[PATCH 1/1] libata-dev: ata_check_atapi_dma() fix for ATA_FLAG_PIO_POLLING LLDDs

[Albert Lee <albertcc@tw.ibm.com>]

ata_check_atapi_dma() fix for LLDDs with the ATA_FLAG_PIO_POLLING flag.

Signed-off-by: Albert Lee <albertcc@tw.ibm.com>
---

We don't support polling DMA.
If the LLDD handles only interrupts in the HSM_ST_LAST state 
(indicated by the ATA_FLAG_PIO_POLLING flag) and the ATAPI device
generates CDB interrupts, we have to use the polling PIO protocol.
(Otherwise, the CDB interrupts might confuse the LLDD.)

We just set the protocol to PIO here in ata_check_atapi_dma().
Later in ata_qc_issue_prot(), the PIO protocol will be set as "polling"
for those LLDDs.

Patch against the irq-pio branch +
(1) irq-pio minor fixes (respin) +
(2) integrate polling pio with irq-pio (respin)

For your review, thanks,

Albert

--- 05_cleanup/drivers/scsi/libata-core.c	2006-03-13 13:29:02.000000000 +0800
+++ 06_check_atapi_dma/drivers/scsi/libata-core.c	2006-03-13 13:29:25.000000000 +0800
@@ -2850,6 +2850,15 @@ int ata_check_atapi_dma(struct ata_queue
 	if (ap->ops->check_atapi_dma)
 		rc = ap->ops->check_atapi_dma(qc);
 
+	/* We don't support polling DMA.
+	 * Use PIO if the LLDD handles only interrupts in
+	 * the HSM_ST_LAST state and the ATAPI device
+	 * generates CDB interrupts.
+	 */
+	if ((ap->flags & ATA_FLAG_PIO_POLLING) &&
+	    (qc->dev->flags & ATA_DFLAG_CDB_INTR))
+		rc = 1;
+
 	return rc;
 }
 /**
@@ -4004,6 +4013,7 @@ unsigned int ata_qc_issue_prot(struct at
 			break;
 		case ATA_PROT_ATAPI_DMA:
 			if (qc->dev->flags & ATA_DFLAG_CDB_INTR)
+				/* see ata_check_atapi_dma() */
 				BUG();
 			break;
 		default: