at rtems.org> wrote:
>>>
>>>
>>>
>>> On Jun 12, 2017 3:10 PM, "Hesham Almatary" <heshamelmatary at gmail.com>
>>> wrote:
>>>
>>>
>>>
>>> On Tue, Jun 13, 2017 at 2:09 AM, Denis Obrezkov <denisobrezkov at gmail.com>
>>> wrote:
>>> > Hello all,
>>> >
>>> > I was able to debug my program on top of HiFive1 board.
>>> > I found out that I can't use next and until commands from gdb,
>>> > they just hang gdb.
>>> > And now I have a problem, this is my backtrace:
>>> >
>>> > #0  0x2041d728 in _User_extensions_Iterate (arg=0x80002eb8,
>>> > visitor=0x2041d64e <_User_extensions_Fatal_visitor>,
>>> > direction=CHAIN_ITERATOR_FORWARD)
>>> >     at
>>> >
>>> > /home/reprofy/Projects/riscv/rtems/development/rtems/kernel/rtems-riscv/c/src/../../cpukit/score/src/userextiterate.c:155
>>> > #1  0x20410ed4 in _User_extensions_Fatal (source=INTERNAL_ERROR_CORE,
>>> > error=2) at
>>> > ../../cpukit/../../../hifive1/lib/include/rtems/score/userextimpl.h:307
>>> > #2  0x20410ef8 in _Terminate (the_source=INTERNAL_ERROR_CORE,
>>> > the_error=2)
>>> > at
>>> >
>>> > /home/reprofy/Projects/riscv/rtems/development/rtems/kernel/rtems-riscv/c/src/../../cpukit/score/src/interr.c:35
>>> > #3  0x20410f40 in _Internal_error
>>> > (core_error=INTERNAL_ERROR_TOO_LITTLE_WORKSPACE) at
>>> >
>>> > /home/reprofy/Projects/riscv/rtems/development/rtems/kernel/rtems-riscv/c/src/../../cpukit/score/src/interr.c:44
>>> > #4  0x2041e484 in _Workspace_Handler_initialization (areas=0x80002f78,
>>> > area_count=1, extend=0x0) at
>>> >
>>> > /home/reprofy/Projects/riscv/rtems/development/rtems/kernel/rtems-riscv/c/src/../../cpukit/score/src/wkspace.c:123
>>> > #5  0x2040031c in bsp_work_area_initialize_default
>>> > (area_begin=0x80001d58,
>>> > area_size=8872) at
>>> > ../../../../../.././hifive1/lib/include/bsp/bootcard.h:147
>>> > #6  0x20400372 in bsp_work_area_initialize () at
>>> >
>>> > /home/reprofy/Projects/riscv/rtems/development/rtems/kernel/rtems-riscv/c/src/lib/libbsp/riscv32/hifive1/../../shared/bspgetworkarea.c:61
>>> > #7  0x2040f3ce in rtems_initialize_executive () at
>>> >
>>> > /home/reprofy/Projects/riscv/rtems/development/rtems/kernel/rtems-riscv/c/src/../../cpukit/sapi/src/exinit.c:95
>>> > #8  0x204002cc in boot_card (cmdline=0x0) at
>>> >
>>> > /home/reprofy/Projects/riscv/rtems/development/rtems/kernel/rtems-riscv/c/src/lib/libbsp/riscv32/hifive1/../../shared/bootcard.c:76
>>> > #9  0x20400064 in _end_clear_bss () at
>>> >
>>> > /home/reprofy/Projects/riscv/rtems/development/rtems/kernel/rtems-riscv/c/src/lib/libbsp/riscv32/hifive1/start/start.S:116
>>> >
>>> > So, I think I should increase a workspace.
>>> > Now, I have some questions:
>>> > I took Hesham's linker file for riscv-generic, and there were no stack
>>> > space,
>>> > I allocated 0x1000 for stack. Should I make it zero again, or should I
>>> > decrease its size?
>>> > what size for stack should be enough?
>>>>
>>>>                 /*
>>>>                  * The stack section will occupy the remaining
>>>> REGION_STACK region and may
>>>>                  * contain the task stacks.  Depending on the region
>>>> distribution this
>>>>                  * section may be of zero size.
>>>>                  */
>>>
>>> This means that the stack_end address is starting at your very top
>>> address space (i.e. 16KiB if you set your linker size to 16KiB). The stack
>>> grows dynamically downwards, so it might overflow other sections (e.g. work,
>>> heap, data, etc) if you don't have enough memory size to fit. You can try to
>>> see how much size each section occupies (riscv32-rtems4.12-objdump -h
>>> hello.exe), especially work space.
>>>
>>>
>>> This is how the SPARC start up stack is done. If you aren't careful, this
>>> may not match the assumption in the default method that carves up the
>>> workspace.
>>>
>>> You need to be careful that the memory regions don't overlap.
>>>
>>>
>>>
>>>
>>> > And why there were no space for stack? is it possible to work without
>>> > stack
>>> > allocated in
>>> > linkcmd file?
>>> > Can RTEMS work with a smaller workspace?
>>> >
>>> As far as I remember, 4KiB was the smallest I could allocate for
>>> workspace to work (with Epiphany). As I suggested before, you might want to
>>> start saving up memory by using the local instruction memory (on HiFive) and
>>> compressed version of RISC-V ISA.
>>>
>>>
>>> Is this a characteristic of this board or the architecture in general? I
>>> had the impression that the RISC-V was quite general purpose and could
>>> handle plenty of RAM even if some boards don't have much.
>>>
>>>
>> RISC-V itself doesn't put such restrictions on the size of memory (e.g.
>> Linux, seL4, RTEMS can run on FPGA and/or simulators). There's a compressed
>> extension for RISC-V (like thumb) for platforms that have small memories
>> same as HiFive. Having such small memory is only HiFive related, not RISC-V.
>>>
>>>
>>> > --
>>> > Regards, Denis Obrezkov
>>>
>>>
>>>
>>> --
>>> Hesha
>>>
>>> _______________________________________________
>>> devel mailing list
>>> devel at rtems.org
>>> http://lists.rtems.org/mailman/listinfo/devel
>>>
>>>
>>
>>
>>
>> --
>> Hesham
>
> I think that RISC-V Compressed ISA can't reduce .bss and .data sections, am
> I right?
> So, it seems that space allocated by rtems for its workspace should be
> reduced.
>
Currently all of the sections (including .bss, .text, .data, etc) are
competing for 16KiB memory, if you reduced the .text (using compressed
RISC-V and/or ITIM) section size, this gives more space for other
sections, hence .work and .stack.

>
> --
> Regards, Denis Obrezkov


-- 
Hesham