2012-01-22 02:27:43

by Larry Finger

[permalink] [raw]
Subject: Re: (Very old) Regression following commit bfc0f59

On 01/21/2012 08:18 PM, Larry Finger wrote:
> I recently discovered that an ancient laptop built by Mtech and containing an
> AMD K6 450/2 CPU was generating errors when using the Cardbus versions of
> Broadcom cards using either b43 of b43legacy. Expecting some change in those
> drivers or ssb, I found a kernel old enough to work without the error, and did a
> bisection. I was very surprised when "x86: merge tsc calibration", which is
> commit bfc0f5947afa5e3a13e55867f4478c8a92c11dca (dated July 1, 2008), turned out
> to be the source of the regression. It was verified by generating a reversion
> patch. One main difference is that the bad code gets a processor speed of
> 214.398 MHz, whereas reverting the commit yields 428.845 MHz, which is more in
> keeping with the stated frequency of 450 MHz. As the machine is more than 10
> years old, the clock may have drifted considerably.

The newest kernel that was running on this laptop was 3.2-rc4. When I updated to
3.3-rc1, the problem above was fixed, thus nothing needs to be done.

Larry