app dumps in dld_malloc when executed under aries, behavior changes after PHSS_39294 is install.

[Srimalik]

Hi,
I have a PA banary which run without problems on PA machine but when it is executed on an IA machine it dumps are core even before main is called.

gdb shows:
0xc3538c9c in dld_malloc+0x80 () from /usr/lib/dld.sl
#1 0xc35328f8 in init_deferred_plt+0xe0 () from /usr/lib/dld.sl
#2 0xc3534730 in form_load_graph+0x230 () from /usr/lib/dld.sl
#3 0xc3534d90 in form_load_graph+0x890 () from /usr/lib/dld.sl
#4 0xc3534d90 in form_load_graph+0x890 () from /usr/lib/dld.sl
#5 0xc3544138 in finish_dld_main+0x1028 () from /usr/lib/dld.sl
#6 0xc35479dc in _dld_main+0x1c8 () from /usr/lib/dld.sl
#7 0x754c in __map_dld+0x674 ()
(gdb)

installed PHSS_39294 and rebooted the machine
now I am getting these messages when executing the binary.

[HP ARIES32]: ***** Log time: Thu Sep 17 00:46:54 MDT 2009 *****
[HP ARIES32]:
[HP ARIES32]: Internal Error (code : 006) [PID : 2105]
[HP ARIES32]: Encountered unimplemented syscall : LWP_SA_PROCINIT (532).
[HP ARIES32]:
[HP ARIES32]: ARIES Version :
[HP ARIES32]: 11.31 (PHSS_39294) Ver: 3.0.00022 [May 20, 2009 12:18:50]
[HP ARIES32]: Invoked command :
[HP ARIES32]: /opt/VRTSob/bin/vxsvc -e
bash-3.2#

[Srimalik]

tusc output(taking too much time to upload, still trying) shows(only including calls related to processes/threads, there are mmaps and sigprocmasks also)
fork, lwp_self, set_userthreadid, mpctl, lwp_getscheduler,

Has anybody seen similar issues before?

[Suprateeka R Hegde]

looking at the aries patch, it looks like you are on 11.31.

did you try installing the latest linker patch? some of the aries related issues are fixed in the latest patch:

11.11: PHSS_39749
11.23: PHSS_39821
11.31: PHSS_39822

[Srimalik]

>looking at the aries patch, it looks like you are on 11.31.

Yes its 11.31, march 2009 release.

>did you try installing the latest linker patch? some of the aries related issues are fixed in the latest patch:

installed PHSS_39822 but its still failing,

The error seems to be pointing to some syscall failure(not found) under aries.

[Rajesh K Chaurasia]

The error from ARIES indicates that MxN threads specific system call is encountered. ARIES forces 1x1 compatibility model by setting environment variables in emulated application's address space. I need to know which pthread patch is installed on your system and what is the value of environment variable PTHREAD_COMPAT_MODE? Does vxsvc application clears environment during exec()?

Does the application work fine with PHSS_38527 ARIES patch?

Regards
-Rajesh

[Dennis Handly]

You now have the latest Aries patch. So if Suprateeka's dld patch suggestion doesn't work, you'll need to contact the Response Center.
This may be the important error:
Encountered unimplemented syscall: LWP_SA_PROCINIT (532).

[Srimalik]

>I need to know which pthread patch is installed on your system and what is the value of environment variable PTHREAD_COMPAT_MODE?

PHCO_38837 was installed. After your post I installed PHCO_39604 but the results did not change. PTHREAD_COMPAT_MODE is set to 1.

Does vxsvc application clears environment during exec()?

No. the process execs itself while running and the value of PTHREAD_COMPAT_MODE just before the exec call is 1.

>Does the application work fine with PHSS_38527 ARIES patch?
No. :-(

>you'll need to contact the Response Center.
>This may be the important error:
>Encountered unimplemented syscall: LWP_SA_PROCINIT (532).

hmmm.
We used to build this app on 11.11(aCC 3.37), I am trying to build it on 11.23(aCC 3.85) the build went fine, the process ran without issues on PA but hit this issue on IA. wanted to check whether there is a patch available for this already.

[Rajesh K Chaurasia]

Thanks for the information on pthread patch.

As documented in aries(5) man page - ARIES does not support true emulation of MxN threads. ARIES emulates MxN thread applications by forcing 1x1 compatibility mode. PA-RISC applications on Integrity should never encounter lwp_sa_procinit() system call. We need to discover origin of this system call and if the caller did not check for compatibility mode. It is unlikely that your code could be calling lwp_sa_procinit() directly.

As Dennis suggested - please submit the issue to HP response center.

Regards
-Rajesh