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 	int i;
184 
185 	for (i=0; i < 3; i++) {
186 		spu->irqs[i] = irq_of_parse_and_map(np, i);
187 		if (!spu->irqs[i])
188 			goto err;
189 	}
190 	return 0;
191 
192 err:
193 	pr_debug("failed to map irq %x for spu %s\n", i, spu->name);
194 	for (; i >= 0; i--) {
195 		if (spu->irqs[i])
196 			irq_dispose_mapping(spu->irqs[i]);
197 	}
198 	return -EINVAL;
199 }
200 
201 static int spu_map_resource(struct spu *spu, int nr,
202 			    void __iomem** virt, unsigned long *phys)
203 {
204 	struct device_node *np = spu->devnode;
205 	struct resource resource = { };
206 	unsigned long len;
207 	int ret;
208 
209 	ret = of_address_to_resource(np, nr, &resource);
210 	if (ret)
211 		return ret;
212 	if (phys)
213 		*phys = resource.start;
214 	len = resource_size(&resource);
215 	*virt = ioremap(resource.start, len);
216 	if (!*virt)
217 		return -EINVAL;
218 	return 0;
219 }
220 
221 static int __init spu_map_device(struct spu *spu)
222 {
223 	struct device_node *np = spu->devnode;
224 	int ret = -ENODEV;
225 
226 	spu->name = of_get_property(np, "name", NULL);
227 	if (!spu->name)
228 		goto out;
229 
230 	ret = spu_map_resource(spu, 0, (void __iomem**)&spu->local_store,
231 			       &spu->local_store_phys);
232 	if (ret) {
233 		pr_debug("spu_new: failed to map %pOF resource 0\n",
234 			 np);
235 		goto out;
236 	}
237 	ret = spu_map_resource(spu, 1, (void __iomem**)&spu->problem,
238 			       &spu->problem_phys);
239 	if (ret) {
240 		pr_debug("spu_new: failed to map %pOF resource 1\n",
241 			 np);
242 		goto out_unmap;
243 	}
244 	ret = spu_map_resource(spu, 2, (void __iomem**)&spu->priv2, NULL);
245 	if (ret) {
246 		pr_debug("spu_new: failed to map %pOF resource 2\n",
247 			 np);
248 		goto out_unmap;
249 	}
250 	if (!firmware_has_feature(FW_FEATURE_LPAR))
251 		ret = spu_map_resource(spu, 3,
252 			       (void __iomem**)&spu->priv1, NULL);
253 	if (ret) {
254 		pr_debug("spu_new: failed to map %pOF resource 3\n",
255 			 np);
256 		goto out_unmap;
257 	}
258 	pr_debug("spu_new: %pOF maps:\n", np);
259 	pr_debug("  local store   : 0x%016lx -> 0x%p\n",
260 		 spu->local_store_phys, spu->local_store);
261 	pr_debug("  problem state : 0x%016lx -> 0x%p\n",
262 		 spu->problem_phys, spu->problem);
263 	pr_debug("  priv2         :                       0x%p\n", spu->priv2);
264 	pr_debug("  priv1         :                       0x%p\n", spu->priv1);
265 
266 	return 0;
267 
268 out_unmap:
269 	spu_unmap(spu);
270 out:
271 	pr_debug("failed to map spe %s: %d\n", spu->name, ret);
272 	return ret;
273 }
274 
275 static int __init of_enumerate_spus(int (*fn)(void *data))
276 {
277 	int ret;
278 	struct device_node *node;
279 	unsigned int n = 0;
280 
281 	ret = -ENODEV;
282 	for_each_node_by_type(node, "spe") {
283 		ret = fn(node);
284 		if (ret) {
285 			printk(KERN_WARNING "%s: Error initializing %pOFn\n",
286 				__func__, node);
287 			of_node_put(node);
288 			break;
289 		}
290 		n++;
291 	}
292 	return ret ? ret : n;
293 }
294 
295 static int __init of_create_spu(struct spu *spu, void *data)
296 {
297 	int ret;
298 	struct device_node *spe = (struct device_node *)data;
299 	static int legacy_map = 0, legacy_irq = 0;
300 
301 	spu->devnode = of_node_get(spe);
302 	spu->spe_id = find_spu_unit_number(spe);
303 
304 	spu->node = of_node_to_nid(spe);
305 	if (spu->node >= MAX_NUMNODES) {
306 		printk(KERN_WARNING "SPE %pOF on node %d ignored,"
307 		       " node number too big\n", spe, spu->node);
308 		printk(KERN_WARNING "Check if CONFIG_NUMA is enabled.\n");
309 		ret = -ENODEV;
310 		goto out;
311 	}
312 
313 	ret = spu_map_device(spu);
314 	if (ret) {
315 		if (!legacy_map) {
316 			legacy_map = 1;
317 			printk(KERN_WARNING "%s: Legacy device tree found, "
318 				"trying to map old style\n", __func__);
319 		}
320 		ret = spu_map_device_old(spu);
321 		if (ret) {
322 			printk(KERN_ERR "Unable to map %s\n",
323 				spu->name);
324 			goto out;
325 		}
326 	}
327 
328 	ret = spu_map_interrupts(spu, spe);
329 	if (ret) {
330 		if (!legacy_irq) {
331 			legacy_irq = 1;
332 			printk(KERN_WARNING "%s: Legacy device tree found, "
333 				"trying old style irq\n", __func__);
334 		}
335 		ret = spu_map_interrupts_old(spu, spe);
336 		if (ret) {
337 			printk(KERN_ERR "%s: could not map interrupts\n",
338 				spu->name);
339 			goto out_unmap;
340 		}
341 	}
342 
343 	pr_debug("Using SPE %s %p %p %p %p %d\n", spu->name,
344 		spu->local_store, spu->problem, spu->priv1,
345 		spu->priv2, spu->number);
346 	goto out;
347 
348 out_unmap:
349 	spu_unmap(spu);
350 out:
351 	return ret;
352 }
353 
354 static int of_destroy_spu(struct spu *spu)
355 {
356 	spu_unmap(spu);
357 	of_node_put(spu->devnode);
358 	return 0;
359 }
360 
361 static void enable_spu_by_master_run(struct spu_context *ctx)
362 {
363 	ctx->ops->master_start(ctx);
364 }
365 
366 static void disable_spu_by_master_run(struct spu_context *ctx)
367 {
368 	ctx->ops->master_stop(ctx);
369 }
370 
371 /* Hardcoded affinity idxs for qs20 */
372 #define QS20_SPES_PER_BE 8
373 static int qs20_reg_idxs[QS20_SPES_PER_BE] =   { 0, 2, 4, 6, 7, 5, 3, 1 };
374 static int qs20_reg_memory[QS20_SPES_PER_BE] = { 1, 1, 0, 0, 0, 0, 0, 0 };
375 
376 static struct spu *spu_lookup_reg(int node, u32 reg)
377 {
378 	struct spu *spu;
379 	const u32 *spu_reg;
380 
381 	list_for_each_entry(spu, &cbe_spu_info[node].spus, cbe_list) {
382 		spu_reg = of_get_property(spu_devnode(spu), "reg", NULL);
383 		if (*spu_reg == reg)
384 			return spu;
385 	}
386 	return NULL;
387 }
388 
389 static void init_affinity_qs20_harcoded(void)
390 {
391 	int node, i;
392 	struct spu *last_spu, *spu;
393 	u32 reg;
394 
395 	for (node = 0; node < MAX_NUMNODES; node++) {
396 		last_spu = NULL;
397 		for (i = 0; i < QS20_SPES_PER_BE; i++) {
398 			reg = qs20_reg_idxs[i];
399 			spu = spu_lookup_reg(node, reg);
400 			if (!spu)
401 				continue;
402 			spu->has_mem_affinity = qs20_reg_memory[reg];
403 			if (last_spu)
404 				list_add_tail(&spu->aff_list,
405 						&last_spu->aff_list);
406 			last_spu = spu;
407 		}
408 	}
409 }
410 
411 static int of_has_vicinity(void)
412 {
413 	struct device_node *dn;
414 
415 	for_each_node_by_type(dn, "spe") {
416 		if (of_find_property(dn, "vicinity", NULL))  {
417 			of_node_put(dn);
418 			return 1;
419 		}
420 	}
421 	return 0;
422 }
423 
424 static struct spu *devnode_spu(int cbe, struct device_node *dn)
425 {
426 	struct spu *spu;
427 
428 	list_for_each_entry(spu, &cbe_spu_info[cbe].spus, cbe_list)
429 		if (spu_devnode(spu) == dn)
430 			return spu;
431 	return NULL;
432 }
433 
434 static struct spu *
435 neighbour_spu(int cbe, struct device_node *target, struct device_node *avoid)
436 {
437 	struct spu *spu;
438 	struct device_node *spu_dn;
439 	const phandle *vic_handles;
440 	int lenp, i;
441 
442 	list_for_each_entry(spu, &cbe_spu_info[cbe].spus, cbe_list) {
443 		spu_dn = spu_devnode(spu);
444 		if (spu_dn == avoid)
445 			continue;
446 		vic_handles = of_get_property(spu_dn, "vicinity", &lenp);
447 		for (i=0; i < (lenp / sizeof(phandle)); i++) {
448 			if (vic_handles[i] == target->phandle)
449 				return spu;
450 		}
451 	}
452 	return NULL;
453 }
454 
455 static void init_affinity_node(int cbe)
456 {
457 	struct spu *spu, *last_spu;
458 	struct device_node *vic_dn, *last_spu_dn;
459 	phandle avoid_ph;
460 	const phandle *vic_handles;
461 	int lenp, i, added;
462 
463 	last_spu = list_first_entry(&cbe_spu_info[cbe].spus, struct spu,
464 								cbe_list);
465 	avoid_ph = 0;
466 	for (added = 1; added < cbe_spu_info[cbe].n_spus; added++) {
467 		last_spu_dn = spu_devnode(last_spu);
468 		vic_handles = of_get_property(last_spu_dn, "vicinity", &lenp);
469 
470 		/*
471 		 * Walk through each phandle in vicinity property of the spu
472 		 * (tipically two vicinity phandles per spe node)
473 		 */
474 		for (i = 0; i < (lenp / sizeof(phandle)); i++) {
475 			if (vic_handles[i] == avoid_ph)
476 				continue;
477 
478 			vic_dn = of_find_node_by_phandle(vic_handles[i]);
479 			if (!vic_dn)
480 				continue;
481 
482 			if (of_node_name_eq(vic_dn, "spe") ) {
483 				spu = devnode_spu(cbe, vic_dn);
484 				avoid_ph = last_spu_dn->phandle;
485 			} else {
486 				/*
487 				 * "mic-tm" and "bif0" nodes do not have
488 				 * vicinity property. So we need to find the
489 				 * spe which has vic_dn as neighbour, but
490 				 * skipping the one we came from (last_spu_dn)
491 				 */
492 				spu = neighbour_spu(cbe, vic_dn, last_spu_dn);
493 				if (!spu)
494 					continue;
495 				if (of_node_name_eq(vic_dn, "mic-tm")) {
496 					last_spu->has_mem_affinity = 1;
497 					spu->has_mem_affinity = 1;
498 				}
499 				avoid_ph = vic_dn->phandle;
500 			}
501 
502 			list_add_tail(&spu->aff_list, &last_spu->aff_list);
503 			last_spu = spu;
504 			break;
505 		}
506 	}
507 }
508 
509 static void init_affinity_fw(void)
510 {
511 	int cbe;
512 
513 	for (cbe = 0; cbe < MAX_NUMNODES; cbe++)
514 		init_affinity_node(cbe);
515 }
516 
517 static int __init init_affinity(void)
518 {
519 	if (of_has_vicinity()) {
520 		init_affinity_fw();
521 	} else {
522 		if (of_machine_is_compatible("IBM,CPBW-1.0"))
523 			init_affinity_qs20_harcoded();
524 		else
525 			printk("No affinity configuration found\n");
526 	}
527 
528 	return 0;
529 }
530 
531 const struct spu_management_ops spu_management_of_ops = {
532 	.enumerate_spus = of_enumerate_spus,
533 	.create_spu = of_create_spu,
534 	.destroy_spu = of_destroy_spu,
535 	.enable_spu = enable_spu_by_master_run,
536 	.disable_spu = disable_spu_by_master_run,
537 	.init_affinity = init_affinity,
538 };
539