iommu/io-pgtable-arm: Convert to IOMMU API TLB sync

PATCH Superseded Robin Murphy 3 participants 5 messages Sep 18, 2017 – Sep 27, 2017

Subsystems (5)

Arm Qualcomm mailing list Arm MediaTek IOMMU Arm SMMU drivers +1 more subsystems

lore

Share & tools

lore

Robin Murphy

author reviewer

2017-09-18 15:21 UTC #0 lore raw

PATCH 1/2 iommu/io-pgtable-arm: Convert to IOMMU API TLB sync

Now that the core API issues its own post-unmap TLB sync call, push that operation out from the io-pgtable-arm internals into the users. For now, we leave the invalidation implicit in the unmap operation, since none of the current users would benefit much from any change to that.

CC Magnus Damm <[email protected]>

CC Laurent Pinchart <[email protected]>

Signed-off-by Robin Murphy <[email protected]>

---
 drivers/iommu/arm-smmu-v3.c    |  6 ++++++
 drivers/iommu/arm-smmu.c       | 18 +++++++++++++-----
 drivers/iommu/io-pgtable-arm.c |  7 +------
 drivers/iommu/ipmmu-vmsa.c     |  6 ++++++
 4 files changed, 26 insertions(+), 11 deletions(-)

drivers/iommu/arm-smmu-v3.c +6 -0

index e67ba6c40faf..6a866e029173 100644
@@ -1743,6 +1743,11 @@ arm_smmu_unmap(struct iommu_domain *domain, unsigned long iova, size_t size)
 	return ops->unmap(ops, iova, size);
 }
 
+static void arm_smmu_iotlb_sync(struct iommu_domain *domain)
+{
+	__arm_smmu_tlb_sync(to_smmu_domain(domain)->smmu);
+}
+
 static phys_addr_t
 arm_smmu_iova_to_phys(struct iommu_domain *domain, dma_addr_t iova)
 {

@@ -1963,6 +1968,7 @@ static struct iommu_ops arm_smmu_ops = {
 	.map			= arm_smmu_map,
 	.unmap			= arm_smmu_unmap,
 	.map_sg			= default_iommu_map_sg,
+	.iotlb_sync		= arm_smmu_iotlb_sync,
 	.iova_to_phys		= arm_smmu_iova_to_phys,
 	.add_device		= arm_smmu_add_device,
 	.remove_device		= arm_smmu_remove_device,

drivers/iommu/arm-smmu.c +13 -5

index 3bdb799d3b4b..c4b7efe08ce4 100644
@@ -250,6 +250,7 @@ enum arm_smmu_domain_stage {
 struct arm_smmu_domain {
 	struct arm_smmu_device		*smmu;
 	struct io_pgtable_ops		*pgtbl_ops;
+	const struct iommu_gather_ops	*tlb_ops;
 	struct arm_smmu_cfg		cfg;
 	enum arm_smmu_domain_stage	stage;
 	struct mutex			init_mutex; /* Protects smmu pointer */

@@ -735,7 +736,6 @@ static int arm_smmu_init_domain_context(struct iommu_domain *domain,
 	enum io_pgtable_fmt fmt;
 	struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
 	struct arm_smmu_cfg *cfg = &smmu_domain->cfg;
-	const struct iommu_gather_ops *tlb_ops;
 
 	mutex_lock(&smmu_domain->init_mutex);
 	if (smmu_domain->smmu)

@@ -813,7 +813,7 @@ static int arm_smmu_init_domain_context(struct iommu_domain *domain,
 			ias = min(ias, 32UL);
 			oas = min(oas, 32UL);
 		}
-		tlb_ops = &arm_smmu_s1_tlb_ops;
+		smmu_domain->tlb_ops = &arm_smmu_s1_tlb_ops;
 		break;
 	case ARM_SMMU_DOMAIN_NESTED:
 		/*

@@ -833,9 +833,9 @@ static int arm_smmu_init_domain_context(struct iommu_domain *domain,
 			oas = min(oas, 40UL);
 		}
 		if (smmu->version == ARM_SMMU_V2)
-			tlb_ops = &arm_smmu_s2_tlb_ops_v2;
+			smmu_domain->tlb_ops = &arm_smmu_s2_tlb_ops_v2;
 		else
-			tlb_ops = &arm_smmu_s2_tlb_ops_v1;
+			smmu_domain->tlb_ops = &arm_smmu_s2_tlb_ops_v1;
 		break;
 	default:
 		ret = -EINVAL;

@@ -863,7 +863,7 @@ static int arm_smmu_init_domain_context(struct iommu_domain *domain,
 		.pgsize_bitmap	= smmu->pgsize_bitmap,
 		.ias		= ias,
 		.oas		= oas,
-		.tlb		= tlb_ops,
+		.tlb		= smmu_domain->tlb_ops,
 		.iommu_dev	= smmu->dev,
 	};

@@ -1259,6 +1259,13 @@ static size_t arm_smmu_unmap(struct iommu_domain *domain, unsigned long iova,
 	return ops->unmap(ops, iova, size);
 }
 
+static void arm_smmu_iotlb_sync(struct iommu_domain *domain)
+{
+	struct arm_smmu_domain *smmu_domain = to_smmu_domain(domain);
+
+	smmu_domain->tlb_ops->tlb_sync(smmu_domain);
+}
+
 static phys_addr_t arm_smmu_iova_to_phys_hard(struct iommu_domain *domain,
 					      dma_addr_t iova)
 {

@@ -1562,6 +1569,7 @@ static struct iommu_ops arm_smmu_ops = {
 	.map			= arm_smmu_map,
 	.unmap			= arm_smmu_unmap,
 	.map_sg			= default_iommu_map_sg,
+	.iotlb_sync		= arm_smmu_iotlb_sync,
 	.iova_to_phys		= arm_smmu_iova_to_phys,
 	.add_device		= arm_smmu_add_device,
 	.remove_device		= arm_smmu_remove_device,

drivers/iommu/io-pgtable-arm.c +1 -6

index e8018a308868..51e5c43caed1 100644
@@ -609,7 +609,6 @@ static int __arm_lpae_unmap(struct arm_lpae_io_pgtable *data,
 static int arm_lpae_unmap(struct io_pgtable_ops *ops, unsigned long iova,
 			  size_t size)
 {
-	size_t unmapped;
 	struct arm_lpae_io_pgtable *data = io_pgtable_ops_to_data(ops);
 	arm_lpae_iopte *ptep = data->pgd;
 	int lvl = ARM_LPAE_START_LVL(data);

@@ -617,11 +616,7 @@ static int arm_lpae_unmap(struct io_pgtable_ops *ops, unsigned long iova,
 	if (WARN_ON(iova >= (1ULL << data->iop.cfg.ias)))
 		return 0;
 
-	unmapped = __arm_lpae_unmap(data, iova, size, lvl, ptep);
-	if (unmapped)
-		io_pgtable_tlb_sync(&data->iop);
-
-	return unmapped;
+	return __arm_lpae_unmap(data, iova, size, lvl, ptep);
 }
 
 static phys_addr_t arm_lpae_iova_to_phys(struct io_pgtable_ops *ops,

drivers/iommu/ipmmu-vmsa.c +6 -0

index 195d6e93ac71..2c045b8d25f8 100644
@@ -619,6 +619,11 @@ static size_t ipmmu_unmap(struct iommu_domain *io_domain, unsigned long iova,
 	return domain->iop->unmap(domain->iop, iova, size);
 }
 
+static void ipmmu_iotlb_sync(struct iommu_domain *io_domain)
+{
+	ipmmu_tlb_flush_all(to_vmsa_domain(io_domain));
+}
+
 static phys_addr_t ipmmu_iova_to_phys(struct iommu_domain *io_domain,
 				      dma_addr_t iova)
 {

@@ -876,6 +881,7 @@ static const struct iommu_ops ipmmu_ops = {
 	.detach_dev = ipmmu_detach_device,
 	.map = ipmmu_map,
 	.unmap = ipmmu_unmap,
+	.iotlb_sync = ipmmu_iotlb_sync,
 	.map_sg = default_iommu_map_sg,
 	.iova_to_phys = ipmmu_iova_to_phys,
 	.add_device = ipmmu_add_device_dma,

Signature · 1 line

2.13.4.dirty

Robin Murphy

author reviewer

2017-09-18 15:21 UTC #1 lore raw

PATCH 2/2 iommu/io-pgtable-arm-v7s: Convert to IOMMU API TLB sync

Now that the core API issues its own post-unmap TLB sync call, push that operation out from the io-pgtable-arm-v7s internals into the users. For now, we leave the invalidation implicit in the unmap operation, since none of the current users would benefit much from any change to that.

Note that the conversion of msm_iommu is implicit, since that apparently has no specific TLB sync operation anyway.

CC Yong Wu <[email protected]>

CC Rob Clark <[email protected]>

Signed-off-by Robin Murphy <[email protected]>

---
 drivers/iommu/io-pgtable-arm-v7s.c |  7 +------
 drivers/iommu/mtk_iommu.c          |  6 ++++++
 drivers/iommu/qcom_iommu.c         | 12 ++++++++++++
 3 files changed, 19 insertions(+), 6 deletions(-)

drivers/iommu/io-pgtable-arm-v7s.c +1 -6

index d665d0dc16e8..397531da8d9c 100644
@@ -660,16 +660,11 @@ static int arm_v7s_unmap(struct io_pgtable_ops *ops, unsigned long iova,
 			 size_t size)
 {
 	struct arm_v7s_io_pgtable *data = io_pgtable_ops_to_data(ops);
-	size_t unmapped;
 
 	if (WARN_ON(upper_32_bits(iova)))
 		return 0;
 
-	unmapped = __arm_v7s_unmap(data, iova, size, 1, data->pgd);
-	if (unmapped)
-		io_pgtable_tlb_sync(&data->iop);
-
-	return unmapped;
+	return __arm_v7s_unmap(data, iova, size, 1, data->pgd);
 }
 
 static phys_addr_t arm_v7s_iova_to_phys(struct io_pgtable_ops *ops,

drivers/iommu/mtk_iommu.c +6 -0

index bd515be5b380..36499e10caa2 100644
@@ -391,6 +391,11 @@ static size_t mtk_iommu_unmap(struct iommu_domain *domain,
 	return unmapsz;
 }
 
+static void mtk_iommu_iotlb_sync(struct iommu_domain *domain)
+{
+	mtk_iommu_tlb_sync(mtk_iommu_get_m4u_data());
+}
+
 static phys_addr_t mtk_iommu_iova_to_phys(struct iommu_domain *domain,
 					  dma_addr_t iova)
 {

@@ -490,6 +495,7 @@ static struct iommu_ops mtk_iommu_ops = {
 	.map		= mtk_iommu_map,
 	.unmap		= mtk_iommu_unmap,
 	.map_sg		= default_iommu_map_sg,
+	.iotlb_sync	= mtk_iommu_iotlb_sync,
 	.iova_to_phys	= mtk_iommu_iova_to_phys,
 	.add_device	= mtk_iommu_add_device,
 	.remove_device	= mtk_iommu_remove_device,

drivers/iommu/qcom_iommu.c +12 -0

index c8a587d034b0..49fb82a9a9c6 100644
@@ -443,6 +443,17 @@ static size_t qcom_iommu_unmap(struct iommu_domain *domain, unsigned long iova,
 	return ret;
 }
 
+static void qcom_iommu_iotlb_sync(struct iommu_domain *domain)
+{
+	struct qcom_iommu_domain *qcom_domain = to_qcom_iommu_domain(domain);
+	struct io_pgtable *pgtable = container_of(qcom_domain->pgtbl_ops,
+						  struct io_pgtable, ops);
+
+	pm_runtime_get_sync(qcom_domain->iommu->dev);
+	qcom_iommu_tlb_sync(pgtable->cookie);
+	pm_runtime_put_sync(qcom_domain->iommu->dev);
+}
+
 static phys_addr_t qcom_iommu_iova_to_phys(struct iommu_domain *domain,
 					   dma_addr_t iova)
 {

@@ -570,6 +581,7 @@ static const struct iommu_ops qcom_iommu_ops = {
 	.map		= qcom_iommu_map,
 	.unmap		= qcom_iommu_unmap,
 	.map_sg		= default_iommu_map_sg,
+	.iotlb_sync	= qcom_iommu_iotlb_sync,
 	.iova_to_phys	= qcom_iommu_iova_to_phys,
 	.add_device	= qcom_iommu_add_device,
 	.remove_device	= qcom_iommu_remove_device,

Signature · 1 line

2.13.4.dirty

Will Deacon

2017-09-22 15:43 UTC #2 lore raw

PATCH 1/2 iommu/io-pgtable-arm: Convert to IOMMU API TLB sync

In reply to Robin Murphy

Quoted 44 lines · Robin Murphy

CC Magnus Damm <[email protected]>

CC Laurent Pinchart <[email protected]>

Signed-off-by Robin Murphy <[email protected]>

---
 drivers/iommu/arm-smmu-v3.c    |  6 ++++++
 drivers/iommu/arm-smmu.c       | 18 +++++++++++++-----
 drivers/iommu/io-pgtable-arm.c |  7 +------
 drivers/iommu/ipmmu-vmsa.c     |  6 ++++++
 4 files changed, 26 insertions(+), 11 deletions(-)

drivers/iommu/arm-smmu-v3.c +6 -0

index e67ba6c40faf..6a866e029173 100644
@@ -1743,6 +1743,11 @@ arm_smmu_unmap(struct iommu_domain *domain, unsigned long iova, size_t size)
 	return ops->unmap(ops, iova, size);
 }
 
+static void arm_smmu_iotlb_sync(struct iommu_domain *domain)
+{
+	__arm_smmu_tlb_sync(to_smmu_domain(domain)->smmu);
+}
+
 static phys_addr_t
 arm_smmu_iova_to_phys(struct iommu_domain *domain, dma_addr_t iova)
 {

@@ -1963,6 +1968,7 @@ static struct iommu_ops arm_smmu_ops = {
 	.map			= arm_smmu_map,
 	.unmap			= arm_smmu_unmap,
 	.map_sg			= default_iommu_map_sg,
+	.iotlb_sync		= arm_smmu_iotlb_sync,
 	.iova_to_phys		= arm_smmu_iova_to_phys,
 	.add_device		= arm_smmu_add_device,
 	.remove_device		= arm_smmu_remove_device,

drivers/iommu/arm-smmu.c +0 -0

index 3bdb799d3b4b..c4b7efe08ce4 100644

drivers/iommu/arm-smmu.c enum arm_smmu_domain_stage { · line 253

Quoted 4 lines

drivers/iommu/arm-smmu.c +1 -0

@@ -250,6 +250,7 @@ enum arm_smmu_domain_stage {
 struct arm_smmu_domain {
 	struct arm_smmu_device		*smmu;
 	struct io_pgtable_ops		*pgtbl_ops;
+	const struct iommu_gather_ops	*tlb_ops;

Hmmm.... I *suppose* that means there's a way to the TLB invalidation functions from the iommu_device field of the io_pgtable_cfg via something like iommu_ops_from_fwnode, but it feels horribly ugly doing that so I'm happy to keep the extra gather_ops pointer in the io_pgtable_cfg for now.

This patch and the subsequent one look fine to me, but I'd like to go a step further and hook-up the add_flush callback too so that it can be (mostly) removed from the io_pgtable code and make the *_fast operations even cheaper for VFIO when it starts using them.

So,

Acked-by Will Deacon <[email protected]>

Joerg, do you reckon it's worth merging this as-is, or should we also hook up add_flush before implementing this?

Will

Joerg Roedel

maintainer

2017-09-27 12:27 UTC #3 lore raw

PATCH 1/2 iommu/io-pgtable-arm: Convert to IOMMU API TLB sync

Hi Will, Robin,

Quoted 2 lines · Will Deacon

Joerg, do you reckon it's worth merging this as-is, or should we also hook up add_flush before implementing this?

The patches implement .iotlb_sync() so that it is okay to not have a .iotlb_range_add() call-back for now. But the patches lack a .flush_iotlb_range() call-back, which is strictly necessary for correctness because it is defined to be synchronous and doesn't require a sync-call afterwards.

Regards,

Joerg

Robin Murphy

author reviewer

2017-09-27 12:34 UTC #4 lore raw

PATCH 1/2 iommu/io-pgtable-arm: Convert to IOMMU API TLB sync

Quoted 11 lines · Joerg Roedel

Hi Will, Robin,
Quoted 2 lines · Will Deacon
Joerg, do you reckon it's worth merging this as-is, or should we also hook up add_flush before implementing this?
The patches implement .iotlb_sync() so that it is okay to not have a .iotlb_range_add() call-back for now. But the patches lack a .flush_iotlb_range() call-back, which is strictly necessary for correctness because it is defined to be synchronous and doesn't require a sync-call afterwards.

Good point - even with the flush still implicit in the unmap, an iommu_unmap_fast()/iommu_flush_iotlb_all() sequence would indeed be missing a sync at the end. I'll fix that up and post a new version shortly.

Thanks,
Robin.