memset in a interrupt service routine ISR

[barryb]

I am using Version 4.084 on a PIC18F458 processor. We have found that using memset within an ISR is a problem. Memset uses RAM to store the size and value for memset. This gets corrupted if an interrupt fires while inside the memset function (and the ISR also has a memset). I thought the compiler is supposed to disable interrupts for its library functions.

[PCM programmer]

Make it easier for us to help you. Post an extremely short, but complete
and compilable program that shows the problem. (like, 10 lines of code).
And with the #include, #fuses, #use delay(), etc.

[barryb]

It is very easy. Put a memset in an isr, and in the main.

As you can see, memset uses RAM at:
159 @MEMSET.P1
15A-15B @MEMSET.P1

The memset does an RCALL to 0x9e, but I don't see any interrupt disable, therefore both the interrupt and the main will use 0x159-0x15B for the memset memory.

Now, I must wonder if the other functions such as memcpy, strcpy, etc have the same problem.

[PCM programmer]

With your version (v4.084), I think the best way to fix the problem is just
to protect calls outside the isr with disable/enable interrupts, as shown
below in bold.
I looked at two different solutions and it seems that in the ASM code for
your version, the compiler still shares RAM locations for parameters. It
doesn't do that in vs. 4.128. That's why solutions that look as if they
might work with the latest version probably won't work with your version.
So the following code is probably the best or easiest way to do it.

[FvM]

Honestly speaking, I won't expect memset used in an ISR. Possibly CCS neither did, because they are usually protecting
all non-relocatable RTL functions with an automatic interrupt disable which is missing in this case.

But it should be easy for CCS to include memset in the list of protected functions.

[PCM programmer]

Here is a thread where it is said that CCS disables interrupts in too
many of the built-in functions:
http://www.ccsinfo.com/forum/viewtopic.php?t=25535
Probaly because of that thread (or an email) CCS added this #device option:

[barryb]

I looked over the bug fixes since v4.084, and I don't see any mention of changing memset (or any other function) for use in interrupts.


Also, the FAQ says the following (so why don't they check their own library functions):
In order to prevent the above problem, the compiler will "protect" the function call to A() from main() by disabling all interrupts before the call to A() and restoring the interrupt state after A() returns. In doing so, the compiler can allow complete sharing of functions between the main program and the interrupt functions.

The programmer must take the following special considerations into account:
1.In the above example, interrupts will be disabled for the entire execution of A(). This will increase the interrupt latency depending on the execution time of A().

2.If the function A() changes the interrupts using ENABLE/DISABLE _INTERRUPTS then the effect may be lost upon the return from A(), since the entire INTCON register is saved before A() is called and restored afterwards. Furthermore, if the global interrupt flag is enabled in A(), the program may execute incorrectly.

3.A program should not depend on the interrupts being disabled in the above situation. The compiler may NOT disable interrupts when the function or any function it calls requires no local RAM.

4.The interrupts may be disabled, as described above for internal compiler functions called by the same manor. For example, multiplication invoked by a simple * may have this effect.