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 "nouveau_drv.h"
25 #include "nouveau_dma.h"
26 #include "nouveau_fence.h"
27 #include "nouveau_vmm.h"
28
29 #include "nv50_display.h"
30
31 #include <nvif/push206e.h>
32
33 #include <nvhw/class/cl826f.h>
34
35 static int
nv84_fence_emit32(struct nouveau_channel * chan,u64 virtual,u32 sequence)36 nv84_fence_emit32(struct nouveau_channel *chan, u64 virtual, u32 sequence)
37 {
38 struct nvif_push *push = chan->chan.push;
39 int ret = PUSH_WAIT(push, 8);
40 if (ret == 0) {
41 PUSH_MTHD(push, NV826F, SET_CONTEXT_DMA_SEMAPHORE, chan->vram.handle);
42
43 PUSH_MTHD(push, NV826F, SEMAPHOREA,
44 NVVAL(NV826F, SEMAPHOREA, OFFSET_UPPER, upper_32_bits(virtual)),
45
46 SEMAPHOREB, lower_32_bits(virtual),
47 SEMAPHOREC, sequence,
48
49 SEMAPHORED,
50 NVDEF(NV826F, SEMAPHORED, OPERATION, RELEASE),
51
52 NON_STALLED_INTERRUPT, 0);
53 PUSH_KICK(push);
54 }
55 return ret;
56 }
57
58 static int
nv84_fence_sync32(struct nouveau_channel * chan,u64 virtual,u32 sequence)59 nv84_fence_sync32(struct nouveau_channel *chan, u64 virtual, u32 sequence)
60 {
61 struct nvif_push *push = chan->chan.push;
62 int ret = PUSH_WAIT(push, 7);
63 if (ret == 0) {
64 PUSH_MTHD(push, NV826F, SET_CONTEXT_DMA_SEMAPHORE, chan->vram.handle);
65
66 PUSH_MTHD(push, NV826F, SEMAPHOREA,
67 NVVAL(NV826F, SEMAPHOREA, OFFSET_UPPER, upper_32_bits(virtual)),
68
69 SEMAPHOREB, lower_32_bits(virtual),
70 SEMAPHOREC, sequence,
71
72 SEMAPHORED,
73 NVDEF(NV826F, SEMAPHORED, OPERATION, ACQ_GEQ));
74 PUSH_KICK(push);
75 }
76 return ret;
77 }
78
79 static int
nv84_fence_emit(struct nouveau_fence * fence)80 nv84_fence_emit(struct nouveau_fence *fence)
81 {
82 struct nouveau_channel *chan = fence->channel;
83 struct nv84_fence_chan *fctx = chan->fence;
84 u64 addr = fctx->vma->addr + chan->chid * 16;
85
86 return fctx->base.emit32(chan, addr, fence->base.seqno);
87 }
88
89 static int
nv84_fence_sync(struct nouveau_fence * fence,struct nouveau_channel * prev,struct nouveau_channel * chan)90 nv84_fence_sync(struct nouveau_fence *fence,
91 struct nouveau_channel *prev, struct nouveau_channel *chan)
92 {
93 struct nv84_fence_chan *fctx = chan->fence;
94 u64 addr = fctx->vma->addr + prev->chid * 16;
95
96 return fctx->base.sync32(chan, addr, fence->base.seqno);
97 }
98
99 static u32
nv84_fence_read(struct nouveau_channel * chan)100 nv84_fence_read(struct nouveau_channel *chan)
101 {
102 struct nv84_fence_priv *priv = chan->drm->fence;
103 return nouveau_bo_rd32(priv->bo, chan->chid * 16/4);
104 }
105
106 static void
nv84_fence_context_del(struct nouveau_channel * chan)107 nv84_fence_context_del(struct nouveau_channel *chan)
108 {
109 struct nv84_fence_priv *priv = chan->drm->fence;
110 struct nv84_fence_chan *fctx = chan->fence;
111
112 nouveau_bo_wr32(priv->bo, chan->chid * 16 / 4, fctx->base.sequence);
113 mutex_lock(&priv->mutex);
114 nouveau_vma_del(&fctx->vma);
115 mutex_unlock(&priv->mutex);
116 nouveau_fence_context_del(&fctx->base);
117 chan->fence = NULL;
118 nouveau_fence_context_free(&fctx->base);
119 }
120
121 int
nv84_fence_context_new(struct nouveau_channel * chan)122 nv84_fence_context_new(struct nouveau_channel *chan)
123 {
124 struct nv84_fence_priv *priv = chan->drm->fence;
125 struct nv84_fence_chan *fctx;
126 int ret;
127
128 fctx = chan->fence = kzalloc(sizeof(*fctx), GFP_KERNEL);
129 if (!fctx)
130 return -ENOMEM;
131
132 nouveau_fence_context_new(chan, &fctx->base);
133 fctx->base.emit = nv84_fence_emit;
134 fctx->base.sync = nv84_fence_sync;
135 fctx->base.read = nv84_fence_read;
136 fctx->base.emit32 = nv84_fence_emit32;
137 fctx->base.sync32 = nv84_fence_sync32;
138 fctx->base.sequence = nv84_fence_read(chan);
139
140 mutex_lock(&priv->mutex);
141 ret = nouveau_vma_new(priv->bo, chan->vmm, &fctx->vma);
142 mutex_unlock(&priv->mutex);
143
144 if (ret)
145 nv84_fence_context_del(chan);
146 return ret;
147 }
148
149 static bool
nv84_fence_suspend(struct nouveau_drm * drm)150 nv84_fence_suspend(struct nouveau_drm *drm)
151 {
152 struct nv84_fence_priv *priv = drm->fence;
153 int i;
154
155 priv->suspend = vmalloc(array_size(sizeof(u32), drm->chan.nr));
156 if (priv->suspend) {
157 for (i = 0; i < drm->chan.nr; i++)
158 priv->suspend[i] = nouveau_bo_rd32(priv->bo, i*4);
159 }
160
161 return priv->suspend != NULL;
162 }
163
164 static void
nv84_fence_resume(struct nouveau_drm * drm)165 nv84_fence_resume(struct nouveau_drm *drm)
166 {
167 struct nv84_fence_priv *priv = drm->fence;
168 int i;
169
170 if (priv->suspend) {
171 for (i = 0; i < drm->chan.nr; i++)
172 nouveau_bo_wr32(priv->bo, i*4, priv->suspend[i]);
173 vfree(priv->suspend);
174 priv->suspend = NULL;
175 }
176 }
177
178 static void
nv84_fence_destroy(struct nouveau_drm * drm)179 nv84_fence_destroy(struct nouveau_drm *drm)
180 {
181 struct nv84_fence_priv *priv = drm->fence;
182 nouveau_bo_unmap(priv->bo);
183 if (priv->bo)
184 nouveau_bo_unpin(priv->bo);
185 nouveau_bo_ref(NULL, &priv->bo);
186 drm->fence = NULL;
187 kfree(priv);
188 }
189
190 int
nv84_fence_create(struct nouveau_drm * drm)191 nv84_fence_create(struct nouveau_drm *drm)
192 {
193 struct nv84_fence_priv *priv;
194 u32 domain;
195 int ret;
196
197 priv = drm->fence = kzalloc(sizeof(*priv), GFP_KERNEL);
198 if (!priv)
199 return -ENOMEM;
200
201 priv->base.dtor = nv84_fence_destroy;
202 priv->base.suspend = nv84_fence_suspend;
203 priv->base.resume = nv84_fence_resume;
204 priv->base.context_new = nv84_fence_context_new;
205 priv->base.context_del = nv84_fence_context_del;
206
207 priv->base.uevent = drm->client.device.info.family < NV_DEVICE_INFO_V0_AMPERE;
208
209 mutex_init(&priv->mutex);
210
211 /* Use VRAM if there is any ; otherwise fallback to system memory */
212 domain = drm->client.device.info.ram_size != 0 ?
213 NOUVEAU_GEM_DOMAIN_VRAM :
214 /*
215 * fences created in sysmem must be non-cached or we
216 * will lose CPU/GPU coherency!
217 */
218 NOUVEAU_GEM_DOMAIN_GART | NOUVEAU_GEM_DOMAIN_COHERENT;
219 ret = nouveau_bo_new(&drm->client, 16 * drm->chan.nr, 0,
220 domain, 0, 0, NULL, NULL, &priv->bo);
221 if (ret == 0) {
222 ret = nouveau_bo_pin(priv->bo, domain, false);
223 if (ret == 0) {
224 ret = nouveau_bo_map(priv->bo);
225 if (ret)
226 nouveau_bo_unpin(priv->bo);
227 }
228 if (ret)
229 nouveau_bo_ref(NULL, &priv->bo);
230 }
231
232 if (ret)
233 nv84_fence_destroy(drm);
234 return ret;
235 }
236