TMS320F28P650DK: Errata about prefetching beyond Valid Memory

Hi,

Johnny Chen said:

(1). Would the errata also affect the CLA core for using the last 8W length of LS9 as CLA program ? Should be NOT, right ?

No, it's only for CPU.

Johnny Chen said:

(2). https://e2e.ti.com/support/microcontrollers/c2000-microcontrollers-group/c2000/f/c2000-microcontrollers-forum/1405856/tms320f28p650dk-the-ls8-and-ls9-ram-for-cla-program/5391776#5391776

In this post, may I reconfirm the conclusion is that user can't assign LS0, LS1, LS8 and LS9 RAM as CLA program memory as a combined memory sector shown as below, right ?

RAMLS_CLA_PROG : origin = 0x004000, length = 0x005000 // LS8~LS9~LS0~LS1 for CLA_PROGRAM

That is correct.

Regards,

Vivek Singh

Hi Vivek,

Thanks for your reply.

Vivek Singh said:

No, it's only for CPU.

To configure LS8 & LS9 as CLA program, user has to copy code from flash that it's done by CPU1 as below,

memcpy((uint32_t *)((uint32_t)&Cla1ProgRunStart + (0x1E000U)), (uint32_t *)&Cla1ProgLoadStart,
(uint32_t)&Cla1ProgLoadSize);

Executing the memcpy line, won't it be a problem here for the errata ? since CPU1 should still own the last 8W length of LS9 at this moment, or my concern is unnecessary actually ?

Vivek Singh said:

That is correct.

So that how would you suggest to configure a memory for LS0_1, and a memory for LS8_9 RAM as CLA program memory in .cmd file and memcpy ?

Should it configure as such ?

.cmd file

RAMLS8_LS9 : origin = 0x004000, length = 0x004000 // LS8~LS9 for CLA_PROGRAM

RAMLS0_LS1 : origin = 0x008000, length = 0x001000 // LS0~LS1 for CLA_PROGRAM

Cla1Prog : LOAD = FLASH_BANK0,
RUN = RAMLS8_LS9 | RAMLS0_LS1,
LOAD_START(Cla1ProgLoadStart),
RUN_START(Cla1ProgRunStart),
LOAD_SIZE(Cla1ProgLoadSize),
ALIGN(4)

.c file to do memcpy

memcpy((uint32_t *)((uint32_t)&Cla1ProgRunStart + (0x1E000U)), (uint32_t *)&Cla1ProgLoadStart,
(uint32_t)&Cla1ProgLoadSize);

Do you see any problem here ? Please correct if there is any wrong.

Thanks,

Johnny

Hi Vivek,

Any chance to get the input here ?

Thanks,

Johnny

Johnny Chen said:

Executing the memcpy line, won't it be a problem here for the errata ? since CPU1 should still own the last 8W length of LS9 at this moment, or my concern is unnecessary actually ?

Issue is for the code fetch and not for reading the data so it should be ok.

Johnny Chen said:

.c file to do memcpy

memcpy((uint32_t *)((uint32_t)&Cla1ProgRunStart + (0x1E000U)), (uint32_t *)&Cla1ProgLoadStart,
(uint32_t)&Cla1ProgLoadSize);

On CPU side, address for LS8/LS9 and LS0/LS1 are not continuous so this memcpy will not work for LS0/LS1. You need to do it in two step. One for LS0 and LS1 and other one for LS8 and LS9.

Regards,

Vivek Singh

Hi Vivek,

Thanks for the reply.

Vivek Singh said:

Issue is for the code fetch and not for reading the data so it should be ok.

Would you kindly clarify more what is the difference between code fetch and reading data ?

Vivek Singh said:

You need to do it in two step. One for LS0 and LS1 and other one for LS8 and LS9.

Do you mean doing as such ? Do twice for that

.cmd file

RAMLS8_LS9 : origin = 0x004000, length = 0x004000 // LS8~LS9 for CLA_PROGRAM

RAMLS0_LS1 : origin = 0x008000, length = 0x001000 // LS0~LS1 for CLA_PROGRAM

Cla1Prog : LOAD = FLASH_BANK0,
RUN = RAMLS8_LS9,
LOAD_START(Cla1ProgLoadStart_LS8_9),
RUN_START(Cla1ProgRunStart_LS8_9),
LOAD_SIZE(Cla1ProgLoadSize_LS8_9),
ALIGN(4)

Cla1Prog : LOAD = FLASH_BANK0,
RUN = RAMLS0_LS1,
LOAD_START(Cla1ProgLoadStart_LS0_1),
RUN_START(Cla1ProgRunStart_LS0_1),
LOAD_SIZE(Cla1ProgLoadSize_LS0_1),
ALIGN(4)

.c file to do memcpy

memcpy((uint32_t *)&Cla1ProgRunStart_LS0_1, (uint32_t *)&Cla1ProgLoadStart_LS0_1,
(uint32_t)&Cla1ProgLoadSize_LS0_1);

memcpy((uint32_t *)((uint32_t)&Cla1ProgRunStart_LS8_9 + (0x1E000U)), (uint32_t *)&Cla1ProgLoadStart_LS8_9,
(uint32_t)&Cla1ProgLoadSize_LS8_9);

Is above looking correct ? Please correct me if there is any wrong.

Thanks and regards,

Johnny