----- Original Message ----- > From: "Ben Hutchings" <ben@xxxxxxxxxxxxxxx> > To: "Dirk Behme" <dirk.behme@xxxxxxxxx> > Cc: "Kamal Mostafa" <kamal@xxxxxxxxxxxxx>, "gregkh" <gregkh@xxxxxxxxxxxxxxxxxxx>, "Mathieu Desnoyers" > <mathieu.desnoyers@xxxxxxxxxxxx>, "stable" <stable@xxxxxxxxxxxxxxx>, "Luis Henriques" > <luis.henriques@xxxxxxxxxxxxx>, "Jiri Slaby" <jslaby@xxxxxxx>, "Roman Pen" <r.peniaev@xxxxxxxxx>, "Steven Rostedt" > <rostedt@xxxxxxxxxxx>, "Frederic Weisbecker" <fweisbec@xxxxxxxxx>, "Ingo Molnar" <mingo@xxxxxxxxxx>, "Jens Axboe" > <axboe@xxxxxx> > Sent: Saturday, May 3, 2014 2:33:56 PM > Subject: Re: "blktrace: fix accounting ..." breaks lttng API in -stable trees > > On Sat, 2014-05-03 at 20:08 +0200, Dirk Behme wrote: > > Am 03.05.2014 18:43, schrieb Ben Hutchings: > > > On Sat, 2014-05-03 at 18:35 +0200, Dirk Behme wrote: > > >> Am 02.05.2014 20:09, schrieb Kamal Mostafa: > > >>> On Fri, 2014-05-02 at 13:30 -0400, gregkh wrote: > > >>>> On Fri, May 02, 2014 at 10:07:33AM -0700, Kamal Mostafa wrote: > > >>>>> Dirk Behme points out that this "Cc: stable" commit breaks the > > >>>>> lttng-modules userspace API when applied to stable kernels. Stable > > >>>>> versions 3.2, 3.8, 3.11, and 3.13 (at least) have all queued it: > > >>>>> > > >>>>> af5040da01ef980670b3741b3e10733ee3e33566 > > >>>>> blktrace: fix accounting of partially completed requests > > >>>>> > > >>>>> > > >>>>> On Thu, 2014-05-01 at 10:28 +0200, Dirk Behme wrote: > > >>>>>> [...] might break the build of the user space lttng-modules > > >>>>>> (lttng-probe-block.c) due the the API change of > > >>>>>> trace_block_rq_complete(). > > >>>>> > > >>>>>> [...] On the other hand, looking into the lttng-modules git > > >>>>>> http://git.lttng.org/?p=lttng-modules.git;a=commitdiff;h=1c53e689434a6bdd7dc3f54c07bfb72750d1d24c > > >>>>>> looks like this is the necessary user space adaption to the kernel > > >>>>>> change? So this looks like that lttng-modules expects a > > >>>>>> KERNEL_VERSION > > >>>>>>> = (3,15,0) to have this commit? > > >>>>> > > >>>>> > > >>>>> My inclination is that we probably need to revert/drop "af5040d > > >>>>> blktrace: fix accounting..." from the stable kernels to unbreak the > > >>>>> userspace API. > > >>>> > > >>>> Then you will run into this issue with 3.15, when it is released. > > >>> > > >>> No, I think "#if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,15,0))" in the > > >>> lttng-modules commit referenced above guards against that. Apparently, > > >>> lttng-modules expects to see the new API in >= 3.15 and the old API in > > >>> the stable kernels. > > >> > > >> I'm not in the position to judge if it's a lttng or a kernel community > > >> issue ;) > > >> > > >> So just for my understanding: > > >> > > >> Having > > >> > > >> #if (LINUX_VERSION_CODE >= KERNEL_VERSION(3,15,0)) > > >> > > >> in lttng-modules commit > > >> > > >> http://git.lttng.org/?p=lttng-modules.git;a=commitdiff;h=1c53e689434a6bdd7dc3f54c07bfb72750d1d24c > > >> > > >> does mean that we don't have a lttng-modules version which will build > > >> against the -stable kernels (3.2.58, 3,8.13.22 etc) with the back > > >> ported commit [1], atm? > > > [...] > > > > > > I don't understand why lttng-modules has its own definitions of > > > tracepoints. However, I suspect that it doesn't use anything beyond the > > > event structure definition, which has *not* changed, making that commit > > > a no-op. > > > > > > Someone who cares should actually experiment with using old > > > lttng-modules with blktrace eventsf rom a patched kernel. > > > > Not sure if this helps, but using -stable kernel 3.2.58 and recent > > lttng-modules.git: > [...] > > CC [M] lttng-modules.git/probes/lttng-probe-block.o > [...] > > Error: Conflicting types for »trace_block_rq_complete« > > In file included from lttng-modules.git/probes/lttng-probe-block.c:31:0: > > include/trace/events/block.h:123:1: Note: Previous definition of > > »trace_block_rq_complete« was here > > make[3]: *** [lttng-modules.git/probes/lttng-probe-block.o] Error 1 > [...] > > So it's rebuilding the OOT module that fails, not trying to run already > existing userland. That's just tough; module API changes on a stable > branch are allowed and have happened many times. (Distributions tend to > care a bit more about this.) I'm currently going back starting from 3.2.xx stable kernels onward to add the missing ifdefs in LTTng to follow the module API changes within those stable branches. Indeed, LTTng being an OOT module will adapt to those upstream changes. I'll push the fixes into LTTng master and stable-2.4 branches. Thanks, Mathieu > > Ben. > > -- > Ben Hutchings > All the simple programs have been written, and all the good names taken. > -- Mathieu Desnoyers EfficiOS Inc. http://www.efficios.com -- To unsubscribe from this list: send the line "unsubscribe stable" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html