In my custom i.mx27 based board am having data_abort exceptions. As I could find from arm library code, a simple dump showing last register values should happen to the console, since it is coded so in data_abort handler. So, why is not i.mx27 triggering the data_abort exception handler?

I did some tests on i.mx27 litekit board, where I forced a data_abort exception trying to access to a NULL pointer. What I found is that dump wasn't shown on this board neither.

Then, I disassembled code in the internal ROM of the i.MX27 where lowest level exception vectors should be placed and I found the following:

0x00000000 0xe59ff00c LDR r15, [r15, #0xc] 0x00000004 0xe51ff11c LDR r15, [r15, #-0x11c] 0x00000008 0xe51ff11c LDR r15, [r15, #-0x11c] 0x0000000c 0xe51ff11c LDR r15, [r15, #-0x11c] 0x00000010 0xe51ff11c LDR r15, [r15, #-0x11c] 0x00000014 0xc0000000 ANDGT r0, r0, r0 0x00000018 0xe51ff120 LDR r15, [r15, #-0x120] 0x0000001c 0xe51ff120 LDR r15, [r15, #-0x120]

This means that u-boot exception vector address should be located in the internal vRAM (0xFFFF_4C00 - 0xFFFF_FFFF). I solved the problem adding an assembler macro in lowlevel_init.S for the imx27lite board:

-- diff --git a/board/logicpd/imx27lite/lowlevel_init.S b/board/logicpd/imx27lite/lowlevel_init.S index e2cdecb..8419471 100644 --- a/board/logicpd/imx27lite/lowlevel_init.S +++ b/board/logicpd/imx27lite/lowlevel_init.S @@ -40,6 +40,19 @@ SDRAM_LOADMODE_CMD_W: .word (ESDCTL_SDE | ESDCTL_SMODE_LOAD_MODE | \ ESDCTL_ROW13 | ESDCTL_COL10) SDRAM_NORMAL_CMD_W: .word SDRAM_ESDCTL_REGISTER_VAL

+.macro init_vram_vectors + /* TODO: + * - Find a proper place to do this for all i.mx27 chips. + * - Calculate Exception vector address in RAM using some parameter. + */ + write32 0xfffffef0, 0xa7f00004 /* Undefined interrupt handler */ + write32 0xfffffef4, 0xa7f00008 /* Software interrupt handler */ + write32 0xfffffef8, 0xa7f0000c /* Prefetch abort handler */ + write32 0xfffffefc, 0xa7f00010 /* Data abort handler */ + write32 0xffffff00, 0xa7f00018 /* Default IRQ handler */ + write32 0xffffff04, 0xa7f0001c /* Default FIQ handler */ +.endm /* init vram vectors */ + .macro init_aipi /* * setup AIPI1 and AIPI2 @@ -167,4 +180,6 @@ lowlevel_init:

sdram_init

+ init_vram_vectors + mov pc,r10 --

I am aware that this patch is not probably the proper way of doing this but I'd like to know:

Has anyone got the same problem? Please, try to force and exception and see if a dump is generated, if not, apply this patch and confirm results.

If this is a generalized problem, which would be a good place to do it?

Comments are much appreciated. Thank you.