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 struct ib_device *ib_port_sysfs_get_ibdev_kobj(struct kobject *kobj,
134 struct ib_port *port = container_of(kobj, struct ib_port, kobj);
136 *port_num = port->port_num;
139 EXPORT_SYMBOL(ib_port_sysfs_get_ibdev_kobj);
141 static const struct sysfs_ops port_sysfs_ops = {
142 .show = port_attr_show,
143 .store = port_attr_store
146 static ssize_t hw_stat_device_show(struct device *dev,
147 struct device_attribute *attr, char *buf)
149 struct hw_stats_device_attribute *stat_attr =
150 container_of(attr, struct hw_stats_device_attribute, attr);
151 struct ib_device *ibdev = container_of(dev, struct ib_device, dev);
153 return stat_attr->show(ibdev, ibdev->hw_stats_data->stats,
154 stat_attr - ibdev->hw_stats_data->attrs, 0, buf);
157 static ssize_t hw_stat_device_store(struct device *dev,
158 struct device_attribute *attr,
159 const char *buf, size_t count)
161 struct hw_stats_device_attribute *stat_attr =
162 container_of(attr, struct hw_stats_device_attribute, attr);
163 struct ib_device *ibdev = container_of(dev, struct ib_device, dev);
165 return stat_attr->store(ibdev, ibdev->hw_stats_data->stats,
166 stat_attr - ibdev->hw_stats_data->attrs, 0, buf,
170 static ssize_t hw_stat_port_show(struct ib_device *ibdev, u32 port_num,
171 struct ib_port_attribute *attr, char *buf)
173 struct hw_stats_port_attribute *stat_attr =
174 container_of(attr, struct hw_stats_port_attribute, attr);
175 struct ib_port *port = ibdev->port_data[port_num].sysfs;
177 return stat_attr->show(ibdev, port->hw_stats_data->stats,
178 stat_attr - port->hw_stats_data->attrs,
179 port->port_num, buf);
182 static ssize_t hw_stat_port_store(struct ib_device *ibdev, u32 port_num,
183 struct ib_port_attribute *attr,
184 const char *buf, size_t count)
186 struct hw_stats_port_attribute *stat_attr =
187 container_of(attr, struct hw_stats_port_attribute, attr);
188 struct ib_port *port = ibdev->port_data[port_num].sysfs;
190 return stat_attr->store(ibdev, port->hw_stats_data->stats,
191 stat_attr - port->hw_stats_data->attrs,
192 port->port_num, buf, count);
195 static ssize_t gid_attr_show(struct kobject *kobj,
196 struct attribute *attr, char *buf)
198 struct ib_port_attribute *port_attr =
199 container_of(attr, struct ib_port_attribute, attr);
200 struct ib_port *p = container_of(kobj, struct gid_attr_group,
203 if (!port_attr->show)
206 return port_attr->show(p->ibdev, p->port_num, port_attr, buf);
209 static const struct sysfs_ops gid_attr_sysfs_ops = {
210 .show = gid_attr_show
213 static ssize_t state_show(struct ib_device *ibdev, u32 port_num,
214 struct ib_port_attribute *unused, char *buf)
216 struct ib_port_attr attr;
219 static const char *state_name[] = {
220 [IB_PORT_NOP] = "NOP",
221 [IB_PORT_DOWN] = "DOWN",
222 [IB_PORT_INIT] = "INIT",
223 [IB_PORT_ARMED] = "ARMED",
224 [IB_PORT_ACTIVE] = "ACTIVE",
225 [IB_PORT_ACTIVE_DEFER] = "ACTIVE_DEFER"
228 ret = ib_query_port(ibdev, port_num, &attr);
232 return sysfs_emit(buf, "%d: %s\n", attr.state,
234 attr.state < ARRAY_SIZE(state_name) ?
235 state_name[attr.state] :
239 static ssize_t lid_show(struct ib_device *ibdev, u32 port_num,
240 struct ib_port_attribute *unused, char *buf)
242 struct ib_port_attr attr;
245 ret = ib_query_port(ibdev, port_num, &attr);
249 return sysfs_emit(buf, "0x%x\n", attr.lid);
252 static ssize_t lid_mask_count_show(struct ib_device *ibdev, u32 port_num,
253 struct ib_port_attribute *unused, char *buf)
255 struct ib_port_attr attr;
258 ret = ib_query_port(ibdev, port_num, &attr);
262 return sysfs_emit(buf, "%u\n", attr.lmc);
265 static ssize_t sm_lid_show(struct ib_device *ibdev, u32 port_num,
266 struct ib_port_attribute *unused, char *buf)
268 struct ib_port_attr attr;
271 ret = ib_query_port(ibdev, port_num, &attr);
275 return sysfs_emit(buf, "0x%x\n", attr.sm_lid);
278 static ssize_t sm_sl_show(struct ib_device *ibdev, u32 port_num,
279 struct ib_port_attribute *unused, char *buf)
281 struct ib_port_attr attr;
284 ret = ib_query_port(ibdev, port_num, &attr);
288 return sysfs_emit(buf, "%u\n", attr.sm_sl);
291 static ssize_t cap_mask_show(struct ib_device *ibdev, u32 port_num,
292 struct ib_port_attribute *unused, char *buf)
294 struct ib_port_attr attr;
297 ret = ib_query_port(ibdev, port_num, &attr);
301 return sysfs_emit(buf, "0x%08x\n", attr.port_cap_flags);
304 static ssize_t rate_show(struct ib_device *ibdev, u32 port_num,
305 struct ib_port_attribute *unused, char *buf)
307 struct ib_port_attr attr;
309 int rate; /* in deci-Gb/sec */
312 ret = ib_query_port(ibdev, port_num, &attr);
316 switch (attr.active_speed) {
346 default: /* default to SDR for invalid rates */
352 rate *= ib_width_enum_to_int(attr.active_width);
356 return sysfs_emit(buf, "%d%s Gb/sec (%dX%s)\n", rate / 10,
357 rate % 10 ? ".5" : "",
358 ib_width_enum_to_int(attr.active_width), speed);
361 static const char *phys_state_to_str(enum ib_port_phys_state phys_state)
363 static const char *phys_state_str[] = {
368 "PortConfigurationTraining",
374 if (phys_state < ARRAY_SIZE(phys_state_str))
375 return phys_state_str[phys_state];
379 static ssize_t phys_state_show(struct ib_device *ibdev, u32 port_num,
380 struct ib_port_attribute *unused, char *buf)
382 struct ib_port_attr attr;
386 ret = ib_query_port(ibdev, port_num, &attr);
390 return sysfs_emit(buf, "%u: %s\n", attr.phys_state,
391 phys_state_to_str(attr.phys_state));
394 static ssize_t link_layer_show(struct ib_device *ibdev, u32 port_num,
395 struct ib_port_attribute *unused, char *buf)
399 switch (rdma_port_get_link_layer(ibdev, port_num)) {
400 case IB_LINK_LAYER_INFINIBAND:
401 output = "InfiniBand";
403 case IB_LINK_LAYER_ETHERNET:
411 return sysfs_emit(buf, "%s\n", output);
414 static IB_PORT_ATTR_RO(state);
415 static IB_PORT_ATTR_RO(lid);
416 static IB_PORT_ATTR_RO(lid_mask_count);
417 static IB_PORT_ATTR_RO(sm_lid);
418 static IB_PORT_ATTR_RO(sm_sl);
419 static IB_PORT_ATTR_RO(cap_mask);
420 static IB_PORT_ATTR_RO(rate);
421 static IB_PORT_ATTR_RO(phys_state);
422 static IB_PORT_ATTR_RO(link_layer);
424 static struct attribute *port_default_attrs[] = {
425 &ib_port_attr_state.attr,
426 &ib_port_attr_lid.attr,
427 &ib_port_attr_lid_mask_count.attr,
428 &ib_port_attr_sm_lid.attr,
429 &ib_port_attr_sm_sl.attr,
430 &ib_port_attr_cap_mask.attr,
431 &ib_port_attr_rate.attr,
432 &ib_port_attr_phys_state.attr,
433 &ib_port_attr_link_layer.attr,
437 static ssize_t print_ndev(const struct ib_gid_attr *gid_attr, char *buf)
439 struct net_device *ndev;
443 ndev = rcu_dereference(gid_attr->ndev);
445 ret = sysfs_emit(buf, "%s\n", ndev->name);
450 static ssize_t print_gid_type(const struct ib_gid_attr *gid_attr, char *buf)
452 return sysfs_emit(buf, "%s\n",
453 ib_cache_gid_type_str(gid_attr->gid_type));
456 static ssize_t _show_port_gid_attr(
457 struct ib_device *ibdev, u32 port_num, struct ib_port_attribute *attr,
459 ssize_t (*print)(const struct ib_gid_attr *gid_attr, char *buf))
461 struct port_table_attribute *tab_attr =
462 container_of(attr, struct port_table_attribute, attr);
463 const struct ib_gid_attr *gid_attr;
466 gid_attr = rdma_get_gid_attr(ibdev, port_num, tab_attr->index);
467 if (IS_ERR(gid_attr))
468 /* -EINVAL is returned for user space compatibility reasons. */
471 ret = print(gid_attr, buf);
472 rdma_put_gid_attr(gid_attr);
476 static ssize_t show_port_gid(struct ib_device *ibdev, u32 port_num,
477 struct ib_port_attribute *attr, char *buf)
479 struct port_table_attribute *tab_attr =
480 container_of(attr, struct port_table_attribute, attr);
481 const struct ib_gid_attr *gid_attr;
484 gid_attr = rdma_get_gid_attr(ibdev, port_num, tab_attr->index);
485 if (IS_ERR(gid_attr)) {
486 const union ib_gid zgid = {};
488 /* If reading GID fails, it is likely due to GID entry being
489 * empty (invalid) or reserved GID in the table. User space
490 * expects to read GID table entries as long as it given index
491 * is within GID table size. Administrative/debugging tool
492 * fails to query rest of the GID entries if it hits error
493 * while querying a GID of the given index. To avoid user
494 * space throwing such error on fail to read gid, return zero
495 * GID as before. This maintains backward compatibility.
497 return sysfs_emit(buf, "%pI6\n", zgid.raw);
500 len = sysfs_emit(buf, "%pI6\n", gid_attr->gid.raw);
501 rdma_put_gid_attr(gid_attr);
505 static ssize_t show_port_gid_attr_ndev(struct ib_device *ibdev, u32 port_num,
506 struct ib_port_attribute *attr,
509 return _show_port_gid_attr(ibdev, port_num, attr, buf, print_ndev);
512 static ssize_t show_port_gid_attr_gid_type(struct ib_device *ibdev,
514 struct ib_port_attribute *attr,
517 return _show_port_gid_attr(ibdev, port_num, attr, buf, print_gid_type);
520 static ssize_t show_port_pkey(struct ib_device *ibdev, u32 port_num,
521 struct ib_port_attribute *attr, char *buf)
523 struct port_table_attribute *tab_attr =
524 container_of(attr, struct port_table_attribute, attr);
528 ret = ib_query_pkey(ibdev, port_num, tab_attr->index, &pkey);
532 return sysfs_emit(buf, "0x%04x\n", pkey);
535 #define PORT_PMA_ATTR(_name, _counter, _width, _offset) \
536 struct port_table_attribute port_pma_attr_##_name = { \
537 .attr = __ATTR(_name, S_IRUGO, show_pma_counter, NULL), \
538 .index = (_offset) | ((_width) << 16) | ((_counter) << 24), \
539 .attr_id = IB_PMA_PORT_COUNTERS, \
542 #define PORT_PMA_ATTR_EXT(_name, _width, _offset) \
543 struct port_table_attribute port_pma_attr_ext_##_name = { \
544 .attr = __ATTR(_name, S_IRUGO, show_pma_counter, NULL), \
545 .index = (_offset) | ((_width) << 16), \
546 .attr_id = IB_PMA_PORT_COUNTERS_EXT, \
550 * Get a Perfmgmt MAD block of data.
551 * Returns error code or the number of bytes retrieved.
553 static int get_perf_mad(struct ib_device *dev, int port_num, __be16 attr,
554 void *data, int offset, size_t size)
556 struct ib_mad *in_mad;
557 struct ib_mad *out_mad;
558 size_t mad_size = sizeof(*out_mad);
559 u16 out_mad_pkey_index = 0;
562 if (!dev->ops.process_mad)
565 in_mad = kzalloc(sizeof(*in_mad), GFP_KERNEL);
566 out_mad = kzalloc(sizeof(*out_mad), GFP_KERNEL);
567 if (!in_mad || !out_mad) {
572 in_mad->mad_hdr.base_version = 1;
573 in_mad->mad_hdr.mgmt_class = IB_MGMT_CLASS_PERF_MGMT;
574 in_mad->mad_hdr.class_version = 1;
575 in_mad->mad_hdr.method = IB_MGMT_METHOD_GET;
576 in_mad->mad_hdr.attr_id = attr;
578 if (attr != IB_PMA_CLASS_PORT_INFO)
579 in_mad->data[41] = port_num; /* PortSelect field */
581 if ((dev->ops.process_mad(dev, IB_MAD_IGNORE_MKEY, port_num, NULL, NULL,
582 in_mad, out_mad, &mad_size,
583 &out_mad_pkey_index) &
584 (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) !=
585 (IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY)) {
589 memcpy(data, out_mad->data + offset, size);
597 static ssize_t show_pma_counter(struct ib_device *ibdev, u32 port_num,
598 struct ib_port_attribute *attr, char *buf)
600 struct port_table_attribute *tab_attr =
601 container_of(attr, struct port_table_attribute, attr);
602 int offset = tab_attr->index & 0xffff;
603 int width = (tab_attr->index >> 16) & 0xff;
608 ret = get_perf_mad(ibdev, port_num, tab_attr->attr_id, &data,
609 40 + offset / 8, sizeof(data));
615 len = sysfs_emit(buf, "%d\n",
616 (*data >> (4 - (offset % 8))) & 0xf);
619 len = sysfs_emit(buf, "%u\n", *data);
622 len = sysfs_emit(buf, "%u\n", be16_to_cpup((__be16 *)data));
625 len = sysfs_emit(buf, "%u\n", be32_to_cpup((__be32 *)data));
628 len = sysfs_emit(buf, "%llu\n", be64_to_cpup((__be64 *)data));
638 static PORT_PMA_ATTR(symbol_error , 0, 16, 32);
639 static PORT_PMA_ATTR(link_error_recovery , 1, 8, 48);
640 static PORT_PMA_ATTR(link_downed , 2, 8, 56);
641 static PORT_PMA_ATTR(port_rcv_errors , 3, 16, 64);
642 static PORT_PMA_ATTR(port_rcv_remote_physical_errors, 4, 16, 80);
643 static PORT_PMA_ATTR(port_rcv_switch_relay_errors , 5, 16, 96);
644 static PORT_PMA_ATTR(port_xmit_discards , 6, 16, 112);
645 static PORT_PMA_ATTR(port_xmit_constraint_errors , 7, 8, 128);
646 static PORT_PMA_ATTR(port_rcv_constraint_errors , 8, 8, 136);
647 static PORT_PMA_ATTR(local_link_integrity_errors , 9, 4, 152);
648 static PORT_PMA_ATTR(excessive_buffer_overrun_errors, 10, 4, 156);
649 static PORT_PMA_ATTR(VL15_dropped , 11, 16, 176);
650 static PORT_PMA_ATTR(port_xmit_data , 12, 32, 192);
651 static PORT_PMA_ATTR(port_rcv_data , 13, 32, 224);
652 static PORT_PMA_ATTR(port_xmit_packets , 14, 32, 256);
653 static PORT_PMA_ATTR(port_rcv_packets , 15, 32, 288);
654 static PORT_PMA_ATTR(port_xmit_wait , 0, 32, 320);
657 * Counters added by extended set
659 static PORT_PMA_ATTR_EXT(port_xmit_data , 64, 64);
660 static PORT_PMA_ATTR_EXT(port_rcv_data , 64, 128);
661 static PORT_PMA_ATTR_EXT(port_xmit_packets , 64, 192);
662 static PORT_PMA_ATTR_EXT(port_rcv_packets , 64, 256);
663 static PORT_PMA_ATTR_EXT(unicast_xmit_packets , 64, 320);
664 static PORT_PMA_ATTR_EXT(unicast_rcv_packets , 64, 384);
665 static PORT_PMA_ATTR_EXT(multicast_xmit_packets , 64, 448);
666 static PORT_PMA_ATTR_EXT(multicast_rcv_packets , 64, 512);
668 static struct attribute *pma_attrs[] = {
669 &port_pma_attr_symbol_error.attr.attr,
670 &port_pma_attr_link_error_recovery.attr.attr,
671 &port_pma_attr_link_downed.attr.attr,
672 &port_pma_attr_port_rcv_errors.attr.attr,
673 &port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
674 &port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
675 &port_pma_attr_port_xmit_discards.attr.attr,
676 &port_pma_attr_port_xmit_constraint_errors.attr.attr,
677 &port_pma_attr_port_rcv_constraint_errors.attr.attr,
678 &port_pma_attr_local_link_integrity_errors.attr.attr,
679 &port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
680 &port_pma_attr_VL15_dropped.attr.attr,
681 &port_pma_attr_port_xmit_data.attr.attr,
682 &port_pma_attr_port_rcv_data.attr.attr,
683 &port_pma_attr_port_xmit_packets.attr.attr,
684 &port_pma_attr_port_rcv_packets.attr.attr,
685 &port_pma_attr_port_xmit_wait.attr.attr,
689 static struct attribute *pma_attrs_ext[] = {
690 &port_pma_attr_symbol_error.attr.attr,
691 &port_pma_attr_link_error_recovery.attr.attr,
692 &port_pma_attr_link_downed.attr.attr,
693 &port_pma_attr_port_rcv_errors.attr.attr,
694 &port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
695 &port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
696 &port_pma_attr_port_xmit_discards.attr.attr,
697 &port_pma_attr_port_xmit_constraint_errors.attr.attr,
698 &port_pma_attr_port_rcv_constraint_errors.attr.attr,
699 &port_pma_attr_local_link_integrity_errors.attr.attr,
700 &port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
701 &port_pma_attr_VL15_dropped.attr.attr,
702 &port_pma_attr_ext_port_xmit_data.attr.attr,
703 &port_pma_attr_ext_port_rcv_data.attr.attr,
704 &port_pma_attr_ext_port_xmit_packets.attr.attr,
705 &port_pma_attr_port_xmit_wait.attr.attr,
706 &port_pma_attr_ext_port_rcv_packets.attr.attr,
707 &port_pma_attr_ext_unicast_rcv_packets.attr.attr,
708 &port_pma_attr_ext_unicast_xmit_packets.attr.attr,
709 &port_pma_attr_ext_multicast_rcv_packets.attr.attr,
710 &port_pma_attr_ext_multicast_xmit_packets.attr.attr,
714 static struct attribute *pma_attrs_noietf[] = {
715 &port_pma_attr_symbol_error.attr.attr,
716 &port_pma_attr_link_error_recovery.attr.attr,
717 &port_pma_attr_link_downed.attr.attr,
718 &port_pma_attr_port_rcv_errors.attr.attr,
719 &port_pma_attr_port_rcv_remote_physical_errors.attr.attr,
720 &port_pma_attr_port_rcv_switch_relay_errors.attr.attr,
721 &port_pma_attr_port_xmit_discards.attr.attr,
722 &port_pma_attr_port_xmit_constraint_errors.attr.attr,
723 &port_pma_attr_port_rcv_constraint_errors.attr.attr,
724 &port_pma_attr_local_link_integrity_errors.attr.attr,
725 &port_pma_attr_excessive_buffer_overrun_errors.attr.attr,
726 &port_pma_attr_VL15_dropped.attr.attr,
727 &port_pma_attr_ext_port_xmit_data.attr.attr,
728 &port_pma_attr_ext_port_rcv_data.attr.attr,
729 &port_pma_attr_ext_port_xmit_packets.attr.attr,
730 &port_pma_attr_ext_port_rcv_packets.attr.attr,
731 &port_pma_attr_port_xmit_wait.attr.attr,
735 static const struct attribute_group pma_group = {
740 static const struct attribute_group pma_group_ext = {
742 .attrs = pma_attrs_ext
745 static const struct attribute_group pma_group_noietf = {
747 .attrs = pma_attrs_noietf
750 static void ib_port_release(struct kobject *kobj)
752 struct ib_port *port = container_of(kobj, struct ib_port, kobj);
755 for (i = 0; i != ARRAY_SIZE(port->groups); i++)
756 kfree(port->groups[i].attrs);
757 if (port->hw_stats_data)
758 rdma_free_hw_stats_struct(port->hw_stats_data->stats);
759 kfree(port->hw_stats_data);
763 static void ib_port_gid_attr_release(struct kobject *kobj)
765 struct gid_attr_group *gid_attr_group =
766 container_of(kobj, struct gid_attr_group, kobj);
769 for (i = 0; i != ARRAY_SIZE(gid_attr_group->groups); i++)
770 kfree(gid_attr_group->groups[i].attrs);
771 kfree(gid_attr_group);
774 static struct kobj_type port_type = {
775 .release = ib_port_release,
776 .sysfs_ops = &port_sysfs_ops,
777 .default_attrs = port_default_attrs
780 static struct kobj_type gid_attr_type = {
781 .sysfs_ops = &gid_attr_sysfs_ops,
782 .release = ib_port_gid_attr_release
786 * Figure out which counter table to use depending on
787 * the device capabilities.
789 static const struct attribute_group *get_counter_table(struct ib_device *dev,
792 struct ib_class_port_info cpi;
794 if (get_perf_mad(dev, port_num, IB_PMA_CLASS_PORT_INFO,
795 &cpi, 40, sizeof(cpi)) >= 0) {
796 if (cpi.capability_mask & IB_PMA_CLASS_CAP_EXT_WIDTH)
797 /* We have extended counters */
798 return &pma_group_ext;
800 if (cpi.capability_mask & IB_PMA_CLASS_CAP_EXT_WIDTH_NOIETF)
801 /* But not the IETF ones */
802 return &pma_group_noietf;
805 /* Fall back to normal counters */
809 static int update_hw_stats(struct ib_device *dev, struct rdma_hw_stats *stats,
810 u32 port_num, int index)
814 if (time_is_after_eq_jiffies(stats->timestamp + stats->lifespan))
816 ret = dev->ops.get_hw_stats(dev, stats, port_num, index);
819 if (ret == stats->num_counters)
820 stats->timestamp = jiffies;
825 static int print_hw_stat(struct ib_device *dev, int port_num,
826 struct rdma_hw_stats *stats, int index, char *buf)
828 u64 v = rdma_counter_get_hwstat_value(dev, port_num, index);
830 return sysfs_emit(buf, "%llu\n", stats->value[index] + v);
833 static ssize_t show_hw_stats(struct ib_device *ibdev,
834 struct rdma_hw_stats *stats, unsigned int index,
835 unsigned int port_num, char *buf)
839 mutex_lock(&stats->lock);
840 ret = update_hw_stats(ibdev, stats, port_num, index);
843 ret = print_hw_stat(ibdev, port_num, stats, index, buf);
845 mutex_unlock(&stats->lock);
850 static ssize_t show_stats_lifespan(struct ib_device *ibdev,
851 struct rdma_hw_stats *stats,
852 unsigned int index, unsigned int port_num,
857 mutex_lock(&stats->lock);
858 msecs = jiffies_to_msecs(stats->lifespan);
859 mutex_unlock(&stats->lock);
861 return sysfs_emit(buf, "%d\n", msecs);
864 static ssize_t set_stats_lifespan(struct ib_device *ibdev,
865 struct rdma_hw_stats *stats,
866 unsigned int index, unsigned int port_num,
867 const char *buf, size_t count)
873 ret = kstrtoint(buf, 10, &msecs);
876 if (msecs < 0 || msecs > 10000)
878 jiffies = msecs_to_jiffies(msecs);
880 mutex_lock(&stats->lock);
881 stats->lifespan = jiffies;
882 mutex_unlock(&stats->lock);
887 static struct hw_stats_device_data *
888 alloc_hw_stats_device(struct ib_device *ibdev)
890 struct hw_stats_device_data *data;
891 struct rdma_hw_stats *stats;
893 if (!ibdev->ops.alloc_hw_device_stats)
894 return ERR_PTR(-EOPNOTSUPP);
895 stats = ibdev->ops.alloc_hw_device_stats(ibdev);
897 return ERR_PTR(-ENOMEM);
898 if (!stats->descs || stats->num_counters <= 0)
902 * Two extra attribue elements here, one for the lifespan entry and
903 * one to NULL terminate the list for the sysfs core code
905 data = kzalloc(struct_size(data, attrs, stats->num_counters + 1),
909 data->group.attrs = kcalloc(stats->num_counters + 2,
910 sizeof(*data->group.attrs), GFP_KERNEL);
911 if (!data->group.attrs)
914 data->group.name = "hw_counters";
921 rdma_free_hw_stats_struct(stats);
922 return ERR_PTR(-ENOMEM);
925 void ib_device_release_hw_stats(struct hw_stats_device_data *data)
927 kfree(data->group.attrs);
928 rdma_free_hw_stats_struct(data->stats);
932 int ib_setup_device_attrs(struct ib_device *ibdev)
934 struct hw_stats_device_attribute *attr;
935 struct hw_stats_device_data *data;
936 bool opstat_skipped = false;
939 data = alloc_hw_stats_device(ibdev);
941 if (PTR_ERR(data) == -EOPNOTSUPP)
943 return PTR_ERR(data);
945 ibdev->hw_stats_data = data;
947 ret = ibdev->ops.get_hw_stats(ibdev, data->stats, 0,
948 data->stats->num_counters);
949 if (ret != data->stats->num_counters) {
950 if (WARN_ON(ret >= 0))
955 data->stats->timestamp = jiffies;
957 for (i = 0; i < data->stats->num_counters; i++) {
958 if (data->stats->descs[i].flags & IB_STAT_FLAG_OPTIONAL) {
959 opstat_skipped = true;
963 WARN_ON(opstat_skipped);
964 attr = &data->attrs[pos];
965 sysfs_attr_init(&attr->attr.attr);
966 attr->attr.attr.name = data->stats->descs[i].name;
967 attr->attr.attr.mode = 0444;
968 attr->attr.show = hw_stat_device_show;
969 attr->show = show_hw_stats;
970 data->group.attrs[pos] = &attr->attr.attr;
974 attr = &data->attrs[pos];
975 sysfs_attr_init(&attr->attr.attr);
976 attr->attr.attr.name = "lifespan";
977 attr->attr.attr.mode = 0644;
978 attr->attr.show = hw_stat_device_show;
979 attr->show = show_stats_lifespan;
980 attr->attr.store = hw_stat_device_store;
981 attr->store = set_stats_lifespan;
982 data->group.attrs[pos] = &attr->attr.attr;
983 for (i = 0; i != ARRAY_SIZE(ibdev->groups); i++)
984 if (!ibdev->groups[i]) {
985 ibdev->groups[i] = &data->group;
988 WARN(true, "struct ib_device->groups is too small");
992 static struct hw_stats_port_data *
993 alloc_hw_stats_port(struct ib_port *port, struct attribute_group *group)
995 struct ib_device *ibdev = port->ibdev;
996 struct hw_stats_port_data *data;
997 struct rdma_hw_stats *stats;
999 if (!ibdev->ops.alloc_hw_port_stats)
1000 return ERR_PTR(-EOPNOTSUPP);
1001 stats = ibdev->ops.alloc_hw_port_stats(port->ibdev, port->port_num);
1003 return ERR_PTR(-ENOMEM);
1004 if (!stats->descs || stats->num_counters <= 0)
1005 goto err_free_stats;
1008 * Two extra attribue elements here, one for the lifespan entry and
1009 * one to NULL terminate the list for the sysfs core code
1011 data = kzalloc(struct_size(data, attrs, stats->num_counters + 1),
1014 goto err_free_stats;
1015 group->attrs = kcalloc(stats->num_counters + 2,
1016 sizeof(*group->attrs), GFP_KERNEL);
1020 group->name = "hw_counters";
1021 data->stats = stats;
1027 rdma_free_hw_stats_struct(stats);
1028 return ERR_PTR(-ENOMEM);
1031 static int setup_hw_port_stats(struct ib_port *port,
1032 struct attribute_group *group)
1034 struct hw_stats_port_attribute *attr;
1035 struct hw_stats_port_data *data;
1036 bool opstat_skipped = false;
1037 int i, ret, pos = 0;
1039 data = alloc_hw_stats_port(port, group);
1041 return PTR_ERR(data);
1043 ret = port->ibdev->ops.get_hw_stats(port->ibdev, data->stats,
1045 data->stats->num_counters);
1046 if (ret != data->stats->num_counters) {
1047 if (WARN_ON(ret >= 0))
1052 data->stats->timestamp = jiffies;
1054 for (i = 0; i < data->stats->num_counters; i++) {
1055 if (data->stats->descs[i].flags & IB_STAT_FLAG_OPTIONAL) {
1056 opstat_skipped = true;
1060 WARN_ON(opstat_skipped);
1061 attr = &data->attrs[pos];
1062 sysfs_attr_init(&attr->attr.attr);
1063 attr->attr.attr.name = data->stats->descs[i].name;
1064 attr->attr.attr.mode = 0444;
1065 attr->attr.show = hw_stat_port_show;
1066 attr->show = show_hw_stats;
1067 group->attrs[pos] = &attr->attr.attr;
1071 attr = &data->attrs[pos];
1072 sysfs_attr_init(&attr->attr.attr);
1073 attr->attr.attr.name = "lifespan";
1074 attr->attr.attr.mode = 0644;
1075 attr->attr.show = hw_stat_port_show;
1076 attr->show = show_stats_lifespan;
1077 attr->attr.store = hw_stat_port_store;
1078 attr->store = set_stats_lifespan;
1079 group->attrs[pos] = &attr->attr.attr;
1081 port->hw_stats_data = data;
1085 struct rdma_hw_stats *ib_get_hw_stats_port(struct ib_device *ibdev,
1088 if (!ibdev->port_data || !rdma_is_port_valid(ibdev, port_num) ||
1089 !ibdev->port_data[port_num].sysfs->hw_stats_data)
1091 return ibdev->port_data[port_num].sysfs->hw_stats_data->stats;
1095 alloc_port_table_group(const char *name, struct attribute_group *group,
1096 struct port_table_attribute *attrs, size_t num,
1097 ssize_t (*show)(struct ib_device *ibdev, u32 port_num,
1098 struct ib_port_attribute *, char *buf))
1100 struct attribute **attr_list;
1103 attr_list = kcalloc(num + 1, sizeof(*attr_list), GFP_KERNEL);
1107 for (i = 0; i < num; i++) {
1108 struct port_table_attribute *element = &attrs[i];
1110 if (snprintf(element->name, sizeof(element->name), "%d", i) >=
1111 sizeof(element->name))
1114 sysfs_attr_init(&element->attr.attr);
1115 element->attr.attr.name = element->name;
1116 element->attr.attr.mode = 0444;
1117 element->attr.show = show;
1120 attr_list[i] = &element->attr.attr;
1123 group->attrs = attr_list;
1132 * ibp0s9/ports/XX/gid_attrs/{ndevs,types}/YYY
1133 * YYY is the gid table index in decimal
1135 static int setup_gid_attrs(struct ib_port *port,
1136 const struct ib_port_attr *attr)
1138 struct gid_attr_group *gid_attr_group;
1141 gid_attr_group = kzalloc(struct_size(gid_attr_group, attrs_list,
1142 attr->gid_tbl_len * 2),
1144 if (!gid_attr_group)
1146 gid_attr_group->port = port;
1147 kobject_init(&gid_attr_group->kobj, &gid_attr_type);
1149 ret = alloc_port_table_group("ndevs", &gid_attr_group->groups[0],
1150 gid_attr_group->attrs_list,
1152 show_port_gid_attr_ndev);
1155 gid_attr_group->groups_list[0] = &gid_attr_group->groups[0];
1157 ret = alloc_port_table_group(
1158 "types", &gid_attr_group->groups[1],
1159 gid_attr_group->attrs_list + attr->gid_tbl_len,
1160 attr->gid_tbl_len, show_port_gid_attr_gid_type);
1163 gid_attr_group->groups_list[1] = &gid_attr_group->groups[1];
1165 ret = kobject_add(&gid_attr_group->kobj, &port->kobj, "gid_attrs");
1168 ret = sysfs_create_groups(&gid_attr_group->kobj,
1169 gid_attr_group->groups_list);
1172 port->gid_attr_group = gid_attr_group;
1176 kobject_del(&gid_attr_group->kobj);
1178 kobject_put(&gid_attr_group->kobj);
1182 static void destroy_gid_attrs(struct ib_port *port)
1184 struct gid_attr_group *gid_attr_group = port->gid_attr_group;
1186 if (!gid_attr_group)
1188 sysfs_remove_groups(&gid_attr_group->kobj, gid_attr_group->groups_list);
1189 kobject_del(&gid_attr_group->kobj);
1190 kobject_put(&gid_attr_group->kobj);
1195 * ibp0s9/ports/XX/{gids,pkeys,counters}/YYY
1197 static struct ib_port *setup_port(struct ib_core_device *coredev, int port_num,
1198 const struct ib_port_attr *attr)
1200 struct ib_device *device = rdma_device_to_ibdev(&coredev->dev);
1201 bool is_full_dev = &device->coredev == coredev;
1202 const struct attribute_group **cur_group;
1206 p = kvzalloc(struct_size(p, attrs_list,
1207 attr->gid_tbl_len + attr->pkey_tbl_len),
1210 return ERR_PTR(-ENOMEM);
1212 p->port_num = port_num;
1213 kobject_init(&p->kobj, &port_type);
1215 cur_group = p->groups_list;
1216 ret = alloc_port_table_group("gids", &p->groups[0], p->attrs_list,
1217 attr->gid_tbl_len, show_port_gid);
1220 *cur_group++ = &p->groups[0];
1222 if (attr->pkey_tbl_len) {
1223 ret = alloc_port_table_group("pkeys", &p->groups[1],
1224 p->attrs_list + attr->gid_tbl_len,
1225 attr->pkey_tbl_len, show_port_pkey);
1228 *cur_group++ = &p->groups[1];
1232 * If port == 0, it means hw_counters are per device and not per
1233 * port, so holder should be device. Therefore skip per port
1234 * counter initialization.
1236 if (port_num && is_full_dev) {
1237 ret = setup_hw_port_stats(p, &p->groups[2]);
1238 if (ret && ret != -EOPNOTSUPP)
1241 *cur_group++ = &p->groups[2];
1244 if (device->ops.process_mad && is_full_dev)
1245 *cur_group++ = get_counter_table(device, port_num);
1247 ret = kobject_add(&p->kobj, coredev->ports_kobj, "%d", port_num);
1250 ret = sysfs_create_groups(&p->kobj, p->groups_list);
1254 ret = sysfs_create_groups(&p->kobj, device->ops.port_groups);
1259 list_add_tail(&p->kobj.entry, &coredev->port_list);
1260 if (device->port_data && is_full_dev)
1261 device->port_data[port_num].sysfs = p;
1266 sysfs_remove_groups(&p->kobj, p->groups_list);
1268 kobject_del(&p->kobj);
1270 kobject_put(&p->kobj);
1271 return ERR_PTR(ret);
1274 static void destroy_port(struct ib_core_device *coredev, struct ib_port *port)
1276 bool is_full_dev = &port->ibdev->coredev == coredev;
1278 if (port->ibdev->port_data &&
1279 port->ibdev->port_data[port->port_num].sysfs == port)
1280 port->ibdev->port_data[port->port_num].sysfs = NULL;
1281 list_del(&port->kobj.entry);
1283 sysfs_remove_groups(&port->kobj, port->ibdev->ops.port_groups);
1284 sysfs_remove_groups(&port->kobj, port->groups_list);
1285 kobject_del(&port->kobj);
1286 kobject_put(&port->kobj);
1289 static const char *node_type_string(int node_type)
1291 switch (node_type) {
1292 case RDMA_NODE_IB_CA:
1294 case RDMA_NODE_IB_SWITCH:
1296 case RDMA_NODE_IB_ROUTER:
1298 case RDMA_NODE_RNIC:
1300 case RDMA_NODE_USNIC:
1302 case RDMA_NODE_USNIC_UDP:
1304 case RDMA_NODE_UNSPECIFIED:
1305 return "unspecified";
1310 static ssize_t node_type_show(struct device *device,
1311 struct device_attribute *attr, char *buf)
1313 struct ib_device *dev = rdma_device_to_ibdev(device);
1315 return sysfs_emit(buf, "%u: %s\n", dev->node_type,
1316 node_type_string(dev->node_type));
1318 static DEVICE_ATTR_RO(node_type);
1320 static ssize_t sys_image_guid_show(struct device *device,
1321 struct device_attribute *dev_attr, char *buf)
1323 struct ib_device *dev = rdma_device_to_ibdev(device);
1324 __be16 *guid = (__be16 *)&dev->attrs.sys_image_guid;
1326 return sysfs_emit(buf, "%04x:%04x:%04x:%04x\n",
1327 be16_to_cpu(guid[0]),
1328 be16_to_cpu(guid[1]),
1329 be16_to_cpu(guid[2]),
1330 be16_to_cpu(guid[3]));
1332 static DEVICE_ATTR_RO(sys_image_guid);
1334 static ssize_t node_guid_show(struct device *device,
1335 struct device_attribute *attr, char *buf)
1337 struct ib_device *dev = rdma_device_to_ibdev(device);
1338 __be16 *node_guid = (__be16 *)&dev->node_guid;
1340 return sysfs_emit(buf, "%04x:%04x:%04x:%04x\n",
1341 be16_to_cpu(node_guid[0]),
1342 be16_to_cpu(node_guid[1]),
1343 be16_to_cpu(node_guid[2]),
1344 be16_to_cpu(node_guid[3]));
1346 static DEVICE_ATTR_RO(node_guid);
1348 static ssize_t node_desc_show(struct device *device,
1349 struct device_attribute *attr, char *buf)
1351 struct ib_device *dev = rdma_device_to_ibdev(device);
1353 return sysfs_emit(buf, "%.64s\n", dev->node_desc);
1356 static ssize_t node_desc_store(struct device *device,
1357 struct device_attribute *attr,
1358 const char *buf, size_t count)
1360 struct ib_device *dev = rdma_device_to_ibdev(device);
1361 struct ib_device_modify desc = {};
1364 if (!dev->ops.modify_device)
1367 memcpy(desc.node_desc, buf, min_t(int, count, IB_DEVICE_NODE_DESC_MAX));
1368 ret = ib_modify_device(dev, IB_DEVICE_MODIFY_NODE_DESC, &desc);
1374 static DEVICE_ATTR_RW(node_desc);
1376 static ssize_t fw_ver_show(struct device *device, struct device_attribute *attr,
1379 struct ib_device *dev = rdma_device_to_ibdev(device);
1380 char version[IB_FW_VERSION_NAME_MAX] = {};
1382 ib_get_device_fw_str(dev, version);
1384 return sysfs_emit(buf, "%s\n", version);
1386 static DEVICE_ATTR_RO(fw_ver);
1388 static struct attribute *ib_dev_attrs[] = {
1389 &dev_attr_node_type.attr,
1390 &dev_attr_node_guid.attr,
1391 &dev_attr_sys_image_guid.attr,
1392 &dev_attr_fw_ver.attr,
1393 &dev_attr_node_desc.attr,
1397 const struct attribute_group ib_dev_attr_group = {
1398 .attrs = ib_dev_attrs,
1401 void ib_free_port_attrs(struct ib_core_device *coredev)
1403 struct kobject *p, *t;
1405 list_for_each_entry_safe(p, t, &coredev->port_list, entry) {
1406 struct ib_port *port = container_of(p, struct ib_port, kobj);
1408 destroy_gid_attrs(port);
1409 destroy_port(coredev, port);
1412 kobject_put(coredev->ports_kobj);
1415 int ib_setup_port_attrs(struct ib_core_device *coredev)
1417 struct ib_device *device = rdma_device_to_ibdev(&coredev->dev);
1421 coredev->ports_kobj = kobject_create_and_add("ports",
1422 &coredev->dev.kobj);
1423 if (!coredev->ports_kobj)
1426 rdma_for_each_port (device, port_num) {
1427 struct ib_port_attr attr;
1428 struct ib_port *port;
1430 ret = ib_query_port(device, port_num, &attr);
1434 port = setup_port(coredev, port_num, &attr);
1436 ret = PTR_ERR(port);
1440 ret = setup_gid_attrs(port, &attr);
1447 ib_free_port_attrs(coredev);
1452 * ib_port_register_client_groups - Add an ib_client's attributes to the port
1454 * @ibdev: IB device to add counters
1455 * @port_num: valid port number
1456 * @groups: Group list of attributes
1458 * Do not use. Only for legacy sysfs compatibility.
1460 int ib_port_register_client_groups(struct ib_device *ibdev, u32 port_num,
1461 const struct attribute_group **groups)
1463 return sysfs_create_groups(&ibdev->port_data[port_num].sysfs->kobj,
1466 EXPORT_SYMBOL(ib_port_register_client_groups);
1468 void ib_port_unregister_client_groups(struct ib_device *ibdev, u32 port_num,
1469 const struct attribute_group **groups)
1471 return sysfs_remove_groups(&ibdev->port_data[port_num].sysfs->kobj,
1474 EXPORT_SYMBOL(ib_port_unregister_client_groups);