I found the gentoo thread after running into this a few times myself (while compiling a kernel -- not running gentoo personally), and determining that it wasn't due to memory defects.
Only on Beta Bios 1.94 did I get crazy BSOD/Kernel Panics Seg faults but that was a while back and I haven't had so much as a sneeze from the system since then.
A lot of the users with issues though seem to be Ryzen 5 + B350 users.
Possibly. I'm also running on b350 (pro4), since two weeks; oddly, I seem to be running into it less with my ryzen 5 [email protected]/1.3125v than with the CPU at stock, though that could simply be due to my looking for patterns in anecdotal evidence (only 8 build attempts in total, the last 4 only resulting in a single segfault). My system's otherwise rock stable though.
It seems theyāre slowly adding support to the bioses for that ā mine got the option with the update to 1.0.0.6 or 6a. Having said that, a lot of Michaelās segfaults appear to be conftests, and so harmless/desirable.
I noticed in the latest ASrock uefi there is an explicit workaround for a gcc return value. So thatās also a good sign. Is it true Ubuntu exhibits this problem more than most other distros?
Is this the part where we discover the three letter agencies read reflections on trusting trust and thereās stuff lurking around our compilers? Lol
Doubt thatās due to anything more than Ubuntuās popularity and/or some versions of GCC being more robust than others wrt the generation of these segfaults (I also ran into it less often on a fc23-based VM than on an fc24/25-based one). Iām still surprised that no windows (non-WSL) users have complained about it, though, and/or that itās run into so little generally ā are people who compile/program shit really that scarce, as a percentage of the general (PC enthusiast/PCMR) population?
Iāve been using Gentoo with gcc though Iām using it on a laptop (of course Iām not using Ryzen) honestly from the headaches Iāve had to deal with from Gentoo. I havenāt had an issue since Iāve changed the package accept keyword from ~amd64 to amd64. though i donāt think many issues are going to be fixed on Gentoo. for a distro that is supposed to be rolling release a lot of packages are fairly outdated compared to other RR distros. and I donāt even want to get into it with Funtoo. itās even worse.
I have been messing around with this today on my 1800x (retail) 1700 (retail) and 1600x (loaner from msi, thanks guys)
I can cause segfaults all day long with the phoronix test suite but they seem to be in conftest? e.g. not real? with the same image, on x299 I get the same segfaults in conftest.
I have experienced that Iāve had to up my SoC voltage a bit with memory overclocks/xmp but I hvae been testing at purely stock settings today. Testing on ASRock Taichi X370. Testing on Fedora 26 only. I am tempted to live boot ubuntu and see if I can get it to segfault.
I am getting segfaults so I had a bit of a freakoutā¦ but then I realized its just conftest? so I think there is something to thise but the signal to noise ratio is too close to 1 for my liking at the moment?
I was running into segfaults on my 1800X + ASUS Prime X370-Pro until I upped the SOC voltage to 1.18v. Now I can go over a day compiling Linux/GCC without segfaults.
Yeah, the conftest stuff is from php ā missed that at first as well, sorry about that.
If you want to be sure, I think the best test to run overnight is still this one https://github.com/suaefar/ryzen-test (and if you get segfaults with the ākillā script, try the āsaveā script next (āI observed that when the processes remain āisolatedā in their respective cache domains there seem to be no segfaults.ā).
Seems Larabel is in over his head, and also more concerned with generating short-term attention for his products over worries about what drawing so much attention to results that he doesnāt fully understand and that partly constitute false positives will do for his rep in the long term. As #45 would say: #Sad.
I can not for the life of me reproduce these results.
At least nothing aside from the stoopid conftest like everyone else. Which is a non-issue
One surefire way however that I can reproduce a much worse error is by lowering my DRAM Voltage to 1.2V and setting Load Line level 3 while at stock VCore and clocks on some potato Corsair DIMMS.
This will lead to Hard Machine Check Exceptions and resets while at idle. (Itās got to do with IDLE C-states and is completely random) Disabling C-States and or setting a lower Load Line level (level 4) completely avoids this problem.
That said a new tool Iāve recently started using for testing ryzen stability is sandsifter, the x86 processor fuzzer.
I will have a post coming up about that sometime once Iāve figured it all out.
So I have just managed to get something while compiling Digikam of all things. I wasnāt even looking for a segfault at the time.
Unfortunately I wasnāt logging any more detailed info on that run and I canāt reproduce it sinceā¦
89%] Linking CXX executable usexmpsidecar
In file included from /usr/include/qt/QtCore/qglobal.h:45:0,
from /usr/include/qt/QtGui/qtguiglobal.h:43,
from /usr/include/qt/QtWidgets/qtwidgetsglobal.h:43,
from /usr/include/qt/QtWidgets/qscrollarea.h:43,
from /usr/include/qt/QtWidgets/QScrollArea:1,
from /home/rgr/Projects/Digikam/dk/core/showfoto/setup/showfotosetupmetadata.h:29,
from /home/rgr/Projects/Digikam/dk/core/showfoto/setup/showfotosetupmetadata.cpp:24:
/usr/include/c++/7.1.1/type_traits: In instantiation of āstruct std::remove_cv<QBrush>ā:
/usr/include/c++/7.1.1/type_traits:328:12: required from āstruct std::is_integral<QBrush>ā
/usr/include/qt/QtGui/qbrush.h:137:1: required from here
/usr/include/c++/7.1.1/type_traits:1577:67: internal compiler error: Segmentation fault
remove_const<typename remove_volatile<_Tp>::type>::type type;
^~~~
Please submit a full bug report,
with preprocessed source if appropriate.
See <https://bugs.archlinux.org/> for instructions.
make[2]: *** [core/showfoto/CMakeFiles/showfoto.dir/build.make:111: core/showfoto/CMakeFiles/showfoto.dir/setup/showfotosetupmetadata.cpp.o] Error 1
make[2]: *** Waiting for unfinished jobs....
Ok now I think Iām getting somewhere, going to go over what I can dig out of this.
The segfault is not happening specifically for a single piece of code but rather randomly at about the same point during the compilation.
The AMD threads are saying theyāre getting segfaults once every 400ish runs on gst-plugins-bad-1.10.4 so, youāre probably going to have to run it a fair few times in order to get the exception again.