This message is from the T13 list server.
[ BC [EMAIL PROTECTED] ]
Fun question, thank you. Last I checked, the standards left this issue of different
master/slave speed ratings undiscussed, whoops. If in the standard the issue remains
undiscussed, then ...
1)
Marketing folk think a faster burst is always better, no matter how slow the max
sustained transfer rate is. The idea is that you have a list of specifications, and
each specification can count as a win, even if some other specification makes it
meaningless. Arguing against that kind of thinking is a losing battle: too many
customers with money are deluded in the same way. You see such nonsense as people
advertising an "all UDMA" system, as if somehow that were a guarantee of more useful
hardware.
2)
As is normal in interfaces, we here have a pair of dueling should's.
The host "should" be designed to keep the bus as quiet as possible. Each target
"should" be designed to reject as much noise as possible. Nobody does what they
should, but if everybody did, then talking faster to one device should work.
Back in the real world, we see people like Microsoft think they have to push the
choice back on to the luser, with checkmarks on the driver designed to let people slow
the host down. I doubt they are here recklessly complicating your life: I imagine
they have a record of real trouble some need to overcome. Also I hear Microsoft
decides for you that single word Dma and multiword Dma 0 is never worth trying to use.
3)
A host that talks at the faster speed is overclocking the slower device, but hoping
the slower device will never misinterpret that talk.
Overclocking the task file is least safe, because the slow device always has to listen
for talk there, to catch changes in the DEV bit, SRST's, etc. Overclocking old ATAPI
devices might be least safe, because around 1996 some of the cheaper ASIC's required
external glue logic to let the device see things like writes of x1F7 Command while
BSY|DRQ set.
Overclocking PIO bursts of the x1F0 Data register is less safe with hosts that fail to
leave the DA bits quietly zeroed throughout such a burst. Myself I haven't made much
sense of the DMA standards, but I remember thinking UDMA renamed yet more signals than
we of the ANSI & SFF committees had already. Parallel to the technical writing
principle of sticking with one term for each concept, to understand when one kind of
traffic looks like another, you need to use one name for each wire. Me, in such
study, I combine pin numbers with pin names. I don't know if UDMA to Master with a
PIO Slave is regarded as safe because lots of people conservatively choose to dedicate
all of a cable to a UDMA target, or if UDMA hosts commonly deactivate enough of the
PIO signals to persuade slow devices that noone is using the bus.
Pat LaVarre
-----Original Message-----
From: Stephane Cattaneo [mailto:[EMAIL PROTECTED]]
Sent: Mon 10/7/2002 1:55 AM
To: [EMAIL PROTECTED]; [EMAIL PROTECTED]
Cc:
Subject: Re: [t13] Master and Slave speed
This message is from the T13 list server.
I mean communicate with each max speed for each device. (and not use the
slowest speed to communicate with both of them) It lookslike it works,
but I want a confirmation by the specialist....
S.
[EMAIL PROTECTED] wrote:
> This message is from the T13 list server.
>
> Regarding when you have two devices, with two different speeds, can I
> use the fastest speed of the two to communicate with both of them ?
>
> Stephane Cattaneo
> Satellite division
> STMicroelectronics
