stm32F746 + sdram problem

[Franzi.Edo]

Posted on December 01, 2015 at 16:24

Hi,

I have an stm32F746 configured to run @ 216-MHz. My primary memory is an sdram configured on the block 2 (address 0xD0000000).

My memory tests (running from the flash) are all OK (many different tests under different environmental conditions (lower voltage, temperature, etc)).

Now, as soon as I try to execute a small code downloaded into this sdram memory (I verified the code in the sdram), I have an immediate exception (Hard Fault).

I tried many usual actions such as the reduction of the cpu speed, increased the memory timing etc. No way, I can safely read/write in the memory but I cannot execute any code.

On my previous targets (stm32F429) I never had this behaviour. The address 0xD0000000 seems not to be under any cache control.

So, any ideas of why I cannot execute programs on the sdram?

Regards

Edo

[Nesrine M_O]

Posted on December 01, 2015 at 17:16

Hi Edo,

“So, any ideas of why I cannot execute programs on the sdram?�

The SDRAM bank 2 starting from 0xD0000000 is a never execute area as it is mentioned in the Cortex®-M7 default mapping.

So you need to modify the region attribute by using The Cortex®-M7 MPU.

-Syrine-

[Franzi.Edo]

Posted on December 01, 2015 at 17:48

Hi Syrine,

Thank you for this information. Of course, you are right. However, and like for my previous M4 designs, I have swapped the bank using the following instruction. Now the memory is seen at the address 0x70000000.

SYSCFG->MEMRMP = 0x400;

With this instruction I still have some strange behaviours (even if the programs do not crache immediately). In this mode I have the impression that it is a cache issue.

I don't know if the M7 at the start-up it has the caches activated. If yes, how to disable them before using the sdram? Of course, I decrease the performances, but at least I can continue to work.

Edo

[Nesrine M_O]

Posted on December 02, 2015 at 09:27

Hi Edo,

1-You are using a custom board or an ST board?

2-“I don't know if the M7 at the start-up it has the caches activated�

By default, the Cortex®-M7 system caches (instruction and data caches) are disabled.

3-I recommended you to have a look to the FMC_SDRAM_MemRemap project under the STM32Cube F7 package, the example guides you through the different configuration steps to use the IS42S32800G SDRAM memory (mounted on STM327x6G-EVAL revB evaluation board) as code execution memory.

STM32Cube_FW_F7_V1.2.0\Projects\STM32756G_EVAL\Examples\FMC\FMC_SDRAM_MemRemap

-Syrine-

[Franzi.Edo]

Posted on December 02, 2015 at 09:30

Hi Syrine,

Thank you. Your information helped me to progress.

For 1. No, I have my own design

For 2. OK

For 3. OK, I will check again.

But in the meantime some progress:

I changed the attributes of a portion of the sdram area and now I can execute program inside the sdram.

I slowly progress but still remain a problem and maybe you will have an advise about.

So, now I can execute code but I cannot use the same physical sdram for locating a stack (or other R/W data).

As soon as my sdram running code access a data my system hangs (no exceptions!!).

Here is my sdram mapping

0xD00000000 - 0xD01FFFFF Heap

0xD02000000 - 0xD05FFFFF Executable area

0xD06000000 - 0xD07FFFFF Some data (images)

In order to make executable the area 0xD02000000 - 0xD05FFFFF I modified the MPU in the following way

// New attributes for the SDRAM area (0xD0200000)

// Change the attribute only in the executable area

MPU->CTRL = 0x00000000; // Disable the MPU

MPU->RNR  = 0x00000000; // Region 0

MPU->RBAR = 0xD0200000 | (1<<4) | (0<<0); // Address, valid & region 0

MPU->RASR = (0<<28) // Instruction fetches enabled

  | (3<<24) // Full access

  | (1<<19) // TEX: 001 Memory attribute

  | (1<<18) // S:     1

  | (1<<17) // C:     1

  | (1<<16) // B:     1 Sharable, Normal,

   // Outer and inner write-back.

// Write and read allocate

  | (0<<8) // Sub-region is enabled

  | (22<<1) // 4-MB

  | (1<<0); // Region enable

MPU->CTRL = (1<<2) // Enable the usage of all the

// default map

  | (0<<1) // MPU disabled during faults

  | (1<<0); // MPU enabled

__asm__ volatile (

''dsb \n''

''isb \n''

);

What is the solution to make possible the usage of the sdram for execution code and simultaneously for storing data without troubles?

Regards

  Edo

[js23]

Posted on May 09, 2016 at 13:05

Hi Edo,

that is exactly the problem I have at the moment. Could you (or somebody else) solve it in the meantime?

Thanks,

Johannes