[PATCH] drm/amdgpu: Off by one sanity checks
Christian König
christian.koenig at amd.com
Wed Jul 12 07:42:14 UTC 2017
Am 11.07.2017 um 21:53 schrieb Dan Carpenter:
> This is just future proofing code, not something that can be triggered
> in real life. We're testing to make sure we don't shift wrap when we
> do "1ull << i" so "i" has to be in the 0-63 range. If it's 64 then we
> have gone too far.
>
> Signed-off-by: Dan Carpenter <dan.carpenter at oracle.com>
Acked-by: Christian König <christian.koenig at amd.com>
>
> diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
> index 3a0b69b09ed6..e63925dffd2a 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v9_0.c
> @@ -2425,7 +2425,7 @@ static int gfx_v9_0_kiq_kcq_enable(struct amdgpu_device *adev)
> /* This situation may be hit in the future if a new HW
> * generation exposes more than 64 queues. If so, the
> * definition of queue_mask needs updating */
> - if (WARN_ON(i > (sizeof(queue_mask)*8))) {
> + if (WARN_ON(i >= (sizeof(queue_mask)*8))) {
> DRM_ERROR("Invalid KCQ enabled: %d\n", i);
> break;
> }
> diff --git a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
> index aa5a50f5eac8..85a79829842e 100644
> --- a/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
> +++ b/drivers/gpu/drm/amd/amdgpu/gfx_v8_0.c
> @@ -4564,7 +4564,7 @@ static int gfx_v8_0_kiq_kcq_enable(struct amdgpu_device *adev)
> /* This situation may be hit in the future if a new HW
> * generation exposes more than 64 queues. If so, the
> * definition of queue_mask needs updating */
> - if (WARN_ON(i > (sizeof(queue_mask)*8))) {
> + if (WARN_ON(i >= (sizeof(queue_mask)*8))) {
> DRM_ERROR("Invalid KCQ enabled: %d\n", i);
> break;
> }
> diff --git a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
> index 955aa304ff48..0def783889cd 100644
> --- a/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
> +++ b/drivers/gpu/drm/amd/amdkfd/kfd_device_queue_manager.c
> @@ -671,7 +671,7 @@ static int set_sched_resources(struct device_queue_manager *dqm)
> /* This situation may be hit in the future if a new HW
> * generation exposes more than 64 queues. If so, the
> * definition of res.queue_mask needs updating */
> - if (WARN_ON(i > (sizeof(res.queue_mask)*8))) {
> + if (WARN_ON(i >= (sizeof(res.queue_mask)*8))) {
> pr_err("Invalid queue enabled by amdgpu: %d\n", i);
> break;
> }
More information about the amd-gfx
mailing list