Some improvements for iommu

PATCH Applied Baolin Wang 3 participants 7 messages Jun 12, 2020 – Jul 24, 2020

Subsystems (9)

DRM drivers DRM misc Arm Qualcomm mailing list Arm MediaTek +5 more subsystems

lore

Share & tools

lore

Baolin Wang

author

2020-06-12 03:40 UTC #0 lore raw

v2 cover Some improvements for iommu

Hi,

The first patch masks some functions as static, and the second patch changes to use the gfp parameter from iommu_ops->map() to allocate ARM page pages. Any comments are welcome. Thanks.

Changes from v1:
- Fix the building errors when enabling CONFIG_IOMMU_IO_PGTABLE_LPAE_SELFTEST

Baolin Wang (2):
iommu: Mark __iommu_map/__iommu_map_sg as static
iommu: Add gfp parameter to io_pgtable_ops->map()

 drivers/gpu/drm/panfrost/panfrost_mmu.c |  2 +-
 drivers/iommu/arm-smmu-v3.c             |  2 +-
 drivers/iommu/arm-smmu.c                |  2 +-
 drivers/iommu/io-pgtable-arm-v7s.c      | 18 +++++++++---------
 drivers/iommu/io-pgtable-arm.c          | 18 +++++++++---------
 drivers/iommu/iommu.c                   | 10 +++++-----
 drivers/iommu/ipmmu-vmsa.c              |  2 +-
 drivers/iommu/msm_iommu.c               |  2 +-
 drivers/iommu/mtk_iommu.c               |  2 +-
 drivers/iommu/qcom_iommu.c              |  2 +-
 include/linux/io-pgtable.h              |  2 +-
 11 files changed, 31 insertions(+), 31 deletions(-)

Signature · 1 line

1.8.3.1

Baolin Wang

author

2020-06-12 03:40 UTC #1 lore raw

PATCH v2 1/2 iommu: Mark __iommu_map/__iommu_map_sg as static

Now the __iommu_map() and __iommu_map_sg() are used only in iommu.c file, so mark them as static.

Signed-off-by Baolin Wang <[email protected]>

---
 drivers/iommu/iommu.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

drivers/iommu/iommu.c +5 -5

index 8584f48..14eca9f 100644
@@ -2174,8 +2174,8 @@ static size_t iommu_pgsize(struct iommu_domain *domain,
 	return pgsize;
 }
 
-int __iommu_map(struct iommu_domain *domain, unsigned long iova,
-	      phys_addr_t paddr, size_t size, int prot, gfp_t gfp)
+static int __iommu_map(struct iommu_domain *domain, unsigned long iova,
+		       phys_addr_t paddr, size_t size, int prot, gfp_t gfp)
 {
 	const struct iommu_ops *ops = domain->ops;
 	unsigned long orig_iova = iova;

@@ -2325,9 +2325,9 @@ size_t iommu_unmap_fast(struct iommu_domain *domain,
 }
 EXPORT_SYMBOL_GPL(iommu_unmap_fast);
 
-size_t __iommu_map_sg(struct iommu_domain *domain, unsigned long iova,
-		    struct scatterlist *sg, unsigned int nents, int prot,
-		    gfp_t gfp)
+static size_t __iommu_map_sg(struct iommu_domain *domain, unsigned long iova,
+			     struct scatterlist *sg, unsigned int nents, int prot,
+			     gfp_t gfp)
 {
 	size_t len = 0, mapped = 0;
 	phys_addr_t start;

Signature · 1 line

1.8.3.1

Will Deacon

maintainer Acked-by on v2 1/2

2020-07-14 08:27 UTC #2 lore raw

Re: [PATCH v2 1/2] iommu: Mark __iommu_map/__iommu_map_sg as static

Quoted 7 lines · Baolin Wang

Now the __iommu_map() and __iommu_map_sg() are used only in iommu.c file, so mark them as static.
Signed-off-by Baolin Wang <[email protected]>
---
 drivers/iommu/iommu.c | 10 +++++-----
 1 file changed, 5 insertions(+), 5 deletions(-)

Acked-by Will Deacon <[email protected]>

(I'm assuming Joerg will pick this up for 5.9)

Thanks,

Will
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Baolin Wang

author

2020-06-12 03:40 UTC #3 lore raw

PATCH v2 2/2 iommu: Add gfp parameter to io_pgtable_ops->map()

In reply to Baolin Wang

Now the ARM page tables are always allocated by GFP_ATOMIC parameter, but the iommu_ops->map() function has been added a gfp_t parameter by commit 781ca2de89ba ("iommu: Add gfp parameter to iommu_ops::map"), thus io_pgtable_ops->map() should use the gfp parameter passed from iommu_ops->map() to allocate page pages, which can avoid wasting the memory allocators atomic pools for some non-atomic contexts.

Signed-off-by Baolin Wang <[email protected]>

---
 drivers/gpu/drm/panfrost/panfrost_mmu.c |  2 +-
 drivers/iommu/arm-smmu-v3.c             |  2 +-
 drivers/iommu/arm-smmu.c                |  2 +-
 drivers/iommu/io-pgtable-arm-v7s.c      | 18 +++++++++---------
 drivers/iommu/io-pgtable-arm.c          | 18 +++++++++---------
 drivers/iommu/ipmmu-vmsa.c              |  2 +-
 drivers/iommu/msm_iommu.c               |  2 +-
 drivers/iommu/mtk_iommu.c               |  2 +-
 drivers/iommu/qcom_iommu.c              |  2 +-
 include/linux/io-pgtable.h              |  2 +-
 10 files changed, 26 insertions(+), 26 deletions(-)

drivers/gpu/drm/panfrost/panfrost_mmu.c +1 -1

index ed28aeb..5a39eee 100644
@@ -262,7 +262,7 @@ static int mmu_map_sg(struct panfrost_device *pfdev, struct panfrost_mmu *mmu,
 		while (len) {
 			size_t pgsize = get_pgsize(iova | paddr, len);
 
-			ops->map(ops, iova, paddr, pgsize, prot);
+			ops->map(ops, iova, paddr, pgsize, prot, GFP_KERNEL);
 			iova += pgsize;
 			paddr += pgsize;
 			len -= pgsize;

drivers/iommu/arm-smmu-v3.c +1 -1

index f578677..7b59f06 100644
@@ -2850,7 +2850,7 @@ static int arm_smmu_map(struct iommu_domain *domain, unsigned long iova,
 	if (!ops)
 		return -ENODEV;
 
-	return ops->map(ops, iova, paddr, size, prot);
+	return ops->map(ops, iova, paddr, size, prot, gfp);
 }
 
 static size_t arm_smmu_unmap(struct iommu_domain *domain, unsigned long iova,

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

index 243bc4c..dc1d253 100644
@@ -1227,7 +1227,7 @@ static int arm_smmu_map(struct iommu_domain *domain, unsigned long iova,
 		return -ENODEV;
 
 	arm_smmu_rpm_get(smmu);
-	ret = ops->map(ops, iova, paddr, size, prot);
+	ret = ops->map(ops, iova, paddr, size, prot, gfp);
 	arm_smmu_rpm_put(smmu);
 
 	return ret;

drivers/iommu/io-pgtable-arm-v7s.c +9 -9

index 4272fe4..a688f22 100644
@@ -470,7 +470,7 @@ static arm_v7s_iopte arm_v7s_install_table(arm_v7s_iopte *table,
 
 static int __arm_v7s_map(struct arm_v7s_io_pgtable *data, unsigned long iova,
 			 phys_addr_t paddr, size_t size, int prot,
-			 int lvl, arm_v7s_iopte *ptep)
+			 int lvl, arm_v7s_iopte *ptep, gfp_t gfp)
 {
 	struct io_pgtable_cfg *cfg = &data->iop.cfg;
 	arm_v7s_iopte pte, *cptep;

@@ -491,7 +491,7 @@ static int __arm_v7s_map(struct arm_v7s_io_pgtable *data, unsigned long iova,
 	/* Grab a pointer to the next level */
 	pte = READ_ONCE(*ptep);
 	if (!pte) {
-		cptep = __arm_v7s_alloc_table(lvl + 1, GFP_ATOMIC, data);
+		cptep = __arm_v7s_alloc_table(lvl + 1, gfp, data);
 		if (!cptep)
 			return -ENOMEM;

@@ -512,11 +512,11 @@ static int __arm_v7s_map(struct arm_v7s_io_pgtable *data, unsigned long iova,
 	}
 
 	/* Rinse, repeat */
-	return __arm_v7s_map(data, iova, paddr, size, prot, lvl + 1, cptep);
+	return __arm_v7s_map(data, iova, paddr, size, prot, lvl + 1, cptep, gfp);
 }
 
 static int arm_v7s_map(struct io_pgtable_ops *ops, unsigned long iova,
-			phys_addr_t paddr, size_t size, int prot)
+			phys_addr_t paddr, size_t size, int prot, gfp_t gfp)
 {
 	struct arm_v7s_io_pgtable *data = io_pgtable_ops_to_data(ops);
 	struct io_pgtable *iop = &data->iop;

@@ -530,7 +530,7 @@ static int arm_v7s_map(struct io_pgtable_ops *ops, unsigned long iova,
 		    paddr >= (1ULL << data->iop.cfg.oas)))
 		return -ERANGE;
 
-	ret = __arm_v7s_map(data, iova, paddr, size, prot, 1, data->pgd);
+	ret = __arm_v7s_map(data, iova, paddr, size, prot, 1, data->pgd, gfp);
 	/*
 	 * Synchronise all PTE updates for the new mapping before there's
 	 * a chance for anything to kick off a table walk for the new iova.

@@ -922,12 +922,12 @@ static int __init arm_v7s_do_selftests(void)
 		if (ops->map(ops, iova, iova, size, IOMMU_READ |
 						    IOMMU_WRITE |
 						    IOMMU_NOEXEC |
-						    IOMMU_CACHE))
+						    IOMMU_CACHE, GFP_KERNEL))
 			return __FAIL(ops);
 
 		/* Overlapping mappings */
 		if (!ops->map(ops, iova, iova + size, size,
-			      IOMMU_READ | IOMMU_NOEXEC))
+			      IOMMU_READ | IOMMU_NOEXEC, GFP_KERNEL))
 			return __FAIL(ops);
 
 		if (ops->iova_to_phys(ops, iova + 42) != (iova + 42))

@@ -946,7 +946,7 @@ static int __init arm_v7s_do_selftests(void)
 			return __FAIL(ops);
 
 		/* Remap of partial unmap */
-		if (ops->map(ops, iova_start + size, size, size, IOMMU_READ))
+		if (ops->map(ops, iova_start + size, size, size, IOMMU_READ, GFP_KERNEL))
 			return __FAIL(ops);
 
 		if (ops->iova_to_phys(ops, iova_start + size + 42)

@@ -967,7 +967,7 @@ static int __init arm_v7s_do_selftests(void)
 			return __FAIL(ops);
 
 		/* Remap full block */
-		if (ops->map(ops, iova, iova, size, IOMMU_WRITE))
+		if (ops->map(ops, iova, iova, size, IOMMU_WRITE, GFP_KERNEL))
 			return __FAIL(ops);
 
 		if (ops->iova_to_phys(ops, iova + 42) != (iova + 42))

drivers/iommu/io-pgtable-arm.c +9 -9

index 04fbd4b..4a5a7b0 100644
@@ -355,7 +355,7 @@ static arm_lpae_iopte arm_lpae_install_table(arm_lpae_iopte *table,
 
 static int __arm_lpae_map(struct arm_lpae_io_pgtable *data, unsigned long iova,
 			  phys_addr_t paddr, size_t size, arm_lpae_iopte prot,
-			  int lvl, arm_lpae_iopte *ptep)
+			  int lvl, arm_lpae_iopte *ptep, gfp_t gfp)
 {
 	arm_lpae_iopte *cptep, pte;
 	size_t block_size = ARM_LPAE_BLOCK_SIZE(lvl, data);

@@ -376,7 +376,7 @@ static int __arm_lpae_map(struct arm_lpae_io_pgtable *data, unsigned long iova,
 	/* Grab a pointer to the next level */
 	pte = READ_ONCE(*ptep);
 	if (!pte) {
-		cptep = __arm_lpae_alloc_pages(tblsz, GFP_ATOMIC, cfg);
+		cptep = __arm_lpae_alloc_pages(tblsz, gfp, cfg);
 		if (!cptep)
 			return -ENOMEM;

@@ -396,7 +396,7 @@ static int __arm_lpae_map(struct arm_lpae_io_pgtable *data, unsigned long iova,
 	}
 
 	/* Rinse, repeat */
-	return __arm_lpae_map(data, iova, paddr, size, prot, lvl + 1, cptep);
+	return __arm_lpae_map(data, iova, paddr, size, prot, lvl + 1, cptep, gfp);
 }
 
 static arm_lpae_iopte arm_lpae_prot_to_pte(struct arm_lpae_io_pgtable *data,

@@ -461,7 +461,7 @@ static arm_lpae_iopte arm_lpae_prot_to_pte(struct arm_lpae_io_pgtable *data,
 }
 
 static int arm_lpae_map(struct io_pgtable_ops *ops, unsigned long iova,
-			phys_addr_t paddr, size_t size, int iommu_prot)
+			phys_addr_t paddr, size_t size, int iommu_prot, gfp_t gfp)
 {
 	struct arm_lpae_io_pgtable *data = io_pgtable_ops_to_data(ops);
 	struct io_pgtable_cfg *cfg = &data->iop.cfg;

@@ -483,7 +483,7 @@ static int arm_lpae_map(struct io_pgtable_ops *ops, unsigned long iova,
 		return -ERANGE;
 
 	prot = arm_lpae_prot_to_pte(data, iommu_prot);
-	ret = __arm_lpae_map(data, iova, paddr, size, prot, lvl, ptep);
+	ret = __arm_lpae_map(data, iova, paddr, size, prot, lvl, ptep, gfp);
 	/*
 	 * Synchronise all PTE updates for the new mapping before there's
 	 * a chance for anything to kick off a table walk for the new iova.

@@ -1178,12 +1178,12 @@ static int __init arm_lpae_run_tests(struct io_pgtable_cfg *cfg)
 			if (ops->map(ops, iova, iova, size, IOMMU_READ |
 							    IOMMU_WRITE |
 							    IOMMU_NOEXEC |
-							    IOMMU_CACHE))
+							    IOMMU_CACHE, GFP_KERNEL))
 				return __FAIL(ops, i);
 
 			/* Overlapping mappings */
 			if (!ops->map(ops, iova, iova + size, size,
-				      IOMMU_READ | IOMMU_NOEXEC))
+				      IOMMU_READ | IOMMU_NOEXEC, GFP_KERNEL))
 				return __FAIL(ops, i);
 
 			if (ops->iova_to_phys(ops, iova + 42) != (iova + 42))

@@ -1198,7 +1198,7 @@ static int __init arm_lpae_run_tests(struct io_pgtable_cfg *cfg)
 			return __FAIL(ops, i);
 
 		/* Remap of partial unmap */
-		if (ops->map(ops, SZ_1G + size, size, size, IOMMU_READ))
+		if (ops->map(ops, SZ_1G + size, size, size, IOMMU_READ, GFP_KERNEL))
 			return __FAIL(ops, i);
 
 		if (ops->iova_to_phys(ops, SZ_1G + size + 42) != (size + 42))

@@ -1216,7 +1216,7 @@ static int __init arm_lpae_run_tests(struct io_pgtable_cfg *cfg)
 				return __FAIL(ops, i);
 
 			/* Remap full block */
-			if (ops->map(ops, iova, iova, size, IOMMU_WRITE))
+			if (ops->map(ops, iova, iova, size, IOMMU_WRITE, GFP_KERNEL))
 				return __FAIL(ops, i);
 
 			if (ops->iova_to_phys(ops, iova + 42) != (iova + 42))

drivers/iommu/ipmmu-vmsa.c +1 -1

index 4c2972f..87475b2 100644
@@ -687,7 +687,7 @@ static int ipmmu_map(struct iommu_domain *io_domain, unsigned long iova,
 	if (!domain)
 		return -ENODEV;
 
-	return domain->iop->map(domain->iop, iova, paddr, size, prot);
+	return domain->iop->map(domain->iop, iova, paddr, size, prot, gfp);
 }
 
 static size_t ipmmu_unmap(struct iommu_domain *io_domain, unsigned long iova,

drivers/iommu/msm_iommu.c +1 -1

index 3d8a635..12ce685 100644
@@ -491,7 +491,7 @@ static int msm_iommu_map(struct iommu_domain *domain, unsigned long iova,
 	int ret;
 
 	spin_lock_irqsave(&priv->pgtlock, flags);
-	ret = priv->iop->map(priv->iop, iova, pa, len, prot);
+	ret = priv->iop->map(priv->iop, iova, pa, len, prot, GFP_ATOMIC);
 	spin_unlock_irqrestore(&priv->pgtlock, flags);
 
 	return ret;

drivers/iommu/mtk_iommu.c +1 -1

index 2be96f1..b7b1641 100644
@@ -397,7 +397,7 @@ static int mtk_iommu_map(struct iommu_domain *domain, unsigned long iova,
 		paddr |= BIT_ULL(32);
 
 	/* Synchronize with the tlb_lock */
-	return dom->iop->map(dom->iop, iova, paddr, size, prot);
+	return dom->iop->map(dom->iop, iova, paddr, size, prot, gfp);
 }
 
 static size_t mtk_iommu_unmap(struct iommu_domain *domain,

drivers/iommu/qcom_iommu.c +1 -1

index c3e1fbd..cfcfd75 100644
@@ -441,7 +441,7 @@ static int qcom_iommu_map(struct iommu_domain *domain, unsigned long iova,
 		return -ENODEV;
 
 	spin_lock_irqsave(&qcom_domain->pgtbl_lock, flags);
-	ret = ops->map(ops, iova, paddr, size, prot);
+	ret = ops->map(ops, iova, paddr, size, prot, GFP_ATOMIC);
 	spin_unlock_irqrestore(&qcom_domain->pgtbl_lock, flags);
 	return ret;
 }

include/linux/io-pgtable.h +1 -1

index 53d53c6..23285ba 100644
@@ -155,7 +155,7 @@ struct io_pgtable_cfg {
  */
 struct io_pgtable_ops {
 	int (*map)(struct io_pgtable_ops *ops, unsigned long iova,
-		   phys_addr_t paddr, size_t size, int prot);
+		   phys_addr_t paddr, size_t size, int prot, gfp_t gfp);
 	size_t (*unmap)(struct io_pgtable_ops *ops, unsigned long iova,
 			size_t size, struct iommu_iotlb_gather *gather);
 	phys_addr_t (*iova_to_phys)(struct io_pgtable_ops *ops,

Signature · 1 line

1.8.3.1

Will Deacon

maintainer Acked-by on v2 2/2

2020-07-14 08:28 UTC #4 lore raw

Re: [PATCH v2 2/2] iommu: Add gfp parameter to io_pgtable_ops->map()

Quoted 8 lines · Baolin Wang

Now the ARM page tables are always allocated by GFP_ATOMIC parameter, but the iommu_ops->map() function has been added a gfp_t parameter by commit 781ca2de89ba ("iommu: Add gfp parameter to iommu_ops::map"), thus io_pgtable_ops->map() should use the gfp parameter passed from iommu_ops->map() to allocate page pages, which can avoid wasting the memory allocators atomic pools for some non-atomic contexts.
Signed-off-by Baolin Wang <[email protected]>

Quoted 12 lines

---
 drivers/gpu/drm/panfrost/panfrost_mmu.c |  2 +-
 drivers/iommu/arm-smmu-v3.c             |  2 +-
 drivers/iommu/arm-smmu.c                |  2 +-
 drivers/iommu/io-pgtable-arm-v7s.c      | 18 +++++++++---------
 drivers/iommu/io-pgtable-arm.c          | 18 +++++++++---------
 drivers/iommu/ipmmu-vmsa.c              |  2 +-
 drivers/iommu/msm_iommu.c               |  2 +-
 drivers/iommu/mtk_iommu.c               |  2 +-
 drivers/iommu/qcom_iommu.c              |  2 +-
 include/linux/io-pgtable.h              |  2 +-
 10 files changed, 26 insertions(+), 26 deletions(-)

I was a bit nervous about us passing GFP_KERNEL with a spinlock held, but it looks like you've checked all the callsites and it looks fine to me, so:

Acked-by Will Deacon <[email protected]>

Joerg -- not sure what you want to do with this one, as it's likely to conflict (trivially) with unrelated driver changes.

Will
_______________________________________________
dri-devel mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/dri-devel

Baolin Wang

author

2020-07-23 11:36 UTC #5 lore raw

Re: [PATCH v2 2/2] iommu: Add gfp parameter to io_pgtable_ops->map()

Quoted 22 lines · Will Deacon

Quoted 8 lines · Baolin Wang
Now the ARM page tables are always allocated by GFP_ATOMIC parameter, but the iommu_ops->map() function has been added a gfp_t parameter by commit 781ca2de89ba ("iommu: Add gfp parameter to iommu_ops::map"), thus io_pgtable_ops->map() should use the gfp parameter passed from iommu_ops->map() to allocate page pages, which can avoid wasting the memory allocators atomic pools for some non-atomic contexts.
Signed-off-by Baolin Wang <[email protected]>
Quoted 12 lines
---
 drivers/gpu/drm/panfrost/panfrost_mmu.c |  2 +-
 drivers/iommu/arm-smmu-v3.c             |  2 +-
 drivers/iommu/arm-smmu.c                |  2 +-
 drivers/iommu/io-pgtable-arm-v7s.c      | 18 +++++++++---------
 drivers/iommu/io-pgtable-arm.c          | 18 +++++++++---------
 drivers/iommu/ipmmu-vmsa.c              |  2 +-
 drivers/iommu/msm_iommu.c               |  2 +-
 drivers/iommu/mtk_iommu.c               |  2 +-
 drivers/iommu/qcom_iommu.c              |  2 +-
 include/linux/io-pgtable.h              |  2 +-
 10 files changed, 26 insertions(+), 26 deletions(-)

Quoted 7 lines

I was a bit nervous about us passing GFP_KERNEL with a spinlock held, but it looks like you've checked all the callsites and it looks fine to me, so:
Acked-by Will Deacon <[email protected]>
Joerg -- not sure what you want to do with this one, as it's likely to conflict (trivially) with unrelated driver changes.

Thanks Will. Joerg, could you apply this 2 patches if no objection from your side? Thanks.

Joerg Roedel

maintainer Applied

2020-07-24 12:30 UTC #6 lore raw

Re: [PATCH v2 2/2] iommu: Add gfp parameter to io_pgtable_ops->map()

Quoted 2 lines · Baolin Wang

Thanks Will. Joerg, could you apply this 2 patches if no objection from your side? Thanks.

Applied both, thanks. Modified the first patch because another patch already made __iommu_map static.