System crashed and throws AXI BUS ERROR

[Arunachalam0810]

Hi,

We are working in IMX8 chip(6.1 kernel) and having an issue which throws log as AXI BUS  ERROR and shows dma stuck at some address, logs are shown below. This issue occurs while surfing UI along with some log storage. (Crashdump attached)

Any ideas on this issue please?

Error logs are as below,

 AXI BUS ERROR

 **************************

***   GPU DRV CONFIG   ***

**************************

Galcore version 6.4.11.608260

Galcore options:

bankSize          = 0x00000000

fastClear         = -1

compression       = 15

powerManagement   = 1

baseAddress       = 0x40000000

physSize          = 0xC0000000

recovery          = 0

stuckDump         = 0

GPU smallBatch    = 1

allMapInOne       = 1

enableNN          = 0xff

userClusterMasks  =

ff,

ff,

ff,

ff,

ff,

ff,

ff,

ff,

ff,

ff,

[  232.334437] ff,

[  232.336286] ff,

[  232.338140] ff,

[  232.339983] ff,

ff,

ff,

ff,

ff,

ff,

ff,

irqs              =

26,

irq2Ds            =

27,

System reserve memory configuration:

contiguousSizes[0] = 0x10000000

contiguousBases[0] = 0x100000000

Registers configuration:

[  232.385909]   bases[0] = 0x38000000, sizes[0] = 0x8000,

[  232.385924]

2D bases[0] = 0x38008000, sizes[0] = 0x8000,

External sRAMBases =

0xffffffffffffffff,

0xffffffffffffffff,

mmu               = 1

mmuPageTablePool  = 1

mmuCmdPool        = 1

mmuDynamicMap     = 1

isrPoll           = 0x00000000

Build options:

gpuTimeout    = 30000

gcdGPU_2D_TIMEOUT = 20000

gcdINTERRUPT_STATISTIC = 1

GPU[0](ChipModel=0x7000 ChipRevision=0x6204):

**************************

***   GPU STATE DUMP   ***

**************************

axi      = 0x0000007F

idle     = 0x7FFFFFFE

 FE not idle

 AQ_HI_CLOCK_CONTROL  = 0x00040100

 DMA appears to be stuck at this address:

 0x010003E0

 dmaLow   = 0x00000000

 dmaHigh  = 0x00000000

 dmaState = 0x00000000

 command state       = 0 (PAR_IDLE_ST)

 command DMA state   = 0 (CMD_IDLE_ST)

 command fetch state = 0 (FET_IDLE_ST)

 DMA request state   = 0 (REQ_IDLE_ST)

 cal state           = 0 (CAL_IDLE_ST)

 VE request state    = 0 (VER_IDLE_ST)

 Debug registers:

 RA[0] debug registers:

 [00] 00000000 00000000 00000000 00000000 031F0000 02000170 048D8630 68508040

[AldoG]

Hello,

Could you share the exact i.MX8 that you are using?
Also, please share the OS that you are running and the version.

Thank you,
Saludos,
Aldo.

[Arunachalam0810]

Hi Aldo,

Thanks for quick reply.

Chip - imx8mp6

Model: MSC SM2S-IMX8MPLUS

Linux version - 6.1.1

root@ARIA123456:~# uname -a
Linux ARIA123456 6.1.1-nxp+6.1.1+1.0.0+ge-mic-aria+g2d6653ea4978 #1 SMP PREEMPT Wed Nov  8 16:36:39 UTC 2023 aarch64 GNU/Linux

Thanks,

Arun

[AldoG]

Hello,

Thank you for sharing more information, I would suggest increasing the CMA size you have defined in your device tree, usually we recommend having at least 25% of the available DDR.

Please try it and let me know of the results,
Saludos,
Aldo.

[hblee]

I'm having exactly the same problem as this issue.
After an inference operation of a self-developed model in pytorch, a crash occurs in Weston. (when HDMI is connected)
The Linux kernel version is 6.1.36.
RAM is 2GB and CMA is 512MB.
Interestingly, when I set up the IMX8MP-EVK board with the same size of RAM and CMA (change boot and kernel dts), no problem occurs.