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