1 /*
2  * spu management operations for of based platforms
3  *
4  * (C) Copyright IBM Deutschland Entwicklung GmbH 2005
5  * Copyright 2006 Sony Corp.
6  * (C) Copyright 2007 TOSHIBA CORPORATION
7  *
8  * This program is free software; you can redistribute it and/or modify
9  * it under the terms of the GNU General Public License as published by
10  * the Free Software Foundation; version 2 of the License.
11  *
12  * This program is distributed in the hope that it will be useful,
13  * but WITHOUT ANY WARRANTY; without even the implied warranty of
14  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
15  * GNU General Public License for more details.
16  *
17  * You should have received a copy of the GNU General Public License along
18  * with this program; if not, write to the Free Software Foundation, Inc.,
19  * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA.
20  */
21 
22 #include <linux/interrupt.h>
23 #include <linux/list.h>
24 #include <linux/export.h>
25 #include <linux/ptrace.h>
26 #include <linux/wait.h>
27 #include <linux/mm.h>
28 #include <linux/io.h>
29 #include <linux/mutex.h>
30 #include <linux/device.h>
31 
32 #include <asm/spu.h>
33 #include <asm/spu_priv1.h>
34 #include <asm/firmware.h>
35 #include <asm/prom.h>
36 
37 #include "spufs/spufs.h"
38 #include "interrupt.h"
39 
40 struct device_node *spu_devnode(struct spu *spu)
41 {
42 	return spu->devnode;
43 }
44 
45 EXPORT_SYMBOL_GPL(spu_devnode);
46 
47 static u64 __init find_spu_unit_number(struct device_node *spe)
48 {
49 	const unsigned int *prop;
50 	int proplen;
51 
52 	/* new device trees should provide the physical-id attribute */
53 	prop = of_get_property(spe, "physical-id", &proplen);
54 	if (proplen == 4)
55 		return (u64)*prop;
56 
57 	/* celleb device tree provides the unit-id */
58 	prop = of_get_property(spe, "unit-id", &proplen);
59 	if (proplen == 4)
60 		return (u64)*prop;
61 
62 	/* legacy device trees provide the id in the reg attribute */
63 	prop = of_get_property(spe, "reg", &proplen);
64 	if (proplen == 4)
65 		return (u64)*prop;
66 
67 	return 0;
68 }
69 
70 static void spu_unmap(struct spu *spu)
71 {
72 	if (!firmware_has_feature(FW_FEATURE_LPAR))
73 		iounmap(spu->priv1);
74 	iounmap(spu->priv2);
75 	iounmap(spu->problem);
76 	iounmap((__force u8 __iomem *)spu->local_store);
77 }
78 
79 static int __init spu_map_interrupts_old(struct spu *spu,
80 	struct device_node *np)
81 {
82 	unsigned int isrc;
83 	const u32 *tmp;
84 	int nid;
85 
86 	/* Get the interrupt source unit from the device-tree */
87 	tmp = of_get_property(np, "isrc", NULL);
88 	if (!tmp)
89 		return -ENODEV;
90 	isrc = tmp[0];
91 
92 	tmp = of_get_property(np->parent->parent, "node-id", NULL);
93 	if (!tmp) {
94 		printk(KERN_WARNING "%s: can't find node-id\n", __func__);
95 		nid = spu->node;
96 	} else
97 		nid = tmp[0];
98 
99 	/* Add the node number */
100 	isrc |= nid << IIC_IRQ_NODE_SHIFT;
101 
102 	/* Now map interrupts of all 3 classes */
103 	spu->irqs[0] = irq_create_mapping(NULL, IIC_IRQ_CLASS_0 | isrc);
104 	spu->irqs[1] = irq_create_mapping(NULL, IIC_IRQ_CLASS_1 | isrc);
105 	spu->irqs[2] = irq_create_mapping(NULL, IIC_IRQ_CLASS_2 | isrc);
106 
107 	/* Right now, we only fail if class 2 failed */
108 	if (!spu->irqs[2])
109 		return -EINVAL;
110 
111 	return 0;
112 }
113 
114 static void __iomem * __init spu_map_prop_old(struct spu *spu,
115 					      struct device_node *n,
116 					      const char *name)
117 {
118 	const struct address_prop {
119 		unsigned long address;
120 		unsigned int len;
121 	} __attribute__((packed)) *prop;
122 	int proplen;
123 
124 	prop = of_get_property(n, name, &proplen);
125 	if (prop == NULL || proplen != sizeof (struct address_prop))
126 		return NULL;
127 
128 	return ioremap(prop->address, prop->len);
129 }
130 
131 static int __init spu_map_device_old(struct spu *spu)
132 {
133 	struct device_node *node = spu->devnode;
134 	const char *prop;
135 	int ret;
136 
137 	ret = -ENODEV;
138 	spu->name = of_get_property(node, "name", NULL);
139 	if (!spu->name)
140 		goto out;
141 
142 	prop = of_get_property(node, "local-store", NULL);
143 	if (!prop)
144 		goto out;
145 	spu->local_store_phys = *(unsigned long *)prop;
146 
147 	/* we use local store as ram, not io memory */
148 	spu->local_store = (void __force *)
149 		spu_map_prop_old(spu, node, "local-store");
150 	if (!spu->local_store)
151 		goto out;
152 
153 	prop = of_get_property(node, "problem", NULL);
154 	if (!prop)
155 		goto out_unmap;
156 	spu->problem_phys = *(unsigned long *)prop;
157 
158 	spu->problem = spu_map_prop_old(spu, node, "problem");
159 	if (!spu->problem)
160 		goto out_unmap;
161 
162 	spu->priv2 = spu_map_prop_old(spu, node, "priv2");
163 	if (!spu->priv2)
164 		goto out_unmap;
165 
166 	if (!firmware_has_feature(FW_FEATURE_LPAR)) {
167 		spu->priv1 = spu_map_prop_old(spu, node, "priv1");
168 		if (!spu->priv1)
169 			goto out_unmap;
170 	}
171 
172 	ret = 0;
173 	goto out;
174 
175 out_unmap:
176 	spu_unmap(spu);
177 out:
178 	return ret;
179 }
180 
181 static int __init spu_map_interrupts(struct spu *spu, struct device_node *np)
182 {
183 	struct of_phandle_args oirq;
184 	int ret;
185 	int i;
186 
187 	for (i=0; i < 3; i++) {
188 		ret = of_irq_parse_one(np, i, &oirq);
189 		if (ret) {
190 			pr_debug("spu_new: failed to get irq %d\n", i);
191 			goto err;
192 		}
193 		ret = -EINVAL;
194 		pr_debug("  irq %d no 0x%x on %pOF\n", i, oirq.args[0],
195 			 oirq.np);
196 		spu->irqs[i] = irq_create_of_mapping(&oirq);
197 		if (!spu->irqs[i]) {
198 			pr_debug("spu_new: failed to map it !\n");
199 			goto err;
200 		}
201 	}
202 	return 0;
203 
204 err:
205 	pr_debug("failed to map irq %x for spu %s\n", *oirq.args,
206 		spu->name);
207 	for (; i >= 0; i--) {
208 		if (spu->irqs[i])
209 			irq_dispose_mapping(spu->irqs[i]);
210 	}
211 	return ret;
212 }
213 
214 static int spu_map_resource(struct spu *spu, int nr,
215 			    void __iomem** virt, unsigned long *phys)
216 {
217 	struct device_node *np = spu->devnode;
218 	struct resource resource = { };
219 	unsigned long len;
220 	int ret;
221 
222 	ret = of_address_to_resource(np, nr, &resource);
223 	if (ret)
224 		return ret;
225 	if (phys)
226 		*phys = resource.start;
227 	len = resource_size(&resource);
228 	*virt = ioremap(resource.start, len);
229 	if (!*virt)
230 		return -EINVAL;
231 	return 0;
232 }
233 
234 static int __init spu_map_device(struct spu *spu)
235 {
236 	struct device_node *np = spu->devnode;
237 	int ret = -ENODEV;
238 
239 	spu->name = of_get_property(np, "name", NULL);
240 	if (!spu->name)
241 		goto out;
242 
243 	ret = spu_map_resource(spu, 0, (void __iomem**)&spu->local_store,
244 			       &spu->local_store_phys);
245 	if (ret) {
246 		pr_debug("spu_new: failed to map %pOF resource 0\n",
247 			 np);
248 		goto out;
249 	}
250 	ret = spu_map_resource(spu, 1, (void __iomem**)&spu->problem,
251 			       &spu->problem_phys);
252 	if (ret) {
253 		pr_debug("spu_new: failed to map %pOF resource 1\n",
254 			 np);
255 		goto out_unmap;
256 	}
257 	ret = spu_map_resource(spu, 2, (void __iomem**)&spu->priv2, NULL);
258 	if (ret) {
259 		pr_debug("spu_new: failed to map %pOF resource 2\n",
260 			 np);
261 		goto out_unmap;
262 	}
263 	if (!firmware_has_feature(FW_FEATURE_LPAR))
264 		ret = spu_map_resource(spu, 3,
265 			       (void __iomem**)&spu->priv1, NULL);
266 	if (ret) {
267 		pr_debug("spu_new: failed to map %pOF resource 3\n",
268 			 np);
269 		goto out_unmap;
270 	}
271 	pr_debug("spu_new: %pOF maps:\n", np);
272 	pr_debug("  local store   : 0x%016lx -> 0x%p\n",
273 		 spu->local_store_phys, spu->local_store);
274 	pr_debug("  problem state : 0x%016lx -> 0x%p\n",
275 		 spu->problem_phys, spu->problem);
276 	pr_debug("  priv2         :                       0x%p\n", spu->priv2);
277 	pr_debug("  priv1         :                       0x%p\n", spu->priv1);
278 
279 	return 0;
280 
281 out_unmap:
282 	spu_unmap(spu);
283 out:
284 	pr_debug("failed to map spe %s: %d\n", spu->name, ret);
285 	return ret;
286 }
287 
288 static int __init of_enumerate_spus(int (*fn)(void *data))
289 {
290 	int ret;
291 	struct device_node *node;
292 	unsigned int n = 0;
293 
294 	ret = -ENODEV;
295 	for_each_node_by_type(node, "spe") {
296 		ret = fn(node);
297 		if (ret) {
298 			printk(KERN_WARNING "%s: Error initializing %s\n",
299 				__func__, node->name);
300 			of_node_put(node);
301 			break;
302 		}
303 		n++;
304 	}
305 	return ret ? ret : n;
306 }
307 
308 static int __init of_create_spu(struct spu *spu, void *data)
309 {
310 	int ret;
311 	struct device_node *spe = (struct device_node *)data;
312 	static int legacy_map = 0, legacy_irq = 0;
313 
314 	spu->devnode = of_node_get(spe);
315 	spu->spe_id = find_spu_unit_number(spe);
316 
317 	spu->node = of_node_to_nid(spe);
318 	if (spu->node >= MAX_NUMNODES) {
319 		printk(KERN_WARNING "SPE %pOF on node %d ignored,"
320 		       " node number too big\n", spe, spu->node);
321 		printk(KERN_WARNING "Check if CONFIG_NUMA is enabled.\n");
322 		ret = -ENODEV;
323 		goto out;
324 	}
325 
326 	ret = spu_map_device(spu);
327 	if (ret) {
328 		if (!legacy_map) {
329 			legacy_map = 1;
330 			printk(KERN_WARNING "%s: Legacy device tree found, "
331 				"trying to map old style\n", __func__);
332 		}
333 		ret = spu_map_device_old(spu);
334 		if (ret) {
335 			printk(KERN_ERR "Unable to map %s\n",
336 				spu->name);
337 			goto out;
338 		}
339 	}
340 
341 	ret = spu_map_interrupts(spu, spe);
342 	if (ret) {
343 		if (!legacy_irq) {
344 			legacy_irq = 1;
345 			printk(KERN_WARNING "%s: Legacy device tree found, "
346 				"trying old style irq\n", __func__);
347 		}
348 		ret = spu_map_interrupts_old(spu, spe);
349 		if (ret) {
350 			printk(KERN_ERR "%s: could not map interrupts\n",
351 				spu->name);
352 			goto out_unmap;
353 		}
354 	}
355 
356 	pr_debug("Using SPE %s %p %p %p %p %d\n", spu->name,
357 		spu->local_store, spu->problem, spu->priv1,
358 		spu->priv2, spu->number);
359 	goto out;
360 
361 out_unmap:
362 	spu_unmap(spu);
363 out:
364 	return ret;
365 }
366 
367 static int of_destroy_spu(struct spu *spu)
368 {
369 	spu_unmap(spu);
370 	of_node_put(spu->devnode);
371 	return 0;
372 }
373 
374 static void enable_spu_by_master_run(struct spu_context *ctx)
375 {
376 	ctx->ops->master_start(ctx);
377 }
378 
379 static void disable_spu_by_master_run(struct spu_context *ctx)
380 {
381 	ctx->ops->master_stop(ctx);
382 }
383 
384 /* Hardcoded affinity idxs for qs20 */
385 #define QS20_SPES_PER_BE 8
386 static int qs20_reg_idxs[QS20_SPES_PER_BE] =   { 0, 2, 4, 6, 7, 5, 3, 1 };
387 static int qs20_reg_memory[QS20_SPES_PER_BE] = { 1, 1, 0, 0, 0, 0, 0, 0 };
388 
389 static struct spu *spu_lookup_reg(int node, u32 reg)
390 {
391 	struct spu *spu;
392 	const u32 *spu_reg;
393 
394 	list_for_each_entry(spu, &cbe_spu_info[node].spus, cbe_list) {
395 		spu_reg = of_get_property(spu_devnode(spu), "reg", NULL);
396 		if (*spu_reg == reg)
397 			return spu;
398 	}
399 	return NULL;
400 }
401 
402 static void init_affinity_qs20_harcoded(void)
403 {
404 	int node, i;
405 	struct spu *last_spu, *spu;
406 	u32 reg;
407 
408 	for (node = 0; node < MAX_NUMNODES; node++) {
409 		last_spu = NULL;
410 		for (i = 0; i < QS20_SPES_PER_BE; i++) {
411 			reg = qs20_reg_idxs[i];
412 			spu = spu_lookup_reg(node, reg);
413 			if (!spu)
414 				continue;
415 			spu->has_mem_affinity = qs20_reg_memory[reg];
416 			if (last_spu)
417 				list_add_tail(&spu->aff_list,
418 						&last_spu->aff_list);
419 			last_spu = spu;
420 		}
421 	}
422 }
423 
424 static int of_has_vicinity(void)
425 {
426 	struct device_node *dn;
427 
428 	for_each_node_by_type(dn, "spe") {
429 		if (of_find_property(dn, "vicinity", NULL))  {
430 			of_node_put(dn);
431 			return 1;
432 		}
433 	}
434 	return 0;
435 }
436 
437 static struct spu *devnode_spu(int cbe, struct device_node *dn)
438 {
439 	struct spu *spu;
440 
441 	list_for_each_entry(spu, &cbe_spu_info[cbe].spus, cbe_list)
442 		if (spu_devnode(spu) == dn)
443 			return spu;
444 	return NULL;
445 }
446 
447 static struct spu *
448 neighbour_spu(int cbe, struct device_node *target, struct device_node *avoid)
449 {
450 	struct spu *spu;
451 	struct device_node *spu_dn;
452 	const phandle *vic_handles;
453 	int lenp, i;
454 
455 	list_for_each_entry(spu, &cbe_spu_info[cbe].spus, cbe_list) {
456 		spu_dn = spu_devnode(spu);
457 		if (spu_dn == avoid)
458 			continue;
459 		vic_handles = of_get_property(spu_dn, "vicinity", &lenp);
460 		for (i=0; i < (lenp / sizeof(phandle)); i++) {
461 			if (vic_handles[i] == target->phandle)
462 				return spu;
463 		}
464 	}
465 	return NULL;
466 }
467 
468 static void init_affinity_node(int cbe)
469 {
470 	struct spu *spu, *last_spu;
471 	struct device_node *vic_dn, *last_spu_dn;
472 	phandle avoid_ph;
473 	const phandle *vic_handles;
474 	const char *name;
475 	int lenp, i, added;
476 
477 	last_spu = list_first_entry(&cbe_spu_info[cbe].spus, struct spu,
478 								cbe_list);
479 	avoid_ph = 0;
480 	for (added = 1; added < cbe_spu_info[cbe].n_spus; added++) {
481 		last_spu_dn = spu_devnode(last_spu);
482 		vic_handles = of_get_property(last_spu_dn, "vicinity", &lenp);
483 
484 		/*
485 		 * Walk through each phandle in vicinity property of the spu
486 		 * (tipically two vicinity phandles per spe node)
487 		 */
488 		for (i = 0; i < (lenp / sizeof(phandle)); i++) {
489 			if (vic_handles[i] == avoid_ph)
490 				continue;
491 
492 			vic_dn = of_find_node_by_phandle(vic_handles[i]);
493 			if (!vic_dn)
494 				continue;
495 
496 			/* a neighbour might be spe, mic-tm, or bif0 */
497 			name = of_get_property(vic_dn, "name", NULL);
498 			if (!name)
499 				continue;
500 
501 			if (strcmp(name, "spe") == 0) {
502 				spu = devnode_spu(cbe, vic_dn);
503 				avoid_ph = last_spu_dn->phandle;
504 			} else {
505 				/*
506 				 * "mic-tm" and "bif0" nodes do not have
507 				 * vicinity property. So we need to find the
508 				 * spe which has vic_dn as neighbour, but
509 				 * skipping the one we came from (last_spu_dn)
510 				 */
511 				spu = neighbour_spu(cbe, vic_dn, last_spu_dn);
512 				if (!spu)
513 					continue;
514 				if (!strcmp(name, "mic-tm")) {
515 					last_spu->has_mem_affinity = 1;
516 					spu->has_mem_affinity = 1;
517 				}
518 				avoid_ph = vic_dn->phandle;
519 			}
520 
521 			list_add_tail(&spu->aff_list, &last_spu->aff_list);
522 			last_spu = spu;
523 			break;
524 		}
525 	}
526 }
527 
528 static void init_affinity_fw(void)
529 {
530 	int cbe;
531 
532 	for (cbe = 0; cbe < MAX_NUMNODES; cbe++)
533 		init_affinity_node(cbe);
534 }
535 
536 static int __init init_affinity(void)
537 {
538 	if (of_has_vicinity()) {
539 		init_affinity_fw();
540 	} else {
541 		if (of_machine_is_compatible("IBM,CPBW-1.0"))
542 			init_affinity_qs20_harcoded();
543 		else
544 			printk("No affinity configuration found\n");
545 	}
546 
547 	return 0;
548 }
549 
550 const struct spu_management_ops spu_management_of_ops = {
551 	.enumerate_spus = of_enumerate_spus,
552 	.create_spu = of_create_spu,
553 	.destroy_spu = of_destroy_spu,
554 	.enable_spu = enable_spu_by_master_run,
555 	.disable_spu = disable_spu_by_master_run,
556 	.init_affinity = init_affinity,
557 };
558