OSDev.org

I'll be brief (but I'm not tired this time, so I'll be descriptive too). I have a small kernel with a working IDT (e.g. interrupts I call from the kernel fire the ISR and return to where they're supposed to). I don't have problems with exceptions now, but I want to build in exception error code support into the ISRs now, so I don't have to screw with them later (except tweaking and refining, of course).

My understanding of error codes is that they are 32 bit values at the top of the stack when the ISR is accessed, and they are only pushed for interrupts 8 and 10-14. So, if my standard, working, ISR looks like this (with no error handling): (Where int_master is a C function int_master(int interrupt)). However with error code handling, it would look like this: Where int_master is now int_master(int errorcode, int interrupt) and %2 is the second argument of the macro creating the ISRs, which is 1 for and exception w/error code and 0 for any other interrupt/exception. The nop if there is no error code is to keep the lengths of the ISRs the same as the function I use to fill the IDT with the addresses of all the ISRs relies on their uniform length. If there is no error code, the trash ecx value can be accepted as the error code argument of int_master will be ignored.

Is this a correct understanding?

trashing ecx doesn't sound like a good idea (how will you restore its value ?)

I would rather go for

Another related question: A "push dword 0" statement is 5 bytes long, right? I only ask because when I changed the code to: The code generates several exceptions (rolled up in the 8th interrupt). Again, the only reason this is an issue is that I have an automated IDT filler-outer that requires the ISRs to be identical in size. But this code: Works like a charm (only because the second push dword 0 is never reached, no exceptions are firing). The only thing that I can guess is that push dword 0's size != times 5 nop's size and thus everything would get offset exponentially from where it should be after the first exception with an error code.

Any thoughts would be appreciated.

1. maybe "%if %2 == 0" could give better results.
2. "push dword 0" is actually 5 byte longs. (68 00 00 00 00) except if you're in a [bits 16] segment -- where it becomes (66 68 00 00 00 00)

<sigh>
That's what I thought. Unfortunately (?) I'm in 32 bits and %if %2 == 0 is the same as %if %2 = 0, just as %if %2 != 0 is the same as %if %2 <> 0. The NASM people added these aliases to make C people feel more comfortable.

Oh well, back to debugging the exceptiona