2 * Copyright (c) 2004, 2005 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005 Mellanox Technologies Ltd. All rights reserved.
4 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
6 * This software is available to you under a choice of one of two
7 * licenses. You may choose to be licensed under the terms of the GNU
8 * General Public License (GPL) Version 2, available from the file
9 * COPYING in the main directory of this source tree, or the
10 * OpenIB.org BSD license below:
12 * Redistribution and use in source and binary forms, with or
13 * without modification, are permitted provided that the following
16 * - Redistributions of source code must retain the above
17 * copyright notice, this list of conditions and the following
20 * - Redistributions in binary form must reproduce the above
21 * copyright notice, this list of conditions and the following
22 * disclaimer in the documentation and/or other materials
23 * provided with the distribution.
25 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
26 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
27 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
28 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
29 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
30 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
31 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
35 #include "core_priv.h"
37 #include <linux/slab.h>
38 #include <linux/stat.h>
39 #include <linux/string.h>
40 #include <linux/netdevice.h>
41 #include <linux/ethtool.h>
43 #include <rdma/ib_mad.h>
44 #include <rdma/ib_pma.h>
45 #include <rdma/ib_cache.h>
46 #include <rdma/rdma_counter.h>
50 struct gid_attr_group {
53 struct attribute_group ndev;
54 struct attribute_group type;
58 struct ib_device *ibdev;
59 struct gid_attr_group *gid_attr_group;
60 struct attribute_group gid_group;
61 struct attribute_group *pkey_group;
62 const struct attribute_group *pma_table;
63 struct attribute_group *hw_stats_ag;
64 struct rdma_hw_stats *hw_stats;
68 struct port_attribute {
69 struct attribute attr;
70 ssize_t (*show)(struct ib_port *, struct port_attribute *, char *buf);
71 ssize_t (*store)(struct ib_port *, struct port_attribute *,
72 const char *buf, size_t count);
75 #define PORT_ATTR(_name, _mode, _show, _store) \
76 struct port_attribute port_attr_##_name = __ATTR(_name, _mode, _show, _store)
78 #define PORT_ATTR_RO(_name) \
79 struct port_attribute port_attr_##_name = __ATTR_RO(_name)
81 struct port_table_attribute {
82 struct port_attribute attr;
88 struct hw_stats_attribute {
89 struct attribute attr;
90 ssize_t (*show)(struct kobject *kobj,
91 struct attribute *attr, char *buf);
92 ssize_t (*store)(struct kobject *kobj,
93 struct attribute *attr,
100 static ssize_t port_attr_show(struct kobject *kobj,
101 struct attribute *attr, char *buf)
103 struct port_attribute *port_attr =
104 container_of(attr, struct port_attribute, attr);
105 struct ib_port *p = container_of(kobj, struct ib_port, kobj);
107 if (!port_attr->show)
110 return port_attr->show(p, port_attr, buf);
113 static ssize_t port_attr_store(struct kobject *kobj,
114 struct attribute *attr,
115 const char *buf, size_t count)
117 struct port_attribute *port_attr =
118 container_of(attr, struct port_attribute, attr);
119 struct ib_port *p = container_of(kobj, struct ib_port, kobj);
121 if (!port_attr->store)
123 return port_attr->store(p, port_attr, buf, count);
126 static const struct sysfs_ops port_sysfs_ops = {
127 .show = port_attr_show,
128 .store = port_attr_store
131 static ssize_t gid_attr_show(struct kobject *kobj,
132 struct attribute *attr, char *buf)
134 struct port_attribute *port_attr =
135 container_of(attr, struct port_attribute, attr);
136 struct ib_port *p = container_of(kobj, struct gid_attr_group,
139 if (!port_attr->show)
142 return port_attr->show(p, port_attr, buf);
145 static const struct sysfs_ops gid_attr_sysfs_ops = {
146 .show = gid_attr_show
149 static ssize_t state_show(struct ib_port *p, struct port_attribute *unused,
152 struct ib_port_attr attr;
155 static const char *state_name[] = {
156 [IB_PORT_NOP] = "NOP",
157 [IB_PORT_DOWN] = "DOWN",
158 [IB_PORT_INIT] = "INIT",
159 [IB_PORT_ARMED] = "ARMED",
160 [IB_PORT_ACTIVE] = "ACTIVE",
161 [IB_PORT_ACTIVE_DEFER] = "ACTIVE_DEFER"
164 ret = ib_query_port(p->ibdev, p->port_num, &attr);
168 return sprintf(buf, "%d: %s\n", attr.state,
169 attr.state >= 0 && attr.state < ARRAY_SIZE(state_name) ?
170 state_name[attr.state] : "UNKNOWN");
173 static ssize_t lid_show(struct ib_port *p, struct port_attribute *unused,
176 struct ib_port_attr attr;
179 ret = ib_query_port(p->ibdev, p->port_num, &attr);
183 return sprintf(buf, "0x%x\n", attr.lid);
186 static ssize_t lid_mask_count_show(struct ib_port *p,
187 struct port_attribute *unused,
190 struct ib_port_attr attr;
193 ret = ib_query_port(p->ibdev, p->port_num, &attr);
197 return sprintf(buf, "%d\n", attr.lmc);
200 static ssize_t sm_lid_show(struct ib_port *p, struct port_attribute *unused,
203 struct ib_port_attr attr;
206 ret = ib_query_port(p->ibdev, p->port_num, &attr);
210 return sprintf(buf, "0x%x\n", attr.sm_lid);
213 static ssize_t sm_sl_show(struct ib_port *p, struct port_attribute *unused,
216 struct ib_port_attr attr;
219 ret = ib_query_port(p->ibdev, p->port_num, &attr);
223 return sprintf(buf, "%d\n", attr.sm_sl);
226 static ssize_t cap_mask_show(struct ib_port *p, struct port_attribute *unused,
229 struct ib_port_attr attr;
232 ret = ib_query_port(p->ibdev, p->port_num, &attr);
236 return sprintf(buf, "0x%08x\n", attr.port_cap_flags);
239 static ssize_t rate_show(struct ib_port *p, struct port_attribute *unused,
242 struct ib_port_attr attr;
244 int rate; /* in deci-Gb/sec */
247 ret = ib_query_port(p->ibdev, p->port_num, &attr);
251 switch (attr.active_speed) {
277 default: /* default to SDR for invalid rates */
283 rate *= ib_width_enum_to_int(attr.active_width);
287 return sprintf(buf, "%d%s Gb/sec (%dX%s)\n",
288 rate / 10, rate % 10 ? ".5" : "",
289 ib_width_enum_to_int(attr.active_width), speed);
292 static const char *phys_state_to_str(enum ib_port_phys_state phys_state)
294 static const char * phys_state_str[] = {
299 "PortConfigurationTraining",
305 if (phys_state < ARRAY_SIZE(phys_state_str))
306 return phys_state_str[phys_state];
310 static ssize_t phys_state_show(struct ib_port *p, struct port_attribute *unused,
313 struct ib_port_attr attr;
317 ret = ib_query_port(p->ibdev, p->port_num, &attr);
321 return sprintf(buf, "%d: %s\n", attr.phys_state,
322 phys_state_to_str(attr.phys_state));
325 static ssize_t link_layer_show(struct ib_port *p, struct port_attribute *unused,
328 switch (rdma_port_get_link_layer(p->ibdev, p->port_num)) {
329 case IB_LINK_LAYER_INFINIBAND:
330 return sprintf(buf, "%s\n", "InfiniBand");
331 case IB_LINK_LAYER_ETHERNET:
332 return sprintf(buf, "%s\n", "Ethernet");
334 return sprintf(buf, "%s\n", "Unknown");
338 static PORT_ATTR_RO(state);
339 static PORT_ATTR_RO(lid);
340 static PORT_ATTR_RO(lid_mask_count);
341 static PORT_ATTR_RO(sm_lid);
342 static PORT_ATTR_RO(sm_sl);
343 static PORT_ATTR_RO(cap_mask);
344 static PORT_ATTR_RO(rate);
345 static PORT_ATTR_RO(phys_state);
346 static PORT_ATTR_RO(link_layer);
348 static struct attribute *port_default_attrs[] = {
349 &port_attr_state.attr,
351 &port_attr_lid_mask_count.attr,
352 &port_attr_sm_lid.attr,
353 &port_attr_sm_sl.attr,
354 &port_attr_cap_mask.attr,
355 &port_attr_rate.attr,
356 &port_attr_phys_state.attr,
357 &port_attr_link_layer.attr,
361 static size_t print_ndev(const struct ib_gid_attr *gid_attr, char *buf)
363 struct net_device *ndev;
364 size_t ret = -EINVAL;
367 ndev = rcu_dereference(gid_attr->ndev);
369 ret = sprintf(buf, "%s\n", ndev->name);
374 static size_t print_gid_type(const struct ib_gid_attr *gid_attr, char *buf)
376 return sprintf(buf, "%s\n", ib_cache_gid_type_str(gid_attr->gid_type));
379 static ssize_t _show_port_gid_attr(
380 struct ib_port *p, struct port_attribute *attr, char *buf,
381 size_t (*print)(const struct ib_gid_attr *gid_attr, char *buf))
383 struct port_table_attribute *tab_attr =
384 container_of(attr, struct port_table_attribute, attr);
385 const struct ib_gid_attr *gid_attr;
388 gid_attr = rdma_get_gid_attr(p->ibdev, p->port_num, tab_attr->index);
389 if (IS_ERR(gid_attr))
390 /* -EINVAL is returned for user space compatibility reasons. */
393 ret = print(gid_attr, buf);
394 rdma_put_gid_attr(gid_attr);
398 static ssize_t show_port_gid(struct ib_port *p, struct port_attribute *attr,
401 struct port_table_attribute *tab_attr =
402 container_of(attr, struct port_table_attribute, attr);
403 const struct ib_gid_attr *gid_attr;
406 gid_attr = rdma_get_gid_attr(p->ibdev, p->port_num, tab_attr->index);
407 if (IS_ERR(gid_attr)) {
408 const union ib_gid zgid = {};
410 /* If reading GID fails, it is likely due to GID entry being
411 * empty (invalid) or reserved GID in the table. User space
412 * expects to read GID table entries as long as it given index
413 * is within GID table size. Administrative/debugging tool
414 * fails to query rest of the GID entries if it hits error
415 * while querying a GID of the given index. To avoid user
416 * space throwing such error on fail to read gid, return zero
417 * GID as before. This maintains backward compatibility.
419 return sprintf(buf, "%pI6\n", zgid.raw);
422 ret = sprintf(buf, "%pI6\n", gid_attr->gid.raw);
423 rdma_put_gid_attr(gid_attr);
427 static ssize_t show_port_gid_attr_ndev(struct ib_port *p,
428 struct port_attribute *attr, char *buf)
430 return _show_port_gid_attr(p, attr, buf, print_ndev);
433 static ssize_t show_port_gid_attr_gid_type(struct ib_port *p,
434 struct port_attribute *attr,
437 return _show_port_gid_attr(p, attr, buf, print_gid_type);
440 static ssize_t show_port_pkey(struct ib_port *p, struct port_attribute *attr,
443 struct port_table_attribute *tab_attr =
444 container_of(attr, struct port_table_attribute, attr);
448 ret = ib_query_pkey(p->ibdev, p->port_num, tab_attr->index, &pkey);
452 return sprintf(buf, "0x%04x\n", pkey);
455 #define PORT_PMA_ATTR(_name, _counter, _width, _offset) \
456 struct port_table_attribute port_pma_attr_##_name = { \
457 .attr = __ATTR(_name, S_IRUGO, show_pma_counter, NULL), \
458 .index = (_offset) | ((_width) << 16) | ((_counter) << 24), \
459 .attr_id = IB_PMA_PORT_COUNTERS , \
462 #define PORT_PMA_ATTR_EXT(_name, _width, _offset) \
463 struct port_table_attribute port_pma_attr_ext_##_name = { \
464 .attr = __ATTR(_name, S_IRUGO, show_pma_counter, NULL), \
465 .index = (_offset) | ((_width) << 16), \
466 .attr_id = IB_PMA_PORT_COUNTERS_EXT , \
470 * Get a Perfmgmt MAD block of data.
471 * Returns error code or the number of bytes retrieved.
473 static int get_perf_mad(struct ib_device *dev, int port_num, __be16 attr,
474 void *data, int offset, size_t size)
476 struct ib_mad *in_mad;
477 struct ib_mad *out_mad;
478 size_t mad_size = sizeof(*out_mad);
479 u16 out_mad_pkey_index = 0;
482 if (!dev->ops.process_mad)
485 in_mad = kzalloc(sizeof(*in_mad), GFP_KERNEL);
486 out_mad = kzalloc(sizeof(*out_mad), GFP_KERNEL);
487 if (!in_mad || !out_mad) {
492 in_mad->mad_hdr.base_version = 1;
493 in_mad->mad_hdr.mgmt_class = IB_MGMT_CLASS_PERF_MGMT;
494 in_mad->mad_hdr.class_version = 1;
495 in_mad->mad_hdr.method = IB_MGMT_METHOD_GET;
496 in_mad->mad_hdr.attr_id = attr;
498 if (attr != IB_PMA_CLASS_PORT_INFO)
499 in_mad->data[41] = port_num; /* PortSelect field */
501 if ((dev->ops.process_mad(dev, IB_MAD_IGNORE_MKEY, port_num, NULL, NULL,
502 in_mad, out_mad, &mad_size,
503 &out_mad_pkey_index) &
504 (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) !=
505 (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) {
509 memcpy(data, out_mad->data + offset, size);
517 static ssize_t show_pma_counter(struct ib_port *p, struct port_attribute *attr,
520 struct port_table_attribute *tab_attr =
521 container_of(attr, struct port_table_attribute, attr);
522 int offset = tab_attr->index & 0xffff;
523 int width = (tab_attr->index >> 16) & 0xff;
527 ret = get_perf_mad(p->ibdev, p->port_num, tab_attr->attr_id, &data,
528 40 + offset / 8, sizeof(data));
534 ret = sprintf(buf, "%u\n", (*data >>
535 (4 - (offset % 8))) & 0xf);
538 ret = sprintf(buf, "%u\n", *data);
541 ret = sprintf(buf, "%u\n",
542 be16_to_cpup((__be16 *)data));
545 ret = sprintf(buf, "%u\n",
546 be32_to_cpup((__be32 *)data));
549 ret = sprintf(buf, "%llu\n",
550 be64_to_cpup((__be64 *)data));
560 static PORT_PMA_ATTR(symbol_error , 0, 16, 32);
561 static PORT_PMA_ATTR(link_error_recovery , 1, 8, 48);
562 static PORT_PMA_ATTR(link_downed , 2, 8, 56);
563 static PORT_PMA_ATTR(port_rcv_errors , 3, 16, 64);
564 static PORT_PMA_ATTR(port_rcv_remote_physical_errors, 4, 16, 80);
565 static PORT_PMA_ATTR(port_rcv_switch_relay_errors , 5, 16, 96);
566 static PORT_PMA_ATTR(port_xmit_discards , 6, 16, 112);
567 static PORT_PMA_ATTR(port_xmit_constraint_errors , 7, 8, 128);
568 static PORT_PMA_ATTR(port_rcv_constraint_errors , 8, 8, 136);
569 static PORT_PMA_ATTR(local_link_integrity_errors , 9, 4, 152);
570 static PORT_PMA_ATTR(excessive_buffer_overrun_errors, 10, 4, 156);
571 static PORT_PMA_ATTR(VL15_dropped , 11, 16, 176);
572 static PORT_PMA_ATTR(port_xmit_data , 12, 32, 192);
573 static PORT_PMA_ATTR(port_rcv_data , 13, 32, 224);
574 static PORT_PMA_ATTR(port_xmit_packets , 14, 32, 256);
575 static PORT_PMA_ATTR(port_rcv_packets , 15, 32, 288);
576 static PORT_PMA_ATTR(port_xmit_wait , 0, 32, 320);
579 * Counters added by extended set
581 static PORT_PMA_ATTR_EXT(port_xmit_data , 64, 64);
582 static PORT_PMA_ATTR_EXT(port_rcv_data , 64, 128);
583 static PORT_PMA_ATTR_EXT(port_xmit_packets , 64, 192);
584 static PORT_PMA_ATTR_EXT(port_rcv_packets , 64, 256);
585 static PORT_PMA_ATTR_EXT(unicast_xmit_packets , 64, 320);
586 static PORT_PMA_ATTR_EXT(unicast_rcv_packets , 64, 384);
587 static PORT_PMA_ATTR_EXT(multicast_xmit_packets , 64, 448);
588 static PORT_PMA_ATTR_EXT(multicast_rcv_packets , 64, 512);
590 static struct attribute *pma_attrs[] = {
591 &port_pma_attr_symbol_error.attr.attr,
592 &port_pma_attr_link_error_recovery.attr.attr,
593 &port_pma_attr_link_downed.attr.attr,
594 &port_pma_attr_port_rcv_errors.attr.attr,
595 &port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
596 &port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
597 &port_pma_attr_port_xmit_discards.attr.attr,
598 &port_pma_attr_port_xmit_constraint_errors.attr.attr,
599 &port_pma_attr_port_rcv_constraint_errors.attr.attr,
600 &port_pma_attr_local_link_integrity_errors.attr.attr,
601 &port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
602 &port_pma_attr_VL15_dropped.attr.attr,
603 &port_pma_attr_port_xmit_data.attr.attr,
604 &port_pma_attr_port_rcv_data.attr.attr,
605 &port_pma_attr_port_xmit_packets.attr.attr,
606 &port_pma_attr_port_rcv_packets.attr.attr,
607 &port_pma_attr_port_xmit_wait.attr.attr,
611 static struct attribute *pma_attrs_ext[] = {
612 &port_pma_attr_symbol_error.attr.attr,
613 &port_pma_attr_link_error_recovery.attr.attr,
614 &port_pma_attr_link_downed.attr.attr,
615 &port_pma_attr_port_rcv_errors.attr.attr,
616 &port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
617 &port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
618 &port_pma_attr_port_xmit_discards.attr.attr,
619 &port_pma_attr_port_xmit_constraint_errors.attr.attr,
620 &port_pma_attr_port_rcv_constraint_errors.attr.attr,
621 &port_pma_attr_local_link_integrity_errors.attr.attr,
622 &port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
623 &port_pma_attr_VL15_dropped.attr.attr,
624 &port_pma_attr_ext_port_xmit_data.attr.attr,
625 &port_pma_attr_ext_port_rcv_data.attr.attr,
626 &port_pma_attr_ext_port_xmit_packets.attr.attr,
627 &port_pma_attr_port_xmit_wait.attr.attr,
628 &port_pma_attr_ext_port_rcv_packets.attr.attr,
629 &port_pma_attr_ext_unicast_rcv_packets.attr.attr,
630 &port_pma_attr_ext_unicast_xmit_packets.attr.attr,
631 &port_pma_attr_ext_multicast_rcv_packets.attr.attr,
632 &port_pma_attr_ext_multicast_xmit_packets.attr.attr,
636 static struct attribute *pma_attrs_noietf[] = {
637 &port_pma_attr_symbol_error.attr.attr,
638 &port_pma_attr_link_error_recovery.attr.attr,
639 &port_pma_attr_link_downed.attr.attr,
640 &port_pma_attr_port_rcv_errors.attr.attr,
641 &port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
642 &port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
643 &port_pma_attr_port_xmit_discards.attr.attr,
644 &port_pma_attr_port_xmit_constraint_errors.attr.attr,
645 &port_pma_attr_port_rcv_constraint_errors.attr.attr,
646 &port_pma_attr_local_link_integrity_errors.attr.attr,
647 &port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
648 &port_pma_attr_VL15_dropped.attr.attr,
649 &port_pma_attr_ext_port_xmit_data.attr.attr,
650 &port_pma_attr_ext_port_rcv_data.attr.attr,
651 &port_pma_attr_ext_port_xmit_packets.attr.attr,
652 &port_pma_attr_ext_port_rcv_packets.attr.attr,
653 &port_pma_attr_port_xmit_wait.attr.attr,
657 static const struct attribute_group pma_group = {
662 static const struct attribute_group pma_group_ext = {
664 .attrs = pma_attrs_ext
667 static const struct attribute_group pma_group_noietf = {
669 .attrs = pma_attrs_noietf
672 static void ib_port_release(struct kobject *kobj)
674 struct ib_port *p = container_of(kobj, struct ib_port, kobj);
678 if (p->gid_group.attrs) {
679 for (i = 0; (a = p->gid_group.attrs[i]); ++i)
682 kfree(p->gid_group.attrs);
686 if (p->pkey_group->attrs) {
687 for (i = 0; (a = p->pkey_group->attrs[i]); ++i)
690 kfree(p->pkey_group->attrs);
693 kfree(p->pkey_group);
694 p->pkey_group = NULL;
700 static void ib_port_gid_attr_release(struct kobject *kobj)
702 struct gid_attr_group *g = container_of(kobj, struct gid_attr_group,
708 for (i = 0; (a = g->ndev.attrs[i]); ++i)
711 kfree(g->ndev.attrs);
715 for (i = 0; (a = g->type.attrs[i]); ++i)
718 kfree(g->type.attrs);
724 static struct kobj_type port_type = {
725 .release = ib_port_release,
726 .sysfs_ops = &port_sysfs_ops,
727 .default_attrs = port_default_attrs
730 static struct kobj_type gid_attr_type = {
731 .sysfs_ops = &gid_attr_sysfs_ops,
732 .release = ib_port_gid_attr_release
735 static struct attribute **
736 alloc_group_attrs(ssize_t (*show)(struct ib_port *,
737 struct port_attribute *, char *buf),
740 struct attribute **tab_attr;
741 struct port_table_attribute *element;
744 tab_attr = kcalloc(1 + len, sizeof(struct attribute *), GFP_KERNEL);
748 for (i = 0; i < len; i++) {
749 element = kzalloc(sizeof(struct port_table_attribute),
754 if (snprintf(element->name, sizeof(element->name),
755 "%d", i) >= sizeof(element->name)) {
760 element->attr.attr.name = element->name;
761 element->attr.attr.mode = S_IRUGO;
762 element->attr.show = show;
764 sysfs_attr_init(&element->attr.attr);
766 tab_attr[i] = &element->attr.attr;
779 * Figure out which counter table to use depending on
780 * the device capabilities.
782 static const struct attribute_group *get_counter_table(struct ib_device *dev,
785 struct ib_class_port_info cpi;
787 if (get_perf_mad(dev, port_num, IB_PMA_CLASS_PORT_INFO,
788 &cpi, 40, sizeof(cpi)) >= 0) {
789 if (cpi.capability_mask & IB_PMA_CLASS_CAP_EXT_WIDTH)
790 /* We have extended counters */
791 return &pma_group_ext;
793 if (cpi.capability_mask & IB_PMA_CLASS_CAP_EXT_WIDTH_NOIETF)
794 /* But not the IETF ones */
795 return &pma_group_noietf;
798 /* Fall back to normal counters */
802 static int update_hw_stats(struct ib_device *dev, struct rdma_hw_stats *stats,
803 u8 port_num, int index)
807 if (time_is_after_eq_jiffies(stats->timestamp + stats->lifespan))
809 ret = dev->ops.get_hw_stats(dev, stats, port_num, index);
812 if (ret == stats->num_counters)
813 stats->timestamp = jiffies;
818 static ssize_t print_hw_stat(struct ib_device *dev, int port_num,
819 struct rdma_hw_stats *stats, int index, char *buf)
821 u64 v = rdma_counter_get_hwstat_value(dev, port_num, index);
823 return sprintf(buf, "%llu\n", stats->value[index] + v);
826 static ssize_t show_hw_stats(struct kobject *kobj, struct attribute *attr,
829 struct ib_device *dev;
830 struct ib_port *port;
831 struct hw_stats_attribute *hsa;
832 struct rdma_hw_stats *stats;
835 hsa = container_of(attr, struct hw_stats_attribute, attr);
836 if (!hsa->port_num) {
837 dev = container_of((struct device *)kobj,
838 struct ib_device, dev);
839 stats = dev->hw_stats;
841 port = container_of(kobj, struct ib_port, kobj);
843 stats = port->hw_stats;
845 mutex_lock(&stats->lock);
846 ret = update_hw_stats(dev, stats, hsa->port_num, hsa->index);
849 ret = print_hw_stat(dev, hsa->port_num, stats, hsa->index, buf);
851 mutex_unlock(&stats->lock);
856 static ssize_t show_stats_lifespan(struct kobject *kobj,
857 struct attribute *attr,
860 struct hw_stats_attribute *hsa;
861 struct rdma_hw_stats *stats;
864 hsa = container_of(attr, struct hw_stats_attribute, attr);
865 if (!hsa->port_num) {
866 struct ib_device *dev = container_of((struct device *)kobj,
867 struct ib_device, dev);
869 stats = dev->hw_stats;
871 struct ib_port *p = container_of(kobj, struct ib_port, kobj);
876 mutex_lock(&stats->lock);
877 msecs = jiffies_to_msecs(stats->lifespan);
878 mutex_unlock(&stats->lock);
880 return sprintf(buf, "%d\n", msecs);
883 static ssize_t set_stats_lifespan(struct kobject *kobj,
884 struct attribute *attr,
885 const char *buf, size_t count)
887 struct hw_stats_attribute *hsa;
888 struct rdma_hw_stats *stats;
893 ret = kstrtoint(buf, 10, &msecs);
896 if (msecs < 0 || msecs > 10000)
898 jiffies = msecs_to_jiffies(msecs);
899 hsa = container_of(attr, struct hw_stats_attribute, attr);
900 if (!hsa->port_num) {
901 struct ib_device *dev = container_of((struct device *)kobj,
902 struct ib_device, dev);
904 stats = dev->hw_stats;
906 struct ib_port *p = container_of(kobj, struct ib_port, kobj);
911 mutex_lock(&stats->lock);
912 stats->lifespan = jiffies;
913 mutex_unlock(&stats->lock);
918 static void free_hsag(struct kobject *kobj, struct attribute_group *attr_group)
920 struct attribute **attr;
922 sysfs_remove_group(kobj, attr_group);
924 for (attr = attr_group->attrs; *attr; attr++)
929 static struct attribute *alloc_hsa(int index, u8 port_num, const char *name)
931 struct hw_stats_attribute *hsa;
933 hsa = kmalloc(sizeof(*hsa), GFP_KERNEL);
937 hsa->attr.name = (char *)name;
938 hsa->attr.mode = S_IRUGO;
939 hsa->show = show_hw_stats;
942 hsa->port_num = port_num;
947 static struct attribute *alloc_hsa_lifespan(char *name, u8 port_num)
949 struct hw_stats_attribute *hsa;
951 hsa = kmalloc(sizeof(*hsa), GFP_KERNEL);
955 hsa->attr.name = name;
956 hsa->attr.mode = S_IWUSR | S_IRUGO;
957 hsa->show = show_stats_lifespan;
958 hsa->store = set_stats_lifespan;
960 hsa->port_num = port_num;
965 static void setup_hw_stats(struct ib_device *device, struct ib_port *port,
968 struct attribute_group *hsag;
969 struct rdma_hw_stats *stats;
972 stats = device->ops.alloc_hw_stats(device, port_num);
977 if (!stats->names || stats->num_counters <= 0)
981 * Two extra attribue elements here, one for the lifespan entry and
982 * one to NULL terminate the list for the sysfs core code
984 hsag = kzalloc(sizeof(*hsag) +
985 sizeof(void *) * (stats->num_counters + 2),
990 ret = device->ops.get_hw_stats(device, stats, port_num,
991 stats->num_counters);
992 if (ret != stats->num_counters)
995 stats->timestamp = jiffies;
997 hsag->name = "hw_counters";
998 hsag->attrs = (void *)hsag + sizeof(*hsag);
1000 for (i = 0; i < stats->num_counters; i++) {
1001 hsag->attrs[i] = alloc_hsa(i, port_num, stats->names[i]);
1002 if (!hsag->attrs[i])
1004 sysfs_attr_init(hsag->attrs[i]);
1007 mutex_init(&stats->lock);
1008 /* treat an error here as non-fatal */
1009 hsag->attrs[i] = alloc_hsa_lifespan("lifespan", port_num);
1011 sysfs_attr_init(hsag->attrs[i]);
1014 struct kobject *kobj = &port->kobj;
1015 ret = sysfs_create_group(kobj, hsag);
1018 port->hw_stats_ag = hsag;
1019 port->hw_stats = stats;
1020 if (device->port_data)
1021 device->port_data[port_num].hw_stats = stats;
1023 struct kobject *kobj = &device->dev.kobj;
1024 ret = sysfs_create_group(kobj, hsag);
1027 device->hw_stats_ag = hsag;
1028 device->hw_stats = stats;
1035 kfree(hsag->attrs[i]);
1043 static int add_port(struct ib_core_device *coredev, int port_num)
1045 struct ib_device *device = rdma_device_to_ibdev(&coredev->dev);
1046 bool is_full_dev = &device->coredev == coredev;
1048 struct ib_port_attr attr;
1052 ret = ib_query_port(device, port_num, &attr);
1056 p = kzalloc(sizeof *p, GFP_KERNEL);
1061 p->port_num = port_num;
1063 ret = kobject_init_and_add(&p->kobj, &port_type,
1064 coredev->ports_kobj,
1070 p->gid_attr_group = kzalloc(sizeof(*p->gid_attr_group), GFP_KERNEL);
1071 if (!p->gid_attr_group) {
1076 p->gid_attr_group->port = p;
1077 ret = kobject_init_and_add(&p->gid_attr_group->kobj, &gid_attr_type,
1078 &p->kobj, "gid_attrs");
1080 goto err_put_gid_attrs;
1083 if (device->ops.process_mad && is_full_dev) {
1084 p->pma_table = get_counter_table(device, port_num);
1085 ret = sysfs_create_group(&p->kobj, p->pma_table);
1087 goto err_put_gid_attrs;
1090 p->gid_group.name = "gids";
1091 p->gid_group.attrs = alloc_group_attrs(show_port_gid, attr.gid_tbl_len);
1092 if (!p->gid_group.attrs) {
1094 goto err_remove_pma;
1097 ret = sysfs_create_group(&p->kobj, &p->gid_group);
1101 p->gid_attr_group->ndev.name = "ndevs";
1102 p->gid_attr_group->ndev.attrs = alloc_group_attrs(show_port_gid_attr_ndev,
1104 if (!p->gid_attr_group->ndev.attrs) {
1106 goto err_remove_gid;
1109 ret = sysfs_create_group(&p->gid_attr_group->kobj,
1110 &p->gid_attr_group->ndev);
1112 goto err_free_gid_ndev;
1114 p->gid_attr_group->type.name = "types";
1115 p->gid_attr_group->type.attrs = alloc_group_attrs(show_port_gid_attr_gid_type,
1117 if (!p->gid_attr_group->type.attrs) {
1119 goto err_remove_gid_ndev;
1122 ret = sysfs_create_group(&p->gid_attr_group->kobj,
1123 &p->gid_attr_group->type);
1125 goto err_free_gid_type;
1127 if (attr.pkey_tbl_len) {
1128 p->pkey_group = kzalloc(sizeof(*p->pkey_group), GFP_KERNEL);
1129 if (!p->pkey_group) {
1131 goto err_remove_gid_type;
1134 p->pkey_group->name = "pkeys";
1135 p->pkey_group->attrs = alloc_group_attrs(show_port_pkey,
1137 if (!p->pkey_group->attrs) {
1139 goto err_free_pkey_group;
1142 ret = sysfs_create_group(&p->kobj, p->pkey_group);
1148 if (device->ops.init_port && is_full_dev) {
1149 ret = device->ops.init_port(device, port_num, &p->kobj);
1151 goto err_remove_pkey;
1155 * If port == 0, it means hw_counters are per device and not per
1156 * port, so holder should be device. Therefore skip per port conunter
1159 if (device->ops.alloc_hw_stats && port_num && is_full_dev)
1160 setup_hw_stats(device, p, port_num);
1162 list_add_tail(&p->kobj.entry, &coredev->port_list);
1164 kobject_uevent(&p->kobj, KOBJ_ADD);
1169 sysfs_remove_group(&p->kobj, p->pkey_group);
1172 if (p->pkey_group) {
1173 for (i = 0; i < attr.pkey_tbl_len; ++i)
1174 kfree(p->pkey_group->attrs[i]);
1176 kfree(p->pkey_group->attrs);
1177 p->pkey_group->attrs = NULL;
1180 err_free_pkey_group:
1181 kfree(p->pkey_group);
1183 err_remove_gid_type:
1184 sysfs_remove_group(&p->gid_attr_group->kobj,
1185 &p->gid_attr_group->type);
1188 for (i = 0; i < attr.gid_tbl_len; ++i)
1189 kfree(p->gid_attr_group->type.attrs[i]);
1191 kfree(p->gid_attr_group->type.attrs);
1192 p->gid_attr_group->type.attrs = NULL;
1194 err_remove_gid_ndev:
1195 sysfs_remove_group(&p->gid_attr_group->kobj,
1196 &p->gid_attr_group->ndev);
1199 for (i = 0; i < attr.gid_tbl_len; ++i)
1200 kfree(p->gid_attr_group->ndev.attrs[i]);
1202 kfree(p->gid_attr_group->ndev.attrs);
1203 p->gid_attr_group->ndev.attrs = NULL;
1206 sysfs_remove_group(&p->kobj, &p->gid_group);
1209 for (i = 0; i < attr.gid_tbl_len; ++i)
1210 kfree(p->gid_group.attrs[i]);
1212 kfree(p->gid_group.attrs);
1213 p->gid_group.attrs = NULL;
1217 sysfs_remove_group(&p->kobj, p->pma_table);
1220 kobject_put(&p->gid_attr_group->kobj);
1223 kobject_put(&p->kobj);
1227 static ssize_t node_type_show(struct device *device,
1228 struct device_attribute *attr, char *buf)
1230 struct ib_device *dev = rdma_device_to_ibdev(device);
1232 switch (dev->node_type) {
1233 case RDMA_NODE_IB_CA: return sprintf(buf, "%d: CA\n", dev->node_type);
1234 case RDMA_NODE_RNIC: return sprintf(buf, "%d: RNIC\n", dev->node_type);
1235 case RDMA_NODE_USNIC: return sprintf(buf, "%d: usNIC\n", dev->node_type);
1236 case RDMA_NODE_USNIC_UDP: return sprintf(buf, "%d: usNIC UDP\n", dev->node_type);
1237 case RDMA_NODE_UNSPECIFIED: return sprintf(buf, "%d: unspecified\n", dev->node_type);
1238 case RDMA_NODE_IB_SWITCH: return sprintf(buf, "%d: switch\n", dev->node_type);
1239 case RDMA_NODE_IB_ROUTER: return sprintf(buf, "%d: router\n", dev->node_type);
1240 default: return sprintf(buf, "%d: <unknown>\n", dev->node_type);
1243 static DEVICE_ATTR_RO(node_type);
1245 static ssize_t sys_image_guid_show(struct device *device,
1246 struct device_attribute *dev_attr, char *buf)
1248 struct ib_device *dev = rdma_device_to_ibdev(device);
1250 return sprintf(buf, "%04x:%04x:%04x:%04x\n",
1251 be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[0]),
1252 be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[1]),
1253 be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[2]),
1254 be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[3]));
1256 static DEVICE_ATTR_RO(sys_image_guid);
1258 static ssize_t node_guid_show(struct device *device,
1259 struct device_attribute *attr, char *buf)
1261 struct ib_device *dev = rdma_device_to_ibdev(device);
1263 return sprintf(buf, "%04x:%04x:%04x:%04x\n",
1264 be16_to_cpu(((__be16 *) &dev->node_guid)[0]),
1265 be16_to_cpu(((__be16 *) &dev->node_guid)[1]),
1266 be16_to_cpu(((__be16 *) &dev->node_guid)[2]),
1267 be16_to_cpu(((__be16 *) &dev->node_guid)[3]));
1269 static DEVICE_ATTR_RO(node_guid);
1271 static ssize_t node_desc_show(struct device *device,
1272 struct device_attribute *attr, char *buf)
1274 struct ib_device *dev = rdma_device_to_ibdev(device);
1276 return sprintf(buf, "%.64s\n", dev->node_desc);
1279 static ssize_t node_desc_store(struct device *device,
1280 struct device_attribute *attr,
1281 const char *buf, size_t count)
1283 struct ib_device *dev = rdma_device_to_ibdev(device);
1284 struct ib_device_modify desc = {};
1287 if (!dev->ops.modify_device)
1290 memcpy(desc.node_desc, buf, min_t(int, count, IB_DEVICE_NODE_DESC_MAX));
1291 ret = ib_modify_device(dev, IB_DEVICE_MODIFY_NODE_DESC, &desc);
1297 static DEVICE_ATTR_RW(node_desc);
1299 static ssize_t fw_ver_show(struct device *device, struct device_attribute *attr,
1302 struct ib_device *dev = rdma_device_to_ibdev(device);
1304 ib_get_device_fw_str(dev, buf);
1305 strlcat(buf, "\n", IB_FW_VERSION_NAME_MAX);
1308 static DEVICE_ATTR_RO(fw_ver);
1310 static struct attribute *ib_dev_attrs[] = {
1311 &dev_attr_node_type.attr,
1312 &dev_attr_node_guid.attr,
1313 &dev_attr_sys_image_guid.attr,
1314 &dev_attr_fw_ver.attr,
1315 &dev_attr_node_desc.attr,
1319 const struct attribute_group ib_dev_attr_group = {
1320 .attrs = ib_dev_attrs,
1323 void ib_free_port_attrs(struct ib_core_device *coredev)
1325 struct ib_device *device = rdma_device_to_ibdev(&coredev->dev);
1326 bool is_full_dev = &device->coredev == coredev;
1327 struct kobject *p, *t;
1329 list_for_each_entry_safe(p, t, &coredev->port_list, entry) {
1330 struct ib_port *port = container_of(p, struct ib_port, kobj);
1332 list_del(&p->entry);
1333 if (port->hw_stats_ag)
1334 free_hsag(&port->kobj, port->hw_stats_ag);
1335 kfree(port->hw_stats);
1336 if (device->port_data && is_full_dev)
1337 device->port_data[port->port_num].hw_stats = NULL;
1339 if (port->pma_table)
1340 sysfs_remove_group(p, port->pma_table);
1341 if (port->pkey_group)
1342 sysfs_remove_group(p, port->pkey_group);
1343 sysfs_remove_group(p, &port->gid_group);
1344 sysfs_remove_group(&port->gid_attr_group->kobj,
1345 &port->gid_attr_group->ndev);
1346 sysfs_remove_group(&port->gid_attr_group->kobj,
1347 &port->gid_attr_group->type);
1348 kobject_put(&port->gid_attr_group->kobj);
1352 kobject_put(coredev->ports_kobj);
1355 int ib_setup_port_attrs(struct ib_core_device *coredev)
1357 struct ib_device *device = rdma_device_to_ibdev(&coredev->dev);
1361 coredev->ports_kobj = kobject_create_and_add("ports",
1362 &coredev->dev.kobj);
1363 if (!coredev->ports_kobj)
1366 rdma_for_each_port (device, port) {
1367 ret = add_port(coredev, port);
1375 ib_free_port_attrs(coredev);
1379 int ib_device_register_sysfs(struct ib_device *device)
1383 ret = ib_setup_port_attrs(&device->coredev);
1387 if (device->ops.alloc_hw_stats)
1388 setup_hw_stats(device, NULL, 0);
1393 void ib_device_unregister_sysfs(struct ib_device *device)
1395 if (device->hw_stats_ag)
1396 free_hsag(&device->dev.kobj, device->hw_stats_ag);
1397 kfree(device->hw_stats);
1399 ib_free_port_attrs(&device->coredev);
1403 * ib_port_register_module_stat - add module counters under relevant port
1406 * @device: IB device to add counters
1407 * @port_num: valid port number
1408 * @kobj: pointer to the kobject to initialize
1409 * @ktype: pointer to the ktype for this kobject.
1410 * @name: the name of the kobject
1412 int ib_port_register_module_stat(struct ib_device *device, u8 port_num,
1413 struct kobject *kobj, struct kobj_type *ktype,
1416 struct kobject *p, *t;
1419 list_for_each_entry_safe(p, t, &device->coredev.port_list, entry) {
1420 struct ib_port *port = container_of(p, struct ib_port, kobj);
1422 if (port->port_num != port_num)
1425 ret = kobject_init_and_add(kobj, ktype, &port->kobj, "%s",
1435 EXPORT_SYMBOL(ib_port_register_module_stat);
1438 * ib_port_unregister_module_stat - release module counters
1439 * @kobj: pointer to the kobject to release
1441 void ib_port_unregister_module_stat(struct kobject *kobj)
1445 EXPORT_SYMBOL(ib_port_unregister_module_stat);