On Wed, 29 Dec 2004 20:12:04 +0000, Alan Cox [off-list ref] wrote:

On Mer, 2004-12-29 at 17:29, Bartlomiej Zolnierkiewicz wrote:

quoted

Let me complain once again :-), libata based driver would be better...

Eventually probably but libata's PATA support is pretty pathetic right
now, it doesn't know anything about PATA drive errata. I did look at it
and when it grows up into a real IDE layer for PATA I'm all for moving
*every* IDE driver to it because some of the IDE error path corner cases

OK

are almost rewrite level fixes (eg DMA changedown, timer/interrupt CD
race oops)

fully agreed

quoted

quoted

+                *      If we are in pass through mode then not much
+                *      needs to be done, but we do bother to clear the
+                *      IRQ mask in case the drives are PIO (eg rev 0x10)
+                *      for now.
+                */

comment or code is wrong, unmask is turned on unconditionally

Disagree. It says "e.g."

"e.g." is okay, "in case the drives are PIO" is not,
unmask also affects DMA (I believe it is not needed for DMA but ...)
 

quoted

quoted

+                       hwif->ide_dma_off_quietly(drive);
+#ifdef CONFIG_IDEDMA_ONLYDISK
+                       if (drive->media == ide_disk)
+#endif
+                               hwif->ide_dma_check(drive);

hack, it looks like fixup code in ide-probe.c need to be moved to probe_hwif()

I'm not sure of the best way to do that cleanly. What do you have in
mind ?

moving ->fixup() execution from probe_hwif_init_with_fixup() to probe_hwif()
(just before code which does PIO / DMA tuning)

From what I see it shouldn't affect current ->fixup users

(and is saner because currently "undecoded slave" is also tuned).

Thanks. I'll go and polish these up.

Thanks.