1 /*
2  * Copyright 2010 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 <subdev/mmu.h>
25 #include <subdev/bar.h>
26 #include <subdev/fb.h>
27 #include <subdev/timer.h>
28 
29 #include <core/engine.h>
30 #include <core/gpuobj.h>
31 
32 struct nv50_mmu_priv {
33 	struct nvkm_mmu base;
34 };
35 
36 static void
37 nv50_vm_map_pgt(struct nvkm_gpuobj *pgd, u32 pde, struct nvkm_gpuobj *pgt[2])
38 {
39 	u64 phys = 0xdeadcafe00000000ULL;
40 	u32 coverage = 0;
41 
42 	if (pgt[0]) {
43 		phys = 0x00000003 | pgt[0]->addr; /* present, 4KiB pages */
44 		coverage = (pgt[0]->size >> 3) << 12;
45 	} else
46 	if (pgt[1]) {
47 		phys = 0x00000001 | pgt[1]->addr; /* present */
48 		coverage = (pgt[1]->size >> 3) << 16;
49 	}
50 
51 	if (phys & 1) {
52 		if (coverage <= 32 * 1024 * 1024)
53 			phys |= 0x60;
54 		else if (coverage <= 64 * 1024 * 1024)
55 			phys |= 0x40;
56 		else if (coverage <= 128 * 1024 * 1024)
57 			phys |= 0x20;
58 	}
59 
60 	nv_wo32(pgd, (pde * 8) + 0, lower_32_bits(phys));
61 	nv_wo32(pgd, (pde * 8) + 4, upper_32_bits(phys));
62 }
63 
64 static inline u64
65 vm_addr(struct nvkm_vma *vma, u64 phys, u32 memtype, u32 target)
66 {
67 	phys |= 1; /* present */
68 	phys |= (u64)memtype << 40;
69 	phys |= target << 4;
70 	if (vma->access & NV_MEM_ACCESS_SYS)
71 		phys |= (1 << 6);
72 	if (!(vma->access & NV_MEM_ACCESS_WO))
73 		phys |= (1 << 3);
74 	return phys;
75 }
76 
77 static void
78 nv50_vm_map(struct nvkm_vma *vma, struct nvkm_gpuobj *pgt,
79 	    struct nvkm_mem *mem, u32 pte, u32 cnt, u64 phys, u64 delta)
80 {
81 	u32 comp = (mem->memtype & 0x180) >> 7;
82 	u32 block, target;
83 	int i;
84 
85 	/* IGPs don't have real VRAM, re-target to stolen system memory */
86 	target = 0;
87 	if (nvkm_fb(vma->vm->mmu)->ram->stolen) {
88 		phys += nvkm_fb(vma->vm->mmu)->ram->stolen;
89 		target = 3;
90 	}
91 
92 	phys  = vm_addr(vma, phys, mem->memtype, target);
93 	pte <<= 3;
94 	cnt <<= 3;
95 
96 	while (cnt) {
97 		u32 offset_h = upper_32_bits(phys);
98 		u32 offset_l = lower_32_bits(phys);
99 
100 		for (i = 7; i >= 0; i--) {
101 			block = 1 << (i + 3);
102 			if (cnt >= block && !(pte & (block - 1)))
103 				break;
104 		}
105 		offset_l |= (i << 7);
106 
107 		phys += block << (vma->node->type - 3);
108 		cnt  -= block;
109 		if (comp) {
110 			u32 tag = mem->tag->offset + ((delta >> 16) * comp);
111 			offset_h |= (tag << 17);
112 			delta    += block << (vma->node->type - 3);
113 		}
114 
115 		while (block) {
116 			nv_wo32(pgt, pte + 0, offset_l);
117 			nv_wo32(pgt, pte + 4, offset_h);
118 			pte += 8;
119 			block -= 8;
120 		}
121 	}
122 }
123 
124 static void
125 nv50_vm_map_sg(struct nvkm_vma *vma, struct nvkm_gpuobj *pgt,
126 	       struct nvkm_mem *mem, u32 pte, u32 cnt, dma_addr_t *list)
127 {
128 	u32 target = (vma->access & NV_MEM_ACCESS_NOSNOOP) ? 3 : 2;
129 	pte <<= 3;
130 	while (cnt--) {
131 		u64 phys = vm_addr(vma, (u64)*list++, mem->memtype, target);
132 		nv_wo32(pgt, pte + 0, lower_32_bits(phys));
133 		nv_wo32(pgt, pte + 4, upper_32_bits(phys));
134 		pte += 8;
135 	}
136 }
137 
138 static void
139 nv50_vm_unmap(struct nvkm_gpuobj *pgt, u32 pte, u32 cnt)
140 {
141 	pte <<= 3;
142 	while (cnt--) {
143 		nv_wo32(pgt, pte + 0, 0x00000000);
144 		nv_wo32(pgt, pte + 4, 0x00000000);
145 		pte += 8;
146 	}
147 }
148 
149 static void
150 nv50_vm_flush(struct nvkm_vm *vm)
151 {
152 	struct nv50_mmu_priv *priv = (void *)vm->mmu;
153 	struct nvkm_bar *bar = nvkm_bar(priv);
154 	struct nvkm_engine *engine;
155 	int i, vme;
156 
157 	bar->flush(bar);
158 
159 	mutex_lock(&nv_subdev(priv)->mutex);
160 	for (i = 0; i < NVDEV_SUBDEV_NR; i++) {
161 		if (!atomic_read(&vm->engref[i]))
162 			continue;
163 
164 		/* unfortunate hw bug workaround... */
165 		engine = nvkm_engine(priv, i);
166 		if (engine && engine->tlb_flush) {
167 			engine->tlb_flush(engine);
168 			continue;
169 		}
170 
171 		switch (i) {
172 		case NVDEV_ENGINE_GR    : vme = 0x00; break;
173 		case NVDEV_ENGINE_VP    :
174 		case NVDEV_ENGINE_MSPDEC: vme = 0x01; break;
175 		case NVDEV_SUBDEV_BAR   : vme = 0x06; break;
176 		case NVDEV_ENGINE_MSPPP :
177 		case NVDEV_ENGINE_MPEG  : vme = 0x08; break;
178 		case NVDEV_ENGINE_BSP   :
179 		case NVDEV_ENGINE_MSVLD : vme = 0x09; break;
180 		case NVDEV_ENGINE_CIPHER:
181 		case NVDEV_ENGINE_SEC   : vme = 0x0a; break;
182 		case NVDEV_ENGINE_CE0   : vme = 0x0d; break;
183 		default:
184 			continue;
185 		}
186 
187 		nv_wr32(priv, 0x100c80, (vme << 16) | 1);
188 		if (!nv_wait(priv, 0x100c80, 0x00000001, 0x00000000))
189 			nv_error(priv, "vm flush timeout: engine %d\n", vme);
190 	}
191 	mutex_unlock(&nv_subdev(priv)->mutex);
192 }
193 
194 static int
195 nv50_vm_create(struct nvkm_mmu *mmu, u64 offset, u64 length,
196 	       u64 mm_offset, struct nvkm_vm **pvm)
197 {
198 	u32 block = (1 << (mmu->pgt_bits + 12));
199 	if (block > length)
200 		block = length;
201 
202 	return nvkm_vm_create(mmu, offset, length, mm_offset, block, pvm);
203 }
204 
205 static int
206 nv50_mmu_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
207 	      struct nvkm_oclass *oclass, void *data, u32 size,
208 	      struct nvkm_object **pobject)
209 {
210 	struct nv50_mmu_priv *priv;
211 	int ret;
212 
213 	ret = nvkm_mmu_create(parent, engine, oclass, "VM", "vm", &priv);
214 	*pobject = nv_object(priv);
215 	if (ret)
216 		return ret;
217 
218 	priv->base.limit = 1ULL << 40;
219 	priv->base.dma_bits = 40;
220 	priv->base.pgt_bits  = 29 - 12;
221 	priv->base.spg_shift = 12;
222 	priv->base.lpg_shift = 16;
223 	priv->base.create = nv50_vm_create;
224 	priv->base.map_pgt = nv50_vm_map_pgt;
225 	priv->base.map = nv50_vm_map;
226 	priv->base.map_sg = nv50_vm_map_sg;
227 	priv->base.unmap = nv50_vm_unmap;
228 	priv->base.flush = nv50_vm_flush;
229 	return 0;
230 }
231 
232 struct nvkm_oclass
233 nv50_mmu_oclass = {
234 	.handle = NV_SUBDEV(MMU, 0x50),
235 	.ofuncs = &(struct nvkm_ofuncs) {
236 		.ctor = nv50_mmu_ctor,
237 		.dtor = _nvkm_mmu_dtor,
238 		.init = _nvkm_mmu_init,
239 		.fini = _nvkm_mmu_fini,
240 	},
241 };
242