Re: aribiter: "broken" master

[Richard Walter <rwalter@corp.auspex.com>]

> 
> Hello,
> 
> PCI spec R2.2, pg. 71, last paragraph, claims that if a master gets a GNT#
> and does not use it within 16 clocks the arbiter may consider it "broken"
> and ignore the master's REQ#.  
> 
> If a master gets GNT# and then "changes it's mind" (well within 16 clocks),

One observation: PCI 2.2, page 70, paragraph 1, sentence 2: "Agents must
only use REQ# to signal a true need to use the bus."  I've always
interpreted this as being that masters are not allowed to change their mind.
If is does, then I'd say it is non-compliant.

> drops the REQ# and never asserts FRAME# are there any arbiters that might
> consider that master "broken?"  Also, if a master does exceed 16 clocks is
> it considered broken for good?  Or just until it releases it's REQ#?
> 
> Are there any known devices whose master exhibits this behavior?
> 
> Thanks in advance,
> Thomas C. Jones
> Sr. MTS
> Sebring Systems, Inc.
> 

-Richard Walter
Auspex Systems, Inc.
Note: I speak for myself, not for Auspex.