[RFC PATCH 4/6] iommu/arm-smmu: Add support for IOMMU_DOMAIN_DMA in SMMUv1/SMMUv2 driver

On Thu, Jan 28, 2016 at 05:28:30PM +0000, Robin Murphy wrote:

On 27/01/16 05:21, Anup Patel wrote:

quoted

To allow use of large memory (> 4Gb) with 32bit devices we need to use
some kind of iommu for such 32bit devices.

This patch extends SMMUv1/SMMUv2 driver to support DMA domains which
in-turn will allows us to use iommu based DMA mappings for 32bit devices.

Signed-off-by: Anup Patel <redacted>
Reviewed-by: Ray Jui <rjui@broadcom.com>
Reviewed-by: Scott Branden <sbranden@broadcom.com>
---
 drivers/iommu/arm-smmu.c | 21 ++++++++++++++++++++-
 1 file changed, 20 insertions(+), 1 deletion(-)

diff --git a/drivers/iommu/arm-smmu.c b/drivers/iommu/arm-smmu.c
index 9bdf6b2..43424fe 100644
--- a/drivers/iommu/arm-smmu.c
+++ b/drivers/iommu/arm-smmu.c

@@ -29,6 +29,7 @@
 #define pr_fmt(fmt) "arm-smmu: " fmt

 #include <linux/delay.h>
+#include <linux/dma-iommu.h>
 #include <linux/dma-mapping.h>
 #include <linux/err.h>
 #include <linux/interrupt.h>

@@ -967,7 +968,7 @@ static struct iommu_domain *arm_smmu_domain_alloc(unsigned type)
 {
 	struct arm_smmu_domain *smmu_domain;

-	if (type != IOMMU_DOMAIN_UNMANAGED)
+	if (type != IOMMU_DOMAIN_UNMANAGED && type != IOMMU_DOMAIN_DMA)
 		return NULL;
 	/*
 	 * Allocate the domain and initialise some of its data structures.

@@ -978,6 +979,12 @@ static struct iommu_domain *arm_smmu_domain_alloc(unsigned type)
 	if (!smmu_domain)
 		return NULL;

+	if (type == IOMMU_DOMAIN_DMA &&
+	    iommu_get_dma_cookie(&smmu_domain->domain)) {
+		kfree(smmu_domain);
+		return NULL;
+	}
+
 	mutex_init(&smmu_domain->init_mutex);
 	spin_lock_init(&smmu_domain->pgtbl_lock);

@@ -992,6 +999,7 @@ static void arm_smmu_domain_free(struct iommu_domain *domain)
 	 * Free the domain resources. We assume that all devices have
 	 * already been detached.
 	 */
+	iommu_put_dma_cookie(domain);
 	arm_smmu_destroy_domain_context(domain);
 	kfree(smmu_domain);
 }

@@ -1361,6 +1369,16 @@ static int arm_smmu_init_platform_device(struct device *dev,
 	return 0;
 }

+int arm_smmu_of_xlate(struct device *dev, struct of_phandle_args *args)
+{
+	/*
+	 * Nothing to do here because SMMU is already aware of all
+	 * MMU masters and their stream IDs using mmu-master attibute
+	 * SMMU DT node.
+	 */

...but on the same hand this will also never get called if there's
no "iommus" property on the master. Maintaining support for existing
users of the "mmu-masters" binding is one thing (namely the thing
that's been slowing down my efforts to clean up the really hacky
generic binding support I did all the DMA stuff with), but having
_both_ bindings in a single DT is something I don't think anybody
wants to see

Indeed. NAK to the mixed case.

Mark.