texenvprogram: fix for ARB_draw_buffers.
[mesa.git] / src / gallium / winsys / drm / vmware / core / vmw_context.c
1 /**********************************************************
2 * Copyright 2009 VMware, Inc. All rights reserved.
3 *
4 * Permission is hereby granted, free of charge, to any person
5 * obtaining a copy of this software and associated documentation
6 * files (the "Software"), to deal in the Software without
7 * restriction, including without limitation the rights to use, copy,
8 * modify, merge, publish, distribute, sublicense, and/or sell copies
9 * of the Software, and to permit persons to whom the Software is
10 * furnished to do so, subject to the following conditions:
11 *
12 * The above copyright notice and this permission notice shall be
13 * included in all copies or substantial portions of the Software.
14 *
15 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
16 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
17 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
18 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
19 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
20 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
21 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
22 * SOFTWARE.
23 *
24 **********************************************************/
25
26
27 #include "svga_cmd.h"
28
29 #include "util/u_debug.h"
30 #include "util/u_memory.h"
31 #include "util/u_debug_stack.h"
32 #include "pipebuffer/pb_buffer.h"
33 #include "pipebuffer/pb_validate.h"
34
35 #include "svga_winsys.h"
36 #include "vmw_context.h"
37 #include "vmw_screen.h"
38 #include "vmw_buffer.h"
39 #include "vmw_surface.h"
40 #include "vmw_fence.h"
41
42 #define VMW_COMMAND_SIZE (64*1024)
43 #define VMW_SURFACE_RELOCS (1024)
44 #define VMW_REGION_RELOCS (512)
45
46 #define VMW_MUST_FLUSH_STACK 8
47
48 struct vmw_region_relocation
49 {
50 struct SVGAGuestPtr *where;
51 struct pb_buffer *buffer;
52 /* TODO: put offset info inside where */
53 uint32 offset;
54 };
55
56 struct vmw_svga_winsys_context
57 {
58 struct svga_winsys_context base;
59
60 struct vmw_winsys_screen *vws;
61
62 #ifdef DEBUG
63 boolean must_flush;
64 struct debug_stack_frame must_flush_stack[VMW_MUST_FLUSH_STACK];
65 #endif
66
67 struct {
68 uint8_t buffer[VMW_COMMAND_SIZE];
69 uint32_t size;
70 uint32_t used;
71 uint32_t reserved;
72 } command;
73
74 struct {
75 struct vmw_svga_winsys_surface *handles[VMW_SURFACE_RELOCS];
76 uint32_t size;
77 uint32_t used;
78 uint32_t staged;
79 uint32_t reserved;
80 } surface;
81
82 struct {
83 struct vmw_region_relocation relocs[VMW_REGION_RELOCS];
84 uint32_t size;
85 uint32_t used;
86 uint32_t staged;
87 uint32_t reserved;
88 } region;
89
90 struct pb_validate *validate;
91
92 uint32_t last_fence;
93
94 /**
95 * The amount of GMR that is referred by the commands currently batched
96 * in the context.
97 */
98 uint32_t seen_regions;
99
100 /**
101 * Whether this context should fail to reserve more commands, not because it
102 * ran out of command space, but because a substantial ammount of GMR was
103 * referred.
104 */
105 boolean preemptive_flush;
106 };
107
108
109 static INLINE struct vmw_svga_winsys_context *
110 vmw_svga_winsys_context(struct svga_winsys_context *swc)
111 {
112 assert(swc);
113 return (struct vmw_svga_winsys_context *)swc;
114 }
115
116
117 static enum pipe_error
118 vmw_swc_flush(struct svga_winsys_context *swc,
119 struct pipe_fence_handle **pfence)
120 {
121 struct vmw_svga_winsys_context *vswc = vmw_svga_winsys_context(swc);
122 struct pipe_fence_handle *fence = NULL;
123 unsigned i;
124 enum pipe_error ret;
125
126 ret = pb_validate_validate(vswc->validate);
127 assert(ret == PIPE_OK);
128 if(ret == PIPE_OK) {
129
130 /* Apply relocations */
131 for(i = 0; i < vswc->region.used; ++i) {
132 struct vmw_region_relocation *reloc = &vswc->region.relocs[i];
133 struct SVGAGuestPtr ptr;
134
135 if(!vmw_gmr_bufmgr_region_ptr(reloc->buffer, &ptr))
136 assert(0);
137
138 ptr.offset += reloc->offset;
139
140 *reloc->where = ptr;
141 }
142
143 if (vswc->command.used)
144 vmw_ioctl_command(vswc->vws,
145 vswc->command.buffer,
146 vswc->command.used,
147 &vswc->last_fence);
148
149 fence = vmw_pipe_fence(vswc->last_fence);
150
151 pb_validate_fence(vswc->validate, fence);
152 }
153
154 vswc->command.used = 0;
155 vswc->command.reserved = 0;
156
157 for(i = 0; i < vswc->surface.used + vswc->surface.staged; ++i) {
158 struct vmw_svga_winsys_surface *vsurf =
159 vswc->surface.handles[i];
160 p_atomic_dec(&vsurf->validated);
161 vmw_svga_winsys_surface_reference(&vswc->surface.handles[i], NULL);
162 }
163
164 vswc->surface.used = 0;
165 vswc->surface.reserved = 0;
166
167 for(i = 0; i < vswc->region.used + vswc->region.staged; ++i) {
168 pb_reference(&vswc->region.relocs[i].buffer, NULL);
169 }
170
171 vswc->region.used = 0;
172 vswc->region.reserved = 0;
173
174 #ifdef DEBUG
175 vswc->must_flush = FALSE;
176 #endif
177 vswc->preemptive_flush = FALSE;
178 vswc->seen_regions = 0;
179
180 if(pfence)
181 *pfence = fence;
182
183 return ret;
184 }
185
186
187 static void *
188 vmw_swc_reserve(struct svga_winsys_context *swc,
189 uint32_t nr_bytes, uint32_t nr_relocs )
190 {
191 struct vmw_svga_winsys_context *vswc = vmw_svga_winsys_context(swc);
192
193 #ifdef DEBUG
194 /* Check if somebody forgot to check the previous failure */
195 if(vswc->must_flush) {
196 debug_printf("Forgot to flush:\n");
197 debug_backtrace_dump(vswc->must_flush_stack, VMW_MUST_FLUSH_STACK);
198 assert(!vswc->must_flush);
199 }
200 #endif
201
202 assert(nr_bytes <= vswc->command.size);
203 if(nr_bytes > vswc->command.size)
204 return NULL;
205
206 if(vswc->preemptive_flush ||
207 vswc->command.used + nr_bytes > vswc->command.size ||
208 vswc->surface.used + nr_relocs > vswc->surface.size ||
209 vswc->region.used + nr_relocs > vswc->region.size) {
210 #ifdef DEBUG
211 vswc->must_flush = TRUE;
212 debug_backtrace_capture(vswc->must_flush_stack, 1,
213 VMW_MUST_FLUSH_STACK);
214 #endif
215 return NULL;
216 }
217
218 assert(vswc->command.used + nr_bytes <= vswc->command.size);
219 assert(vswc->surface.used + nr_relocs <= vswc->surface.size);
220 assert(vswc->region.used + nr_relocs <= vswc->region.size);
221
222 vswc->command.reserved = nr_bytes;
223 vswc->surface.reserved = nr_relocs;
224 vswc->surface.staged = 0;
225 vswc->region.reserved = nr_relocs;
226 vswc->region.staged = 0;
227
228 return vswc->command.buffer + vswc->command.used;
229 }
230
231
232 static void
233 vmw_swc_surface_relocation(struct svga_winsys_context *swc,
234 uint32 *where,
235 struct svga_winsys_surface *surface,
236 unsigned flags)
237 {
238 struct vmw_svga_winsys_context *vswc = vmw_svga_winsys_context(swc);
239 struct vmw_svga_winsys_surface *vsurf;
240
241 if(!surface) {
242 *where = SVGA3D_INVALID_ID;
243 return;
244 }
245
246 assert(vswc->surface.staged < vswc->surface.reserved);
247
248 vsurf = vmw_svga_winsys_surface(surface);
249
250 *where = vsurf->sid;
251
252 vmw_svga_winsys_surface_reference(&vswc->surface.handles[vswc->surface.used + vswc->surface.staged], vsurf);
253 p_atomic_inc(&vsurf->validated);
254 ++vswc->surface.staged;
255 }
256
257
258 static void
259 vmw_swc_region_relocation(struct svga_winsys_context *swc,
260 struct SVGAGuestPtr *where,
261 struct svga_winsys_buffer *buffer,
262 uint32 offset,
263 unsigned flags)
264 {
265 struct vmw_svga_winsys_context *vswc = vmw_svga_winsys_context(swc);
266 struct vmw_region_relocation *reloc;
267 enum pipe_error ret;
268
269 assert(vswc->region.staged < vswc->region.reserved);
270
271 reloc = &vswc->region.relocs[vswc->region.used + vswc->region.staged];
272 reloc->where = where;
273 pb_reference(&reloc->buffer, vmw_pb_buffer(buffer));
274 reloc->offset = offset;
275
276 ++vswc->region.staged;
277
278 ret = pb_validate_add_buffer(vswc->validate, reloc->buffer, flags);
279 /* TODO: Update pipebuffer to reserve buffers and not fail here */
280 assert(ret == PIPE_OK);
281
282 /*
283 * Flush preemptively the FIFO commands to keep the GMR working set within
284 * the GMR pool size.
285 *
286 * This is necessary for applications like SPECviewperf that generate huge
287 * amounts of immediate vertex data, so that we don't pile up too much of
288 * that vertex data neither in the guest nor in the host.
289 *
290 * Note that in the current implementation if a region is referred twice in
291 * a command stream, it will be accounted twice. We could detect repeated
292 * regions and count only once, but there is no incentive to do that, since
293 * regions are typically short-lived; always referred in a single command;
294 * and at the worst we just flush the commands a bit sooner, which for the
295 * SVGA virtual device it's not a performance issue since flushing commands
296 * to the FIFO won't cause flushing in the host.
297 */
298 vswc->seen_regions += reloc->buffer->base.size;
299 if(vswc->seen_regions >= VMW_GMR_POOL_SIZE/2)
300 vswc->preemptive_flush = TRUE;
301 }
302
303
304 static void
305 vmw_swc_commit(struct svga_winsys_context *swc)
306 {
307 struct vmw_svga_winsys_context *vswc = vmw_svga_winsys_context(swc);
308
309 assert(vswc->command.reserved);
310 assert(vswc->command.used + vswc->command.reserved <= vswc->command.size);
311 vswc->command.used += vswc->command.reserved;
312 vswc->command.reserved = 0;
313
314 assert(vswc->surface.staged <= vswc->surface.reserved);
315 assert(vswc->surface.used + vswc->surface.staged <= vswc->surface.size);
316 vswc->surface.used += vswc->surface.staged;
317 vswc->surface.staged = 0;
318 vswc->surface.reserved = 0;
319
320 assert(vswc->region.staged <= vswc->region.reserved);
321 assert(vswc->region.used + vswc->region.staged <= vswc->region.size);
322 vswc->region.used += vswc->region.staged;
323 vswc->region.staged = 0;
324 vswc->region.reserved = 0;
325 }
326
327
328 static void
329 vmw_swc_destroy(struct svga_winsys_context *swc)
330 {
331 struct vmw_svga_winsys_context *vswc = vmw_svga_winsys_context(swc);
332 unsigned i;
333
334 for(i = 0; i < vswc->region.used; ++i) {
335 pb_reference(&vswc->region.relocs[i].buffer, NULL);
336 }
337
338 for(i = 0; i < vswc->surface.used; ++i) {
339 p_atomic_dec(&vswc->surface.handles[i]->validated);
340 vmw_svga_winsys_surface_reference(&vswc->surface.handles[i], NULL);
341 }
342 pb_validate_destroy(vswc->validate);
343 vmw_ioctl_context_destroy(vswc->vws, swc->cid);
344 FREE(vswc);
345 }
346
347
348 struct svga_winsys_context *
349 vmw_svga_winsys_context_create(struct svga_winsys_screen *sws)
350 {
351 struct vmw_winsys_screen *vws = vmw_winsys_screen(sws);
352 struct vmw_svga_winsys_context *vswc;
353
354 vswc = CALLOC_STRUCT(vmw_svga_winsys_context);
355 if(!vswc)
356 return NULL;
357
358 vswc->base.destroy = vmw_swc_destroy;
359 vswc->base.reserve = vmw_swc_reserve;
360 vswc->base.surface_relocation = vmw_swc_surface_relocation;
361 vswc->base.region_relocation = vmw_swc_region_relocation;
362 vswc->base.commit = vmw_swc_commit;
363 vswc->base.flush = vmw_swc_flush;
364
365 vswc->base.cid = vmw_ioctl_context_create(vws);
366
367 vswc->vws = vws;
368
369 vswc->command.size = VMW_COMMAND_SIZE;
370 vswc->surface.size = VMW_SURFACE_RELOCS;
371 vswc->region.size = VMW_REGION_RELOCS;
372
373 vswc->validate = pb_validate_create();
374 if(!vswc->validate) {
375 FREE(vswc);
376 return NULL;
377 }
378
379 return &vswc->base;
380 }
381
382