1 /*
2 * Copyright (c) 2016 Hisilicon Limited.
3 * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
4 *
5 * This software is available to you under a choice of one of two
6 * licenses. You may choose to be licensed under the terms of the GNU
7 * General Public License (GPL) Version 2, available from the file
8 * COPYING in the main directory of this source tree, or the
9 * OpenIB.org BSD license below:
10 *
11 * Redistribution and use in source and binary forms, with or
12 * without modification, are permitted provided that the following
13 * conditions are met:
14 *
15 * - Redistributions of source code must retain the above
16 * copyright notice, this list of conditions and the following
17 * disclaimer.
18 *
19 * - Redistributions in binary form must reproduce the above
20 * copyright notice, this list of conditions and the following
21 * disclaimer in the documentation and/or other materials
22 * provided with the distribution.
23 *
24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
31 * SOFTWARE.
32 */
33 #include <linux/acpi.h>
34 #include <linux/module.h>
35 #include <rdma/ib_addr.h>
36 #include <rdma/ib_smi.h>
37 #include <rdma/ib_user_verbs.h>
38 #include <rdma/ib_cache.h>
39 #include "hns_roce_common.h"
40 #include "hns_roce_device.h"
41 #include "hns_roce_hem.h"
42 #include "hns_roce_hw_v2.h"
43 #include "hns_roce_bond.h"
44
hns_roce_set_mac(struct hns_roce_dev * hr_dev,u32 port,const u8 * addr)45 static int hns_roce_set_mac(struct hns_roce_dev *hr_dev, u32 port,
46 const u8 *addr)
47 {
48 u8 phy_port;
49 u32 i;
50
51 if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09)
52 return 0;
53
54 if (!memcmp(hr_dev->dev_addr[port], addr, ETH_ALEN))
55 return 0;
56
57 for (i = 0; i < ETH_ALEN; i++)
58 hr_dev->dev_addr[port][i] = addr[i];
59
60 phy_port = hr_dev->iboe.phy_port[port];
61 return hr_dev->hw->set_mac(hr_dev, phy_port, addr);
62 }
63
hns_roce_add_gid(const struct ib_gid_attr * attr,void ** context)64 static int hns_roce_add_gid(const struct ib_gid_attr *attr, void **context)
65 {
66 struct hns_roce_dev *hr_dev = to_hr_dev(attr->device);
67 u32 port = attr->port_num - 1;
68 int ret;
69
70 if (port >= hr_dev->caps.num_ports)
71 return -EINVAL;
72
73 ret = hr_dev->hw->set_gid(hr_dev, attr->index, &attr->gid, attr);
74
75 return ret;
76 }
77
hns_roce_del_gid(const struct ib_gid_attr * attr,void ** context)78 static int hns_roce_del_gid(const struct ib_gid_attr *attr, void **context)
79 {
80 struct hns_roce_dev *hr_dev = to_hr_dev(attr->device);
81 u32 port = attr->port_num - 1;
82 int ret;
83
84 if (port >= hr_dev->caps.num_ports)
85 return -EINVAL;
86
87 ret = hr_dev->hw->set_gid(hr_dev, attr->index, NULL, NULL);
88
89 return ret;
90 }
91
hns_roce_get_port_state(struct hns_roce_dev * hr_dev,u32 port_num,enum ib_port_state * state)92 static int hns_roce_get_port_state(struct hns_roce_dev *hr_dev, u32 port_num,
93 enum ib_port_state *state)
94 {
95 struct hns_roce_bond_group *bond_grp;
96 u8 bus_num = get_hr_bus_num(hr_dev);
97 struct net_device *net_dev;
98
99 net_dev = ib_device_get_netdev(&hr_dev->ib_dev, port_num);
100 if (!net_dev)
101 return -ENODEV;
102
103 if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_BOND) {
104 bond_grp = hns_roce_get_bond_grp(net_dev, bus_num);
105 if (bond_grp) {
106 *state = ib_get_curr_port_state(bond_grp->upper_dev);
107 goto out;
108 }
109 }
110
111 *state = ib_get_curr_port_state(net_dev);
112 out:
113 dev_put(net_dev);
114 return 0;
115 }
116
handle_en_event(struct net_device * netdev,struct hns_roce_dev * hr_dev,u32 port,unsigned long event)117 static int handle_en_event(struct net_device *netdev,
118 struct hns_roce_dev *hr_dev,
119 u32 port, unsigned long event)
120 {
121 struct ib_device *ibdev = &hr_dev->ib_dev;
122 struct device *dev = hr_dev->dev;
123 enum ib_port_state curr_state;
124 struct ib_event ibevent;
125 int ret = 0;
126
127 if (!netdev) {
128 dev_err(dev, "can't find netdev on port(%u)!\n", port);
129 return -ENODEV;
130 }
131
132 switch (event) {
133 case NETDEV_REGISTER:
134 case NETDEV_CHANGEADDR:
135 ret = hns_roce_set_mac(hr_dev, port, netdev->dev_addr);
136 break;
137 case NETDEV_UP:
138 case NETDEV_CHANGE:
139 ret = hns_roce_set_mac(hr_dev, port, netdev->dev_addr);
140 if (ret)
141 return ret;
142 fallthrough;
143 case NETDEV_DOWN:
144 if (!netif_is_lag_master(netdev))
145 break;
146 curr_state = ib_get_curr_port_state(netdev);
147
148 write_lock_irq(&ibdev->cache_lock);
149 if (ibdev->port_data[port].cache.last_port_state == curr_state) {
150 write_unlock_irq(&ibdev->cache_lock);
151 return 0;
152 }
153 ibdev->port_data[port].cache.last_port_state = curr_state;
154 write_unlock_irq(&ibdev->cache_lock);
155
156 ibevent.event = (curr_state == IB_PORT_DOWN) ?
157 IB_EVENT_PORT_ERR : IB_EVENT_PORT_ACTIVE;
158 ibevent.device = ibdev;
159 ibevent.element.port_num = port + 1;
160 ib_dispatch_event(&ibevent);
161 break;
162 default:
163 dev_dbg(dev, "NETDEV event = 0x%x!\n", (u32)(event));
164 break;
165 }
166
167 return ret;
168 }
169
hns_roce_netdev_event(struct notifier_block * self,unsigned long event,void * ptr)170 static int hns_roce_netdev_event(struct notifier_block *self,
171 unsigned long event, void *ptr)
172 {
173 struct net_device *dev = netdev_notifier_info_to_dev(ptr);
174 struct hns_roce_bond_group *bond_grp;
175 struct hns_roce_ib_iboe *iboe = NULL;
176 struct hns_roce_dev *hr_dev = NULL;
177 struct net_device *upper = NULL;
178 int ret;
179 u32 port;
180
181 hr_dev = container_of(self, struct hns_roce_dev, iboe.nb);
182 iboe = &hr_dev->iboe;
183 if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_BOND) {
184 bond_grp = hns_roce_get_bond_grp(get_hr_netdev(hr_dev, 0),
185 get_hr_bus_num(hr_dev));
186 upper = bond_grp ? bond_grp->upper_dev : NULL;
187 }
188
189 for (port = 0; port < hr_dev->caps.num_ports; port++) {
190 if ((!upper && dev == iboe->netdevs[port]) ||
191 (upper && dev == upper)) {
192 ret = handle_en_event(dev, hr_dev, port, event);
193 if (ret)
194 return NOTIFY_DONE;
195 break;
196 }
197 }
198
199 return NOTIFY_DONE;
200 }
201
hns_roce_setup_mtu_mac(struct hns_roce_dev * hr_dev)202 static int hns_roce_setup_mtu_mac(struct hns_roce_dev *hr_dev)
203 {
204 struct net_device *net_dev;
205 int ret;
206 u8 i;
207
208 for (i = 0; i < hr_dev->caps.num_ports; i++) {
209 net_dev = get_hr_netdev(hr_dev, i);
210 ret = hns_roce_set_mac(hr_dev, i, net_dev->dev_addr);
211 if (ret)
212 return ret;
213 }
214
215 return 0;
216 }
217
hns_roce_query_device(struct ib_device * ib_dev,struct ib_device_attr * props,struct ib_udata * uhw)218 static int hns_roce_query_device(struct ib_device *ib_dev,
219 struct ib_device_attr *props,
220 struct ib_udata *uhw)
221 {
222 struct hns_roce_dev *hr_dev = to_hr_dev(ib_dev);
223
224 memset(props, 0, sizeof(*props));
225
226 props->fw_ver = hr_dev->caps.fw_ver;
227 props->sys_image_guid = cpu_to_be64(hr_dev->sys_image_guid);
228 props->max_mr_size = (u64)(~(0ULL));
229 props->page_size_cap = hr_dev->caps.page_size_cap;
230 props->vendor_id = hr_dev->vendor_id;
231 props->vendor_part_id = hr_dev->vendor_part_id;
232 props->hw_ver = hr_dev->hw_rev;
233 props->max_qp = hr_dev->caps.num_qps;
234 props->max_qp_wr = hr_dev->caps.max_wqes;
235 props->device_cap_flags = IB_DEVICE_PORT_ACTIVE_EVENT |
236 IB_DEVICE_RC_RNR_NAK_GEN;
237 props->max_send_sge = hr_dev->caps.max_sq_sg;
238 props->max_recv_sge = hr_dev->caps.max_rq_sg;
239 props->max_sge_rd = hr_dev->caps.max_sq_sg;
240 props->max_cq = hr_dev->caps.num_cqs;
241 props->max_cqe = hr_dev->caps.max_cqes;
242 props->max_mr = hr_dev->caps.num_mtpts;
243 props->max_pd = hr_dev->caps.num_pds;
244 props->max_qp_rd_atom = hr_dev->caps.max_qp_dest_rdma;
245 props->max_qp_init_rd_atom = hr_dev->caps.max_qp_init_rdma;
246 props->atomic_cap = hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_ATOMIC ?
247 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
248 props->max_pkeys = 1;
249 props->local_ca_ack_delay = hr_dev->caps.local_ca_ack_delay;
250 props->max_ah = INT_MAX;
251 props->cq_caps.max_cq_moderation_period = HNS_ROCE_MAX_CQ_PERIOD;
252 props->cq_caps.max_cq_moderation_count = HNS_ROCE_MAX_CQ_COUNT;
253 if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08)
254 props->cq_caps.max_cq_moderation_period = HNS_ROCE_MAX_CQ_PERIOD_HIP08;
255
256 if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ) {
257 props->max_srq = hr_dev->caps.num_srqs;
258 props->max_srq_wr = hr_dev->caps.max_srq_wrs;
259 props->max_srq_sge = hr_dev->caps.max_srq_sges;
260 }
261
262 if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_FRMR &&
263 hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09) {
264 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
265 props->max_fast_reg_page_list_len = HNS_ROCE_FRMR_MAX_PA;
266 }
267
268 if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_XRC)
269 props->device_cap_flags |= IB_DEVICE_XRC;
270
271 return 0;
272 }
273
hns_roce_query_port(struct ib_device * ib_dev,u32 port_num,struct ib_port_attr * props)274 static int hns_roce_query_port(struct ib_device *ib_dev, u32 port_num,
275 struct ib_port_attr *props)
276 {
277 struct hns_roce_dev *hr_dev = to_hr_dev(ib_dev);
278 struct net_device *net_dev;
279 enum ib_mtu mtu;
280 u32 port;
281 int ret;
282
283 port = port_num - 1;
284
285 /* props being zeroed by the caller, avoid zeroing it here */
286
287 props->max_mtu = hr_dev->caps.max_mtu;
288 props->gid_tbl_len = hr_dev->caps.gid_table_len[port];
289 props->port_cap_flags = IB_PORT_CM_SUP | IB_PORT_REINIT_SUP |
290 IB_PORT_VENDOR_CLASS_SUP |
291 IB_PORT_BOOT_MGMT_SUP;
292 props->max_msg_sz = HNS_ROCE_MAX_MSG_LEN;
293 props->pkey_tbl_len = 1;
294 ret = ib_get_eth_speed(ib_dev, port_num, &props->active_speed,
295 &props->active_width);
296 if (ret)
297 ibdev_warn(ib_dev, "failed to get speed, ret = %d.\n", ret);
298
299 net_dev = ib_device_get_netdev(ib_dev, port_num);
300 if (!net_dev) {
301 ibdev_err(ib_dev, "find netdev %u failed!\n", port);
302 return -EINVAL;
303 }
304
305 mtu = iboe_get_mtu(net_dev->mtu);
306 props->active_mtu = mtu ? min(props->max_mtu, mtu) : IB_MTU_256;
307
308 dev_put(net_dev);
309
310 ret = hns_roce_get_port_state(hr_dev, port_num, &props->state);
311 if (ret) {
312 ibdev_err(ib_dev, "failed to get port state.\n");
313 return ret;
314 }
315
316 props->phys_state = props->state == IB_PORT_ACTIVE ?
317 IB_PORT_PHYS_STATE_LINK_UP :
318 IB_PORT_PHYS_STATE_DISABLED;
319 return 0;
320 }
321
hns_roce_get_link_layer(struct ib_device * device,u32 port_num)322 static enum rdma_link_layer hns_roce_get_link_layer(struct ib_device *device,
323 u32 port_num)
324 {
325 return IB_LINK_LAYER_ETHERNET;
326 }
327
hns_roce_query_pkey(struct ib_device * ib_dev,u32 port,u16 index,u16 * pkey)328 static int hns_roce_query_pkey(struct ib_device *ib_dev, u32 port, u16 index,
329 u16 *pkey)
330 {
331 if (index > 0)
332 return -EINVAL;
333
334 *pkey = PKEY_ID;
335
336 return 0;
337 }
338
hns_roce_modify_device(struct ib_device * ib_dev,int mask,struct ib_device_modify * props)339 static int hns_roce_modify_device(struct ib_device *ib_dev, int mask,
340 struct ib_device_modify *props)
341 {
342 unsigned long flags;
343
344 if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
345 return -EOPNOTSUPP;
346
347 if (mask & IB_DEVICE_MODIFY_NODE_DESC) {
348 spin_lock_irqsave(&to_hr_dev(ib_dev)->sm_lock, flags);
349 memcpy(ib_dev->node_desc, props->node_desc, NODE_DESC_SIZE);
350 spin_unlock_irqrestore(&to_hr_dev(ib_dev)->sm_lock, flags);
351 }
352
353 return 0;
354 }
355
356 struct hns_user_mmap_entry *
hns_roce_user_mmap_entry_insert(struct ib_ucontext * ucontext,u64 address,size_t length,enum hns_roce_mmap_type mmap_type)357 hns_roce_user_mmap_entry_insert(struct ib_ucontext *ucontext, u64 address,
358 size_t length,
359 enum hns_roce_mmap_type mmap_type)
360 {
361 struct hns_user_mmap_entry *entry;
362 int ret;
363
364 entry = kzalloc(sizeof(*entry), GFP_KERNEL);
365 if (!entry)
366 return NULL;
367
368 entry->address = address;
369 entry->mmap_type = mmap_type;
370
371 switch (mmap_type) {
372 /* pgoff 0 must be used by DB for compatibility */
373 case HNS_ROCE_MMAP_TYPE_DB:
374 ret = rdma_user_mmap_entry_insert_exact(
375 ucontext, &entry->rdma_entry, length, 0);
376 break;
377 case HNS_ROCE_MMAP_TYPE_DWQE:
378 ret = rdma_user_mmap_entry_insert_range(
379 ucontext, &entry->rdma_entry, length, 1,
380 U32_MAX);
381 break;
382 default:
383 ret = -EINVAL;
384 break;
385 }
386
387 if (ret) {
388 kfree(entry);
389 return NULL;
390 }
391
392 return entry;
393 }
394
hns_roce_dealloc_uar_entry(struct hns_roce_ucontext * context)395 static void hns_roce_dealloc_uar_entry(struct hns_roce_ucontext *context)
396 {
397 if (context->db_mmap_entry)
398 rdma_user_mmap_entry_remove(
399 &context->db_mmap_entry->rdma_entry);
400 }
401
hns_roce_alloc_uar_entry(struct ib_ucontext * uctx)402 static int hns_roce_alloc_uar_entry(struct ib_ucontext *uctx)
403 {
404 struct hns_roce_ucontext *context = to_hr_ucontext(uctx);
405 u64 address;
406
407 address = context->uar.pfn << PAGE_SHIFT;
408 context->db_mmap_entry = hns_roce_user_mmap_entry_insert(
409 uctx, address, PAGE_SIZE, HNS_ROCE_MMAP_TYPE_DB);
410 if (!context->db_mmap_entry)
411 return -ENOMEM;
412
413 return 0;
414 }
415
hns_roce_alloc_ucontext(struct ib_ucontext * uctx,struct ib_udata * udata)416 static int hns_roce_alloc_ucontext(struct ib_ucontext *uctx,
417 struct ib_udata *udata)
418 {
419 struct hns_roce_ucontext *context = to_hr_ucontext(uctx);
420 struct hns_roce_dev *hr_dev = to_hr_dev(uctx->device);
421 struct hns_roce_ib_alloc_ucontext_resp resp = {};
422 struct hns_roce_ib_alloc_ucontext ucmd = {};
423 int ret = -EAGAIN;
424
425 if (!hr_dev->active)
426 goto error_out;
427
428 resp.qp_tab_size = hr_dev->caps.num_qps;
429 resp.srq_tab_size = hr_dev->caps.num_srqs;
430
431 ret = ib_copy_from_udata(&ucmd, udata,
432 min(udata->inlen, sizeof(ucmd)));
433 if (ret)
434 goto error_out;
435
436 if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09)
437 context->config = ucmd.config & HNS_ROCE_EXSGE_FLAGS;
438
439 if (context->config & HNS_ROCE_EXSGE_FLAGS) {
440 resp.config |= HNS_ROCE_RSP_EXSGE_FLAGS;
441 resp.max_inline_data = hr_dev->caps.max_sq_inline;
442 }
443
444 if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_RQ_INLINE) {
445 context->config |= ucmd.config & HNS_ROCE_RQ_INLINE_FLAGS;
446 if (context->config & HNS_ROCE_RQ_INLINE_FLAGS)
447 resp.config |= HNS_ROCE_RSP_RQ_INLINE_FLAGS;
448 }
449
450 if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_CQE_INLINE) {
451 context->config |= ucmd.config & HNS_ROCE_CQE_INLINE_FLAGS;
452 if (context->config & HNS_ROCE_CQE_INLINE_FLAGS)
453 resp.config |= HNS_ROCE_RSP_CQE_INLINE_FLAGS;
454 }
455
456 if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09)
457 resp.congest_type = hr_dev->caps.cong_cap;
458
459 ret = hns_roce_uar_alloc(hr_dev, &context->uar);
460 if (ret)
461 goto error_out;
462
463 ret = hns_roce_alloc_uar_entry(uctx);
464 if (ret)
465 goto error_fail_uar_entry;
466
467 if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_CQ_RECORD_DB ||
468 hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_RECORD_DB) {
469 INIT_LIST_HEAD(&context->page_list);
470 mutex_init(&context->page_mutex);
471 }
472
473 resp.cqe_size = hr_dev->caps.cqe_sz;
474
475 ret = ib_copy_to_udata(udata, &resp,
476 min(udata->outlen, sizeof(resp)));
477 if (ret)
478 goto error_fail_copy_to_udata;
479
480 hns_roce_get_cq_bankid_for_uctx(context);
481
482 return 0;
483
484 error_fail_copy_to_udata:
485 if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_CQ_RECORD_DB ||
486 hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_RECORD_DB)
487 mutex_destroy(&context->page_mutex);
488 hns_roce_dealloc_uar_entry(context);
489
490 error_fail_uar_entry:
491 ida_free(&hr_dev->uar_ida.ida, (int)context->uar.logic_idx);
492
493 error_out:
494 atomic64_inc(&hr_dev->dfx_cnt[HNS_ROCE_DFX_UCTX_ALLOC_ERR_CNT]);
495
496 return ret;
497 }
498
hns_roce_dealloc_ucontext(struct ib_ucontext * ibcontext)499 static void hns_roce_dealloc_ucontext(struct ib_ucontext *ibcontext)
500 {
501 struct hns_roce_ucontext *context = to_hr_ucontext(ibcontext);
502 struct hns_roce_dev *hr_dev = to_hr_dev(ibcontext->device);
503
504 hns_roce_put_cq_bankid_for_uctx(context);
505
506 if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_CQ_RECORD_DB ||
507 hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_RECORD_DB)
508 mutex_destroy(&context->page_mutex);
509
510 hns_roce_dealloc_uar_entry(context);
511
512 ida_free(&hr_dev->uar_ida.ida, (int)context->uar.logic_idx);
513 }
514
hns_roce_mmap(struct ib_ucontext * uctx,struct vm_area_struct * vma)515 static int hns_roce_mmap(struct ib_ucontext *uctx, struct vm_area_struct *vma)
516 {
517 struct hns_roce_dev *hr_dev = to_hr_dev(uctx->device);
518 struct rdma_user_mmap_entry *rdma_entry;
519 struct hns_user_mmap_entry *entry;
520 phys_addr_t pfn;
521 pgprot_t prot;
522 int ret;
523
524 if (hr_dev->dis_db) {
525 atomic64_inc(&hr_dev->dfx_cnt[HNS_ROCE_DFX_MMAP_ERR_CNT]);
526 return -EPERM;
527 }
528
529 rdma_entry = rdma_user_mmap_entry_get_pgoff(uctx, vma->vm_pgoff);
530 if (!rdma_entry) {
531 atomic64_inc(&hr_dev->dfx_cnt[HNS_ROCE_DFX_MMAP_ERR_CNT]);
532 return -EINVAL;
533 }
534
535 entry = to_hns_mmap(rdma_entry);
536 pfn = entry->address >> PAGE_SHIFT;
537
538 switch (entry->mmap_type) {
539 case HNS_ROCE_MMAP_TYPE_DB:
540 case HNS_ROCE_MMAP_TYPE_DWQE:
541 prot = pgprot_device(vma->vm_page_prot);
542 break;
543 default:
544 ret = -EINVAL;
545 goto out;
546 }
547
548 ret = rdma_user_mmap_io(uctx, vma, pfn, rdma_entry->npages * PAGE_SIZE,
549 prot, rdma_entry);
550
551 out:
552 rdma_user_mmap_entry_put(rdma_entry);
553 if (ret)
554 atomic64_inc(&hr_dev->dfx_cnt[HNS_ROCE_DFX_MMAP_ERR_CNT]);
555
556 return ret;
557 }
558
hns_roce_free_mmap(struct rdma_user_mmap_entry * rdma_entry)559 static void hns_roce_free_mmap(struct rdma_user_mmap_entry *rdma_entry)
560 {
561 struct hns_user_mmap_entry *entry = to_hns_mmap(rdma_entry);
562
563 kfree(entry);
564 }
565
hns_roce_port_immutable(struct ib_device * ib_dev,u32 port_num,struct ib_port_immutable * immutable)566 static int hns_roce_port_immutable(struct ib_device *ib_dev, u32 port_num,
567 struct ib_port_immutable *immutable)
568 {
569 struct ib_port_attr attr;
570 int ret;
571
572 ret = ib_query_port(ib_dev, port_num, &attr);
573 if (ret)
574 return ret;
575
576 immutable->pkey_tbl_len = attr.pkey_tbl_len;
577 immutable->gid_tbl_len = attr.gid_tbl_len;
578
579 immutable->max_mad_size = IB_MGMT_MAD_SIZE;
580 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
581 if (to_hr_dev(ib_dev)->caps.flags & HNS_ROCE_CAP_FLAG_ROCE_V1_V2)
582 immutable->core_cap_flags |= RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
583
584 return 0;
585 }
586
hns_roce_disassociate_ucontext(struct ib_ucontext * ibcontext)587 static void hns_roce_disassociate_ucontext(struct ib_ucontext *ibcontext)
588 {
589 }
590
hns_roce_get_fw_ver(struct ib_device * device,char * str)591 static void hns_roce_get_fw_ver(struct ib_device *device, char *str)
592 {
593 u64 fw_ver = to_hr_dev(device)->caps.fw_ver;
594 unsigned int major, minor, sub_minor;
595
596 major = upper_32_bits(fw_ver);
597 minor = high_16_bits(lower_32_bits(fw_ver));
598 sub_minor = low_16_bits(fw_ver);
599
600 snprintf(str, IB_FW_VERSION_NAME_MAX, "%u.%u.%04u", major, minor,
601 sub_minor);
602 }
603
604 #define HNS_ROCE_HW_CNT(ename, cname) \
605 [HNS_ROCE_HW_##ename##_CNT].name = cname
606
607 static const struct rdma_stat_desc hns_roce_port_stats_descs[] = {
608 HNS_ROCE_HW_CNT(RX_RC_PKT, "rx_rc_pkt"),
609 HNS_ROCE_HW_CNT(RX_UC_PKT, "rx_uc_pkt"),
610 HNS_ROCE_HW_CNT(RX_UD_PKT, "rx_ud_pkt"),
611 HNS_ROCE_HW_CNT(RX_XRC_PKT, "rx_xrc_pkt"),
612 HNS_ROCE_HW_CNT(RX_PKT, "rx_pkt"),
613 HNS_ROCE_HW_CNT(RX_ERR_PKT, "rx_err_pkt"),
614 HNS_ROCE_HW_CNT(RX_CNP_PKT, "rx_cnp_pkt"),
615 HNS_ROCE_HW_CNT(TX_RC_PKT, "tx_rc_pkt"),
616 HNS_ROCE_HW_CNT(TX_UC_PKT, "tx_uc_pkt"),
617 HNS_ROCE_HW_CNT(TX_UD_PKT, "tx_ud_pkt"),
618 HNS_ROCE_HW_CNT(TX_XRC_PKT, "tx_xrc_pkt"),
619 HNS_ROCE_HW_CNT(TX_PKT, "tx_pkt"),
620 HNS_ROCE_HW_CNT(TX_ERR_PKT, "tx_err_pkt"),
621 HNS_ROCE_HW_CNT(TX_CNP_PKT, "tx_cnp_pkt"),
622 HNS_ROCE_HW_CNT(TRP_GET_MPT_ERR_PKT, "trp_get_mpt_err_pkt"),
623 HNS_ROCE_HW_CNT(TRP_GET_IRRL_ERR_PKT, "trp_get_irrl_err_pkt"),
624 HNS_ROCE_HW_CNT(ECN_DB, "ecn_doorbell"),
625 HNS_ROCE_HW_CNT(RX_BUF, "rx_buffer"),
626 HNS_ROCE_HW_CNT(TRP_RX_SOF, "trp_rx_sof"),
627 HNS_ROCE_HW_CNT(CQ_CQE, "cq_cqe"),
628 HNS_ROCE_HW_CNT(CQ_POE, "cq_poe"),
629 HNS_ROCE_HW_CNT(CQ_NOTIFY, "cq_notify"),
630 };
631
hns_roce_alloc_hw_port_stats(struct ib_device * device,u32 port_num)632 static struct rdma_hw_stats *hns_roce_alloc_hw_port_stats(
633 struct ib_device *device, u32 port_num)
634 {
635 struct hns_roce_dev *hr_dev = to_hr_dev(device);
636
637 if (port_num > hr_dev->caps.num_ports) {
638 ibdev_err(device, "invalid port num.\n");
639 return NULL;
640 }
641
642 return rdma_alloc_hw_stats_struct(hns_roce_port_stats_descs,
643 ARRAY_SIZE(hns_roce_port_stats_descs),
644 RDMA_HW_STATS_DEFAULT_LIFESPAN);
645 }
646
hns_roce_get_hw_stats(struct ib_device * device,struct rdma_hw_stats * stats,u32 port,int index)647 static int hns_roce_get_hw_stats(struct ib_device *device,
648 struct rdma_hw_stats *stats,
649 u32 port, int index)
650 {
651 struct hns_roce_dev *hr_dev = to_hr_dev(device);
652 int num_counters = HNS_ROCE_HW_CNT_TOTAL;
653 int ret;
654
655 if (port == 0)
656 return 0;
657
658 if (port > hr_dev->caps.num_ports)
659 return -EINVAL;
660
661 ret = hr_dev->hw->query_hw_counter(hr_dev, stats->value, port,
662 &num_counters);
663 if (ret) {
664 ibdev_err(device, "failed to query hw counter, ret = %d\n",
665 ret);
666 return ret;
667 }
668
669 return num_counters;
670 }
671
672 static void
hns_roce_unregister_bond_cleanup(struct hns_roce_dev * hr_dev,struct hns_roce_bond_group * bond_grp)673 hns_roce_unregister_bond_cleanup(struct hns_roce_dev *hr_dev,
674 struct hns_roce_bond_group *bond_grp)
675 {
676 struct net_device *net_dev;
677 int i;
678
679 /* To avoid the loss of other slave devices when main_hr_dev
680 * is unregistered, re-initialize the remaining slaves before
681 * the bond resources cleanup.
682 */
683 bond_grp->bond_state = HNS_ROCE_BOND_NOT_BONDED;
684 for (i = 0; i < ROCE_BOND_FUNC_MAX; i++) {
685 net_dev = bond_grp->bond_func_info[i].net_dev;
686 if (net_dev && net_dev != get_hr_netdev(hr_dev, 0))
687 hns_roce_bond_init_client(bond_grp, i);
688 }
689
690 hns_roce_cleanup_bond(bond_grp);
691 }
692
hns_roce_unregister_device(struct hns_roce_dev * hr_dev,bool bond_cleanup)693 static void hns_roce_unregister_device(struct hns_roce_dev *hr_dev,
694 bool bond_cleanup)
695 {
696 struct net_device *net_dev = get_hr_netdev(hr_dev, 0);
697 struct hns_roce_ib_iboe *iboe = &hr_dev->iboe;
698 struct hns_roce_bond_group *bond_grp;
699 u8 bus_num = get_hr_bus_num(hr_dev);
700
701 if (bond_cleanup && hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_BOND) {
702 bond_grp = hns_roce_get_bond_grp(net_dev, bus_num);
703 if (bond_grp)
704 hns_roce_unregister_bond_cleanup(hr_dev, bond_grp);
705 }
706
707 hr_dev->active = false;
708 unregister_netdevice_notifier(&iboe->nb);
709 ib_unregister_device(&hr_dev->ib_dev);
710 }
711
712 static const struct ib_device_ops hns_roce_dev_ops = {
713 .owner = THIS_MODULE,
714 .driver_id = RDMA_DRIVER_HNS,
715 .uverbs_abi_ver = 1,
716 .uverbs_no_driver_id_binding = 1,
717
718 .get_dev_fw_str = hns_roce_get_fw_ver,
719 .add_gid = hns_roce_add_gid,
720 .alloc_pd = hns_roce_alloc_pd,
721 .alloc_ucontext = hns_roce_alloc_ucontext,
722 .create_ah = hns_roce_create_ah,
723 .create_user_ah = hns_roce_create_ah,
724 .create_cq = hns_roce_create_cq,
725 .create_qp = hns_roce_create_qp,
726 .dealloc_pd = hns_roce_dealloc_pd,
727 .dealloc_ucontext = hns_roce_dealloc_ucontext,
728 .del_gid = hns_roce_del_gid,
729 .dereg_mr = hns_roce_dereg_mr,
730 .destroy_ah = hns_roce_destroy_ah,
731 .destroy_cq = hns_roce_destroy_cq,
732 .disassociate_ucontext = hns_roce_disassociate_ucontext,
733 .get_dma_mr = hns_roce_get_dma_mr,
734 .get_link_layer = hns_roce_get_link_layer,
735 .get_port_immutable = hns_roce_port_immutable,
736 .mmap = hns_roce_mmap,
737 .mmap_free = hns_roce_free_mmap,
738 .modify_device = hns_roce_modify_device,
739 .modify_qp = hns_roce_modify_qp,
740 .query_ah = hns_roce_query_ah,
741 .query_device = hns_roce_query_device,
742 .query_pkey = hns_roce_query_pkey,
743 .query_port = hns_roce_query_port,
744 .reg_user_mr = hns_roce_reg_user_mr,
745
746 INIT_RDMA_OBJ_SIZE(ib_ah, hns_roce_ah, ibah),
747 INIT_RDMA_OBJ_SIZE(ib_cq, hns_roce_cq, ib_cq),
748 INIT_RDMA_OBJ_SIZE(ib_pd, hns_roce_pd, ibpd),
749 INIT_RDMA_OBJ_SIZE(ib_qp, hns_roce_qp, ibqp),
750 INIT_RDMA_OBJ_SIZE(ib_ucontext, hns_roce_ucontext, ibucontext),
751 };
752
753 static const struct ib_device_ops hns_roce_dev_hw_stats_ops = {
754 .alloc_hw_port_stats = hns_roce_alloc_hw_port_stats,
755 .get_hw_stats = hns_roce_get_hw_stats,
756 };
757
758 static const struct ib_device_ops hns_roce_dev_mr_ops = {
759 .rereg_user_mr = hns_roce_rereg_user_mr,
760 };
761
762 static const struct ib_device_ops hns_roce_dev_frmr_ops = {
763 .alloc_mr = hns_roce_alloc_mr,
764 .map_mr_sg = hns_roce_map_mr_sg,
765 };
766
767 static const struct ib_device_ops hns_roce_dev_srq_ops = {
768 .create_srq = hns_roce_create_srq,
769 .destroy_srq = hns_roce_destroy_srq,
770
771 INIT_RDMA_OBJ_SIZE(ib_srq, hns_roce_srq, ibsrq),
772 };
773
774 static const struct ib_device_ops hns_roce_dev_xrcd_ops = {
775 .alloc_xrcd = hns_roce_alloc_xrcd,
776 .dealloc_xrcd = hns_roce_dealloc_xrcd,
777
778 INIT_RDMA_OBJ_SIZE(ib_xrcd, hns_roce_xrcd, ibxrcd),
779 };
780
781 static const struct ib_device_ops hns_roce_dev_restrack_ops = {
782 .fill_res_cq_entry = hns_roce_fill_res_cq_entry,
783 .fill_res_cq_entry_raw = hns_roce_fill_res_cq_entry_raw,
784 .fill_res_qp_entry = hns_roce_fill_res_qp_entry,
785 .fill_res_qp_entry_raw = hns_roce_fill_res_qp_entry_raw,
786 .fill_res_mr_entry = hns_roce_fill_res_mr_entry,
787 .fill_res_mr_entry_raw = hns_roce_fill_res_mr_entry_raw,
788 .fill_res_srq_entry = hns_roce_fill_res_srq_entry,
789 .fill_res_srq_entry_raw = hns_roce_fill_res_srq_entry_raw,
790 };
791
hns_roce_register_device(struct hns_roce_dev * hr_dev)792 static int hns_roce_register_device(struct hns_roce_dev *hr_dev)
793 {
794 struct hns_roce_ib_iboe *iboe = NULL;
795 struct device *dev = hr_dev->dev;
796 struct ib_device *ib_dev = NULL;
797 struct net_device *net_dev;
798 unsigned int i;
799 int ret;
800
801 iboe = &hr_dev->iboe;
802 spin_lock_init(&iboe->lock);
803
804 ib_dev = &hr_dev->ib_dev;
805
806 ib_dev->node_type = RDMA_NODE_IB_CA;
807 ib_dev->dev.parent = dev;
808
809 ib_dev->phys_port_cnt = hr_dev->caps.num_ports;
810 ib_dev->local_dma_lkey = hr_dev->caps.reserved_lkey;
811 ib_dev->num_comp_vectors = hr_dev->caps.num_comp_vectors;
812
813 if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_REREG_MR)
814 ib_set_device_ops(ib_dev, &hns_roce_dev_mr_ops);
815
816 if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_FRMR)
817 ib_set_device_ops(ib_dev, &hns_roce_dev_frmr_ops);
818
819 if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ) {
820 ib_set_device_ops(ib_dev, &hns_roce_dev_srq_ops);
821 ib_set_device_ops(ib_dev, hr_dev->hw->hns_roce_dev_srq_ops);
822 }
823
824 if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_XRC)
825 ib_set_device_ops(ib_dev, &hns_roce_dev_xrcd_ops);
826
827 if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09 &&
828 !hr_dev->is_vf)
829 ib_set_device_ops(ib_dev, &hns_roce_dev_hw_stats_ops);
830
831 ib_set_device_ops(ib_dev, hr_dev->hw->hns_roce_dev_ops);
832 ib_set_device_ops(ib_dev, &hns_roce_dev_ops);
833 ib_set_device_ops(ib_dev, &hns_roce_dev_restrack_ops);
834
835 dma_set_max_seg_size(dev, SZ_2G);
836
837 if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_BOND) {
838 ret = hns_roce_alloc_bond_grp(hr_dev);
839 if (ret) {
840 dev_err(dev, "failed to alloc bond_grp for bus %u, ret = %d\n",
841 get_hr_bus_num(hr_dev), ret);
842 return ret;
843 }
844 }
845
846 if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_BOND &&
847 hns_roce_bond_is_active(hr_dev)) {
848 ret = hns_roce_bond_init(hr_dev);
849 if (ret) {
850 dev_err(dev, "failed to init bond!\n");
851 return ret;
852 }
853 ret = ib_register_device(ib_dev, "hns_bond_%d", dev);
854 } else {
855 for (i = 0; i < hr_dev->caps.num_ports; i++) {
856 net_dev = get_hr_netdev(hr_dev, i);
857 if (!net_dev)
858 continue;
859
860 ret = ib_device_set_netdev(ib_dev, net_dev, i + 1);
861 if (ret)
862 return ret;
863 }
864 ret = ib_register_device(ib_dev, "hns_%d", dev);
865 }
866 if (ret) {
867 dev_err(dev, "ib_register_device failed!\n");
868 return ret;
869 }
870
871 ret = hns_roce_setup_mtu_mac(hr_dev);
872 if (ret) {
873 dev_err(dev, "setup_mtu_mac failed!\n");
874 goto error_failed_setup_mtu_mac;
875 }
876
877 iboe->nb.notifier_call = hns_roce_netdev_event;
878 ret = register_netdevice_notifier(&iboe->nb);
879 if (ret) {
880 dev_err(dev, "register_netdevice_notifier failed!\n");
881 goto error_failed_setup_mtu_mac;
882 }
883
884 hr_dev->active = true;
885 return 0;
886
887 error_failed_setup_mtu_mac:
888 ib_unregister_device(ib_dev);
889
890 return ret;
891 }
892
hns_roce_init_hem(struct hns_roce_dev * hr_dev)893 static int hns_roce_init_hem(struct hns_roce_dev *hr_dev)
894 {
895 struct device *dev = hr_dev->dev;
896 int ret;
897
898 ret = hns_roce_init_hem_table(hr_dev, &hr_dev->mr_table.mtpt_table,
899 HEM_TYPE_MTPT, hr_dev->caps.mtpt_entry_sz,
900 hr_dev->caps.num_mtpts);
901 if (ret) {
902 dev_err(dev, "failed to init MTPT context memory, aborting.\n");
903 return ret;
904 }
905
906 ret = hns_roce_init_hem_table(hr_dev, &hr_dev->qp_table.qp_table,
907 HEM_TYPE_QPC, hr_dev->caps.qpc_sz,
908 hr_dev->caps.num_qps);
909 if (ret) {
910 dev_err(dev, "failed to init QP context memory, aborting.\n");
911 goto err_unmap_dmpt;
912 }
913
914 ret = hns_roce_init_hem_table(hr_dev, &hr_dev->qp_table.irrl_table,
915 HEM_TYPE_IRRL,
916 hr_dev->caps.irrl_entry_sz *
917 hr_dev->caps.max_qp_init_rdma,
918 hr_dev->caps.num_qps);
919 if (ret) {
920 dev_err(dev, "failed to init irrl_table memory, aborting.\n");
921 goto err_unmap_qp;
922 }
923
924 if (hr_dev->caps.trrl_entry_sz) {
925 ret = hns_roce_init_hem_table(hr_dev,
926 &hr_dev->qp_table.trrl_table,
927 HEM_TYPE_TRRL,
928 hr_dev->caps.trrl_entry_sz *
929 hr_dev->caps.max_qp_dest_rdma,
930 hr_dev->caps.num_qps);
931 if (ret) {
932 dev_err(dev,
933 "failed to init trrl_table memory, aborting.\n");
934 goto err_unmap_irrl;
935 }
936 }
937
938 ret = hns_roce_init_hem_table(hr_dev, &hr_dev->cq_table.table,
939 HEM_TYPE_CQC, hr_dev->caps.cqc_entry_sz,
940 hr_dev->caps.num_cqs);
941 if (ret) {
942 dev_err(dev, "failed to init CQ context memory, aborting.\n");
943 goto err_unmap_trrl;
944 }
945
946 if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ) {
947 ret = hns_roce_init_hem_table(hr_dev, &hr_dev->srq_table.table,
948 HEM_TYPE_SRQC,
949 hr_dev->caps.srqc_entry_sz,
950 hr_dev->caps.num_srqs);
951 if (ret) {
952 dev_err(dev,
953 "failed to init SRQ context memory, aborting.\n");
954 goto err_unmap_cq;
955 }
956 }
957
958 if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_FLOW_CTRL) {
959 ret = hns_roce_init_hem_table(hr_dev,
960 &hr_dev->qp_table.sccc_table,
961 HEM_TYPE_SCCC,
962 hr_dev->caps.sccc_sz,
963 hr_dev->caps.num_qps);
964 if (ret) {
965 dev_err(dev,
966 "failed to init SCC context memory, aborting.\n");
967 goto err_unmap_srq;
968 }
969 }
970
971 if (hr_dev->caps.qpc_timer_entry_sz) {
972 ret = hns_roce_init_hem_table(hr_dev, &hr_dev->qpc_timer_table,
973 HEM_TYPE_QPC_TIMER,
974 hr_dev->caps.qpc_timer_entry_sz,
975 hr_dev->caps.qpc_timer_bt_num);
976 if (ret) {
977 dev_err(dev,
978 "failed to init QPC timer memory, aborting.\n");
979 goto err_unmap_ctx;
980 }
981 }
982
983 if (hr_dev->caps.cqc_timer_entry_sz) {
984 ret = hns_roce_init_hem_table(hr_dev, &hr_dev->cqc_timer_table,
985 HEM_TYPE_CQC_TIMER,
986 hr_dev->caps.cqc_timer_entry_sz,
987 hr_dev->caps.cqc_timer_bt_num);
988 if (ret) {
989 dev_err(dev,
990 "failed to init CQC timer memory, aborting.\n");
991 goto err_unmap_qpc_timer;
992 }
993 }
994
995 if (hr_dev->caps.gmv_entry_sz) {
996 ret = hns_roce_init_hem_table(hr_dev, &hr_dev->gmv_table,
997 HEM_TYPE_GMV,
998 hr_dev->caps.gmv_entry_sz,
999 hr_dev->caps.gmv_entry_num);
1000 if (ret) {
1001 dev_err(dev,
1002 "failed to init gmv table memory, ret = %d\n",
1003 ret);
1004 goto err_unmap_cqc_timer;
1005 }
1006 }
1007
1008 return 0;
1009
1010 err_unmap_cqc_timer:
1011 if (hr_dev->caps.cqc_timer_entry_sz)
1012 hns_roce_cleanup_hem_table(hr_dev, &hr_dev->cqc_timer_table);
1013
1014 err_unmap_qpc_timer:
1015 if (hr_dev->caps.qpc_timer_entry_sz)
1016 hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qpc_timer_table);
1017
1018 err_unmap_ctx:
1019 if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_FLOW_CTRL)
1020 hns_roce_cleanup_hem_table(hr_dev,
1021 &hr_dev->qp_table.sccc_table);
1022 err_unmap_srq:
1023 if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ)
1024 hns_roce_cleanup_hem_table(hr_dev, &hr_dev->srq_table.table);
1025
1026 err_unmap_cq:
1027 hns_roce_cleanup_hem_table(hr_dev, &hr_dev->cq_table.table);
1028
1029 err_unmap_trrl:
1030 if (hr_dev->caps.trrl_entry_sz)
1031 hns_roce_cleanup_hem_table(hr_dev,
1032 &hr_dev->qp_table.trrl_table);
1033
1034 err_unmap_irrl:
1035 hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qp_table.irrl_table);
1036
1037 err_unmap_qp:
1038 hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qp_table.qp_table);
1039
1040 err_unmap_dmpt:
1041 hns_roce_cleanup_hem_table(hr_dev, &hr_dev->mr_table.mtpt_table);
1042
1043 return ret;
1044 }
1045
hns_roce_teardown_hca(struct hns_roce_dev * hr_dev)1046 static void hns_roce_teardown_hca(struct hns_roce_dev *hr_dev)
1047 {
1048 hns_roce_cleanup_bitmap(hr_dev);
1049 mutex_destroy(&hr_dev->pgdir_mutex);
1050 }
1051
1052 /**
1053 * hns_roce_setup_hca - setup host channel adapter
1054 * @hr_dev: pointer to hns roce device
1055 * Return : int
1056 */
hns_roce_setup_hca(struct hns_roce_dev * hr_dev)1057 static int hns_roce_setup_hca(struct hns_roce_dev *hr_dev)
1058 {
1059 struct device *dev = hr_dev->dev;
1060 int ret;
1061
1062 spin_lock_init(&hr_dev->sm_lock);
1063
1064 INIT_LIST_HEAD(&hr_dev->qp_list);
1065 spin_lock_init(&hr_dev->qp_list_lock);
1066
1067 INIT_LIST_HEAD(&hr_dev->pgdir_list);
1068 mutex_init(&hr_dev->pgdir_mutex);
1069
1070 hns_roce_init_uar_table(hr_dev);
1071
1072 ret = hns_roce_uar_alloc(hr_dev, &hr_dev->priv_uar);
1073 if (ret) {
1074 dev_err(dev, "failed to allocate priv_uar.\n");
1075 goto err_uar_table_free;
1076 }
1077
1078 ret = hns_roce_init_qp_table(hr_dev);
1079 if (ret) {
1080 dev_err(dev, "failed to init qp_table.\n");
1081 goto err_uar_table_free;
1082 }
1083
1084 hns_roce_init_pd_table(hr_dev);
1085
1086 if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_XRC)
1087 hns_roce_init_xrcd_table(hr_dev);
1088
1089 hns_roce_init_mr_table(hr_dev);
1090
1091 hns_roce_init_cq_table(hr_dev);
1092
1093 if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ)
1094 hns_roce_init_srq_table(hr_dev);
1095
1096 return 0;
1097
1098 err_uar_table_free:
1099 ida_destroy(&hr_dev->uar_ida.ida);
1100 mutex_destroy(&hr_dev->pgdir_mutex);
1101
1102 return ret;
1103 }
1104
check_and_get_armed_cq(struct list_head * cq_list,struct ib_cq * cq)1105 static void check_and_get_armed_cq(struct list_head *cq_list, struct ib_cq *cq)
1106 {
1107 struct hns_roce_cq *hr_cq = to_hr_cq(cq);
1108 unsigned long flags;
1109
1110 spin_lock_irqsave(&hr_cq->lock, flags);
1111 if (cq->comp_handler) {
1112 if (!hr_cq->is_armed) {
1113 hr_cq->is_armed = 1;
1114 list_add_tail(&hr_cq->node, cq_list);
1115 }
1116 }
1117 spin_unlock_irqrestore(&hr_cq->lock, flags);
1118 }
1119
hns_roce_handle_device_err(struct hns_roce_dev * hr_dev)1120 void hns_roce_handle_device_err(struct hns_roce_dev *hr_dev)
1121 {
1122 struct hns_roce_qp *hr_qp;
1123 struct hns_roce_cq *hr_cq;
1124 struct list_head cq_list;
1125 unsigned long flags_qp;
1126 unsigned long flags;
1127
1128 INIT_LIST_HEAD(&cq_list);
1129
1130 spin_lock_irqsave(&hr_dev->qp_list_lock, flags);
1131 list_for_each_entry(hr_qp, &hr_dev->qp_list, node) {
1132 spin_lock_irqsave(&hr_qp->sq.lock, flags_qp);
1133 if (hr_qp->sq.tail != hr_qp->sq.head)
1134 check_and_get_armed_cq(&cq_list, hr_qp->ibqp.send_cq);
1135 spin_unlock_irqrestore(&hr_qp->sq.lock, flags_qp);
1136
1137 spin_lock_irqsave(&hr_qp->rq.lock, flags_qp);
1138 if ((!hr_qp->ibqp.srq) && (hr_qp->rq.tail != hr_qp->rq.head))
1139 check_and_get_armed_cq(&cq_list, hr_qp->ibqp.recv_cq);
1140 spin_unlock_irqrestore(&hr_qp->rq.lock, flags_qp);
1141 }
1142
1143 list_for_each_entry(hr_cq, &cq_list, node)
1144 hns_roce_cq_completion(hr_dev, hr_cq->cqn);
1145
1146 spin_unlock_irqrestore(&hr_dev->qp_list_lock, flags);
1147 }
1148
hns_roce_alloc_dfx_cnt(struct hns_roce_dev * hr_dev)1149 static int hns_roce_alloc_dfx_cnt(struct hns_roce_dev *hr_dev)
1150 {
1151 hr_dev->dfx_cnt = kvcalloc(HNS_ROCE_DFX_CNT_TOTAL, sizeof(atomic64_t),
1152 GFP_KERNEL);
1153 if (!hr_dev->dfx_cnt)
1154 return -ENOMEM;
1155
1156 return 0;
1157 }
1158
hns_roce_dealloc_dfx_cnt(struct hns_roce_dev * hr_dev)1159 static void hns_roce_dealloc_dfx_cnt(struct hns_roce_dev *hr_dev)
1160 {
1161 kvfree(hr_dev->dfx_cnt);
1162 }
1163
hns_roce_init(struct hns_roce_dev * hr_dev)1164 int hns_roce_init(struct hns_roce_dev *hr_dev)
1165 {
1166 struct device *dev = hr_dev->dev;
1167 int ret;
1168
1169 hr_dev->is_reset = false;
1170
1171 ret = hns_roce_alloc_dfx_cnt(hr_dev);
1172 if (ret)
1173 return ret;
1174
1175 if (hr_dev->hw->cmq_init) {
1176 ret = hr_dev->hw->cmq_init(hr_dev);
1177 if (ret) {
1178 dev_err(dev, "init RoCE Command Queue failed!\n");
1179 goto error_failed_alloc_dfx_cnt;
1180 }
1181 }
1182
1183 ret = hr_dev->hw->hw_profile(hr_dev);
1184 if (ret) {
1185 dev_err(dev, "get RoCE engine profile failed!\n");
1186 goto error_failed_cmd_init;
1187 }
1188
1189 ret = hns_roce_cmd_init(hr_dev);
1190 if (ret) {
1191 dev_err(dev, "cmd init failed!\n");
1192 goto error_failed_cmd_init;
1193 }
1194
1195 /* EQ depends on poll mode, event mode depends on EQ */
1196 ret = hr_dev->hw->init_eq(hr_dev);
1197 if (ret) {
1198 dev_err(dev, "eq init failed!\n");
1199 goto error_failed_eq_table;
1200 }
1201
1202 if (hr_dev->cmd_mod) {
1203 ret = hns_roce_cmd_use_events(hr_dev);
1204 if (ret)
1205 dev_warn(dev,
1206 "Cmd event mode failed, set back to poll!\n");
1207 }
1208
1209 ret = hns_roce_init_hem(hr_dev);
1210 if (ret) {
1211 dev_err(dev, "init HEM(Hardware Entry Memory) failed!\n");
1212 goto error_failed_init_hem;
1213 }
1214
1215 ret = hns_roce_setup_hca(hr_dev);
1216 if (ret) {
1217 dev_err(dev, "setup hca failed!\n");
1218 goto error_failed_setup_hca;
1219 }
1220
1221 if (hr_dev->hw->hw_init) {
1222 ret = hr_dev->hw->hw_init(hr_dev);
1223 if (ret) {
1224 dev_err(dev, "hw_init failed!\n");
1225 goto error_failed_engine_init;
1226 }
1227 }
1228
1229 ret = hns_roce_register_device(hr_dev);
1230 if (ret)
1231 goto error_failed_register_device;
1232
1233 hns_roce_register_debugfs(hr_dev);
1234
1235 return 0;
1236
1237 error_failed_register_device:
1238 if (hr_dev->hw->hw_exit)
1239 hr_dev->hw->hw_exit(hr_dev);
1240
1241 error_failed_engine_init:
1242 hns_roce_teardown_hca(hr_dev);
1243
1244 error_failed_setup_hca:
1245 hns_roce_cleanup_hem(hr_dev);
1246
1247 error_failed_init_hem:
1248 if (hr_dev->cmd_mod)
1249 hns_roce_cmd_use_polling(hr_dev);
1250 hr_dev->hw->cleanup_eq(hr_dev);
1251
1252 error_failed_eq_table:
1253 hns_roce_cmd_cleanup(hr_dev);
1254
1255 error_failed_cmd_init:
1256 if (hr_dev->hw->cmq_exit)
1257 hr_dev->hw->cmq_exit(hr_dev);
1258
1259 error_failed_alloc_dfx_cnt:
1260 hns_roce_dealloc_dfx_cnt(hr_dev);
1261
1262 return ret;
1263 }
1264
hns_roce_exit(struct hns_roce_dev * hr_dev,bool bond_cleanup)1265 void hns_roce_exit(struct hns_roce_dev *hr_dev, bool bond_cleanup)
1266 {
1267 hns_roce_unregister_debugfs(hr_dev);
1268 hns_roce_unregister_device(hr_dev, bond_cleanup);
1269
1270 if (hr_dev->hw->hw_exit)
1271 hr_dev->hw->hw_exit(hr_dev);
1272 hns_roce_teardown_hca(hr_dev);
1273 hns_roce_cleanup_hem(hr_dev);
1274
1275 if (hr_dev->cmd_mod)
1276 hns_roce_cmd_use_polling(hr_dev);
1277
1278 hr_dev->hw->cleanup_eq(hr_dev);
1279 hns_roce_cmd_cleanup(hr_dev);
1280 if (hr_dev->hw->cmq_exit)
1281 hr_dev->hw->cmq_exit(hr_dev);
1282 hns_roce_dealloc_dfx_cnt(hr_dev);
1283 }
1284
1285 MODULE_LICENSE("Dual BSD/GPL");
1286 MODULE_AUTHOR("Wei Hu <xavier.huwei@huawei.com>");
1287 MODULE_AUTHOR("Nenglong Zhao <zhaonenglong@hisilicon.com>");
1288 MODULE_AUTHOR("Lijun Ou <oulijun@huawei.com>");
1289 MODULE_DESCRIPTION("HNS RoCE Driver");
1290