@@ -420,12 +420,15 @@ int vsp1_du_atomic_update(struct device *dev, unsigned int rpf_index,
rpf->location.left = dst->left;
rpf->location.top = dst->top;
- /* Set the memory buffer address. */
+ /* Set the memory buffer address but don't apply the values to the
+ * hardware as the crop offsets haven't been computed yet.
+ */
memory.num_planes = fmtinfo->planes;
memory.addr[0] = mem[0];
memory.addr[1] = mem[1];
+ memory.addr[2] = 0;
- rpf->ops->set_memory(rpf, &memory);
+ vsp1_rwpf_set_memory(rpf, &memory, false);
spin_lock_irqsave(&pipe->irqlock, flags);
@@ -69,25 +69,20 @@ static int rpf_s_stream(struct v4l2_subdev *subdev, int enable)
pstride = format->plane_fmt[0].bytesperline
<< VI6_RPF_SRCM_PSTRIDE_Y_SHIFT;
- vsp1_rpf_write(rpf, VI6_RPF_SRCM_ADDR_Y,
- rpf->buf_addr[0] + rpf->offsets[0]);
-
if (format->num_planes > 1) {
rpf->offsets[1] = crop->top * format->plane_fmt[1].bytesperline
+ crop->left * fmtinfo->bpp[1] / 8;
pstride |= format->plane_fmt[1].bytesperline
<< VI6_RPF_SRCM_PSTRIDE_C_SHIFT;
-
- vsp1_rpf_write(rpf, VI6_RPF_SRCM_ADDR_C0,
- rpf->buf_addr[1] + rpf->offsets[1]);
-
- if (format->num_planes > 2)
- vsp1_rpf_write(rpf, VI6_RPF_SRCM_ADDR_C1,
- rpf->buf_addr[2] + rpf->offsets[1]);
+ } else {
+ rpf->offsets[1] = 0;
}
vsp1_rpf_write(rpf, VI6_RPF_SRCM_PSTRIDE, pstride);
+ /* Now that the offsets have been computed program the DMA addresses. */
+ rpf->ops->set_memory(rpf);
+
/* Format */
infmt = VI6_RPF_INFMT_CIPM
| (fmtinfo->hwfmt << VI6_RPF_INFMT_RDFMT_SHIFT);
@@ -154,24 +149,14 @@ static struct v4l2_subdev_ops rpf_ops = {
* Video Device Operations
*/
-static void rpf_set_memory(struct vsp1_rwpf *rpf, struct vsp1_rwpf_memory *mem)
+static void rpf_set_memory(struct vsp1_rwpf *rpf)
{
- unsigned int i;
-
- for (i = 0; i < 3; ++i)
- rpf->buf_addr[i] = mem->addr[i];
-
- if (!vsp1_entity_is_streaming(&rpf->entity))
- return;
-
vsp1_rpf_write(rpf, VI6_RPF_SRCM_ADDR_Y,
- mem->addr[0] + rpf->offsets[0]);
- if (mem->num_planes > 1)
- vsp1_rpf_write(rpf, VI6_RPF_SRCM_ADDR_C0,
- mem->addr[1] + rpf->offsets[1]);
- if (mem->num_planes > 2)
- vsp1_rpf_write(rpf, VI6_RPF_SRCM_ADDR_C1,
- mem->addr[2] + rpf->offsets[1]);
+ rpf->buf_addr[0] + rpf->offsets[0]);
+ vsp1_rpf_write(rpf, VI6_RPF_SRCM_ADDR_C0,
+ rpf->buf_addr[1] + rpf->offsets[1]);
+ vsp1_rpf_write(rpf, VI6_RPF_SRCM_ADDR_C1,
+ rpf->buf_addr[2] + rpf->offsets[1]);
}
static const struct vsp1_rwpf_operations rpf_vdev_ops = {
@@ -265,3 +265,29 @@ int vsp1_rwpf_init_ctrls(struct vsp1_rwpf *rwpf)
return rwpf->ctrls.error;
}
+
+/* -----------------------------------------------------------------------------
+ * Buffers
+ */
+
+/**
+ * vsp1_rwpf_set_memory - Configure DMA addresses for a [RW]PF
+ * @rwpf: the [RW]PF instance
+ * @mem: DMA memory addresses
+ * @apply: whether to apply the configuration to the hardware
+ *
+ * This function stores the DMA addresses for all planes in the rwpf instance
+ * and optionally applies the configuration to hardware registers if the apply
+ * argument is set to true.
+ */
+void vsp1_rwpf_set_memory(struct vsp1_rwpf *rwpf, struct vsp1_rwpf_memory *mem,
+ bool apply)
+{
+ unsigned int i;
+
+ for (i = 0; i < 3; ++i)
+ rwpf->buf_addr[i] = mem->addr[i];
+
+ if (apply)
+ rwpf->ops->set_memory(rwpf);
+}
@@ -34,9 +34,13 @@ struct vsp1_rwpf_memory {
unsigned int length[3];
};
+/**
+ * struct vsp1_rwpf_operations - RPF and WPF operations
+ * @set_memory: Setup memory buffer access. This operation applies the settings
+ * stored in the rwpf buf_addr field to the hardware.
+ */
struct vsp1_rwpf_operations {
- void (*set_memory)(struct vsp1_rwpf *rwpf,
- struct vsp1_rwpf_memory *mem);
+ void (*set_memory)(struct vsp1_rwpf *rwpf);
};
struct vsp1_rwpf {
@@ -93,4 +97,7 @@ int vsp1_rwpf_set_selection(struct v4l2_subdev *subdev,
struct v4l2_subdev_pad_config *cfg,
struct v4l2_subdev_selection *sel);
+void vsp1_rwpf_set_memory(struct vsp1_rwpf *rwpf, struct vsp1_rwpf_memory *mem,
+ bool apply);
+
#endif /* __VSP1_RWPF_H__ */
@@ -449,7 +449,7 @@ static void vsp1_video_frame_end(struct vsp1_pipeline *pipe,
spin_lock_irqsave(&pipe->irqlock, flags);
- video->rwpf->ops->set_memory(video->rwpf, &buf->mem);
+ vsp1_rwpf_set_memory(video->rwpf, &buf->mem, true);
pipe->buffers_ready |= 1 << video->pipe_index;
spin_unlock_irqrestore(&pipe->irqlock, flags);
@@ -530,6 +530,11 @@ static int vsp1_video_buffer_prepare(struct vb2_buffer *vb)
return -EINVAL;
}
+ for ( ; i < 3; ++i) {
+ buf->mem.addr[i] = 0;
+ buf->mem.length[i] = 0;
+ }
+
return 0;
}
@@ -552,7 +557,7 @@ static void vsp1_video_buffer_queue(struct vb2_buffer *vb)
spin_lock_irqsave(&pipe->irqlock, flags);
- video->rwpf->ops->set_memory(video->rwpf, &buf->mem);
+ vsp1_rwpf_set_memory(video->rwpf, &buf->mem, true);
pipe->buffers_ready |= 1 << video->pipe_index;
if (vb2_is_streaming(&video->queue) &&
@@ -157,13 +157,11 @@ static struct v4l2_subdev_ops wpf_ops = {
* Video Device Operations
*/
-static void wpf_set_memory(struct vsp1_rwpf *wpf, struct vsp1_rwpf_memory *mem)
+static void wpf_set_memory(struct vsp1_rwpf *wpf)
{
- vsp1_wpf_write(wpf, VI6_WPF_DSTM_ADDR_Y, mem->addr[0]);
- if (mem->num_planes > 1)
- vsp1_wpf_write(wpf, VI6_WPF_DSTM_ADDR_C0, mem->addr[1]);
- if (mem->num_planes > 2)
- vsp1_wpf_write(wpf, VI6_WPF_DSTM_ADDR_C1, mem->addr[2]);
+ vsp1_wpf_write(wpf, VI6_WPF_DSTM_ADDR_Y, wpf->buf_addr[0]);
+ vsp1_wpf_write(wpf, VI6_WPF_DSTM_ADDR_C0, wpf->buf_addr[1]);
+ vsp1_wpf_write(wpf, VI6_WPF_DSTM_ADDR_C1, wpf->buf_addr[2]);
}
static const struct vsp1_rwpf_operations wpf_vdev_ops = {
The RPF source memory pointers need to be offset to take the crop rectangle into account. Offsets are computed in the RPF stream start, which can happen (when using the DRM pipeline) after calling the RPF .set_memory() operation that programs the buffer addresses. The .set_memory() operation tries to guard against the problem by skipping programming of the registers when the module isn't streaming. This will however only protect the first use of an RPF in a DRM pipeline, as in all subsequent uses the module streaming flag will be set and the .set_memory() operation will use potentially incorrect offsets. Fix this by allowing the caller to decide whether to program the hardware immediately or just cache the addresses. While at it refactor the memory set code and create a new vsp1_rwpf_set_memory() that cache addresses and calls the .set_memory() operation to apply them to the hardware. As a side effect the driver now writes all three DMA address registers regardless of the number of planes, and initializes unused addresses to zero. Signed-off-by: Laurent Pinchart <laurent.pinchart+renesas@ideasonboard.com> --- drivers/media/platform/vsp1/vsp1_drm.c | 7 ++++-- drivers/media/platform/vsp1/vsp1_rpf.c | 37 ++++++++++---------------------- drivers/media/platform/vsp1/vsp1_rwpf.c | 26 ++++++++++++++++++++++ drivers/media/platform/vsp1/vsp1_rwpf.h | 11 ++++++++-- drivers/media/platform/vsp1/vsp1_video.c | 9 ++++++-- drivers/media/platform/vsp1/vsp1_wpf.c | 10 ++++----- 6 files changed, 62 insertions(+), 38 deletions(-)