Skip to contentSkip to conversation

Misc bug fixes for VT-d SVM

PATCH Applied Jacob Pan 3 participants 12 messages Mar 20, 2020 – Mar 27, 2020 IOMMU

lore
Share & tools
lore

Jacob Pan

author
#0 lore raw

cover Misc bug fixes for VT-d SVM

Just a few bug fixes while testing Intel SVM code.

Jacob Pan (3):
iommu/vt-d: Remove redundant IOTLB flush
iommu/vt-d: Fix mm reference leak
iommu/vt-d: Add build dependency on IOASID

 drivers/iommu/Kconfig     |  1 +
 drivers/iommu/intel-svm.c | 13 ++++++-------
 2 files changed, 7 insertions(+), 7 deletions(-)
Signature · 1 line
2.7.4
author
#7 lore raw

PATCH 2/3 iommu/vt-d: Fix mm reference leak

Move canonical address check before mmget_not_zero() to avoid mm reference leak.

Fixes 9d8c3af31607 ("iommu/vt-d: IOMMU Page Request needs to check if

address is canonical.")

---
 drivers/iommu/intel-svm.c | 7 ++++---
 1 file changed, 4 insertions(+), 3 deletions(-)
drivers/iommu/intel-svm.c +4 -3
index 1483f1845762..56253c59ca10 100644
@@ -861,14 +861,15 @@ static irqreturn_t prq_event_thread(int irq, void *d)
 		 * any faults on kernel addresses. */
 		if (!svm->mm)
 			goto bad_req;
-		/* If the mm is already defunct, don't handle faults. */
-		if (!mmget_not_zero(svm->mm))
-			goto bad_req;
 
 		/* If address is not canonical, return invalid response */
 		if (!is_canonical_address(address))
 			goto bad_req;
 
+		/* If the mm is already defunct, don't handle faults. */
+		if (!mmget_not_zero(svm->mm))
+			goto bad_req;
+
 		down_read(&svm->mm->mmap_sem);
 		vma = find_extend_vma(svm->mm, address);
 		if (!vma || address < vma->vm_start)
Signature · 1 line
2.7.4
author
#9 lore raw

PATCH 3/3 iommu/vt-d: Add build dependency on IOASID

In reply to Jacob Pan

IOASID code is needed by VT-d scalable mode for PASID allocation. Add explicit dependency such that IOASID is built-in whenever Intel IOMMU is enabled.
Otherwise, aux domain code will fail when IOMMU is built-in and IOASID is compiled as a module.

---
 drivers/iommu/Kconfig | 1 +
 1 file changed, 1 insertion(+)
drivers/iommu/Kconfig +1 -0
index d2fade984999..25149544d57c 100644
@@ -188,6 +188,7 @@ config INTEL_IOMMU
 	select NEED_DMA_MAP_STATE
 	select DMAR_TABLE
 	select SWIOTLB
+	select IOASID
 	help
 	  DMA remapping (DMAR) devices support enables independent address
 	  translations for Direct Memory Access (DMA) from devices.
Signature · 1 line
2.7.4
author
#1 lore raw

PATCH 1/3 iommu/vt-d: Remove redundant IOTLB flush

In reply to Jacob Pan

IOTLB flush already included in the PASID tear down process. There is no need to flush again.

---
 drivers/iommu/intel-svm.c | 6 ++----
 1 file changed, 2 insertions(+), 4 deletions(-)
drivers/iommu/intel-svm.c +2 -4
index 8f42d717d8d7..1483f1845762 100644
@@ -268,10 +268,9 @@ static void intel_mm_release(struct mmu_notifier *mn, struct mm_struct *mm)
 	 * *has* to handle gracefully without affecting other processes.
 	 */
 	rcu_read_lock();
-	list_for_each_entry_rcu(sdev, &svm->devs, list) {
+	list_for_each_entry_rcu(sdev, &svm->devs, list)
 		intel_pasid_tear_down_entry(svm->iommu, sdev->dev, svm->pasid);
-		intel_flush_svm_range_dev(svm, sdev, 0, -1, 0);
-	}
+
 	rcu_read_unlock();
 
 }
@@ -731,7 +730,6 @@ int intel_svm_unbind_mm(struct device *dev, int pasid)
 			 * large and has to be physically contiguous. So it's
 			 * hard to be as defensive as we might like. */
 			intel_pasid_tear_down_entry(iommu, dev, svm->pasid);
-			intel_flush_svm_range_dev(svm, sdev, 0, -1, 0);
 			kfree_rcu(sdev, rcu);
 
 			if (list_empty(&svm->devs)) {
Signature · 6 lines
2.7.4

_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu
#2 lore raw

Re: [PATCH 1/3] iommu/vt-d: Remove redundant IOTLB flush

Quoted 2 lines · Jacob Pan

IOTLB flush already included in the PASID tear down process. There is no need to flush again.

It seems that intel_pasid_tear_down_entry() doesn't flush the pasid based device TLB?

Best regards,
baolu

Quoted 33 lines
---
  drivers/iommu/intel-svm.c | 6 ++----
  1 file changed, 2 insertions(+), 4 deletions(-)
drivers/iommu/intel-svm.c +2 -4
index 8f42d717d8d7..1483f1845762 100644
@@ -268,10 +268,9 @@ static void intel_mm_release(struct mmu_notifier *mn, struct mm_struct *mm)
  	 * *has* to handle gracefully without affecting other processes.
  	 */
  	rcu_read_lock();
-	list_for_each_entry_rcu(sdev, &svm->devs, list) {
+	list_for_each_entry_rcu(sdev, &svm->devs, list)
  		intel_pasid_tear_down_entry(svm->iommu, sdev->dev, svm->pasid);
-		intel_flush_svm_range_dev(svm, sdev, 0, -1, 0);
-	}
+
  	rcu_read_unlock();
 
  }
@@ -731,7 +730,6 @@ int intel_svm_unbind_mm(struct device *dev, int pasid)
  			 * large and has to be physically contiguous. So it's
  			 * hard to be as defensive as we might like. */
  			intel_pasid_tear_down_entry(iommu, dev, svm->pasid);
-			intel_flush_svm_range_dev(svm, sdev, 0, -1, 0);
  			kfree_rcu(sdev, rcu);
 
  			if (list_empty(&svm->devs)) {
Acked-by on v1 2/3
#8 lore raw

Re: [PATCH 2/3] iommu/vt-d: Fix mm reference leak

In reply to Jacob Pan
Quoted 7 lines · Jacob Pan

Move canonical address check before mmget_not_zero() to avoid mm reference leak.

Fixes 9d8c3af31607 ("iommu/vt-d: IOMMU Page Request needs to check if

address is canonical.")

Best regards,
baolu

Quoted 28 lines
---
  drivers/iommu/intel-svm.c | 7 ++++---
  1 file changed, 4 insertions(+), 3 deletions(-)
drivers/iommu/intel-svm.c +4 -3
index 1483f1845762..56253c59ca10 100644
@@ -861,14 +861,15 @@ static irqreturn_t prq_event_thread(int irq, void *d)
  		 * any faults on kernel addresses. */
  		if (!svm->mm)
  			goto bad_req;
-		/* If the mm is already defunct, don't handle faults. */
-		if (!mmget_not_zero(svm->mm))
-			goto bad_req;
 
  		/* If address is not canonical, return invalid response */
  		if (!is_canonical_address(address))
  			goto bad_req;
 
+		/* If the mm is already defunct, don't handle faults. */
+		if (!mmget_not_zero(svm->mm))
+			goto bad_req;
+
  		down_read(&svm->mm->mmap_sem);
  		vma = find_extend_vma(svm->mm, address);
  		if (!vma || address < vma->vm_start)
Acked-by on v1 3/3
#10 lore raw

Re: [PATCH 3/3] iommu/vt-d: Add build dependency on IOASID

In reply to Jacob Pan
Quoted 7 lines · Jacob Pan

IOASID code is needed by VT-d scalable mode for PASID allocation. Add explicit dependency such that IOASID is built-in whenever Intel IOMMU is enabled.
Otherwise, aux domain code will fail when IOMMU is built-in and IOASID is compiled as a module.

Fixes 59a623374dc38 ("iommu/vt-d: Replace Intel specific PASID allocator with IOASID")

Best regards,
baolu

Quoted 17 lines
---
  drivers/iommu/Kconfig | 1 +
  1 file changed, 1 insertion(+)
drivers/iommu/Kconfig +1 -0
index d2fade984999..25149544d57c 100644
@@ -188,6 +188,7 @@ config INTEL_IOMMU
  	select NEED_DMA_MAP_STATE
  	select DMAR_TABLE
  	select SWIOTLB
+	select IOASID
  	help
  	  DMA remapping (DMAR) devices support enables independent address
  	  translations for Direct Memory Access (DMA) from devices.
author
#3 lore raw

Re: [PATCH 1/3] iommu/vt-d: Remove redundant IOTLB flush

In reply to Lu Baolu
Quoted 7 lines · On Fri, 20 Mar 2020 21:45:26 +0800 Lu Baolu <[email protected]> wrote:
Quoted 2 lines · Jacob Pan

IOTLB flush already included in the PASID tear down process. There is no need to flush again.

It seems that intel_pasid_tear_down_entry() doesn't flush the pasid based device TLB?

I saw this code in intel_pasid_tear_down_entry(). Isn't the last line flush the devtlb? Not in guest of course since the passdown tlb flush is inclusive.

	pasid_cache_invalidation_with_pasid(iommu, did, pasid);
	iotlb_invalidation_with_pasid(iommu, did, pasid);

	/* Device IOTLB doesn't need to be flushed in caching mode. */
	if (!cap_caching_mode(iommu->cap))
		devtlb_invalidation_with_pasid(iommu, dev, pasid);
Quoted 38 lines

Best regards,
baolu

Quoted 35 lines
---
  drivers/iommu/intel-svm.c | 6 ++----
  1 file changed, 2 insertions(+), 4 deletions(-)
drivers/iommu/intel-svm.c +0 -0
index 8f42d717d8d7..1483f1845762 100644
@@ -268,10 +268,9 @@ static void intel_mm_release(struct
 mmu_notifier *mn, struct mm_struct *mm)
  	 * *has* to handle gracefully without affecting other

processes. */
rcu_read_lock();

drivers/iommu/intel-svm.c +1 -1
-	list_for_each_entry_rcu(sdev, &svm->devs, list) {
+	list_for_each_entry_rcu(sdev, &svm->devs, list)
  		intel_pasid_tear_down_entry(svm->iommu,

sdev->dev, svm->pasid);

drivers/iommu/intel-svm.c +1 -2
-		intel_flush_svm_range_dev(svm, sdev, 0, -1, 0);
-	}
+
  	rcu_read_unlock();
 
  }
@@ -731,7 +730,6 @@ int intel_svm_unbind_mm(struct device *dev, int
 pasid)
  			 * large and has to be physically

contiguous. So it's
* hard to be as defensive as we might
like. */ intel_pasid_tear_down_entry(iommu, dev, svm->pasid);

drivers/iommu/intel-svm.c +0 -2
-			intel_flush_svm_range_dev(svm, sdev, 0,
-1, 0); kfree_rcu(sdev, rcu);
Quoted 2 lines
Quoted 2 lines
  			if (list_empty(&svm->devs)) {

[Jacob Pan]

#4 lore raw

Re: [PATCH 1/3] iommu/vt-d: Remove redundant IOTLB flush

Quoted 14 lines · Jacob Pan
Quoted 7 lines · On Fri, 20 Mar 2020 21:45:26 +0800 Lu Baolu <[email protected]> wrote:
Quoted 2 lines · Jacob Pan

IOTLB flush already included in the PASID tear down process. There is no need to flush again.

It seems that intel_pasid_tear_down_entry() doesn't flush the pasid based device TLB?

I saw this code in intel_pasid_tear_down_entry(). Isn't the last line flush the devtlb? Not in guest of course since the passdown tlb flush is inclusive.

Quoted 7 lines
	pasid_cache_invalidation_with_pasid(iommu, did, pasid);
	iotlb_invalidation_with_pasid(iommu, did, pasid);

	/* Device IOTLB doesn't need to be flushed in caching mode. */
	if (!cap_caching_mode(iommu->cap))
		devtlb_invalidation_with_pasid(iommu, dev, pasid);

But devtlb_invalidation_with_pasid() doesn't do the right thing, it flushes the device tlb, instead of pasid-based device tlb.

static void
devtlb_invalidation_with_pasid(struct intel_iommu *iommu,
struct device *dev, int pasid)

{
        struct device_domain_info *info;
        u16 sid, qdep, pfsid;

        info = dev->archdata.iommu;
        if (!info || !info->ats_enabled)
                return;

        sid = info->bus << 8 | info->devfn;
        qdep = info->ats_qdep;
        pfsid = info->pfsid;

        qi_flush_dev_iotlb(iommu, sid, pfsid, qdep, 0, 64 - 

VTD_PAGE_SHIFT);
}

Best regards,
baolu

Quoted 43 lines
Quoted 38 lines

Best regards,
baolu

Quoted 35 lines
---
   drivers/iommu/intel-svm.c | 6 ++----
   1 file changed, 2 insertions(+), 4 deletions(-)
drivers/iommu/intel-svm.c +0 -0
index 8f42d717d8d7..1483f1845762 100644
@@ -268,10 +268,9 @@ static void intel_mm_release(struct
 mmu_notifier *mn, struct mm_struct *mm)
   	 * *has* to handle gracefully without affecting other

processes. */
rcu_read_lock();

drivers/iommu/intel-svm.c +1 -1
-	list_for_each_entry_rcu(sdev, &svm->devs, list) {
+	list_for_each_entry_rcu(sdev, &svm->devs, list)
   		intel_pasid_tear_down_entry(svm->iommu,

sdev->dev, svm->pasid);

drivers/iommu/intel-svm.c +1 -2
-		intel_flush_svm_range_dev(svm, sdev, 0, -1, 0);
-	}
+
   	rcu_read_unlock();
 
   }
@@ -731,7 +730,6 @@ int intel_svm_unbind_mm(struct device *dev, int
 pasid)
   			 * large and has to be physically

contiguous. So it's
* hard to be as defensive as we might
like. */ intel_pasid_tear_down_entry(iommu, dev, svm->pasid);

drivers/iommu/intel-svm.c +0 -2
-			intel_flush_svm_range_dev(svm, sdev, 0,
-1, 0); kfree_rcu(sdev, rcu);
Quoted 2 lines
Quoted 2 lines
   			if (list_empty(&svm->devs)) {

[Jacob Pan]

author
#5 lore raw

Re: [PATCH 1/3] iommu/vt-d: Remove redundant IOTLB flush

Quoted 18 lines · On Sat, 21 Mar 2020 09:32:45 +0800 Lu Baolu <[email protected]> wrote:
Quoted 17 lines · Jacob Pan
Quoted 7 lines · On Fri, 20 Mar 2020 21:45:26 +0800 Lu Baolu <[email protected]> wrote:
Quoted 2 lines · Jacob Pan

IOTLB flush already included in the PASID tear down process. There is no need to flush again.

It seems that intel_pasid_tear_down_entry() doesn't flush the pasid based device TLB?

I saw this code in intel_pasid_tear_down_entry(). Isn't the last line flush the devtlb? Not in guest of course since the passdown tlb flush is inclusive.

	pasid_cache_invalidation_with_pasid(iommu, did, pasid);
	iotlb_invalidation_with_pasid(iommu, did, pasid);
Quoted 8 lines
Quoted 4 lines
	/* Device IOTLB doesn't need to be flushed in caching mode.
*/ if (!cap_caching_mode(iommu->cap))
		devtlb_invalidation_with_pasid(iommu, dev, pasid);

But devtlb_invalidation_with_pasid() doesn't do the right thing, it flushes the device tlb, instead of pasid-based device tlb.

Hmm, you are right. But the function name is misleading, pasid argument is not used, is there a reason why?
This is used for PASID based device IOTLB flush, right?

Quoted 16 lines

static void
devtlb_invalidation_with_pasid(struct intel_iommu *iommu,
struct device *dev, int pasid)

{
         struct device_domain_info *info;
         u16 sid, qdep, pfsid;

         info = dev->archdata.iommu;
         if (!info || !info->ats_enabled)
                 return;

         sid = info->bus << 8 | info->devfn;
         qdep = info->ats_qdep;
         pfsid = info->pfsid;

         qi_flush_dev_iotlb(iommu, sid, pfsid, qdep, 0, 64 - 
Quoted 49 lines

VTD_PAGE_SHIFT);
}

Best regards,
baolu

Quoted 43 lines
Quoted 38 lines

Best regards,
baolu

Quoted 35 lines
---
   drivers/iommu/intel-svm.c | 6 ++----
   1 file changed, 2 insertions(+), 4 deletions(-)
drivers/iommu/intel-svm.c +0 -0
index 8f42d717d8d7..1483f1845762 100644
@@ -268,10 +268,9 @@ static void intel_mm_release(struct
 mmu_notifier *mn, struct mm_struct *mm)
   	 * *has* to handle gracefully without affecting other

processes. */
rcu_read_lock();

drivers/iommu/intel-svm.c +1 -1
-	list_for_each_entry_rcu(sdev, &svm->devs, list) {
+	list_for_each_entry_rcu(sdev, &svm->devs, list)
   		intel_pasid_tear_down_entry(svm->iommu,

sdev->dev, svm->pasid);

drivers/iommu/intel-svm.c +1 -2
-		intel_flush_svm_range_dev(svm, sdev, 0, -1, 0);
-	}
+
   	rcu_read_unlock();
 
   }
@@ -731,7 +730,6 @@ int intel_svm_unbind_mm(struct device *dev,
 int pasid)
   			 * large and has to be physically

contiguous. So it's
* hard to be as defensive as we might
like. */ intel_pasid_tear_down_entry(iommu, dev, svm->pasid);

drivers/iommu/intel-svm.c +0 -2
-			intel_flush_svm_range_dev(svm, sdev, 0,
-1, 0); kfree_rcu(sdev, rcu);
Quoted 2 lines
Quoted 2 lines
   			if (list_empty(&svm->devs)) {

[Jacob Pan]

[Jacob Pan]

#6 lore raw

Re: [PATCH 1/3] iommu/vt-d: Remove redundant IOTLB flush

Quoted 24 lines · Jacob Pan
Quoted 21 lines · On Sat, 21 Mar 2020 09:32:45 +0800 Lu Baolu<[email protected]> wrote:
Quoted 20 lines · Jacob Pan
Quoted 6 lines · On Fri, 20 Mar 2020 21:45:26 +0800 Lu Baolu<[email protected]> wrote:
Quoted 2 lines · Jacob Pan

IOTLB flush already included in the PASID tear down process. There is no need to flush again.

It seems that intel_pasid_tear_down_entry() doesn't flush the pasid based device TLB?

I saw this code in intel_pasid_tear_down_entry(). Isn't the last line flush the devtlb? Not in guest of course since the passdown tlb flush is inclusive.

	pasid_cache_invalidation_with_pasid(iommu, did, pasid);
	iotlb_invalidation_with_pasid(iommu, did, pasid);

	/* Device IOTLB doesn't need to be flushed in caching mode.
*/ if (!cap_caching_mode(iommu->cap))
		devtlb_invalidation_with_pasid(iommu, dev, pasid);
Quoted 7 lines
Quoted 3 lines

But devtlb_invalidation_with_pasid() doesn't do the right thing, it flushes the device tlb, instead of pasid-based device tlb.

Hmm, you are right. But the function name is misleading, pasid argument is not used, is there a reason why?
This is used for PASID based device IOTLB flush, right?

Yes. I will fix and put your patch after it.

Best regards,
baolu
_______________________________________________
iommu mailing list
[email protected]
https://lists.linuxfoundation.org/mailman/listinfo/iommu

maintainer Applied
#11 lore raw
In reply to Jacob Pan
Quoted 2 lines · Jacob Pan

iommu/vt-d: Fix mm reference leak
iommu/vt-d: Add build dependency on IOASID

Applied these two, thanks.