xref: /netbsd-src/sys/external/bsd/drm2/dist/drm/nouveau/nouveau_nv84_fence.c (revision 33881f779a77dce6440bdc44610d94de75bebefe)
1 /*	$NetBSD: nouveau_nv84_fence.c,v 1.7 2020/02/14 04:35:20 riastradh Exp $	*/
2 
3 /*
4  * Copyright 2012 Red Hat Inc.
5  *
6  * Permission is hereby granted, free of charge, to any person obtaining a
7  * copy of this software and associated documentation files (the "Software"),
8  * to deal in the Software without restriction, including without limitation
9  * the rights to use, copy, modify, merge, publish, distribute, sublicense,
10  * and/or sell copies of the Software, and to permit persons to whom the
11  * Software is furnished to do so, subject to the following conditions:
12  *
13  * The above copyright notice and this permission notice shall be included in
14  * all copies or substantial portions of the Software.
15  *
16  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
17  * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
18  * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
19  * THE COPYRIGHT HOLDER(S) OR AUTHOR(S) BE LIABLE FOR ANY CLAIM, DAMAGES OR
20  * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
21  * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR
22  * OTHER DEALINGS IN THE SOFTWARE.
23  *
24  * Authors: Ben Skeggs
25  */
26 
27 #include <sys/cdefs.h>
28 __KERNEL_RCSID(0, "$NetBSD: nouveau_nv84_fence.c,v 1.7 2020/02/14 04:35:20 riastradh Exp $");
29 
30 #include "nouveau_drm.h"
31 #include "nouveau_dma.h"
32 #include "nouveau_fence.h"
33 
34 #include "nv50_display.h"
35 
36 #include <linux/nbsd-namespace.h>
37 
38 u64
39 nv84_fence_crtc(struct nouveau_channel *chan, int crtc)
40 {
41 	struct nv84_fence_chan *fctx = chan->fence;
42 	return fctx->dispc_vma[crtc].offset;
43 }
44 
45 static int
46 nv84_fence_emit32(struct nouveau_channel *chan, u64 virtual, u32 sequence)
47 {
48 	int ret = RING_SPACE(chan, 8);
49 	if (ret == 0) {
50 		BEGIN_NV04(chan, 0, NV11_SUBCHAN_DMA_SEMAPHORE, 1);
51 		OUT_RING  (chan, chan->vram.handle);
52 		BEGIN_NV04(chan, 0, NV84_SUBCHAN_SEMAPHORE_ADDRESS_HIGH, 5);
53 		OUT_RING  (chan, upper_32_bits(virtual));
54 		OUT_RING  (chan, lower_32_bits(virtual));
55 		OUT_RING  (chan, sequence);
56 		OUT_RING  (chan, NV84_SUBCHAN_SEMAPHORE_TRIGGER_WRITE_LONG);
57 		OUT_RING  (chan, 0x00000000);
58 		FIRE_RING (chan);
59 	}
60 	return ret;
61 }
62 
63 static int
64 nv84_fence_sync32(struct nouveau_channel *chan, u64 virtual, u32 sequence)
65 {
66 	int ret = RING_SPACE(chan, 7);
67 	if (ret == 0) {
68 		BEGIN_NV04(chan, 0, NV11_SUBCHAN_DMA_SEMAPHORE, 1);
69 		OUT_RING  (chan, chan->vram.handle);
70 		BEGIN_NV04(chan, 0, NV84_SUBCHAN_SEMAPHORE_ADDRESS_HIGH, 4);
71 		OUT_RING  (chan, upper_32_bits(virtual));
72 		OUT_RING  (chan, lower_32_bits(virtual));
73 		OUT_RING  (chan, sequence);
74 		OUT_RING  (chan, NV84_SUBCHAN_SEMAPHORE_TRIGGER_ACQUIRE_GEQUAL);
75 		FIRE_RING (chan);
76 	}
77 	return ret;
78 }
79 
80 static int
81 nv84_fence_emit(struct nouveau_fence *fence)
82 {
83 	struct nouveau_channel *chan = fence->channel;
84 	struct nv84_fence_chan *fctx = chan->fence;
85 	u64 addr = chan->chid * 16;
86 
87 	if (fence->sysmem)
88 		addr += fctx->vma_gart.offset;
89 	else
90 		addr += fctx->vma.offset;
91 
92 	return fctx->base.emit32(chan, addr, fence->base.seqno);
93 }
94 
95 static int
96 nv84_fence_sync(struct nouveau_fence *fence,
97 		struct nouveau_channel *prev, struct nouveau_channel *chan)
98 {
99 	struct nv84_fence_chan *fctx = chan->fence;
100 	u64 addr = prev->chid * 16;
101 
102 	if (fence->sysmem)
103 		addr += fctx->vma_gart.offset;
104 	else
105 		addr += fctx->vma.offset;
106 
107 	return fctx->base.sync32(chan, addr, fence->base.seqno);
108 }
109 
110 static u32
111 nv84_fence_read(struct nouveau_channel *chan)
112 {
113 	struct nv84_fence_priv *priv = chan->drm->fence;
114 	return nouveau_bo_rd32(priv->bo, chan->chid * 16/4);
115 }
116 
117 static void
118 nv84_fence_context_del(struct nouveau_channel *chan)
119 {
120 	struct drm_device *dev = chan->drm->dev;
121 	struct nv84_fence_priv *priv = chan->drm->fence;
122 	struct nv84_fence_chan *fctx = chan->fence;
123 	int i;
124 
125 	for (i = 0; i < dev->mode_config.num_crtc; i++) {
126 		struct nouveau_bo *bo = nv50_display_crtc_sema(dev, i);
127 		nouveau_bo_vma_del(bo, &fctx->dispc_vma[i]);
128 	}
129 
130 	nouveau_bo_wr32(priv->bo, chan->chid * 16 / 4, fctx->base.sequence);
131 	mutex_lock(&priv->mutex);
132 	nouveau_bo_vma_del(priv->bo, &fctx->vma_gart);
133 	nouveau_bo_vma_del(priv->bo, &fctx->vma);
134 	mutex_unlock(&priv->mutex);
135 	nouveau_fence_context_del(&fctx->base);
136 	chan->fence = NULL;
137 	nouveau_fence_context_free(&fctx->base);
138 }
139 
140 int
141 nv84_fence_context_new(struct nouveau_channel *chan)
142 {
143 	struct nouveau_cli *cli = (void *)chan->user.client;
144 	struct nv84_fence_priv *priv = chan->drm->fence;
145 	struct nv84_fence_chan *fctx;
146 	int ret, i;
147 
148 	fctx = chan->fence = kzalloc(sizeof(*fctx), GFP_KERNEL);
149 	if (!fctx)
150 		return -ENOMEM;
151 
152 	nouveau_fence_context_new(chan, &fctx->base);
153 	fctx->base.emit = nv84_fence_emit;
154 	fctx->base.sync = nv84_fence_sync;
155 	fctx->base.read = nv84_fence_read;
156 	fctx->base.emit32 = nv84_fence_emit32;
157 	fctx->base.sync32 = nv84_fence_sync32;
158 	fctx->base.sequence = nv84_fence_read(chan);
159 
160 	mutex_lock(&priv->mutex);
161 	ret = nouveau_bo_vma_add(priv->bo, cli->vm, &fctx->vma);
162 	if (ret == 0) {
163 		ret = nouveau_bo_vma_add(priv->bo_gart, cli->vm,
164 					&fctx->vma_gart);
165 	}
166 	mutex_unlock(&priv->mutex);
167 
168 	/* map display semaphore buffers into channel's vm */
169 	for (i = 0; !ret && i < chan->drm->dev->mode_config.num_crtc; i++) {
170 		struct nouveau_bo *bo = nv50_display_crtc_sema(chan->drm->dev, i);
171 		ret = nouveau_bo_vma_add(bo, cli->vm, &fctx->dispc_vma[i]);
172 	}
173 
174 	if (ret)
175 		nv84_fence_context_del(chan);
176 	return ret;
177 }
178 
179 static bool
180 nv84_fence_suspend(struct nouveau_drm *drm)
181 {
182 	struct nv84_fence_priv *priv = drm->fence;
183 	int i;
184 
185 	priv->suspend = vmalloc(priv->base.contexts * sizeof(u32));
186 	if (priv->suspend) {
187 		for (i = 0; i < priv->base.contexts; i++)
188 			priv->suspend[i] = nouveau_bo_rd32(priv->bo, i*4);
189 	}
190 
191 	return priv->suspend != NULL;
192 }
193 
194 static void
195 nv84_fence_resume(struct nouveau_drm *drm)
196 {
197 	struct nv84_fence_priv *priv = drm->fence;
198 	int i;
199 
200 	if (priv->suspend) {
201 		for (i = 0; i < priv->base.contexts; i++)
202 			nouveau_bo_wr32(priv->bo, i*4, priv->suspend[i]);
203 		vfree(priv->suspend);
204 		priv->suspend = NULL;
205 	}
206 }
207 
208 static void
209 nv84_fence_destroy(struct nouveau_drm *drm)
210 {
211 	struct nv84_fence_priv *priv = drm->fence;
212 
213 	nouveau_bo_unmap(priv->bo_gart);
214 	if (priv->bo_gart)
215 		nouveau_bo_unpin(priv->bo_gart);
216 	nouveau_bo_ref(NULL, &priv->bo_gart);
217 	nouveau_bo_unmap(priv->bo);
218 	if (priv->bo)
219 		nouveau_bo_unpin(priv->bo);
220 	nouveau_bo_ref(NULL, &priv->bo);
221 	drm->fence = NULL;
222 	mutex_destroy(&priv->mutex);
223 	kfree(priv);
224 }
225 
226 int
227 nv84_fence_create(struct nouveau_drm *drm)
228 {
229 	struct nvkm_fifo *fifo = nvxx_fifo(&drm->device);
230 	struct nv84_fence_priv *priv;
231 	u32 domain;
232 	int ret;
233 
234 	priv = drm->fence = kzalloc(sizeof(*priv), GFP_KERNEL);
235 	if (!priv)
236 		return -ENOMEM;
237 
238 	priv->base.dtor = nv84_fence_destroy;
239 	priv->base.suspend = nv84_fence_suspend;
240 	priv->base.resume = nv84_fence_resume;
241 	priv->base.context_new = nv84_fence_context_new;
242 	priv->base.context_del = nv84_fence_context_del;
243 
244 	priv->base.contexts = fifo->nr;
245 	priv->base.context_base = fence_context_alloc(priv->base.contexts);
246 	priv->base.uevent = true;
247 
248 	mutex_init(&priv->mutex);
249 
250 	/* Use VRAM if there is any ; otherwise fallback to system memory */
251 	domain = drm->device.info.ram_size != 0 ? TTM_PL_FLAG_VRAM :
252 			 /*
253 			  * fences created in sysmem must be non-cached or we
254 			  * will lose CPU/GPU coherency!
255 			  */
256 			 TTM_PL_FLAG_TT | TTM_PL_FLAG_UNCACHED;
257 	ret = nouveau_bo_new(drm->dev, 16 * priv->base.contexts, 0, domain, 0,
258 			     0, NULL, NULL, &priv->bo);
259 	if (ret == 0) {
260 		ret = nouveau_bo_pin(priv->bo, domain, false);
261 		if (ret == 0) {
262 			ret = nouveau_bo_map(priv->bo);
263 			if (ret)
264 				nouveau_bo_unpin(priv->bo);
265 		}
266 		if (ret)
267 			nouveau_bo_ref(NULL, &priv->bo);
268 	}
269 
270 	if (ret == 0)
271 		ret = nouveau_bo_new(drm->dev, 16 * priv->base.contexts, 0,
272 				     TTM_PL_FLAG_TT | TTM_PL_FLAG_UNCACHED, 0,
273 				     0, NULL, NULL, &priv->bo_gart);
274 	if (ret == 0) {
275 		ret = nouveau_bo_pin(priv->bo_gart, TTM_PL_FLAG_TT, false);
276 		if (ret == 0) {
277 			ret = nouveau_bo_map(priv->bo_gart);
278 			if (ret)
279 				nouveau_bo_unpin(priv->bo_gart);
280 		}
281 		if (ret)
282 			nouveau_bo_ref(NULL, &priv->bo_gart);
283 	}
284 
285 	if (ret)
286 		nv84_fence_destroy(drm);
287 	return ret;
288 }
289