Re: hwclock still not working
On Sunday, 8 Nov, John Goerzen wrote:
>
> No. Hmm. I guess that the need for /dev/rtc should have been a hint
> that I need this, but I don't understand why. I've never used it on
> the PC, and the configure description says nothing about it
> interacting with the CMOS clock.... hmm.... I'll enable it and then
> see what happens.
I believe the PC approach is rotten. A user process is not supposed to
talk to hardware directly - that's what kernel drivers are for. Neither
is it supposed to disable interrupts to ensure atomicity of access.
Especially bearing in mind that disabling interrupts won't achieve
atomicity on SMP systems.
The deficiencies of this technique are especially explicit on Alphas.
Archives of axp-list@redhat.com can show how many problems it has
caused. But there is already a driver for that in the kernel; besides
doing things the right way, enabling it is also easier than inventing
crazy workarounds.
However, I don't see why Configure.help should mention things like "CMOS
clock" - are people really more familiar with "CMOS clock" rather than
"realtime clock"? While both terms are a bit inappropriate, "CMOS
clock" is also misleading: it's not the technology that's important
here, and besides, everything is in CMOS nowadays anyway. I know it's
been actively used, at least on PCs, but it's the worst name for that
battery-powered clock chip I've heard so far...
Nikita
Reply to: