Skip to contentSkip to conversation

PCI: vmd: Prevent recursive locking on interrupt allocation

PATCH Applied Thomas Gleixner 5 participants 5 messages Feb 13, 2022 – Mar 1, 2022

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 PCI: vmd: Prevent recursive locking on interrupt allocation

Tejas reported the following recursive locking issue:

swapper/0/1 is trying to acquire lock:
ffff8881074fd0a0 (&md->mutex){+.+.}-{3:3}, at: msi_get_virq+0x30/0xc0

but task is already holding lock:
ffff8881017cd6a0 (&md->mutex){+.+.}-{3:3}, at: __pci_enable_msi_range+0xf2/0x290

 stack backtrace:
  __mutex_lock+0x9d/0x920
  msi_get_virq+0x30/0xc0
  pci_irq_vector+0x26/0x30
  vmd_msi_init+0xcc/0x210
  msi_domain_alloc+0xbf/0x150
  msi_domain_alloc_irqs_descs_locked+0x3e/0xb0
  __pci_enable_msi_range+0x155/0x290
  pci_alloc_irq_vectors_affinity+0xba/0x100
  pcie_port_device_register+0x307/0x550
  pcie_portdrv_probe+0x3c/0xd0
  pci_device_probe+0x95/0x110

This is caused by the VMD MSI code which does a lookup of the Linux interrupt number for an VMD managed MSI[X] vector. The lookup function tries to acquire the already held mutex.

Avoid that by caching the Linux interrupt number at initialization time instead of looking it up over and over.

Fixes 82ff8e6b78fc ("PCI/MSI: Use msi_get_virq() in pci_get_vector()")
---
 drivers/pci/controller/vmd.c |   14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)
drivers/pci/controller/vmd.c +7 -7
@@ -99,11 +99,13 @@ struct vmd_irq {
  * @srcu:	SRCU struct for local synchronization.
  * @count:	number of child IRQs assigned to this vector; used to track
  *		sharing.
+ * @virq:	The underlying VMD Linux interrupt number
  */
 struct vmd_irq_list {
 	struct list_head	irq_list;
 	struct srcu_struct	srcu;
 	unsigned int		count;
+	unsigned int		virq;
 };
 
 struct vmd_dev {
@@ -253,7 +255,6 @@ static int vmd_msi_init(struct irq_domai
 	struct msi_desc *desc = arg->desc;
 	struct vmd_dev *vmd = vmd_from_bus(msi_desc_to_pci_dev(desc)->bus);
 	struct vmd_irq *vmdirq = kzalloc(sizeof(*vmdirq), GFP_KERNEL);
-	unsigned int index, vector;
 
 	if (!vmdirq)
 		return -ENOMEM;
@@ -261,10 +262,8 @@ static int vmd_msi_init(struct irq_domai
 	INIT_LIST_HEAD(&vmdirq->node);
 	vmdirq->irq = vmd_next_irq(vmd, desc);
 	vmdirq->virq = virq;
-	index = index_from_irqs(vmd, vmdirq->irq);
-	vector = pci_irq_vector(vmd->dev, index);
 
-	irq_domain_set_info(domain, virq, vector, info->chip, vmdirq,
+	irq_domain_set_info(domain, virq, vmdirq->irq->virq, info->chip, vmdirq,
 			    handle_untracked_irq, vmd, NULL);
 	return 0;
 }
@@ -685,7 +684,8 @@ static int vmd_alloc_irqs(struct vmd_dev
 			return err;
 
 		INIT_LIST_HEAD(&vmd->irqs[i].irq_list);
-		err = devm_request_irq(&dev->dev, pci_irq_vector(dev, i),
+		vmd->irqs[i].virq = pci_irq_vector(dev, i);
+		err = devm_request_irq(&dev->dev, vmd->irqs[i].virq,
 				       vmd_irq, IRQF_NO_THREAD,
 				       vmd->name, &vmd->irqs[i]);
 		if (err)
@@ -969,7 +969,7 @@ static int vmd_suspend(struct device *de
 	int i;
 
 	for (i = 0; i < vmd->msix_count; i++)
-		devm_free_irq(dev, pci_irq_vector(pdev, i), &vmd->irqs[i]);
+		devm_free_irq(dev, vmd->irqs[i].virq, &vmd->irqs[i]);
 
 	return 0;
 }
@@ -981,7 +981,7 @@ static int vmd_resume(struct device *dev
 	int err, i;
 
 	for (i = 0; i < vmd->msix_count; i++) {
-		err = devm_request_irq(dev, pci_irq_vector(pdev, i),
+		err = devm_request_irq(dev, vmd->irqs[i].virq,
 				       vmd_irq, IRQF_NO_THREAD,
 				       vmd->name, &vmd->irqs[i]);
 		if (err)
#1 lore raw
The following commit has been merged into the irq/urgent branch of tip:

The following commit has been merged into the irq/urgent branch of tip:

Commit-ID:     ba1366f3d039e7c3ca1fc29ed00ce3ed2b8fd32f
Gitweb:        https://git.kernel.org/tip/ba1366f3d039e7c3ca1fc29ed00ce3ed2b8fd32f
Author:        Thomas Gleixner <[email protected]>
AuthorDate:    Sun, 13 Feb 2022 14:54:05 +01:00
Committer:     Thomas Gleixner <[email protected]>
CommitterDate: Mon, 21 Feb 2022 08:26:53 +01:00

PCI: vmd: Prevent recursive locking on interrupt allocation

Tejas reported the following recursive locking issue:

swapper/0/1 is trying to acquire lock:
ffff8881074fd0a0 (&md->mutex){+.+.}-{3:3}, at: msi_get_virq+0x30/0xc0

but task is already holding lock:
ffff8881017cd6a0 (&md->mutex){+.+.}-{3:3}, at: __pci_enable_msi_range+0xf2/0x290

 stack backtrace:
  __mutex_lock+0x9d/0x920
  msi_get_virq+0x30/0xc0
  pci_irq_vector+0x26/0x30
  vmd_msi_init+0xcc/0x210
  msi_domain_alloc+0xbf/0x150
  msi_domain_alloc_irqs_descs_locked+0x3e/0xb0
  __pci_enable_msi_range+0x155/0x290
  pci_alloc_irq_vectors_affinity+0xba/0x100
  pcie_port_device_register+0x307/0x550
  pcie_portdrv_probe+0x3c/0xd0
  pci_device_probe+0x95/0x110

This is caused by the VMD MSI code which does a lookup of the Linux interrupt number for an VMD managed MSI[X] vector. The lookup function tries to acquire the already held mutex.

Avoid that by caching the Linux interrupt number at initialization time instead of looking it up over and over.

---
 drivers/pci/controller/vmd.c | 14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)
drivers/pci/controller/vmd.c +7 -7
index cc166c6..eb05cce 100644
@@ -99,11 +99,13 @@ struct vmd_irq {
  * @srcu:	SRCU struct for local synchronization.
  * @count:	number of child IRQs assigned to this vector; used to track
  *		sharing.
+ * @virq:	The underlying VMD Linux interrupt number
  */
 struct vmd_irq_list {
 	struct list_head	irq_list;
 	struct srcu_struct	srcu;
 	unsigned int		count;
+	unsigned int		virq;
 };
 
 struct vmd_dev {
@@ -253,7 +255,6 @@ static int vmd_msi_init(struct irq_domain *domain, struct msi_domain_info *info,
 	struct msi_desc *desc = arg->desc;
 	struct vmd_dev *vmd = vmd_from_bus(msi_desc_to_pci_dev(desc)->bus);
 	struct vmd_irq *vmdirq = kzalloc(sizeof(*vmdirq), GFP_KERNEL);
-	unsigned int index, vector;
 
 	if (!vmdirq)
 		return -ENOMEM;
@@ -261,10 +262,8 @@ static int vmd_msi_init(struct irq_domain *domain, struct msi_domain_info *info,
 	INIT_LIST_HEAD(&vmdirq->node);
 	vmdirq->irq = vmd_next_irq(vmd, desc);
 	vmdirq->virq = virq;
-	index = index_from_irqs(vmd, vmdirq->irq);
-	vector = pci_irq_vector(vmd->dev, index);
 
-	irq_domain_set_info(domain, virq, vector, info->chip, vmdirq,
+	irq_domain_set_info(domain, virq, vmdirq->irq->virq, info->chip, vmdirq,
 			    handle_untracked_irq, vmd, NULL);
 	return 0;
 }
@@ -685,7 +684,8 @@ static int vmd_alloc_irqs(struct vmd_dev *vmd)
 			return err;
 
 		INIT_LIST_HEAD(&vmd->irqs[i].irq_list);
-		err = devm_request_irq(&dev->dev, pci_irq_vector(dev, i),
+		vmd->irqs[i].virq = pci_irq_vector(dev, i);
+		err = devm_request_irq(&dev->dev, vmd->irqs[i].virq,
 				       vmd_irq, IRQF_NO_THREAD,
 				       vmd->name, &vmd->irqs[i]);
 		if (err)
@@ -969,7 +969,7 @@ static int vmd_suspend(struct device *dev)
 	int i;
 
 	for (i = 0; i < vmd->msix_count; i++)
-		devm_free_irq(dev, pci_irq_vector(pdev, i), &vmd->irqs[i]);
+		devm_free_irq(dev, vmd->irqs[i].virq, &vmd->irqs[i]);
 
 	return 0;
 }
@@ -981,7 +981,7 @@ static int vmd_resume(struct device *dev)
 	int err, i;
 
 	for (i = 0; i < vmd->msix_count; i++) {
-		err = devm_request_irq(dev, pci_irq_vector(pdev, i),
+		err = devm_request_irq(dev, vmd->irqs[i].virq,
 				       vmd_irq, IRQF_NO_THREAD,
 				       vmd->name, &vmd->irqs[i]);
 		if (err)
Acked-by on v1 1/1
#2 lore raw
In reply to Thomas Gleixner
Quoted 25 lines · Thomas Gleixner

Tejas reported the following recursive locking issue:

swapper/0/1 is trying to acquire lock:
ffff8881074fd0a0 (&md->mutex){+.+.}-{3:3}, at: msi_get_virq+0x30/0xc0

but task is already holding lock:
ffff8881017cd6a0 (&md->mutex){+.+.}-{3:3}, at: __pci_enable_msi_range+0xf2/0x290

 stack backtrace:
  __mutex_lock+0x9d/0x920
  msi_get_virq+0x30/0xc0
  pci_irq_vector+0x26/0x30
  vmd_msi_init+0xcc/0x210
  msi_domain_alloc+0xbf/0x150
  msi_domain_alloc_irqs_descs_locked+0x3e/0xb0
  __pci_enable_msi_range+0x155/0x290
  pci_alloc_irq_vectors_affinity+0xba/0x100
  pcie_port_device_register+0x307/0x550
  pcie_portdrv_probe+0x3c/0xd0
  pci_device_probe+0x95/0x110

This is caused by the VMD MSI code which does a lookup of the Linux interrupt number for an VMD managed MSI[X] vector. The lookup function tries to acquire the already held mutex.

Quoted 6 lines

Avoid that by caching the Linux interrupt number at initialization time instead of looking it up over and over.

Fixes 82ff8e6b78fc ("PCI/MSI: Use msi_get_virq() in pci_get_vector()")

Bjorn, this is a fix for a patch we merged in the last cycle, if possible we should be sending it before v5.17 is released, please.

Thanks,
Lorenzo

Quoted 67 lines
 drivers/pci/controller/vmd.c |   14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)
drivers/pci/controller/vmd.c +7 -7
@@ -99,11 +99,13 @@ struct vmd_irq {
  * @srcu:	SRCU struct for local synchronization.
  * @count:	number of child IRQs assigned to this vector; used to track
  *		sharing.
+ * @virq:	The underlying VMD Linux interrupt number
  */
 struct vmd_irq_list {
 	struct list_head	irq_list;
 	struct srcu_struct	srcu;
 	unsigned int		count;
+	unsigned int		virq;
 };
 
 struct vmd_dev {
@@ -253,7 +255,6 @@ static int vmd_msi_init(struct irq_domai
 	struct msi_desc *desc = arg->desc;
 	struct vmd_dev *vmd = vmd_from_bus(msi_desc_to_pci_dev(desc)->bus);
 	struct vmd_irq *vmdirq = kzalloc(sizeof(*vmdirq), GFP_KERNEL);
-	unsigned int index, vector;
 
 	if (!vmdirq)
 		return -ENOMEM;
@@ -261,10 +262,8 @@ static int vmd_msi_init(struct irq_domai
 	INIT_LIST_HEAD(&vmdirq->node);
 	vmdirq->irq = vmd_next_irq(vmd, desc);
 	vmdirq->virq = virq;
-	index = index_from_irqs(vmd, vmdirq->irq);
-	vector = pci_irq_vector(vmd->dev, index);
 
-	irq_domain_set_info(domain, virq, vector, info->chip, vmdirq,
+	irq_domain_set_info(domain, virq, vmdirq->irq->virq, info->chip, vmdirq,
 			    handle_untracked_irq, vmd, NULL);
 	return 0;
 }
@@ -685,7 +684,8 @@ static int vmd_alloc_irqs(struct vmd_dev
 			return err;
 
 		INIT_LIST_HEAD(&vmd->irqs[i].irq_list);
-		err = devm_request_irq(&dev->dev, pci_irq_vector(dev, i),
+		vmd->irqs[i].virq = pci_irq_vector(dev, i);
+		err = devm_request_irq(&dev->dev, vmd->irqs[i].virq,
 				       vmd_irq, IRQF_NO_THREAD,
 				       vmd->name, &vmd->irqs[i]);
 		if (err)
@@ -969,7 +969,7 @@ static int vmd_suspend(struct device *de
 	int i;
 
 	for (i = 0; i < vmd->msix_count; i++)
-		devm_free_irq(dev, pci_irq_vector(pdev, i), &vmd->irqs[i]);
+		devm_free_irq(dev, vmd->irqs[i].virq, &vmd->irqs[i]);
 
 	return 0;
 }
@@ -981,7 +981,7 @@ static int vmd_resume(struct device *dev
 	int err, i;
 
 	for (i = 0; i < vmd->msix_count; i++) {
-		err = devm_request_irq(dev, pci_irq_vector(pdev, i),
+		err = devm_request_irq(dev, vmd->irqs[i].virq,
 				       vmd_irq, IRQF_NO_THREAD,
 				       vmd->name, &vmd->irqs[i]);
 		if (err)
#3 lore raw
Quoted 33 lines · Lorenzo Pieralisi
Quoted 25 lines · Thomas Gleixner

Tejas reported the following recursive locking issue:

swapper/0/1 is trying to acquire lock:
ffff8881074fd0a0 (&md->mutex){+.+.}-{3:3}, at: msi_get_virq+0x30/0xc0

but task is already holding lock:
ffff8881017cd6a0 (&md->mutex){+.+.}-{3:3}, at: __pci_enable_msi_range+0xf2/0x290

 stack backtrace:
  __mutex_lock+0x9d/0x920
  msi_get_virq+0x30/0xc0
  pci_irq_vector+0x26/0x30
  vmd_msi_init+0xcc/0x210
  msi_domain_alloc+0xbf/0x150
  msi_domain_alloc_irqs_descs_locked+0x3e/0xb0
  __pci_enable_msi_range+0x155/0x290
  pci_alloc_irq_vectors_affinity+0xba/0x100
  pcie_port_device_register+0x307/0x550
  pcie_portdrv_probe+0x3c/0xd0
  pci_device_probe+0x95/0x110

This is caused by the VMD MSI code which does a lookup of the Linux interrupt number for an VMD managed MSI[X] vector. The lookup function tries to acquire the already held mutex.

Quoted 6 lines

Avoid that by caching the Linux interrupt number at initialization time instead of looking it up over and over.

Fixes 82ff8e6b78fc ("PCI/MSI: Use msi_get_virq() in pci_get_vector()")
Quoted 5 lines

Bjorn, this is a fix for a patch we merged in the last cycle, if possible we should be sending it before v5.17 is released, please.

Agreed. I think Thomas merged 82ff8e6b78fc, and it looks like he's just merged this fix to irq/urgent of tip, so I think this should be already taken care of:

https://lore.kernel.org/r/164542867635.16921.13795049956787158926.tip-bot2@tip-bot2

Quoted 67 lines
Quoted 67 lines
 drivers/pci/controller/vmd.c |   14 +++++++-------
 1 file changed, 7 insertions(+), 7 deletions(-)
drivers/pci/controller/vmd.c +7 -7
@@ -99,11 +99,13 @@ struct vmd_irq {
  * @srcu:	SRCU struct for local synchronization.
  * @count:	number of child IRQs assigned to this vector; used to track
  *		sharing.
+ * @virq:	The underlying VMD Linux interrupt number
  */
 struct vmd_irq_list {
 	struct list_head	irq_list;
 	struct srcu_struct	srcu;
 	unsigned int		count;
+	unsigned int		virq;
 };
 
 struct vmd_dev {
@@ -253,7 +255,6 @@ static int vmd_msi_init(struct irq_domai
 	struct msi_desc *desc = arg->desc;
 	struct vmd_dev *vmd = vmd_from_bus(msi_desc_to_pci_dev(desc)->bus);
 	struct vmd_irq *vmdirq = kzalloc(sizeof(*vmdirq), GFP_KERNEL);
-	unsigned int index, vector;
 
 	if (!vmdirq)
 		return -ENOMEM;
@@ -261,10 +262,8 @@ static int vmd_msi_init(struct irq_domai
 	INIT_LIST_HEAD(&vmdirq->node);
 	vmdirq->irq = vmd_next_irq(vmd, desc);
 	vmdirq->virq = virq;
-	index = index_from_irqs(vmd, vmdirq->irq);
-	vector = pci_irq_vector(vmd->dev, index);
 
-	irq_domain_set_info(domain, virq, vector, info->chip, vmdirq,
+	irq_domain_set_info(domain, virq, vmdirq->irq->virq, info->chip, vmdirq,
 			    handle_untracked_irq, vmd, NULL);
 	return 0;
 }
@@ -685,7 +684,8 @@ static int vmd_alloc_irqs(struct vmd_dev
 			return err;
 
 		INIT_LIST_HEAD(&vmd->irqs[i].irq_list);
-		err = devm_request_irq(&dev->dev, pci_irq_vector(dev, i),
+		vmd->irqs[i].virq = pci_irq_vector(dev, i);
+		err = devm_request_irq(&dev->dev, vmd->irqs[i].virq,
 				       vmd_irq, IRQF_NO_THREAD,
 				       vmd->name, &vmd->irqs[i]);
 		if (err)
@@ -969,7 +969,7 @@ static int vmd_suspend(struct device *de
 	int i;
 
 	for (i = 0; i < vmd->msix_count; i++)
-		devm_free_irq(dev, pci_irq_vector(pdev, i), &vmd->irqs[i]);
+		devm_free_irq(dev, vmd->irqs[i].virq, &vmd->irqs[i]);
 
 	return 0;
 }
@@ -981,7 +981,7 @@ static int vmd_resume(struct device *dev
 	int err, i;
 
 	for (i = 0; i < vmd->msix_count; i++) {
-		err = devm_request_irq(dev, pci_irq_vector(pdev, i),
+		err = devm_request_irq(dev, vmd->irqs[i].virq,
 				       vmd_irq, IRQF_NO_THREAD,
 				       vmd->name, &vmd->irqs[i]);
 		if (err)
reviewer Reviewed-by on v1 1/1
#4 lore raw
In reply to Thomas Gleixner

This is how it used to be before I removed the member to better cache align the vmd_irq_list struct, but then the scru_struct grew to negate that benefit.

So this is good

Quoted 100 lines · Thomas Gleixner

Tejas reported the following recursive locking issue:

swapper/0/1 is trying to acquire lock:
ffff8881074fd0a0 (&md->mutex){+.+.}-{3:3}, at: msi_get_virq+0x30/0xc0

but task is already holding lock:
ffff8881017cd6a0 (&md->mutex){+.+.}-{3:3}, at: __pci_enable_msi_range+0xf2/0x290

  stack backtrace:
   __mutex_lock+0x9d/0x920
   msi_get_virq+0x30/0xc0
   pci_irq_vector+0x26/0x30
   vmd_msi_init+0xcc/0x210
   msi_domain_alloc+0xbf/0x150
   msi_domain_alloc_irqs_descs_locked+0x3e/0xb0
   __pci_enable_msi_range+0x155/0x290
   pci_alloc_irq_vectors_affinity+0xba/0x100
   pcie_port_device_register+0x307/0x550
   pcie_portdrv_probe+0x3c/0xd0
   pci_device_probe+0x95/0x110

This is caused by the VMD MSI code which does a lookup of the Linux interrupt number for an VMD managed MSI[X] vector. The lookup function tries to acquire the already held mutex.

Avoid that by caching the Linux interrupt number at initialization time instead of looking it up over and over.

Fixes 82ff8e6b78fc ("PCI/MSI: Use msi_get_virq() in pci_get_vector()")
---
  drivers/pci/controller/vmd.c |   14 +++++++-------
  1 file changed, 7 insertions(+), 7 deletions(-)
drivers/pci/controller/vmd.c +7 -7
@@ -99,11 +99,13 @@ struct vmd_irq {
   * @srcu:	SRCU struct for local synchronization.
   * @count:	number of child IRQs assigned to this vector; used to track
   *		sharing.
+ * @virq:	The underlying VMD Linux interrupt number
   */
  struct vmd_irq_list {
  	struct list_head	irq_list;
  	struct srcu_struct	srcu;
  	unsigned int		count;
+	unsigned int		virq;
  };
 
  struct vmd_dev {
@@ -253,7 +255,6 @@ static int vmd_msi_init(struct irq_domai
  	struct msi_desc *desc = arg->desc;
  	struct vmd_dev *vmd = vmd_from_bus(msi_desc_to_pci_dev(desc)->bus);
  	struct vmd_irq *vmdirq = kzalloc(sizeof(*vmdirq), GFP_KERNEL);
-	unsigned int index, vector;
 
  	if (!vmdirq)
  		return -ENOMEM;
@@ -261,10 +262,8 @@ static int vmd_msi_init(struct irq_domai
  	INIT_LIST_HEAD(&vmdirq->node);
  	vmdirq->irq = vmd_next_irq(vmd, desc);
  	vmdirq->virq = virq;
-	index = index_from_irqs(vmd, vmdirq->irq);
-	vector = pci_irq_vector(vmd->dev, index);
 
-	irq_domain_set_info(domain, virq, vector, info->chip, vmdirq,
+	irq_domain_set_info(domain, virq, vmdirq->irq->virq, info->chip, vmdirq,
  			    handle_untracked_irq, vmd, NULL);
  	return 0;
  }
@@ -685,7 +684,8 @@ static int vmd_alloc_irqs(struct vmd_dev
  			return err;
 
  		INIT_LIST_HEAD(&vmd->irqs[i].irq_list);
-		err = devm_request_irq(&dev->dev, pci_irq_vector(dev, i),
+		vmd->irqs[i].virq = pci_irq_vector(dev, i);
+		err = devm_request_irq(&dev->dev, vmd->irqs[i].virq,
  				       vmd_irq, IRQF_NO_THREAD,
  				       vmd->name, &vmd->irqs[i]);
  		if (err)
@@ -969,7 +969,7 @@ static int vmd_suspend(struct device *de
  	int i;
 
  	for (i = 0; i < vmd->msix_count; i++)
-		devm_free_irq(dev, pci_irq_vector(pdev, i), &vmd->irqs[i]);
+		devm_free_irq(dev, vmd->irqs[i].virq, &vmd->irqs[i]);
 
  	return 0;
  }
@@ -981,7 +981,7 @@ static int vmd_resume(struct device *dev
  	int err, i;
 
  	for (i = 0; i < vmd->msix_count; i++) {
-		err = devm_request_irq(dev, pci_irq_vector(pdev, i),
+		err = devm_request_irq(dev, vmd->irqs[i].virq,
  				       vmd_irq, IRQF_NO_THREAD,
  				       vmd->name, &vmd->irqs[i]);
  		if (err)