From root Wed Nov 28 11:05:51 1984
>From hpfcll Wed Nov 28 10:41:46 1984 remote from hpfcla
>From mear Wed Nov 28 10:38:36 1984 remote from hpfcll
To: hpfcla!rob
Subject: Firefox memory controller
Status: R

Rob,


My mind has the  following  algorithm for error  handling in the firefox
memory  controller.  Can you please  correct  any  misconceptions  I may
have?  The  memory  controller  can  detect  all single and  double  bit
errors.  Triple (and more) bit errors may or may not be detected.

Single bit  errors  can always be  corrected.  The  syndrome  is used to
identify  the  faulty  bit.  If  the  syndrome   matches  the  value  in
IO_MAP_SYNDROME (an hard  error has been mapped  out), the  error is not
reported.

I am  uncertain of the  following.  If a hard error has been mapped out,
double bit errors can  sometimes be corrected.  The known hard error bit
identified by  IO_MAP_SYNDROME  is corrected, and the resulting bits are
sent back through the ECC logic.  This makes it important  that software
make sure that an error is hard  before it maps it out.  Can double  bit
errors ever be incorrectly "fixed" via this mechanism.  For example, can
a detectable double bit error be turned into an undetectable  triple bit
error via this mechanism?

The Firefox memory  controller  allows one memory chip to be mapped outp
per 2 Mbyte bank.  How does the memory  controller handle the 4Mb board?
Can two hard errors be mapped out, one per bank?  Perhaps a little  more
information on the organization of the board would clear this up for me.

Has anyone given some thought about including the address when computing
the ECC bits.  This would  ensure that the data being read, was what was
written to that address.  This could help isolate addressing failures at
the expense of a few  additional  bits  (hopefully no delays).  Do these
kinds of addressing errors occur in practice?

Here are some possible  errors in the Firefox memory board IRS.  This is
included  for your  information  only, and  should not be  considered  a
critique.


	page 3, first paragraph: 

		will align to a boundary corresponding to *there*
	
	page 6, first paragraph:
		
		Module error will be a 1 if the memory controller detects a
		module error.  Firefox will never have a module error.
		
		Confusing to first time reader?

	page 6, third paragraph:

		When a mappable *hard* error has been determined to be a
		hard error.
	
	page 10, third paragraph:
		
		This type of hard error is detected by counting soft errors
		and at some threshold considering it a *soft* error.

	page 13, section 1.5.2.1.4
		
		(bits 25-32)  

	page 13, section 1.5.2.2
		
		(bits 33-39)


				Charlie Mear
				hpfcla!mear



