Hi Greg,
Am 08.08.2024 um 18:05 schrieb Greg Ungerer:
Hi Michael, Finn,
On 8/8/24 11:57, Finn Thain wrote:
Hello Michael
On Thu, 8 Aug 2024, Michael Schmitz wrote:
Well, that __clear_user() bug is no longer theoretical. I accidentally
bumped into it when I sent a ^C to a shell script I wrote to test some
mac_scsi driver patches...
...
I'm afraid I've lost track of where we're at with this patch series.
Does it need more work, or more bug reports such as the one below?
Apparently the series is waiting for some testing on a Coldfire system
with MMU.
Ok, I am in a state that I can do that now (I managed to fix my M5475EVB
board).
If I test the v4 versions of this patch set that should do the job?
Yes, v4 is the one to test.
I'll see that I can dig out my test code to reproduce the bug.
Thanks,
Michael
Regards
Greg
The previous bug reports might be considered somewhat contrived but this
one's from 'real' user space code, and none too complex at that?
Right. That code was as follows. There's nothing here aimed at
arch/m68k/lib/uaccess.c in particular, just IO to a block device and a
tmpfs filesystem.
#!/bin/bash
set -e -u
filename=$1
rand=/tmp/rand_test_data
zero=/tmp/zero_test_data
len=512K
dd if=/dev/urandom of=$rand bs=$len count=1 2>/dev/null
dd if=/dev/zero of=$zero bs=$len count=1 2>/dev/null
write() {
dd $args if=$orig of=$filename # oflag=direct
}
compare() {
echo 3 > /proc/sys/vm/drop_caches
if ! cmp -n $len $orig $filename ; then
diff -u <(hexdump -C < $orig) <(hexdump -C < $filename)
fi
}
while true; do
for args in "bs=512k count=1" "bs=64k count=8" "bs=4k
count=128" "bs=512 count=1k" ; do
echo $args
orig=$rand ; write ; compare
orig=$zero ; write ; compare
sync
echo
done
done
