Re: [PATCH v5, 14/15] media: mtk-vcodec: support stateless VP8 decoding
From: AngeloGioacchino Del Regno
Date: Thu Jan 20 2022 - 10:02:52 EST
Il 17/01/22 10:40, Yunfei Dong ha scritto:
Add support for VP8 decoding using the stateless API,
as supported by MT8192.
Signed-off-by: Yunfei Dong <yunfei.dong@xxxxxxxxxxxx>
---
drivers/media/platform/mtk-vcodec/Makefile | 1 +
.../mtk-vcodec/mtk_vcodec_dec_stateless.c | 24 +-
.../platform/mtk-vcodec/mtk_vcodec_drv.h | 1 +
.../mtk-vcodec/vdec/vdec_vp8_req_if.c | 440 ++++++++++++++++++
.../media/platform/mtk-vcodec/vdec_drv_if.c | 4 +
.../media/platform/mtk-vcodec/vdec_drv_if.h | 1 +
6 files changed, 469 insertions(+), 2 deletions(-)
create mode 100644 drivers/media/platform/mtk-vcodec/vdec/vdec_vp8_req_if.c
Hello Yunfei,
I've found some issues in this patch, and there are also some other considerations
to tidy it up.
(....snip....)
diff --git a/drivers/media/platform/mtk-vcodec/vdec/vdec_vp8_req_if.c b/drivers/media/platform/mtk-vcodec/vdec/vdec_vp8_req_if.c
new file mode 100644
index 000000000000..969568b98251
--- /dev/null
+++ b/drivers/media/platform/mtk-vcodec/vdec/vdec_vp8_req_if.c
@@ -0,0 +1,440 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Copyright (c) 2021 MediaTek Inc.
+ * Author: Yunfei Dong <yunfei.dong@xxxxxxxxxxxx>
+ */
+
+#include <linux/slab.h>
+#include <media/v4l2-mem2mem.h>
+#include <media/videobuf2-dma-contig.h>
+#include <uapi/linux/v4l2-controls.h>
+
+#include "../mtk_vcodec_util.h"
+#include "../mtk_vcodec_dec.h"
+#include "../mtk_vcodec_intr.h"
+#include "../vdec_drv_base.h"
+#include "../vdec_drv_if.h"
+#include "../vdec_vpu_if.h"
+
+/* Decoding picture buffer size (3 reference frames plus current frame) */
+#define VP8_DPB_SIZE 4
+
+/* HW working buffer size (bytes) */
+#define VP8_SEG_ID_SZ (256 * 1024)
This is SZ_256K
+#define VP8_PP_WRAPY_SZ (64 * 1024)
+#define VP8_PP_WRAPC_SZ (64 * 1024)
+#define VP8_VLD_PRED_SZ (64 * 1024)
And these are all SZ_64K.
+
+/**
+ * struct vdec_vp8_slice_info - decode misc information
+ * @vld_wrapper_dma : vld wrapper dma address
+ * @seg_id_buf_dma : seg id dma address
+ * @wrap_y_dma : wrap y dma address
+ * @wrap_c_dma : wrap y dma address
+ * @cur_y_fb_dma : current plane Y frame buffer dma address
+ * @cur_c_fb_dma : current plane C frame buffer dma address
+ * @bs_dma : bitstream dma address
+ * @bs_sz : bitstream size
+ * @resolution_changed: resolution change flag 1 - changed, 0 - not change
+ * @frame_header_type : current frame header type
+ * @wait_key_frame : wait key frame coming
+ * @crc : used to check whether hardware's status is right
+ * @timeout : decode timeout: 1 timeout, 0 no timeount
There's no `timeout` in this structure, did you forget to remove the
documentation for this one?
+ * @reserved: : reserved, currently unused
+ */
+struct vdec_vp8_slice_info {
+ u64 vld_wrapper_dma;
+ u64 seg_id_buf_dma;
+ u64 wrap_y_dma;
+ u64 wrap_c_dma;
+ u64 cur_y_fb_dma;
+ u64 cur_c_fb_dma;
+ u64 bs_dma;
+ u32 bs_sz;
+ u32 resolution_changed;
+ u32 frame_header_type;
+ u32 crc[8];
+ u32 reserved;
+};
+
+/**
+ * struct vdec_vp8_slice_dpb_info - vp8 reference information
+ * @y_dma_addr : Y bitstream physical address
+ * @c_dma_addr : CbCr bitstream physical address
+ * @reference_flag: reference picture flag
+ * @reserved : 64bit align
+ */
+struct vdec_vp8_slice_dpb_info {
+ dma_addr_t y_dma_addr;
+ dma_addr_t c_dma_addr;
+ int reference_flag;
+ int reserved;
+};
+
+/**
+ * struct vdec_vp8_slice_vsi - VPU shared information
+ * @dec : decoding information
+ * @pic : picture information
+ * @vp8_dpb_info : reference buffer information
+ */
+struct vdec_vp8_slice_vsi {
+ struct vdec_vp8_slice_info dec;
+ struct vdec_pic_info pic;
+ struct vdec_vp8_slice_dpb_info vp8_dpb_info[3];
+};
+
+/**
+ * struct vdec_vp8_slice_inst - VP8 decoder instance
+ * @seg_id_buf : seg buffer
+ * @wrap_y_buf : wrapper y buffer
+ * @wrap_c_buf : wrapper c buffer
+ * @vld_wrapper_buf: vld wrapper buffer
+ * @ctx : V4L2 context
+ * @vpu : VPU instance for decoder
+ * @vsi : VPU share information
+ */
+struct vdec_vp8_slice_inst {
+ struct mtk_vcodec_mem seg_id_buf;
+ struct mtk_vcodec_mem wrap_y_buf;
+ struct mtk_vcodec_mem wrap_c_buf;
+ struct mtk_vcodec_mem vld_wrapper_buf;
+ struct mtk_vcodec_ctx *ctx;
+ struct vdec_vpu_inst vpu;
+ struct vdec_vp8_slice_vsi *vsi;
+};
+
+static void *vdec_vp8_slice_get_ctrl_ptr(struct mtk_vcodec_ctx *ctx, int id)
+{
+ struct v4l2_ctrl *ctrl = v4l2_ctrl_find(&ctx->ctrl_hdl, id);
+
You should check if ctrl is NULL here, and eventually return a ERR_PTR(-EINVAL)
if that ever occurs... or you may get a NULL pointer kernel panic...
+ return ctrl->p_cur.p;
+}
+
...snip...
+
+static void vdec_vp8_slice_get_decode_parameters(struct vdec_vp8_slice_inst *inst)
static int ..... ?
+{
+ const struct v4l2_ctrl_vp8_frame *frame_header =
+ vdec_vp8_slice_get_ctrl_ptr(inst->ctx, V4L2_CTRL_TYPE_VP8_FRAME);
This is crashing the kernel, because it ends up being NULL...
... the solution to that is to look for V4L2_CID_STATELESS_VP8_FRAME instead of
V4L2_CTRL_TYPE_VP8_FRAME, but you should really do an error check here and
eventually bail out, in case anything goes horribly wrong...
+ struct mtk_vcodec_ctx *ctx = inst->ctx;
+ struct vb2_queue *vq;
+ struct vb2_buffer *vb;
+ u64 referenct_ts;
+ int index, vb2_index;
+
...so you'd be checking for error pointer of frame_header here and returning.
+ vq = v4l2_m2m_get_vq(ctx->m2m_ctx, V4L2_BUF_TYPE_VIDEO_CAPTURE_MPLANE);
+ for (index = 0; index < 3; index++) {
+ referenct_ts = vdec_vp8_slice_get_ref_by_ts(frame_header, index);
+ vb2_index = vb2_find_timestamp(vq, referenct_ts, 0);
+ if (vb2_index < 0) {
+ if (!V4L2_VP8_FRAME_IS_KEY_FRAME(frame_header))
+ mtk_vcodec_err(inst, "reference invalid: index(%d) ts(%lld)",
+ index, referenct_ts);
+ inst->vsi->vp8_dpb_info[index].reference_flag = 0;
+ continue;
+ }
+ inst->vsi->vp8_dpb_info[index].reference_flag = 1;
+
+ vb = vq->bufs[vb2_index];
+ inst->vsi->vp8_dpb_info[index].y_dma_addr =
+ vb2_dma_contig_plane_dma_addr(vb, 0);
+ if (ctx->q_data[MTK_Q_DATA_DST].fmt->num_planes == 2)
+ inst->vsi->vp8_dpb_info[index].c_dma_addr =
+ vb2_dma_contig_plane_dma_addr(vb, 1);
+ else
+ inst->vsi->vp8_dpb_info[index].c_dma_addr =
+ inst->vsi->vp8_dpb_info[index].y_dma_addr +
+ ctx->picinfo.fb_sz[0];
+ }
+
+ inst->vsi->dec.frame_header_type = frame_header->flags >> 1;
+}
+
+static int vdec_vp8_slice_init(struct mtk_vcodec_ctx *ctx)
+{
+ struct vdec_vp8_slice_inst *inst;
+ int err;
+
+ inst = kzalloc(sizeof(*inst), GFP_KERNEL);
+ if (!inst)
+ return -ENOMEM;
+
+ inst->ctx = ctx;
+
+ inst->vpu.id = SCP_IPI_VDEC_LAT;
+ inst->vpu.core_id = SCP_IPI_VDEC_CORE;
+ inst->vpu.ctx = ctx;
+ inst->vpu.codec_type = ctx->current_codec;
+ inst->vpu.capture_type = ctx->capture_fourcc;
+
+ err = vpu_dec_init(&inst->vpu);
+ if (err) {
+ mtk_vcodec_err(inst, "vdec_vp8 init err=%d", err);
+ goto error_free_inst;
+ }
+
+ inst->vsi = inst->vpu.vsi;
+ err = vdec_vp8_slice_alloc_working_buf(inst);
+ if (err)
+ goto error_deinit;
+
+ mtk_vcodec_debug(inst, "vp8 struct size = %d vsi: %d\n",
+ (int)sizeof(struct v4l2_ctrl_vp8_frame),
+ (int)sizeof(struct vdec_vp8_slice_vsi));
+ mtk_vcodec_debug(inst, "vp8:%p, codec_type = 0x%x vsi: 0x%p",
+ inst, inst->vpu.codec_type, inst->vpu.vsi);
+
+ ctx->drv_handle = inst;
+ return 0;
+
+error_deinit:
+ vpu_dec_deinit(&inst->vpu);
+error_free_inst:
+ kfree(inst);
+ return err;
+}
+
+static int vdec_vp8_slice_decode(void *h_vdec, struct mtk_vcodec_mem *bs,
+ struct vdec_fb *fb, bool *res_chg)
+{
+ struct vdec_vp8_slice_inst *inst = h_vdec;
+ struct vdec_vpu_inst *vpu = &inst->vpu;
+ struct mtk_video_dec_buf *src_buf_info, *dst_buf_info;
+ unsigned int data;
+ u64 y_fb_dma, c_fb_dma;
+ int err, timeout;
+
+ /* Resolution changes are never initiated by us */
+ *res_chg = false;
+
+ /* bs NULL means flush decoder */
+ if (!bs)
+ return vpu_dec_reset(vpu);
+
+ src_buf_info = container_of(bs, struct mtk_video_dec_buf, bs_buffer);
+
+ fb = inst->ctx->dev->vdec_pdata->get_cap_buffer(inst->ctx);
+ dst_buf_info = container_of(fb, struct mtk_video_dec_buf, frame_buffer);
+
+ y_fb_dma = fb ? (u64)fb->base_y.dma_addr : 0;
+ if (inst->ctx->q_data[MTK_Q_DATA_DST].fmt->num_planes == 1)
+ c_fb_dma = y_fb_dma +
+ inst->ctx->picinfo.buf_w * inst->ctx->picinfo.buf_h;
+ else
+ c_fb_dma = fb ? (u64)fb->base_c.dma_addr : 0;
+
+ inst->vsi->dec.bs_dma = (unsigned long)bs->dma_addr;
inst->vsi->dec.bs_dma is u64... so please cast to u64.
+ inst->vsi->dec.bs_sz = bs->size;
+ inst->vsi->dec.cur_y_fb_dma = y_fb_dma;
+ inst->vsi->dec.cur_c_fb_dma = c_fb_dma;
+
+ mtk_vcodec_debug(inst, "frame[%d] bs(%zu 0x%lx) y/c(0x%llx 0x%llx)",
+ inst->ctx->decoded_frame_cnt,
+ bs->size, (unsigned long)bs->dma_addr,
...and it would be useful if it was u64 here too, obviously.
+ y_fb_dma, c_fb_dma);
+
+ v4l2_m2m_buf_copy_metadata(&src_buf_info->m2m_buf.vb,
+ &dst_buf_info->m2m_buf.vb, true);
+
+ vdec_vp8_slice_get_decode_parameters(inst);
+ err = vpu_dec_start(vpu, &data, 1);
+ if (err) {
+ mtk_vcodec_debug(inst, "vp8 dec start err!");
+ goto error;
+ }
Regards,
Angelo