uC for Indirect Execution

On 11 May 2006 17:14:29 -0700, "rektide" wrote in comp.arch.embedded:

I don't know about the Blackfin, but ARM9 processors can most certainly execute code from RAM, both from the internal RAM that they have, and from external RAM, if they have an external bus.

But I have no idea what you mean by "execute code indirectly".

"rektide" skrev i meddelandet news: snipped-for-privacy@y43g2000cwc.googlegroups.com...

Why uCLinux for an ARM9 with an MMU. Curious to know why you want to run Linux without any Ethernet nor LCD?

People often run ucLinux in embedded systems even if there is an MMU. An MMU provides two things - flexibility for things like swap space, which is seldom used in embedded systems, and memory protection between processes. In most embedded systems, you have (or should have!) total control over which processes and programs are running, and don't need this sort of protection. MMU's are not free - they often add latency to memory access, and they can greatly increase processes switching overhead, so there are good reasons not to use one even when it is present. Of course, it all depends entirely on your application - there are plenty of good reasons *for* using the MMU as well.

That presumption is quite universally wrong. Harvard architecture means that the CPU *can* strictly separate code and data memory, because it's designed that way. It doesn't mean that it *has* to be used that way. People have been setting up memory chip selection circuitry for Harvard CPUs to behave like von-Neumann ones since the dawn of the business.

Just consider it: all current PC CPUs are actually Harvard architectures, with separate primary memories for code and data (called the "level 0 cache" or other misguiding names), yet they exactly reproduce the behaviour of the original von-Neumann i386 they've replaced.

The only thing that can actually keep you from executing code out of RAM is if the chip doesn't export its code memory bus to the outside world at all. But that has nothing to do with von Neumann vs. Harvard.

you are wrong. Blackfin most certainly can (and usually does for speed reasons) execute from SDRAM.

Bob

And the ones that _do_ strictly separate their on-board RAM into program and data spaces, such as the Motorola 56000 and Analog's ADSP21xx series, have special instructions to access program memory as data, or to load program RAM from external flash, or both.

The ARM9 datacache gets disabled when you dont have the MMU enabled. This is a real performance killer.

That seems a strange limitation to me - especially as the paper I read showing something like ten times improvement in process switching speed for linux with the MMU disabled was running on an ARM (I forget which). But then, I have never looked at the details of ARMs (at least, not since the days of the Archimedes when I was still at school), so I have no idea of the details.

For some reason, I was under the impression that Harvard uc's often did have this limitation, that it was only more full blooded cpu's that didnt have this restriction.

As to your peripherial query, USB powered wifi.

Although I hadnt noticed there was no ethernet. Ouch. Ah well.

Good info on the datacache, I probably will run real linux.

What limitation?

What are you talking about?

Please quote appropriately.