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>
47 #include <rdma/ib_sysfs.h>
49 struct port_table_attribute {
50 struct ib_port_attribute attr;
56 struct gid_attr_group {
59 struct attribute_group groups[2];
60 const struct attribute_group *groups_list[3];
61 struct port_table_attribute attrs_list[];
66 struct ib_device *ibdev;
67 struct gid_attr_group *gid_attr_group;
68 struct hw_stats_port_data *hw_stats_data;
70 struct attribute_group groups[3];
71 const struct attribute_group *groups_list[5];
73 struct port_table_attribute attrs_list[];
76 struct hw_stats_device_attribute {
77 struct device_attribute attr;
78 ssize_t (*show)(struct ib_device *ibdev, struct rdma_hw_stats *stats,
79 unsigned int index, unsigned int port_num, char *buf);
80 ssize_t (*store)(struct ib_device *ibdev, struct rdma_hw_stats *stats,
81 unsigned int index, unsigned int port_num,
82 const char *buf, size_t count);
85 struct hw_stats_port_attribute {
86 struct ib_port_attribute attr;
87 ssize_t (*show)(struct ib_device *ibdev, struct rdma_hw_stats *stats,
88 unsigned int index, unsigned int port_num, char *buf);
89 ssize_t (*store)(struct ib_device *ibdev, struct rdma_hw_stats *stats,
90 unsigned int index, unsigned int port_num,
91 const char *buf, size_t count);
94 struct hw_stats_device_data {
95 struct attribute_group group;
96 struct rdma_hw_stats *stats;
97 struct hw_stats_device_attribute attrs[];
100 struct hw_stats_port_data {
101 struct rdma_hw_stats *stats;
102 struct hw_stats_port_attribute attrs[];
105 static ssize_t port_attr_show(struct kobject *kobj,
106 struct attribute *attr, char *buf)
108 struct ib_port_attribute *port_attr =
109 container_of(attr, struct ib_port_attribute, attr);
110 struct ib_port *p = container_of(kobj, struct ib_port, kobj);
112 if (!port_attr->show)
115 return port_attr->show(p->ibdev, p->port_num, port_attr, buf);
118 static ssize_t port_attr_store(struct kobject *kobj,
119 struct attribute *attr,
120 const char *buf, size_t count)
122 struct ib_port_attribute *port_attr =
123 container_of(attr, struct ib_port_attribute, attr);
124 struct ib_port *p = container_of(kobj, struct ib_port, kobj);
126 if (!port_attr->store)
128 return port_attr->store(p->ibdev, p->port_num, port_attr, buf, count);
131 int ib_port_sysfs_create_groups(struct ib_device *ibdev, u32 port_num,
132 const struct attribute_group **groups)
134 return sysfs_create_groups(&ibdev->port_data[port_num].sysfs->kobj,
137 EXPORT_SYMBOL_GPL(ib_port_sysfs_create_groups);
139 void ib_port_sysfs_remove_groups(struct ib_device *ibdev, u32 port_num,
140 const struct attribute_group **groups)
142 return sysfs_remove_groups(&ibdev->port_data[port_num].sysfs->kobj,
145 EXPORT_SYMBOL_GPL(ib_port_sysfs_remove_groups);
147 struct ib_device *ib_port_sysfs_get_ibdev_kobj(struct kobject *kobj,
150 struct ib_port *port = container_of(kobj, struct ib_port, kobj);
152 *port_num = port->port_num;
155 EXPORT_SYMBOL(ib_port_sysfs_get_ibdev_kobj);
157 static const struct sysfs_ops port_sysfs_ops = {
158 .show = port_attr_show,
159 .store = port_attr_store
162 static ssize_t hw_stat_device_show(struct device *dev,
163 struct device_attribute *attr, char *buf)
165 struct hw_stats_device_attribute *stat_attr =
166 container_of(attr, struct hw_stats_device_attribute, attr);
167 struct ib_device *ibdev = container_of(dev, struct ib_device, dev);
169 return stat_attr->show(ibdev, ibdev->hw_stats_data->stats,
170 stat_attr - ibdev->hw_stats_data->attrs, 0, buf);
173 static ssize_t hw_stat_device_store(struct device *dev,
174 struct device_attribute *attr,
175 const char *buf, size_t count)
177 struct hw_stats_device_attribute *stat_attr =
178 container_of(attr, struct hw_stats_device_attribute, attr);
179 struct ib_device *ibdev = container_of(dev, struct ib_device, dev);
181 return stat_attr->store(ibdev, ibdev->hw_stats_data->stats,
182 stat_attr - ibdev->hw_stats_data->attrs, 0, buf,
186 static ssize_t hw_stat_port_show(struct ib_device *ibdev, u32 port_num,
187 struct ib_port_attribute *attr, char *buf)
189 struct hw_stats_port_attribute *stat_attr =
190 container_of(attr, struct hw_stats_port_attribute, attr);
191 struct ib_port *port = ibdev->port_data[port_num].sysfs;
193 return stat_attr->show(ibdev, port->hw_stats_data->stats,
194 stat_attr - port->hw_stats_data->attrs,
195 port->port_num, buf);
198 static ssize_t hw_stat_port_store(struct ib_device *ibdev, u32 port_num,
199 struct ib_port_attribute *attr,
200 const char *buf, size_t count)
202 struct hw_stats_port_attribute *stat_attr =
203 container_of(attr, struct hw_stats_port_attribute, attr);
204 struct ib_port *port = ibdev->port_data[port_num].sysfs;
206 return stat_attr->store(ibdev, port->hw_stats_data->stats,
207 stat_attr - port->hw_stats_data->attrs,
208 port->port_num, buf, count);
211 static ssize_t gid_attr_show(struct kobject *kobj,
212 struct attribute *attr, char *buf)
214 struct ib_port_attribute *port_attr =
215 container_of(attr, struct ib_port_attribute, attr);
216 struct ib_port *p = container_of(kobj, struct gid_attr_group,
219 if (!port_attr->show)
222 return port_attr->show(p->ibdev, p->port_num, port_attr, buf);
225 static const struct sysfs_ops gid_attr_sysfs_ops = {
226 .show = gid_attr_show
229 static ssize_t state_show(struct ib_device *ibdev, u32 port_num,
230 struct ib_port_attribute *unused, char *buf)
232 struct ib_port_attr attr;
235 static const char *state_name[] = {
236 [IB_PORT_NOP] = "NOP",
237 [IB_PORT_DOWN] = "DOWN",
238 [IB_PORT_INIT] = "INIT",
239 [IB_PORT_ARMED] = "ARMED",
240 [IB_PORT_ACTIVE] = "ACTIVE",
241 [IB_PORT_ACTIVE_DEFER] = "ACTIVE_DEFER"
244 ret = ib_query_port(ibdev, port_num, &attr);
248 return sysfs_emit(buf, "%d: %s\n", attr.state,
250 attr.state < ARRAY_SIZE(state_name) ?
251 state_name[attr.state] :
255 static ssize_t lid_show(struct ib_device *ibdev, u32 port_num,
256 struct ib_port_attribute *unused, char *buf)
258 struct ib_port_attr attr;
261 ret = ib_query_port(ibdev, port_num, &attr);
265 return sysfs_emit(buf, "0x%x\n", attr.lid);
268 static ssize_t lid_mask_count_show(struct ib_device *ibdev, u32 port_num,
269 struct ib_port_attribute *unused, char *buf)
271 struct ib_port_attr attr;
274 ret = ib_query_port(ibdev, port_num, &attr);
278 return sysfs_emit(buf, "%d\n", attr.lmc);
281 static ssize_t sm_lid_show(struct ib_device *ibdev, u32 port_num,
282 struct ib_port_attribute *unused, char *buf)
284 struct ib_port_attr attr;
287 ret = ib_query_port(ibdev, port_num, &attr);
291 return sysfs_emit(buf, "0x%x\n", attr.sm_lid);
294 static ssize_t sm_sl_show(struct ib_device *ibdev, u32 port_num,
295 struct ib_port_attribute *unused, char *buf)
297 struct ib_port_attr attr;
300 ret = ib_query_port(ibdev, port_num, &attr);
304 return sysfs_emit(buf, "%d\n", attr.sm_sl);
307 static ssize_t cap_mask_show(struct ib_device *ibdev, u32 port_num,
308 struct ib_port_attribute *unused, char *buf)
310 struct ib_port_attr attr;
313 ret = ib_query_port(ibdev, port_num, &attr);
317 return sysfs_emit(buf, "0x%08x\n", attr.port_cap_flags);
320 static ssize_t rate_show(struct ib_device *ibdev, u32 port_num,
321 struct ib_port_attribute *unused, char *buf)
323 struct ib_port_attr attr;
325 int rate; /* in deci-Gb/sec */
328 ret = ib_query_port(ibdev, port_num, &attr);
332 switch (attr.active_speed) {
362 default: /* default to SDR for invalid rates */
368 rate *= ib_width_enum_to_int(attr.active_width);
372 return sysfs_emit(buf, "%d%s Gb/sec (%dX%s)\n", rate / 10,
373 rate % 10 ? ".5" : "",
374 ib_width_enum_to_int(attr.active_width), speed);
377 static const char *phys_state_to_str(enum ib_port_phys_state phys_state)
379 static const char *phys_state_str[] = {
384 "PortConfigurationTraining",
390 if (phys_state < ARRAY_SIZE(phys_state_str))
391 return phys_state_str[phys_state];
395 static ssize_t phys_state_show(struct ib_device *ibdev, u32 port_num,
396 struct ib_port_attribute *unused, char *buf)
398 struct ib_port_attr attr;
402 ret = ib_query_port(ibdev, port_num, &attr);
406 return sysfs_emit(buf, "%d: %s\n", attr.phys_state,
407 phys_state_to_str(attr.phys_state));
410 static ssize_t link_layer_show(struct ib_device *ibdev, u32 port_num,
411 struct ib_port_attribute *unused, char *buf)
415 switch (rdma_port_get_link_layer(ibdev, port_num)) {
416 case IB_LINK_LAYER_INFINIBAND:
417 output = "InfiniBand";
419 case IB_LINK_LAYER_ETHERNET:
427 return sysfs_emit(buf, "%s\n", output);
430 static IB_PORT_ATTR_RO(state);
431 static IB_PORT_ATTR_RO(lid);
432 static IB_PORT_ATTR_RO(lid_mask_count);
433 static IB_PORT_ATTR_RO(sm_lid);
434 static IB_PORT_ATTR_RO(sm_sl);
435 static IB_PORT_ATTR_RO(cap_mask);
436 static IB_PORT_ATTR_RO(rate);
437 static IB_PORT_ATTR_RO(phys_state);
438 static IB_PORT_ATTR_RO(link_layer);
440 static struct attribute *port_default_attrs[] = {
441 &ib_port_attr_state.attr,
442 &ib_port_attr_lid.attr,
443 &ib_port_attr_lid_mask_count.attr,
444 &ib_port_attr_sm_lid.attr,
445 &ib_port_attr_sm_sl.attr,
446 &ib_port_attr_cap_mask.attr,
447 &ib_port_attr_rate.attr,
448 &ib_port_attr_phys_state.attr,
449 &ib_port_attr_link_layer.attr,
453 static ssize_t print_ndev(const struct ib_gid_attr *gid_attr, char *buf)
455 struct net_device *ndev;
459 ndev = rcu_dereference(gid_attr->ndev);
461 ret = sysfs_emit(buf, "%s\n", ndev->name);
466 static ssize_t print_gid_type(const struct ib_gid_attr *gid_attr, char *buf)
468 return sysfs_emit(buf, "%s\n",
469 ib_cache_gid_type_str(gid_attr->gid_type));
472 static ssize_t _show_port_gid_attr(
473 struct ib_device *ibdev, u32 port_num, struct ib_port_attribute *attr,
475 ssize_t (*print)(const struct ib_gid_attr *gid_attr, char *buf))
477 struct port_table_attribute *tab_attr =
478 container_of(attr, struct port_table_attribute, attr);
479 const struct ib_gid_attr *gid_attr;
482 gid_attr = rdma_get_gid_attr(ibdev, port_num, tab_attr->index);
483 if (IS_ERR(gid_attr))
484 /* -EINVAL is returned for user space compatibility reasons. */
487 ret = print(gid_attr, buf);
488 rdma_put_gid_attr(gid_attr);
492 static ssize_t show_port_gid(struct ib_device *ibdev, u32 port_num,
493 struct ib_port_attribute *attr, char *buf)
495 struct port_table_attribute *tab_attr =
496 container_of(attr, struct port_table_attribute, attr);
497 const struct ib_gid_attr *gid_attr;
500 gid_attr = rdma_get_gid_attr(ibdev, port_num, tab_attr->index);
501 if (IS_ERR(gid_attr)) {
502 const union ib_gid zgid = {};
504 /* If reading GID fails, it is likely due to GID entry being
505 * empty (invalid) or reserved GID in the table. User space
506 * expects to read GID table entries as long as it given index
507 * is within GID table size. Administrative/debugging tool
508 * fails to query rest of the GID entries if it hits error
509 * while querying a GID of the given index. To avoid user
510 * space throwing such error on fail to read gid, return zero
511 * GID as before. This maintains backward compatibility.
513 return sysfs_emit(buf, "%pI6\n", zgid.raw);
516 len = sysfs_emit(buf, "%pI6\n", gid_attr->gid.raw);
517 rdma_put_gid_attr(gid_attr);
521 static ssize_t show_port_gid_attr_ndev(struct ib_device *ibdev, u32 port_num,
522 struct ib_port_attribute *attr,
525 return _show_port_gid_attr(ibdev, port_num, attr, buf, print_ndev);
528 static ssize_t show_port_gid_attr_gid_type(struct ib_device *ibdev,
530 struct ib_port_attribute *attr,
533 return _show_port_gid_attr(ibdev, port_num, attr, buf, print_gid_type);
536 static ssize_t show_port_pkey(struct ib_device *ibdev, u32 port_num,
537 struct ib_port_attribute *attr, char *buf)
539 struct port_table_attribute *tab_attr =
540 container_of(attr, struct port_table_attribute, attr);
544 ret = ib_query_pkey(ibdev, port_num, tab_attr->index, &pkey);
548 return sysfs_emit(buf, "0x%04x\n", pkey);
551 #define PORT_PMA_ATTR(_name, _counter, _width, _offset) \
552 struct port_table_attribute port_pma_attr_##_name = { \
553 .attr = __ATTR(_name, S_IRUGO, show_pma_counter, NULL), \
554 .index = (_offset) | ((_width) << 16) | ((_counter) << 24), \
555 .attr_id = IB_PMA_PORT_COUNTERS, \
558 #define PORT_PMA_ATTR_EXT(_name, _width, _offset) \
559 struct port_table_attribute port_pma_attr_ext_##_name = { \
560 .attr = __ATTR(_name, S_IRUGO, show_pma_counter, NULL), \
561 .index = (_offset) | ((_width) << 16), \
562 .attr_id = IB_PMA_PORT_COUNTERS_EXT, \
566 * Get a Perfmgmt MAD block of data.
567 * Returns error code or the number of bytes retrieved.
569 static int get_perf_mad(struct ib_device *dev, int port_num, __be16 attr,
570 void *data, int offset, size_t size)
572 struct ib_mad *in_mad;
573 struct ib_mad *out_mad;
574 size_t mad_size = sizeof(*out_mad);
575 u16 out_mad_pkey_index = 0;
578 if (!dev->ops.process_mad)
581 in_mad = kzalloc(sizeof(*in_mad), GFP_KERNEL);
582 out_mad = kzalloc(sizeof(*out_mad), GFP_KERNEL);
583 if (!in_mad || !out_mad) {
588 in_mad->mad_hdr.base_version = 1;
589 in_mad->mad_hdr.mgmt_class = IB_MGMT_CLASS_PERF_MGMT;
590 in_mad->mad_hdr.class_version = 1;
591 in_mad->mad_hdr.method = IB_MGMT_METHOD_GET;
592 in_mad->mad_hdr.attr_id = attr;
594 if (attr != IB_PMA_CLASS_PORT_INFO)
595 in_mad->data[41] = port_num; /* PortSelect field */
597 if ((dev->ops.process_mad(dev, IB_MAD_IGNORE_MKEY, port_num, NULL, NULL,
598 in_mad, out_mad, &mad_size,
599 &out_mad_pkey_index) &
600 (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) !=
601 (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) {
605 memcpy(data, out_mad->data + offset, size);
613 static ssize_t show_pma_counter(struct ib_device *ibdev, u32 port_num,
614 struct ib_port_attribute *attr, char *buf)
616 struct port_table_attribute *tab_attr =
617 container_of(attr, struct port_table_attribute, attr);
618 int offset = tab_attr->index & 0xffff;
619 int width = (tab_attr->index >> 16) & 0xff;
624 ret = get_perf_mad(ibdev, port_num, tab_attr->attr_id, &data,
625 40 + offset / 8, sizeof(data));
631 len = sysfs_emit(buf, "%u\n",
632 (*data >> (4 - (offset % 8))) & 0xf);
635 len = sysfs_emit(buf, "%u\n", *data);
638 len = sysfs_emit(buf, "%u\n", be16_to_cpup((__be16 *)data));
641 len = sysfs_emit(buf, "%u\n", be32_to_cpup((__be32 *)data));
644 len = sysfs_emit(buf, "%llu\n", be64_to_cpup((__be64 *)data));
654 static PORT_PMA_ATTR(symbol_error , 0, 16, 32);
655 static PORT_PMA_ATTR(link_error_recovery , 1, 8, 48);
656 static PORT_PMA_ATTR(link_downed , 2, 8, 56);
657 static PORT_PMA_ATTR(port_rcv_errors , 3, 16, 64);
658 static PORT_PMA_ATTR(port_rcv_remote_physical_errors, 4, 16, 80);
659 static PORT_PMA_ATTR(port_rcv_switch_relay_errors , 5, 16, 96);
660 static PORT_PMA_ATTR(port_xmit_discards , 6, 16, 112);
661 static PORT_PMA_ATTR(port_xmit_constraint_errors , 7, 8, 128);
662 static PORT_PMA_ATTR(port_rcv_constraint_errors , 8, 8, 136);
663 static PORT_PMA_ATTR(local_link_integrity_errors , 9, 4, 152);
664 static PORT_PMA_ATTR(excessive_buffer_overrun_errors, 10, 4, 156);
665 static PORT_PMA_ATTR(VL15_dropped , 11, 16, 176);
666 static PORT_PMA_ATTR(port_xmit_data , 12, 32, 192);
667 static PORT_PMA_ATTR(port_rcv_data , 13, 32, 224);
668 static PORT_PMA_ATTR(port_xmit_packets , 14, 32, 256);
669 static PORT_PMA_ATTR(port_rcv_packets , 15, 32, 288);
670 static PORT_PMA_ATTR(port_xmit_wait , 0, 32, 320);
673 * Counters added by extended set
675 static PORT_PMA_ATTR_EXT(port_xmit_data , 64, 64);
676 static PORT_PMA_ATTR_EXT(port_rcv_data , 64, 128);
677 static PORT_PMA_ATTR_EXT(port_xmit_packets , 64, 192);
678 static PORT_PMA_ATTR_EXT(port_rcv_packets , 64, 256);
679 static PORT_PMA_ATTR_EXT(unicast_xmit_packets , 64, 320);
680 static PORT_PMA_ATTR_EXT(unicast_rcv_packets , 64, 384);
681 static PORT_PMA_ATTR_EXT(multicast_xmit_packets , 64, 448);
682 static PORT_PMA_ATTR_EXT(multicast_rcv_packets , 64, 512);
684 static struct attribute *pma_attrs[] = {
685 &port_pma_attr_symbol_error.attr.attr,
686 &port_pma_attr_link_error_recovery.attr.attr,
687 &port_pma_attr_link_downed.attr.attr,
688 &port_pma_attr_port_rcv_errors.attr.attr,
689 &port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
690 &port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
691 &port_pma_attr_port_xmit_discards.attr.attr,
692 &port_pma_attr_port_xmit_constraint_errors.attr.attr,
693 &port_pma_attr_port_rcv_constraint_errors.attr.attr,
694 &port_pma_attr_local_link_integrity_errors.attr.attr,
695 &port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
696 &port_pma_attr_VL15_dropped.attr.attr,
697 &port_pma_attr_port_xmit_data.attr.attr,
698 &port_pma_attr_port_rcv_data.attr.attr,
699 &port_pma_attr_port_xmit_packets.attr.attr,
700 &port_pma_attr_port_rcv_packets.attr.attr,
701 &port_pma_attr_port_xmit_wait.attr.attr,
705 static struct attribute *pma_attrs_ext[] = {
706 &port_pma_attr_symbol_error.attr.attr,
707 &port_pma_attr_link_error_recovery.attr.attr,
708 &port_pma_attr_link_downed.attr.attr,
709 &port_pma_attr_port_rcv_errors.attr.attr,
710 &port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
711 &port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
712 &port_pma_attr_port_xmit_discards.attr.attr,
713 &port_pma_attr_port_xmit_constraint_errors.attr.attr,
714 &port_pma_attr_port_rcv_constraint_errors.attr.attr,
715 &port_pma_attr_local_link_integrity_errors.attr.attr,
716 &port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
717 &port_pma_attr_VL15_dropped.attr.attr,
718 &port_pma_attr_ext_port_xmit_data.attr.attr,
719 &port_pma_attr_ext_port_rcv_data.attr.attr,
720 &port_pma_attr_ext_port_xmit_packets.attr.attr,
721 &port_pma_attr_port_xmit_wait.attr.attr,
722 &port_pma_attr_ext_port_rcv_packets.attr.attr,
723 &port_pma_attr_ext_unicast_rcv_packets.attr.attr,
724 &port_pma_attr_ext_unicast_xmit_packets.attr.attr,
725 &port_pma_attr_ext_multicast_rcv_packets.attr.attr,
726 &port_pma_attr_ext_multicast_xmit_packets.attr.attr,
730 static struct attribute *pma_attrs_noietf[] = {
731 &port_pma_attr_symbol_error.attr.attr,
732 &port_pma_attr_link_error_recovery.attr.attr,
733 &port_pma_attr_link_downed.attr.attr,
734 &port_pma_attr_port_rcv_errors.attr.attr,
735 &port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
736 &port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
737 &port_pma_attr_port_xmit_discards.attr.attr,
738 &port_pma_attr_port_xmit_constraint_errors.attr.attr,
739 &port_pma_attr_port_rcv_constraint_errors.attr.attr,
740 &port_pma_attr_local_link_integrity_errors.attr.attr,
741 &port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
742 &port_pma_attr_VL15_dropped.attr.attr,
743 &port_pma_attr_ext_port_xmit_data.attr.attr,
744 &port_pma_attr_ext_port_rcv_data.attr.attr,
745 &port_pma_attr_ext_port_xmit_packets.attr.attr,
746 &port_pma_attr_ext_port_rcv_packets.attr.attr,
747 &port_pma_attr_port_xmit_wait.attr.attr,
751 static const struct attribute_group pma_group = {
756 static const struct attribute_group pma_group_ext = {
758 .attrs = pma_attrs_ext
761 static const struct attribute_group pma_group_noietf = {
763 .attrs = pma_attrs_noietf
766 static void ib_port_release(struct kobject *kobj)
768 struct ib_port *port = container_of(kobj, struct ib_port, kobj);
771 for (i = 0; i != ARRAY_SIZE(port->groups); i++)
772 kfree(port->groups[i].attrs);
773 if (port->hw_stats_data)
774 kfree(port->hw_stats_data->stats);
775 kfree(port->hw_stats_data);
779 static void ib_port_gid_attr_release(struct kobject *kobj)
781 struct gid_attr_group *gid_attr_group =
782 container_of(kobj, struct gid_attr_group, kobj);
785 for (i = 0; i != ARRAY_SIZE(gid_attr_group->groups); i++)
786 kfree(gid_attr_group->groups[i].attrs);
787 kfree(gid_attr_group);
790 static struct kobj_type port_type = {
791 .release = ib_port_release,
792 .sysfs_ops = &port_sysfs_ops,
793 .default_attrs = port_default_attrs
796 static struct kobj_type gid_attr_type = {
797 .sysfs_ops = &gid_attr_sysfs_ops,
798 .release = ib_port_gid_attr_release
802 * Figure out which counter table to use depending on
803 * the device capabilities.
805 static const struct attribute_group *get_counter_table(struct ib_device *dev,
808 struct ib_class_port_info cpi;
810 if (get_perf_mad(dev, port_num, IB_PMA_CLASS_PORT_INFO,
811 &cpi, 40, sizeof(cpi)) >= 0) {
812 if (cpi.capability_mask & IB_PMA_CLASS_CAP_EXT_WIDTH)
813 /* We have extended counters */
814 return &pma_group_ext;
816 if (cpi.capability_mask & IB_PMA_CLASS_CAP_EXT_WIDTH_NOIETF)
817 /* But not the IETF ones */
818 return &pma_group_noietf;
821 /* Fall back to normal counters */
825 static int update_hw_stats(struct ib_device *dev, struct rdma_hw_stats *stats,
826 u32 port_num, int index)
830 if (time_is_after_eq_jiffies(stats->timestamp + stats->lifespan))
832 ret = dev->ops.get_hw_stats(dev, stats, port_num, index);
835 if (ret == stats->num_counters)
836 stats->timestamp = jiffies;
841 static int print_hw_stat(struct ib_device *dev, int port_num,
842 struct rdma_hw_stats *stats, int index, char *buf)
844 u64 v = rdma_counter_get_hwstat_value(dev, port_num, index);
846 return sysfs_emit(buf, "%llu\n", stats->value[index] + v);
849 static ssize_t show_hw_stats(struct ib_device *ibdev,
850 struct rdma_hw_stats *stats, unsigned int index,
851 unsigned int port_num, char *buf)
855 mutex_lock(&stats->lock);
856 ret = update_hw_stats(ibdev, stats, port_num, index);
859 ret = print_hw_stat(ibdev, port_num, stats, index, buf);
861 mutex_unlock(&stats->lock);
866 static ssize_t show_stats_lifespan(struct ib_device *ibdev,
867 struct rdma_hw_stats *stats,
868 unsigned int index, unsigned int port_num,
873 mutex_lock(&stats->lock);
874 msecs = jiffies_to_msecs(stats->lifespan);
875 mutex_unlock(&stats->lock);
877 return sysfs_emit(buf, "%d\n", msecs);
880 static ssize_t set_stats_lifespan(struct ib_device *ibdev,
881 struct rdma_hw_stats *stats,
882 unsigned int index, unsigned int port_num,
883 const char *buf, size_t count)
889 ret = kstrtoint(buf, 10, &msecs);
892 if (msecs < 0 || msecs > 10000)
894 jiffies = msecs_to_jiffies(msecs);
896 mutex_lock(&stats->lock);
897 stats->lifespan = jiffies;
898 mutex_unlock(&stats->lock);
903 static struct hw_stats_device_data *
904 alloc_hw_stats_device(struct ib_device *ibdev)
906 struct hw_stats_device_data *data;
907 struct rdma_hw_stats *stats;
909 if (!ibdev->ops.alloc_hw_device_stats)
910 return ERR_PTR(-EOPNOTSUPP);
911 stats = ibdev->ops.alloc_hw_device_stats(ibdev);
913 return ERR_PTR(-ENOMEM);
914 if (!stats->names || stats->num_counters <= 0)
918 * Two extra attribue elements here, one for the lifespan entry and
919 * one to NULL terminate the list for the sysfs core code
921 data = kzalloc(struct_size(data, attrs, stats->num_counters + 1),
925 data->group.attrs = kcalloc(stats->num_counters + 2,
926 sizeof(*data->group.attrs), GFP_KERNEL);
927 if (!data->group.attrs)
930 mutex_init(&stats->lock);
931 data->group.name = "hw_counters";
939 return ERR_PTR(-ENOMEM);
942 void ib_device_release_hw_stats(struct hw_stats_device_data *data)
944 kfree(data->group.attrs);
949 int ib_setup_device_attrs(struct ib_device *ibdev)
951 struct hw_stats_device_attribute *attr;
952 struct hw_stats_device_data *data;
955 data = alloc_hw_stats_device(ibdev);
957 if (PTR_ERR(data) == -EOPNOTSUPP)
959 return PTR_ERR(data);
961 ibdev->hw_stats_data = data;
963 ret = ibdev->ops.get_hw_stats(ibdev, data->stats, 0,
964 data->stats->num_counters);
965 if (ret != data->stats->num_counters) {
966 if (WARN_ON(ret >= 0))
971 data->stats->timestamp = jiffies;
973 for (i = 0; i < data->stats->num_counters; i++) {
974 attr = &data->attrs[i];
975 sysfs_attr_init(&attr->attr.attr);
976 attr->attr.attr.name = data->stats->names[i];
977 attr->attr.attr.mode = 0444;
978 attr->attr.show = hw_stat_device_show;
979 attr->show = show_hw_stats;
980 data->group.attrs[i] = &attr->attr.attr;
983 attr = &data->attrs[i];
984 sysfs_attr_init(&attr->attr.attr);
985 attr->attr.attr.name = "lifespan";
986 attr->attr.attr.mode = 0644;
987 attr->attr.show = hw_stat_device_show;
988 attr->show = show_stats_lifespan;
989 attr->attr.store = hw_stat_device_store;
990 attr->store = set_stats_lifespan;
991 data->group.attrs[i] = &attr->attr.attr;
992 for (i = 0; i != ARRAY_SIZE(ibdev->groups); i++)
993 if (!ibdev->groups[i]) {
994 ibdev->groups[i] = &data->group;
997 WARN(true, "struct ib_device->groups is too small");
1001 static struct hw_stats_port_data *
1002 alloc_hw_stats_port(struct ib_port *port, struct attribute_group *group)
1004 struct ib_device *ibdev = port->ibdev;
1005 struct hw_stats_port_data *data;
1006 struct rdma_hw_stats *stats;
1008 if (!ibdev->ops.alloc_hw_port_stats)
1009 return ERR_PTR(-EOPNOTSUPP);
1010 stats = ibdev->ops.alloc_hw_port_stats(port->ibdev, port->port_num);
1012 return ERR_PTR(-ENOMEM);
1013 if (!stats->names || stats->num_counters <= 0)
1014 goto err_free_stats;
1017 * Two extra attribue elements here, one for the lifespan entry and
1018 * one to NULL terminate the list for the sysfs core code
1020 data = kzalloc(struct_size(data, attrs, stats->num_counters + 1),
1023 goto err_free_stats;
1024 group->attrs = kcalloc(stats->num_counters + 2,
1025 sizeof(*group->attrs), GFP_KERNEL);
1029 mutex_init(&stats->lock);
1030 group->name = "hw_counters";
1031 data->stats = stats;
1038 return ERR_PTR(-ENOMEM);
1041 static int setup_hw_port_stats(struct ib_port *port,
1042 struct attribute_group *group)
1044 struct hw_stats_port_attribute *attr;
1045 struct hw_stats_port_data *data;
1048 data = alloc_hw_stats_port(port, group);
1050 return PTR_ERR(data);
1052 ret = port->ibdev->ops.get_hw_stats(port->ibdev, data->stats,
1054 data->stats->num_counters);
1055 if (ret != data->stats->num_counters) {
1056 if (WARN_ON(ret >= 0))
1061 data->stats->timestamp = jiffies;
1063 for (i = 0; i < data->stats->num_counters; i++) {
1064 attr = &data->attrs[i];
1065 sysfs_attr_init(&attr->attr.attr);
1066 attr->attr.attr.name = data->stats->names[i];
1067 attr->attr.attr.mode = 0444;
1068 attr->attr.show = hw_stat_port_show;
1069 attr->show = show_hw_stats;
1070 group->attrs[i] = &attr->attr.attr;
1073 attr = &data->attrs[i];
1074 sysfs_attr_init(&attr->attr.attr);
1075 attr->attr.attr.name = "lifespan";
1076 attr->attr.attr.mode = 0644;
1077 attr->attr.show = hw_stat_port_show;
1078 attr->show = show_stats_lifespan;
1079 attr->attr.store = hw_stat_port_store;
1080 attr->store = set_stats_lifespan;
1081 group->attrs[i] = &attr->attr.attr;
1083 port->hw_stats_data = data;
1087 struct rdma_hw_stats *ib_get_hw_stats_port(struct ib_device *ibdev,
1090 if (!ibdev->port_data || !rdma_is_port_valid(ibdev, port_num) ||
1091 !ibdev->port_data[port_num].sysfs->hw_stats_data)
1093 return ibdev->port_data[port_num].sysfs->hw_stats_data->stats;
1097 alloc_port_table_group(const char *name, struct attribute_group *group,
1098 struct port_table_attribute *attrs, size_t num,
1099 ssize_t (*show)(struct ib_device *ibdev, u32 port_num,
1100 struct ib_port_attribute *, char *buf))
1102 struct attribute **attr_list;
1105 attr_list = kcalloc(num + 1, sizeof(*attr_list), GFP_KERNEL);
1109 for (i = 0; i < num; i++) {
1110 struct port_table_attribute *element = &attrs[i];
1112 if (snprintf(element->name, sizeof(element->name), "%d", i) >=
1113 sizeof(element->name))
1116 sysfs_attr_init(&element->attr.attr);
1117 element->attr.attr.name = element->name;
1118 element->attr.attr.mode = 0444;
1119 element->attr.show = show;
1122 attr_list[i] = &element->attr.attr;
1125 group->attrs = attr_list;
1134 * ibp0s9/ports/XX/gid_attrs/{ndevs,types}/YYY
1135 * YYY is the gid table index in decimal
1137 static int setup_gid_attrs(struct ib_port *port,
1138 const struct ib_port_attr *attr)
1140 struct gid_attr_group *gid_attr_group;
1143 gid_attr_group = kzalloc(struct_size(gid_attr_group, attrs_list,
1144 attr->gid_tbl_len * 2),
1146 if (!gid_attr_group)
1148 gid_attr_group->port = port;
1149 kobject_init(&gid_attr_group->kobj, &gid_attr_type);
1151 ret = alloc_port_table_group("ndevs", &gid_attr_group->groups[0],
1152 gid_attr_group->attrs_list,
1154 show_port_gid_attr_ndev);
1157 gid_attr_group->groups_list[0] = &gid_attr_group->groups[0];
1159 ret = alloc_port_table_group(
1160 "types", &gid_attr_group->groups[1],
1161 gid_attr_group->attrs_list + attr->gid_tbl_len,
1162 attr->gid_tbl_len, show_port_gid_attr_gid_type);
1165 gid_attr_group->groups_list[1] = &gid_attr_group->groups[1];
1167 ret = kobject_add(&gid_attr_group->kobj, &port->kobj, "gid_attrs");
1170 ret = sysfs_create_groups(&gid_attr_group->kobj,
1171 gid_attr_group->groups_list);
1174 port->gid_attr_group = gid_attr_group;
1178 kobject_del(&gid_attr_group->kobj);
1180 kobject_put(&gid_attr_group->kobj);
1184 static void destroy_gid_attrs(struct ib_port *port)
1186 struct gid_attr_group *gid_attr_group = port->gid_attr_group;
1188 if (!gid_attr_group)
1190 sysfs_remove_groups(&gid_attr_group->kobj, gid_attr_group->groups_list);
1191 kobject_del(&gid_attr_group->kobj);
1192 kobject_put(&gid_attr_group->kobj);
1197 * ibp0s9/ports/XX/{gids,pkeys,counters}/YYY
1199 static struct ib_port *setup_port(struct ib_core_device *coredev, int port_num,
1200 const struct ib_port_attr *attr)
1202 struct ib_device *device = rdma_device_to_ibdev(&coredev->dev);
1203 bool is_full_dev = &device->coredev == coredev;
1204 const struct attribute_group **cur_group;
1208 p = kzalloc(struct_size(p, attrs_list,
1209 attr->gid_tbl_len + attr->pkey_tbl_len),
1212 return ERR_PTR(-ENOMEM);
1214 p->port_num = port_num;
1215 kobject_init(&p->kobj, &port_type);
1217 cur_group = p->groups_list;
1218 ret = alloc_port_table_group("gids", &p->groups[0], p->attrs_list,
1219 attr->gid_tbl_len, show_port_gid);
1222 *cur_group++ = &p->groups[0];
1224 if (attr->pkey_tbl_len) {
1225 ret = alloc_port_table_group("pkeys", &p->groups[1],
1226 p->attrs_list + attr->gid_tbl_len,
1227 attr->pkey_tbl_len, show_port_pkey);
1230 *cur_group++ = &p->groups[1];
1234 * If port == 0, it means hw_counters are per device and not per
1235 * port, so holder should be device. Therefore skip per port
1236 * counter initialization.
1238 if (port_num && is_full_dev) {
1239 ret = setup_hw_port_stats(p, &p->groups[2]);
1240 if (ret && ret != -EOPNOTSUPP)
1243 *cur_group++ = &p->groups[2];
1246 if (device->ops.process_mad && is_full_dev)
1247 *cur_group++ = get_counter_table(device, port_num);
1249 ret = kobject_add(&p->kobj, coredev->ports_kobj, "%d", port_num);
1252 ret = sysfs_create_groups(&p->kobj, p->groups_list);
1256 list_add_tail(&p->kobj.entry, &coredev->port_list);
1257 if (device->port_data && is_full_dev)
1258 device->port_data[port_num].sysfs = p;
1263 kobject_del(&p->kobj);
1265 kobject_put(&p->kobj);
1266 return ERR_PTR(ret);
1269 static void destroy_port(struct ib_port *port)
1271 if (port->ibdev->port_data &&
1272 port->ibdev->port_data[port->port_num].sysfs == port)
1273 port->ibdev->port_data[port->port_num].sysfs = NULL;
1274 list_del(&port->kobj.entry);
1275 sysfs_remove_groups(&port->kobj, port->groups_list);
1276 kobject_del(&port->kobj);
1277 kobject_put(&port->kobj);
1280 static const char *node_type_string(int node_type)
1282 switch (node_type) {
1283 case RDMA_NODE_IB_CA:
1285 case RDMA_NODE_IB_SWITCH:
1287 case RDMA_NODE_IB_ROUTER:
1289 case RDMA_NODE_RNIC:
1291 case RDMA_NODE_USNIC:
1293 case RDMA_NODE_USNIC_UDP:
1295 case RDMA_NODE_UNSPECIFIED:
1296 return "unspecified";
1301 static ssize_t node_type_show(struct device *device,
1302 struct device_attribute *attr, char *buf)
1304 struct ib_device *dev = rdma_device_to_ibdev(device);
1306 return sysfs_emit(buf, "%d: %s\n", dev->node_type,
1307 node_type_string(dev->node_type));
1309 static DEVICE_ATTR_RO(node_type);
1311 static ssize_t sys_image_guid_show(struct device *device,
1312 struct device_attribute *dev_attr, char *buf)
1314 struct ib_device *dev = rdma_device_to_ibdev(device);
1315 __be16 *guid = (__be16 *)&dev->attrs.sys_image_guid;
1317 return sysfs_emit(buf, "%04x:%04x:%04x:%04x\n",
1318 be16_to_cpu(guid[0]),
1319 be16_to_cpu(guid[1]),
1320 be16_to_cpu(guid[2]),
1321 be16_to_cpu(guid[3]));
1323 static DEVICE_ATTR_RO(sys_image_guid);
1325 static ssize_t node_guid_show(struct device *device,
1326 struct device_attribute *attr, char *buf)
1328 struct ib_device *dev = rdma_device_to_ibdev(device);
1329 __be16 *node_guid = (__be16 *)&dev->node_guid;
1331 return sysfs_emit(buf, "%04x:%04x:%04x:%04x\n",
1332 be16_to_cpu(node_guid[0]),
1333 be16_to_cpu(node_guid[1]),
1334 be16_to_cpu(node_guid[2]),
1335 be16_to_cpu(node_guid[3]));
1337 static DEVICE_ATTR_RO(node_guid);
1339 static ssize_t node_desc_show(struct device *device,
1340 struct device_attribute *attr, char *buf)
1342 struct ib_device *dev = rdma_device_to_ibdev(device);
1344 return sysfs_emit(buf, "%.64s\n", dev->node_desc);
1347 static ssize_t node_desc_store(struct device *device,
1348 struct device_attribute *attr,
1349 const char *buf, size_t count)
1351 struct ib_device *dev = rdma_device_to_ibdev(device);
1352 struct ib_device_modify desc = {};
1355 if (!dev->ops.modify_device)
1358 memcpy(desc.node_desc, buf, min_t(int, count, IB_DEVICE_NODE_DESC_MAX));
1359 ret = ib_modify_device(dev, IB_DEVICE_MODIFY_NODE_DESC, &desc);
1365 static DEVICE_ATTR_RW(node_desc);
1367 static ssize_t fw_ver_show(struct device *device, struct device_attribute *attr,
1370 struct ib_device *dev = rdma_device_to_ibdev(device);
1371 char version[IB_FW_VERSION_NAME_MAX] = {};
1373 ib_get_device_fw_str(dev, version);
1375 return sysfs_emit(buf, "%s\n", version);
1377 static DEVICE_ATTR_RO(fw_ver);
1379 static struct attribute *ib_dev_attrs[] = {
1380 &dev_attr_node_type.attr,
1381 &dev_attr_node_guid.attr,
1382 &dev_attr_sys_image_guid.attr,
1383 &dev_attr_fw_ver.attr,
1384 &dev_attr_node_desc.attr,
1388 const struct attribute_group ib_dev_attr_group = {
1389 .attrs = ib_dev_attrs,
1392 void ib_free_port_attrs(struct ib_core_device *coredev)
1394 struct kobject *p, *t;
1396 list_for_each_entry_safe(p, t, &coredev->port_list, entry) {
1397 struct ib_port *port = container_of(p, struct ib_port, kobj);
1399 destroy_gid_attrs(port);
1403 kobject_put(coredev->ports_kobj);
1406 int ib_setup_port_attrs(struct ib_core_device *coredev)
1408 struct ib_device *device = rdma_device_to_ibdev(&coredev->dev);
1409 bool is_full_dev = &device->coredev == coredev;
1413 coredev->ports_kobj = kobject_create_and_add("ports",
1414 &coredev->dev.kobj);
1415 if (!coredev->ports_kobj)
1418 rdma_for_each_port (device, port_num) {
1419 struct ib_port_attr attr;
1420 struct ib_port *port;
1422 ret = ib_query_port(device, port_num, &attr);
1426 port = setup_port(coredev, port_num, &attr);
1428 ret = PTR_ERR(port);
1432 ret = setup_gid_attrs(port, &attr);
1436 if (device->ops.init_port && is_full_dev) {
1437 ret = device->ops.init_port(device, port_num,
1446 ib_free_port_attrs(coredev);
1451 * ib_port_register_module_stat - add module counters under relevant port
1454 * @device: IB device to add counters
1455 * @port_num: valid port number
1456 * @kobj: pointer to the kobject to initialize
1457 * @ktype: pointer to the ktype for this kobject.
1458 * @name: the name of the kobject
1460 int ib_port_register_module_stat(struct ib_device *device, u32 port_num,
1461 struct kobject *kobj, struct kobj_type *ktype,
1464 struct kobject *p, *t;
1467 list_for_each_entry_safe(p, t, &device->coredev.port_list, entry) {
1468 struct ib_port *port = container_of(p, struct ib_port, kobj);
1470 if (port->port_num != port_num)
1473 ret = kobject_init_and_add(kobj, ktype, &port->kobj, "%s",
1483 EXPORT_SYMBOL(ib_port_register_module_stat);
1486 * ib_port_unregister_module_stat - release module counters
1487 * @kobj: pointer to the kobject to release
1489 void ib_port_unregister_module_stat(struct kobject *kobj)
1493 EXPORT_SYMBOL(ib_port_unregister_module_stat);