Skip to contentSkip to conversation

scsi: lpfc: fix calls to dma_set_mask_and_coherent()

PATCH New Ewan D. Milne 5 participants 8 messages Feb 11, 2019 – Feb 13, 2019

Subsystems (4)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 scsi: lpfc: fix calls to dma_set_mask_and_coherent()

The change to use dma_set_mask_and_coherent() incorrectly made a second call with the 32 bit DMA mask value when the call with the 64 bit DMA mask value succeeded. This resulted in NVMe/FC connections failing due to corrupted data buffers, and various other SCSI/FCP I/O errors.

Fixes f30e1bfd6154 ("scsi: lpfc: use dma_set_mask_and_coherent")
Cc <stable at vger.kernel.org>
Suggested-by Don Dutile <ddutile at redhat.com>
Signed-off-by Ewan D. Milne <emilne at redhat.com>
---
 drivers/scsi/lpfc/lpfc_init.c | 12 ++++++------
 1 file changed, 6 insertions(+), 6 deletions(-)
drivers/scsi/lpfc/lpfc_init.c +6 -6
index bede11e..60c178d 100644
@@ -7367,9 +7367,9 @@ struct lpfc_rpi_hdr *
 		return error;
 
 	/* Set the device DMA mask size */
-	if (dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)) ||
-	    dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)))
-		return error;
+	if (dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)) != 0)
+		if (dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)) != 0)
+			return error;
 
 	/* Get the bus address of Bar0 and Bar2 and the number of bytes
 	 * required by each mapping.
@@ -9745,9 +9745,9 @@ struct lpfc_cq_event *
 		return error;
 
 	/* Set the device DMA mask size */
-	if (dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)) ||
-	    dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)))
-		return error;
+	if (dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)) != 0)
+		if (dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)) != 0)
+			return error;
 
 	/*
 	 * The BARs and register set definitions and offset locations are
Signature · 1 line
1.8.3.1
#1 lore raw

PATCH 1/1 scsi: lpfc: fix calls to dma_set_mask_and_coherent()

Quoted 33 lines · Ewan D. Milne

The change to use dma_set_mask_and_coherent() incorrectly made a second call with the 32 bit DMA mask value when the call with the 64 bit DMA mask value succeeded. This resulted in NVMe/FC connections failing due to corrupted data buffers, and various other SCSI/FCP I/O errors.

Fixes f30e1bfd6154 ("scsi: lpfc: use dma_set_mask_and_coherent")
Cc <stable at vger.kernel.org>
Suggested-by Don Dutile <ddutile at redhat.com>
Signed-off-by Ewan D. Milne <emilne at redhat.com>
---
  drivers/scsi/lpfc/lpfc_init.c | 12 ++++++------
  1 file changed, 6 insertions(+), 6 deletions(-)
drivers/scsi/lpfc/lpfc_init.c +3 -3
index bede11e..60c178d 100644
@@ -7367,9 +7367,9 @@ struct lpfc_rpi_hdr *
  		return error;
 
  	/* Set the device DMA mask size */
-	if (dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)) ||
-	    dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)))
-		return error;
+	if (dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)) != 0)
+		if (dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)) != 0)
+			return error;
 
  	/* Get the bus address of Bar0 and Bar2 and the number of bytes
  	 * required by each mapping.
@@ -9745,9 +9745,9 @@ struct lpfc_cq_event *
  		return error;
 

drivers/scsi/lpfc/lpfc_init.c struct lpfc_cq_event * · line 9753

Quoted 10 lines
drivers/scsi/lpfc/lpfc_init.c +3 -3
  	/* Set the device DMA mask size */
-	if (dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)) ||
-	    dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)))
-		return error;
+	if (dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)) != 0)
+		if (dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)) != 0)
+			return error;
 
  	/*
  	 * The BARs and register set definitions and offset locations are

Thanks Ewan.

Signed-off-by ? James Smart? <james.smart at broadcom.com>

-- james

#2 lore raw

PATCH 1/1 scsi: lpfc: fix calls to dma_set_mask_and_coherent()

In reply to Ewan D. Milne
Quoted 4 lines · Ewan D. Milne

The change to use dma_set_mask_and_coherent() incorrectly made a second call with the 32 bit DMA mask value when the call with the 64 bit DMA mask value succeeded. This resulted in NVMe/FC connections failing due to corrupted data buffers, and various other SCSI/FCP I/O errors.

Ooops, sorry.

drivers/scsi/lpfc/lpfc_init.c struct lpfc_rpi_hdr * · line 7371

Quoted 6 lines
drivers/scsi/lpfc/lpfc_init.c +2 -3
 	/* Set the device DMA mask size */
-	if (dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)) ||
-	    dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)))
-		return error;
+	if (dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)) != 0)
+		if (dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)) != 0)

But this still looks obsfuctating, why not:

	error = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
	if (error)
		error = dma_set_mask_and_coherent(&pdev->dev,
				DMA_BIT_MASK(32)));
	if (error)
		return error;
#3 lore raw

PATCH 1/1 scsi: lpfc: fix calls to dma_set_mask_and_coherent()

Quoted 17 lines · Christoph Hellwig
Quoted 4 lines · Ewan D. Milne

The change to use dma_set_mask_and_coherent() incorrectly made a second call with the 32 bit DMA mask value when the call with the 64 bit DMA mask value succeeded. This resulted in NVMe/FC connections failing due to corrupted data buffers, and various other SCSI/FCP I/O errors.

Ooops, sorry.

drivers/scsi/lpfc/lpfc_init.c struct lpfc_rpi_hdr * · line 7371

Quoted 6 lines
drivers/scsi/lpfc/lpfc_init.c +2 -3
 	/* Set the device DMA mask size */
-	if (dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)) ||
-	    dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)))
-		return error;
+	if (dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)) != 0)
+		if (dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32)) != 0)

But this still looks obsfuctating, why not:

Quoted 6 lines
	error = dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64));
	if (error)
		error = dma_set_mask_and_coherent(&pdev->dev,
				DMA_BIT_MASK(32)));
	if (error)
		return error;

I agree that making the flow of control explicit would be good, but...

It looks like this would introduce a different problem. "error" is set to -ENODEV earlier in the two functions so that the various error paths will return that value. If it is overwritten with a successful result from the call to dma_set_mask_and_coherent() but a later call to e.g. ioremap() fails, the functions will incorrectly return a value indicating that they have succeeded.

It also appears as if the patches to hptiop, hisi_sas and bfa need to be fixed up. I don't have a test environment for these, although I might be able to modify the test environment for bfa. Martin/James, what about the others?

-Ewan

#4 lore raw

PATCH 1/1 scsi: lpfc: fix calls to dma_set_mask_and_coherent()

Quoted 10 lines · Ewan D. Milne

It looks like this would introduce a different problem. "error" is set to -ENODEV earlier in the two functions so that the various error paths will return that value. If it is overwritten with a successful result from the call to dma_set_mask_and_coherent() but a later call to e.g. ioremap() fails, the functions will incorrectly return a value indicating that they have succeeded.

It also appears as if the patches to hptiop, hisi_sas and bfa need to be fixed up. I don't have a test environment for these, although I might be able to modify the test environment for bfa. Martin/James, what about the others?

I don't have a test environment for them either, and I doubt many people have. If you don't feel comfortable I can send a fix for them.

#5 lore raw

PATCH 1/1 scsi: lpfc: fix calls to dma_set_mask_and_coherent()

Quoted 8 lines · Christoph Hellwig
Quoted 7 lines · Ewan D. Milne

It looks like this would introduce a different problem. "error" is set to -ENODEV earlier in the two functions so that the various error paths will return that value. If it is overwritten with a successful result from the call to dma_set_mask_and_coherent() but a later call to e.g. ioremap() fails, the functions will incorrectly return a value indicating that they have succeeded.

Quoted 7 lines
Quoted 3 lines

It also appears as if the patches to hptiop, hisi_sas and bfa need to be fixed up. I don't have a test environment for these, although I might be able to modify the test environment for bfa. Martin/James, what about the others?

I don't have a test environment for them either, and I doubt many people have. If you don't feel comfortable I can send a fix for them.

[ removed cc: stable ]

Thanks, I have the patches written but Hannes has posted already. I prefer to test before posting but in this case testing one driver with a similar pattern is better than not having a fix available, I think. As long as it gets enough review.

-Ewan

#6 lore raw

PATCH 1/1 scsi: lpfc: fix calls to dma_set_mask_and_coherent()

In reply to Ewan D. Milne

Ewan,

Quoted 4 lines

It also appears as if the patches to hptiop, hisi_sas and bfa need to be fixed up. I don't have a test environment for these, although I might be able to modify the test environment for bfa. Martin/James, what about the others?

I'm afraid I don't have any of those. But I'm sure John would be happy to check hisi_sas.

Signature · 1 line
Martin K. Petersen	Oracle Linux Engineering
#7 lore raw

PATCH 1/1 scsi: lpfc: fix calls to dma_set_mask_and_coherent()

Quoted 11 lines · Martin K. Petersen

Ewan,

Quoted 4 lines

It also appears as if the patches to hptiop, hisi_sas and bfa need to be fixed up. I don't have a test environment for these, although I might be able to modify the test environment for bfa. Martin/James, what about the others?

I'm afraid I don't have any of those. But I'm sure John would be happy to check hisi_sas.

Happy to test. Or I can just send the fix myself. So what is the preferred pattern for setting the DMA mask?

John