Author | Tokens | Token Proportion | Commits | Commit Proportion |
---|---|---|---|---|
Fabien Dessenne | 5968 | 99.80% | 4 | 50.00% |
Jia-Ju Bai | 6 | 0.10% | 2 | 25.00% |
Krzysztof Kozlowski | 4 | 0.07% | 1 | 12.50% |
Benjamin Gaignard | 2 | 0.03% | 1 | 12.50% |
Total | 5980 | 8 |
// SPDX-License-Identifier: GPL-2.0 /* * Copyright (C) STMicroelectronics SA 2014 * Authors: Fabien Dessenne <fabien.dessenne@st.com> for STMicroelectronics. */ #include <linux/delay.h> #include "bdisp.h" #include "bdisp-filter.h" #include "bdisp-reg.h" /* Max width of the source frame in a single node */ #define MAX_SRC_WIDTH 2048 /* Reset & boot poll config */ #define POLL_RST_MAX 500 #define POLL_RST_DELAY_MS 2 enum bdisp_target_plan { BDISP_RGB, BDISP_Y, BDISP_CBCR }; struct bdisp_op_cfg { bool cconv; /* RGB - YUV conversion */ bool hflip; /* Horizontal flip */ bool vflip; /* Vertical flip */ bool wide; /* Wide (>MAX_SRC_WIDTH) */ bool scale; /* Scale */ u16 h_inc; /* Horizontal increment in 6.10 format */ u16 v_inc; /* Vertical increment in 6.10 format */ bool src_interlaced; /* is the src an interlaced buffer */ u8 src_nbp; /* nb of planes of the src */ bool src_yuv; /* is the src a YUV color format */ bool src_420; /* is the src 4:2:0 chroma subsampled */ u8 dst_nbp; /* nb of planes of the dst */ bool dst_yuv; /* is the dst a YUV color format */ bool dst_420; /* is the dst 4:2:0 chroma subsampled */ }; struct bdisp_filter_addr { u16 min; /* Filter min scale factor (6.10 fixed point) */ u16 max; /* Filter max scale factor (6.10 fixed point) */ void *virt; /* Virtual address for filter table */ dma_addr_t paddr; /* Physical address for filter table */ }; static const struct bdisp_filter_h_spec bdisp_h_spec[] = { { .min = 0, .max = 921, .coef = { 0x00, 0x00, 0x00, 0x00, 0x40, 0x00, 0x00, 0x00, 0x00, 0x00, 0xff, 0x07, 0x3d, 0xfc, 0x01, 0x00, 0x00, 0x01, 0xfd, 0x11, 0x36, 0xf9, 0x02, 0x00, 0x00, 0x01, 0xfb, 0x1b, 0x2e, 0xf9, 0x02, 0x00, 0x00, 0x01, 0xf9, 0x26, 0x26, 0xf9, 0x01, 0x00, 0x00, 0x02, 0xf9, 0x30, 0x19, 0xfb, 0x01, 0x00, 0x00, 0x02, 0xf9, 0x39, 0x0e, 0xfd, 0x01, 0x00, 0x00, 0x01, 0xfc, 0x3e, 0x06, 0xff, 0x00, 0x00 } }, { .min = 921, .max = 1024, .coef = { 0x00, 0x00, 0x00, 0x00, 0x40, 0x00, 0x00, 0x00, 0xff, 0x03, 0xfd, 0x08, 0x3e, 0xf9, 0x04, 0xfe, 0xfd, 0x06, 0xf8, 0x13, 0x3b, 0xf4, 0x07, 0xfc, 0xfb, 0x08, 0xf5, 0x1f, 0x34, 0xf1, 0x09, 0xfb, 0xfb, 0x09, 0xf2, 0x2b, 0x2a, 0xf1, 0x09, 0xfb, 0xfb, 0x09, 0xf2, 0x35, 0x1e, 0xf4, 0x08, 0xfb, 0xfc, 0x07, 0xf5, 0x3c, 0x12, 0xf7, 0x06, 0xfd, 0xfe, 0x04, 0xfa, 0x3f, 0x07, 0xfc, 0x03, 0xff } }, { .min = 1024, .max = 1126, .coef = { 0x00, 0x00, 0x00, 0x00, 0x40, 0x00, 0x00, 0x00, 0xff, 0x03, 0xfd, 0x08, 0x3e, 0xf9, 0x04, 0xfe, 0xfd, 0x06, 0xf8, 0x13, 0x3b, 0xf4, 0x07, 0xfc, 0xfb, 0x08, 0xf5, 0x1f, 0x34, 0xf1, 0x09, 0xfb, 0xfb, 0x09, 0xf2, 0x2b, 0x2a, 0xf1, 0x09, 0xfb, 0xfb, 0x09, 0xf2, 0x35, 0x1e, 0xf4, 0x08, 0xfb, 0xfc, 0x07, 0xf5, 0x3c, 0x12, 0xf7, 0x06, 0xfd, 0xfe, 0x04, 0xfa, 0x3f, 0x07, 0xfc, 0x03, 0xff } }, { .min = 1126, .max = 1228, .coef = { 0x00, 0x00, 0x00, 0x00, 0x40, 0x00, 0x00, 0x00, 0xff, 0x03, 0xfd, 0x08, 0x3e, 0xf9, 0x04, 0xfe, 0xfd, 0x06, 0xf8, 0x13, 0x3b, 0xf4, 0x07, 0xfc, 0xfb, 0x08, 0xf5, 0x1f, 0x34, 0xf1, 0x09, 0xfb, 0xfb, 0x09, 0xf2, 0x2b, 0x2a, 0xf1, 0x09, 0xfb, 0xfb, 0x09, 0xf2, 0x35, 0x1e, 0xf4, 0x08, 0xfb, 0xfc, 0x07, 0xf5, 0x3c, 0x12, 0xf7, 0x06, 0xfd, 0xfe, 0x04, 0xfa, 0x3f, 0x07, 0xfc, 0x03, 0xff } }, { .min = 1228, .max = 1331, .coef = { 0xfd, 0x04, 0xfc, 0x05, 0x39, 0x05, 0xfc, 0x04, 0xfc, 0x06, 0xf9, 0x0c, 0x39, 0xfe, 0x00, 0x02, 0xfb, 0x08, 0xf6, 0x17, 0x35, 0xf9, 0x02, 0x00, 0xfc, 0x08, 0xf4, 0x20, 0x30, 0xf4, 0x05, 0xff, 0xfd, 0x07, 0xf4, 0x29, 0x28, 0xf3, 0x07, 0xfd, 0xff, 0x05, 0xf5, 0x31, 0x1f, 0xf3, 0x08, 0xfc, 0x00, 0x02, 0xf9, 0x38, 0x14, 0xf6, 0x08, 0xfb, 0x02, 0x00, 0xff, 0x3a, 0x0b, 0xf8, 0x06, 0xfc } }, { .min = 1331, .max = 1433, .coef = { 0xfc, 0x06, 0xf9, 0x09, 0x34, 0x09, 0xf9, 0x06, 0xfd, 0x07, 0xf7, 0x10, 0x32, 0x02, 0xfc, 0x05, 0xfe, 0x07, 0xf6, 0x17, 0x2f, 0xfc, 0xff, 0x04, 0xff, 0x06, 0xf5, 0x20, 0x2a, 0xf9, 0x01, 0x02, 0x00, 0x04, 0xf6, 0x27, 0x25, 0xf6, 0x04, 0x00, 0x02, 0x01, 0xf9, 0x2d, 0x1d, 0xf5, 0x06, 0xff, 0x04, 0xff, 0xfd, 0x31, 0x15, 0xf5, 0x07, 0xfe, 0x05, 0xfc, 0x02, 0x35, 0x0d, 0xf7, 0x07, 0xfd } }, { .min = 1433, .max = 1536, .coef = { 0xfe, 0x06, 0xf8, 0x0b, 0x30, 0x0b, 0xf8, 0x06, 0xff, 0x06, 0xf7, 0x12, 0x2d, 0x05, 0xfa, 0x06, 0x00, 0x04, 0xf6, 0x18, 0x2c, 0x00, 0xfc, 0x06, 0x01, 0x02, 0xf7, 0x1f, 0x27, 0xfd, 0xff, 0x04, 0x03, 0x00, 0xf9, 0x24, 0x24, 0xf9, 0x00, 0x03, 0x04, 0xff, 0xfd, 0x29, 0x1d, 0xf7, 0x02, 0x01, 0x06, 0xfc, 0x00, 0x2d, 0x17, 0xf6, 0x04, 0x00, 0x06, 0xfa, 0x05, 0x30, 0x0f, 0xf7, 0x06, 0xff } }, { .min = 1536, .max = 2048, .coef = { 0x05, 0xfd, 0xfb, 0x13, 0x25, 0x13, 0xfb, 0xfd, 0x05, 0xfc, 0xfd, 0x17, 0x24, 0x0f, 0xf9, 0xff, 0x04, 0xfa, 0xff, 0x1b, 0x24, 0x0b, 0xf9, 0x00, 0x03, 0xf9, 0x01, 0x1f, 0x23, 0x08, 0xf8, 0x01, 0x02, 0xf9, 0x04, 0x22, 0x20, 0x04, 0xf9, 0x02, 0x01, 0xf8, 0x08, 0x25, 0x1d, 0x01, 0xf9, 0x03, 0x00, 0xf9, 0x0c, 0x25, 0x1a, 0xfe, 0xfa, 0x04, 0xff, 0xf9, 0x10, 0x26, 0x15, 0xfc, 0xfc, 0x05 } }, { .min = 2048, .max = 3072, .coef = { 0xfc, 0xfd, 0x06, 0x13, 0x18, 0x13, 0x06, 0xfd, 0xfc, 0xfe, 0x08, 0x15, 0x17, 0x12, 0x04, 0xfc, 0xfb, 0xfe, 0x0a, 0x16, 0x18, 0x10, 0x03, 0xfc, 0xfb, 0x00, 0x0b, 0x18, 0x17, 0x0f, 0x01, 0xfb, 0xfb, 0x00, 0x0d, 0x19, 0x17, 0x0d, 0x00, 0xfb, 0xfb, 0x01, 0x0f, 0x19, 0x16, 0x0b, 0x00, 0xfb, 0xfc, 0x03, 0x11, 0x19, 0x15, 0x09, 0xfe, 0xfb, 0xfc, 0x04, 0x12, 0x1a, 0x12, 0x08, 0xfe, 0xfc } }, { .min = 3072, .max = 4096, .coef = { 0xfe, 0x02, 0x09, 0x0f, 0x0e, 0x0f, 0x09, 0x02, 0xff, 0x02, 0x09, 0x0f, 0x10, 0x0e, 0x08, 0x01, 0xff, 0x03, 0x0a, 0x10, 0x10, 0x0d, 0x07, 0x00, 0x00, 0x04, 0x0b, 0x10, 0x0f, 0x0c, 0x06, 0x00, 0x00, 0x05, 0x0c, 0x10, 0x0e, 0x0c, 0x05, 0x00, 0x00, 0x06, 0x0c, 0x11, 0x0e, 0x0b, 0x04, 0x00, 0x00, 0x07, 0x0d, 0x11, 0x0f, 0x0a, 0x03, 0xff, 0x01, 0x08, 0x0e, 0x11, 0x0e, 0x09, 0x02, 0xff } }, { .min = 4096, .max = 5120, .coef = { 0x00, 0x04, 0x09, 0x0c, 0x0e, 0x0c, 0x09, 0x04, 0x01, 0x05, 0x09, 0x0c, 0x0d, 0x0c, 0x08, 0x04, 0x01, 0x05, 0x0a, 0x0c, 0x0e, 0x0b, 0x08, 0x03, 0x02, 0x06, 0x0a, 0x0d, 0x0c, 0x0b, 0x07, 0x03, 0x02, 0x07, 0x0a, 0x0d, 0x0d, 0x0a, 0x07, 0x02, 0x03, 0x07, 0x0b, 0x0d, 0x0c, 0x0a, 0x06, 0x02, 0x03, 0x08, 0x0b, 0x0d, 0x0d, 0x0a, 0x05, 0x01, 0x04, 0x08, 0x0c, 0x0d, 0x0c, 0x09, 0x05, 0x01 } }, { .min = 5120, .max = 65535, .coef = { 0x03, 0x06, 0x09, 0x0b, 0x09, 0x0b, 0x09, 0x06, 0x03, 0x06, 0x09, 0x0b, 0x0c, 0x0a, 0x08, 0x05, 0x03, 0x06, 0x09, 0x0b, 0x0c, 0x0a, 0x08, 0x05, 0x04, 0x07, 0x09, 0x0b, 0x0b, 0x0a, 0x08, 0x04, 0x04, 0x07, 0x0a, 0x0b, 0x0b, 0x0a, 0x07, 0x04, 0x04, 0x08, 0x0a, 0x0b, 0x0b, 0x09, 0x07, 0x04, 0x05, 0x08, 0x0a, 0x0b, 0x0c, 0x09, 0x06, 0x03, 0x05, 0x08, 0x0a, 0x0b, 0x0c, 0x09, 0x06, 0x03 } } }; #define NB_H_FILTER ARRAY_SIZE(bdisp_h_spec) static const struct bdisp_filter_v_spec bdisp_v_spec[] = { { .min = 0, .max = 1024, .coef = { 0x00, 0x00, 0x40, 0x00, 0x00, 0x00, 0x06, 0x3d, 0xfd, 0x00, 0xfe, 0x0f, 0x38, 0xfb, 0x00, 0xfd, 0x19, 0x2f, 0xfb, 0x00, 0xfc, 0x24, 0x24, 0xfc, 0x00, 0xfb, 0x2f, 0x19, 0xfd, 0x00, 0xfb, 0x38, 0x0f, 0xfe, 0x00, 0xfd, 0x3d, 0x06, 0x00, 0x00 } }, { .min = 1024, .max = 1331, .coef = { 0xfc, 0x05, 0x3e, 0x05, 0xfc, 0xf8, 0x0e, 0x3b, 0xff, 0x00, 0xf5, 0x18, 0x38, 0xf9, 0x02, 0xf4, 0x21, 0x31, 0xf5, 0x05, 0xf4, 0x2a, 0x27, 0xf4, 0x07, 0xf6, 0x30, 0x1e, 0xf4, 0x08, 0xf9, 0x35, 0x15, 0xf6, 0x07, 0xff, 0x37, 0x0b, 0xf9, 0x06 } }, { .min = 1331, .max = 1433, .coef = { 0xf8, 0x0a, 0x3c, 0x0a, 0xf8, 0xf6, 0x12, 0x3b, 0x02, 0xfb, 0xf4, 0x1b, 0x35, 0xfd, 0xff, 0xf4, 0x23, 0x30, 0xf8, 0x01, 0xf6, 0x29, 0x27, 0xf6, 0x04, 0xf9, 0x2e, 0x1e, 0xf5, 0x06, 0xfd, 0x31, 0x16, 0xf6, 0x06, 0x02, 0x32, 0x0d, 0xf8, 0x07 } }, { .min = 1433, .max = 1536, .coef = { 0xf6, 0x0e, 0x38, 0x0e, 0xf6, 0xf5, 0x15, 0x38, 0x06, 0xf8, 0xf5, 0x1d, 0x33, 0x00, 0xfb, 0xf6, 0x23, 0x2d, 0xfc, 0xfe, 0xf9, 0x28, 0x26, 0xf9, 0x00, 0xfc, 0x2c, 0x1e, 0xf7, 0x03, 0x00, 0x2e, 0x18, 0xf6, 0x04, 0x05, 0x2e, 0x11, 0xf7, 0x05 } }, { .min = 1536, .max = 2048, .coef = { 0xfb, 0x13, 0x24, 0x13, 0xfb, 0xfd, 0x17, 0x23, 0x0f, 0xfa, 0xff, 0x1a, 0x23, 0x0b, 0xf9, 0x01, 0x1d, 0x22, 0x07, 0xf9, 0x04, 0x20, 0x1f, 0x04, 0xf9, 0x07, 0x22, 0x1c, 0x01, 0xfa, 0x0b, 0x24, 0x17, 0xff, 0xfb, 0x0f, 0x24, 0x14, 0xfd, 0xfc } }, { .min = 2048, .max = 3072, .coef = { 0x05, 0x10, 0x16, 0x10, 0x05, 0x06, 0x11, 0x16, 0x0f, 0x04, 0x08, 0x13, 0x15, 0x0e, 0x02, 0x09, 0x14, 0x16, 0x0c, 0x01, 0x0b, 0x15, 0x15, 0x0b, 0x00, 0x0d, 0x16, 0x13, 0x0a, 0x00, 0x0f, 0x17, 0x13, 0x08, 0xff, 0x11, 0x18, 0x12, 0x07, 0xfe } }, { .min = 3072, .max = 4096, .coef = { 0x09, 0x0f, 0x10, 0x0f, 0x09, 0x09, 0x0f, 0x12, 0x0e, 0x08, 0x0a, 0x10, 0x11, 0x0e, 0x07, 0x0b, 0x11, 0x11, 0x0d, 0x06, 0x0c, 0x11, 0x12, 0x0c, 0x05, 0x0d, 0x12, 0x11, 0x0c, 0x04, 0x0e, 0x12, 0x11, 0x0b, 0x04, 0x0f, 0x13, 0x11, 0x0a, 0x03 } }, { .min = 4096, .max = 5120, .coef = { 0x0a, 0x0e, 0x10, 0x0e, 0x0a, 0x0b, 0x0e, 0x0f, 0x0e, 0x0a, 0x0b, 0x0f, 0x10, 0x0d, 0x09, 0x0c, 0x0f, 0x10, 0x0d, 0x08, 0x0d, 0x0f, 0x0f, 0x0d, 0x08, 0x0d, 0x10, 0x10, 0x0c, 0x07, 0x0e, 0x10, 0x0f, 0x0c, 0x07, 0x0f, 0x10, 0x10, 0x0b, 0x06 } }, { .min = 5120, .max = 65535, .coef = { 0x0b, 0x0e, 0x0e, 0x0e, 0x0b, 0x0b, 0x0e, 0x0f, 0x0d, 0x0b, 0x0c, 0x0e, 0x0f, 0x0d, 0x0a, 0x0c, 0x0e, 0x0f, 0x0d, 0x0a, 0x0d, 0x0f, 0x0e, 0x0d, 0x09, 0x0d, 0x0f, 0x0f, 0x0c, 0x09, 0x0e, 0x0f, 0x0e, 0x0c, 0x09, 0x0e, 0x0f, 0x0f, 0x0c, 0x08 } } }; #define NB_V_FILTER ARRAY_SIZE(bdisp_v_spec) static struct bdisp_filter_addr bdisp_h_filter[NB_H_FILTER]; static struct bdisp_filter_addr bdisp_v_filter[NB_V_FILTER]; /** * bdisp_hw_reset * @bdisp: bdisp entity * * Resets HW * * RETURNS: * 0 on success. */ int bdisp_hw_reset(struct bdisp_dev *bdisp) { unsigned int i; dev_dbg(bdisp->dev, "%s\n", __func__); /* Mask Interrupt */ writel(0, bdisp->regs + BLT_ITM0); /* Reset */ writel(readl(bdisp->regs + BLT_CTL) | BLT_CTL_RESET, bdisp->regs + BLT_CTL); writel(0, bdisp->regs + BLT_CTL); /* Wait for reset done */ for (i = 0; i < POLL_RST_MAX; i++) { if (readl(bdisp->regs + BLT_STA1) & BLT_STA1_IDLE) break; udelay(POLL_RST_DELAY_MS * 1000); } if (i == POLL_RST_MAX) dev_err(bdisp->dev, "Reset timeout\n"); return (i == POLL_RST_MAX) ? -EAGAIN : 0; } /** * bdisp_hw_get_and_clear_irq * @bdisp: bdisp entity * * Read then reset interrupt status * * RETURNS: * 0 if expected interrupt was raised. */ int bdisp_hw_get_and_clear_irq(struct bdisp_dev *bdisp) { u32 its; its = readl(bdisp->regs + BLT_ITS); /* Check for the only expected IT: LastNode of AQ1 */ if (!(its & BLT_ITS_AQ1_LNA)) { dev_dbg(bdisp->dev, "Unexpected IT status: 0x%08X\n", its); writel(its, bdisp->regs + BLT_ITS); return -1; } /* Clear and mask */ writel(its, bdisp->regs + BLT_ITS); writel(0, bdisp->regs + BLT_ITM0); return 0; } /** * bdisp_hw_free_nodes * @ctx: bdisp context * * Free node memory * * RETURNS: * None */ void bdisp_hw_free_nodes(struct bdisp_ctx *ctx) { if (ctx && ctx->node[0]) dma_free_attrs(ctx->bdisp_dev->dev, sizeof(struct bdisp_node) * MAX_NB_NODE, ctx->node[0], ctx->node_paddr[0], DMA_ATTR_WRITE_COMBINE); } /** * bdisp_hw_alloc_nodes * @ctx: bdisp context * * Allocate dma memory for nodes * * RETURNS: * 0 on success */ int bdisp_hw_alloc_nodes(struct bdisp_ctx *ctx) { struct device *dev = ctx->bdisp_dev->dev; unsigned int i, node_size = sizeof(struct bdisp_node); void *base; dma_addr_t paddr; /* Allocate all the nodes within a single memory page */ base = dma_alloc_attrs(dev, node_size * MAX_NB_NODE, &paddr, GFP_KERNEL, DMA_ATTR_WRITE_COMBINE); if (!base) { dev_err(dev, "%s no mem\n", __func__); return -ENOMEM; } memset(base, 0, node_size * MAX_NB_NODE); for (i = 0; i < MAX_NB_NODE; i++) { ctx->node[i] = base; ctx->node_paddr[i] = paddr; dev_dbg(dev, "node[%d]=0x%p (paddr=%pad)\n", i, ctx->node[i], &paddr); base += node_size; paddr += node_size; } return 0; } /** * bdisp_hw_free_filters * @dev: device * * Free filters memory * * RETURNS: * None */ void bdisp_hw_free_filters(struct device *dev) { int size = (BDISP_HF_NB * NB_H_FILTER) + (BDISP_VF_NB * NB_V_FILTER); if (bdisp_h_filter[0].virt) dma_free_attrs(dev, size, bdisp_h_filter[0].virt, bdisp_h_filter[0].paddr, DMA_ATTR_WRITE_COMBINE); } /** * bdisp_hw_alloc_filters * @dev: device * * Allocate dma memory for filters * * RETURNS: * 0 on success */ int bdisp_hw_alloc_filters(struct device *dev) { unsigned int i, size; void *base; dma_addr_t paddr; /* Allocate all the filters within a single memory page */ size = (BDISP_HF_NB * NB_H_FILTER) + (BDISP_VF_NB * NB_V_FILTER); base = dma_alloc_attrs(dev, size, &paddr, GFP_KERNEL, DMA_ATTR_WRITE_COMBINE); if (!base) return -ENOMEM; /* Setup filter addresses */ for (i = 0; i < NB_H_FILTER; i++) { bdisp_h_filter[i].min = bdisp_h_spec[i].min; bdisp_h_filter[i].max = bdisp_h_spec[i].max; memcpy(base, bdisp_h_spec[i].coef, BDISP_HF_NB); bdisp_h_filter[i].virt = base; bdisp_h_filter[i].paddr = paddr; base += BDISP_HF_NB; paddr += BDISP_HF_NB; } for (i = 0; i < NB_V_FILTER; i++) { bdisp_v_filter[i].min = bdisp_v_spec[i].min; bdisp_v_filter[i].max = bdisp_v_spec[i].max; memcpy(base, bdisp_v_spec[i].coef, BDISP_VF_NB); bdisp_v_filter[i].virt = base; bdisp_v_filter[i].paddr = paddr; base += BDISP_VF_NB; paddr += BDISP_VF_NB; } return 0; } /** * bdisp_hw_get_hf_addr * @inc: resize increment * * Find the horizontal filter table that fits the resize increment * * RETURNS: * table physical address */ static dma_addr_t bdisp_hw_get_hf_addr(u16 inc) { unsigned int i; for (i = NB_H_FILTER - 1; i > 0; i--) if ((bdisp_h_filter[i].min < inc) && (inc <= bdisp_h_filter[i].max)) break; return bdisp_h_filter[i].paddr; } /** * bdisp_hw_get_vf_addr * @inc: resize increment * * Find the vertical filter table that fits the resize increment * * RETURNS: * table physical address */ static dma_addr_t bdisp_hw_get_vf_addr(u16 inc) { unsigned int i; for (i = NB_V_FILTER - 1; i > 0; i--) if ((bdisp_v_filter[i].min < inc) && (inc <= bdisp_v_filter[i].max)) break; return bdisp_v_filter[i].paddr; } /** * bdisp_hw_get_inc * @from: input size * @to: output size * @inc: resize increment in 6.10 format * * Computes the increment (inverse of scale) in 6.10 format * * RETURNS: * 0 on success */ static int bdisp_hw_get_inc(u32 from, u32 to, u16 *inc) { u32 tmp; if (!to) return -EINVAL; if (to == from) { *inc = 1 << 10; return 0; } tmp = (from << 10) / to; if ((tmp > 0xFFFF) || (!tmp)) /* overflow (downscale x 63) or too small (upscale x 1024) */ return -EINVAL; *inc = (u16)tmp; return 0; } /** * bdisp_hw_get_hv_inc * @ctx: device context * @h_inc: horizontal increment * @v_inc: vertical increment * * Computes the horizontal & vertical increments (inverse of scale) * * RETURNS: * 0 on success */ static int bdisp_hw_get_hv_inc(struct bdisp_ctx *ctx, u16 *h_inc, u16 *v_inc) { u32 src_w, src_h, dst_w, dst_h; src_w = ctx->src.crop.width; src_h = ctx->src.crop.height; dst_w = ctx->dst.crop.width; dst_h = ctx->dst.crop.height; if (bdisp_hw_get_inc(src_w, dst_w, h_inc) || bdisp_hw_get_inc(src_h, dst_h, v_inc)) { dev_err(ctx->bdisp_dev->dev, "scale factors failed (%dx%d)->(%dx%d)\n", src_w, src_h, dst_w, dst_h); return -EINVAL; } return 0; } /** * bdisp_hw_get_op_cfg * @ctx: device context * @c: operation configuration * * Check which blitter operations are expected and sets the scaling increments * * RETURNS: * 0 on success */ static int bdisp_hw_get_op_cfg(struct bdisp_ctx *ctx, struct bdisp_op_cfg *c) { struct device *dev = ctx->bdisp_dev->dev; struct bdisp_frame *src = &ctx->src; struct bdisp_frame *dst = &ctx->dst; if (src->width > MAX_SRC_WIDTH * MAX_VERTICAL_STRIDES) { dev_err(dev, "Image width out of HW caps\n"); return -EINVAL; } c->wide = src->width > MAX_SRC_WIDTH; c->hflip = ctx->hflip; c->vflip = ctx->vflip; c->src_interlaced = (src->field == V4L2_FIELD_INTERLACED); c->src_nbp = src->fmt->nb_planes; c->src_yuv = (src->fmt->pixelformat == V4L2_PIX_FMT_NV12) || (src->fmt->pixelformat == V4L2_PIX_FMT_YUV420); c->src_420 = c->src_yuv; c->dst_nbp = dst->fmt->nb_planes; c->dst_yuv = (dst->fmt->pixelformat == V4L2_PIX_FMT_NV12) || (dst->fmt->pixelformat == V4L2_PIX_FMT_YUV420); c->dst_420 = c->dst_yuv; c->cconv = (c->src_yuv != c->dst_yuv); if (bdisp_hw_get_hv_inc(ctx, &c->h_inc, &c->v_inc)) { dev_err(dev, "Scale factor out of HW caps\n"); return -EINVAL; } /* Deinterlacing adjustment : stretch a field to a frame */ if (c->src_interlaced) c->v_inc /= 2; if ((c->h_inc != (1 << 10)) || (c->v_inc != (1 << 10))) c->scale = true; else c->scale = false; return 0; } /** * bdisp_hw_color_format * @pixelformat: v4l2 pixel format * * v4l2 to bdisp pixel format convert * * RETURNS: * bdisp pixel format */ static u32 bdisp_hw_color_format(u32 pixelformat) { u32 ret; switch (pixelformat) { case V4L2_PIX_FMT_YUV420: ret = (BDISP_YUV_3B << BLT_TTY_COL_SHIFT); break; case V4L2_PIX_FMT_NV12: ret = (BDISP_NV12 << BLT_TTY_COL_SHIFT) | BLT_TTY_BIG_END; break; case V4L2_PIX_FMT_RGB565: ret = (BDISP_RGB565 << BLT_TTY_COL_SHIFT); break; case V4L2_PIX_FMT_XBGR32: /* This V4L format actually refers to xRGB */ ret = (BDISP_XRGB8888 << BLT_TTY_COL_SHIFT); break; case V4L2_PIX_FMT_RGB24: /* RGB888 format */ ret = (BDISP_RGB888 << BLT_TTY_COL_SHIFT) | BLT_TTY_BIG_END; break; case V4L2_PIX_FMT_ABGR32: /* This V4L format actually refers to ARGB */ default: ret = (BDISP_ARGB8888 << BLT_TTY_COL_SHIFT) | BLT_TTY_ALPHA_R; break; } return ret; } /** * bdisp_hw_build_node * @ctx: device context * @cfg: operation configuration * @node: node to be set * @t_plan: whether the node refers to a RGB/Y or a CbCr plane * @src_x_offset: x offset in the source image * * Build a node * * RETURNS: * None */ static void bdisp_hw_build_node(struct bdisp_ctx *ctx, struct bdisp_op_cfg *cfg, struct bdisp_node *node, enum bdisp_target_plan t_plan, int src_x_offset) { struct bdisp_frame *src = &ctx->src; struct bdisp_frame *dst = &ctx->dst; u16 h_inc, v_inc, yh_inc, yv_inc; struct v4l2_rect src_rect = src->crop; struct v4l2_rect dst_rect = dst->crop; int dst_x_offset; s32 dst_width = dst->crop.width; u32 src_fmt, dst_fmt; const u32 *ivmx; dev_dbg(ctx->bdisp_dev->dev, "%s\n", __func__); memset(node, 0, sizeof(*node)); /* Adjust src and dst areas wrt src_x_offset */ src_rect.left += src_x_offset; src_rect.width -= src_x_offset; src_rect.width = min_t(__s32, MAX_SRC_WIDTH, src_rect.width); dst_x_offset = (src_x_offset * dst_width) / ctx->src.crop.width; dst_rect.left += dst_x_offset; dst_rect.width = (src_rect.width * dst_width) / ctx->src.crop.width; /* General */ src_fmt = src->fmt->pixelformat; dst_fmt = dst->fmt->pixelformat; node->nip = 0; node->cic = BLT_CIC_ALL_GRP; node->ack = BLT_ACK_BYPASS_S2S3; switch (cfg->src_nbp) { case 1: /* Src2 = RGB / Src1 = Src3 = off */ node->ins = BLT_INS_S1_OFF | BLT_INS_S2_MEM | BLT_INS_S3_OFF; break; case 2: /* Src3 = Y * Src2 = CbCr or ColorFill if writing the Y plane * Src1 = off */ node->ins = BLT_INS_S1_OFF | BLT_INS_S3_MEM; if (t_plan == BDISP_Y) node->ins |= BLT_INS_S2_CF; else node->ins |= BLT_INS_S2_MEM; break; case 3: default: /* Src3 = Y * Src2 = Cb or ColorFill if writing the Y plane * Src1 = Cr or ColorFill if writing the Y plane */ node->ins = BLT_INS_S3_MEM; if (t_plan == BDISP_Y) node->ins |= BLT_INS_S2_CF | BLT_INS_S1_CF; else node->ins |= BLT_INS_S2_MEM | BLT_INS_S1_MEM; break; } /* Color convert */ node->ins |= cfg->cconv ? BLT_INS_IVMX : 0; /* Scale needed if scaling OR 4:2:0 up/downsampling */ node->ins |= (cfg->scale || cfg->src_420 || cfg->dst_420) ? BLT_INS_SCALE : 0; /* Target */ node->tba = (t_plan == BDISP_CBCR) ? dst->paddr[1] : dst->paddr[0]; node->tty = dst->bytesperline; node->tty |= bdisp_hw_color_format(dst_fmt); node->tty |= BLT_TTY_DITHER; node->tty |= (t_plan == BDISP_CBCR) ? BLT_TTY_CHROMA : 0; node->tty |= cfg->hflip ? BLT_TTY_HSO : 0; node->tty |= cfg->vflip ? BLT_TTY_VSO : 0; if (cfg->dst_420 && (t_plan == BDISP_CBCR)) { /* 420 chroma downsampling */ dst_rect.height /= 2; dst_rect.width /= 2; dst_rect.left /= 2; dst_rect.top /= 2; dst_x_offset /= 2; dst_width /= 2; } node->txy = cfg->vflip ? (dst_rect.height - 1) : dst_rect.top; node->txy <<= 16; node->txy |= cfg->hflip ? (dst_width - dst_x_offset - 1) : dst_rect.left; node->tsz = dst_rect.height << 16 | dst_rect.width; if (cfg->src_interlaced) { /* handle only the top field which is half height of a frame */ src_rect.top /= 2; src_rect.height /= 2; } if (cfg->src_nbp == 1) { /* Src 2 : RGB */ node->s2ba = src->paddr[0]; node->s2ty = src->bytesperline; if (cfg->src_interlaced) node->s2ty *= 2; node->s2ty |= bdisp_hw_color_format(src_fmt); node->s2xy = src_rect.top << 16 | src_rect.left; node->s2sz = src_rect.height << 16 | src_rect.width; } else { /* Src 2 : Cb or CbCr */ if (cfg->src_420) { /* 420 chroma upsampling */ src_rect.top /= 2; src_rect.left /= 2; src_rect.width /= 2; src_rect.height /= 2; } node->s2ba = src->paddr[1]; node->s2ty = src->bytesperline; if (cfg->src_nbp == 3) node->s2ty /= 2; if (cfg->src_interlaced) node->s2ty *= 2; node->s2ty |= bdisp_hw_color_format(src_fmt); node->s2xy = src_rect.top << 16 | src_rect.left; node->s2sz = src_rect.height << 16 | src_rect.width; if (cfg->src_nbp == 3) { /* Src 1 : Cr */ node->s1ba = src->paddr[2]; node->s1ty = node->s2ty; node->s1xy = node->s2xy; } /* Src 3 : Y */ node->s3ba = src->paddr[0]; node->s3ty = src->bytesperline; if (cfg->src_interlaced) node->s3ty *= 2; node->s3ty |= bdisp_hw_color_format(src_fmt); if ((t_plan != BDISP_CBCR) && cfg->src_420) { /* No chroma upsampling for output RGB / Y plane */ node->s3xy = node->s2xy * 2; node->s3sz = node->s2sz * 2; } else { /* No need to read Y (Src3) when writing Chroma */ node->s3ty |= BLT_S3TY_BLANK_ACC; node->s3xy = node->s2xy; node->s3sz = node->s2sz; } } /* Resize (scale OR 4:2:0: chroma up/downsampling) */ if (node->ins & BLT_INS_SCALE) { /* no need to compute Y when writing CbCr from RGB input */ bool skip_y = (t_plan == BDISP_CBCR) && !cfg->src_yuv; /* FCTL */ if (cfg->scale) { node->fctl = BLT_FCTL_HV_SCALE; if (!skip_y) node->fctl |= BLT_FCTL_Y_HV_SCALE; } else { node->fctl = BLT_FCTL_HV_SAMPLE; if (!skip_y) node->fctl |= BLT_FCTL_Y_HV_SAMPLE; } /* RSF - Chroma may need to be up/downsampled */ h_inc = cfg->h_inc; v_inc = cfg->v_inc; if (!cfg->src_420 && cfg->dst_420 && (t_plan == BDISP_CBCR)) { /* RGB to 4:2:0 for Chroma: downsample */ h_inc *= 2; v_inc *= 2; } else if (cfg->src_420 && !cfg->dst_420) { /* 4:2:0: to RGB: upsample*/ h_inc /= 2; v_inc /= 2; } node->rsf = v_inc << 16 | h_inc; /* RZI */ node->rzi = BLT_RZI_DEFAULT; /* Filter table physical addr */ node->hfp = bdisp_hw_get_hf_addr(h_inc); node->vfp = bdisp_hw_get_vf_addr(v_inc); /* Y version */ if (!skip_y) { yh_inc = cfg->h_inc; yv_inc = cfg->v_inc; node->y_rsf = yv_inc << 16 | yh_inc; node->y_rzi = BLT_RZI_DEFAULT; node->y_hfp = bdisp_hw_get_hf_addr(yh_inc); node->y_vfp = bdisp_hw_get_vf_addr(yv_inc); } } /* Versatile matrix for RGB / YUV conversion */ if (cfg->cconv) { ivmx = cfg->src_yuv ? bdisp_yuv_to_rgb : bdisp_rgb_to_yuv; node->ivmx0 = ivmx[0]; node->ivmx1 = ivmx[1]; node->ivmx2 = ivmx[2]; node->ivmx3 = ivmx[3]; } } /** * bdisp_hw_build_all_nodes * @ctx: device context * * Build all the nodes for the blitter operation * * RETURNS: * 0 on success */ static int bdisp_hw_build_all_nodes(struct bdisp_ctx *ctx) { struct bdisp_op_cfg cfg; unsigned int i, nid = 0; int src_x_offset = 0; for (i = 0; i < MAX_NB_NODE; i++) if (!ctx->node[i]) { dev_err(ctx->bdisp_dev->dev, "node %d is null\n", i); return -EINVAL; } /* Get configuration (scale, flip, ...) */ if (bdisp_hw_get_op_cfg(ctx, &cfg)) return -EINVAL; /* Split source in vertical strides (HW constraint) */ for (i = 0; i < MAX_VERTICAL_STRIDES; i++) { /* Build RGB/Y node and link it to the previous node */ bdisp_hw_build_node(ctx, &cfg, ctx->node[nid], cfg.dst_nbp == 1 ? BDISP_RGB : BDISP_Y, src_x_offset); if (nid) ctx->node[nid - 1]->nip = ctx->node_paddr[nid]; nid++; /* Build additional Cb(Cr) node, link it to the previous one */ if (cfg.dst_nbp > 1) { bdisp_hw_build_node(ctx, &cfg, ctx->node[nid], BDISP_CBCR, src_x_offset); ctx->node[nid - 1]->nip = ctx->node_paddr[nid]; nid++; } /* Next stride until full width covered */ src_x_offset += MAX_SRC_WIDTH; if (src_x_offset >= ctx->src.crop.width) break; } /* Mark last node as the last */ ctx->node[nid - 1]->nip = 0; return 0; } /** * bdisp_hw_save_request * @ctx: device context * * Save a copy of the request and of the built nodes * * RETURNS: * None */ static void bdisp_hw_save_request(struct bdisp_ctx *ctx) { struct bdisp_node **copy_node = ctx->bdisp_dev->dbg.copy_node; struct bdisp_request *request = &ctx->bdisp_dev->dbg.copy_request; struct bdisp_node **node = ctx->node; int i; /* Request copy */ request->src = ctx->src; request->dst = ctx->dst; request->hflip = ctx->hflip; request->vflip = ctx->vflip; request->nb_req++; /* Nodes copy */ for (i = 0; i < MAX_NB_NODE; i++) { /* Allocate memory if not done yet */ if (!copy_node[i]) { copy_node[i] = devm_kzalloc(ctx->bdisp_dev->dev, sizeof(*copy_node[i]), GFP_ATOMIC); if (!copy_node[i]) return; } *copy_node[i] = *node[i]; } } /** * bdisp_hw_update * @ctx: device context * * Send the request to the HW * * RETURNS: * 0 on success */ int bdisp_hw_update(struct bdisp_ctx *ctx) { int ret; struct bdisp_dev *bdisp = ctx->bdisp_dev; struct device *dev = bdisp->dev; unsigned int node_id; dev_dbg(dev, "%s\n", __func__); /* build nodes */ ret = bdisp_hw_build_all_nodes(ctx); if (ret) { dev_err(dev, "cannot build nodes (%d)\n", ret); return ret; } /* Save a copy of the request */ bdisp_hw_save_request(ctx); /* Configure interrupt to 'Last Node Reached for AQ1' */ writel(BLT_AQ1_CTL_CFG, bdisp->regs + BLT_AQ1_CTL); writel(BLT_ITS_AQ1_LNA, bdisp->regs + BLT_ITM0); /* Write first node addr */ writel(ctx->node_paddr[0], bdisp->regs + BLT_AQ1_IP); /* Find and write last node addr : this starts the HW processing */ for (node_id = 0; node_id < MAX_NB_NODE - 1; node_id++) { if (!ctx->node[node_id]->nip) break; } writel(ctx->node_paddr[node_id], bdisp->regs + BLT_AQ1_LNA); return 0; }
Information contained on this website is for historical information purposes only and does not indicate or represent copyright ownership.
Created with Cregit http://github.com/cregit/cregit
Version 2.0-RC1