1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
|
/**************************************************************************
*
* Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
* All Rights Reserved.
*
* Permission is hereby granted, free of charge, to any person obtaining a
* copy of this software and associated documentation files (the
* "Software"), to deal in the Software without restriction, including
* without limitation the rights to use, copy, modify, merge, publish,
* distribute, sub license, and/or sell copies of the Software, and to
* permit persons to whom the Software is furnished to do so, subject to
* the following conditions:
*
* The above copyright notice and this permission notice (including the
* next paragraph) shall be included in all copies or substantial portions
* of the Software.
*
* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
* OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
* IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
* ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
* TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
* SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
*
**************************************************************************/
#include "intel_batchbuffer.h"
#include "intel_ioctl.h"
#include "intel_reg.h"
/* Relocations in kernel space:
* - pass dma buffer seperately
* - memory manager knows how to patch
* - pass list of dependent buffers
* - pass relocation list
*
* Either:
* - get back an offset for buffer to fire
* - memory manager knows how to fire buffer
*
* Really want the buffer to be AGP and pinned.
*
*/
/* Cliprect fence: The highest fence protecting a dma buffer
* containing explicit cliprect information. Like the old drawable
* lock but irq-driven. X server must wait for this fence to expire
* before changing cliprects [and then doing sw rendering?]. For
* other dma buffers, the scheduler will grab current cliprect info
* and mix into buffer. X server must hold the lock while changing
* cliprects??? Make per-drawable. Need cliprects in shared memory
* -- beats storing them with every cmd buffer in the queue.
*
* ==> X server must wait for this fence to expire before touching the
* framebuffer with new cliprects.
*
* ==> Cliprect-dependent buffers associated with a
* cliprect-timestamp. All of the buffers associated with a timestamp
* must go to hardware before any buffer with a newer timestamp.
*
* ==> Dma should be queued per-drawable for correct X/GL
* synchronization. Or can fences be used for this?
*
* Applies to: Blit operations, metaops, X server operations -- X
* server automatically waits on its own dma to complete before
* modifying cliprects ???
*/
static void
intel_dump_batchbuffer(GLuint offset, GLuint * ptr, GLuint count)
{
int i;
fprintf(stderr, "\n\n\nSTART BATCH (%d dwords):\n", count / 4);
for (i = 0; i < count / 4; i += 4)
fprintf(stderr, "0x%x:\t0x%08x 0x%08x 0x%08x 0x%08x\n",
offset + i * 4, ptr[i], ptr[i + 1], ptr[i + 2], ptr[i + 3]);
fprintf(stderr, "END BATCH\n\n\n");
}
void
intel_batchbuffer_reset(struct intel_batchbuffer *batch)
{
int i;
/*
* Get a new, free batchbuffer.
*/
batch->size = BATCH_SZ;
driBOData(batch->buffer, batch->size, NULL, 0);
driBOResetList(&batch->list);
/*
* Unreference buffers previously on the relocation list.
*/
for (i = 0; i < batch->nr_relocs; i++) {
struct buffer_reloc *r = &batch->reloc[i];
driBOUnReference(r->buf);
}
batch->list_count = 0;
batch->nr_relocs = 0;
batch->flags = 0;
/*
* We don't refcount the batchbuffer itself since we can't destroy it
* while it's on the list.
*/
driBOAddListItem(&batch->list, batch->buffer,
DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_EXE,
DRM_BO_MASK_MEM | DRM_BO_FLAG_EXE);
batch->map = driBOMap(batch->buffer, DRM_BO_FLAG_WRITE, 0);
batch->ptr = batch->map;
}
/*======================================================================
* Public functions
*/
struct intel_batchbuffer *
intel_batchbuffer_alloc(struct intel_context *intel)
{
struct intel_batchbuffer *batch = calloc(sizeof(*batch), 1);
batch->intel = intel;
driGenBuffers(intel->intelScreen->batchPool, "batchbuffer", 1,
&batch->buffer, 4096,
DRM_BO_FLAG_MEM_TT | DRM_BO_FLAG_EXE, 0);
batch->last_fence = NULL;
driBOCreateList(20, &batch->list);
intel_batchbuffer_reset(batch);
return batch;
}
void
intel_batchbuffer_free(struct intel_batchbuffer *batch)
{
if (batch->last_fence) {
driFenceFinish(batch->last_fence,
DRM_FENCE_TYPE_EXE | DRM_I915_FENCE_TYPE_RW, GL_FALSE);
driFenceUnReference(batch->last_fence);
batch->last_fence = NULL;
}
if (batch->map) {
driBOUnmap(batch->buffer);
batch->map = NULL;
}
driBOUnReference(batch->buffer);
batch->buffer = NULL;
free(batch);
}
/* TODO: Push this whole function into bufmgr.
*/
static void
do_flush_locked(struct intel_batchbuffer *batch,
GLuint used,
GLboolean ignore_cliprects, GLboolean allow_unlock)
{
GLuint *ptr;
GLuint i;
struct intel_context *intel = batch->intel;
unsigned fenceFlags;
struct _DriFenceObject *fo;
driBOValidateList(batch->intel->driFd, &batch->list);
/* Apply the relocations. This nasty map indicates to me that the
* whole task should be done internally by the memory manager, and
* that dma buffers probably need to be pinned within agp space.
*/
ptr = (GLuint *) driBOMap(batch->buffer, DRM_BO_FLAG_WRITE,
DRM_BO_HINT_ALLOW_UNFENCED_MAP);
for (i = 0; i < batch->nr_relocs; i++) {
struct buffer_reloc *r = &batch->reloc[i];
ptr[r->offset / 4] = driBOOffset(r->buf) + r->delta;
}
if (INTEL_DEBUG & DEBUG_BATCH)
intel_dump_batchbuffer(0, ptr, used);
driBOUnmap(batch->buffer);
batch->map = NULL;
/* Throw away non-effective packets. Won't work once we have
* hardware contexts which would preserve statechanges beyond a
* single buffer.
*/
if (!(intel->numClipRects == 0 && !ignore_cliprects)) {
intel_batch_ioctl(batch->intel,
driBOOffset(batch->buffer),
used, ignore_cliprects, allow_unlock);
}
/*
* Kernel fencing. The flags tells the kernel that we've
* programmed an MI_FLUSH.
*/
fenceFlags = DRM_I915_FENCE_FLAG_FLUSHED;
fo = driFenceBuffers(batch->intel->driFd,
"Batch fence", fenceFlags);
/*
* User space fencing.
*/
driBOFence(batch->buffer, fo);
if (driFenceType(fo) == DRM_FENCE_TYPE_EXE) {
/*
* Oops. We only validated a batch buffer. This means we
* didn't do any proper rendering. Discard this fence object.
*/
driFenceUnReference(fo);
} else {
driFenceUnReference(batch->last_fence);
batch->last_fence = fo;
for (i = 0; i < batch->nr_relocs; i++) {
struct buffer_reloc *r = &batch->reloc[i];
driBOFence(r->buf, fo);
}
}
if (intel->numClipRects == 0 && !ignore_cliprects) {
if (allow_unlock) {
UNLOCK_HARDWARE(intel);
sched_yield();
LOCK_HARDWARE(intel);
}
// intel->vtbl.lost_hardware(intel);
}
}
struct _DriFenceObject *
intel_batchbuffer_flush(struct intel_batchbuffer *batch)
{
struct intel_context *intel = batch->intel;
GLuint used = batch->ptr - batch->map;
GLboolean was_locked = intel->locked;
if (used == 0)
return batch->last_fence;
#define MI_FLUSH ((0<<29)|(4<<23))
/* Add the MI_BATCH_BUFFER_END. Always add an MI_FLUSH - this is a
* performance drain that we would like to avoid.
*/
if (used & 4) {
((int *) batch->ptr)[0] = MI_FLUSH;
((int *) batch->ptr)[1] = 0;
((int *) batch->ptr)[2] = MI_BATCH_BUFFER_END;
used += 12;
}
else {
((int *) batch->ptr)[0] = MI_FLUSH;
((int *) batch->ptr)[1] = MI_BATCH_BUFFER_END;
used += 8;
}
driBOUnmap(batch->buffer);
batch->ptr = NULL;
batch->map = NULL;
/* TODO: Just pass the relocation list and dma buffer up to the
* kernel.
*/
if (!was_locked)
LOCK_HARDWARE(intel);
do_flush_locked(batch, used, !(batch->flags & INTEL_BATCH_CLIPRECTS),
GL_FALSE);
if (!was_locked)
UNLOCK_HARDWARE(intel);
/* Reset the buffer:
*/
intel_batchbuffer_reset(batch);
return batch->last_fence;
}
void
intel_batchbuffer_finish(struct intel_batchbuffer *batch)
{
struct _DriFenceObject *fence = intel_batchbuffer_flush(batch);
driFenceReference(fence);
driFenceFinish(fence, 3, GL_FALSE);
driFenceUnReference(fence);
}
/* This is the only way buffers get added to the validate list.
*/
GLboolean
intel_batchbuffer_emit_reloc(struct intel_batchbuffer *batch,
struct _DriBufferObject *buffer,
GLuint flags, GLuint mask, GLuint delta)
{
assert(batch->nr_relocs < MAX_RELOCS);
driBOAddListItem(&batch->list, buffer, flags, mask);
{
struct buffer_reloc *r = &batch->reloc[batch->nr_relocs++];
driBOReference(buffer);
r->buf = buffer;
r->offset = batch->ptr - batch->map;
r->delta = delta;
}
batch->ptr += 4;
return GL_TRUE;
}
void
intel_batchbuffer_data(struct intel_batchbuffer *batch,
const void *data, GLuint bytes, GLuint flags)
{
assert((bytes & 3) == 0);
intel_batchbuffer_require_space(batch, bytes, flags);
memcpy(batch->ptr, data, bytes);
batch->ptr += bytes;
}
|