Message-ID: <55779FB0.7000809@huawei.com>
Date:	Wed, 10 Jun 2015 10:23:44 +0800
From:	"Wangnan (F)" <wangnan0@...wei.com>
To:	Alexei Starovoitov <ast@...mgrid.com>, <acme@...nel.org>,
	<brendan.d.gregg@...il.com>, <daniel@...earbox.net>,
	<namhyung@...nel.org>, <masami.hiramatsu.pt@...achi.com>,
	<paulus@...ba.org>, <a.p.zijlstra@...llo.nl>, <mingo@...hat.com>,
	<jolsa@...nel.org>, <dsahern@...il.com>
CC:	<linux-kernel@...r.kernel.org>, <lizefan@...wei.com>,
	<hekuang@...wei.com>, <xiakaixu@...wei.com>, <pi3orama@....com>
Subject: Re: [RFC PATCH v6 25/32] perf tools: Add 'bpf.' config section to
 perf default config



On 2015/6/10 9:09, Alexei Starovoitov wrote:
> On 6/9/15 5:47 PM, Wangnan (F) wrote:
>>
>>
>> On 2015/6/10 7:43, Alexei Starovoitov wrote:
>>> On 6/8/15 10:50 PM, Wang Nan wrote:
>>>> perf_bpf_config() is added to parse 'bpf' section in perf config file.
>>>> Following is an example:
>>>>
>>>>   [bpf]
>>>>     clang-path = /llvm/bin/x86_64-linux-clang"
>>>>     llc-path = /llvm/bin/x86_64-linux-llc"
>>>>     clang-opt = "-nostdinc -isystem /llvm/lib/clang/include
>>>> -I/kernel/arch/x86/include ..."
>>>>     llc-opt = ""
>>>
>>> a section to specify -I flags to compile prog.c is useful,
>>> but users shouldn't be populating it manually for kernel headers.
>>> How about adding a script that can figure out $(LINUXINCLUDE)
>>> automatically ?
>>> You can even invoke such flag detector from perf via something like:
>>> f = open /tmpdir/Makefile
>>> fprintf(f, "obj-y := dummy.o\n");
>>> fprintf(f, "$(obj)/%%.o: $(src)/%%.c\n");
>>> fprintf(f, "\t@...o -n \"$(NOSTDINC_FLAGS) $(LINUXINCLUDE)
>>> $(EXTRA_CFLAGS)\" > %s\n", cachefile);
>>> snprintf(cmd, "make -s -C /lib/modules/%s/build M=%s dummy.o",
>>>    uts_release, tmpdir);
>>> system(cmd);
>>> read flags from cachefile and cache it for the future.
>>> ...
>>> or as independent script that populates .perfconfig
>>>
>>
>> That's cool, but could I implement it later? Introducing such stuffs
>
> I think --clang-opts shouldn't be introduced without
> automatic flag finder. Ease of use is important.
>

What about automatic finder failed? It is possible that user environment 
doesn't have
kbuild directory installed, should we make it another deadend? Then 
these 'ease of use'
design is useless for me because in embedded area usecases are always rare.

I think we can hide --clang-opt from cmdline, make it reside in 
.perfconfig only, and
pop some messages on it when failure. Most of the time users need to 
config only once.
Still not very hard, right?

>> also introduces
>> a lot of trouble tasks:
>>
>> 1. Dependency on make and kernel build. We should search make and kbuild
>> dir dynamically
>>     and also gives users the right to specify them by theirselves. A lot
>> of configuration
>>     options should be appended: --make-path=/path/to/make
>> --kbuild-dir=kernel/build/dir
>>     in cmdline and
>>     [kbuild]
>>        make_path = /path/to/make
>>        kbuild_dir = /path/to/kbuild
>>     in .perfconfig.
>
> they're not mandatory. If 'make' is not in a PATH, it's dead end.
> kernel build is very likely installed in /lib/modules/
>
>> 2. Selection of architectures. Although currently we want it to work
>> only when we dynamically
>>     compile a script, I think finally we should consider cross compiling
>> bpf objects. Then cmdline
>>     generation becomes complex. Also, --arch and [kbuild.arch] should
>> also be introduced.
>
> that's a rare use case. This one can be added later.
>
>> 3. Kernel dependency. Consider if kernel decides to change its 
>> interface...
>
> change what interface? 'make M=' ? sure. then lots of scripts will be
> broken.
>

'make M=' is unlikely to be changed. However it is possible for kernel 
to append
more variable for specifying include, or devide $(NOSTDINC_FLAGS) 
$(LINUXINCLUDE)
and $(EXTRA_CFLAGS) into smaller pieces. I'm not quite sure whether 
those make
variables are part of kbuild interface.

>> I think currently we can pop some messages to let user know how to get
>> include dirs manually,
>> let further patches to do it for them automatically.
>>
>> P.S.
>>
>> Have you tested your Makefile? It doesn't work for me:
>
> works as a charm:
> $ cat /home/ast/ff/Makefile
> obj-y := dummy.o
>
> $(obj)/%.o: $(src)/%.c
>     @echo -n "$(NOSTDINC_FLAGS) $(LINUXINCLUDE) $(EXTRA_CFLAGS)" > 
> /tmp/xxx
> $ make -s -C /w/net-next/bld_x64 M=/home/ast/ff dummy.o
> $ cat /tmp/xxx
>  -nostdinc -isystem /usr/lib/gcc/x86_64-linux-gnu/4.7/include 
> -I../arch/x86/include -Iarch/x86/include/generated/uapi 
> -Iarch/x86/include/generated  -I../include -Iinclude 
> -I../arch/x86/include/uapi -Iarch/x86/include/generated/uapi 
> -I../include/uapi -Iinclude/generated/uapi -include 
> ../include/linux/kconfig.h
>

Finally I made it work. dummy.c must exist on current directory.

What we are talking about seems not very suit to be implemented using C. 
It should be implemented
using a shell script:

#!/usr/bin/env sh
TMPDIR=`mktemp -d`
cat << EOF > $TMPDIR/Makefile
obj-y := dummy.o

\$(obj)/%.o: \$(src)/%.c
         @echo -n "\$(NOSTDINC_FLAGS) \$(LINUXINCLUDE) \$(EXTRA_CFLAGS)" 
 > $TMPDIR/cflags
EOF

touch $TMPDIR/dummy.c

DEFAULT_KBUILD_DIR=/lib/modules/`uname -r`/build

if ! test -d "$KBUILD_DIR"
then
     KBUILD_DIR=$DEFAULT_KBUILD_DIR
fi

make -s -C $KBUILD_DIR M=$TMPDIR $OTHER_OPTS dummy.o 2>/dev/null

cat $TMPDIR/cflags 2>/dev/null
RET=$?

rm -rf $TMPDIR
exit $RET

In this script, we can inject KBUILD_DIR and OTHER_OPTS to select another
kbuild directory and "ARCH=" option.

I'll embed this script in my next version.

Thaknk you.

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists