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/fb.h>
26 #include <subdev/timer.h>
27 
28 #include <core/engine.h>
29 #include <core/gpuobj.h>
30 
31 static void
32 nv50_vm_map_pgt(struct nvkm_gpuobj *pgd, u32 pde, struct nvkm_memory *pgt[2])
33 {
34 	u64 phys = 0xdeadcafe00000000ULL;
35 	u32 coverage = 0;
36 
37 	if (pgt[0]) {
38 		/* present, 4KiB pages */
39 		phys = 0x00000003 | nvkm_memory_addr(pgt[0]);
40 		coverage = (nvkm_memory_size(pgt[0]) >> 3) << 12;
41 	} else
42 	if (pgt[1]) {
43 		/* present, 64KiB pages  */
44 		phys = 0x00000001 | nvkm_memory_addr(pgt[1]);
45 		coverage = (nvkm_memory_size(pgt[1]) >> 3) << 16;
46 	}
47 
48 	if (phys & 1) {
49 		if (coverage <= 32 * 1024 * 1024)
50 			phys |= 0x60;
51 		else if (coverage <= 64 * 1024 * 1024)
52 			phys |= 0x40;
53 		else if (coverage <= 128 * 1024 * 1024)
54 			phys |= 0x20;
55 	}
56 
57 	nvkm_kmap(pgd);
58 	nvkm_wo32(pgd, (pde * 8) + 0, lower_32_bits(phys));
59 	nvkm_wo32(pgd, (pde * 8) + 4, upper_32_bits(phys));
60 	nvkm_done(pgd);
61 }
62 
63 static inline u64
64 vm_addr(struct nvkm_vma *vma, u64 phys, u32 memtype, u32 target)
65 {
66 	phys |= 1; /* present */
67 	phys |= (u64)memtype << 40;
68 	phys |= target << 4;
69 	if (vma->access & NV_MEM_ACCESS_SYS)
70 		phys |= (1 << 6);
71 	if (!(vma->access & NV_MEM_ACCESS_WO))
72 		phys |= (1 << 3);
73 	return phys;
74 }
75 
76 static void
77 nv50_vm_map(struct nvkm_vma *vma, struct nvkm_memory *pgt,
78 	    struct nvkm_mem *mem, u32 pte, u32 cnt, u64 phys, u64 delta)
79 {
80 	u32 comp = (mem->memtype & 0x180) >> 7;
81 	u32 block, target;
82 	int i;
83 
84 	/* IGPs don't have real VRAM, re-target to stolen system memory */
85 	target = 0;
86 	if (nvkm_fb(vma->vm->mmu)->ram->stolen) {
87 		phys += nvkm_fb(vma->vm->mmu)->ram->stolen;
88 		target = 3;
89 	}
90 
91 	phys  = vm_addr(vma, phys, mem->memtype, target);
92 	pte <<= 3;
93 	cnt <<= 3;
94 
95 	nvkm_kmap(pgt);
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 			nvkm_wo32(pgt, pte + 0, offset_l);
117 			nvkm_wo32(pgt, pte + 4, offset_h);
118 			pte += 8;
119 			block -= 8;
120 		}
121 	}
122 	nvkm_done(pgt);
123 }
124 
125 static void
126 nv50_vm_map_sg(struct nvkm_vma *vma, struct nvkm_memory *pgt,
127 	       struct nvkm_mem *mem, u32 pte, u32 cnt, dma_addr_t *list)
128 {
129 	u32 target = (vma->access & NV_MEM_ACCESS_NOSNOOP) ? 3 : 2;
130 	pte <<= 3;
131 	nvkm_kmap(pgt);
132 	while (cnt--) {
133 		u64 phys = vm_addr(vma, (u64)*list++, mem->memtype, target);
134 		nvkm_wo32(pgt, pte + 0, lower_32_bits(phys));
135 		nvkm_wo32(pgt, pte + 4, upper_32_bits(phys));
136 		pte += 8;
137 	}
138 	nvkm_done(pgt);
139 }
140 
141 static void
142 nv50_vm_unmap(struct nvkm_vma *vma, struct nvkm_memory *pgt, u32 pte, u32 cnt)
143 {
144 	pte <<= 3;
145 	nvkm_kmap(pgt);
146 	while (cnt--) {
147 		nvkm_wo32(pgt, pte + 0, 0x00000000);
148 		nvkm_wo32(pgt, pte + 4, 0x00000000);
149 		pte += 8;
150 	}
151 	nvkm_done(pgt);
152 }
153 
154 static void
155 nv50_vm_flush(struct nvkm_vm *vm)
156 {
157 	struct nvkm_mmu *mmu = (void *)vm->mmu;
158 	struct nvkm_subdev *subdev = &mmu->subdev;
159 	struct nvkm_device *device = subdev->device;
160 	struct nvkm_engine *engine;
161 	int i, vme;
162 
163 	mutex_lock(&subdev->mutex);
164 	for (i = 0; i < NVDEV_SUBDEV_NR; i++) {
165 		if (!atomic_read(&vm->engref[i]))
166 			continue;
167 
168 		/* unfortunate hw bug workaround... */
169 		engine = nvkm_engine(mmu, i);
170 		if (engine && engine->tlb_flush) {
171 			engine->tlb_flush(engine);
172 			continue;
173 		}
174 
175 		switch (i) {
176 		case NVDEV_ENGINE_GR    : vme = 0x00; break;
177 		case NVDEV_ENGINE_VP    :
178 		case NVDEV_ENGINE_MSPDEC: vme = 0x01; break;
179 		case NVDEV_SUBDEV_BAR   : vme = 0x06; break;
180 		case NVDEV_ENGINE_MSPPP :
181 		case NVDEV_ENGINE_MPEG  : vme = 0x08; break;
182 		case NVDEV_ENGINE_BSP   :
183 		case NVDEV_ENGINE_MSVLD : vme = 0x09; break;
184 		case NVDEV_ENGINE_CIPHER:
185 		case NVDEV_ENGINE_SEC   : vme = 0x0a; break;
186 		case NVDEV_ENGINE_CE0   : vme = 0x0d; break;
187 		default:
188 			continue;
189 		}
190 
191 		nvkm_wr32(device, 0x100c80, (vme << 16) | 1);
192 		if (nvkm_msec(device, 2000,
193 			if (!(nvkm_rd32(device, 0x100c80) & 0x00000001))
194 				break;
195 		) < 0)
196 			nvkm_error(subdev, "vm flush timeout: engine %d\n", vme);
197 	}
198 	mutex_unlock(&subdev->mutex);
199 }
200 
201 static int
202 nv50_vm_create(struct nvkm_mmu *mmu, u64 offset, u64 length, u64 mm_offset,
203 	       struct lock_class_key *key, struct nvkm_vm **pvm)
204 {
205 	u32 block = (1 << (mmu->pgt_bits + 12));
206 	if (block > length)
207 		block = length;
208 
209 	return nvkm_vm_create(mmu, offset, length, mm_offset, block, key, pvm);
210 }
211 
212 static int
213 nv50_mmu_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
214 	      struct nvkm_oclass *oclass, void *data, u32 size,
215 	      struct nvkm_object **pobject)
216 {
217 	struct nvkm_mmu *mmu;
218 	int ret;
219 
220 	ret = nvkm_mmu_create(parent, engine, oclass, "VM", "mmu", &mmu);
221 	*pobject = nv_object(mmu);
222 	if (ret)
223 		return ret;
224 
225 	mmu->limit = 1ULL << 40;
226 	mmu->dma_bits = 40;
227 	mmu->pgt_bits  = 29 - 12;
228 	mmu->spg_shift = 12;
229 	mmu->lpg_shift = 16;
230 	mmu->create = nv50_vm_create;
231 	mmu->map_pgt = nv50_vm_map_pgt;
232 	mmu->map = nv50_vm_map;
233 	mmu->map_sg = nv50_vm_map_sg;
234 	mmu->unmap = nv50_vm_unmap;
235 	mmu->flush = nv50_vm_flush;
236 	return 0;
237 }
238 
239 struct nvkm_oclass
240 nv50_mmu_oclass = {
241 	.handle = NV_SUBDEV(MMU, 0x50),
242 	.ofuncs = &(struct nvkm_ofuncs) {
243 		.ctor = nv50_mmu_ctor,
244 		.dtor = _nvkm_mmu_dtor,
245 		.init = _nvkm_mmu_init,
246 		.fini = _nvkm_mmu_fini,
247 	},
248 };
249