On Wed, Feb 16, 2011 at 07:01:06PM +0000, Pawel Moll wrote:

/ # dd if=/dev/mmcblk0 of=/dev/null bs=128k count=10
10+0 records in
10+0 records out
1310720 bytes (1.3MB) copied, 46.539866 seconds, 27.5KB/s
/ # sleep 30
/ # dd if=/dev/mmcblk0 of=/dev/null bs=128k count=10
10+0 records in
10+0 records out
1310720 bytes (1.3MB) copied, 46.540215 seconds, 27.5KB/s


So it does the right thing with decreasing the clock rate in face of
problems, I just can't see it clocking it back up...

You need at least 100 requests before it'll consider clocking back up.
Once it has 100 requests, no more than 5% of them at the current clock
rate must have failed for it to consider clocking up.  I found with
fewer requests, it was forever clocking down, up, down very frequently.