libunwind lacks AArch64 support

Isn't this just a config.sub update?

config.sub in libunwind (apt-get source in quantal)

# Free Software Foundation, Inc.
timestamp='2009-06-11'

W dniu 18.01.2013 01:39, Alexander Sack pisze:
> config.sub in libunwind (apt-get source in quantal)
>
> # Free Software Foundation, Inc.
> timestamp='2009-06-11'

No, it is not. Under OE old gnu-config files are not issue cause
autoreconf is called.

After adding aarch64 support into configure.ac:

| make[2]: Entering directory `/tmp/OE/tmp/work/aarch64-oe-linux/libunwind/1.1-r0/libunwind-1.1/src'
| ../aarch64-oe-linux-libtool --tag=CC --mode=compile aarch64-oe-linux-gcc --sysroot=/home/hrw/HDD/devel/canonical/aarch64/openembedded/build/tmp-eglibc/sy
sroots/genericarmv8 -DHAVE_CONFIG_H -I. -I../include -I../include -I../include/tdep-aarch64 -I. -D_GNU_SOURCE -DNDEBUG -O2 -pipe -g -feliminate-unused-debug
-types -fexceptions -Wall -Wsign-compare -c -o elf64.lo elf64.c
| make[2]: *** No rule to make target `libunwind-aarch64.la', needed by `libunwind-setjmp.la'. Stop.
| make[2]: *** Waiting for unfinished jobs....
| aarch64-oe-linux-libtool: compile: aarch64-oe-linux-gcc --sysroot=/home/hrw/HDD/devel/canonical/aarch64/openembedded/build/tmp-eglibc/sysroots/genericarmv8
 -DHAVE_CONFIG_H -I. -I../include -I../include -I../include/tdep-aarch64 -I. -D_GNU_SOURCE -DNDEBUG -O2 -pipe -g -feliminate-unused-debug-types -fexceptions -
Wall -Wsign-compare -c elf64.c -fPIC -DPIC -o .libs/elf64.o
| In file included from ../include/libunwind_i.h:52:0,
| from elfxx.h:44,
| from elf64.h:7,
| from elf64.c:2:
| ../include/libunwind.h:25:3: error: #error "Unsupported arch"
| In file included from elfxx.h:44:0,
| from elf64.h:7,
| from elf64.c:2:
| ../include/libunwind_i.h:154:19: error: unknown type name 'UNW_PASTE'
| ../include/libunwind_i.h:154:19: error: unknown type name 'full_mask'
| ../include/libunwind_i.h:203:12: error: expected ')' before '(' token
| ../include/libunwind_i.h:207:12: error: expected ')' before '(' token
| ../include/libunwind_i.h:213:13: error: expected ')' before '(' token
| ../include/libunwind_i.h:219:12: error: expected ')' before '(' token
| ../include/libunwind_i.h:224:13: error: expected ')' before '(' token
| ../include/libunwind_i.h:227:12: error: expected ')' before '(' token
| ../include/libunwind_i.h:229:1: error: unknown type name 'unw_dyn_info_list_t'
| ../include/libunwind_i.h:266:13: error: unknown type name 'UNW_PASTE'
| ../include/libunwind_i.h:266:13: error: unknown type name 'mi_init'
| ../include/libunwind_i.h:267:1: error: unknown type name 'unw_word_t'
| ../include/libunwind_i.h:280:5: error: unknown type name 'unw_dyn_info_t'
| ../include/libunwind_i.h:281:5: error: unknown type name 'unw_dyn_info_t'
| ../include/libunwind_i.h: In function 'invalidate_edi':
| ../include/libunwind_i.h:295:16: error: request for member 'format' in something not a structure or union
| ../include/libunwind_i.h:296:16: error: request for member 'format' in something not a structure or union
| In file included from ../include/libunwind_i.h:320:0,
| from elfxx.h:44,
| from elf64.h:7,
| from elf64.c:2:
| ../include/tdep/libunwind_i.h: At top level:
| ../include/tdep/libunwind_i.h:25:3: error: #error "Unsupported arch"
| In file included from elf64.h:7:0,
| from elf64.c:2:
| elfxx.h:46:35: error: unknown type name 'unw_addr_space_t'
| elfxx.h:47:18: error: unknown type name 'unw_word_t'
| elfxx.h:49:7: error: unknown type name '...

configure.ac patch used

src/arm$ wc *
...
 2152 8635 64623 total

2K lines and 64KB of code needed to implement ARM support, Aarch64 probably similar. This is probably something the toolchain team can do better than we can.

12:10 < hrw> mgrettondann: can someone from TCWG take care of bug 1099810?
12:11 < mgrettondann> hrw, not immediately - but certainly in a week or two
12:12 < hrw> mgrettondann: sure. just wanted to grab attention

libunwind isn't a showstopper for getting HHVM to build. The JIT functionality depends on libunwind, but we should be able to get the interpreter to work without it.

Yvan: please add comment when you will have something for integration.

Marcin: Yes sure, I've a couple of things to fix before upstream submission, but will let you know the status.

The push to May is caused by upstreaming or porting. I need to know to update some JIRA cards for the HHVM work.

Porting, I was preempted by multiple merge reviews and releases tasks, but I would be able to post the patch for review by tomorrow evening. Regarding JIRA, I update the TCWG related one, is it sufficient or should I update thsi bug entry and/or LEG cards too ?

On 11 April 2013 13:19, Yvan Roux <email address hidden> wrote:
> Porting, I was preempted by multiple merge reviews and releases tasks,
> but I would be able to post the patch for review by tomorrow evening.
> Regarding JIRA, I update the TCWG related one, is it sufficient or
> should I update thsi bug entry and/or LEG cards too ?

Don't worry about the JIRA LEG cards. I will do that.

Thanks,

anmar

Ok, thanks.

This is the patch I sent for review, thus it will probably evolved but you can use it for early testing.

Yvan: few notes (already given on irc):

15:03 < hrw> yroux: you need to update (C) info in some files
15:06 < yroux> hrw, indeed thanks
15:24 < hrw> yroux: do: s/unreachable()/_builtin_unreachable()/ in src/aarch64/Gresume.c and it will work
15:24 < hrw> yroux: that's for 1.1 version

backport from the trunk for the unreachable() call issue

Here is a second version of the patch after upstream comments, no new features are enabled, but unwinding is now made with eh_frame information instead of debug frame ones. Notice that it hasn't been approved yet.

Patch commited upstream