[PATCH 10/14] drm/amdkfd: Use ref count to prevent kfd_process destruction
Felix Kuehling
felix.kuehling at amd.com
Fri Dec 1 21:17:30 UTC 2017
On 2017-11-28 04:52 AM, Christian König wrote:
> Am 28.11.2017 um 00:29 schrieb Felix Kuehling:
>> Use a reference counter instead of a lock to prevent process
>> destruction while functions running out of process context are using
>> the kfd_process structure. In many cases these functions don't need
>> the structure to be locked. In the few cases that really do need the
>> process lock, take it explicitly.
>>
>> This helps simplify lock dependencies between the process lock and
>> other locks, particularly amdgpu and mm_struct locks. This will be
>> important when amdgpu calls back to amdkfd for memory evictions.
>
> Actually that is not only an optimization or cleanup, but a rather
> important bug fix.
>
> Using a mutex as protection to prevent object deletion is illegal
> because mutex_unlock() can accesses the mutex object even after it is
> unlocked.
>
> See this LWN article as well https://lwn.net/Articles/575460/.
>
> If you have other use cases like this in the KFD it should better be
> fixed as well.
I'm not aware of other misuses of Mutexes in KFD.
The article sounded like this was likely to get fixed in the mutex
rather than hoping to track down all incorrect uses of Mutexes. Quote:
>
> As of this writing, no patches have been posted. It would be
> surprising, though, if a fix for this particular problem did not
> surface by the time the 3.14 merge window opens. Locking problems are
> hard enough to deal with when the locking primitives have simple and
> easily understood behavior; having subtle traps built into that layer
> of the kernel is a recipe for a lot of long-term pain.
>
I haven't found such a fix. That said, in the discussion under that
article some argued that the example would be broken even with a
spinlock. So maybe there is no such general fix.
Regards,
Felix
>
>> Signed-off-by: Felix Kuehling <Felix.Kuehling at amd.com>
>
> Acked-by: Christian König <christian.koenig at amd.com>
>
> Regards,
> Christian.
>
>> ---
>> drivers/gpu/drm/amd/amdkfd/kfd_events.c | 14 +++++++-------
>> drivers/gpu/drm/amd/amdkfd/kfd_priv.h | 1 +
>> drivers/gpu/drm/amd/amdkfd/kfd_process.c | 16 +++++++++++++---
>> 3 files changed, 21 insertions(+), 10 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_events.c
>> b/drivers/gpu/drm/amd/amdkfd/kfd_events.c
>> index cb92d4b..93aae5c 100644
>> --- a/drivers/gpu/drm/amd/amdkfd/kfd_events.c
>> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_events.c
>> @@ -441,7 +441,7 @@ void kfd_signal_event_interrupt(unsigned int
>> pasid, uint32_t partial_id,
>> /*
>> * Because we are called from arbitrary context (workqueue) as
>> opposed
>> * to process context, kfd_process could attempt to exit while
>> we are
>> - * running so the lookup function returns a locked process.
>> + * running so the lookup function increments the process ref count.
>> */
>> struct kfd_process *p = kfd_lookup_process_by_pasid(pasid);
>> @@ -493,7 +493,7 @@ void kfd_signal_event_interrupt(unsigned int
>> pasid, uint32_t partial_id,
>> }
>> mutex_unlock(&p->event_mutex);
>> - mutex_unlock(&p->mutex);
>> + kfd_unref_process(p);
>> }
>> static struct kfd_event_waiter *alloc_event_waiters(uint32_t
>> num_events)
>> @@ -847,7 +847,7 @@ void kfd_signal_iommu_event(struct kfd_dev *dev,
>> unsigned int pasid,
>> /*
>> * Because we are called from arbitrary context (workqueue) as
>> opposed
>> * to process context, kfd_process could attempt to exit while
>> we are
>> - * running so the lookup function returns a locked process.
>> + * running so the lookup function increments the process ref count.
>> */
>> struct kfd_process *p = kfd_lookup_process_by_pasid(pasid);
>> struct mm_struct *mm;
>> @@ -860,7 +860,7 @@ void kfd_signal_iommu_event(struct kfd_dev *dev,
>> unsigned int pasid,
>> */
>> mm = get_task_mm(p->lead_thread);
>> if (!mm) {
>> - mutex_unlock(&p->mutex);
>> + kfd_unref_process(p);
>> return; /* Process is exiting */
>> }
>> @@ -903,7 +903,7 @@ void kfd_signal_iommu_event(struct kfd_dev
>> *dev, unsigned int pasid,
>> &memory_exception_data);
>> mutex_unlock(&p->event_mutex);
>> - mutex_unlock(&p->mutex);
>> + kfd_unref_process(p);
>> }
>> void kfd_signal_hw_exception_event(unsigned int pasid)
>> @@ -911,7 +911,7 @@ void kfd_signal_hw_exception_event(unsigned int
>> pasid)
>> /*
>> * Because we are called from arbitrary context (workqueue) as
>> opposed
>> * to process context, kfd_process could attempt to exit while
>> we are
>> - * running so the lookup function returns a locked process.
>> + * running so the lookup function increments the process ref count.
>> */
>> struct kfd_process *p = kfd_lookup_process_by_pasid(pasid);
>> @@ -924,5 +924,5 @@ void kfd_signal_hw_exception_event(unsigned int
>> pasid)
>> lookup_events_by_type_and_signal(p,
>> KFD_EVENT_TYPE_HW_EXCEPTION, NULL);
>> mutex_unlock(&p->event_mutex);
>> - mutex_unlock(&p->mutex);
>> + kfd_unref_process(p);
>> }
>> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
>> b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
>> index 248e4f5..0c96a6b 100644
>> --- a/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
>> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_priv.h
>> @@ -606,6 +606,7 @@ void kfd_process_destroy_wq(void);
>> struct kfd_process *kfd_create_process(struct file *filep);
>> struct kfd_process *kfd_get_process(const struct task_struct *);
>> struct kfd_process *kfd_lookup_process_by_pasid(unsigned int pasid);
>> +void kfd_unref_process(struct kfd_process *p);
>> struct kfd_process_device *kfd_bind_process_to_device(struct
>> kfd_dev *dev,
>> struct kfd_process *p);
>> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_process.c
>> b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
>> index e02e8a2..509f987 100644
>> --- a/drivers/gpu/drm/amd/amdkfd/kfd_process.c
>> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_process.c
>> @@ -49,6 +49,7 @@ DEFINE_STATIC_SRCU(kfd_processes_srcu);
>> static struct workqueue_struct *kfd_process_wq;
>> static struct kfd_process *find_process(const struct task_struct
>> *thread);
>> +static void kfd_process_ref_release(struct kref *ref);
>> static struct kfd_process *create_process(const struct task_struct
>> *thread);
>> static int kfd_process_init_cwsr(struct kfd_process *p, struct file
>> *filep);
>> @@ -146,6 +147,11 @@ static struct kfd_process *find_process(const
>> struct task_struct *thread)
>> return p;
>> }
>> +void kfd_unref_process(struct kfd_process *p)
>> +{
>> + kref_put(&p->ref, kfd_process_ref_release);
>> +}
>> +
>> /* No process locking is needed in this function, because the process
>> * is not findable any more. We must assume that no other thread is
>> * using it any more, otherwise we couldn't safely free the process
>> @@ -201,7 +207,7 @@ static void kfd_process_destroy_delayed(struct
>> rcu_head *rcu)
>> {
>> struct kfd_process *p = container_of(rcu, struct kfd_process,
>> rcu);
>> - kref_put(&p->ref, kfd_process_ref_release);
>> + kfd_unref_process(p);
>> }
>> static void kfd_process_notifier_release(struct mmu_notifier *mn,
>> @@ -525,6 +531,8 @@ void kfd_process_iommu_unbind_callback(struct
>> kfd_dev *dev, unsigned int pasid)
>> mutex_unlock(kfd_get_dbgmgr_mutex());
>> + mutex_lock(&p->mutex);
>> +
>> pdd = kfd_get_process_device_data(dev, p);
>> if (pdd)
>> /* For GPU relying on IOMMU, we need to dequeue here
>> @@ -533,6 +541,8 @@ void kfd_process_iommu_unbind_callback(struct
>> kfd_dev *dev, unsigned int pasid)
>> kfd_process_dequeue_from_device(pdd);
>> mutex_unlock(&p->mutex);
>> +
>> + kfd_unref_process(p);
>> }
>> struct kfd_process_device *kfd_get_first_process_device_data(
>> @@ -557,7 +567,7 @@ bool kfd_has_process_device_data(struct
>> kfd_process *p)
>> return !(list_empty(&p->per_device_data));
>> }
>> -/* This returns with process->mutex locked. */
>> +/* This increments the process->ref counter. */
>> struct kfd_process *kfd_lookup_process_by_pasid(unsigned int pasid)
>> {
>> struct kfd_process *p;
>> @@ -567,7 +577,7 @@ struct kfd_process
>> *kfd_lookup_process_by_pasid(unsigned int pasid)
>> hash_for_each_rcu(kfd_processes_table, temp, p, kfd_processes) {
>> if (p->pasid == pasid) {
>> - mutex_lock(&p->mutex);
>> + kref_get(&p->ref);
>> break;
>> }
>> }
>
More information about the amd-gfx
mailing list