NetNews Usenet Archive 1992 #19/NN_1992_19.iso/spool/comp/os/vxworks/780

Internet Message Format | 1992-09-01 | 1.4 KB

Path: sparky!uunet!zaphod.mps.ohio-state.edu!cs.utexas.edu!ut-emx!shrike!flash
From: [e-mail] (James W. Melton)
Newsgroups: comp.os.vxworks
Subject: 68040 A32D16 gotcha
Message-ID: <[e-mail]>
Date: 2 Sep 92 02:17:30 GMT
Organization: "Lockheed Austin Division, 6800 Burleson Rd, Austin, TX 78744
Lines: 23

Recently, we had the joy of porting a small, custom device driver
that was developed on an MV147 to a Tadpole TP41. There was a
little gotcha that should be of interest to the net at large.

The device is an ancient beast that operates in the A32D16 VME
space. We have 1 register that contains a 32 bit value.
Apparently, (according to a guy at Tadpole), prior versions of the
680x0 processor family would detect if a 32-bit write was destined
for a D16 address and break the transfer into two 16-bit transfers
to sequential word addresses. Not so with the 68040. If you ask for
a 32-bit write to an address, it attempts to comply. If the target
address is in D16 space, only the low-order 16 bits will be written
to the target address.

If all the technical details are incorrect, please free to correct
me. The bottom line is, when in D16, only do 16-bit transfers.

----
--
Jim Melton, novice guru
email:      [e-mail] | "So far as we know, our
voice mail: [phone]            |  computer has never had
fax:        [phone]            |  an undetected error"