Date: Fri, 22 Aug 1997 10:01:57 +1100
From: Bill Currie <billc AT blackmagic DOT tait DOT co DOT nz>
Subject: Re: Function Sizes (how to in asm?)
In-reply-to: <UNUWnCAZEI$zEwON@talula.demon.co.uk>
To: Shawn Hargreaves <Shawn AT talula DOT demon DOT co DOT uk>, djgpp AT delorie DOT com
Message-id: <199708212206.KAA25797@teleng1.tait.co.nz gatekeeper.tait.co.nz>
Organization: Tait Electronics Limited
MIME-version: 1.0
Content-type: text/plain; charset=US-ASCII
Content-transfer-encoding: 7BIT
Comments: Authenticated sender is <billc AT blackmagic DOT tait DOT co DOT nz>
Precedence: bulk

On 21 Aug 97 at 18:55, Shawn Hargreaves wrote:
> The functions in src/libc/go32/gopint.c 
> (_go32_dpmi_allocate_iret_wrapper() and friends), allocate a custom
> stack for the interrupt handler, which is locked. So it doesn't
> matter however much data the compiler decides to put on the stack,
> as long as it doesn't overflow the stack size (the default seems to
> be just under 32k, and I can't imagine an interrupt handler ever
> using more than that!).

Nor can I, but you never know.

> 
> As far as I can see the only problems with C code are needing the
> dummy marker functions (seems to work at present but could break
> with future versions of the compiler), and the possibility that the

Or even the present using -O3 as things get inlined out of existence. 
However, this might not be an issue as I believe gcc leaves a copy of 
an inlined function if its address is taken or it is non static. And 
since all functions called by the interrupt must be locked, the 
inlined copies are locked as well and thus won't cause any problems.

> generated code might use external helper functions or global data
> (eg. floating point constants, jump tables for a switch statement

Jump tables are either inlined in the function's code. eg:

	#first part of function
	jmp  jump_table(,%eax,4)
jump_table
	.long case_1
	.long case_2
case_1:
	#code for case 1
	jmp	switch_end
case_2:
	#code for case 2
	jmp	switch_end
switch_end:
	#rest of function

or don't exist at all as a binary search (?? something like it 
anyway) using cmp and jae/be/e/ne for small (and/or sparse?) switch 
statements,

> or helper functions for working with the "long long" data type). As

Could be a problem.  I guess that for interrupt routines, inline asm 
should be used for `long long's.

> long as you are careful in how you lay things out, though, C
> interrupt handlers can be quite reliable, and it's certainly a lot
> easier than doing everything in asm!

Most definitly! It was a tad easier writing my ide driver in C than 
my serial port driver was in asm.  Mind you, getting all those 
volatiles right was a problem.

Bill
--
Leave others their otherness.