Allow userland to specify a syncobj that is waited on before a render job starts processing.
v2: Use 0 as invalid syncobj to drop flag (Eric) Drop extra newline (Eric)
Signed-off-by: Stefan Schake stschake@gmail.com --- drivers/gpu/drm/vc4/vc4_drv.h | 1 + drivers/gpu/drm/vc4/vc4_gem.c | 30 +++++++++++++++++++++++++----- include/uapi/drm/vc4_drm.h | 7 +++---- 3 files changed, 29 insertions(+), 9 deletions(-)
diff --git a/drivers/gpu/drm/vc4/vc4_drv.h b/drivers/gpu/drm/vc4/vc4_drv.h index 22589d3..554a4e8 100644 --- a/drivers/gpu/drm/vc4/vc4_drv.h +++ b/drivers/gpu/drm/vc4/vc4_drv.h @@ -11,6 +11,7 @@ #include <drm/drm_encoder.h> #include <drm/drm_gem_cma_helper.h> #include <drm/drm_atomic.h> +#include <drm/drm_syncobj.h>
#include "uapi/drm/vc4_drm.h"
diff --git a/drivers/gpu/drm/vc4/vc4_gem.c b/drivers/gpu/drm/vc4/vc4_gem.c index 2107b0d..e305ccde 100644 --- a/drivers/gpu/drm/vc4/vc4_gem.c +++ b/drivers/gpu/drm/vc4/vc4_gem.c @@ -27,6 +27,7 @@ #include <linux/device.h> #include <linux/io.h> #include <linux/sched/signal.h> +#include <linux/dma-fence-array.h>
#include "uapi/drm/vc4_drm.h" #include "vc4_drv.h" @@ -1115,6 +1116,7 @@ vc4_submit_cl_ioctl(struct drm_device *dev, void *data, struct drm_vc4_submit_cl *args = data; struct vc4_exec_info *exec; struct ww_acquire_ctx acquire_ctx; + struct dma_fence *in_fence; int ret = 0;
if ((args->flags & ~(VC4_SUBMIT_CL_USE_CLEAR_COLOR | @@ -1125,11 +1127,6 @@ vc4_submit_cl_ioctl(struct drm_device *dev, void *data, return -EINVAL; }
- if (args->pad2 != 0) { - DRM_DEBUG("->pad2 must be set to zero\n"); - return -EINVAL; - } - exec = kcalloc(1, sizeof(*exec), GFP_KERNEL); if (!exec) { DRM_ERROR("malloc failure on exec struct\n"); @@ -1164,6 +1161,29 @@ vc4_submit_cl_ioctl(struct drm_device *dev, void *data, } }
+ if (args->in_sync) { + ret = drm_syncobj_find_fence(file_priv, args->in_sync, + &in_fence); + if (ret) + goto fail; + + /* When the fence (or fence array) is exclusively from our + * context we can skip the wait since jobs are executed in + * order of their submission through this ioctl and this can + * only have fences from a prior job. + */ + if (!dma_fence_match_context(in_fence, + vc4->dma_fence_context)) { + ret = dma_fence_wait(in_fence, true); + if (ret) { + dma_fence_put(in_fence); + goto fail; + } + } + + dma_fence_put(in_fence); + } + if (exec->args->bin_cl_size != 0) { ret = vc4_get_bcl(dev, exec); if (ret) diff --git a/include/uapi/drm/vc4_drm.h b/include/uapi/drm/vc4_drm.h index b95a0e1..d97065b 100644 --- a/include/uapi/drm/vc4_drm.h +++ b/include/uapi/drm/vc4_drm.h @@ -183,11 +183,10 @@ struct drm_vc4_submit_cl { /* ID of the perfmon to attach to this job. 0 means no perfmon. */ __u32 perfmonid;
- /* Unused field to align this struct on 64 bits. Must be set to 0. - * If one ever needs to add an u32 field to this struct, this field - * can be used. + /* Syncobj handle to wait on. If set, processing of this render job + * will not start until the syncobj is signaled. 0 means ignore. */ - __u32 pad2; + __u32 in_sync; };
/**