1 files changed, 41 insertions, 24 deletions

diff --git a/drivers/media/platform/amphion/vpu_malone.c b/drivers/media/platform/amphion/vpu_malone.c
index 4769c053c6c2..80802975c4f1 100644
--- a/drivers/media/platform/amphion/vpu_malone.c
+++ b/drivers/media/platform/amphion/vpu_malone.c
@@ -3,6 +3,7 @@
  * Copyright 2020-2021 NXP
  */
 
+#include <linux/bitfield.h>
 #include <linux/init.h>
 #include <linux/interconnect.h>
 #include <linux/ioctl.h>
@@ -25,6 +26,10 @@
 #include "vpu_imx8q.h"
 #include "vpu_malone.h"
 
+static bool low_latency;
+module_param(low_latency, bool, 0644);
+MODULE_PARM_DESC(low_latency, "Set low latency frame flush mode: 0 (disable) or 1 (enable)");
+
 #define CMD_SIZE			25600
 #define MSG_SIZE			25600
 #define CODEC_SIZE			0x1000
@@ -68,6 +73,12 @@
 
 #define MALONE_DEC_FMT_RV_MASK			BIT(21)
 
+#define MALONE_VERSION_MASK			0xFFFFF
+#define MALONE_VERSION(maj, min, inc)		\
+		(FIELD_PREP(0xF0000, maj) | FIELD_PREP(0xFF00, min) | FIELD_PREP(0xFF, inc))
+#define CHECK_VERSION(iface, maj, min)		\
+		(FIELD_GET(MALONE_VERSION_MASK, (iface)->fw_version) >= MALONE_VERSION(maj, min, 0))
+
 enum vpu_malone_stream_input_mode {
 	INVALID_MODE = 0,
 	FRAME_LVL,
@@ -332,6 +343,8 @@ struct vpu_dec_ctrl {
 	u32 buf_addr[VID_API_NUM_STREAMS];
 };
 
+static const struct malone_padding_scode *get_padding_scode(u32 type, u32 fmt);
+
 u32 vpu_malone_get_data_size(void)
 {
 	return sizeof(struct vpu_dec_ctrl);
@@ -654,9 +667,15 @@ static int vpu_malone_set_params(struct vpu_shared_addr *shared,
 		hc->jpg[instance].jpg_mjpeg_interlaced = 0;
 	}
 
-	hc->codec_param[instance].disp_imm = params->display_delay_enable ? 1 : 0;
-	if (malone_format != MALONE_FMT_AVC)
+	if (params->display_delay_enable &&
+	    get_padding_scode(SCODE_PADDING_BUFFLUSH, params->codec_format))
+		hc->codec_param[instance].disp_imm = 1;
+	else
 		hc->codec_param[instance].disp_imm = 0;
+
+	if (params->codec_format == V4L2_PIX_FMT_HEVC && !CHECK_VERSION(iface, 1, 9))
+		hc->codec_param[instance].disp_imm = 0;
+
 	hc->codec_param[instance].dbglog_enable = 0;
 	iface->dbglog_desc.level = 0;
 
@@ -889,7 +908,8 @@ static void vpu_malone_unpack_seq_hdr(struct vpu_rpc_event *pkt,
 	info->frame_rate.numerator = 1000;
 	info->frame_rate.denominator = pkt->data[8];
 	info->dsp_asp_ratio = pkt->data[9];
-	info->level_idc = pkt->data[10];
+	info->profile_idc = (pkt->data[10] >> 8) & 0xff;
+	info->level_idc = pkt->data[10] & 0xff;
 	info->bit_depth_luma = pkt->data[13];
 	info->bit_depth_chroma = pkt->data[14];
 	info->chroma_fmt = pkt->data[15];
@@ -906,6 +926,8 @@ static void vpu_malone_unpack_seq_hdr(struct vpu_rpc_event *pkt,
 		info->pixfmt = V4L2_PIX_FMT_NV12M_10BE_8L128;
 	else
 		info->pixfmt = V4L2_PIX_FMT_NV12M_8L128;
+	if (pkt->hdr.num > 28)
+		info->constraint_set_flags = pkt->data[28];
 	if (info->frame_rate.numerator && info->frame_rate.denominator) {
 		unsigned long n, d;
 
@@ -1023,6 +1045,7 @@ static const struct malone_padding_scode padding_scodes[] = {
 	{SCODE_PADDING_EOS,      V4L2_PIX_FMT_JPEG,        {0x0, 0x0}},
 	{SCODE_PADDING_BUFFLUSH, V4L2_PIX_FMT_H264,        {0x15010000, 0x0}},
 	{SCODE_PADDING_BUFFLUSH, V4L2_PIX_FMT_H264_MVC,    {0x15010000, 0x0}},
+	{SCODE_PADDING_BUFFLUSH, V4L2_PIX_FMT_HEVC,        {0x3e010000, 0x20}},
 };
 
 static const struct malone_padding_scode padding_scode_dft = {0x0, 0x0};
@@ -1057,8 +1080,11 @@ static int vpu_malone_add_padding_scode(struct vpu_buffer *stream_buffer,
 	int ret;
 
 	ps = get_padding_scode(scode_type, pixelformat);
-	if (!ps)
+	if (!ps) {
+		if (scode_type == SCODE_PADDING_BUFFLUSH)
+			return 0;
 		return -EINVAL;
+	}
 
 	wptr = readl(&str_buf->wptr);
 	if (wptr < stream_buffer->phys || wptr > stream_buffer->phys + stream_buffer->length)
@@ -1311,22 +1337,18 @@ static int vpu_malone_insert_scode_vc1_g_seq(struct malone_scode_t *scode)
 {
 	if (!scode->inst->total_input_count)
 		return 0;
-	if (vpu_vb_is_codecconfig(to_vb2_v4l2_buffer(scode->vb)))
-		scode->need_data = 0;
 	return 0;
 }
 
 static int vpu_malone_insert_scode_vc1_g_pic(struct malone_scode_t *scode)
 {
-	struct vb2_v4l2_buffer *vbuf;
 	u8 nal_hdr[MALONE_VC1_NAL_HEADER_LEN];
 	u32 *data = NULL;
 	int ret;
 
-	vbuf = to_vb2_v4l2_buffer(scode->vb);
 	data = vb2_plane_vaddr(scode->vb, 0);
 
-	if (scode->inst->total_input_count == 0 || vpu_vb_is_codecconfig(vbuf))
+	if (scode->inst->total_input_count == 0)
 		return 0;
 	if (MALONE_VC1_CONTAIN_NAL(*data))
 		return 0;
@@ -1347,8 +1369,6 @@ static int vpu_malone_insert_scode_vc1_l_seq(struct malone_scode_t *scode)
 	int size = 0;
 	u8 rcv_seqhdr[MALONE_VC1_RCV_SEQ_HEADER_LEN];
 
-	if (vpu_vb_is_codecconfig(to_vb2_v4l2_buffer(scode->vb)))
-		scode->need_data = 0;
 	if (scode->inst->total_input_count)
 		return 0;
 	scode->need_data = 0;
@@ -1534,7 +1554,7 @@ static int vpu_malone_input_frame_data(struct vpu_malone_str_buffer __iomem *str
 	scode.vb = vb;
 	scode.wptr = wptr;
 	scode.need_data = 1;
-	if (vbuf->sequence == 0 || vpu_vb_is_codecconfig(vbuf))
+	if (vbuf->sequence == 0)
 		ret = vpu_malone_insert_scode(&scode, SCODE_SEQUENCE);
 
 	if (ret < 0)
@@ -1562,7 +1582,15 @@ static int vpu_malone_input_frame_data(struct vpu_malone_str_buffer __iomem *str
 
 	vpu_malone_update_wptr(str_buf, wptr);
 
-	if (disp_imm && !vpu_vb_is_codecconfig(vbuf)) {
+	/*
+	 * Enable the low latency flush mode if display delay is set to 0
+	 * or the low latency frame flush mode if it is set to 1.
+	 * The low latency flush mode requires some padding data to be appended to each frame,
+	 * but there must not be any padding data between the sequence header and the frame.
+	 * This module is currently only supported for the H264 and HEVC formats,
+	 * for other formats, vpu_malone_add_scode() will return 0.
+	 */
+	if (disp_imm || low_latency) {
 		ret = vpu_malone_add_scode(inst->core->iface,
 					   inst->id,
 					   &inst->stream_buffer,
@@ -1609,7 +1637,6 @@ int vpu_malone_input_frame(struct vpu_shared_addr *shared,
 			   struct vpu_inst *inst, struct vb2_buffer *vb)
 {
 	struct vpu_dec_ctrl *hc = shared->priv;
-	struct vb2_v4l2_buffer *vbuf;
 	struct vpu_malone_str_buffer __iomem *str_buf = hc->str_buf[inst->id];
 	u32 disp_imm = hc->codec_param[inst->id].disp_imm;
 	u32 size;
@@ -1623,16 +1650,6 @@ int vpu_malone_input_frame(struct vpu_shared_addr *shared,
 		return ret;
 	size = ret;
 
-	/*
-	 * if buffer only contain codec data, and the timestamp is invalid,
-	 * don't put the invalid timestamp to resync
-	 * merge the data to next frame
-	 */
-	vbuf = to_vb2_v4l2_buffer(vb);
-	if (vpu_vb_is_codecconfig(vbuf)) {
-		inst->extra_size += size;
-		return 0;
-	}
 	if (inst->extra_size) {
 		size += inst->extra_size;
 		inst->extra_size = 0;