On Thu, Dec 20, 2001 at 10:19:04AM -0500, Tom Perkins wrote:
> Been busy testing the patches (in mass) that we've previously discussed.
Great.
> The "patch" is getting pretty large. The coding is complete I think but
> debugging isn't. Would you like to review the "patch" before I finished
> my testing? And what if anything to you want with it? There are code
> paths I don't have the hardware to test (nothing in true ECP for ex.).
How large? If you like, send it to me in private mail and I'll put it
on people.redhat.com for people to look and and play with.
> And finally, I think the fifo stuck code in function
> "parport_pc_compat_write_block_pio" should drop any attempt at recovery.
Well, we should do something to make it reliable. Whether that should
just be to reset the FIFO and take off the FIFO depth from the byte
count I don't know.
> My reasoning is as follows: In compatible mode your not performing a
> true block operation like in ECP your just sending a window on a stream.
It's the same thing. It's just a byte stream in both cases. (Okay,
ECP has an address-or-data tag for each byte, but still.)
> When the "fifo sticks" in this mode there doesn't appear to be a
> reliable (across all chipsets) way of stopping the transfer and putting
> the chip in ECR_TST.
I think we agreed that ECR_TST was the wrong way to go. You have to
go through mode 000 or 001 before doing anything, and that means
resetting the FIFO. We don't do that now; I think you said your patch
would fix that.
> However, since we've working in compatible mode the
> fifo should be already full at time-out occurrence.
It won't necessarily be full. The printer might be taken off-line
half-way through the last full FIFO run. But it's a small window I
suppose.
> Now if on time-out we immediately switch to ECR_PS2 mode (while we
> have the fifo full condition) we already know the recovery point and
> this would minimize the window for data corruption to microseconds
> at best.
Yes. At least we know that will work, and most of the cases will be
covered.
Tim.
*/
-- To unsubscribe, send mail to: linux-parport-request@torque.net --
-- with the single word "unsubscribe" in the body of the message. --
This archive was generated by hypermail 2b29 : Thu Dec 20 2001 - 11:01:41 EST