Internal compiler error: Segmentation fault. [solved?]

Message

pablo_supertux · Post by **pablo_supertux** » Sun Jan 31, 2016 5:46 pm

Hi

In the last couple of months, I've been getting this error x86_64-pc-linux-gnu-g++: Internal error: Segmentation fault but extremly sporadically, once every 2-3 months.

I've found this thread here https://forums.gentoo.org/viewtopic-t-9 ... art-0.html and from reading it, I may have problems with my ram.

I have "keepwork" in my FEATURES variable so that /var/tmp/portage isn't wipped out and I can continue building if an error happens. When I then run "make" manually (without -j9), the building process resumes most of the times without a problem. Sometimes I get again Internal error: Segmentation fault at some different file but waiting a couple of minutes "solves" the problem. From this behaviour I could conclude that I indeed have a ram issue, but wouldn't I also be experiencing more gltches than just this g++ error? I mean, appart of this error all other running processes seem to be running just fine and even starting new programs does not lead to unexpected results (I would've expected crashing and segfaults all over the place). But no, everything just keep running fine.

The problem is that I am not able to reproduce this error. Like I said, it happens once every 2 or 3 months, last time I had this issue was late October 2015; I don't have spare ram dimms and buying new ones just for checking an issue I cannot reproduce seems not a good idea.

Another obvservation I've made is that this only happens when I'm compiling some "heavy" c++ packages that use a lot of c++11 or include boost. Compiling C programs so far have never generated this error, C++ do. Could it be that there is something wrong with my toolchain?

I'm going to run memtest over night and see if it shows something, somehow I don't have the feeling this is a ram issue, but I may be wrong.

Has anybody else had this issue without it beeing caused by defective ram? What else could it be happening here?

Post by **asturm** » Sun Jan 31, 2016 5:50 pm

boost build uses a lot of RAM, so it creates exactly the circumstances where you would hit some faulty bit 'high up' in your RAM.

pablo_supertux · Post by **pablo_supertux** » Sun Jan 31, 2016 6:07 pm

hmmm... so I may have a ram proble, after all... that's not good

Post by **NeddySeagoon** » Sun Jan 31, 2016 6:11 pm

pablo_supertux,

Boot memtest86 and run a few cycles.

If you get random errors - not repeatable, its probably not RAM.
If you get the same error at the same address it MAY be RAM.

Don't go buying replacement RAM just yet.

pablo_supertux · Post by **pablo_supertux** » Sun Jan 31, 2016 6:20 pm

thanks, that's a good idea. I'll run memtest a couple of times and see what happens.

pablo_supertux · Post by **pablo_supertux** » Mon Feb 01, 2016 7:11 pm

Well, it seems I do have a ram problem:

http://sakuranohana.org/pics/mem01.jpg
http://sakuranohana.org/pics/mem02.jpg

what is the meaning of the Good and Bad columns?

Post by **NeddySeagoon** » Mon Feb 01, 2016 7:29 pm

pablo_supertux,

Good is the data that was written, bad is the data that was read.
Err-Bits shows the bits that differ. 0x20000 (that's the same one bit) in both cases.

Some things to try.

It looks like you have 4x4G sticks of RAM,
Remove one pair and label them with their RAM slot positions.
Memtest the remaining 8G.
Remove the second pair and label them with their RAM slot positions.
Fit the first pair of DIMMs back in their original slots.
Memtest that 8G.

As its the same data bit failing in the tests, it may be the motherboard or the CPU.

Some motherboards insist on DIMMs being fitted in one pair of of sockets first or they won't boot.
If you have one of those, you need to test each pair of DIMMs in that pair of sockets.

pablo_supertux · Post by **pablo_supertux** » Mon Feb 01, 2016 7:43 pm

NeddySeagoon wrote:pablo_supertux,

It looks like you have 4x4G sticks of RAM,
Remove one pair and label them with their RAM slot positions.
Memtest the remaining 8G.
Remove the second pair and label them with their RAM slot positions.
Fit the first pair of DIMMs back in their original slots.
Memtest that 8G.

I have actually 2x 8GB sticks. The manual of the MB says in this case I should use DIMM_A1 and DIMM_B1. I'll test the sticks separately and see what happens.

pablo_supertux · Post by **pablo_supertux** » Tue Feb 02, 2016 1:59 pm

I checked stick # 1 (alone) on it's original slot and on the other slot, where stick 2 was. In both cases memtest failed to find any error, after one pass the error list remained empty.

I then checked stick #2 in all 4 slots and in every slot memtest found an error (same address, different bit from the day before yesterday) almost immediately in every slot.

So I don't think the problem is the MB, thank goodness, changing the ram is much easier than changing the whole MB.

Post by **NeddySeagoon** » Tue Feb 02, 2016 5:36 pm

pablo_supertux,

That's fairly conclusive.

pablo_supertux · Post by **pablo_supertux** » Tue Feb 02, 2016 8:11 pm

I've installed the new ram I bought today and let memtest have a pass at it, this time it concluded without error. It is the first time to me that a ram stick has failed.

Anyway thank you for the advice. I hope that the error doesn't bite me again.

Post by **asturm** » Tue Feb 02, 2016 8:24 pm

Over the years, I've had so many RAMs fail that I outright refuse to buy any notebook that comes with soldered-on RAM.