[opensuse-packaging] tests are failing with illegal instruction (SIGILL)
Hi, I'm trying to package an updated version of ethereum-cpp. You can find my package at https://build.opensuse.org/package/show/home:etamPL:branches:network:cryptoc... . I've come to a point where it finally builds successfully (yay!). Unfortunately tests are failing with illegal instruction (SIGILL) errors. I've managed to run them under gdb, but even with debugging symbols installed it doesn't show any specific location in code. I'm not experienced with assembly so I have no idea what is responsible for the problem and where to look for solutions. Thanks for any help. -- Adam Mizerski
On Thursday 2019-02-07 16:22, Adam Mizerski wrote:
I've managed to run them under gdb, but even with debugging symbols installed it doesn't show any specific location in code.
(gdb) p $rip (gdb) disas look for around the ==> marker -- To unsubscribe, e-mail: opensuse-packaging+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-packaging+owner@opensuse.org
Hi Adam! On 2/7/19 4:22 PM, Adam Mizerski wrote:
I've come to a point where it finally builds successfully (yay!). Unfortunately tests are failing with illegal instruction (SIGILL) errors.
I've managed to run them under gdb, but even with debugging symbols installed it doesn't show any specific location in code.
It's not unusual for crypto libraries to detect the processor's capabilities by running code with certain instructions and see whether it triggers the SIGILL trap. OpenSSL does that, for examplel When the SIGILL trap occurs, try typing "continue" in GDB to see if it just continues so that you get to the actual point where the crash happens. Adrian -- To unsubscribe, e-mail: opensuse-packaging+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-packaging+owner@opensuse.org
On Thursday 2019-02-07 16:33, John Paul Adrian Glaubitz wrote:
On 2/7/19 4:22 PM, Adam Mizerski wrote:
I've come to a point where it finally builds successfully (yay!). Unfortunately tests are failing with illegal instruction (SIGILL) errors.
I've managed to run them under gdb, but even with debugging symbols installed it doesn't show any specific location in code.
It's not unusual for crypto libraries to detect the processor's capabilities by running code with certain instructions and see whether it triggers the SIGILL trap. OpenSSL does that, for examplel
I suspect the report was made because the SIGILL was followed by a program abort and getting thrown back to the shell prompt. -- To unsubscribe, e-mail: opensuse-packaging+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-packaging+owner@opensuse.org
On Feb 07 2019, Adam Mizerski
I've managed to run them under gdb, but even with debugging symbols installed it doesn't show any specific location in code.
How does it look like? Andreas. -- Andreas Schwab, SUSE Labs, schwab@suse.de GPG Key fingerprint = 0196 BAD8 1CE9 1970 F4BE 1748 E4D4 88E3 0EEA B9D7 "And now for something completely different." -- To unsubscribe, e-mail: opensuse-packaging+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-packaging+owner@opensuse.org
W dniu 07.02.2019 o 16:22, Adam Mizerski pisze:
Hi,
I'm trying to package an updated version of ethereum-cpp. You can find my package at https://build.opensuse.org/package/show/home:etamPL:branches:network:cryptoc... .
I've come to a point where it finally builds successfully (yay!). Unfortunately tests are failing with illegal instruction (SIGILL) errors.
I've managed to run them under gdb, but even with debugging symbols installed it doesn't show any specific location in code.
I'm not experienced with assembly so I have no idea what is responsible for the problem and where to look for solutions.
Thanks for any help.
-- Adam Mizerski
Thank you all for your comments. I don't know what has changed, but this time backtrace gave me a hint which library might be a problem. And it was correct: RocksDB (https://build.opensuse.org/package/show/home:etamPL:branches:network:cryptoc...) by default appends "-march=native", which obviously leads to problems. Tests are running now. Thanks. -- Adam Mizerski
El 07-02-2019 a las 17:14, Adam Mizerski escribió:
W dniu 07.02.2019 o 16:22, Adam Mizerski pisze:
Tests are running now. Thanks.
..And the issue around having an automatic filter for this compiler options comes up again.. at least one that is active during build of tumbleweed.. not sure how to implement that.. either by a hackish enviroment variable, a post build check, or implementing a compiler flag that is stronger than -msomething . -mfixed-instruction-set=x86_64 or similar..sigh. -- To unsubscribe, e-mail: opensuse-packaging+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-packaging+owner@opensuse.org
On Thu, 7 Feb 2019, Cristian Rodríguez wrote:
El 07-02-2019 a las 17:14, Adam Mizerski escribió:
W dniu 07.02.2019 o 16:22, Adam Mizerski pisze:
Tests are running now. Thanks.
..And the issue around having an automatic filter for this compiler options comes up again.. at least one that is active during build of tumbleweed.. not sure how to implement that.. either by a hackish enviroment variable, a post build check, or implementing a compiler flag that is stronger than -msomething . -mfixed-instruction-set=x86_64 or similar..sigh.
Make sure the build VMs are dumbed down to bare x86-64 ISA support?
IIRC qemu can handle this.
Richard.
--
Richard Biener
Am 08.02.19 um 09:03 schrieb Richard Biener:
On Thu, 7 Feb 2019, Cristian Rodríguez wrote:
..And the issue around having an automatic filter for this compiler options comes up again.. at least one that is active during build of tumbleweed.. not sure how to implement that.. either by a hackish enviroment variable, a post build check, or implementing a compiler flag that is stronger than -msomething . -mfixed-instruction-set=x86_64 or similar..sigh.
Make sure the build VMs are dumbed down to bare x86-64 ISA support? IIRC qemu can handle this.
It's not that easy, because `cpuid' instruction is non-emulated, and available to unprivileged users IIRC, so if someone just uses it to determine what comands to use, it will still fail. (It might be that this is only true for Xen and not qemu/kvm, but I remember having real life problems because of this some time ago). -- Stefan Seyfried "For a successful technology, reality must take precedence over public relations, for nature cannot be fooled." -- Richard Feynman -- To unsubscribe, e-mail: opensuse-packaging+unsubscribe@opensuse.org To contact the owner, e-mail: opensuse-packaging+owner@opensuse.org
participants (7)
-
Adam Mizerski -
Andreas Schwab -
Cristian Rodríguez -
Jan Engelhardt -
John Paul Adrian Glaubitz -
Richard Biener -
Stefan Seyfried