On Thu, 2 Feb 2023 03:46:59 +0000
"Liu, Yi L" <yi.l.liu@xxxxxxxxx> wrote:
> > From: Alex Williamson <alex.williamson@xxxxxxxxxx>
> > Sent: Thursday, February 2, 2023 7:28 AM
> >
> > On Wed, 1 Feb 2023 14:20:10 -0500
> > Matthew Rosato <mjrosato@xxxxxxxxxxxxx> wrote:
> >
> > > 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")
> > > Reported-by: Anthony Krowiak <akrowiak@xxxxxxxxxxxxx>
> > > Suggested-by: Jason Gunthorpe <jgg@xxxxxxxxxx>
> > > Signed-off-by: Matthew Rosato <mjrosato@xxxxxxxxxxxxx>
> > > ---
> > > Changes from v1:
> > > * use spin_lock instead of spin_lock_irqsave (Jason)
> > > * clear device->kvm_put as part of vfio_kvm_put_kvm (Yi)
> > > * Re-arrange code to avoid referencing the group contents from within
> > > vfio_main (Kevin) which meant moving most of the code in this patch
> > > to group.c along with getting/dropping of the dev_set lock
> > > ---
> > > drivers/vfio/group.c | 90
> > +++++++++++++++++++++++++++++++++++++---
> > > drivers/vfio/vfio.h | 1 +
> > > drivers/vfio/vfio_main.c | 11 ++---
> > > include/linux/vfio.h | 2 +-
> > > 4 files changed, 91 insertions(+), 13 deletions(-)
> > >
> > > diff --git a/drivers/vfio/group.c b/drivers/vfio/group.c
> > > index bb24b2f0271e..52f434861294 100644
> > > --- a/drivers/vfio/group.c
> > > +++ b/drivers/vfio/group.c
> > > @@ -13,6 +13,9 @@
> > > #include <linux/vfio.h>
> > > #include <linux/iommufd.h>
> > > #include <linux/anon_inodes.h>
> > > +#ifdef CONFIG_HAVE_KVM
> > > +#include <linux/kvm_host.h>
> > > +#endif
> > > #include "vfio.h"
> > >
> > > static struct vfio {
> > > @@ -154,6 +157,55 @@ static int vfio_group_ioctl_set_container(struct
> > vfio_group *group,
> > > return ret;
> > > }
> > >
> > > +#ifdef CONFIG_HAVE_KVM
> > > +static bool vfio_kvm_get_kvm_safe(struct vfio_device *device, struct
> > kvm *kvm)
> >
> > I'm tempted to name these vfio_device_get_kvm_safe() and only pass the
> > vfio_device, where of course we can get the kvm pointer from the group
> > internally.
> >
> > > +{
> > > + void (*pfn)(struct kvm *kvm);
> > > + bool (*fn)(struct kvm *kvm);
> > > + bool ret;
> > > +
> >
> > We should assert_lockdep_held(&device->dev_set->lock) in both of these
> > since that seems to be what's protecting device->kvm and
> > device->put_kvm.
> >
> > If we change as above to get the kvm pointer from the group within this
> > function, we can also move the kvm_ref_lock here, which seems to
> > simplify the caller quite a bit.
> >
> > > + 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;
> >
> > It simplifies the caller if we can use this even in the !device->kvm
> > case.
> >
> > > +
> > > + device->put_kvm(device->kvm);
> > > + device->put_kvm = NULL;
> > > + 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_group_open(struct vfio_device *device)
> > > {
> > > int ret;
> > > @@ -164,14 +216,32 @@ static int vfio_device_group_open(struct
> > vfio_device *device)
> > > goto out_unlock;
> > > }
> > >
> > > + mutex_lock(&device->dev_set->lock);
> > > +
> > > /*
> > > - * 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.
> > > + * Before the first device open, 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 again. Save
> > > + * the pointer in the device for use by drivers.
> > > */
> > > + if (device->open_count == 0) {
> > > + spin_lock(&device->group->kvm_ref_lock);
> > > + if (device->group->kvm &&
> > > + vfio_kvm_get_kvm_safe(device, device->group->kvm))
> > > + device->kvm = device->group->kvm;
> > > + spin_unlock(&device->group->kvm_ref_lock);
> > > + }
> > > +
> > > ret = vfio_device_open(device, device->group->iommufd,
> > > device->group->kvm);
> >
> > We're using device->group->kvm outside of kvm_ref_lock here, it should
> > be using device->kvm.
>
> Existing code set device->kvm in the vfio_device_first_open() which is
> called by vfio_device_open(). After above change, seems not necessary
> to pass kvm pointer into the call chain. Isn't it?
Yes, we can get it from the device. I didn't check how much this
bloats the patch though. As a fix, it might make sense to save that
refactoring for a follow-on patch. Thanks,
Alex
