From mboxrd@z Thu Jan  1 00:00:00 1970
Date: Wed,  1 Nov 1995 07:20:55 -0500
From: forsyth@plan9.cs.york.ac.uk forsyth@plan9.cs.york.ac.uk
Subject: kernel page faults on moving mouse
Topicbox-Message-UUID: 31ff3e38-eac8-11e9-9e20-41e7f4b1d025
Content-Type: text/plain; charset=UTF-8
Message-ID: <19951101122055.vewTkoMjpnRj5NdO0lEJaMARGDuFS1Ln-rc2VAasIUM@z>

has anyone else seen kernel traps on a PC when the mouse is moved
when there is a fair amount of i/o and (probably) external cache
replacement?  i can reproduce the fault fairly regularly
by scrubbing the scroll bar up & down in mothra while it is
converting a moderately large gif for display.
the traps always name 8½ as victim.

the page fault normally shows a bad virtual address of either
0x65 or 0x0, the contents of EAX.  the PC points into the kernel
data space: i believe it is a bbmalloc arena.
the stack traces suggest that the trap occurs
when the mouse handler is moving the software
cursor in an interrupt that seems to be
interrupting a main level graphics operation and/or vgaupdate.
it does seem to be important that something moderately strenuous
be happening in the system at the same time:
disc I/O or ethernet I/O seem equally effective.
i suspect mothra helps to trigger it because it
runs an fb/gif2pic | fb/3to1 pipeline (or similar)
to convert GIFs for display.

the curious thing is that so far i have only managed to
crash AMD DX4/100 machines.
i have not crashed a Cyrix or Intel 386 processor.
unfortunately i haven't got an Intel 486 with the same
configuration that i can try.

i've tried two different makes of motherboard,
both VLB, one with ET4000/W32p VLB and the other with CL5429 VLB.
SVGA is set to either 1024x768x8 or 1152x900x8.
hardware cursors are disabled.

i've looked fairly hard for software
bugs in the devbit.c/devvga.c/bbmalloc.c/libgnot
combination, but i begin to wonder whether the problem is that
the handling of the cache on either the processor or motherboard
is wrong for compiling bitblt.

at this point, i thought i might as well ask the list/group
if anyone else was having problems.