Valgrind fails on rdrand when the cpu supports it (haswell)

This upstream bug looks related (affecting rdrand) but the instruction bytes are different:

https://bugs.kde.org/show_bug.cgi?id=353370

Status changed to 'Confirmed' because the bug affects multiple users.

I get the exact same error and bytes.
unhandled instruction bytes: 0xF 0xC7 0xF0 0x89 0x6 0xF 0x42 0xC1

It happens when using std::random_device with GCC 5.2 on willy.

It's weird GCC stopped using /dev/urandom, a very good randomness source that makes use of RDRAND

I also get the same error on Ubuntu Wily + Valgrind 3.11 + g++ 5.2

vex amd64->IR: unhandled instruction bytes: 0xF 0xC7 0xF0 0x89 0x6 0xF 0x42 0xC1
vex amd64->IR: REX=0 REX.W=0 REX.R=0 REX.X=0 REX.B=0
vex amd64->IR: VEX=0 VEX.L=0 VEX.nVVVV=0x0 ESC=0F
vex amd64->IR: PFX.66=0 PFX.F2=0 PFX.F3=0
==13334== valgrind: Unrecognised instruction at address 0x510dec5.
==13334== at 0x510DEC5: ??? (in /usr/lib/x86_64-linux-gnu/libstdc++.so.6.0.21)
==13334== by 0x510E061: std::random_device::_M_getval() (in /usr/lib/x86_64-linux-gnu/libstdc++.so.6.0.21)
==13334== by 0x477C8D: std::random_device::operator()() (random.h:1619)

I can confirm the problem on 16.04 and I can also confirm that the patch posted in kde-bugs #353370 report seems to fix the issue (valgrind compiled from source with and without the patch)

We are seeing this bug with valgrind 1:3.11.0-1ubuntu4.1 on 16.04 in our Ceph testing.

Installing valgrind 1:3.12.0~svn20160714-1ubuntu2 from yakkety onto my xenial machine seems to resolve the issue, too. But probably including the patch from https://bugs.kde.org/show_bug.cgi?id=353370#c9 will be easier than doing an update in xenial, right?

please could somebody confirm that backporting the patch for this issue solved the problem?

Hi Matthias,
I was affected by this bug and can confirm, that applying the linked patch fixes the issue on ubuntu 16.04 (Xenial Xerus) on a i7-6600U.

It would be nice if the patched could be backported to Xenial as it's tedious to recompile valgrind on every update to get it functional again.

Status changed to 'Confirmed' because the bug affects multiple users.

I just ran into this problem on xenial, but I wanted to inquire whether there is any plan to backport the patch before I try to compile it myself. Can we expect the fix in xenial any time soon?