Weird memory corruption issues observed on Cortex-M33 of i.MX93EVK

[Vikas7Bal]

Hi,
I am working with iMX93EVK board (especially bare metal programming of Cortex-M33)

I have tried the examples (ethosu_apps) available as part of sdk_2_16 and I observed that, the code is executing from ITCM memory, M33 Code TCM (0FFE_0000 - 0FFF_FFFF) ~128KB.

I am trying to build my application based on "ethosu_apps" and I am facing weird issues, where if I add some debug prints at some point in the code, the build crashes at previous points (which worked before) and I started seeing garbage strings in the logs that are printed.

Not only with the prints, even if I add few lines of code, then the build crashes uexpectedly on Cortex-M33

I have placed my linker script code here. (I couldn't find any file attach options, so directly pasting the code)

/*

** ###################################################################

**     Processor:           MIMX9352DVNxL_cm33

**     Compiler:            GNU C Compiler

**     Reference manual:    IMX93RM, Internal, November. 2021

**     Version:             rev. 1.0, 2021-11-16

**     Build:               b220308

**

**     Abstract:

**         Linker file for the GNU C Compiler

**

**     Copyright 2016 Freescale Semiconductor, Inc.

**     Copyright 2016-2022 NXP

**     All rights reserved.

**

**     SPDX-License-Identifier: BSD-3-Clause

**

**     http:                 www.nxp.com

**     mail:                 support@nxp.com

**

** ###################################################################

*/

/* Entry Point */   

ENTRY(Reset_Handler)

HEAP_SIZE  = DEFINED(__heap_size__)  ? __heap_size__  : 0x4000;

STACK_SIZE = DEFINED(__stack_size__) ? __stack_size__ : 0x6000;

M_VECTOR_RAM_SIZE = DEFINED(__ram_vector_table__) ? 0x0478 : 0x0;

/* Specify the memory areas */

MEMORY

{

  m_interrupts          (RX)  : ORIGIN = 0x0FFE0000, LENGTH = 0x00000478

  m_text                (RX)  : ORIGIN = 0x0FFE0478, LENGTH = 0x0001FB88

  m_data                (RW)  : ORIGIN = 0x20003000, LENGTH = 0x0001B000

  m_m33_suspend_ram     (RW)  : ORIGIN = 0x20000000, LENGTH = 0x00002000

  m_a55_suspend_ram     (RW)  : ORIGIN = 0x20002000, LENGTH = 0x00001000

  m_ocram               (RW)  : ORIGIN = 0x20480000, LENGTH = 0x00060000

  m_rsc_tbl             (RW)  : ORIGIN = 0x2001E000, LENGTH = 0x00001000

}

/* Define output sections */

SECTIONS

{

  /* Interrupt vector table */

  .interrupts :

  {

    __VECTOR_TABLE = .;

    __Vectors = .;

    . = ALIGN(4);

    KEEP(*(.isr_vector))

    . = ALIGN(4);

  } > m_interrupts

  .resource_table :

  {

    . = ALIGN(8);

    KEEP(*(.resource_table))

    . = ALIGN(8);

  } > m_text

  /* Program code */

  .text :

  {

    . = ALIGN(4);

    *(.text)

    *(.text*)

    *(.glue_7)

    *(.glue_7t)

    *(.eh_frame)

    KEEP (*(.init))

    KEEP (*(.fini))

    . = ALIGN(4);

  } > m_text

  .ARM.extab : { *(.ARM.extab* .gnu.linkonce.armextab.*) } > m_text

  .ARM :

  {

    __exidx_start = .;

    *(.ARM.exidx*)

    __exidx_end = .;

  } > m_text

  .ctors :

  {

    __CTOR_LIST__ = .;

    KEEP (*crtbegin.o(.ctors))

    KEEP (*crtbegin?.o(.ctors))

    KEEP (*(EXCLUDE_FILE(*crtend?.o *crtend.o) .ctors))

    KEEP (*(SORT(.ctors.*)))

    KEEP (*(.ctors))

    __CTOR_END__ = .;

  } > m_text

  .dtors :

  {

    __DTOR_LIST__ = .;

    KEEP (*crtbegin.o(.dtors))

    KEEP (*crtbegin?.o(.dtors))

    KEEP (*(EXCLUDE_FILE(*crtend?.o *crtend.o) .dtors))

    KEEP (*(SORT(.dtors.*)))

    KEEP (*(.dtors))

    __DTOR_END__ = .;

  } > m_text

  .preinit_array :

  {

    PROVIDE_HIDDEN (__preinit_array_start = .);

    KEEP (*(.preinit_array*))

    PROVIDE_HIDDEN (__preinit_array_end = .);

  } > m_text

  .init_array :

  {

    PROVIDE_HIDDEN (__init_array_start = .);

    KEEP (*(SORT(.init_array.*)))

    KEEP (*(.init_array*))

    PROVIDE_HIDDEN (__init_array_end = .);

  } > m_text

  .fini_array :

  {

    PROVIDE_HIDDEN (__fini_array_start = .);

    KEEP (*(SORT(.fini_array.*)))

    KEEP (*(.fini_array*))

    PROVIDE_HIDDEN (__fini_array_end = .);

  } > m_text

  __etext = .;

  __DATA_ROM = .;

  .interrupts_ram :

  {

    . = ALIGN(4);

    __VECTOR_RAM__ = .;

    __interrupts_ram_start__ = .;

    *(.m_interrupts_ram)

    . += M_VECTOR_RAM_SIZE;

    . = ALIGN(4);

    __interrupts_ram_end__ = .;

  } > m_ocram

  __VECTOR_RAM = DEFINED(__ram_vector_table__) ? __VECTOR_RAM__ : ORIGIN(m_interrupts);

  __RAM_VECTOR_TABLE_SIZE_BYTES = DEFINED(__ram_vector_table__) ? (__interrupts_ram_end__ - __interrupts_ram_start__) : 0x0;

  .data : AT(__DATA_ROM)

  {

    . = ALIGN(4);

    __DATA_RAM = .;

    __data_start__ = .;

    *(.data)

    *(.data*)

    *(.rodata)

    *(.rodata*)

    *(.got)    

    *(.got.*)   

    . = ALIGN(16);

    KEEP(*(.jcr*))

    . = ALIGN(4);

    __data_end__ = .;

  } > m_ocram

  __DATA_END = __DATA_ROM + (__data_end__ - __data_start__);

  text_end = ORIGIN(m_text) + LENGTH(m_text);

  .ocram.init : AT(__DATA_END)

  {

    __ocram_data_start__ = .;

    . = ALIGN(16);

    *(.bss.tensor_arena_ocram)

    __ocram_data_end__ = .;

  } > m_ocram

  .bss :

  {

    . = ALIGN(4);

    __START_BSS = .;

    __bss_start__ = .;

    *(.bss)

    *(.bss*)

    *(COMMON)

    . = ALIGN(4);

    __bss_end__ = .;

    __END_BSS = .;

  } > m_ocram

  .m33_suspend :

  {

    *(M33SuspendRam)

    . = ALIGN(4);

  } > m_m33_suspend_ram

  .heap :

  {

    . = ALIGN(8);

    __end__ = .;

    PROVIDE(end = .);

    __HeapBase = .;

    . += HEAP_SIZE;

    __HeapLimit = .;

    __heap_limit = .;

  } > m_ocram

  .stack (NOLOAD) :

  {

  __StackLimit = .;

  . += STACK_SIZE;

  __StackTop = .;

  PROVIDE(__stack = __StackTop);

  } > m_data

  __StackTop   = ORIGIN(m_data) + LENGTH(m_data);

  __StackLimit = __StackTop - STACK_SIZE;

  PROVIDE(__stack = __StackTop);

  __RscTblStart = ORIGIN(m_rsc_tbl);

  __RscTblEnd = ORIGIN(m_rsc_tbl) + LENGTH(m_rsc_tbl);

  .ARM.attributes 0 : { *(.ARM.attributes) }

  #ASSERT(__StackLimit >= __HeapLimit, "region m_data overflowed with stack and heap")

}

I can confirm that, there is no corruption between stack and heap, as I have kept stack in m_data (TCM memory) and heap in m_ocram (OCRAM memory)

So, can someone please provide any suggestions/ insights on whether I am missing something here?


Thanks!

[Manuel_Salas]

Hello @Vikas7Bal 

I hope you are doing very well.

It appears to be a memory issue, as you know the TCM total size is 256KB but divided in two parts of 128kB.

Your application seems to be equals or higher to 128kB.

Can you try reducing the app size just for test?

Best regards,

Salas.