MLK-11444 ata: imx: cmd buf corruption errata bug fix

errata:
When a read command returns less data than specified in the PRDs (for
example, there are two PRDs for this command, but the device returns a
number of bytes which is less than in the first PRD), the second PRD of
this command is  not read out of the PRD FIFO, causing the next command
to use this PRD erroneously.

workaround
- forces sg_tablesize = 1
- modified the sg_io function in block/scsi_ioctl.c to use a 64k buffer
  allocated with dma_alloc_coherent during the probe in ahci_imx
- In order to fix the scsi/sata hang, when CD_ROM and HDD are
  accessed simultaneously after the workaround is applied.
  Do not go to sleep in scsi_eh_handler, when there is host failed.

Signed-off-by: Richard Zhu <Richard.Zhu@freescale.com>

4 files changed, 76 insertions, 6 deletions

diff --git a/block/blk-map.c b/block/blk-map.c
index e31be14da8ea..dcfb489a0c8d 100644
--- a/block/blk-map.c
+++ b/block/blk-map.c
@@ -206,6 +206,12 @@ int blk_rq_unmap_user(struct bio *bio)
 }
 EXPORT_SYMBOL(blk_rq_unmap_user);
 
+#ifdef CONFIG_AHCI_IMX
+extern void *sg_io_buffer_hack;
+#else
+#define sg_io_buffer_hack NULL
+#endif
+
 /**
  * blk_rq_map_kern - map kernel data to a request, for passthrough requests
  * @q:		request queue where request should be inserted
@@ -233,7 +239,14 @@ int blk_rq_map_kern(struct request_queue *q, struct request *rq, void *kbuf,
 	if (!len || !kbuf)
 		return -EINVAL;
 
-	do_copy = !blk_rq_aligned(q, addr, len) || object_is_on_stack(kbuf);
+#ifdef CONFIG_AHCI_IMX
+	if (kbuf == sg_io_buffer_hack)
+		do_copy = 0;
+	else
+#endif
+		do_copy = !blk_rq_aligned(q, addr, len)
+			|| object_is_on_stack(kbuf);
+
 	if (do_copy)
 		bio = bio_copy_kern(q, kbuf, len, gfp_mask, reading);
 	else
diff --git a/block/scsi_ioctl.c b/block/scsi_ioctl.c
index 7440de44dd85..6ed73c02a04d 100644
--- a/block/scsi_ioctl.c
+++ b/block/scsi_ioctl.c
@@ -253,6 +253,12 @@ static int blk_fill_sghdr_rq(struct request_queue *q, struct request *rq,
 	return 0;
 }
 
+#ifdef CONFIG_AHCI_IMX
+extern void *sg_io_buffer_hack;
+#else
+#define sg_io_buffer_hack NULL
+#endif
+
 static int blk_complete_sghdr_rq(struct request *rq, struct sg_io_hdr *hdr,
 				 struct bio *bio)
 {
@@ -282,7 +288,12 @@ static int blk_complete_sghdr_rq(struct request *rq, struct sg_io_hdr *hdr,
 			ret = -EFAULT;
 	}
 
-	r = blk_rq_unmap_user(bio);
+	if (sg_io_buffer_hack && !hdr->iovec_count)
+		r = copy_to_user(hdr->dxferp, sg_io_buffer_hack,
+				hdr->dxfer_len);
+	else
+		r = blk_rq_unmap_user(bio);
+
 	if (!ret)
 		ret = r;
 
@@ -306,6 +317,9 @@ static int sg_io(struct request_queue *q, struct gendisk *bd_disk,
 	if (hdr->dxfer_len > (queue_max_hw_sectors(q) << 9))
 		return -EIO;
 
+	if (sg_io_buffer_hack && hdr->dxfer_len > 0x10000)
+		return -EIO;
+
 	if (hdr->dxfer_len)
 		switch (hdr->dxfer_direction) {
 		default:
@@ -353,9 +367,14 @@ static int sg_io(struct request_queue *q, struct gendisk *bd_disk,
 
 		ret = blk_rq_map_user_iov(q, rq, NULL, &i, GFP_KERNEL);
 		kfree(iov);
-	} else if (hdr->dxfer_len)
-		ret = blk_rq_map_user(q, rq, NULL, hdr->dxferp, hdr->dxfer_len,
-				      GFP_KERNEL);
+	} else if (hdr->dxfer_len) {
+		if (sg_io_buffer_hack)
+			ret = blk_rq_map_kern(q, rq, sg_io_buffer_hack,
+					hdr->dxfer_len, GFP_KERNEL);
+		else
+			ret = blk_rq_map_user(q, rq, NULL, hdr->dxferp,
+					hdr->dxfer_len, GFP_KERNEL);
+	}
 
 	if (ret)
 		goto out_free_cdb;
diff --git a/drivers/ata/ahci_imx.c b/drivers/ata/ahci_imx.c
index 787567e840bd..8da450ba2eaa 100644
--- a/drivers/ata/ahci_imx.c
+++ b/drivers/ata/ahci_imx.c
@@ -72,6 +72,8 @@ struct imx_ahci_priv {
 	u32 phy_params;
 };
 
+void *sg_io_buffer_hack;
+
 static int ahci_imx_hotplug;
 module_param_named(hotplug, ahci_imx_hotplug, int, 0644);
 MODULE_PARM_DESC(hotplug, "AHCI IMX hot-plug support (0=Don't support, 1=support)");
@@ -818,6 +820,24 @@ static int imx_ahci_probe(struct platform_device *pdev)
 	reg_val = clk_get_rate(imxpriv->ahb_clk) / 1000;
 	writel(reg_val, hpriv->mmio + IMX_TIMER1MS);
 
+	/*
+	* Due to IP bug on the Synopsis 3.00 SATA version,
+	* which is present on mx6q, and not on mx53,
+	* we should use sg_tablesize = 1 for reliable operation
+	*/
+	if (imxpriv->type == AHCI_IMX6Q) {
+		dma_addr_t dma;
+
+		ahci_platform_sht.sg_tablesize = 1;
+
+		sg_io_buffer_hack = dma_alloc_coherent(NULL, 0x10000,
+				&dma, GFP_KERNEL);
+		if (!sg_io_buffer_hack) {
+			ret = -ENOMEM;
+			goto disable_sata;
+		}
+	}
+
 	ret = ahci_platform_init_host(pdev, hpriv, &ahci_imx_port_info,
 				      &ahci_platform_sht);
 	if (ret)
diff --git a/drivers/scsi/scsi_error.c b/drivers/scsi/scsi_error.c
index cf70f0bb8375..99ef1a58f478 100644
--- a/drivers/scsi/scsi_error.c
+++ b/drivers/scsi/scsi_error.c
@@ -61,6 +61,12 @@ static int scsi_eh_try_stu(struct scsi_cmnd *scmd);
 static int scsi_try_to_abort_cmd(struct scsi_host_template *,
 				 struct scsi_cmnd *);
 
+#ifdef CONFIG_AHCI_IMX
+extern void *sg_io_buffer_hack;
+#else
+#define sg_io_buffer_hack NULL
+#endif
+
 /* called with shost->host_lock held */
 void scsi_eh_wakeup(struct Scsi_Host *shost)
 {
@@ -69,6 +75,11 @@ void scsi_eh_wakeup(struct Scsi_Host *shost)
 		wake_up_process(shost->ehandler);
 		SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost,
 			"Waking error handler thread\n"));
+	} else if ((shost->host_failed > 0) || (sg_io_buffer_hack != NULL)) {
+		trace_scsi_eh_wakeup(shost);
+		wake_up_process(shost->ehandler);
+		SCSI_LOG_ERROR_RECOVERY(5, shost_printk(KERN_INFO, shost,
+			"Waking error handler thread\n"));
 	}
 }
 
@@ -2141,8 +2152,15 @@ int scsi_error_handler(void *data)
 		if (kthread_should_stop())
 			break;
 
+		/*
+		 * Do not go to sleep, when there is host_failed when the
+		 * one-PRD per command workaroud is tiggered.
+		 * Because that ata/scsi subsystem maybe hang, when CD_ROM
+		 * and HDD are accessed simultaneously.
+		 */
 		if ((shost->host_failed == 0 && shost->host_eh_scheduled == 0) ||
-		    shost->host_failed != atomic_read(&shost->host_busy)) {
+		    ((shost->host_failed != atomic_read(&shost->host_busy)) &&
+		    (sg_io_buffer_hack == NULL) && (shost->host_failed > 0))) {
 			SCSI_LOG_ERROR_RECOVERY(1,
 				shost_printk(KERN_INFO, shost,
 					     "scsi_eh_%d: sleeping\n",