1 /*
2 * Copyright 2012 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 #include "nv04.h"
25 #include "channv04.h"
26 #include "regsnv04.h"
27
28 #include <core/client.h>
29 #include <core/ramht.h>
30 #include <subdev/instmem.h>
31 #include <subdev/timer.h>
32 #include <engine/sw.h>
33
34 static const struct nv04_fifo_ramfc
35 nv04_fifo_ramfc[] = {
36 { 32, 0, 0x00, 0, NV04_PFIFO_CACHE1_DMA_PUT },
37 { 32, 0, 0x04, 0, NV04_PFIFO_CACHE1_DMA_GET },
38 { 16, 0, 0x08, 0, NV04_PFIFO_CACHE1_DMA_INSTANCE },
39 { 16, 16, 0x08, 0, NV04_PFIFO_CACHE1_DMA_DCOUNT },
40 { 32, 0, 0x0c, 0, NV04_PFIFO_CACHE1_DMA_STATE },
41 { 32, 0, 0x10, 0, NV04_PFIFO_CACHE1_DMA_FETCH },
42 { 32, 0, 0x14, 0, NV04_PFIFO_CACHE1_ENGINE },
43 { 32, 0, 0x18, 0, NV04_PFIFO_CACHE1_PULL1 },
44 {}
45 };
46
47 void
nv04_fifo_pause(struct nvkm_fifo * base,unsigned long * pflags)48 nv04_fifo_pause(struct nvkm_fifo *base, unsigned long *pflags)
49 __acquires(fifo->base.lock)
50 {
51 struct nv04_fifo *fifo = nv04_fifo(base);
52 struct nvkm_device *device = fifo->base.engine.subdev.device;
53 unsigned long flags;
54
55 spin_lock_irqsave(&fifo->base.lock, flags);
56 *pflags = flags;
57
58 nvkm_wr32(device, NV03_PFIFO_CACHES, 0x00000000);
59 nvkm_mask(device, NV04_PFIFO_CACHE1_PULL0, 0x00000001, 0x00000000);
60
61 /* in some cases the puller may be left in an inconsistent state
62 * if you try to stop it while it's busy translating handles.
63 * sometimes you get a CACHE_ERROR, sometimes it just fails
64 * silently; sending incorrect instance offsets to PGRAPH after
65 * it's started up again.
66 *
67 * to avoid this, we invalidate the most recently calculated
68 * instance.
69 */
70 nvkm_msec(device, 2000,
71 u32 tmp = nvkm_rd32(device, NV04_PFIFO_CACHE1_PULL0);
72 if (!(tmp & NV04_PFIFO_CACHE1_PULL0_HASH_BUSY))
73 break;
74 );
75
76 if (nvkm_rd32(device, NV04_PFIFO_CACHE1_PULL0) &
77 NV04_PFIFO_CACHE1_PULL0_HASH_FAILED)
78 nvkm_wr32(device, NV03_PFIFO_INTR_0, NV_PFIFO_INTR_CACHE_ERROR);
79
80 nvkm_wr32(device, NV04_PFIFO_CACHE1_HASH, 0x00000000);
81 }
82
83 void
nv04_fifo_start(struct nvkm_fifo * base,unsigned long * pflags)84 nv04_fifo_start(struct nvkm_fifo *base, unsigned long *pflags)
85 __releases(fifo->base.lock)
86 {
87 struct nv04_fifo *fifo = nv04_fifo(base);
88 struct nvkm_device *device = fifo->base.engine.subdev.device;
89 unsigned long flags = *pflags;
90
91 nvkm_mask(device, NV04_PFIFO_CACHE1_PULL0, 0x00000001, 0x00000001);
92 nvkm_wr32(device, NV03_PFIFO_CACHES, 0x00000001);
93
94 spin_unlock_irqrestore(&fifo->base.lock, flags);
95 }
96
97 static const char *
nv_dma_state_err(u32 state)98 nv_dma_state_err(u32 state)
99 {
100 static const char * const desc[] = {
101 "NONE", "CALL_SUBR_ACTIVE", "INVALID_MTHD", "RET_SUBR_INACTIVE",
102 "INVALID_CMD", "IB_EMPTY"/* NV50+ */, "MEM_FAULT", "UNK"
103 };
104 return desc[(state >> 29) & 0x7];
105 }
106
107 static bool
nv04_fifo_swmthd(struct nvkm_device * device,u32 chid,u32 addr,u32 data)108 nv04_fifo_swmthd(struct nvkm_device *device, u32 chid, u32 addr, u32 data)
109 {
110 struct nvkm_sw *sw = device->sw;
111 const int subc = (addr & 0x0000e000) >> 13;
112 const int mthd = (addr & 0x00001ffc);
113 const u32 mask = 0x0000000f << (subc * 4);
114 u32 engine = nvkm_rd32(device, 0x003280);
115 bool handled = false;
116
117 switch (mthd) {
118 case 0x0000 ... 0x0000: /* subchannel's engine -> software */
119 nvkm_wr32(device, 0x003280, (engine &= ~mask));
120 fallthrough;
121 case 0x0180 ... 0x01fc: /* handle -> instance */
122 data = nvkm_rd32(device, 0x003258) & 0x0000ffff;
123 fallthrough;
124 case 0x0100 ... 0x017c:
125 case 0x0200 ... 0x1ffc: /* pass method down to sw */
126 if (!(engine & mask) && sw)
127 handled = nvkm_sw_mthd(sw, chid, subc, mthd, data);
128 break;
129 default:
130 break;
131 }
132
133 return handled;
134 }
135
136 static void
nv04_fifo_cache_error(struct nv04_fifo * fifo,u32 chid,u32 get)137 nv04_fifo_cache_error(struct nv04_fifo *fifo, u32 chid, u32 get)
138 {
139 struct nvkm_subdev *subdev = &fifo->base.engine.subdev;
140 struct nvkm_device *device = subdev->device;
141 struct nvkm_fifo_chan *chan;
142 unsigned long flags;
143 u32 pull0 = nvkm_rd32(device, 0x003250);
144 u32 mthd, data;
145 int ptr;
146
147 /* NV_PFIFO_CACHE1_GET actually goes to 0xffc before wrapping on my
148 * G80 chips, but CACHE1 isn't big enough for this much data.. Tests
149 * show that it wraps around to the start at GET=0x800.. No clue as to
150 * why..
151 */
152 ptr = (get & 0x7ff) >> 2;
153
154 if (device->card_type < NV_40) {
155 mthd = nvkm_rd32(device, NV04_PFIFO_CACHE1_METHOD(ptr));
156 data = nvkm_rd32(device, NV04_PFIFO_CACHE1_DATA(ptr));
157 } else {
158 mthd = nvkm_rd32(device, NV40_PFIFO_CACHE1_METHOD(ptr));
159 data = nvkm_rd32(device, NV40_PFIFO_CACHE1_DATA(ptr));
160 }
161
162 if (!(pull0 & 0x00000100) ||
163 !nv04_fifo_swmthd(device, chid, mthd, data)) {
164 chan = nvkm_fifo_chan_chid(&fifo->base, chid, &flags);
165 nvkm_error(subdev, "CACHE_ERROR - "
166 "ch %d [%s] subc %d mthd %04x data %08x\n",
167 chid, chan ? chan->object.client->name : "unknown",
168 (mthd >> 13) & 7, mthd & 0x1ffc, data);
169 nvkm_fifo_chan_put(&fifo->base, flags, &chan);
170 }
171
172 nvkm_wr32(device, NV04_PFIFO_CACHE1_DMA_PUSH, 0);
173 nvkm_wr32(device, NV03_PFIFO_INTR_0, NV_PFIFO_INTR_CACHE_ERROR);
174
175 nvkm_wr32(device, NV03_PFIFO_CACHE1_PUSH0,
176 nvkm_rd32(device, NV03_PFIFO_CACHE1_PUSH0) & ~1);
177 nvkm_wr32(device, NV03_PFIFO_CACHE1_GET, get + 4);
178 nvkm_wr32(device, NV03_PFIFO_CACHE1_PUSH0,
179 nvkm_rd32(device, NV03_PFIFO_CACHE1_PUSH0) | 1);
180 nvkm_wr32(device, NV04_PFIFO_CACHE1_HASH, 0);
181
182 nvkm_wr32(device, NV04_PFIFO_CACHE1_DMA_PUSH,
183 nvkm_rd32(device, NV04_PFIFO_CACHE1_DMA_PUSH) | 1);
184 nvkm_wr32(device, NV04_PFIFO_CACHE1_PULL0, 1);
185 }
186
187 static void
nv04_fifo_dma_pusher(struct nv04_fifo * fifo,u32 chid)188 nv04_fifo_dma_pusher(struct nv04_fifo *fifo, u32 chid)
189 {
190 struct nvkm_subdev *subdev = &fifo->base.engine.subdev;
191 struct nvkm_device *device = subdev->device;
192 u32 dma_get = nvkm_rd32(device, 0x003244);
193 u32 dma_put = nvkm_rd32(device, 0x003240);
194 u32 push = nvkm_rd32(device, 0x003220);
195 u32 state = nvkm_rd32(device, 0x003228);
196 struct nvkm_fifo_chan *chan;
197 unsigned long flags;
198 const char *name;
199
200 chan = nvkm_fifo_chan_chid(&fifo->base, chid, &flags);
201 name = chan ? chan->object.client->name : "unknown";
202 if (device->card_type == NV_50) {
203 u32 ho_get = nvkm_rd32(device, 0x003328);
204 u32 ho_put = nvkm_rd32(device, 0x003320);
205 u32 ib_get = nvkm_rd32(device, 0x003334);
206 u32 ib_put = nvkm_rd32(device, 0x003330);
207
208 nvkm_error(subdev, "DMA_PUSHER - "
209 "ch %d [%s] get %02x%08x put %02x%08x ib_get %08x "
210 "ib_put %08x state %08x (err: %s) push %08x\n",
211 chid, name, ho_get, dma_get, ho_put, dma_put,
212 ib_get, ib_put, state, nv_dma_state_err(state),
213 push);
214
215 /* METHOD_COUNT, in DMA_STATE on earlier chipsets */
216 nvkm_wr32(device, 0x003364, 0x00000000);
217 if (dma_get != dma_put || ho_get != ho_put) {
218 nvkm_wr32(device, 0x003244, dma_put);
219 nvkm_wr32(device, 0x003328, ho_put);
220 } else
221 if (ib_get != ib_put)
222 nvkm_wr32(device, 0x003334, ib_put);
223 } else {
224 nvkm_error(subdev, "DMA_PUSHER - ch %d [%s] get %08x put %08x "
225 "state %08x (err: %s) push %08x\n",
226 chid, name, dma_get, dma_put, state,
227 nv_dma_state_err(state), push);
228
229 if (dma_get != dma_put)
230 nvkm_wr32(device, 0x003244, dma_put);
231 }
232 nvkm_fifo_chan_put(&fifo->base, flags, &chan);
233
234 nvkm_wr32(device, 0x003228, 0x00000000);
235 nvkm_wr32(device, 0x003220, 0x00000001);
236 nvkm_wr32(device, 0x002100, NV_PFIFO_INTR_DMA_PUSHER);
237 }
238
239 void
nv04_fifo_intr(struct nvkm_fifo * base)240 nv04_fifo_intr(struct nvkm_fifo *base)
241 {
242 struct nv04_fifo *fifo = nv04_fifo(base);
243 struct nvkm_subdev *subdev = &fifo->base.engine.subdev;
244 struct nvkm_device *device = subdev->device;
245 u32 mask = nvkm_rd32(device, NV03_PFIFO_INTR_EN_0);
246 u32 stat = nvkm_rd32(device, NV03_PFIFO_INTR_0) & mask;
247 u32 reassign, chid, get, sem;
248
249 reassign = nvkm_rd32(device, NV03_PFIFO_CACHES) & 1;
250 nvkm_wr32(device, NV03_PFIFO_CACHES, 0);
251
252 chid = nvkm_rd32(device, NV03_PFIFO_CACHE1_PUSH1) & (fifo->base.nr - 1);
253 get = nvkm_rd32(device, NV03_PFIFO_CACHE1_GET);
254
255 if (stat & NV_PFIFO_INTR_CACHE_ERROR) {
256 nv04_fifo_cache_error(fifo, chid, get);
257 stat &= ~NV_PFIFO_INTR_CACHE_ERROR;
258 }
259
260 if (stat & NV_PFIFO_INTR_DMA_PUSHER) {
261 nv04_fifo_dma_pusher(fifo, chid);
262 stat &= ~NV_PFIFO_INTR_DMA_PUSHER;
263 }
264
265 if (stat & NV_PFIFO_INTR_SEMAPHORE) {
266 stat &= ~NV_PFIFO_INTR_SEMAPHORE;
267 nvkm_wr32(device, NV03_PFIFO_INTR_0, NV_PFIFO_INTR_SEMAPHORE);
268
269 sem = nvkm_rd32(device, NV10_PFIFO_CACHE1_SEMAPHORE);
270 nvkm_wr32(device, NV10_PFIFO_CACHE1_SEMAPHORE, sem | 0x1);
271
272 nvkm_wr32(device, NV03_PFIFO_CACHE1_GET, get + 4);
273 nvkm_wr32(device, NV04_PFIFO_CACHE1_PULL0, 1);
274 }
275
276 if (device->card_type == NV_50) {
277 if (stat & 0x00000010) {
278 stat &= ~0x00000010;
279 nvkm_wr32(device, 0x002100, 0x00000010);
280 }
281
282 if (stat & 0x40000000) {
283 nvkm_wr32(device, 0x002100, 0x40000000);
284 nvkm_fifo_uevent(&fifo->base);
285 stat &= ~0x40000000;
286 }
287 }
288
289 if (stat) {
290 nvkm_warn(subdev, "intr %08x\n", stat);
291 nvkm_mask(device, NV03_PFIFO_INTR_EN_0, stat, 0x00000000);
292 nvkm_wr32(device, NV03_PFIFO_INTR_0, stat);
293 }
294
295 nvkm_wr32(device, NV03_PFIFO_CACHES, reassign);
296 }
297
298 void
nv04_fifo_init(struct nvkm_fifo * base)299 nv04_fifo_init(struct nvkm_fifo *base)
300 {
301 struct nv04_fifo *fifo = nv04_fifo(base);
302 struct nvkm_device *device = fifo->base.engine.subdev.device;
303 struct nvkm_instmem *imem = device->imem;
304 struct nvkm_ramht *ramht = imem->ramht;
305 struct nvkm_memory *ramro = imem->ramro;
306 struct nvkm_memory *ramfc = imem->ramfc;
307
308 nvkm_wr32(device, NV04_PFIFO_DELAY_0, 0x000000ff);
309 nvkm_wr32(device, NV04_PFIFO_DMA_TIMESLICE, 0x0101ffff);
310
311 nvkm_wr32(device, NV03_PFIFO_RAMHT, (0x03 << 24) /* search 128 */ |
312 ((ramht->bits - 9) << 16) |
313 (ramht->gpuobj->addr >> 8));
314 nvkm_wr32(device, NV03_PFIFO_RAMRO, nvkm_memory_addr(ramro) >> 8);
315 nvkm_wr32(device, NV03_PFIFO_RAMFC, nvkm_memory_addr(ramfc) >> 8);
316
317 nvkm_wr32(device, NV03_PFIFO_CACHE1_PUSH1, fifo->base.nr - 1);
318
319 nvkm_wr32(device, NV03_PFIFO_INTR_0, 0xffffffff);
320 nvkm_wr32(device, NV03_PFIFO_INTR_EN_0, 0xffffffff);
321
322 nvkm_wr32(device, NV03_PFIFO_CACHE1_PUSH0, 1);
323 nvkm_wr32(device, NV04_PFIFO_CACHE1_PULL0, 1);
324 nvkm_wr32(device, NV03_PFIFO_CACHES, 1);
325 }
326
327 int
nv04_fifo_new_(const struct nvkm_fifo_func * func,struct nvkm_device * device,int index,int nr,const struct nv04_fifo_ramfc * ramfc,struct nvkm_fifo ** pfifo)328 nv04_fifo_new_(const struct nvkm_fifo_func *func, struct nvkm_device *device,
329 int index, int nr, const struct nv04_fifo_ramfc *ramfc,
330 struct nvkm_fifo **pfifo)
331 {
332 struct nv04_fifo *fifo;
333 int ret;
334
335 if (!(fifo = kzalloc(sizeof(*fifo), GFP_KERNEL)))
336 return -ENOMEM;
337 fifo->ramfc = ramfc;
338 *pfifo = &fifo->base;
339
340 ret = nvkm_fifo_ctor(func, device, index, nr, &fifo->base);
341 if (ret)
342 return ret;
343
344 set_bit(nr - 1, fifo->base.mask); /* inactive channel */
345 return 0;
346 }
347
348 static const struct nvkm_fifo_func
349 nv04_fifo = {
350 .init = nv04_fifo_init,
351 .intr = nv04_fifo_intr,
352 .pause = nv04_fifo_pause,
353 .start = nv04_fifo_start,
354 .chan = {
355 &nv04_fifo_dma_oclass,
356 NULL
357 },
358 };
359
360 int
nv04_fifo_new(struct nvkm_device * device,int index,struct nvkm_fifo ** pfifo)361 nv04_fifo_new(struct nvkm_device *device, int index, struct nvkm_fifo **pfifo)
362 {
363 return nv04_fifo_new_(&nv04_fifo, device, index, 16,
364 nv04_fifo_ramfc, pfifo);
365 }
366