Re: [PATCH] drm/panthor: Lock XArray when getting entries for heap and VM

[Mihail Atanassov <mihail.atanassov@arm.com>]

On 06/11/2024 12:14, Mihail Atanassov wrote:
> Hi Liviu,
> 
> On 06/11/2024 12:07, Liviu Dudau wrote:
>> Similar to cac075706f29 ("drm/panthor: Fix race when converting
>> group handle to group object") we need to use the XArray's internal
>> locking when retrieving a pointer from there for heap and vm.
>>
>> Reported-by: Jann Horn <jannh@google.com>
>> Cc: Boris Brezillon <boris.brezillon@collabora.com>
>> Cc: Steven Price <steven.price@arm.com>
>> Signed-off-by: Liviu Dudau <liviu.dudau@arm.com>
>> ---
>>   drivers/gpu/drm/panthor/panthor_heap.c | 15 +++++++++++++--
>>   drivers/gpu/drm/panthor/panthor_mmu.c  |  2 ++
>>   2 files changed, 15 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/gpu/drm/panthor/panthor_heap.c b/drivers/gpu/drm/ 
>> panthor/panthor_heap.c
>> index 3796a9eb22af2..fe0bcb6837f74 100644
>> --- a/drivers/gpu/drm/panthor/panthor_heap.c
>> +++ b/drivers/gpu/drm/panthor/panthor_heap.c
>> @@ -351,6 +351,17 @@ int panthor_heap_create(struct panthor_heap_pool 
>> *pool,
>>       return ret;
>>   }
>> +static struct panthor_heap *panthor_heap_from_id(struct 
>> pathor_heap_pool *pool, u32 id)

s/pathor/panthor/, but you already know that.

>> +{
>> +    struct panthor_heap *heap;
>> +
>> +    xa_lock(&pool->xa);
>> +    heap = xa_load(&pool->xa, id);
>> +    xa_unlock(&pool->va);

s/va/xa

>> +
>> +    return heap;
>> +}
>> +
>>   /**
>>    * panthor_heap_return_chunk() - Return an unused heap chunk
>>    * @pool: The pool this heap belongs to.
>> @@ -375,7 +386,7 @@ int panthor_heap_return_chunk(struct 
>> panthor_heap_pool *pool,
>>           return -EINVAL;
>>       down_read(&pool->lock);
>> -    heap = xa_load(&pool->xa, heap_id);
>> +    heap = panthor_heap_from_id(pool, heap_id);
>>       if (!heap) {
>>           ret = -EINVAL;
>>           goto out_unlock;
>> @@ -438,7 +449,7 @@ int panthor_heap_grow(struct panthor_heap_pool *pool,
>>           return -EINVAL;
>>       down_read(&pool->lock);
>> -    heap = xa_load(&pool->xa, heap_id);
>> +    heap = panthor_heap_from_id(pool, heap_id);
>>       if (!heap) {
>>           ret = -EINVAL;
>>           goto out_unlock;
>> diff --git a/drivers/gpu/drm/panthor/panthor_mmu.c b/drivers/gpu/drm/ 
>> panthor/panthor_mmu.c
>> index 8ca85526491e6..8b5cda9d21768 100644
>> --- a/drivers/gpu/drm/panthor/panthor_mmu.c
>> +++ b/drivers/gpu/drm/panthor/panthor_mmu.c
>> @@ -1580,7 +1580,9 @@ panthor_vm_pool_get_vm(struct panthor_vm_pool 
>> *pool, u32 handle)
>>   {
>>       struct panthor_vm *vm;
>> +    xa_lock(&pool->xa);
>>       vm = panthor_vm_get(xa_load(&pool->xa, handle));
>> +    xa_unlock(&pool->va);

s/va/xa/

>>       return vm;
>>   }
> 
> Reviewed-by: Mihail Atanassov <mihail.atanassov@arm.com>

Lesson learned for me -- at least build-test what you review :).

With the typos fixed up so this patch builds, I can't observe the race 
in `panthor_vm_pool_get_vm` any more.

The other comments on this patch notwithstanding,

Tested-by: Mihail Atanassov <mihail.atanassov@arm.com>

> 

-- 
Mihail Atanassov <mihail.atanassov@arm.com>