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 #define nv50_instmem(p) container_of((p), struct nv50_instmem, base)
25 #include "priv.h"
26 
27 #include <core/memory.h>
28 #include <subdev/bar.h>
29 #include <subdev/fb.h>
30 #include <subdev/mmu.h>
31 
32 struct nv50_instmem {
33 	struct nvkm_instmem base;
34 	u64 addr;
35 
36 	/* Mappings that can be evicted when BAR2 space has been exhausted. */
37 	struct list_head lru;
38 };
39 
40 /******************************************************************************
41  * instmem object implementation
42  *****************************************************************************/
43 #define nv50_instobj(p) container_of((p), struct nv50_instobj, base.memory)
44 
45 struct nv50_instobj {
46 	struct nvkm_instobj base;
47 	struct nv50_instmem *imem;
48 	struct nvkm_memory *ram;
49 	struct nvkm_vma *bar;
50 	refcount_t maps;
51 	void *map;
52 	struct list_head lru;
53 };
54 
55 static void
56 nv50_instobj_wr32_slow(struct nvkm_memory *memory, u64 offset, u32 data)
57 {
58 	struct nv50_instobj *iobj = nv50_instobj(memory);
59 	struct nv50_instmem *imem = iobj->imem;
60 	struct nvkm_device *device = imem->base.subdev.device;
61 	u64 base = (nvkm_memory_addr(iobj->ram) + offset) & 0xffffff00000ULL;
62 	u64 addr = (nvkm_memory_addr(iobj->ram) + offset) & 0x000000fffffULL;
63 	unsigned long flags;
64 
65 	spin_lock_irqsave(&imem->base.lock, flags);
66 	if (unlikely(imem->addr != base)) {
67 		nvkm_wr32(device, 0x001700, base >> 16);
68 		imem->addr = base;
69 	}
70 	nvkm_wr32(device, 0x700000 + addr, data);
71 	spin_unlock_irqrestore(&imem->base.lock, flags);
72 }
73 
74 static u32
75 nv50_instobj_rd32_slow(struct nvkm_memory *memory, u64 offset)
76 {
77 	struct nv50_instobj *iobj = nv50_instobj(memory);
78 	struct nv50_instmem *imem = iobj->imem;
79 	struct nvkm_device *device = imem->base.subdev.device;
80 	u64 base = (nvkm_memory_addr(iobj->ram) + offset) & 0xffffff00000ULL;
81 	u64 addr = (nvkm_memory_addr(iobj->ram) + offset) & 0x000000fffffULL;
82 	u32 data;
83 	unsigned long flags;
84 
85 	spin_lock_irqsave(&imem->base.lock, flags);
86 	if (unlikely(imem->addr != base)) {
87 		nvkm_wr32(device, 0x001700, base >> 16);
88 		imem->addr = base;
89 	}
90 	data = nvkm_rd32(device, 0x700000 + addr);
91 	spin_unlock_irqrestore(&imem->base.lock, flags);
92 	return data;
93 }
94 
95 static const struct nvkm_memory_ptrs
96 nv50_instobj_slow = {
97 	.rd32 = nv50_instobj_rd32_slow,
98 	.wr32 = nv50_instobj_wr32_slow,
99 };
100 
101 static void
102 nv50_instobj_wr32(struct nvkm_memory *memory, u64 offset, u32 data)
103 {
104 	iowrite32_native(data, nv50_instobj(memory)->map + offset);
105 }
106 
107 static u32
108 nv50_instobj_rd32(struct nvkm_memory *memory, u64 offset)
109 {
110 	return ioread32_native(nv50_instobj(memory)->map + offset);
111 }
112 
113 static const struct nvkm_memory_ptrs
114 nv50_instobj_fast = {
115 	.rd32 = nv50_instobj_rd32,
116 	.wr32 = nv50_instobj_wr32,
117 };
118 
119 static void
120 nv50_instobj_kmap(struct nv50_instobj *iobj, struct nvkm_vmm *vmm)
121 {
122 	struct nv50_instmem *imem = iobj->imem;
123 	struct nv50_instobj *eobj;
124 	struct nvkm_memory *memory = &iobj->base.memory;
125 	struct nvkm_subdev *subdev = &imem->base.subdev;
126 	struct nvkm_device *device = subdev->device;
127 	struct nvkm_vma *bar = NULL, *ebar;
128 	u64 size = nvkm_memory_size(memory);
129 	void *emap;
130 	int ret;
131 
132 	/* Attempt to allocate BAR2 address-space and map the object
133 	 * into it.  The lock has to be dropped while doing this due
134 	 * to the possibility of recursion for page table allocation.
135 	 */
136 	mutex_unlock(&subdev->mutex);
137 	while ((ret = nvkm_vmm_get(vmm, 12, size, &bar))) {
138 		/* Evict unused mappings, and keep retrying until we either
139 		 * succeed,or there's no more objects left on the LRU.
140 		 */
141 		mutex_lock(&subdev->mutex);
142 		eobj = list_first_entry_or_null(&imem->lru, typeof(*eobj), lru);
143 		if (eobj) {
144 			nvkm_debug(subdev, "evict %016llx %016llx @ %016llx\n",
145 				   nvkm_memory_addr(&eobj->base.memory),
146 				   nvkm_memory_size(&eobj->base.memory),
147 				   eobj->bar->addr);
148 			list_del_init(&eobj->lru);
149 			ebar = eobj->bar;
150 			eobj->bar = NULL;
151 			emap = eobj->map;
152 			eobj->map = NULL;
153 		}
154 		mutex_unlock(&subdev->mutex);
155 		if (!eobj)
156 			break;
157 		iounmap(emap);
158 		nvkm_vmm_put(vmm, &ebar);
159 	}
160 
161 	if (ret == 0)
162 		ret = nvkm_memory_map(memory, 0, vmm, bar, NULL, 0);
163 	mutex_lock(&subdev->mutex);
164 	if (ret || iobj->bar) {
165 		/* We either failed, or another thread beat us. */
166 		mutex_unlock(&subdev->mutex);
167 		nvkm_vmm_put(vmm, &bar);
168 		mutex_lock(&subdev->mutex);
169 		return;
170 	}
171 
172 	/* Make the mapping visible to the host. */
173 	iobj->bar = bar;
174 	iobj->map = ioremap_wc(device->func->resource_addr(device, 3) +
175 			       (u32)iobj->bar->addr, size);
176 	if (!iobj->map) {
177 		nvkm_warn(subdev, "PRAMIN ioremap failed\n");
178 		nvkm_vmm_put(vmm, &iobj->bar);
179 	}
180 }
181 
182 static int
183 nv50_instobj_map(struct nvkm_memory *memory, u64 offset, struct nvkm_vmm *vmm,
184 		 struct nvkm_vma *vma, void *argv, u32 argc)
185 {
186 	memory = nv50_instobj(memory)->ram;
187 	return nvkm_memory_map(memory, offset, vmm, vma, argv, argc);
188 }
189 
190 static void
191 nv50_instobj_release(struct nvkm_memory *memory)
192 {
193 	struct nv50_instobj *iobj = nv50_instobj(memory);
194 	struct nv50_instmem *imem = iobj->imem;
195 	struct nvkm_subdev *subdev = &imem->base.subdev;
196 
197 	wmb();
198 	nvkm_bar_flush(subdev->device->bar);
199 
200 	if (refcount_dec_and_mutex_lock(&iobj->maps, &subdev->mutex)) {
201 		/* Add the now-unused mapping to the LRU instead of directly
202 		 * unmapping it here, in case we need to map it again later.
203 		 */
204 		if (likely(iobj->lru.next) && iobj->map) {
205 			BUG_ON(!list_empty(&iobj->lru));
206 			list_add_tail(&iobj->lru, &imem->lru);
207 		}
208 
209 		/* Switch back to NULL accessors when last map is gone. */
210 		iobj->base.memory.ptrs = NULL;
211 		mutex_unlock(&subdev->mutex);
212 	}
213 }
214 
215 static void __iomem *
216 nv50_instobj_acquire(struct nvkm_memory *memory)
217 {
218 	struct nv50_instobj *iobj = nv50_instobj(memory);
219 	struct nvkm_instmem *imem = &iobj->imem->base;
220 	struct nvkm_vmm *vmm;
221 	void __iomem *map = NULL;
222 
223 	/* Already mapped? */
224 	if (refcount_inc_not_zero(&iobj->maps))
225 		return iobj->map;
226 
227 	/* Take the lock, and re-check that another thread hasn't
228 	 * already mapped the object in the meantime.
229 	 */
230 	mutex_lock(&imem->subdev.mutex);
231 	if (refcount_inc_not_zero(&iobj->maps)) {
232 		mutex_unlock(&imem->subdev.mutex);
233 		return iobj->map;
234 	}
235 
236 	/* Attempt to get a direct CPU mapping of the object. */
237 	if ((vmm = nvkm_bar_bar2_vmm(imem->subdev.device))) {
238 		if (!iobj->map)
239 			nv50_instobj_kmap(iobj, vmm);
240 		map = iobj->map;
241 	}
242 
243 	if (!refcount_inc_not_zero(&iobj->maps)) {
244 		/* Exclude object from eviction while it's being accessed. */
245 		if (likely(iobj->lru.next))
246 			list_del_init(&iobj->lru);
247 
248 		if (map)
249 			iobj->base.memory.ptrs = &nv50_instobj_fast;
250 		else
251 			iobj->base.memory.ptrs = &nv50_instobj_slow;
252 		refcount_set(&iobj->maps, 1);
253 	}
254 
255 	mutex_unlock(&imem->subdev.mutex);
256 	return map;
257 }
258 
259 static void
260 nv50_instobj_boot(struct nvkm_memory *memory, struct nvkm_vmm *vmm)
261 {
262 	struct nv50_instobj *iobj = nv50_instobj(memory);
263 	struct nvkm_instmem *imem = &iobj->imem->base;
264 
265 	/* Exclude bootstrapped objects (ie. the page tables for the
266 	 * instmem BAR itself) from eviction.
267 	 */
268 	mutex_lock(&imem->subdev.mutex);
269 	if (likely(iobj->lru.next)) {
270 		list_del_init(&iobj->lru);
271 		iobj->lru.next = NULL;
272 	}
273 
274 	nv50_instobj_kmap(iobj, vmm);
275 	nvkm_instmem_boot(imem);
276 	mutex_unlock(&imem->subdev.mutex);
277 }
278 
279 static u64
280 nv50_instobj_size(struct nvkm_memory *memory)
281 {
282 	return nvkm_memory_size(nv50_instobj(memory)->ram);
283 }
284 
285 static u64
286 nv50_instobj_addr(struct nvkm_memory *memory)
287 {
288 	return nvkm_memory_addr(nv50_instobj(memory)->ram);
289 }
290 
291 static enum nvkm_memory_target
292 nv50_instobj_target(struct nvkm_memory *memory)
293 {
294 	return nvkm_memory_target(nv50_instobj(memory)->ram);
295 }
296 
297 static void *
298 nv50_instobj_dtor(struct nvkm_memory *memory)
299 {
300 	struct nv50_instobj *iobj = nv50_instobj(memory);
301 	struct nvkm_instmem *imem = &iobj->imem->base;
302 	struct nvkm_vma *bar;
303 	void *map = map;
304 
305 	mutex_lock(&imem->subdev.mutex);
306 	if (likely(iobj->lru.next))
307 		list_del(&iobj->lru);
308 	map = iobj->map;
309 	bar = iobj->bar;
310 	mutex_unlock(&imem->subdev.mutex);
311 
312 	if (map) {
313 		struct nvkm_vmm *vmm = nvkm_bar_bar2_vmm(imem->subdev.device);
314 		iounmap(map);
315 		if (likely(vmm)) /* Can be NULL during BAR destructor. */
316 			nvkm_vmm_put(vmm, &bar);
317 	}
318 
319 	nvkm_memory_unref(&iobj->ram);
320 	nvkm_instobj_dtor(imem, &iobj->base);
321 	return iobj;
322 }
323 
324 static const struct nvkm_memory_func
325 nv50_instobj_func = {
326 	.dtor = nv50_instobj_dtor,
327 	.target = nv50_instobj_target,
328 	.size = nv50_instobj_size,
329 	.addr = nv50_instobj_addr,
330 	.boot = nv50_instobj_boot,
331 	.acquire = nv50_instobj_acquire,
332 	.release = nv50_instobj_release,
333 	.map = nv50_instobj_map,
334 };
335 
336 static int
337 nv50_instobj_new(struct nvkm_instmem *base, u32 size, u32 align, bool zero,
338 		 struct nvkm_memory **pmemory)
339 {
340 	struct nv50_instmem *imem = nv50_instmem(base);
341 	struct nv50_instobj *iobj;
342 	struct nvkm_device *device = imem->base.subdev.device;
343 	u8 page = max(order_base_2(align), 12);
344 
345 	if (!(iobj = kzalloc(sizeof(*iobj), GFP_KERNEL)))
346 		return -ENOMEM;
347 	*pmemory = &iobj->base.memory;
348 
349 	nvkm_instobj_ctor(&nv50_instobj_func, &imem->base, &iobj->base);
350 	iobj->imem = imem;
351 	refcount_set(&iobj->maps, 0);
352 	INIT_LIST_HEAD(&iobj->lru);
353 
354 	return nvkm_ram_get(device, 0, 1, page, size, true, true, &iobj->ram);
355 }
356 
357 /******************************************************************************
358  * instmem subdev implementation
359  *****************************************************************************/
360 
361 static void
362 nv50_instmem_fini(struct nvkm_instmem *base)
363 {
364 	nv50_instmem(base)->addr = ~0ULL;
365 }
366 
367 static const struct nvkm_instmem_func
368 nv50_instmem = {
369 	.fini = nv50_instmem_fini,
370 	.memory_new = nv50_instobj_new,
371 	.zero = false,
372 };
373 
374 int
375 nv50_instmem_new(struct nvkm_device *device, int index,
376 		 struct nvkm_instmem **pimem)
377 {
378 	struct nv50_instmem *imem;
379 
380 	if (!(imem = kzalloc(sizeof(*imem), GFP_KERNEL)))
381 		return -ENOMEM;
382 	nvkm_instmem_ctor(&nv50_instmem, device, index, &imem->base);
383 	INIT_LIST_HEAD(&imem->lru);
384 	*pimem = &imem->base;
385 	return 0;
386 }
387