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 static void
33 nv50_vm_map_pgt(struct nvkm_gpuobj *pgd, u32 pde, struct nvkm_gpuobj *pgt[2])
34 {
35 	u64 phys = 0xdeadcafe00000000ULL;
36 	u32 coverage = 0;
37 
38 	if (pgt[0]) {
39 		phys = 0x00000003 | pgt[0]->addr; /* present, 4KiB pages */
40 		coverage = (pgt[0]->size >> 3) << 12;
41 	} else
42 	if (pgt[1]) {
43 		phys = 0x00000001 | pgt[1]->addr; /* present */
44 		coverage = (pgt[1]->size >> 3) << 16;
45 	}
46 
47 	if (phys & 1) {
48 		if (coverage <= 32 * 1024 * 1024)
49 			phys |= 0x60;
50 		else if (coverage <= 64 * 1024 * 1024)
51 			phys |= 0x40;
52 		else if (coverage <= 128 * 1024 * 1024)
53 			phys |= 0x20;
54 	}
55 
56 	nvkm_kmap(pgd);
57 	nvkm_wo32(pgd, (pde * 8) + 0, lower_32_bits(phys));
58 	nvkm_wo32(pgd, (pde * 8) + 4, upper_32_bits(phys));
59 	nvkm_done(pgd);
60 }
61 
62 static inline u64
63 vm_addr(struct nvkm_vma *vma, u64 phys, u32 memtype, u32 target)
64 {
65 	phys |= 1; /* present */
66 	phys |= (u64)memtype << 40;
67 	phys |= target << 4;
68 	if (vma->access & NV_MEM_ACCESS_SYS)
69 		phys |= (1 << 6);
70 	if (!(vma->access & NV_MEM_ACCESS_WO))
71 		phys |= (1 << 3);
72 	return phys;
73 }
74 
75 static void
76 nv50_vm_map(struct nvkm_vma *vma, struct nvkm_gpuobj *pgt,
77 	    struct nvkm_mem *mem, u32 pte, u32 cnt, u64 phys, u64 delta)
78 {
79 	u32 comp = (mem->memtype & 0x180) >> 7;
80 	u32 block, target;
81 	int i;
82 
83 	/* IGPs don't have real VRAM, re-target to stolen system memory */
84 	target = 0;
85 	if (nvkm_fb(vma->vm->mmu)->ram->stolen) {
86 		phys += nvkm_fb(vma->vm->mmu)->ram->stolen;
87 		target = 3;
88 	}
89 
90 	phys  = vm_addr(vma, phys, mem->memtype, target);
91 	pte <<= 3;
92 	cnt <<= 3;
93 
94 	nvkm_kmap(pgt);
95 	while (cnt) {
96 		u32 offset_h = upper_32_bits(phys);
97 		u32 offset_l = lower_32_bits(phys);
98 
99 		for (i = 7; i >= 0; i--) {
100 			block = 1 << (i + 3);
101 			if (cnt >= block && !(pte & (block - 1)))
102 				break;
103 		}
104 		offset_l |= (i << 7);
105 
106 		phys += block << (vma->node->type - 3);
107 		cnt  -= block;
108 		if (comp) {
109 			u32 tag = mem->tag->offset + ((delta >> 16) * comp);
110 			offset_h |= (tag << 17);
111 			delta    += block << (vma->node->type - 3);
112 		}
113 
114 		while (block) {
115 			nvkm_wo32(pgt, pte + 0, offset_l);
116 			nvkm_wo32(pgt, pte + 4, offset_h);
117 			pte += 8;
118 			block -= 8;
119 		}
120 	}
121 	nvkm_done(pgt);
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 	nvkm_kmap(pgt);
131 	while (cnt--) {
132 		u64 phys = vm_addr(vma, (u64)*list++, mem->memtype, target);
133 		nvkm_wo32(pgt, pte + 0, lower_32_bits(phys));
134 		nvkm_wo32(pgt, pte + 4, upper_32_bits(phys));
135 		pte += 8;
136 	}
137 	nvkm_done(pgt);
138 }
139 
140 static void
141 nv50_vm_unmap(struct nvkm_gpuobj *pgt, u32 pte, u32 cnt)
142 {
143 	pte <<= 3;
144 	nvkm_kmap(pgt);
145 	while (cnt--) {
146 		nvkm_wo32(pgt, pte + 0, 0x00000000);
147 		nvkm_wo32(pgt, pte + 4, 0x00000000);
148 		pte += 8;
149 	}
150 	nvkm_done(pgt);
151 }
152 
153 static void
154 nv50_vm_flush(struct nvkm_vm *vm)
155 {
156 	struct nvkm_mmu *mmu = (void *)vm->mmu;
157 	struct nvkm_subdev *subdev = &mmu->subdev;
158 	struct nvkm_device *device = subdev->device;
159 	struct nvkm_bar *bar = device->bar;
160 	struct nvkm_engine *engine;
161 	int i, vme;
162 
163 	bar->flush(bar);
164 
165 	mutex_lock(&subdev->mutex);
166 	for (i = 0; i < NVDEV_SUBDEV_NR; i++) {
167 		if (!atomic_read(&vm->engref[i]))
168 			continue;
169 
170 		/* unfortunate hw bug workaround... */
171 		engine = nvkm_engine(mmu, i);
172 		if (engine && engine->tlb_flush) {
173 			engine->tlb_flush(engine);
174 			continue;
175 		}
176 
177 		switch (i) {
178 		case NVDEV_ENGINE_GR    : vme = 0x00; break;
179 		case NVDEV_ENGINE_VP    :
180 		case NVDEV_ENGINE_MSPDEC: vme = 0x01; break;
181 		case NVDEV_SUBDEV_BAR   : vme = 0x06; break;
182 		case NVDEV_ENGINE_MSPPP :
183 		case NVDEV_ENGINE_MPEG  : vme = 0x08; break;
184 		case NVDEV_ENGINE_BSP   :
185 		case NVDEV_ENGINE_MSVLD : vme = 0x09; break;
186 		case NVDEV_ENGINE_CIPHER:
187 		case NVDEV_ENGINE_SEC   : vme = 0x0a; break;
188 		case NVDEV_ENGINE_CE0   : vme = 0x0d; break;
189 		default:
190 			continue;
191 		}
192 
193 		nvkm_wr32(device, 0x100c80, (vme << 16) | 1);
194 		if (nvkm_msec(device, 2000,
195 			if (!(nvkm_rd32(device, 0x100c80) & 0x00000001))
196 				break;
197 		) < 0)
198 			nvkm_error(subdev, "vm flush timeout: engine %d\n", vme);
199 	}
200 	mutex_unlock(&subdev->mutex);
201 }
202 
203 static int
204 nv50_vm_create(struct nvkm_mmu *mmu, u64 offset, u64 length,
205 	       u64 mm_offset, struct nvkm_vm **pvm)
206 {
207 	u32 block = (1 << (mmu->pgt_bits + 12));
208 	if (block > length)
209 		block = length;
210 
211 	return nvkm_vm_create(mmu, offset, length, mm_offset, block, pvm);
212 }
213 
214 static int
215 nv50_mmu_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
216 	      struct nvkm_oclass *oclass, void *data, u32 size,
217 	      struct nvkm_object **pobject)
218 {
219 	struct nvkm_mmu *mmu;
220 	int ret;
221 
222 	ret = nvkm_mmu_create(parent, engine, oclass, "VM", "mmu", &mmu);
223 	*pobject = nv_object(mmu);
224 	if (ret)
225 		return ret;
226 
227 	mmu->limit = 1ULL << 40;
228 	mmu->dma_bits = 40;
229 	mmu->pgt_bits  = 29 - 12;
230 	mmu->spg_shift = 12;
231 	mmu->lpg_shift = 16;
232 	mmu->create = nv50_vm_create;
233 	mmu->map_pgt = nv50_vm_map_pgt;
234 	mmu->map = nv50_vm_map;
235 	mmu->map_sg = nv50_vm_map_sg;
236 	mmu->unmap = nv50_vm_unmap;
237 	mmu->flush = nv50_vm_flush;
238 	return 0;
239 }
240 
241 struct nvkm_oclass
242 nv50_mmu_oclass = {
243 	.handle = NV_SUBDEV(MMU, 0x50),
244 	.ofuncs = &(struct nvkm_ofuncs) {
245 		.ctor = nv50_mmu_ctor,
246 		.dtor = _nvkm_mmu_dtor,
247 		.init = _nvkm_mmu_init,
248 		.fini = _nvkm_mmu_fini,
249 	},
250 };
251