On Fri, Feb 03, 2023 at 11:55:20PM -0300, Leonardo Brás wrote:
> On Fri, 2023-02-03 at 10:55 -0800, Paul E. McKenney wrote:
> > On Fri, Feb 03, 2023 at 03:45:05PM -0300, Leonardo Brás wrote:
> > > On Fri, 2023-02-03 at 10:19 -0800, Paul E. McKenney wrote:
> > > > On Fri, Feb 03, 2023 at 02:32:35PM -0300, Leonardo Brás wrote:
> > > > > On Fri, 2023-02-03 at 08:47 -0800, Paul E. McKenney wrote:
> > > > > > On Sat, Feb 04, 2023 at 12:23:25AM +0900, Akira Yokosawa wrote:
> > > > > > > Under shallow clone of perfbook repo, "git describe" can't work.
> > > > > > > Catch the error and use "git log" to obtain the commit id instead.
> > > > > > >
> > > > > > > Following is the comparison of autodate.tex.
> > > > > > >
> > > > > > > Before:
> > > > > > >
> > > > > > > \date{February 3, 2023 \\ Tag: \texttt{} }
> > > > > > > \newcommand{\commityear}{2023}
> > > > > > > \newcommand{\commitid}{}
> > > > > > > \IfQqzBg{}{\setboolean{qqzbg}{false}}
> > > > > > > % tcolorbox version: 5.1
> > > > > > > \tcbsetforeverylayer{autoparskip}
> > > > > > >
> > > > > > > After:
> > > > > > >
> > > > > > > \date{February 3, 2023 \\ Commit: \texttt{g25bd138c30bf} (shallow clone) }
> > > > > > > \newcommand{\commityear}{2023}
> > > > > > > \newcommand{\commitid}{g25bd138c30bf}
> > > > > > > \IfQqzBg{}{\setboolean{qqzbg}{false}}
> > > > > > > % tcolorbox version: 5.1
> > > > > > > \tcbsetforeverylayer{autoparskip}
> > > > > > >
> > > > > > > Signed-off-by: Akira Yokosawa <akiyks@xxxxxxxxx>
> > > > > > > Cc: Leonardo Bras <leobras.c@xxxxxxxxx>
> > > > > >
> > > > > > Thank you, Akira, and I have queued this.
> > > > > >
> > > > > > Leonardo, could you please check to see whether this works for you in
> > > > > > your particular environment?
> > > > > >
> > > > >
> > > > > Sure,
> > > > > I reconfigured it as a shallow git clone, and then pushed master + this patch on
> > > > > leobras_ci, and it seems to have worked just fine. Please see:
> > > > > https://gitlab.com/linux-kernel/perfbook/-/jobs/3712415008/artifacts/file/perfbook.pdf
> > > > >
> > > > >
> > > > > > If it does, please include something
> > > > > > like this in your reply:
> > > > > >
> > > > > >
> > > > > >
> > > > >
> > > > > So, yeah:
> > > > > Tested-by: Leonardo Bras <leobras.c@xxxxxxxxx>
> > > >
> > > > Applied, thank you very much!
> > > >
> > > > > Paul, could you please help me test some other feature?
> > > > > It would require you to push master + this patch on a test branch in your
> > > > > kernel.org repository.
> > > > >
> > > > > If everything works as intended, it should automatically trigger the build +
> > > > > testing for that branch.
> > > >
> > > > One approach is to go into the CodeSamples/count directory and invoke
> > > > all the programs with no arguments. If they all exit with zero status,
> > > > life is good.
> > >
> > > I think there was a misunderstanding:
> > > I asked help for testing a Gitlab feature, which is buiding any test branches
> > > from your kernel.org repository automatically when you push to a new branch.
> > >
> > > >
> > > > Would you be interested in adding a "make check" target to the
> > > > CodeSamples/count/Makefile to support this?
> > >
> > > Sure, I could do that!
> > > Just noticed by running `make -C CodeSamples/` it doesn't generate anything in
> > > CodeSamples/count:
> > >
> > > ### None in count/ can be built on arch: 'unknown'.
> > >
> > > I did some hacking, and then tried:
> > > make target=x86 -C CodeSamples/
> > >
> > > And suddenly I see errors such as:
> > >
> > > cc -g -o lockdeq -DTEST lockdeq.c -lpthread
> > > lockdeq.c:78:26: error: expected ‘:’, ‘,’, ‘;’, ‘}’ or ‘__attribute__’ before
> > > ‘____cacheline_internodealigned_in_smp’
> > > 78 | spinlock_t rlock ____cacheline_internodealigned_in_smp;
> > > | ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
> > > lockdeq.c: In function ‘init_pdeq’:
> > > lockdeq.c:100:12: error: ‘struct pdeq’ has no member named ‘ridx’; did you mean
> > > ‘lidx’?
> > > 100 | d->ridx = 1;
> > > | ^~~~
> > > | lidx
> > >
> > > and warnings such as :
> > > cc -g -Wall -o xchglock xchglock.c -lpthread
> > > xchglock.c: In function ‘main’:
> > > xchglock.c:87:9: warning: implicit declaration of function ‘smp_mb’; did you
> > > mean ‘cmm_mb’? [-Wimplicit-function-declaration]
> > > 87 | smp_mb();
> > > | ^~~~~~
> > > | cmm_mb
> > > /usr/bin/ld: /tmp/ccDvA7Yo.o: in function `main':
> > > /dev/shm/perfbook/CodeSamples/locking/xchglock.c:87: undefined reference to
> > > `smp_mb'
> > > /usr/bin/ld: /dev/shm/perfbook/CodeSamples/locking/xchglock.c:91: undefined
> > > reference to `smp_mb'
> > > /usr/bin/ld: /dev/shm/perfbook/CodeSamples/locking/xchglock.c:93: undefined
> > > reference to `smp_mb'
> > > /usr/bin/ld: /dev/shm/perfbook/CodeSamples/locking/xchglock.c:95: undefined
> > > reference to `smp_mb'
> > > collect2: error: ld returned 1 exit status
> > > make[1]: *** [Makefile:21: xchglock] Error 1
> > >
> > > Is that the correct way of calling the make on CodeSamples?
> >
> > Ah, the top-level Makefile has some historical baggage, which is entirely
> > my doing. Something about me not expecting this to be a publicly
> > available project when I first set it up. ;-)
> >
> > So I suggest something like this:
> >
> > cd CodeSamples
> > make pthreads
> > cd count
> > make
> >
> > Please let me know how this goes.
>
> No luck:
>
> ###
> [perfbook]$ cd CodeSamples/
> [CodeSamples]$ make pthreads
> echo "#ifndef __PERFBOOK_API_H__" > api.h
> echo "#define __PERFBOOK_API_H__" >> api.h
> echo "/* MECHANICALLY GENERATED, DO NOT EDIT!!! */" >> api.h
> echo "" >> api.h
> cat linux/common.h >> api.h
> echo "" >> api.h
> cat api-pthreads/api-pthreads.h | \
> sed '/begin{snippet}/d' | \
> sed '/end{snippet}/d' >> api.h
> echo "" >> api.h
> cat api-pthreads/api-gcc.h >> api.h
> cat api-pthreads/api-linux.h >> api.h
> echo "" >> api.h
> if test -f /usr/include/urcu-call-rcu.h -o \
> -f /usr/local/include/urcu-call-rcu.h -o \
> -f /usr/include//urcu-call-rcu.h -o \
> -f /usr/include/urcu-call-rcu.h ; \
> then \
> echo "#define _LGPL_SOURCE" >> api.h; \
> echo "#include <urcu/rculist.h>" >> api.h; \
> echo "#include <urcu/rcuhlist.h>" >> api.h; \
> echo "#include <urcu-pointer.h>" >> api.h; \
> elif test -d /usr/include/urcu -o -d /usr/local/include/urcu ; \
> then \
> echo "#define _LGPL_SOURCE" >> api.h; \
> echo "#include <urcu/rculist.h>" >> api.h; \
> echo "#include <urcu/rcuhlist.h>" >> api.h; \
> echo "#include <urcu-pointer-static.h>" >> api.h; \
> else \
> echo "### Missing liburcu package!!! Install it. ###"; \
> exit 1; \
> fi
> echo "#endif /* #ifndef __PERFBOOK_API_H__ */" >> api.h
> echo "# MECHANICALLY GENERATED, DO NOT EDIT!!!" > Makefile.arch
> echo "" >> Makefile.arch
> [CodeSamples]$ cd count/
> [count]$ make
> ### None in count/ can be built on arch: 'unknown'.
> ###
>
>
> So, for some reason, it couldn't still get the architecture.
>
> Oh, I noticed depends.mk use `uname -p` for getting the arch, which is non-
> portable:
>
> ###
> uname --help
> Usage: uname [OPTION]...
> Print certain system information. With no OPTION, same as -s.
> [...]
> -m, --machine print the machine hardware name
> -p, --processor print the processor type (non-portable)
> ###
>
>
> I heard some distros patch this by doing uname -p the same as uname -m, which
> seems to have the correct output. I patched this and then now it works:
>
> ###
> make
> make -C .. Makefile.arch
> make[1]: Entering directory '/dev/shm/perfbook/CodeSamples'
> make pthreads
> make[2]: Entering directory '/dev/shm/perfbook/CodeSamples'
> echo "#ifndef __PERFBOOK_API_H__" > api.h
> echo "#define __PERFBOOK_API_H__" >> api.h
> echo "/* MECHANICALLY GENERATED, DO NOT EDIT!!! */" >> api.h
> echo "" >> api.h
> cat linux/common.h >> api.h
> echo "" >> api.h
> cat arch-x86/arch-x86.h >> api.h
> echo "" >> api.h
> cat api-pthreads/api-pthreads.h | \
> sed '/begin{snippet}/d' | \
> sed '/end{snippet}/d' >> api.h
> echo "" >> api.h
> cat api-pthreads/api-gcc.h >> api.h
> cat api-pthreads/api-linux.h >> api.h
> echo "" >> api.h
> if test -f /usr/include/urcu-call-rcu.h -o \
> -f /usr/local/include/urcu-call-rcu.h -o \
> -f /usr/include/x86_64-linux-gnu/urcu-call-rcu.h -o \
> -f /usr/include/urcu-call-rcu.h ; \
> then \
> echo "#define _LGPL_SOURCE" >> api.h; \
> echo "#include <urcu/rculist.h>" >> api.h; \
> echo "#include <urcu/rcuhlist.h>" >> api.h; \
> echo "#include <urcu-pointer.h>" >> api.h; \
> elif test -d /usr/include/urcu -o -d /usr/local/include/urcu ; \
> then \
> echo "#define _LGPL_SOURCE" >> api.h; \
> echo "#include <urcu/rculist.h>" >> api.h; \
> echo "#include <urcu/rcuhlist.h>" >> api.h; \
> echo "#include <urcu-pointer-static.h>" >> api.h; \
> else \
> echo "### Missing liburcu package!!! Install it. ###"; \
> exit 1; \
> fi
> echo "#endif /* #ifndef __PERFBOOK_API_H__ */" >> api.h
> echo "# MECHANICALLY GENERATED, DO NOT EDIT!!!" > Makefile.arch
> echo "" >> Makefile.arch
> cat arch-x86/Makefile.arch >> Makefile.arch
> make[2]: Leaving directory '/dev/shm/perfbook/CodeSamples'
> make[1]: Leaving directory '/dev/shm/perfbook/CodeSamples'
> cc -g -O3 -Wall -o count_atomic count_atomic.c -lpthread
> cc -g -O3 -Wall -o count_end count_end.c -lpthread
> cc -g -O3 -Wall -o count_end_rcu count_end_rcu.c -lpthread
> cc -g -O3 -Wall -o count_lim count_lim.c -lpthread
> cc -g -O3 -Wall -o count_lim_app count_lim_app.c -lpthread
> cc -g -O3 -Wall -o count_lim_atomic count_lim_atomic.c -lpthread
> cc -g -O3 -Wall -o count_lim_sig count_lim_sig.c -lpthread
> cc -g -O3 -Wall -o count_limd count_limd.c -lpthread
> cc -g -O3 -Wall -o count_nonatomic count_nonatomic.c -lpthread
> cc -g -O3 -Wall -o count_stack count_stack.c -lpthread
> cc -g -O3 -Wall -o count_stat count_stat.c -lpthread
> cc -g -O3 -Wall -o count_stat_atomic count_stat_atomic.c -lpthread
> cc -g -O3 -Wall -o count_stat_eventual count_stat_eventual.c -lpthread
> cc -g -O3 -Wall -o count_tstat count_tstat.c -lpthread
> ###
>
> I will send a patch fixing the uname soon :)
Thank you for digging into this!
And yes, portability is easier than it was 40 years ago, but still not
easy. :-/
Thanx, Paul
> > This should probably be documented somewhere, perhaps in howto/howto.tex
> > or maybe in BUILD-FAQ.txt. Thoughts?
> >
> > Thanx, Paul
>
> Best regards,
> Leo
>
