Red Hat Bugzilla – Bug 83762
kswapd consumes cpu/memory/swap during gcc/g++ compile
Last modified: 2007-04-18 12:50:52 EDT
From Bugzilla Helper: User-Agent: Mozilla/5.0 Galeon/1.2.6 (X11; Linux i686; U;) Gecko/20020830 Description of problem: I am running Red Hat linux v8.0, kernel 18-24, running on a P4 with 750+mb of ram and 1.5gb of swap. Sometimes when I 'make' a large package (such as alsa9), my computer freezes and my hard drive starts spinning. I cannot move the mouse, keyboard strokes don't work, and I can't even log on from another terminal. This problem has occurred only while I was doing a 'make (gcc/g++) on programs. It didn't seem to matter WHICH programs I was compiling. After running 'top' to spy on processes, I found that, during a make, the kernel daemon,'kswapd', was taking over 85% of the CPU time and 100% of memory and swap space. Here's the output of top: --- 3:59pm up 1 day, 2:01, 2 users, load average: 22.91, 14.28, 6.48 140 processes: 129 sleeping, 9 running, 2 zombie, 0 stopped CPU states: 2.9% user, 81.7% system, 0.0% nice, 15.2% idle Mem: 764400K av, 757412K used, 6988K free, 0K shrd, 848K buff Swap: 1566328K av, 1566328K used, 0K free 10620K cached 5 root 25 0 0 0 0 RW 78.7 0.0 10:38 kswapd 10560 swells 15 0 4036 3056 844 R 1.1 0.3 1:13 xmms 10568 swells 15 0 256 140 88 S 0.6 0.0 0:12 esd 1002 root 5 -10 151M 4492 780 S < 0.3 0.5 326:28 X 15702 swells 15 0 548 512 264 R 0.3 0.0 0:24 top 6289 swells 15 0 17592 8784 1420 S 0.2 1.1 3:41 galeon-bin 25907 mailman 15 0 1856 1840 576 R 0.2 0.2 0:00 python 25912 mailman 15 0 1620 1620 532 R 0.2 0.2 0:00 python 1898 swells 15 0 38372 3840 392 D 0.1 0.5 0:23 gnome-panel 1906 swells 15 0 10140 2868 584 D 0.1 0.3 2:01 rhn-applet-gui 10647 swells 15 0 3864 1492 648 D 0.1 0.1 0:12 evolution-mail 25908 mailman 15 0 1616 1616 532 D 0.1 0.2 0:00 python 25910 mailman 15 0 1620 1596 532 R 0.1 0.2 0:00 python 25916 mailman 15 0 1376 1372 520 R 0.1 0.1 0:00 python 25922 lp 15 0 472 360 276 D 0.1 0.0 0:00 lpd 1 root 15 0 88 44 28 S 0.0 0.0 0:05 init 2 root 15 0 0 0 0 SW 0.0 0.0 0:00 keventd There are a few more processes, but they seem to be irrelevent When the process finishes, I get: 4:08pm up 1 day, 2:09, 2 users, load average: 7.45, 21.66, 14.78 119 processes: 114 sleeping, 3 running, 2 zombie, 0 stopped CPU states: 0.7% user, 0.5% system, 0.0% nice, 98.6% idle Mem: 764400K av, 102644K used, 661756K free, 0K shrd, 2368K buff Swap: 1566328K av, 146008K used, 1420320K free 30452K cached I am using gcc/g++ version 3.2. Version-Release number of selected component (if applicable): How reproducible: Sometimes Steps to Reproduce: 1.d/l alsa9 2.Run ./configure, then make 3.It might take a few 'makes' before it happens. If so, do a 'make clean' between each make Actual Results: The computer freezes...kswapd eats all my resources. Expected Results: I have plenty of memory and 1.5gb of swap. so this should not have happened. Additional info:
can you use the "M" key in top to sort by biggest memory user and see what's using most ram in teh problem case ?
Here is the printout of 'top' with 'M' set. 11:53am up 4 days, 21:55, 2 users, load average: 12.53, 6.28, 2.95 123 processes: 115 sleeping, 5 running, 2 zombie, 1 stopped CPU states: 0.2% user, 83.5% system, 0.0% nice, 16.1% idle Mem: 764400K av, 757512K used, 6888K free, 0K shrd, 944K buff Swap: 1566328K av, 1566328K used, 0K free 4172K cached PID USER PRI NI SIZE RSS SHARE STAT %CPU %MEM TIME COMMAND 26283 swells 15 0 1971M 651M 200 D 0.0 87.2 1:34 gnome-terminal 25561 swells 15 0 15424 15M 124 D 0.0 2.0 0:01 cc1 1898 swells 15 0 95368 6820 220 D 0.0 0.8 1:04 gnome-panel 1002 root 5 -10 151M 4976 208 D < 0.2 0.6 1318m X 1906 swells 15 0 13512 4488 384 D 0.0 0.5 4:39 rhn-applet-gui 1880 swells 15 0 28408 4260 244 D 0.0 0.5 1:08 metacity 25434 swells 15 0 3496 3492 52 S 0.0 0.4 0:00 make 24697 swells 15 0 2852 2440 92 S 0.0 0.3 0:00 evolution-addre 25564 mailman 15 0 2248 2248 416 D 0.1 0.2 0:00 python 24644 swells 15 0 2548 1840 140 D 0.0 0.2 0:00 evolution 25536 swells 15 0 1340 1340 52 S 0.0 0.1 0:00 make 24714 swells 15 0 3056 1056 92 S 0.0 0.1 0:00 evolution-mail 1900 swells 15 0 24708 900 256 S 0.0 0.1 0:30 nautilus 1077 swells 15 0 3740 828 140 S 0.0 0.1 0:08 gconfd-2 10594 swells 15 0 1372 580 88 S 0.0 0.0 0:00 wombat 25566 mailman 15 0 564 564 404 D 0.0 0.0 0:00 python 24047 swells 16 0 484 440 196 R 0.2 0.0 0:11 top
METOO I can reproduce this bug on one out of five boxes. During a large C++ compile, kswapd occasionally pegs the CPU at 85%, and generates constant disk activity. The system quickly stops responding to all mouse and keyboard input. This does not occur on all compiles. The bug can be reliably reproduced by attempting to open a new gnome-terminal window while the compile is in progress. If I quickly CTRL-ALT-BACKSPACE as soon as I notice kswapd going off the deepend, the system usually recovers about two minutes later. Otherwise a reboot is required. The affected box is a P-II class CPU with 256MB RAM and 512MB swap partition, and an IDE hard disk. The boxes that never exhibit this bug are all SMP boxes with P-III class CPUs, with either 256MB or 512MB RAM, and appropriately-sized swap, and SCSI hard drives. Attaching dmesg + PCI listing from the box exhibit this bug.
Created attachment 90280 [details] dmesg foutput
Created attachment 90281 [details] lspci output. This is an Intel 440BX motherboard.
This problem also happens eventually with gedit (Gnome Edit) as well as Gnome Terminal. I have a P-IV 2.0GHZ with 764MB RAM and 1.5GB swap partition, and an IDE hard disk.
The ever expanding gnome terminal memory hog bug is fixed by RH 9. We've also made changes over time to try and get better performance under thrashing loads. In the cases above it seems the big problem with the gnome-terminal leaking memory making g++ tip it over.