memory corruption under heavy load?

Marek Michalkiewicz (marekm@i17linuxb.ists.pwr.wroc.pl)
Mon, 1 Apr 1996 12:45:25 +0200 (MET DST)

Trying to stress-test some new hardware, I did "make -j" on the kernel
sources. It runs for a while, starts swapping a lot, load average goes
up to 30 or so, then cc1 gets fatal signal 11. There is still plenty
of free swap space at this point (20MB total, about 10MB free).

Before everyone will tell me: "it's a hardware problem, read the signal
11 FAQ, Linux has no bugs", read on...

This never happens on the same machine under 1.2.13, 1.3.45 and 1.3.58
(it runs happily until it runs out of swap space). I can reproduce it
every time I try "make -j" on any large source package, under 1.3.80
and 1.3.74. Normal compiles work fine under 1.3.80, just not "make -j"
so I don't think it's a hardware problem.

The symptom is usually signal 11, but sometimes also syntax errors in
perfectly good include files. It looks like memory corruption caused
by some changes between 1.3.58 and 1.3.74. I suspect the new swapping
or page cache code. I can do a binary search, to determine exactly
which patchlevel broke things - just tell me if this is necessary.

Hardware: GigaByte GA-5486AL PCI motherboard with ALI1489 chipset (BTW,
it seems to work fine with normal IDE driver, specifying ide0=ali14xx
slows it down 3MB/s -> 1MB/s!), AMD 5x86-160 CPU, WD AC2850 hard drive
(814MB EIDE), 8MB RAM (two 70ns 4MB PS/2 SIMMs).

Feel free to ask if you need more information. Thanks in advance for
fixing it, and keep up the good work!

Regards,

Marek