On 2018-07-16 4:10 PM, Lorenzo Pieralisi wrote:

On Tue, Jul 10, 2018 at 05:13:45PM +0100, Robin Murphy wrote:

quoted

IORT revision D allows PCI root complex nodes to specify a memory
address size limit equivalently to named components, to help describe
straightforward integrations which don't really warrant a full-blown
_DMA method. Now that our headers are up-to-date, plumb it in.

Signed-off-by: Robin Murphy <robin.murphy@arm.com>
---
  drivers/acpi/arm64/iort.c | 25 +++++++++++++++++++++++--
  1 file changed, 23 insertions(+), 2 deletions(-)

diff --git a/drivers/acpi/arm64/iort.c b/drivers/acpi/arm64/iort.c
index 7a3a541046ed..4a66896e2aa3 100644
--- a/drivers/acpi/arm64/iort.c
+++ b/drivers/acpi/arm64/iort.c

@@ -947,6 +947,24 @@ static int nc_dma_get_range(struct device *dev, u64 *size)
  	return 0;
  }
  
+static int rc_dma_get_range(struct device *dev, u64 *size)
+{
+	struct acpi_iort_node *node;
+	struct acpi_iort_root_complex *rc;
+
+	node = iort_scan_node(ACPI_IORT_NODE_PCI_ROOT_COMPLEX,
+			      iort_match_node_callback, dev);
+	if (!node || node->revision < 1)
+		return -ENODEV;
+
+	rc = (struct acpi_iort_root_complex *)node->node_data;
+
+	*size = rc->memory_address_limit >= 64 ? U64_MAX :
+			1ULL<<rc->memory_address_limit;
+
+	return 0;
+}
+
  /**
   * iort_dma_setup() - Set-up device DMA parameters.
   *

@@ -975,10 +993,13 @@ void iort_dma_setup(struct device *dev, u64 *dma_addr, u64 *dma_size)
  
  	size = max(dev->coherent_dma_mask, dev->coherent_dma_mask + 1);
  
-	if (dev_is_pci(dev))
+	if (dev_is_pci(dev)) {
  		ret = acpi_dma_get_range(dev, &dmaaddr, &offset, &size);
-	else
+		if (ret == -ENODEV)
+			ret = rc_dma_get_range(dev, &size);

Thank you for putting together the patch.

The question is whether it is OK to ignore the IORT address limits
when _DMA is actually specified. It is a sort of grey area that
has to be clarified, maybe we can add a check to detect a size
mismatch, I do not know if something should be added at IORT spec
level to clarify its relation to the _DMA object, if present.

Yeah, I'm assuming that _DMA would be used to describe conditions more 
specific than the simple address size limit (i.e. bridge windows), so 
even if both are present, the range inferred from _DMA will always be 
less than or equal to that inferred from IORT, and thus rather than 
explicitly calculating the intersection of the two we can simply do this 
short-circuit.

If IORT accurately reflects the total number of usable address bits, 
then I can't see that it would ever make sense for _DMA to specify an 
address range which exceeds that; I guess it comes down to how much 
effort we want to spend verifying firmware instead of trusting it.

Robin.