On 18/02/2019 07:21, Hannes Reinecke wrote:
On 2/15/19 10:16 AM, John Garry wrote:
On 15/02/2019 08:08, Christoph Hellwig wrote:
On Fri, Feb 15, 2019 at 08:43:55AM +0100, Christoph Hellwig wrote:
On Fri, Feb 15, 2019 at 07:55:39AM +0100, Hannes Reinecke wrote:
Yeah, there is a few more. And the sad part is as of a few kernel
release ago we shouldn't even need the fallback 32-bit dma mask
anymore - we've cleaned up all the mess that required it.
Care to elaborate?
Can you point me to the respective commits facilitating that?
It mostly has been that way for a long time, but we still had a few
oddball architectures checking for an exact 32-bit match in their
arch specific direct mapping routines. With the move to the generic
dma-direct implementation all those got fixed.
I thought that many SCSI drivers were changed over to stop using the
PCI DMA APIs at the end of last year, like this:
commit c22b332d811b90448e090c7fb487448afb039fcc
Author: Christoph Hellwig <hch@xxxxxx>
Date: Wed Oct 10 18:34:51 2018 +0200
scsi: csiostor: switch to generic DMA API
Switch from the legacy PCI DMA API to the generic DMA API.
Signed-off-by: Christoph Hellwig <hch@xxxxxx>
Reviewed-by: Johannes Thumshirn <jthumshirn@xxxxxxx>
Signed-off-by: Martin K. Petersen <martin.petersen@xxxxxxxxxx>
diff --git a/drivers/scsi/csiostor/csio_init.c
b/drivers/scsi/csiostor/csio_init.c
index ed2dae657964..aa04e4a7aed5 100644
--- a/drivers/scsi/csiostor/csio_init.c
+++ b/drivers/scsi/csiostor/csio_init.c
@@ -210,11 +210,8 @@ csio_pci_init(struct pci_dev *pdev, int *bars)
pci_set_master(pdev);
pci_try_set_mwi(pdev);
- if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(64))) {
- pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
- } else if (!pci_set_dma_mask(pdev, DMA_BIT_MASK(32))) {
- pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
- } else {
+ if (dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(64)) ||
+ dma_set_mask_and_coherent(&pdev->dev, DMA_BIT_MASK(32))) {
dev_err(&pdev->dev, "No suitable DMA available.\n");
goto err_release_regions;
}
... which is precisely the error we're encountering.
If the first call to dma_set_mask_and_coherent() succeeds it will return
'0', and the next call to dma_set_mask_and_coherent() will be executed.
Sure, I know this. What's more, if the first fails, then we bail out and
won't even try the second.
Thanks,
John
Cheers,
Hannes
