Path: utzoo!attcan!uunet!ccicpg!turnkey!conexch!rob
From: rob@conexch.UUCP (Robert Collins)
Newsgroups: comp.sys.intel
Subject: Re: Returning the 80286 to Real Mode
Message-ID: <12365@conexch.UUCP>
Date: 11 Nov 88 08:12:15 GMT
References: <1081@xenon.idec.stc.co.uk> <11367@conexch.UUCP> <3493@pt.cs.cmu.edu>
Reply-To: rob@conexch.UUCP (Robert Collins)
Organization: The Consultants' Exchange, Orange County, CA.  (714) 842-6348
Lines: 59

In article <3493@pt.cs.cmu.edu> ralf@b.gp.cs.cmu.edu (Ralf Brown) writes:
+(someone from Intel will no doubt correct any errors in the following)
+
+In article <11367@conexch.UUCP> rob@conexch.UUCP (Robert Collins) writes:
+}If memory serves me correctly, IBM states that the processor shutdown
+}will typically take 60ms to take place.  One the processor shuts down,
+
+More like 600 microseconds.  If it were 60ms, VDISK could only service 16
+requests per second, for a maximum possible throughput of 64K/s (8 512-byte
+sectors).
+
Thank you, as I said...if memory serves me correctly.  You are very much
right about the 600 microseconds.  But, I might point out that 600uS
is the time for the keyboard controller to respond to the command.  But,
I suppose the time to actually shut down the processor is negligable.
We found, that the ACTUAL time to perform the shutdown was closer to
1mS (as opposed to 600uS).

+}So, the bottom line is this:  Kludge #3 takes ~60ms.  Kludge #4 takes
+}another few ms.  In other words, it's SLOW!  There is another MUCH
+}more elegant way to cause a processor shutdown without using the
+}keyboard controller.  In fact, it takes ~1ms (or less).  This method
+}involves telling the processor you will service NO interrupts, (lidt to
+}a idt w/ 0 entries), then generate an interrupt.  Since the processor
+}can't service the interrupt, it generates an exception.  The exception
+}also can't be service (since you hosed IDT), which will in turn will
+}reset the processor.  The former method is rather well documented in
+
+The processor is only reset by an external signal.  What happens is that
+the triple exception generated by the above puts the processor in SHUTDOWN
+mode.  The PC/AT contains external circuitry to assert the reset line when
+the processor goes into SHUTDOWN.  Since the processor is reset, you still
+need kludges #1, 2, and 4 (magic CMOS byte, return address, FAR jump in
+BIOS).
+

Sorry if I said "reset" I meant shutdown.  Though, I'll admit, I didn't
know (nor know of the) difference between the two.  And sorry if I implied
the other kludges weren't necesary.  They most certainly are necesary, but
using the keyboard controller isn't!

>}executing a 386 instruction on the 286 will cause the desired exception.
>It generates an "undefined opcode" exception.

And that's the whole key.  Generating the "invalid opcode" exception
will lead to a processor shutdown.  Which switches back to real mode,
and starts execution back at 0f000:0fff0.

When we did the timing tests, we found that this method was about 88%
faster than using the keyboard controller...the time was closer to
400uS from the time the invalid ipcode was executed to the time control
was returned to the original program.


-- 
"Worship the Lord your God, and serve him only."  Mat. 4:10
Robert Collins                 UUCP:  ucbvax!ucivax!icnvax!conexch!rob
HOMENET:  (714) 995-7344       UUCP:  uunet!ccicpg!turnkey!conexch!rob
WORKNET:  (714) 229-0284