Skip to contentSkip to conversation

vfio: fix deadlock between group lock and kvm lock

PATCH Superseded Matthew Rosato 5 participants 11 messages Jan 31, 2023 – Feb 2, 2023

Subsystems (3)

v1 of 4 Superseded Latest is v4 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 vfio: fix deadlock between group lock and kvm lock

After 51cdc8bc120e, we have another deadlock scenario between the kvm->lock and the vfio group_lock with two different codepaths acquiring the locks in different order. Specifically in vfio_open_device, vfio holds the vfio group_lock when issuing device->ops->open_device but some drivers (like vfio-ap) need to acquire kvm->lock during their open_device routine; Meanwhile, kvm_vfio_release will acquire the kvm->lock first before calling vfio_file_set_kvm which will acquire the vfio group_lock.

To resolve this, let's remove the need for the vfio group_lock from the kvm_vfio_release codepath. This is done by introducing a new spinlock to protect modifications to the vfio group kvm pointer, and acquiring a kvm ref from within vfio while holding this spinlock, with the reference held until the last close for the device in question.

Fixes 51cdc8bc120e ("kvm/vfio: Fix potential deadlock on vfio group_lock")
---
 drivers/vfio/group.c     | 14 +++----
 drivers/vfio/vfio.h      |  5 ++-
 drivers/vfio/vfio_main.c | 88 ++++++++++++++++++++++++++++++++++++----
 include/linux/vfio.h     |  2 +-
 4 files changed, 88 insertions(+), 21 deletions(-)
drivers/vfio/group.c +5 -9
index bb24b2f0271e..8f67d33e0e8d 100644
@@ -164,13 +164,7 @@ static int vfio_device_group_open(struct vfio_device *device)
 		goto out_unlock;
 	}
 
-	/*
-	 * Here we pass the KVM pointer with the group under the lock.  If the
-	 * device driver will use it, it must obtain a reference and release it
-	 * during close_device.
-	 */
-	ret = vfio_device_open(device, device->group->iommufd,
-			       device->group->kvm);
+	ret = vfio_device_open(device, device->group);
 
 out_unlock:
 	mutex_unlock(&device->group->group_lock);
@@ -450,6 +444,7 @@ static struct vfio_group *vfio_group_alloc(struct iommu_group *iommu_group,
 
 	refcount_set(&group->drivers, 1);
 	mutex_init(&group->group_lock);
+	spin_lock_init(&group->kvm_ref_lock);
 	INIT_LIST_HEAD(&group->device_list);
 	mutex_init(&group->device_lock);
 	group->iommu_group = iommu_group;
@@ -799,13 +794,14 @@ EXPORT_SYMBOL_GPL(vfio_file_enforced_coherent);
 void vfio_file_set_kvm(struct file *file, struct kvm *kvm)
 {
 	struct vfio_group *group = file->private_data;
+	unsigned long flags;
 
 	if (!vfio_file_is_group(file))
 		return;
 
-	mutex_lock(&group->group_lock);
+	spin_lock_irqsave(&group->kvm_ref_lock, flags);
 	group->kvm = kvm;
-	mutex_unlock(&group->group_lock);
+	spin_unlock_irqrestore(&group->kvm_ref_lock, flags);
 }
 EXPORT_SYMBOL_GPL(vfio_file_set_kvm);
 
drivers/vfio/vfio.h +3 -2
index f8219a438bfb..d153ad35de24 100644
@@ -15,11 +15,11 @@ struct iommufd_ctx;
 struct iommu_group;
 struct vfio_device;
 struct vfio_container;
+struct vfio_group;
 
 void vfio_device_put_registration(struct vfio_device *device);
 bool vfio_device_try_get_registration(struct vfio_device *device);
-int vfio_device_open(struct vfio_device *device,
-		     struct iommufd_ctx *iommufd, struct kvm *kvm);
+int vfio_device_open(struct vfio_device *device, struct vfio_group *group);
 void vfio_device_close(struct vfio_device *device,
 		       struct iommufd_ctx *iommufd);
 
@@ -74,6 +74,7 @@ struct vfio_group {
 	struct file			*opened_file;
 	struct blocking_notifier_head	notifier;
 	struct iommufd_ctx		*iommufd;
+	spinlock_t			kvm_ref_lock;
 };
 
 int vfio_device_set_group(struct vfio_device *device,
drivers/vfio/vfio_main.c +79 -9

132 lines not on this page · raw message

include/linux/vfio.h +1 -1
index 35be78e9ae57..87ff862ff555 100644
@@ -46,7 +46,6 @@ struct vfio_device {
 	struct vfio_device_set *dev_set;
 	struct list_head dev_set_list;
 	unsigned int migration_flags;
-	/* Driver must reference the kvm during open_device or never touch it */
 	struct kvm *kvm;
 
 	/* Members below here are private, not for driver use */
@@ -58,6 +57,7 @@ struct vfio_device {
 	struct list_head group_next;
 	struct list_head iommu_entry;
 	struct iommufd_access *iommufd_access;
+	void (*put_kvm)(struct kvm *kvm);
 #if IS_ENABLED(CONFIG_IOMMUFD)
 	struct iommufd_device *iommufd_device;
 	struct iommufd_ctx *iommufd_ictx;
Signature · 1 line
2.39.1
#1 lore raw

[Intel-gfx] ✗ Fi.CI.CHECKPATCH: warning for vfio: fix deadlock between group lock and kvm lock

== Series Details ==

== Series Details ==

Series: vfio: fix deadlock between group lock and kvm lock
URL : https://patchwork.freedesktop.org/series/113535/
State : warning

== Summary ==

Error: dim checkpatch failed
22a38a2690cd vfio: fix deadlock between group lock and kvm lock -:91: CHECK:UNCOMMENTED_DEFINITION: spinlock_t definition without comment #91: FILE: drivers/vfio/vfio.h:77:
+ spinlock_t kvm_ref_lock;

total: 0 errors, 0 warnings, 1 checks, 203 lines checked

Reviewed-by on v1 1/1
#2 lore raw
In reply to Matthew Rosato
Quoted 4 lines · Matthew Rosato
drivers/vfio/group.c +0 -0
@@ -799,13 +794,14 @@ EXPORT_SYMBOL_GPL(vfio_file_enforced_coherent);
 void vfio_file_set_kvm(struct file *file, struct kvm *kvm)
 {
 	struct vfio_group *group = file->private_data;

drivers/vfio/group.c EXPORT_SYMBOL_GPL(vfio_file_enforced_coherent); · line 804

Quoted 10 lines
drivers/vfio/group.c +3 -2
+	unsigned long flags;
 
 	if (!vfio_file_is_group(file))
 		return;
 
-	mutex_lock(&group->group_lock);
+	spin_lock_irqsave(&group->kvm_ref_lock, flags);
 	group->kvm = kvm;
-	mutex_unlock(&group->group_lock);
+	spin_unlock_irqrestore(&group->kvm_ref_lock, flags);

We know we are in a sleeping context here so these are just 'spin_lock()', same with the other one

Otherwise it seems Ok to me

Jason

#3 lore raw
Quoted 14 lines

From: Jason Gunthorpe <[email protected]>
Sent: Wednesday, February 1, 2023 4:26 AM

Review

Quoted 1 line · Matthew Rosato
(quoted diff) +0 -0
@@ -799,13 +794,14 @@

EXPORT_SYMBOL_GPL(vfio_file_enforced_coherent);

Quoted 8 lines
drivers/vfio/group.c +1 -0
 void vfio_file_set_kvm(struct file *file, struct kvm *kvm)
 {
 	struct vfio_group *group = file->private_data;
+	unsigned long flags;
 
 	if (!vfio_file_is_group(file))
 		return;
 
Quoted 8 lines

drivers/vfio/group.c EXPORT_SYMBOL_GPL(vfio_file_enforced_coherent); · line 804

Quoted 5 lines
drivers/vfio/group.c +2 -2
-	mutex_lock(&group->group_lock);
+	spin_lock_irqsave(&group->kvm_ref_lock, flags);
 	group->kvm = kvm;
-	mutex_unlock(&group->group_lock);
+	spin_unlock_irqrestore(&group->kvm_ref_lock, flags);

We know we are in a sleeping context here so these are just 'spin_lock()', same with the other one

a dumb question. Why spinlock is required here? 😊

Regards,
Yi Liu

#4 lore raw
Quoted 23 lines · Yi L
Quoted 14 lines

From: Jason Gunthorpe <[email protected]>
Sent: Wednesday, February 1, 2023 4:26 AM

Review

Quoted 1 line · Matthew Rosato
(quoted diff) +0 -0
@@ -799,13 +794,14 @@

EXPORT_SYMBOL_GPL(vfio_file_enforced_coherent);

Quoted 8 lines
drivers/vfio/group.c +1 -0
 void vfio_file_set_kvm(struct file *file, struct kvm *kvm)
 {
 	struct vfio_group *group = file->private_data;
+	unsigned long flags;
 
 	if (!vfio_file_is_group(file))
 		return;
 
Quoted 8 lines

drivers/vfio/group.c EXPORT_SYMBOL_GPL(vfio_file_enforced_coherent); · line 804

Quoted 5 lines
drivers/vfio/group.c +2 -2
-	mutex_lock(&group->group_lock);
+	spin_lock_irqsave(&group->kvm_ref_lock, flags);
 	group->kvm = kvm;
-	mutex_unlock(&group->group_lock);
+	spin_unlock_irqrestore(&group->kvm_ref_lock, flags);

We know we are in a sleeping context here so these are just 'spin_lock()', same with the other one

Quoted 2 lines

a dumb question. Why spinlock is required here? 😊

You mean as opposed to another mutex? I don't think it's required per se (we are replacing a mutex so we could have again used another mutex here), but all current users of this new lock hold it over a very short window (e.g. set a pointer as above, or refcount++ and copy the pointer as in the first device_open)

#5 lore raw
Quoted 27 lines

From: Matthew Rosato <[email protected]>
Sent: Wednesday, February 1, 2023 10:43 PM

Quoted 23 lines · Yi L
Quoted 22 lines

From: Jason Gunthorpe <[email protected]>
Sent: Wednesday, February 1, 2023 4:26 AM

Review

Quoted 1 line · Matthew Rosato
(quoted diff) +0 -0
@@ -799,13 +794,14 @@

EXPORT_SYMBOL_GPL(vfio_file_enforced_coherent);

drivers/vfio/group.c EXPORT_SYMBOL_GPL(vfio_file_enforced_coherent); · line 804

Quoted 13 lines
drivers/vfio/group.c +3 -2
 void vfio_file_set_kvm(struct file *file, struct kvm *kvm)
 {
 	struct vfio_group *group = file->private_data;
+	unsigned long flags;
 
 	if (!vfio_file_is_group(file))
 		return;
 
-	mutex_lock(&group->group_lock);
+	spin_lock_irqsave(&group->kvm_ref_lock, flags);
 	group->kvm = kvm;
-	mutex_unlock(&group->group_lock);
+	spin_unlock_irqrestore(&group->kvm_ref_lock, flags);

We know we are in a sleeping context here so these are just 'spin_lock()', same with the other one

Quoted 8 lines
Quoted 2 lines

a dumb question. Why spinlock is required here? 😊

You mean as opposed to another mutex? I don't think it's required per se (we are replacing a mutex so we could have again used another mutex here), but all current users of this new lock hold it over a very short window (e.g. set a pointer as above, or refcount++ and copy the pointer as in the first device_open)

I see. Just not sure if spinlock is required for a special reason.

Regards,
Yi Liu

#6 lore raw

[Intel-gfx] ✗ Fi.CI.BAT: failure for vfio: fix deadlock between group lock and kvm lock

In reply to Matthew Rosato
== Series Details ==

== Series Details ==

Series: vfio: fix deadlock between group lock and kvm lock
URL : https://patchwork.freedesktop.org/series/113535/
State : failure

== Summary ==

CI Bug Log - changes from CI_DRM_12674 -> Patchwork_113535v1
====================================================

Summary
-------

**FAILURE**

Serious unknown changes coming with Patchwork_113535v1 absolutely need to be verified manually.

If you think the reported changes have nothing to do with the changes introduced in Patchwork_113535v1, please notify your bug team to allow them to document this new failure mode, which will reduce false positives in CI.

External URL: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_113535v1/index.html

Participating hosts (27 -> 26)
------------------------------

Additional (1): fi-kbl-soraka
Missing (2): bat-atsm-1 fi-snb-2520m

Possible new issues
-------------------

Here are the unknown changes that may have been introduced in Patchwork_113535v1:

### IGT changes ###

#### Possible regressions ####

* igt@i915_selftest@live@execlists:
- fi-bsw-n3050: [PASS][1] -> [ABORT][2]
[1]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12674/fi-bsw-n3050/igt@i915_selftest@[email protected]
[2]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_113535v1/fi-bsw-n3050/igt@i915_selftest@[email protected]

#### Suppressed ####

The following results come from untrusted machines, tests, or statuses. They do not affect the overall result.

* igt@i915_selftest@live@slpc:
- {bat-rpls-1}: NOTRUN -> [DMESG-FAIL][3]
[3]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_113535v1/bat-rpls-1/igt@i915_selftest@[email protected]

Known issues
------------

Here are the changes found in Patchwork_113535v1 that come from known issues:

### IGT changes ###

#### Issues hit ####

* igt@fbdev@write:
- fi-blb-e6850: [PASS][4] -> [SKIP][5] ([fdo#109271]) +4 similar issues
[4]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12674/fi-blb-e6850/igt@[email protected]
[5]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_113535v1/fi-blb-e6850/igt@[email protected]

* igt@gem_huc_copy@huc-copy:
- fi-kbl-soraka: NOTRUN -> [SKIP][6] ([fdo#109271] / [i915#2190])
[6]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_113535v1/fi-kbl-soraka/igt@[email protected]

* igt@gem_lmem_swapping@basic:
- fi-kbl-soraka: NOTRUN -> [SKIP][7] ([fdo#109271] / [i915#4613]) +3 similar issues
[7]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_113535v1/fi-kbl-soraka/igt@[email protected]

* igt@i915_selftest@live@execlists:
- fi-kbl-soraka: NOTRUN -> [INCOMPLETE][8] ([i915#7156])
[8]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_113535v1/fi-kbl-soraka/igt@i915_selftest@[email protected]
- fi-bsw-nick: [PASS][9] -> [ABORT][10] ([i915#7911])
[9]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12674/fi-bsw-nick/igt@i915_selftest@[email protected]
[10]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_113535v1/fi-bsw-nick/igt@i915_selftest@[email protected]

* igt@i915_selftest@live@gt_heartbeat:
- fi-kbl-soraka: NOTRUN -> [DMESG-FAIL][11] ([i915#5334] / [i915#7872])
[11]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_113535v1/fi-kbl-soraka/igt@i915_selftest@live@gt_heartbeat.html
- fi-kbl-7567u: [PASS][12] -> [DMESG-FAIL][13] ([i915#5334] / [i915#7872])
[12]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12674/fi-kbl-7567u/igt@i915_selftest@live@gt_heartbeat.html
[13]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_113535v1/fi-kbl-7567u/igt@i915_selftest@live@gt_heartbeat.html

* igt@i915_selftest@live@gt_pm:
- fi-kbl-soraka: NOTRUN -> [DMESG-FAIL][14] ([i915#1886])
[14]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_113535v1/fi-kbl-soraka/igt@i915_selftest@live@gt_pm.html

* igt@kms_chamelium_frames@hdmi-crc-fast:
- fi-kbl-soraka: NOTRUN -> [SKIP][15] ([fdo#109271]) +15 similar issues
[15]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_113535v1/fi-kbl-soraka/igt@[email protected]

#### Possible fixes ####

* igt@i915_selftest@live@hangcheck:
- {bat-dg2-11}: [ABORT][16] ([i915#7913]) -> [PASS][17]
[16]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12674/bat-dg2-11/igt@i915_selftest@[email protected]
[17]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_113535v1/bat-dg2-11/igt@i915_selftest@[email protected]

* igt@i915_selftest@live@requests:
- {bat-rpls-1}: [ABORT][18] ([i915#4983] / [i915#7981]) -> [PASS][19]
[18]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12674/bat-rpls-1/igt@i915_selftest@[email protected]
[19]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_113535v1/bat-rpls-1/igt@i915_selftest@[email protected]

* igt@kms_cursor_legacy@basic-busy-flip-before-cursor@atomic-transitions:
- fi-bsw-n3050: [FAIL][20] ([i915#6298]) -> [PASS][21]
[20]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_12674/fi-bsw-n3050/igt@kms_cursor_legacy@[email protected]
[21]: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_113535v1/fi-bsw-n3050/igt@kms_cursor_legacy@[email protected]

{name}: This element is suppressed. This means it is ignored when computing
the status of the difference (SUCCESS, WARNING, or FAILURE).

[fdo#109271]: https://bugs.freedesktop.org/show_bug.cgi?id=109271
[i915#1886]: https://gitlab.freedesktop.org/drm/intel/issues/1886
[i915#2190]: https://gitlab.freedesktop.org/drm/intel/issues/2190
[i915#4613]: https://gitlab.freedesktop.org/drm/intel/issues/4613
[i915#4983]: https://gitlab.freedesktop.org/drm/intel/issues/4983
[i915#5334]: https://gitlab.freedesktop.org/drm/intel/issues/5334
[i915#6298]: https://gitlab.freedesktop.org/drm/intel/issues/6298
[i915#6311]: https://gitlab.freedesktop.org/drm/intel/issues/6311
[i915#7156]: https://gitlab.freedesktop.org/drm/intel/issues/7156
[i915#7359]: https://gitlab.freedesktop.org/drm/intel/issues/7359
[i915#7828]: https://gitlab.freedesktop.org/drm/intel/issues/7828
[i915#7872]: https://gitlab.freedesktop.org/drm/intel/issues/7872
[i915#7911]: https://gitlab.freedesktop.org/drm/intel/issues/7911
[i915#7913]: https://gitlab.freedesktop.org/drm/intel/issues/7913
[i915#7981]: https://gitlab.freedesktop.org/drm/intel/issues/7981

Build changes
-------------

* Linux: CI_DRM_12674 -> Patchwork_113535v1

CI-20190529: 20190529
CI_DRM_12674: abcd161e6541aaf1e5c23e16019d02905c0e50fd @ git://anongit.freedesktop.org/gfx-ci/linux
IGT_7143: c7b12dcc460fc2348e1fa7f4dcb791bb82e29e44 @ https://gitlab.freedesktop.org/drm/igt-gpu-tools.git
Patchwork_113535v1: abcd161e6541aaf1e5c23e16019d02905c0e50fd @ git://anongit.freedesktop.org/gfx-ci/linux

### Linux commits

852ae00fca95 vfio: fix deadlock between group lock and kvm lock

== Logs ==

For more details see: https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_113535v1/index.html

#7 lore raw
In reply to Matthew Rosato
Quoted 5 lines

From: Matthew Rosato <[email protected]>
Sent: Wednesday, February 1, 2023 4:07 AM

drivers/vfio/vfio_main.c +1 -1
-	device->kvm = kvm;
+	/*

drivers/vfio/vfio_main.c static int vfio_device_first_open(struct vfio_device *device, · line 427

Quoted 10 lines
drivers/vfio/vfio_main.c +10 -0
+	 * Get the KVM pointer currently associated with the group, if there
+	 * is one, and obtain a reference now that will be held until the
+	 * open_count reaches 0.  Save the pointer in the device for use by
+	 * drivers.
+	 */
+	spin_lock_irqsave(&group->kvm_ref_lock, flags);
+	if (group->kvm && vfio_kvm_get_kvm_safe(device, group->kvm))
+		device->kvm = group->kvm;
+	spin_unlock_irqrestore(&group->kvm_ref_lock, flags);
+

No group reference in vfio_main.c.

btw Yi, looks the deadlock issue also exists in your cdev work.

kvm_vfio_release() holds kvm lock and then try to acquire device->device_set->lock in vfio_device_file_set_kvm().

vfio_device_ioctl_bind_iommufd() holds device->device_set->lock and then call vfio_device_open() which finally hit kvm lock acquisition in driver's open_device routine (e.g. vfio-ap).

A similar fix is required in your series.

Thanks
Kevin

#8 lore raw
Quoted 17 lines · Kevin
Quoted 5 lines

From: Matthew Rosato <[email protected]>
Sent: Wednesday, February 1, 2023 4:07 AM

drivers/vfio/vfio_main.c +1 -1
-	device->kvm = kvm;
+	/*

drivers/vfio/vfio_main.c static int vfio_device_first_open(struct vfio_device *device, · line 427

Quoted 10 lines
drivers/vfio/vfio_main.c +10 -0
+	 * Get the KVM pointer currently associated with the group, if there
+	 * is one, and obtain a reference now that will be held until the
+	 * open_count reaches 0.  Save the pointer in the device for use by
+	 * drivers.
+	 */
+	spin_lock_irqsave(&group->kvm_ref_lock, flags);
+	if (group->kvm && vfio_kvm_get_kvm_safe(device, group->kvm))
+		device->kvm = group->kvm;
+	spin_unlock_irqrestore(&group->kvm_ref_lock, flags);
+

No group reference in vfio_main.c.

OK -- I think to do that I'll have to move the lock/unlock of the dev_set lock to group.c right before we call vfio_device_{open,close} and check open_count there to make decisions about the kvm ref (before calling vfio_device_open to decide to get kvm ref, after returning from vfio_device_open to see if we to drop ref on error, after close to see if we need to drop ref).

Quoted 14 lines

btw Yi, looks the deadlock issue also exists in your cdev work.

kvm_vfio_release() holds kvm lock and then try to acquire device->device_set->lock in vfio_device_file_set_kvm().

vfio_device_ioctl_bind_iommufd() holds device->device_set->lock and then call vfio_device_open() which finally hit kvm lock acquisition in driver's open_device routine (e.g. vfio-ap).

A similar fix is required in your series.

Thanks
Kevin

#9 lore raw
In reply to Matthew Rosato
Quoted 205 lines

From: Matthew Rosato <[email protected]>
Sent: Wednesday, February 1, 2023 4:07 AM

After 51cdc8bc120e, we have another deadlock scenario between the kvm->lock and the vfio group_lock with two different codepaths acquiring the locks in different order. Specifically in vfio_open_device, vfio holds the vfio group_lock when issuing device->ops->open_device but some
drivers (like vfio-ap) need to acquire kvm->lock during their open_device routine; Meanwhile, kvm_vfio_release will acquire the kvm->lock first before calling vfio_file_set_kvm which will acquire the vfio group_lock.

To resolve this, let's remove the need for the vfio group_lock from the kvm_vfio_release codepath. This is done by introducing a new spinlock to protect modifications to the vfio group kvm pointer, and acquiring a kvm ref from within vfio while holding this spinlock, with the reference held until the last close for the device in question.

Fixes 51cdc8bc120e ("kvm/vfio: Fix potential deadlock on vfio group_lock")
---
 drivers/vfio/group.c     | 14 +++----
 drivers/vfio/vfio.h      |  5 ++-
 drivers/vfio/vfio_main.c | 88
++++++++++++++++++++++++++++++++++++----
 include/linux/vfio.h     |  2 +-
 4 files changed, 88 insertions(+), 21 deletions(-)
drivers/vfio/group.c +0 -2
index bb24b2f0271e..8f67d33e0e8d 100644
@@ -164,13 +164,7 @@ static int vfio_device_group_open(struct
 vfio_device *device)
 		goto out_unlock;
 	}
 
-	/*
-	 * Here we pass the KVM pointer with the group under the lock.  If

the

drivers/vfio/group.c +2 -5
-	 * device driver will use it, it must obtain a reference and release it
-	 * during close_device.
-	 */
-	ret = vfio_device_open(device, device->group->iommufd,
-			       device->group->kvm);
+	ret = vfio_device_open(device, device->group);
 
 out_unlock:
 	mutex_unlock(&device->group->group_lock);
@@ -450,6 +444,7 @@ static struct vfio_group *vfio_group_alloc(struct
 iommu_group *iommu_group,
 
 	refcount_set(&group->drivers, 1);
 	mutex_init(&group->group_lock);
+	spin_lock_init(&group->kvm_ref_lock);
 	INIT_LIST_HEAD(&group->device_list);
 	mutex_init(&group->device_lock);
drivers/vfio/group.c +3 -2
 	group->iommu_group = iommu_group;
@@ -799,13 +794,14 @@
 EXPORT_SYMBOL_GPL(vfio_file_enforced_coherent);
 void vfio_file_set_kvm(struct file *file, struct kvm *kvm)
 {
 	struct vfio_group *group = file->private_data;
+	unsigned long flags;
 
 	if (!vfio_file_is_group(file))
 		return;
 
-	mutex_lock(&group->group_lock);
+	spin_lock_irqsave(&group->kvm_ref_lock, flags);
 	group->kvm = kvm;
-	mutex_unlock(&group->group_lock);
+	spin_unlock_irqrestore(&group->kvm_ref_lock, flags);
 }
 EXPORT_SYMBOL_GPL(vfio_file_set_kvm);
drivers/vfio/vfio.h +3 -2
index f8219a438bfb..d153ad35de24 100644
@@ -15,11 +15,11 @@ struct iommufd_ctx;
 struct iommu_group;
 struct vfio_device;
 struct vfio_container;
+struct vfio_group;
 
 void vfio_device_put_registration(struct vfio_device *device);
 bool vfio_device_try_get_registration(struct vfio_device *device);
-int vfio_device_open(struct vfio_device *device,
-		     struct iommufd_ctx *iommufd, struct kvm *kvm);
+int vfio_device_open(struct vfio_device *device, struct vfio_group
 *group);
 void vfio_device_close(struct vfio_device *device,
 		       struct iommufd_ctx *iommufd);
@@ -74,6 +74,7 @@ struct vfio_group {
 	struct file			*opened_file;
 	struct blocking_notifier_head	notifier;
 	struct iommufd_ctx		*iommufd;
+	spinlock_t			kvm_ref_lock;
 };
 
 int vfio_device_set_group(struct vfio_device *device,
drivers/vfio/vfio_main.c +65 -3
index 5177bb061b17..b0defb0d0d87 100644
@@ -16,6 +16,9 @@
 #include <linux/fs.h>
 #include <linux/idr.h>
 #include <linux/iommu.h>
+#ifdef CONFIG_HAVE_KVM
+#include <linux/kvm_host.h>
+#endif
 #include <linux/list.h>
 #include <linux/miscdevice.h>
 #include <linux/module.h>
@@ -344,9 +347,59 @@ static bool vfio_assert_device_open(struct
 vfio_device *device)
 	return !WARN_ON_ONCE(!READ_ONCE(device->open_count));
 }
 
+#ifdef CONFIG_HAVE_KVM
+static bool vfio_kvm_get_kvm_safe(struct vfio_device *device, struct kvm
 *kvm)
+{
+	void (*pfn)(struct kvm *kvm);
+	bool (*fn)(struct kvm *kvm);
+	bool ret;
+
+	pfn = symbol_get(kvm_put_kvm);
+	if (WARN_ON(!pfn))
+		return false;
+
+	fn = symbol_get(kvm_get_kvm_safe);
+	if (WARN_ON(!fn)) {
+		symbol_put(kvm_put_kvm);
+		return false;
+	}
+
+	ret = fn(kvm);
+	if (ret)
+		device->put_kvm = pfn;
+	else
+		symbol_put(kvm_put_kvm);
+
+	symbol_put(kvm_get_kvm_safe);
+
+	return ret;
+}
+
+static void vfio_kvm_put_kvm(struct vfio_device *device)
+{
+	if (WARN_ON(!device->kvm || !device->put_kvm))
+		return;
+
+	device->put_kvm(device->kvm);
+	symbol_put(kvm_put_kvm);
+}
+
+#else
+static bool vfio_kvm_get_kvm_safe(struct vfio_device *device, struct kvm
 *kvm)
+{
+	return false;
+}
+
+static void vfio_kvm_put_kvm(struct vfio_device *device)
+{
+}
+#endif
+
 static int vfio_device_first_open(struct vfio_device *device,
-				  struct iommufd_ctx *iommufd, struct kvm
 *kvm)
+				  struct vfio_group *group)
 {
+	struct iommufd_ctx *iommufd = group->iommufd;
+	unsigned long flags;
 	int ret;
 
 	lockdep_assert_held(&device->dev_set->lock);
@@ -361,16 +414,30 @@ static int vfio_device_first_open(struct
 vfio_device *device,
 	if (ret)
 		goto err_module_put;
 
-	device->kvm = kvm;
+	/*
+	 * Get the KVM pointer currently associated with the group, if there
+	 * is one, and obtain a reference now that will be held until the
+	 * open_count reaches 0.  Save the pointer in the device for use by
+	 * drivers.
+	 */
+	spin_lock_irqsave(&group->kvm_ref_lock, flags);
+	if (group->kvm && vfio_kvm_get_kvm_safe(device, group->kvm))
+		device->kvm = group->kvm;
+	spin_unlock_irqrestore(&group->kvm_ref_lock, flags);
+
 	if (device->ops->open_device) {
 		ret = device->ops->open_device(device);
 		if (ret)
-			goto err_unuse_iommu;

drivers/vfio/vfio_main.c static int vfio_device_first_open(struct · line 439

Quoted 10 lines
drivers/vfio/vfio_main.c +5 -2
+			goto err_put_kvm;
 	}
 	return 0;
 
-err_unuse_iommu:
-	device->kvm = NULL;
+err_put_kvm:
+	if (device->kvm) {
+		vfio_kvm_put_kvm(device);
+		device->put_kvm = NULL;

Can "device->put_kvm = NULL;" be part of vfio_kvm_put_kvm()?

Regards,
Yi Liu

#10 lore raw
Quoted 1 line · Yi L
Quoted 1 line

From: Matthew Rosato <[email protected]>

...

Quoted 5 lines

drivers/vfio/vfio_main.c static int vfio_device_first_open(struct vfio_device *device, · line 438

Quoted 3 lines
drivers/vfio/vfio_main.c +3 -0
+	if (device->kvm) {
+		vfio_kvm_put_kvm(device);
+		device->put_kvm = NULL;

Can "device->put_kvm = NULL;" be part of vfio_kvm_put_kvm()?

Sure