1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Hantro G1 post-processor support 4 * 5 * Copyright (C) 2019 Collabora, Ltd. 6 */ 7 8 #include <linux/dma-mapping.h> 9 #include <linux/types.h> 10 11 #include "hantro.h" 12 #include "hantro_hw.h" 13 #include "hantro_g1_regs.h" 14 #include "hantro_g2_regs.h" 15 #include "hantro_v4l2.h" 16 17 #define HANTRO_PP_REG_WRITE(vpu, reg_name, val) \ 18 { \ 19 hantro_reg_write(vpu, \ 20 &hantro_g1_postproc_regs.reg_name, \ 21 val); \ 22 } 23 24 #define HANTRO_PP_REG_WRITE_RELAXED(vpu, reg_name, val) \ 25 { \ 26 hantro_reg_write_relaxed(vpu, \ 27 &hantro_g1_postproc_regs.reg_name, \ 28 val); \ 29 } 30 31 #define VPU_PP_IN_YUYV 0x0 32 #define VPU_PP_IN_NV12 0x1 33 #define VPU_PP_IN_YUV420 0x2 34 #define VPU_PP_IN_YUV240_TILED 0x5 35 #define VPU_PP_OUT_RGB 0x0 36 #define VPU_PP_OUT_YUYV 0x3 37 38 static const struct hantro_postproc_regs hantro_g1_postproc_regs = { 39 .pipeline_en = {G1_REG_PP_INTERRUPT, 1, 0x1}, 40 .max_burst = {G1_REG_PP_DEV_CONFIG, 0, 0x1f}, 41 .clk_gate = {G1_REG_PP_DEV_CONFIG, 1, 0x1}, 42 .out_swap32 = {G1_REG_PP_DEV_CONFIG, 5, 0x1}, 43 .out_endian = {G1_REG_PP_DEV_CONFIG, 6, 0x1}, 44 .out_luma_base = {G1_REG_PP_OUT_LUMA_BASE, 0, 0xffffffff}, 45 .input_width = {G1_REG_PP_INPUT_SIZE, 0, 0x1ff}, 46 .input_height = {G1_REG_PP_INPUT_SIZE, 9, 0x1ff}, 47 .output_width = {G1_REG_PP_CONTROL, 4, 0x7ff}, 48 .output_height = {G1_REG_PP_CONTROL, 15, 0x7ff}, 49 .input_fmt = {G1_REG_PP_CONTROL, 29, 0x7}, 50 .output_fmt = {G1_REG_PP_CONTROL, 26, 0x7}, 51 .orig_width = {G1_REG_PP_MASK1_ORIG_WIDTH, 23, 0x1ff}, 52 .display_width = {G1_REG_PP_DISPLAY_WIDTH, 0, 0xfff}, 53 }; 54 55 bool hantro_needs_postproc(const struct hantro_ctx *ctx, 56 const struct hantro_fmt *fmt) 57 { 58 if (ctx->is_encoder) 59 return false; 60 61 if (ctx->need_postproc) 62 return true; 63 64 return fmt->postprocessed; 65 } 66 67 static void hantro_postproc_g1_enable(struct hantro_ctx *ctx) 68 { 69 struct hantro_dev *vpu = ctx->dev; 70 struct vb2_v4l2_buffer *dst_buf; 71 u32 src_pp_fmt, dst_pp_fmt; 72 dma_addr_t dst_dma; 73 74 /* Turn on pipeline mode. Must be done first. */ 75 HANTRO_PP_REG_WRITE(vpu, pipeline_en, 0x1); 76 77 src_pp_fmt = VPU_PP_IN_NV12; 78 79 switch (ctx->vpu_dst_fmt->fourcc) { 80 case V4L2_PIX_FMT_YUYV: 81 dst_pp_fmt = VPU_PP_OUT_YUYV; 82 break; 83 default: 84 WARN(1, "output format %d not supported by the post-processor, this wasn't expected.", 85 ctx->vpu_dst_fmt->fourcc); 86 dst_pp_fmt = 0; 87 break; 88 } 89 90 dst_buf = v4l2_m2m_next_dst_buf(ctx->fh.m2m_ctx); 91 dst_dma = vb2_dma_contig_plane_dma_addr(&dst_buf->vb2_buf, 0); 92 93 HANTRO_PP_REG_WRITE(vpu, clk_gate, 0x1); 94 HANTRO_PP_REG_WRITE(vpu, out_endian, 0x1); 95 HANTRO_PP_REG_WRITE(vpu, out_swap32, 0x1); 96 HANTRO_PP_REG_WRITE(vpu, max_burst, 16); 97 HANTRO_PP_REG_WRITE(vpu, out_luma_base, dst_dma); 98 HANTRO_PP_REG_WRITE(vpu, input_width, MB_WIDTH(ctx->dst_fmt.width)); 99 HANTRO_PP_REG_WRITE(vpu, input_height, MB_HEIGHT(ctx->dst_fmt.height)); 100 HANTRO_PP_REG_WRITE(vpu, input_fmt, src_pp_fmt); 101 HANTRO_PP_REG_WRITE(vpu, output_fmt, dst_pp_fmt); 102 HANTRO_PP_REG_WRITE(vpu, output_width, ctx->dst_fmt.width); 103 HANTRO_PP_REG_WRITE(vpu, output_height, ctx->dst_fmt.height); 104 HANTRO_PP_REG_WRITE(vpu, orig_width, MB_WIDTH(ctx->dst_fmt.width)); 105 HANTRO_PP_REG_WRITE(vpu, display_width, ctx->dst_fmt.width); 106 } 107 108 static int down_scale_factor(struct hantro_ctx *ctx) 109 { 110 if (ctx->src_fmt.width == ctx->dst_fmt.width) 111 return 0; 112 113 return DIV_ROUND_CLOSEST(ctx->src_fmt.width, ctx->dst_fmt.width); 114 } 115 116 static void hantro_postproc_g2_enable(struct hantro_ctx *ctx) 117 { 118 struct hantro_dev *vpu = ctx->dev; 119 struct vb2_v4l2_buffer *dst_buf; 120 int down_scale = down_scale_factor(ctx); 121 int out_depth; 122 size_t chroma_offset; 123 dma_addr_t dst_dma; 124 125 dst_buf = hantro_get_dst_buf(ctx); 126 dst_dma = vb2_dma_contig_plane_dma_addr(&dst_buf->vb2_buf, 0); 127 chroma_offset = ctx->dst_fmt.plane_fmt[0].bytesperline * 128 ctx->dst_fmt.height; 129 130 if (down_scale) { 131 hantro_reg_write(vpu, &g2_down_scale_e, 1); 132 hantro_reg_write(vpu, &g2_down_scale_y, down_scale >> 2); 133 hantro_reg_write(vpu, &g2_down_scale_x, down_scale >> 2); 134 hantro_write_addr(vpu, G2_DS_DST, dst_dma); 135 hantro_write_addr(vpu, G2_DS_DST_CHR, dst_dma + (chroma_offset >> down_scale)); 136 } else { 137 hantro_write_addr(vpu, G2_RS_OUT_LUMA_ADDR, dst_dma); 138 hantro_write_addr(vpu, G2_RS_OUT_CHROMA_ADDR, dst_dma + chroma_offset); 139 } 140 141 out_depth = hantro_get_format_depth(ctx->dst_fmt.pixelformat); 142 if (ctx->dev->variant->legacy_regs) { 143 u8 pp_shift = 0; 144 145 if (out_depth > 8) 146 pp_shift = 16 - out_depth; 147 148 hantro_reg_write(ctx->dev, &g2_rs_out_bit_depth, out_depth); 149 hantro_reg_write(ctx->dev, &g2_pp_pix_shift, pp_shift); 150 } else { 151 hantro_reg_write(vpu, &g2_output_8_bits, out_depth > 8 ? 0 : 1); 152 hantro_reg_write(vpu, &g2_output_format, out_depth > 8 ? 1 : 0); 153 } 154 hantro_reg_write(vpu, &g2_out_rs_e, 1); 155 } 156 157 static int hantro_postproc_g2_enum_framesizes(struct hantro_ctx *ctx, 158 struct v4l2_frmsizeenum *fsize) 159 { 160 /** 161 * G2 scaler can scale down by 0, 2, 4 or 8 162 * use fsize->index has power of 2 diviser 163 **/ 164 if (fsize->index > 3) 165 return -EINVAL; 166 167 if (!ctx->src_fmt.width || !ctx->src_fmt.height) 168 return -EINVAL; 169 170 fsize->type = V4L2_FRMSIZE_TYPE_DISCRETE; 171 fsize->discrete.width = ctx->src_fmt.width >> fsize->index; 172 fsize->discrete.height = ctx->src_fmt.height >> fsize->index; 173 174 return 0; 175 } 176 177 void hantro_postproc_free(struct hantro_ctx *ctx) 178 { 179 struct hantro_dev *vpu = ctx->dev; 180 unsigned int i; 181 182 for (i = 0; i < VB2_MAX_FRAME; ++i) { 183 struct hantro_aux_buf *priv = &ctx->postproc.dec_q[i]; 184 185 if (priv->cpu) { 186 dma_free_attrs(vpu->dev, priv->size, priv->cpu, 187 priv->dma, priv->attrs); 188 priv->cpu = NULL; 189 } 190 } 191 } 192 193 int hantro_postproc_alloc(struct hantro_ctx *ctx) 194 { 195 struct hantro_dev *vpu = ctx->dev; 196 struct v4l2_m2m_ctx *m2m_ctx = ctx->fh.m2m_ctx; 197 struct vb2_queue *cap_queue = &m2m_ctx->cap_q_ctx.q; 198 unsigned int num_buffers = cap_queue->num_buffers; 199 struct v4l2_pix_format_mplane pix_mp; 200 const struct hantro_fmt *fmt; 201 unsigned int i, buf_size; 202 203 /* this should always pick native format */ 204 fmt = hantro_get_default_fmt(ctx, false, ctx->bit_depth, HANTRO_AUTO_POSTPROC); 205 if (!fmt) 206 return -EINVAL; 207 v4l2_fill_pixfmt_mp(&pix_mp, fmt->fourcc, ctx->src_fmt.width, 208 ctx->src_fmt.height); 209 210 buf_size = pix_mp.plane_fmt[0].sizeimage; 211 if (ctx->vpu_src_fmt->fourcc == V4L2_PIX_FMT_H264_SLICE) 212 buf_size += hantro_h264_mv_size(pix_mp.width, 213 pix_mp.height); 214 else if (ctx->vpu_src_fmt->fourcc == V4L2_PIX_FMT_VP9_FRAME) 215 buf_size += hantro_vp9_mv_size(pix_mp.width, 216 pix_mp.height); 217 else if (ctx->vpu_src_fmt->fourcc == V4L2_PIX_FMT_HEVC_SLICE) 218 buf_size += hantro_hevc_mv_size(pix_mp.width, 219 pix_mp.height); 220 else if (ctx->vpu_src_fmt->fourcc == V4L2_PIX_FMT_AV1_FRAME) 221 buf_size += hantro_av1_mv_size(pix_mp.width, 222 pix_mp.height); 223 224 for (i = 0; i < num_buffers; ++i) { 225 struct hantro_aux_buf *priv = &ctx->postproc.dec_q[i]; 226 227 /* 228 * The buffers on this queue are meant as intermediate 229 * buffers for the decoder, so no mapping is needed. 230 */ 231 priv->attrs = DMA_ATTR_NO_KERNEL_MAPPING; 232 priv->cpu = dma_alloc_attrs(vpu->dev, buf_size, &priv->dma, 233 GFP_KERNEL, priv->attrs); 234 if (!priv->cpu) 235 return -ENOMEM; 236 priv->size = buf_size; 237 } 238 return 0; 239 } 240 241 static void hantro_postproc_g1_disable(struct hantro_ctx *ctx) 242 { 243 struct hantro_dev *vpu = ctx->dev; 244 245 HANTRO_PP_REG_WRITE(vpu, pipeline_en, 0x0); 246 } 247 248 static void hantro_postproc_g2_disable(struct hantro_ctx *ctx) 249 { 250 struct hantro_dev *vpu = ctx->dev; 251 252 hantro_reg_write(vpu, &g2_out_rs_e, 0); 253 } 254 255 void hantro_postproc_disable(struct hantro_ctx *ctx) 256 { 257 struct hantro_dev *vpu = ctx->dev; 258 259 if (vpu->variant->postproc_ops && vpu->variant->postproc_ops->disable) 260 vpu->variant->postproc_ops->disable(ctx); 261 } 262 263 void hantro_postproc_enable(struct hantro_ctx *ctx) 264 { 265 struct hantro_dev *vpu = ctx->dev; 266 267 if (vpu->variant->postproc_ops && vpu->variant->postproc_ops->enable) 268 vpu->variant->postproc_ops->enable(ctx); 269 } 270 271 int hanto_postproc_enum_framesizes(struct hantro_ctx *ctx, 272 struct v4l2_frmsizeenum *fsize) 273 { 274 struct hantro_dev *vpu = ctx->dev; 275 276 if (vpu->variant->postproc_ops && vpu->variant->postproc_ops->enum_framesizes) 277 return vpu->variant->postproc_ops->enum_framesizes(ctx, fsize); 278 279 return -EINVAL; 280 } 281 282 const struct hantro_postproc_ops hantro_g1_postproc_ops = { 283 .enable = hantro_postproc_g1_enable, 284 .disable = hantro_postproc_g1_disable, 285 }; 286 287 const struct hantro_postproc_ops hantro_g2_postproc_ops = { 288 .enable = hantro_postproc_g2_enable, 289 .disable = hantro_postproc_g2_disable, 290 .enum_framesizes = hantro_postproc_g2_enum_framesizes, 291 }; 292