perf tools: Don't clone maps from parent when synthesizing forks
[linux/fpc-iii.git] / drivers / infiniband / hw / vmw_pvrdma / pvrdma_main.c
blob398443f43dc350c6aa1c5b6fefd35743912e0843
1 /*
2 * Copyright (c) 2012-2016 VMware, Inc. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of EITHER the GNU General Public License
6 * version 2 as published by the Free Software Foundation or the BSD
7 * 2-Clause License. This program is distributed in the hope that it
8 * will be useful, but WITHOUT ANY WARRANTY; WITHOUT EVEN THE IMPLIED
9 * WARRANTY OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE.
10 * See the GNU General Public License version 2 for more details at
11 * http://www.gnu.org/licenses/old-licenses/gpl-2.0.en.html.
13 * You should have received a copy of the GNU General Public License
14 * along with this program available in the file COPYING in the main
15 * directory of this source tree.
17 * The BSD 2-Clause License
19 * Redistribution and use in source and binary forms, with or
20 * without modification, are permitted provided that the following
21 * conditions are met:
23 * - Redistributions of source code must retain the above
24 * copyright notice, this list of conditions and the following
25 * disclaimer.
27 * - Redistributions in binary form must reproduce the above
28 * copyright notice, this list of conditions and the following
29 * disclaimer in the documentation and/or other materials
30 * provided with the distribution.
32 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
33 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
34 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
35 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
36 * COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT,
37 * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
38 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
39 * SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
40 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
41 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
42 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
43 * OF THE POSSIBILITY OF SUCH DAMAGE.
46 #include <linux/errno.h>
47 #include <linux/inetdevice.h>
48 #include <linux/init.h>
49 #include <linux/module.h>
50 #include <linux/slab.h>
51 #include <rdma/ib_addr.h>
52 #include <rdma/ib_smi.h>
53 #include <rdma/ib_user_verbs.h>
54 #include <net/addrconf.h>
56 #include "pvrdma.h"
58 #define DRV_NAME "vmw_pvrdma"
59 #define DRV_VERSION "1.0.1.0-k"
61 static DEFINE_MUTEX(pvrdma_device_list_lock);
62 static LIST_HEAD(pvrdma_device_list);
63 static struct workqueue_struct *event_wq;
65 static int pvrdma_add_gid(const struct ib_gid_attr *attr, void **context);
66 static int pvrdma_del_gid(const struct ib_gid_attr *attr, void **context);
68 static ssize_t hca_type_show(struct device *device,
69 struct device_attribute *attr, char *buf)
71 return sprintf(buf, "VMW_PVRDMA-%s\n", DRV_VERSION);
73 static DEVICE_ATTR_RO(hca_type);
75 static ssize_t hw_rev_show(struct device *device,
76 struct device_attribute *attr, char *buf)
78 return sprintf(buf, "%d\n", PVRDMA_REV_ID);
80 static DEVICE_ATTR_RO(hw_rev);
82 static ssize_t board_id_show(struct device *device,
83 struct device_attribute *attr, char *buf)
85 return sprintf(buf, "%d\n", PVRDMA_BOARD_ID);
87 static DEVICE_ATTR_RO(board_id);
89 static struct attribute *pvrdma_class_attributes[] = {
90 &dev_attr_hw_rev.attr,
91 &dev_attr_hca_type.attr,
92 &dev_attr_board_id.attr,
93 NULL,
96 static const struct attribute_group pvrdma_attr_group = {
97 .attrs = pvrdma_class_attributes,
100 static void pvrdma_get_fw_ver_str(struct ib_device *device, char *str)
102 struct pvrdma_dev *dev =
103 container_of(device, struct pvrdma_dev, ib_dev);
104 snprintf(str, IB_FW_VERSION_NAME_MAX, "%d.%d.%d\n",
105 (int) (dev->dsr->caps.fw_ver >> 32),
106 (int) (dev->dsr->caps.fw_ver >> 16) & 0xffff,
107 (int) dev->dsr->caps.fw_ver & 0xffff);
110 static int pvrdma_init_device(struct pvrdma_dev *dev)
112 /* Initialize some device related stuff */
113 spin_lock_init(&dev->cmd_lock);
114 sema_init(&dev->cmd_sema, 1);
115 atomic_set(&dev->num_qps, 0);
116 atomic_set(&dev->num_srqs, 0);
117 atomic_set(&dev->num_cqs, 0);
118 atomic_set(&dev->num_pds, 0);
119 atomic_set(&dev->num_ahs, 0);
121 return 0;
124 static int pvrdma_port_immutable(struct ib_device *ibdev, u8 port_num,
125 struct ib_port_immutable *immutable)
127 struct pvrdma_dev *dev = to_vdev(ibdev);
128 struct ib_port_attr attr;
129 int err;
131 if (dev->dsr->caps.gid_types == PVRDMA_GID_TYPE_FLAG_ROCE_V1)
132 immutable->core_cap_flags |= RDMA_CORE_PORT_IBA_ROCE;
133 else if (dev->dsr->caps.gid_types == PVRDMA_GID_TYPE_FLAG_ROCE_V2)
134 immutable->core_cap_flags |= RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
136 err = ib_query_port(ibdev, port_num, &attr);
137 if (err)
138 return err;
140 immutable->pkey_tbl_len = attr.pkey_tbl_len;
141 immutable->gid_tbl_len = attr.gid_tbl_len;
142 immutable->max_mad_size = IB_MGMT_MAD_SIZE;
143 return 0;
146 static struct net_device *pvrdma_get_netdev(struct ib_device *ibdev,
147 u8 port_num)
149 struct net_device *netdev;
150 struct pvrdma_dev *dev = to_vdev(ibdev);
152 if (port_num != 1)
153 return NULL;
155 rcu_read_lock();
156 netdev = dev->netdev;
157 if (netdev)
158 dev_hold(netdev);
159 rcu_read_unlock();
161 return netdev;
164 static int pvrdma_register_device(struct pvrdma_dev *dev)
166 int ret = -1;
168 dev->ib_dev.node_guid = dev->dsr->caps.node_guid;
169 dev->sys_image_guid = dev->dsr->caps.sys_image_guid;
170 dev->flags = 0;
171 dev->ib_dev.owner = THIS_MODULE;
172 dev->ib_dev.num_comp_vectors = 1;
173 dev->ib_dev.dev.parent = &dev->pdev->dev;
174 dev->ib_dev.uverbs_abi_ver = PVRDMA_UVERBS_ABI_VERSION;
175 dev->ib_dev.uverbs_cmd_mask =
176 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
177 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
178 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
179 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
180 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
181 (1ull << IB_USER_VERBS_CMD_REG_MR) |
182 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
183 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
184 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
185 (1ull << IB_USER_VERBS_CMD_POLL_CQ) |
186 (1ull << IB_USER_VERBS_CMD_REQ_NOTIFY_CQ) |
187 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
188 (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
189 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
190 (1ull << IB_USER_VERBS_CMD_QUERY_QP) |
191 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
192 (1ull << IB_USER_VERBS_CMD_POST_SEND) |
193 (1ull << IB_USER_VERBS_CMD_POST_RECV) |
194 (1ull << IB_USER_VERBS_CMD_CREATE_AH) |
195 (1ull << IB_USER_VERBS_CMD_DESTROY_AH);
197 dev->ib_dev.node_type = RDMA_NODE_IB_CA;
198 dev->ib_dev.phys_port_cnt = dev->dsr->caps.phys_port_cnt;
200 dev->ib_dev.query_device = pvrdma_query_device;
201 dev->ib_dev.query_port = pvrdma_query_port;
202 dev->ib_dev.query_gid = pvrdma_query_gid;
203 dev->ib_dev.query_pkey = pvrdma_query_pkey;
204 dev->ib_dev.modify_port = pvrdma_modify_port;
205 dev->ib_dev.alloc_ucontext = pvrdma_alloc_ucontext;
206 dev->ib_dev.dealloc_ucontext = pvrdma_dealloc_ucontext;
207 dev->ib_dev.mmap = pvrdma_mmap;
208 dev->ib_dev.alloc_pd = pvrdma_alloc_pd;
209 dev->ib_dev.dealloc_pd = pvrdma_dealloc_pd;
210 dev->ib_dev.create_ah = pvrdma_create_ah;
211 dev->ib_dev.destroy_ah = pvrdma_destroy_ah;
212 dev->ib_dev.create_qp = pvrdma_create_qp;
213 dev->ib_dev.modify_qp = pvrdma_modify_qp;
214 dev->ib_dev.query_qp = pvrdma_query_qp;
215 dev->ib_dev.destroy_qp = pvrdma_destroy_qp;
216 dev->ib_dev.post_send = pvrdma_post_send;
217 dev->ib_dev.post_recv = pvrdma_post_recv;
218 dev->ib_dev.create_cq = pvrdma_create_cq;
219 dev->ib_dev.destroy_cq = pvrdma_destroy_cq;
220 dev->ib_dev.poll_cq = pvrdma_poll_cq;
221 dev->ib_dev.req_notify_cq = pvrdma_req_notify_cq;
222 dev->ib_dev.get_dma_mr = pvrdma_get_dma_mr;
223 dev->ib_dev.reg_user_mr = pvrdma_reg_user_mr;
224 dev->ib_dev.dereg_mr = pvrdma_dereg_mr;
225 dev->ib_dev.alloc_mr = pvrdma_alloc_mr;
226 dev->ib_dev.map_mr_sg = pvrdma_map_mr_sg;
227 dev->ib_dev.add_gid = pvrdma_add_gid;
228 dev->ib_dev.del_gid = pvrdma_del_gid;
229 dev->ib_dev.get_netdev = pvrdma_get_netdev;
230 dev->ib_dev.get_port_immutable = pvrdma_port_immutable;
231 dev->ib_dev.get_link_layer = pvrdma_port_link_layer;
232 dev->ib_dev.get_dev_fw_str = pvrdma_get_fw_ver_str;
234 mutex_init(&dev->port_mutex);
235 spin_lock_init(&dev->desc_lock);
237 dev->cq_tbl = kcalloc(dev->dsr->caps.max_cq, sizeof(struct pvrdma_cq *),
238 GFP_KERNEL);
239 if (!dev->cq_tbl)
240 return ret;
241 spin_lock_init(&dev->cq_tbl_lock);
243 dev->qp_tbl = kcalloc(dev->dsr->caps.max_qp, sizeof(struct pvrdma_qp *),
244 GFP_KERNEL);
245 if (!dev->qp_tbl)
246 goto err_cq_free;
247 spin_lock_init(&dev->qp_tbl_lock);
249 /* Check if SRQ is supported by backend */
250 if (dev->dsr->caps.max_srq) {
251 dev->ib_dev.uverbs_cmd_mask |=
252 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) |
253 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) |
254 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) |
255 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) |
256 (1ull << IB_USER_VERBS_CMD_POST_SRQ_RECV);
258 dev->ib_dev.create_srq = pvrdma_create_srq;
259 dev->ib_dev.modify_srq = pvrdma_modify_srq;
260 dev->ib_dev.query_srq = pvrdma_query_srq;
261 dev->ib_dev.destroy_srq = pvrdma_destroy_srq;
263 dev->srq_tbl = kcalloc(dev->dsr->caps.max_srq,
264 sizeof(struct pvrdma_srq *),
265 GFP_KERNEL);
266 if (!dev->srq_tbl)
267 goto err_qp_free;
269 dev->ib_dev.driver_id = RDMA_DRIVER_VMW_PVRDMA;
270 spin_lock_init(&dev->srq_tbl_lock);
271 rdma_set_device_sysfs_group(&dev->ib_dev, &pvrdma_attr_group);
273 ret = ib_register_device(&dev->ib_dev, "vmw_pvrdma%d", NULL);
274 if (ret)
275 goto err_srq_free;
277 dev->ib_active = true;
279 return 0;
281 err_srq_free:
282 kfree(dev->srq_tbl);
283 err_qp_free:
284 kfree(dev->qp_tbl);
285 err_cq_free:
286 kfree(dev->cq_tbl);
288 return ret;
291 static irqreturn_t pvrdma_intr0_handler(int irq, void *dev_id)
293 u32 icr = PVRDMA_INTR_CAUSE_RESPONSE;
294 struct pvrdma_dev *dev = dev_id;
296 dev_dbg(&dev->pdev->dev, "interrupt 0 (response) handler\n");
298 if (!dev->pdev->msix_enabled) {
299 /* Legacy intr */
300 icr = pvrdma_read_reg(dev, PVRDMA_REG_ICR);
301 if (icr == 0)
302 return IRQ_NONE;
305 if (icr == PVRDMA_INTR_CAUSE_RESPONSE)
306 complete(&dev->cmd_done);
308 return IRQ_HANDLED;
311 static void pvrdma_qp_event(struct pvrdma_dev *dev, u32 qpn, int type)
313 struct pvrdma_qp *qp;
314 unsigned long flags;
316 spin_lock_irqsave(&dev->qp_tbl_lock, flags);
317 qp = dev->qp_tbl[qpn % dev->dsr->caps.max_qp];
318 if (qp)
319 refcount_inc(&qp->refcnt);
320 spin_unlock_irqrestore(&dev->qp_tbl_lock, flags);
322 if (qp && qp->ibqp.event_handler) {
323 struct ib_qp *ibqp = &qp->ibqp;
324 struct ib_event e;
326 e.device = ibqp->device;
327 e.element.qp = ibqp;
328 e.event = type; /* 1:1 mapping for now. */
329 ibqp->event_handler(&e, ibqp->qp_context);
331 if (qp) {
332 if (refcount_dec_and_test(&qp->refcnt))
333 complete(&qp->free);
337 static void pvrdma_cq_event(struct pvrdma_dev *dev, u32 cqn, int type)
339 struct pvrdma_cq *cq;
340 unsigned long flags;
342 spin_lock_irqsave(&dev->cq_tbl_lock, flags);
343 cq = dev->cq_tbl[cqn % dev->dsr->caps.max_cq];
344 if (cq)
345 refcount_inc(&cq->refcnt);
346 spin_unlock_irqrestore(&dev->cq_tbl_lock, flags);
348 if (cq && cq->ibcq.event_handler) {
349 struct ib_cq *ibcq = &cq->ibcq;
350 struct ib_event e;
352 e.device = ibcq->device;
353 e.element.cq = ibcq;
354 e.event = type; /* 1:1 mapping for now. */
355 ibcq->event_handler(&e, ibcq->cq_context);
357 if (cq) {
358 if (refcount_dec_and_test(&cq->refcnt))
359 complete(&cq->free);
363 static void pvrdma_srq_event(struct pvrdma_dev *dev, u32 srqn, int type)
365 struct pvrdma_srq *srq;
366 unsigned long flags;
368 spin_lock_irqsave(&dev->srq_tbl_lock, flags);
369 if (dev->srq_tbl)
370 srq = dev->srq_tbl[srqn % dev->dsr->caps.max_srq];
371 else
372 srq = NULL;
373 if (srq)
374 refcount_inc(&srq->refcnt);
375 spin_unlock_irqrestore(&dev->srq_tbl_lock, flags);
377 if (srq && srq->ibsrq.event_handler) {
378 struct ib_srq *ibsrq = &srq->ibsrq;
379 struct ib_event e;
381 e.device = ibsrq->device;
382 e.element.srq = ibsrq;
383 e.event = type; /* 1:1 mapping for now. */
384 ibsrq->event_handler(&e, ibsrq->srq_context);
386 if (srq) {
387 if (refcount_dec_and_test(&srq->refcnt))
388 complete(&srq->free);
392 static void pvrdma_dispatch_event(struct pvrdma_dev *dev, int port,
393 enum ib_event_type event)
395 struct ib_event ib_event;
397 memset(&ib_event, 0, sizeof(ib_event));
398 ib_event.device = &dev->ib_dev;
399 ib_event.element.port_num = port;
400 ib_event.event = event;
401 ib_dispatch_event(&ib_event);
404 static void pvrdma_dev_event(struct pvrdma_dev *dev, u8 port, int type)
406 if (port < 1 || port > dev->dsr->caps.phys_port_cnt) {
407 dev_warn(&dev->pdev->dev, "event on port %d\n", port);
408 return;
411 pvrdma_dispatch_event(dev, port, type);
414 static inline struct pvrdma_eqe *get_eqe(struct pvrdma_dev *dev, unsigned int i)
416 return (struct pvrdma_eqe *)pvrdma_page_dir_get_ptr(
417 &dev->async_pdir,
418 PAGE_SIZE +
419 sizeof(struct pvrdma_eqe) * i);
422 static irqreturn_t pvrdma_intr1_handler(int irq, void *dev_id)
424 struct pvrdma_dev *dev = dev_id;
425 struct pvrdma_ring *ring = &dev->async_ring_state->rx;
426 int ring_slots = (dev->dsr->async_ring_pages.num_pages - 1) *
427 PAGE_SIZE / sizeof(struct pvrdma_eqe);
428 unsigned int head;
430 dev_dbg(&dev->pdev->dev, "interrupt 1 (async event) handler\n");
433 * Don't process events until the IB device is registered. Otherwise
434 * we'll try to ib_dispatch_event() on an invalid device.
436 if (!dev->ib_active)
437 return IRQ_HANDLED;
439 while (pvrdma_idx_ring_has_data(ring, ring_slots, &head) > 0) {
440 struct pvrdma_eqe *eqe;
442 eqe = get_eqe(dev, head);
444 switch (eqe->type) {
445 case PVRDMA_EVENT_QP_FATAL:
446 case PVRDMA_EVENT_QP_REQ_ERR:
447 case PVRDMA_EVENT_QP_ACCESS_ERR:
448 case PVRDMA_EVENT_COMM_EST:
449 case PVRDMA_EVENT_SQ_DRAINED:
450 case PVRDMA_EVENT_PATH_MIG:
451 case PVRDMA_EVENT_PATH_MIG_ERR:
452 case PVRDMA_EVENT_QP_LAST_WQE_REACHED:
453 pvrdma_qp_event(dev, eqe->info, eqe->type);
454 break;
456 case PVRDMA_EVENT_CQ_ERR:
457 pvrdma_cq_event(dev, eqe->info, eqe->type);
458 break;
460 case PVRDMA_EVENT_SRQ_ERR:
461 case PVRDMA_EVENT_SRQ_LIMIT_REACHED:
462 pvrdma_srq_event(dev, eqe->info, eqe->type);
463 break;
465 case PVRDMA_EVENT_PORT_ACTIVE:
466 case PVRDMA_EVENT_PORT_ERR:
467 case PVRDMA_EVENT_LID_CHANGE:
468 case PVRDMA_EVENT_PKEY_CHANGE:
469 case PVRDMA_EVENT_SM_CHANGE:
470 case PVRDMA_EVENT_CLIENT_REREGISTER:
471 case PVRDMA_EVENT_GID_CHANGE:
472 pvrdma_dev_event(dev, eqe->info, eqe->type);
473 break;
475 case PVRDMA_EVENT_DEVICE_FATAL:
476 pvrdma_dev_event(dev, 1, eqe->type);
477 break;
479 default:
480 break;
483 pvrdma_idx_ring_inc(&ring->cons_head, ring_slots);
486 return IRQ_HANDLED;
489 static inline struct pvrdma_cqne *get_cqne(struct pvrdma_dev *dev,
490 unsigned int i)
492 return (struct pvrdma_cqne *)pvrdma_page_dir_get_ptr(
493 &dev->cq_pdir,
494 PAGE_SIZE +
495 sizeof(struct pvrdma_cqne) * i);
498 static irqreturn_t pvrdma_intrx_handler(int irq, void *dev_id)
500 struct pvrdma_dev *dev = dev_id;
501 struct pvrdma_ring *ring = &dev->cq_ring_state->rx;
502 int ring_slots = (dev->dsr->cq_ring_pages.num_pages - 1) * PAGE_SIZE /
503 sizeof(struct pvrdma_cqne);
504 unsigned int head;
505 unsigned long flags;
507 dev_dbg(&dev->pdev->dev, "interrupt x (completion) handler\n");
509 while (pvrdma_idx_ring_has_data(ring, ring_slots, &head) > 0) {
510 struct pvrdma_cqne *cqne;
511 struct pvrdma_cq *cq;
513 cqne = get_cqne(dev, head);
514 spin_lock_irqsave(&dev->cq_tbl_lock, flags);
515 cq = dev->cq_tbl[cqne->info % dev->dsr->caps.max_cq];
516 if (cq)
517 refcount_inc(&cq->refcnt);
518 spin_unlock_irqrestore(&dev->cq_tbl_lock, flags);
520 if (cq && cq->ibcq.comp_handler)
521 cq->ibcq.comp_handler(&cq->ibcq, cq->ibcq.cq_context);
522 if (cq) {
523 if (refcount_dec_and_test(&cq->refcnt))
524 complete(&cq->free);
526 pvrdma_idx_ring_inc(&ring->cons_head, ring_slots);
529 return IRQ_HANDLED;
532 static void pvrdma_free_irq(struct pvrdma_dev *dev)
534 int i;
536 dev_dbg(&dev->pdev->dev, "freeing interrupts\n");
537 for (i = 0; i < dev->nr_vectors; i++)
538 free_irq(pci_irq_vector(dev->pdev, i), dev);
541 static void pvrdma_enable_intrs(struct pvrdma_dev *dev)
543 dev_dbg(&dev->pdev->dev, "enable interrupts\n");
544 pvrdma_write_reg(dev, PVRDMA_REG_IMR, 0);
547 static void pvrdma_disable_intrs(struct pvrdma_dev *dev)
549 dev_dbg(&dev->pdev->dev, "disable interrupts\n");
550 pvrdma_write_reg(dev, PVRDMA_REG_IMR, ~0);
553 static int pvrdma_alloc_intrs(struct pvrdma_dev *dev)
555 struct pci_dev *pdev = dev->pdev;
556 int ret = 0, i;
558 ret = pci_alloc_irq_vectors(pdev, 1, PVRDMA_MAX_INTERRUPTS,
559 PCI_IRQ_MSIX);
560 if (ret < 0) {
561 ret = pci_alloc_irq_vectors(pdev, 1, 1,
562 PCI_IRQ_MSI | PCI_IRQ_LEGACY);
563 if (ret < 0)
564 return ret;
566 dev->nr_vectors = ret;
568 ret = request_irq(pci_irq_vector(dev->pdev, 0), pvrdma_intr0_handler,
569 pdev->msix_enabled ? 0 : IRQF_SHARED, DRV_NAME, dev);
570 if (ret) {
571 dev_err(&dev->pdev->dev,
572 "failed to request interrupt 0\n");
573 goto out_free_vectors;
576 for (i = 1; i < dev->nr_vectors; i++) {
577 ret = request_irq(pci_irq_vector(dev->pdev, i),
578 i == 1 ? pvrdma_intr1_handler :
579 pvrdma_intrx_handler,
580 0, DRV_NAME, dev);
581 if (ret) {
582 dev_err(&dev->pdev->dev,
583 "failed to request interrupt %d\n", i);
584 goto free_irqs;
588 return 0;
590 free_irqs:
591 while (--i >= 0)
592 free_irq(pci_irq_vector(dev->pdev, i), dev);
593 out_free_vectors:
594 pci_free_irq_vectors(pdev);
595 return ret;
598 static void pvrdma_free_slots(struct pvrdma_dev *dev)
600 struct pci_dev *pdev = dev->pdev;
602 if (dev->resp_slot)
603 dma_free_coherent(&pdev->dev, PAGE_SIZE, dev->resp_slot,
604 dev->dsr->resp_slot_dma);
605 if (dev->cmd_slot)
606 dma_free_coherent(&pdev->dev, PAGE_SIZE, dev->cmd_slot,
607 dev->dsr->cmd_slot_dma);
610 static int pvrdma_add_gid_at_index(struct pvrdma_dev *dev,
611 const union ib_gid *gid,
612 u8 gid_type,
613 int index)
615 int ret;
616 union pvrdma_cmd_req req;
617 struct pvrdma_cmd_create_bind *cmd_bind = &req.create_bind;
619 if (!dev->sgid_tbl) {
620 dev_warn(&dev->pdev->dev, "sgid table not initialized\n");
621 return -EINVAL;
624 memset(cmd_bind, 0, sizeof(*cmd_bind));
625 cmd_bind->hdr.cmd = PVRDMA_CMD_CREATE_BIND;
626 memcpy(cmd_bind->new_gid, gid->raw, 16);
627 cmd_bind->mtu = ib_mtu_enum_to_int(IB_MTU_1024);
628 cmd_bind->vlan = 0xfff;
629 cmd_bind->index = index;
630 cmd_bind->gid_type = gid_type;
632 ret = pvrdma_cmd_post(dev, &req, NULL, 0);
633 if (ret < 0) {
634 dev_warn(&dev->pdev->dev,
635 "could not create binding, error: %d\n", ret);
636 return -EFAULT;
638 memcpy(&dev->sgid_tbl[index], gid, sizeof(*gid));
639 return 0;
642 static int pvrdma_add_gid(const struct ib_gid_attr *attr, void **context)
644 struct pvrdma_dev *dev = to_vdev(attr->device);
646 return pvrdma_add_gid_at_index(dev, &attr->gid,
647 ib_gid_type_to_pvrdma(attr->gid_type),
648 attr->index);
651 static int pvrdma_del_gid_at_index(struct pvrdma_dev *dev, int index)
653 int ret;
654 union pvrdma_cmd_req req;
655 struct pvrdma_cmd_destroy_bind *cmd_dest = &req.destroy_bind;
657 /* Update sgid table. */
658 if (!dev->sgid_tbl) {
659 dev_warn(&dev->pdev->dev, "sgid table not initialized\n");
660 return -EINVAL;
663 memset(cmd_dest, 0, sizeof(*cmd_dest));
664 cmd_dest->hdr.cmd = PVRDMA_CMD_DESTROY_BIND;
665 memcpy(cmd_dest->dest_gid, &dev->sgid_tbl[index], 16);
666 cmd_dest->index = index;
668 ret = pvrdma_cmd_post(dev, &req, NULL, 0);
669 if (ret < 0) {
670 dev_warn(&dev->pdev->dev,
671 "could not destroy binding, error: %d\n", ret);
672 return ret;
674 memset(&dev->sgid_tbl[index], 0, 16);
675 return 0;
678 static int pvrdma_del_gid(const struct ib_gid_attr *attr, void **context)
680 struct pvrdma_dev *dev = to_vdev(attr->device);
682 dev_dbg(&dev->pdev->dev, "removing gid at index %u from %s",
683 attr->index, dev->netdev->name);
685 return pvrdma_del_gid_at_index(dev, attr->index);
688 static void pvrdma_netdevice_event_handle(struct pvrdma_dev *dev,
689 struct net_device *ndev,
690 unsigned long event)
692 struct pci_dev *pdev_net;
693 unsigned int slot;
695 switch (event) {
696 case NETDEV_REBOOT:
697 case NETDEV_DOWN:
698 pvrdma_dispatch_event(dev, 1, IB_EVENT_PORT_ERR);
699 break;
700 case NETDEV_UP:
701 pvrdma_write_reg(dev, PVRDMA_REG_CTL,
702 PVRDMA_DEVICE_CTL_UNQUIESCE);
704 mb();
706 if (pvrdma_read_reg(dev, PVRDMA_REG_ERR))
707 dev_err(&dev->pdev->dev,
708 "failed to activate device during link up\n");
709 else
710 pvrdma_dispatch_event(dev, 1, IB_EVENT_PORT_ACTIVE);
711 break;
712 case NETDEV_UNREGISTER:
713 dev_put(dev->netdev);
714 dev->netdev = NULL;
715 break;
716 case NETDEV_REGISTER:
717 /* vmxnet3 will have same bus, slot. But func will be 0 */
718 slot = PCI_SLOT(dev->pdev->devfn);
719 pdev_net = pci_get_slot(dev->pdev->bus,
720 PCI_DEVFN(slot, 0));
721 if ((dev->netdev == NULL) &&
722 (pci_get_drvdata(pdev_net) == ndev)) {
723 /* this is our netdev */
724 dev->netdev = ndev;
725 dev_hold(ndev);
727 pci_dev_put(pdev_net);
728 break;
730 default:
731 dev_dbg(&dev->pdev->dev, "ignore netdevice event %ld on %s\n",
732 event, dev_name(&dev->ib_dev.dev));
733 break;
737 static void pvrdma_netdevice_event_work(struct work_struct *work)
739 struct pvrdma_netdevice_work *netdev_work;
740 struct pvrdma_dev *dev;
742 netdev_work = container_of(work, struct pvrdma_netdevice_work, work);
744 mutex_lock(&pvrdma_device_list_lock);
745 list_for_each_entry(dev, &pvrdma_device_list, device_link) {
746 if ((netdev_work->event == NETDEV_REGISTER) ||
747 (dev->netdev == netdev_work->event_netdev)) {
748 pvrdma_netdevice_event_handle(dev,
749 netdev_work->event_netdev,
750 netdev_work->event);
751 break;
754 mutex_unlock(&pvrdma_device_list_lock);
756 kfree(netdev_work);
759 static int pvrdma_netdevice_event(struct notifier_block *this,
760 unsigned long event, void *ptr)
762 struct net_device *event_netdev = netdev_notifier_info_to_dev(ptr);
763 struct pvrdma_netdevice_work *netdev_work;
765 netdev_work = kmalloc(sizeof(*netdev_work), GFP_ATOMIC);
766 if (!netdev_work)
767 return NOTIFY_BAD;
769 INIT_WORK(&netdev_work->work, pvrdma_netdevice_event_work);
770 netdev_work->event_netdev = event_netdev;
771 netdev_work->event = event;
772 queue_work(event_wq, &netdev_work->work);
774 return NOTIFY_DONE;
777 static int pvrdma_pci_probe(struct pci_dev *pdev,
778 const struct pci_device_id *id)
780 struct pci_dev *pdev_net;
781 struct pvrdma_dev *dev;
782 int ret;
783 unsigned long start;
784 unsigned long len;
785 dma_addr_t slot_dma = 0;
787 dev_dbg(&pdev->dev, "initializing driver %s\n", pci_name(pdev));
789 /* Allocate zero-out device */
790 dev = (struct pvrdma_dev *)ib_alloc_device(sizeof(*dev));
791 if (!dev) {
792 dev_err(&pdev->dev, "failed to allocate IB device\n");
793 return -ENOMEM;
796 mutex_lock(&pvrdma_device_list_lock);
797 list_add(&dev->device_link, &pvrdma_device_list);
798 mutex_unlock(&pvrdma_device_list_lock);
800 ret = pvrdma_init_device(dev);
801 if (ret)
802 goto err_free_device;
804 dev->pdev = pdev;
805 pci_set_drvdata(pdev, dev);
807 ret = pci_enable_device(pdev);
808 if (ret) {
809 dev_err(&pdev->dev, "cannot enable PCI device\n");
810 goto err_free_device;
813 dev_dbg(&pdev->dev, "PCI resource flags BAR0 %#lx\n",
814 pci_resource_flags(pdev, 0));
815 dev_dbg(&pdev->dev, "PCI resource len %#llx\n",
816 (unsigned long long)pci_resource_len(pdev, 0));
817 dev_dbg(&pdev->dev, "PCI resource start %#llx\n",
818 (unsigned long long)pci_resource_start(pdev, 0));
819 dev_dbg(&pdev->dev, "PCI resource flags BAR1 %#lx\n",
820 pci_resource_flags(pdev, 1));
821 dev_dbg(&pdev->dev, "PCI resource len %#llx\n",
822 (unsigned long long)pci_resource_len(pdev, 1));
823 dev_dbg(&pdev->dev, "PCI resource start %#llx\n",
824 (unsigned long long)pci_resource_start(pdev, 1));
826 if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM) ||
827 !(pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
828 dev_err(&pdev->dev, "PCI BAR region not MMIO\n");
829 ret = -ENOMEM;
830 goto err_free_device;
833 ret = pci_request_regions(pdev, DRV_NAME);
834 if (ret) {
835 dev_err(&pdev->dev, "cannot request PCI resources\n");
836 goto err_disable_pdev;
839 /* Enable 64-Bit DMA */
840 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
841 ret = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
842 if (ret != 0) {
843 dev_err(&pdev->dev,
844 "pci_set_consistent_dma_mask failed\n");
845 goto err_free_resource;
847 } else {
848 ret = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
849 if (ret != 0) {
850 dev_err(&pdev->dev,
851 "pci_set_dma_mask failed\n");
852 goto err_free_resource;
856 pci_set_master(pdev);
858 /* Map register space */
859 start = pci_resource_start(dev->pdev, PVRDMA_PCI_RESOURCE_REG);
860 len = pci_resource_len(dev->pdev, PVRDMA_PCI_RESOURCE_REG);
861 dev->regs = ioremap(start, len);
862 if (!dev->regs) {
863 dev_err(&pdev->dev, "register mapping failed\n");
864 ret = -ENOMEM;
865 goto err_free_resource;
868 /* Setup per-device UAR. */
869 dev->driver_uar.index = 0;
870 dev->driver_uar.pfn =
871 pci_resource_start(dev->pdev, PVRDMA_PCI_RESOURCE_UAR) >>
872 PAGE_SHIFT;
873 dev->driver_uar.map =
874 ioremap(dev->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
875 if (!dev->driver_uar.map) {
876 dev_err(&pdev->dev, "failed to remap UAR pages\n");
877 ret = -ENOMEM;
878 goto err_unmap_regs;
881 dev->dsr_version = pvrdma_read_reg(dev, PVRDMA_REG_VERSION);
882 dev_info(&pdev->dev, "device version %d, driver version %d\n",
883 dev->dsr_version, PVRDMA_VERSION);
885 dev->dsr = dma_zalloc_coherent(&pdev->dev, sizeof(*dev->dsr),
886 &dev->dsrbase, GFP_KERNEL);
887 if (!dev->dsr) {
888 dev_err(&pdev->dev, "failed to allocate shared region\n");
889 ret = -ENOMEM;
890 goto err_uar_unmap;
893 /* Setup the shared region */
894 dev->dsr->driver_version = PVRDMA_VERSION;
895 dev->dsr->gos_info.gos_bits = sizeof(void *) == 4 ?
896 PVRDMA_GOS_BITS_32 :
897 PVRDMA_GOS_BITS_64;
898 dev->dsr->gos_info.gos_type = PVRDMA_GOS_TYPE_LINUX;
899 dev->dsr->gos_info.gos_ver = 1;
900 dev->dsr->uar_pfn = dev->driver_uar.pfn;
902 /* Command slot. */
903 dev->cmd_slot = dma_alloc_coherent(&pdev->dev, PAGE_SIZE,
904 &slot_dma, GFP_KERNEL);
905 if (!dev->cmd_slot) {
906 ret = -ENOMEM;
907 goto err_free_dsr;
910 dev->dsr->cmd_slot_dma = (u64)slot_dma;
912 /* Response slot. */
913 dev->resp_slot = dma_alloc_coherent(&pdev->dev, PAGE_SIZE,
914 &slot_dma, GFP_KERNEL);
915 if (!dev->resp_slot) {
916 ret = -ENOMEM;
917 goto err_free_slots;
920 dev->dsr->resp_slot_dma = (u64)slot_dma;
922 /* Async event ring */
923 dev->dsr->async_ring_pages.num_pages = PVRDMA_NUM_RING_PAGES;
924 ret = pvrdma_page_dir_init(dev, &dev->async_pdir,
925 dev->dsr->async_ring_pages.num_pages, true);
926 if (ret)
927 goto err_free_slots;
928 dev->async_ring_state = dev->async_pdir.pages[0];
929 dev->dsr->async_ring_pages.pdir_dma = dev->async_pdir.dir_dma;
931 /* CQ notification ring */
932 dev->dsr->cq_ring_pages.num_pages = PVRDMA_NUM_RING_PAGES;
933 ret = pvrdma_page_dir_init(dev, &dev->cq_pdir,
934 dev->dsr->cq_ring_pages.num_pages, true);
935 if (ret)
936 goto err_free_async_ring;
937 dev->cq_ring_state = dev->cq_pdir.pages[0];
938 dev->dsr->cq_ring_pages.pdir_dma = dev->cq_pdir.dir_dma;
941 * Write the PA of the shared region to the device. The writes must be
942 * ordered such that the high bits are written last. When the writes
943 * complete, the device will have filled out the capabilities.
946 pvrdma_write_reg(dev, PVRDMA_REG_DSRLOW, (u32)dev->dsrbase);
947 pvrdma_write_reg(dev, PVRDMA_REG_DSRHIGH,
948 (u32)((u64)(dev->dsrbase) >> 32));
950 /* Make sure the write is complete before reading status. */
951 mb();
953 /* The driver supports RoCE V1 and V2. */
954 if (!PVRDMA_SUPPORTED(dev)) {
955 dev_err(&pdev->dev, "driver needs RoCE v1 or v2 support\n");
956 ret = -EFAULT;
957 goto err_free_cq_ring;
960 /* Paired vmxnet3 will have same bus, slot. But func will be 0 */
961 pdev_net = pci_get_slot(pdev->bus, PCI_DEVFN(PCI_SLOT(pdev->devfn), 0));
962 if (!pdev_net) {
963 dev_err(&pdev->dev, "failed to find paired net device\n");
964 ret = -ENODEV;
965 goto err_free_cq_ring;
968 if (pdev_net->vendor != PCI_VENDOR_ID_VMWARE ||
969 pdev_net->device != PCI_DEVICE_ID_VMWARE_VMXNET3) {
970 dev_err(&pdev->dev, "failed to find paired vmxnet3 device\n");
971 pci_dev_put(pdev_net);
972 ret = -ENODEV;
973 goto err_free_cq_ring;
976 dev->netdev = pci_get_drvdata(pdev_net);
977 pci_dev_put(pdev_net);
978 if (!dev->netdev) {
979 dev_err(&pdev->dev, "failed to get vmxnet3 device\n");
980 ret = -ENODEV;
981 goto err_free_cq_ring;
983 dev_hold(dev->netdev);
985 dev_info(&pdev->dev, "paired device to %s\n", dev->netdev->name);
987 /* Interrupt setup */
988 ret = pvrdma_alloc_intrs(dev);
989 if (ret) {
990 dev_err(&pdev->dev, "failed to allocate interrupts\n");
991 ret = -ENOMEM;
992 goto err_free_cq_ring;
995 /* Allocate UAR table. */
996 ret = pvrdma_uar_table_init(dev);
997 if (ret) {
998 dev_err(&pdev->dev, "failed to allocate UAR table\n");
999 ret = -ENOMEM;
1000 goto err_free_intrs;
1003 /* Allocate GID table */
1004 dev->sgid_tbl = kcalloc(dev->dsr->caps.gid_tbl_len,
1005 sizeof(union ib_gid), GFP_KERNEL);
1006 if (!dev->sgid_tbl) {
1007 ret = -ENOMEM;
1008 goto err_free_uar_table;
1010 dev_dbg(&pdev->dev, "gid table len %d\n", dev->dsr->caps.gid_tbl_len);
1012 pvrdma_enable_intrs(dev);
1014 /* Activate pvrdma device */
1015 pvrdma_write_reg(dev, PVRDMA_REG_CTL, PVRDMA_DEVICE_CTL_ACTIVATE);
1017 /* Make sure the write is complete before reading status. */
1018 mb();
1020 /* Check if device was successfully activated */
1021 ret = pvrdma_read_reg(dev, PVRDMA_REG_ERR);
1022 if (ret != 0) {
1023 dev_err(&pdev->dev, "failed to activate device\n");
1024 ret = -EFAULT;
1025 goto err_disable_intr;
1028 /* Register IB device */
1029 ret = pvrdma_register_device(dev);
1030 if (ret) {
1031 dev_err(&pdev->dev, "failed to register IB device\n");
1032 goto err_disable_intr;
1035 dev->nb_netdev.notifier_call = pvrdma_netdevice_event;
1036 ret = register_netdevice_notifier(&dev->nb_netdev);
1037 if (ret) {
1038 dev_err(&pdev->dev, "failed to register netdevice events\n");
1039 goto err_unreg_ibdev;
1042 dev_info(&pdev->dev, "attached to device\n");
1043 return 0;
1045 err_unreg_ibdev:
1046 ib_unregister_device(&dev->ib_dev);
1047 err_disable_intr:
1048 pvrdma_disable_intrs(dev);
1049 kfree(dev->sgid_tbl);
1050 err_free_uar_table:
1051 pvrdma_uar_table_cleanup(dev);
1052 err_free_intrs:
1053 pvrdma_free_irq(dev);
1054 pci_free_irq_vectors(pdev);
1055 err_free_cq_ring:
1056 if (dev->netdev) {
1057 dev_put(dev->netdev);
1058 dev->netdev = NULL;
1060 pvrdma_page_dir_cleanup(dev, &dev->cq_pdir);
1061 err_free_async_ring:
1062 pvrdma_page_dir_cleanup(dev, &dev->async_pdir);
1063 err_free_slots:
1064 pvrdma_free_slots(dev);
1065 err_free_dsr:
1066 dma_free_coherent(&pdev->dev, sizeof(*dev->dsr), dev->dsr,
1067 dev->dsrbase);
1068 err_uar_unmap:
1069 iounmap(dev->driver_uar.map);
1070 err_unmap_regs:
1071 iounmap(dev->regs);
1072 err_free_resource:
1073 pci_release_regions(pdev);
1074 err_disable_pdev:
1075 pci_disable_device(pdev);
1076 pci_set_drvdata(pdev, NULL);
1077 err_free_device:
1078 mutex_lock(&pvrdma_device_list_lock);
1079 list_del(&dev->device_link);
1080 mutex_unlock(&pvrdma_device_list_lock);
1081 ib_dealloc_device(&dev->ib_dev);
1082 return ret;
1085 static void pvrdma_pci_remove(struct pci_dev *pdev)
1087 struct pvrdma_dev *dev = pci_get_drvdata(pdev);
1089 if (!dev)
1090 return;
1092 dev_info(&pdev->dev, "detaching from device\n");
1094 unregister_netdevice_notifier(&dev->nb_netdev);
1095 dev->nb_netdev.notifier_call = NULL;
1097 flush_workqueue(event_wq);
1099 if (dev->netdev) {
1100 dev_put(dev->netdev);
1101 dev->netdev = NULL;
1104 /* Unregister ib device */
1105 ib_unregister_device(&dev->ib_dev);
1107 mutex_lock(&pvrdma_device_list_lock);
1108 list_del(&dev->device_link);
1109 mutex_unlock(&pvrdma_device_list_lock);
1111 pvrdma_disable_intrs(dev);
1112 pvrdma_free_irq(dev);
1113 pci_free_irq_vectors(pdev);
1115 /* Deactivate pvrdma device */
1116 pvrdma_write_reg(dev, PVRDMA_REG_CTL, PVRDMA_DEVICE_CTL_RESET);
1117 pvrdma_page_dir_cleanup(dev, &dev->cq_pdir);
1118 pvrdma_page_dir_cleanup(dev, &dev->async_pdir);
1119 pvrdma_free_slots(dev);
1121 iounmap(dev->regs);
1122 kfree(dev->sgid_tbl);
1123 kfree(dev->cq_tbl);
1124 kfree(dev->srq_tbl);
1125 kfree(dev->qp_tbl);
1126 pvrdma_uar_table_cleanup(dev);
1127 iounmap(dev->driver_uar.map);
1129 ib_dealloc_device(&dev->ib_dev);
1131 /* Free pci resources */
1132 pci_release_regions(pdev);
1133 pci_disable_device(pdev);
1134 pci_set_drvdata(pdev, NULL);
1137 static const struct pci_device_id pvrdma_pci_table[] = {
1138 { PCI_DEVICE(PCI_VENDOR_ID_VMWARE, PCI_DEVICE_ID_VMWARE_PVRDMA), },
1139 { 0 },
1142 MODULE_DEVICE_TABLE(pci, pvrdma_pci_table);
1144 static struct pci_driver pvrdma_driver = {
1145 .name = DRV_NAME,
1146 .id_table = pvrdma_pci_table,
1147 .probe = pvrdma_pci_probe,
1148 .remove = pvrdma_pci_remove,
1151 static int __init pvrdma_init(void)
1153 int err;
1155 event_wq = alloc_ordered_workqueue("pvrdma_event_wq", WQ_MEM_RECLAIM);
1156 if (!event_wq)
1157 return -ENOMEM;
1159 err = pci_register_driver(&pvrdma_driver);
1160 if (err)
1161 destroy_workqueue(event_wq);
1163 return err;
1166 static void __exit pvrdma_cleanup(void)
1168 pci_unregister_driver(&pvrdma_driver);
1170 destroy_workqueue(event_wq);
1173 module_init(pvrdma_init);
1174 module_exit(pvrdma_cleanup);
1176 MODULE_AUTHOR("VMware, Inc");
1177 MODULE_DESCRIPTION("VMware Paravirtual RDMA driver");
1178 MODULE_LICENSE("Dual BSD/GPL");