OpenOCD Failed to read memory

chivy · August 2, 2022, 7:27am

I use the OpenOCD to debug my kernel. The OpenOCD connect to board succeed, I target remote my board, Then use ‘file kernel.elf’, OpenOCD get the error messages:

Info : accepting 'gdb' connection on tcp/3333
Info : New GDB Connection: 1, Target riscv.cpu, state: halted
Warn : Prefer GDB command "target extended-remote 3333" instead of "target remote 3333"
Warn : Failed to read memory via program buffer.
Warn : Failed to read memory via abstract access.
Error: Target riscv.cpu: Failed to read memory (addr=0x49883ae0ce8bfd4e)
Error:   progbuf=failed, sysbus=skipped (unsupported size), abstract=failed
Error: Target riscv.cpu: Failed to read memory (addr=0x49883ae0ce8bfd4e)
Error:   progbuf=failed, sysbus=skipped (unsupported size), abstract=failed

I find the issue:

github.com/riscv/riscv-openocd

src/target/riscv/riscv-013.c:read_memory() failure corrupts x8/s0/fp register

opened 11:48AM - 09 Dec 20 UTC

closed 08:39PM - 14 Dec 20 UTC

TommyMurphyTM1234

During debugging I found that the frame pointer (x8/s0/fp), which is also used b…y the debugger for certain operations, was getting corrupted. I added some logging to the above file and see this: ``` Info : read_memory:3560 Begin x8/s0/fp = 0x8013000 Warn : Failed to read memory via program buffer. Warn : Failed to read memory via abstract access. Error: Target mpfs.hart0_e51: Failed to read memory (addr=0x500000000000000) Error: progbuf=failed, sysbus=skipped (unsupported size), abstract=failed Info : read_memory:3621 End (failure) x8/s0/fp = 0x500000000000000 ``` So it seems that a failed memory read leaves the frame pointer corrupted with obvious concomitant negative impact on further execution of the program being debugged. I guess that at least one code path fails to restore the x8/s0/fp register used by openocd to do some work. There are several places where it saves and restores the original value but in this case I guess that the restore never happens. One obvious solution would be to save it at the start of read_memory() and restore it at the end (success and failure cases?) but maybe this is not the appropriate place? @timsifive - any suggestions before I submit a patch/PR? Thanks. Thanks a lot Cheers Tommy Edit: the reason for the attempted read of memory at 0x0500_0000_0000_0000 is that the stack contains a pointer variable which has not been initialized yet and happens to assume that value on entry and (in Eclipse) when the Variables view is active or the user mouses over the variable in the Editor view the debugger attempts to read that memory. Edit 2: I don't know if the same or a similar issue might also affect riscv-011.c?

And this issue had been solved by OpenOCD:

It seems that SiFive OpenOCD didn’t merge this branch, this issue repaired on Dec 15, 2020. SiFive OpenOCD only provide the bin file, the last time is on Dec 1, 2020.

RalphF · August 4, 2022, 6:13pm

Hi Chivy,

The problem you’re experiencing might be caused by the issue you identified, but we can’t be certain. We don’t have a newer version of OpenOCD available publicly, so we suggest rebuilding OpenOCD yourself from the following repo:

GitHub - riscv/riscv-openocd: Fork of OpenOCD that has RISC-V support

If you are a SiFive IP customer, please email support@sifive.com from your employer’s email domain and we can get you a newer version of the tools.

Regards,
Ralph

Topic		Replies	Views
Debugging a program through openOCD RISC-V	16	10684	September 3, 2019
OPENOCD error HiFive1 Rev B	9	4082	June 10, 2021
Socket Problem HiFive1 Rev B	0	1763	July 28, 2020
Debug Linux Kernel with OpenOCD & GDB HiFive Unmatched	5	3541	August 3, 2022
Problem with Debug Connection running openocd HiFive Unmatched	1	2264	July 30, 2021

OpenOCD Failed to read memory

Related Topics