Has anyone noticed problems with io accounting in /proc/diskstats?
I wrote a simple program to open a large block special device and
then read blocks at random byte offsets in a hard loop. For a minute
or so the counters for the device in /proc/diskstats show the expected
increments, but then inexplicably all the counters stop incrementing.
This has been observed with 2.6.12-rc3 and various recent git trees.
I'm guessing it's related to the recent io accounting changes in
ll_rw_blk.c but haven't investigated any further at this stage.
Thanks
-- Mark