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>
49 struct gid_attr_group {
52 struct attribute_group ndev;
53 struct attribute_group type;
57 struct ib_device *ibdev;
58 struct gid_attr_group *gid_attr_group;
59 struct attribute_group gid_group;
60 struct attribute_group pkey_group;
61 struct attribute_group *pma_table;
62 struct attribute_group *hw_stats_ag;
63 struct rdma_hw_stats *hw_stats;
67 struct port_attribute {
68 struct attribute attr;
69 ssize_t (*show)(struct ib_port *, struct port_attribute *, char *buf);
70 ssize_t (*store)(struct ib_port *, struct port_attribute *,
71 const char *buf, size_t count);
74 #define PORT_ATTR(_name, _mode, _show, _store) \
75 struct port_attribute port_attr_##_name = __ATTR(_name, _mode, _show, _store)
77 #define PORT_ATTR_RO(_name) \
78 struct port_attribute port_attr_##_name = __ATTR_RO(_name)
80 struct port_table_attribute {
81 struct port_attribute attr;
87 struct hw_stats_attribute {
88 struct attribute attr;
89 ssize_t (*show)(struct kobject *kobj,
90 struct attribute *attr, char *buf);
91 ssize_t (*store)(struct kobject *kobj,
92 struct attribute *attr,
99 static ssize_t port_attr_show(struct kobject *kobj,
100 struct attribute *attr, char *buf)
102 struct port_attribute *port_attr =
103 container_of(attr, struct port_attribute, attr);
104 struct ib_port *p = container_of(kobj, struct ib_port, kobj);
106 if (!port_attr->show)
109 return port_attr->show(p, port_attr, buf);
112 static ssize_t port_attr_store(struct kobject *kobj,
113 struct attribute *attr,
114 const char *buf, size_t count)
116 struct port_attribute *port_attr =
117 container_of(attr, struct port_attribute, attr);
118 struct ib_port *p = container_of(kobj, struct ib_port, kobj);
120 if (!port_attr->store)
122 return port_attr->store(p, port_attr, buf, count);
125 static const struct sysfs_ops port_sysfs_ops = {
126 .show = port_attr_show,
127 .store = port_attr_store
130 static ssize_t gid_attr_show(struct kobject *kobj,
131 struct attribute *attr, char *buf)
133 struct port_attribute *port_attr =
134 container_of(attr, struct port_attribute, attr);
135 struct ib_port *p = container_of(kobj, struct gid_attr_group,
138 if (!port_attr->show)
141 return port_attr->show(p, port_attr, buf);
144 static const struct sysfs_ops gid_attr_sysfs_ops = {
145 .show = gid_attr_show
148 static ssize_t state_show(struct ib_port *p, struct port_attribute *unused,
151 struct ib_port_attr attr;
154 static const char *state_name[] = {
155 [IB_PORT_NOP] = "NOP",
156 [IB_PORT_DOWN] = "DOWN",
157 [IB_PORT_INIT] = "INIT",
158 [IB_PORT_ARMED] = "ARMED",
159 [IB_PORT_ACTIVE] = "ACTIVE",
160 [IB_PORT_ACTIVE_DEFER] = "ACTIVE_DEFER"
163 ret = ib_query_port(p->ibdev, p->port_num, &attr);
167 return sprintf(buf, "%d: %s\n", attr.state,
168 attr.state >= 0 && attr.state < ARRAY_SIZE(state_name) ?
169 state_name[attr.state] : "UNKNOWN");
172 static ssize_t lid_show(struct ib_port *p, struct port_attribute *unused,
175 struct ib_port_attr attr;
178 ret = ib_query_port(p->ibdev, p->port_num, &attr);
182 return sprintf(buf, "0x%x\n", attr.lid);
185 static ssize_t lid_mask_count_show(struct ib_port *p,
186 struct port_attribute *unused,
189 struct ib_port_attr attr;
192 ret = ib_query_port(p->ibdev, p->port_num, &attr);
196 return sprintf(buf, "%d\n", attr.lmc);
199 static ssize_t sm_lid_show(struct ib_port *p, struct port_attribute *unused,
202 struct ib_port_attr attr;
205 ret = ib_query_port(p->ibdev, p->port_num, &attr);
209 return sprintf(buf, "0x%x\n", attr.sm_lid);
212 static ssize_t sm_sl_show(struct ib_port *p, struct port_attribute *unused,
215 struct ib_port_attr attr;
218 ret = ib_query_port(p->ibdev, p->port_num, &attr);
222 return sprintf(buf, "%d\n", attr.sm_sl);
225 static ssize_t cap_mask_show(struct ib_port *p, struct port_attribute *unused,
228 struct ib_port_attr attr;
231 ret = ib_query_port(p->ibdev, p->port_num, &attr);
235 return sprintf(buf, "0x%08x\n", attr.port_cap_flags);
238 static ssize_t rate_show(struct ib_port *p, struct port_attribute *unused,
241 struct ib_port_attr attr;
243 int rate; /* in deci-Gb/sec */
246 ret = ib_query_port(p->ibdev, p->port_num, &attr);
250 switch (attr.active_speed) {
276 default: /* default to SDR for invalid rates */
282 rate *= ib_width_enum_to_int(attr.active_width);
286 return sprintf(buf, "%d%s Gb/sec (%dX%s)\n",
287 rate / 10, rate % 10 ? ".5" : "",
288 ib_width_enum_to_int(attr.active_width), speed);
291 static ssize_t phys_state_show(struct ib_port *p, struct port_attribute *unused,
294 struct ib_port_attr attr;
298 ret = ib_query_port(p->ibdev, p->port_num, &attr);
302 switch (attr.phys_state) {
303 case 1: return sprintf(buf, "1: Sleep\n");
304 case 2: return sprintf(buf, "2: Polling\n");
305 case 3: return sprintf(buf, "3: Disabled\n");
306 case 4: return sprintf(buf, "4: PortConfigurationTraining\n");
307 case 5: return sprintf(buf, "5: LinkUp\n");
308 case 6: return sprintf(buf, "6: LinkErrorRecovery\n");
309 case 7: return sprintf(buf, "7: Phy Test\n");
310 default: return sprintf(buf, "%d: <unknown>\n", attr.phys_state);
314 static ssize_t link_layer_show(struct ib_port *p, struct port_attribute *unused,
317 switch (rdma_port_get_link_layer(p->ibdev, p->port_num)) {
318 case IB_LINK_LAYER_INFINIBAND:
319 return sprintf(buf, "%s\n", "InfiniBand");
320 case IB_LINK_LAYER_ETHERNET:
321 return sprintf(buf, "%s\n", "Ethernet");
323 return sprintf(buf, "%s\n", "Unknown");
327 static PORT_ATTR_RO(state);
328 static PORT_ATTR_RO(lid);
329 static PORT_ATTR_RO(lid_mask_count);
330 static PORT_ATTR_RO(sm_lid);
331 static PORT_ATTR_RO(sm_sl);
332 static PORT_ATTR_RO(cap_mask);
333 static PORT_ATTR_RO(rate);
334 static PORT_ATTR_RO(phys_state);
335 static PORT_ATTR_RO(link_layer);
337 static struct attribute *port_default_attrs[] = {
338 &port_attr_state.attr,
340 &port_attr_lid_mask_count.attr,
341 &port_attr_sm_lid.attr,
342 &port_attr_sm_sl.attr,
343 &port_attr_cap_mask.attr,
344 &port_attr_rate.attr,
345 &port_attr_phys_state.attr,
346 &port_attr_link_layer.attr,
350 static size_t print_ndev(const struct ib_gid_attr *gid_attr, char *buf)
352 struct net_device *ndev;
353 size_t ret = -EINVAL;
356 ndev = rcu_dereference(gid_attr->ndev);
358 ret = sprintf(buf, "%s\n", ndev->name);
363 static size_t print_gid_type(const struct ib_gid_attr *gid_attr, char *buf)
365 return sprintf(buf, "%s\n", ib_cache_gid_type_str(gid_attr->gid_type));
368 static ssize_t _show_port_gid_attr(
369 struct ib_port *p, struct port_attribute *attr, char *buf,
370 size_t (*print)(const struct ib_gid_attr *gid_attr, char *buf))
372 struct port_table_attribute *tab_attr =
373 container_of(attr, struct port_table_attribute, attr);
374 const struct ib_gid_attr *gid_attr;
377 gid_attr = rdma_get_gid_attr(p->ibdev, p->port_num, tab_attr->index);
378 if (IS_ERR(gid_attr))
379 return PTR_ERR(gid_attr);
381 ret = print(gid_attr, buf);
382 rdma_put_gid_attr(gid_attr);
386 static ssize_t show_port_gid(struct ib_port *p, struct port_attribute *attr,
389 struct port_table_attribute *tab_attr =
390 container_of(attr, struct port_table_attribute, attr);
391 const struct ib_gid_attr *gid_attr;
394 gid_attr = rdma_get_gid_attr(p->ibdev, p->port_num, tab_attr->index);
395 if (IS_ERR(gid_attr)) {
396 const union ib_gid zgid = {};
398 /* If reading GID fails, it is likely due to GID entry being
399 * empty (invalid) or reserved GID in the table. User space
400 * expects to read GID table entries as long as it given index
401 * is within GID table size. Administrative/debugging tool
402 * fails to query rest of the GID entries if it hits error
403 * while querying a GID of the given index. To avoid user
404 * space throwing such error on fail to read gid, return zero
405 * GID as before. This maintains backward compatibility.
407 return sprintf(buf, "%pI6\n", zgid.raw);
410 ret = sprintf(buf, "%pI6\n", gid_attr->gid.raw);
411 rdma_put_gid_attr(gid_attr);
415 static ssize_t show_port_gid_attr_ndev(struct ib_port *p,
416 struct port_attribute *attr, char *buf)
418 return _show_port_gid_attr(p, attr, buf, print_ndev);
421 static ssize_t show_port_gid_attr_gid_type(struct ib_port *p,
422 struct port_attribute *attr,
425 return _show_port_gid_attr(p, attr, buf, print_gid_type);
428 static ssize_t show_port_pkey(struct ib_port *p, struct port_attribute *attr,
431 struct port_table_attribute *tab_attr =
432 container_of(attr, struct port_table_attribute, attr);
436 ret = ib_query_pkey(p->ibdev, p->port_num, tab_attr->index, &pkey);
440 return sprintf(buf, "0x%04x\n", pkey);
443 #define PORT_PMA_ATTR(_name, _counter, _width, _offset) \
444 struct port_table_attribute port_pma_attr_##_name = { \
445 .attr = __ATTR(_name, S_IRUGO, show_pma_counter, NULL), \
446 .index = (_offset) | ((_width) << 16) | ((_counter) << 24), \
447 .attr_id = IB_PMA_PORT_COUNTERS , \
450 #define PORT_PMA_ATTR_EXT(_name, _width, _offset) \
451 struct port_table_attribute port_pma_attr_ext_##_name = { \
452 .attr = __ATTR(_name, S_IRUGO, show_pma_counter, NULL), \
453 .index = (_offset) | ((_width) << 16), \
454 .attr_id = IB_PMA_PORT_COUNTERS_EXT , \
458 * Get a Perfmgmt MAD block of data.
459 * Returns error code or the number of bytes retrieved.
461 static int get_perf_mad(struct ib_device *dev, int port_num, __be16 attr,
462 void *data, int offset, size_t size)
464 struct ib_mad *in_mad;
465 struct ib_mad *out_mad;
466 size_t mad_size = sizeof(*out_mad);
467 u16 out_mad_pkey_index = 0;
470 if (!dev->ops.process_mad)
473 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
474 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
475 if (!in_mad || !out_mad) {
480 in_mad->mad_hdr.base_version = 1;
481 in_mad->mad_hdr.mgmt_class = IB_MGMT_CLASS_PERF_MGMT;
482 in_mad->mad_hdr.class_version = 1;
483 in_mad->mad_hdr.method = IB_MGMT_METHOD_GET;
484 in_mad->mad_hdr.attr_id = attr;
486 if (attr != IB_PMA_CLASS_PORT_INFO)
487 in_mad->data[41] = port_num; /* PortSelect field */
489 if ((dev->ops.process_mad(dev, IB_MAD_IGNORE_MKEY,
490 port_num, NULL, NULL,
491 (const struct ib_mad_hdr *)in_mad, mad_size,
492 (struct ib_mad_hdr *)out_mad, &mad_size,
493 &out_mad_pkey_index) &
494 (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) !=
495 (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) {
499 memcpy(data, out_mad->data + offset, size);
507 static ssize_t show_pma_counter(struct ib_port *p, struct port_attribute *attr,
510 struct port_table_attribute *tab_attr =
511 container_of(attr, struct port_table_attribute, attr);
512 int offset = tab_attr->index & 0xffff;
513 int width = (tab_attr->index >> 16) & 0xff;
517 ret = get_perf_mad(p->ibdev, p->port_num, tab_attr->attr_id, &data,
518 40 + offset / 8, sizeof(data));
524 ret = sprintf(buf, "%u\n", (*data >>
525 (4 - (offset % 8))) & 0xf);
528 ret = sprintf(buf, "%u\n", *data);
531 ret = sprintf(buf, "%u\n",
532 be16_to_cpup((__be16 *)data));
535 ret = sprintf(buf, "%u\n",
536 be32_to_cpup((__be32 *)data));
539 ret = sprintf(buf, "%llu\n",
540 be64_to_cpup((__be64 *)data));
550 static PORT_PMA_ATTR(symbol_error , 0, 16, 32);
551 static PORT_PMA_ATTR(link_error_recovery , 1, 8, 48);
552 static PORT_PMA_ATTR(link_downed , 2, 8, 56);
553 static PORT_PMA_ATTR(port_rcv_errors , 3, 16, 64);
554 static PORT_PMA_ATTR(port_rcv_remote_physical_errors, 4, 16, 80);
555 static PORT_PMA_ATTR(port_rcv_switch_relay_errors , 5, 16, 96);
556 static PORT_PMA_ATTR(port_xmit_discards , 6, 16, 112);
557 static PORT_PMA_ATTR(port_xmit_constraint_errors , 7, 8, 128);
558 static PORT_PMA_ATTR(port_rcv_constraint_errors , 8, 8, 136);
559 static PORT_PMA_ATTR(local_link_integrity_errors , 9, 4, 152);
560 static PORT_PMA_ATTR(excessive_buffer_overrun_errors, 10, 4, 156);
561 static PORT_PMA_ATTR(VL15_dropped , 11, 16, 176);
562 static PORT_PMA_ATTR(port_xmit_data , 12, 32, 192);
563 static PORT_PMA_ATTR(port_rcv_data , 13, 32, 224);
564 static PORT_PMA_ATTR(port_xmit_packets , 14, 32, 256);
565 static PORT_PMA_ATTR(port_rcv_packets , 15, 32, 288);
566 static PORT_PMA_ATTR(port_xmit_wait , 0, 32, 320);
569 * Counters added by extended set
571 static PORT_PMA_ATTR_EXT(port_xmit_data , 64, 64);
572 static PORT_PMA_ATTR_EXT(port_rcv_data , 64, 128);
573 static PORT_PMA_ATTR_EXT(port_xmit_packets , 64, 192);
574 static PORT_PMA_ATTR_EXT(port_rcv_packets , 64, 256);
575 static PORT_PMA_ATTR_EXT(unicast_xmit_packets , 64, 320);
576 static PORT_PMA_ATTR_EXT(unicast_rcv_packets , 64, 384);
577 static PORT_PMA_ATTR_EXT(multicast_xmit_packets , 64, 448);
578 static PORT_PMA_ATTR_EXT(multicast_rcv_packets , 64, 512);
580 static struct attribute *pma_attrs[] = {
581 &port_pma_attr_symbol_error.attr.attr,
582 &port_pma_attr_link_error_recovery.attr.attr,
583 &port_pma_attr_link_downed.attr.attr,
584 &port_pma_attr_port_rcv_errors.attr.attr,
585 &port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
586 &port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
587 &port_pma_attr_port_xmit_discards.attr.attr,
588 &port_pma_attr_port_xmit_constraint_errors.attr.attr,
589 &port_pma_attr_port_rcv_constraint_errors.attr.attr,
590 &port_pma_attr_local_link_integrity_errors.attr.attr,
591 &port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
592 &port_pma_attr_VL15_dropped.attr.attr,
593 &port_pma_attr_port_xmit_data.attr.attr,
594 &port_pma_attr_port_rcv_data.attr.attr,
595 &port_pma_attr_port_xmit_packets.attr.attr,
596 &port_pma_attr_port_rcv_packets.attr.attr,
597 &port_pma_attr_port_xmit_wait.attr.attr,
601 static struct attribute *pma_attrs_ext[] = {
602 &port_pma_attr_symbol_error.attr.attr,
603 &port_pma_attr_link_error_recovery.attr.attr,
604 &port_pma_attr_link_downed.attr.attr,
605 &port_pma_attr_port_rcv_errors.attr.attr,
606 &port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
607 &port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
608 &port_pma_attr_port_xmit_discards.attr.attr,
609 &port_pma_attr_port_xmit_constraint_errors.attr.attr,
610 &port_pma_attr_port_rcv_constraint_errors.attr.attr,
611 &port_pma_attr_local_link_integrity_errors.attr.attr,
612 &port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
613 &port_pma_attr_VL15_dropped.attr.attr,
614 &port_pma_attr_ext_port_xmit_data.attr.attr,
615 &port_pma_attr_ext_port_rcv_data.attr.attr,
616 &port_pma_attr_ext_port_xmit_packets.attr.attr,
617 &port_pma_attr_port_xmit_wait.attr.attr,
618 &port_pma_attr_ext_port_rcv_packets.attr.attr,
619 &port_pma_attr_ext_unicast_rcv_packets.attr.attr,
620 &port_pma_attr_ext_unicast_xmit_packets.attr.attr,
621 &port_pma_attr_ext_multicast_rcv_packets.attr.attr,
622 &port_pma_attr_ext_multicast_xmit_packets.attr.attr,
626 static struct attribute *pma_attrs_noietf[] = {
627 &port_pma_attr_symbol_error.attr.attr,
628 &port_pma_attr_link_error_recovery.attr.attr,
629 &port_pma_attr_link_downed.attr.attr,
630 &port_pma_attr_port_rcv_errors.attr.attr,
631 &port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
632 &port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
633 &port_pma_attr_port_xmit_discards.attr.attr,
634 &port_pma_attr_port_xmit_constraint_errors.attr.attr,
635 &port_pma_attr_port_rcv_constraint_errors.attr.attr,
636 &port_pma_attr_local_link_integrity_errors.attr.attr,
637 &port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
638 &port_pma_attr_VL15_dropped.attr.attr,
639 &port_pma_attr_ext_port_xmit_data.attr.attr,
640 &port_pma_attr_ext_port_rcv_data.attr.attr,
641 &port_pma_attr_ext_port_xmit_packets.attr.attr,
642 &port_pma_attr_ext_port_rcv_packets.attr.attr,
643 &port_pma_attr_port_xmit_wait.attr.attr,
647 static struct attribute_group pma_group = {
652 static struct attribute_group pma_group_ext = {
654 .attrs = pma_attrs_ext
657 static struct attribute_group pma_group_noietf = {
659 .attrs = pma_attrs_noietf
662 static void ib_port_release(struct kobject *kobj)
664 struct ib_port *p = container_of(kobj, struct ib_port, kobj);
668 if (p->gid_group.attrs) {
669 for (i = 0; (a = p->gid_group.attrs[i]); ++i)
672 kfree(p->gid_group.attrs);
675 if (p->pkey_group.attrs) {
676 for (i = 0; (a = p->pkey_group.attrs[i]); ++i)
679 kfree(p->pkey_group.attrs);
685 static void ib_port_gid_attr_release(struct kobject *kobj)
687 struct gid_attr_group *g = container_of(kobj, struct gid_attr_group,
693 for (i = 0; (a = g->ndev.attrs[i]); ++i)
696 kfree(g->ndev.attrs);
700 for (i = 0; (a = g->type.attrs[i]); ++i)
703 kfree(g->type.attrs);
709 static struct kobj_type port_type = {
710 .release = ib_port_release,
711 .sysfs_ops = &port_sysfs_ops,
712 .default_attrs = port_default_attrs
715 static struct kobj_type gid_attr_type = {
716 .sysfs_ops = &gid_attr_sysfs_ops,
717 .release = ib_port_gid_attr_release
720 static struct attribute **
721 alloc_group_attrs(ssize_t (*show)(struct ib_port *,
722 struct port_attribute *, char *buf),
725 struct attribute **tab_attr;
726 struct port_table_attribute *element;
729 tab_attr = kcalloc(1 + len, sizeof(struct attribute *), GFP_KERNEL);
733 for (i = 0; i < len; i++) {
734 element = kzalloc(sizeof(struct port_table_attribute),
739 if (snprintf(element->name, sizeof(element->name),
740 "%d", i) >= sizeof(element->name)) {
745 element->attr.attr.name = element->name;
746 element->attr.attr.mode = S_IRUGO;
747 element->attr.show = show;
749 sysfs_attr_init(&element->attr.attr);
751 tab_attr[i] = &element->attr.attr;
764 * Figure out which counter table to use depending on
765 * the device capabilities.
767 static struct attribute_group *get_counter_table(struct ib_device *dev,
770 struct ib_class_port_info cpi;
772 if (get_perf_mad(dev, port_num, IB_PMA_CLASS_PORT_INFO,
773 &cpi, 40, sizeof(cpi)) >= 0) {
774 if (cpi.capability_mask & IB_PMA_CLASS_CAP_EXT_WIDTH)
775 /* We have extended counters */
776 return &pma_group_ext;
778 if (cpi.capability_mask & IB_PMA_CLASS_CAP_EXT_WIDTH_NOIETF)
779 /* But not the IETF ones */
780 return &pma_group_noietf;
783 /* Fall back to normal counters */
787 static int update_hw_stats(struct ib_device *dev, struct rdma_hw_stats *stats,
788 u8 port_num, int index)
792 if (time_is_after_eq_jiffies(stats->timestamp + stats->lifespan))
794 ret = dev->ops.get_hw_stats(dev, stats, port_num, index);
797 if (ret == stats->num_counters)
798 stats->timestamp = jiffies;
803 static ssize_t print_hw_stat(struct rdma_hw_stats *stats, int index, char *buf)
805 return sprintf(buf, "%llu\n", stats->value[index]);
808 static ssize_t show_hw_stats(struct kobject *kobj, struct attribute *attr,
811 struct ib_device *dev;
812 struct ib_port *port;
813 struct hw_stats_attribute *hsa;
814 struct rdma_hw_stats *stats;
817 hsa = container_of(attr, struct hw_stats_attribute, attr);
818 if (!hsa->port_num) {
819 dev = container_of((struct device *)kobj,
820 struct ib_device, dev);
821 stats = dev->hw_stats;
823 port = container_of(kobj, struct ib_port, kobj);
825 stats = port->hw_stats;
827 mutex_lock(&stats->lock);
828 ret = update_hw_stats(dev, stats, hsa->port_num, hsa->index);
831 ret = print_hw_stat(stats, hsa->index, buf);
833 mutex_unlock(&stats->lock);
838 static ssize_t show_stats_lifespan(struct kobject *kobj,
839 struct attribute *attr,
842 struct hw_stats_attribute *hsa;
843 struct rdma_hw_stats *stats;
846 hsa = container_of(attr, struct hw_stats_attribute, attr);
847 if (!hsa->port_num) {
848 struct ib_device *dev = container_of((struct device *)kobj,
849 struct ib_device, dev);
851 stats = dev->hw_stats;
853 struct ib_port *p = container_of(kobj, struct ib_port, kobj);
858 mutex_lock(&stats->lock);
859 msecs = jiffies_to_msecs(stats->lifespan);
860 mutex_unlock(&stats->lock);
862 return sprintf(buf, "%d\n", msecs);
865 static ssize_t set_stats_lifespan(struct kobject *kobj,
866 struct attribute *attr,
867 const char *buf, size_t count)
869 struct hw_stats_attribute *hsa;
870 struct rdma_hw_stats *stats;
875 ret = kstrtoint(buf, 10, &msecs);
878 if (msecs < 0 || msecs > 10000)
880 jiffies = msecs_to_jiffies(msecs);
881 hsa = container_of(attr, struct hw_stats_attribute, attr);
882 if (!hsa->port_num) {
883 struct ib_device *dev = container_of((struct device *)kobj,
884 struct ib_device, dev);
886 stats = dev->hw_stats;
888 struct ib_port *p = container_of(kobj, struct ib_port, kobj);
893 mutex_lock(&stats->lock);
894 stats->lifespan = jiffies;
895 mutex_unlock(&stats->lock);
900 static void free_hsag(struct kobject *kobj, struct attribute_group *attr_group)
902 struct attribute **attr;
904 sysfs_remove_group(kobj, attr_group);
906 for (attr = attr_group->attrs; *attr; attr++)
911 static struct attribute *alloc_hsa(int index, u8 port_num, const char *name)
913 struct hw_stats_attribute *hsa;
915 hsa = kmalloc(sizeof(*hsa), GFP_KERNEL);
919 hsa->attr.name = (char *)name;
920 hsa->attr.mode = S_IRUGO;
921 hsa->show = show_hw_stats;
924 hsa->port_num = port_num;
929 static struct attribute *alloc_hsa_lifespan(char *name, u8 port_num)
931 struct hw_stats_attribute *hsa;
933 hsa = kmalloc(sizeof(*hsa), GFP_KERNEL);
937 hsa->attr.name = name;
938 hsa->attr.mode = S_IWUSR | S_IRUGO;
939 hsa->show = show_stats_lifespan;
940 hsa->store = set_stats_lifespan;
942 hsa->port_num = port_num;
947 static void setup_hw_stats(struct ib_device *device, struct ib_port *port,
950 struct attribute_group *hsag;
951 struct rdma_hw_stats *stats;
954 stats = device->ops.alloc_hw_stats(device, port_num);
959 if (!stats->names || stats->num_counters <= 0)
963 * Two extra attribue elements here, one for the lifespan entry and
964 * one to NULL terminate the list for the sysfs core code
966 hsag = kzalloc(sizeof(*hsag) +
967 sizeof(void *) * (stats->num_counters + 2),
972 ret = device->ops.get_hw_stats(device, stats, port_num,
973 stats->num_counters);
974 if (ret != stats->num_counters)
977 stats->timestamp = jiffies;
979 hsag->name = "hw_counters";
980 hsag->attrs = (void *)hsag + sizeof(*hsag);
982 for (i = 0; i < stats->num_counters; i++) {
983 hsag->attrs[i] = alloc_hsa(i, port_num, stats->names[i]);
986 sysfs_attr_init(hsag->attrs[i]);
989 mutex_init(&stats->lock);
990 /* treat an error here as non-fatal */
991 hsag->attrs[i] = alloc_hsa_lifespan("lifespan", port_num);
993 sysfs_attr_init(hsag->attrs[i]);
996 struct kobject *kobj = &port->kobj;
997 ret = sysfs_create_group(kobj, hsag);
1000 port->hw_stats_ag = hsag;
1001 port->hw_stats = stats;
1003 struct kobject *kobj = &device->dev.kobj;
1004 ret = sysfs_create_group(kobj, hsag);
1007 device->hw_stats_ag = hsag;
1008 device->hw_stats = stats;
1015 kfree(hsag->attrs[i]);
1023 static int add_port(struct ib_core_device *coredev, int port_num)
1025 struct ib_device *device = rdma_device_to_ibdev(&coredev->dev);
1026 bool is_full_dev = &device->coredev == coredev;
1028 struct ib_port_attr attr;
1032 ret = ib_query_port(device, port_num, &attr);
1036 p = kzalloc(sizeof *p, GFP_KERNEL);
1041 p->port_num = port_num;
1043 ret = kobject_init_and_add(&p->kobj, &port_type,
1044 coredev->ports_kobj,
1051 p->gid_attr_group = kzalloc(sizeof(*p->gid_attr_group), GFP_KERNEL);
1052 if (!p->gid_attr_group) {
1057 p->gid_attr_group->port = p;
1058 ret = kobject_init_and_add(&p->gid_attr_group->kobj, &gid_attr_type,
1059 &p->kobj, "gid_attrs");
1061 kfree(p->gid_attr_group);
1065 if (device->ops.process_mad && is_full_dev) {
1066 p->pma_table = get_counter_table(device, port_num);
1067 ret = sysfs_create_group(&p->kobj, p->pma_table);
1069 goto err_put_gid_attrs;
1072 p->gid_group.name = "gids";
1073 p->gid_group.attrs = alloc_group_attrs(show_port_gid, attr.gid_tbl_len);
1074 if (!p->gid_group.attrs) {
1076 goto err_remove_pma;
1079 ret = sysfs_create_group(&p->kobj, &p->gid_group);
1083 p->gid_attr_group->ndev.name = "ndevs";
1084 p->gid_attr_group->ndev.attrs = alloc_group_attrs(show_port_gid_attr_ndev,
1086 if (!p->gid_attr_group->ndev.attrs) {
1088 goto err_remove_gid;
1091 ret = sysfs_create_group(&p->gid_attr_group->kobj,
1092 &p->gid_attr_group->ndev);
1094 goto err_free_gid_ndev;
1096 p->gid_attr_group->type.name = "types";
1097 p->gid_attr_group->type.attrs = alloc_group_attrs(show_port_gid_attr_gid_type,
1099 if (!p->gid_attr_group->type.attrs) {
1101 goto err_remove_gid_ndev;
1104 ret = sysfs_create_group(&p->gid_attr_group->kobj,
1105 &p->gid_attr_group->type);
1107 goto err_free_gid_type;
1109 p->pkey_group.name = "pkeys";
1110 p->pkey_group.attrs = alloc_group_attrs(show_port_pkey,
1112 if (!p->pkey_group.attrs) {
1114 goto err_remove_gid_type;
1117 ret = sysfs_create_group(&p->kobj, &p->pkey_group);
1121 if (device->ops.init_port && is_full_dev) {
1122 ret = device->ops.init_port(device, port_num, &p->kobj);
1124 goto err_remove_pkey;
1128 * If port == 0, it means hw_counters are per device and not per
1129 * port, so holder should be device. Therefore skip per port conunter
1132 if (device->ops.alloc_hw_stats && port_num && is_full_dev)
1133 setup_hw_stats(device, p, port_num);
1135 list_add_tail(&p->kobj.entry, &coredev->port_list);
1137 kobject_uevent(&p->kobj, KOBJ_ADD);
1141 sysfs_remove_group(&p->kobj, &p->pkey_group);
1144 for (i = 0; i < attr.pkey_tbl_len; ++i)
1145 kfree(p->pkey_group.attrs[i]);
1147 kfree(p->pkey_group.attrs);
1148 p->pkey_group.attrs = NULL;
1150 err_remove_gid_type:
1151 sysfs_remove_group(&p->gid_attr_group->kobj,
1152 &p->gid_attr_group->type);
1155 for (i = 0; i < attr.gid_tbl_len; ++i)
1156 kfree(p->gid_attr_group->type.attrs[i]);
1158 kfree(p->gid_attr_group->type.attrs);
1159 p->gid_attr_group->type.attrs = NULL;
1161 err_remove_gid_ndev:
1162 sysfs_remove_group(&p->gid_attr_group->kobj,
1163 &p->gid_attr_group->ndev);
1166 for (i = 0; i < attr.gid_tbl_len; ++i)
1167 kfree(p->gid_attr_group->ndev.attrs[i]);
1169 kfree(p->gid_attr_group->ndev.attrs);
1170 p->gid_attr_group->ndev.attrs = NULL;
1173 sysfs_remove_group(&p->kobj, &p->gid_group);
1176 for (i = 0; i < attr.gid_tbl_len; ++i)
1177 kfree(p->gid_group.attrs[i]);
1179 kfree(p->gid_group.attrs);
1180 p->gid_group.attrs = NULL;
1184 sysfs_remove_group(&p->kobj, p->pma_table);
1187 kobject_put(&p->gid_attr_group->kobj);
1190 kobject_put(&p->kobj);
1194 static ssize_t node_type_show(struct device *device,
1195 struct device_attribute *attr, char *buf)
1197 struct ib_device *dev = rdma_device_to_ibdev(device);
1199 switch (dev->node_type) {
1200 case RDMA_NODE_IB_CA: return sprintf(buf, "%d: CA\n", dev->node_type);
1201 case RDMA_NODE_RNIC: return sprintf(buf, "%d: RNIC\n", dev->node_type);
1202 case RDMA_NODE_USNIC: return sprintf(buf, "%d: usNIC\n", dev->node_type);
1203 case RDMA_NODE_USNIC_UDP: return sprintf(buf, "%d: usNIC UDP\n", dev->node_type);
1204 case RDMA_NODE_UNSPECIFIED: return sprintf(buf, "%d: unspecified\n", dev->node_type);
1205 case RDMA_NODE_IB_SWITCH: return sprintf(buf, "%d: switch\n", dev->node_type);
1206 case RDMA_NODE_IB_ROUTER: return sprintf(buf, "%d: router\n", dev->node_type);
1207 default: return sprintf(buf, "%d: <unknown>\n", dev->node_type);
1210 static DEVICE_ATTR_RO(node_type);
1212 static ssize_t sys_image_guid_show(struct device *device,
1213 struct device_attribute *dev_attr, char *buf)
1215 struct ib_device *dev = rdma_device_to_ibdev(device);
1217 return sprintf(buf, "%04x:%04x:%04x:%04x\n",
1218 be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[0]),
1219 be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[1]),
1220 be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[2]),
1221 be16_to_cpu(((__be16 *) &dev->attrs.sys_image_guid)[3]));
1223 static DEVICE_ATTR_RO(sys_image_guid);
1225 static ssize_t node_guid_show(struct device *device,
1226 struct device_attribute *attr, char *buf)
1228 struct ib_device *dev = rdma_device_to_ibdev(device);
1230 return sprintf(buf, "%04x:%04x:%04x:%04x\n",
1231 be16_to_cpu(((__be16 *) &dev->node_guid)[0]),
1232 be16_to_cpu(((__be16 *) &dev->node_guid)[1]),
1233 be16_to_cpu(((__be16 *) &dev->node_guid)[2]),
1234 be16_to_cpu(((__be16 *) &dev->node_guid)[3]));
1236 static DEVICE_ATTR_RO(node_guid);
1238 static ssize_t node_desc_show(struct device *device,
1239 struct device_attribute *attr, char *buf)
1241 struct ib_device *dev = rdma_device_to_ibdev(device);
1243 return sprintf(buf, "%.64s\n", dev->node_desc);
1246 static ssize_t node_desc_store(struct device *device,
1247 struct device_attribute *attr,
1248 const char *buf, size_t count)
1250 struct ib_device *dev = rdma_device_to_ibdev(device);
1251 struct ib_device_modify desc = {};
1254 if (!dev->ops.modify_device)
1257 memcpy(desc.node_desc, buf, min_t(int, count, IB_DEVICE_NODE_DESC_MAX));
1258 ret = ib_modify_device(dev, IB_DEVICE_MODIFY_NODE_DESC, &desc);
1264 static DEVICE_ATTR_RW(node_desc);
1266 static ssize_t fw_ver_show(struct device *device, struct device_attribute *attr,
1269 struct ib_device *dev = rdma_device_to_ibdev(device);
1271 ib_get_device_fw_str(dev, buf);
1272 strlcat(buf, "\n", IB_FW_VERSION_NAME_MAX);
1275 static DEVICE_ATTR_RO(fw_ver);
1277 static struct attribute *ib_dev_attrs[] = {
1278 &dev_attr_node_type.attr,
1279 &dev_attr_node_guid.attr,
1280 &dev_attr_sys_image_guid.attr,
1281 &dev_attr_fw_ver.attr,
1282 &dev_attr_node_desc.attr,
1286 const struct attribute_group ib_dev_attr_group = {
1287 .attrs = ib_dev_attrs,
1290 void ib_free_port_attrs(struct ib_core_device *coredev)
1292 struct kobject *p, *t;
1294 list_for_each_entry_safe(p, t, &coredev->port_list, entry) {
1295 struct ib_port *port = container_of(p, struct ib_port, kobj);
1297 list_del(&p->entry);
1298 if (port->hw_stats_ag)
1299 free_hsag(&port->kobj, port->hw_stats_ag);
1300 kfree(port->hw_stats);
1302 if (port->pma_table)
1303 sysfs_remove_group(p, port->pma_table);
1304 sysfs_remove_group(p, &port->pkey_group);
1305 sysfs_remove_group(p, &port->gid_group);
1306 sysfs_remove_group(&port->gid_attr_group->kobj,
1307 &port->gid_attr_group->ndev);
1308 sysfs_remove_group(&port->gid_attr_group->kobj,
1309 &port->gid_attr_group->type);
1310 kobject_put(&port->gid_attr_group->kobj);
1314 kobject_put(coredev->ports_kobj);
1317 int ib_setup_port_attrs(struct ib_core_device *coredev)
1319 struct ib_device *device = rdma_device_to_ibdev(&coredev->dev);
1323 coredev->ports_kobj = kobject_create_and_add("ports",
1324 &coredev->dev.kobj);
1325 if (!coredev->ports_kobj)
1328 rdma_for_each_port (device, port) {
1329 ret = add_port(coredev, port);
1337 ib_free_port_attrs(coredev);
1341 int ib_device_register_sysfs(struct ib_device *device)
1345 ret = ib_setup_port_attrs(&device->coredev);
1349 if (device->ops.alloc_hw_stats)
1350 setup_hw_stats(device, NULL, 0);
1355 void ib_device_unregister_sysfs(struct ib_device *device)
1357 if (device->hw_stats_ag)
1358 free_hsag(&device->dev.kobj, device->hw_stats_ag);
1359 kfree(device->hw_stats);
1361 ib_free_port_attrs(&device->coredev);
1365 * ib_port_register_module_stat - add module counters under relevant port
1368 * @device: IB device to add counters
1369 * @port_num: valid port number
1370 * @kobj: pointer to the kobject to initialize
1371 * @ktype: pointer to the ktype for this kobject.
1372 * @name: the name of the kobject
1374 int ib_port_register_module_stat(struct ib_device *device, u8 port_num,
1375 struct kobject *kobj, struct kobj_type *ktype,
1378 struct kobject *p, *t;
1381 list_for_each_entry_safe(p, t, &device->coredev.port_list, entry) {
1382 struct ib_port *port = container_of(p, struct ib_port, kobj);
1384 if (port->port_num != port_num)
1387 ret = kobject_init_and_add(kobj, ktype, &port->kobj, "%s",
1395 EXPORT_SYMBOL(ib_port_register_module_stat);
1398 * ib_port_unregister_module_stat - release module counters
1399 * @kobj: pointer to the kobject to release
1401 void ib_port_unregister_module_stat(struct kobject *kobj)
1405 EXPORT_SYMBOL(ib_port_unregister_module_stat);