Re: Device drivers in Ada (was Re: Ada The Best Language?)

[Kilgallen@eisner.decus.org.nospam (Larry Kilgallen)]

In article <Nlh67.385623$p33.7761469@news1.sttls1.wa.home.com>, "Mark Lundquist" <up.yerz@nospam.com> writes:
> 
> "McDoobie" <chris@dont.spam.me> wrote in message
> news:0Bh57.24338$JN6.5004065@news1.rdc1.mi.home.com...
> 
>> I use Ada95 because for a lot of jobs it's the right tool for me. I write
>> applications using Ada95, I do device drives with Assembler
> 
> Just out of curiosity... have you ever tried writing a device driver in Ada?
> 
> My experience is that Ada is really well-suited for device-level
> programming, and I'm curious about the experiences of others.
> 
> Why do you write device drivers in assembler instead of a higher-level
> language?  Just curious, I'm not being judgemental or anything...

I don't recall what operating system (if any) was specified in the post
from McDoobie, but on VAX VMS device drivers require special use of the
registers.  Even Bliss, with it's fine-grained control over register
usage could only be used in the drivers for pseudo-devices because it
lacked what I would call the wait-for-interrupt-and-keep-channel-linkage.
In Macro one would do something like:

	BSBW	<routine>
	.WORD   168
	<other instructions>

where the inline word was an offset to an alternate return address.

Longword size for all instructions on Alpha eliminated that problem :-),
but the register usage issues remained until Alpha VMS V6.2 when the
ability to write drivers in C was supported, mainly by adding calls
that made standard use of registers.

At this point one could write Alpha VMS device drivers in Ada, given
an Ada compiler that supports 64-bit addressing.  I am not sure that
GNAT does, and I recognize demand is low.  I have had good luck using
DEC Ada against the sorts of VMS data structures that drivers use,
in particular because a representation clause allows one to specify
where the discriminant goes, making it match the standard location
specified by the operating system for determining the nature of a
block of nonpaged pool.

 _think_ (but have not gotten far enough to know for sure) that I
would prefer a mechanism in Ada to specify where the cell (dope vector?)
goes that tells one tagged type from another.  For most drivers tagged
types should not be necessary to meet the requirements set by the
operating system, but I would prefer some control in that regard.
(I realize compiler vendors do not want to be constrained, and tagged
types are certainly more useful within pure Ada code rather than for
interfacing to already-established operating system conventions.