On Thu, Aug 08, 2024 at 12:35:01AM +0200, Niklas Cassel wrote:
On Fri, Aug 02, 2024 at 02:30:45AM +0000, Hongxing Zhu wrote:
Does this solve your problem: diff --git a/drivers/ata/libahci_platform.c b/drivers/ata/libahci_platform.c index 581704e61f28..fc86e2c8c42b 100644 --- a/drivers/ata/libahci_platform.c +++ b/drivers/ata/libahci_platform.c @@ -747,12 +747,11 @@ int ahci_platform_init_host(struct platform_device *pdev, ap->ops = &ata_dummy_port_ops; }
if (hpriv->cap & HOST_CAP_64) {rc = dma_coerce_mask_and_coherent(dev,DMA_BIT_MASK(64));
if (rc) {dev_err(dev, "Failed to enable 64-bit DMA.\n");return rc;}
rc = dma_coerce_mask_and_coherent(dev,DMA_BIT_MASK((hpriv->cap & HOST_CAP_64) ? 64 :32));
if (rc) {dev_err(dev, "DMA enable failed\n");return rc; } rc = ahci_reset_controller(host);Hi Niklas: I'm so sorry to reply late. About the 32bit DMA limitation of i.MX8QM AHCI SATA. It's seems that one "dma-ranges" property in the DT can let i.MX8QM SATA works fine in my past days tests without this commit. How about drop these driver changes, and add "dma-ranges" for i.MX8QM SATA? Thanks a lot for your kindly help.
Hello Richard,
did you try my suggested patch above?
If you look at dma-ranges: https://devicetree-specification.readthedocs.io/en/latest/chapter2-devicetre...
"dma-ranges" property should be used on a bus device node (such as PCI host bridges).
Yes, 32bit is limited by internal bus farbic, not AHCI controller.
It look like hsio-bus { dma-ranges = <low 4G space> sata@addr { ... } }
This should be correct reflect hardware behavior. force ahci to 32bit dma mask should be a wordaround.
Frank
It does not seem correct to add this property (describing the DMA limit of the AHCI controller, a PCI endpoint) on the PCI host bridge/controller.
This property belongs to the AHCI controller, not the upstream PCI host bridge/controller.
AHCI has a specific register to describe if the hardware can support 64-bit DMA addresses or not, so if my suggested patch works for you, it seems like a more elegant solution (which also avoids having to abuse device tree properties).
Kind regards, Niklas