[PATCH -tip tracing/kprobes] PPC: Powerpc port of the kprobe-based event tracer

[Michael Neuling]

In message <4B317324.3000708 at redhat.com> you wrote:
> Michael Neuling wrote:
> > In message <4B2B934C.1060807 at redhat.com> you wrote:
> >>
> >>
> >> Mahesh Jagannath Salgaonkar wrote:
> >>> Michael Neuling wrote:
> >>>> In message <4B29EE5F.9020801 at linux.vnet.ibm.com> you wrote:
> >>>>> Hi Michael,
> >>>>>
> >>>>> Michael Neuling wrote:
> >>>>>>> + * regs_get_argument_nth() - get Nth argument at function call
> >>>>>>> + * @regs:    pt_regs which contains registers at function entry.
> >>>>>>> + * @n:        argument number.
> >>>>>>> + *
> >>>>>>> + * regs_get_argument_nth() returns @n th argument of a function call
.
> >>>>>>> + * Since usually the kernel stack will be changed right after
> >>>>>>> function en
> >>>> try
> >>>>>> ,
> >>>>>>> + * you must use this at function entry. If the @n th entry is NOT
> >>>>>>> in the
> >>>>>>> + * kernel stack or pt_regs, this returns 0.
> >>>>>>> + */
> >>>>>>> +unsigned long regs_get_argument_nth(struct pt_regs *regs, unsigned
> >>>>>>> int n)
> >>>>>>> +{
> >>>>>>> +    if (n < ARRAY_SIZE(arg_offs_table))
> >>>>>>> +        return *(unsigned long *)((char *)regs + arg_offs_table[n]);
> >>>>>>> +    else {
> >>>>>>> +        /*
> >>>>>>> +         * If more arguments are passed that can be stored in
> >>>>>>> +         * registers, the remaining arguments are stored in the
> >>>>>>> +         * parameter save area located at fixed offset from stack
> >>>>>>> +         * pointer.
> >>>>>>> +         * Following the PowerPC ABI, the first few arguments are
> >>>>>>> +         * actually passed in registers (r3-r10), with equivalent
> >>>>>>> space
> >>>>>>> +         * left unused in the parameter save area.
> >>>>>>> +         */
> >>>>>>> +        n += (PARAMETER_SAVE_AREA_OFFSET / sizeof(unsigned long));
> >>>>>>> +        return regs_get_kernel_stack_nth(regs, n);
> >>>>>> How do we handle FP args?
> >>>>> Currently this patch does not support FP args.
> >>>>
> >>>> This might be OK.  I don't think we use floating point parameters in any
> >>>> function definitions in the kernel. 
> >>>> We do use altivec in the raid6 driver (drivers/md/raid6altivec.uc) but
> >>>> they are static inline, so they probably don't even end up as
> >>>> functions. 
> >>>> I guess we need to make sure that we're not limiting the interface in
> >>>> such a way that we can't support it later if the above changes. 
> >>>> regs_get_argument_nth returns an unsigned long which makes returning a
> >>>> 128 bit VMX register impossible.  This might be a show stopper for me.
> >>>> How are the x86 guys dealing with this?
> >>>>
> >>> Nope, x86 does not deal with bigger registers (Masami, correct me if I
> >>> am wrong). The return data type is opaque to user. Hence this enables us
> >>> to handle any such situations in future without effecting user space API.
> >>
> >> Right, we don't use those bigger registers in the kernel context.
> >> (some special functions use it inside, but most of codes
> >>  are just using general regs)
> > 
> > Sure, but kernel interfaces are for now.  What happens if this changes in
> > the future?
> > 
> >> And regs_get_argument_nth is just an accessor of pt_regs field.
> >> This means that it just provides field in pt_regs.
> > 
> > Sure, that's how it's implemented, but that's not what the name of the
> > function suggests it does.
> 
> Hmm, OK, maybe I see what you think.
> Actually, this function doesn't cover all functions in the kernel
> even on x86, because each ABI depends on how function is defined,
> e.g. func(fmt,...) and asmlinkage(i386) function will put all
> arguments on the stack.
> 
> Originally, I had introduced this function only for helping user
> of kprobe-tracer who wants to trace function arguments without
> any arch specific knowledge. However, now we have perf-probe which
> can analyze debuginfo to find function arguments and local variables.
> So, this function is a bit out-of-dated :-)
> 
> How about removing this function from this patch? I can also
> update kprobe-tracer to drop the function argument support.
> Instead of that, I think it is enough to add a description of
> how to get function arguments in Documentation/trace/kprobetrace.txt

Sounds like a better solution to me.  

Cheers,
Mikey

> Thank you,
> 
> > 
> > Mikey
> > 
> >>
> >>>>>>> +    }
> >>>>>>> +}
> >>>>>>> +/*
> >>>>>>>   * does not yet catch signals sent when the child dies.
> >>>>>>>   * in exit.c or in signal.c.
> >>>>>>>   */
> >>>>>>> Index: linux-2.6-tip/kernel/trace/Kconfig
> >>>>>>> ===================================================================
> >>>>>>> --- linux-2.6-tip.orig/kernel/trace/Kconfig
> >>>>>>> +++ linux-2.6-tip/kernel/trace/Kconfig
> >>>>>>> @@ -464,7 +464,7 @@ config BLK_DEV_IO_TRACE
> >>>>>>>  
> >>>>>>>  config KPROBE_EVENT
> >>>>>>>      depends on KPROBES
> >>>>>>> -    depends on X86
> >>>>>>> +    depends on X86 || PPC
> >>>>>>>      bool "Enable kprobes-based dynamic events"
> >>>>>>>      select TRACING
> >>>>>>>      default y
> >>>>>>>
> >>>>>>> _______________________________________________
> >>>>>>> Linuxppc-dev mailing list
> >>>>>>> Linuxppc-dev at lists.ozlabs.org
> >>>>>>> https://lists.ozlabs.org/listinfo/linuxppc-dev
> >>>>>>>
> >>>>> Thanks for reviewing.
> >>>>
> >>>> We are creating a new user space API here, so I'm keen for others to tak
e
> >>>> a good look at the interface before we commit to something we are going
> >>>> to have to keep forever. 
> >>>> Who is the main consumer of this (/me is pretty ignorant of kprobes)?
> >>>> What do they think of the interface?
> >>>>
> >>> The user space API are already present in the upstream kernel and
> >>> currently only supported architecture is x86. This patch provides ppc
> >>> architecture specific interfaces that enables powerpc also in par with x8
6.
> >>
> >> Yes, there is a kprobe tracer in ftrace (see Documentation/trace/kprobetra
ce.
> > txt).
> >> and this tracer is only for probing kernel (not userspace).
> >>
> >>>
> >>> The main consumer would be kernel developers who would like to see
> >>> register values, arguments and stack when the probe hits at given text
> >>> address.
> >>
> >> Right.
> >>
> >> BTW, there is a user-space tools we have (tools/perf/builtin-probe.c).
> >> Currently, it's only for x86. Mahesh, You just need to add a register
> >> translation table in tools/perf/util/probe-finder.c for ppc support.
> >>
> >> Thank you!
> >>
> >> -- 
> >> Masami Hiramatsu
> >>
> >> Software Engineer
> >> Hitachi Computer Products (America), Inc.
> >> Software Solutions Division
> >>
> >> e-mail: mhiramat at redhat.com
> >>
> 
> -- 
> Masami Hiramatsu
> 
> Software Engineer
> Hitachi Computer Products (America), Inc.
> Software Solutions Division
> 
> e-mail: mhiramat at redhat.com
>