OpenOCD return error when using GDB to hit hardware breakpoints on hardware that doesn't support them.

[Alan-19950616]

I think it's better to report a warning when this behaviour is detected than to report an error outright.

reporting an error will cause GDB to disconnect

riscv-openocd/src/server/gdb_server.c

Line 1832 in cb2b394

return ERROR_FAIL;

[en-sc]

You mention hardware breakpoints in the name of the issue. However, the line you are referring to is related to processing of software breakpoints:

riscv-openocd/src/server/gdb_server.c

Line 1829 in cb2b394

if (target->rtos && bp_type == BKPT_SOFT) {

Are you sure you have determined the source of the issue correctly?

Please, attach OpenOCD -d3 log of the reproduction if it's not too much trouble.

[Alan-19950616]

Sorry, I got the wrong line number.

riscv-openocd/src/server/gdb_server.c

Line 1859 in cb2b394

retval = gdb_error(connection, retval);

[en-sc]

AFAIU, ERROR_NOT_IMPLEMENTED should be returned here by watchpoint_add().

Please, attach OpenOCD -d3 log so it is possible to figure out why watchpoint_add() does not return ERROR_NOT_IMPLEMENTED in the case in question.

[Alan-19950616]

OpenOCD log

Info : [riscv.cpu] Found 0 triggers
Error: Couldn't find an available hardware trigger.
Error: can't add breakpoint: resource not available

OpenOCD -d3 log

84595 Error: 84593 31200 riscv.c:673 add_trigger(): Couldn't find an available hardware trigger.
84596 Error: 84594 31200 breakpoints.c:93 breakpoint_add_internal(): can't add breakpoint: resource not available
84597 Debug: 84595 31200 gdb_server.c:1495 gdb_error(): Reporting -308 to GDB as generic error

[aap-sc]

ERROR_NOT_IMPLEMENTED is returned if OpenOCD does not implement something (not the target). Here the behavior is as expected. You try to set a HW breakpoint -> you have no available resources -> you get an error. GDB should not disconnect in such a case. Which GDB version are you using?

[Alan-19950616]

GDB version GNU gdb (GDB) 13.2.90.20230712-git

I want to make sure that the error code is #define ERROR_TARGET_RESOURCE_NOT_AVAILABLE (-308) and not #define ERROR_NOT_IMPLEMENTED (-7).

[aap-sc]

I want to make sure that the error code is 

Again, OpenOCD should return ERROR_TARGET_RESOURCE_NOT_AVAILABLE here since no HW resources available. ERROR_NOT_IMPLEMENTED is for the cases when something is not impleneted in OpenOCD itself.

Here is the debug session from gdb 13.2:

(gdb) watch *(int*)0x0
Hardware watchpoint 1: *(int*)0x0
(gdb) c
Continuing.
Warning:
Could not insert hardware watchpoint 1.
Could not insert hardware breakpoints:
You may have requested too many hardware breakpoints/watchpoints.

Command aborted.
(gdb) q

GDB is not disconnected. it simply aborts the "continue" command since it can't set HW watchpoint.

[Alan-19950616]

That's the problem I'm having, and I'm not expressing it clearly.

At this point it is not possible to use commands such as continue or single-step unless the hardware breakpoint is removed, is it possible to just warn without returning an error?

[aap-sc]

At this point it is not possible to use commands such as continue or single-step unless the hardware breakpoint is removed, is it possible to just warn without returning an error?

But that's the intended behavior for GDB. You've explicitly asked to set a HW watchpoint. GDB is unable to do so. GDB is unable to proceed unless you cancel this request.

[TommyMurphyTM1234]

If you're trying to report and analyse an OpenOCD issue then it might make more sense to use OpenOCD standalone and its Telnet interface to interact with the program/target obviating the need for GDB and thus simplifying the diagnosis?

Also, the code that you're referring to in gdb_server.c is target neutral code in OpenOCD so, almost certainly, should not be changed specifically for RISC-V. Such a modification is unlikely to be accepted upstream in any case.

[aap-sc]

I presume that the question is answered. The behavior you have is the expected one. Let us know if you think that there is something that needs to be changed.