RE: Glitching of PCIRSTN

["Weng Tianxiang" <WTX@umem.com>]

Hi Neal,
Good point!!! I really miss the point when nRESET has a short glitch.

It will take one day or more for me to give a correct answer.

Weng

-----Original Message-----
From: Neal Palmer [mailto:neal@dinigroup.com] 
Sent: Tuesday, November 27, 2001 6:54 PM
To: Weng Tianxiang
Cc: pci-sig@znyx.com
Subject: RE: Glitching of PCIRSTN



Weng,

  you missed the problem of "Glitching of PCIRSTN" in your logic.  If
you get a short glitch of PCIRSTN way after (5 clock cycles or more) the
initial deassertion, then you are back to the problem of having only
some of your FFs reset.

-- Neal

On Tue, 27 Nov 2001, Weng Tianxiang wrote:

> Hi Andrew,
> Thank you for your comments.
>
> The following is the part of my new code in VHDL that absorbs your and

> Neal's advices. I think it will be benefitial to any PCI core 
> designers:
>
> -- at concurrent space
> nRESETGlobal <= nRESET and nRESETBuffer(4);
>
> -- for any FF signals
> AnyProcess : Process(nRESETGlobal, CLK66M)
> Begin
>   if(nRESETGlobal = '0') then
>      -- initialization code
>   elsif(CLK66M'event and CLK66M = '1') then
>      -- register values
>   end if;
> end process;
>
> -- nRESETBuffer(4) is used as internal nRESET deassertion synchronous 
> signal nRESETBufferA : Process(CLK66M)
> Begin
>   if(CLK66M'event and CLK66M = '1') then
>      nRESETBuffer(4 downto 0) <= nRESETBuffer(3 downto 0) & nRESET;
>   end if;
> end process;
>
> 1. What cause us to worry about nRESET is the rising edge of nRESET 
> that may happen at the same time or very close time range with clock 
> rising edge, generating metastability situation with FF. Shift 
> register eliminates the situation.
>
> 2. nRESETGlobal signal is an "AND" operation that guarantees that 
> nRESET assertion will be immediately effective while its rising edge 
> is masked by synchronous signal nRESETBuffer(4), leaving no any clock 
> free running for other FF.
>
> 3. After PCI normal operations, re-assertion of nRESET will cause no 
> trouble. The reason is nGNT will be first deasserted to any PCI 
> boards, leaving all PCI master to linquish PCI bus first.
>
> 4. I only use nRESET signal as nREQ_OE signal. Other OE signals 
> driving to PCI bus has more complex logic, but no nRESET is directly 
> involved.
>
> Any comments and advices are welcome.
>
> Weng
>
> -----Original Message-----
> From: Ingraham, Andrew [mailto:Andrew.Ingraham@compaq.com]
> Sent: Tuesday, November 27, 2001 1:02 PM
> To: Weng Tianxiang
> Cc: pci-sig@znyx.com
> Subject: Re: Glitching of PCIRSTN
>
>
> > The method I will use to get around the situation is to use a 5 
> > levels
>
> > of shift registers as nRESET buffers and use the 5th FF signal as 
> > global synchronous reset signal ...
>
> I think you want to use the OR of both the raw RESET and the 
> shifted/synchronized version.
>
> Or connect the raw version to your output buffer tri-states, and the 
> sync'd version to your state machine logic.
>
> You need your ASIC to go into Reset immediately when nRESET does 
> assert. Especially important that your outputs tri-state and get off 
> the bus right away (see the "Trst-off" spec) and not wait for 5 more 
> clocks to come, which may or may not happen for a while if at all.
>
> Andy
>
>

-- 
-- Neal Palmer

The Dini Group
1010 Pearl St #6
La Jolla, CA 92037
(858) 454-3419 x16
(858) 454-1728 (Fax)