libata: Better timeout recovery

index c1db2f2..fa48031 100644
@@ -515,7 +515,7 @@ void ata_scsi_error(struct Scsi_Host *host)
 
 	/* For new EH, all qcs are finished in one of three ways -
 	 * normal completion, error completion, and SCSI timeout.
-	 * Both cmpletions can race against SCSI timeout.  When normal
+	 * Both completions can race against SCSI timeout.  When normal
 	 * completion wins, the qc never reaches EH.  When error
 	 * completion wins, the qc has ATA_QCFLAG_FAILED set.
 	 *

@@ -530,7 +530,19 @@ void ata_scsi_error(struct Scsi_Host *host)
 		int nr_timedout = 0;
 
 		spin_lock_irqsave(ap->lock, flags);
-
+		
+		/* This must occur under the ap->lock as we don't want
+		   a polled recovery to race the real interrupt handler
+		   
+		   The lost_interrupt handler checks for any completed but
+		   non-notified command and completes much like an IRQ handler.
+		   
+		   We then fall into the error recovery code which will treat
+		   this as if normal completion won the race */
+
+		if (ap->ops->lost_interrupt)
+			ap->ops->lost_interrupt(ap);
+			
 		list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
 			struct ata_queued_cmd *qc;
 

@@ -574,6 +586,9 @@ void ata_scsi_error(struct Scsi_Host *host)
 		ap->eh_tries = ATA_EH_MAX_TRIES;
 	} else
 		spin_unlock_wait(ap->lock);
+		
+	/* If we timed raced normal completion and there is nothing to
+	   recover nr_timedout == 0 why exactly are we doing error recovery ? */
 
  repeat:
 	/* invoke error handler */

142 lines not on this page · raw message

index 15cdb91..eb50be8 100644
@@ -17,7 +17,7 @@
 #include <linux/libata.h>
 
 #define DRV_NAME "pata_isapnp"
-#define DRV_VERSION "0.2.2"
+#define DRV_VERSION "0.2.5"
 
 static struct scsi_host_template isapnp_sht = {
 	ATA_PIO_SHT(DRV_NAME),

@@ -28,6 +28,13 @@ static struct ata_port_operations isapnp_port_ops = {
 	.cable_detect	= ata_cable_40wire,
 };
 
+static struct ata_port_operations isapnp_noalt_port_ops = {
+	.inherits	= &ata_sff_port_ops,
+	.cable_detect	= ata_cable_40wire,
+	/* No altstatus so we don't want to use the lost interrupt poll */
+	.lost_interrupt = ATA_OP_NULL,
+};
+
 /**
  *	isapnp_init_one		-	attach an isapnp interface
  *	@idev: PnP device

@@ -65,7 +72,7 @@ static int isapnp_init_one(struct pnp_dev *idev, const struct pnp_device_id *dev
 
 	ap = host->ports[0];
 
-	ap->ops = &isapnp_port_ops;
+	ap->ops = &isapnp_noalt_port_ops;
 	ap->pio_mask = 1;
 	ap->flags |= ATA_FLAG_SLAVE_POSS;
 

@@ -76,6 +83,7 @@ static int isapnp_init_one(struct pnp_dev *idev, const struct pnp_device_id *dev
 					   pnp_port_start(idev, 1), 1);
 		ap->ioaddr.altstatus_addr = ctl_addr;
 		ap->ioaddr.ctl_addr = ctl_addr;
+		ap->ops = &isapnp_port_ops;
 	}
 
 	ata_sff_std_ports(&ap->ioaddr);

index d3f2c0d..d240d08 100644
@@ -42,7 +42,7 @@
 
 
 #define DRV_NAME "pata_pcmcia"
-#define DRV_VERSION "0.3.3"
+#define DRV_VERSION "0.3.5"
 
 /*
  *	Private data structure to glue stuff together

@@ -126,6 +126,38 @@ static unsigned int ata_data_xfer_8bit(struct ata_device *dev,
 	return buflen;
 }
 
+/**
+ *	pcmcia_8bit_drain_fifo - Stock FIFO drain logic for SFF controllers
+ *	@ap: port to drain
+ *	@qc: command
+ *
+ *	Drain the FIFO and device of any stuck data following a command
+ *	failing to complete. In some cases this is neccessary before a
+ *	reset will recover the device.
+ *
+ */
+ 
+void pcmcia_8bit_drain_fifo(struct ata_queued_cmd *qc)
+{
+	int count;
+	struct ata_port *ap;
+
+	/* We only need to flush incoming data when a command was running */
+	if (qc == NULL || qc->dma_dir == DMA_TO_DEVICE)
+		return;
+
+	ap = qc->ap;
+
+	/* Drain up to 64K of data before we give up this recovery method */
+	for (count = 0; (ap->ops->sff_check_status(ap) & ATA_DRQ)
+							&& count < 65536;)
+		ioread8(ap->ioaddr.data_addr);
+
+	/* Can become DEBUG later */
+	ata_port_printk(ap, KERN_WARNING, "drained %d bytes to clear DRQ.\n",
+								count);
+
+}
 
 static struct scsi_host_template pcmcia_sht = {
 	ATA_PIO_SHT(DRV_NAME),

@@ -143,6 +175,7 @@ static struct ata_port_operations pcmcia_8bit_port_ops = {
 	.sff_data_xfer	= ata_data_xfer_8bit,
 	.cable_detect	= ata_cable_40wire,
 	.set_mode	= pcmcia_set_mode_8bit,
+	.drain_fifo	= pcmcia_8bit_drain_fifo,
 };
 
 #define CS_CHECK(fn, ret) \

index 225bfc5..e4a72f6 100644
@@ -767,6 +767,7 @@ struct ata_port_operations {
 	ata_reset_fn_t		pmp_hardreset;
 	ata_postreset_fn_t	pmp_postreset;
 	void (*error_handler)(struct ata_port *ap);
+	void (*lost_interrupt)(struct ata_port *ap);
 	void (*post_internal_cmd)(struct ata_queued_cmd *qc);
 
 	/*

@@ -808,6 +809,8 @@ struct ata_port_operations {
 	void (*bmdma_start)(struct ata_queued_cmd *qc);
 	void (*bmdma_stop)(struct ata_queued_cmd *qc);
 	u8   (*bmdma_status)(struct ata_port *ap);
+
+	void (*drain_fifo)(struct ata_queued_cmd *qc);
 #endif /* CONFIG_ATA_SFF */
 
 	ssize_t (*em_show)(struct ata_port *ap, char *buf);

@@ -1516,6 +1519,7 @@ extern bool ata_sff_qc_fill_rtf(struct ata_queued_cmd *qc);
 extern unsigned int ata_sff_host_intr(struct ata_port *ap,
 				      struct ata_queued_cmd *qc);
 extern irqreturn_t ata_sff_interrupt(int irq, void *dev_instance);
+extern void ata_sff_lost_interrupt(struct ata_port *ap);
 extern void ata_sff_freeze(struct ata_port *ap);
 extern void ata_sff_thaw(struct ata_port *ap);
 extern int ata_sff_prereset(struct ata_link *link, unsigned long deadline);

@@ -1528,6 +1532,7 @@ extern int ata_sff_softreset(struct ata_link *link, unsigned int *classes,
 extern int sata_sff_hardreset(struct ata_link *link, unsigned int *class,
 			       unsigned long deadline);
 extern void ata_sff_postreset(struct ata_link *link, unsigned int *classes);
+extern void ata_sff_drain_fifo(struct ata_queued_cmd *qc);
 extern void ata_sff_error_handler(struct ata_port *ap);
 extern void ata_sff_post_internal_cmd(struct ata_queued_cmd *qc);
 extern int ata_sff_port_start(struct ata_port *ap);