Running code out of RAM

Even if you're code is in ROM, I presume you're used to have data and stack in RAM, right?

Just as it would have with code in ROM.

Same as with code in ROM: pay attention to what you're doing during design and coding -- then test your software.

On Sun, 23 May 2004 22:46:09 -0400, "Haran Vela" wrote in comp.arch.embedded:

In addition to what Grant said, you can (and probably should) test your image in RAM.

Use your development tools to separate writeable sections (initialized variable data, stack space, uninitialized data, etc.) from non-writable sections (code, constant data), that is the ones that are still in EPROM or flash right now.

At system start up, when your initialization code copies the code and constant sections from original storage to SDRAM, compute some soft of checksum or CRC on the image. Then have a background task periodically repeat this calculation. If the background task finds that the image area has been changed, it should lock the system in some sort of fail safe condition and indicate the reason.

Broken code is broken code, whereever it runs.

Buffers should always be checked for overruns. If you are concerned, I think a code review is in order.

Are you worried about the actual reliability of RAM vs. ROM or do you suspect that you have buggy code that is rying to overwrite program memory and failing because it it read-only?

of

of

and

Assuming your code is in one contiguous block of RAM, then if you are really paranoid, you can make a debug board that monitors the address bus / write line and generates an external interrupt on any attempt to write to the code area. Tanya

The rouge problem is cured by forbidding makeup to the sisters :-) After that, write correct code. About the only thing that can harm is cosmic rays, and you can detect most of those (very rare) events with CRC checksums over the load module, or with ECC memory.

Some CPUs have memory protection hardware, so that you can make sections of memory read-only. If this capability exists it's well worth using it; and you can catch hardware exceptions when these illegal writes occur in order to detect the culprits.

There are also poor man's memory protection methods as well, where write enable lines to the RAM pass through a GPIO gateway (with some address logic as well).

Actually both. As most of the postings have indicated, buggy code has to be dealt with through better development and testing processes but what are your views on RAM reliability?

Thanks

first

of

of

and

been a

On Mon, 24 May 2004 22:49:21 GMT, Darin Johnson wrote in comp.arch.embedded:

But none of these features provides any protection against soft errors or actual RAM hardware errors. With any type of DRAM, in particular, it pays to be careful.

Very reliable.

Unless the device is working in some extreme conditions, RAM reliability would never be an issue. You're more likely to have a dicky power supply or ill-designed power-down circuitry.

Ken.

be

Interesting that you should say this as I've had three pieces of "consumer kit" (firewall, router, external disc drive) all fail because of power supply circuit problems recently. I suspect that the power supply tends to get forgotten in a lot of designs (reference designs from which they are derived tending to be "the logic" and the power supply an afterthought).

Andrew

These people make good power supplies:

Since the original code is in (((E)E)P)ROM, why not just copy it from ROM to RAM at frequent intervals ? Some simple memory word parity could be used to trigger the copying operation in addition to the periodic copying. Frequent copying (in order to avoid more than one SEU in a memory word) should not be a problem compared to RAM CRC calculations, unless the RAM:ROM speed ratio is larger than 10:1.

If you do not have a reliable code backup copy in the system, then using ECC RAM and frequently reading and writing back (with corrected memory words and ECC bits) every memory word should largely prevent the formation of two or more bit errors in the same word, which would not be correctable.

Paul

Thank you all for some very useful suggestions. I feel a lot more comfortable now and have some strategies to pursue.

Thanks. Haran.

of

of

and

These people make good power supplies: