ide: don't allow DMA to be enabled if CONFIG_IDEDMA_{ICS,PCI}_AUTO=n

[PATCH] ide: don't allow DMA to be enabled if CONFIG_IDEDMA_{ICS,PCI}_AUTO=n

For CONFIG_IDEDMA_{ICS,PCI}_AUTO=n and/or "ide=nodma" option the host/device are not programmed for DMA and it is also explicitly disabled by ide_set_dma() (->ide_dma_check returns "-1"). However the code responsible for manually enabling DMA ("hdparm -d 1") has a bug which results in DMA being erroneously enabled - ide_set_dma() incorrectly passes "0" return value to set_using_dma(). This may work if BIOS/firmware configured the host/device for DMA and chipset allows independent configuration of DMA/PIO modes but won't work after suspend and is generally unsafe on many chipsets (possibly including data corruption if the same registers are used for DMA/PIO timings).

This patch fixes kernel bugzilla bug #8169 (piix host driver fixes for setting PIO mode exposed the problem described above). The side-effect of the fix is that some rare configuration may be forced to PIO mode when DMA mode was previously used - this is addressed by the next patch which removes CONFIG_IDEDMA_{PCI,ICS}_AUTO config option completely.

Thanks goes out to Patrick Horn for reporting the issue, narrowing it down to the specific commit and testing the fix. Also thanks to Sergei Shtylyov for help in debugging the problem.

---

 drivers/ide/ide-dma.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Index: b/drivers/ide/ide-dma.c
===================================================================
@@ -767,7 +767,7 @@ int ide_set_dma(ide_drive_t *drive)
 	switch(rc) {
 	case -1: /* DMA needs to be disabled */
 		hwif->dma_off_quietly(drive);
-		return 0;
+		return -1;
 	case  0: /* DMA needs to be enabled */
 		return hwif->ide_dma_on(drive);
 	case  1: /* DMA setting cannot be changed */

Hello.

Not only that: it also would affect CompactFlash with no DMA support but
with DMA autotuning enabled in IDE core...

Heh, I've just run into this myself while trying to retro-fit Palmchip
BK3710 IDE driver into 2.6.18 properly.

Timeouts ensued in my case...

WBR, Sergei