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 
26 #include <core/device.h>
27 #include <core/gpuobj.h>
28 #include <core/option.h>
29 #include <subdev/timer.h>
30 
31 #define NV44_GART_SIZE (512 * 1024 * 1024)
32 #define NV44_GART_PAGE (  4 * 1024)
33 
34 /*******************************************************************************
35  * VM map/unmap callbacks
36  ******************************************************************************/
37 
38 static void
39 nv44_vm_fill(struct nvkm_gpuobj *pgt, dma_addr_t null,
40 	     dma_addr_t *list, u32 pte, u32 cnt)
41 {
42 	u32 base = (pte << 2) & ~0x0000000f;
43 	u32 tmp[4];
44 
45 	tmp[0] = nv_ro32(pgt, base + 0x0);
46 	tmp[1] = nv_ro32(pgt, base + 0x4);
47 	tmp[2] = nv_ro32(pgt, base + 0x8);
48 	tmp[3] = nv_ro32(pgt, base + 0xc);
49 
50 	while (cnt--) {
51 		u32 addr = list ? (*list++ >> 12) : (null >> 12);
52 		switch (pte++ & 0x3) {
53 		case 0:
54 			tmp[0] &= ~0x07ffffff;
55 			tmp[0] |= addr;
56 			break;
57 		case 1:
58 			tmp[0] &= ~0xf8000000;
59 			tmp[0] |= addr << 27;
60 			tmp[1] &= ~0x003fffff;
61 			tmp[1] |= addr >> 5;
62 			break;
63 		case 2:
64 			tmp[1] &= ~0xffc00000;
65 			tmp[1] |= addr << 22;
66 			tmp[2] &= ~0x0001ffff;
67 			tmp[2] |= addr >> 10;
68 			break;
69 		case 3:
70 			tmp[2] &= ~0xfffe0000;
71 			tmp[2] |= addr << 17;
72 			tmp[3] &= ~0x00000fff;
73 			tmp[3] |= addr >> 15;
74 			break;
75 		}
76 	}
77 
78 	nv_wo32(pgt, base + 0x0, tmp[0]);
79 	nv_wo32(pgt, base + 0x4, tmp[1]);
80 	nv_wo32(pgt, base + 0x8, tmp[2]);
81 	nv_wo32(pgt, base + 0xc, tmp[3] | 0x40000000);
82 }
83 
84 static void
85 nv44_vm_map_sg(struct nvkm_vma *vma, struct nvkm_gpuobj *pgt,
86 	       struct nvkm_mem *mem, u32 pte, u32 cnt, dma_addr_t *list)
87 {
88 	struct nv04_mmu_priv *priv = (void *)vma->vm->mmu;
89 	u32 tmp[4];
90 	int i;
91 
92 	if (pte & 3) {
93 		u32  max = 4 - (pte & 3);
94 		u32 part = (cnt > max) ? max : cnt;
95 		nv44_vm_fill(pgt, priv->null, list, pte, part);
96 		pte  += part;
97 		list += part;
98 		cnt  -= part;
99 	}
100 
101 	while (cnt >= 4) {
102 		for (i = 0; i < 4; i++)
103 			tmp[i] = *list++ >> 12;
104 		nv_wo32(pgt, pte++ * 4, tmp[0] >>  0 | tmp[1] << 27);
105 		nv_wo32(pgt, pte++ * 4, tmp[1] >>  5 | tmp[2] << 22);
106 		nv_wo32(pgt, pte++ * 4, tmp[2] >> 10 | tmp[3] << 17);
107 		nv_wo32(pgt, pte++ * 4, tmp[3] >> 15 | 0x40000000);
108 		cnt -= 4;
109 	}
110 
111 	if (cnt)
112 		nv44_vm_fill(pgt, priv->null, list, pte, cnt);
113 }
114 
115 static void
116 nv44_vm_unmap(struct nvkm_gpuobj *pgt, u32 pte, u32 cnt)
117 {
118 	struct nv04_mmu_priv *priv = (void *)nvkm_mmu(pgt);
119 
120 	if (pte & 3) {
121 		u32  max = 4 - (pte & 3);
122 		u32 part = (cnt > max) ? max : cnt;
123 		nv44_vm_fill(pgt, priv->null, NULL, pte, part);
124 		pte  += part;
125 		cnt  -= part;
126 	}
127 
128 	while (cnt >= 4) {
129 		nv_wo32(pgt, pte++ * 4, 0x00000000);
130 		nv_wo32(pgt, pte++ * 4, 0x00000000);
131 		nv_wo32(pgt, pte++ * 4, 0x00000000);
132 		nv_wo32(pgt, pte++ * 4, 0x00000000);
133 		cnt -= 4;
134 	}
135 
136 	if (cnt)
137 		nv44_vm_fill(pgt, priv->null, NULL, pte, cnt);
138 }
139 
140 static void
141 nv44_vm_flush(struct nvkm_vm *vm)
142 {
143 	struct nv04_mmu_priv *priv = (void *)vm->mmu;
144 	nv_wr32(priv, 0x100814, priv->base.limit - NV44_GART_PAGE);
145 	nv_wr32(priv, 0x100808, 0x00000020);
146 	if (!nv_wait(priv, 0x100808, 0x00000001, 0x00000001))
147 		nv_error(priv, "timeout: 0x%08x\n", nv_rd32(priv, 0x100808));
148 	nv_wr32(priv, 0x100808, 0x00000000);
149 }
150 
151 /*******************************************************************************
152  * MMU subdev
153  ******************************************************************************/
154 
155 static int
156 nv44_mmu_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
157 	      struct nvkm_oclass *oclass, void *data, u32 size,
158 	      struct nvkm_object **pobject)
159 {
160 	struct nvkm_device *device = nv_device(parent);
161 	struct nv04_mmu_priv *priv;
162 	int ret;
163 
164 	if (pci_find_capability(device->pdev, PCI_CAP_ID_AGP) ||
165 	    !nvkm_boolopt(device->cfgopt, "NvPCIE", true)) {
166 		return nvkm_object_ctor(parent, engine, &nv04_mmu_oclass,
167 					data, size, pobject);
168 	}
169 
170 	ret = nvkm_mmu_create(parent, engine, oclass, "PCIEGART",
171 			      "pciegart", &priv);
172 	*pobject = nv_object(priv);
173 	if (ret)
174 		return ret;
175 
176 	priv->base.create = nv04_vm_create;
177 	priv->base.limit = NV44_GART_SIZE;
178 	priv->base.dma_bits = 39;
179 	priv->base.pgt_bits = 32 - 12;
180 	priv->base.spg_shift = 12;
181 	priv->base.lpg_shift = 12;
182 	priv->base.map_sg = nv44_vm_map_sg;
183 	priv->base.unmap = nv44_vm_unmap;
184 	priv->base.flush = nv44_vm_flush;
185 
186 	priv->nullp = pci_alloc_consistent(device->pdev, 16 * 1024, &priv->null);
187 	if (!priv->nullp) {
188 		nv_error(priv, "unable to allocate dummy pages\n");
189 		return -ENOMEM;
190 	}
191 
192 	ret = nvkm_vm_create(&priv->base, 0, NV44_GART_SIZE, 0, 4096,
193 			     &priv->vm);
194 	if (ret)
195 		return ret;
196 
197 	ret = nvkm_gpuobj_new(nv_object(priv), NULL,
198 			      (NV44_GART_SIZE / NV44_GART_PAGE) * 4,
199 			      512 * 1024, NVOBJ_FLAG_ZERO_ALLOC,
200 			      &priv->vm->pgt[0].obj[0]);
201 	priv->vm->pgt[0].refcount[0] = 1;
202 	if (ret)
203 		return ret;
204 
205 	return 0;
206 }
207 
208 static int
209 nv44_mmu_init(struct nvkm_object *object)
210 {
211 	struct nv04_mmu_priv *priv = (void *)object;
212 	struct nvkm_gpuobj *gart = priv->vm->pgt[0].obj[0];
213 	u32 addr;
214 	int ret;
215 
216 	ret = nvkm_mmu_init(&priv->base);
217 	if (ret)
218 		return ret;
219 
220 	/* calculate vram address of this PRAMIN block, object must be
221 	 * allocated on 512KiB alignment, and not exceed a total size
222 	 * of 512KiB for this to work correctly
223 	 */
224 	addr  = nv_rd32(priv, 0x10020c);
225 	addr -= ((gart->addr >> 19) + 1) << 19;
226 
227 	nv_wr32(priv, 0x100850, 0x80000000);
228 	nv_wr32(priv, 0x100818, priv->null);
229 	nv_wr32(priv, 0x100804, NV44_GART_SIZE);
230 	nv_wr32(priv, 0x100850, 0x00008000);
231 	nv_mask(priv, 0x10008c, 0x00000200, 0x00000200);
232 	nv_wr32(priv, 0x100820, 0x00000000);
233 	nv_wr32(priv, 0x10082c, 0x00000001);
234 	nv_wr32(priv, 0x100800, addr | 0x00000010);
235 	return 0;
236 }
237 
238 struct nvkm_oclass
239 nv44_mmu_oclass = {
240 	.handle = NV_SUBDEV(MMU, 0x44),
241 	.ofuncs = &(struct nvkm_ofuncs) {
242 		.ctor = nv44_mmu_ctor,
243 		.dtor = nv04_mmu_dtor,
244 		.init = nv44_mmu_init,
245 		.fini = _nvkm_mmu_fini,
246 	},
247 };
248