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/gpuobj.h>
27 #include <core/option.h>
28 #include <subdev/timer.h>
29 
30 #define NV44_GART_SIZE (512 * 1024 * 1024)
31 #define NV44_GART_PAGE (  4 * 1024)
32 
33 /*******************************************************************************
34  * VM map/unmap callbacks
35  ******************************************************************************/
36 
37 static void
38 nv44_vm_fill(struct nvkm_memory *pgt, dma_addr_t null,
39 	     dma_addr_t *list, u32 pte, u32 cnt)
40 {
41 	u32 base = (pte << 2) & ~0x0000000f;
42 	u32 tmp[4];
43 
44 	tmp[0] = nvkm_ro32(pgt, base + 0x0);
45 	tmp[1] = nvkm_ro32(pgt, base + 0x4);
46 	tmp[2] = nvkm_ro32(pgt, base + 0x8);
47 	tmp[3] = nvkm_ro32(pgt, base + 0xc);
48 
49 	while (cnt--) {
50 		u32 addr = list ? (*list++ >> 12) : (null >> 12);
51 		switch (pte++ & 0x3) {
52 		case 0:
53 			tmp[0] &= ~0x07ffffff;
54 			tmp[0] |= addr;
55 			break;
56 		case 1:
57 			tmp[0] &= ~0xf8000000;
58 			tmp[0] |= addr << 27;
59 			tmp[1] &= ~0x003fffff;
60 			tmp[1] |= addr >> 5;
61 			break;
62 		case 2:
63 			tmp[1] &= ~0xffc00000;
64 			tmp[1] |= addr << 22;
65 			tmp[2] &= ~0x0001ffff;
66 			tmp[2] |= addr >> 10;
67 			break;
68 		case 3:
69 			tmp[2] &= ~0xfffe0000;
70 			tmp[2] |= addr << 17;
71 			tmp[3] &= ~0x00000fff;
72 			tmp[3] |= addr >> 15;
73 			break;
74 		}
75 	}
76 
77 	nvkm_wo32(pgt, base + 0x0, tmp[0]);
78 	nvkm_wo32(pgt, base + 0x4, tmp[1]);
79 	nvkm_wo32(pgt, base + 0x8, tmp[2]);
80 	nvkm_wo32(pgt, base + 0xc, tmp[3] | 0x40000000);
81 }
82 
83 static void
84 nv44_vm_map_sg(struct nvkm_vma *vma, struct nvkm_memory *pgt,
85 	       struct nvkm_mem *mem, u32 pte, u32 cnt, dma_addr_t *list)
86 {
87 	struct nv04_mmu *mmu = nv04_mmu(vma->vm->mmu);
88 	u32 tmp[4];
89 	int i;
90 
91 	nvkm_kmap(pgt);
92 	if (pte & 3) {
93 		u32  max = 4 - (pte & 3);
94 		u32 part = (cnt > max) ? max : cnt;
95 		nv44_vm_fill(pgt, mmu->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 		nvkm_wo32(pgt, pte++ * 4, tmp[0] >>  0 | tmp[1] << 27);
105 		nvkm_wo32(pgt, pte++ * 4, tmp[1] >>  5 | tmp[2] << 22);
106 		nvkm_wo32(pgt, pte++ * 4, tmp[2] >> 10 | tmp[3] << 17);
107 		nvkm_wo32(pgt, pte++ * 4, tmp[3] >> 15 | 0x40000000);
108 		cnt -= 4;
109 	}
110 
111 	if (cnt)
112 		nv44_vm_fill(pgt, mmu->null, list, pte, cnt);
113 	nvkm_done(pgt);
114 }
115 
116 static void
117 nv44_vm_unmap(struct nvkm_vma *vma, struct nvkm_memory *pgt, u32 pte, u32 cnt)
118 {
119 	struct nv04_mmu *mmu = nv04_mmu(vma->vm->mmu);
120 
121 	nvkm_kmap(pgt);
122 	if (pte & 3) {
123 		u32  max = 4 - (pte & 3);
124 		u32 part = (cnt > max) ? max : cnt;
125 		nv44_vm_fill(pgt, mmu->null, NULL, pte, part);
126 		pte  += part;
127 		cnt  -= part;
128 	}
129 
130 	while (cnt >= 4) {
131 		nvkm_wo32(pgt, pte++ * 4, 0x00000000);
132 		nvkm_wo32(pgt, pte++ * 4, 0x00000000);
133 		nvkm_wo32(pgt, pte++ * 4, 0x00000000);
134 		nvkm_wo32(pgt, pte++ * 4, 0x00000000);
135 		cnt -= 4;
136 	}
137 
138 	if (cnt)
139 		nv44_vm_fill(pgt, mmu->null, NULL, pte, cnt);
140 	nvkm_done(pgt);
141 }
142 
143 static void
144 nv44_vm_flush(struct nvkm_vm *vm)
145 {
146 	struct nv04_mmu *mmu = nv04_mmu(vm->mmu);
147 	struct nvkm_device *device = mmu->base.subdev.device;
148 	nvkm_wr32(device, 0x100814, mmu->base.limit - NV44_GART_PAGE);
149 	nvkm_wr32(device, 0x100808, 0x00000020);
150 	nvkm_msec(device, 2000,
151 		if (nvkm_rd32(device, 0x100808) & 0x00000001)
152 			break;
153 	);
154 	nvkm_wr32(device, 0x100808, 0x00000000);
155 }
156 
157 /*******************************************************************************
158  * MMU subdev
159  ******************************************************************************/
160 
161 static int
162 nv44_mmu_ctor(struct nvkm_object *parent, struct nvkm_object *engine,
163 	      struct nvkm_oclass *oclass, void *data, u32 size,
164 	      struct nvkm_object **pobject)
165 {
166 	struct nvkm_device *device = nv_device(parent);
167 	struct nv04_mmu *mmu;
168 	int ret;
169 
170 	if (pci_find_capability(device->pdev, PCI_CAP_ID_AGP) ||
171 	    !nvkm_boolopt(device->cfgopt, "NvPCIE", true)) {
172 		return nvkm_object_old(parent, engine, &nv04_mmu_oclass,
173 					data, size, pobject);
174 	}
175 
176 	ret = nvkm_mmu_create(parent, engine, oclass, "PCIEGART",
177 			      "mmu", &mmu);
178 	*pobject = nv_object(mmu);
179 	if (ret)
180 		return ret;
181 
182 	mmu->base.create = nv04_vm_create;
183 	mmu->base.limit = NV44_GART_SIZE;
184 	mmu->base.dma_bits = 39;
185 	mmu->base.pgt_bits = 32 - 12;
186 	mmu->base.spg_shift = 12;
187 	mmu->base.lpg_shift = 12;
188 	mmu->base.map_sg = nv44_vm_map_sg;
189 	mmu->base.unmap = nv44_vm_unmap;
190 	mmu->base.flush = nv44_vm_flush;
191 
192 	mmu->nullp = pci_alloc_consistent(device->pdev, 16 * 1024, &mmu->null);
193 	if (!mmu->nullp) {
194 		nvkm_warn(&mmu->base.subdev, "unable to allocate dummy pages\n");
195 		mmu->null = 0;
196 	}
197 
198 	ret = nvkm_vm_create(&mmu->base, 0, NV44_GART_SIZE, 0, 4096, NULL,
199 			     &mmu->vm);
200 	if (ret)
201 		return ret;
202 
203 	ret = nvkm_memory_new(device, NVKM_MEM_TARGET_INST,
204 			      (NV44_GART_SIZE / NV44_GART_PAGE) * 4,
205 			      512 * 1024, true,
206 			      &mmu->vm->pgt[0].mem[0]);
207 	mmu->vm->pgt[0].refcount[0] = 1;
208 	if (ret)
209 		return ret;
210 
211 	return 0;
212 }
213 
214 static int
215 nv44_mmu_init(struct nvkm_object *object)
216 {
217 	struct nv04_mmu *mmu = (void *)object;
218 	struct nvkm_device *device = mmu->base.subdev.device;
219 	struct nvkm_memory *gart = mmu->vm->pgt[0].mem[0];
220 	u32 addr;
221 	int ret;
222 
223 	ret = nvkm_mmu_init(&mmu->base);
224 	if (ret)
225 		return ret;
226 
227 	/* calculate vram address of this PRAMIN block, object must be
228 	 * allocated on 512KiB alignment, and not exceed a total size
229 	 * of 512KiB for this to work correctly
230 	 */
231 	addr  = nvkm_rd32(device, 0x10020c);
232 	addr -= ((nvkm_memory_addr(gart) >> 19) + 1) << 19;
233 
234 	nvkm_wr32(device, 0x100850, 0x80000000);
235 	nvkm_wr32(device, 0x100818, mmu->null);
236 	nvkm_wr32(device, 0x100804, NV44_GART_SIZE);
237 	nvkm_wr32(device, 0x100850, 0x00008000);
238 	nvkm_mask(device, 0x10008c, 0x00000200, 0x00000200);
239 	nvkm_wr32(device, 0x100820, 0x00000000);
240 	nvkm_wr32(device, 0x10082c, 0x00000001);
241 	nvkm_wr32(device, 0x100800, addr | 0x00000010);
242 	return 0;
243 }
244 
245 struct nvkm_oclass
246 nv44_mmu_oclass = {
247 	.handle = NV_SUBDEV(MMU, 0x44),
248 	.ofuncs = &(struct nvkm_ofuncs) {
249 		.ctor = nv44_mmu_ctor,
250 		.dtor = nv04_mmu_dtor,
251 		.init = nv44_mmu_init,
252 		.fini = _nvkm_mmu_fini,
253 	},
254 };
255