nvc0_graph.c revision e425e0b33990575fd1c41671725b36247a325ea9
1/*
2 * Copyright 2010 Red Hat Inc.
3 *
4 * Permission is hereby granted, free of charge, to any person obtaining a
5 * copy of this software and associated documentation files (the "Software"),
6 * to deal in the Software without restriction, including without limitation
7 * the rights to use, copy, modify, merge, publish, distribute, sublicense,
8 * and/or sell copies of the Software, and to permit persons to whom the
9 * Software is furnished to do so, subject to the following conditions:
10 *
11 * The above copyright notice and this permission notice shall be included in
12 * all copies or substantial portions of the Software.
13 *
14 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
15 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
16 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
17 * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
18 * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
19 * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
20 * OTHER DEALINGS IN THE SOFTWARE.
21 *
22 * Authors: Ben Skeggs
23 */
24
25#include <linux/firmware.h>
26
27#include "drmP.h"
28
29#include "nouveau_drv.h"
30#include "nouveau_mm.h"
31
32#include "nvc0_graph.h"
33#include "nvc0_grhub.fuc.h"
34#include "nvc0_grgpc.fuc.h"
35
36static void
37nvc0_graph_ctxctl_debug_unit(struct drm_device *dev, u32 base)
38{
39	NV_INFO(dev, "PGRAPH: %06x - done 0x%08x\n", base,
40		nv_rd32(dev, base + 0x400));
41	NV_INFO(dev, "PGRAPH: %06x - stat 0x%08x 0x%08x 0x%08x 0x%08x\n", base,
42		nv_rd32(dev, base + 0x800), nv_rd32(dev, base + 0x804),
43		nv_rd32(dev, base + 0x808), nv_rd32(dev, base + 0x80c));
44	NV_INFO(dev, "PGRAPH: %06x - stat 0x%08x 0x%08x 0x%08x 0x%08x\n", base,
45		nv_rd32(dev, base + 0x810), nv_rd32(dev, base + 0x814),
46		nv_rd32(dev, base + 0x818), nv_rd32(dev, base + 0x81c));
47}
48
49static void
50nvc0_graph_ctxctl_debug(struct drm_device *dev)
51{
52	u32 gpcnr = nv_rd32(dev, 0x409604) & 0xffff;
53	u32 gpc;
54
55	nvc0_graph_ctxctl_debug_unit(dev, 0x409000);
56	for (gpc = 0; gpc < gpcnr; gpc++)
57		nvc0_graph_ctxctl_debug_unit(dev, 0x502000 + (gpc * 0x8000));
58}
59
60static int
61nvc0_graph_load_context(struct nouveau_channel *chan)
62{
63	struct drm_device *dev = chan->dev;
64
65	nv_wr32(dev, 0x409840, 0x00000030);
66	nv_wr32(dev, 0x409500, 0x80000000 | chan->ramin->vinst >> 12);
67	nv_wr32(dev, 0x409504, 0x00000003);
68	if (!nv_wait(dev, 0x409800, 0x00000010, 0x00000010))
69		NV_ERROR(dev, "PGRAPH: load_ctx timeout\n");
70
71	return 0;
72}
73
74static int
75nvc0_graph_unload_context_to(struct drm_device *dev, u64 chan)
76{
77	nv_wr32(dev, 0x409840, 0x00000003);
78	nv_wr32(dev, 0x409500, 0x80000000 | chan >> 12);
79	nv_wr32(dev, 0x409504, 0x00000009);
80	if (!nv_wait(dev, 0x409800, 0x00000001, 0x00000000)) {
81		NV_ERROR(dev, "PGRAPH: unload_ctx timeout\n");
82		return -EBUSY;
83	}
84
85	return 0;
86}
87
88static int
89nvc0_graph_construct_context(struct nouveau_channel *chan)
90{
91	struct drm_nouveau_private *dev_priv = chan->dev->dev_private;
92	struct nvc0_graph_priv *priv = nv_engine(chan->dev, NVOBJ_ENGINE_GR);
93	struct nvc0_graph_chan *grch = chan->engctx[NVOBJ_ENGINE_GR];
94	struct drm_device *dev = chan->dev;
95	int ret, i;
96	u32 *ctx;
97
98	ctx = kmalloc(priv->grctx_size, GFP_KERNEL);
99	if (!ctx)
100		return -ENOMEM;
101
102	if (!nouveau_ctxfw) {
103		nv_wr32(dev, 0x409840, 0x80000000);
104		nv_wr32(dev, 0x409500, 0x80000000 | chan->ramin->vinst >> 12);
105		nv_wr32(dev, 0x409504, 0x00000001);
106		if (!nv_wait(dev, 0x409800, 0x80000000, 0x80000000)) {
107			NV_ERROR(dev, "PGRAPH: HUB_SET_CHAN timeout\n");
108			nvc0_graph_ctxctl_debug(dev);
109			ret = -EBUSY;
110			goto err;
111		}
112	} else {
113		nvc0_graph_load_context(chan);
114
115		nv_wo32(grch->grctx, 0x1c, 1);
116		nv_wo32(grch->grctx, 0x20, 0);
117		nv_wo32(grch->grctx, 0x28, 0);
118		nv_wo32(grch->grctx, 0x2c, 0);
119		dev_priv->engine.instmem.flush(dev);
120	}
121
122	ret = nvc0_grctx_generate(chan);
123	if (ret)
124		goto err;
125
126	if (!nouveau_ctxfw) {
127		nv_wr32(dev, 0x409840, 0x80000000);
128		nv_wr32(dev, 0x409500, 0x80000000 | chan->ramin->vinst >> 12);
129		nv_wr32(dev, 0x409504, 0x00000002);
130		if (!nv_wait(dev, 0x409800, 0x80000000, 0x80000000)) {
131			NV_ERROR(dev, "PGRAPH: HUB_CTX_SAVE timeout\n");
132			nvc0_graph_ctxctl_debug(dev);
133			ret = -EBUSY;
134			goto err;
135		}
136	} else {
137		ret = nvc0_graph_unload_context_to(dev, chan->ramin->vinst);
138		if (ret)
139			goto err;
140	}
141
142	for (i = 0; i < priv->grctx_size; i += 4)
143		ctx[i / 4] = nv_ro32(grch->grctx, i);
144
145	priv->grctx_vals = ctx;
146	return 0;
147
148err:
149	kfree(ctx);
150	return ret;
151}
152
153static int
154nvc0_graph_create_context_mmio_list(struct nouveau_channel *chan)
155{
156	struct nvc0_graph_priv *priv = nv_engine(chan->dev, NVOBJ_ENGINE_GR);
157	struct nvc0_graph_chan *grch = chan->engctx[NVOBJ_ENGINE_GR];
158	struct drm_device *dev = chan->dev;
159	int i = 0, gpc, tp, ret;
160	u32 magic;
161
162	ret = nouveau_gpuobj_new(dev, chan, 0x2000, 256, NVOBJ_FLAG_VM,
163				 &grch->unk408004);
164	if (ret)
165		return ret;
166
167	ret = nouveau_gpuobj_new(dev, chan, 0x8000, 256, NVOBJ_FLAG_VM,
168				 &grch->unk40800c);
169	if (ret)
170		return ret;
171
172	ret = nouveau_gpuobj_new(dev, chan, 384 * 1024, 4096,
173				 NVOBJ_FLAG_VM | NVOBJ_FLAG_VM_USER,
174				 &grch->unk418810);
175	if (ret)
176		return ret;
177
178	ret = nouveau_gpuobj_new(dev, chan, 0x1000, 0, NVOBJ_FLAG_VM,
179				 &grch->mmio);
180	if (ret)
181		return ret;
182
183
184	nv_wo32(grch->mmio, i++ * 4, 0x00408004);
185	nv_wo32(grch->mmio, i++ * 4, grch->unk408004->linst >> 8);
186	nv_wo32(grch->mmio, i++ * 4, 0x00408008);
187	nv_wo32(grch->mmio, i++ * 4, 0x80000018);
188
189	nv_wo32(grch->mmio, i++ * 4, 0x0040800c);
190	nv_wo32(grch->mmio, i++ * 4, grch->unk40800c->linst >> 8);
191	nv_wo32(grch->mmio, i++ * 4, 0x00408010);
192	nv_wo32(grch->mmio, i++ * 4, 0x80000000);
193
194	nv_wo32(grch->mmio, i++ * 4, 0x00418810);
195	nv_wo32(grch->mmio, i++ * 4, 0x80000000 | grch->unk418810->linst >> 12);
196	nv_wo32(grch->mmio, i++ * 4, 0x00419848);
197	nv_wo32(grch->mmio, i++ * 4, 0x10000000 | grch->unk418810->linst >> 12);
198
199	nv_wo32(grch->mmio, i++ * 4, 0x00419004);
200	nv_wo32(grch->mmio, i++ * 4, grch->unk40800c->linst >> 8);
201	nv_wo32(grch->mmio, i++ * 4, 0x00419008);
202	nv_wo32(grch->mmio, i++ * 4, 0x00000000);
203
204	nv_wo32(grch->mmio, i++ * 4, 0x00418808);
205	nv_wo32(grch->mmio, i++ * 4, grch->unk408004->linst >> 8);
206	nv_wo32(grch->mmio, i++ * 4, 0x0041880c);
207	nv_wo32(grch->mmio, i++ * 4, 0x80000018);
208
209	magic = 0x02180000;
210	nv_wo32(grch->mmio, i++ * 4, 0x00405830);
211	nv_wo32(grch->mmio, i++ * 4, magic);
212	for (gpc = 0; gpc < priv->gpc_nr; gpc++) {
213		for (tp = 0; tp < priv->tp_nr[gpc]; tp++, magic += 0x0324) {
214			u32 reg = 0x504520 + (gpc * 0x8000) + (tp * 0x0800);
215			nv_wo32(grch->mmio, i++ * 4, reg);
216			nv_wo32(grch->mmio, i++ * 4, magic);
217		}
218	}
219
220	grch->mmio_nr = i / 2;
221	return 0;
222}
223
224static int
225nvc0_graph_context_new(struct nouveau_channel *chan, int engine)
226{
227	struct drm_device *dev = chan->dev;
228	struct drm_nouveau_private *dev_priv = dev->dev_private;
229	struct nouveau_instmem_engine *pinstmem = &dev_priv->engine.instmem;
230	struct nvc0_graph_priv *priv = nv_engine(dev, engine);
231	struct nvc0_graph_chan *grch;
232	struct nouveau_gpuobj *grctx;
233	int ret, i;
234
235	grch = kzalloc(sizeof(*grch), GFP_KERNEL);
236	if (!grch)
237		return -ENOMEM;
238	chan->engctx[NVOBJ_ENGINE_GR] = grch;
239
240	ret = nouveau_gpuobj_new(dev, chan, priv->grctx_size, 256,
241				 NVOBJ_FLAG_VM | NVOBJ_FLAG_ZERO_ALLOC,
242				 &grch->grctx);
243	if (ret)
244		goto error;
245	grctx = grch->grctx;
246
247	ret = nvc0_graph_create_context_mmio_list(chan);
248	if (ret)
249		goto error;
250
251	nv_wo32(chan->ramin, 0x0210, lower_32_bits(grctx->linst) | 4);
252	nv_wo32(chan->ramin, 0x0214, upper_32_bits(grctx->linst));
253	pinstmem->flush(dev);
254
255	if (!priv->grctx_vals) {
256		ret = nvc0_graph_construct_context(chan);
257		if (ret)
258			goto error;
259	}
260
261	for (i = 0; i < priv->grctx_size; i += 4)
262		nv_wo32(grctx, i, priv->grctx_vals[i / 4]);
263
264	if (!nouveau_ctxfw) {
265		nv_wo32(grctx, 0x00, grch->mmio_nr);
266		nv_wo32(grctx, 0x04, grch->mmio->linst >> 8);
267	} else {
268		nv_wo32(grctx, 0xf4, 0);
269		nv_wo32(grctx, 0xf8, 0);
270		nv_wo32(grctx, 0x10, grch->mmio_nr);
271		nv_wo32(grctx, 0x14, lower_32_bits(grch->mmio->linst));
272		nv_wo32(grctx, 0x18, upper_32_bits(grch->mmio->linst));
273		nv_wo32(grctx, 0x1c, 1);
274		nv_wo32(grctx, 0x20, 0);
275		nv_wo32(grctx, 0x28, 0);
276		nv_wo32(grctx, 0x2c, 0);
277	}
278	pinstmem->flush(dev);
279	return 0;
280
281error:
282	priv->base.context_del(chan, engine);
283	return ret;
284}
285
286static void
287nvc0_graph_context_del(struct nouveau_channel *chan, int engine)
288{
289	struct nvc0_graph_chan *grch = chan->engctx[engine];
290
291	nouveau_gpuobj_ref(NULL, &grch->mmio);
292	nouveau_gpuobj_ref(NULL, &grch->unk418810);
293	nouveau_gpuobj_ref(NULL, &grch->unk40800c);
294	nouveau_gpuobj_ref(NULL, &grch->unk408004);
295	nouveau_gpuobj_ref(NULL, &grch->grctx);
296	chan->engctx[engine] = NULL;
297}
298
299static int
300nvc0_graph_object_new(struct nouveau_channel *chan, int engine,
301		      u32 handle, u16 class)
302{
303	return 0;
304}
305
306static int
307nvc0_graph_fini(struct drm_device *dev, int engine, bool suspend)
308{
309	return 0;
310}
311
312static int
313nvc0_graph_mthd_page_flip(struct nouveau_channel *chan,
314			  u32 class, u32 mthd, u32 data)
315{
316	nouveau_finish_page_flip(chan, NULL);
317	return 0;
318}
319
320static void
321nvc0_graph_init_obj418880(struct drm_device *dev)
322{
323	struct nvc0_graph_priv *priv = nv_engine(dev, NVOBJ_ENGINE_GR);
324	int i;
325
326	nv_wr32(dev, GPC_BCAST(0x0880), 0x00000000);
327	nv_wr32(dev, GPC_BCAST(0x08a4), 0x00000000);
328	for (i = 0; i < 4; i++)
329		nv_wr32(dev, GPC_BCAST(0x0888) + (i * 4), 0x00000000);
330	nv_wr32(dev, GPC_BCAST(0x08b4), priv->unk4188b4->vinst >> 8);
331	nv_wr32(dev, GPC_BCAST(0x08b8), priv->unk4188b8->vinst >> 8);
332}
333
334static void
335nvc0_graph_init_regs(struct drm_device *dev)
336{
337	nv_wr32(dev, 0x400080, 0x003083c2);
338	nv_wr32(dev, 0x400088, 0x00006fe7);
339	nv_wr32(dev, 0x40008c, 0x00000000);
340	nv_wr32(dev, 0x400090, 0x00000030);
341	nv_wr32(dev, 0x40013c, 0x013901f7);
342	nv_wr32(dev, 0x400140, 0x00000100);
343	nv_wr32(dev, 0x400144, 0x00000000);
344	nv_wr32(dev, 0x400148, 0x00000110);
345	nv_wr32(dev, 0x400138, 0x00000000);
346	nv_wr32(dev, 0x400130, 0x00000000);
347	nv_wr32(dev, 0x400134, 0x00000000);
348	nv_wr32(dev, 0x400124, 0x00000002);
349}
350
351static void
352nvc0_graph_init_gpc_0(struct drm_device *dev)
353{
354	struct nvc0_graph_priv *priv = nv_engine(dev, NVOBJ_ENGINE_GR);
355	const u32 magicgpc918 = DIV_ROUND_UP(0x00800000, priv->tp_total);
356	u32 data[TP_MAX / 8];
357	u8  tpnr[GPC_MAX];
358	int i, gpc, tpc;
359
360	/*
361	 *      TP      ROP UNKVAL(magic_not_rop_nr)
362	 * 450: 4/0/0/0 2        3
363	 * 460: 3/4/0/0 4        1
364	 * 465: 3/4/4/0 4        7
365	 * 470: 3/3/4/4 5        5
366	 * 480: 3/4/4/4 6        6
367	 */
368
369	memset(data, 0x00, sizeof(data));
370	memcpy(tpnr, priv->tp_nr, sizeof(priv->tp_nr));
371	for (i = 0, gpc = -1; i < priv->tp_total; i++) {
372		do {
373			gpc = (gpc + 1) % priv->gpc_nr;
374		} while (!tpnr[gpc]);
375		tpc = priv->tp_nr[gpc] - tpnr[gpc]--;
376
377		data[i / 8] |= tpc << ((i % 8) * 4);
378	}
379
380	nv_wr32(dev, GPC_BCAST(0x0980), data[0]);
381	nv_wr32(dev, GPC_BCAST(0x0984), data[1]);
382	nv_wr32(dev, GPC_BCAST(0x0988), data[2]);
383	nv_wr32(dev, GPC_BCAST(0x098c), data[3]);
384
385	for (gpc = 0; gpc < priv->gpc_nr; gpc++) {
386		nv_wr32(dev, GPC_UNIT(gpc, 0x0914), priv->magic_not_rop_nr << 8 |
387						  priv->tp_nr[gpc]);
388		nv_wr32(dev, GPC_UNIT(gpc, 0x0910), 0x00040000 | priv->tp_total);
389		nv_wr32(dev, GPC_UNIT(gpc, 0x0918), magicgpc918);
390	}
391
392	nv_wr32(dev, GPC_BCAST(0x1bd4), magicgpc918);
393	nv_wr32(dev, GPC_BCAST(0x08ac), nv_rd32(dev, 0x100800));
394}
395
396static void
397nvc0_graph_init_units(struct drm_device *dev)
398{
399	nv_wr32(dev, 0x409c24, 0x000f0000);
400	nv_wr32(dev, 0x404000, 0xc0000000); /* DISPATCH */
401	nv_wr32(dev, 0x404600, 0xc0000000); /* M2MF */
402	nv_wr32(dev, 0x408030, 0xc0000000);
403	nv_wr32(dev, 0x40601c, 0xc0000000);
404	nv_wr32(dev, 0x404490, 0xc0000000); /* MACRO */
405	nv_wr32(dev, 0x406018, 0xc0000000);
406	nv_wr32(dev, 0x405840, 0xc0000000);
407	nv_wr32(dev, 0x405844, 0x00ffffff);
408	nv_mask(dev, 0x419cc0, 0x00000008, 0x00000008);
409	nv_mask(dev, 0x419eb4, 0x00001000, 0x00001000);
410}
411
412static void
413nvc0_graph_init_gpc_1(struct drm_device *dev)
414{
415	struct nvc0_graph_priv *priv = nv_engine(dev, NVOBJ_ENGINE_GR);
416	int gpc, tp;
417
418	for (gpc = 0; gpc < priv->gpc_nr; gpc++) {
419		nv_wr32(dev, GPC_UNIT(gpc, 0x0420), 0xc0000000);
420		nv_wr32(dev, GPC_UNIT(gpc, 0x0900), 0xc0000000);
421		nv_wr32(dev, GPC_UNIT(gpc, 0x1028), 0xc0000000);
422		nv_wr32(dev, GPC_UNIT(gpc, 0x0824), 0xc0000000);
423		for (tp = 0; tp < priv->tp_nr[gpc]; tp++) {
424			nv_wr32(dev, TP_UNIT(gpc, tp, 0x508), 0xffffffff);
425			nv_wr32(dev, TP_UNIT(gpc, tp, 0x50c), 0xffffffff);
426			nv_wr32(dev, TP_UNIT(gpc, tp, 0x224), 0xc0000000);
427			nv_wr32(dev, TP_UNIT(gpc, tp, 0x48c), 0xc0000000);
428			nv_wr32(dev, TP_UNIT(gpc, tp, 0x084), 0xc0000000);
429			nv_wr32(dev, TP_UNIT(gpc, tp, 0x644), 0x001ffffe);
430			nv_wr32(dev, TP_UNIT(gpc, tp, 0x64c), 0x0000000f);
431		}
432		nv_wr32(dev, GPC_UNIT(gpc, 0x2c90), 0xffffffff);
433		nv_wr32(dev, GPC_UNIT(gpc, 0x2c94), 0xffffffff);
434	}
435}
436
437static void
438nvc0_graph_init_rop(struct drm_device *dev)
439{
440	struct nvc0_graph_priv *priv = nv_engine(dev, NVOBJ_ENGINE_GR);
441	int rop;
442
443	for (rop = 0; rop < priv->rop_nr; rop++) {
444		nv_wr32(dev, ROP_UNIT(rop, 0x144), 0xc0000000);
445		nv_wr32(dev, ROP_UNIT(rop, 0x070), 0xc0000000);
446		nv_wr32(dev, ROP_UNIT(rop, 0x204), 0xffffffff);
447		nv_wr32(dev, ROP_UNIT(rop, 0x208), 0xffffffff);
448	}
449}
450
451static void
452nvc0_graph_init_fuc(struct drm_device *dev, u32 fuc_base,
453		    struct nvc0_graph_fuc *code, struct nvc0_graph_fuc *data)
454{
455	int i;
456
457	nv_wr32(dev, fuc_base + 0x01c0, 0x01000000);
458	for (i = 0; i < data->size / 4; i++)
459		nv_wr32(dev, fuc_base + 0x01c4, data->data[i]);
460
461	nv_wr32(dev, fuc_base + 0x0180, 0x01000000);
462	for (i = 0; i < code->size / 4; i++) {
463		if ((i & 0x3f) == 0)
464			nv_wr32(dev, fuc_base + 0x0188, i >> 6);
465		nv_wr32(dev, fuc_base + 0x0184, code->data[i]);
466	}
467}
468
469static int
470nvc0_graph_init_ctxctl(struct drm_device *dev)
471{
472	struct drm_nouveau_private *dev_priv = dev->dev_private;
473	struct nvc0_graph_priv *priv = nv_engine(dev, NVOBJ_ENGINE_GR);
474	u32 r000260;
475	int i;
476
477	if (!nouveau_ctxfw) {
478		/* load HUB microcode */
479		r000260 = nv_mask(dev, 0x000260, 0x00000001, 0x00000000);
480		nv_wr32(dev, 0x4091c0, 0x01000000);
481		for (i = 0; i < sizeof(nvc0_grhub_data) / 4; i++)
482			nv_wr32(dev, 0x4091c4, nvc0_grhub_data[i]);
483
484		nv_wr32(dev, 0x409180, 0x01000000);
485		for (i = 0; i < sizeof(nvc0_grhub_code) / 4; i++) {
486			if ((i & 0x3f) == 0)
487				nv_wr32(dev, 0x409188, i >> 6);
488			nv_wr32(dev, 0x409184, nvc0_grhub_code[i]);
489		}
490
491		/* load GPC microcode */
492		nv_wr32(dev, 0x41a1c0, 0x01000000);
493		for (i = 0; i < sizeof(nvc0_grgpc_data) / 4; i++)
494			nv_wr32(dev, 0x41a1c4, nvc0_grgpc_data[i]);
495
496		nv_wr32(dev, 0x41a180, 0x01000000);
497		for (i = 0; i < sizeof(nvc0_grgpc_code) / 4; i++) {
498			if ((i & 0x3f) == 0)
499				nv_wr32(dev, 0x41a188, i >> 6);
500			nv_wr32(dev, 0x41a184, nvc0_grgpc_code[i]);
501		}
502		nv_wr32(dev, 0x000260, r000260);
503
504		/* start HUB ucode running, it'll init the GPCs */
505		nv_wr32(dev, 0x409800, dev_priv->chipset);
506		nv_wr32(dev, 0x40910c, 0x00000000);
507		nv_wr32(dev, 0x409100, 0x00000002);
508		if (!nv_wait(dev, 0x409800, 0x80000000, 0x80000000)) {
509			NV_ERROR(dev, "PGRAPH: HUB_INIT timed out\n");
510			nvc0_graph_ctxctl_debug(dev);
511			return -EBUSY;
512		}
513
514		priv->grctx_size = nv_rd32(dev, 0x409804);
515		return 0;
516	}
517
518	/* load fuc microcode */
519	r000260 = nv_mask(dev, 0x000260, 0x00000001, 0x00000000);
520	nvc0_graph_init_fuc(dev, 0x409000, &priv->fuc409c, &priv->fuc409d);
521	nvc0_graph_init_fuc(dev, 0x41a000, &priv->fuc41ac, &priv->fuc41ad);
522	nv_wr32(dev, 0x000260, r000260);
523
524	/* start both of them running */
525	nv_wr32(dev, 0x409840, 0xffffffff);
526	nv_wr32(dev, 0x41a10c, 0x00000000);
527	nv_wr32(dev, 0x40910c, 0x00000000);
528	nv_wr32(dev, 0x41a100, 0x00000002);
529	nv_wr32(dev, 0x409100, 0x00000002);
530	if (!nv_wait(dev, 0x409800, 0x00000001, 0x00000001))
531		NV_INFO(dev, "0x409800 wait failed\n");
532
533	nv_wr32(dev, 0x409840, 0xffffffff);
534	nv_wr32(dev, 0x409500, 0x7fffffff);
535	nv_wr32(dev, 0x409504, 0x00000021);
536
537	nv_wr32(dev, 0x409840, 0xffffffff);
538	nv_wr32(dev, 0x409500, 0x00000000);
539	nv_wr32(dev, 0x409504, 0x00000010);
540	if (!nv_wait_ne(dev, 0x409800, 0xffffffff, 0x00000000)) {
541		NV_ERROR(dev, "fuc09 req 0x10 timeout\n");
542		return -EBUSY;
543	}
544	priv->grctx_size = nv_rd32(dev, 0x409800);
545
546	nv_wr32(dev, 0x409840, 0xffffffff);
547	nv_wr32(dev, 0x409500, 0x00000000);
548	nv_wr32(dev, 0x409504, 0x00000016);
549	if (!nv_wait_ne(dev, 0x409800, 0xffffffff, 0x00000000)) {
550		NV_ERROR(dev, "fuc09 req 0x16 timeout\n");
551		return -EBUSY;
552	}
553
554	nv_wr32(dev, 0x409840, 0xffffffff);
555	nv_wr32(dev, 0x409500, 0x00000000);
556	nv_wr32(dev, 0x409504, 0x00000025);
557	if (!nv_wait_ne(dev, 0x409800, 0xffffffff, 0x00000000)) {
558		NV_ERROR(dev, "fuc09 req 0x25 timeout\n");
559		return -EBUSY;
560	}
561
562	return 0;
563}
564
565static int
566nvc0_graph_init(struct drm_device *dev, int engine)
567{
568	int ret;
569
570	nv_mask(dev, 0x000200, 0x18001000, 0x00000000);
571	nv_mask(dev, 0x000200, 0x18001000, 0x18001000);
572
573	nvc0_graph_init_obj418880(dev);
574	nvc0_graph_init_regs(dev);
575	/*nvc0_graph_init_unitplemented_magics(dev);*/
576	nvc0_graph_init_gpc_0(dev);
577	/*nvc0_graph_init_unitplemented_c242(dev);*/
578
579	nv_wr32(dev, 0x400500, 0x00010001);
580	nv_wr32(dev, 0x400100, 0xffffffff);
581	nv_wr32(dev, 0x40013c, 0xffffffff);
582
583	nvc0_graph_init_units(dev);
584	nvc0_graph_init_gpc_1(dev);
585	nvc0_graph_init_rop(dev);
586
587	nv_wr32(dev, 0x400108, 0xffffffff);
588	nv_wr32(dev, 0x400138, 0xffffffff);
589	nv_wr32(dev, 0x400118, 0xffffffff);
590	nv_wr32(dev, 0x400130, 0xffffffff);
591	nv_wr32(dev, 0x40011c, 0xffffffff);
592	nv_wr32(dev, 0x400134, 0xffffffff);
593	nv_wr32(dev, 0x400054, 0x34ce3464);
594
595	ret = nvc0_graph_init_ctxctl(dev);
596	if (ret)
597		return ret;
598
599	return 0;
600}
601
602int
603nvc0_graph_isr_chid(struct drm_device *dev, u64 inst)
604{
605	struct drm_nouveau_private *dev_priv = dev->dev_private;
606	struct nouveau_channel *chan;
607	unsigned long flags;
608	int i;
609
610	spin_lock_irqsave(&dev_priv->channels.lock, flags);
611	for (i = 0; i < dev_priv->engine.fifo.channels; i++) {
612		chan = dev_priv->channels.ptr[i];
613		if (!chan || !chan->ramin)
614			continue;
615
616		if (inst == chan->ramin->vinst)
617			break;
618	}
619	spin_unlock_irqrestore(&dev_priv->channels.lock, flags);
620	return i;
621}
622
623static void
624nvc0_graph_ctxctl_isr(struct drm_device *dev)
625{
626	u32 ustat = nv_rd32(dev, 0x409c18);
627
628	if (ustat & 0x00000001)
629		NV_INFO(dev, "PGRAPH: CTXCTRL ucode error\n");
630	if (ustat & 0x00080000)
631		NV_INFO(dev, "PGRAPH: CTXCTRL watchdog timeout\n");
632	if (ustat & ~0x00080001)
633		NV_INFO(dev, "PGRAPH: CTXCTRL 0x%08x\n", ustat);
634
635	nvc0_graph_ctxctl_debug(dev);
636	nv_wr32(dev, 0x409c20, ustat);
637}
638
639static void
640nvc0_graph_isr(struct drm_device *dev)
641{
642	u64 inst = (u64)(nv_rd32(dev, 0x409b00) & 0x0fffffff) << 12;
643	u32 chid = nvc0_graph_isr_chid(dev, inst);
644	u32 stat = nv_rd32(dev, 0x400100);
645	u32 addr = nv_rd32(dev, 0x400704);
646	u32 mthd = (addr & 0x00003ffc);
647	u32 subc = (addr & 0x00070000) >> 16;
648	u32 data = nv_rd32(dev, 0x400708);
649	u32 code = nv_rd32(dev, 0x400110);
650	u32 class = nv_rd32(dev, 0x404200 + (subc * 4));
651
652	if (stat & 0x00000010) {
653		if (nouveau_gpuobj_mthd_call2(dev, chid, class, mthd, data)) {
654			NV_INFO(dev, "PGRAPH: ILLEGAL_MTHD ch %d [0x%010llx] "
655				     "subc %d class 0x%04x mthd 0x%04x "
656				     "data 0x%08x\n",
657				chid, inst, subc, class, mthd, data);
658		}
659		nv_wr32(dev, 0x400100, 0x00000010);
660		stat &= ~0x00000010;
661	}
662
663	if (stat & 0x00000020) {
664		NV_INFO(dev, "PGRAPH: ILLEGAL_CLASS ch %d [0x%010llx] subc %d "
665			     "class 0x%04x mthd 0x%04x data 0x%08x\n",
666			chid, inst, subc, class, mthd, data);
667		nv_wr32(dev, 0x400100, 0x00000020);
668		stat &= ~0x00000020;
669	}
670
671	if (stat & 0x00100000) {
672		NV_INFO(dev, "PGRAPH: DATA_ERROR [");
673		nouveau_enum_print(nv50_data_error_names, code);
674		printk("] ch %d [0x%010llx] subc %d class 0x%04x "
675		       "mthd 0x%04x data 0x%08x\n",
676		       chid, inst, subc, class, mthd, data);
677		nv_wr32(dev, 0x400100, 0x00100000);
678		stat &= ~0x00100000;
679	}
680
681	if (stat & 0x00200000) {
682		u32 trap = nv_rd32(dev, 0x400108);
683		NV_INFO(dev, "PGRAPH: TRAP ch %d status 0x%08x\n", chid, trap);
684		nv_wr32(dev, 0x400108, trap);
685		nv_wr32(dev, 0x400100, 0x00200000);
686		stat &= ~0x00200000;
687	}
688
689	if (stat & 0x00080000) {
690		nvc0_graph_ctxctl_isr(dev);
691		nv_wr32(dev, 0x400100, 0x00080000);
692		stat &= ~0x00080000;
693	}
694
695	if (stat) {
696		NV_INFO(dev, "PGRAPH: unknown stat 0x%08x\n", stat);
697		nv_wr32(dev, 0x400100, stat);
698	}
699
700	nv_wr32(dev, 0x400500, 0x00010001);
701}
702
703static void
704nvc0_runk140_isr(struct drm_device *dev)
705{
706	u32 units = nv_rd32(dev, 0x00017c) & 0x1f;
707
708	while (units) {
709		u32 unit = ffs(units) - 1;
710		u32 reg = 0x140000 + unit * 0x2000;
711		u32 st0 = nv_mask(dev, reg + 0x1020, 0, 0);
712		u32 st1 = nv_mask(dev, reg + 0x1420, 0, 0);
713
714		NV_DEBUG(dev, "PRUNK140: %d 0x%08x 0x%08x\n", unit, st0, st1);
715		units &= ~(1 << unit);
716	}
717}
718
719static int
720nvc0_graph_create_fw(struct drm_device *dev, const char *fwname,
721		     struct nvc0_graph_fuc *fuc)
722{
723	struct drm_nouveau_private *dev_priv = dev->dev_private;
724	const struct firmware *fw;
725	char f[32];
726	int ret;
727
728	snprintf(f, sizeof(f), "nouveau/nv%02x_%s", dev_priv->chipset, fwname);
729	ret = request_firmware(&fw, f, &dev->pdev->dev);
730	if (ret) {
731		snprintf(f, sizeof(f), "nouveau/%s", fwname);
732		ret = request_firmware(&fw, f, &dev->pdev->dev);
733		if (ret) {
734			NV_ERROR(dev, "failed to load %s\n", fwname);
735			return ret;
736		}
737	}
738
739	fuc->size = fw->size;
740	fuc->data = kmemdup(fw->data, fuc->size, GFP_KERNEL);
741	release_firmware(fw);
742	return (fuc->data != NULL) ? 0 : -ENOMEM;
743}
744
745static void
746nvc0_graph_destroy_fw(struct nvc0_graph_fuc *fuc)
747{
748	if (fuc->data) {
749		kfree(fuc->data);
750		fuc->data = NULL;
751	}
752}
753
754static void
755nvc0_graph_destroy(struct drm_device *dev, int engine)
756{
757	struct nvc0_graph_priv *priv = nv_engine(dev, engine);
758
759	if (nouveau_ctxfw) {
760		nvc0_graph_destroy_fw(&priv->fuc409c);
761		nvc0_graph_destroy_fw(&priv->fuc409d);
762		nvc0_graph_destroy_fw(&priv->fuc41ac);
763		nvc0_graph_destroy_fw(&priv->fuc41ad);
764	}
765
766	nouveau_irq_unregister(dev, 12);
767	nouveau_irq_unregister(dev, 25);
768
769	nouveau_gpuobj_ref(NULL, &priv->unk4188b8);
770	nouveau_gpuobj_ref(NULL, &priv->unk4188b4);
771
772	if (priv->grctx_vals)
773		kfree(priv->grctx_vals);
774
775	NVOBJ_ENGINE_DEL(dev, GR);
776	kfree(priv);
777}
778
779int
780nvc0_graph_create(struct drm_device *dev)
781{
782	struct drm_nouveau_private *dev_priv = dev->dev_private;
783	struct nvc0_graph_priv *priv;
784	int ret, gpc, i;
785	u32 fermi;
786
787	fermi = nvc0_graph_class(dev);
788	if (!fermi) {
789		NV_ERROR(dev, "PGRAPH: unsupported chipset, please report!\n");
790		return 0;
791	}
792
793	priv = kzalloc(sizeof(*priv), GFP_KERNEL);
794	if (!priv)
795		return -ENOMEM;
796
797	priv->base.destroy = nvc0_graph_destroy;
798	priv->base.init = nvc0_graph_init;
799	priv->base.fini = nvc0_graph_fini;
800	priv->base.context_new = nvc0_graph_context_new;
801	priv->base.context_del = nvc0_graph_context_del;
802	priv->base.object_new = nvc0_graph_object_new;
803
804	NVOBJ_ENGINE_ADD(dev, GR, &priv->base);
805	nouveau_irq_register(dev, 12, nvc0_graph_isr);
806	nouveau_irq_register(dev, 25, nvc0_runk140_isr);
807
808	if (nouveau_ctxfw) {
809		NV_INFO(dev, "PGRAPH: using external firmware\n");
810		if (nvc0_graph_create_fw(dev, "fuc409c", &priv->fuc409c) ||
811		    nvc0_graph_create_fw(dev, "fuc409d", &priv->fuc409d) ||
812		    nvc0_graph_create_fw(dev, "fuc41ac", &priv->fuc41ac) ||
813		    nvc0_graph_create_fw(dev, "fuc41ad", &priv->fuc41ad)) {
814			ret = 0;
815			goto error;
816		}
817	}
818
819	ret = nouveau_gpuobj_new(dev, NULL, 0x1000, 256, 0, &priv->unk4188b4);
820	if (ret)
821		goto error;
822
823	ret = nouveau_gpuobj_new(dev, NULL, 0x1000, 256, 0, &priv->unk4188b8);
824	if (ret)
825		goto error;
826
827	for (i = 0; i < 0x1000; i += 4) {
828		nv_wo32(priv->unk4188b4, i, 0x00000010);
829		nv_wo32(priv->unk4188b8, i, 0x00000010);
830	}
831
832	priv->gpc_nr  =  nv_rd32(dev, 0x409604) & 0x0000001f;
833	priv->rop_nr = (nv_rd32(dev, 0x409604) & 0x001f0000) >> 16;
834	for (gpc = 0; gpc < priv->gpc_nr; gpc++) {
835		priv->tp_nr[gpc] = nv_rd32(dev, GPC_UNIT(gpc, 0x2608));
836		priv->tp_total += priv->tp_nr[gpc];
837	}
838
839	/*XXX: these need figuring out... */
840	switch (dev_priv->chipset) {
841	case 0xc0:
842		if (priv->tp_total == 11) { /* 465, 3/4/4/0, 4 */
843			priv->magic_not_rop_nr = 0x07;
844		} else
845		if (priv->tp_total == 14) { /* 470, 3/3/4/4, 5 */
846			priv->magic_not_rop_nr = 0x05;
847		} else
848		if (priv->tp_total == 15) { /* 480, 3/4/4/4, 6 */
849			priv->magic_not_rop_nr = 0x06;
850		}
851		break;
852	case 0xc3: /* 450, 4/0/0/0, 2 */
853		priv->magic_not_rop_nr = 0x03;
854		break;
855	case 0xc4: /* 460, 3/4/0/0, 4 */
856		priv->magic_not_rop_nr = 0x01;
857		break;
858	case 0xc1: /* 2/0/0/0, 1 */
859		priv->magic_not_rop_nr = 0x01;
860		break;
861	case 0xc8: /* 4/4/3/4, 5 */
862		priv->magic_not_rop_nr = 0x06;
863		break;
864	case 0xce: /* 4/4/0/0, 4 */
865		priv->magic_not_rop_nr = 0x03;
866		break;
867	case 0xcf: /* 4/0/0/0, 3 */
868		priv->magic_not_rop_nr = 0x03;
869		break;
870	}
871
872	if (!priv->magic_not_rop_nr) {
873		NV_ERROR(dev, "PGRAPH: unknown config: %d/%d/%d/%d, %d\n",
874			 priv->tp_nr[0], priv->tp_nr[1], priv->tp_nr[2],
875			 priv->tp_nr[3], priv->rop_nr);
876		/* use 0xc3's values... */
877		priv->magic_not_rop_nr = 0x03;
878	}
879
880	NVOBJ_CLASS(dev, 0x902d, GR); /* 2D */
881	NVOBJ_CLASS(dev, 0x9039, GR); /* M2MF */
882	NVOBJ_MTHD (dev, 0x9039, 0x0500, nvc0_graph_mthd_page_flip);
883	NVOBJ_CLASS(dev, 0x9097, GR); /* 3D */
884	if (fermi >= 0x9197)
885		NVOBJ_CLASS(dev, 0x9197, GR); /* 3D (NVC1-) */
886	if (fermi >= 0x9297)
887		NVOBJ_CLASS(dev, 0x9297, GR); /* 3D (NVC8-) */
888	NVOBJ_CLASS(dev, 0x90c0, GR); /* COMPUTE */
889	return 0;
890
891error:
892	nvc0_graph_destroy(dev, NVOBJ_ENGINE_GR);
893	return ret;
894}
895
896MODULE_FIRMWARE("nouveau/nvc0_fuc409c");
897MODULE_FIRMWARE("nouveau/nvc0_fuc409d");
898MODULE_FIRMWARE("nouveau/nvc0_fuc41ac");
899MODULE_FIRMWARE("nouveau/nvc0_fuc41ad");
900MODULE_FIRMWARE("nouveau/nvc3_fuc409c");
901MODULE_FIRMWARE("nouveau/nvc3_fuc409d");
902MODULE_FIRMWARE("nouveau/nvc3_fuc41ac");
903MODULE_FIRMWARE("nouveau/nvc3_fuc41ad");
904MODULE_FIRMWARE("nouveau/nvc4_fuc409c");
905MODULE_FIRMWARE("nouveau/nvc4_fuc409d");
906MODULE_FIRMWARE("nouveau/nvc4_fuc41ac");
907MODULE_FIRMWARE("nouveau/nvc4_fuc41ad");
908MODULE_FIRMWARE("nouveau/fuc409c");
909MODULE_FIRMWARE("nouveau/fuc409d");
910MODULE_FIRMWARE("nouveau/fuc41ac");
911MODULE_FIRMWARE("nouveau/fuc41ad");
912