M487: Crash report test failed in IAR

[ccli8]

Description

Target:
NUMAKER_PFM_M487

Toolchain:
IAR

mbed-os:
0a832dd (HEAD -> master, origin/master, origin/HEAD) Merge pull request #9051 from michalpasztamobica/esp8266_busy_signal

The test item mbed-os-tests-mbed_platform-crash_reporting / mbed-os-tests-mbed_platform-crash_reporting is passed in ARM/ARMC6/GCC_ARM, but failed in IAR. Per my checking, the crash data ram base/limit __CRASH_DATA_RAM_START__/__CRASH_DATA_RAM_END__ calculated in M487.icf are incorrect because IAR linker doesn't arrange memory as simply expected.

In mbed-os/targets/TARGET_NUVOTON/TARGET_M480/device/TOOLCHAIN_IAR\M487.icf:

/* Define Crash Data Symbols */
define exported symbol __CRASH_DATA_RAM_START__ = __ICFEDIT_region_IRAM_start__ + __ICFEDIT_size_cstack__ + __ICFEDIT_size_intvec__;
define exported symbol __CRASH_DATA_RAM_END__ = __ICFEDIT_region_IRAM_start__ + __ICFEDIT_size_cstack__ + __ICFEDIT_size_intvec__ + __ICFEDIT_size_crash_data__;

In mbed-os/platform/mbed_crash_data_offsets.h:

#if MBED_CONF_PLATFORM_CRASH_CAPTURE_ENABLED
#if defined(__CC_ARM) || (defined(__ARMCC_VERSION) && (__ARMCC_VERSION >= 6010050))
extern uint32_t Image$$RW_m_crash_data$$ZI$$Base[];
extern uint32_t Image$$RW_m_crash_data$$ZI$$Size;
#define __CRASH_DATA_RAM_START__    Image$$RW_m_crash_data$$ZI$$Base
#define __CRASH_DATA_RAM_SIZE__     Image$$RW_m_crash_data$$ZI$$Size
#elif defined(__ICCARM__)
extern uint32_t __CRASH_DATA_RAM_START__[];
extern uint32_t __CRASH_DATA_RAM_END__[];
#define __CRASH_DATA_RAM_SIZE__     (__CRASH_DATA_RAM_END__ - __CRASH_DATA_RAM_START__)
#elif defined(__GNUC__)
extern uint32_t __CRASH_DATA_RAM_START__[];
extern uint32_t __CRASH_DATA_RAM_END__[];
#define __CRASH_DATA_RAM_SIZE__     (__CRASH_DATA_RAM_END__ - __CRASH_DATA_RAM_START__)
#endif /* defined(__CC_ARM) */

If I change __CRASH_DATA_RAM_START__/__CRASH_DATA_RAM_END__ as below, the test is passed. But this change cannot apply to all targets because CRASH_DATA_RAM is defined only in NUMAKER_PFM_M487 target or some.

#elif defined(__ICCARM__)
extern uint32_t CRASH_DATA_RAM$$Base[];
extern uint32_t CRASH_DATA_RAM$$Limit[];
#define __CRASH_DATA_RAM_START__    CRASH_DATA_RAM$$Base
#define __CRASH_DATA_RAM_END__      CRASH_DATA_RAM$$Limit
#define __CRASH_DATA_RAM_SIZE__     (__CRASH_DATA_RAM_END__ - __CRASH_DATA_RAM_START__)

Besides, checking mbed-os/platform/mbed_crash_data_offsets.h, the unit of CRASH_DATA_RAM_SIZE doesn't match across all toolchains. In ARM/ARMC6, it is byte, but in IAR/GCC_ARM, it is word.

Issue request type

[ ] Question
[ ] Enhancement
[x] Bug

[ciarmcom]

Internal Jira reference: https://jira.arm.com/browse/MBOCUSTRIA-303

[0xc0170]

Just to confirm, this is the issue specific to M487 only ?

[ccli8]

@0xc0170 The issue happens on M487 and possibly on other Nuvoton targets if crash report support is added as M487. This is because the common code __CRASH_DATA_RAM_START__/__CRASH_DATA_RAM_END__ cannot be simply calculated out in M487 IAR .icf file. For other non-Nuvoton targets, I cannot confirm because IAR linker may or may not arrange memory as __CRASH_DATA_RAM_START__/__CRASH_DATA_RAM_END__ can be simply calculated out.

[cmonr]

@SenRamakri Fyi ^^^

[samchuarm]

Hi @ccli8 , is this still an issue? What's pending? Thanks.

[ccli8]

is this still an issue? What's pending? Thanks.

@samchuarm This issue happens on M487 and might also happen on other targets. Expect mbed team could come out with a general solution.

[0xc0170]

@samchuarm This issue happens on M487 and might also happen on other targets. Expect mbed team could come out with a general solution.

@ARMmbed/mbed-os-core Can you review this issue please?

[SenRamakri]

@ccli8 - I'm trying to investigate this. Unfortunately I don't have M487 board and so cant really try. DO you know if this is reproducible in any other Nuvoton boards?

Btw, I tried to compile for M487 and I do see the below symbols defined in Map file and they are 0x100 apart. So, I'm thinking symbols are exported as required and of course, they are defined as the compilation succeeds. Its possible that icf is not properly configured. I'm looking more into this.

CRASH_DATA_RAM_END {Abs}
0x200006c0 Data Gb command line/config [164]
CRASH_DATA_RAM_START {Abs}
0x200005c0 Data Gb command line/config [164]

[ccli8]

@SenRamakri Following M487, I privately add crash report support to NUMAKER_PFM_NUC472 and can reproduce the same issue and its fix. I think all Nuvoton targets would suffer the same issue. In my opinion, like my fix above, all targets supporting crash report shall define the CRASH_DATA_RAM block. __CRASH_DATA_RAM_START__/__CRASH_DATA_RAM_END__ can then be derived from the block rather than calculated manually.

#elif defined(__ICCARM__)

extern uint32_t CRASH_DATA_RAM$$Base[];
extern uint32_t CRASH_DATA_RAM$$Limit[];

#define __CRASH_DATA_RAM_START__    CRASH_DATA_RAM$$Base
#define __CRASH_DATA_RAM_END__      CRASH_DATA_RAM$$Limit
#define __CRASH_DATA_RAM_SIZE__     (__CRASH_DATA_RAM_END__ - __CRASH_DATA_RAM_START__)

[SenRamakri]

@ccli8 - I see your point although we are not seeing this with other targets. I think for some reason the alignment we are using for different blocks in M487.icf is messing up with placement of crash_data block. I'm looking more into this now.

[sathysreenath]

Hi

We are seeing this test fail as well on the k64 frdm board with IAR. Not sure if this helps, but here is the info from verbose when the test errors out.

++ MbedOS Error Info ++
[1550526552.16][CONN][RXD] Error Status: 0x80FF011F Code: 287 Module: 255
[1550526552.21][CONN][RXD] Error Message: Executing crash reporting test.
[1550526552.23][CONN][RXD] Location: 0x0
[1550526552.25][CONN][RXD] Error Value: 0xDEADBAD
[1550526552.37][CONN][RXD] Current Thread: main Id: 0x200002B0 Entry: 0x1C81 StackSize: 0x1000 StackMem: 0x20000310 SP: 0x200012A0
[1550526552.46][CONN][RXD] For more info, visit: https://armmbed.github.io/mbedos-error/?error=0x80FF011F
[1550526552.48][CONN][RXD] -- MbedOS Error Info --

Thanks,
Sathy