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 <linux/prefetch.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         prefetch((void *) front);
  54         prefetch((void *) back);
  55
  56         for (j = 0; j < width; j++) {
  57                 if (back[j] != front[j]) {
  58                         start = j;
  59                         break;
  60                 }
  61         }
  62
  63         for (k = width - 1; k > j; k--) {
  64                 if (back[k] != front[k]) {
  65                         end = k+1;
  66                         break;
  67                 }
  68         }
  69
  70         identical = start + (width - end);
  71         *bfront = (u8 *) &front[start];
  72         *width_bytes = (end - start) * sizeof(unsigned long);
  73
  74         return identical * sizeof(unsigned long);
  75 }
  76 #endif
  77
  78 static inline u16 pixel32_to_be16(const uint32_t pixel)
  79 {
  80         return (((pixel >> 3) & 0x001f) |
  81                 ((pixel >> 5) & 0x07e0) |
  82                 ((pixel >> 8) & 0xf800));
  83 }
  84
  85 static inline u16 get_pixel_val16(const uint8_t *pixel, int bpp)
  86 {
  87         u16 pixel_val16 = 0;
  88         if (bpp == 2)
  89                 pixel_val16 = *(const uint16_t *)pixel;
  90         else if (bpp == 4)
  91                 pixel_val16 = pixel32_to_be16(*(const uint32_t *)pixel);
  92         return pixel_val16;
  93 }
  94
  95 /*
  96  * Render a command stream for an encoded horizontal line segment of pixels.
  97  *
  98  * A command buffer holds several commands.
  99  * It always begins with a fresh command header
 100  * (the protocol doesn't require this, but we enforce it to allow
 101  * multiple buffers to be potentially encoded and sent in parallel).
 102  * A single command encodes one contiguous horizontal line of pixels
 103  *
 104  * The function relies on the client to do all allocation, so that
 105  * rendering can be done directly to output buffers (e.g. USB URBs).
 106  * The function fills the supplied command buffer, providing information
 107  * on where it left off, so the client may call in again with additional
 108  * buffers if the line will take several buffers to complete.
 109  *
 110  * A single command can transmit a maximum of 256 pixels,
 111  * regardless of the compression ratio (protocol design limit).
 112  * To the hardware, 0 for a size byte means 256
 113  *
 114  * Rather than 256 pixel commands which are either rl or raw encoded,
 115  * the rlx command simply assumes alternating raw and rl spans within one cmd.
 116  * This has a slightly larger header overhead, but produces more even results.
 117  * It also processes all data (read and write) in a single pass.
 118  * Performance benchmarks of common cases show it having just slightly better
 119  * compression than 256 pixel raw or rle commands, with similar CPU consumpion.
 120  * But for very rl friendly data, will compress not quite as well.
 121  */
 122 static void udl_compress_hline16(
 123         const u8 **pixel_start_ptr,
 124         const u8 *const pixel_end,
 125         uint32_t *device_address_ptr,
 126         uint8_t **command_buffer_ptr,
 127         const uint8_t *const cmd_buffer_end, int bpp)
 128 {
 129         const u8 *pixel = *pixel_start_ptr;
 130         uint32_t dev_addr  = *device_address_ptr;
 131         uint8_t *cmd = *command_buffer_ptr;
 132
 133         while ((pixel_end > pixel) &&
 134                (cmd_buffer_end - MIN_RLX_CMD_BYTES > cmd)) {
 135                 uint8_t *raw_pixels_count_byte = NULL;
 136                 uint8_t *cmd_pixels_count_byte = NULL;
 137                 const u8 *raw_pixel_start = NULL;
 138                 const u8 *cmd_pixel_start, *cmd_pixel_end = NULL;
 139                 uint16_t pixel_val16;
 140
 141                 prefetchw((void *) cmd); /* pull in one cache line at least */
 142
 143                 *cmd++ = 0xaf;
 144                 *cmd++ = 0x6b;
 145                 *cmd++ = (uint8_t) ((dev_addr >> 16) & 0xFF);
 146                 *cmd++ = (uint8_t) ((dev_addr >> 8) & 0xFF);
 147                 *cmd++ = (uint8_t) ((dev_addr) & 0xFF);
 148
 149                 cmd_pixels_count_byte = cmd++; /*  we'll know this later */
 150                 cmd_pixel_start = pixel;
 151
 152                 raw_pixels_count_byte = cmd++; /*  we'll know this later */
 153                 raw_pixel_start = pixel;
 154
 155                 cmd_pixel_end = pixel + (min(MAX_CMD_PIXELS + 1,
 156                         min((int)(pixel_end - pixel) / bpp,
 157                             (int)(cmd_buffer_end - cmd) / 2))) * bpp;
 158
 159                 prefetch_range((void *) pixel, (cmd_pixel_end - pixel) * bpp);
 160                 pixel_val16 = get_pixel_val16(pixel, bpp);
 161
 162                 while (pixel < cmd_pixel_end) {
 163                         const u8 *const start = pixel;
 164                         const uint16_t repeating_pixel_val16 = pixel_val16;
 165
 166                         *(uint16_t *)cmd = cpu_to_be16(pixel_val16);
 167
 168                         cmd += 2;
 169                         pixel += bpp;
 170
 171                         while (pixel < cmd_pixel_end) {
 172                                 pixel_val16 = get_pixel_val16(pixel, bpp);
 173                                 if (pixel_val16 != repeating_pixel_val16)
 174                                         break;
 175                                 pixel += bpp;
 176                         }
 177
 178                         if (unlikely(pixel > start + bpp)) {
 179                                 /* go back and fill in raw pixel count */
 180                                 *raw_pixels_count_byte = (((start -
 181                                                 raw_pixel_start) / bpp) + 1) & 0xFF;
 182
 183                                 /* immediately after raw data is repeat byte */
 184                                 *cmd++ = (((pixel - start) / bpp) - 1) & 0xFF;
 185
 186                                 /* Then start another raw pixel span */
 187                                 raw_pixel_start = pixel;
 188                                 raw_pixels_count_byte = cmd++;
 189                         }
 190                 }
 191
 192                 if (pixel > raw_pixel_start) {
 193                         /* finalize last RAW span */
 194                         *raw_pixels_count_byte = ((pixel-raw_pixel_start) / bpp) & 0xFF;
 195                 }
 196
 197                 *cmd_pixels_count_byte = ((pixel - cmd_pixel_start) / bpp) & 0xFF;
 198                 dev_addr += ((pixel - cmd_pixel_start) / bpp) * 2;
 199         }
 200
 201         if (cmd_buffer_end <= MIN_RLX_CMD_BYTES + cmd) {
 202                 /* Fill leftover bytes with no-ops */
 203                 if (cmd_buffer_end > cmd)
 204                         memset(cmd, 0xAF, cmd_buffer_end - cmd);
 205                 cmd = (uint8_t *) cmd_buffer_end;
 206         }
 207
 208         *command_buffer_ptr = cmd;
 209         *pixel_start_ptr = pixel;
 210         *device_address_ptr = dev_addr;
 211
 212         return;
 213 }
 214
 215 /*
 216  * There are 3 copies of every pixel: The front buffer that the fbdev
 217  * client renders to, the actual framebuffer across the USB bus in hardware
 218  * (that we can only write to, slowly, and can never read), and (optionally)
 219  * our shadow copy that tracks what's been sent to that hardware buffer.
 220  */
 221 int udl_render_hline(struct drm_device *dev, int bpp, struct urb **urb_ptr,
 222                      const char *front, char **urb_buf_ptr,
 223                      u32 byte_offset, u32 device_byte_offset,
 224                      u32 byte_width,
 225                      int *ident_ptr, int *sent_ptr)
 226 {
 227         const u8 *line_start, *line_end, *next_pixel;
 228         u32 base16 = 0 + (device_byte_offset / bpp) * 2;
 229         struct urb *urb = *urb_ptr;
 230         u8 *cmd = *urb_buf_ptr;
 231         u8 *cmd_end = (u8 *) urb->transfer_buffer + urb->transfer_buffer_length;
 232
 233         BUG_ON(!(bpp == 2 || bpp == 4));
 234
 235         line_start = (u8 *) (front + byte_offset);
 236         next_pixel = line_start;
 237         line_end = next_pixel + byte_width;
 238
 239         while (next_pixel < line_end) {
 240
 241                 udl_compress_hline16(&next_pixel,
 242                              line_end, &base16,
 243                              (u8 **) &cmd, (u8 *) cmd_end, bpp);
 244
 245                 if (cmd >= cmd_end) {
 246                         int len = cmd - (u8 *) urb->transfer_buffer;
 247                         if (udl_submit_urb(dev, urb, len))
 248                                 return 1; /* lost pixels is set */
 249                         *sent_ptr += len;
 250                         urb = udl_get_urb(dev);
 251                         if (!urb)
 252                                 return 1; /* lost_pixels is set */
 253                         *urb_ptr = urb;
 254                         cmd = urb->transfer_buffer;
 255                         cmd_end = &cmd[urb->transfer_buffer_length];
 256                 }
 257         }
 258
 259         *urb_buf_ptr = cmd;
 260
 261         return 0;
 262 }
 263