Message-ID: <55C04D60.2050101@huawei.com>
Date:	Tue, 4 Aug 2015 13:28:00 +0800
From:	"Wangnan (F)" <wangnan0@...wei.com>
To:	Arnaldo Carvalho de Melo <acme@...nel.org>,
	pi3orama <pi3orama@....com>
CC:	He Kuang <hekuang@...wei.com>, Li Zefan <lizefan@...wei.com>,
	"Alexei Starovoitov" <ast@...mgrid.com>,
	Xia Kaixu <xiakaixu@...wei.com>,
	<linux-kernel@...r.kernel.org>, Ingo Molnar <mingo@...nel.org>,
	Jiri Olsa <jolsa@...hat.com>, David Ahern <dsahern@...il.com>,
	Namhyung Kim <namhyung@...nel.org>
Subject: Re: perf eBPF patch ordering.  was: Re: perf test LLVM was: Re: [GIT
 PULL 00/39] perf tools: filtering events using eBPF programs

Hi Arnaldo,

The following changes since commit 922cc21746202956acb41c89a6190bb50805fa31:

   perf tools: Introduce llvm config options (2015-07-31 12:17:50 -0300)

are available in the git repository at:

   https://github.com/WangNan0/linux.git ebpf

for you to fetch changes up to d85bf4b6470b8d860bbae25418e5ae3ccd9711e8:

   perf tools: Support attach BPF program on uprobe events (2015-08-04 
04:59:20 +0000)

----------------------------------------------------------------

The new cset has following improvements:

1. Improve error message: now don't dump LLVM environment setting messages
    if clang is found. Also, describe how to pre-compile .c file into .o.

    See: perf tools: Call clang to compile C source to object code
https://github.com/WangNan0/linux/commit/264676a5b922aaf1e9be3800fe06d5b67b06cd12

2. Reorder patches, so when 'perf record' is able to accept '--event 
file.c', the BPF filter
    should work. Also, an example BPF script file is provided, and the 
compilation
    method is described in commit message.

    See:
         perf tools: Infrastructure for compiling scriptlets when 
passing '.c' to --event
https://github.com/WangNan0/linux/commit/eca622f4a88e1a791fc2405c398256ad572eba54

3. Introduce 'perf test BPF', which uses previous introduced scriptlet, 
fork a
    'perf record' to utilise it and uses 'perf report' to check the result.

    See: perf tests: Enforce LLVM test for BPF test
https://github.com/WangNan0/linux/commit/a7cdab453863c580446dc2c3a3f3a86f21b770ce

         perf test: Enable 'perf test' run as test targets
https://github.com/WangNan0/linux/commit/b14f2627e95d348be5ec19bd24a5117e8c2ffe46
     and
         perf test: Add 'perf test BPF'
https://github.com/WangNan0/linux/commit/8414217dbfa57df4dbb55642dc26205e1c7cbdf1

4. Fix a bug that if the filename doesn't contain '/' it is recongnised 
as event name then
    failed to be applied by order adjusting in parse-events.l: bring 
{bpf_object} and
    {bpf_source} ahead.

You need to pop 9 patches from your perf/ebpf tree and rebase my tree. 
However, until
patch "perf tools: Enable passing bpf object file to --event" the 
changes is tiny. Please
check.

Thank you.

On 2015/8/4 3:49, Arnaldo Carvalho de Melo wrote:
> Em Mon, Aug 03, 2015 at 01:11:16PM -0300, Arnaldo Carvalho de Melo escreveu:
>>>>   ERROR:    unable to compile ./foo.c
>>>>   Hint:    Check error message shown above.
>>>>          LLVM 3.7 or newer is required. Which can be found from http://llvm.org
>>>>          You may want to try git trunk:
>>>>              git clone http://llvm.org/git/llvm.git
>>>>                   and
> <SNIP>
>>>>       or: perf record [<options>] -- <command> [<options>]
>>>>       -e, --event <event>   event selector. use 'perf list' to list available events
>>>>   [root@...icio ~]#
>>>> Now to find a hello.c BPF scriptlet...
> So, we do not need to provide all this LLVM environment installation
> hints when we get to any error, i.e. the one above was just becasuse
> "./foo.c" doesn't exist, clang ran successfully, so no need for telling
> the user how to install it.
>
> The following error also shouldn't emit those hints:
>
>    [root@...icio ~]# perf record -e ./lock_page.bpf.c sleep 1
>    /root/./lock_page.bpf.c:1:5: error: expected parameter declarator
>    SEC("lock_page=__lock_page page->flags")
>        ^
>    /root/./lock_page.bpf.c:1:5: error: expected ')'
>    /root/./lock_page.bpf.c:1:4: note: to match this '('
>    SEC("lock_page=__lock_page page->flags")
>       ^
>    /root/./lock_page.bpf.c:1:1: warning: type specifier missing, defaults to 'int' [-Wimplicit-int]
>    SEC("lock_page=__lock_page page->flags")
>    ^
>    /root/./lock_page.bpf.c:1:41: error: expected ';' after top level declarator
>    SEC("lock_page=__lock_page page->flags")
>                                            ^
>                                            ;
>    /root/./lock_page.bpf.c:2:22: warning: declaration of 'struct pt_regs' will not be visible outside of this function [-Wvisibility]
>    int lock_page(struct pt_regs *ctx, int err, unsigned long flags)
>                       ^
>    2 warnings and 3 errors generated.
>    ERROR:	unable to compile ./lock_page.bpf.c
>    Hint:	Check error message shown above.
>       	LLVM 3.7 or newer is required. Which can be found from http://llvm.org
>       	You may want to try git trunk:
>       		git clone http://llvm.org/git/llvm.git
>       		     and
>       		git clone http://llvm.org/git/clang.git
>
>       	Or fetch the latest clang/llvm 3.7 from pre-built llvm packages for
>       	debian/ubuntu:
>       		http://llvm.org/apt
>
>       	If you are using old version of clang, change 'clang-bpf-cmd-template'
>       	option in [llvm] section of ~/.perfconfig to:
>
>       	  "$CLANG_EXEC $CLANG_OPTIONS $KERNEL_INC_OPTIONS \
>       	     -working-directory $WORKING_DIR -c $CLANG_SOURCE \
>       	     -emit-llvm -o - | /path/to/llc -march=bpf -filetype=obj -o -"
>       	(Replace /path/to/llc with path to your llc)
>
>    Hint:	You can also pre-compile it into .o
>    invalid or unsupported event: './lock_page.bpf.c'
>    Run 'perf list' for a list of valid events
>
>     usage: perf record [<options>] [<command>]
>        or: perf record [<options>] -- <command> [<options>]
>
>        -e, --event <event>   event selector. use 'perf list' to list available events
>    [root@...icio ~]#
>
> And I am interested in that "Hint: You can also pre-compile it into .o", seems
> like a useful one, if it  was on the screen, i.e. how can I pre-compile this
> into a .o, its some random piece of eBPF "scriptlet" I found in one of your
> csets, after this cset in your patchkit:
>
>    [root@...icio ~]# cat lock_page.bpf.c
>    SEC("lock_page=__lock_page page->flags")
>    int lock_page(struct pt_regs *ctx, int err, unsigned long flags)
>    {
>    	return 1;
>    }
>    [root@...icio ~]#
>
> Also it would be nice to tell the user, when compilation fails, where to look
> for a eBPF scriptlet primer, i.e. where can I find some documentation on how
> to write such scriptlets?
>
> At this point I expect it to build the .c into a .o, then, as the wiring up is
> not there, tell the user that all went well, but more infrastructure is needed,
> apply the following patches from Wang Nan! 8-)
>
> - Arnaldo


--
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