[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Sarge, kernel-image, and i586-SMP

Steaphan Greene wrote:
>I would not mind it being slow.  However, since there seems to be
>absolutely no SMP support for 586 OR BELOW, I must roll my own or
>ignore one of my processors.  That's what bothers me.  Even a 386-smp
>kernel would be fine by me.  I just would like to be able to

In case anyone tries this, Debian *really* has no support for SMP on 386; one of the kernel patches which is specfically required for real-i386 is not multi-processor-safe.

(Long story. The patch emulates 486 instructions, one of which is an atomic instruction, but does not do so atomically on multi-processor machines. The patch is required to support i486+-based atomicity.h, which is part of the binary compatibility standard for C++ used by all other Linux distributions, without breaking C++ binary compatibility between i386 and i486 on Debian. We can't comfortably break either type of binary compatibility.)

There's no real reason I can see not to build a 486-smp kernel. There actually are some pretty good arguments for a 586 kernel, because 586 instruction scheduling is wildly different from both 486 and 686 (which are fairly similar to each other).

Christoph Hellwig wrote:
>There were SMP systems based on i386 CPUs from Sequent, and 486-based
>SMP systems both based on the Intel MP spec were aviable from various
>vendors, aswell as stranger designs like the NCR Voyager (also
>supported by Linux)

Well, Debian sarge can't support those Sequent systems, but the 486 systems could be supported.

Reply to: