[Top] [All Lists]

Re: Bye, bye, "generic kernels"

To: "Bradley D. LaRonde" <>
Subject: Re: Bye, bye, "generic kernels"
From: Harald Koerfgen <>
Date: Mon, 19 Jul 1999 21:54:41 +0200 (MEST)
Cc:, SGI Linux <>,
In-reply-to: <006401bed20f$37ebdf60$>
Organization: none
Reply-to: "Harald Koerfgen" <>
On 19-Jul-99 Bradley D. LaRonde wrote:
> Off topic:  I'm wondering - why did I need to apply the R3000 patch for my
> R4000 processor build?  Is it just called the R3000 patch but really it's
> more than that?

Well, the VR4111 doesn't support ll/sc instructions and without the R3000 patch 
you will have no chance to get a working kernel.

> On topic:  I like the fine-grained CPU configuration.  By any chance did you
> include Vr4111 in there?

Not exactly. I was more thinking along the lines of:

"Hey, I have this wonderful CPU with an R4000 core (i.e. R4000 exception
handling) but it doesn't support ll/sc instructions so use the R3000 variant
for atomic operations"


"My CPU is based on an R3000 but it does have ll/sc instructions so use them"

or even

"Yes, my box has an R3000 variant but linux doesn't have to care about the
writeback buffer, some magical hardware does this".

That is already implemented (No big deal, to be honest. Just some
hacking and replacing some "#if (_MIPS_ISA == _MIPS_ISA_MIPS1)" with "#if
!defined (CONFIG_CPU_HAS_LLSC)").

This concept can easily be expanded to, let's say, DMA snooping or "cache
coherent I/O", if needed.

An additional idea is to hide all these details from Joe Average, based on
assumptions from the machine and CPU type, but to make them available to


<Prev in Thread] Current Thread [Next in Thread>