See below. Also, this patch only fixes 1 of the 2 places, we need to fix lpfc_sli4_pci_mem_setup() also. With the same code applied in both places, the NVMe/FC issue is fixed. I expect the SCSI/FCP tests would pass too, but those take a couple of days.
-Ewan
On Wed, 2019-02-13 at 12:42 +0100, Hannes Reinecke wrote:
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@vger.kernel.org Suggested-by: Don Dutile ddutile@redhat.com Signed-off-by: Ewan D. Milne emilne@redhat.com Signed-off-by: Hannes Reinecke hare@suse.com
drivers/scsi/lpfc/lpfc_init.c | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-)
diff --git a/drivers/scsi/lpfc/lpfc_init.c b/drivers/scsi/lpfc/lpfc_init.c index bede11e16349..8a4b096ffe47 100644 --- a/drivers/scsi/lpfc/lpfc_init.c +++ b/drivers/scsi/lpfc/lpfc_init.c @@ -7361,15 +7361,18 @@ lpfc_sli_pci_mem_setup(struct lpfc_hba *phba) unsigned long bar0map_len, bar2map_len; int i, hbq_count; void *ptr;
- int error = -ENODEV;
- int error;
if (!pdev) return error;
Since this patch removes the initialization of "error" above, this return statement returns an undefined value. Change it to return -ENODEV?
/* 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)))
- 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;
- error = -ENODEV;
/* Get the bus address of Bar0 and Bar2 and the number of bytes * required by each mapping.