The mdev remove callback for the vfio_ap device driver bails out with
-EBUSY if the mdev is in use by a KVM guest. The intended purpose was
to prevent the mdev from being removed while in use; however, returning a
non-zero rc does not prevent removal of the mdev. Consequently, the memory
for the resources allocated when the mdev was created are leaked.
To fix this issue:
* The remove callback will not terminate with -EBUSY when the mdev is in
use by a KVM guest.
* The memory for the resources allocated when the mdev was created will
be freed.
* Since the struct ap_matrix_mdev now gets freed while the guest is
is still running, we need to ensure that the pointer to the function
that handles interception of the PQAP instruction executed on the guest
is not accessed while it is being set to NULL. To prevent this, a r/w
lock is introduced that protects the function pointer.
Change log:
v5 -> v6:
--------
* Replaced struct kvm_s390_module_hook with function pointer
int (*crypto_hook)(struct kvm_vcpu *vcpu)
Tony Krowiak (2):
s390/vfio-ap: clean up mdev resources when remove callback invoked
s390/vfio-ap: r/w lock for PQAP interception handler function pointer
arch/s390/include/asm/kvm_host.h | 8 +++----
arch/s390/kvm/kvm-s390.c | 1 +
arch/s390/kvm/priv.c | 10 ++++----
drivers/s390/crypto/vfio_ap_ops.c | 33 ++++++++++++++-------------
drivers/s390/crypto/vfio_ap_private.h | 2 +-
5 files changed, 28 insertions(+), 26 deletions(-)
--
2.30.2
The function pointer to the interception handler for the PQAP instruction
can get changed during the interception process. Let's add a
semaphore to struct kvm_s390_crypto to control read/write access to the
function pointer contained therein.
The semaphore must be locked for write access by the vfio_ap device driver
when notified that the KVM pointer has been set or cleared. It must be
locked for read access by the interception framework when the PQAP
instruction is intercepted.
Signed-off-by: Tony Krowiak <[email protected]>
Reviewed-by: Jason Gunthorpe <[email protected]>
Cc: [email protected]
---
arch/s390/include/asm/kvm_host.h | 8 +++-----
arch/s390/kvm/kvm-s390.c | 1 +
arch/s390/kvm/priv.c | 10 ++++++----
drivers/s390/crypto/vfio_ap_ops.c | 23 +++++++++++++++++------
drivers/s390/crypto/vfio_ap_private.h | 2 +-
5 files changed, 28 insertions(+), 16 deletions(-)
diff --git a/arch/s390/include/asm/kvm_host.h b/arch/s390/include/asm/kvm_host.h
index 8925f3969478..36441669ffba 100644
--- a/arch/s390/include/asm/kvm_host.h
+++ b/arch/s390/include/asm/kvm_host.h
@@ -803,14 +803,12 @@ struct kvm_s390_cpu_model {
unsigned short ibc;
};
-struct kvm_s390_module_hook {
- int (*hook)(struct kvm_vcpu *vcpu);
- struct module *owner;
-};
+typedef int (*crypto_hook)(struct kvm_vcpu *vcpu);
struct kvm_s390_crypto {
struct kvm_s390_crypto_cb *crycb;
- struct kvm_s390_module_hook *pqap_hook;
+ struct rw_semaphore pqap_hook_rwsem;
+ crypto_hook *pqap_hook;
__u32 crycbd;
__u8 aes_kw;
__u8 dea_kw;
diff --git a/arch/s390/kvm/kvm-s390.c b/arch/s390/kvm/kvm-s390.c
index 1296fc10f80c..418d910df569 100644
--- a/arch/s390/kvm/kvm-s390.c
+++ b/arch/s390/kvm/kvm-s390.c
@@ -2606,6 +2606,7 @@ static void kvm_s390_crypto_init(struct kvm *kvm)
{
kvm->arch.crypto.crycb = &kvm->arch.sie_page2->crycb;
kvm_s390_set_crycb_format(kvm);
+ init_rwsem(&kvm->arch.crypto.pqap_hook_rwsem);
if (!test_kvm_facility(kvm, 76))
return;
diff --git a/arch/s390/kvm/priv.c b/arch/s390/kvm/priv.c
index 9928f785c677..6bed9406c1f3 100644
--- a/arch/s390/kvm/priv.c
+++ b/arch/s390/kvm/priv.c
@@ -610,6 +610,7 @@ static int handle_io_inst(struct kvm_vcpu *vcpu)
static int handle_pqap(struct kvm_vcpu *vcpu)
{
struct ap_queue_status status = {};
+ crypto_hook pqap_hook;
unsigned long reg0;
int ret;
uint8_t fc;
@@ -657,15 +658,16 @@ static int handle_pqap(struct kvm_vcpu *vcpu)
* Verify that the hook callback is registered, lock the owner
* and call the hook.
*/
+ down_read(&vcpu->kvm->arch.crypto.pqap_hook_rwsem);
if (vcpu->kvm->arch.crypto.pqap_hook) {
- if (!try_module_get(vcpu->kvm->arch.crypto.pqap_hook->owner))
- return -EOPNOTSUPP;
- ret = vcpu->kvm->arch.crypto.pqap_hook->hook(vcpu);
- module_put(vcpu->kvm->arch.crypto.pqap_hook->owner);
+ pqap_hook = *vcpu->kvm->arch.crypto.pqap_hook;
+ ret = pqap_hook(vcpu);
if (!ret && vcpu->run->s.regs.gprs[1] & 0x00ff0000)
kvm_s390_set_psw_cc(vcpu, 3);
+ up_read(&vcpu->kvm->arch.crypto.pqap_hook_rwsem);
return ret;
}
+ up_read(&vcpu->kvm->arch.crypto.pqap_hook_rwsem);
/*
* A vfio_driver must register a hook.
* No hook means no driver to enable the SIE CRYCB and no queues.
diff --git a/drivers/s390/crypto/vfio_ap_ops.c b/drivers/s390/crypto/vfio_ap_ops.c
index 122c85c22469..742277bc8d1c 100644
--- a/drivers/s390/crypto/vfio_ap_ops.c
+++ b/drivers/s390/crypto/vfio_ap_ops.c
@@ -352,8 +352,7 @@ static int vfio_ap_mdev_create(struct mdev_device *mdev)
vfio_ap_matrix_init(&matrix_dev->info, &matrix_mdev->matrix);
init_waitqueue_head(&matrix_mdev->wait_for_kvm);
mdev_set_drvdata(mdev, matrix_mdev);
- matrix_mdev->pqap_hook.hook = handle_pqap;
- matrix_mdev->pqap_hook.owner = THIS_MODULE;
+ matrix_mdev->pqap_hook = handle_pqap;
mutex_lock(&matrix_dev->lock);
list_add(&matrix_mdev->node, &matrix_dev->mdev_list);
mutex_unlock(&matrix_dev->lock);
@@ -1115,15 +1114,20 @@ static int vfio_ap_mdev_set_kvm(struct ap_matrix_mdev *matrix_mdev,
}
kvm_get_kvm(kvm);
+ matrix_mdev->kvm = kvm;
matrix_mdev->kvm_busy = true;
mutex_unlock(&matrix_dev->lock);
+
+ down_write(&matrix_mdev->kvm->arch.crypto.pqap_hook_rwsem);
+ kvm->arch.crypto.pqap_hook = &matrix_mdev->pqap_hook;
+ up_write(&matrix_mdev->kvm->arch.crypto.pqap_hook_rwsem);
+
kvm_arch_crypto_set_masks(kvm,
matrix_mdev->matrix.apm,
matrix_mdev->matrix.aqm,
matrix_mdev->matrix.adm);
+
mutex_lock(&matrix_dev->lock);
- kvm->arch.crypto.pqap_hook = &matrix_mdev->pqap_hook;
- matrix_mdev->kvm = kvm;
matrix_mdev->kvm_busy = false;
wake_up_all(&matrix_mdev->wait_for_kvm);
}
@@ -1189,10 +1193,17 @@ static void vfio_ap_mdev_unset_kvm(struct ap_matrix_mdev *matrix_mdev)
if (matrix_mdev->kvm) {
matrix_mdev->kvm_busy = true;
mutex_unlock(&matrix_dev->lock);
- kvm_arch_crypto_clear_masks(matrix_mdev->kvm);
+
+ if (matrix_mdev->kvm->arch.crypto.crycbd) {
+ down_write(&matrix_mdev->kvm->arch.crypto.pqap_hook_rwsem);
+ matrix_mdev->kvm->arch.crypto.pqap_hook = NULL;
+ up_write(&matrix_mdev->kvm->arch.crypto.pqap_hook_rwsem);
+
+ kvm_arch_crypto_clear_masks(matrix_mdev->kvm);
+ }
+
mutex_lock(&matrix_dev->lock);
vfio_ap_mdev_reset_queues(matrix_mdev->mdev);
- matrix_mdev->kvm->arch.crypto.pqap_hook = NULL;
kvm_put_kvm(matrix_mdev->kvm);
matrix_mdev->kvm = NULL;
matrix_mdev->kvm_busy = false;
diff --git a/drivers/s390/crypto/vfio_ap_private.h b/drivers/s390/crypto/vfio_ap_private.h
index f82a6396acae..e12218e5a629 100644
--- a/drivers/s390/crypto/vfio_ap_private.h
+++ b/drivers/s390/crypto/vfio_ap_private.h
@@ -86,7 +86,7 @@ struct ap_matrix_mdev {
bool kvm_busy;
wait_queue_head_t wait_for_kvm;
struct kvm *kvm;
- struct kvm_s390_module_hook pqap_hook;
+ crypto_hook pqap_hook;
struct mdev_device *mdev;
};
--
2.30.2
The mdev remove callback for the vfio_ap device driver bails out with
-EBUSY if the mdev is in use by a KVM guest (i.e., the KVM pointer in the
struct ap_matrix_mdev is not NULL). The intended purpose was
to prevent the mdev from being removed while in use. There are two
problems with this scenario:
1. Returning a non-zero return code from the remove callback does not
prevent the removal of the mdev.
2. The KVM pointer in the struct ap_matrix_mdev will always be NULL because
the remove callback will not get invoked until the mdev fd is closed.
When the mdev fd is closed, the mdev release callback is invoked and
clears the KVM pointer from the struct ap_matrix_mdev.
Let's go ahead and remove the check for KVM in the remove callback and
allow the cleanup of mdev resources to proceed.
Signed-off-by: Tony Krowiak <[email protected]>
Reviewed-by: Jason Gunthorpe <[email protected]>
Cc: [email protected]
---
drivers/s390/crypto/vfio_ap_ops.c | 10 ----------
1 file changed, 10 deletions(-)
diff --git a/drivers/s390/crypto/vfio_ap_ops.c b/drivers/s390/crypto/vfio_ap_ops.c
index b2c7e10dfdcd..122c85c22469 100644
--- a/drivers/s390/crypto/vfio_ap_ops.c
+++ b/drivers/s390/crypto/vfio_ap_ops.c
@@ -366,16 +366,6 @@ static int vfio_ap_mdev_remove(struct mdev_device *mdev)
struct ap_matrix_mdev *matrix_mdev = mdev_get_drvdata(mdev);
mutex_lock(&matrix_dev->lock);
-
- /*
- * If the KVM pointer is in flux or the guest is running, disallow
- * un-assignment of control domain.
- */
- if (matrix_mdev->kvm_busy || matrix_mdev->kvm) {
- mutex_unlock(&matrix_dev->lock);
- return -EBUSY;
- }
-
vfio_ap_mdev_reset_queues(mdev);
list_del(&matrix_mdev->node);
kfree(matrix_mdev);
--
2.30.2
On 21.06.21 17:57, Tony Krowiak wrote:
> The mdev remove callback for the vfio_ap device driver bails out with
> -EBUSY if the mdev is in use by a KVM guest (i.e., the KVM pointer in the
> struct ap_matrix_mdev is not NULL). The intended purpose was
> to prevent the mdev from being removed while in use. There are two
> problems with this scenario:
>
> 1. Returning a non-zero return code from the remove callback does not
> prevent the removal of the mdev.
>
> 2. The KVM pointer in the struct ap_matrix_mdev will always be NULL because
> the remove callback will not get invoked until the mdev fd is closed.
> When the mdev fd is closed, the mdev release callback is invoked and
> clears the KVM pointer from the struct ap_matrix_mdev.
>
> Let's go ahead and remove the check for KVM in the remove callback and
> allow the cleanup of mdev resources to proceed.
>
> Signed-off-by: Tony Krowiak <[email protected]>
> Reviewed-by: Jason Gunthorpe <[email protected]>
> Cc: [email protected]
This one is already queued on
https://git.kernel.org/pub/scm/linux/kernel/git/s390/linux.git/log/?h=fixes
Jason. Do you want this in stable? Then we should write a mail after merging.
> ---
> drivers/s390/crypto/vfio_ap_ops.c | 10 ----------
> 1 file changed, 10 deletions(-)
>
> diff --git a/drivers/s390/crypto/vfio_ap_ops.c b/drivers/s390/crypto/vfio_ap_ops.c
> index b2c7e10dfdcd..122c85c22469 100644
> --- a/drivers/s390/crypto/vfio_ap_ops.c
> +++ b/drivers/s390/crypto/vfio_ap_ops.c
> @@ -366,16 +366,6 @@ static int vfio_ap_mdev_remove(struct mdev_device *mdev)
> struct ap_matrix_mdev *matrix_mdev = mdev_get_drvdata(mdev);
>
> mutex_lock(&matrix_dev->lock);
> -
> - /*
> - * If the KVM pointer is in flux or the guest is running, disallow
> - * un-assignment of control domain.
> - */
> - if (matrix_mdev->kvm_busy || matrix_mdev->kvm) {
> - mutex_unlock(&matrix_dev->lock);
> - return -EBUSY;
> - }
> -
> vfio_ap_mdev_reset_queues(mdev);
> list_del(&matrix_mdev->node);
> kfree(matrix_mdev);
>
On Mon, Jun 21, 2021 at 06:04:54PM +0200, Christian Borntraeger wrote:
>
>
> On 21.06.21 17:57, Tony Krowiak wrote:
> > The mdev remove callback for the vfio_ap device driver bails out with
> > -EBUSY if the mdev is in use by a KVM guest (i.e., the KVM pointer in the
> > struct ap_matrix_mdev is not NULL). The intended purpose was
> > to prevent the mdev from being removed while in use. There are two
> > problems with this scenario:
> >
> > 1. Returning a non-zero return code from the remove callback does not
> > prevent the removal of the mdev.
> >
> > 2. The KVM pointer in the struct ap_matrix_mdev will always be NULL because
> > the remove callback will not get invoked until the mdev fd is closed.
> > When the mdev fd is closed, the mdev release callback is invoked and
> > clears the KVM pointer from the struct ap_matrix_mdev.
> >
> > Let's go ahead and remove the check for KVM in the remove callback and
> > allow the cleanup of mdev resources to proceed.
> >
> > Signed-off-by: Tony Krowiak <[email protected]>
> > Reviewed-by: Jason Gunthorpe <[email protected]>
> > Cc: [email protected]
>
> This one is already queued on
>
> https://git.kernel.org/pub/scm/linux/kernel/git/s390/linux.git/log/?h=fixes
>
> Jason. Do you want this in stable? Then we should write a mail after
> merging.
It is fine as is, we are already at rc7, so long as it goes to some
tree for this merge window
Jason
I assumed that this patch would get queued along with the other one in
this series,
but it looks like that was an erroneous assumption. Should this also be
queued?
On 6/21/21 11:57 AM, Tony Krowiak wrote:
> The function pointer to the interception handler for the PQAP instruction
> can get changed during the interception process. Let's add a
> semaphore to struct kvm_s390_crypto to control read/write access to the
> function pointer contained therein.
>
> The semaphore must be locked for write access by the vfio_ap device driver
> when notified that the KVM pointer has been set or cleared. It must be
> locked for read access by the interception framework when the PQAP
> instruction is intercepted.
>
> Signed-off-by: Tony Krowiak <[email protected]>
> Reviewed-by: Jason Gunthorpe <[email protected]>
> Cc: [email protected]
> ---
> arch/s390/include/asm/kvm_host.h | 8 +++-----
> arch/s390/kvm/kvm-s390.c | 1 +
> arch/s390/kvm/priv.c | 10 ++++++----
> drivers/s390/crypto/vfio_ap_ops.c | 23 +++++++++++++++++------
> drivers/s390/crypto/vfio_ap_private.h | 2 +-
> 5 files changed, 28 insertions(+), 16 deletions(-)
>
> diff --git a/arch/s390/include/asm/kvm_host.h b/arch/s390/include/asm/kvm_host.h
> index 8925f3969478..36441669ffba 100644
> --- a/arch/s390/include/asm/kvm_host.h
> +++ b/arch/s390/include/asm/kvm_host.h
> @@ -803,14 +803,12 @@ struct kvm_s390_cpu_model {
> unsigned short ibc;
> };
>
> -struct kvm_s390_module_hook {
> - int (*hook)(struct kvm_vcpu *vcpu);
> - struct module *owner;
> -};
> +typedef int (*crypto_hook)(struct kvm_vcpu *vcpu);
>
> struct kvm_s390_crypto {
> struct kvm_s390_crypto_cb *crycb;
> - struct kvm_s390_module_hook *pqap_hook;
> + struct rw_semaphore pqap_hook_rwsem;
> + crypto_hook *pqap_hook;
> __u32 crycbd;
> __u8 aes_kw;
> __u8 dea_kw;
> diff --git a/arch/s390/kvm/kvm-s390.c b/arch/s390/kvm/kvm-s390.c
> index 1296fc10f80c..418d910df569 100644
> --- a/arch/s390/kvm/kvm-s390.c
> +++ b/arch/s390/kvm/kvm-s390.c
> @@ -2606,6 +2606,7 @@ static void kvm_s390_crypto_init(struct kvm *kvm)
> {
> kvm->arch.crypto.crycb = &kvm->arch.sie_page2->crycb;
> kvm_s390_set_crycb_format(kvm);
> + init_rwsem(&kvm->arch.crypto.pqap_hook_rwsem);
>
> if (!test_kvm_facility(kvm, 76))
> return;
> diff --git a/arch/s390/kvm/priv.c b/arch/s390/kvm/priv.c
> index 9928f785c677..6bed9406c1f3 100644
> --- a/arch/s390/kvm/priv.c
> +++ b/arch/s390/kvm/priv.c
> @@ -610,6 +610,7 @@ static int handle_io_inst(struct kvm_vcpu *vcpu)
> static int handle_pqap(struct kvm_vcpu *vcpu)
> {
> struct ap_queue_status status = {};
> + crypto_hook pqap_hook;
> unsigned long reg0;
> int ret;
> uint8_t fc;
> @@ -657,15 +658,16 @@ static int handle_pqap(struct kvm_vcpu *vcpu)
> * Verify that the hook callback is registered, lock the owner
> * and call the hook.
> */
> + down_read(&vcpu->kvm->arch.crypto.pqap_hook_rwsem);
> if (vcpu->kvm->arch.crypto.pqap_hook) {
> - if (!try_module_get(vcpu->kvm->arch.crypto.pqap_hook->owner))
> - return -EOPNOTSUPP;
> - ret = vcpu->kvm->arch.crypto.pqap_hook->hook(vcpu);
> - module_put(vcpu->kvm->arch.crypto.pqap_hook->owner);
> + pqap_hook = *vcpu->kvm->arch.crypto.pqap_hook;
> + ret = pqap_hook(vcpu);
> if (!ret && vcpu->run->s.regs.gprs[1] & 0x00ff0000)
> kvm_s390_set_psw_cc(vcpu, 3);
> + up_read(&vcpu->kvm->arch.crypto.pqap_hook_rwsem);
> return ret;
> }
> + up_read(&vcpu->kvm->arch.crypto.pqap_hook_rwsem);
> /*
> * A vfio_driver must register a hook.
> * No hook means no driver to enable the SIE CRYCB and no queues.
> diff --git a/drivers/s390/crypto/vfio_ap_ops.c b/drivers/s390/crypto/vfio_ap_ops.c
> index 122c85c22469..742277bc8d1c 100644
> --- a/drivers/s390/crypto/vfio_ap_ops.c
> +++ b/drivers/s390/crypto/vfio_ap_ops.c
> @@ -352,8 +352,7 @@ static int vfio_ap_mdev_create(struct mdev_device *mdev)
> vfio_ap_matrix_init(&matrix_dev->info, &matrix_mdev->matrix);
> init_waitqueue_head(&matrix_mdev->wait_for_kvm);
> mdev_set_drvdata(mdev, matrix_mdev);
> - matrix_mdev->pqap_hook.hook = handle_pqap;
> - matrix_mdev->pqap_hook.owner = THIS_MODULE;
> + matrix_mdev->pqap_hook = handle_pqap;
> mutex_lock(&matrix_dev->lock);
> list_add(&matrix_mdev->node, &matrix_dev->mdev_list);
> mutex_unlock(&matrix_dev->lock);
> @@ -1115,15 +1114,20 @@ static int vfio_ap_mdev_set_kvm(struct ap_matrix_mdev *matrix_mdev,
> }
>
> kvm_get_kvm(kvm);
> + matrix_mdev->kvm = kvm;
> matrix_mdev->kvm_busy = true;
> mutex_unlock(&matrix_dev->lock);
> +
> + down_write(&matrix_mdev->kvm->arch.crypto.pqap_hook_rwsem);
> + kvm->arch.crypto.pqap_hook = &matrix_mdev->pqap_hook;
> + up_write(&matrix_mdev->kvm->arch.crypto.pqap_hook_rwsem);
> +
> kvm_arch_crypto_set_masks(kvm,
> matrix_mdev->matrix.apm,
> matrix_mdev->matrix.aqm,
> matrix_mdev->matrix.adm);
> +
> mutex_lock(&matrix_dev->lock);
> - kvm->arch.crypto.pqap_hook = &matrix_mdev->pqap_hook;
> - matrix_mdev->kvm = kvm;
> matrix_mdev->kvm_busy = false;
> wake_up_all(&matrix_mdev->wait_for_kvm);
> }
> @@ -1189,10 +1193,17 @@ static void vfio_ap_mdev_unset_kvm(struct ap_matrix_mdev *matrix_mdev)
> if (matrix_mdev->kvm) {
> matrix_mdev->kvm_busy = true;
> mutex_unlock(&matrix_dev->lock);
> - kvm_arch_crypto_clear_masks(matrix_mdev->kvm);
> +
> + if (matrix_mdev->kvm->arch.crypto.crycbd) {
> + down_write(&matrix_mdev->kvm->arch.crypto.pqap_hook_rwsem);
> + matrix_mdev->kvm->arch.crypto.pqap_hook = NULL;
> + up_write(&matrix_mdev->kvm->arch.crypto.pqap_hook_rwsem);
> +
> + kvm_arch_crypto_clear_masks(matrix_mdev->kvm);
> + }
> +
> mutex_lock(&matrix_dev->lock);
> vfio_ap_mdev_reset_queues(matrix_mdev->mdev);
> - matrix_mdev->kvm->arch.crypto.pqap_hook = NULL;
> kvm_put_kvm(matrix_mdev->kvm);
> matrix_mdev->kvm = NULL;
> matrix_mdev->kvm_busy = false;
> diff --git a/drivers/s390/crypto/vfio_ap_private.h b/drivers/s390/crypto/vfio_ap_private.h
> index f82a6396acae..e12218e5a629 100644
> --- a/drivers/s390/crypto/vfio_ap_private.h
> +++ b/drivers/s390/crypto/vfio_ap_private.h
> @@ -86,7 +86,7 @@ struct ap_matrix_mdev {
> bool kvm_busy;
> wait_queue_head_t wait_for_kvm;
> struct kvm *kvm;
> - struct kvm_s390_module_hook pqap_hook;
> + crypto_hook pqap_hook;
> struct mdev_device *mdev;
> };
>
On 6/21/21 11:57 AM, Tony Krowiak wrote:
> The function pointer to the interception handler for the PQAP instruction
> can get changed during the interception process. Let's add a
> semaphore to struct kvm_s390_crypto to control read/write access to the
> function pointer contained therein.
>
> The semaphore must be locked for write access by the vfio_ap device driver
> when notified that the KVM pointer has been set or cleared. It must be
> locked for read access by the interception framework when the PQAP
> instruction is intercepted.
>
> Signed-off-by: Tony Krowiak <[email protected]>
> Reviewed-by: Jason Gunthorpe <[email protected]>
> Cc: [email protected]
> ---
> arch/s390/include/asm/kvm_host.h | 8 +++-----
> arch/s390/kvm/kvm-s390.c | 1 +
> arch/s390/kvm/priv.c | 10 ++++++----
> drivers/s390/crypto/vfio_ap_ops.c | 23 +++++++++++++++++------
> drivers/s390/crypto/vfio_ap_private.h | 2 +-
> 5 files changed, 28 insertions(+), 16 deletions(-)
>
> diff --git a/arch/s390/include/asm/kvm_host.h b/arch/s390/include/asm/kvm_host.h
> index 8925f3969478..36441669ffba 100644
> --- a/arch/s390/include/asm/kvm_host.h
> +++ b/arch/s390/include/asm/kvm_host.h
> @@ -803,14 +803,12 @@ struct kvm_s390_cpu_model {
> unsigned short ibc;
> };
>
> -struct kvm_s390_module_hook {
> - int (*hook)(struct kvm_vcpu *vcpu);
> - struct module *owner;
> -};
> +typedef int (*crypto_hook)(struct kvm_vcpu *vcpu);
>
> struct kvm_s390_crypto {
> struct kvm_s390_crypto_cb *crycb;
> - struct kvm_s390_module_hook *pqap_hook;
> + struct rw_semaphore pqap_hook_rwsem;
> + crypto_hook *pqap_hook;
> __u32 crycbd;
> __u8 aes_kw;
> __u8 dea_kw;
> diff --git a/arch/s390/kvm/kvm-s390.c b/arch/s390/kvm/kvm-s390.c
> index 1296fc10f80c..418d910df569 100644
> --- a/arch/s390/kvm/kvm-s390.c
> +++ b/arch/s390/kvm/kvm-s390.c
> @@ -2606,6 +2606,7 @@ static void kvm_s390_crypto_init(struct kvm *kvm)
> {
> kvm->arch.crypto.crycb = &kvm->arch.sie_page2->crycb;
> kvm_s390_set_crycb_format(kvm);
> + init_rwsem(&kvm->arch.crypto.pqap_hook_rwsem);
>
> if (!test_kvm_facility(kvm, 76))
> return;
> diff --git a/arch/s390/kvm/priv.c b/arch/s390/kvm/priv.c
> index 9928f785c677..6bed9406c1f3 100644
> --- a/arch/s390/kvm/priv.c
> +++ b/arch/s390/kvm/priv.c
> @@ -610,6 +610,7 @@ static int handle_io_inst(struct kvm_vcpu *vcpu)
> static int handle_pqap(struct kvm_vcpu *vcpu)
> {
> struct ap_queue_status status = {};
> + crypto_hook pqap_hook;
> unsigned long reg0;
> int ret;
> uint8_t fc;
> @@ -657,15 +658,16 @@ static int handle_pqap(struct kvm_vcpu *vcpu)
> * Verify that the hook callback is registered, lock the owner
> * and call the hook.
> */
> + down_read(&vcpu->kvm->arch.crypto.pqap_hook_rwsem);
> if (vcpu->kvm->arch.crypto.pqap_hook) {
> - if (!try_module_get(vcpu->kvm->arch.crypto.pqap_hook->owner))
> - return -EOPNOTSUPP;
> - ret = vcpu->kvm->arch.crypto.pqap_hook->hook(vcpu);
> - module_put(vcpu->kvm->arch.crypto.pqap_hook->owner);
> + pqap_hook = *vcpu->kvm->arch.crypto.pqap_hook;
> + ret = pqap_hook(vcpu);
> if (!ret && vcpu->run->s.regs.gprs[1] & 0x00ff0000)
> kvm_s390_set_psw_cc(vcpu, 3);
> + up_read(&vcpu->kvm->arch.crypto.pqap_hook_rwsem);
> return ret;
> }
> + up_read(&vcpu->kvm->arch.crypto.pqap_hook_rwsem);
> /*
> * A vfio_driver must register a hook.
> * No hook means no driver to enable the SIE CRYCB and no queues.
> diff --git a/drivers/s390/crypto/vfio_ap_ops.c b/drivers/s390/crypto/vfio_ap_ops.c
> index 122c85c22469..742277bc8d1c 100644
> --- a/drivers/s390/crypto/vfio_ap_ops.c
> +++ b/drivers/s390/crypto/vfio_ap_ops.c
> @@ -352,8 +352,7 @@ static int vfio_ap_mdev_create(struct mdev_device *mdev)
> vfio_ap_matrix_init(&matrix_dev->info, &matrix_mdev->matrix);
> init_waitqueue_head(&matrix_mdev->wait_for_kvm);
> mdev_set_drvdata(mdev, matrix_mdev);
> - matrix_mdev->pqap_hook.hook = handle_pqap;
> - matrix_mdev->pqap_hook.owner = THIS_MODULE;
> + matrix_mdev->pqap_hook = handle_pqap;
> mutex_lock(&matrix_dev->lock);
> list_add(&matrix_mdev->node, &matrix_dev->mdev_list);
> mutex_unlock(&matrix_dev->lock);
> @@ -1115,15 +1114,20 @@ static int vfio_ap_mdev_set_kvm(struct ap_matrix_mdev *matrix_mdev,
> }
>
> kvm_get_kvm(kvm);
> + matrix_mdev->kvm = kvm;
> matrix_mdev->kvm_busy = true;
> mutex_unlock(&matrix_dev->lock);
> +
> + down_write(&matrix_mdev->kvm->arch.crypto.pqap_hook_rwsem);
> + kvm->arch.crypto.pqap_hook = &matrix_mdev->pqap_hook;
Setting the hook using a field in matrix_mdev while not
holding the matrix_dev->lock opens us up to a possible race
condition.
> + up_write(&matrix_mdev->kvm->arch.crypto.pqap_hook_rwsem);
> +
> kvm_arch_crypto_set_masks(kvm,
> matrix_mdev->matrix.apm,
> matrix_mdev->matrix.aqm,
> matrix_mdev->matrix.adm);
> +
> mutex_lock(&matrix_dev->lock);
> - kvm->arch.crypto.pqap_hook = &matrix_mdev->pqap_hook;
> - matrix_mdev->kvm = kvm;
> matrix_mdev->kvm_busy = false;
> wake_up_all(&matrix_mdev->wait_for_kvm);
> }
> @@ -1189,10 +1193,17 @@ static void vfio_ap_mdev_unset_kvm(struct ap_matrix_mdev *matrix_mdev)
> if (matrix_mdev->kvm) {
> matrix_mdev->kvm_busy = true;
> mutex_unlock(&matrix_dev->lock);
> - kvm_arch_crypto_clear_masks(matrix_mdev->kvm);
> +
> + if (matrix_mdev->kvm->arch.crypto.crycbd) {
> + down_write(&matrix_mdev->kvm->arch.crypto.pqap_hook_rwsem);
> + matrix_mdev->kvm->arch.crypto.pqap_hook = NULL;
> + up_write(&matrix_mdev->kvm->arch.crypto.pqap_hook_rwsem);
> +
> + kvm_arch_crypto_clear_masks(matrix_mdev->kvm);
> + }
> +
> mutex_lock(&matrix_dev->lock);
> vfio_ap_mdev_reset_queues(matrix_mdev->mdev);
> - matrix_mdev->kvm->arch.crypto.pqap_hook = NULL;
> kvm_put_kvm(matrix_mdev->kvm);
> matrix_mdev->kvm = NULL;
> matrix_mdev->kvm_busy = false;
> diff --git a/drivers/s390/crypto/vfio_ap_private.h b/drivers/s390/crypto/vfio_ap_private.h
> index f82a6396acae..e12218e5a629 100644
> --- a/drivers/s390/crypto/vfio_ap_private.h
> +++ b/drivers/s390/crypto/vfio_ap_private.h
> @@ -86,7 +86,7 @@ struct ap_matrix_mdev {
> bool kvm_busy;
> wait_queue_head_t wait_for_kvm;
> struct kvm *kvm;
> - struct kvm_s390_module_hook pqap_hook;
> + crypto_hook pqap_hook;
> struct mdev_device *mdev;
> };
>
On 30.06.21 17:18, Tony Krowiak wrote:
> I assumed that this patch would get queued along with the other one in this series,
> but it looks like that was an erroneous assumption. Should this also be queued?
Sorry, this is on my todo list.
On 7/1/21 11:25 AM, Christian Borntraeger wrote:
> On 30.06.21 17:18, Tony Krowiak wrote:
>> I assumed that this patch would get queued along with the other one
>> in this series,
>> but it looks like that was an erroneous assumption. Should this also
>> be queued?
>
> Sorry, this is on my todo list.
That's fine, it is probably a good thing it wasn't merged. I found a
problem with the patch and am working on a replacement. I'll be posting
it soon.
On 7/1/21 11:25 AM, Christian Borntraeger wrote:
> On 30.06.21 17:18, Tony Krowiak wrote:
>> I assumed that this patch would get queued along with the other one
>> in this series,
>> but it looks like that was an erroneous assumption. Should this also
>> be queued?
>
> Sorry, this is on my todo list.
I rolled this up into the patch I posted today:
Message ID: <[email protected]>
s390/vfio-ap: do not open code locks for VFIO_GROUP_NOTIFY_SET_KVM
notification