[PATCH] mark struct file_operations const 2
[linux-2.6/linux-acpi-2.6/ibm-acpi-2.6.git] / arch / ia64 / sn / kernel / sn2 / sn_hwperf.c
blob6da9854751cd9be20d0035fa32b037f6ef6590c0
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.
6 * Copyright (C) 2004-2006 Silicon Graphics, Inc. All rights reserved.
8 * SGI Altix topology and hardware performance monitoring API.
9 * Mark Goodwin <markgw@sgi.com>.
11 * Creates /proc/sgi_sn/sn_topology (read-only) to export
12 * info about Altix nodes, routers, CPUs and NumaLink
13 * interconnection/topology.
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.
22 * This API is used by SGI Performance Co-Pilot and other
23 * tools, see http://oss.sgi.com/projects/pcp
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>
37 #include <asm/processor.h>
38 #include <asm/topology.h>
39 #include <asm/semaphore.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>
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 DECLARE_MUTEX(sn_hwperf_init_mutex);
55 #define cnode_possible(n) ((n) < num_cnodes)
57 static int sn_hwperf_enum_objects(int *nobj, struct sn_hwperf_object_info **ret)
59 int e;
60 u64 sz;
61 struct sn_hwperf_object_info *objbuf = NULL;
63 if ((e = sn_hwperf_init()) < 0) {
64 printk(KERN_ERR "sn_hwperf_init failed: err %d\n", e);
65 goto out;
68 sz = sn_hwperf_obj_cnt * sizeof(struct sn_hwperf_object_info);
69 if ((objbuf = (struct sn_hwperf_object_info *) vmalloc(sz)) == NULL) {
70 printk("sn_hwperf_enum_objects: vmalloc(%d) failed\n", (int)sz);
71 e = -ENOMEM;
72 goto out;
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);
82 out:
83 *nobj = sn_hwperf_obj_cnt;
84 *ret = objbuf;
85 return e;
88 static int sn_hwperf_location_to_bpos(char *location,
89 int *rack, int *bay, int *slot, int *slab)
91 char type;
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;
105 static int sn_hwperf_geoid_to_cnode(char *location)
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;
113 if (sn_hwperf_location_to_bpos(location, &rack, &bay, &slot, &slab))
114 return -1;
117 * FIXME: replace with cleaner for_each_XXX macro which addresses
118 * both compute and IO nodes once ACPI3.0 is available.
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;
133 return cnode_possible(cnode) ? cnode : -1;
136 static int sn_hwperf_obj_to_cnode(struct sn_hwperf_object_info * obj)
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);
145 static int sn_hwperf_generic_ordinal(struct sn_hwperf_object_info *obj,
146 struct sn_hwperf_object_info *objs)
148 int ordinal;
149 struct sn_hwperf_object_info *p;
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++;
158 return ordinal;
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 */
166 static const char *sn_hwperf_get_slabname(struct sn_hwperf_object_info *obj,
167 struct sn_hwperf_object_info *objs, int *ordinal)
169 int isnode;
170 const char *slabname = slabname_other;
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);
176 else {
177 *ordinal = sn_hwperf_generic_ordinal(obj, objs);
178 if (SN_HWPERF_IS_ROUTER(obj))
179 slabname = slabname_router;
182 return slabname;
185 static void print_pci_topology(struct seq_file *s)
187 char *p;
188 size_t sz;
189 int e;
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;
203 static inline int sn_hwperf_has_cpus(cnodeid_t node)
205 return node < MAX_NUMNODES && node_online(node) && nr_cpus_node(node);
208 static inline int sn_hwperf_has_mem(cnodeid_t node)
210 return node < MAX_NUMNODES && node_online(node) && NODE_DATA(node)->node_present_pages;
213 static struct sn_hwperf_object_info *
214 sn_hwperf_findobj_id(struct sn_hwperf_object_info *objbuf,
215 int nobj, int id)
217 int i;
218 struct sn_hwperf_object_info *p = objbuf;
220 for (i=0; i < nobj; i++, p++) {
221 if (p->id == id)
222 return p;
225 return NULL;
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)
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;
243 if (!cnode_possible(node))
244 return -EINVAL;
246 if (sn_hwperf_has_cpus(node)) {
247 if (near_cpu_node)
248 *near_cpu_node = node;
249 found_cpu++;
252 if (sn_hwperf_has_mem(node)) {
253 if (near_mem_node)
254 *near_mem_node = node;
255 found_mem++;
258 if (found_cpu && found_mem)
259 return 0; /* trivially successful */
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;
270 if (!nodeobj) {
271 e = -ENOENT;
272 goto err;
275 /* get it's interconnect topology */
276 sz = op->ports * sizeof(struct sn_hwperf_port_info);
277 if (sz > sizeof(ptdata))
278 BUG();
279 e = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
280 SN_HWPERF_ENUM_PORTS, nodeobj->id, sz,
281 (u64)&ptdata, 0, 0, NULL);
282 if (e != SN_HWPERF_OP_OK) {
283 e = -EINVAL;
284 goto err;
287 /* find nearest node with cpus and nearest memory */
288 for (router=NULL, j=0; j < op->ports; j++) {
289 dest = sn_hwperf_findobj_id(objbuf, nobj, ptdata[j].conn_id);
290 if (dest && SN_HWPERF_IS_ROUTER(dest))
291 router = dest;
292 if (!dest || SN_HWPERF_FOREIGN(dest) ||
293 !SN_HWPERF_IS_NODE(dest) || SN_HWPERF_IS_IONODE(dest)) {
294 continue;
296 c = sn_hwperf_obj_to_cnode(dest);
297 if (!found_cpu && sn_hwperf_has_cpus(c)) {
298 if (near_cpu_node)
299 *near_cpu_node = c;
300 found_cpu++;
302 if (!found_mem && sn_hwperf_has_mem(c)) {
303 if (near_mem_node)
304 *near_mem_node = c;
305 found_mem++;
309 if (router && (!found_cpu || !found_mem)) {
310 /* search for a node connected to the same router */
311 sz = router->ports * sizeof(struct sn_hwperf_port_info);
312 if (sz > sizeof(ptdata))
313 BUG();
314 e = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
315 SN_HWPERF_ENUM_PORTS, router->id, sz,
316 (u64)&ptdata, 0, 0, NULL);
317 if (e != SN_HWPERF_OP_OK) {
318 e = -EINVAL;
319 goto err;
321 for (j=0; j < router->ports; j++) {
322 dest = sn_hwperf_findobj_id(objbuf, nobj,
323 ptdata[j].conn_id);
324 if (!dest || dest->id == node ||
325 SN_HWPERF_FOREIGN(dest) ||
326 !SN_HWPERF_IS_NODE(dest) ||
327 SN_HWPERF_IS_IONODE(dest)) {
328 continue;
330 c = sn_hwperf_obj_to_cnode(dest);
331 if (!found_cpu && sn_hwperf_has_cpus(c)) {
332 if (near_cpu_node)
333 *near_cpu_node = c;
334 found_cpu++;
336 if (!found_mem && sn_hwperf_has_mem(c)) {
337 if (near_mem_node)
338 *near_mem_node = c;
339 found_mem++;
341 if (found_cpu && found_mem)
342 break;
346 if (!found_cpu || !found_mem) {
347 /* resort to _any_ node with CPUs and memory */
348 for (i=0, op=objbuf; i < nobj; i++, op++) {
349 if (SN_HWPERF_FOREIGN(op) ||
350 SN_HWPERF_IS_IONODE(op) ||
351 !SN_HWPERF_IS_NODE(op)) {
352 continue;
354 c = sn_hwperf_obj_to_cnode(op);
355 if (!found_cpu && sn_hwperf_has_cpus(c)) {
356 if (near_cpu_node)
357 *near_cpu_node = c;
358 found_cpu++;
360 if (!found_mem && sn_hwperf_has_mem(c)) {
361 if (near_mem_node)
362 *near_mem_node = c;
363 found_mem++;
365 if (found_cpu && found_mem)
366 break;
370 if (!found_cpu || !found_mem)
371 e = -ENODATA;
373 err:
374 return e;
378 static int sn_topology_show(struct seq_file *s, void *d)
380 int sz;
381 int pt;
382 int e = 0;
383 int i;
384 int j;
385 const char *slabname;
386 int ordinal;
387 cpumask_t cpumask;
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;
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");
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;
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",
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);
431 print_pci_topology(s);
434 if (SN_HWPERF_FOREIGN(obj)) {
435 /* private in another partition: not interesting */
436 return 0;
439 for (i = 0; i < SN_HWPERF_MAXSTRING && obj->name[i]; i++) {
440 if (obj->name[i] == ' ')
441 obj->name[i] = '_';
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);
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;
454 seq_printf(s, ", nasid 0x%x", cnodeid_to_nasid(ordinal));
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);
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));
469 seq_putc(s, '\n');
472 * CPUs on this node, if any
474 if (!SN_HWPERF_IS_IONODE(obj)) {
475 cpumask = node_to_cpumask(ordinal);
476 for_each_online_cpu(i) {
477 if (cpu_isset(i, cpumask)) {
478 slice = 'a' + cpuid_to_slice(i);
479 c = cpu_data(i);
480 seq_printf(s, "cpu %d %s%c local"
481 " freq %luMHz, arch ia64",
482 i, obj->location, slice,
483 c->proc_freq / 1000000);
484 for_each_online_cpu(j) {
485 seq_printf(s, j ? ":%d" : ", dist %d",
486 node_distance(
487 cpu_to_node(i),
488 cpu_to_node(j)));
490 seq_putc(s, '\n');
496 if (obj->ports) {
498 * numalink ports
500 sz = obj->ports * sizeof(struct sn_hwperf_port_info);
501 if ((ptdata = kmalloc(sz, GFP_KERNEL)) == NULL)
502 return -ENOMEM;
503 e = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
504 SN_HWPERF_ENUM_PORTS, obj->id, sz,
505 (u64) ptdata, 0, 0, NULL);
506 if (e != SN_HWPERF_OP_OK)
507 return -EINVAL;
508 for (ordinal=0, p=objs; p != obj; p++) {
509 if (!SN_HWPERF_FOREIGN(p))
510 ordinal += p->ports;
512 for (pt = 0; pt < obj->ports; pt++) {
513 for (p = objs, i = 0; i < sn_hwperf_obj_cnt; i++, p++) {
514 if (ptdata[pt].conn_id == p->id) {
515 break;
518 seq_printf(s, "numalink %d %s-%d",
519 ordinal+pt, obj->location, ptdata[pt].port);
521 if (i >= sn_hwperf_obj_cnt) {
522 /* no connection */
523 seq_puts(s, " local endpoint disconnected"
524 ", protocol unknown\n");
525 continue;
528 if (obj->sn_hwp_this_part && p->sn_hwp_this_part)
529 /* both ends local to this partition */
530 seq_puts(s, " local");
531 else if (SN_HWPERF_FOREIGN(p))
532 /* both ends of the link in foreign partiton */
533 seq_puts(s, " foreign");
534 else
535 /* link straddles a partition */
536 seq_puts(s, " shared");
539 * Unlikely, but strictly should query the LLP config
540 * registers because an NL4R can be configured to run
541 * NL3 protocol, even when not talking to an NL3 router.
542 * Ditto for node-node.
544 seq_printf(s, " endpoint %s-%d, protocol %s\n",
545 p->location, ptdata[pt].conn_port,
546 (SN_HWPERF_IS_NL3ROUTER(obj) ||
547 SN_HWPERF_IS_NL3ROUTER(p)) ? "LLP3" : "LLP4");
549 kfree(ptdata);
552 return 0;
555 static void *sn_topology_start(struct seq_file *s, loff_t * pos)
557 struct sn_hwperf_object_info *objs = s->private;
559 if (*pos < sn_hwperf_obj_cnt)
560 return (void *)(objs + *pos);
562 return NULL;
565 static void *sn_topology_next(struct seq_file *s, void *v, loff_t * pos)
567 ++*pos;
568 return sn_topology_start(s, pos);
571 static void sn_topology_stop(struct seq_file *m, void *v)
573 return;
577 * /proc/sgi_sn/sn_topology, read-only using seq_file
579 static struct seq_operations sn_topology_seq_ops = {
580 .start = sn_topology_start,
581 .next = sn_topology_next,
582 .stop = sn_topology_stop,
583 .show = sn_topology_show
586 struct sn_hwperf_op_info {
587 u64 op;
588 struct sn_hwperf_ioctl_args *a;
589 void *p;
590 int *v0;
591 int ret;
594 static void sn_hwperf_call_sal(void *info)
596 struct sn_hwperf_op_info *op_info = info;
597 int r;
599 r = ia64_sn_hwperf_op(sn_hwperf_master_nasid, op_info->op,
600 op_info->a->arg, op_info->a->sz,
601 (u64) op_info->p, 0, 0, op_info->v0);
602 op_info->ret = r;
605 static int sn_hwperf_op_cpu(struct sn_hwperf_op_info *op_info)
607 u32 cpu;
608 u32 use_ipi;
609 int r = 0;
610 cpumask_t save_allowed;
612 cpu = (op_info->a->arg & SN_HWPERF_ARG_CPU_MASK) >> 32;
613 use_ipi = op_info->a->arg & SN_HWPERF_ARG_USE_IPI_MASK;
614 op_info->a->arg &= SN_HWPERF_ARG_OBJID_MASK;
616 if (cpu != SN_HWPERF_ARG_ANY_CPU) {
617 if (cpu >= NR_CPUS || !cpu_online(cpu)) {
618 r = -EINVAL;
619 goto out;
623 if (cpu == SN_HWPERF_ARG_ANY_CPU || cpu == get_cpu()) {
624 /* don't care, or already on correct cpu */
625 sn_hwperf_call_sal(op_info);
627 else {
628 if (use_ipi) {
629 /* use an interprocessor interrupt to call SAL */
630 smp_call_function_single(cpu, sn_hwperf_call_sal,
631 op_info, 1, 1);
633 else {
634 /* migrate the task before calling SAL */
635 save_allowed = current->cpus_allowed;
636 set_cpus_allowed(current, cpumask_of_cpu(cpu));
637 sn_hwperf_call_sal(op_info);
638 set_cpus_allowed(current, save_allowed);
641 r = op_info->ret;
643 out:
644 return r;
647 /* map SAL hwperf error code to system error code */
648 static int sn_hwperf_map_err(int hwperf_err)
650 int e;
652 switch(hwperf_err) {
653 case SN_HWPERF_OP_OK:
654 e = 0;
655 break;
657 case SN_HWPERF_OP_NOMEM:
658 e = -ENOMEM;
659 break;
661 case SN_HWPERF_OP_NO_PERM:
662 e = -EPERM;
663 break;
665 case SN_HWPERF_OP_IO_ERROR:
666 e = -EIO;
667 break;
669 case SN_HWPERF_OP_BUSY:
670 e = -EBUSY;
671 break;
673 case SN_HWPERF_OP_RECONFIGURE:
674 e = -EAGAIN;
675 break;
677 case SN_HWPERF_OP_INVAL:
678 default:
679 e = -EINVAL;
680 break;
683 return e;
687 * ioctl for "sn_hwperf" misc device
689 static int
690 sn_hwperf_ioctl(struct inode *in, struct file *fp, u32 op, u64 arg)
692 struct sn_hwperf_ioctl_args a;
693 struct cpuinfo_ia64 *cdata;
694 struct sn_hwperf_object_info *objs;
695 struct sn_hwperf_object_info *cpuobj;
696 struct sn_hwperf_op_info op_info;
697 void *p = NULL;
698 int nobj;
699 char slice;
700 int node;
701 int r;
702 int v0;
703 int i;
704 int j;
706 unlock_kernel();
708 /* only user requests are allowed here */
709 if ((op & SN_HWPERF_OP_MASK) < 10) {
710 r = -EINVAL;
711 goto error;
713 r = copy_from_user(&a, (const void __user *)arg,
714 sizeof(struct sn_hwperf_ioctl_args));
715 if (r != 0) {
716 r = -EFAULT;
717 goto error;
721 * Allocate memory to hold a kernel copy of the user buffer. The
722 * buffer contents are either copied in or out (or both) of user
723 * space depending on the flags encoded in the requested operation.
725 if (a.ptr) {
726 p = vmalloc(a.sz);
727 if (!p) {
728 r = -ENOMEM;
729 goto error;
733 if (op & SN_HWPERF_OP_MEM_COPYIN) {
734 r = copy_from_user(p, (const void __user *)a.ptr, a.sz);
735 if (r != 0) {
736 r = -EFAULT;
737 goto error;
741 switch (op) {
742 case SN_HWPERF_GET_CPU_INFO:
743 if (a.sz == sizeof(u64)) {
744 /* special case to get size needed */
745 *(u64 *) p = (u64) num_online_cpus() *
746 sizeof(struct sn_hwperf_object_info);
747 } else
748 if (a.sz < num_online_cpus() * sizeof(struct sn_hwperf_object_info)) {
749 r = -ENOMEM;
750 goto error;
751 } else
752 if ((r = sn_hwperf_enum_objects(&nobj, &objs)) == 0) {
753 memset(p, 0, a.sz);
754 for (i = 0; i < nobj; i++) {
755 int cpuobj_index = 0;
756 if (!SN_HWPERF_IS_NODE(objs + i))
757 continue;
758 node = sn_hwperf_obj_to_cnode(objs + i);
759 for_each_online_cpu(j) {
760 if (node != cpu_to_node(j))
761 continue;
762 cpuobj = (struct sn_hwperf_object_info *) p + cpuobj_index++;
763 slice = 'a' + cpuid_to_slice(j);
764 cdata = cpu_data(j);
765 cpuobj->id = j;
766 snprintf(cpuobj->name,
767 sizeof(cpuobj->name),
768 "CPU %luMHz %s",
769 cdata->proc_freq / 1000000,
770 cdata->vendor);
771 snprintf(cpuobj->location,
772 sizeof(cpuobj->location),
773 "%s%c", objs[i].location,
774 slice);
778 vfree(objs);
780 break;
782 case SN_HWPERF_GET_NODE_NASID:
783 if (a.sz != sizeof(u64) ||
784 (node = a.arg) < 0 || !cnode_possible(node)) {
785 r = -EINVAL;
786 goto error;
788 *(u64 *)p = (u64)cnodeid_to_nasid(node);
789 break;
791 case SN_HWPERF_GET_OBJ_NODE:
792 if (a.sz != sizeof(u64) || a.arg < 0) {
793 r = -EINVAL;
794 goto error;
796 if ((r = sn_hwperf_enum_objects(&nobj, &objs)) == 0) {
797 if (a.arg >= nobj) {
798 r = -EINVAL;
799 vfree(objs);
800 goto error;
802 if (objs[(i = a.arg)].id != a.arg) {
803 for (i = 0; i < nobj; i++) {
804 if (objs[i].id == a.arg)
805 break;
808 if (i == nobj) {
809 r = -EINVAL;
810 vfree(objs);
811 goto error;
814 if (!SN_HWPERF_IS_NODE(objs + i) &&
815 !SN_HWPERF_IS_IONODE(objs + i)) {
816 r = -ENOENT;
817 vfree(objs);
818 goto error;
821 *(u64 *)p = (u64)sn_hwperf_obj_to_cnode(objs + i);
822 vfree(objs);
824 break;
826 case SN_HWPERF_GET_MMRS:
827 case SN_HWPERF_SET_MMRS:
828 case SN_HWPERF_OBJECT_DISTANCE:
829 op_info.p = p;
830 op_info.a = &a;
831 op_info.v0 = &v0;
832 op_info.op = op;
833 r = sn_hwperf_op_cpu(&op_info);
834 if (r) {
835 r = sn_hwperf_map_err(r);
836 a.v0 = v0;
837 goto error;
839 break;
841 default:
842 /* all other ops are a direct SAL call */
843 r = ia64_sn_hwperf_op(sn_hwperf_master_nasid, op,
844 a.arg, a.sz, (u64) p, 0, 0, &v0);
845 if (r) {
846 r = sn_hwperf_map_err(r);
847 goto error;
849 a.v0 = v0;
850 break;
853 if (op & SN_HWPERF_OP_MEM_COPYOUT) {
854 r = copy_to_user((void __user *)a.ptr, p, a.sz);
855 if (r != 0) {
856 r = -EFAULT;
857 goto error;
861 error:
862 vfree(p);
864 lock_kernel();
865 return r;
868 static const struct file_operations sn_hwperf_fops = {
869 .ioctl = sn_hwperf_ioctl,
872 static struct miscdevice sn_hwperf_dev = {
873 MISC_DYNAMIC_MINOR,
874 "sn_hwperf",
875 &sn_hwperf_fops
878 static int sn_hwperf_init(void)
880 u64 v;
881 int salr;
882 int e = 0;
884 /* single threaded, once-only initialization */
885 down(&sn_hwperf_init_mutex);
887 if (sn_hwperf_salheap) {
888 up(&sn_hwperf_init_mutex);
889 return e;
893 * The PROM code needs a fixed reference node. For convenience the
894 * same node as the console I/O is used.
896 sn_hwperf_master_nasid = (nasid_t) ia64_sn_get_console_nasid();
899 * Request the needed size and install the PROM scratch area.
900 * The PROM keeps various tracking bits in this memory area.
902 salr = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
903 (u64) SN_HWPERF_GET_HEAPSIZE, 0,
904 (u64) sizeof(u64), (u64) &v, 0, 0, NULL);
905 if (salr != SN_HWPERF_OP_OK) {
906 e = -EINVAL;
907 goto out;
910 if ((sn_hwperf_salheap = vmalloc(v)) == NULL) {
911 e = -ENOMEM;
912 goto out;
914 salr = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
915 SN_HWPERF_INSTALL_HEAP, 0, v,
916 (u64) sn_hwperf_salheap, 0, 0, NULL);
917 if (salr != SN_HWPERF_OP_OK) {
918 e = -EINVAL;
919 goto out;
922 salr = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
923 SN_HWPERF_OBJECT_COUNT, 0,
924 sizeof(u64), (u64) &v, 0, 0, NULL);
925 if (salr != SN_HWPERF_OP_OK) {
926 e = -EINVAL;
927 goto out;
929 sn_hwperf_obj_cnt = (int)v;
931 out:
932 if (e < 0 && sn_hwperf_salheap) {
933 vfree(sn_hwperf_salheap);
934 sn_hwperf_salheap = NULL;
935 sn_hwperf_obj_cnt = 0;
937 up(&sn_hwperf_init_mutex);
938 return e;
941 int sn_topology_open(struct inode *inode, struct file *file)
943 int e;
944 struct seq_file *seq;
945 struct sn_hwperf_object_info *objbuf;
946 int nobj;
948 if ((e = sn_hwperf_enum_objects(&nobj, &objbuf)) == 0) {
949 e = seq_open(file, &sn_topology_seq_ops);
950 seq = file->private_data;
951 seq->private = objbuf;
954 return e;
957 int sn_topology_release(struct inode *inode, struct file *file)
959 struct seq_file *seq = file->private_data;
961 vfree(seq->private);
962 return seq_release(inode, file);
965 int sn_hwperf_get_nearest_node(cnodeid_t node,
966 cnodeid_t *near_mem_node, cnodeid_t *near_cpu_node)
968 int e;
969 int nobj;
970 struct sn_hwperf_object_info *objbuf;
972 if ((e = sn_hwperf_enum_objects(&nobj, &objbuf)) == 0) {
973 e = sn_hwperf_get_nearest_node_objdata(objbuf, nobj,
974 node, near_mem_node, near_cpu_node);
975 vfree(objbuf);
978 return e;
981 static int __devinit sn_hwperf_misc_register_init(void)
983 int e;
985 if (!ia64_platform_is("sn2"))
986 return 0;
988 sn_hwperf_init();
991 * Register a dynamic misc device for hwperf ioctls. Platforms
992 * supporting hotplug will create /dev/sn_hwperf, else user
993 * can to look up the minor number in /proc/misc.
995 if ((e = misc_register(&sn_hwperf_dev)) != 0) {
996 printk(KERN_ERR "sn_hwperf_misc_register_init: failed to "
997 "register misc device for \"%s\"\n", sn_hwperf_dev.name);
1000 return e;
1003 device_initcall(sn_hwperf_misc_register_init); /* after misc_init() */
1004 EXPORT_SYMBOL(sn_hwperf_get_nearest_node);