drivers/gpu/drm/udl/udl_transfer.c

   1 /*
   2  * Copyright (C) 2012 Red Hat
   3  * based in parts on udlfb.c:
   4  * Copyright (C) 2009 Roberto De Ioris <roberto@unbit.it>
   5  * Copyright (C) 2009 Jaya Kumar <jayakumar.lkml@gmail.com>
   6  * Copyright (C) 2009 Bernie Thompson <bernie@plugable.com>
   7  *
   8  * This file is subject to the terms and conditions of the GNU General Public
   9  * License v2. See the file COPYING in the main directory of this archive for
  10  * more details.
  11  */
  12
  13 #include <linux/module.h>
  14 #include <linux/slab.h>
  15 #include <linux/fb.h>
  16 #include <asm/unaligned.h>
  17
  18 #include <drm/drmP.h>
  19 #include "udl_drv.h"
  20
  21 #define MAX_CMD_PIXELS          255
  22
  23 #define RLX_HEADER_BYTES        7
  24 #define MIN_RLX_PIX_BYTES       4
  25 #define MIN_RLX_CMD_BYTES       (RLX_HEADER_BYTES + MIN_RLX_PIX_BYTES)
  26
  27 #define RLE_HEADER_BYTES        6
  28 #define MIN_RLE_PIX_BYTES       3
  29 #define MIN_RLE_CMD_BYTES       (RLE_HEADER_BYTES + MIN_RLE_PIX_BYTES)
  30
  31 #define RAW_HEADER_BYTES        6
  32 #define MIN_RAW_PIX_BYTES       2
  33 #define MIN_RAW_CMD_BYTES       (RAW_HEADER_BYTES + MIN_RAW_PIX_BYTES)
  34
  35 /*
  36  * Trims identical data from front and back of line
  37  * Sets new front buffer address and width
  38  * And returns byte count of identical pixels
  39  * Assumes CPU natural alignment (unsigned long)
  40  * for back and front buffer ptrs and width
  41  */
  42 #if 0
  43 static int udl_trim_hline(const u8 *bback, const u8 **bfront, int *width_bytes)
  44 {
  45         int j, k;
  46         const unsigned long *back = (const unsigned long *) bback;
  47         const unsigned long *front = (const unsigned long *) *bfront;
  48         const int width = *width_bytes / sizeof(unsigned long);
  49         int identical = width;
  50         int start = width;
  51         int end = width;
  52
  53         for (j = 0; j < width; j++) {
  54                 if (back[j] != front[j]) {
  55                         start = j;
  56                         break;
  57                 }
  58         }
  59
  60         for (k = width - 1; k > j; k--) {
  61                 if (back[k] != front[k]) {
  62                         end = k+1;
  63                         break;
  64                 }
  65         }
  66
  67         identical = start + (width - end);
  68         *bfront = (u8 *) &front[start];
  69         *width_bytes = (end - start) * sizeof(unsigned long);
  70
  71         return identical * sizeof(unsigned long);
  72 }
  73 #endif
  74
  75 static inline u16 pixel32_to_be16(const uint32_t pixel)
  76 {
  77         return (((pixel >> 3) & 0x001f) |
  78                 ((pixel >> 5) & 0x07e0) |
  79                 ((pixel >> 8) & 0xf800));
  80 }
  81
  82 static inline u16 get_pixel_val16(const uint8_t *pixel, int log_bpp)
  83 {
  84         u16 pixel_val16;
  85         if (log_bpp == 1)
  86                 pixel_val16 = *(const uint16_t *)pixel;
  87         else
  88                 pixel_val16 = pixel32_to_be16(*(const uint32_t *)pixel);
  89         return pixel_val16;
  90 }
  91
  92 /*
  93  * Render a command stream for an encoded horizontal line segment of pixels.
  94  *
  95  * A command buffer holds several commands.
  96  * It always begins with a fresh command header
  97  * (the protocol doesn't require this, but we enforce it to allow
  98  * multiple buffers to be potentially encoded and sent in parallel).
  99  * A single command encodes one contiguous horizontal line of pixels
 100  *
 101  * The function relies on the client to do all allocation, so that
 102  * rendering can be done directly to output buffers (e.g. USB URBs).
 103  * The function fills the supplied command buffer, providing information
 104  * on where it left off, so the client may call in again with additional
 105  * buffers if the line will take several buffers to complete.
 106  *
 107  * A single command can transmit a maximum of 256 pixels,
 108  * regardless of the compression ratio (protocol design limit).
 109  * To the hardware, 0 for a size byte means 256
 110  *
 111  * Rather than 256 pixel commands which are either rl or raw encoded,
 112  * the rlx command simply assumes alternating raw and rl spans within one cmd.
 113  * This has a slightly larger header overhead, but produces more even results.
 114  * It also processes all data (read and write) in a single pass.
 115  * Performance benchmarks of common cases show it having just slightly better
 116  * compression than 256 pixel raw or rle commands, with similar CPU consumpion.
 117  * But for very rl friendly data, will compress not quite as well.
 118  */
 119 static void udl_compress_hline16(
 120         const u8 **pixel_start_ptr,
 121         const u8 *const pixel_end,
 122         uint32_t *device_address_ptr,
 123         uint8_t **command_buffer_ptr,
 124         const uint8_t *const cmd_buffer_end, int log_bpp)
 125 {
 126         const int bpp = 1 << log_bpp;
 127         const u8 *pixel = *pixel_start_ptr;
 128         uint32_t dev_addr  = *device_address_ptr;
 129         uint8_t *cmd = *command_buffer_ptr;
 130
 131         while ((pixel_end > pixel) &&
 132                (cmd_buffer_end - MIN_RLX_CMD_BYTES > cmd)) {
 133                 uint8_t *raw_pixels_count_byte = NULL;
 134                 uint8_t *cmd_pixels_count_byte = NULL;
 135                 const u8 *raw_pixel_start = NULL;
 136                 const u8 *cmd_pixel_start, *cmd_pixel_end = NULL;
 137                 uint16_t pixel_val16;
 138
 139                 *cmd++ = 0xaf;
 140                 *cmd++ = 0x6b;
 141                 *cmd++ = (uint8_t) ((dev_addr >> 16) & 0xFF);
 142                 *cmd++ = (uint8_t) ((dev_addr >> 8) & 0xFF);
 143                 *cmd++ = (uint8_t) ((dev_addr) & 0xFF);
 144
 145                 cmd_pixels_count_byte = cmd++; /*  we'll know this later */
 146                 cmd_pixel_start = pixel;
 147
 148                 raw_pixels_count_byte = cmd++; /*  we'll know this later */
 149                 raw_pixel_start = pixel;
 150
 151                 cmd_pixel_end = pixel + (min3(MAX_CMD_PIXELS + 1UL,
 152                                         (unsigned long)(pixel_end - pixel) >> log_bpp,
 153                                         (unsigned long)(cmd_buffer_end - 1 - cmd) / 2) << log_bpp);
 154
 155                 pixel_val16 = get_pixel_val16(pixel, log_bpp);
 156
 157                 while (pixel < cmd_pixel_end) {
 158                         const u8 *const start = pixel;
 159                         const uint16_t repeating_pixel_val16 = pixel_val16;
 160
 161                         put_unaligned_be16(pixel_val16, cmd);
 162
 163                         cmd += 2;
 164                         pixel += bpp;
 165
 166                         while (pixel < cmd_pixel_end) {
 167                                 pixel_val16 = get_pixel_val16(pixel, log_bpp);
 168                                 if (pixel_val16 != repeating_pixel_val16)
 169                                         break;
 170                                 pixel += bpp;
 171                         }
 172
 173                         if (unlikely(pixel > start + bpp)) {
 174                                 /* go back and fill in raw pixel count */
 175                                 *raw_pixels_count_byte = (((start -
 176                                                 raw_pixel_start) >> log_bpp) + 1) & 0xFF;
 177
 178                                 /* immediately after raw data is repeat byte */
 179                                 *cmd++ = (((pixel - start) >> log_bpp) - 1) & 0xFF;
 180
 181                                 /* Then start another raw pixel span */
 182                                 raw_pixel_start = pixel;
 183                                 raw_pixels_count_byte = cmd++;
 184                         }
 185                 }
 186
 187                 if (pixel > raw_pixel_start) {
 188                         /* finalize last RAW span */
 189                         *raw_pixels_count_byte = ((pixel - raw_pixel_start) >> log_bpp) & 0xFF;
 190                 } else {
 191                         /* undo unused byte */
 192                         cmd--;
 193                 }
 194
 195                 *cmd_pixels_count_byte = ((pixel - cmd_pixel_start) >> log_bpp) & 0xFF;
 196                 dev_addr += ((pixel - cmd_pixel_start) >> log_bpp) * 2;
 197         }
 198
 199         if (cmd_buffer_end <= MIN_RLX_CMD_BYTES + cmd) {
 200                 /* Fill leftover bytes with no-ops */
 201                 if (cmd_buffer_end > cmd)
 202                         memset(cmd, 0xAF, cmd_buffer_end - cmd);
 203                 cmd = (uint8_t *) cmd_buffer_end;
 204         }
 205
 206         *command_buffer_ptr = cmd;
 207         *pixel_start_ptr = pixel;
 208         *device_address_ptr = dev_addr;
 209
 210         return;
 211 }
 212
 213 /*
 214  * There are 3 copies of every pixel: The front buffer that the fbdev
 215  * client renders to, the actual framebuffer across the USB bus in hardware
 216  * (that we can only write to, slowly, and can never read), and (optionally)
 217  * our shadow copy that tracks what's been sent to that hardware buffer.
 218  */
 219 int udl_render_hline(struct drm_device *dev, int log_bpp, struct urb **urb_ptr,
 220                      const char *front, char **urb_buf_ptr,
 221                      u32 byte_offset, u32 device_byte_offset,
 222                      u32 byte_width,
 223                      int *ident_ptr, int *sent_ptr)
 224 {
 225         const u8 *line_start, *line_end, *next_pixel;
 226         u32 base16 = 0 + (device_byte_offset >> log_bpp) * 2;
 227         struct urb *urb = *urb_ptr;
 228         u8 *cmd = *urb_buf_ptr;
 229         u8 *cmd_end = (u8 *) urb->transfer_buffer + urb->transfer_buffer_length;
 230
 231         BUG_ON(!(log_bpp == 1 || log_bpp == 2));
 232
 233         line_start = (u8 *) (front + byte_offset);
 234         next_pixel = line_start;
 235         line_end = next_pixel + byte_width;
 236
 237         while (next_pixel < line_end) {
 238
 239                 udl_compress_hline16(&next_pixel,
 240                              line_end, &base16,
 241                              (u8 **) &cmd, (u8 *) cmd_end, log_bpp);
 242
 243                 if (cmd >= cmd_end) {
 244                         int len = cmd - (u8 *) urb->transfer_buffer;
 245                         if (udl_submit_urb(dev, urb, len))
 246                                 return 1; /* lost pixels is set */
 247                         *sent_ptr += len;
 248                         urb = udl_get_urb(dev);
 249                         if (!urb)
 250                                 return 1; /* lost_pixels is set */
 251                         *urb_ptr = urb;
 252                         cmd = urb->transfer_buffer;
 253                         cmd_end = &cmd[urb->transfer_buffer_length];
 254                 }
 255         }
 256
 257         *urb_buf_ptr = cmd;
 258
 259         return 0;
 260 }
 261