summaryrefslogtreecommitdiffstats
path: root/kernel/drivers/gpu/drm/nouveau/nouveau_dma.c
blob: d168c63533c1797b3ee994b51752a19965082c9d (plain)
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
/*
 * Copyright (C) 2007 Ben Skeggs.
 * 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, sublicense, 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 NONINFRINGEMENT.
 * IN NO EVENT SHALL THE COPYRIGHT OWNER(S) 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 "nouveau_drm.h"
#include "nouveau_dma.h"

void
OUT_RINGp(struct nouveau_channel *chan, const void *data, unsigned nr_dwords)
{
	bool is_iomem;
	u32 *mem = ttm_kmap_obj_virtual(&chan->push.buffer->kmap, &is_iomem);
	mem = &mem[chan->dma.cur];
	if (is_iomem)
		memcpy_toio((void __force __iomem *)mem, data, nr_dwords * 4);
	else
		memcpy(mem, data, nr_dwords * 4);
	chan->dma.cur += nr_dwords;
}

/* Fetch and adjust GPU GET pointer
 *
 * Returns:
 *  value >= 0, the adjusted GET pointer
 *  -EINVAL if GET pointer currently outside main push buffer
 *  -EBUSY if timeout exceeded
 */
static inline int
READ_GET(struct nouveau_channel *chan, uint64_t *prev_get, int *timeout)
{
	uint64_t val;

	val = nvif_rd32(&chan->user, chan->user_get);
        if (chan->user_get_hi)
                val |= (uint64_t)nvif_rd32(&chan->user, chan->user_get_hi) << 32;

	/* reset counter as long as GET is still advancing, this is
	 * to avoid misdetecting a GPU lockup if the GPU happens to
	 * just be processing an operation that takes a long time
	 */
	if (val != *prev_get) {
		*prev_get = val;
		*timeout = 0;
	}

	if ((++*timeout & 0xff) == 0) {
		udelay(1);
		if (*timeout > 100000)
			return -EBUSY;
	}

	if (val < chan->push.vma.offset ||
	    val > chan->push.vma.offset + (chan->dma.max << 2))
		return -EINVAL;

	return (val - chan->push.vma.offset) >> 2;
}

void
nv50_dma_push(struct nouveau_channel *chan, struct nouveau_bo *bo,
	      int delta, int length)
{
	struct nouveau_cli *cli = (void *)chan->user.client;
	struct nouveau_bo *pb = chan->push.buffer;
	struct nvkm_vma *vma;
	int ip = (chan->dma.ib_put * 2) + chan->dma.ib_base;
	u64 offset;

	vma = nouveau_bo_vma_find(bo, cli->vm);
	BUG_ON(!vma);
	offset = vma->offset + delta;

	BUG_ON(chan->dma.ib_free < 1);

	nouveau_bo_wr32(pb, ip++, lower_32_bits(offset));
	nouveau_bo_wr32(pb, ip++, upper_32_bits(offset) | length << 8);

	chan->dma.ib_put = (chan->dma.ib_put + 1) & chan->dma.ib_max;

	mb();
	/* Flush writes. */
	nouveau_bo_rd32(pb, 0);

	nvif_wr32(&chan->user, 0x8c, chan->dma.ib_put);
	chan->dma.ib_free--;
}

static int
nv50_dma_push_wait(struct nouveau_channel *chan, int count)
{
	uint32_t cnt = 0, prev_get = 0;

	while (chan->dma.ib_free < count) {
		uint32_t get = nvif_rd32(&chan->user, 0x88);
		if (get != prev_get) {
			prev_get = get;
			cnt = 0;
		}

		if ((++cnt & 0xff) == 0) {
			DRM_UDELAY(1);
			if (cnt > 100000)
				return -EBUSY;
		}

		chan->dma.ib_free = get - chan->dma.ib_put;
		if (chan->dma.ib_free <= 0)
			chan->dma.ib_free += chan->dma.ib_max;
	}

	return 0;
}

static int
nv50_dma_wait(struct nouveau_channel *chan, int slots, int count)
{
	uint64_t prev_get = 0;
	int ret, cnt = 0;

	ret = nv50_dma_push_wait(chan, slots + 1);
	if (unlikely(ret))
		return ret;

	while (chan->dma.free < count) {
		int get = READ_GET(chan, &prev_get, &cnt);
		if (unlikely(get < 0)) {
			if (get == -EINVAL)
				continue;

			return get;
		}

		if (get <= chan->dma.cur) {
			chan->dma.free = chan->dma.max - chan->dma.cur;
			if (chan->dma.free >= count)
				break;

			FIRE_RING(chan);
			do {
				get = READ_GET(chan, &prev_get, &cnt);
				if (unlikely(get < 0)) {
					if (get == -EINVAL)
						continue;
					return get;
				}
			} while (get == 0);
			chan->dma.cur = 0;
			chan->dma.put = 0;
		}

		chan->dma.free = get - chan->dma.cur - 1;
	}

	return 0;
}

int
nouveau_dma_wait(struct nouveau_channel *chan, int slots, int size)
{
	uint64_t prev_get = 0;
	int cnt = 0, get;

	if (chan->dma.ib_max)
		return nv50_dma_wait(chan, slots, size);

	while (chan->dma.free < size) {
		get = READ_GET(chan, &prev_get, &cnt);
		if (unlikely(get == -EBUSY))
			return -EBUSY;

		/* loop until we have a usable GET pointer.  the value
		 * we read from the GPU may be outside the main ring if
		 * PFIFO is processing a buffer called from the main ring,
		 * discard these values until something sensible is seen.
		 *
		 * the other case we discard GET is while the GPU is fetching
		 * from the SKIPS area, so the code below doesn't have to deal
		 * with some fun corner cases.
		 */
		if (unlikely(get == -EINVAL) || get < NOUVEAU_DMA_SKIPS)
			continue;

		if (get <= chan->dma.cur) {
			/* engine is fetching behind us, or is completely
			 * idle (GET == PUT) so we have free space up until
			 * the end of the push buffer
			 *
			 * we can only hit that path once per call due to
			 * looping back to the beginning of the push buffer,
			 * we'll hit the fetching-ahead-of-us path from that
			 * point on.
			 *
			 * the *one* exception to that rule is if we read
			 * GET==PUT, in which case the below conditional will
			 * always succeed and break us out of the wait loop.
			 */
			chan->dma.free = chan->dma.max - chan->dma.cur;
			if (chan->dma.free >= size)
				break;

			/* not enough space left at the end of the push buffer,
			 * instruct the GPU to jump back to the start right
			 * after processing the currently pending commands.
			 */
			OUT_RING(chan, chan->push.vma.offset | 0x20000000);

			/* wait for GET to depart from the skips area.
			 * prevents writing GET==PUT and causing a race
			 * condition that causes us to think the GPU is
			 * idle when it's not.
			 */
			do {
				get = READ_GET(chan, &prev_get, &cnt);
				if (unlikely(get == -EBUSY))
					return -EBUSY;
				if (unlikely(get == -EINVAL))
					continue;
			} while (get <= NOUVEAU_DMA_SKIPS);
			WRITE_PUT(NOUVEAU_DMA_SKIPS);

			/* we're now submitting commands at the start of
			 * the push buffer.
			 */
			chan->dma.cur  =
			chan->dma.put  = NOUVEAU_DMA_SKIPS;
		}

		/* engine fetching ahead of us, we have space up until the
		 * current GET pointer.  the "- 1" is to ensure there's
		 * space left to emit a jump back to the beginning of the
		 * push buffer if we require it.  we can never get GET == PUT
		 * here, so this is safe.
		 */
		chan->dma.free = get - chan->dma.cur - 1;
	}

	return 0;
}