xref: /linux/drivers/infiniband/hw/hns/hns_roce_main.c (revision dd91b5e1d6448794c07378d1be12e3261c8769e7)
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 #include "hns_roce_hw_v2.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 
handle_en_event(struct hns_roce_dev * hr_dev,u32 port,unsigned long event)92 static int handle_en_event(struct hns_roce_dev *hr_dev, u32 port,
93 			   unsigned long event)
94 {
95 	struct device *dev = hr_dev->dev;
96 	struct net_device *netdev;
97 	int ret = 0;
98 
99 	netdev = hr_dev->iboe.netdevs[port];
100 	if (!netdev) {
101 		dev_err(dev, "can't find netdev on port(%u)!\n", port);
102 		return -ENODEV;
103 	}
104 
105 	switch (event) {
106 	case NETDEV_UP:
107 	case NETDEV_CHANGE:
108 	case NETDEV_REGISTER:
109 	case NETDEV_CHANGEADDR:
110 		ret = hns_roce_set_mac(hr_dev, port, netdev->dev_addr);
111 		break;
112 	case NETDEV_DOWN:
113 		/*
114 		 * In v1 engine, only support all ports closed together.
115 		 */
116 		break;
117 	default:
118 		dev_dbg(dev, "NETDEV event = 0x%x!\n", (u32)(event));
119 		break;
120 	}
121 
122 	return ret;
123 }
124 
hns_roce_netdev_event(struct notifier_block * self,unsigned long event,void * ptr)125 static int hns_roce_netdev_event(struct notifier_block *self,
126 				 unsigned long event, void *ptr)
127 {
128 	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
129 	struct hns_roce_ib_iboe *iboe = NULL;
130 	struct hns_roce_dev *hr_dev = NULL;
131 	int ret;
132 	u32 port;
133 
134 	hr_dev = container_of(self, struct hns_roce_dev, iboe.nb);
135 	iboe = &hr_dev->iboe;
136 
137 	for (port = 0; port < hr_dev->caps.num_ports; port++) {
138 		if (dev == iboe->netdevs[port]) {
139 			ret = handle_en_event(hr_dev, port, event);
140 			if (ret)
141 				return NOTIFY_DONE;
142 			break;
143 		}
144 	}
145 
146 	return NOTIFY_DONE;
147 }
148 
hns_roce_setup_mtu_mac(struct hns_roce_dev * hr_dev)149 static int hns_roce_setup_mtu_mac(struct hns_roce_dev *hr_dev)
150 {
151 	int ret;
152 	u8 i;
153 
154 	for (i = 0; i < hr_dev->caps.num_ports; i++) {
155 		ret = hns_roce_set_mac(hr_dev, i,
156 				       hr_dev->iboe.netdevs[i]->dev_addr);
157 		if (ret)
158 			return ret;
159 	}
160 
161 	return 0;
162 }
163 
hns_roce_query_device(struct ib_device * ib_dev,struct ib_device_attr * props,struct ib_udata * uhw)164 static int hns_roce_query_device(struct ib_device *ib_dev,
165 				 struct ib_device_attr *props,
166 				 struct ib_udata *uhw)
167 {
168 	struct hns_roce_dev *hr_dev = to_hr_dev(ib_dev);
169 
170 	memset(props, 0, sizeof(*props));
171 
172 	props->fw_ver = hr_dev->caps.fw_ver;
173 	props->sys_image_guid = cpu_to_be64(hr_dev->sys_image_guid);
174 	props->max_mr_size = (u64)(~(0ULL));
175 	props->page_size_cap = hr_dev->caps.page_size_cap;
176 	props->vendor_id = hr_dev->vendor_id;
177 	props->vendor_part_id = hr_dev->vendor_part_id;
178 	props->hw_ver = hr_dev->hw_rev;
179 	props->max_qp = hr_dev->caps.num_qps;
180 	props->max_qp_wr = hr_dev->caps.max_wqes;
181 	props->device_cap_flags = IB_DEVICE_PORT_ACTIVE_EVENT |
182 				  IB_DEVICE_RC_RNR_NAK_GEN;
183 	props->max_send_sge = hr_dev->caps.max_sq_sg;
184 	props->max_recv_sge = hr_dev->caps.max_rq_sg;
185 	props->max_sge_rd = hr_dev->caps.max_sq_sg;
186 	props->max_cq = hr_dev->caps.num_cqs;
187 	props->max_cqe = hr_dev->caps.max_cqes;
188 	props->max_mr = hr_dev->caps.num_mtpts;
189 	props->max_pd = hr_dev->caps.num_pds;
190 	props->max_qp_rd_atom = hr_dev->caps.max_qp_dest_rdma;
191 	props->max_qp_init_rd_atom = hr_dev->caps.max_qp_init_rdma;
192 	props->atomic_cap = hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_ATOMIC ?
193 			    IB_ATOMIC_HCA : IB_ATOMIC_NONE;
194 	props->max_pkeys = 1;
195 	props->local_ca_ack_delay = hr_dev->caps.local_ca_ack_delay;
196 	props->max_ah = INT_MAX;
197 	props->cq_caps.max_cq_moderation_period = HNS_ROCE_MAX_CQ_PERIOD;
198 	props->cq_caps.max_cq_moderation_count = HNS_ROCE_MAX_CQ_COUNT;
199 	if (hr_dev->pci_dev->revision == PCI_REVISION_ID_HIP08)
200 		props->cq_caps.max_cq_moderation_period = HNS_ROCE_MAX_CQ_PERIOD_HIP08;
201 
202 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ) {
203 		props->max_srq = hr_dev->caps.num_srqs;
204 		props->max_srq_wr = hr_dev->caps.max_srq_wrs;
205 		props->max_srq_sge = hr_dev->caps.max_srq_sges;
206 	}
207 
208 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_FRMR &&
209 	    hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09) {
210 		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
211 		props->max_fast_reg_page_list_len = HNS_ROCE_FRMR_MAX_PA;
212 	}
213 
214 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_XRC)
215 		props->device_cap_flags |= IB_DEVICE_XRC;
216 
217 	return 0;
218 }
219 
hns_roce_query_port(struct ib_device * ib_dev,u32 port_num,struct ib_port_attr * props)220 static int hns_roce_query_port(struct ib_device *ib_dev, u32 port_num,
221 			       struct ib_port_attr *props)
222 {
223 	struct hns_roce_dev *hr_dev = to_hr_dev(ib_dev);
224 	struct device *dev = hr_dev->dev;
225 	struct net_device *net_dev;
226 	unsigned long flags;
227 	enum ib_mtu mtu;
228 	u32 port;
229 	int ret;
230 
231 	port = port_num - 1;
232 
233 	/* props being zeroed by the caller, avoid zeroing it here */
234 
235 	props->max_mtu = hr_dev->caps.max_mtu;
236 	props->gid_tbl_len = hr_dev->caps.gid_table_len[port];
237 	props->port_cap_flags = IB_PORT_CM_SUP | IB_PORT_REINIT_SUP |
238 				IB_PORT_VENDOR_CLASS_SUP |
239 				IB_PORT_BOOT_MGMT_SUP;
240 	props->max_msg_sz = HNS_ROCE_MAX_MSG_LEN;
241 	props->pkey_tbl_len = 1;
242 	ret = ib_get_eth_speed(ib_dev, port_num, &props->active_speed,
243 			       &props->active_width);
244 	if (ret)
245 		ibdev_warn(ib_dev, "failed to get speed, ret = %d.\n", ret);
246 
247 	spin_lock_irqsave(&hr_dev->iboe.lock, flags);
248 
249 	net_dev = hr_dev->iboe.netdevs[port];
250 	if (!net_dev) {
251 		spin_unlock_irqrestore(&hr_dev->iboe.lock, flags);
252 		dev_err(dev, "find netdev %u failed!\n", port);
253 		return -EINVAL;
254 	}
255 
256 	mtu = iboe_get_mtu(net_dev->mtu);
257 	props->active_mtu = mtu ? min(props->max_mtu, mtu) : IB_MTU_256;
258 	props->state = netif_running(net_dev) && netif_carrier_ok(net_dev) ?
259 			       IB_PORT_ACTIVE :
260 			       IB_PORT_DOWN;
261 	props->phys_state = props->state == IB_PORT_ACTIVE ?
262 				    IB_PORT_PHYS_STATE_LINK_UP :
263 				    IB_PORT_PHYS_STATE_DISABLED;
264 
265 	spin_unlock_irqrestore(&hr_dev->iboe.lock, flags);
266 
267 	return 0;
268 }
269 
hns_roce_get_link_layer(struct ib_device * device,u32 port_num)270 static enum rdma_link_layer hns_roce_get_link_layer(struct ib_device *device,
271 						    u32 port_num)
272 {
273 	return IB_LINK_LAYER_ETHERNET;
274 }
275 
hns_roce_query_pkey(struct ib_device * ib_dev,u32 port,u16 index,u16 * pkey)276 static int hns_roce_query_pkey(struct ib_device *ib_dev, u32 port, u16 index,
277 			       u16 *pkey)
278 {
279 	if (index > 0)
280 		return -EINVAL;
281 
282 	*pkey = PKEY_ID;
283 
284 	return 0;
285 }
286 
hns_roce_modify_device(struct ib_device * ib_dev,int mask,struct ib_device_modify * props)287 static int hns_roce_modify_device(struct ib_device *ib_dev, int mask,
288 				  struct ib_device_modify *props)
289 {
290 	unsigned long flags;
291 
292 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
293 		return -EOPNOTSUPP;
294 
295 	if (mask & IB_DEVICE_MODIFY_NODE_DESC) {
296 		spin_lock_irqsave(&to_hr_dev(ib_dev)->sm_lock, flags);
297 		memcpy(ib_dev->node_desc, props->node_desc, NODE_DESC_SIZE);
298 		spin_unlock_irqrestore(&to_hr_dev(ib_dev)->sm_lock, flags);
299 	}
300 
301 	return 0;
302 }
303 
304 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)305 hns_roce_user_mmap_entry_insert(struct ib_ucontext *ucontext, u64 address,
306 				size_t length,
307 				enum hns_roce_mmap_type mmap_type)
308 {
309 	struct hns_user_mmap_entry *entry;
310 	int ret;
311 
312 	entry = kzalloc(sizeof(*entry), GFP_KERNEL);
313 	if (!entry)
314 		return NULL;
315 
316 	entry->address = address;
317 	entry->mmap_type = mmap_type;
318 
319 	switch (mmap_type) {
320 	/* pgoff 0 must be used by DB for compatibility */
321 	case HNS_ROCE_MMAP_TYPE_DB:
322 		ret = rdma_user_mmap_entry_insert_exact(
323 				ucontext, &entry->rdma_entry, length, 0);
324 		break;
325 	case HNS_ROCE_MMAP_TYPE_DWQE:
326 		ret = rdma_user_mmap_entry_insert_range(
327 				ucontext, &entry->rdma_entry, length, 1,
328 				U32_MAX);
329 		break;
330 	default:
331 		ret = -EINVAL;
332 		break;
333 	}
334 
335 	if (ret) {
336 		kfree(entry);
337 		return NULL;
338 	}
339 
340 	return entry;
341 }
342 
hns_roce_dealloc_uar_entry(struct hns_roce_ucontext * context)343 static void hns_roce_dealloc_uar_entry(struct hns_roce_ucontext *context)
344 {
345 	if (context->db_mmap_entry)
346 		rdma_user_mmap_entry_remove(
347 			&context->db_mmap_entry->rdma_entry);
348 }
349 
hns_roce_alloc_uar_entry(struct ib_ucontext * uctx)350 static int hns_roce_alloc_uar_entry(struct ib_ucontext *uctx)
351 {
352 	struct hns_roce_ucontext *context = to_hr_ucontext(uctx);
353 	u64 address;
354 
355 	address = context->uar.pfn << PAGE_SHIFT;
356 	context->db_mmap_entry = hns_roce_user_mmap_entry_insert(
357 		uctx, address, PAGE_SIZE, HNS_ROCE_MMAP_TYPE_DB);
358 	if (!context->db_mmap_entry)
359 		return -ENOMEM;
360 
361 	return 0;
362 }
363 
hns_roce_alloc_ucontext(struct ib_ucontext * uctx,struct ib_udata * udata)364 static int hns_roce_alloc_ucontext(struct ib_ucontext *uctx,
365 				   struct ib_udata *udata)
366 {
367 	struct hns_roce_ucontext *context = to_hr_ucontext(uctx);
368 	struct hns_roce_dev *hr_dev = to_hr_dev(uctx->device);
369 	struct hns_roce_ib_alloc_ucontext_resp resp = {};
370 	struct hns_roce_ib_alloc_ucontext ucmd = {};
371 	int ret = -EAGAIN;
372 
373 	if (!hr_dev->active)
374 		goto error_out;
375 
376 	resp.qp_tab_size = hr_dev->caps.num_qps;
377 	resp.srq_tab_size = hr_dev->caps.num_srqs;
378 
379 	ret = ib_copy_from_udata(&ucmd, udata,
380 				 min(udata->inlen, sizeof(ucmd)));
381 	if (ret)
382 		goto error_out;
383 
384 	if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09)
385 		context->config = ucmd.config & HNS_ROCE_EXSGE_FLAGS;
386 
387 	if (context->config & HNS_ROCE_EXSGE_FLAGS) {
388 		resp.config |= HNS_ROCE_RSP_EXSGE_FLAGS;
389 		resp.max_inline_data = hr_dev->caps.max_sq_inline;
390 	}
391 
392 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_RQ_INLINE) {
393 		context->config |= ucmd.config & HNS_ROCE_RQ_INLINE_FLAGS;
394 		if (context->config & HNS_ROCE_RQ_INLINE_FLAGS)
395 			resp.config |= HNS_ROCE_RSP_RQ_INLINE_FLAGS;
396 	}
397 
398 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_CQE_INLINE) {
399 		context->config |= ucmd.config & HNS_ROCE_CQE_INLINE_FLAGS;
400 		if (context->config & HNS_ROCE_CQE_INLINE_FLAGS)
401 			resp.config |= HNS_ROCE_RSP_CQE_INLINE_FLAGS;
402 	}
403 
404 	if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09)
405 		resp.congest_type = hr_dev->caps.cong_cap;
406 
407 	ret = hns_roce_uar_alloc(hr_dev, &context->uar);
408 	if (ret)
409 		goto error_out;
410 
411 	ret = hns_roce_alloc_uar_entry(uctx);
412 	if (ret)
413 		goto error_fail_uar_entry;
414 
415 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_CQ_RECORD_DB ||
416 	    hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_RECORD_DB) {
417 		INIT_LIST_HEAD(&context->page_list);
418 		mutex_init(&context->page_mutex);
419 	}
420 
421 	resp.cqe_size = hr_dev->caps.cqe_sz;
422 
423 	ret = ib_copy_to_udata(udata, &resp,
424 			       min(udata->outlen, sizeof(resp)));
425 	if (ret)
426 		goto error_fail_copy_to_udata;
427 
428 	return 0;
429 
430 error_fail_copy_to_udata:
431 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_CQ_RECORD_DB ||
432 	    hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_RECORD_DB)
433 		mutex_destroy(&context->page_mutex);
434 	hns_roce_dealloc_uar_entry(context);
435 
436 error_fail_uar_entry:
437 	ida_free(&hr_dev->uar_ida.ida, (int)context->uar.logic_idx);
438 
439 error_out:
440 	atomic64_inc(&hr_dev->dfx_cnt[HNS_ROCE_DFX_UCTX_ALLOC_ERR_CNT]);
441 
442 	return ret;
443 }
444 
hns_roce_dealloc_ucontext(struct ib_ucontext * ibcontext)445 static void hns_roce_dealloc_ucontext(struct ib_ucontext *ibcontext)
446 {
447 	struct hns_roce_ucontext *context = to_hr_ucontext(ibcontext);
448 	struct hns_roce_dev *hr_dev = to_hr_dev(ibcontext->device);
449 
450 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_CQ_RECORD_DB ||
451 	    hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_RECORD_DB)
452 		mutex_destroy(&context->page_mutex);
453 
454 	hns_roce_dealloc_uar_entry(context);
455 
456 	ida_free(&hr_dev->uar_ida.ida, (int)context->uar.logic_idx);
457 }
458 
hns_roce_mmap(struct ib_ucontext * uctx,struct vm_area_struct * vma)459 static int hns_roce_mmap(struct ib_ucontext *uctx, struct vm_area_struct *vma)
460 {
461 	struct hns_roce_dev *hr_dev = to_hr_dev(uctx->device);
462 	struct rdma_user_mmap_entry *rdma_entry;
463 	struct hns_user_mmap_entry *entry;
464 	phys_addr_t pfn;
465 	pgprot_t prot;
466 	int ret;
467 
468 	if (hr_dev->dis_db) {
469 		atomic64_inc(&hr_dev->dfx_cnt[HNS_ROCE_DFX_MMAP_ERR_CNT]);
470 		return -EPERM;
471 	}
472 
473 	rdma_entry = rdma_user_mmap_entry_get_pgoff(uctx, vma->vm_pgoff);
474 	if (!rdma_entry) {
475 		atomic64_inc(&hr_dev->dfx_cnt[HNS_ROCE_DFX_MMAP_ERR_CNT]);
476 		return -EINVAL;
477 	}
478 
479 	entry = to_hns_mmap(rdma_entry);
480 	pfn = entry->address >> PAGE_SHIFT;
481 
482 	switch (entry->mmap_type) {
483 	case HNS_ROCE_MMAP_TYPE_DB:
484 	case HNS_ROCE_MMAP_TYPE_DWQE:
485 		prot = pgprot_device(vma->vm_page_prot);
486 		break;
487 	default:
488 		ret = -EINVAL;
489 		goto out;
490 	}
491 
492 	ret = rdma_user_mmap_io(uctx, vma, pfn, rdma_entry->npages * PAGE_SIZE,
493 				prot, rdma_entry);
494 
495 out:
496 	rdma_user_mmap_entry_put(rdma_entry);
497 	if (ret)
498 		atomic64_inc(&hr_dev->dfx_cnt[HNS_ROCE_DFX_MMAP_ERR_CNT]);
499 
500 	return ret;
501 }
502 
hns_roce_free_mmap(struct rdma_user_mmap_entry * rdma_entry)503 static void hns_roce_free_mmap(struct rdma_user_mmap_entry *rdma_entry)
504 {
505 	struct hns_user_mmap_entry *entry = to_hns_mmap(rdma_entry);
506 
507 	kfree(entry);
508 }
509 
hns_roce_port_immutable(struct ib_device * ib_dev,u32 port_num,struct ib_port_immutable * immutable)510 static int hns_roce_port_immutable(struct ib_device *ib_dev, u32 port_num,
511 				   struct ib_port_immutable *immutable)
512 {
513 	struct ib_port_attr attr;
514 	int ret;
515 
516 	ret = ib_query_port(ib_dev, port_num, &attr);
517 	if (ret)
518 		return ret;
519 
520 	immutable->pkey_tbl_len = attr.pkey_tbl_len;
521 	immutable->gid_tbl_len = attr.gid_tbl_len;
522 
523 	immutable->max_mad_size = IB_MGMT_MAD_SIZE;
524 	immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
525 	if (to_hr_dev(ib_dev)->caps.flags & HNS_ROCE_CAP_FLAG_ROCE_V1_V2)
526 		immutable->core_cap_flags |= RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
527 
528 	return 0;
529 }
530 
hns_roce_disassociate_ucontext(struct ib_ucontext * ibcontext)531 static void hns_roce_disassociate_ucontext(struct ib_ucontext *ibcontext)
532 {
533 }
534 
hns_roce_get_fw_ver(struct ib_device * device,char * str)535 static void hns_roce_get_fw_ver(struct ib_device *device, char *str)
536 {
537 	u64 fw_ver = to_hr_dev(device)->caps.fw_ver;
538 	unsigned int major, minor, sub_minor;
539 
540 	major = upper_32_bits(fw_ver);
541 	minor = high_16_bits(lower_32_bits(fw_ver));
542 	sub_minor = low_16_bits(fw_ver);
543 
544 	snprintf(str, IB_FW_VERSION_NAME_MAX, "%u.%u.%04u", major, minor,
545 		 sub_minor);
546 }
547 
548 #define HNS_ROCE_HW_CNT(ename, cname) \
549 	[HNS_ROCE_HW_##ename##_CNT].name = cname
550 
551 static const struct rdma_stat_desc hns_roce_port_stats_descs[] = {
552 	HNS_ROCE_HW_CNT(RX_RC_PKT, "rx_rc_pkt"),
553 	HNS_ROCE_HW_CNT(RX_UC_PKT, "rx_uc_pkt"),
554 	HNS_ROCE_HW_CNT(RX_UD_PKT, "rx_ud_pkt"),
555 	HNS_ROCE_HW_CNT(RX_XRC_PKT, "rx_xrc_pkt"),
556 	HNS_ROCE_HW_CNT(RX_PKT, "rx_pkt"),
557 	HNS_ROCE_HW_CNT(RX_ERR_PKT, "rx_err_pkt"),
558 	HNS_ROCE_HW_CNT(RX_CNP_PKT, "rx_cnp_pkt"),
559 	HNS_ROCE_HW_CNT(TX_RC_PKT, "tx_rc_pkt"),
560 	HNS_ROCE_HW_CNT(TX_UC_PKT, "tx_uc_pkt"),
561 	HNS_ROCE_HW_CNT(TX_UD_PKT, "tx_ud_pkt"),
562 	HNS_ROCE_HW_CNT(TX_XRC_PKT, "tx_xrc_pkt"),
563 	HNS_ROCE_HW_CNT(TX_PKT, "tx_pkt"),
564 	HNS_ROCE_HW_CNT(TX_ERR_PKT, "tx_err_pkt"),
565 	HNS_ROCE_HW_CNT(TX_CNP_PKT, "tx_cnp_pkt"),
566 	HNS_ROCE_HW_CNT(TRP_GET_MPT_ERR_PKT, "trp_get_mpt_err_pkt"),
567 	HNS_ROCE_HW_CNT(TRP_GET_IRRL_ERR_PKT, "trp_get_irrl_err_pkt"),
568 	HNS_ROCE_HW_CNT(ECN_DB, "ecn_doorbell"),
569 	HNS_ROCE_HW_CNT(RX_BUF, "rx_buffer"),
570 	HNS_ROCE_HW_CNT(TRP_RX_SOF, "trp_rx_sof"),
571 	HNS_ROCE_HW_CNT(CQ_CQE, "cq_cqe"),
572 	HNS_ROCE_HW_CNT(CQ_POE, "cq_poe"),
573 	HNS_ROCE_HW_CNT(CQ_NOTIFY, "cq_notify"),
574 };
575 
hns_roce_alloc_hw_port_stats(struct ib_device * device,u32 port_num)576 static struct rdma_hw_stats *hns_roce_alloc_hw_port_stats(
577 				struct ib_device *device, u32 port_num)
578 {
579 	struct hns_roce_dev *hr_dev = to_hr_dev(device);
580 
581 	if (port_num > hr_dev->caps.num_ports) {
582 		ibdev_err(device, "invalid port num.\n");
583 		return NULL;
584 	}
585 
586 	return rdma_alloc_hw_stats_struct(hns_roce_port_stats_descs,
587 					  ARRAY_SIZE(hns_roce_port_stats_descs),
588 					  RDMA_HW_STATS_DEFAULT_LIFESPAN);
589 }
590 
hns_roce_get_hw_stats(struct ib_device * device,struct rdma_hw_stats * stats,u32 port,int index)591 static int hns_roce_get_hw_stats(struct ib_device *device,
592 				 struct rdma_hw_stats *stats,
593 				 u32 port, int index)
594 {
595 	struct hns_roce_dev *hr_dev = to_hr_dev(device);
596 	int num_counters = HNS_ROCE_HW_CNT_TOTAL;
597 	int ret;
598 
599 	if (port == 0)
600 		return 0;
601 
602 	if (port > hr_dev->caps.num_ports)
603 		return -EINVAL;
604 
605 	ret = hr_dev->hw->query_hw_counter(hr_dev, stats->value, port,
606 					   &num_counters);
607 	if (ret) {
608 		ibdev_err(device, "failed to query hw counter, ret = %d\n",
609 			  ret);
610 		return ret;
611 	}
612 
613 	return num_counters;
614 }
615 
hns_roce_unregister_device(struct hns_roce_dev * hr_dev)616 static void hns_roce_unregister_device(struct hns_roce_dev *hr_dev)
617 {
618 	struct hns_roce_ib_iboe *iboe = &hr_dev->iboe;
619 
620 	hr_dev->active = false;
621 	unregister_netdevice_notifier(&iboe->nb);
622 	ib_unregister_device(&hr_dev->ib_dev);
623 }
624 
625 static const struct ib_device_ops hns_roce_dev_ops = {
626 	.owner = THIS_MODULE,
627 	.driver_id = RDMA_DRIVER_HNS,
628 	.uverbs_abi_ver = 1,
629 	.uverbs_no_driver_id_binding = 1,
630 
631 	.get_dev_fw_str = hns_roce_get_fw_ver,
632 	.add_gid = hns_roce_add_gid,
633 	.alloc_pd = hns_roce_alloc_pd,
634 	.alloc_ucontext = hns_roce_alloc_ucontext,
635 	.create_ah = hns_roce_create_ah,
636 	.create_user_ah = hns_roce_create_ah,
637 	.create_cq = hns_roce_create_cq,
638 	.create_qp = hns_roce_create_qp,
639 	.dealloc_pd = hns_roce_dealloc_pd,
640 	.dealloc_ucontext = hns_roce_dealloc_ucontext,
641 	.del_gid = hns_roce_del_gid,
642 	.dereg_mr = hns_roce_dereg_mr,
643 	.destroy_ah = hns_roce_destroy_ah,
644 	.destroy_cq = hns_roce_destroy_cq,
645 	.disassociate_ucontext = hns_roce_disassociate_ucontext,
646 	.get_dma_mr = hns_roce_get_dma_mr,
647 	.get_link_layer = hns_roce_get_link_layer,
648 	.get_port_immutable = hns_roce_port_immutable,
649 	.mmap = hns_roce_mmap,
650 	.mmap_free = hns_roce_free_mmap,
651 	.modify_device = hns_roce_modify_device,
652 	.modify_qp = hns_roce_modify_qp,
653 	.query_ah = hns_roce_query_ah,
654 	.query_device = hns_roce_query_device,
655 	.query_pkey = hns_roce_query_pkey,
656 	.query_port = hns_roce_query_port,
657 	.reg_user_mr = hns_roce_reg_user_mr,
658 
659 	INIT_RDMA_OBJ_SIZE(ib_ah, hns_roce_ah, ibah),
660 	INIT_RDMA_OBJ_SIZE(ib_cq, hns_roce_cq, ib_cq),
661 	INIT_RDMA_OBJ_SIZE(ib_pd, hns_roce_pd, ibpd),
662 	INIT_RDMA_OBJ_SIZE(ib_qp, hns_roce_qp, ibqp),
663 	INIT_RDMA_OBJ_SIZE(ib_ucontext, hns_roce_ucontext, ibucontext),
664 };
665 
666 static const struct ib_device_ops hns_roce_dev_hw_stats_ops = {
667 	.alloc_hw_port_stats = hns_roce_alloc_hw_port_stats,
668 	.get_hw_stats = hns_roce_get_hw_stats,
669 };
670 
671 static const struct ib_device_ops hns_roce_dev_mr_ops = {
672 	.rereg_user_mr = hns_roce_rereg_user_mr,
673 };
674 
675 static const struct ib_device_ops hns_roce_dev_mw_ops = {
676 	.alloc_mw = hns_roce_alloc_mw,
677 	.dealloc_mw = hns_roce_dealloc_mw,
678 
679 	INIT_RDMA_OBJ_SIZE(ib_mw, hns_roce_mw, ibmw),
680 };
681 
682 static const struct ib_device_ops hns_roce_dev_frmr_ops = {
683 	.alloc_mr = hns_roce_alloc_mr,
684 	.map_mr_sg = hns_roce_map_mr_sg,
685 };
686 
687 static const struct ib_device_ops hns_roce_dev_srq_ops = {
688 	.create_srq = hns_roce_create_srq,
689 	.destroy_srq = hns_roce_destroy_srq,
690 
691 	INIT_RDMA_OBJ_SIZE(ib_srq, hns_roce_srq, ibsrq),
692 };
693 
694 static const struct ib_device_ops hns_roce_dev_xrcd_ops = {
695 	.alloc_xrcd = hns_roce_alloc_xrcd,
696 	.dealloc_xrcd = hns_roce_dealloc_xrcd,
697 
698 	INIT_RDMA_OBJ_SIZE(ib_xrcd, hns_roce_xrcd, ibxrcd),
699 };
700 
701 static const struct ib_device_ops hns_roce_dev_restrack_ops = {
702 	.fill_res_cq_entry = hns_roce_fill_res_cq_entry,
703 	.fill_res_cq_entry_raw = hns_roce_fill_res_cq_entry_raw,
704 	.fill_res_qp_entry = hns_roce_fill_res_qp_entry,
705 	.fill_res_qp_entry_raw = hns_roce_fill_res_qp_entry_raw,
706 	.fill_res_mr_entry = hns_roce_fill_res_mr_entry,
707 	.fill_res_mr_entry_raw = hns_roce_fill_res_mr_entry_raw,
708 	.fill_res_srq_entry = hns_roce_fill_res_srq_entry,
709 	.fill_res_srq_entry_raw = hns_roce_fill_res_srq_entry_raw,
710 };
711 
hns_roce_register_device(struct hns_roce_dev * hr_dev)712 static int hns_roce_register_device(struct hns_roce_dev *hr_dev)
713 {
714 	int ret;
715 	struct hns_roce_ib_iboe *iboe = NULL;
716 	struct ib_device *ib_dev = NULL;
717 	struct device *dev = hr_dev->dev;
718 	unsigned int i;
719 
720 	iboe = &hr_dev->iboe;
721 	spin_lock_init(&iboe->lock);
722 
723 	ib_dev = &hr_dev->ib_dev;
724 
725 	ib_dev->node_type = RDMA_NODE_IB_CA;
726 	ib_dev->dev.parent = dev;
727 
728 	ib_dev->phys_port_cnt = hr_dev->caps.num_ports;
729 	ib_dev->local_dma_lkey = hr_dev->caps.reserved_lkey;
730 	ib_dev->num_comp_vectors = hr_dev->caps.num_comp_vectors;
731 
732 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_REREG_MR)
733 		ib_set_device_ops(ib_dev, &hns_roce_dev_mr_ops);
734 
735 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_MW)
736 		ib_set_device_ops(ib_dev, &hns_roce_dev_mw_ops);
737 
738 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_FRMR)
739 		ib_set_device_ops(ib_dev, &hns_roce_dev_frmr_ops);
740 
741 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ) {
742 		ib_set_device_ops(ib_dev, &hns_roce_dev_srq_ops);
743 		ib_set_device_ops(ib_dev, hr_dev->hw->hns_roce_dev_srq_ops);
744 	}
745 
746 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_XRC)
747 		ib_set_device_ops(ib_dev, &hns_roce_dev_xrcd_ops);
748 
749 	if (hr_dev->pci_dev->revision >= PCI_REVISION_ID_HIP09 &&
750 	    !hr_dev->is_vf)
751 		ib_set_device_ops(ib_dev, &hns_roce_dev_hw_stats_ops);
752 
753 	ib_set_device_ops(ib_dev, hr_dev->hw->hns_roce_dev_ops);
754 	ib_set_device_ops(ib_dev, &hns_roce_dev_ops);
755 	ib_set_device_ops(ib_dev, &hns_roce_dev_restrack_ops);
756 	for (i = 0; i < hr_dev->caps.num_ports; i++) {
757 		if (!hr_dev->iboe.netdevs[i])
758 			continue;
759 
760 		ret = ib_device_set_netdev(ib_dev, hr_dev->iboe.netdevs[i],
761 					   i + 1);
762 		if (ret)
763 			return ret;
764 	}
765 	dma_set_max_seg_size(dev, SZ_2G);
766 	ret = ib_register_device(ib_dev, "hns_%d", dev);
767 	if (ret) {
768 		dev_err(dev, "ib_register_device failed!\n");
769 		return ret;
770 	}
771 
772 	ret = hns_roce_setup_mtu_mac(hr_dev);
773 	if (ret) {
774 		dev_err(dev, "setup_mtu_mac failed!\n");
775 		goto error_failed_setup_mtu_mac;
776 	}
777 
778 	iboe->nb.notifier_call = hns_roce_netdev_event;
779 	ret = register_netdevice_notifier(&iboe->nb);
780 	if (ret) {
781 		dev_err(dev, "register_netdevice_notifier failed!\n");
782 		goto error_failed_setup_mtu_mac;
783 	}
784 
785 	hr_dev->active = true;
786 	return 0;
787 
788 error_failed_setup_mtu_mac:
789 	ib_unregister_device(ib_dev);
790 
791 	return ret;
792 }
793 
hns_roce_init_hem(struct hns_roce_dev * hr_dev)794 static int hns_roce_init_hem(struct hns_roce_dev *hr_dev)
795 {
796 	struct device *dev = hr_dev->dev;
797 	int ret;
798 
799 	ret = hns_roce_init_hem_table(hr_dev, &hr_dev->mr_table.mtpt_table,
800 				      HEM_TYPE_MTPT, hr_dev->caps.mtpt_entry_sz,
801 				      hr_dev->caps.num_mtpts);
802 	if (ret) {
803 		dev_err(dev, "failed to init MTPT context memory, aborting.\n");
804 		return ret;
805 	}
806 
807 	ret = hns_roce_init_hem_table(hr_dev, &hr_dev->qp_table.qp_table,
808 				      HEM_TYPE_QPC, hr_dev->caps.qpc_sz,
809 				      hr_dev->caps.num_qps);
810 	if (ret) {
811 		dev_err(dev, "failed to init QP context memory, aborting.\n");
812 		goto err_unmap_dmpt;
813 	}
814 
815 	ret = hns_roce_init_hem_table(hr_dev, &hr_dev->qp_table.irrl_table,
816 				      HEM_TYPE_IRRL,
817 				      hr_dev->caps.irrl_entry_sz *
818 				      hr_dev->caps.max_qp_init_rdma,
819 				      hr_dev->caps.num_qps);
820 	if (ret) {
821 		dev_err(dev, "failed to init irrl_table memory, aborting.\n");
822 		goto err_unmap_qp;
823 	}
824 
825 	if (hr_dev->caps.trrl_entry_sz) {
826 		ret = hns_roce_init_hem_table(hr_dev,
827 					      &hr_dev->qp_table.trrl_table,
828 					      HEM_TYPE_TRRL,
829 					      hr_dev->caps.trrl_entry_sz *
830 					      hr_dev->caps.max_qp_dest_rdma,
831 					      hr_dev->caps.num_qps);
832 		if (ret) {
833 			dev_err(dev,
834 				"failed to init trrl_table memory, aborting.\n");
835 			goto err_unmap_irrl;
836 		}
837 	}
838 
839 	ret = hns_roce_init_hem_table(hr_dev, &hr_dev->cq_table.table,
840 				      HEM_TYPE_CQC, hr_dev->caps.cqc_entry_sz,
841 				      hr_dev->caps.num_cqs);
842 	if (ret) {
843 		dev_err(dev, "failed to init CQ context memory, aborting.\n");
844 		goto err_unmap_trrl;
845 	}
846 
847 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ) {
848 		ret = hns_roce_init_hem_table(hr_dev, &hr_dev->srq_table.table,
849 					      HEM_TYPE_SRQC,
850 					      hr_dev->caps.srqc_entry_sz,
851 					      hr_dev->caps.num_srqs);
852 		if (ret) {
853 			dev_err(dev,
854 				"failed to init SRQ context memory, aborting.\n");
855 			goto err_unmap_cq;
856 		}
857 	}
858 
859 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_FLOW_CTRL) {
860 		ret = hns_roce_init_hem_table(hr_dev,
861 					      &hr_dev->qp_table.sccc_table,
862 					      HEM_TYPE_SCCC,
863 					      hr_dev->caps.sccc_sz,
864 					      hr_dev->caps.num_qps);
865 		if (ret) {
866 			dev_err(dev,
867 				"failed to init SCC context memory, aborting.\n");
868 			goto err_unmap_srq;
869 		}
870 	}
871 
872 	if (hr_dev->caps.qpc_timer_entry_sz) {
873 		ret = hns_roce_init_hem_table(hr_dev, &hr_dev->qpc_timer_table,
874 					      HEM_TYPE_QPC_TIMER,
875 					      hr_dev->caps.qpc_timer_entry_sz,
876 					      hr_dev->caps.qpc_timer_bt_num);
877 		if (ret) {
878 			dev_err(dev,
879 				"failed to init QPC timer memory, aborting.\n");
880 			goto err_unmap_ctx;
881 		}
882 	}
883 
884 	if (hr_dev->caps.cqc_timer_entry_sz) {
885 		ret = hns_roce_init_hem_table(hr_dev, &hr_dev->cqc_timer_table,
886 					      HEM_TYPE_CQC_TIMER,
887 					      hr_dev->caps.cqc_timer_entry_sz,
888 					      hr_dev->caps.cqc_timer_bt_num);
889 		if (ret) {
890 			dev_err(dev,
891 				"failed to init CQC timer memory, aborting.\n");
892 			goto err_unmap_qpc_timer;
893 		}
894 	}
895 
896 	if (hr_dev->caps.gmv_entry_sz) {
897 		ret = hns_roce_init_hem_table(hr_dev, &hr_dev->gmv_table,
898 					      HEM_TYPE_GMV,
899 					      hr_dev->caps.gmv_entry_sz,
900 					      hr_dev->caps.gmv_entry_num);
901 		if (ret) {
902 			dev_err(dev,
903 				"failed to init gmv table memory, ret = %d\n",
904 				ret);
905 			goto err_unmap_cqc_timer;
906 		}
907 	}
908 
909 	return 0;
910 
911 err_unmap_cqc_timer:
912 	if (hr_dev->caps.cqc_timer_entry_sz)
913 		hns_roce_cleanup_hem_table(hr_dev, &hr_dev->cqc_timer_table);
914 
915 err_unmap_qpc_timer:
916 	if (hr_dev->caps.qpc_timer_entry_sz)
917 		hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qpc_timer_table);
918 
919 err_unmap_ctx:
920 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_FLOW_CTRL)
921 		hns_roce_cleanup_hem_table(hr_dev,
922 					   &hr_dev->qp_table.sccc_table);
923 err_unmap_srq:
924 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ)
925 		hns_roce_cleanup_hem_table(hr_dev, &hr_dev->srq_table.table);
926 
927 err_unmap_cq:
928 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->cq_table.table);
929 
930 err_unmap_trrl:
931 	if (hr_dev->caps.trrl_entry_sz)
932 		hns_roce_cleanup_hem_table(hr_dev,
933 					   &hr_dev->qp_table.trrl_table);
934 
935 err_unmap_irrl:
936 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qp_table.irrl_table);
937 
938 err_unmap_qp:
939 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->qp_table.qp_table);
940 
941 err_unmap_dmpt:
942 	hns_roce_cleanup_hem_table(hr_dev, &hr_dev->mr_table.mtpt_table);
943 
944 	return ret;
945 }
946 
hns_roce_teardown_hca(struct hns_roce_dev * hr_dev)947 static void hns_roce_teardown_hca(struct hns_roce_dev *hr_dev)
948 {
949 	hns_roce_cleanup_bitmap(hr_dev);
950 
951 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_CQ_RECORD_DB ||
952 	    hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_RECORD_DB)
953 		mutex_destroy(&hr_dev->pgdir_mutex);
954 }
955 
956 /**
957  * hns_roce_setup_hca - setup host channel adapter
958  * @hr_dev: pointer to hns roce device
959  * Return : int
960  */
hns_roce_setup_hca(struct hns_roce_dev * hr_dev)961 static int hns_roce_setup_hca(struct hns_roce_dev *hr_dev)
962 {
963 	struct device *dev = hr_dev->dev;
964 	int ret;
965 
966 	spin_lock_init(&hr_dev->sm_lock);
967 
968 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_CQ_RECORD_DB ||
969 	    hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_RECORD_DB) {
970 		INIT_LIST_HEAD(&hr_dev->pgdir_list);
971 		mutex_init(&hr_dev->pgdir_mutex);
972 	}
973 
974 	hns_roce_init_uar_table(hr_dev);
975 
976 	ret = hns_roce_uar_alloc(hr_dev, &hr_dev->priv_uar);
977 	if (ret) {
978 		dev_err(dev, "failed to allocate priv_uar.\n");
979 		goto err_uar_table_free;
980 	}
981 
982 	ret = hns_roce_init_qp_table(hr_dev);
983 	if (ret) {
984 		dev_err(dev, "failed to init qp_table.\n");
985 		goto err_uar_table_free;
986 	}
987 
988 	hns_roce_init_pd_table(hr_dev);
989 
990 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_XRC)
991 		hns_roce_init_xrcd_table(hr_dev);
992 
993 	hns_roce_init_mr_table(hr_dev);
994 
995 	hns_roce_init_cq_table(hr_dev);
996 
997 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_SRQ)
998 		hns_roce_init_srq_table(hr_dev);
999 
1000 	return 0;
1001 
1002 err_uar_table_free:
1003 	ida_destroy(&hr_dev->uar_ida.ida);
1004 	if (hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_CQ_RECORD_DB ||
1005 	    hr_dev->caps.flags & HNS_ROCE_CAP_FLAG_QP_RECORD_DB)
1006 		mutex_destroy(&hr_dev->pgdir_mutex);
1007 
1008 	return ret;
1009 }
1010 
check_and_get_armed_cq(struct list_head * cq_list,struct ib_cq * cq)1011 static void check_and_get_armed_cq(struct list_head *cq_list, struct ib_cq *cq)
1012 {
1013 	struct hns_roce_cq *hr_cq = to_hr_cq(cq);
1014 	unsigned long flags;
1015 
1016 	spin_lock_irqsave(&hr_cq->lock, flags);
1017 	if (cq->comp_handler) {
1018 		if (!hr_cq->is_armed) {
1019 			hr_cq->is_armed = 1;
1020 			list_add_tail(&hr_cq->node, cq_list);
1021 		}
1022 	}
1023 	spin_unlock_irqrestore(&hr_cq->lock, flags);
1024 }
1025 
hns_roce_handle_device_err(struct hns_roce_dev * hr_dev)1026 void hns_roce_handle_device_err(struct hns_roce_dev *hr_dev)
1027 {
1028 	struct hns_roce_qp *hr_qp;
1029 	struct hns_roce_cq *hr_cq;
1030 	struct list_head cq_list;
1031 	unsigned long flags_qp;
1032 	unsigned long flags;
1033 
1034 	INIT_LIST_HEAD(&cq_list);
1035 
1036 	spin_lock_irqsave(&hr_dev->qp_list_lock, flags);
1037 	list_for_each_entry(hr_qp, &hr_dev->qp_list, node) {
1038 		spin_lock_irqsave(&hr_qp->sq.lock, flags_qp);
1039 		if (hr_qp->sq.tail != hr_qp->sq.head)
1040 			check_and_get_armed_cq(&cq_list, hr_qp->ibqp.send_cq);
1041 		spin_unlock_irqrestore(&hr_qp->sq.lock, flags_qp);
1042 
1043 		spin_lock_irqsave(&hr_qp->rq.lock, flags_qp);
1044 		if ((!hr_qp->ibqp.srq) && (hr_qp->rq.tail != hr_qp->rq.head))
1045 			check_and_get_armed_cq(&cq_list, hr_qp->ibqp.recv_cq);
1046 		spin_unlock_irqrestore(&hr_qp->rq.lock, flags_qp);
1047 	}
1048 
1049 	list_for_each_entry(hr_cq, &cq_list, node)
1050 		hns_roce_cq_completion(hr_dev, hr_cq->cqn);
1051 
1052 	spin_unlock_irqrestore(&hr_dev->qp_list_lock, flags);
1053 }
1054 
hns_roce_alloc_dfx_cnt(struct hns_roce_dev * hr_dev)1055 static int hns_roce_alloc_dfx_cnt(struct hns_roce_dev *hr_dev)
1056 {
1057 	hr_dev->dfx_cnt = kvcalloc(HNS_ROCE_DFX_CNT_TOTAL, sizeof(atomic64_t),
1058 				   GFP_KERNEL);
1059 	if (!hr_dev->dfx_cnt)
1060 		return -ENOMEM;
1061 
1062 	return 0;
1063 }
1064 
hns_roce_dealloc_dfx_cnt(struct hns_roce_dev * hr_dev)1065 static void hns_roce_dealloc_dfx_cnt(struct hns_roce_dev *hr_dev)
1066 {
1067 	kvfree(hr_dev->dfx_cnt);
1068 }
1069 
hns_roce_init(struct hns_roce_dev * hr_dev)1070 int hns_roce_init(struct hns_roce_dev *hr_dev)
1071 {
1072 	struct device *dev = hr_dev->dev;
1073 	int ret;
1074 
1075 	hr_dev->is_reset = false;
1076 
1077 	ret = hns_roce_alloc_dfx_cnt(hr_dev);
1078 	if (ret)
1079 		return ret;
1080 
1081 	if (hr_dev->hw->cmq_init) {
1082 		ret = hr_dev->hw->cmq_init(hr_dev);
1083 		if (ret) {
1084 			dev_err(dev, "init RoCE Command Queue failed!\n");
1085 			goto error_failed_alloc_dfx_cnt;
1086 		}
1087 	}
1088 
1089 	ret = hr_dev->hw->hw_profile(hr_dev);
1090 	if (ret) {
1091 		dev_err(dev, "get RoCE engine profile failed!\n");
1092 		goto error_failed_cmd_init;
1093 	}
1094 
1095 	ret = hns_roce_cmd_init(hr_dev);
1096 	if (ret) {
1097 		dev_err(dev, "cmd init failed!\n");
1098 		goto error_failed_cmd_init;
1099 	}
1100 
1101 	/* EQ depends on poll mode, event mode depends on EQ */
1102 	ret = hr_dev->hw->init_eq(hr_dev);
1103 	if (ret) {
1104 		dev_err(dev, "eq init failed!\n");
1105 		goto error_failed_eq_table;
1106 	}
1107 
1108 	if (hr_dev->cmd_mod) {
1109 		ret = hns_roce_cmd_use_events(hr_dev);
1110 		if (ret)
1111 			dev_warn(dev,
1112 				 "Cmd event  mode failed, set back to poll!\n");
1113 	}
1114 
1115 	ret = hns_roce_init_hem(hr_dev);
1116 	if (ret) {
1117 		dev_err(dev, "init HEM(Hardware Entry Memory) failed!\n");
1118 		goto error_failed_init_hem;
1119 	}
1120 
1121 	ret = hns_roce_setup_hca(hr_dev);
1122 	if (ret) {
1123 		dev_err(dev, "setup hca failed!\n");
1124 		goto error_failed_setup_hca;
1125 	}
1126 
1127 	if (hr_dev->hw->hw_init) {
1128 		ret = hr_dev->hw->hw_init(hr_dev);
1129 		if (ret) {
1130 			dev_err(dev, "hw_init failed!\n");
1131 			goto error_failed_engine_init;
1132 		}
1133 	}
1134 
1135 	INIT_LIST_HEAD(&hr_dev->qp_list);
1136 	spin_lock_init(&hr_dev->qp_list_lock);
1137 
1138 	ret = hns_roce_register_device(hr_dev);
1139 	if (ret)
1140 		goto error_failed_register_device;
1141 
1142 	hns_roce_register_debugfs(hr_dev);
1143 
1144 	return 0;
1145 
1146 error_failed_register_device:
1147 	if (hr_dev->hw->hw_exit)
1148 		hr_dev->hw->hw_exit(hr_dev);
1149 
1150 error_failed_engine_init:
1151 	hns_roce_teardown_hca(hr_dev);
1152 
1153 error_failed_setup_hca:
1154 	hns_roce_cleanup_hem(hr_dev);
1155 
1156 error_failed_init_hem:
1157 	if (hr_dev->cmd_mod)
1158 		hns_roce_cmd_use_polling(hr_dev);
1159 	hr_dev->hw->cleanup_eq(hr_dev);
1160 
1161 error_failed_eq_table:
1162 	hns_roce_cmd_cleanup(hr_dev);
1163 
1164 error_failed_cmd_init:
1165 	if (hr_dev->hw->cmq_exit)
1166 		hr_dev->hw->cmq_exit(hr_dev);
1167 
1168 error_failed_alloc_dfx_cnt:
1169 	hns_roce_dealloc_dfx_cnt(hr_dev);
1170 
1171 	return ret;
1172 }
1173 
hns_roce_exit(struct hns_roce_dev * hr_dev)1174 void hns_roce_exit(struct hns_roce_dev *hr_dev)
1175 {
1176 	hns_roce_unregister_debugfs(hr_dev);
1177 	hns_roce_unregister_device(hr_dev);
1178 
1179 	if (hr_dev->hw->hw_exit)
1180 		hr_dev->hw->hw_exit(hr_dev);
1181 	hns_roce_teardown_hca(hr_dev);
1182 	hns_roce_cleanup_hem(hr_dev);
1183 
1184 	if (hr_dev->cmd_mod)
1185 		hns_roce_cmd_use_polling(hr_dev);
1186 
1187 	hr_dev->hw->cleanup_eq(hr_dev);
1188 	hns_roce_cmd_cleanup(hr_dev);
1189 	if (hr_dev->hw->cmq_exit)
1190 		hr_dev->hw->cmq_exit(hr_dev);
1191 	hns_roce_dealloc_dfx_cnt(hr_dev);
1192 }
1193 
1194 MODULE_LICENSE("Dual BSD/GPL");
1195 MODULE_AUTHOR("Wei Hu <xavier.huwei@huawei.com>");
1196 MODULE_AUTHOR("Nenglong Zhao <zhaonenglong@hisilicon.com>");
1197 MODULE_AUTHOR("Lijun Ou <oulijun@huawei.com>");
1198 MODULE_DESCRIPTION("HNS RoCE Driver");
1199