[IA64-SGI] - Reduce overhead of reading sn_topology
[linux-2.6/cjktty.git] / arch / ia64 / sn / kernel / sn2 / sn_hwperf.c
blob739c948dc504d4a6089b795afe329002cef1f056
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 <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>
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);
54 static int sn_hwperf_enum_objects(int *nobj, struct sn_hwperf_object_info **ret)
56 int e;
57 u64 sz;
58 struct sn_hwperf_object_info *objbuf = NULL;
60 if ((e = sn_hwperf_init()) < 0) {
61 printk(KERN_ERR "sn_hwperf_init failed: err %d\n", e);
62 goto out;
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;
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);
79 out:
80 *nobj = sn_hwperf_obj_cnt;
81 *ret = objbuf;
82 return e;
85 static int sn_hwperf_location_to_bpos(char *location,
86 int *rack, int *bay, int *slot, int *slab)
88 char type;
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;
102 static int sn_hwperf_geoid_to_cnode(char *location)
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;
110 if (sn_hwperf_location_to_bpos(location, &rack, &bay, &slot, &slab))
111 return -1;
114 * FIXME: replace with cleaner for_each_XXX macro which addresses
115 * both compute and IO nodes once ACPI3.0 is available.
117 for (cnode = 0; cnode < num_cnodes; cnode++) {
118 geoid = cnodeid_get_geoid(cnode);
119 module_id = geo_module(geoid);
120 this_rack = MODULE_GET_RACK(module_id);
121 this_bay = MODULE_GET_BPOS(module_id);
122 this_slot = geo_slot(geoid);
123 this_slab = geo_slab(geoid);
124 if (rack == this_rack && bay == this_bay &&
125 slot == this_slot && slab == this_slab) {
126 break;
130 return node_possible(cnode) ? cnode : -1;
133 static int sn_hwperf_obj_to_cnode(struct sn_hwperf_object_info * obj)
135 if (!SN_HWPERF_IS_NODE(obj) && !SN_HWPERF_IS_IONODE(obj))
136 BUG();
137 if (!obj->sn_hwp_this_part)
138 return -1;
139 return sn_hwperf_geoid_to_cnode(obj->location);
142 static int sn_hwperf_generic_ordinal(struct sn_hwperf_object_info *obj,
143 struct sn_hwperf_object_info *objs)
145 int ordinal;
146 struct sn_hwperf_object_info *p;
148 for (ordinal=0, p=objs; p != obj; p++) {
149 if (SN_HWPERF_FOREIGN(p))
150 continue;
151 if (SN_HWPERF_SAME_OBJTYPE(p, obj))
152 ordinal++;
155 return ordinal;
158 static const char *slabname_node = "node"; /* SHub asic */
159 static const char *slabname_ionode = "ionode"; /* TIO asic */
160 static const char *slabname_router = "router"; /* NL3R or NL4R */
161 static const char *slabname_other = "other"; /* unknown asic */
163 static const char *sn_hwperf_get_slabname(struct sn_hwperf_object_info *obj,
164 struct sn_hwperf_object_info *objs, int *ordinal)
166 int isnode;
167 const char *slabname = slabname_other;
169 if ((isnode = SN_HWPERF_IS_NODE(obj)) || SN_HWPERF_IS_IONODE(obj)) {
170 slabname = isnode ? slabname_node : slabname_ionode;
171 *ordinal = sn_hwperf_obj_to_cnode(obj);
173 else {
174 *ordinal = sn_hwperf_generic_ordinal(obj, objs);
175 if (SN_HWPERF_IS_ROUTER(obj))
176 slabname = slabname_router;
179 return slabname;
182 static void print_pci_topology(struct seq_file *s)
184 char *p;
185 size_t sz;
186 int e;
188 for (sz = PAGE_SIZE; sz < 16 * PAGE_SIZE; sz += PAGE_SIZE) {
189 if (!(p = (char *)kmalloc(sz, GFP_KERNEL)))
190 break;
191 e = ia64_sn_ioif_get_pci_topology(__pa(p), sz);
192 if (e == SALRET_OK)
193 seq_puts(s, p);
194 kfree(p);
195 if (e == SALRET_OK || e == SALRET_NOT_IMPLEMENTED)
196 break;
200 static inline int sn_hwperf_has_cpus(cnodeid_t node)
202 return node_online(node) && nr_cpus_node(node);
205 static inline int sn_hwperf_has_mem(cnodeid_t node)
207 return node_online(node) && NODE_DATA(node)->node_present_pages;
210 static struct sn_hwperf_object_info *
211 sn_hwperf_findobj_id(struct sn_hwperf_object_info *objbuf,
212 int nobj, int id)
214 int i;
215 struct sn_hwperf_object_info *p = objbuf;
217 for (i=0; i < nobj; i++, p++) {
218 if (p->id == id)
219 return p;
222 return NULL;
226 static int sn_hwperf_get_nearest_node_objdata(struct sn_hwperf_object_info *objbuf,
227 int nobj, cnodeid_t node, cnodeid_t *near_mem_node, cnodeid_t *near_cpu_node)
229 int e;
230 struct sn_hwperf_object_info *nodeobj = NULL;
231 struct sn_hwperf_object_info *op;
232 struct sn_hwperf_object_info *dest;
233 struct sn_hwperf_object_info *router;
234 struct sn_hwperf_port_info ptdata[16];
235 int sz, i, j;
236 cnodeid_t c;
237 int found_mem = 0;
238 int found_cpu = 0;
240 if (!node_possible(node))
241 return -EINVAL;
243 if (sn_hwperf_has_cpus(node)) {
244 if (near_cpu_node)
245 *near_cpu_node = node;
246 found_cpu++;
249 if (sn_hwperf_has_mem(node)) {
250 if (near_mem_node)
251 *near_mem_node = node;
252 found_mem++;
255 if (found_cpu && found_mem)
256 return 0; /* trivially successful */
258 /* find the argument node object */
259 for (i=0, op=objbuf; i < nobj; i++, op++) {
260 if (!SN_HWPERF_IS_NODE(op) && !SN_HWPERF_IS_IONODE(op))
261 continue;
262 if (node == sn_hwperf_obj_to_cnode(op)) {
263 nodeobj = op;
264 break;
267 if (!nodeobj) {
268 e = -ENOENT;
269 goto err;
272 /* get it's interconnect topology */
273 sz = op->ports * sizeof(struct sn_hwperf_port_info);
274 if (sz > sizeof(ptdata))
275 BUG();
276 e = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
277 SN_HWPERF_ENUM_PORTS, nodeobj->id, sz,
278 (u64)&ptdata, 0, 0, NULL);
279 if (e != SN_HWPERF_OP_OK) {
280 e = -EINVAL;
281 goto err;
284 /* find nearest node with cpus and nearest memory */
285 for (router=NULL, j=0; j < op->ports; j++) {
286 dest = sn_hwperf_findobj_id(objbuf, nobj, ptdata[j].conn_id);
287 if (dest && SN_HWPERF_IS_ROUTER(dest))
288 router = dest;
289 if (!dest || SN_HWPERF_FOREIGN(dest) ||
290 !SN_HWPERF_IS_NODE(dest) || SN_HWPERF_IS_IONODE(dest)) {
291 continue;
293 c = sn_hwperf_obj_to_cnode(dest);
294 if (!found_cpu && sn_hwperf_has_cpus(c)) {
295 if (near_cpu_node)
296 *near_cpu_node = c;
297 found_cpu++;
299 if (!found_mem && sn_hwperf_has_mem(c)) {
300 if (near_mem_node)
301 *near_mem_node = c;
302 found_mem++;
306 if (router && (!found_cpu || !found_mem)) {
307 /* search for a node connected to the same router */
308 sz = router->ports * sizeof(struct sn_hwperf_port_info);
309 if (sz > sizeof(ptdata))
310 BUG();
311 e = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
312 SN_HWPERF_ENUM_PORTS, router->id, sz,
313 (u64)&ptdata, 0, 0, NULL);
314 if (e != SN_HWPERF_OP_OK) {
315 e = -EINVAL;
316 goto err;
318 for (j=0; j < router->ports; j++) {
319 dest = sn_hwperf_findobj_id(objbuf, nobj,
320 ptdata[j].conn_id);
321 if (!dest || dest->id == node ||
322 SN_HWPERF_FOREIGN(dest) ||
323 !SN_HWPERF_IS_NODE(dest) ||
324 SN_HWPERF_IS_IONODE(dest)) {
325 continue;
327 c = sn_hwperf_obj_to_cnode(dest);
328 if (!found_cpu && sn_hwperf_has_cpus(c)) {
329 if (near_cpu_node)
330 *near_cpu_node = c;
331 found_cpu++;
333 if (!found_mem && sn_hwperf_has_mem(c)) {
334 if (near_mem_node)
335 *near_mem_node = c;
336 found_mem++;
338 if (found_cpu && found_mem)
339 break;
343 if (!found_cpu || !found_mem) {
344 /* resort to _any_ node with CPUs and memory */
345 for (i=0, op=objbuf; i < nobj; i++, op++) {
346 if (SN_HWPERF_FOREIGN(op) ||
347 SN_HWPERF_IS_IONODE(op) ||
348 !SN_HWPERF_IS_NODE(op)) {
349 continue;
351 c = sn_hwperf_obj_to_cnode(op);
352 if (!found_cpu && sn_hwperf_has_cpus(c)) {
353 if (near_cpu_node)
354 *near_cpu_node = c;
355 found_cpu++;
357 if (!found_mem && sn_hwperf_has_mem(c)) {
358 if (near_mem_node)
359 *near_mem_node = c;
360 found_mem++;
362 if (found_cpu && found_mem)
363 break;
367 if (!found_cpu || !found_mem)
368 e = -ENODATA;
370 err:
371 return e;
375 static int sn_topology_show(struct seq_file *s, void *d)
377 int sz;
378 int pt;
379 int e = 0;
380 int i;
381 int j;
382 const char *slabname;
383 int ordinal;
384 cpumask_t cpumask;
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;
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");
406 if (ia64_sn_get_sn_info(0,
407 &shubtype, &nasid_mask, &nasid_shift, &system_size,
408 &sharing_size, &partid, &coher, &region_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;
414 seq_printf(s, "partition %u %s local "
415 "shubtype %s, "
416 "nasid_mask 0x%016lx, "
417 "nasid_bits %d:%d, "
418 "system_size %d, "
419 "sharing_size %d, "
420 "coherency_domain %d, "
421 "region_size %d\n",
423 partid, system_utsname.nodename,
424 shubtype ? "shub2" : "shub1",
425 (u64)nasid_mask << nasid_shift, nasid_msb, nasid_shift,
426 system_size, sharing_size, coher, region_size);
428 print_pci_topology(s);
431 if (SN_HWPERF_FOREIGN(obj)) {
432 /* private in another partition: not interesting */
433 return 0;
436 for (i = 0; i < SN_HWPERF_MAXSTRING && obj->name[i]; i++) {
437 if (obj->name[i] == ' ')
438 obj->name[i] = '_';
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);
445 if (!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;
451 seq_printf(s, ", nasid 0x%x", cnodeid_to_nasid(ordinal));
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);
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));
466 seq_putc(s, '\n');
469 * CPUs on this node, if any
471 cpumask = node_to_cpumask(ordinal);
472 for_each_online_cpu(i) {
473 if (cpu_isset(i, cpumask)) {
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)));
486 seq_putc(s, '\n');
491 if (obj->ports) {
493 * numalink ports
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;
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;
513 seq_printf(s, "numalink %d %s-%d",
514 ordinal+pt, obj->location, ptdata[pt].port);
516 if (i >= sn_hwperf_obj_cnt) {
517 /* no connection */
518 seq_puts(s, " local endpoint disconnected"
519 ", protocol unknown\n");
520 continue;
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 (!obj->sn_hwp_this_part && !p->sn_hwp_this_part)
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");
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.
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");
544 kfree(ptdata);
547 return 0;
550 static void *sn_topology_start(struct seq_file *s, loff_t * pos)
552 struct sn_hwperf_object_info *objs = s->private;
554 if (*pos < sn_hwperf_obj_cnt)
555 return (void *)(objs + *pos);
557 return NULL;
560 static void *sn_topology_next(struct seq_file *s, void *v, loff_t * pos)
562 ++*pos;
563 return sn_topology_start(s, pos);
566 static void sn_topology_stop(struct seq_file *m, void *v)
568 return;
572 * /proc/sgi_sn/sn_topology, read-only using seq_file
574 static 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
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;
589 static void sn_hwperf_call_sal(void *info)
591 struct sn_hwperf_op_info *op_info = info;
592 int r;
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;
600 static int sn_hwperf_op_cpu(struct sn_hwperf_op_info *op_info)
602 u32 cpu;
603 u32 use_ipi;
604 int r = 0;
605 cpumask_t save_allowed;
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;
611 if (cpu != SN_HWPERF_ARG_ANY_CPU) {
612 if (cpu >= NR_CPUS || !cpu_online(cpu)) {
613 r = -EINVAL;
614 goto out;
618 if (cpu == SN_HWPERF_ARG_ANY_CPU || cpu == get_cpu()) {
619 /* don't care, or already on correct cpu */
620 sn_hwperf_call_sal(op_info);
622 else {
623 if (use_ipi) {
624 /* use an interprocessor interrupt to call SAL */
625 smp_call_function_single(cpu, sn_hwperf_call_sal,
626 op_info, 1, 1);
628 else {
629 /* migrate the task before calling SAL */
630 save_allowed = current->cpus_allowed;
631 set_cpus_allowed(current, cpumask_of_cpu(cpu));
632 sn_hwperf_call_sal(op_info);
633 set_cpus_allowed(current, save_allowed);
636 r = op_info->ret;
638 out:
639 return r;
642 /* map SAL hwperf error code to system error code */
643 static int sn_hwperf_map_err(int hwperf_err)
645 int e;
647 switch(hwperf_err) {
648 case SN_HWPERF_OP_OK:
649 e = 0;
650 break;
652 case SN_HWPERF_OP_NOMEM:
653 e = -ENOMEM;
654 break;
656 case SN_HWPERF_OP_NO_PERM:
657 e = -EPERM;
658 break;
660 case SN_HWPERF_OP_IO_ERROR:
661 e = -EIO;
662 break;
664 case SN_HWPERF_OP_BUSY:
665 e = -EBUSY;
666 break;
668 case SN_HWPERF_OP_RECONFIGURE:
669 e = -EAGAIN;
670 break;
672 case SN_HWPERF_OP_INVAL:
673 default:
674 e = -EINVAL;
675 break;
678 return e;
682 * ioctl for "sn_hwperf" misc device
684 static int
685 sn_hwperf_ioctl(struct inode *in, struct file *fp, u32 op, u64 arg)
687 struct sn_hwperf_ioctl_args a;
688 struct cpuinfo_ia64 *cdata;
689 struct sn_hwperf_object_info *objs;
690 struct sn_hwperf_object_info *cpuobj;
691 struct sn_hwperf_op_info op_info;
692 void *p = NULL;
693 int nobj;
694 char slice;
695 int node;
696 int r;
697 int v0;
698 int i;
699 int j;
701 unlock_kernel();
703 /* only user requests are allowed here */
704 if ((op & SN_HWPERF_OP_MASK) < 10) {
705 r = -EINVAL;
706 goto error;
708 r = copy_from_user(&a, (const void __user *)arg,
709 sizeof(struct sn_hwperf_ioctl_args));
710 if (r != 0) {
711 r = -EFAULT;
712 goto error;
716 * Allocate memory to hold a kernel copy of the user buffer. The
717 * buffer contents are either copied in or out (or both) of user
718 * space depending on the flags encoded in the requested operation.
720 if (a.ptr) {
721 p = vmalloc(a.sz);
722 if (!p) {
723 r = -ENOMEM;
724 goto error;
728 if (op & SN_HWPERF_OP_MEM_COPYIN) {
729 r = copy_from_user(p, (const void __user *)a.ptr, a.sz);
730 if (r != 0) {
731 r = -EFAULT;
732 goto error;
736 switch (op) {
737 case SN_HWPERF_GET_CPU_INFO:
738 if (a.sz == sizeof(u64)) {
739 /* special case to get size needed */
740 *(u64 *) p = (u64) num_online_cpus() *
741 sizeof(struct sn_hwperf_object_info);
742 } else
743 if (a.sz < num_online_cpus() * sizeof(struct sn_hwperf_object_info)) {
744 r = -ENOMEM;
745 goto error;
746 } else
747 if ((r = sn_hwperf_enum_objects(&nobj, &objs)) == 0) {
748 memset(p, 0, a.sz);
749 for (i = 0; i < nobj; i++) {
750 int cpuobj_index = 0;
751 if (!SN_HWPERF_IS_NODE(objs + i))
752 continue;
753 node = sn_hwperf_obj_to_cnode(objs + i);
754 for_each_online_cpu(j) {
755 if (node != cpu_to_node(j))
756 continue;
757 cpuobj = (struct sn_hwperf_object_info *) p + cpuobj_index++;
758 slice = 'a' + cpuid_to_slice(j);
759 cdata = cpu_data(j);
760 cpuobj->id = j;
761 snprintf(cpuobj->name,
762 sizeof(cpuobj->name),
763 "CPU %luMHz %s",
764 cdata->proc_freq / 1000000,
765 cdata->vendor);
766 snprintf(cpuobj->location,
767 sizeof(cpuobj->location),
768 "%s%c", objs[i].location,
769 slice);
773 vfree(objs);
775 break;
777 case SN_HWPERF_GET_NODE_NASID:
778 if (a.sz != sizeof(u64) ||
779 (node = a.arg) < 0 || !node_possible(node)) {
780 r = -EINVAL;
781 goto error;
783 *(u64 *)p = (u64)cnodeid_to_nasid(node);
784 break;
786 case SN_HWPERF_GET_OBJ_NODE:
787 if (a.sz != sizeof(u64) || a.arg < 0) {
788 r = -EINVAL;
789 goto error;
791 if ((r = sn_hwperf_enum_objects(&nobj, &objs)) == 0) {
792 if (a.arg >= nobj) {
793 r = -EINVAL;
794 vfree(objs);
795 goto error;
797 if (objs[(i = a.arg)].id != a.arg) {
798 for (i = 0; i < nobj; i++) {
799 if (objs[i].id == a.arg)
800 break;
803 if (i == nobj) {
804 r = -EINVAL;
805 vfree(objs);
806 goto error;
809 if (!SN_HWPERF_IS_NODE(objs + i) &&
810 !SN_HWPERF_IS_IONODE(objs + i)) {
811 r = -ENOENT;
812 vfree(objs);
813 goto error;
816 *(u64 *)p = (u64)sn_hwperf_obj_to_cnode(objs + i);
817 vfree(objs);
819 break;
821 case SN_HWPERF_GET_MMRS:
822 case SN_HWPERF_SET_MMRS:
823 case SN_HWPERF_OBJECT_DISTANCE:
824 op_info.p = p;
825 op_info.a = &a;
826 op_info.v0 = &v0;
827 op_info.op = op;
828 r = sn_hwperf_op_cpu(&op_info);
829 if (r) {
830 r = sn_hwperf_map_err(r);
831 a.v0 = v0;
832 goto error;
834 break;
836 default:
837 /* all other ops are a direct SAL call */
838 r = ia64_sn_hwperf_op(sn_hwperf_master_nasid, op,
839 a.arg, a.sz, (u64) p, 0, 0, &v0);
840 if (r) {
841 r = sn_hwperf_map_err(r);
842 goto error;
844 a.v0 = v0;
845 break;
848 if (op & SN_HWPERF_OP_MEM_COPYOUT) {
849 r = copy_to_user((void __user *)a.ptr, p, a.sz);
850 if (r != 0) {
851 r = -EFAULT;
852 goto error;
856 error:
857 vfree(p);
859 lock_kernel();
860 return r;
863 static struct file_operations sn_hwperf_fops = {
864 .ioctl = sn_hwperf_ioctl,
867 static struct miscdevice sn_hwperf_dev = {
868 MISC_DYNAMIC_MINOR,
869 "sn_hwperf",
870 &sn_hwperf_fops
873 static int sn_hwperf_init(void)
875 u64 v;
876 int salr;
877 int e = 0;
879 /* single threaded, once-only initialization */
880 down(&sn_hwperf_init_mutex);
882 if (sn_hwperf_salheap) {
883 up(&sn_hwperf_init_mutex);
884 return e;
888 * The PROM code needs a fixed reference node. For convenience the
889 * same node as the console I/O is used.
891 sn_hwperf_master_nasid = (nasid_t) ia64_sn_get_console_nasid();
894 * Request the needed size and install the PROM scratch area.
895 * The PROM keeps various tracking bits in this memory area.
897 salr = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
898 (u64) SN_HWPERF_GET_HEAPSIZE, 0,
899 (u64) sizeof(u64), (u64) &v, 0, 0, NULL);
900 if (salr != SN_HWPERF_OP_OK) {
901 e = -EINVAL;
902 goto out;
905 if ((sn_hwperf_salheap = vmalloc(v)) == NULL) {
906 e = -ENOMEM;
907 goto out;
909 salr = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
910 SN_HWPERF_INSTALL_HEAP, 0, v,
911 (u64) sn_hwperf_salheap, 0, 0, NULL);
912 if (salr != SN_HWPERF_OP_OK) {
913 e = -EINVAL;
914 goto out;
917 salr = ia64_sn_hwperf_op(sn_hwperf_master_nasid,
918 SN_HWPERF_OBJECT_COUNT, 0,
919 sizeof(u64), (u64) &v, 0, 0, NULL);
920 if (salr != SN_HWPERF_OP_OK) {
921 e = -EINVAL;
922 goto out;
924 sn_hwperf_obj_cnt = (int)v;
926 out:
927 if (e < 0 && sn_hwperf_salheap) {
928 vfree(sn_hwperf_salheap);
929 sn_hwperf_salheap = NULL;
930 sn_hwperf_obj_cnt = 0;
932 up(&sn_hwperf_init_mutex);
933 return e;
936 int sn_topology_open(struct inode *inode, struct file *file)
938 int e;
939 struct seq_file *seq;
940 struct sn_hwperf_object_info *objbuf;
941 int nobj;
943 if ((e = sn_hwperf_enum_objects(&nobj, &objbuf)) == 0) {
944 e = seq_open(file, &sn_topology_seq_ops);
945 seq = file->private_data;
946 seq->private = objbuf;
949 return e;
952 int sn_topology_release(struct inode *inode, struct file *file)
954 struct seq_file *seq = file->private_data;
956 vfree(seq->private);
957 return seq_release(inode, file);
960 int sn_hwperf_get_nearest_node(cnodeid_t node,
961 cnodeid_t *near_mem_node, cnodeid_t *near_cpu_node)
963 int e;
964 int nobj;
965 struct sn_hwperf_object_info *objbuf;
967 if ((e = sn_hwperf_enum_objects(&nobj, &objbuf)) == 0) {
968 e = sn_hwperf_get_nearest_node_objdata(objbuf, nobj,
969 node, near_mem_node, near_cpu_node);
970 vfree(objbuf);
973 return e;
976 static int __devinit sn_hwperf_misc_register_init(void)
978 int e;
980 if (!ia64_platform_is("sn2"))
981 return 0;
983 sn_hwperf_init();
986 * Register a dynamic misc device for hwperf ioctls. Platforms
987 * supporting hotplug will create /dev/sn_hwperf, else user
988 * can to look up the minor number in /proc/misc.
990 if ((e = misc_register(&sn_hwperf_dev)) != 0) {
991 printk(KERN_ERR "sn_hwperf_misc_register_init: failed to "
992 "register misc device for \"%s\"\n", sn_hwperf_dev.name);
995 return e;
998 device_initcall(sn_hwperf_misc_register_init); /* after misc_init() */
999 EXPORT_SYMBOL(sn_hwperf_get_nearest_node);