Merge branch 'juju' of git://git.kernel.org/pub/scm/linux/kernel/git/ieee1394/linux13...
[linux-2.6] / drivers / video / nvidia / nv_accel.c
1  /***************************************************************************\
2 |*                                                                           *|
3 |*       Copyright 1993-2003 NVIDIA, Corporation.  All rights reserved.      *|
4 |*                                                                           *|
5 |*     NOTICE TO USER:   The source code  is copyrighted under  U.S. and     *|
6 |*     international laws.  Users and possessors of this source code are     *|
7 |*     hereby granted a nonexclusive,  royalty-free copyright license to     *|
8 |*     use this code in individual and commercial software.                  *|
9 |*                                                                           *|
10 |*     Any use of this source code must include,  in the user documenta-     *|
11 |*     tion and  internal comments to the code,  notices to the end user     *|
12 |*     as follows:                                                           *|
13 |*                                                                           *|
14 |*       Copyright 1993-2003 NVIDIA, Corporation.  All rights reserved.      *|
15 |*                                                                           *|
16 |*     NVIDIA, CORPORATION MAKES NO REPRESENTATION ABOUT THE SUITABILITY     *|
17 |*     OF  THIS SOURCE  CODE  FOR ANY PURPOSE.  IT IS  PROVIDED  "AS IS"     *|
18 |*     WITHOUT EXPRESS OR IMPLIED WARRANTY OF ANY KIND.  NVIDIA, CORPOR-     *|
19 |*     ATION DISCLAIMS ALL WARRANTIES  WITH REGARD  TO THIS SOURCE CODE,     *|
20 |*     INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY, NONINFRINGE-     *|
21 |*     MENT,  AND FITNESS  FOR A PARTICULAR PURPOSE.   IN NO EVENT SHALL     *|
22 |*     NVIDIA, CORPORATION  BE LIABLE FOR ANY SPECIAL,  INDIRECT,  INCI-     *|
23 |*     DENTAL, OR CONSEQUENTIAL DAMAGES,  OR ANY DAMAGES  WHATSOEVER RE-     *|
24 |*     SULTING FROM LOSS OF USE,  DATA OR PROFITS,  WHETHER IN AN ACTION     *|
25 |*     OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION,  ARISING OUT OF     *|
26 |*     OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOURCE CODE.     *|
27 |*                                                                           *|
28 |*     U.S. Government  End  Users.   This source code  is a "commercial     *|
29 |*     item,"  as that  term is  defined at  48 C.F.R. 2.101 (OCT 1995),     *|
30 |*     consisting  of "commercial  computer  software"  and  "commercial     *|
31 |*     computer  software  documentation,"  as such  terms  are  used in     *|
32 |*     48 C.F.R. 12.212 (SEPT 1995)  and is provided to the U.S. Govern-     *|
33 |*     ment only as  a commercial end item.   Consistent with  48 C.F.R.     *|
34 |*     12.212 and  48 C.F.R. 227.7202-1 through  227.7202-4 (JUNE 1995),     *|
35 |*     all U.S. Government End Users  acquire the source code  with only     *|
36 |*     those rights set forth herein.                                        *|
37 |*                                                                           *|
38  \***************************************************************************/
39
40 /*
41  * GPL Licensing Note - According to Mark Vojkovich, author of the Xorg/
42  * XFree86 'nv' driver, this source code is provided under MIT-style licensing
43  * where the source code is provided "as is" without warranty of any kind.
44  * The only usage restriction is for the copyright notices to be retained
45  * whenever code is used.
46  *
47  * Antonino Daplas <adaplas@pol.net> 2005-03-11
48  */
49
50 #include <linux/fb.h>
51 #include "nv_type.h"
52 #include "nv_proto.h"
53 #include "nv_dma.h"
54 #include "nv_local.h"
55
56 /* There is a HW race condition with videoram command buffers.
57    You can't jump to the location of your put offset.  We write put
58    at the jump offset + SKIPS dwords with noop padding in between
59    to solve this problem */
60 #define SKIPS  8
61
62 static const int NVCopyROP[16] = {
63         0xCC,                   /* copy   */
64         0x55                    /* invert */
65 };
66
67 static const int NVCopyROP_PM[16] = {
68         0xCA,                   /* copy  */
69         0x5A,                   /* invert */
70 };
71
72 static inline void nvidiafb_safe_mode(struct fb_info *info)
73 {
74         struct nvidia_par *par = info->par;
75
76         touch_softlockup_watchdog();
77         info->pixmap.scan_align = 1;
78         par->lockup = 1;
79 }
80
81 static inline void NVFlush(struct fb_info *info)
82 {
83         struct nvidia_par *par = info->par;
84         int count = 1000000000;
85
86         while (--count && READ_GET(par) != par->dmaPut) ;
87
88         if (!count) {
89                 printk("nvidiafb: DMA Flush lockup\n");
90                 nvidiafb_safe_mode(info);
91         }
92 }
93
94 static inline void NVSync(struct fb_info *info)
95 {
96         struct nvidia_par *par = info->par;
97         int count = 1000000000;
98
99         while (--count && NV_RD32(par->PGRAPH, 0x0700)) ;
100
101         if (!count) {
102                 printk("nvidiafb: DMA Sync lockup\n");
103                 nvidiafb_safe_mode(info);
104         }
105 }
106
107 static void NVDmaKickoff(struct nvidia_par *par)
108 {
109         if (par->dmaCurrent != par->dmaPut) {
110                 par->dmaPut = par->dmaCurrent;
111                 WRITE_PUT(par, par->dmaPut);
112         }
113 }
114
115 static void NVDmaWait(struct fb_info *info, int size)
116 {
117         struct nvidia_par *par = info->par;
118         int dmaGet;
119         int count = 1000000000, cnt;
120         size++;
121
122         while (par->dmaFree < size && --count && !par->lockup) {
123                 dmaGet = READ_GET(par);
124
125                 if (par->dmaPut >= dmaGet) {
126                         par->dmaFree = par->dmaMax - par->dmaCurrent;
127                         if (par->dmaFree < size) {
128                                 NVDmaNext(par, 0x20000000);
129                                 if (dmaGet <= SKIPS) {
130                                         if (par->dmaPut <= SKIPS)
131                                                 WRITE_PUT(par, SKIPS + 1);
132                                         cnt = 1000000000;
133                                         do {
134                                                 dmaGet = READ_GET(par);
135                                         } while (--cnt && dmaGet <= SKIPS);
136                                         if (!cnt) {
137                                                 printk("DMA Get lockup\n");
138                                                 par->lockup = 1;
139                                         }
140                                 }
141                                 WRITE_PUT(par, SKIPS);
142                                 par->dmaCurrent = par->dmaPut = SKIPS;
143                                 par->dmaFree = dmaGet - (SKIPS + 1);
144                         }
145                 } else
146                         par->dmaFree = dmaGet - par->dmaCurrent - 1;
147         }
148
149         if (!count) {
150                 printk("nvidiafb: DMA Wait Lockup\n");
151                 nvidiafb_safe_mode(info);
152         }
153 }
154
155 static void NVSetPattern(struct fb_info *info, u32 clr0, u32 clr1,
156                          u32 pat0, u32 pat1)
157 {
158         struct nvidia_par *par = info->par;
159
160         NVDmaStart(info, par, PATTERN_COLOR_0, 4);
161         NVDmaNext(par, clr0);
162         NVDmaNext(par, clr1);
163         NVDmaNext(par, pat0);
164         NVDmaNext(par, pat1);
165 }
166
167 static void NVSetRopSolid(struct fb_info *info, u32 rop, u32 planemask)
168 {
169         struct nvidia_par *par = info->par;
170
171         if (planemask != ~0) {
172                 NVSetPattern(info, 0, planemask, ~0, ~0);
173                 if (par->currentRop != (rop + 32)) {
174                         NVDmaStart(info, par, ROP_SET, 1);
175                         NVDmaNext(par, NVCopyROP_PM[rop]);
176                         par->currentRop = rop + 32;
177                 }
178         } else if (par->currentRop != rop) {
179                 if (par->currentRop >= 16)
180                         NVSetPattern(info, ~0, ~0, ~0, ~0);
181                 NVDmaStart(info, par, ROP_SET, 1);
182                 NVDmaNext(par, NVCopyROP[rop]);
183                 par->currentRop = rop;
184         }
185 }
186
187 static void NVSetClippingRectangle(struct fb_info *info, int x1, int y1,
188                                    int x2, int y2)
189 {
190         struct nvidia_par *par = info->par;
191         int h = y2 - y1 + 1;
192         int w = x2 - x1 + 1;
193
194         NVDmaStart(info, par, CLIP_POINT, 2);
195         NVDmaNext(par, (y1 << 16) | x1);
196         NVDmaNext(par, (h << 16) | w);
197 }
198
199 void NVResetGraphics(struct fb_info *info)
200 {
201         struct nvidia_par *par = info->par;
202         u32 surfaceFormat, patternFormat, rectFormat, lineFormat;
203         int pitch, i;
204
205         pitch = info->fix.line_length;
206
207         par->dmaBase = (u32 __iomem *) (&par->FbStart[par->FbUsableSize]);
208
209         for (i = 0; i < SKIPS; i++)
210                 NV_WR32(&par->dmaBase[i], 0, 0x00000000);
211
212         NV_WR32(&par->dmaBase[0x0 + SKIPS], 0, 0x00040000);
213         NV_WR32(&par->dmaBase[0x1 + SKIPS], 0, 0x80000010);
214         NV_WR32(&par->dmaBase[0x2 + SKIPS], 0, 0x00042000);
215         NV_WR32(&par->dmaBase[0x3 + SKIPS], 0, 0x80000011);
216         NV_WR32(&par->dmaBase[0x4 + SKIPS], 0, 0x00044000);
217         NV_WR32(&par->dmaBase[0x5 + SKIPS], 0, 0x80000012);
218         NV_WR32(&par->dmaBase[0x6 + SKIPS], 0, 0x00046000);
219         NV_WR32(&par->dmaBase[0x7 + SKIPS], 0, 0x80000013);
220         NV_WR32(&par->dmaBase[0x8 + SKIPS], 0, 0x00048000);
221         NV_WR32(&par->dmaBase[0x9 + SKIPS], 0, 0x80000014);
222         NV_WR32(&par->dmaBase[0xA + SKIPS], 0, 0x0004A000);
223         NV_WR32(&par->dmaBase[0xB + SKIPS], 0, 0x80000015);
224         NV_WR32(&par->dmaBase[0xC + SKIPS], 0, 0x0004C000);
225         NV_WR32(&par->dmaBase[0xD + SKIPS], 0, 0x80000016);
226         NV_WR32(&par->dmaBase[0xE + SKIPS], 0, 0x0004E000);
227         NV_WR32(&par->dmaBase[0xF + SKIPS], 0, 0x80000017);
228
229         par->dmaPut = 0;
230         par->dmaCurrent = 16 + SKIPS;
231         par->dmaMax = 8191;
232         par->dmaFree = par->dmaMax - par->dmaCurrent;
233
234         switch (info->var.bits_per_pixel) {
235         case 32:
236         case 24:
237                 surfaceFormat = SURFACE_FORMAT_DEPTH24;
238                 patternFormat = PATTERN_FORMAT_DEPTH24;
239                 rectFormat = RECT_FORMAT_DEPTH24;
240                 lineFormat = LINE_FORMAT_DEPTH24;
241                 break;
242         case 16:
243                 surfaceFormat = SURFACE_FORMAT_DEPTH16;
244                 patternFormat = PATTERN_FORMAT_DEPTH16;
245                 rectFormat = RECT_FORMAT_DEPTH16;
246                 lineFormat = LINE_FORMAT_DEPTH16;
247                 break;
248         default:
249                 surfaceFormat = SURFACE_FORMAT_DEPTH8;
250                 patternFormat = PATTERN_FORMAT_DEPTH8;
251                 rectFormat = RECT_FORMAT_DEPTH8;
252                 lineFormat = LINE_FORMAT_DEPTH8;
253                 break;
254         }
255
256         NVDmaStart(info, par, SURFACE_FORMAT, 4);
257         NVDmaNext(par, surfaceFormat);
258         NVDmaNext(par, pitch | (pitch << 16));
259         NVDmaNext(par, 0);
260         NVDmaNext(par, 0);
261
262         NVDmaStart(info, par, PATTERN_FORMAT, 1);
263         NVDmaNext(par, patternFormat);
264
265         NVDmaStart(info, par, RECT_FORMAT, 1);
266         NVDmaNext(par, rectFormat);
267
268         NVDmaStart(info, par, LINE_FORMAT, 1);
269         NVDmaNext(par, lineFormat);
270
271         par->currentRop = ~0;   /* set to something invalid */
272         NVSetRopSolid(info, ROP_COPY, ~0);
273
274         NVSetClippingRectangle(info, 0, 0, info->var.xres_virtual,
275                                info->var.yres_virtual);
276
277         NVDmaKickoff(par);
278 }
279
280 int nvidiafb_sync(struct fb_info *info)
281 {
282         struct nvidia_par *par = info->par;
283
284         if (info->state != FBINFO_STATE_RUNNING)
285                 return 0;
286
287         if (!par->lockup)
288                 NVFlush(info);
289
290         if (!par->lockup)
291                 NVSync(info);
292
293         return 0;
294 }
295
296 void nvidiafb_copyarea(struct fb_info *info, const struct fb_copyarea *region)
297 {
298         struct nvidia_par *par = info->par;
299
300         if (info->state != FBINFO_STATE_RUNNING)
301                 return;
302
303         if (par->lockup)
304                 return cfb_copyarea(info, region);
305
306         NVDmaStart(info, par, BLIT_POINT_SRC, 3);
307         NVDmaNext(par, (region->sy << 16) | region->sx);
308         NVDmaNext(par, (region->dy << 16) | region->dx);
309         NVDmaNext(par, (region->height << 16) | region->width);
310
311         NVDmaKickoff(par);
312 }
313
314 void nvidiafb_fillrect(struct fb_info *info, const struct fb_fillrect *rect)
315 {
316         struct nvidia_par *par = info->par;
317         u32 color;
318
319         if (info->state != FBINFO_STATE_RUNNING)
320                 return;
321
322         if (par->lockup)
323                 return cfb_fillrect(info, rect);
324
325         if (info->var.bits_per_pixel == 8)
326                 color = rect->color;
327         else
328                 color = ((u32 *) info->pseudo_palette)[rect->color];
329
330         if (rect->rop != ROP_COPY)
331                 NVSetRopSolid(info, rect->rop, ~0);
332
333         NVDmaStart(info, par, RECT_SOLID_COLOR, 1);
334         NVDmaNext(par, color);
335
336         NVDmaStart(info, par, RECT_SOLID_RECTS(0), 2);
337         NVDmaNext(par, (rect->dx << 16) | rect->dy);
338         NVDmaNext(par, (rect->width << 16) | rect->height);
339
340         NVDmaKickoff(par);
341
342         if (rect->rop != ROP_COPY)
343                 NVSetRopSolid(info, ROP_COPY, ~0);
344 }
345
346 static void nvidiafb_mono_color_expand(struct fb_info *info,
347                                        const struct fb_image *image)
348 {
349         struct nvidia_par *par = info->par;
350         u32 fg, bg, mask = ~(~0 >> (32 - info->var.bits_per_pixel));
351         u32 dsize, width, *data = (u32 *) image->data, tmp;
352         int j, k = 0;
353
354         width = (image->width + 31) & ~31;
355         dsize = (width * image->height) >> 5;
356
357         if (info->var.bits_per_pixel == 8) {
358                 fg = image->fg_color | mask;
359                 bg = image->bg_color | mask;
360         } else {
361                 fg = ((u32 *) info->pseudo_palette)[image->fg_color] | mask;
362                 bg = ((u32 *) info->pseudo_palette)[image->bg_color] | mask;
363         }
364
365         NVDmaStart(info, par, RECT_EXPAND_TWO_COLOR_CLIP, 7);
366         NVDmaNext(par, (image->dy << 16) | (image->dx & 0xffff));
367         NVDmaNext(par, ((image->dy + image->height) << 16) |
368                   ((image->dx + image->width) & 0xffff));
369         NVDmaNext(par, bg);
370         NVDmaNext(par, fg);
371         NVDmaNext(par, (image->height << 16) | width);
372         NVDmaNext(par, (image->height << 16) | width);
373         NVDmaNext(par, (image->dy << 16) | (image->dx & 0xffff));
374
375         while (dsize >= RECT_EXPAND_TWO_COLOR_DATA_MAX_DWORDS) {
376                 NVDmaStart(info, par, RECT_EXPAND_TWO_COLOR_DATA(0),
377                            RECT_EXPAND_TWO_COLOR_DATA_MAX_DWORDS);
378
379                 for (j = RECT_EXPAND_TWO_COLOR_DATA_MAX_DWORDS; j--;) {
380                         tmp = data[k++];
381                         reverse_order(&tmp);
382                         NVDmaNext(par, tmp);
383                 }
384
385                 dsize -= RECT_EXPAND_TWO_COLOR_DATA_MAX_DWORDS;
386         }
387
388         if (dsize) {
389                 NVDmaStart(info, par, RECT_EXPAND_TWO_COLOR_DATA(0), dsize);
390
391                 for (j = dsize; j--;) {
392                         tmp = data[k++];
393                         reverse_order(&tmp);
394                         NVDmaNext(par, tmp);
395                 }
396         }
397
398         NVDmaKickoff(par);
399 }
400
401 void nvidiafb_imageblit(struct fb_info *info, const struct fb_image *image)
402 {
403         struct nvidia_par *par = info->par;
404
405         if (info->state != FBINFO_STATE_RUNNING)
406                 return;
407
408         if (image->depth == 1 && !par->lockup)
409                 nvidiafb_mono_color_expand(info, image);
410         else
411                 cfb_imageblit(info, image);
412 }