• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * This file is subject to the terms and conditions of the GNU General Public
3  * License.  See the file "COPYING" in the main directory of this archive
4  * for more details.
5  *
6  * Copyright (C) 2004-2006 Silicon Graphics, Inc. All rights reserved.
7  *
8  * SGI Altix topology and hardware performance monitoring API.
9  * Mark Goodwin <markgw@sgi.com>.
10  *
11  * Creates /proc/sgi_sn/sn_topology (read-only) to export
12  * info about Altix nodes, routers, CPUs and NumaLink
13  * interconnection/topology.
14  *
15  * Also creates a dynamic misc device named "sn_hwperf"
16  * that supports an ioctl interface to call down into SAL
17  * to discover hw objects, topology and to read/write
18  * memory mapped registers, e.g. for performance monitoring.
19  * The "sn_hwperf" device is registered only after the procfs
20  * file is first opened, i.e. only if/when it's needed.
21  *
22  * This API is used by SGI Performance Co-Pilot and other
23  * tools, see http://oss.sgi.com/projects/pcp
24  */
25 
26 #include <linux/fs.h>
27 #include <linux/slab.h>
28 #include <linux/vmalloc.h>
29 #include <linux/seq_file.h>
30 #include <linux/miscdevice.h>
31 #include <linux/utsname.h>
32 #include <linux/cpumask.h>
33 #include <linux/smp_lock.h>
34 #include <linux/nodemask.h>
35 #include <linux/smp.h>
36 #include <linux/mutex.h>
37 
38 #include <asm/processor.h>
39 #include <asm/topology.h>
40 #include <asm/uaccess.h>
41 #include <asm/sal.h>
42 #include <asm/sn/io.h>
43 #include <asm/sn/sn_sal.h>
44 #include <asm/sn/module.h>
45 #include <asm/sn/geo.h>
46 #include <asm/sn/sn2/sn_hwperf.h>
47 #include <asm/sn/addrs.h>
48 
49 static void *sn_hwperf_salheap = NULL;
50 static int sn_hwperf_obj_cnt = 0;
51 static nasid_t sn_hwperf_master_nasid = INVALID_NASID;
52 static int sn_hwperf_init(void);
53 static DEFINE_MUTEX(sn_hwperf_init_mutex);
54 
55 #define cnode_possible(n)	((n) < num_cnodes)
56 
sn_hwperf_enum_objects(int * nobj,struct sn_hwperf_object_info ** ret)57 static int sn_hwperf_enum_objects(int *nobj, struct sn_hwperf_object_info **ret)
58 {
59 	int e;
60 	u64 sz;
61 	struct sn_hwperf_object_info *objbuf = NULL;
62 
63 	if ((e = sn_hwperf_init()) < 0) {
64 		printk(KERN_ERR "sn_hwperf_init failed: err %d\n", e);
65 		goto out;
66 	}
67 
68 	sz = sn_hwperf_obj_cnt * sizeof(struct sn_hwperf_object_info);
69 	objbuf = vmalloc(sz);
70 	if (objbuf == NULL) {
71 		printk("sn_hwperf_enum_objects: vmalloc(%d) failed\n", (int)sz);
72 		e = -ENOMEM;
73 		goto out;
74 	}
75 
76 	e = ia64_sn_hwperf_op(sn_hwperf_master_nasid, SN_HWPERF_ENUM_OBJECTS,
77 		0, sz, (u64) objbuf, 0, 0, NULL);
78 	if (e != SN_HWPERF_OP_OK) {
79 		e = -EINVAL;
80 		vfree(objbuf);
81 	}
82 
83 out:
84 	*nobj = sn_hwperf_obj_cnt;
85 	*ret = objbuf;
86 	return e;
87 }
88 
sn_hwperf_location_to_bpos(char * location,int * rack,int * bay,int * slot,int * slab)89 static int sn_hwperf_location_to_bpos(char *location,
90 	int *rack, int *bay, int *slot, int *slab)
91 {
92 	char type;
93 
94 	/* first scan for an old style geoid string */
95 	if (sscanf(location, "%03d%c%02d#%d",
96 		rack, &type, bay, slab) == 4)
97 		*slot = 0;
98 	else /* scan for a new bladed geoid string */
99 	if (sscanf(location, "%03d%c%02d^%02d#%d",
100 		rack, &type, bay, slot, slab) != 5)
101 		return -1;
102 	/* success */
103 	return 0;
104 }
105 
sn_hwperf_geoid_to_cnode(char * location)106 static int sn_hwperf_geoid_to_cnode(char *location)
107 {
108 	int cnode;
109 	geoid_t geoid;
110 	moduleid_t module_id;
111 	int rack, bay, slot, slab;
112 	int this_rack, this_bay, this_slot, this_slab;
113 
114 	if (sn_hwperf_location_to_bpos(location, &rack, &bay, &slot, &slab))
115 		return -1;
116 
117 	/*
118 	 * FIXME: replace with cleaner for_each_XXX macro which addresses
119 	 * both compute and IO nodes once ACPI3.0 is available.
120 	 */
121 	for (cnode = 0; cnode < num_cnodes; cnode++) {
122 		geoid = cnodeid_get_geoid(cnode);
123 		module_id = geo_module(geoid);
124 		this_rack = MODULE_GET_RACK(module_id);
125 		this_bay = MODULE_GET_BPOS(module_id);
126 		this_slot = geo_slot(geoid);
127 		this_slab = geo_slab(geoid);
128 		if (rack == this_rack && bay == this_bay &&
129 			slot == this_slot && slab == this_slab) {
130 			break;
131 		}
132 	}
133 
134 	return cnode_possible(cnode) ? cnode : -1;
135 }
136 
sn_hwperf_obj_to_cnode(struct sn_hwperf_object_info * obj)137 static int sn_hwperf_obj_to_cnode(struct sn_hwperf_object_info * obj)
138 {
139 	if (!SN_HWPERF_IS_NODE(obj) && !SN_HWPERF_IS_IONODE(obj))
140 		BUG();
141 	if (SN_HWPERF_FOREIGN(obj))
142 		return -1;
143 	return sn_hwperf_geoid_to_cnode(obj->location);
144 }
145 
sn_hwperf_generic_ordinal(struct sn_hwperf_object_info * obj,struct sn_hwperf_object_info * objs)146 static int sn_hwperf_generic_ordinal(struct sn_hwperf_object_info *obj,
147 				struct sn_hwperf_object_info *objs)
148 {
149 	int ordinal;
150 	struct sn_hwperf_object_info *p;
151 
152 	for (ordinal=0, p=objs; p != obj; p++) {
153 		if (SN_HWPERF_FOREIGN(p))
154 			continue;
155 		if (SN_HWPERF_SAME_OBJTYPE(p, obj))
156 			ordinal++;
157 	}
158 
159 	return ordinal;
160 }
161 
162 static const char *slabname_node =	"node"; /* SHub asic */
163 static const char *slabname_ionode =	"ionode"; /* TIO asic */
164 static const char *slabname_router =	"router"; /* NL3R or NL4R */
165 static const char *slabname_other =	"other"; /* unknown asic */
166 
sn_hwperf_get_slabname(struct sn_hwperf_object_info * obj,struct sn_hwperf_object_info * objs,int * ordinal)167 static const char *sn_hwperf_get_slabname(struct sn_hwperf_object_info *obj,
168 			struct sn_hwperf_object_info *objs, int *ordinal)
169 {
170 	int isnode;
171 	const char *slabname = slabname_other;
172 
173 	if ((isnode = SN_HWPERF_IS_NODE(obj)) || SN_HWPERF_IS_IONODE(obj)) {
174 	    	slabname = isnode ? slabname_node : slabname_ionode;
175 		*ordinal = sn_hwperf_obj_to_cnode(obj);
176 	}
177 	else {
178 		*ordinal = sn_hwperf_generic_ordinal(obj, objs);
179 		if (SN_HWPERF_IS_ROUTER(obj))
180 			slabname = slabname_router;
181 	}
182 
183 	return slabname;
184 }
185 
print_pci_topology(struct seq_file * s)186 static void print_pci_topology(struct seq_file *s)
187 {
188 	char *p;
189 	size_t sz;
190 	int e;
191 
192 	for (sz = PAGE_SIZE; sz < 16 * PAGE_SIZE; sz += PAGE_SIZE) {
193 		if (!(p = kmalloc(sz, GFP_KERNEL)))
194 			break;
195 		e = ia64_sn_ioif_get_pci_topology(__pa(p), sz);
196 		if (e == SALRET_OK)
197 			seq_puts(s, p);
198 		kfree(p);
199 		if (e == SALRET_OK || e == SALRET_NOT_IMPLEMENTED)
200 			break;
201 	}
202 }
203 
sn_hwperf_has_cpus(cnodeid_t node)204 static inline int sn_hwperf_has_cpus(cnodeid_t node)
205 {
206 	return node < MAX_NUMNODES && node_online(node) && nr_cpus_node(node);
207 }
208 
sn_hwperf_has_mem(cnodeid_t node)209 static inline int sn_hwperf_has_mem(cnodeid_t node)
210 {
211 	return node < MAX_NUMNODES && node_online(node) && NODE_DATA(node)->node_present_pages;
212 }
213 
214 static struct sn_hwperf_object_info *
sn_hwperf_findobj_id(struct sn_hwperf_object_info * objbuf,int nobj,int id)215 sn_hwperf_findobj_id(struct sn_hwperf_object_info *objbuf,
216 	int nobj, int id)
217 {
218 	int i;
219 	struct sn_hwperf_object_info *p = objbuf;
220 
221 	for (i=0; i < nobj; i++, p++) {
222 		if (p->id == id)
223 			return p;
224 	}
225 
226 	return NULL;
227 
228 }
229 
sn_hwperf_get_nearest_node_objdata(struct sn_hwperf_object_info * objbuf,int nobj,cnodeid_t node,cnodeid_t * near_mem_node,cnodeid_t * near_cpu_node)230 static int sn_hwperf_get_nearest_node_objdata(struct sn_hwperf_object_info *objbuf,
231 	int nobj, cnodeid_t node, cnodeid_t *near_mem_node, cnodeid_t *near_cpu_node)
232 {
233 	int e;
234 	struct sn_hwperf_object_info *nodeobj = NULL;
235 	struct sn_hwperf_object_info *op;
236 	struct sn_hwperf_object_info *dest;
237 	struct sn_hwperf_object_info *router;
238 	struct sn_hwperf_port_info ptdata[16];
239 	int sz, i, j;
240 	cnodeid_t c;
241 	int found_mem = 0;
242 	int found_cpu = 0;
243 
244 	if (!cnode_possible(node))
245 		return -EINVAL;
246 
247 	if (sn_hwperf_has_cpus(node)) {
248 		if (near_cpu_node)
249 			*near_cpu_node = node;
250 		found_cpu++;
251 	}
252 
253 	if (sn_hwperf_has_mem(node)) {
254 		if (near_mem_node)
255 			*near_mem_node = node;
256 		found_mem++;
257 	}
258 
259 	if (found_cpu && found_mem)
260 		return 0; /* trivially successful */
261 
262 	/* find the argument node object */
263 	for (i=0, op=objbuf; i < nobj; i++, op++) {
264 		if (!SN_HWPERF_IS_NODE(op) && !SN_HWPERF_IS_IONODE(op))
265 			continue;
266 		if (node == sn_hwperf_obj_to_cnode(op)) {
267 			nodeobj = op;
268 			break;
269 		}
270 	}
271 	if (!nodeobj) {
272 		e = -ENOENT;
273 		goto err;
274 	}
275 
276 	/* get it's interconnect topology */
277 	sz = op->ports * sizeof(struct sn_hwperf_port_info);
278 	if (sz > sizeof(ptdata))
279 		BUG();
280 	e = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
281 			      SN_HWPERF_ENUM_PORTS, nodeobj->id, sz,
282 			      (u64)&ptdata, 0, 0, NULL);
283 	if (e != SN_HWPERF_OP_OK) {
284 		e = -EINVAL;
285 		goto err;
286 	}
287 
288 	/* find nearest node with cpus and nearest memory */
289 	for (router=NULL, j=0; j < op->ports; j++) {
290 		dest = sn_hwperf_findobj_id(objbuf, nobj, ptdata[j].conn_id);
291 		if (dest && SN_HWPERF_IS_ROUTER(dest))
292 			router = dest;
293 		if (!dest || SN_HWPERF_FOREIGN(dest) ||
294 		    !SN_HWPERF_IS_NODE(dest) || SN_HWPERF_IS_IONODE(dest)) {
295 			continue;
296 		}
297 		c = sn_hwperf_obj_to_cnode(dest);
298 		if (!found_cpu && sn_hwperf_has_cpus(c)) {
299 			if (near_cpu_node)
300 				*near_cpu_node = c;
301 			found_cpu++;
302 		}
303 		if (!found_mem && sn_hwperf_has_mem(c)) {
304 			if (near_mem_node)
305 				*near_mem_node = c;
306 			found_mem++;
307 		}
308 	}
309 
310 	if (router && (!found_cpu || !found_mem)) {
311 		/* search for a node connected to the same router */
312 		sz = router->ports * sizeof(struct sn_hwperf_port_info);
313 		if (sz > sizeof(ptdata))
314 			BUG();
315 		e = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
316 				      SN_HWPERF_ENUM_PORTS, router->id, sz,
317 				      (u64)&ptdata, 0, 0, NULL);
318 		if (e != SN_HWPERF_OP_OK) {
319 			e = -EINVAL;
320 			goto err;
321 		}
322 		for (j=0; j < router->ports; j++) {
323 			dest = sn_hwperf_findobj_id(objbuf, nobj,
324 				ptdata[j].conn_id);
325 			if (!dest || dest->id == node ||
326 			    SN_HWPERF_FOREIGN(dest) ||
327 			    !SN_HWPERF_IS_NODE(dest) ||
328 			    SN_HWPERF_IS_IONODE(dest)) {
329 				continue;
330 			}
331 			c = sn_hwperf_obj_to_cnode(dest);
332 			if (!found_cpu && sn_hwperf_has_cpus(c)) {
333 				if (near_cpu_node)
334 					*near_cpu_node = c;
335 				found_cpu++;
336 			}
337 			if (!found_mem && sn_hwperf_has_mem(c)) {
338 				if (near_mem_node)
339 					*near_mem_node = c;
340 				found_mem++;
341 			}
342 			if (found_cpu && found_mem)
343 				break;
344 		}
345 	}
346 
347 	if (!found_cpu || !found_mem) {
348 		/* resort to _any_ node with CPUs and memory */
349 		for (i=0, op=objbuf; i < nobj; i++, op++) {
350 			if (SN_HWPERF_FOREIGN(op) ||
351 			    SN_HWPERF_IS_IONODE(op) ||
352 			    !SN_HWPERF_IS_NODE(op)) {
353 				continue;
354 			}
355 			c = sn_hwperf_obj_to_cnode(op);
356 			if (!found_cpu && sn_hwperf_has_cpus(c)) {
357 				if (near_cpu_node)
358 					*near_cpu_node = c;
359 				found_cpu++;
360 			}
361 			if (!found_mem && sn_hwperf_has_mem(c)) {
362 				if (near_mem_node)
363 					*near_mem_node = c;
364 				found_mem++;
365 			}
366 			if (found_cpu && found_mem)
367 				break;
368 		}
369 	}
370 
371 	if (!found_cpu || !found_mem)
372 		e = -ENODATA;
373 
374 err:
375 	return e;
376 }
377 
378 
sn_topology_show(struct seq_file * s,void * d)379 static int sn_topology_show(struct seq_file *s, void *d)
380 {
381 	int sz;
382 	int pt;
383 	int e = 0;
384 	int i;
385 	int j;
386 	const char *slabname;
387 	int ordinal;
388 	char slice;
389 	struct cpuinfo_ia64 *c;
390 	struct sn_hwperf_port_info *ptdata;
391 	struct sn_hwperf_object_info *p;
392 	struct sn_hwperf_object_info *obj = d;	/* this object */
393 	struct sn_hwperf_object_info *objs = s->private; /* all objects */
394 	u8 shubtype;
395 	u8 system_size;
396 	u8 sharing_size;
397 	u8 partid;
398 	u8 coher;
399 	u8 nasid_shift;
400 	u8 region_size;
401 	u16 nasid_mask;
402 	int nasid_msb;
403 
404 	if (obj == objs) {
405 		seq_printf(s, "# sn_topology version 2\n");
406 		seq_printf(s, "# objtype ordinal location partition"
407 			" [attribute value [, ...]]\n");
408 
409 		if (ia64_sn_get_sn_info(0,
410 			&shubtype, &nasid_mask, &nasid_shift, &system_size,
411 			&sharing_size, &partid, &coher, &region_size))
412 			BUG();
413 		for (nasid_msb=63; nasid_msb > 0; nasid_msb--) {
414 			if (((u64)nasid_mask << nasid_shift) & (1ULL << nasid_msb))
415 				break;
416 		}
417 		seq_printf(s, "partition %u %s local "
418 			"shubtype %s, "
419 			"nasid_mask 0x%016lx, "
420 			"nasid_bits %d:%d, "
421 			"system_size %d, "
422 			"sharing_size %d, "
423 			"coherency_domain %d, "
424 			"region_size %d\n",
425 
426 			partid, utsname()->nodename,
427 			shubtype ? "shub2" : "shub1",
428 			(u64)nasid_mask << nasid_shift, nasid_msb, nasid_shift,
429 			system_size, sharing_size, coher, region_size);
430 
431 		print_pci_topology(s);
432 	}
433 
434 	if (SN_HWPERF_FOREIGN(obj)) {
435 		/* private in another partition: not interesting */
436 		return 0;
437 	}
438 
439 	for (i = 0; i < SN_HWPERF_MAXSTRING && obj->name[i]; i++) {
440 		if (obj->name[i] == ' ')
441 			obj->name[i] = '_';
442 	}
443 
444 	slabname = sn_hwperf_get_slabname(obj, objs, &ordinal);
445 	seq_printf(s, "%s %d %s %s asic %s", slabname, ordinal, obj->location,
446 		obj->sn_hwp_this_part ? "local" : "shared", obj->name);
447 
448 	if (ordinal < 0 || (!SN_HWPERF_IS_NODE(obj) && !SN_HWPERF_IS_IONODE(obj)))
449 		seq_putc(s, '\n');
450 	else {
451 		cnodeid_t near_mem = -1;
452 		cnodeid_t near_cpu = -1;
453 
454 		seq_printf(s, ", nasid 0x%x", cnodeid_to_nasid(ordinal));
455 
456 		if (sn_hwperf_get_nearest_node_objdata(objs, sn_hwperf_obj_cnt,
457 			ordinal, &near_mem, &near_cpu) == 0) {
458 			seq_printf(s, ", near_mem_nodeid %d, near_cpu_nodeid %d",
459 				near_mem, near_cpu);
460 		}
461 
462 		if (!SN_HWPERF_IS_IONODE(obj)) {
463 			for_each_online_node(i) {
464 				seq_printf(s, i ? ":%d" : ", dist %d",
465 					node_distance(ordinal, i));
466 			}
467 		}
468 
469 		seq_putc(s, '\n');
470 
471 		/*
472 		 * CPUs on this node, if any
473 		 */
474 		if (!SN_HWPERF_IS_IONODE(obj)) {
475 			for_each_cpu_and(i, cpu_online_mask,
476 					 cpumask_of_node(ordinal)) {
477 				slice = 'a' + cpuid_to_slice(i);
478 				c = cpu_data(i);
479 				seq_printf(s, "cpu %d %s%c local"
480 					   " freq %luMHz, arch ia64",
481 					   i, obj->location, slice,
482 					   c->proc_freq / 1000000);
483 				for_each_online_cpu(j) {
484 					seq_printf(s, j ? ":%d" : ", dist %d",
485 						   node_distance(
486 						    	cpu_to_node(i),
487 						    	cpu_to_node(j)));
488 				}
489 				seq_putc(s, '\n');
490 			}
491 		}
492 	}
493 
494 	if (obj->ports) {
495 		/*
496 		 * numalink ports
497 		 */
498 		sz = obj->ports * sizeof(struct sn_hwperf_port_info);
499 		if ((ptdata = kmalloc(sz, GFP_KERNEL)) == NULL)
500 			return -ENOMEM;
501 		e = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
502 				      SN_HWPERF_ENUM_PORTS, obj->id, sz,
503 				      (u64) ptdata, 0, 0, NULL);
504 		if (e != SN_HWPERF_OP_OK)
505 			return -EINVAL;
506 		for (ordinal=0, p=objs; p != obj; p++) {
507 			if (!SN_HWPERF_FOREIGN(p))
508 				ordinal += p->ports;
509 		}
510 		for (pt = 0; pt < obj->ports; pt++) {
511 			for (p = objs, i = 0; i < sn_hwperf_obj_cnt; i++, p++) {
512 				if (ptdata[pt].conn_id == p->id) {
513 					break;
514 				}
515 			}
516 			seq_printf(s, "numalink %d %s-%d",
517 			    ordinal+pt, obj->location, ptdata[pt].port);
518 
519 			if (i >= sn_hwperf_obj_cnt) {
520 				/* no connection */
521 				seq_puts(s, " local endpoint disconnected"
522 					    ", protocol unknown\n");
523 				continue;
524 			}
525 
526 			if (obj->sn_hwp_this_part && p->sn_hwp_this_part)
527 				/* both ends local to this partition */
528 				seq_puts(s, " local");
529 			else if (SN_HWPERF_FOREIGN(p))
530 				/* both ends of the link in foreign partiton */
531 				seq_puts(s, " foreign");
532 			else
533 				/* link straddles a partition */
534 				seq_puts(s, " shared");
535 
536 			/*
537 			 * Unlikely, but strictly should query the LLP config
538 			 * registers because an NL4R can be configured to run
539 			 * NL3 protocol, even when not talking to an NL3 router.
540 			 * Ditto for node-node.
541 			 */
542 			seq_printf(s, " endpoint %s-%d, protocol %s\n",
543 				p->location, ptdata[pt].conn_port,
544 				(SN_HWPERF_IS_NL3ROUTER(obj) ||
545 				SN_HWPERF_IS_NL3ROUTER(p)) ?  "LLP3" : "LLP4");
546 		}
547 		kfree(ptdata);
548 	}
549 
550 	return 0;
551 }
552 
sn_topology_start(struct seq_file * s,loff_t * pos)553 static void *sn_topology_start(struct seq_file *s, loff_t * pos)
554 {
555 	struct sn_hwperf_object_info *objs = s->private;
556 
557 	if (*pos < sn_hwperf_obj_cnt)
558 		return (void *)(objs + *pos);
559 
560 	return NULL;
561 }
562 
sn_topology_next(struct seq_file * s,void * v,loff_t * pos)563 static void *sn_topology_next(struct seq_file *s, void *v, loff_t * pos)
564 {
565 	++*pos;
566 	return sn_topology_start(s, pos);
567 }
568 
sn_topology_stop(struct seq_file * m,void * v)569 static void sn_topology_stop(struct seq_file *m, void *v)
570 {
571 	return;
572 }
573 
574 /*
575  * /proc/sgi_sn/sn_topology, read-only using seq_file
576  */
577 static const struct seq_operations sn_topology_seq_ops = {
578 	.start = sn_topology_start,
579 	.next = sn_topology_next,
580 	.stop = sn_topology_stop,
581 	.show = sn_topology_show
582 };
583 
584 struct sn_hwperf_op_info {
585 	u64 op;
586 	struct sn_hwperf_ioctl_args *a;
587 	void *p;
588 	int *v0;
589 	int ret;
590 };
591 
sn_hwperf_call_sal(void * info)592 static void sn_hwperf_call_sal(void *info)
593 {
594 	struct sn_hwperf_op_info *op_info = info;
595 	int r;
596 
597 	r = ia64_sn_hwperf_op(sn_hwperf_master_nasid, op_info->op,
598 		      op_info->a->arg, op_info->a->sz,
599 		      (u64) op_info->p, 0, 0, op_info->v0);
600 	op_info->ret = r;
601 }
602 
sn_hwperf_op_cpu(struct sn_hwperf_op_info * op_info)603 static int sn_hwperf_op_cpu(struct sn_hwperf_op_info *op_info)
604 {
605 	u32 cpu;
606 	u32 use_ipi;
607 	int r = 0;
608 	cpumask_t save_allowed;
609 
610 	cpu = (op_info->a->arg & SN_HWPERF_ARG_CPU_MASK) >> 32;
611 	use_ipi = op_info->a->arg & SN_HWPERF_ARG_USE_IPI_MASK;
612 	op_info->a->arg &= SN_HWPERF_ARG_OBJID_MASK;
613 
614 	if (cpu != SN_HWPERF_ARG_ANY_CPU) {
615 		if (cpu >= NR_CPUS || !cpu_online(cpu)) {
616 			r = -EINVAL;
617 			goto out;
618 		}
619 	}
620 
621 	if (cpu == SN_HWPERF_ARG_ANY_CPU || cpu == get_cpu()) {
622 		/* don't care, or already on correct cpu */
623 		sn_hwperf_call_sal(op_info);
624 	}
625 	else {
626 		if (use_ipi) {
627 			/* use an interprocessor interrupt to call SAL */
628 			smp_call_function_single(cpu, sn_hwperf_call_sal,
629 				op_info, 1);
630 		}
631 		else {
632 			/* migrate the task before calling SAL */
633 			save_allowed = current->cpus_allowed;
634 			set_cpus_allowed(current, cpumask_of_cpu(cpu));
635 			sn_hwperf_call_sal(op_info);
636 			set_cpus_allowed(current, save_allowed);
637 		}
638 	}
639 	r = op_info->ret;
640 
641 out:
642 	return r;
643 }
644 
645 /* map SAL hwperf error code to system error code */
sn_hwperf_map_err(int hwperf_err)646 static int sn_hwperf_map_err(int hwperf_err)
647 {
648 	int e;
649 
650 	switch(hwperf_err) {
651 	case SN_HWPERF_OP_OK:
652 		e = 0;
653 		break;
654 
655 	case SN_HWPERF_OP_NOMEM:
656 		e = -ENOMEM;
657 		break;
658 
659 	case SN_HWPERF_OP_NO_PERM:
660 		e = -EPERM;
661 		break;
662 
663 	case SN_HWPERF_OP_IO_ERROR:
664 		e = -EIO;
665 		break;
666 
667 	case SN_HWPERF_OP_BUSY:
668 		e = -EBUSY;
669 		break;
670 
671 	case SN_HWPERF_OP_RECONFIGURE:
672 		e = -EAGAIN;
673 		break;
674 
675 	case SN_HWPERF_OP_INVAL:
676 	default:
677 		e = -EINVAL;
678 		break;
679 	}
680 
681 	return e;
682 }
683 
684 /*
685  * ioctl for "sn_hwperf" misc device
686  */
687 static int
sn_hwperf_ioctl(struct inode * in,struct file * fp,u32 op,u64 arg)688 sn_hwperf_ioctl(struct inode *in, struct file *fp, u32 op, u64 arg)
689 {
690 	struct sn_hwperf_ioctl_args a;
691 	struct cpuinfo_ia64 *cdata;
692 	struct sn_hwperf_object_info *objs;
693 	struct sn_hwperf_object_info *cpuobj;
694 	struct sn_hwperf_op_info op_info;
695 	void *p = NULL;
696 	int nobj;
697 	char slice;
698 	int node;
699 	int r;
700 	int v0;
701 	int i;
702 	int j;
703 
704 	unlock_kernel();
705 
706 	/* only user requests are allowed here */
707 	if ((op & SN_HWPERF_OP_MASK) < 10) {
708 		r = -EINVAL;
709 		goto error;
710 	}
711 	r = copy_from_user(&a, (const void __user *)arg,
712 		sizeof(struct sn_hwperf_ioctl_args));
713 	if (r != 0) {
714 		r = -EFAULT;
715 		goto error;
716 	}
717 
718 	/*
719 	 * Allocate memory to hold a kernel copy of the user buffer. The
720 	 * buffer contents are either copied in or out (or both) of user
721 	 * space depending on the flags encoded in the requested operation.
722 	 */
723 	if (a.ptr) {
724 		p = vmalloc(a.sz);
725 		if (!p) {
726 			r = -ENOMEM;
727 			goto error;
728 		}
729 	}
730 
731 	if (op & SN_HWPERF_OP_MEM_COPYIN) {
732 		r = copy_from_user(p, (const void __user *)a.ptr, a.sz);
733 		if (r != 0) {
734 			r = -EFAULT;
735 			goto error;
736 		}
737 	}
738 
739 	switch (op) {
740 	case SN_HWPERF_GET_CPU_INFO:
741 		if (a.sz == sizeof(u64)) {
742 			/* special case to get size needed */
743 			*(u64 *) p = (u64) num_online_cpus() *
744 				sizeof(struct sn_hwperf_object_info);
745 		} else
746 		if (a.sz < num_online_cpus() * sizeof(struct sn_hwperf_object_info)) {
747 			r = -ENOMEM;
748 			goto error;
749 		} else
750 		if ((r = sn_hwperf_enum_objects(&nobj, &objs)) == 0) {
751 			int cpuobj_index = 0;
752 
753 			memset(p, 0, a.sz);
754 			for (i = 0; i < nobj; i++) {
755 				if (!SN_HWPERF_IS_NODE(objs + i))
756 					continue;
757 				node = sn_hwperf_obj_to_cnode(objs + i);
758 				for_each_online_cpu(j) {
759 					if (node != cpu_to_node(j))
760 						continue;
761 					cpuobj = (struct sn_hwperf_object_info *) p + cpuobj_index++;
762 					slice = 'a' + cpuid_to_slice(j);
763 					cdata = cpu_data(j);
764 					cpuobj->id = j;
765 					snprintf(cpuobj->name,
766 						 sizeof(cpuobj->name),
767 						 "CPU %luMHz %s",
768 						 cdata->proc_freq / 1000000,
769 						 cdata->vendor);
770 					snprintf(cpuobj->location,
771 						 sizeof(cpuobj->location),
772 						 "%s%c", objs[i].location,
773 						 slice);
774 				}
775 			}
776 
777 			vfree(objs);
778 		}
779 		break;
780 
781 	case SN_HWPERF_GET_NODE_NASID:
782 		if (a.sz != sizeof(u64) ||
783 		   (node = a.arg) < 0 || !cnode_possible(node)) {
784 			r = -EINVAL;
785 			goto error;
786 		}
787 		*(u64 *)p = (u64)cnodeid_to_nasid(node);
788 		break;
789 
790 	case SN_HWPERF_GET_OBJ_NODE:
791 		if (a.sz != sizeof(u64) || a.arg < 0) {
792 			r = -EINVAL;
793 			goto error;
794 		}
795 		if ((r = sn_hwperf_enum_objects(&nobj, &objs)) == 0) {
796 			if (a.arg >= nobj) {
797 				r = -EINVAL;
798 				vfree(objs);
799 				goto error;
800 			}
801 			if (objs[(i = a.arg)].id != a.arg) {
802 				for (i = 0; i < nobj; i++) {
803 					if (objs[i].id == a.arg)
804 						break;
805 				}
806 			}
807 			if (i == nobj) {
808 				r = -EINVAL;
809 				vfree(objs);
810 				goto error;
811 			}
812 
813 			if (!SN_HWPERF_IS_NODE(objs + i) &&
814 			    !SN_HWPERF_IS_IONODE(objs + i)) {
815 			    	r = -ENOENT;
816 				vfree(objs);
817 				goto error;
818 			}
819 
820 			*(u64 *)p = (u64)sn_hwperf_obj_to_cnode(objs + i);
821 			vfree(objs);
822 		}
823 		break;
824 
825 	case SN_HWPERF_GET_MMRS:
826 	case SN_HWPERF_SET_MMRS:
827 	case SN_HWPERF_OBJECT_DISTANCE:
828 		op_info.p = p;
829 		op_info.a = &a;
830 		op_info.v0 = &v0;
831 		op_info.op = op;
832 		r = sn_hwperf_op_cpu(&op_info);
833 		if (r) {
834 			r = sn_hwperf_map_err(r);
835 			a.v0 = v0;
836 			goto error;
837 		}
838 		break;
839 
840 	default:
841 		/* all other ops are a direct SAL call */
842 		r = ia64_sn_hwperf_op(sn_hwperf_master_nasid, op,
843 			      a.arg, a.sz, (u64) p, 0, 0, &v0);
844 		if (r) {
845 			r = sn_hwperf_map_err(r);
846 			goto error;
847 		}
848 		a.v0 = v0;
849 		break;
850 	}
851 
852 	if (op & SN_HWPERF_OP_MEM_COPYOUT) {
853 		r = copy_to_user((void __user *)a.ptr, p, a.sz);
854 		if (r != 0) {
855 			r = -EFAULT;
856 			goto error;
857 		}
858 	}
859 
860 error:
861 	vfree(p);
862 
863 	lock_kernel();
864 	return r;
865 }
866 
867 static const struct file_operations sn_hwperf_fops = {
868 	.ioctl = sn_hwperf_ioctl,
869 };
870 
871 static struct miscdevice sn_hwperf_dev = {
872 	MISC_DYNAMIC_MINOR,
873 	"sn_hwperf",
874 	&sn_hwperf_fops
875 };
876 
sn_hwperf_init(void)877 static int sn_hwperf_init(void)
878 {
879 	u64 v;
880 	int salr;
881 	int e = 0;
882 
883 	/* single threaded, once-only initialization */
884 	mutex_lock(&sn_hwperf_init_mutex);
885 
886 	if (sn_hwperf_salheap) {
887 		mutex_unlock(&sn_hwperf_init_mutex);
888 		return e;
889 	}
890 
891 	/*
892 	 * The PROM code needs a fixed reference node. For convenience the
893 	 * same node as the console I/O is used.
894 	 */
895 	sn_hwperf_master_nasid = (nasid_t) ia64_sn_get_console_nasid();
896 
897 	/*
898 	 * Request the needed size and install the PROM scratch area.
899 	 * The PROM keeps various tracking bits in this memory area.
900 	 */
901 	salr = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
902 				 (u64) SN_HWPERF_GET_HEAPSIZE, 0,
903 				 (u64) sizeof(u64), (u64) &v, 0, 0, NULL);
904 	if (salr != SN_HWPERF_OP_OK) {
905 		e = -EINVAL;
906 		goto out;
907 	}
908 
909 	if ((sn_hwperf_salheap = vmalloc(v)) == NULL) {
910 		e = -ENOMEM;
911 		goto out;
912 	}
913 	salr = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
914 				 SN_HWPERF_INSTALL_HEAP, 0, v,
915 				 (u64) sn_hwperf_salheap, 0, 0, NULL);
916 	if (salr != SN_HWPERF_OP_OK) {
917 		e = -EINVAL;
918 		goto out;
919 	}
920 
921 	salr = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
922 				 SN_HWPERF_OBJECT_COUNT, 0,
923 				 sizeof(u64), (u64) &v, 0, 0, NULL);
924 	if (salr != SN_HWPERF_OP_OK) {
925 		e = -EINVAL;
926 		goto out;
927 	}
928 	sn_hwperf_obj_cnt = (int)v;
929 
930 out:
931 	if (e < 0 && sn_hwperf_salheap) {
932 		vfree(sn_hwperf_salheap);
933 		sn_hwperf_salheap = NULL;
934 		sn_hwperf_obj_cnt = 0;
935 	}
936 	mutex_unlock(&sn_hwperf_init_mutex);
937 	return e;
938 }
939 
sn_topology_open(struct inode * inode,struct file * file)940 int sn_topology_open(struct inode *inode, struct file *file)
941 {
942 	int e;
943 	struct seq_file *seq;
944 	struct sn_hwperf_object_info *objbuf;
945 	int nobj;
946 
947 	if ((e = sn_hwperf_enum_objects(&nobj, &objbuf)) == 0) {
948 		e = seq_open(file, &sn_topology_seq_ops);
949 		seq = file->private_data;
950 		seq->private = objbuf;
951 	}
952 
953 	return e;
954 }
955 
sn_topology_release(struct inode * inode,struct file * file)956 int sn_topology_release(struct inode *inode, struct file *file)
957 {
958 	struct seq_file *seq = file->private_data;
959 
960 	vfree(seq->private);
961 	return seq_release(inode, file);
962 }
963 
sn_hwperf_get_nearest_node(cnodeid_t node,cnodeid_t * near_mem_node,cnodeid_t * near_cpu_node)964 int sn_hwperf_get_nearest_node(cnodeid_t node,
965 	cnodeid_t *near_mem_node, cnodeid_t *near_cpu_node)
966 {
967 	int e;
968 	int nobj;
969 	struct sn_hwperf_object_info *objbuf;
970 
971 	if ((e = sn_hwperf_enum_objects(&nobj, &objbuf)) == 0) {
972 		e = sn_hwperf_get_nearest_node_objdata(objbuf, nobj,
973 			node, near_mem_node, near_cpu_node);
974 		vfree(objbuf);
975 	}
976 
977 	return e;
978 }
979 
sn_hwperf_misc_register_init(void)980 static int __devinit sn_hwperf_misc_register_init(void)
981 {
982 	int e;
983 
984 	if (!ia64_platform_is("sn2"))
985 		return 0;
986 
987 	sn_hwperf_init();
988 
989 	/*
990 	 * Register a dynamic misc device for hwperf ioctls. Platforms
991 	 * supporting hotplug will create /dev/sn_hwperf, else user
992 	 * can to look up the minor number in /proc/misc.
993 	 */
994 	if ((e = misc_register(&sn_hwperf_dev)) != 0) {
995 		printk(KERN_ERR "sn_hwperf_misc_register_init: failed to "
996 		"register misc device for \"%s\"\n", sn_hwperf_dev.name);
997 	}
998 
999 	return e;
1000 }
1001 
1002 device_initcall(sn_hwperf_misc_register_init); /* after misc_init() */
1003 EXPORT_SYMBOL(sn_hwperf_get_nearest_node);
1004