[syslinux] com32: custom int3 handler

[Jeff Kalikstein]

--- "H. Peter Anvin" <hpa at zytor.com> wrote:
> At the simplest, this could
> look like:
> 
> 	.globl int3_entry
> 	.type int3_entry, @function
> int3_entry:
> 	pushal
> 	call int3
> 	popal
> 	iret
> 	.size int3_entry, .-int3_entry
> 

Call me stubborn, but I just really like keeping my
code inside of C files.  I took your suggestion above
and hacked it into the following (which works!):

void init_handlers()
{
	__asm__(
		"jmp int3_entrydone		\n"
		// the symbol below will not be visible to the C
code,
		// so we will calculate it from the start of
init_handlers.
		// The symbol is placed here, so it can be seen from
the
		// disassembly of the .elf file
		"int3_entry:			\n"
		"pusha				\n"
		"call int3_handler		\n"
		"popa				\n"
		"iret				\n"
		"int3_entrydone:		\n"); 

	struct {
		unsigned long limit	: 16;
		unsigned long base	: 32;
	} __attribute__((packed)) idtr;

	struct {
		unsigned long off_low	: 16;
		unsigned long cs	: 16;
		unsigned long reserved	: 8;
		unsigned long type	: 4;
		unsigned long zero	: 1;
		unsigned long dpl	: 2;
		unsigned long present	: 1;
		unsigned long off_high	: 16;
	} __attribute__((packed)) *idt;
	
	__asm__("sidt (%0)" :: "r"(&idtr));

	idt = (void*)idtr.base;
	
	unsigned short cs;
	__asm__("movw %%cs,%0" : "=rm" (cs));

	// int3_entry calculation based on disassembly of
inline entry
	// code, above
	unsigned long int3_entry = (unsigned
long)init_handlers + 5;
	idt[3].off_low = (unsigned long)int3_entry;
	idt[3].off_high = (unsigned long)int3_entry >> 16;
	idt[3].cs = cs;
	idt[3].type = 0xe;	// 32-bit trap gate
	idt[3].dpl = 0x3;
	idt[3].present = 1;
	idt[3].reserved = 0;
	idt[3].zero = 0;
	
	put_str("trying int3");
	__asm__("int3");
	put_str("DONE!");
}

I couldn't figure out how to let the gcc inline
assembler export it's labels to the rest of the C
code, so you see the ugly little hack I did there.  Is
there a better way?  I'm pretty new to this low level
gcc stuff, and I really appreciate your insight.

> I don't want to turn all of this into API functions,
> because a) I think 
> it's a rare need, b) it can be done without code in
> syslinux, and c) I 
> think it's asking for people who don't understand
> the limitations to try 
> to use it.

I understand and respect your syslinux design and
implementation decisions.  I think syslinux is one of
the slickest open-source projects around.  Thanks!