[PATCH] libata: implement qc->result_tf

Add qc->result_tf and ATA_QCFLAG_RESULT_TF.  This moves the
responsibility of loading result TF from post-compltion path to qc
execution path.  qc->result_tf is loaded if explicitly requested or
the qc failsa.  This allows more efficient completion implementation
and correct handling of result TF for controllers which don't have
global TF representation such as sil3124/32.

Signed-off-by: Tejun Heo <htejun@gmail.com>
diff --git a/drivers/scsi/libata-core.c b/drivers/scsi/libata-core.c
index 19ae3fa..51cb9ca 100644
--- a/drivers/scsi/libata-core.c
+++ b/drivers/scsi/libata-core.c
@@ -955,7 +955,6 @@
 {
 	struct completion *waiting = qc->private_data;
 
-	qc->ap->ops->tf_read(qc->ap, &qc->tf);
 	complete(waiting);
 }
 
@@ -997,6 +996,7 @@
 	qc->tf = *tf;
 	if (cdb)
 		memcpy(qc->cdb, cdb, ATAPI_CDB_LEN);
+	qc->flags |= ATA_QCFLAG_RESULT_TF;
 	qc->dma_dir = dma_dir;
 	if (dma_dir != DMA_NONE) {
 		ata_sg_init_one(qc, buf, buflen);
@@ -1034,7 +1034,7 @@
 	/* finish up */
 	spin_lock_irqsave(&ap->host_set->lock, flags);
 
-	*tf = qc->tf;
+	*tf = qc->result_tf;
 	err_mask = qc->err_mask;
 
 	ata_qc_free(qc);
diff --git a/drivers/scsi/libata-scsi.c b/drivers/scsi/libata-scsi.c
index b0c83c2..ce90b63 100644
--- a/drivers/scsi/libata-scsi.c
+++ b/drivers/scsi/libata-scsi.c
@@ -537,7 +537,7 @@
 void ata_gen_ata_desc_sense(struct ata_queued_cmd *qc)
 {
 	struct scsi_cmnd *cmd = qc->scsicmd;
-	struct ata_taskfile *tf = &qc->tf;
+	struct ata_taskfile *tf = &qc->result_tf;
 	unsigned char *sb = cmd->sense_buffer;
 	unsigned char *desc = sb + 8;
 
@@ -608,7 +608,7 @@
 void ata_gen_fixed_sense(struct ata_queued_cmd *qc)
 {
 	struct scsi_cmnd *cmd = qc->scsicmd;
-	struct ata_taskfile *tf = &qc->tf;
+	struct ata_taskfile *tf = &qc->result_tf;
 	unsigned char *sb = cmd->sense_buffer;
 
 	memset(sb, 0, SCSI_SENSE_BUFFERSIZE);
@@ -1199,14 +1199,11 @@
 	 */
 	if (((cdb[0] == ATA_16) || (cdb[0] == ATA_12)) &&
  	    ((cdb[2] & 0x20) || need_sense)) {
-		qc->ap->ops->tf_read(qc->ap, &qc->tf);
  		ata_gen_ata_desc_sense(qc);
 	} else {
 		if (!need_sense) {
 			cmd->result = SAM_STAT_GOOD;
 		} else {
-			qc->ap->ops->tf_read(qc->ap, &qc->tf);
-
 			/* TODO: decide which descriptor format to use
 			 * for 48b LBA devices and call that here
 			 * instead of the fixed desc, which is only
@@ -1217,10 +1214,8 @@
 		}
 	}
 
-	if (need_sense) {
-		/* The ata_gen_..._sense routines fill in tf */
-		ata_dump_status(qc->ap->id, &qc->tf);
-	}
+	if (need_sense)
+		ata_dump_status(qc->ap->id, &qc->result_tf);
 
 	qc->scsidone(cmd);
 
@@ -2004,7 +1999,6 @@
 		 * a sense descriptors, since that's only
 		 * correct for ATA, not ATAPI
 		 */
-		qc->ap->ops->tf_read(qc->ap, &qc->tf);
 		ata_gen_ata_desc_sense(qc);
 	}
 
@@ -2080,7 +2074,6 @@
 		 * a sense descriptors, since that's only
 		 * correct for ATA, not ATAPI
 		 */
-		qc->ap->ops->tf_read(qc->ap, &qc->tf);
 		ata_gen_ata_desc_sense(qc);
 	} else {
 		u8 *scsicmd = cmd->cmnd;
@@ -2361,6 +2354,9 @@
 	 */
 	qc->nsect = cmd->bufflen / ATA_SECT_SIZE;
 
+	/* request result TF */
+	qc->flags |= ATA_QCFLAG_RESULT_TF;
+
 	return 0;
 
  invalid_fld:
diff --git a/include/linux/libata.h b/include/linux/libata.h
index 0e1a3be..a4b8a41 100644
--- a/include/linux/libata.h
+++ b/include/linux/libata.h
@@ -162,7 +162,9 @@
 	ATA_QCFLAG_SINGLE	= (1 << 2), /* no s/g, just a single buffer */
 	ATA_QCFLAG_DMAMAP	= ATA_QCFLAG_SG | ATA_QCFLAG_SINGLE,
 	ATA_QCFLAG_IO		= (1 << 3), /* standard IO command */
-	ATA_QCFLAG_EH_SCHEDULED = (1 << 4), /* EH scheduled */
+	ATA_QCFLAG_RESULT_TF	= (1 << 4), /* result TF requested */
+
+	ATA_QCFLAG_EH_SCHEDULED = (1 << 16), /* EH scheduled */
 
 	/* host set flags */
 	ATA_HOST_SIMPLEX	= (1 << 0),	/* Host is simplex, one DMA channel per host_set only */
@@ -343,7 +345,7 @@
 	struct scatterlist	*__sg;
 
 	unsigned int		err_mask;
-
+	struct ata_taskfile	result_tf;
 	ata_qc_cb_t		complete_fn;
 
 	void			*private_data;
@@ -824,6 +826,10 @@
 	qc->err_mask = 0;
 
 	ata_tf_init(qc->ap, &qc->tf, qc->dev->devno);
+
+	/* init result_tf such that it indicates normal completion */
+	qc->result_tf.command = ATA_DRDY;
+	qc->result_tf.feature = 0;
 }
 
 /**
@@ -839,9 +845,15 @@
  */
 static inline void ata_qc_complete(struct ata_queued_cmd *qc)
 {
+	struct ata_port *ap = qc->ap;
+
 	if (unlikely(qc->flags & ATA_QCFLAG_EH_SCHEDULED))
 		return;
 
+	/* read result TF if failed or requested */
+	if (qc->err_mask || qc->flags & ATA_QCFLAG_RESULT_TF)
+		ap->ops->tf_read(ap, &qc->result_tf);
+
 	__ata_qc_complete(qc);
 }