1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Copyright (C) 2012 Red Hat 4 * based in parts on udlfb.c: 5 * Copyright (C) 2009 Roberto De Ioris <roberto@unbit.it> 6 * Copyright (C) 2009 Jaya Kumar <jayakumar.lkml@gmail.com> 7 * Copyright (C) 2009 Bernie Thompson <bernie@plugable.com> 8 */ 9 10 #include <asm/unaligned.h> 11 12 #include "udl_drv.h" 13 14 #define MAX_CMD_PIXELS 255 15 16 #define RLX_HEADER_BYTES 7 17 #define MIN_RLX_PIX_BYTES 4 18 #define MIN_RLX_CMD_BYTES (RLX_HEADER_BYTES + MIN_RLX_PIX_BYTES) 19 20 #define RLE_HEADER_BYTES 6 21 #define MIN_RLE_PIX_BYTES 3 22 #define MIN_RLE_CMD_BYTES (RLE_HEADER_BYTES + MIN_RLE_PIX_BYTES) 23 24 #define RAW_HEADER_BYTES 6 25 #define MIN_RAW_PIX_BYTES 2 26 #define MIN_RAW_CMD_BYTES (RAW_HEADER_BYTES + MIN_RAW_PIX_BYTES) 27 28 static inline u16 pixel32_to_be16(const uint32_t pixel) 29 { 30 return (((pixel >> 3) & 0x001f) | 31 ((pixel >> 5) & 0x07e0) | 32 ((pixel >> 8) & 0xf800)); 33 } 34 35 static inline u16 get_pixel_val16(const uint8_t *pixel, int log_bpp) 36 { 37 u16 pixel_val16; 38 if (log_bpp == 1) 39 pixel_val16 = *(const uint16_t *)pixel; 40 else 41 pixel_val16 = pixel32_to_be16(*(const uint32_t *)pixel); 42 return pixel_val16; 43 } 44 45 /* 46 * Render a command stream for an encoded horizontal line segment of pixels. 47 * 48 * A command buffer holds several commands. 49 * It always begins with a fresh command header 50 * (the protocol doesn't require this, but we enforce it to allow 51 * multiple buffers to be potentially encoded and sent in parallel). 52 * A single command encodes one contiguous horizontal line of pixels 53 * 54 * The function relies on the client to do all allocation, so that 55 * rendering can be done directly to output buffers (e.g. USB URBs). 56 * The function fills the supplied command buffer, providing information 57 * on where it left off, so the client may call in again with additional 58 * buffers if the line will take several buffers to complete. 59 * 60 * A single command can transmit a maximum of 256 pixels, 61 * regardless of the compression ratio (protocol design limit). 62 * To the hardware, 0 for a size byte means 256 63 * 64 * Rather than 256 pixel commands which are either rl or raw encoded, 65 * the rlx command simply assumes alternating raw and rl spans within one cmd. 66 * This has a slightly larger header overhead, but produces more even results. 67 * It also processes all data (read and write) in a single pass. 68 * Performance benchmarks of common cases show it having just slightly better 69 * compression than 256 pixel raw or rle commands, with similar CPU consumpion. 70 * But for very rl friendly data, will compress not quite as well. 71 */ 72 static void udl_compress_hline16( 73 const u8 **pixel_start_ptr, 74 const u8 *const pixel_end, 75 uint32_t *device_address_ptr, 76 uint8_t **command_buffer_ptr, 77 const uint8_t *const cmd_buffer_end, int log_bpp) 78 { 79 const int bpp = 1 << log_bpp; 80 const u8 *pixel = *pixel_start_ptr; 81 uint32_t dev_addr = *device_address_ptr; 82 uint8_t *cmd = *command_buffer_ptr; 83 84 while ((pixel_end > pixel) && 85 (cmd_buffer_end - MIN_RLX_CMD_BYTES > cmd)) { 86 uint8_t *raw_pixels_count_byte = NULL; 87 uint8_t *cmd_pixels_count_byte = NULL; 88 const u8 *raw_pixel_start = NULL; 89 const u8 *cmd_pixel_start, *cmd_pixel_end = NULL; 90 uint16_t pixel_val16; 91 92 *cmd++ = 0xaf; 93 *cmd++ = 0x6b; 94 *cmd++ = (uint8_t) ((dev_addr >> 16) & 0xFF); 95 *cmd++ = (uint8_t) ((dev_addr >> 8) & 0xFF); 96 *cmd++ = (uint8_t) ((dev_addr) & 0xFF); 97 98 cmd_pixels_count_byte = cmd++; /* we'll know this later */ 99 cmd_pixel_start = pixel; 100 101 raw_pixels_count_byte = cmd++; /* we'll know this later */ 102 raw_pixel_start = pixel; 103 104 cmd_pixel_end = pixel + (min3(MAX_CMD_PIXELS + 1UL, 105 (unsigned long)(pixel_end - pixel) >> log_bpp, 106 (unsigned long)(cmd_buffer_end - 1 - cmd) / 2) << log_bpp); 107 108 pixel_val16 = get_pixel_val16(pixel, log_bpp); 109 110 while (pixel < cmd_pixel_end) { 111 const u8 *const start = pixel; 112 const uint16_t repeating_pixel_val16 = pixel_val16; 113 114 put_unaligned_be16(pixel_val16, cmd); 115 116 cmd += 2; 117 pixel += bpp; 118 119 while (pixel < cmd_pixel_end) { 120 pixel_val16 = get_pixel_val16(pixel, log_bpp); 121 if (pixel_val16 != repeating_pixel_val16) 122 break; 123 pixel += bpp; 124 } 125 126 if (unlikely(pixel > start + bpp)) { 127 /* go back and fill in raw pixel count */ 128 *raw_pixels_count_byte = (((start - 129 raw_pixel_start) >> log_bpp) + 1) & 0xFF; 130 131 /* immediately after raw data is repeat byte */ 132 *cmd++ = (((pixel - start) >> log_bpp) - 1) & 0xFF; 133 134 /* Then start another raw pixel span */ 135 raw_pixel_start = pixel; 136 raw_pixels_count_byte = cmd++; 137 } 138 } 139 140 if (pixel > raw_pixel_start) { 141 /* finalize last RAW span */ 142 *raw_pixels_count_byte = ((pixel - raw_pixel_start) >> log_bpp) & 0xFF; 143 } else { 144 /* undo unused byte */ 145 cmd--; 146 } 147 148 *cmd_pixels_count_byte = ((pixel - cmd_pixel_start) >> log_bpp) & 0xFF; 149 dev_addr += ((pixel - cmd_pixel_start) >> log_bpp) * 2; 150 } 151 152 if (cmd_buffer_end <= MIN_RLX_CMD_BYTES + cmd) { 153 /* Fill leftover bytes with no-ops */ 154 if (cmd_buffer_end > cmd) 155 memset(cmd, 0xAF, cmd_buffer_end - cmd); 156 cmd = (uint8_t *) cmd_buffer_end; 157 } 158 159 *command_buffer_ptr = cmd; 160 *pixel_start_ptr = pixel; 161 *device_address_ptr = dev_addr; 162 163 return; 164 } 165 166 /* 167 * There are 3 copies of every pixel: The front buffer that the fbdev 168 * client renders to, the actual framebuffer across the USB bus in hardware 169 * (that we can only write to, slowly, and can never read), and (optionally) 170 * our shadow copy that tracks what's been sent to that hardware buffer. 171 */ 172 int udl_render_hline(struct drm_device *dev, int log_bpp, struct urb **urb_ptr, 173 const char *front, char **urb_buf_ptr, 174 u32 byte_offset, u32 device_byte_offset, 175 u32 byte_width) 176 { 177 const u8 *line_start, *line_end, *next_pixel; 178 u32 base16 = 0 + (device_byte_offset >> log_bpp) * 2; 179 struct urb *urb = *urb_ptr; 180 u8 *cmd = *urb_buf_ptr; 181 u8 *cmd_end = (u8 *) urb->transfer_buffer + urb->transfer_buffer_length; 182 183 if (WARN_ON(!(log_bpp == 1 || log_bpp == 2))) { 184 /* need to finish URB at error from this function */ 185 udl_urb_completion(urb); 186 return -EINVAL; 187 } 188 189 line_start = (u8 *) (front + byte_offset); 190 next_pixel = line_start; 191 line_end = next_pixel + byte_width; 192 193 while (next_pixel < line_end) { 194 195 udl_compress_hline16(&next_pixel, 196 line_end, &base16, 197 (u8 **) &cmd, (u8 *) cmd_end, log_bpp); 198 199 if (cmd >= cmd_end) { 200 int len = cmd - (u8 *) urb->transfer_buffer; 201 int ret = udl_submit_urb(dev, urb, len); 202 if (ret) 203 return ret; 204 urb = udl_get_urb(dev); 205 if (!urb) 206 return -EAGAIN; 207 *urb_ptr = urb; 208 cmd = urb->transfer_buffer; 209 cmd_end = &cmd[urb->transfer_buffer_length]; 210 } 211 } 212 213 *urb_buf_ptr = cmd; 214 215 return 0; 216 } 217