smp_call_function: get rid of the unused nonatomic/retry argument
[deliverable/linux.git] / arch / ia64 / sn / kernel / sn2 / sn_hwperf.c
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
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
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
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
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
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
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
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
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
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 *
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
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
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 cpumask_t cpumask;
389 char slice;
390 struct cpuinfo_ia64 *c;
391 struct sn_hwperf_port_info *ptdata;
392 struct sn_hwperf_object_info *p;
393 struct sn_hwperf_object_info *obj = d; /* this object */
394 struct sn_hwperf_object_info *objs = s->private; /* all objects */
395 u8 shubtype;
396 u8 system_size;
397 u8 sharing_size;
398 u8 partid;
399 u8 coher;
400 u8 nasid_shift;
401 u8 region_size;
402 u16 nasid_mask;
403 int nasid_msb;
404
405 if (obj == objs) {
406 seq_printf(s, "# sn_topology version 2\n");
407 seq_printf(s, "# objtype ordinal location partition"
408 " [attribute value [, ...]]\n");
409
410 if (ia64_sn_get_sn_info(0,
411 &shubtype, &nasid_mask, &nasid_shift, &system_size,
412 &sharing_size, &partid, &coher, &region_size))
413 BUG();
414 for (nasid_msb=63; nasid_msb > 0; nasid_msb--) {
415 if (((u64)nasid_mask << nasid_shift) & (1ULL << nasid_msb))
416 break;
417 }
418 seq_printf(s, "partition %u %s local "
419 "shubtype %s, "
420 "nasid_mask 0x%016lx, "
421 "nasid_bits %d:%d, "
422 "system_size %d, "
423 "sharing_size %d, "
424 "coherency_domain %d, "
425 "region_size %d\n",
426
427 partid, utsname()->nodename,
428 shubtype ? "shub2" : "shub1",
429 (u64)nasid_mask << nasid_shift, nasid_msb, nasid_shift,
430 system_size, sharing_size, coher, region_size);
431
432 print_pci_topology(s);
433 }
434
435 if (SN_HWPERF_FOREIGN(obj)) {
436 /* private in another partition: not interesting */
437 return 0;
438 }
439
440 for (i = 0; i < SN_HWPERF_MAXSTRING && obj->name[i]; i++) {
441 if (obj->name[i] == ' ')
442 obj->name[i] = '_';
443 }
444
445 slabname = sn_hwperf_get_slabname(obj, objs, &ordinal);
446 seq_printf(s, "%s %d %s %s asic %s", slabname, ordinal, obj->location,
447 obj->sn_hwp_this_part ? "local" : "shared", obj->name);
448
449 if (ordinal < 0 || (!SN_HWPERF_IS_NODE(obj) && !SN_HWPERF_IS_IONODE(obj)))
450 seq_putc(s, '\n');
451 else {
452 cnodeid_t near_mem = -1;
453 cnodeid_t near_cpu = -1;
454
455 seq_printf(s, ", nasid 0x%x", cnodeid_to_nasid(ordinal));
456
457 if (sn_hwperf_get_nearest_node_objdata(objs, sn_hwperf_obj_cnt,
458 ordinal, &near_mem, &near_cpu) == 0) {
459 seq_printf(s, ", near_mem_nodeid %d, near_cpu_nodeid %d",
460 near_mem, near_cpu);
461 }
462
463 if (!SN_HWPERF_IS_IONODE(obj)) {
464 for_each_online_node(i) {
465 seq_printf(s, i ? ":%d" : ", dist %d",
466 node_distance(ordinal, i));
467 }
468 }
469
470 seq_putc(s, '\n');
471
472 /*
473 * CPUs on this node, if any
474 */
475 if (!SN_HWPERF_IS_IONODE(obj)) {
476 cpumask = node_to_cpumask(ordinal);
477 for_each_online_cpu(i) {
478 if (cpu_isset(i, cpumask)) {
479 slice = 'a' + cpuid_to_slice(i);
480 c = cpu_data(i);
481 seq_printf(s, "cpu %d %s%c local"
482 " freq %luMHz, arch ia64",
483 i, obj->location, slice,
484 c->proc_freq / 1000000);
485 for_each_online_cpu(j) {
486 seq_printf(s, j ? ":%d" : ", dist %d",
487 node_distance(
488 cpu_to_node(i),
489 cpu_to_node(j)));
490 }
491 seq_putc(s, '\n');
492 }
493 }
494 }
495 }
496
497 if (obj->ports) {
498 /*
499 * numalink ports
500 */
501 sz = obj->ports * sizeof(struct sn_hwperf_port_info);
502 if ((ptdata = kmalloc(sz, GFP_KERNEL)) == NULL)
503 return -ENOMEM;
504 e = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
505 SN_HWPERF_ENUM_PORTS, obj->id, sz,
506 (u64) ptdata, 0, 0, NULL);
507 if (e != SN_HWPERF_OP_OK)
508 return -EINVAL;
509 for (ordinal=0, p=objs; p != obj; p++) {
510 if (!SN_HWPERF_FOREIGN(p))
511 ordinal += p->ports;
512 }
513 for (pt = 0; pt < obj->ports; pt++) {
514 for (p = objs, i = 0; i < sn_hwperf_obj_cnt; i++, p++) {
515 if (ptdata[pt].conn_id == p->id) {
516 break;
517 }
518 }
519 seq_printf(s, "numalink %d %s-%d",
520 ordinal+pt, obj->location, ptdata[pt].port);
521
522 if (i >= sn_hwperf_obj_cnt) {
523 /* no connection */
524 seq_puts(s, " local endpoint disconnected"
525 ", protocol unknown\n");
526 continue;
527 }
528
529 if (obj->sn_hwp_this_part && p->sn_hwp_this_part)
530 /* both ends local to this partition */
531 seq_puts(s, " local");
532 else if (SN_HWPERF_FOREIGN(p))
533 /* both ends of the link in foreign partiton */
534 seq_puts(s, " foreign");
535 else
536 /* link straddles a partition */
537 seq_puts(s, " shared");
538
539 /*
540 * Unlikely, but strictly should query the LLP config
541 * registers because an NL4R can be configured to run
542 * NL3 protocol, even when not talking to an NL3 router.
543 * Ditto for node-node.
544 */
545 seq_printf(s, " endpoint %s-%d, protocol %s\n",
546 p->location, ptdata[pt].conn_port,
547 (SN_HWPERF_IS_NL3ROUTER(obj) ||
548 SN_HWPERF_IS_NL3ROUTER(p)) ? "LLP3" : "LLP4");
549 }
550 kfree(ptdata);
551 }
552
553 return 0;
554 }
555
556 static void *sn_topology_start(struct seq_file *s, loff_t * pos)
557 {
558 struct sn_hwperf_object_info *objs = s->private;
559
560 if (*pos < sn_hwperf_obj_cnt)
561 return (void *)(objs + *pos);
562
563 return NULL;
564 }
565
566 static void *sn_topology_next(struct seq_file *s, void *v, loff_t * pos)
567 {
568 ++*pos;
569 return sn_topology_start(s, pos);
570 }
571
572 static void sn_topology_stop(struct seq_file *m, void *v)
573 {
574 return;
575 }
576
577 /*
578 * /proc/sgi_sn/sn_topology, read-only using seq_file
579 */
580 static const struct seq_operations sn_topology_seq_ops = {
581 .start = sn_topology_start,
582 .next = sn_topology_next,
583 .stop = sn_topology_stop,
584 .show = sn_topology_show
585 };
586
587 struct sn_hwperf_op_info {
588 u64 op;
589 struct sn_hwperf_ioctl_args *a;
590 void *p;
591 int *v0;
592 int ret;
593 };
594
595 static void sn_hwperf_call_sal(void *info)
596 {
597 struct sn_hwperf_op_info *op_info = info;
598 int r;
599
600 r = ia64_sn_hwperf_op(sn_hwperf_master_nasid, op_info->op,
601 op_info->a->arg, op_info->a->sz,
602 (u64) op_info->p, 0, 0, op_info->v0);
603 op_info->ret = r;
604 }
605
606 static int sn_hwperf_op_cpu(struct sn_hwperf_op_info *op_info)
607 {
608 u32 cpu;
609 u32 use_ipi;
610 int r = 0;
611 cpumask_t save_allowed;
612
613 cpu = (op_info->a->arg & SN_HWPERF_ARG_CPU_MASK) >> 32;
614 use_ipi = op_info->a->arg & SN_HWPERF_ARG_USE_IPI_MASK;
615 op_info->a->arg &= SN_HWPERF_ARG_OBJID_MASK;
616
617 if (cpu != SN_HWPERF_ARG_ANY_CPU) {
618 if (cpu >= NR_CPUS || !cpu_online(cpu)) {
619 r = -EINVAL;
620 goto out;
621 }
622 }
623
624 if (cpu == SN_HWPERF_ARG_ANY_CPU || cpu == get_cpu()) {
625 /* don't care, or already on correct cpu */
626 sn_hwperf_call_sal(op_info);
627 }
628 else {
629 if (use_ipi) {
630 /* use an interprocessor interrupt to call SAL */
631 smp_call_function_single(cpu, sn_hwperf_call_sal,
632 op_info, 1);
633 }
634 else {
635 /* migrate the task before calling SAL */
636 save_allowed = current->cpus_allowed;
637 set_cpus_allowed(current, cpumask_of_cpu(cpu));
638 sn_hwperf_call_sal(op_info);
639 set_cpus_allowed(current, save_allowed);
640 }
641 }
642 r = op_info->ret;
643
644 out:
645 return r;
646 }
647
648 /* map SAL hwperf error code to system error code */
649 static int sn_hwperf_map_err(int hwperf_err)
650 {
651 int e;
652
653 switch(hwperf_err) {
654 case SN_HWPERF_OP_OK:
655 e = 0;
656 break;
657
658 case SN_HWPERF_OP_NOMEM:
659 e = -ENOMEM;
660 break;
661
662 case SN_HWPERF_OP_NO_PERM:
663 e = -EPERM;
664 break;
665
666 case SN_HWPERF_OP_IO_ERROR:
667 e = -EIO;
668 break;
669
670 case SN_HWPERF_OP_BUSY:
671 e = -EBUSY;
672 break;
673
674 case SN_HWPERF_OP_RECONFIGURE:
675 e = -EAGAIN;
676 break;
677
678 case SN_HWPERF_OP_INVAL:
679 default:
680 e = -EINVAL;
681 break;
682 }
683
684 return e;
685 }
686
687 /*
688 * ioctl for "sn_hwperf" misc device
689 */
690 static int
691 sn_hwperf_ioctl(struct inode *in, struct file *fp, u32 op, u64 arg)
692 {
693 struct sn_hwperf_ioctl_args a;
694 struct cpuinfo_ia64 *cdata;
695 struct sn_hwperf_object_info *objs;
696 struct sn_hwperf_object_info *cpuobj;
697 struct sn_hwperf_op_info op_info;
698 void *p = NULL;
699 int nobj;
700 char slice;
701 int node;
702 int r;
703 int v0;
704 int i;
705 int j;
706
707 unlock_kernel();
708
709 /* only user requests are allowed here */
710 if ((op & SN_HWPERF_OP_MASK) < 10) {
711 r = -EINVAL;
712 goto error;
713 }
714 r = copy_from_user(&a, (const void __user *)arg,
715 sizeof(struct sn_hwperf_ioctl_args));
716 if (r != 0) {
717 r = -EFAULT;
718 goto error;
719 }
720
721 /*
722 * Allocate memory to hold a kernel copy of the user buffer. The
723 * buffer contents are either copied in or out (or both) of user
724 * space depending on the flags encoded in the requested operation.
725 */
726 if (a.ptr) {
727 p = vmalloc(a.sz);
728 if (!p) {
729 r = -ENOMEM;
730 goto error;
731 }
732 }
733
734 if (op & SN_HWPERF_OP_MEM_COPYIN) {
735 r = copy_from_user(p, (const void __user *)a.ptr, a.sz);
736 if (r != 0) {
737 r = -EFAULT;
738 goto error;
739 }
740 }
741
742 switch (op) {
743 case SN_HWPERF_GET_CPU_INFO:
744 if (a.sz == sizeof(u64)) {
745 /* special case to get size needed */
746 *(u64 *) p = (u64) num_online_cpus() *
747 sizeof(struct sn_hwperf_object_info);
748 } else
749 if (a.sz < num_online_cpus() * sizeof(struct sn_hwperf_object_info)) {
750 r = -ENOMEM;
751 goto error;
752 } else
753 if ((r = sn_hwperf_enum_objects(&nobj, &objs)) == 0) {
754 int cpuobj_index = 0;
755
756 memset(p, 0, a.sz);
757 for (i = 0; i < nobj; i++) {
758 if (!SN_HWPERF_IS_NODE(objs + i))
759 continue;
760 node = sn_hwperf_obj_to_cnode(objs + i);
761 for_each_online_cpu(j) {
762 if (node != cpu_to_node(j))
763 continue;
764 cpuobj = (struct sn_hwperf_object_info *) p + cpuobj_index++;
765 slice = 'a' + cpuid_to_slice(j);
766 cdata = cpu_data(j);
767 cpuobj->id = j;
768 snprintf(cpuobj->name,
769 sizeof(cpuobj->name),
770 "CPU %luMHz %s",
771 cdata->proc_freq / 1000000,
772 cdata->vendor);
773 snprintf(cpuobj->location,
774 sizeof(cpuobj->location),
775 "%s%c", objs[i].location,
776 slice);
777 }
778 }
779
780 vfree(objs);
781 }
782 break;
783
784 case SN_HWPERF_GET_NODE_NASID:
785 if (a.sz != sizeof(u64) ||
786 (node = a.arg) < 0 || !cnode_possible(node)) {
787 r = -EINVAL;
788 goto error;
789 }
790 *(u64 *)p = (u64)cnodeid_to_nasid(node);
791 break;
792
793 case SN_HWPERF_GET_OBJ_NODE:
794 if (a.sz != sizeof(u64) || a.arg < 0) {
795 r = -EINVAL;
796 goto error;
797 }
798 if ((r = sn_hwperf_enum_objects(&nobj, &objs)) == 0) {
799 if (a.arg >= nobj) {
800 r = -EINVAL;
801 vfree(objs);
802 goto error;
803 }
804 if (objs[(i = a.arg)].id != a.arg) {
805 for (i = 0; i < nobj; i++) {
806 if (objs[i].id == a.arg)
807 break;
808 }
809 }
810 if (i == nobj) {
811 r = -EINVAL;
812 vfree(objs);
813 goto error;
814 }
815
816 if (!SN_HWPERF_IS_NODE(objs + i) &&
817 !SN_HWPERF_IS_IONODE(objs + i)) {
818 r = -ENOENT;
819 vfree(objs);
820 goto error;
821 }
822
823 *(u64 *)p = (u64)sn_hwperf_obj_to_cnode(objs + i);
824 vfree(objs);
825 }
826 break;
827
828 case SN_HWPERF_GET_MMRS:
829 case SN_HWPERF_SET_MMRS:
830 case SN_HWPERF_OBJECT_DISTANCE:
831 op_info.p = p;
832 op_info.a = &a;
833 op_info.v0 = &v0;
834 op_info.op = op;
835 r = sn_hwperf_op_cpu(&op_info);
836 if (r) {
837 r = sn_hwperf_map_err(r);
838 a.v0 = v0;
839 goto error;
840 }
841 break;
842
843 default:
844 /* all other ops are a direct SAL call */
845 r = ia64_sn_hwperf_op(sn_hwperf_master_nasid, op,
846 a.arg, a.sz, (u64) p, 0, 0, &v0);
847 if (r) {
848 r = sn_hwperf_map_err(r);
849 goto error;
850 }
851 a.v0 = v0;
852 break;
853 }
854
855 if (op & SN_HWPERF_OP_MEM_COPYOUT) {
856 r = copy_to_user((void __user *)a.ptr, p, a.sz);
857 if (r != 0) {
858 r = -EFAULT;
859 goto error;
860 }
861 }
862
863 error:
864 vfree(p);
865
866 lock_kernel();
867 return r;
868 }
869
870 static const struct file_operations sn_hwperf_fops = {
871 .ioctl = sn_hwperf_ioctl,
872 };
873
874 static struct miscdevice sn_hwperf_dev = {
875 MISC_DYNAMIC_MINOR,
876 "sn_hwperf",
877 &sn_hwperf_fops
878 };
879
880 static int sn_hwperf_init(void)
881 {
882 u64 v;
883 int salr;
884 int e = 0;
885
886 /* single threaded, once-only initialization */
887 mutex_lock(&sn_hwperf_init_mutex);
888
889 if (sn_hwperf_salheap) {
890 mutex_unlock(&sn_hwperf_init_mutex);
891 return e;
892 }
893
894 /*
895 * The PROM code needs a fixed reference node. For convenience the
896 * same node as the console I/O is used.
897 */
898 sn_hwperf_master_nasid = (nasid_t) ia64_sn_get_console_nasid();
899
900 /*
901 * Request the needed size and install the PROM scratch area.
902 * The PROM keeps various tracking bits in this memory area.
903 */
904 salr = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
905 (u64) SN_HWPERF_GET_HEAPSIZE, 0,
906 (u64) sizeof(u64), (u64) &v, 0, 0, NULL);
907 if (salr != SN_HWPERF_OP_OK) {
908 e = -EINVAL;
909 goto out;
910 }
911
912 if ((sn_hwperf_salheap = vmalloc(v)) == NULL) {
913 e = -ENOMEM;
914 goto out;
915 }
916 salr = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
917 SN_HWPERF_INSTALL_HEAP, 0, v,
918 (u64) sn_hwperf_salheap, 0, 0, NULL);
919 if (salr != SN_HWPERF_OP_OK) {
920 e = -EINVAL;
921 goto out;
922 }
923
924 salr = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
925 SN_HWPERF_OBJECT_COUNT, 0,
926 sizeof(u64), (u64) &v, 0, 0, NULL);
927 if (salr != SN_HWPERF_OP_OK) {
928 e = -EINVAL;
929 goto out;
930 }
931 sn_hwperf_obj_cnt = (int)v;
932
933 out:
934 if (e < 0 && sn_hwperf_salheap) {
935 vfree(sn_hwperf_salheap);
936 sn_hwperf_salheap = NULL;
937 sn_hwperf_obj_cnt = 0;
938 }
939 mutex_unlock(&sn_hwperf_init_mutex);
940 return e;
941 }
942
943 int sn_topology_open(struct inode *inode, struct file *file)
944 {
945 int e;
946 struct seq_file *seq;
947 struct sn_hwperf_object_info *objbuf;
948 int nobj;
949
950 if ((e = sn_hwperf_enum_objects(&nobj, &objbuf)) == 0) {
951 e = seq_open(file, &sn_topology_seq_ops);
952 seq = file->private_data;
953 seq->private = objbuf;
954 }
955
956 return e;
957 }
958
959 int sn_topology_release(struct inode *inode, struct file *file)
960 {
961 struct seq_file *seq = file->private_data;
962
963 vfree(seq->private);
964 return seq_release(inode, file);
965 }
966
967 int sn_hwperf_get_nearest_node(cnodeid_t node,
968 cnodeid_t *near_mem_node, cnodeid_t *near_cpu_node)
969 {
970 int e;
971 int nobj;
972 struct sn_hwperf_object_info *objbuf;
973
974 if ((e = sn_hwperf_enum_objects(&nobj, &objbuf)) == 0) {
975 e = sn_hwperf_get_nearest_node_objdata(objbuf, nobj,
976 node, near_mem_node, near_cpu_node);
977 vfree(objbuf);
978 }
979
980 return e;
981 }
982
983 static int __devinit sn_hwperf_misc_register_init(void)
984 {
985 int e;
986
987 if (!ia64_platform_is("sn2"))
988 return 0;
989
990 sn_hwperf_init();
991
992 /*
993 * Register a dynamic misc device for hwperf ioctls. Platforms
994 * supporting hotplug will create /dev/sn_hwperf, else user
995 * can to look up the minor number in /proc/misc.
996 */
997 if ((e = misc_register(&sn_hwperf_dev)) != 0) {
998 printk(KERN_ERR "sn_hwperf_misc_register_init: failed to "
999 "register misc device for \"%s\"\n", sn_hwperf_dev.name);
1000 }
1001
1002 return e;
1003 }
1004
1005 device_initcall(sn_hwperf_misc_register_init); /* after misc_init() */
1006 EXPORT_SYMBOL(sn_hwperf_get_nearest_node);
This page took 0.052402 seconds and 6 git commands to generate.