RDMA/hns: Fix init failure of RoCE VF and HIP08
[platform/kernel/linux-starfive.git] / drivers / infiniband / hw / hns / hns_roce_main.c
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 <linux/pci.h>
36 #include <rdma/ib_addr.h>
37 #include <rdma/ib_smi.h>
38 #include <rdma/ib_user_verbs.h>
39 #include <rdma/ib_cache.h>
40 #include "hns_roce_common.h"
41 #include "hns_roce_device.h"
42 #include "hns_roce_hem.h"
43
44 static int hns_roce_set_mac(struct hns_roce_dev *hr_dev, u32 port,
45                             const u8 *addr)
46 {
47         u8 phy_port;
48         u32 i;
49
50         if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09)
51                 return 0;
52
53         if (!memcmp(hr_dev->dev_addr[port], addr, ETH_ALEN))
54                 return 0;
55
56         for (i = 0; i < ETH_ALEN; i++)
57                 hr_dev->dev_addr[port][i] = addr[i];
58
59         phy_port = hr_dev->iboe.phy_port[port];
60         return hr_dev->hw->set_mac(hr_dev, phy_port, addr);
61 }
62
63 static int hns_roce_add_gid(const struct ib_gid_attr *attr, void **context)
64 {
65         struct hns_roce_dev *hr_dev = to_hr_dev(attr->device);
66         u32 port = attr->port_num - 1;
67         int ret;
68
69         if (port >= hr_dev->caps.num_ports)
70                 return -EINVAL;
71
72         ret = hr_dev->hw->set_gid(hr_dev, attr->index, &attr->gid, attr);
73
74         return ret;
75 }
76
77 static int hns_roce_del_gid(const struct ib_gid_attr *attr, void **context)
78 {
79         struct hns_roce_dev *hr_dev = to_hr_dev(attr->device);
80         u32 port = attr->port_num - 1;
81         int ret;
82
83         if (port >= hr_dev->caps.num_ports)
84                 return -EINVAL;
85
86         ret = hr_dev->hw->set_gid(hr_dev, attr->index, NULL, NULL);
87
88         return ret;
89 }
90
91 static int handle_en_event(struct hns_roce_dev *hr_dev, u32 port,
92                            unsigned long event)
93 {
94         struct device *dev = hr_dev->dev;
95         struct net_device *netdev;
96         int ret = 0;
97
98         netdev = hr_dev->iboe.netdevs[port];
99         if (!netdev) {
100                 dev_err(dev, "can't find netdev on port(%u)!\n", port);
101                 return -ENODEV;
102         }
103
104         switch (event) {
105         case NETDEV_UP:
106         case NETDEV_CHANGE:
107         case NETDEV_REGISTER:
108         case NETDEV_CHANGEADDR:
109                 ret = hns_roce_set_mac(hr_dev, port, netdev->dev_addr);
110                 break;
111         case NETDEV_DOWN:
112                 /*
113                  * In v1 engine, only support all ports closed together.
114                  */
115                 break;
116         default:
117                 dev_dbg(dev, "NETDEV event = 0x%x!\n", (u32)(event));
118                 break;
119         }
120
121         return ret;
122 }
123
124 static int hns_roce_netdev_event(struct notifier_block *self,
125                                  unsigned long event, void *ptr)
126 {
127         struct net_device *dev = netdev_notifier_info_to_dev(ptr);
128         struct hns_roce_ib_iboe *iboe = NULL;
129         struct hns_roce_dev *hr_dev = NULL;
130         int ret;
131         u32 port;
132
133         hr_dev = container_of(self, struct hns_roce_dev, iboe.nb);
134         iboe = &hr_dev->iboe;
135
136         for (port = 0; port < hr_dev->caps.num_ports; port++) {
137                 if (dev == iboe->netdevs[port]) {
138                         ret = handle_en_event(hr_dev, port, event);
139                         if (ret)
140                                 return NOTIFY_DONE;
141                         break;
142                 }
143         }
144
145         return NOTIFY_DONE;
146 }
147
148 static int hns_roce_setup_mtu_mac(struct hns_roce_dev *hr_dev)
149 {
150         int ret;
151         u8 i;
152
153         for (i = 0; i < hr_dev->caps.num_ports; i++) {
154                 ret = hns_roce_set_mac(hr_dev, i,
155                                        hr_dev->iboe.netdevs[i]->dev_addr);
156                 if (ret)
157                         return ret;
158         }
159
160         return 0;
161 }
162
163 static int hns_roce_query_device(struct ib_device *ib_dev,
164                                  struct ib_device_attr *props,
165                                  struct ib_udata *uhw)
166 {
167         struct hns_roce_dev *hr_dev = to_hr_dev(ib_dev);
168
169         memset(props, 0, sizeof(*props));
170
171         props->fw_ver = hr_dev->caps.fw_ver;
172         props->sys_image_guid = cpu_to_be64(hr_dev->sys_image_guid);
173         props->max_mr_size = (u64)(~(0ULL));
174         props->page_size_cap = hr_dev->caps.page_size_cap;
175         props->vendor_id = hr_dev->vendor_id;
176         props->vendor_part_id = hr_dev->vendor_part_id;
177         props->hw_ver = hr_dev->hw_rev;
178         props->max_qp = hr_dev->caps.num_qps;
179         props->max_qp_wr = hr_dev->caps.max_wqes;
180         props->device_cap_flags = IB_DEVICE_PORT_ACTIVE_EVENT |
181                                   IB_DEVICE_RC_RNR_NAK_GEN;
182         props->max_send_sge = hr_dev->caps.max_sq_sg;
183         props->max_recv_sge = hr_dev->caps.max_rq_sg;
184         props->max_sge_rd = 1;
185         props->max_cq = hr_dev->caps.num_cqs;
186         props->max_cqe = hr_dev->caps.max_cqes;
187         props->max_mr = hr_dev->caps.num_mtpts;
188         props->max_pd = hr_dev->caps.num_pds;
189         props->max_qp_rd_atom = hr_dev->caps.max_qp_dest_rdma;
190         props->max_qp_init_rd_atom = hr_dev->caps.max_qp_init_rdma;
191         props->atomic_cap = hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_ATOMIC ?
192                             IB_ATOMIC_HCA : IB_ATOMIC_NONE;
193         props->max_pkeys = 1;
194         props->local_ca_ack_delay = hr_dev->caps.local_ca_ack_delay;
195         if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ) {
196                 props->max_srq = hr_dev->caps.num_srqs;
197                 props->max_srq_wr = hr_dev->caps.max_srq_wrs;
198                 props->max_srq_sge = hr_dev->caps.max_srq_sges;
199         }
200
201         if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_FRMR &&
202             hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09) {
203                 props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
204                 props->max_fast_reg_page_list_len = HNS_ROCE_FRMR_MAX_PA;
205         }
206
207         if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_XRC)
208                 props->device_cap_flags |= IB_DEVICE_XRC;
209
210         return 0;
211 }
212
213 static int hns_roce_query_port(struct ib_device *ib_dev, u32 port_num,
214                                struct ib_port_attr *props)
215 {
216         struct hns_roce_dev *hr_dev = to_hr_dev(ib_dev);
217         struct device *dev = hr_dev->dev;
218         struct net_device *net_dev;
219         unsigned long flags;
220         enum ib_mtu mtu;
221         u32 port;
222         int ret;
223
224         port = port_num - 1;
225
226         /* props being zeroed by the caller, avoid zeroing it here */
227
228         props->max_mtu = hr_dev->caps.max_mtu;
229         props->gid_tbl_len = hr_dev->caps.gid_table_len[port];
230         props->port_cap_flags = IB_PORT_CM_SUP | IB_PORT_REINIT_SUP |
231                                 IB_PORT_VENDOR_CLASS_SUP |
232                                 IB_PORT_BOOT_MGMT_SUP;
233         props->max_msg_sz = HNS_ROCE_MAX_MSG_LEN;
234         props->pkey_tbl_len = 1;
235         ret = ib_get_eth_speed(ib_dev, port_num, &props->active_speed,
236                                &props->active_width);
237         if (ret)
238                 ibdev_warn(ib_dev, "failed to get speed, ret = %d.\n", ret);
239
240         spin_lock_irqsave(&hr_dev->iboe.lock, flags);
241
242         net_dev = hr_dev->iboe.netdevs[port];
243         if (!net_dev) {
244                 spin_unlock_irqrestore(&hr_dev->iboe.lock, flags);
245                 dev_err(dev, "find netdev %u failed!\n", port);
246                 return -EINVAL;
247         }
248
249         mtu = iboe_get_mtu(net_dev->mtu);
250         props->active_mtu = mtu ? min(props->max_mtu, mtu) : IB_MTU_256;
251         props->state = netif_running(net_dev) && netif_carrier_ok(net_dev) ?
252                                IB_PORT_ACTIVE :
253                                IB_PORT_DOWN;
254         props->phys_state = props->state == IB_PORT_ACTIVE ?
255                                     IB_PORT_PHYS_STATE_LINK_UP :
256                                     IB_PORT_PHYS_STATE_DISABLED;
257
258         spin_unlock_irqrestore(&hr_dev->iboe.lock, flags);
259
260         return 0;
261 }
262
263 static enum rdma_link_layer hns_roce_get_link_layer(struct ib_device *device,
264                                                     u32 port_num)
265 {
266         return IB_LINK_LAYER_ETHERNET;
267 }
268
269 static int hns_roce_query_pkey(struct ib_device *ib_dev, u32 port, u16 index,
270                                u16 *pkey)
271 {
272         if (index > 0)
273                 return -EINVAL;
274
275         *pkey = PKEY_ID;
276
277         return 0;
278 }
279
280 static int hns_roce_modify_device(struct ib_device *ib_dev, int mask,
281                                   struct ib_device_modify *props)
282 {
283         unsigned long flags;
284
285         if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
286                 return -EOPNOTSUPP;
287
288         if (mask & IB_DEVICE_MODIFY_NODE_DESC) {
289                 spin_lock_irqsave(&to_hr_dev(ib_dev)->sm_lock, flags);
290                 memcpy(ib_dev->node_desc, props->node_desc, NODE_DESC_SIZE);
291                 spin_unlock_irqrestore(&to_hr_dev(ib_dev)->sm_lock, flags);
292         }
293
294         return 0;
295 }
296
297 struct hns_user_mmap_entry *
298 hns_roce_user_mmap_entry_insert(struct ib_ucontext *ucontext, u64 address,
299                                 size_t length,
300                                 enum hns_roce_mmap_type mmap_type)
301 {
302         struct hns_user_mmap_entry *entry;
303         int ret;
304
305         entry = kzalloc(sizeof(*entry), GFP_KERNEL);
306         if (!entry)
307                 return NULL;
308
309         entry->address = address;
310         entry->mmap_type = mmap_type;
311
312         switch (mmap_type) {
313         /* pgoff 0 must be used by DB for compatibility */
314         case HNS_ROCE_MMAP_TYPE_DB:
315                 ret = rdma_user_mmap_entry_insert_exact(
316                                 ucontext, &entry->rdma_entry, length, 0);
317                 break;
318         case HNS_ROCE_MMAP_TYPE_DWQE:
319                 ret = rdma_user_mmap_entry_insert_range(
320                                 ucontext, &entry->rdma_entry, length, 1,
321                                 U32_MAX);
322                 break;
323         default:
324                 ret = -EINVAL;
325                 break;
326         }
327
328         if (ret) {
329                 kfree(entry);
330                 return NULL;
331         }
332
333         return entry;
334 }
335
336 static void hns_roce_dealloc_uar_entry(struct hns_roce_ucontext *context)
337 {
338         if (context->db_mmap_entry)
339                 rdma_user_mmap_entry_remove(
340                         &context->db_mmap_entry->rdma_entry);
341 }
342
343 static int hns_roce_alloc_uar_entry(struct ib_ucontext *uctx)
344 {
345         struct hns_roce_ucontext *context = to_hr_ucontext(uctx);
346         u64 address;
347
348         address = context->uar.pfn << PAGE_SHIFT;
349         context->db_mmap_entry = hns_roce_user_mmap_entry_insert(
350                 uctx, address, PAGE_SIZE, HNS_ROCE_MMAP_TYPE_DB);
351         if (!context->db_mmap_entry)
352                 return -ENOMEM;
353
354         return 0;
355 }
356
357 static int hns_roce_alloc_ucontext(struct ib_ucontext *uctx,
358                                    struct ib_udata *udata)
359 {
360         struct hns_roce_ucontext *context = to_hr_ucontext(uctx);
361         struct hns_roce_dev *hr_dev = to_hr_dev(uctx->device);
362         struct hns_roce_ib_alloc_ucontext_resp resp = {};
363         struct hns_roce_ib_alloc_ucontext ucmd = {};
364         int ret;
365
366         if (!hr_dev->active)
367                 return -EAGAIN;
368
369         resp.qp_tab_size = hr_dev->caps.num_qps;
370         resp.srq_tab_size = hr_dev->caps.num_srqs;
371
372         ret = ib_copy_from_udata(&ucmd, udata,
373                                  min(udata->inlen, sizeof(ucmd)));
374         if (ret)
375                 return ret;
376
377         if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09)
378                 context->config = ucmd.config & HNS_ROCE_EXSGE_FLAGS;
379
380         if (context->config & HNS_ROCE_EXSGE_FLAGS) {
381                 resp.config |= HNS_ROCE_RSP_EXSGE_FLAGS;
382                 resp.max_inline_data = hr_dev->caps.max_sq_inline;
383         }
384
385         if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_RQ_INLINE) {
386                 context->config |= ucmd.config & HNS_ROCE_RQ_INLINE_FLAGS;
387                 if (context->config & HNS_ROCE_RQ_INLINE_FLAGS)
388                         resp.config |= HNS_ROCE_RSP_RQ_INLINE_FLAGS;
389         }
390
391         if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_CQE_INLINE) {
392                 context->config |= ucmd.config & HNS_ROCE_CQE_INLINE_FLAGS;
393                 if (context->config & HNS_ROCE_CQE_INLINE_FLAGS)
394                         resp.config |= HNS_ROCE_RSP_CQE_INLINE_FLAGS;
395         }
396
397         ret = hns_roce_uar_alloc(hr_dev, &context->uar);
398         if (ret)
399                 goto error_fail_uar_alloc;
400
401         ret = hns_roce_alloc_uar_entry(uctx);
402         if (ret)
403                 goto error_fail_uar_entry;
404
405         if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_CQ_RECORD_DB ||
406             hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_RECORD_DB) {
407                 INIT_LIST_HEAD(&context->page_list);
408                 mutex_init(&context->page_mutex);
409         }
410
411         resp.cqe_size = hr_dev->caps.cqe_sz;
412
413         ret = ib_copy_to_udata(udata, &resp,
414                                min(udata->outlen, sizeof(resp)));
415         if (ret)
416                 goto error_fail_copy_to_udata;
417
418         return 0;
419
420 error_fail_copy_to_udata:
421         hns_roce_dealloc_uar_entry(context);
422
423 error_fail_uar_entry:
424         ida_free(&hr_dev->uar_ida.ida, (int)context->uar.logic_idx);
425
426 error_fail_uar_alloc:
427         return ret;
428 }
429
430 static void hns_roce_dealloc_ucontext(struct ib_ucontext *ibcontext)
431 {
432         struct hns_roce_ucontext *context = to_hr_ucontext(ibcontext);
433         struct hns_roce_dev *hr_dev = to_hr_dev(ibcontext->device);
434
435         hns_roce_dealloc_uar_entry(context);
436
437         ida_free(&hr_dev->uar_ida.ida, (int)context->uar.logic_idx);
438 }
439
440 static int hns_roce_mmap(struct ib_ucontext *uctx, struct vm_area_struct *vma)
441 {
442         struct rdma_user_mmap_entry *rdma_entry;
443         struct hns_user_mmap_entry *entry;
444         phys_addr_t pfn;
445         pgprot_t prot;
446         int ret;
447
448         rdma_entry = rdma_user_mmap_entry_get_pgoff(uctx, vma->vm_pgoff);
449         if (!rdma_entry)
450                 return -EINVAL;
451
452         entry = to_hns_mmap(rdma_entry);
453         pfn = entry->address >> PAGE_SHIFT;
454
455         switch (entry->mmap_type) {
456         case HNS_ROCE_MMAP_TYPE_DB:
457         case HNS_ROCE_MMAP_TYPE_DWQE:
458                 prot = pgprot_device(vma->vm_page_prot);
459                 break;
460         default:
461                 ret = -EINVAL;
462                 goto out;
463         }
464
465         ret = rdma_user_mmap_io(uctx, vma, pfn, rdma_entry->npages * PAGE_SIZE,
466                                 prot, rdma_entry);
467
468 out:
469         rdma_user_mmap_entry_put(rdma_entry);
470         return ret;
471 }
472
473 static void hns_roce_free_mmap(struct rdma_user_mmap_entry *rdma_entry)
474 {
475         struct hns_user_mmap_entry *entry = to_hns_mmap(rdma_entry);
476
477         kfree(entry);
478 }
479
480 static int hns_roce_port_immutable(struct ib_device *ib_dev, u32 port_num,
481                                    struct ib_port_immutable *immutable)
482 {
483         struct ib_port_attr attr;
484         int ret;
485
486         ret = ib_query_port(ib_dev, port_num, &attr);
487         if (ret)
488                 return ret;
489
490         immutable->pkey_tbl_len = attr.pkey_tbl_len;
491         immutable->gid_tbl_len = attr.gid_tbl_len;
492
493         immutable->max_mad_size = IB_MGMT_MAD_SIZE;
494         immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
495         if (to_hr_dev(ib_dev)->caps.flags & HNS_ROCE_CAP_FLAG_ROCE_V1_V2)
496                 immutable->core_cap_flags |= RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
497
498         return 0;
499 }
500
501 static void hns_roce_disassociate_ucontext(struct ib_ucontext *ibcontext)
502 {
503 }
504
505 static void hns_roce_get_fw_ver(struct ib_device *device, char *str)
506 {
507         u64 fw_ver = to_hr_dev(device)->caps.fw_ver;
508         unsigned int major, minor, sub_minor;
509
510         major = upper_32_bits(fw_ver);
511         minor = high_16_bits(lower_32_bits(fw_ver));
512         sub_minor = low_16_bits(fw_ver);
513
514         snprintf(str, IB_FW_VERSION_NAME_MAX, "%u.%u.%04u", major, minor,
515                  sub_minor);
516 }
517
518 #define HNS_ROCE_HW_CNT(ename, cname) \
519         [HNS_ROCE_HW_##ename##_CNT].name = cname
520
521 static const struct rdma_stat_desc hns_roce_port_stats_descs[] = {
522         HNS_ROCE_HW_CNT(RX_RC_PKT, "rx_rc_pkt"),
523         HNS_ROCE_HW_CNT(RX_UC_PKT, "rx_uc_pkt"),
524         HNS_ROCE_HW_CNT(RX_UD_PKT, "rx_ud_pkt"),
525         HNS_ROCE_HW_CNT(RX_XRC_PKT, "rx_xrc_pkt"),
526         HNS_ROCE_HW_CNT(RX_PKT, "rx_pkt"),
527         HNS_ROCE_HW_CNT(RX_ERR_PKT, "rx_err_pkt"),
528         HNS_ROCE_HW_CNT(RX_CNP_PKT, "rx_cnp_pkt"),
529         HNS_ROCE_HW_CNT(TX_RC_PKT, "tx_rc_pkt"),
530         HNS_ROCE_HW_CNT(TX_UC_PKT, "tx_uc_pkt"),
531         HNS_ROCE_HW_CNT(TX_UD_PKT, "tx_ud_pkt"),
532         HNS_ROCE_HW_CNT(TX_XRC_PKT, "tx_xrc_pkt"),
533         HNS_ROCE_HW_CNT(TX_PKT, "tx_pkt"),
534         HNS_ROCE_HW_CNT(TX_ERR_PKT, "tx_err_pkt"),
535         HNS_ROCE_HW_CNT(TX_CNP_PKT, "tx_cnp_pkt"),
536         HNS_ROCE_HW_CNT(TRP_GET_MPT_ERR_PKT, "trp_get_mpt_err_pkt"),
537         HNS_ROCE_HW_CNT(TRP_GET_IRRL_ERR_PKT, "trp_get_irrl_err_pkt"),
538         HNS_ROCE_HW_CNT(ECN_DB, "ecn_doorbell"),
539         HNS_ROCE_HW_CNT(RX_BUF, "rx_buffer"),
540         HNS_ROCE_HW_CNT(TRP_RX_SOF, "trp_rx_sof"),
541         HNS_ROCE_HW_CNT(CQ_CQE, "cq_cqe"),
542         HNS_ROCE_HW_CNT(CQ_POE, "cq_poe"),
543         HNS_ROCE_HW_CNT(CQ_NOTIFY, "cq_notify"),
544 };
545
546 static struct rdma_hw_stats *hns_roce_alloc_hw_port_stats(
547                                 struct ib_device *device, u32 port_num)
548 {
549         struct hns_roce_dev *hr_dev = to_hr_dev(device);
550
551         if (port_num > hr_dev->caps.num_ports) {
552                 ibdev_err(device, "invalid port num.\n");
553                 return NULL;
554         }
555
556         return rdma_alloc_hw_stats_struct(hns_roce_port_stats_descs,
557                                           ARRAY_SIZE(hns_roce_port_stats_descs),
558                                           RDMA_HW_STATS_DEFAULT_LIFESPAN);
559 }
560
561 static int hns_roce_get_hw_stats(struct ib_device *device,
562                                  struct rdma_hw_stats *stats,
563                                  u32 port, int index)
564 {
565         struct hns_roce_dev *hr_dev = to_hr_dev(device);
566         int num_counters = HNS_ROCE_HW_CNT_TOTAL;
567         int ret;
568
569         if (port == 0)
570                 return 0;
571
572         if (port > hr_dev->caps.num_ports)
573                 return -EINVAL;
574
575         ret = hr_dev->hw->query_hw_counter(hr_dev, stats->value, port,
576                                            &num_counters);
577         if (ret) {
578                 ibdev_err(device, "failed to query hw counter, ret = %d\n",
579                           ret);
580                 return ret;
581         }
582
583         return num_counters;
584 }
585
586 static void hns_roce_unregister_device(struct hns_roce_dev *hr_dev)
587 {
588         struct hns_roce_ib_iboe *iboe = &hr_dev->iboe;
589
590         hr_dev->active = false;
591         unregister_netdevice_notifier(&iboe->nb);
592         ib_unregister_device(&hr_dev->ib_dev);
593 }
594
595 static const struct ib_device_ops hns_roce_dev_ops = {
596         .owner = THIS_MODULE,
597         .driver_id = RDMA_DRIVER_HNS,
598         .uverbs_abi_ver = 1,
599         .uverbs_no_driver_id_binding = 1,
600
601         .get_dev_fw_str = hns_roce_get_fw_ver,
602         .add_gid = hns_roce_add_gid,
603         .alloc_pd = hns_roce_alloc_pd,
604         .alloc_ucontext = hns_roce_alloc_ucontext,
605         .create_ah = hns_roce_create_ah,
606         .create_user_ah = hns_roce_create_ah,
607         .create_cq = hns_roce_create_cq,
608         .create_qp = hns_roce_create_qp,
609         .dealloc_pd = hns_roce_dealloc_pd,
610         .dealloc_ucontext = hns_roce_dealloc_ucontext,
611         .del_gid = hns_roce_del_gid,
612         .dereg_mr = hns_roce_dereg_mr,
613         .destroy_ah = hns_roce_destroy_ah,
614         .destroy_cq = hns_roce_destroy_cq,
615         .disassociate_ucontext = hns_roce_disassociate_ucontext,
616         .get_dma_mr = hns_roce_get_dma_mr,
617         .get_link_layer = hns_roce_get_link_layer,
618         .get_port_immutable = hns_roce_port_immutable,
619         .mmap = hns_roce_mmap,
620         .mmap_free = hns_roce_free_mmap,
621         .modify_device = hns_roce_modify_device,
622         .modify_qp = hns_roce_modify_qp,
623         .query_ah = hns_roce_query_ah,
624         .query_device = hns_roce_query_device,
625         .query_pkey = hns_roce_query_pkey,
626         .query_port = hns_roce_query_port,
627         .reg_user_mr = hns_roce_reg_user_mr,
628
629         INIT_RDMA_OBJ_SIZE(ib_ah, hns_roce_ah, ibah),
630         INIT_RDMA_OBJ_SIZE(ib_cq, hns_roce_cq, ib_cq),
631         INIT_RDMA_OBJ_SIZE(ib_pd, hns_roce_pd, ibpd),
632         INIT_RDMA_OBJ_SIZE(ib_qp, hns_roce_qp, ibqp),
633         INIT_RDMA_OBJ_SIZE(ib_ucontext, hns_roce_ucontext, ibucontext),
634 };
635
636 static const struct ib_device_ops hns_roce_dev_hw_stats_ops = {
637         .alloc_hw_port_stats = hns_roce_alloc_hw_port_stats,
638         .get_hw_stats = hns_roce_get_hw_stats,
639 };
640
641 static const struct ib_device_ops hns_roce_dev_mr_ops = {
642         .rereg_user_mr = hns_roce_rereg_user_mr,
643 };
644
645 static const struct ib_device_ops hns_roce_dev_mw_ops = {
646         .alloc_mw = hns_roce_alloc_mw,
647         .dealloc_mw = hns_roce_dealloc_mw,
648
649         INIT_RDMA_OBJ_SIZE(ib_mw, hns_roce_mw, ibmw),
650 };
651
652 static const struct ib_device_ops hns_roce_dev_frmr_ops = {
653         .alloc_mr = hns_roce_alloc_mr,
654         .map_mr_sg = hns_roce_map_mr_sg,
655 };
656
657 static const struct ib_device_ops hns_roce_dev_srq_ops = {
658         .create_srq = hns_roce_create_srq,
659         .destroy_srq = hns_roce_destroy_srq,
660
661         INIT_RDMA_OBJ_SIZE(ib_srq, hns_roce_srq, ibsrq),
662 };
663
664 static const struct ib_device_ops hns_roce_dev_xrcd_ops = {
665         .alloc_xrcd = hns_roce_alloc_xrcd,
666         .dealloc_xrcd = hns_roce_dealloc_xrcd,
667
668         INIT_RDMA_OBJ_SIZE(ib_xrcd, hns_roce_xrcd, ibxrcd),
669 };
670
671 static const struct ib_device_ops hns_roce_dev_restrack_ops = {
672         .fill_res_cq_entry = hns_roce_fill_res_cq_entry,
673         .fill_res_cq_entry_raw = hns_roce_fill_res_cq_entry_raw,
674         .fill_res_qp_entry = hns_roce_fill_res_qp_entry,
675         .fill_res_qp_entry_raw = hns_roce_fill_res_qp_entry_raw,
676         .fill_res_mr_entry = hns_roce_fill_res_mr_entry,
677         .fill_res_mr_entry_raw = hns_roce_fill_res_mr_entry_raw,
678 };
679
680 static int hns_roce_register_device(struct hns_roce_dev *hr_dev)
681 {
682         int ret;
683         struct hns_roce_ib_iboe *iboe = NULL;
684         struct ib_device *ib_dev = NULL;
685         struct device *dev = hr_dev->dev;
686         unsigned int i;
687
688         iboe = &hr_dev->iboe;
689         spin_lock_init(&iboe->lock);
690
691         ib_dev = &hr_dev->ib_dev;
692
693         ib_dev->node_type = RDMA_NODE_IB_CA;
694         ib_dev->dev.parent = dev;
695
696         ib_dev->phys_port_cnt = hr_dev->caps.num_ports;
697         ib_dev->local_dma_lkey = hr_dev->caps.reserved_lkey;
698         ib_dev->num_comp_vectors = hr_dev->caps.num_comp_vectors;
699
700         if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_REREG_MR)
701                 ib_set_device_ops(ib_dev, &hns_roce_dev_mr_ops);
702
703         if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_MW)
704                 ib_set_device_ops(ib_dev, &hns_roce_dev_mw_ops);
705
706         if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_FRMR)
707                 ib_set_device_ops(ib_dev, &hns_roce_dev_frmr_ops);
708
709         if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ) {
710                 ib_set_device_ops(ib_dev, &hns_roce_dev_srq_ops);
711                 ib_set_device_ops(ib_dev, hr_dev->hw->hns_roce_dev_srq_ops);
712         }
713
714         if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_XRC)
715                 ib_set_device_ops(ib_dev, &hns_roce_dev_xrcd_ops);
716
717         if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09 &&
718             !hr_dev->is_vf)
719                 ib_set_device_ops(ib_dev, &hns_roce_dev_hw_stats_ops);
720
721         ib_set_device_ops(ib_dev, hr_dev->hw->hns_roce_dev_ops);
722         ib_set_device_ops(ib_dev, &hns_roce_dev_ops);
723         ib_set_device_ops(ib_dev, &hns_roce_dev_restrack_ops);
724         for (i = 0; i < hr_dev->caps.num_ports; i++) {
725                 if (!hr_dev->iboe.netdevs[i])
726                         continue;
727
728                 ret = ib_device_set_netdev(ib_dev, hr_dev->iboe.netdevs[i],
729                                            i + 1);
730                 if (ret)
731                         return ret;
732         }
733         dma_set_max_seg_size(dev, UINT_MAX);
734         ret = ib_register_device(ib_dev, "hns_%d", dev);
735         if (ret) {
736                 dev_err(dev, "ib_register_device failed!\n");
737                 return ret;
738         }
739
740         ret = hns_roce_setup_mtu_mac(hr_dev);
741         if (ret) {
742                 dev_err(dev, "setup_mtu_mac failed!\n");
743                 goto error_failed_setup_mtu_mac;
744         }
745
746         iboe->nb.notifier_call = hns_roce_netdev_event;
747         ret = register_netdevice_notifier(&iboe->nb);
748         if (ret) {
749                 dev_err(dev, "register_netdevice_notifier failed!\n");
750                 goto error_failed_setup_mtu_mac;
751         }
752
753         hr_dev->active = true;
754         return 0;
755
756 error_failed_setup_mtu_mac:
757         ib_unregister_device(ib_dev);
758
759         return ret;
760 }
761
762 static int hns_roce_init_hem(struct hns_roce_dev *hr_dev)
763 {
764         struct device *dev = hr_dev->dev;
765         int ret;
766
767         ret = hns_roce_init_hem_table(hr_dev, &hr_dev->mr_table.mtpt_table,
768                                       HEM_TYPE_MTPT, hr_dev->caps.mtpt_entry_sz,
769                                       hr_dev->caps.num_mtpts);
770         if (ret) {
771                 dev_err(dev, "failed to init MTPT context memory, aborting.\n");
772                 return ret;
773         }
774
775         ret = hns_roce_init_hem_table(hr_dev, &hr_dev->qp_table.qp_table,
776                                       HEM_TYPE_QPC, hr_dev->caps.qpc_sz,
777                                       hr_dev->caps.num_qps);
778         if (ret) {
779                 dev_err(dev, "failed to init QP context memory, aborting.\n");
780                 goto err_unmap_dmpt;
781         }
782
783         ret = hns_roce_init_hem_table(hr_dev, &hr_dev->qp_table.irrl_table,
784                                       HEM_TYPE_IRRL,
785                                       hr_dev->caps.irrl_entry_sz *
786                                       hr_dev->caps.max_qp_init_rdma,
787                                       hr_dev->caps.num_qps);
788         if (ret) {
789                 dev_err(dev, "failed to init irrl_table memory, aborting.\n");
790                 goto err_unmap_qp;
791         }
792
793         if (hr_dev->caps.trrl_entry_sz) {
794                 ret = hns_roce_init_hem_table(hr_dev,
795                                               &hr_dev->qp_table.trrl_table,
796                                               HEM_TYPE_TRRL,
797                                               hr_dev->caps.trrl_entry_sz *
798                                               hr_dev->caps.max_qp_dest_rdma,
799                                               hr_dev->caps.num_qps);
800                 if (ret) {
801                         dev_err(dev,
802                                 "failed to init trrl_table memory, aborting.\n");
803                         goto err_unmap_irrl;
804                 }
805         }
806
807         ret = hns_roce_init_hem_table(hr_dev, &hr_dev->cq_table.table,
808                                       HEM_TYPE_CQC, hr_dev->caps.cqc_entry_sz,
809                                       hr_dev->caps.num_cqs);
810         if (ret) {
811                 dev_err(dev, "failed to init CQ context memory, aborting.\n");
812                 goto err_unmap_trrl;
813         }
814
815         if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ) {
816                 ret = hns_roce_init_hem_table(hr_dev, &hr_dev->srq_table.table,
817                                               HEM_TYPE_SRQC,
818                                               hr_dev->caps.srqc_entry_sz,
819                                               hr_dev->caps.num_srqs);
820                 if (ret) {
821                         dev_err(dev,
822                                 "failed to init SRQ context memory, aborting.\n");
823                         goto err_unmap_cq;
824                 }
825         }
826
827         if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_FLOW_CTRL) {
828                 ret = hns_roce_init_hem_table(hr_dev,
829                                               &hr_dev->qp_table.sccc_table,
830                                               HEM_TYPE_SCCC,
831                                               hr_dev->caps.sccc_sz,
832                                               hr_dev->caps.num_qps);
833                 if (ret) {
834                         dev_err(dev,
835                                 "failed to init SCC context memory, aborting.\n");
836                         goto err_unmap_srq;
837                 }
838         }
839
840         if (hr_dev->caps.qpc_timer_entry_sz) {
841                 ret = hns_roce_init_hem_table(hr_dev, &hr_dev->qpc_timer_table,
842                                               HEM_TYPE_QPC_TIMER,
843                                               hr_dev->caps.qpc_timer_entry_sz,
844                                               hr_dev->caps.qpc_timer_bt_num);
845                 if (ret) {
846                         dev_err(dev,
847                                 "failed to init QPC timer memory, aborting.\n");
848                         goto err_unmap_ctx;
849                 }
850         }
851
852         if (hr_dev->caps.cqc_timer_entry_sz) {
853                 ret = hns_roce_init_hem_table(hr_dev, &hr_dev->cqc_timer_table,
854                                               HEM_TYPE_CQC_TIMER,
855                                               hr_dev->caps.cqc_timer_entry_sz,
856                                               hr_dev->caps.cqc_timer_bt_num);
857                 if (ret) {
858                         dev_err(dev,
859                                 "failed to init CQC timer memory, aborting.\n");
860                         goto err_unmap_qpc_timer;
861                 }
862         }
863
864         if (hr_dev->caps.gmv_entry_sz) {
865                 ret = hns_roce_init_hem_table(hr_dev, &hr_dev->gmv_table,
866                                               HEM_TYPE_GMV,
867                                               hr_dev->caps.gmv_entry_sz,
868                                               hr_dev->caps.gmv_entry_num);
869                 if (ret) {
870                         dev_err(dev,
871                                 "failed to init gmv table memory, ret = %d\n",
872                                 ret);
873                         goto err_unmap_cqc_timer;
874                 }
875         }
876
877         return 0;
878
879 err_unmap_cqc_timer:
880         if (hr_dev->caps.cqc_timer_entry_sz)
881                 hns_roce_cleanup_hem_table(hr_dev, &hr_dev->cqc_timer_table);
882
883 err_unmap_qpc_timer:
884         if (hr_dev->caps.qpc_timer_entry_sz)
885                 hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qpc_timer_table);
886
887 err_unmap_ctx:
888         if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_FLOW_CTRL)
889                 hns_roce_cleanup_hem_table(hr_dev,
890                                            &hr_dev->qp_table.sccc_table);
891 err_unmap_srq:
892         if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ)
893                 hns_roce_cleanup_hem_table(hr_dev, &hr_dev->srq_table.table);
894
895 err_unmap_cq:
896         hns_roce_cleanup_hem_table(hr_dev, &hr_dev->cq_table.table);
897
898 err_unmap_trrl:
899         if (hr_dev->caps.trrl_entry_sz)
900                 hns_roce_cleanup_hem_table(hr_dev,
901                                            &hr_dev->qp_table.trrl_table);
902
903 err_unmap_irrl:
904         hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qp_table.irrl_table);
905
906 err_unmap_qp:
907         hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qp_table.qp_table);
908
909 err_unmap_dmpt:
910         hns_roce_cleanup_hem_table(hr_dev, &hr_dev->mr_table.mtpt_table);
911
912         return ret;
913 }
914
915 /**
916  * hns_roce_setup_hca - setup host channel adapter
917  * @hr_dev: pointer to hns roce device
918  * Return : int
919  */
920 static int hns_roce_setup_hca(struct hns_roce_dev *hr_dev)
921 {
922         struct device *dev = hr_dev->dev;
923         int ret;
924
925         spin_lock_init(&hr_dev->sm_lock);
926
927         if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_CQ_RECORD_DB ||
928             hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_RECORD_DB) {
929                 INIT_LIST_HEAD(&hr_dev->pgdir_list);
930                 mutex_init(&hr_dev->pgdir_mutex);
931         }
932
933         hns_roce_init_uar_table(hr_dev);
934
935         ret = hns_roce_uar_alloc(hr_dev, &hr_dev->priv_uar);
936         if (ret) {
937                 dev_err(dev, "failed to allocate priv_uar.\n");
938                 goto err_uar_table_free;
939         }
940
941         ret = hns_roce_init_qp_table(hr_dev);
942         if (ret) {
943                 dev_err(dev, "failed to init qp_table.\n");
944                 goto err_uar_table_free;
945         }
946
947         hns_roce_init_pd_table(hr_dev);
948
949         if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_XRC)
950                 hns_roce_init_xrcd_table(hr_dev);
951
952         hns_roce_init_mr_table(hr_dev);
953
954         hns_roce_init_cq_table(hr_dev);
955
956         if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ)
957                 hns_roce_init_srq_table(hr_dev);
958
959         return 0;
960
961 err_uar_table_free:
962         ida_destroy(&hr_dev->uar_ida.ida);
963         return ret;
964 }
965
966 static void check_and_get_armed_cq(struct list_head *cq_list, struct ib_cq *cq)
967 {
968         struct hns_roce_cq *hr_cq = to_hr_cq(cq);
969         unsigned long flags;
970
971         spin_lock_irqsave(&hr_cq->lock, flags);
972         if (cq->comp_handler) {
973                 if (!hr_cq->is_armed) {
974                         hr_cq->is_armed = 1;
975                         list_add_tail(&hr_cq->node, cq_list);
976                 }
977         }
978         spin_unlock_irqrestore(&hr_cq->lock, flags);
979 }
980
981 void hns_roce_handle_device_err(struct hns_roce_dev *hr_dev)
982 {
983         struct hns_roce_qp *hr_qp;
984         struct hns_roce_cq *hr_cq;
985         struct list_head cq_list;
986         unsigned long flags_qp;
987         unsigned long flags;
988
989         INIT_LIST_HEAD(&cq_list);
990
991         spin_lock_irqsave(&hr_dev->qp_list_lock, flags);
992         list_for_each_entry(hr_qp, &hr_dev->qp_list, node) {
993                 spin_lock_irqsave(&hr_qp->sq.lock, flags_qp);
994                 if (hr_qp->sq.tail != hr_qp->sq.head)
995                         check_and_get_armed_cq(&cq_list, hr_qp->ibqp.send_cq);
996                 spin_unlock_irqrestore(&hr_qp->sq.lock, flags_qp);
997
998                 spin_lock_irqsave(&hr_qp->rq.lock, flags_qp);
999                 if ((!hr_qp->ibqp.srq) && (hr_qp->rq.tail != hr_qp->rq.head))
1000                         check_and_get_armed_cq(&cq_list, hr_qp->ibqp.recv_cq);
1001                 spin_unlock_irqrestore(&hr_qp->rq.lock, flags_qp);
1002         }
1003
1004         list_for_each_entry(hr_cq, &cq_list, node)
1005                 hns_roce_cq_completion(hr_dev, hr_cq->cqn);
1006
1007         spin_unlock_irqrestore(&hr_dev->qp_list_lock, flags);
1008 }
1009
1010 int hns_roce_init(struct hns_roce_dev *hr_dev)
1011 {
1012         struct device *dev = hr_dev->dev;
1013         int ret;
1014
1015         hr_dev->is_reset = false;
1016
1017         if (hr_dev->hw->cmq_init) {
1018                 ret = hr_dev->hw->cmq_init(hr_dev);
1019                 if (ret) {
1020                         dev_err(dev, "init RoCE Command Queue failed!\n");
1021                         return ret;
1022                 }
1023         }
1024
1025         ret = hr_dev->hw->hw_profile(hr_dev);
1026         if (ret) {
1027                 dev_err(dev, "get RoCE engine profile failed!\n");
1028                 goto error_failed_cmd_init;
1029         }
1030
1031         ret = hns_roce_cmd_init(hr_dev);
1032         if (ret) {
1033                 dev_err(dev, "cmd init failed!\n");
1034                 goto error_failed_cmd_init;
1035         }
1036
1037         /* EQ depends on poll mode, event mode depends on EQ */
1038         ret = hr_dev->hw->init_eq(hr_dev);
1039         if (ret) {
1040                 dev_err(dev, "eq init failed!\n");
1041                 goto error_failed_eq_table;
1042         }
1043
1044         if (hr_dev->cmd_mod) {
1045                 ret = hns_roce_cmd_use_events(hr_dev);
1046                 if (ret)
1047                         dev_warn(dev,
1048                                  "Cmd event  mode failed, set back to poll!\n");
1049         }
1050
1051         ret = hns_roce_init_hem(hr_dev);
1052         if (ret) {
1053                 dev_err(dev, "init HEM(Hardware Entry Memory) failed!\n");
1054                 goto error_failed_init_hem;
1055         }
1056
1057         ret = hns_roce_setup_hca(hr_dev);
1058         if (ret) {
1059                 dev_err(dev, "setup hca failed!\n");
1060                 goto error_failed_setup_hca;
1061         }
1062
1063         if (hr_dev->hw->hw_init) {
1064                 ret = hr_dev->hw->hw_init(hr_dev);
1065                 if (ret) {
1066                         dev_err(dev, "hw_init failed!\n");
1067                         goto error_failed_engine_init;
1068                 }
1069         }
1070
1071         INIT_LIST_HEAD(&hr_dev->qp_list);
1072         spin_lock_init(&hr_dev->qp_list_lock);
1073         INIT_LIST_HEAD(&hr_dev->dip_list);
1074         spin_lock_init(&hr_dev->dip_list_lock);
1075
1076         ret = hns_roce_register_device(hr_dev);
1077         if (ret)
1078                 goto error_failed_register_device;
1079
1080         return 0;
1081
1082 error_failed_register_device:
1083         if (hr_dev->hw->hw_exit)
1084                 hr_dev->hw->hw_exit(hr_dev);
1085
1086 error_failed_engine_init:
1087         hns_roce_cleanup_bitmap(hr_dev);
1088
1089 error_failed_setup_hca:
1090         hns_roce_cleanup_hem(hr_dev);
1091
1092 error_failed_init_hem:
1093         if (hr_dev->cmd_mod)
1094                 hns_roce_cmd_use_polling(hr_dev);
1095         hr_dev->hw->cleanup_eq(hr_dev);
1096
1097 error_failed_eq_table:
1098         hns_roce_cmd_cleanup(hr_dev);
1099
1100 error_failed_cmd_init:
1101         if (hr_dev->hw->cmq_exit)
1102                 hr_dev->hw->cmq_exit(hr_dev);
1103
1104         return ret;
1105 }
1106
1107 void hns_roce_exit(struct hns_roce_dev *hr_dev)
1108 {
1109         hns_roce_unregister_device(hr_dev);
1110
1111         if (hr_dev->hw->hw_exit)
1112                 hr_dev->hw->hw_exit(hr_dev);
1113         hns_roce_cleanup_bitmap(hr_dev);
1114         hns_roce_cleanup_hem(hr_dev);
1115
1116         if (hr_dev->cmd_mod)
1117                 hns_roce_cmd_use_polling(hr_dev);
1118
1119         hr_dev->hw->cleanup_eq(hr_dev);
1120         hns_roce_cmd_cleanup(hr_dev);
1121         if (hr_dev->hw->cmq_exit)
1122                 hr_dev->hw->cmq_exit(hr_dev);
1123 }
1124
1125 MODULE_LICENSE("Dual BSD/GPL");
1126 MODULE_AUTHOR("Wei Hu <xavier.huwei@huawei.com>");
1127 MODULE_AUTHOR("Nenglong Zhao <zhaonenglong@hisilicon.com>");
1128 MODULE_AUTHOR("Lijun Ou <oulijun@huawei.com>");
1129 MODULE_DESCRIPTION("HNS RoCE Driver");