Skip to contentSkip to conversation

iommu/dma: Fix MSI reservation allocation

PATCH Superseded Marc Zyngier 3 participants 4 messages Mar 3, 2020

Subsystems (3)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 iommu/dma: Fix MSI reservation allocation

The way cookie_init_hw_msi_region() allocates the iommu_dma_msi_page structures doesn't match the way iommu_put_dma_cookie() frees them.

The former performs a single allocation of all the required structures, while the latter tries to free them one at a time. It doesn't quite work for the main use case (the GICv3 ITS where the range is 64kB) when the base ganule size is 4kB.

This leads to a nice slab corruption on teardown, which is easily observable by simply creating a VF on a SRIOV-capable device, and tearing it down immediately (no need to even make use of it).

Fix it by allocating iommu_dma_msi_page structures one at a time.

Fixes 7c1b058c8b5a3 ("iommu/dma: Handle IOMMU API reserved regions")
Cc Robin Murphy <[email protected]>
Cc Joerg Roedel <[email protected]>
Cc Eric Auger <[email protected]>
Cc Will Deacon <[email protected]>
---
 drivers/iommu/dma-iommu.c | 36 ++++++++++++++++++++++++------------
 1 file changed, 24 insertions(+), 12 deletions(-)
drivers/iommu/dma-iommu.c +24 -12
index a2e96a5fd9a7..01fa64856c12 100644
@@ -171,25 +171,37 @@ static int cookie_init_hw_msi_region(struct iommu_dma_cookie *cookie,
 		phys_addr_t start, phys_addr_t end)
 {
 	struct iova_domain *iovad = &cookie->iovad;
-	struct iommu_dma_msi_page *msi_page;
-	int i, num_pages;
+	struct iommu_dma_msi_page *msi_page, *tmp;
+	int i, num_pages, ret = 0;
+	phys_addr_t base;
 
-	start -= iova_offset(iovad, start);
+	base = start -= iova_offset(iovad, start);
 	num_pages = iova_align(iovad, end - start) >> iova_shift(iovad);
 
-	msi_page = kcalloc(num_pages, sizeof(*msi_page), GFP_KERNEL);
-	if (!msi_page)
-		return -ENOMEM;
-
 	for (i = 0; i < num_pages; i++) {
-		msi_page[i].phys = start;
-		msi_page[i].iova = start;
-		INIT_LIST_HEAD(&msi_page[i].list);
-		list_add(&msi_page[i].list, &cookie->msi_page_list);
+		msi_page = kmalloc(sizeof(*msi_page), GFP_KERNEL);
+		if (!msi_page) {
+			ret = -ENOMEM;
+			break;
+		}
+		msi_page->phys = start;
+		msi_page->iova = start;
+		INIT_LIST_HEAD(&msi_page->list);
+		list_add(&msi_page->list, &cookie->msi_page_list);
 		start += iovad->granule;
 	}
 
-	return 0;
+	if (ret) {
+		list_for_each_entry_safe(msi_page, tmp,
+					 &cookie->msi_page_list, list) {
+			if (msi_page->phys >= base && msi_page->phys < start) {
+				list_del(&msi_page->list);
+				kfree(msi_page);
+			}
+		}
+	}
+
+	return ret;
 }
 
 static int iova_reserve_pci_windows(struct pci_dev *dev,
Signature · 1 line
2.20.1
Reviewed-by on v1 1/1
#1 lore raw

Hi Marc,

Quoted 8 lines · Marc Zyngier

The way cookie_init_hw_msi_region() allocates the iommu_dma_msi_page structures doesn't match the way iommu_put_dma_cookie() frees them.

The former performs a single allocation of all the required structures, while the latter tries to free them one at a time. It doesn't quite work for the main use case (the GICv3 ITS where the range is 64kB) when the base ganule size is 4kB.

Quoted 7 lines

This leads to a nice slab corruption on teardown, which is easily observable by simply creating a VF on a SRIOV-capable device, and tearing it down immediately (no need to even make use of it).

Fix it by allocating iommu_dma_msi_page structures one at a time.

Fixes 7c1b058c8b5a3 ("iommu/dma: Handle IOMMU API reserved regions")

Thanks

Eric

Quoted 65 lines
Cc Robin Murphy <[email protected]>
Cc Joerg Roedel <[email protected]>
Cc Eric Auger <[email protected]>
Cc Will Deacon <[email protected]>
---
 drivers/iommu/dma-iommu.c | 36 ++++++++++++++++++++++++------------
 1 file changed, 24 insertions(+), 12 deletions(-)
drivers/iommu/dma-iommu.c +24 -12
index a2e96a5fd9a7..01fa64856c12 100644
@@ -171,25 +171,37 @@ static int cookie_init_hw_msi_region(struct iommu_dma_cookie *cookie,
 		phys_addr_t start, phys_addr_t end)
 {
 	struct iova_domain *iovad = &cookie->iovad;
-	struct iommu_dma_msi_page *msi_page;
-	int i, num_pages;
+	struct iommu_dma_msi_page *msi_page, *tmp;
+	int i, num_pages, ret = 0;
+	phys_addr_t base;
 
-	start -= iova_offset(iovad, start);
+	base = start -= iova_offset(iovad, start);
 	num_pages = iova_align(iovad, end - start) >> iova_shift(iovad);
 
-	msi_page = kcalloc(num_pages, sizeof(*msi_page), GFP_KERNEL);
-	if (!msi_page)
-		return -ENOMEM;
-
 	for (i = 0; i < num_pages; i++) {
-		msi_page[i].phys = start;
-		msi_page[i].iova = start;
-		INIT_LIST_HEAD(&msi_page[i].list);
-		list_add(&msi_page[i].list, &cookie->msi_page_list);
+		msi_page = kmalloc(sizeof(*msi_page), GFP_KERNEL);
+		if (!msi_page) {
+			ret = -ENOMEM;
+			break;
+		}
+		msi_page->phys = start;
+		msi_page->iova = start;
+		INIT_LIST_HEAD(&msi_page->list);
+		list_add(&msi_page->list, &cookie->msi_page_list);
 		start += iovad->granule;
 	}
 
-	return 0;
+	if (ret) {
+		list_for_each_entry_safe(msi_page, tmp,
+					 &cookie->msi_page_list, list) {
+			if (msi_page->phys >= base && msi_page->phys < start) {
+				list_del(&msi_page->list);
+				kfree(msi_page);
+			}
+		}
+	}
+
+	return ret;
 }
 
 static int iova_reserve_pci_windows(struct pci_dev *dev,
maintainer
#2 lore raw
In reply to Marc Zyngier
Quoted 8 lines · Marc Zyngier

The way cookie_init_hw_msi_region() allocates the iommu_dma_msi_page structures doesn't match the way iommu_put_dma_cookie() frees them.

The former performs a single allocation of all the required structures, while the latter tries to free them one at a time. It doesn't quite work for the main use case (the GICv3 ITS where the range is 64kB) when the base ganule size is 4kB.

Quoted 5 lines

This leads to a nice slab corruption on teardown, which is easily observable by simply creating a VF on a SRIOV-capable device, and tearing it down immediately (no need to even make use of it).

Fix it by allocating iommu_dma_msi_page structures one at a time.

Bleh, you know you're supposed to be using 64K pages on those things, right? :P

Quoted 31 lines
Fixes 7c1b058c8b5a3 ("iommu/dma: Handle IOMMU API reserved regions")
Cc Robin Murphy <[email protected]>
Cc Joerg Roedel <[email protected]>
Cc Eric Auger <[email protected]>
Cc Will Deacon <[email protected]>
---
  drivers/iommu/dma-iommu.c | 36 ++++++++++++++++++++++++------------
  1 file changed, 24 insertions(+), 12 deletions(-)
drivers/iommu/dma-iommu.c +4 -5
index a2e96a5fd9a7..01fa64856c12 100644
@@ -171,25 +171,37 @@ static int cookie_init_hw_msi_region(struct iommu_dma_cookie *cookie,
  		phys_addr_t start, phys_addr_t end)
  {
  	struct iova_domain *iovad = &cookie->iovad;
-	struct iommu_dma_msi_page *msi_page;
-	int i, num_pages;
+	struct iommu_dma_msi_page *msi_page, *tmp;
+	int i, num_pages, ret = 0;
+	phys_addr_t base;
 
-	start -= iova_offset(iovad, start);
+	base = start -= iova_offset(iovad, start);
  	num_pages = iova_align(iovad, end - start) >> iova_shift(iovad);
 
-	msi_page = kcalloc(num_pages, sizeof(*msi_page), GFP_KERNEL);
-	if (!msi_page)

drivers/iommu/dma-iommu.c static int cookie_init_hw_msi_region(struct iommu_dma_cookie *cookie, · line 184

Quoted 10 lines
drivers/iommu/dma-iommu.c +3 -6
-		return -ENOMEM;
-
  	for (i = 0; i < num_pages; i++) {
-		msi_page[i].phys = start;
-		msi_page[i].iova = start;
-		INIT_LIST_HEAD(&msi_page[i].list);
-		list_add(&msi_page[i].list, &cookie->msi_page_list);
+		msi_page = kmalloc(sizeof(*msi_page), GFP_KERNEL);
+		if (!msi_page) {
+			ret = -ENOMEM;

I think we can just return here and skip the cleanup below - by the time we get here the cookie itself has already been allocated and initialised, so even if iommu_dma_init_domain() fails someone else has already accepted the responsibility of calling iommu_put_dma_cookie() at some point later, which will clean up properly.

Cheers,
Robin.

Quoted 25 lines
drivers/iommu/dma-iommu.c +17 -1
+			break;
+		}
+		msi_page->phys = start;
+		msi_page->iova = start;
+		INIT_LIST_HEAD(&msi_page->list);
+		list_add(&msi_page->list, &cookie->msi_page_list);
  		start += iovad->granule;
  	}
 
-	return 0;
+	if (ret) {
+		list_for_each_entry_safe(msi_page, tmp,
+					 &cookie->msi_page_list, list) {
+			if (msi_page->phys >= base && msi_page->phys < start) {
+				list_del(&msi_page->list);
+				kfree(msi_page);
+			}
+		}
+	}
+
+	return ret;
  }
 
  static int iova_reserve_pci_windows(struct pci_dev *dev,
#3 lore raw
Quoted 9 lines · Robin Murphy
Quoted 8 lines · Marc Zyngier

The way cookie_init_hw_msi_region() allocates the iommu_dma_msi_page structures doesn't match the way iommu_put_dma_cookie() frees them.

The former performs a single allocation of all the required structures, while the latter tries to free them one at a time. It doesn't quite work for the main use case (the GICv3 ITS where the range is 64kB) when the base ganule size is 4kB.

Quoted 7 lines
Quoted 5 lines

This leads to a nice slab corruption on teardown, which is easily observable by simply creating a VF on a SRIOV-capable device, and tearing it down immediately (no need to even make use of it).

Fix it by allocating iommu_dma_msi_page structures one at a time.

Bleh, you know you're supposed to be using 64K pages on those things, right? :P

lalalala... ;-)

[...]

Quoted 8 lines

drivers/iommu/dma-iommu.c static int cookie_init_hw_msi_region(struct iommu_dma_cookie *cookie, · line 184

Quoted 2 lines
drivers/iommu/dma-iommu.c +2 -0
+		if (!msi_page) {
+			ret = -ENOMEM;

I think we can just return here and skip the cleanup below - by the time we get here the cookie itself has already been allocated and initialised, so even if iommu_dma_init_domain() fails someone else has already accepted the responsibility of calling iommu_put_dma_cookie() at some point later, which will clean up properly.

Ah, that's a very good point. I'll refresh the patch with a simplified error handling.

Thanks,

        M.
Signature · 1 line
Jazz is not dead. It just smells funny...