Messages in this thread | | | Date | Fri, 27 Jun 1997 15:05:45 -0500 | From | Mark Salter <> | Subject | User app corrupting buffer cache. |
| |
I am running into a strange problem where a user-level program is causing the buffer cache to be corrupted. I have verified this corruption on a several 2.0 kernels including 2.0.30, and on 2.1.40. The binary that I am running is a postscript interpreter which sends a rendered bitmap to stdout. I am sure that the application binary has bugs :-) but am concerned that it corrupts the buffer cache.
I am running into this problem on a PPro with 64MB of RAM. I have turned off swapping and am running from the login console with no X windows or networking running.
Here is a typical failure:
> ps_app -obitmap -z foo.ps >/dev/null Segmentation fault
This looks straightforward enough. Just an application bug, you say. Well, the problem is that the buffer cache is now corrupted. If I "cksum ps_app", I'll get a different result than when I cksum it before running. Same goes for foo.ps. If I then:
> cp foo.ps foo.ps.bad
and reboot, foo.ps and foo.ps.bad differ. foo.ps has only printable ascii characters in it, but foo.ps.bad will have some number (usually less than 100) of non-ascii bytes scattered throughout it.
Other parts of the buffer cache will be corrupted also. If I am running emacs and then run ps_app, emacs will often times crash or give various error messages. Trying to start any other apps after running ps_app will usually fail.
I am certain (almost) that the problem is not hardware. This PC running 2.0.30 kernel will run flawlessly under varying and often heavy loads as long as I don't run this particular app.
Any comments or suggestions on how to track down this problem?
-- Mark Salter Member of Technical Staff QMS, Inc. Mobile, AL 36618
| |