SMP enabled on UP kernel
Dave Jones
davej at redhat.com
Thu Feb 23 22:09:06 UTC 2006
On Thu, Feb 23, 2006 at 10:50:49PM +0100, Mostafa Z. Afgani wrote:
> After the last kernel update (2.6.15-1.1831_FC4) the rt2500 (RaLink
> WiFi) driver is causing the system to freeze on modprobe -- on
> restarting the PC the boot process hangs and hence renders the machine
> unusable.
>
> A post on the rt2x00 driver forum led me to the information that SMP is
> enabled even on single processor FC kernels (at least from 2.6.15 on). I
> just though it must be a mistake and dutifully filed a BUG report:
>
> https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=182640
>
> But, within minutes of posting the report, I got a reply that the it is
> NOTABUG & CLOSED! The reason provided was "The performance loss from
> running SMP on a UP x86_64 machine is negligable."
For FC4, this was a mistake. A regression that slipped in when
rebasing a rawhide kernel to FC4 during the 2.6.15 transition.
The update-in-progress (See http://people.redhat.com/davej/kernels/FC4/)
has it reverted, and the UP kernel will behave as UP.
> Now, what I don't understand is what good is SMP on a UP kernel?? If I
> want SMP, I'll install the SMP kernel. I assume I'm not the only one who
> needs to run a UP kernel to make certain hardware work that won't
> otherwise work with SMP! Someone please tell me why anyone would want
> SMP enabled on UP kernels?
For FC5 and onward, for x86-64, there is no -smp kernel.
There are a number of reasons for this
- Cuts down on an extra set of images for the ISOs.
- Increases code-coverage, Everyone runs the same code.
- Is only a tiny performance hit, and work is underway
to make this completely go away.
- Over time, it'll become more difficult to buy a UP x86-64,
when dual-core & hyperthreading become more widespread.
If drivers don't work in SMP, they need to be fixed to do so.
Dave
More information about the fedora-list
mailing list