Re: Make Stap compatible with Android

[Alexander Lochmann <alexander dot lochmann at tu-dortmund dot de>]

On 30.04.2018 22:45, David Smith wrote:
> I took a quick look at your patch. Comments:
> 
> 1) build_android.sh: That script seems quite specific to your setup
> (your choice of BUILD_PREFIX and OUTPUT_DIR for instance). I think I
> would suggest we put something like that on the systemtap wiki in an
> article about building for android, not in the source tree.
I can move this script to another repo.
> 
> 2) I think there's a typo around line 116 in your patch. You've got:
> 
> ====
> @@ -580,10 +587,14 @@ AC_MSG_NOTICE([stap will link $stap_LIBS])
>  if test $build_elfutils = no; then
>    save_LIBS="$LIBS"
>    dnl this will only succeed with elfutils 0.142+
> -  AC_CHECK_LIB(elf,elf_getshdrstrndx,[],[
> -    AC_MSG_FAILURE([libelf too old, need 0.142+])])
> -  staprun_LIBS="$staprun_LIBS -lelf"
> -  stapbpf_LIBS="$stapbpf_LIBS -lelf"
> +  if test "${have_android}" == "bo"; then
> + AC_CHECK_LIB(elf,elf_getshdrstrndx,
> + [staprun_LIBS="$staprun_LIBS -lelf"
> + LIBS="$save_LIBS"],
> + [AC_MSG_FAILURE([libelf too old, need 0.142+])])
> + staprun_LIBS="$staprun_LIBS -lelf"
> + stapbpf_LIBS="$stapbpf_LIBS -lelf"
> +  fi
> ====
> 
> Shouldn't that test be checking for the string "no", not "bo"? I also
> don't understand why you changed the original code. I understand
> surrounding it with the 'if have_android == "no"' test, but not the
> changes themselves. I would have expected just to see whitespace
> changes to the original code there.
Oh yeah... Indeed, it should be "no". Thanks for pointing that out.
But these are just white space changes. Do I miss something?
> 
> 3) I'm also not sure about the approach you are taking here. You seem
> to assume that android will *never* have an elfutils 0.142+. Why don't
> you test the elfutils version and change the configure macro from
> HAVE_ANDROID to something like HAVE_ANDROID_OLD_ELFUTILS (feel free to
> use a better name) if the elfutils version isn't new enough?
As far as I know, neither my MentorGraphics compiler nor the Android
EABI compiler
(https://android.googlesource.com/platform/prebuilts/gcc/linux-x86/arm/arm-eabi-4.7)
have support for elfutils at all.

- Alex
> 
> 
> On Mon, Apr 30, 2018 at 5:44 AM, Alexander Lochmann
> <alexander.lochmann@tu-dortmund.de> wrote:
>> Hi folks,
>>
>> it's been a long time since my first approach to submit my Android patch
>> set.
>> I re-worked my code 'a bit' and stripped it down to the bare essentials.
>> The result is attached to this mail. The patch is based on release-3.2.
>> It basically adds a new configure options: --with-android. This option
>> makes the configure libelf check optional and disables some parts of
>> stap{run,io} which depend on libelf.
>> The result is two binaries, staprun and stapio, which can be executed on
>> Android.
>>
>> What do you think about these changes?
>>
>> Regards,
>> Alex
>>
>> --
>> Technische Universität Dortmund
>> Alexander Lochmann                PGP key: 0xBC3EF6FD
>> Otto-Hahn-Str. 16                 phone:  +49.231.7556141
>> D-44227 Dortmund                  fax:    +49.231.7556116
>> http://ess.cs.tu-dortmund.de/Staff/al
> 
> 
> 

-- 
Technische Universität Dortmund
Alexander Lochmann                PGP key: 0xBC3EF6FD
Otto-Hahn-Str. 16                 phone:  +49.231.7556141
D-44227 Dortmund                  fax:    +49.231.7556116
http://ess.cs.tu-dortmund.de/Staff/al

Attachment: signature.asc
Description: OpenPGP digital signature