mm: treewide: remove GFP_TEMPORARY allocation flag
[linux/fpc-iii.git] / drivers / gpu / drm / udl / udl_transfer.c
blob0c87b1ac6b68f0d41cfd01851a14b9a092455f4f
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>
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.
13 #include <linux/module.h>
14 #include <linux/slab.h>
15 #include <linux/fb.h>
16 #include <linux/prefetch.h>
17 #include <asm/unaligned.h>
19 #include <drm/drmP.h>
20 #include "udl_drv.h"
22 #define MAX_CMD_PIXELS 255
24 #define RLX_HEADER_BYTES 7
25 #define MIN_RLX_PIX_BYTES 4
26 #define MIN_RLX_CMD_BYTES (RLX_HEADER_BYTES + MIN_RLX_PIX_BYTES)
28 #define RLE_HEADER_BYTES 6
29 #define MIN_RLE_PIX_BYTES 3
30 #define MIN_RLE_CMD_BYTES (RLE_HEADER_BYTES + MIN_RLE_PIX_BYTES)
32 #define RAW_HEADER_BYTES 6
33 #define MIN_RAW_PIX_BYTES 2
34 #define MIN_RAW_CMD_BYTES (RAW_HEADER_BYTES + MIN_RAW_PIX_BYTES)
37 * Trims identical data from front and back of line
38 * Sets new front buffer address and width
39 * And returns byte count of identical pixels
40 * Assumes CPU natural alignment (unsigned long)
41 * for back and front buffer ptrs and width
43 #if 0
44 static int udl_trim_hline(const u8 *bback, const u8 **bfront, int *width_bytes)
46 int j, k;
47 const unsigned long *back = (const unsigned long *) bback;
48 const unsigned long *front = (const unsigned long *) *bfront;
49 const int width = *width_bytes / sizeof(unsigned long);
50 int identical = width;
51 int start = width;
52 int end = width;
54 prefetch((void *) front);
55 prefetch((void *) back);
57 for (j = 0; j < width; j++) {
58 if (back[j] != front[j]) {
59 start = j;
60 break;
64 for (k = width - 1; k > j; k--) {
65 if (back[k] != front[k]) {
66 end = k+1;
67 break;
71 identical = start + (width - end);
72 *bfront = (u8 *) &front[start];
73 *width_bytes = (end - start) * sizeof(unsigned long);
75 return identical * sizeof(unsigned long);
77 #endif
79 static inline u16 pixel32_to_be16(const uint32_t pixel)
81 return (((pixel >> 3) & 0x001f) |
82 ((pixel >> 5) & 0x07e0) |
83 ((pixel >> 8) & 0xf800));
86 static inline u16 get_pixel_val16(const uint8_t *pixel, int bpp)
88 u16 pixel_val16 = 0;
89 if (bpp == 2)
90 pixel_val16 = *(const uint16_t *)pixel;
91 else if (bpp == 4)
92 pixel_val16 = pixel32_to_be16(*(const uint32_t *)pixel);
93 return pixel_val16;
97 * Render a command stream for an encoded horizontal line segment of pixels.
99 * A command buffer holds several commands.
100 * It always begins with a fresh command header
101 * (the protocol doesn't require this, but we enforce it to allow
102 * multiple buffers to be potentially encoded and sent in parallel).
103 * A single command encodes one contiguous horizontal line of pixels
105 * The function relies on the client to do all allocation, so that
106 * rendering can be done directly to output buffers (e.g. USB URBs).
107 * The function fills the supplied command buffer, providing information
108 * on where it left off, so the client may call in again with additional
109 * buffers if the line will take several buffers to complete.
111 * A single command can transmit a maximum of 256 pixels,
112 * regardless of the compression ratio (protocol design limit).
113 * To the hardware, 0 for a size byte means 256
115 * Rather than 256 pixel commands which are either rl or raw encoded,
116 * the rlx command simply assumes alternating raw and rl spans within one cmd.
117 * This has a slightly larger header overhead, but produces more even results.
118 * It also processes all data (read and write) in a single pass.
119 * Performance benchmarks of common cases show it having just slightly better
120 * compression than 256 pixel raw or rle commands, with similar CPU consumpion.
121 * But for very rl friendly data, will compress not quite as well.
123 static void udl_compress_hline16(
124 const u8 **pixel_start_ptr,
125 const u8 *const pixel_end,
126 uint32_t *device_address_ptr,
127 uint8_t **command_buffer_ptr,
128 const uint8_t *const cmd_buffer_end, int bpp)
130 const u8 *pixel = *pixel_start_ptr;
131 uint32_t dev_addr = *device_address_ptr;
132 uint8_t *cmd = *command_buffer_ptr;
134 while ((pixel_end > pixel) &&
135 (cmd_buffer_end - MIN_RLX_CMD_BYTES > cmd)) {
136 uint8_t *raw_pixels_count_byte = NULL;
137 uint8_t *cmd_pixels_count_byte = NULL;
138 const u8 *raw_pixel_start = NULL;
139 const u8 *cmd_pixel_start, *cmd_pixel_end = NULL;
140 uint16_t pixel_val16;
142 prefetchw((void *) cmd); /* pull in one cache line at least */
144 *cmd++ = 0xaf;
145 *cmd++ = 0x6b;
146 *cmd++ = (uint8_t) ((dev_addr >> 16) & 0xFF);
147 *cmd++ = (uint8_t) ((dev_addr >> 8) & 0xFF);
148 *cmd++ = (uint8_t) ((dev_addr) & 0xFF);
150 cmd_pixels_count_byte = cmd++; /* we'll know this later */
151 cmd_pixel_start = pixel;
153 raw_pixels_count_byte = cmd++; /* we'll know this later */
154 raw_pixel_start = pixel;
156 cmd_pixel_end = pixel + (min(MAX_CMD_PIXELS + 1,
157 min((int)(pixel_end - pixel) / bpp,
158 (int)(cmd_buffer_end - cmd) / 2))) * bpp;
160 prefetch_range((void *) pixel, (cmd_pixel_end - pixel) * bpp);
161 pixel_val16 = get_pixel_val16(pixel, bpp);
163 while (pixel < cmd_pixel_end) {
164 const u8 *const start = pixel;
165 const uint16_t repeating_pixel_val16 = pixel_val16;
167 put_unaligned_be16(pixel_val16, cmd);
169 cmd += 2;
170 pixel += bpp;
172 while (pixel < cmd_pixel_end) {
173 pixel_val16 = get_pixel_val16(pixel, bpp);
174 if (pixel_val16 != repeating_pixel_val16)
175 break;
176 pixel += bpp;
179 if (unlikely(pixel > start + bpp)) {
180 /* go back and fill in raw pixel count */
181 *raw_pixels_count_byte = (((start -
182 raw_pixel_start) / bpp) + 1) & 0xFF;
184 /* immediately after raw data is repeat byte */
185 *cmd++ = (((pixel - start) / bpp) - 1) & 0xFF;
187 /* Then start another raw pixel span */
188 raw_pixel_start = pixel;
189 raw_pixels_count_byte = cmd++;
193 if (pixel > raw_pixel_start) {
194 /* finalize last RAW span */
195 *raw_pixels_count_byte = ((pixel-raw_pixel_start) / bpp) & 0xFF;
198 *cmd_pixels_count_byte = ((pixel - cmd_pixel_start) / bpp) & 0xFF;
199 dev_addr += ((pixel - cmd_pixel_start) / bpp) * 2;
202 if (cmd_buffer_end <= MIN_RLX_CMD_BYTES + cmd) {
203 /* Fill leftover bytes with no-ops */
204 if (cmd_buffer_end > cmd)
205 memset(cmd, 0xAF, cmd_buffer_end - cmd);
206 cmd = (uint8_t *) cmd_buffer_end;
209 *command_buffer_ptr = cmd;
210 *pixel_start_ptr = pixel;
211 *device_address_ptr = dev_addr;
213 return;
217 * There are 3 copies of every pixel: The front buffer that the fbdev
218 * client renders to, the actual framebuffer across the USB bus in hardware
219 * (that we can only write to, slowly, and can never read), and (optionally)
220 * our shadow copy that tracks what's been sent to that hardware buffer.
222 int udl_render_hline(struct drm_device *dev, int bpp, struct urb **urb_ptr,
223 const char *front, char **urb_buf_ptr,
224 u32 byte_offset, u32 device_byte_offset,
225 u32 byte_width,
226 int *ident_ptr, int *sent_ptr)
228 const u8 *line_start, *line_end, *next_pixel;
229 u32 base16 = 0 + (device_byte_offset / bpp) * 2;
230 struct urb *urb = *urb_ptr;
231 u8 *cmd = *urb_buf_ptr;
232 u8 *cmd_end = (u8 *) urb->transfer_buffer + urb->transfer_buffer_length;
234 BUG_ON(!(bpp == 2 || bpp == 4));
236 line_start = (u8 *) (front + byte_offset);
237 next_pixel = line_start;
238 line_end = next_pixel + byte_width;
240 while (next_pixel < line_end) {
242 udl_compress_hline16(&next_pixel,
243 line_end, &base16,
244 (u8 **) &cmd, (u8 *) cmd_end, bpp);
246 if (cmd >= cmd_end) {
247 int len = cmd - (u8 *) urb->transfer_buffer;
248 if (udl_submit_urb(dev, urb, len))
249 return 1; /* lost pixels is set */
250 *sent_ptr += len;
251 urb = udl_get_urb(dev);
252 if (!urb)
253 return 1; /* lost_pixels is set */
254 *urb_ptr = urb;
255 cmd = urb->transfer_buffer;
256 cmd_end = &cmd[urb->transfer_buffer_length];
260 *urb_buf_ptr = cmd;
262 return 0;