This message is from the T13 list server.

> From: Hale Landis
> ... call to order ...

I'm trying to reply chronologically, so I will reply
to that next.

> From: Larry Barras [mailto:[EMAIL PROTECTED]]
> Sent: Mon 1/13/2003 6:31 PM
...
> the drivers should be able to figure out how to
> accomplish this without hanging the drive/ system/
> bus, etc.
>
> If you are going to play like a driver, then you
> have to take on all responsibility of a driver,
> including figuring out what might work and what
> might not, how not to leave hardware in a bad
> state, how to respond to errors and so forth.

I think this is key, thank you.

I'm saying counting data bytes matters because I'm
saying the author of a standard device driver, or
other bridge, cannot meet these challenges except by
sending only Cdb's authored by itself to only
compliant devices.  But that hypothetical limitation
is impractical: it precludes old proprietary and
newly standard ops.  It requires more collusion than
business permits between PCI/ATAPI host folk and
ATAPI device folk.  It doesn't actually happen.

What happens instead is that programmers privately &
individually decide to swallow a little imprecision:
they let the byte counts go sloppy, and damn the
consequences, which are difficult to characterise
because they occur only when host & device folk
misunderstand each other, which is rare except for
subtleties.

If it doesn't make the O.S. crash more often than
usual, is it bug?

> Subject: [t13] USB vs. ATA/ATAPI data xfer matrix
> From: Hale Landis [mailto:[EMAIL PROTECTED]]
> Sent: Tue 1/14/2003 9:29 AM

This I was delighted to see.

Surely a quick way we can learn some of what might be
usefully added in t13 is to review what other
transport folk have thought was worth specifying.

> Subject: [t13] the right place?
> Sent: Tue 1/14/2003 10:29 AM

This I was less delighted to see.

If this is the wrong place, then how about we just
let it die.  We don't have to reply to tell people
they are boring us: silence will speak volumes.

On the other hand, if every time we mention the
subject we find new people to say new things, then
maybe there is something real here after all.

> Is this the right place to discuss other interfaces

Yes, when we reason by analogy.

> Is this the right place to discuss ...
> broken OS driver stacks?

Yes, when "everyone" is "broken" in the same way.

If we find we have specified that everyone is broken,
then we have to ask if our spec is broken instead.

On this issue in particular, we can't meaningfully
specify that the host & the device always understand
each other: that's not measurable.  We can say (and
here, in effect, we have said) the effect is
unspecified when the host models the device
inaccurately.  But that neglect doesn't help us all
interoperate.

> A host could be a traditional PCI bus ATA host
> adapter and the associated OS device driver stack.

Also known as a chain of soft "bridge"s terminating
in the PCI/IDE hardware "bridge".

> we should agree on how the ATA/ATAPI interface
> works (for good or bad - it works the way it works)

Yes.

Demonstrably, in bus traces, on the desktop, ATAPI
does Not actually work by having the host & device
agree always over how many data bytes were copied
which way.  Instead, the host actually glosses over
disagreements on a vendor-specific basis (Larry said
"reconciles" when "possible"), with a complete
standard solution available only in ATAPI PIO, not
DMA, as yet.

> ...

After all, posting to tell people they are boring
often reproduces the result that online
metadiscussion grows without bound.

I'm liking the one post per author per day limit for
this topic.  I think in this round our signal to
noise ratio has gone up.

> From: [offline]
> Sent: Tue 1/14/2003 1:09 PM
>
> I just asked the questions because it seems Larry
> ... and Hale have missed the points.

I don't entirely agree.  I think we've gotten farther
along this time than last.  This time I see people
asking key questions.

Q1: Does any desktop O.S. pass thru just the Cdb,
rather than Cdb with Direction and MaxLength of data
to copy?

A: No, never, none, "everybody knows" the Cdb
doesn't intelligibly tell you Direction and
MaxLength.

Q2: Can all PCI/IDE bridges copy a count of bytes?

A: No.  Today some round up to multiples of 2 or 4.
We can be sure multiples of 8 will come next.

Q3: Does the host own the job of keeping the bus
alive, and not accessing unallocated memory, no
matter how creatively the device decides Direction
and MaxLength, in response to the ATAPI bus itself
having passed thru only the Cdb, rather than the
whole command?

A: Yes.

> Subject: RE: [t13] is ATAPI DMA broken?
> From: Hale Landis [mailto:[EMAIL PROTECTED]]
> Sent: Tue 1/14/2003 11:00 AM
...
>  Here is a message I sent to Pat some days ago:

Online now!  Thank you.  Difficult for me to make
sense online when half the conversation occurs
offline.

By the way, I'm not saying ATAPI DMA is "broken".

I'm saying T13 ATAPI DMA standardises less of
transport than does ATAPI PIO.  I'm saying T13 ATAPI
DMA could be more perfect, because as yet it can't be
used in a purely standard way to reproduce precisely
the effects that ATAPI PIO actually has on memory.

Pat LaVarre

Reply via email to