Merge branch 'master' of /usr/src/ntfs-2.6/
[linux-2.6] / 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) 2004-2005 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 <asm/processor.h>
36 #include <asm/topology.h>
37 #include <asm/smp.h>
38 #include <asm/semaphore.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 DECLARE_MUTEX(sn_hwperf_init_mutex);
53
54 static int sn_hwperf_enum_objects(int *nobj, struct sn_hwperf_object_info **ret)
55 {
56         int e;
57         u64 sz;
58         struct sn_hwperf_object_info *objbuf = NULL;
59
60         if ((e = sn_hwperf_init()) < 0) {
61                 printk(KERN_ERR "sn_hwperf_init failed: err %d\n", e);
62                 goto out;
63         }
64
65         sz = sn_hwperf_obj_cnt * sizeof(struct sn_hwperf_object_info);
66         if ((objbuf = (struct sn_hwperf_object_info *) vmalloc(sz)) == NULL) {
67                 printk("sn_hwperf_enum_objects: vmalloc(%d) failed\n", (int)sz);
68                 e = -ENOMEM;
69                 goto out;
70         }
71
72         e = ia64_sn_hwperf_op(sn_hwperf_master_nasid, SN_HWPERF_ENUM_OBJECTS,
73                 0, sz, (u64) objbuf, 0, 0, NULL);
74         if (e != SN_HWPERF_OP_OK) {
75                 e = -EINVAL;
76                 vfree(objbuf);
77         }
78
79 out:
80         *nobj = sn_hwperf_obj_cnt;
81         *ret = objbuf;
82         return e;
83 }
84
85 static int sn_hwperf_location_to_bpos(char *location,
86         int *rack, int *bay, int *slot, int *slab)
87 {
88         char type;
89
90         /* first scan for an old style geoid string */
91         if (sscanf(location, "%03d%c%02d#%d",
92                 rack, &type, bay, slab) == 4)
93                 *slot = 0; 
94         else /* scan for a new bladed geoid string */
95         if (sscanf(location, "%03d%c%02d^%02d#%d",
96                 rack, &type, bay, slot, slab) != 5)
97                 return -1; 
98         /* success */
99         return 0;
100 }
101
102 static int sn_hwperf_geoid_to_cnode(char *location)
103 {
104         int cnode;
105         geoid_t geoid;
106         moduleid_t module_id;
107         int rack, bay, slot, slab;
108         int this_rack, this_bay, this_slot, this_slab;
109
110         if (sn_hwperf_location_to_bpos(location, &rack, &bay, &slot, &slab))
111                 return -1;
112
113         for_each_node(cnode) {
114                 geoid = cnodeid_get_geoid(cnode);
115                 module_id = geo_module(geoid);
116                 this_rack = MODULE_GET_RACK(module_id);
117                 this_bay = MODULE_GET_BPOS(module_id);
118                 this_slot = geo_slot(geoid);
119                 this_slab = geo_slab(geoid);
120                 if (rack == this_rack && bay == this_bay &&
121                         slot == this_slot && slab == this_slab) {
122                         break;
123                 }
124         }
125
126         return node_possible(cnode) ? cnode : -1;
127 }
128
129 static int sn_hwperf_obj_to_cnode(struct sn_hwperf_object_info * obj)
130 {
131         if (!SN_HWPERF_IS_NODE(obj) && !SN_HWPERF_IS_IONODE(obj))
132                 BUG();
133         if (!obj->sn_hwp_this_part)
134                 return -1;
135         return sn_hwperf_geoid_to_cnode(obj->location);
136 }
137
138 static int sn_hwperf_generic_ordinal(struct sn_hwperf_object_info *obj,
139                                 struct sn_hwperf_object_info *objs)
140 {
141         int ordinal;
142         struct sn_hwperf_object_info *p;
143
144         for (ordinal=0, p=objs; p != obj; p++) {
145                 if (SN_HWPERF_FOREIGN(p))
146                         continue;
147                 if (SN_HWPERF_SAME_OBJTYPE(p, obj))
148                         ordinal++;
149         }
150
151         return ordinal;
152 }
153
154 static const char *slabname_node =      "node"; /* SHub asic */
155 static const char *slabname_ionode =    "ionode"; /* TIO asic */
156 static const char *slabname_router =    "router"; /* NL3R or NL4R */
157 static const char *slabname_other =     "other"; /* unknown asic */
158
159 static const char *sn_hwperf_get_slabname(struct sn_hwperf_object_info *obj,
160                         struct sn_hwperf_object_info *objs, int *ordinal)
161 {
162         int isnode;
163         const char *slabname = slabname_other;
164
165         if ((isnode = SN_HWPERF_IS_NODE(obj)) || SN_HWPERF_IS_IONODE(obj)) {
166                 slabname = isnode ? slabname_node : slabname_ionode;
167                 *ordinal = sn_hwperf_obj_to_cnode(obj);
168         }
169         else {
170                 *ordinal = sn_hwperf_generic_ordinal(obj, objs);
171                 if (SN_HWPERF_IS_ROUTER(obj))
172                         slabname = slabname_router;
173         }
174
175         return slabname;
176 }
177
178 static void print_pci_topology(struct seq_file *s)
179 {
180         char *p;
181         size_t sz;
182         int e;
183
184         for (sz = PAGE_SIZE; sz < 16 * PAGE_SIZE; sz += PAGE_SIZE) {
185                 if (!(p = (char *)kmalloc(sz, GFP_KERNEL)))
186                         break;
187                 e = ia64_sn_ioif_get_pci_topology(__pa(p), sz);
188                 if (e == SALRET_OK)
189                         seq_puts(s, p);
190                 kfree(p);
191                 if (e == SALRET_OK || e == SALRET_NOT_IMPLEMENTED)
192                         break;
193         }
194 }
195
196 static inline int sn_hwperf_has_cpus(cnodeid_t node)
197 {
198         return node_online(node) && nr_cpus_node(node);
199 }
200
201 static inline int sn_hwperf_has_mem(cnodeid_t node)
202 {
203         return node_online(node) && NODE_DATA(node)->node_present_pages;
204 }
205
206 static struct sn_hwperf_object_info *
207 sn_hwperf_findobj_id(struct sn_hwperf_object_info *objbuf,
208         int nobj, int id)
209 {
210         int i;
211         struct sn_hwperf_object_info *p = objbuf;
212
213         for (i=0; i < nobj; i++, p++) {
214                 if (p->id == id)
215                         return p;
216         }
217
218         return NULL;
219
220 }
221
222 static int sn_hwperf_get_nearest_node_objdata(struct sn_hwperf_object_info *objbuf,
223         int nobj, cnodeid_t node, cnodeid_t *near_mem_node, cnodeid_t *near_cpu_node)
224 {
225         int e;
226         struct sn_hwperf_object_info *nodeobj = NULL;
227         struct sn_hwperf_object_info *op;
228         struct sn_hwperf_object_info *dest;
229         struct sn_hwperf_object_info *router;
230         struct sn_hwperf_port_info ptdata[16];
231         int sz, i, j;
232         cnodeid_t c;
233         int found_mem = 0;
234         int found_cpu = 0;
235
236         if (!node_possible(node))
237                 return -EINVAL;
238
239         if (sn_hwperf_has_cpus(node)) {
240                 if (near_cpu_node)
241                         *near_cpu_node = node;
242                 found_cpu++;
243         }
244
245         if (sn_hwperf_has_mem(node)) {
246                 if (near_mem_node)
247                         *near_mem_node = node;
248                 found_mem++;
249         }
250
251         if (found_cpu && found_mem)
252                 return 0; /* trivially successful */
253
254         /* find the argument node object */
255         for (i=0, op=objbuf; i < nobj; i++, op++) {
256                 if (!SN_HWPERF_IS_NODE(op) && !SN_HWPERF_IS_IONODE(op))
257                         continue;
258                 if (node == sn_hwperf_obj_to_cnode(op)) {
259                         nodeobj = op;
260                         break;
261                 }
262         }
263         if (!nodeobj) {
264                 e = -ENOENT;
265                 goto err;
266         }
267
268         /* get it's interconnect topology */
269         sz = op->ports * sizeof(struct sn_hwperf_port_info);
270         if (sz > sizeof(ptdata))
271                 BUG();
272         e = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
273                               SN_HWPERF_ENUM_PORTS, nodeobj->id, sz,
274                               (u64)&ptdata, 0, 0, NULL);
275         if (e != SN_HWPERF_OP_OK) {
276                 e = -EINVAL;
277                 goto err;
278         }
279
280         /* find nearest node with cpus and nearest memory */
281         for (router=NULL, j=0; j < op->ports; j++) {
282                 dest = sn_hwperf_findobj_id(objbuf, nobj, ptdata[j].conn_id);
283                 if (!dest || SN_HWPERF_FOREIGN(dest) ||
284                     !SN_HWPERF_IS_NODE(dest) || SN_HWPERF_IS_IONODE(dest)) {
285                         continue;
286                 }
287                 c = sn_hwperf_obj_to_cnode(dest);
288                 if (!found_cpu && sn_hwperf_has_cpus(c)) {
289                         if (near_cpu_node)
290                                 *near_cpu_node = c;
291                         found_cpu++;
292                 }
293                 if (!found_mem && sn_hwperf_has_mem(c)) {
294                         if (near_mem_node)
295                                 *near_mem_node = c;
296                         found_mem++;
297                 }
298                 if (SN_HWPERF_IS_ROUTER(dest))
299                         router = dest;
300         }
301
302         if (router && (!found_cpu || !found_mem)) {
303                 /* search for a node connected to the same router */
304                 sz = router->ports * sizeof(struct sn_hwperf_port_info);
305                 if (sz > sizeof(ptdata))
306                         BUG();
307                 e = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
308                                       SN_HWPERF_ENUM_PORTS, router->id, sz,
309                                       (u64)&ptdata, 0, 0, NULL);
310                 if (e != SN_HWPERF_OP_OK) {
311                         e = -EINVAL;
312                         goto err;
313                 }
314                 for (j=0; j < router->ports; j++) {
315                         dest = sn_hwperf_findobj_id(objbuf, nobj,
316                                 ptdata[j].conn_id);
317                         if (!dest || dest->id == node ||
318                             SN_HWPERF_FOREIGN(dest) ||
319                             !SN_HWPERF_IS_NODE(dest) ||
320                             SN_HWPERF_IS_IONODE(dest)) {
321                                 continue;
322                         }
323                         c = sn_hwperf_obj_to_cnode(dest);
324                         if (!found_cpu && sn_hwperf_has_cpus(c)) {
325                                 if (near_cpu_node)
326                                         *near_cpu_node = c;
327                                 found_cpu++;
328                         }
329                         if (!found_mem && sn_hwperf_has_mem(c)) {
330                                 if (near_mem_node)
331                                         *near_mem_node = c;
332                                 found_mem++;
333                         }
334                         if (found_cpu && found_mem)
335                                 break;
336                 }
337         }
338
339         if (!found_cpu || !found_mem) {
340                 /* resort to _any_ node with CPUs and memory */
341                 for (i=0, op=objbuf; i < nobj; i++, op++) {
342                         if (SN_HWPERF_FOREIGN(op) ||
343                             SN_HWPERF_IS_IONODE(op) ||
344                             !SN_HWPERF_IS_NODE(op)) {
345                                 continue;
346                         }
347                         c = sn_hwperf_obj_to_cnode(op);
348                         if (!found_cpu && sn_hwperf_has_cpus(c)) {
349                                 if (near_cpu_node)
350                                         *near_cpu_node = c;
351                                 found_cpu++;
352                         }
353                         if (!found_mem && sn_hwperf_has_mem(c)) {
354                                 if (near_mem_node)
355                                         *near_mem_node = c;
356                                 found_mem++;
357                         }
358                         if (found_cpu && found_mem)
359                                 break;
360                 }
361         }
362
363         if (!found_cpu || !found_mem)
364                 e = -ENODATA;
365
366 err:
367         return e;
368 }
369
370
371 static int sn_topology_show(struct seq_file *s, void *d)
372 {
373         int sz;
374         int pt;
375         int e = 0;
376         int i;
377         int j;
378         const char *slabname;
379         int ordinal;
380         cpumask_t cpumask;
381         char slice;
382         struct cpuinfo_ia64 *c;
383         struct sn_hwperf_port_info *ptdata;
384         struct sn_hwperf_object_info *p;
385         struct sn_hwperf_object_info *obj = d;  /* this object */
386         struct sn_hwperf_object_info *objs = s->private; /* all objects */
387         u8 shubtype;
388         u8 system_size;
389         u8 sharing_size;
390         u8 partid;
391         u8 coher;
392         u8 nasid_shift;
393         u8 region_size;
394         u16 nasid_mask;
395         int nasid_msb;
396
397         if (obj == objs) {
398                 seq_printf(s, "# sn_topology version 2\n");
399                 seq_printf(s, "# objtype ordinal location partition"
400                         " [attribute value [, ...]]\n");
401
402                 if (ia64_sn_get_sn_info(0,
403                         &shubtype, &nasid_mask, &nasid_shift, &system_size,
404                         &sharing_size, &partid, &coher, &region_size))
405                         BUG();
406                 for (nasid_msb=63; nasid_msb > 0; nasid_msb--) {
407                         if (((u64)nasid_mask << nasid_shift) & (1ULL << nasid_msb))
408                                 break;
409                 }
410                 seq_printf(s, "partition %u %s local "
411                         "shubtype %s, "
412                         "nasid_mask 0x%016lx, "
413                         "nasid_bits %d:%d, "
414                         "system_size %d, "
415                         "sharing_size %d, "
416                         "coherency_domain %d, "
417                         "region_size %d\n",
418
419                         partid, system_utsname.nodename,
420                         shubtype ? "shub2" : "shub1", 
421                         (u64)nasid_mask << nasid_shift, nasid_msb, nasid_shift,
422                         system_size, sharing_size, coher, region_size);
423
424                 print_pci_topology(s);
425         }
426
427         if (SN_HWPERF_FOREIGN(obj)) {
428                 /* private in another partition: not interesting */
429                 return 0;
430         }
431
432         for (i = 0; i < SN_HWPERF_MAXSTRING && obj->name[i]; i++) {
433                 if (obj->name[i] == ' ')
434                         obj->name[i] = '_';
435         }
436
437         slabname = sn_hwperf_get_slabname(obj, objs, &ordinal);
438         seq_printf(s, "%s %d %s %s asic %s", slabname, ordinal, obj->location,
439                 obj->sn_hwp_this_part ? "local" : "shared", obj->name);
440
441         if (!SN_HWPERF_IS_NODE(obj) && !SN_HWPERF_IS_IONODE(obj))
442                 seq_putc(s, '\n');
443         else {
444                 cnodeid_t near_mem = -1;
445                 cnodeid_t near_cpu = -1;
446
447                 seq_printf(s, ", nasid 0x%x", cnodeid_to_nasid(ordinal));
448
449                 if (sn_hwperf_get_nearest_node_objdata(objs, sn_hwperf_obj_cnt,
450                         ordinal, &near_mem, &near_cpu) == 0) {
451                         seq_printf(s, ", near_mem_nodeid %d, near_cpu_nodeid %d",
452                                 near_mem, near_cpu);
453                 }
454
455                 if (!SN_HWPERF_IS_IONODE(obj)) {
456                         for_each_online_node(i) {
457                                 seq_printf(s, i ? ":%d" : ", dist %d",
458                                         node_distance(ordinal, i));
459                         }
460                 }
461
462                 seq_putc(s, '\n');
463
464                 /*
465                  * CPUs on this node, if any
466                  */
467                 cpumask = node_to_cpumask(ordinal);
468                 for_each_online_cpu(i) {
469                         if (cpu_isset(i, cpumask)) {
470                                 slice = 'a' + cpuid_to_slice(i);
471                                 c = cpu_data(i);
472                                 seq_printf(s, "cpu %d %s%c local"
473                                         " freq %luMHz, arch ia64",
474                                         i, obj->location, slice,
475                                         c->proc_freq / 1000000);
476                                 for_each_online_cpu(j) {
477                                         seq_printf(s, j ? ":%d" : ", dist %d",
478                                                 node_distance(
479                                                     cpu_to_node(i),
480                                                     cpu_to_node(j)));
481                                 }
482                                 seq_putc(s, '\n');
483                         }
484                 }
485         }
486
487         if (obj->ports) {
488                 /*
489                  * numalink ports
490                  */
491                 sz = obj->ports * sizeof(struct sn_hwperf_port_info);
492                 if ((ptdata = vmalloc(sz)) == NULL)
493                         return -ENOMEM;
494                 e = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
495                                       SN_HWPERF_ENUM_PORTS, obj->id, sz,
496                                       (u64) ptdata, 0, 0, NULL);
497                 if (e != SN_HWPERF_OP_OK)
498                         return -EINVAL;
499                 for (ordinal=0, p=objs; p != obj; p++) {
500                         if (!SN_HWPERF_FOREIGN(p))
501                                 ordinal += p->ports;
502                 }
503                 for (pt = 0; pt < obj->ports; pt++) {
504                         for (p = objs, i = 0; i < sn_hwperf_obj_cnt; i++, p++) {
505                                 if (ptdata[pt].conn_id == p->id) {
506                                         break;
507                                 }
508                         }
509                         seq_printf(s, "numalink %d %s-%d",
510                             ordinal+pt, obj->location, ptdata[pt].port);
511
512                         if (i >= sn_hwperf_obj_cnt) {
513                                 /* no connection */
514                                 seq_puts(s, " local endpoint disconnected"
515                                             ", protocol unknown\n");
516                                 continue;
517                         }
518
519                         if (obj->sn_hwp_this_part && p->sn_hwp_this_part)
520                                 /* both ends local to this partition */
521                                 seq_puts(s, " local");
522                         else if (!obj->sn_hwp_this_part && !p->sn_hwp_this_part)
523                                 /* both ends of the link in foreign partiton */
524                                 seq_puts(s, " foreign");
525                         else
526                                 /* link straddles a partition */
527                                 seq_puts(s, " shared");
528
529                         /*
530                          * Unlikely, but strictly should query the LLP config
531                          * registers because an NL4R can be configured to run
532                          * NL3 protocol, even when not talking to an NL3 router.
533                          * Ditto for node-node.
534                          */
535                         seq_printf(s, " endpoint %s-%d, protocol %s\n",
536                                 p->location, ptdata[pt].conn_port,
537                                 (SN_HWPERF_IS_NL3ROUTER(obj) ||
538                                 SN_HWPERF_IS_NL3ROUTER(p)) ?  "LLP3" : "LLP4");
539                 }
540                 vfree(ptdata);
541         }
542
543         return 0;
544 }
545
546 static void *sn_topology_start(struct seq_file *s, loff_t * pos)
547 {
548         struct sn_hwperf_object_info *objs = s->private;
549
550         if (*pos < sn_hwperf_obj_cnt)
551                 return (void *)(objs + *pos);
552
553         return NULL;
554 }
555
556 static void *sn_topology_next(struct seq_file *s, void *v, loff_t * pos)
557 {
558         ++*pos;
559         return sn_topology_start(s, pos);
560 }
561
562 static void sn_topology_stop(struct seq_file *m, void *v)
563 {
564         return;
565 }
566
567 /*
568  * /proc/sgi_sn/sn_topology, read-only using seq_file
569  */
570 static struct seq_operations sn_topology_seq_ops = {
571         .start = sn_topology_start,
572         .next = sn_topology_next,
573         .stop = sn_topology_stop,
574         .show = sn_topology_show
575 };
576
577 struct sn_hwperf_op_info {
578         u64 op;
579         struct sn_hwperf_ioctl_args *a;
580         void *p;
581         int *v0;
582         int ret;
583 };
584
585 static void sn_hwperf_call_sal(void *info)
586 {
587         struct sn_hwperf_op_info *op_info = info;
588         int r;
589
590         r = ia64_sn_hwperf_op(sn_hwperf_master_nasid, op_info->op,
591                       op_info->a->arg, op_info->a->sz,
592                       (u64) op_info->p, 0, 0, op_info->v0);
593         op_info->ret = r;
594 }
595
596 static int sn_hwperf_op_cpu(struct sn_hwperf_op_info *op_info)
597 {
598         u32 cpu;
599         u32 use_ipi;
600         int r = 0;
601         cpumask_t save_allowed;
602         
603         cpu = (op_info->a->arg & SN_HWPERF_ARG_CPU_MASK) >> 32;
604         use_ipi = op_info->a->arg & SN_HWPERF_ARG_USE_IPI_MASK;
605         op_info->a->arg &= SN_HWPERF_ARG_OBJID_MASK;
606
607         if (cpu != SN_HWPERF_ARG_ANY_CPU) {
608                 if (cpu >= num_online_cpus() || !cpu_online(cpu)) {
609                         r = -EINVAL;
610                         goto out;
611                 }
612         }
613
614         if (cpu == SN_HWPERF_ARG_ANY_CPU || cpu == get_cpu()) {
615                 /* don't care, or already on correct cpu */
616                 sn_hwperf_call_sal(op_info);
617         }
618         else {
619                 if (use_ipi) {
620                         /* use an interprocessor interrupt to call SAL */
621                         smp_call_function_single(cpu, sn_hwperf_call_sal,
622                                 op_info, 1, 1);
623                 }
624                 else {
625                         /* migrate the task before calling SAL */ 
626                         save_allowed = current->cpus_allowed;
627                         set_cpus_allowed(current, cpumask_of_cpu(cpu));
628                         sn_hwperf_call_sal(op_info);
629                         set_cpus_allowed(current, save_allowed);
630                 }
631         }
632         r = op_info->ret;
633
634 out:
635         return r;
636 }
637
638 /* map SAL hwperf error code to system error code */
639 static int sn_hwperf_map_err(int hwperf_err)
640 {
641         int e;
642
643         switch(hwperf_err) {
644         case SN_HWPERF_OP_OK:
645                 e = 0;
646                 break;
647
648         case SN_HWPERF_OP_NOMEM:
649                 e = -ENOMEM;
650                 break;
651
652         case SN_HWPERF_OP_NO_PERM:
653                 e = -EPERM;
654                 break;
655
656         case SN_HWPERF_OP_IO_ERROR:
657                 e = -EIO;
658                 break;
659
660         case SN_HWPERF_OP_BUSY:
661                 e = -EBUSY;
662                 break;
663
664         case SN_HWPERF_OP_RECONFIGURE:
665                 e = -EAGAIN;
666                 break;
667
668         case SN_HWPERF_OP_INVAL:
669         default:
670                 e = -EINVAL;
671                 break;
672         }
673
674         return e;
675 }
676
677 /*
678  * ioctl for "sn_hwperf" misc device
679  */
680 static int
681 sn_hwperf_ioctl(struct inode *in, struct file *fp, u32 op, u64 arg)
682 {
683         struct sn_hwperf_ioctl_args a;
684         struct cpuinfo_ia64 *cdata;
685         struct sn_hwperf_object_info *objs;
686         struct sn_hwperf_object_info *cpuobj;
687         struct sn_hwperf_op_info op_info;
688         void *p = NULL;
689         int nobj;
690         char slice;
691         int node;
692         int r;
693         int v0;
694         int i;
695         int j;
696
697         unlock_kernel();
698
699         /* only user requests are allowed here */
700         if ((op & SN_HWPERF_OP_MASK) < 10) {
701                 r = -EINVAL;
702                 goto error;
703         }
704         r = copy_from_user(&a, (const void __user *)arg,
705                 sizeof(struct sn_hwperf_ioctl_args));
706         if (r != 0) {
707                 r = -EFAULT;
708                 goto error;
709         }
710
711         /*
712          * Allocate memory to hold a kernel copy of the user buffer. The
713          * buffer contents are either copied in or out (or both) of user
714          * space depending on the flags encoded in the requested operation.
715          */
716         if (a.ptr) {
717                 p = vmalloc(a.sz);
718                 if (!p) {
719                         r = -ENOMEM;
720                         goto error;
721                 }
722         }
723
724         if (op & SN_HWPERF_OP_MEM_COPYIN) {
725                 r = copy_from_user(p, (const void __user *)a.ptr, a.sz);
726                 if (r != 0) {
727                         r = -EFAULT;
728                         goto error;
729                 }
730         }
731
732         switch (op) {
733         case SN_HWPERF_GET_CPU_INFO:
734                 if (a.sz == sizeof(u64)) {
735                         /* special case to get size needed */
736                         *(u64 *) p = (u64) num_online_cpus() *
737                                 sizeof(struct sn_hwperf_object_info);
738                 } else
739                 if (a.sz < num_online_cpus() * sizeof(struct sn_hwperf_object_info)) {
740                         r = -ENOMEM;
741                         goto error;
742                 } else
743                 if ((r = sn_hwperf_enum_objects(&nobj, &objs)) == 0) {
744                         memset(p, 0, a.sz);
745                         for (i = 0; i < nobj; i++) {
746                                 if (!SN_HWPERF_IS_NODE(objs + i))
747                                         continue;
748                                 node = sn_hwperf_obj_to_cnode(objs + i);
749                                 for_each_online_cpu(j) {
750                                         if (node != cpu_to_node(j))
751                                                 continue;
752                                         cpuobj = (struct sn_hwperf_object_info *) p + j;
753                                         slice = 'a' + cpuid_to_slice(j);
754                                         cdata = cpu_data(j);
755                                         cpuobj->id = j;
756                                         snprintf(cpuobj->name,
757                                                  sizeof(cpuobj->name),
758                                                  "CPU %luMHz %s",
759                                                  cdata->proc_freq / 1000000,
760                                                  cdata->vendor);
761                                         snprintf(cpuobj->location,
762                                                  sizeof(cpuobj->location),
763                                                  "%s%c", objs[i].location,
764                                                  slice);
765                                 }
766                         }
767
768                         vfree(objs);
769                 }
770                 break;
771
772         case SN_HWPERF_GET_NODE_NASID:
773                 if (a.sz != sizeof(u64) ||
774                    (node = a.arg) < 0 || !node_possible(node)) {
775                         r = -EINVAL;
776                         goto error;
777                 }
778                 *(u64 *)p = (u64)cnodeid_to_nasid(node);
779                 break;
780
781         case SN_HWPERF_GET_OBJ_NODE:
782                 if (a.sz != sizeof(u64) || a.arg < 0) {
783                         r = -EINVAL;
784                         goto error;
785                 }
786                 if ((r = sn_hwperf_enum_objects(&nobj, &objs)) == 0) {
787                         if (a.arg >= nobj) {
788                                 r = -EINVAL;
789                                 vfree(objs);
790                                 goto error;
791                         }
792                         if (objs[(i = a.arg)].id != a.arg) {
793                                 for (i = 0; i < nobj; i++) {
794                                         if (objs[i].id == a.arg)
795                                                 break;
796                                 }
797                         }
798                         if (i == nobj) {
799                                 r = -EINVAL;
800                                 vfree(objs);
801                                 goto error;
802                         }
803
804                         if (!SN_HWPERF_IS_NODE(objs + i) &&
805                             !SN_HWPERF_IS_IONODE(objs + i)) {
806                                 r = -ENOENT;
807                                 vfree(objs);
808                                 goto error;
809                         }
810
811                         *(u64 *)p = (u64)sn_hwperf_obj_to_cnode(objs + i);
812                         vfree(objs);
813                 }
814                 break;
815
816         case SN_HWPERF_GET_MMRS:
817         case SN_HWPERF_SET_MMRS:
818         case SN_HWPERF_OBJECT_DISTANCE:
819                 op_info.p = p;
820                 op_info.a = &a;
821                 op_info.v0 = &v0;
822                 op_info.op = op;
823                 r = sn_hwperf_op_cpu(&op_info);
824                 if (r) {
825                         r = sn_hwperf_map_err(r);
826                         a.v0 = v0;
827                         goto error;
828                 }
829                 break;
830
831         default:
832                 /* all other ops are a direct SAL call */
833                 r = ia64_sn_hwperf_op(sn_hwperf_master_nasid, op,
834                               a.arg, a.sz, (u64) p, 0, 0, &v0);
835                 if (r) {
836                         r = sn_hwperf_map_err(r);
837                         goto error;
838                 }
839                 a.v0 = v0;
840                 break;
841         }
842
843         if (op & SN_HWPERF_OP_MEM_COPYOUT) {
844                 r = copy_to_user((void __user *)a.ptr, p, a.sz);
845                 if (r != 0) {
846                         r = -EFAULT;
847                         goto error;
848                 }
849         }
850
851 error:
852         vfree(p);
853
854         lock_kernel();
855         return r;
856 }
857
858 static struct file_operations sn_hwperf_fops = {
859         .ioctl = sn_hwperf_ioctl,
860 };
861
862 static struct miscdevice sn_hwperf_dev = {
863         MISC_DYNAMIC_MINOR,
864         "sn_hwperf",
865         &sn_hwperf_fops
866 };
867
868 static int sn_hwperf_init(void)
869 {
870         u64 v;
871         int salr;
872         int e = 0;
873
874         /* single threaded, once-only initialization */
875         down(&sn_hwperf_init_mutex);
876
877         if (sn_hwperf_salheap) {
878                 up(&sn_hwperf_init_mutex);
879                 return e;
880         }
881
882         /*
883          * The PROM code needs a fixed reference node. For convenience the
884          * same node as the console I/O is used.
885          */
886         sn_hwperf_master_nasid = (nasid_t) ia64_sn_get_console_nasid();
887
888         /*
889          * Request the needed size and install the PROM scratch area.
890          * The PROM keeps various tracking bits in this memory area.
891          */
892         salr = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
893                                  (u64) SN_HWPERF_GET_HEAPSIZE, 0,
894                                  (u64) sizeof(u64), (u64) &v, 0, 0, NULL);
895         if (salr != SN_HWPERF_OP_OK) {
896                 e = -EINVAL;
897                 goto out;
898         }
899
900         if ((sn_hwperf_salheap = vmalloc(v)) == NULL) {
901                 e = -ENOMEM;
902                 goto out;
903         }
904         salr = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
905                                  SN_HWPERF_INSTALL_HEAP, 0, v,
906                                  (u64) sn_hwperf_salheap, 0, 0, NULL);
907         if (salr != SN_HWPERF_OP_OK) {
908                 e = -EINVAL;
909                 goto out;
910         }
911
912         salr = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
913                                  SN_HWPERF_OBJECT_COUNT, 0,
914                                  sizeof(u64), (u64) &v, 0, 0, NULL);
915         if (salr != SN_HWPERF_OP_OK) {
916                 e = -EINVAL;
917                 goto out;
918         }
919         sn_hwperf_obj_cnt = (int)v;
920
921 out:
922         if (e < 0 && sn_hwperf_salheap) {
923                 vfree(sn_hwperf_salheap);
924                 sn_hwperf_salheap = NULL;
925                 sn_hwperf_obj_cnt = 0;
926         }
927         up(&sn_hwperf_init_mutex);
928         return e;
929 }
930
931 int sn_topology_open(struct inode *inode, struct file *file)
932 {
933         int e;
934         struct seq_file *seq;
935         struct sn_hwperf_object_info *objbuf;
936         int nobj;
937
938         if ((e = sn_hwperf_enum_objects(&nobj, &objbuf)) == 0) {
939                 e = seq_open(file, &sn_topology_seq_ops);
940                 seq = file->private_data;
941                 seq->private = objbuf;
942         }
943
944         return e;
945 }
946
947 int sn_topology_release(struct inode *inode, struct file *file)
948 {
949         struct seq_file *seq = file->private_data;
950
951         vfree(seq->private);
952         return seq_release(inode, file);
953 }
954
955 int sn_hwperf_get_nearest_node(cnodeid_t node,
956         cnodeid_t *near_mem_node, cnodeid_t *near_cpu_node)
957 {
958         int e;
959         int nobj;
960         struct sn_hwperf_object_info *objbuf;
961
962         if ((e = sn_hwperf_enum_objects(&nobj, &objbuf)) == 0) {
963                 e = sn_hwperf_get_nearest_node_objdata(objbuf, nobj,
964                         node, near_mem_node, near_cpu_node);
965                 vfree(objbuf);
966         }
967
968         return e;
969 }
970
971 static int __devinit sn_hwperf_misc_register_init(void)
972 {
973         int e;
974
975         sn_hwperf_init();
976
977         /*
978          * Register a dynamic misc device for hwperf ioctls. Platforms
979          * supporting hotplug will create /dev/sn_hwperf, else user
980          * can to look up the minor number in /proc/misc.
981          */
982         if ((e = misc_register(&sn_hwperf_dev)) != 0) {
983                 printk(KERN_ERR "sn_hwperf_misc_register_init: failed to "
984                 "register misc device for \"%s\"\n", sn_hwperf_dev.name);
985         }
986
987         return e;
988 }
989
990 device_initcall(sn_hwperf_misc_register_init); /* after misc_init() */
991 EXPORT_SYMBOL(sn_hwperf_get_nearest_node);