[RT1166] Issue with moving data from SDRAM to ocram1/ocram2.

[JaagupAverin]

Hello. We want to move data from sdram to dtcm and/or ocram.

What works

Moving data to dtcm with this overlay works:

/ {
    chosen {
        zephyr,sram = &dtcm;
    };

    /delete-node/ memory@80000000;
};

What doesn't work

However, moving it to ocram does not work:

/ {
    chosen {
        zephyr,sram = &ocram1;
    };

    /delete-node/ memory@80000000;
};

The application crashes early inside z_bss_zero():

#ifdef CONFIG_NOCACHE_MEMORY
z_early_memset(&_nocache_ram_start, 0,
		   (uintptr_t) &_nocache_ram_end
		   - (uintptr_t) &_nocache_ram_start);
#endif

The registers inside z_early_memset() appear okay:

r0      0x20240000   539230208
r1      0x0          0
r2      0x40000      262144
r3      0x20000000   536870912
r4      0xe000ed00   3758157056
r5      0x2024aeec   539274988
r6      0x1632a3     1454755
r7      0x2024aeec   539274988
r8      0x3c6a9559   1013618009
r9      0x30001040   805310528
r10     0xc35a       50010
r11     0x300293c1   805475265
r12     0x2024ae80   539274880
sp      0x20299b0b   539594507   <z_interrupt_stacks+2040>
lr      0x3004a26b   805627243   <z_bss_zero+38>
pc      0x3006b9b2   805741618   <z_early_memset>
xpsr    0x21000000   553648128
fpscr   0x0          0
msp     0x2029a240   539599424
psp     0x20299b0b   539594507
control 0x2          2
faultmask 0x0        0

i.e it is trying to set 0x40000 bytes at 0x20240000 to zero. As per zephyr.map, this is indeed the .nocache area. And as per RM this is a valid ocram1 region:

However from there it jumps into the fault handler. Register dump at this point:

r0       0x20240000    539230208
r1       0x0           0
r2       0x20280000    539492352
r3       0x0           0
r4       0x0000ed00    375817056
r5       0x2024aeec    539274988
r6       0x1632a3      1454755
r7       0x2024aeec    539274988
r8       0x3c6a9559    1013610841
r9       0x30001040    805310528
r10      0xc35a        50010
r11      0x300293c1    805475265
r12      0x2024ae80    539274880
sp       0x2029a240    539599424    <z_main_stack+4096>
lr       0xffffffff    4294967295
pc       0x300293ac    805475244    <z_arm_usage_fault>
xpsr     0x61000006    1627389958
fpscr    0x0           0
msp      0x2029a240    539599424
psp      0x20290998    539594904
control  0x0           0
faultmask 0x0          0
primask   0x0          0
cycles    0x0          0
fpccr     0xc0000000   3221225472
fpcar     0xffffffff8  4294967288
fpdscr    0x0          0
mvfr0     0x10110221   269550113
mvfr1     0x12000011   301989905

What I've also tried

• Redefining OCRAM1 with a 0x10000 offset as some sources suggest this area is reserved after boot; although I do not think this is the case by reaching Zephyr.
• Using OCRAM2;

Question

Are ocram1/ocram2 even supported as zephyr,sram candidates? Nothing in the reference manual stands out to be that they shouldn't be.

Thanks,
Jaagup

[DerekSnell]

Hi @JaagupAverin ,
Is this related to the MPU? What do the fault registers in the ARM CPU tell you is the cause of the fault? You could try disabling the MPU to see if that avoids the issue. I believe CONFIG_ARM_MPU=n will disable the MPU.
Let us know what you find. Best regards

[jacob-wienecke-nxp]

@JaagupAverin

I believe your issue could be caused by an overflow of OCRAM1. From the image you provided from the RM, OCRAM1 with ECC enabled is 64KB. Zephyr Device tree currently thinks OCRAM1 is sized at 512KB (which is true for RT1170, but not for RT1160)

I submitted a PR to address this at: #94363

My assumption based on the error provided is that the size of data is greater than 64KB. I'd recommend shifting the remaining data to DTCM and the other OCRAM regions. Alternatively, you could utilize the FlexRAM (512KB), although this will require some additional steps.

These methods are covered in AN14579, depending on which direction you want to go.