everything failing with "Bus Error"

ExecutorElassus

OK, maybe not everything, but certainly all my web browsers, the internal handshake with BOINCmgr, when using the 3.14.5 kernel also startx ("unable to load the NVIDIA module," though it could yesterday), etc etc.

Maybe I should back up.

Since last night, I've been getting weird failures with applications requiring access to eth0, or that work on a client-server model. Both Opera and chromium, for example, fail to start and return only "Bus Error" as their error message. When I try connecting to the router using konqueror, the connection is refused. BOINC Manager also segfaults when I try to connect to the local process, as does xscreensaver-daemon.

I suspect there's something going on with managing user sessions, but I have no idea where to start. My internet seems to be working fine (I can ping yahoo, at least, but even that fails sometimes due to a "DNS lookup" failure.

Any advice where I might start with troubleshooting?

Cheers,

EE
PS- one addendum which might help: firefox, which connects exclusively through a local tor proxy, works fine. Transmission also seems to be working fine.
PPS- also, I can run opera as root, and my syslog shows a whole lot of DNS errors starting at about 4:30, as well as opnvpn failing TLS key negotiation. I'm not sure if that's related or not

eccerr0r · Posted: Tue Feb 03, 2015 6:48 am Post subject:

Bus error is actually a type of segmentation fault, it's pointing to unimplemented memory.

Usually this means either a binary got corrupted or the CPU is having some problems.

Is there any information in dmesg?
Did you update anything recently?

Does a reboot help?
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

ExecutorElassus · Posted: Tue Feb 03, 2015 7:56 am Post subject:

Well, I recently added the cpu_features to make.conf according to the recent news announcement, but didn't remove them from the USE list. I've updates world a couple times, but nothing major. I'll give a fuller list when I'm back at my desk.

As said, opera works for root.

Stay tuned,

EE
UPDATE: there's nothing in dmesg, but messages has the failed handshakes and a failure message from NVRM: "nv_alloc_contig_pages: DMA address not in adressible range of device 0000:01:00 (0x12aa56000, 0x0-0xffffffff)", which is likely why startx failed. Also, rebooting does not help. Any ideas?

eccerr0r · Posted: Tue Feb 03, 2015 3:07 pm Post subject:

What CPU do you have and what -march/cpuflags are you using?
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

ExecutorElassus · Posted: Tue Feb 03, 2015 8:17 pm Post subject:

Cflags are "-march=athlon64 -O2 -pipe"
Cpu_flags_x86="3dnow 3dnowext mmx mmxext popcnt sse sse2 sse3 sse4a", all of which were already a part of USE.

That doesn't much explain why this started happening this morning, since I did nothing in between. Last emerge was a manual emerge of some packages returned by @preserved-rebuild: mplayer2, transmission (which runs but does not open connections), cdrkit, mesa-progs, libreoffice, and ntp.

Cheers,

EE

ExecutorElassus · Posted: Tue Feb 03, 2015 10:33 pm Post subject:

More info: minitube runs fine, unless I try to go fullscreen, and then it bus-errors out as well. Also, I'm running the 3.18.* kernel, which nvidia doesn't officially support (but which has inexplicably always worked fine). Also, boinc-mgr crashes the moment I start trying to type something into the password field when trying to connect to the locally running boinc process. The field stays blank.

ALSO: I just ran strace on boinc-mgr. At the moment I start trying to type into the password field, the terminal fills with error messages like this:

ExecutorElassus · Posted: Wed Feb 04, 2015 7:25 am Post subject:

UPDATE: after deleting all those gibberish files, and re-running lxappearance to set my theme, opera now starts without a bus error. I can now also fullscreen minitube, konqueror does not crash out when typing in the location bar, and boinc-mgr no longer crashes out when trying to enter its password. However, transmission still won't open connections (but apparently deluge runs fine).

So, "bus error" seems to mean not just the CPU having issues running, but also apparently weird stuff with strangely-named icons (and in my case, something going wrong with Qt-based themes).

Anybody have a guess what's going wrong with transmission?

Cheers,

EE

eccerr0r · Posted: Wed Feb 04, 2015 9:29 pm Post subject:

Yes it does look like you have serious corruption going on there... the former [possible] reason for bus errors. Unfortunately it can be very difficult to figure out what got corrupted. Best to figure out why it's corrupting (bad hardware? etc.) and then re-emerge everything, possibly world with --emptytree and redo everything...
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?