CCS/D3-3P-TDA3X-SK: Cannot load program to the EVE of TDA3 RVP in CCS.

I also tried to remove the startup gels of IPU and dap, but I realized I cannot connect DSP if I don't run the TDA3xx_MULTICORE_EnableAllCores. The DSP was always in reset mode. I am using CCS 9.3.

Hi Frank,

In the post http://e2e.ti.com/support/processors/f/791/t/866655?TDA3XEVM-TDA3-RVP-loads-eve-binary-failed-when-debugging-on-CCS, the customer was booting first from sdcard so all the system initialization was taken care by the boot loader and there was no need to load the gels.

But in your case, if you are not booting from the sdcard first, then you need to keep all the gels and select  TDA3xx_MULTICORE_EnableAllCores. Please follow the instructions in the eve_sw_getting_started.pdf in paragraph 'Setup EVE in CCS'.

regards,

Victor

Hi Victor,

    Thank you for your quick reply.

    I followed the instructions in the  eve_sw_getting_started.pdf. But when I reset the CPU for EVE, the program will be stuck in resetting (I guess).

ARP32_EVE_1: GEL Output: --->>> Resetting ARP32 ... <<<---

    There is no new information indicating the resetting done. And I got the same error when I loaded the program in this case.

Best,

Frank

Hi Frank,

Can you try going through the VisionSDK_UserGuide_TDA3xx_RVP.pdf, chapter 3.4 "Load using CCS" . I think for RVP, the gel file TDA3xx_ddr_config.gel is different.

regards,

Victor

Hi Victor,

     I tried, but still failed to reset and load program to EVE.

Best,

Frank

Do you have problem with C6x as well or is it only EVE ?

regards,

Victor

Only EVE. 

I saw in the error message:

ARP32_EVE_1: File Loader: Verification failed: Target failed to read 0xA1000000@Data

It is strange that it tried to access 0xA1000000 . Is this address present in the your EVE linker command file ? If yes then the gel function SetupEveMmuEntry() needs to be updated to include this address in the EVE MMU TLB table.

regards,

Victor

Can you try to build the EVE software release and load any *.out file, for example apps/census/test/elf_out/test.census.eve.out .

regards,

Victor

Hi Victor,

   Here is my eve memory map from vision_sdk_arp32_1_release.xearp32F.map

Best,

Feiran

No, I cannot load the apps/census/test/elf_out/test.census.eve.out. It will be stuck at resetting.

Hi,

To load a *.out file that was built inside the vision SDK, please follow the instructions in the  VisionSDK_UserGuide_TDA3xx_RVP.pdf document. Do not follow the instructions in the eve_sw_getting_started.pdf as they are for loading a *.out file built for EVE standalone, outside of Vision SDK environment.

For vision SDK, You will need to set the variable VISION_SDK_CONFIG to 1 in TDA3xx_multicore_reset.gel . 

regards,

Victor

Another point is that the address 0xA1000000 doesn't appear to be one of the default memory section in vision SDK. Did you make any changes ?

In this case the EVE MMU programming in the gel function EVE_MMU0_Config() need to be updated accordingly.

regards,

Victor 

Hi Victor,

      Can you please let me know what I should change in the gel function EVE_MMU0_Config()? I changed the vec_base to 0xA1000000 based on my memory map but I cannot connect to EVE core in CCS after run TDA3xx_MULTICORE_EnableAllcore script. Do I need to modify other eve mmu entry in the function EVE_MMU0_Config()? I did enable vision_sdk_vision.

Best,

Frank

Hi Frank,

Setting vec_base to 0xA1000000 should be sufficient.

From line 576 of EVE_MMU0_Config, you should have this:

else
{
if(cpu_num == 1)
{
vec_base = 0xA1000000;
}
add_while1_at_resetVector(vec_base);
SetupEveMmuEntry(cpu_num, mmu_num, 1, 0x00000000, vec_base, EVE_MMU_PAGESIZE_4K);
SetupEveMmuEntry(cpu_num, mmu_num, 2, vec_base, vec_base, EVE_MMU_PAGESIZE_16M);

You don't need to modify other entries because each page is 16M and  the line:

 SetupEveMmuEntry(cpu_num, mmu_num, 2, vec_base, vec_base, EVE_MMU_PAGESIZE_16M);

Should cover all the sections for EVE.

regards,

Victor

Can you access the memory 0xA1000000 from the Cortex M3 in a memory window ?

Hi Victor,

    No, I cannot access that memory.

    I cannot connect to the EVE core after I changed the vec_base to 0xA1000000.

    Do you think I can use a sd card to launch the TDA and use ccs to debug the program?

Best,

Frank

Hi Frank,

I would like you to verify the accessibility of this memory location from M4.

After connection to M4, can you open a memory browser window and view the location: 0x80000000 

You should be able to see some values without any issue.

Now open another memory browser window and view the location 0xa1000000. You should also see some values there. Let me know if you have troubles with these.

regards,

Victor

Yes, I can get the access to both 0x80000000 and 0xa1000000.

Thank you,

Frank

Now close the 0x80000000 window and select the script Tda3xx_MULTICORE_EnableAllCores(). After the script completes, connect to ARP32_EVE_1. And look at the content of the memory window 0xa1000000 . Do you see anything ?

It is important to close the window pointing at 0x80000000 because this memory location will no longer be mapped and any attempt from EVE to access it will result in error.  

The value in 0xa1000000 was FFFFFFF before running TDA3xx_MULTICORE_EnableAllCores. Then it will be set to 00000100 after selecting the script. When I connected to EVE, nothing changed in that address. 

Yes this is expected. Since you are able to view the content at 0xa1000000 from ARP32, it means the MMU setup was successful. I guess your issue is that you are not able the *.out file ? You actually have to perform a CPU reset (Ctrl+Shift+R) of the ARP32 before loading the *.out file.

Hi Victor,

        Actually, I cannot connect to the EVE core after I changed the vec_base to 0xa1000000. It will stuck at connecting.

Best,

Frank

What happens if you don't change the vec_base ? Do you still have problems connecting ?

Victor,

       If I don't change the vec_base, I can connect but fail to load EVE program. This is the original post in this issue. 

Thank you,

Frank

Ok. Changing vec_base to 0xa1000000 shouldn't affect the ability to connect to ARP32.

Please select menu Run->Debug Configuration and verify that the options for the tab target ARP-EVE 1 are the same as my screenshot below. In particular, make sure that 'Reset the target on connect' is unchecked.

Hi Victor,

    I just verified it and it is the same as your screenshot.

Thanks,

Frank

Hi Victor,

    I know what makes my EVE vec_base change to 0xa1000000. It is the DDR_MEM setting in cfg.mk file. According to the TDA3xx RVP user guide, this should be set to  DDR_MEM_1024M. If I change this parameter to DDR_MEM_512M, and EMIFMODE=SINGLE_EMIF_512MB, the vec_base address became 0x80000000 and the ccs can load programs to all the cores. However, the application cannot run properly. I got the error shown in the picture even after I load new bootloader to TDA3. 

Please help me figure out how to correctly configure the DDR_MEM. I'm using RVP and it has 1GB DDR. Can I configure it to only use 512 MB? Everytime I change the DDR_MEM, do I need to load a new bootloader to TDA3?

Best,

Frank

Frank,

If RVP has SBL (bootloader) programmed in Flash, it will initialize the system after power-on.

You have to remove all GEL files in CCS for M4 and EVE from target configurations.

If the GEL files are not removed, the GEL script will try to program EVE MMU again and that will crash EVE so it can't be connected in CCS.

Are you trying to debug the AppImage loaded from SD card? Or, do you want to load binary in CCS and then debug?

Steps are different for these 2 scenarios.

Regards,

Stanley

Hi Stanley,

    I would like to load binary in CCS and then debug.

    Now I can load binary to CCS with DDR_MEM=DDR_MEM_512M. What I did is removing the EMIFMODE configuration in cfg.mk,  and this fix the issue with DDR_MEM_512M. But according to the radar RVP user guide, I should configure the EMIFMODE in cfg.mk. Please correct me if I was wrong. And could you please let me know what is the default value for the EMIFMODE? Do I need to configure it in C:\PROCESSOR_SDK_RADAR_03_08_00_00\vision_sdk\apps\configs\tda3xx_rvp_bios_radar\cfg.mk?

Best,

Frank

Hi Frank,

The default EMIFMODE = SINGLE_EMIF_512MB if not already defined in cfg.mk file.

Are you trying to load Radar SDK binary?

Regards,
Stanley

Yes, I'm trying to load radar sdk and I'm using D3 RVP board with TDA3.

Hi Stanley,

     Got it. Make sense. Thank you. 

Best,

Frank