Messages in this thread | | | Date | Sun, 4 Jan 2015 12:57:09 -0800 | Subject | Re: [PATCH] Revert 9fc2105aeaaf56b0cf75296a84702d0f9e64437b to fix pyaudio (and probably more) | From | Linus Torvalds <> |
| |
On Sun, Jan 4, 2015 at 12:45 PM, Nicolas Pitre <nicolas.pitre@linaro.org> wrote: > > Because we're discussing a choice between two evils.
.. and Pavel pointed you to several screenfuls of google hits for people complaining about this.
End of discussion. Seriously. Your whinging about "support costs" is just crying over the fact that you have users. Deal with it.
I've reverted the change for now, since that's clearly better than leaving things broken. And you should acknowledge that. In fact, you should *more* than acknowledge that, you need to internalize it and *understand* it, instead of arguing.
The fact is, we have bogomips on x86 too, and they've been around for a hell of a lot longer than ARM has been. The number may not be "meaningful" any more (since it's TSC cycles rather than anything else), but it's there in /proc anyway. We had it through the times when it fluctuated, and your made-up arguments about how it's a bad thing are just bogus. It wasn't a big deal, and having a reasonable value is not a problem.
And no, we don't set it to some "obviously bogus value". That just makes breakage even more subtle.
You could make it be something that is roughly on the order of the CPU frequency. It doesn't have to be exact, but from a QoI standpoint, try to do a reasonable job.
And dammit, I really never *ever* want to hear arguments against fixing regressions ever again. It really is the #1 rule for the kernel. There is *no* excuse for that NAK. There is only "sorry".
Linus
| |