Commit | Line | Data |
---|---|---|
6ee73861 BS |
1 | /* |
2 | * Copyright (C) 2007 Ben Skeggs. | |
3 | * All Rights Reserved. | |
4 | * | |
5 | * Permission is hereby granted, free of charge, to any person obtaining | |
6 | * a copy of this software and associated documentation files (the | |
7 | * "Software"), to deal in the Software without restriction, including | |
8 | * without limitation the rights to use, copy, modify, merge, publish, | |
9 | * distribute, sublicense, and/or sell copies of the Software, and to | |
10 | * permit persons to whom the Software is furnished to do so, subject to | |
11 | * the following conditions: | |
12 | * | |
13 | * The above copyright notice and this permission notice (including the | |
14 | * next paragraph) shall be included in all copies or substantial | |
15 | * portions of the Software. | |
16 | * | |
17 | * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, | |
18 | * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF | |
19 | * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. | |
20 | * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE | |
21 | * LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION | |
22 | * OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION | |
23 | * WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE. | |
24 | * | |
25 | */ | |
26 | ||
27 | #include "drmP.h" | |
28 | #include "drm.h" | |
29 | #include "nouveau_drv.h" | |
c420b2dc | 30 | #include "nouveau_fifo.h" |
a8eaebc6 | 31 | #include "nouveau_ramht.h" |
332b242f | 32 | #include "nouveau_dma.h" |
a11c3198 | 33 | #include "nouveau_vm.h" |
332b242f | 34 | #include "nv50_evo.h" |
6ee73861 | 35 | |
2703c21a BS |
36 | struct nv50_graph_engine { |
37 | struct nouveau_exec_engine base; | |
38 | u32 ctxprog[512]; | |
39 | u32 ctxprog_size; | |
40 | u32 grctx_size; | |
41 | }; | |
42 | ||
6d6538a0 BS |
43 | static int |
44 | nv50_graph_init(struct drm_device *dev, int engine) | |
6ee73861 | 45 | { |
304424e1 | 46 | struct drm_nouveau_private *dev_priv = dev->dev_private; |
6d6538a0 BS |
47 | struct nv50_graph_engine *pgraph = nv_engine(dev, engine); |
48 | u32 units = nv_rd32(dev, 0x001540); | |
304424e1 MK |
49 | int i; |
50 | ||
6ee73861 BS |
51 | NV_DEBUG(dev, "\n"); |
52 | ||
6d6538a0 | 53 | /* master reset */ |
dce411cd BS |
54 | nv_mask(dev, 0x000200, 0x00201000, 0x00000000); |
55 | nv_mask(dev, 0x000200, 0x00201000, 0x00201000); | |
6d6538a0 BS |
56 | nv_wr32(dev, 0x40008c, 0x00000004); /* HW_CTX_SWITCH_ENABLED */ |
57 | ||
58 | /* reset/enable traps and interrupts */ | |
6ee73861 BS |
59 | nv_wr32(dev, 0x400804, 0xc0000000); |
60 | nv_wr32(dev, 0x406800, 0xc0000000); | |
61 | nv_wr32(dev, 0x400c04, 0xc0000000); | |
716abaa8 | 62 | nv_wr32(dev, 0x401800, 0xc0000000); |
6ee73861 BS |
63 | nv_wr32(dev, 0x405018, 0xc0000000); |
64 | nv_wr32(dev, 0x402000, 0xc0000000); | |
304424e1 | 65 | for (i = 0; i < 16; i++) { |
6d6538a0 BS |
66 | if (!(units & (1 << i))) |
67 | continue; | |
68 | ||
69 | if (dev_priv->chipset < 0xa0) { | |
70 | nv_wr32(dev, 0x408900 + (i << 12), 0xc0000000); | |
71 | nv_wr32(dev, 0x408e08 + (i << 12), 0xc0000000); | |
72 | nv_wr32(dev, 0x408314 + (i << 12), 0xc0000000); | |
73 | } else { | |
74 | nv_wr32(dev, 0x408600 + (i << 11), 0xc0000000); | |
75 | nv_wr32(dev, 0x408708 + (i << 11), 0xc0000000); | |
76 | nv_wr32(dev, 0x40831c + (i << 11), 0xc0000000); | |
304424e1 MK |
77 | } |
78 | } | |
79 | ||
6ee73861 | 80 | nv_wr32(dev, 0x400108, 0xffffffff); |
6d6538a0 BS |
81 | nv_wr32(dev, 0x400138, 0xffffffff); |
82 | nv_wr32(dev, 0x400100, 0xffffffff); | |
83 | nv_wr32(dev, 0x40013c, 0xffffffff); | |
6ee73861 | 84 | nv_wr32(dev, 0x400500, 0x00010001); |
6ee73861 | 85 | |
6d6538a0 BS |
86 | /* upload context program, initialise ctxctl defaults */ |
87 | nv_wr32(dev, 0x400324, 0x00000000); | |
88 | for (i = 0; i < pgraph->ctxprog_size; i++) | |
89 | nv_wr32(dev, 0x400328, pgraph->ctxprog[i]); | |
90 | nv_wr32(dev, 0x400824, 0x00000000); | |
91 | nv_wr32(dev, 0x400828, 0x00000000); | |
92 | nv_wr32(dev, 0x40082c, 0x00000000); | |
93 | nv_wr32(dev, 0x400830, 0x00000000); | |
94 | nv_wr32(dev, 0x400724, 0x00000000); | |
95 | nv_wr32(dev, 0x40032c, 0x00000000); | |
96 | nv_wr32(dev, 0x400320, 4); /* CTXCTL_CMD = NEWCTXDMA */ | |
97 | ||
98 | /* some unknown zcull magic */ | |
562af10c BS |
99 | switch (dev_priv->chipset & 0xf0) { |
100 | case 0x50: | |
101 | case 0x80: | |
102 | case 0x90: | |
103 | nv_wr32(dev, 0x402ca8, 0x00000800); | |
104 | break; | |
105 | case 0xa0: | |
106 | default: | |
107 | nv_wr32(dev, 0x402cc0, 0x00000000); | |
108 | if (dev_priv->chipset == 0xa0 || | |
109 | dev_priv->chipset == 0xaa || | |
110 | dev_priv->chipset == 0xac) { | |
111 | nv_wr32(dev, 0x402ca8, 0x00000802); | |
112 | } else { | |
113 | nv_wr32(dev, 0x402cc0, 0x00000000); | |
114 | nv_wr32(dev, 0x402ca8, 0x00000002); | |
115 | } | |
116 | ||
117 | break; | |
118 | } | |
119 | ||
120 | /* zero out zcull regions */ | |
121 | for (i = 0; i < 8; i++) { | |
122 | nv_wr32(dev, 0x402c20 + (i * 8), 0x00000000); | |
123 | nv_wr32(dev, 0x402c24 + (i * 8), 0x00000000); | |
124 | nv_wr32(dev, 0x402c28 + (i * 8), 0x00000000); | |
125 | nv_wr32(dev, 0x402c2c + (i * 8), 0x00000000); | |
126 | } | |
6ee73861 BS |
127 | |
128 | return 0; | |
129 | } | |
130 | ||
2703c21a | 131 | static int |
6c320fef | 132 | nv50_graph_fini(struct drm_device *dev, int engine, bool suspend) |
6ee73861 | 133 | { |
274fec93 | 134 | nv_wr32(dev, 0x40013c, 0x00000000); |
2703c21a | 135 | return 0; |
6ee73861 BS |
136 | } |
137 | ||
2703c21a BS |
138 | static int |
139 | nv50_graph_context_new(struct nouveau_channel *chan, int engine) | |
6ee73861 BS |
140 | { |
141 | struct drm_device *dev = chan->dev; | |
142 | struct drm_nouveau_private *dev_priv = dev->dev_private; | |
a8eaebc6 | 143 | struct nouveau_gpuobj *ramin = chan->ramin; |
2703c21a BS |
144 | struct nouveau_gpuobj *grctx = NULL; |
145 | struct nv50_graph_engine *pgraph = nv_engine(dev, engine); | |
6ee73861 BS |
146 | int hdr, ret; |
147 | ||
148 | NV_DEBUG(dev, "ch%d\n", chan->id); | |
149 | ||
2703c21a | 150 | ret = nouveau_gpuobj_new(dev, NULL, pgraph->grctx_size, 0, |
a8eaebc6 | 151 | NVOBJ_FLAG_ZERO_ALLOC | |
2703c21a | 152 | NVOBJ_FLAG_ZERO_FREE, &grctx); |
6ee73861 BS |
153 | if (ret) |
154 | return ret; | |
6ee73861 | 155 | |
ac94a343 | 156 | hdr = (dev_priv->chipset == 0x50) ? 0x200 : 0x20; |
b3beb167 | 157 | nv_wo32(ramin, hdr + 0x00, 0x00190002); |
2703c21a BS |
158 | nv_wo32(ramin, hdr + 0x04, grctx->vinst + grctx->size - 1); |
159 | nv_wo32(ramin, hdr + 0x08, grctx->vinst); | |
b3beb167 BS |
160 | nv_wo32(ramin, hdr + 0x0c, 0); |
161 | nv_wo32(ramin, hdr + 0x10, 0); | |
162 | nv_wo32(ramin, hdr + 0x14, 0x00010000); | |
6ee73861 | 163 | |
d58086de | 164 | nv50_grctx_fill(dev, grctx); |
2703c21a | 165 | nv_wo32(grctx, 0x00000, chan->ramin->vinst >> 12); |
6ee73861 | 166 | |
f56cb86f | 167 | dev_priv->engine.instmem.flush(dev); |
2703c21a BS |
168 | |
169 | atomic_inc(&chan->vm->engref[NVOBJ_ENGINE_GR]); | |
170 | chan->engctx[NVOBJ_ENGINE_GR] = grctx; | |
6ee73861 BS |
171 | return 0; |
172 | } | |
173 | ||
2703c21a BS |
174 | static void |
175 | nv50_graph_context_del(struct nouveau_channel *chan, int engine) | |
6ee73861 | 176 | { |
2703c21a | 177 | struct nouveau_gpuobj *grctx = chan->engctx[engine]; |
6ee73861 BS |
178 | struct drm_device *dev = chan->dev; |
179 | struct drm_nouveau_private *dev_priv = dev->dev_private; | |
ac94a343 | 180 | int i, hdr = (dev_priv->chipset == 0x50) ? 0x200 : 0x20; |
3945e475 | 181 | |
6ee73861 | 182 | for (i = hdr; i < hdr + 24; i += 4) |
a8eaebc6 | 183 | nv_wo32(chan->ramin, i, 0); |
f56cb86f | 184 | dev_priv->engine.instmem.flush(dev); |
6ee73861 | 185 | |
2703c21a | 186 | atomic_dec(&chan->vm->engref[engine]); |
5511d490 | 187 | nouveau_gpuobj_ref(NULL, &grctx); |
2703c21a | 188 | chan->engctx[engine] = NULL; |
6ee73861 BS |
189 | } |
190 | ||
191 | static int | |
2703c21a BS |
192 | nv50_graph_object_new(struct nouveau_channel *chan, int engine, |
193 | u32 handle, u16 class) | |
4ea52f89 BS |
194 | { |
195 | struct drm_device *dev = chan->dev; | |
196 | struct drm_nouveau_private *dev_priv = dev->dev_private; | |
197 | struct nouveau_gpuobj *obj = NULL; | |
198 | int ret; | |
199 | ||
200 | ret = nouveau_gpuobj_new(dev, chan, 16, 16, NVOBJ_FLAG_ZERO_FREE, &obj); | |
201 | if (ret) | |
202 | return ret; | |
203 | obj->engine = 1; | |
204 | obj->class = class; | |
205 | ||
206 | nv_wo32(obj, 0x00, class); | |
207 | nv_wo32(obj, 0x04, 0x00000000); | |
208 | nv_wo32(obj, 0x08, 0x00000000); | |
209 | nv_wo32(obj, 0x0c, 0x00000000); | |
210 | dev_priv->engine.instmem.flush(dev); | |
211 | ||
212 | ret = nouveau_ramht_insert(chan, handle, obj); | |
213 | nouveau_gpuobj_ref(NULL, &obj); | |
214 | return ret; | |
215 | } | |
216 | ||
2703c21a BS |
217 | static void |
218 | nv50_graph_tlb_flush(struct drm_device *dev, int engine) | |
56ac7475 | 219 | { |
a11c3198 | 220 | nv50_vm_flush_engine(dev, 0); |
56ac7475 BS |
221 | } |
222 | ||
2703c21a BS |
223 | static void |
224 | nv84_graph_tlb_flush(struct drm_device *dev, int engine) | |
56ac7475 BS |
225 | { |
226 | struct drm_nouveau_private *dev_priv = dev->dev_private; | |
227 | struct nouveau_timer_engine *ptimer = &dev_priv->engine.timer; | |
228 | bool idle, timeout = false; | |
229 | unsigned long flags; | |
230 | u64 start; | |
231 | u32 tmp; | |
232 | ||
233 | spin_lock_irqsave(&dev_priv->context_switch_lock, flags); | |
234 | nv_mask(dev, 0x400500, 0x00000001, 0x00000000); | |
235 | ||
236 | start = ptimer->read(dev); | |
237 | do { | |
238 | idle = true; | |
239 | ||
240 | for (tmp = nv_rd32(dev, 0x400380); tmp && idle; tmp >>= 3) { | |
241 | if ((tmp & 7) == 1) | |
242 | idle = false; | |
243 | } | |
244 | ||
245 | for (tmp = nv_rd32(dev, 0x400384); tmp && idle; tmp >>= 3) { | |
246 | if ((tmp & 7) == 1) | |
247 | idle = false; | |
248 | } | |
249 | ||
250 | for (tmp = nv_rd32(dev, 0x400388); tmp && idle; tmp >>= 3) { | |
251 | if ((tmp & 7) == 1) | |
252 | idle = false; | |
253 | } | |
254 | } while (!idle && !(timeout = ptimer->read(dev) - start > 2000000000)); | |
255 | ||
256 | if (timeout) { | |
257 | NV_ERROR(dev, "PGRAPH TLB flush idle timeout fail: " | |
258 | "0x%08x 0x%08x 0x%08x 0x%08x\n", | |
259 | nv_rd32(dev, 0x400700), nv_rd32(dev, 0x400380), | |
260 | nv_rd32(dev, 0x400384), nv_rd32(dev, 0x400388)); | |
261 | } | |
262 | ||
a11c3198 | 263 | nv50_vm_flush_engine(dev, 0); |
56ac7475 BS |
264 | |
265 | nv_mask(dev, 0x400500, 0x00000001, 0x00000001); | |
266 | spin_unlock_irqrestore(&dev_priv->context_switch_lock, flags); | |
267 | } | |
274fec93 | 268 | |
f9ec8f6c | 269 | static struct nouveau_enum nv50_mp_exec_error_names[] = { |
bb9b18a3 BS |
270 | { 3, "STACK_UNDERFLOW", NULL }, |
271 | { 4, "QUADON_ACTIVE", NULL }, | |
272 | { 8, "TIMEOUT", NULL }, | |
273 | { 0x10, "INVALID_OPCODE", NULL }, | |
274 | { 0x40, "BREAKPOINT", NULL }, | |
274fec93 BS |
275 | {} |
276 | }; | |
277 | ||
278 | static struct nouveau_bitfield nv50_graph_trap_m2mf[] = { | |
279 | { 0x00000001, "NOTIFY" }, | |
280 | { 0x00000002, "IN" }, | |
281 | { 0x00000004, "OUT" }, | |
282 | {} | |
283 | }; | |
284 | ||
285 | static struct nouveau_bitfield nv50_graph_trap_vfetch[] = { | |
286 | { 0x00000001, "FAULT" }, | |
287 | {} | |
288 | }; | |
289 | ||
290 | static struct nouveau_bitfield nv50_graph_trap_strmout[] = { | |
291 | { 0x00000001, "FAULT" }, | |
292 | {} | |
293 | }; | |
294 | ||
295 | static struct nouveau_bitfield nv50_graph_trap_ccache[] = { | |
296 | { 0x00000001, "FAULT" }, | |
297 | {} | |
298 | }; | |
299 | ||
300 | /* There must be a *lot* of these. Will take some time to gather them up. */ | |
6effe393 | 301 | struct nouveau_enum nv50_data_error_names[] = { |
bb9b18a3 BS |
302 | { 0x00000003, "INVALID_QUERY_OR_TEXTURE", NULL }, |
303 | { 0x00000004, "INVALID_VALUE", NULL }, | |
304 | { 0x00000005, "INVALID_ENUM", NULL }, | |
305 | { 0x00000008, "INVALID_OBJECT", NULL }, | |
306 | { 0x00000009, "READ_ONLY_OBJECT", NULL }, | |
307 | { 0x0000000a, "SUPERVISOR_OBJECT", NULL }, | |
308 | { 0x0000000b, "INVALID_ADDRESS_ALIGNMENT", NULL }, | |
309 | { 0x0000000c, "INVALID_BITFIELD", NULL }, | |
310 | { 0x0000000d, "BEGIN_END_ACTIVE", NULL }, | |
311 | { 0x0000000e, "SEMANTIC_COLOR_BACK_OVER_LIMIT", NULL }, | |
312 | { 0x0000000f, "VIEWPORT_ID_NEEDS_GP", NULL }, | |
313 | { 0x00000010, "RT_DOUBLE_BIND", NULL }, | |
314 | { 0x00000011, "RT_TYPES_MISMATCH", NULL }, | |
315 | { 0x00000012, "RT_LINEAR_WITH_ZETA", NULL }, | |
316 | { 0x00000015, "FP_TOO_FEW_REGS", NULL }, | |
317 | { 0x00000016, "ZETA_FORMAT_CSAA_MISMATCH", NULL }, | |
318 | { 0x00000017, "RT_LINEAR_WITH_MSAA", NULL }, | |
319 | { 0x00000018, "FP_INTERPOLANT_START_OVER_LIMIT", NULL }, | |
320 | { 0x00000019, "SEMANTIC_LAYER_OVER_LIMIT", NULL }, | |
321 | { 0x0000001a, "RT_INVALID_ALIGNMENT", NULL }, | |
322 | { 0x0000001b, "SAMPLER_OVER_LIMIT", NULL }, | |
323 | { 0x0000001c, "TEXTURE_OVER_LIMIT", NULL }, | |
324 | { 0x0000001e, "GP_TOO_MANY_OUTPUTS", NULL }, | |
325 | { 0x0000001f, "RT_BPP128_WITH_MS8", NULL }, | |
326 | { 0x00000021, "Z_OUT_OF_BOUNDS", NULL }, | |
327 | { 0x00000023, "XY_OUT_OF_BOUNDS", NULL }, | |
547e6c7f | 328 | { 0x00000024, "VP_ZERO_INPUTS", NULL }, |
bb9b18a3 BS |
329 | { 0x00000027, "CP_MORE_PARAMS_THAN_SHARED", NULL }, |
330 | { 0x00000028, "CP_NO_REG_SPACE_STRIPED", NULL }, | |
331 | { 0x00000029, "CP_NO_REG_SPACE_PACKED", NULL }, | |
332 | { 0x0000002a, "CP_NOT_ENOUGH_WARPS", NULL }, | |
333 | { 0x0000002b, "CP_BLOCK_SIZE_MISMATCH", NULL }, | |
334 | { 0x0000002c, "CP_NOT_ENOUGH_LOCAL_WARPS", NULL }, | |
335 | { 0x0000002d, "CP_NOT_ENOUGH_STACK_WARPS", NULL }, | |
336 | { 0x0000002e, "CP_NO_BLOCKDIM_LATCH", NULL }, | |
337 | { 0x00000031, "ENG2D_FORMAT_MISMATCH", NULL }, | |
338 | { 0x0000003f, "PRIMITIVE_ID_NEEDS_GP", NULL }, | |
339 | { 0x00000044, "SEMANTIC_VIEWPORT_OVER_LIMIT", NULL }, | |
340 | { 0x00000045, "SEMANTIC_COLOR_FRONT_OVER_LIMIT", NULL }, | |
341 | { 0x00000046, "LAYER_ID_NEEDS_GP", NULL }, | |
342 | { 0x00000047, "SEMANTIC_CLIP_OVER_LIMIT", NULL }, | |
343 | { 0x00000048, "SEMANTIC_PTSZ_OVER_LIMIT", NULL }, | |
274fec93 BS |
344 | {} |
345 | }; | |
346 | ||
347 | static struct nouveau_bitfield nv50_graph_intr[] = { | |
348 | { 0x00000001, "NOTIFY" }, | |
349 | { 0x00000002, "COMPUTE_QUERY" }, | |
350 | { 0x00000010, "ILLEGAL_MTHD" }, | |
351 | { 0x00000020, "ILLEGAL_CLASS" }, | |
352 | { 0x00000040, "DOUBLE_NOTIFY" }, | |
353 | { 0x00001000, "CONTEXT_SWITCH" }, | |
354 | { 0x00010000, "BUFFER_NOTIFY" }, | |
355 | { 0x00100000, "DATA_ERROR" }, | |
356 | { 0x00200000, "TRAP" }, | |
357 | { 0x01000000, "SINGLE_STEP" }, | |
358 | {} | |
359 | }; | |
360 | ||
361 | static void | |
362 | nv50_pgraph_mp_trap(struct drm_device *dev, int tpid, int display) | |
363 | { | |
364 | struct drm_nouveau_private *dev_priv = dev->dev_private; | |
365 | uint32_t units = nv_rd32(dev, 0x1540); | |
366 | uint32_t addr, mp10, status, pc, oplow, ophigh; | |
367 | int i; | |
368 | int mps = 0; | |
369 | for (i = 0; i < 4; i++) { | |
370 | if (!(units & 1 << (i+24))) | |
371 | continue; | |
372 | if (dev_priv->chipset < 0xa0) | |
373 | addr = 0x408200 + (tpid << 12) + (i << 7); | |
374 | else | |
375 | addr = 0x408100 + (tpid << 11) + (i << 7); | |
376 | mp10 = nv_rd32(dev, addr + 0x10); | |
377 | status = nv_rd32(dev, addr + 0x14); | |
378 | if (!status) | |
379 | continue; | |
380 | if (display) { | |
381 | nv_rd32(dev, addr + 0x20); | |
382 | pc = nv_rd32(dev, addr + 0x24); | |
383 | oplow = nv_rd32(dev, addr + 0x70); | |
0b89a072 | 384 | ophigh = nv_rd32(dev, addr + 0x74); |
274fec93 BS |
385 | NV_INFO(dev, "PGRAPH_TRAP_MP_EXEC - " |
386 | "TP %d MP %d: ", tpid, i); | |
387 | nouveau_enum_print(nv50_mp_exec_error_names, status); | |
388 | printk(" at %06x warp %d, opcode %08x %08x\n", | |
389 | pc&0xffffff, pc >> 24, | |
390 | oplow, ophigh); | |
391 | } | |
392 | nv_wr32(dev, addr + 0x10, mp10); | |
393 | nv_wr32(dev, addr + 0x14, 0); | |
394 | mps++; | |
395 | } | |
396 | if (!mps && display) | |
397 | NV_INFO(dev, "PGRAPH_TRAP_MP_EXEC - TP %d: " | |
398 | "No MPs claiming errors?\n", tpid); | |
399 | } | |
400 | ||
401 | static void | |
402 | nv50_pgraph_tp_trap(struct drm_device *dev, int type, uint32_t ustatus_old, | |
403 | uint32_t ustatus_new, int display, const char *name) | |
404 | { | |
405 | struct drm_nouveau_private *dev_priv = dev->dev_private; | |
406 | int tps = 0; | |
407 | uint32_t units = nv_rd32(dev, 0x1540); | |
408 | int i, r; | |
409 | uint32_t ustatus_addr, ustatus; | |
410 | for (i = 0; i < 16; i++) { | |
411 | if (!(units & (1 << i))) | |
412 | continue; | |
413 | if (dev_priv->chipset < 0xa0) | |
414 | ustatus_addr = ustatus_old + (i << 12); | |
415 | else | |
416 | ustatus_addr = ustatus_new + (i << 11); | |
417 | ustatus = nv_rd32(dev, ustatus_addr) & 0x7fffffff; | |
418 | if (!ustatus) | |
419 | continue; | |
420 | tps++; | |
421 | switch (type) { | |
422 | case 6: /* texture error... unknown for now */ | |
274fec93 BS |
423 | if (display) { |
424 | NV_ERROR(dev, "magic set %d:\n", i); | |
425 | for (r = ustatus_addr + 4; r <= ustatus_addr + 0x10; r += 4) | |
426 | NV_ERROR(dev, "\t0x%08x: 0x%08x\n", r, | |
427 | nv_rd32(dev, r)); | |
428 | } | |
429 | break; | |
430 | case 7: /* MP error */ | |
c983e6f6 | 431 | if (ustatus & 0x04030000) { |
274fec93 | 432 | nv50_pgraph_mp_trap(dev, i, display); |
c983e6f6 | 433 | ustatus &= ~0x04030000; |
274fec93 BS |
434 | } |
435 | break; | |
436 | case 8: /* TPDMA error */ | |
437 | { | |
438 | uint32_t e0c = nv_rd32(dev, ustatus_addr + 4); | |
439 | uint32_t e10 = nv_rd32(dev, ustatus_addr + 8); | |
440 | uint32_t e14 = nv_rd32(dev, ustatus_addr + 0xc); | |
441 | uint32_t e18 = nv_rd32(dev, ustatus_addr + 0x10); | |
442 | uint32_t e1c = nv_rd32(dev, ustatus_addr + 0x14); | |
443 | uint32_t e20 = nv_rd32(dev, ustatus_addr + 0x18); | |
444 | uint32_t e24 = nv_rd32(dev, ustatus_addr + 0x1c); | |
274fec93 BS |
445 | /* 2d engine destination */ |
446 | if (ustatus & 0x00000010) { | |
447 | if (display) { | |
448 | NV_INFO(dev, "PGRAPH_TRAP_TPDMA_2D - TP %d - Unknown fault at address %02x%08x\n", | |
449 | i, e14, e10); | |
450 | NV_INFO(dev, "PGRAPH_TRAP_TPDMA_2D - TP %d - e0c: %08x, e18: %08x, e1c: %08x, e20: %08x, e24: %08x\n", | |
451 | i, e0c, e18, e1c, e20, e24); | |
452 | } | |
453 | ustatus &= ~0x00000010; | |
454 | } | |
455 | /* Render target */ | |
456 | if (ustatus & 0x00000040) { | |
457 | if (display) { | |
458 | NV_INFO(dev, "PGRAPH_TRAP_TPDMA_RT - TP %d - Unknown fault at address %02x%08x\n", | |
459 | i, e14, e10); | |
460 | NV_INFO(dev, "PGRAPH_TRAP_TPDMA_RT - TP %d - e0c: %08x, e18: %08x, e1c: %08x, e20: %08x, e24: %08x\n", | |
461 | i, e0c, e18, e1c, e20, e24); | |
462 | } | |
463 | ustatus &= ~0x00000040; | |
464 | } | |
465 | /* CUDA memory: l[], g[] or stack. */ | |
466 | if (ustatus & 0x00000080) { | |
467 | if (display) { | |
468 | if (e18 & 0x80000000) { | |
469 | /* g[] read fault? */ | |
470 | NV_INFO(dev, "PGRAPH_TRAP_TPDMA - TP %d - Global read fault at address %02x%08x\n", | |
471 | i, e14, e10 | ((e18 >> 24) & 0x1f)); | |
472 | e18 &= ~0x1f000000; | |
473 | } else if (e18 & 0xc) { | |
474 | /* g[] write fault? */ | |
475 | NV_INFO(dev, "PGRAPH_TRAP_TPDMA - TP %d - Global write fault at address %02x%08x\n", | |
476 | i, e14, e10 | ((e18 >> 7) & 0x1f)); | |
477 | e18 &= ~0x00000f80; | |
478 | } else { | |
479 | NV_INFO(dev, "PGRAPH_TRAP_TPDMA - TP %d - Unknown CUDA fault at address %02x%08x\n", | |
480 | i, e14, e10); | |
481 | } | |
482 | NV_INFO(dev, "PGRAPH_TRAP_TPDMA - TP %d - e0c: %08x, e18: %08x, e1c: %08x, e20: %08x, e24: %08x\n", | |
483 | i, e0c, e18, e1c, e20, e24); | |
484 | } | |
485 | ustatus &= ~0x00000080; | |
486 | } | |
487 | } | |
488 | break; | |
489 | } | |
490 | if (ustatus) { | |
491 | if (display) | |
492 | NV_INFO(dev, "%s - TP%d: Unhandled ustatus 0x%08x\n", name, i, ustatus); | |
493 | } | |
494 | nv_wr32(dev, ustatus_addr, 0xc0000000); | |
495 | } | |
496 | ||
497 | if (!tps && display) | |
498 | NV_INFO(dev, "%s - No TPs claiming errors?\n", name); | |
499 | } | |
500 | ||
501 | static int | |
502 | nv50_pgraph_trap_handler(struct drm_device *dev, u32 display, u64 inst, u32 chid) | |
503 | { | |
504 | u32 status = nv_rd32(dev, 0x400108); | |
505 | u32 ustatus; | |
506 | ||
507 | if (!status && display) { | |
508 | NV_INFO(dev, "PGRAPH - TRAP: no units reporting traps?\n"); | |
509 | return 1; | |
510 | } | |
511 | ||
512 | /* DISPATCH: Relays commands to other units and handles NOTIFY, | |
513 | * COND, QUERY. If you get a trap from it, the command is still stuck | |
514 | * in DISPATCH and you need to do something about it. */ | |
515 | if (status & 0x001) { | |
516 | ustatus = nv_rd32(dev, 0x400804) & 0x7fffffff; | |
517 | if (!ustatus && display) { | |
518 | NV_INFO(dev, "PGRAPH_TRAP_DISPATCH - no ustatus?\n"); | |
519 | } | |
520 | ||
521 | nv_wr32(dev, 0x400500, 0x00000000); | |
522 | ||
523 | /* Known to be triggered by screwed up NOTIFY and COND... */ | |
524 | if (ustatus & 0x00000001) { | |
525 | u32 addr = nv_rd32(dev, 0x400808); | |
526 | u32 subc = (addr & 0x00070000) >> 16; | |
527 | u32 mthd = (addr & 0x00001ffc); | |
528 | u32 datal = nv_rd32(dev, 0x40080c); | |
529 | u32 datah = nv_rd32(dev, 0x400810); | |
530 | u32 class = nv_rd32(dev, 0x400814); | |
531 | u32 r848 = nv_rd32(dev, 0x400848); | |
532 | ||
533 | NV_INFO(dev, "PGRAPH - TRAP DISPATCH_FAULT\n"); | |
534 | if (display && (addr & 0x80000000)) { | |
535 | NV_INFO(dev, "PGRAPH - ch %d (0x%010llx) " | |
536 | "subc %d class 0x%04x mthd 0x%04x " | |
537 | "data 0x%08x%08x " | |
538 | "400808 0x%08x 400848 0x%08x\n", | |
539 | chid, inst, subc, class, mthd, datah, | |
540 | datal, addr, r848); | |
541 | } else | |
542 | if (display) { | |
543 | NV_INFO(dev, "PGRAPH - no stuck command?\n"); | |
544 | } | |
545 | ||
546 | nv_wr32(dev, 0x400808, 0); | |
547 | nv_wr32(dev, 0x4008e8, nv_rd32(dev, 0x4008e8) & 3); | |
548 | nv_wr32(dev, 0x400848, 0); | |
549 | ustatus &= ~0x00000001; | |
550 | } | |
551 | ||
552 | if (ustatus & 0x00000002) { | |
553 | u32 addr = nv_rd32(dev, 0x40084c); | |
554 | u32 subc = (addr & 0x00070000) >> 16; | |
555 | u32 mthd = (addr & 0x00001ffc); | |
556 | u32 data = nv_rd32(dev, 0x40085c); | |
557 | u32 class = nv_rd32(dev, 0x400814); | |
558 | ||
559 | NV_INFO(dev, "PGRAPH - TRAP DISPATCH_QUERY\n"); | |
560 | if (display && (addr & 0x80000000)) { | |
561 | NV_INFO(dev, "PGRAPH - ch %d (0x%010llx) " | |
562 | "subc %d class 0x%04x mthd 0x%04x " | |
563 | "data 0x%08x 40084c 0x%08x\n", | |
564 | chid, inst, subc, class, mthd, | |
565 | data, addr); | |
566 | } else | |
567 | if (display) { | |
568 | NV_INFO(dev, "PGRAPH - no stuck command?\n"); | |
569 | } | |
570 | ||
571 | nv_wr32(dev, 0x40084c, 0); | |
572 | ustatus &= ~0x00000002; | |
573 | } | |
574 | ||
575 | if (ustatus && display) { | |
576 | NV_INFO(dev, "PGRAPH - TRAP_DISPATCH (unknown " | |
577 | "0x%08x)\n", ustatus); | |
578 | } | |
579 | ||
580 | nv_wr32(dev, 0x400804, 0xc0000000); | |
581 | nv_wr32(dev, 0x400108, 0x001); | |
582 | status &= ~0x001; | |
583 | if (!status) | |
584 | return 0; | |
585 | } | |
586 | ||
587 | /* M2MF: Memory to memory copy engine. */ | |
588 | if (status & 0x002) { | |
589 | u32 ustatus = nv_rd32(dev, 0x406800) & 0x7fffffff; | |
590 | if (display) { | |
591 | NV_INFO(dev, "PGRAPH - TRAP_M2MF"); | |
592 | nouveau_bitfield_print(nv50_graph_trap_m2mf, ustatus); | |
593 | printk("\n"); | |
594 | NV_INFO(dev, "PGRAPH - TRAP_M2MF %08x %08x %08x %08x\n", | |
595 | nv_rd32(dev, 0x406804), nv_rd32(dev, 0x406808), | |
596 | nv_rd32(dev, 0x40680c), nv_rd32(dev, 0x406810)); | |
597 | ||
598 | } | |
599 | ||
600 | /* No sane way found yet -- just reset the bugger. */ | |
601 | nv_wr32(dev, 0x400040, 2); | |
602 | nv_wr32(dev, 0x400040, 0); | |
603 | nv_wr32(dev, 0x406800, 0xc0000000); | |
604 | nv_wr32(dev, 0x400108, 0x002); | |
605 | status &= ~0x002; | |
606 | } | |
607 | ||
608 | /* VFETCH: Fetches data from vertex buffers. */ | |
609 | if (status & 0x004) { | |
610 | u32 ustatus = nv_rd32(dev, 0x400c04) & 0x7fffffff; | |
611 | if (display) { | |
612 | NV_INFO(dev, "PGRAPH - TRAP_VFETCH"); | |
613 | nouveau_bitfield_print(nv50_graph_trap_vfetch, ustatus); | |
614 | printk("\n"); | |
615 | NV_INFO(dev, "PGRAPH - TRAP_VFETCH %08x %08x %08x %08x\n", | |
616 | nv_rd32(dev, 0x400c00), nv_rd32(dev, 0x400c08), | |
617 | nv_rd32(dev, 0x400c0c), nv_rd32(dev, 0x400c10)); | |
618 | } | |
619 | ||
620 | nv_wr32(dev, 0x400c04, 0xc0000000); | |
621 | nv_wr32(dev, 0x400108, 0x004); | |
622 | status &= ~0x004; | |
623 | } | |
624 | ||
625 | /* STRMOUT: DirectX streamout / OpenGL transform feedback. */ | |
626 | if (status & 0x008) { | |
627 | ustatus = nv_rd32(dev, 0x401800) & 0x7fffffff; | |
628 | if (display) { | |
629 | NV_INFO(dev, "PGRAPH - TRAP_STRMOUT"); | |
630 | nouveau_bitfield_print(nv50_graph_trap_strmout, ustatus); | |
631 | printk("\n"); | |
632 | NV_INFO(dev, "PGRAPH - TRAP_STRMOUT %08x %08x %08x %08x\n", | |
633 | nv_rd32(dev, 0x401804), nv_rd32(dev, 0x401808), | |
634 | nv_rd32(dev, 0x40180c), nv_rd32(dev, 0x401810)); | |
635 | ||
636 | } | |
637 | ||
638 | /* No sane way found yet -- just reset the bugger. */ | |
639 | nv_wr32(dev, 0x400040, 0x80); | |
640 | nv_wr32(dev, 0x400040, 0); | |
641 | nv_wr32(dev, 0x401800, 0xc0000000); | |
642 | nv_wr32(dev, 0x400108, 0x008); | |
643 | status &= ~0x008; | |
644 | } | |
645 | ||
646 | /* CCACHE: Handles code and c[] caches and fills them. */ | |
647 | if (status & 0x010) { | |
648 | ustatus = nv_rd32(dev, 0x405018) & 0x7fffffff; | |
649 | if (display) { | |
650 | NV_INFO(dev, "PGRAPH - TRAP_CCACHE"); | |
651 | nouveau_bitfield_print(nv50_graph_trap_ccache, ustatus); | |
652 | printk("\n"); | |
653 | NV_INFO(dev, "PGRAPH - TRAP_CCACHE %08x %08x %08x %08x" | |
654 | " %08x %08x %08x\n", | |
4dcf905c MS |
655 | nv_rd32(dev, 0x405000), nv_rd32(dev, 0x405004), |
656 | nv_rd32(dev, 0x405008), nv_rd32(dev, 0x40500c), | |
657 | nv_rd32(dev, 0x405010), nv_rd32(dev, 0x405014), | |
658 | nv_rd32(dev, 0x40501c)); | |
274fec93 BS |
659 | |
660 | } | |
661 | ||
662 | nv_wr32(dev, 0x405018, 0xc0000000); | |
663 | nv_wr32(dev, 0x400108, 0x010); | |
664 | status &= ~0x010; | |
665 | } | |
666 | ||
667 | /* Unknown, not seen yet... 0x402000 is the only trap status reg | |
668 | * remaining, so try to handle it anyway. Perhaps related to that | |
669 | * unknown DMA slot on tesla? */ | |
670 | if (status & 0x20) { | |
671 | ustatus = nv_rd32(dev, 0x402000) & 0x7fffffff; | |
672 | if (display) | |
673 | NV_INFO(dev, "PGRAPH - TRAP_UNKC04 0x%08x\n", ustatus); | |
674 | nv_wr32(dev, 0x402000, 0xc0000000); | |
675 | /* no status modifiction on purpose */ | |
676 | } | |
677 | ||
678 | /* TEXTURE: CUDA texturing units */ | |
679 | if (status & 0x040) { | |
680 | nv50_pgraph_tp_trap(dev, 6, 0x408900, 0x408600, display, | |
681 | "PGRAPH - TRAP_TEXTURE"); | |
682 | nv_wr32(dev, 0x400108, 0x040); | |
683 | status &= ~0x040; | |
684 | } | |
685 | ||
686 | /* MP: CUDA execution engines. */ | |
687 | if (status & 0x080) { | |
688 | nv50_pgraph_tp_trap(dev, 7, 0x408314, 0x40831c, display, | |
689 | "PGRAPH - TRAP_MP"); | |
690 | nv_wr32(dev, 0x400108, 0x080); | |
691 | status &= ~0x080; | |
692 | } | |
693 | ||
694 | /* TPDMA: Handles TP-initiated uncached memory accesses: | |
695 | * l[], g[], stack, 2d surfaces, render targets. */ | |
696 | if (status & 0x100) { | |
697 | nv50_pgraph_tp_trap(dev, 8, 0x408e08, 0x408708, display, | |
698 | "PGRAPH - TRAP_TPDMA"); | |
699 | nv_wr32(dev, 0x400108, 0x100); | |
700 | status &= ~0x100; | |
701 | } | |
702 | ||
703 | if (status) { | |
704 | if (display) | |
705 | NV_INFO(dev, "PGRAPH - TRAP: unknown 0x%08x\n", status); | |
706 | nv_wr32(dev, 0x400108, status); | |
707 | } | |
708 | ||
709 | return 1; | |
710 | } | |
711 | ||
7ff5441e | 712 | int |
274fec93 BS |
713 | nv50_graph_isr_chid(struct drm_device *dev, u64 inst) |
714 | { | |
c420b2dc | 715 | struct nouveau_fifo_priv *pfifo = nv_engine(dev, NVOBJ_ENGINE_FIFO); |
274fec93 BS |
716 | struct drm_nouveau_private *dev_priv = dev->dev_private; |
717 | struct nouveau_channel *chan; | |
718 | unsigned long flags; | |
719 | int i; | |
720 | ||
721 | spin_lock_irqsave(&dev_priv->channels.lock, flags); | |
c420b2dc | 722 | for (i = 0; i < pfifo->channels; i++) { |
274fec93 BS |
723 | chan = dev_priv->channels.ptr[i]; |
724 | if (!chan || !chan->ramin) | |
725 | continue; | |
726 | ||
727 | if (inst == chan->ramin->vinst) | |
728 | break; | |
729 | } | |
730 | spin_unlock_irqrestore(&dev_priv->channels.lock, flags); | |
731 | return i; | |
732 | } | |
733 | ||
734 | static void | |
735 | nv50_graph_isr(struct drm_device *dev) | |
736 | { | |
737 | u32 stat; | |
738 | ||
739 | while ((stat = nv_rd32(dev, 0x400100))) { | |
740 | u64 inst = (u64)(nv_rd32(dev, 0x40032c) & 0x0fffffff) << 12; | |
741 | u32 chid = nv50_graph_isr_chid(dev, inst); | |
742 | u32 addr = nv_rd32(dev, NV04_PGRAPH_TRAPPED_ADDR); | |
743 | u32 subc = (addr & 0x00070000) >> 16; | |
744 | u32 mthd = (addr & 0x00001ffc); | |
745 | u32 data = nv_rd32(dev, NV04_PGRAPH_TRAPPED_DATA); | |
746 | u32 class = nv_rd32(dev, 0x400814); | |
747 | u32 show = stat; | |
748 | ||
749 | if (stat & 0x00000010) { | |
750 | if (!nouveau_gpuobj_mthd_call2(dev, chid, class, | |
751 | mthd, data)) | |
752 | show &= ~0x00000010; | |
753 | } | |
754 | ||
274fec93 BS |
755 | show = (show && nouveau_ratelimit()) ? show : 0; |
756 | ||
757 | if (show & 0x00100000) { | |
758 | u32 ecode = nv_rd32(dev, 0x400110); | |
759 | NV_INFO(dev, "PGRAPH - DATA_ERROR "); | |
760 | nouveau_enum_print(nv50_data_error_names, ecode); | |
761 | printk("\n"); | |
762 | } | |
763 | ||
764 | if (stat & 0x00200000) { | |
765 | if (!nv50_pgraph_trap_handler(dev, show, inst, chid)) | |
766 | show &= ~0x00200000; | |
767 | } | |
768 | ||
769 | nv_wr32(dev, 0x400100, stat); | |
770 | nv_wr32(dev, 0x400500, 0x00010001); | |
771 | ||
772 | if (show) { | |
773 | NV_INFO(dev, "PGRAPH -"); | |
774 | nouveau_bitfield_print(nv50_graph_intr, show); | |
775 | printk("\n"); | |
776 | NV_INFO(dev, "PGRAPH - ch %d (0x%010llx) subc %d " | |
777 | "class 0x%04x mthd 0x%04x data 0x%08x\n", | |
778 | chid, inst, subc, class, mthd, data); | |
6fdb383e | 779 | nv50_fb_vm_trap(dev, 1); |
274fec93 BS |
780 | } |
781 | } | |
782 | ||
783 | if (nv_rd32(dev, 0x400824) & (1 << 31)) | |
784 | nv_wr32(dev, 0x400824, nv_rd32(dev, 0x400824) & ~(1 << 31)); | |
785 | } | |
2703c21a BS |
786 | |
787 | static void | |
788 | nv50_graph_destroy(struct drm_device *dev, int engine) | |
789 | { | |
790 | struct nv50_graph_engine *pgraph = nv_engine(dev, engine); | |
791 | ||
792 | NVOBJ_ENGINE_DEL(dev, GR); | |
793 | ||
794 | nouveau_irq_unregister(dev, 12); | |
795 | kfree(pgraph); | |
796 | } | |
797 | ||
798 | int | |
799 | nv50_graph_create(struct drm_device *dev) | |
800 | { | |
801 | struct drm_nouveau_private *dev_priv = dev->dev_private; | |
802 | struct nv50_graph_engine *pgraph; | |
2703c21a BS |
803 | int ret; |
804 | ||
805 | pgraph = kzalloc(sizeof(*pgraph),GFP_KERNEL); | |
806 | if (!pgraph) | |
807 | return -ENOMEM; | |
808 | ||
d58086de BS |
809 | ret = nv50_grctx_init(dev, pgraph->ctxprog, ARRAY_SIZE(pgraph->ctxprog), |
810 | &pgraph->ctxprog_size, | |
811 | &pgraph->grctx_size); | |
2703c21a BS |
812 | if (ret) { |
813 | NV_ERROR(dev, "PGRAPH: ctxprog build failed\n"); | |
2703c21a BS |
814 | kfree(pgraph); |
815 | return 0; | |
816 | } | |
817 | ||
2703c21a BS |
818 | pgraph->base.destroy = nv50_graph_destroy; |
819 | pgraph->base.init = nv50_graph_init; | |
820 | pgraph->base.fini = nv50_graph_fini; | |
821 | pgraph->base.context_new = nv50_graph_context_new; | |
822 | pgraph->base.context_del = nv50_graph_context_del; | |
823 | pgraph->base.object_new = nv50_graph_object_new; | |
824 | if (dev_priv->chipset == 0x50 || dev_priv->chipset == 0xac) | |
825 | pgraph->base.tlb_flush = nv50_graph_tlb_flush; | |
826 | else | |
827 | pgraph->base.tlb_flush = nv84_graph_tlb_flush; | |
828 | ||
829 | nouveau_irq_register(dev, 12, nv50_graph_isr); | |
830 | ||
2703c21a BS |
831 | NVOBJ_ENGINE_ADD(dev, GR, &pgraph->base); |
832 | NVOBJ_CLASS(dev, 0x0030, GR); /* null */ | |
833 | NVOBJ_CLASS(dev, 0x5039, GR); /* m2mf */ | |
834 | NVOBJ_CLASS(dev, 0x502d, GR); /* 2d */ | |
835 | ||
836 | /* tesla */ | |
837 | if (dev_priv->chipset == 0x50) | |
838 | NVOBJ_CLASS(dev, 0x5097, GR); /* tesla (nv50) */ | |
839 | else | |
840 | if (dev_priv->chipset < 0xa0) | |
841 | NVOBJ_CLASS(dev, 0x8297, GR); /* tesla (nv8x/nv9x) */ | |
842 | else { | |
843 | switch (dev_priv->chipset) { | |
844 | case 0xa0: | |
845 | case 0xaa: | |
846 | case 0xac: | |
847 | NVOBJ_CLASS(dev, 0x8397, GR); | |
848 | break; | |
849 | case 0xa3: | |
850 | case 0xa5: | |
851 | case 0xa8: | |
852 | NVOBJ_CLASS(dev, 0x8597, GR); | |
853 | break; | |
854 | case 0xaf: | |
855 | NVOBJ_CLASS(dev, 0x8697, GR); | |
856 | break; | |
857 | } | |
858 | } | |
859 | ||
860 | /* compute */ | |
861 | NVOBJ_CLASS(dev, 0x50c0, GR); | |
862 | if (dev_priv->chipset > 0xa0 && | |
863 | dev_priv->chipset != 0xaa && | |
864 | dev_priv->chipset != 0xac) | |
865 | NVOBJ_CLASS(dev, 0x85c0, GR); | |
866 | ||
867 | return 0; | |
868 | } |