linux-zen-desktop/drivers/media/platform/mediatek/mdp3/mtk-mdp3-comp.c

1209 lines
34 KiB
C

// SPDX-License-Identifier: GPL-2.0-only
/*
* Copyright (c) 2022 MediaTek Inc.
* Author: Ping-Hsun Wu <ping-hsun.wu@mediatek.com>
*/
#include <linux/clk.h>
#include <linux/of_platform.h>
#include <linux/of_address.h>
#include <linux/pm_runtime.h>
#include "mtk-mdp3-cfg.h"
#include "mtk-mdp3-comp.h"
#include "mtk-mdp3-core.h"
#include "mtk-mdp3-regs.h"
#include "mdp_reg_rdma.h"
#include "mdp_reg_ccorr.h"
#include "mdp_reg_rsz.h"
#include "mdp_reg_wrot.h"
#include "mdp_reg_wdma.h"
static u32 mdp_comp_alias_id[MDP_COMP_TYPE_COUNT];
static int p_id;
static inline const struct mdp_platform_config *
__get_plat_cfg(const struct mdp_comp_ctx *ctx)
{
if (!ctx)
return NULL;
return ctx->comp->mdp_dev->mdp_data->mdp_cfg;
}
static s64 get_comp_flag(const struct mdp_comp_ctx *ctx)
{
const struct mdp_platform_config *mdp_cfg = __get_plat_cfg(ctx);
u32 rdma0, rsz1;
rdma0 = mdp_cfg_get_id_inner(ctx->comp->mdp_dev, MDP_COMP_RDMA0);
rsz1 = mdp_cfg_get_id_inner(ctx->comp->mdp_dev, MDP_COMP_RSZ1);
if (!rdma0 || !rsz1)
return MDP_COMP_NONE;
if (mdp_cfg && mdp_cfg->rdma_rsz1_sram_sharing)
if (ctx->comp->inner_id == rdma0)
return BIT(rdma0) | BIT(rsz1);
return BIT(ctx->comp->inner_id);
}
static int init_rdma(struct mdp_comp_ctx *ctx, struct mdp_cmdq_cmd *cmd)
{
const struct mdp_platform_config *mdp_cfg = __get_plat_cfg(ctx);
phys_addr_t base = ctx->comp->reg_base;
u8 subsys_id = ctx->comp->subsys_id;
s32 rdma0;
rdma0 = mdp_cfg_get_id_inner(ctx->comp->mdp_dev, MDP_COMP_RDMA0);
if (!rdma0)
return -EINVAL;
if (mdp_cfg && mdp_cfg->rdma_support_10bit) {
struct mdp_comp *prz1 = ctx->comp->mdp_dev->comp[MDP_COMP_RSZ1];
/* Disable RSZ1 */
if (ctx->comp->inner_id == rdma0 && prz1)
MM_REG_WRITE(cmd, subsys_id, prz1->reg_base, PRZ_ENABLE,
0x0, BIT(0));
}
/* Reset RDMA */
MM_REG_WRITE(cmd, subsys_id, base, MDP_RDMA_RESET, BIT(0), BIT(0));
MM_REG_POLL(cmd, subsys_id, base, MDP_RDMA_MON_STA_1, BIT(8), BIT(8));
MM_REG_WRITE(cmd, subsys_id, base, MDP_RDMA_RESET, 0x0, BIT(0));
return 0;
}
static int config_rdma_frame(struct mdp_comp_ctx *ctx,
struct mdp_cmdq_cmd *cmd,
const struct v4l2_rect *compose)
{
const struct mdp_platform_config *mdp_cfg = __get_plat_cfg(ctx);
u32 colorformat = ctx->input->buffer.format.colorformat;
bool block10bit = MDP_COLOR_IS_10BIT_PACKED(colorformat);
bool en_ufo = MDP_COLOR_IS_UFP(colorformat);
phys_addr_t base = ctx->comp->reg_base;
u8 subsys_id = ctx->comp->subsys_id;
u32 reg = 0;
if (mdp_cfg && mdp_cfg->rdma_support_10bit) {
if (block10bit)
MM_REG_WRITE(cmd, subsys_id, base,
MDP_RDMA_RESV_DUMMY_0, 0x7, 0x7);
else
MM_REG_WRITE(cmd, subsys_id, base,
MDP_RDMA_RESV_DUMMY_0, 0x0, 0x7);
}
/* Setup smi control */
MM_REG_WRITE(cmd, subsys_id, base, MDP_RDMA_GMCIF_CON,
(7 << 4) + //burst type to 8
(1 << 16), //enable pre-ultra
0x00030071);
/* Setup source frame info */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, rdma.src_ctrl);
MM_REG_WRITE(cmd, subsys_id, base, MDP_RDMA_SRC_CON, reg,
0x03C8FE0F);
if (mdp_cfg)
if (mdp_cfg->rdma_support_10bit && en_ufo) {
/* Setup source buffer base */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, rdma.ufo_dec_y);
MM_REG_WRITE(cmd, subsys_id,
base, MDP_RDMA_UFO_DEC_LENGTH_BASE_Y,
reg, 0xFFFFFFFF);
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, rdma.ufo_dec_c);
MM_REG_WRITE(cmd, subsys_id,
base, MDP_RDMA_UFO_DEC_LENGTH_BASE_C,
reg, 0xFFFFFFFF);
/* Set 10bit source frame pitch */
if (block10bit) {
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, rdma.mf_bkgd_in_pxl);
MM_REG_WRITE(cmd, subsys_id,
base, MDP_RDMA_MF_BKGD_SIZE_IN_PXL,
reg, 0x001FFFFF);
}
}
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, rdma.control);
MM_REG_WRITE(cmd, subsys_id, base, MDP_RDMA_CON, reg,
0x1110);
/* Setup source buffer base */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, rdma.iova[0]);
MM_REG_WRITE(cmd, subsys_id, base, MDP_RDMA_SRC_BASE_0, reg,
0xFFFFFFFF);
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, rdma.iova[1]);
MM_REG_WRITE(cmd, subsys_id, base, MDP_RDMA_SRC_BASE_1, reg,
0xFFFFFFFF);
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, rdma.iova[2]);
MM_REG_WRITE(cmd, subsys_id, base, MDP_RDMA_SRC_BASE_2, reg,
0xFFFFFFFF);
/* Setup source buffer end */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, rdma.iova_end[0]);
MM_REG_WRITE(cmd, subsys_id, base, MDP_RDMA_SRC_END_0,
reg, 0xFFFFFFFF);
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, rdma.iova_end[1]);
MM_REG_WRITE(cmd, subsys_id, base, MDP_RDMA_SRC_END_1,
reg, 0xFFFFFFFF);
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, rdma.iova_end[2]);
MM_REG_WRITE(cmd, subsys_id, base, MDP_RDMA_SRC_END_2,
reg, 0xFFFFFFFF);
/* Setup source frame pitch */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, rdma.mf_bkgd);
MM_REG_WRITE(cmd, subsys_id, base, MDP_RDMA_MF_BKGD_SIZE_IN_BYTE,
reg, 0x001FFFFF);
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, rdma.sf_bkgd);
MM_REG_WRITE(cmd, subsys_id, base, MDP_RDMA_SF_BKGD_SIZE_IN_BYTE,
reg, 0x001FFFFF);
/* Setup color transform */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, rdma.transform);
MM_REG_WRITE(cmd, subsys_id, base, MDP_RDMA_TRANSFORM_0,
reg, 0x0F110000);
return 0;
}
static int config_rdma_subfrm(struct mdp_comp_ctx *ctx,
struct mdp_cmdq_cmd *cmd, u32 index)
{
const struct mdp_platform_config *mdp_cfg = __get_plat_cfg(ctx);
u32 colorformat = ctx->input->buffer.format.colorformat;
bool block10bit = MDP_COLOR_IS_10BIT_PACKED(colorformat);
bool en_ufo = MDP_COLOR_IS_UFP(colorformat);
phys_addr_t base = ctx->comp->reg_base;
u8 subsys_id = ctx->comp->subsys_id;
u32 csf_l = 0, csf_r = 0;
u32 reg = 0;
/* Enable RDMA */
MM_REG_WRITE(cmd, subsys_id, base, MDP_RDMA_EN, BIT(0), BIT(0));
/* Set Y pixel offset */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, rdma.subfrms[index].offset[0]);
MM_REG_WRITE(cmd, subsys_id, base, MDP_RDMA_SRC_OFFSET_0,
reg, 0xFFFFFFFF);
/* Set 10bit UFO mode */
if (mdp_cfg) {
if (mdp_cfg->rdma_support_10bit && block10bit && en_ufo) {
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, rdma.subfrms[index].offset_0_p);
MM_REG_WRITE(cmd, subsys_id, base,
MDP_RDMA_SRC_OFFSET_0_P,
reg, 0xFFFFFFFF);
}
}
/* Set U pixel offset */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, rdma.subfrms[index].offset[1]);
MM_REG_WRITE(cmd, subsys_id, base, MDP_RDMA_SRC_OFFSET_1,
reg, 0xFFFFFFFF);
/* Set V pixel offset */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, rdma.subfrms[index].offset[2]);
MM_REG_WRITE(cmd, subsys_id, base, MDP_RDMA_SRC_OFFSET_2,
reg, 0xFFFFFFFF);
/* Set source size */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, rdma.subfrms[index].src);
MM_REG_WRITE(cmd, subsys_id, base, MDP_RDMA_MF_SRC_SIZE, reg,
0x1FFF1FFF);
/* Set target size */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, rdma.subfrms[index].clip);
MM_REG_WRITE(cmd, subsys_id, base, MDP_RDMA_MF_CLIP_SIZE,
reg, 0x1FFF1FFF);
/* Set crop offset */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, rdma.subfrms[index].clip_ofst);
MM_REG_WRITE(cmd, subsys_id, base, MDP_RDMA_MF_OFFSET_1,
reg, 0x003F001F);
if (CFG_CHECK(MT8183, p_id)) {
csf_l = CFG_COMP(MT8183, ctx->param, subfrms[index].in.left);
csf_r = CFG_COMP(MT8183, ctx->param, subfrms[index].in.right);
}
if (mdp_cfg && mdp_cfg->rdma_upsample_repeat_only)
if ((csf_r - csf_l + 1) > 320)
MM_REG_WRITE(cmd, subsys_id, base,
MDP_RDMA_RESV_DUMMY_0, BIT(2), BIT(2));
return 0;
}
static int wait_rdma_event(struct mdp_comp_ctx *ctx, struct mdp_cmdq_cmd *cmd)
{
struct device *dev = &ctx->comp->mdp_dev->pdev->dev;
phys_addr_t base = ctx->comp->reg_base;
u8 subsys_id = ctx->comp->subsys_id;
if (ctx->comp->alias_id == 0)
MM_REG_WAIT(cmd, ctx->comp->gce_event[MDP_GCE_EVENT_EOF]);
else
dev_err(dev, "Do not support RDMA1_DONE event\n");
/* Disable RDMA */
MM_REG_WRITE(cmd, subsys_id, base, MDP_RDMA_EN, 0x0, BIT(0));
return 0;
}
static const struct mdp_comp_ops rdma_ops = {
.get_comp_flag = get_comp_flag,
.init_comp = init_rdma,
.config_frame = config_rdma_frame,
.config_subfrm = config_rdma_subfrm,
.wait_comp_event = wait_rdma_event,
};
static int init_rsz(struct mdp_comp_ctx *ctx, struct mdp_cmdq_cmd *cmd)
{
phys_addr_t base = ctx->comp->reg_base;
u8 subsys_id = ctx->comp->subsys_id;
/* Reset RSZ */
MM_REG_WRITE(cmd, subsys_id, base, PRZ_ENABLE, 0x10000, BIT(16));
MM_REG_WRITE(cmd, subsys_id, base, PRZ_ENABLE, 0x0, BIT(16));
/* Enable RSZ */
MM_REG_WRITE(cmd, subsys_id, base, PRZ_ENABLE, BIT(0), BIT(0));
return 0;
}
static int config_rsz_frame(struct mdp_comp_ctx *ctx,
struct mdp_cmdq_cmd *cmd,
const struct v4l2_rect *compose)
{
phys_addr_t base = ctx->comp->reg_base;
u8 subsys_id = ctx->comp->subsys_id;
bool bypass = FALSE;
u32 reg = 0;
if (CFG_CHECK(MT8183, p_id))
bypass = CFG_COMP(MT8183, ctx->param, frame.bypass);
if (bypass) {
/* Disable RSZ */
MM_REG_WRITE(cmd, subsys_id, base, PRZ_ENABLE, 0x0, BIT(0));
return 0;
}
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, rsz.control1);
MM_REG_WRITE(cmd, subsys_id, base, PRZ_CONTROL_1, reg,
0x03FFFDF3);
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, rsz.control2);
MM_REG_WRITE(cmd, subsys_id, base, PRZ_CONTROL_2, reg,
0x0FFFC290);
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, rsz.coeff_step_x);
MM_REG_WRITE(cmd, subsys_id, base, PRZ_HORIZONTAL_COEFF_STEP,
reg, 0x007FFFFF);
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, rsz.coeff_step_y);
MM_REG_WRITE(cmd, subsys_id, base, PRZ_VERTICAL_COEFF_STEP,
reg, 0x007FFFFF);
return 0;
}
static int config_rsz_subfrm(struct mdp_comp_ctx *ctx,
struct mdp_cmdq_cmd *cmd, u32 index)
{
const struct mdp_platform_config *mdp_cfg = __get_plat_cfg(ctx);
phys_addr_t base = ctx->comp->reg_base;
u8 subsys_id = ctx->comp->subsys_id;
u32 csf_l = 0, csf_r = 0;
u32 reg = 0;
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, rsz.subfrms[index].control2);
MM_REG_WRITE(cmd, subsys_id, base, PRZ_CONTROL_2, reg,
0x00003800);
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, rsz.subfrms[index].src);
MM_REG_WRITE(cmd, subsys_id, base, PRZ_INPUT_IMAGE, reg,
0xFFFFFFFF);
if (CFG_CHECK(MT8183, p_id)) {
csf_l = CFG_COMP(MT8183, ctx->param, subfrms[index].in.left);
csf_r = CFG_COMP(MT8183, ctx->param, subfrms[index].in.right);
}
if (mdp_cfg && mdp_cfg->rsz_disable_dcm_small_sample)
if ((csf_r - csf_l + 1) <= 16)
MM_REG_WRITE(cmd, subsys_id, base, PRZ_CONTROL_1,
BIT(27), BIT(27));
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, subfrms[index].luma.left);
MM_REG_WRITE(cmd, subsys_id, base, PRZ_LUMA_HORIZONTAL_INTEGER_OFFSET,
reg, 0xFFFF);
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, subfrms[index].luma.left_subpix);
MM_REG_WRITE(cmd, subsys_id,
base, PRZ_LUMA_HORIZONTAL_SUBPIXEL_OFFSET,
reg, 0x1FFFFF);
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, subfrms[index].luma.top);
MM_REG_WRITE(cmd, subsys_id, base, PRZ_LUMA_VERTICAL_INTEGER_OFFSET,
reg, 0xFFFF);
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, subfrms[index].luma.top_subpix);
MM_REG_WRITE(cmd, subsys_id, base, PRZ_LUMA_VERTICAL_SUBPIXEL_OFFSET,
reg, 0x1FFFFF);
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, subfrms[index].chroma.left);
MM_REG_WRITE(cmd, subsys_id,
base, PRZ_CHROMA_HORIZONTAL_INTEGER_OFFSET,
reg, 0xFFFF);
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, subfrms[index].chroma.left_subpix);
MM_REG_WRITE(cmd, subsys_id,
base, PRZ_CHROMA_HORIZONTAL_SUBPIXEL_OFFSET,
reg, 0x1FFFFF);
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, rsz.subfrms[index].clip);
MM_REG_WRITE(cmd, subsys_id, base, PRZ_OUTPUT_IMAGE, reg,
0xFFFFFFFF);
return 0;
}
static int advance_rsz_subfrm(struct mdp_comp_ctx *ctx,
struct mdp_cmdq_cmd *cmd, u32 index)
{
const struct mdp_platform_config *mdp_cfg = __get_plat_cfg(ctx);
if (mdp_cfg && mdp_cfg->rsz_disable_dcm_small_sample) {
phys_addr_t base = ctx->comp->reg_base;
u8 subsys_id = ctx->comp->subsys_id;
u32 csf_l = 0, csf_r = 0;
if (CFG_CHECK(MT8183, p_id)) {
csf_l = CFG_COMP(MT8183, ctx->param, subfrms[index].in.left);
csf_r = CFG_COMP(MT8183, ctx->param, subfrms[index].in.right);
}
if ((csf_r - csf_l + 1) <= 16)
MM_REG_WRITE(cmd, subsys_id, base, PRZ_CONTROL_1, 0x0,
BIT(27));
}
return 0;
}
static const struct mdp_comp_ops rsz_ops = {
.get_comp_flag = get_comp_flag,
.init_comp = init_rsz,
.config_frame = config_rsz_frame,
.config_subfrm = config_rsz_subfrm,
.advance_subfrm = advance_rsz_subfrm,
};
static int init_wrot(struct mdp_comp_ctx *ctx, struct mdp_cmdq_cmd *cmd)
{
phys_addr_t base = ctx->comp->reg_base;
u8 subsys_id = ctx->comp->subsys_id;
/* Reset WROT */
MM_REG_WRITE(cmd, subsys_id, base, VIDO_SOFT_RST, BIT(0), BIT(0));
MM_REG_POLL(cmd, subsys_id, base, VIDO_SOFT_RST_STAT, BIT(0), BIT(0));
MM_REG_WRITE(cmd, subsys_id, base, VIDO_SOFT_RST, 0x0, BIT(0));
MM_REG_POLL(cmd, subsys_id, base, VIDO_SOFT_RST_STAT, 0x0, BIT(0));
return 0;
}
static int config_wrot_frame(struct mdp_comp_ctx *ctx,
struct mdp_cmdq_cmd *cmd,
const struct v4l2_rect *compose)
{
const struct mdp_platform_config *mdp_cfg = __get_plat_cfg(ctx);
phys_addr_t base = ctx->comp->reg_base;
u8 subsys_id = ctx->comp->subsys_id;
u32 reg = 0;
/* Write frame base address */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wrot.iova[0]);
MM_REG_WRITE(cmd, subsys_id, base, VIDO_BASE_ADDR, reg,
0xFFFFFFFF);
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wrot.iova[1]);
MM_REG_WRITE(cmd, subsys_id, base, VIDO_BASE_ADDR_C, reg,
0xFFFFFFFF);
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wrot.iova[2]);
MM_REG_WRITE(cmd, subsys_id, base, VIDO_BASE_ADDR_V, reg,
0xFFFFFFFF);
/* Write frame related registers */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wrot.control);
MM_REG_WRITE(cmd, subsys_id, base, VIDO_CTRL, reg,
0xF131510F);
/* Write frame Y pitch */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wrot.stride[0]);
MM_REG_WRITE(cmd, subsys_id, base, VIDO_STRIDE, reg,
0x0000FFFF);
/* Write frame UV pitch */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wrot.stride[1]);
MM_REG_WRITE(cmd, subsys_id, base, VIDO_STRIDE_C, reg,
0xFFFF);
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wrot.stride[2]);
MM_REG_WRITE(cmd, subsys_id, base, VIDO_STRIDE_V, reg,
0xFFFF);
/* Write matrix control */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wrot.mat_ctrl);
MM_REG_WRITE(cmd, subsys_id, base, VIDO_MAT_CTRL, reg, 0xF3);
/* Set the fixed ALPHA as 0xFF */
MM_REG_WRITE(cmd, subsys_id, base, VIDO_DITHER, 0xFF000000,
0xFF000000);
/* Set VIDO_EOL_SEL */
MM_REG_WRITE(cmd, subsys_id, base, VIDO_RSV_1, BIT(31), BIT(31));
/* Set VIDO_FIFO_TEST */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wrot.fifo_test);
if (reg != 0)
MM_REG_WRITE(cmd, subsys_id, base, VIDO_FIFO_TEST,
reg, 0xFFF);
/* Filter enable */
if (mdp_cfg && mdp_cfg->wrot_filter_constraint) {
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wrot.filter);
MM_REG_WRITE(cmd, subsys_id, base, VIDO_MAIN_BUF_SIZE,
reg, 0x77);
}
return 0;
}
static int config_wrot_subfrm(struct mdp_comp_ctx *ctx,
struct mdp_cmdq_cmd *cmd, u32 index)
{
phys_addr_t base = ctx->comp->reg_base;
u8 subsys_id = ctx->comp->subsys_id;
u32 reg = 0;
/* Write Y pixel offset */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wrot.subfrms[index].offset[0]);
MM_REG_WRITE(cmd, subsys_id, base, VIDO_OFST_ADDR,
reg, 0x0FFFFFFF);
/* Write U pixel offset */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wrot.subfrms[index].offset[1]);
MM_REG_WRITE(cmd, subsys_id, base, VIDO_OFST_ADDR_C,
reg, 0x0FFFFFFF);
/* Write V pixel offset */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wrot.subfrms[index].offset[2]);
MM_REG_WRITE(cmd, subsys_id, base, VIDO_OFST_ADDR_V,
reg, 0x0FFFFFFF);
/* Write source size */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wrot.subfrms[index].src);
MM_REG_WRITE(cmd, subsys_id, base, VIDO_IN_SIZE, reg,
0x1FFF1FFF);
/* Write target size */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wrot.subfrms[index].clip);
MM_REG_WRITE(cmd, subsys_id, base, VIDO_TAR_SIZE, reg,
0x1FFF1FFF);
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wrot.subfrms[index].clip_ofst);
MM_REG_WRITE(cmd, subsys_id, base, VIDO_CROP_OFST, reg,
0x1FFF1FFF);
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wrot.subfrms[index].main_buf);
MM_REG_WRITE(cmd, subsys_id, base, VIDO_MAIN_BUF_SIZE,
reg, 0x1FFF7F00);
/* Enable WROT */
MM_REG_WRITE(cmd, subsys_id, base, VIDO_ROT_EN, BIT(0), BIT(0));
return 0;
}
static int wait_wrot_event(struct mdp_comp_ctx *ctx, struct mdp_cmdq_cmd *cmd)
{
const struct mdp_platform_config *mdp_cfg = __get_plat_cfg(ctx);
struct device *dev = &ctx->comp->mdp_dev->pdev->dev;
phys_addr_t base = ctx->comp->reg_base;
u8 subsys_id = ctx->comp->subsys_id;
if (ctx->comp->alias_id == 0)
MM_REG_WAIT(cmd, ctx->comp->gce_event[MDP_GCE_EVENT_EOF]);
else
dev_err(dev, "Do not support WROT1_DONE event\n");
if (mdp_cfg && mdp_cfg->wrot_filter_constraint)
MM_REG_WRITE(cmd, subsys_id, base, VIDO_MAIN_BUF_SIZE, 0x0,
0x77);
/* Disable WROT */
MM_REG_WRITE(cmd, subsys_id, base, VIDO_ROT_EN, 0x0, BIT(0));
return 0;
}
static const struct mdp_comp_ops wrot_ops = {
.get_comp_flag = get_comp_flag,
.init_comp = init_wrot,
.config_frame = config_wrot_frame,
.config_subfrm = config_wrot_subfrm,
.wait_comp_event = wait_wrot_event,
};
static int init_wdma(struct mdp_comp_ctx *ctx, struct mdp_cmdq_cmd *cmd)
{
phys_addr_t base = ctx->comp->reg_base;
u8 subsys_id = ctx->comp->subsys_id;
/* Reset WDMA */
MM_REG_WRITE(cmd, subsys_id, base, WDMA_RST, BIT(0), BIT(0));
MM_REG_POLL(cmd, subsys_id, base, WDMA_FLOW_CTRL_DBG, BIT(0), BIT(0));
MM_REG_WRITE(cmd, subsys_id, base, WDMA_RST, 0x0, BIT(0));
return 0;
}
static int config_wdma_frame(struct mdp_comp_ctx *ctx,
struct mdp_cmdq_cmd *cmd,
const struct v4l2_rect *compose)
{
phys_addr_t base = ctx->comp->reg_base;
u8 subsys_id = ctx->comp->subsys_id;
u32 reg = 0;
MM_REG_WRITE(cmd, subsys_id, base, WDMA_BUF_CON2, 0x10101050,
0xFFFFFFFF);
/* Setup frame information */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wdma.wdma_cfg);
MM_REG_WRITE(cmd, subsys_id, base, WDMA_CFG, reg,
0x0F01B8F0);
/* Setup frame base address */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wdma.iova[0]);
MM_REG_WRITE(cmd, subsys_id, base, WDMA_DST_ADDR, reg,
0xFFFFFFFF);
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wdma.iova[1]);
MM_REG_WRITE(cmd, subsys_id, base, WDMA_DST_U_ADDR, reg,
0xFFFFFFFF);
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wdma.iova[2]);
MM_REG_WRITE(cmd, subsys_id, base, WDMA_DST_V_ADDR, reg,
0xFFFFFFFF);
/* Setup Y pitch */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wdma.w_in_byte);
MM_REG_WRITE(cmd, subsys_id, base, WDMA_DST_W_IN_BYTE,
reg, 0x0000FFFF);
/* Setup UV pitch */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wdma.uv_stride);
MM_REG_WRITE(cmd, subsys_id, base, WDMA_DST_UV_PITCH,
reg, 0x0000FFFF);
/* Set the fixed ALPHA as 0xFF */
MM_REG_WRITE(cmd, subsys_id, base, WDMA_ALPHA, 0x800000FF,
0x800000FF);
return 0;
}
static int config_wdma_subfrm(struct mdp_comp_ctx *ctx,
struct mdp_cmdq_cmd *cmd, u32 index)
{
phys_addr_t base = ctx->comp->reg_base;
u8 subsys_id = ctx->comp->subsys_id;
u32 reg = 0;
/* Write Y pixel offset */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wdma.subfrms[index].offset[0]);
MM_REG_WRITE(cmd, subsys_id, base, WDMA_DST_ADDR_OFFSET,
reg, 0x0FFFFFFF);
/* Write U pixel offset */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wdma.subfrms[index].offset[1]);
MM_REG_WRITE(cmd, subsys_id, base, WDMA_DST_U_ADDR_OFFSET,
reg, 0x0FFFFFFF);
/* Write V pixel offset */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wdma.subfrms[index].offset[2]);
MM_REG_WRITE(cmd, subsys_id, base, WDMA_DST_V_ADDR_OFFSET,
reg, 0x0FFFFFFF);
/* Write source size */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wdma.subfrms[index].src);
MM_REG_WRITE(cmd, subsys_id, base, WDMA_SRC_SIZE, reg,
0x3FFF3FFF);
/* Write target size */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wdma.subfrms[index].clip);
MM_REG_WRITE(cmd, subsys_id, base, WDMA_CLIP_SIZE, reg,
0x3FFF3FFF);
/* Write clip offset */
if (CFG_CHECK(MT8183, p_id))
reg = CFG_COMP(MT8183, ctx->param, wdma.subfrms[index].clip_ofst);
MM_REG_WRITE(cmd, subsys_id, base, WDMA_CLIP_COORD, reg,
0x3FFF3FFF);
/* Enable WDMA */
MM_REG_WRITE(cmd, subsys_id, base, WDMA_EN, BIT(0), BIT(0));
return 0;
}
static int wait_wdma_event(struct mdp_comp_ctx *ctx, struct mdp_cmdq_cmd *cmd)
{
phys_addr_t base = ctx->comp->reg_base;
u8 subsys_id = ctx->comp->subsys_id;
MM_REG_WAIT(cmd, ctx->comp->gce_event[MDP_GCE_EVENT_EOF]);
/* Disable WDMA */
MM_REG_WRITE(cmd, subsys_id, base, WDMA_EN, 0x0, BIT(0));
return 0;
}
static const struct mdp_comp_ops wdma_ops = {
.get_comp_flag = get_comp_flag,
.init_comp = init_wdma,
.config_frame = config_wdma_frame,
.config_subfrm = config_wdma_subfrm,
.wait_comp_event = wait_wdma_event,
};
static int init_ccorr(struct mdp_comp_ctx *ctx, struct mdp_cmdq_cmd *cmd)
{
phys_addr_t base = ctx->comp->reg_base;
u8 subsys_id = ctx->comp->subsys_id;
/* CCORR enable */
MM_REG_WRITE(cmd, subsys_id, base, MDP_CCORR_EN, BIT(0), BIT(0));
/* Relay mode */
MM_REG_WRITE(cmd, subsys_id, base, MDP_CCORR_CFG, BIT(0), BIT(0));
return 0;
}
static int config_ccorr_subfrm(struct mdp_comp_ctx *ctx,
struct mdp_cmdq_cmd *cmd, u32 index)
{
phys_addr_t base = ctx->comp->reg_base;
u8 subsys_id = ctx->comp->subsys_id;
u32 csf_l = 0, csf_r = 0;
u32 csf_t = 0, csf_b = 0;
u32 hsize, vsize;
if (CFG_CHECK(MT8183, p_id)) {
csf_l = CFG_COMP(MT8183, ctx->param, subfrms[index].in.left);
csf_r = CFG_COMP(MT8183, ctx->param, subfrms[index].in.right);
csf_t = CFG_COMP(MT8183, ctx->param, subfrms[index].in.top);
csf_b = CFG_COMP(MT8183, ctx->param, subfrms[index].in.bottom);
}
hsize = csf_r - csf_l + 1;
vsize = csf_b - csf_t + 1;
MM_REG_WRITE(cmd, subsys_id, base, MDP_CCORR_SIZE,
(hsize << 16) + (vsize << 0), 0x1FFF1FFF);
return 0;
}
static const struct mdp_comp_ops ccorr_ops = {
.get_comp_flag = get_comp_flag,
.init_comp = init_ccorr,
.config_subfrm = config_ccorr_subfrm,
};
static const struct mdp_comp_ops *mdp_comp_ops[MDP_COMP_TYPE_COUNT] = {
[MDP_COMP_TYPE_RDMA] = &rdma_ops,
[MDP_COMP_TYPE_RSZ] = &rsz_ops,
[MDP_COMP_TYPE_WROT] = &wrot_ops,
[MDP_COMP_TYPE_WDMA] = &wdma_ops,
[MDP_COMP_TYPE_CCORR] = &ccorr_ops,
};
static const struct of_device_id mdp_comp_dt_ids[] = {
{
.compatible = "mediatek,mt8183-mdp3-rdma",
.data = (void *)MDP_COMP_TYPE_RDMA,
}, {
.compatible = "mediatek,mt8183-mdp3-ccorr",
.data = (void *)MDP_COMP_TYPE_CCORR,
}, {
.compatible = "mediatek,mt8183-mdp3-rsz",
.data = (void *)MDP_COMP_TYPE_RSZ,
}, {
.compatible = "mediatek,mt8183-mdp3-wrot",
.data = (void *)MDP_COMP_TYPE_WROT,
}, {
.compatible = "mediatek,mt8183-mdp3-wdma",
.data = (void *)MDP_COMP_TYPE_WDMA,
},
{}
};
static inline bool is_dma_capable(const enum mdp_comp_type type)
{
return (type == MDP_COMP_TYPE_RDMA ||
type == MDP_COMP_TYPE_WROT ||
type == MDP_COMP_TYPE_WDMA);
}
static inline bool is_bypass_gce_event(const enum mdp_comp_type type)
{
/*
* Subcomponent PATH is only used for the direction of data flow and
* dose not need to wait for GCE event.
*/
return (type == MDP_COMP_TYPE_PATH);
}
static int mdp_comp_get_id(struct mdp_dev *mdp, enum mdp_comp_type type, u32 alias_id)
{
int i;
for (i = 0; i < mdp->mdp_data->comp_data_len; i++)
if (mdp->mdp_data->comp_data[i].match.type == type &&
mdp->mdp_data->comp_data[i].match.alias_id == alias_id)
return i;
return -ENODEV;
}
int mdp_comp_clock_on(struct device *dev, struct mdp_comp *comp)
{
int i, ret;
/* Only DMA capable components need the pm control */
if (comp->comp_dev && is_dma_capable(comp->type)) {
ret = pm_runtime_resume_and_get(comp->comp_dev);
if (ret < 0) {
dev_err(dev,
"Failed to get power, err %d. type:%d id:%d\n",
ret, comp->type, comp->inner_id);
return ret;
}
}
for (i = 0; i < comp->clk_num; i++) {
if (IS_ERR_OR_NULL(comp->clks[i]))
continue;
ret = clk_prepare_enable(comp->clks[i]);
if (ret) {
dev_err(dev,
"Failed to enable clk %d. type:%d id:%d\n",
i, comp->type, comp->inner_id);
goto err_revert;
}
}
return 0;
err_revert:
while (--i >= 0) {
if (IS_ERR_OR_NULL(comp->clks[i]))
continue;
clk_disable_unprepare(comp->clks[i]);
}
if (comp->comp_dev && is_dma_capable(comp->type))
pm_runtime_put_sync(comp->comp_dev);
return ret;
}
void mdp_comp_clock_off(struct device *dev, struct mdp_comp *comp)
{
int i;
for (i = 0; i < comp->clk_num; i++) {
if (IS_ERR_OR_NULL(comp->clks[i]))
continue;
clk_disable_unprepare(comp->clks[i]);
}
if (comp->comp_dev && is_dma_capable(comp->type))
pm_runtime_put(comp->comp_dev);
}
int mdp_comp_clocks_on(struct device *dev, struct mdp_comp *comps, int num)
{
int i, ret;
for (i = 0; i < num; i++) {
ret = mdp_comp_clock_on(dev, &comps[i]);
if (ret)
return ret;
}
return 0;
}
void mdp_comp_clocks_off(struct device *dev, struct mdp_comp *comps, int num)
{
int i;
for (i = 0; i < num; i++)
mdp_comp_clock_off(dev, &comps[i]);
}
static int mdp_get_subsys_id(struct mdp_dev *mdp, struct device *dev,
struct device_node *node, struct mdp_comp *comp)
{
struct platform_device *comp_pdev;
struct cmdq_client_reg cmdq_reg;
int ret = 0;
int index = 0;
if (!dev || !node || !comp)
return -EINVAL;
comp_pdev = of_find_device_by_node(node);
if (!comp_pdev) {
dev_err(dev, "get comp_pdev fail! comp public id=%d, inner id=%d, type=%d\n",
comp->public_id, comp->inner_id, comp->type);
return -ENODEV;
}
index = mdp->mdp_data->comp_data[comp->public_id].info.dts_reg_ofst;
ret = cmdq_dev_get_client_reg(&comp_pdev->dev, &cmdq_reg, index);
if (ret != 0) {
dev_err(&comp_pdev->dev, "cmdq_dev_get_subsys fail!\n");
put_device(&comp_pdev->dev);
return -EINVAL;
}
comp->subsys_id = cmdq_reg.subsys;
dev_dbg(&comp_pdev->dev, "subsys id=%d\n", cmdq_reg.subsys);
put_device(&comp_pdev->dev);
return 0;
}
static void __mdp_comp_init(struct mdp_dev *mdp, struct device_node *node,
struct mdp_comp *comp)
{
struct resource res;
phys_addr_t base;
int index;
index = mdp->mdp_data->comp_data[comp->public_id].info.dts_reg_ofst;
if (of_address_to_resource(node, index, &res) < 0)
base = 0L;
else
base = res.start;
comp->mdp_dev = mdp;
comp->regs = of_iomap(node, 0);
comp->reg_base = base;
}
static int mdp_comp_init(struct mdp_dev *mdp, struct device_node *node,
struct mdp_comp *comp, enum mtk_mdp_comp_id id)
{
struct device *dev = &mdp->pdev->dev;
struct platform_device *pdev_c;
int clk_ofst;
int i;
s32 event;
if (id < 0 || id >= MDP_MAX_COMP_COUNT) {
dev_err(dev, "Invalid component id %d\n", id);
return -EINVAL;
}
pdev_c = of_find_device_by_node(node);
if (!pdev_c) {
dev_warn(dev, "can't find platform device of node:%s\n",
node->name);
return -ENODEV;
}
comp->comp_dev = &pdev_c->dev;
comp->public_id = id;
comp->type = mdp->mdp_data->comp_data[id].match.type;
comp->inner_id = mdp->mdp_data->comp_data[id].match.inner_id;
comp->alias_id = mdp->mdp_data->comp_data[id].match.alias_id;
comp->ops = mdp_comp_ops[comp->type];
__mdp_comp_init(mdp, node, comp);
comp->clk_num = mdp->mdp_data->comp_data[id].info.clk_num;
comp->clks = devm_kzalloc(dev, sizeof(struct clk *) * comp->clk_num,
GFP_KERNEL);
if (!comp->clks)
return -ENOMEM;
clk_ofst = mdp->mdp_data->comp_data[id].info.clk_ofst;
for (i = 0; i < comp->clk_num; i++) {
comp->clks[i] = of_clk_get(node, i + clk_ofst);
if (IS_ERR(comp->clks[i]))
break;
}
mdp_get_subsys_id(mdp, dev, node, comp);
/* Set GCE SOF event */
if (is_bypass_gce_event(comp->type) ||
of_property_read_u32_index(node, "mediatek,gce-events",
MDP_GCE_EVENT_SOF, &event))
event = MDP_GCE_NO_EVENT;
comp->gce_event[MDP_GCE_EVENT_SOF] = event;
/* Set GCE EOF event */
if (is_dma_capable(comp->type)) {
if (of_property_read_u32_index(node, "mediatek,gce-events",
MDP_GCE_EVENT_EOF, &event)) {
dev_err(dev, "Component id %d has no EOF\n", id);
return -EINVAL;
}
} else {
event = MDP_GCE_NO_EVENT;
}
comp->gce_event[MDP_GCE_EVENT_EOF] = event;
return 0;
}
static void mdp_comp_deinit(struct mdp_comp *comp)
{
if (!comp)
return;
if (comp->comp_dev && comp->clks) {
devm_kfree(&comp->mdp_dev->pdev->dev, comp->clks);
comp->clks = NULL;
}
if (comp->regs)
iounmap(comp->regs);
}
static struct mdp_comp *mdp_comp_create(struct mdp_dev *mdp,
struct device_node *node,
enum mtk_mdp_comp_id id)
{
struct device *dev = &mdp->pdev->dev;
struct mdp_comp *comp;
int ret;
if (mdp->comp[id])
return ERR_PTR(-EEXIST);
comp = devm_kzalloc(dev, sizeof(*comp), GFP_KERNEL);
if (!comp)
return ERR_PTR(-ENOMEM);
ret = mdp_comp_init(mdp, node, comp, id);
if (ret) {
devm_kfree(dev, comp);
return ERR_PTR(ret);
}
mdp->comp[id] = comp;
mdp->comp[id]->mdp_dev = mdp;
dev_dbg(dev, "%s type:%d alias:%d public id:%d inner id:%d base:%#x regs:%p\n",
dev->of_node->name, comp->type, comp->alias_id, id, comp->inner_id,
(u32)comp->reg_base, comp->regs);
return comp;
}
static int mdp_comp_sub_create(struct mdp_dev *mdp)
{
struct device *dev = &mdp->pdev->dev;
struct device_node *node, *parent;
int ret = 0;
parent = dev->of_node->parent;
for_each_child_of_node(parent, node) {
const struct of_device_id *of_id;
enum mdp_comp_type type;
int id, alias_id;
struct mdp_comp *comp;
of_id = of_match_node(mdp->mdp_data->mdp_sub_comp_dt_ids, node);
if (!of_id)
continue;
if (!of_device_is_available(node)) {
dev_dbg(dev, "Skipping disabled sub comp. %pOF\n",
node);
continue;
}
type = (enum mdp_comp_type)(uintptr_t)of_id->data;
alias_id = mdp_comp_alias_id[type];
id = mdp_comp_get_id(mdp, type, alias_id);
if (id < 0) {
dev_err(dev,
"Fail to get sub comp. id: type %d alias %d\n",
type, alias_id);
ret = -EINVAL;
goto err_free_node;
}
mdp_comp_alias_id[type]++;
comp = mdp_comp_create(mdp, node, id);
if (IS_ERR(comp)) {
ret = PTR_ERR(comp);
goto err_free_node;
}
}
return ret;
err_free_node:
of_node_put(node);
return ret;
}
void mdp_comp_destroy(struct mdp_dev *mdp)
{
int i;
for (i = 0; i < ARRAY_SIZE(mdp->comp); i++) {
if (mdp->comp[i]) {
if (is_dma_capable(mdp->comp[i]->type))
pm_runtime_disable(mdp->comp[i]->comp_dev);
mdp_comp_deinit(mdp->comp[i]);
devm_kfree(mdp->comp[i]->comp_dev, mdp->comp[i]);
mdp->comp[i] = NULL;
}
}
}
int mdp_comp_config(struct mdp_dev *mdp)
{
struct device *dev = &mdp->pdev->dev;
struct device_node *node, *parent;
int ret;
memset(mdp_comp_alias_id, 0, sizeof(mdp_comp_alias_id));
p_id = mdp->mdp_data->mdp_plat_id;
parent = dev->of_node->parent;
/* Iterate over sibling MDP function blocks */
for_each_child_of_node(parent, node) {
const struct of_device_id *of_id;
enum mdp_comp_type type;
int id, alias_id;
struct mdp_comp *comp;
of_id = of_match_node(mdp_comp_dt_ids, node);
if (!of_id)
continue;
if (!of_device_is_available(node)) {
dev_dbg(dev, "Skipping disabled component %pOF\n",
node);
continue;
}
type = (enum mdp_comp_type)(uintptr_t)of_id->data;
alias_id = mdp_comp_alias_id[type];
id = mdp_comp_get_id(mdp, type, alias_id);
if (id < 0) {
dev_err(dev,
"Fail to get component id: type %d alias %d\n",
type, alias_id);
continue;
}
mdp_comp_alias_id[type]++;
comp = mdp_comp_create(mdp, node, id);
if (IS_ERR(comp)) {
ret = PTR_ERR(comp);
goto err_init_comps;
}
/* Only DMA capable components need the pm control */
if (!is_dma_capable(comp->type))
continue;
pm_runtime_enable(comp->comp_dev);
}
ret = mdp_comp_sub_create(mdp);
if (ret)
goto err_init_comps;
return 0;
err_init_comps:
mdp_comp_destroy(mdp);
return ret;
}
int mdp_comp_ctx_config(struct mdp_dev *mdp, struct mdp_comp_ctx *ctx,
const struct img_compparam *param,
const struct img_ipi_frameparam *frame)
{
struct device *dev = &mdp->pdev->dev;
enum mtk_mdp_comp_id public_id = MDP_COMP_NONE;
u32 arg;
int i, idx;
if (!param) {
dev_err(dev, "Invalid component param");
return -EINVAL;
}
if (CFG_CHECK(MT8183, p_id))
arg = CFG_COMP(MT8183, param, type);
else
return -EINVAL;
public_id = mdp_cfg_get_id_public(mdp, arg);
if (public_id < 0) {
dev_err(dev, "Invalid component id %d", public_id);
return -EINVAL;
}
ctx->comp = mdp->comp[public_id];
if (!ctx->comp) {
dev_err(dev, "Uninit component inner id %d", arg);
return -EINVAL;
}
ctx->param = param;
if (CFG_CHECK(MT8183, p_id))
arg = CFG_COMP(MT8183, param, input);
else
return -EINVAL;
ctx->input = &frame->inputs[arg];
if (CFG_CHECK(MT8183, p_id))
idx = CFG_COMP(MT8183, param, num_outputs);
else
return -EINVAL;
for (i = 0; i < idx; i++) {
if (CFG_CHECK(MT8183, p_id))
arg = CFG_COMP(MT8183, param, outputs[i]);
else
return -EINVAL;
ctx->outputs[i] = &frame->outputs[arg];
}
return 0;
}