qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Qemu-devel] [ltt-dev] [PATCH 2/3] trace: [ust] Do not use 'm' in ev


From: Stefan Hajnoczi
Subject: Re: [Qemu-devel] [ltt-dev] [PATCH 2/3] trace: [ust] Do not use 'm' in event argument names (used by ust macros)
Date: Wed, 21 Sep 2011 09:28:55 +0100

On Tue, Sep 20, 2011 at 5:22 PM, Mathieu Desnoyers
<address@hidden> wrote:
> * Stefan Hajnoczi (address@hidden) wrote:
>> 2011/9/20 Mathieu Desnoyers <address@hidden>:
>> > * Lluís Vilanova (address@hidden) wrote:
>> >> Stefan Hajnoczi writes:
>> >>
>> >> > On Fri, Sep 16, 2011 at 06:59:38PM +0200, Lluís Vilanova wrote:
>> >> >> Signed-off-by: Lluís Vilanova <address@hidden>
>> >> >> ---
>> >> >> trace-events |    4 ++--
>> >> >> 1 files changed, 2 insertions(+), 2 deletions(-)
>> >> >>
>> >> >> diff --git a/trace-events b/trace-events
>> >> >> index 9d1fbbb..b653d70 100644
>> >> >> --- a/trace-events
>> >> >> +++ b/trace-events
>> >> >> @@ -418,8 +418,8 @@ milkymist_pfpu_pulse_irq(void) "Pulse IRQ"
>> >> >> # hw/milkymist-softusb.c
>> >> >> milkymist_softusb_memory_read(uint32_t addr, uint32_t value) "addr 
>> >> >> %08x value %08x"
>> >> >> milkymist_softusb_memory_write(uint32_t addr, uint32_t value) "addr 
>> >> >> %08x value %08x"
>> >> >> -milkymist_softusb_mevt(uint8_t m) "m %d"
>> >> >> -milkymist_softusb_kevt(uint8_t m) "m %d"
>> >> >> +milkymist_softusb_mevt(uint8_t _m) "m %d"
>> >> >> +milkymist_softusb_kevt(uint8_t _m) "m %d"
>> >>
>> >> > The LTTng community has been very responsive in addressing namespace
>> >> > issues with libust.  Let's post more details and see if it can be fixed
>> >> > in libust.
>> >>
>> >> > Could you please post your gcc and libust versions?
>> >>
>> >> > I have not been able to reproduce the problem on Debian libust-dev
>> >> > 0.15-3.  My gcc version is Debian gcc 4.6.1-4.
>> >>
>> >> Yup, I was using libust-dev 0.5 (debian stable). After switching to 0.15 
>> >> (debian
>> >> testing) other problems arise.
>> >>
>> >> For example, in my box compiling osdep.c yields lots of "variable 
>> >> ‘__tp_cb_data’
>> >> set but not used", but using a minimal example does not raise such 
>> >> problems, so
>> >> I'm assuming this is related to some namespace problems related to some 
>> >> other
>> >> code in qemu's headers.
>> >
>> > If you can find a few minutes to provide:
>> >
>> > 1) a link to the git repository/commit ID you are testing
>> > 2) the exact configuration you use (detailed way to reproduce the
>> >   problem, as a sequence of commands from a pristine repository)
>> >
>> > It will allow me to look into this warning and fix it.
>>
>> AFAICT the only problem with libust is the __tp_cb_data set but not
>> used warning that gcc 4.6 emits, see my test program:
>>
>> #include <stdint.h>
>> #include <ust/tracepoint.h>
>>
>> DECLARE_TRACE(ust_milkymist_softusb_mevt, TP_PROTO(uint8_t m),
>> TP_ARGS(m));
>>
>> int main(int argc, char **argv)
>> {
>>        return 0;
>> }
>>
>> You will get an error with gcc 4.6 and libust 0.15:
>>
>> $ gcc -o a -Wall a.c
>
> Ah, I see. This is due to the backward compatibility API I added in
> 0.15 for qemu. It skips using the __tp_cb_data private data. So it does
> make sense that the compiler complains about this. I can add a
> __attribute__((unused)), as proposed by Lluís.
>
>>
>> QEMU itself does have one problem with libust 0.15.  We do
>> DECLARE_TRACE(name, TP_PROTO(void), TP_ARGS()).  This results in a
>> compiler error but we can fix this my changing our tracepoints to use
>> DECLARE_TRACE(name, TP_PROTO(), TP_ARGS()) I think.  Perhaps libust's
>> behavior here has changed, it shouldn't be hard to update QEMU's
>> tracetool tracepoint generator though.
>
> For tracepoints without arguments, you need to use something like:
>
> TRACEPOINT_EVENT_NOARGS(name, TP_FIELDS())
>
> (note: the full support of TP_FIELDS() declaration will come in UST 2.0.
> For UST 0.15, it's just ignored)
>
> The _NOARGS variant lets you handle the (void) 0 parameter case.

Okay, this is the fault of QEMU's tracetool script, which generates
the LTTng UST tracepoints.  We can move to the new TRACEPOINT*()
macros and use _NOARGS as appropriate.

Stefan



reply via email to

[Prev in Thread] Current Thread [Next in Thread]