[haiku-bugs] Re: [Haiku] #6139: [PATCH] ppc: Implement arch_debug_get_caller
- From: "bonefish" <trac@xxxxxxxxxxxx>
- Date: Sun, 06 Jun 2010 13:46:07 -0000
#6139: [PATCH] ppc: Implement arch_debug_get_caller
----------------------------+-----------------------------------------------
Reporter: andreasf | Owner: axeld
Type: enhancement | Status: closed
Priority: normal | Milestone: R1
Component: System/Kernel | Version: R1/Development
Resolution: fixed | Keywords:
Platform: PowerPC | Blockedby:
Patch: 1 | Blocking:
----------------------------+-----------------------------------------------
Changes (by bonefish):
* status: new => closed
* resolution: => fixed
Comment:
Replying to [comment:5 andreasf]:
> Right, it would've returned the requesting function as caller.
>
> As can be seen in [comment:ticket:6129:5 #6129], {{{stack_trace}}}
itself is included in the stack trace. This is remedied accordingly in
patch 2.
Patches applied in r37032 and r37033.
> Ideally I would like to see everything including {{{panic}}} vanish from
the regular stack trace, but that would seem to require major kernel
debugger API refactoring, pushing the callsite of
{{{debug_call_with_fault_handler}}} to arch-specific code to strip known-
preceding frames.[[BR]]
Actually it can be done significantly easier. If you look at
enter_kernel_debugger() and debug_trap_cpu_in_kdl(), you'll see that they
call arch_debug_save_registers(), which stores the registers state at that
point (not implemented for PPC yet). The x86 debug code already uses that
for stack traces on other CPUs. I haven't enabled it for the current CPU
yet mostly out of concern that any problem in that code would totally
break stack traces, though I guess that's really a non-issue. That would
already simplify stack traces somewhat. The remaining superfluous stack
frames could be skipped, e.g. by passing the number of those stack traces
to arch_debug_save_registers() or saving it directly in the structure.
Alternatively the outmost kernel debugger function (panic() or
kernel_debugger()) could get its caller, which would be stored and serve
as a marker for skipping stack frames. Or finally the complete range of
kernel debugger code could be skipped at the beginning of the stack trace.
> Inside ppc it might be feasible to split {{{stack_trace}}} in one
function with {{{stack_frame*}}} argument and a wrapper with the present
{{{int argc, char** argv}}} args to handle the
{{{evaluate_debug_command}}} case.
The same goes for x86, too. OTOH, arch_debug_stack_trace() is a kernel
debugger function. Printing the same stack trace as the `bt` should be the
desired behavior anyway. ATM the stack trace printed automatically on a
panic differs from the one `bt` prints, which is seriously annoying,
because it also doesn't match the `call` stack frame numbering.
--
Ticket URL: <http://dev.haiku-os.org/ticket/6139#comment:6>
Haiku <http://dev.haiku-os.org>
Haiku - the operating system.
Other related posts:
