Commit | Line | Data |
---|---|---|
5320918b DA |
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 | ||
760285e7 | 18 | #include <drm/drmP.h> |
5320918b DA |
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_be16p(const uint8_t *pixel) | |
79 | { | |
80 | uint32_t pix = *(uint32_t *)pixel; | |
81 | u16 retval; | |
82 | ||
83 | retval = (((pix >> 3) & 0x001f) | | |
84 | ((pix >> 5) & 0x07e0) | | |
85 | ((pix >> 8) & 0xf800)); | |
86 | return retval; | |
87 | } | |
88 | ||
89 | /* | |
90 | * Render a command stream for an encoded horizontal line segment of pixels. | |
91 | * | |
92 | * A command buffer holds several commands. | |
93 | * It always begins with a fresh command header | |
94 | * (the protocol doesn't require this, but we enforce it to allow | |
95 | * multiple buffers to be potentially encoded and sent in parallel). | |
96 | * A single command encodes one contiguous horizontal line of pixels | |
97 | * | |
98 | * The function relies on the client to do all allocation, so that | |
99 | * rendering can be done directly to output buffers (e.g. USB URBs). | |
100 | * The function fills the supplied command buffer, providing information | |
101 | * on where it left off, so the client may call in again with additional | |
102 | * buffers if the line will take several buffers to complete. | |
103 | * | |
104 | * A single command can transmit a maximum of 256 pixels, | |
105 | * regardless of the compression ratio (protocol design limit). | |
106 | * To the hardware, 0 for a size byte means 256 | |
107 | * | |
108 | * Rather than 256 pixel commands which are either rl or raw encoded, | |
109 | * the rlx command simply assumes alternating raw and rl spans within one cmd. | |
110 | * This has a slightly larger header overhead, but produces more even results. | |
111 | * It also processes all data (read and write) in a single pass. | |
112 | * Performance benchmarks of common cases show it having just slightly better | |
113 | * compression than 256 pixel raw or rle commands, with similar CPU consumpion. | |
114 | * But for very rl friendly data, will compress not quite as well. | |
115 | */ | |
116 | static void udl_compress_hline16( | |
117 | const u8 **pixel_start_ptr, | |
118 | const u8 *const pixel_end, | |
119 | uint32_t *device_address_ptr, | |
120 | uint8_t **command_buffer_ptr, | |
121 | const uint8_t *const cmd_buffer_end, int bpp) | |
122 | { | |
123 | const u8 *pixel = *pixel_start_ptr; | |
124 | uint32_t dev_addr = *device_address_ptr; | |
125 | uint8_t *cmd = *command_buffer_ptr; | |
126 | ||
127 | while ((pixel_end > pixel) && | |
128 | (cmd_buffer_end - MIN_RLX_CMD_BYTES > cmd)) { | |
129 | uint8_t *raw_pixels_count_byte = 0; | |
130 | uint8_t *cmd_pixels_count_byte = 0; | |
131 | const u8 *raw_pixel_start = 0; | |
132 | const u8 *cmd_pixel_start, *cmd_pixel_end = 0; | |
133 | ||
134 | prefetchw((void *) cmd); /* pull in one cache line at least */ | |
135 | ||
136 | *cmd++ = 0xaf; | |
137 | *cmd++ = 0x6b; | |
138 | *cmd++ = (uint8_t) ((dev_addr >> 16) & 0xFF); | |
139 | *cmd++ = (uint8_t) ((dev_addr >> 8) & 0xFF); | |
140 | *cmd++ = (uint8_t) ((dev_addr) & 0xFF); | |
141 | ||
142 | cmd_pixels_count_byte = cmd++; /* we'll know this later */ | |
143 | cmd_pixel_start = pixel; | |
144 | ||
145 | raw_pixels_count_byte = cmd++; /* we'll know this later */ | |
146 | raw_pixel_start = pixel; | |
147 | ||
148 | cmd_pixel_end = pixel + (min(MAX_CMD_PIXELS + 1, | |
149 | min((int)(pixel_end - pixel) / bpp, | |
150 | (int)(cmd_buffer_end - cmd) / 2))) * bpp; | |
151 | ||
152 | prefetch_range((void *) pixel, (cmd_pixel_end - pixel) * bpp); | |
153 | ||
154 | while (pixel < cmd_pixel_end) { | |
155 | const u8 * const repeating_pixel = pixel; | |
156 | ||
157 | if (bpp == 2) | |
158 | *(uint16_t *)cmd = cpu_to_be16p((uint16_t *)pixel); | |
159 | else if (bpp == 4) | |
160 | *(uint16_t *)cmd = cpu_to_be16(pixel32_to_be16p(pixel)); | |
161 | ||
162 | cmd += 2; | |
163 | pixel += bpp; | |
164 | ||
165 | if (unlikely((pixel < cmd_pixel_end) && | |
166 | (!memcmp(pixel, repeating_pixel, bpp)))) { | |
167 | /* go back and fill in raw pixel count */ | |
168 | *raw_pixels_count_byte = (((repeating_pixel - | |
169 | raw_pixel_start) / bpp) + 1) & 0xFF; | |
170 | ||
171 | while ((pixel < cmd_pixel_end) | |
172 | && (!memcmp(pixel, repeating_pixel, bpp))) { | |
173 | pixel += bpp; | |
174 | } | |
175 | ||
176 | /* immediately after raw data is repeat byte */ | |
177 | *cmd++ = (((pixel - repeating_pixel) / bpp) - 1) & 0xFF; | |
178 | ||
179 | /* Then start another raw pixel span */ | |
180 | raw_pixel_start = pixel; | |
181 | raw_pixels_count_byte = cmd++; | |
182 | } | |
183 | } | |
184 | ||
185 | if (pixel > raw_pixel_start) { | |
186 | /* finalize last RAW span */ | |
187 | *raw_pixels_count_byte = ((pixel-raw_pixel_start) / bpp) & 0xFF; | |
188 | } | |
189 | ||
190 | *cmd_pixels_count_byte = ((pixel - cmd_pixel_start) / bpp) & 0xFF; | |
191 | dev_addr += ((pixel - cmd_pixel_start) / bpp) * 2; | |
192 | } | |
193 | ||
194 | if (cmd_buffer_end <= MIN_RLX_CMD_BYTES + cmd) { | |
195 | /* Fill leftover bytes with no-ops */ | |
196 | if (cmd_buffer_end > cmd) | |
197 | memset(cmd, 0xAF, cmd_buffer_end - cmd); | |
198 | cmd = (uint8_t *) cmd_buffer_end; | |
199 | } | |
200 | ||
201 | *command_buffer_ptr = cmd; | |
202 | *pixel_start_ptr = pixel; | |
203 | *device_address_ptr = dev_addr; | |
204 | ||
205 | return; | |
206 | } | |
207 | ||
208 | /* | |
209 | * There are 3 copies of every pixel: The front buffer that the fbdev | |
210 | * client renders to, the actual framebuffer across the USB bus in hardware | |
211 | * (that we can only write to, slowly, and can never read), and (optionally) | |
212 | * our shadow copy that tracks what's been sent to that hardware buffer. | |
213 | */ | |
214 | int udl_render_hline(struct drm_device *dev, int bpp, struct urb **urb_ptr, | |
215 | const char *front, char **urb_buf_ptr, | |
216 | u32 byte_offset, u32 byte_width, | |
217 | int *ident_ptr, int *sent_ptr) | |
218 | { | |
219 | const u8 *line_start, *line_end, *next_pixel; | |
220 | u32 base16 = 0 + (byte_offset / bpp) * 2; | |
221 | struct urb *urb = *urb_ptr; | |
222 | u8 *cmd = *urb_buf_ptr; | |
223 | u8 *cmd_end = (u8 *) urb->transfer_buffer + urb->transfer_buffer_length; | |
224 | ||
225 | line_start = (u8 *) (front + byte_offset); | |
226 | next_pixel = line_start; | |
227 | line_end = next_pixel + byte_width; | |
228 | ||
229 | while (next_pixel < line_end) { | |
230 | ||
231 | udl_compress_hline16(&next_pixel, | |
232 | line_end, &base16, | |
233 | (u8 **) &cmd, (u8 *) cmd_end, bpp); | |
234 | ||
235 | if (cmd >= cmd_end) { | |
236 | int len = cmd - (u8 *) urb->transfer_buffer; | |
237 | if (udl_submit_urb(dev, urb, len)) | |
238 | return 1; /* lost pixels is set */ | |
239 | *sent_ptr += len; | |
240 | urb = udl_get_urb(dev); | |
241 | if (!urb) | |
242 | return 1; /* lost_pixels is set */ | |
243 | *urb_ptr = urb; | |
244 | cmd = urb->transfer_buffer; | |
245 | cmd_end = &cmd[urb->transfer_buffer_length]; | |
246 | } | |
247 | } | |
248 | ||
249 | *urb_buf_ptr = cmd; | |
250 | ||
251 | return 0; | |
252 | } | |
253 |