From: Christian König christian.koenig@amd.com
At least inside the same client we should stop waiting for a buffer to be idle, but rather wait for a specific command submission to complete.
Signed-off-by: Christian König christian.koenig@amd.com --- drivers/gpu/drm/radeon/radeon.h | 2 ++ drivers/gpu/drm/radeon/radeon_gem.c | 26 ++++++++++++++++++++++++++ drivers/gpu/drm/radeon/radeon_kms.c | 1 + include/uapi/drm/radeon_drm.h | 7 +++++++ 4 files changed, 36 insertions(+)
diff --git a/drivers/gpu/drm/radeon/radeon.h b/drivers/gpu/drm/radeon/radeon.h index 1529afb..d8bf3a7 100644 --- a/drivers/gpu/drm/radeon/radeon.h +++ b/drivers/gpu/drm/radeon/radeon.h @@ -2230,6 +2230,8 @@ int radeon_gem_busy_ioctl(struct drm_device *dev, void *data, struct drm_file *filp); int radeon_gem_wait_idle_ioctl(struct drm_device *dev, void *data, struct drm_file *filp); +int radeon_gem_wait_cs_ioctl(struct drm_device *dev, void *data, + struct drm_file *filp); int radeon_gem_va_ioctl(struct drm_device *dev, void *data, struct drm_file *filp); int radeon_gem_op_ioctl(struct drm_device *dev, void *data, diff --git a/drivers/gpu/drm/radeon/radeon_gem.c b/drivers/gpu/drm/radeon/radeon_gem.c index dd45611..297f327 100644 --- a/drivers/gpu/drm/radeon/radeon_gem.c +++ b/drivers/gpu/drm/radeon/radeon_gem.c @@ -494,6 +494,32 @@ int radeon_gem_wait_idle_ioctl(struct drm_device *dev, void *data, return r; }
+int radeon_gem_wait_cs_ioctl(struct drm_device *dev, void *data, + struct drm_file *filp) +{ + struct radeon_fpriv *fpriv = filp->driver_priv; + struct drm_radeon_gem_wait_cs *args = data; + struct radeon_fence *fence; + unsigned long timeout; + long r; + + mutex_lock(&fpriv->seq_lock); + fence = radeon_fence_ref(radeon_seq_query(fpriv->seq, args->id)); + mutex_unlock(&fpriv->seq_lock); + + timeout = nsecs_to_jiffies(args->timeout); + r = fence_wait_timeout(&fence->base, true, timeout); + radeon_fence_unref(&fence); + + if (r == 0) + return -EBUSY; + + if (r < 0) + return r; + + return 0; +} + int radeon_gem_set_tiling_ioctl(struct drm_device *dev, void *data, struct drm_file *filp) { diff --git a/drivers/gpu/drm/radeon/radeon_kms.c b/drivers/gpu/drm/radeon/radeon_kms.c index db5c986..69b74a8 100644 --- a/drivers/gpu/drm/radeon/radeon_kms.c +++ b/drivers/gpu/drm/radeon/radeon_kms.c @@ -892,5 +892,6 @@ const struct drm_ioctl_desc radeon_ioctls_kms[] = { DRM_IOCTL_DEF_DRV(RADEON_GEM_VA, radeon_gem_va_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW), DRM_IOCTL_DEF_DRV(RADEON_GEM_OP, radeon_gem_op_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW), DRM_IOCTL_DEF_DRV(RADEON_GEM_USERPTR, radeon_gem_userptr_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW), + DRM_IOCTL_DEF_DRV(RADEON_GEM_WAIT_CS, radeon_gem_wait_cs_ioctl, DRM_AUTH|DRM_UNLOCKED|DRM_RENDER_ALLOW), }; int radeon_max_kms_ioctl = ARRAY_SIZE(radeon_ioctls_kms); diff --git a/include/uapi/drm/radeon_drm.h b/include/uapi/drm/radeon_drm.h index 2c50838..d700c06 100644 --- a/include/uapi/drm/radeon_drm.h +++ b/include/uapi/drm/radeon_drm.h @@ -512,6 +512,7 @@ typedef struct { #define DRM_RADEON_GEM_VA 0x2b #define DRM_RADEON_GEM_OP 0x2c #define DRM_RADEON_GEM_USERPTR 0x2d +#define DRM_RADEON_GEM_WAIT_CS 0x2e
#define DRM_IOCTL_RADEON_CP_INIT DRM_IOW( DRM_COMMAND_BASE + DRM_RADEON_CP_INIT, drm_radeon_init_t) #define DRM_IOCTL_RADEON_CP_START DRM_IO( DRM_COMMAND_BASE + DRM_RADEON_CP_START) @@ -556,6 +557,7 @@ typedef struct { #define DRM_IOCTL_RADEON_GEM_VA DRM_IOWR(DRM_COMMAND_BASE + DRM_RADEON_GEM_VA, struct drm_radeon_gem_va) #define DRM_IOCTL_RADEON_GEM_OP DRM_IOWR(DRM_COMMAND_BASE + DRM_RADEON_GEM_OP, struct drm_radeon_gem_op) #define DRM_IOCTL_RADEON_GEM_USERPTR DRM_IOWR(DRM_COMMAND_BASE + DRM_RADEON_GEM_USERPTR, struct drm_radeon_gem_userptr) +#define DRM_IOCTL_RADEON_GEM_WAIT_CS DRM_IOWR(DRM_COMMAND_BASE + DRM_RADEON_GEM_WAIT_CS, struct drm_radeon_gem_wait_cs)
typedef struct drm_radeon_init { enum { @@ -880,6 +882,11 @@ struct drm_radeon_gem_wait_idle { uint32_t pad; };
+struct drm_radeon_gem_wait_cs { + uint64_t id; + uint64_t timeout; +}; + struct drm_radeon_gem_busy { uint32_t handle; uint32_t domain;