Created attachment 83929 [details] Indepth details of valgrind build and execution on target Immediatly get Unrecognised instruction errors, which result in reboot, upon startup of in-house developed app, running on TMS320DM816x under Android. uname -a: Linux Group-Series 2.6.37+ #1 PREEMPT Sun Dec 1 13:03:34 CST 2013 armv7l GNU/Linux Valgrind version: valgrind-3.9.0 See attachment: valgrin3.9_on_ARM_error.txt for full details
In your description 'Unpacked: tar xf valgrind-3.7.0.tar.bz2" - 3.7 - is a misprint? Did you try compile from the latest trunk?
My apologies; I failed to update the unpack statement. I actually untarred valgrind-3.9.0.tar.bz2 and after my ARM build of valgrind executing the following on target platform: => ./valgrind --version valgrind-3.9.0 -----Original Message----- From: bugzilla_noreply@kde.org [mailto:bugzilla_noreply@kde.org] On Behalf Of Vasily Sent: Thursday, December 05, 2013 3:30 AM To: Liccese, Joe Subject: [valgrind] [Bug 328423] Unrecognised instructions: _fips_armv7_tick and _armv7_tick https://bugs.kde.org/show_bug.cgi?id=328423 Vasily <vasily.golubev@gmail.com> changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |vasily.golubev@gmail.com --- Comment #1 from Vasily <vasily.golubev@gmail.com> --- In your description 'Unpacked: tar xf valgrind-3.7.0.tar.bz2" - 3.7 - is a misprint? Did you try compile from the latest trunk? -- You are receiving this mail because: You reported the bug.
Pulled latest using the following command: svn co svn://svn.valgrind.org/valgrind/trunk valgrind and rebuilt for ARM as noted in initial submit. Still seeing the same errors. Is there anyway to simply force valgrind to ignore these two errors?
(In reply to comment #3) > Is there anyway to simply force valgrind to ignore these two errors? No. You'll need to disassemble the object and find the offending instruction, and report back here. The armv7 support in the trunk is fairly complete so I'm a bit surprised it failed.
So you need more than the following that is reported by Valgrind? disInstr(arm): unhandled instruction: 0xEE190F1D cond=14(0xE) 27:20=225(0xE1) 4:4=1 3:0=13(0xD)
disassembled ARM instruction causing unhandled instruction error: 0006f888 <_fips_armv7_tick>: 6f888: ee190f1d mrc 15, 0, r0, cr9, cr13, {0} 6f88c: e12fff1e .word 0xe12fff1e 0008a708 <_armv7_tick>: 8a708: ee190f1d mrc 15, 0, r0, cr9, cr13, {0} 8a70c: e12fff1e .word 0xe12fff1e
The first of those (0xEE190F1D) is an attempt to read the cycle counter: see also bug 331178 for another instance of that (also on Android; I guess Android exposes the counter to userspace, unlike stock Linux?) 0xe12fff1e is just the usual Thumb return insn "BX LR", which should be no problem for Valgrind.
I included the entire disassembly for routines: _fips_armv7_tick and _armv7_tick and as you noted the 0xe12fff1e is not an issue. The 0xEE190F1D is the offending instruction. Bug 331178 notes the offending instruction is within libcrypto.so, which is the same library we're using here that is causing the error. As I noted in my original attachment; Valgrind indicates that "Valgrind will now raise a SIGILL signal which will probably kill your program" and in fact it does terminate. Is there any workaround for this? -----Original Message----- From: bugzilla_noreply@kde.org [mailto:bugzilla_noreply@kde.org] On Behalf Of Peter Maydell Sent: Thursday, April 03, 2014 9:13 AM To: Liccese, Joe Subject: [valgrind] [Bug 328423] Unrecognised instructions: _fips_armv7_tick and _armv7_tick https://bugs.kde.org/show_bug.cgi?id=328423 Peter Maydell <peter.maydell@linaro.org> changed: What |Removed |Added ---------------------------------------------------------------------------- CC| |peter.maydell@linaro.org --- Comment #7 from Peter Maydell <peter.maydell@linaro.org> --- The first of those (0xEE190F1D) is an attempt to read the cycle counter: see also bug 331178 for another instance of that (also on Android; I guess Android exposes the counter to userspace, unlike stock Linux?) 0xe12fff1e is just the usual Thumb return insn "BX LR", which should be no problem for Valgrind. -- You are receiving this mail because: You reported the bug.
I included the entire disassembly for routines: _fips_armv7_tick and _armv7_tick and as you noted the 0xe12fff1e is not an issue. The 0xEE190F1D is the offending instruction. Bug 331178 notes the offending instruction is within libcrypto.so, which is the same library we're using here that is causing the error. As I noted in my original attachment; Valgrind indicates that "Valgrind will now raise a SIGILL signal which will probably kill your program" and in fact it does terminate. Is there any workaround for this?
Closing this as a dup of 331178; that bug has a somewhat better analysis/discussion. *** This bug has been marked as a duplicate of bug 331178 ***