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
23 * - Redistributions of source code must retain the above
24 * copyright notice, this list of conditions and the following
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>
58 #define DRV_NAME "vmw_pvrdma"
59 #define DRV_VERSION "1.0.0.0-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(struct ib_device *ibdev,
68 const union ib_gid *gid,
69 const struct ib_gid_attr *attr,
71 static int pvrdma_del_gid(struct ib_device *ibdev,
77 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
80 return sprintf(buf, "VMW_PVRDMA-%s\n", DRV_VERSION);
83 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
86 return sprintf(buf, "%d\n", PVRDMA_REV_ID);
89 static ssize_t show_board(struct device *device, struct device_attribute *attr,
92 return sprintf(buf, "%d\n", PVRDMA_BOARD_ID);
95 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
96 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
97 static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
99 static struct device_attribute *pvrdma_class_attributes[] = {
105 static void pvrdma_get_fw_ver_str(struct ib_device *device, char *str,
108 struct pvrdma_dev *dev =
109 container_of(device, struct pvrdma_dev, ib_dev);
110 snprintf(str, str_len, "%d.%d.%d\n",
111 (int) (dev->dsr->caps.fw_ver >> 32),
112 (int) (dev->dsr->caps.fw_ver >> 16) & 0xffff,
113 (int) dev->dsr->caps.fw_ver & 0xffff);
116 static int pvrdma_init_device(struct pvrdma_dev *dev)
118 /* Initialize some device related stuff */
119 spin_lock_init(&dev->cmd_lock);
120 sema_init(&dev->cmd_sema, 1);
121 atomic_set(&dev->num_qps, 0);
122 atomic_set(&dev->num_cqs, 0);
123 atomic_set(&dev->num_pds, 0);
124 atomic_set(&dev->num_ahs, 0);
129 static int pvrdma_port_immutable(struct ib_device *ibdev, u8 port_num,
130 struct ib_port_immutable *immutable)
132 struct ib_port_attr attr;
135 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE;
137 err = ib_query_port(ibdev, port_num, &attr);
141 immutable->pkey_tbl_len = attr.pkey_tbl_len;
142 immutable->gid_tbl_len = attr.gid_tbl_len;
143 immutable->max_mad_size = IB_MGMT_MAD_SIZE;
147 static struct net_device *pvrdma_get_netdev(struct ib_device *ibdev,
150 struct net_device *netdev;
151 struct pvrdma_dev *dev = to_vdev(ibdev);
157 netdev = dev->netdev;
165 static int pvrdma_register_device(struct pvrdma_dev *dev)
170 strlcpy(dev->ib_dev.name, "vmw_pvrdma%d", IB_DEVICE_NAME_MAX);
171 dev->ib_dev.node_guid = dev->dsr->caps.node_guid;
172 dev->sys_image_guid = dev->dsr->caps.sys_image_guid;
174 dev->ib_dev.owner = THIS_MODULE;
175 dev->ib_dev.num_comp_vectors = 1;
176 dev->ib_dev.dev.parent = &dev->pdev->dev;
177 dev->ib_dev.uverbs_abi_ver = PVRDMA_UVERBS_ABI_VERSION;
178 dev->ib_dev.uverbs_cmd_mask =
179 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
180 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
181 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
182 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
183 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
184 (1ull << IB_USER_VERBS_CMD_REG_MR) |
185 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
186 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
187 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
188 (1ull << IB_USER_VERBS_CMD_POLL_CQ) |
189 (1ull << IB_USER_VERBS_CMD_REQ_NOTIFY_CQ) |
190 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
191 (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
192 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
193 (1ull << IB_USER_VERBS_CMD_QUERY_QP) |
194 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
195 (1ull << IB_USER_VERBS_CMD_POST_SEND) |
196 (1ull << IB_USER_VERBS_CMD_POST_RECV) |
197 (1ull << IB_USER_VERBS_CMD_CREATE_AH) |
198 (1ull << IB_USER_VERBS_CMD_DESTROY_AH);
200 dev->ib_dev.node_type = RDMA_NODE_IB_CA;
201 dev->ib_dev.phys_port_cnt = dev->dsr->caps.phys_port_cnt;
203 dev->ib_dev.query_device = pvrdma_query_device;
204 dev->ib_dev.query_port = pvrdma_query_port;
205 dev->ib_dev.query_gid = pvrdma_query_gid;
206 dev->ib_dev.query_pkey = pvrdma_query_pkey;
207 dev->ib_dev.modify_port = pvrdma_modify_port;
208 dev->ib_dev.alloc_ucontext = pvrdma_alloc_ucontext;
209 dev->ib_dev.dealloc_ucontext = pvrdma_dealloc_ucontext;
210 dev->ib_dev.mmap = pvrdma_mmap;
211 dev->ib_dev.alloc_pd = pvrdma_alloc_pd;
212 dev->ib_dev.dealloc_pd = pvrdma_dealloc_pd;
213 dev->ib_dev.create_ah = pvrdma_create_ah;
214 dev->ib_dev.destroy_ah = pvrdma_destroy_ah;
215 dev->ib_dev.create_qp = pvrdma_create_qp;
216 dev->ib_dev.modify_qp = pvrdma_modify_qp;
217 dev->ib_dev.query_qp = pvrdma_query_qp;
218 dev->ib_dev.destroy_qp = pvrdma_destroy_qp;
219 dev->ib_dev.post_send = pvrdma_post_send;
220 dev->ib_dev.post_recv = pvrdma_post_recv;
221 dev->ib_dev.create_cq = pvrdma_create_cq;
222 dev->ib_dev.modify_cq = pvrdma_modify_cq;
223 dev->ib_dev.resize_cq = pvrdma_resize_cq;
224 dev->ib_dev.destroy_cq = pvrdma_destroy_cq;
225 dev->ib_dev.poll_cq = pvrdma_poll_cq;
226 dev->ib_dev.req_notify_cq = pvrdma_req_notify_cq;
227 dev->ib_dev.get_dma_mr = pvrdma_get_dma_mr;
228 dev->ib_dev.reg_user_mr = pvrdma_reg_user_mr;
229 dev->ib_dev.dereg_mr = pvrdma_dereg_mr;
230 dev->ib_dev.alloc_mr = pvrdma_alloc_mr;
231 dev->ib_dev.map_mr_sg = pvrdma_map_mr_sg;
232 dev->ib_dev.add_gid = pvrdma_add_gid;
233 dev->ib_dev.del_gid = pvrdma_del_gid;
234 dev->ib_dev.get_netdev = pvrdma_get_netdev;
235 dev->ib_dev.get_port_immutable = pvrdma_port_immutable;
236 dev->ib_dev.get_link_layer = pvrdma_port_link_layer;
237 dev->ib_dev.get_dev_fw_str = pvrdma_get_fw_ver_str;
239 mutex_init(&dev->port_mutex);
240 spin_lock_init(&dev->desc_lock);
242 dev->cq_tbl = kcalloc(dev->dsr->caps.max_cq, sizeof(void *),
246 spin_lock_init(&dev->cq_tbl_lock);
248 dev->qp_tbl = kcalloc(dev->dsr->caps.max_qp, sizeof(void *),
252 spin_lock_init(&dev->qp_tbl_lock);
254 ret = ib_register_device(&dev->ib_dev, NULL);
258 for (i = 0; i < ARRAY_SIZE(pvrdma_class_attributes); ++i) {
259 ret = device_create_file(&dev->ib_dev.dev,
260 pvrdma_class_attributes[i]);
265 dev->ib_active = true;
270 ib_unregister_device(&dev->ib_dev);
279 static irqreturn_t pvrdma_intr0_handler(int irq, void *dev_id)
281 u32 icr = PVRDMA_INTR_CAUSE_RESPONSE;
282 struct pvrdma_dev *dev = dev_id;
284 dev_dbg(&dev->pdev->dev, "interrupt 0 (response) handler\n");
286 if (!dev->pdev->msix_enabled) {
288 icr = pvrdma_read_reg(dev, PVRDMA_REG_ICR);
293 if (icr == PVRDMA_INTR_CAUSE_RESPONSE)
294 complete(&dev->cmd_done);
299 static void pvrdma_qp_event(struct pvrdma_dev *dev, u32 qpn, int type)
301 struct pvrdma_qp *qp;
304 spin_lock_irqsave(&dev->qp_tbl_lock, flags);
305 qp = dev->qp_tbl[qpn % dev->dsr->caps.max_qp];
307 atomic_inc(&qp->refcnt);
308 spin_unlock_irqrestore(&dev->qp_tbl_lock, flags);
310 if (qp && qp->ibqp.event_handler) {
311 struct ib_qp *ibqp = &qp->ibqp;
314 e.device = ibqp->device;
316 e.event = type; /* 1:1 mapping for now. */
317 ibqp->event_handler(&e, ibqp->qp_context);
320 atomic_dec(&qp->refcnt);
321 if (atomic_read(&qp->refcnt) == 0)
326 static void pvrdma_cq_event(struct pvrdma_dev *dev, u32 cqn, int type)
328 struct pvrdma_cq *cq;
331 spin_lock_irqsave(&dev->cq_tbl_lock, flags);
332 cq = dev->cq_tbl[cqn % dev->dsr->caps.max_cq];
334 atomic_inc(&cq->refcnt);
335 spin_unlock_irqrestore(&dev->cq_tbl_lock, flags);
337 if (cq && cq->ibcq.event_handler) {
338 struct ib_cq *ibcq = &cq->ibcq;
341 e.device = ibcq->device;
343 e.event = type; /* 1:1 mapping for now. */
344 ibcq->event_handler(&e, ibcq->cq_context);
347 atomic_dec(&cq->refcnt);
348 if (atomic_read(&cq->refcnt) == 0)
353 static void pvrdma_dispatch_event(struct pvrdma_dev *dev, int port,
354 enum ib_event_type event)
356 struct ib_event ib_event;
358 memset(&ib_event, 0, sizeof(ib_event));
359 ib_event.device = &dev->ib_dev;
360 ib_event.element.port_num = port;
361 ib_event.event = event;
362 ib_dispatch_event(&ib_event);
365 static void pvrdma_dev_event(struct pvrdma_dev *dev, u8 port, int type)
367 if (port < 1 || port > dev->dsr->caps.phys_port_cnt) {
368 dev_warn(&dev->pdev->dev, "event on port %d\n", port);
372 pvrdma_dispatch_event(dev, port, type);
375 static inline struct pvrdma_eqe *get_eqe(struct pvrdma_dev *dev, unsigned int i)
377 return (struct pvrdma_eqe *)pvrdma_page_dir_get_ptr(
380 sizeof(struct pvrdma_eqe) * i);
383 static irqreturn_t pvrdma_intr1_handler(int irq, void *dev_id)
385 struct pvrdma_dev *dev = dev_id;
386 struct pvrdma_ring *ring = &dev->async_ring_state->rx;
387 int ring_slots = (dev->dsr->async_ring_pages.num_pages - 1) *
388 PAGE_SIZE / sizeof(struct pvrdma_eqe);
391 dev_dbg(&dev->pdev->dev, "interrupt 1 (async event) handler\n");
394 * Don't process events until the IB device is registered. Otherwise
395 * we'll try to ib_dispatch_event() on an invalid device.
400 while (pvrdma_idx_ring_has_data(ring, ring_slots, &head) > 0) {
401 struct pvrdma_eqe *eqe;
403 eqe = get_eqe(dev, head);
406 case PVRDMA_EVENT_QP_FATAL:
407 case PVRDMA_EVENT_QP_REQ_ERR:
408 case PVRDMA_EVENT_QP_ACCESS_ERR:
409 case PVRDMA_EVENT_COMM_EST:
410 case PVRDMA_EVENT_SQ_DRAINED:
411 case PVRDMA_EVENT_PATH_MIG:
412 case PVRDMA_EVENT_PATH_MIG_ERR:
413 case PVRDMA_EVENT_QP_LAST_WQE_REACHED:
414 pvrdma_qp_event(dev, eqe->info, eqe->type);
417 case PVRDMA_EVENT_CQ_ERR:
418 pvrdma_cq_event(dev, eqe->info, eqe->type);
421 case PVRDMA_EVENT_SRQ_ERR:
422 case PVRDMA_EVENT_SRQ_LIMIT_REACHED:
425 case PVRDMA_EVENT_PORT_ACTIVE:
426 case PVRDMA_EVENT_PORT_ERR:
427 case PVRDMA_EVENT_LID_CHANGE:
428 case PVRDMA_EVENT_PKEY_CHANGE:
429 case PVRDMA_EVENT_SM_CHANGE:
430 case PVRDMA_EVENT_CLIENT_REREGISTER:
431 case PVRDMA_EVENT_GID_CHANGE:
432 pvrdma_dev_event(dev, eqe->info, eqe->type);
435 case PVRDMA_EVENT_DEVICE_FATAL:
436 pvrdma_dev_event(dev, 1, eqe->type);
443 pvrdma_idx_ring_inc(&ring->cons_head, ring_slots);
449 static inline struct pvrdma_cqne *get_cqne(struct pvrdma_dev *dev,
452 return (struct pvrdma_cqne *)pvrdma_page_dir_get_ptr(
455 sizeof(struct pvrdma_cqne) * i);
458 static irqreturn_t pvrdma_intrx_handler(int irq, void *dev_id)
460 struct pvrdma_dev *dev = dev_id;
461 struct pvrdma_ring *ring = &dev->cq_ring_state->rx;
462 int ring_slots = (dev->dsr->cq_ring_pages.num_pages - 1) * PAGE_SIZE /
463 sizeof(struct pvrdma_cqne);
467 dev_dbg(&dev->pdev->dev, "interrupt x (completion) handler\n");
469 while (pvrdma_idx_ring_has_data(ring, ring_slots, &head) > 0) {
470 struct pvrdma_cqne *cqne;
471 struct pvrdma_cq *cq;
473 cqne = get_cqne(dev, head);
474 spin_lock_irqsave(&dev->cq_tbl_lock, flags);
475 cq = dev->cq_tbl[cqne->info % dev->dsr->caps.max_cq];
477 atomic_inc(&cq->refcnt);
478 spin_unlock_irqrestore(&dev->cq_tbl_lock, flags);
480 if (cq && cq->ibcq.comp_handler)
481 cq->ibcq.comp_handler(&cq->ibcq, cq->ibcq.cq_context);
483 atomic_dec(&cq->refcnt);
484 if (atomic_read(&cq->refcnt))
487 pvrdma_idx_ring_inc(&ring->cons_head, ring_slots);
493 static void pvrdma_free_irq(struct pvrdma_dev *dev)
497 dev_dbg(&dev->pdev->dev, "freeing interrupts\n");
498 for (i = 0; i < dev->nr_vectors; i++)
499 free_irq(pci_irq_vector(dev->pdev, i), dev);
502 static void pvrdma_enable_intrs(struct pvrdma_dev *dev)
504 dev_dbg(&dev->pdev->dev, "enable interrupts\n");
505 pvrdma_write_reg(dev, PVRDMA_REG_IMR, 0);
508 static void pvrdma_disable_intrs(struct pvrdma_dev *dev)
510 dev_dbg(&dev->pdev->dev, "disable interrupts\n");
511 pvrdma_write_reg(dev, PVRDMA_REG_IMR, ~0);
514 static int pvrdma_alloc_intrs(struct pvrdma_dev *dev)
516 struct pci_dev *pdev = dev->pdev;
519 ret = pci_alloc_irq_vectors(pdev, 1, PVRDMA_MAX_INTERRUPTS,
522 ret = pci_alloc_irq_vectors(pdev, 1, 1,
523 PCI_IRQ_MSI | PCI_IRQ_LEGACY);
527 dev->nr_vectors = ret;
529 ret = request_irq(pci_irq_vector(dev->pdev, 0), pvrdma_intr0_handler,
530 pdev->msix_enabled ? 0 : IRQF_SHARED, DRV_NAME, dev);
532 dev_err(&dev->pdev->dev,
533 "failed to request interrupt 0\n");
534 goto out_free_vectors;
537 for (i = 1; i < dev->nr_vectors; i++) {
538 ret = request_irq(pci_irq_vector(dev->pdev, i),
539 i == 1 ? pvrdma_intr1_handler :
540 pvrdma_intrx_handler,
543 dev_err(&dev->pdev->dev,
544 "failed to request interrupt %d\n", i);
553 free_irq(pci_irq_vector(dev->pdev, i), dev);
555 pci_free_irq_vectors(pdev);
559 static void pvrdma_free_slots(struct pvrdma_dev *dev)
561 struct pci_dev *pdev = dev->pdev;
564 dma_free_coherent(&pdev->dev, PAGE_SIZE, dev->resp_slot,
565 dev->dsr->resp_slot_dma);
567 dma_free_coherent(&pdev->dev, PAGE_SIZE, dev->cmd_slot,
568 dev->dsr->cmd_slot_dma);
571 static int pvrdma_add_gid_at_index(struct pvrdma_dev *dev,
572 const union ib_gid *gid,
576 union pvrdma_cmd_req req;
577 struct pvrdma_cmd_create_bind *cmd_bind = &req.create_bind;
579 if (!dev->sgid_tbl) {
580 dev_warn(&dev->pdev->dev, "sgid table not initialized\n");
584 memset(cmd_bind, 0, sizeof(*cmd_bind));
585 cmd_bind->hdr.cmd = PVRDMA_CMD_CREATE_BIND;
586 memcpy(cmd_bind->new_gid, gid->raw, 16);
587 cmd_bind->mtu = ib_mtu_enum_to_int(IB_MTU_1024);
588 cmd_bind->vlan = 0xfff;
589 cmd_bind->index = index;
590 cmd_bind->gid_type = PVRDMA_GID_TYPE_FLAG_ROCE_V1;
592 ret = pvrdma_cmd_post(dev, &req, NULL, 0);
594 dev_warn(&dev->pdev->dev,
595 "could not create binding, error: %d\n", ret);
598 memcpy(&dev->sgid_tbl[index], gid, sizeof(*gid));
602 static int pvrdma_add_gid(struct ib_device *ibdev,
605 const union ib_gid *gid,
606 const struct ib_gid_attr *attr,
609 struct pvrdma_dev *dev = to_vdev(ibdev);
611 return pvrdma_add_gid_at_index(dev, gid, index);
614 static int pvrdma_del_gid_at_index(struct pvrdma_dev *dev, int index)
617 union pvrdma_cmd_req req;
618 struct pvrdma_cmd_destroy_bind *cmd_dest = &req.destroy_bind;
620 /* Update sgid table. */
621 if (!dev->sgid_tbl) {
622 dev_warn(&dev->pdev->dev, "sgid table not initialized\n");
626 memset(cmd_dest, 0, sizeof(*cmd_dest));
627 cmd_dest->hdr.cmd = PVRDMA_CMD_DESTROY_BIND;
628 memcpy(cmd_dest->dest_gid, &dev->sgid_tbl[index], 16);
629 cmd_dest->index = index;
631 ret = pvrdma_cmd_post(dev, &req, NULL, 0);
633 dev_warn(&dev->pdev->dev,
634 "could not destroy binding, error: %d\n", ret);
637 memset(&dev->sgid_tbl[index], 0, 16);
641 static int pvrdma_del_gid(struct ib_device *ibdev,
646 struct pvrdma_dev *dev = to_vdev(ibdev);
648 dev_dbg(&dev->pdev->dev, "removing gid at index %u from %s",
649 index, dev->netdev->name);
651 return pvrdma_del_gid_at_index(dev, index);
654 static void pvrdma_netdevice_event_handle(struct pvrdma_dev *dev,
660 pvrdma_dispatch_event(dev, 1, IB_EVENT_PORT_ERR);
663 pvrdma_dispatch_event(dev, 1, IB_EVENT_PORT_ACTIVE);
666 dev_dbg(&dev->pdev->dev, "ignore netdevice event %ld on %s\n",
667 event, dev->ib_dev.name);
672 static void pvrdma_netdevice_event_work(struct work_struct *work)
674 struct pvrdma_netdevice_work *netdev_work;
675 struct pvrdma_dev *dev;
677 netdev_work = container_of(work, struct pvrdma_netdevice_work, work);
679 mutex_lock(&pvrdma_device_list_lock);
680 list_for_each_entry(dev, &pvrdma_device_list, device_link) {
681 if (dev->netdev == netdev_work->event_netdev) {
682 pvrdma_netdevice_event_handle(dev, netdev_work->event);
686 mutex_unlock(&pvrdma_device_list_lock);
691 static int pvrdma_netdevice_event(struct notifier_block *this,
692 unsigned long event, void *ptr)
694 struct net_device *event_netdev = netdev_notifier_info_to_dev(ptr);
695 struct pvrdma_netdevice_work *netdev_work;
697 netdev_work = kmalloc(sizeof(*netdev_work), GFP_ATOMIC);
701 INIT_WORK(&netdev_work->work, pvrdma_netdevice_event_work);
702 netdev_work->event_netdev = event_netdev;
703 netdev_work->event = event;
704 queue_work(event_wq, &netdev_work->work);
709 static int pvrdma_pci_probe(struct pci_dev *pdev,
710 const struct pci_device_id *id)
712 struct pci_dev *pdev_net;
713 struct pvrdma_dev *dev;
717 unsigned int version;
718 dma_addr_t slot_dma = 0;
720 dev_dbg(&pdev->dev, "initializing driver %s\n", pci_name(pdev));
722 /* Allocate zero-out device */
723 dev = (struct pvrdma_dev *)ib_alloc_device(sizeof(*dev));
725 dev_err(&pdev->dev, "failed to allocate IB device\n");
729 mutex_lock(&pvrdma_device_list_lock);
730 list_add(&dev->device_link, &pvrdma_device_list);
731 mutex_unlock(&pvrdma_device_list_lock);
733 ret = pvrdma_init_device(dev);
735 goto err_free_device;
738 pci_set_drvdata(pdev, dev);
740 ret = pci_enable_device(pdev);
742 dev_err(&pdev->dev, "cannot enable PCI device\n");
743 goto err_free_device;
746 dev_dbg(&pdev->dev, "PCI resource flags BAR0 %#lx\n",
747 pci_resource_flags(pdev, 0));
748 dev_dbg(&pdev->dev, "PCI resource len %#llx\n",
749 (unsigned long long)pci_resource_len(pdev, 0));
750 dev_dbg(&pdev->dev, "PCI resource start %#llx\n",
751 (unsigned long long)pci_resource_start(pdev, 0));
752 dev_dbg(&pdev->dev, "PCI resource flags BAR1 %#lx\n",
753 pci_resource_flags(pdev, 1));
754 dev_dbg(&pdev->dev, "PCI resource len %#llx\n",
755 (unsigned long long)pci_resource_len(pdev, 1));
756 dev_dbg(&pdev->dev, "PCI resource start %#llx\n",
757 (unsigned long long)pci_resource_start(pdev, 1));
759 if (!(pci_resource_flags(pdev, 0) & IORESOURCE_MEM) ||
760 !(pci_resource_flags(pdev, 1) & IORESOURCE_MEM)) {
761 dev_err(&pdev->dev, "PCI BAR region not MMIO\n");
763 goto err_free_device;
766 ret = pci_request_regions(pdev, DRV_NAME);
768 dev_err(&pdev->dev, "cannot request PCI resources\n");
769 goto err_disable_pdev;
772 /* Enable 64-Bit DMA */
773 if (pci_set_dma_mask(pdev, DMA_BIT_MASK(64)) == 0) {
774 ret = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
777 "pci_set_consistent_dma_mask failed\n");
778 goto err_free_resource;
781 ret = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
784 "pci_set_dma_mask failed\n");
785 goto err_free_resource;
789 pci_set_master(pdev);
791 /* Map register space */
792 start = pci_resource_start(dev->pdev, PVRDMA_PCI_RESOURCE_REG);
793 len = pci_resource_len(dev->pdev, PVRDMA_PCI_RESOURCE_REG);
794 dev->regs = ioremap(start, len);
796 dev_err(&pdev->dev, "register mapping failed\n");
798 goto err_free_resource;
801 /* Setup per-device UAR. */
802 dev->driver_uar.index = 0;
803 dev->driver_uar.pfn =
804 pci_resource_start(dev->pdev, PVRDMA_PCI_RESOURCE_UAR) >>
806 dev->driver_uar.map =
807 ioremap(dev->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
808 if (!dev->driver_uar.map) {
809 dev_err(&pdev->dev, "failed to remap UAR pages\n");
814 version = pvrdma_read_reg(dev, PVRDMA_REG_VERSION);
815 dev_info(&pdev->dev, "device version %d, driver version %d\n",
816 version, PVRDMA_VERSION);
817 if (version < PVRDMA_VERSION) {
818 dev_err(&pdev->dev, "incompatible device version\n");
822 dev->dsr = dma_alloc_coherent(&pdev->dev, sizeof(*dev->dsr),
823 &dev->dsrbase, GFP_KERNEL);
825 dev_err(&pdev->dev, "failed to allocate shared region\n");
830 /* Setup the shared region */
831 memset(dev->dsr, 0, sizeof(*dev->dsr));
832 dev->dsr->driver_version = PVRDMA_VERSION;
833 dev->dsr->gos_info.gos_bits = sizeof(void *) == 4 ?
836 dev->dsr->gos_info.gos_type = PVRDMA_GOS_TYPE_LINUX;
837 dev->dsr->gos_info.gos_ver = 1;
838 dev->dsr->uar_pfn = dev->driver_uar.pfn;
841 dev->cmd_slot = dma_alloc_coherent(&pdev->dev, PAGE_SIZE,
842 &slot_dma, GFP_KERNEL);
843 if (!dev->cmd_slot) {
848 dev->dsr->cmd_slot_dma = (u64)slot_dma;
851 dev->resp_slot = dma_alloc_coherent(&pdev->dev, PAGE_SIZE,
852 &slot_dma, GFP_KERNEL);
853 if (!dev->resp_slot) {
858 dev->dsr->resp_slot_dma = (u64)slot_dma;
860 /* Async event ring */
861 dev->dsr->async_ring_pages.num_pages = 4;
862 ret = pvrdma_page_dir_init(dev, &dev->async_pdir,
863 dev->dsr->async_ring_pages.num_pages, true);
866 dev->async_ring_state = dev->async_pdir.pages[0];
867 dev->dsr->async_ring_pages.pdir_dma = dev->async_pdir.dir_dma;
869 /* CQ notification ring */
870 dev->dsr->cq_ring_pages.num_pages = 4;
871 ret = pvrdma_page_dir_init(dev, &dev->cq_pdir,
872 dev->dsr->cq_ring_pages.num_pages, true);
874 goto err_free_async_ring;
875 dev->cq_ring_state = dev->cq_pdir.pages[0];
876 dev->dsr->cq_ring_pages.pdir_dma = dev->cq_pdir.dir_dma;
879 * Write the PA of the shared region to the device. The writes must be
880 * ordered such that the high bits are written last. When the writes
881 * complete, the device will have filled out the capabilities.
884 pvrdma_write_reg(dev, PVRDMA_REG_DSRLOW, (u32)dev->dsrbase);
885 pvrdma_write_reg(dev, PVRDMA_REG_DSRHIGH,
886 (u32)((u64)(dev->dsrbase) >> 32));
888 /* Make sure the write is complete before reading status. */
891 /* Currently, the driver only supports RoCE mode. */
892 if (dev->dsr->caps.mode != PVRDMA_DEVICE_MODE_ROCE) {
893 dev_err(&pdev->dev, "unsupported transport %d\n",
894 dev->dsr->caps.mode);
896 goto err_free_cq_ring;
899 /* Currently, the driver only supports RoCE V1. */
900 if (!(dev->dsr->caps.gid_types & PVRDMA_GID_TYPE_FLAG_ROCE_V1)) {
901 dev_err(&pdev->dev, "driver needs RoCE v1 support\n");
903 goto err_free_cq_ring;
906 /* Paired vmxnet3 will have same bus, slot. But func will be 0 */
907 pdev_net = pci_get_slot(pdev->bus, PCI_DEVFN(PCI_SLOT(pdev->devfn), 0));
909 dev_err(&pdev->dev, "failed to find paired net device\n");
911 goto err_free_cq_ring;
914 if (pdev_net->vendor != PCI_VENDOR_ID_VMWARE ||
915 pdev_net->device != PCI_DEVICE_ID_VMWARE_VMXNET3) {
916 dev_err(&pdev->dev, "failed to find paired vmxnet3 device\n");
917 pci_dev_put(pdev_net);
919 goto err_free_cq_ring;
922 dev->netdev = pci_get_drvdata(pdev_net);
923 pci_dev_put(pdev_net);
925 dev_err(&pdev->dev, "failed to get vmxnet3 device\n");
927 goto err_free_cq_ring;
930 dev_info(&pdev->dev, "paired device to %s\n", dev->netdev->name);
932 /* Interrupt setup */
933 ret = pvrdma_alloc_intrs(dev);
935 dev_err(&pdev->dev, "failed to allocate interrupts\n");
937 goto err_free_cq_ring;
940 /* Allocate UAR table. */
941 ret = pvrdma_uar_table_init(dev);
943 dev_err(&pdev->dev, "failed to allocate UAR table\n");
948 /* Allocate GID table */
949 dev->sgid_tbl = kcalloc(dev->dsr->caps.gid_tbl_len,
950 sizeof(union ib_gid), GFP_KERNEL);
951 if (!dev->sgid_tbl) {
953 goto err_free_uar_table;
955 dev_dbg(&pdev->dev, "gid table len %d\n", dev->dsr->caps.gid_tbl_len);
957 pvrdma_enable_intrs(dev);
959 /* Activate pvrdma device */
960 pvrdma_write_reg(dev, PVRDMA_REG_CTL, PVRDMA_DEVICE_CTL_ACTIVATE);
962 /* Make sure the write is complete before reading status. */
965 /* Check if device was successfully activated */
966 ret = pvrdma_read_reg(dev, PVRDMA_REG_ERR);
968 dev_err(&pdev->dev, "failed to activate device\n");
970 goto err_disable_intr;
973 /* Register IB device */
974 ret = pvrdma_register_device(dev);
976 dev_err(&pdev->dev, "failed to register IB device\n");
977 goto err_disable_intr;
980 dev->nb_netdev.notifier_call = pvrdma_netdevice_event;
981 ret = register_netdevice_notifier(&dev->nb_netdev);
983 dev_err(&pdev->dev, "failed to register netdevice events\n");
984 goto err_unreg_ibdev;
987 dev_info(&pdev->dev, "attached to device\n");
991 ib_unregister_device(&dev->ib_dev);
993 pvrdma_disable_intrs(dev);
994 kfree(dev->sgid_tbl);
996 pvrdma_uar_table_cleanup(dev);
998 pvrdma_free_irq(dev);
999 pci_free_irq_vectors(pdev);
1001 pvrdma_page_dir_cleanup(dev, &dev->cq_pdir);
1002 err_free_async_ring:
1003 pvrdma_page_dir_cleanup(dev, &dev->async_pdir);
1005 pvrdma_free_slots(dev);
1007 dma_free_coherent(&pdev->dev, sizeof(*dev->dsr), dev->dsr,
1010 iounmap(dev->driver_uar.map);
1014 pci_release_regions(pdev);
1016 pci_disable_device(pdev);
1017 pci_set_drvdata(pdev, NULL);
1019 mutex_lock(&pvrdma_device_list_lock);
1020 list_del(&dev->device_link);
1021 mutex_unlock(&pvrdma_device_list_lock);
1022 ib_dealloc_device(&dev->ib_dev);
1026 static void pvrdma_pci_remove(struct pci_dev *pdev)
1028 struct pvrdma_dev *dev = pci_get_drvdata(pdev);
1033 dev_info(&pdev->dev, "detaching from device\n");
1035 unregister_netdevice_notifier(&dev->nb_netdev);
1036 dev->nb_netdev.notifier_call = NULL;
1038 flush_workqueue(event_wq);
1040 /* Unregister ib device */
1041 ib_unregister_device(&dev->ib_dev);
1043 mutex_lock(&pvrdma_device_list_lock);
1044 list_del(&dev->device_link);
1045 mutex_unlock(&pvrdma_device_list_lock);
1047 pvrdma_disable_intrs(dev);
1048 pvrdma_free_irq(dev);
1049 pci_free_irq_vectors(pdev);
1051 /* Deactivate pvrdma device */
1052 pvrdma_write_reg(dev, PVRDMA_REG_CTL, PVRDMA_DEVICE_CTL_RESET);
1053 pvrdma_page_dir_cleanup(dev, &dev->cq_pdir);
1054 pvrdma_page_dir_cleanup(dev, &dev->async_pdir);
1055 pvrdma_free_slots(dev);
1058 kfree(dev->sgid_tbl);
1061 pvrdma_uar_table_cleanup(dev);
1062 iounmap(dev->driver_uar.map);
1064 ib_dealloc_device(&dev->ib_dev);
1066 /* Free pci resources */
1067 pci_release_regions(pdev);
1068 pci_disable_device(pdev);
1069 pci_set_drvdata(pdev, NULL);
1072 static struct pci_device_id pvrdma_pci_table[] = {
1073 { PCI_DEVICE(PCI_VENDOR_ID_VMWARE, PCI_DEVICE_ID_VMWARE_PVRDMA), },
1077 MODULE_DEVICE_TABLE(pci, pvrdma_pci_table);
1079 static struct pci_driver pvrdma_driver = {
1081 .id_table = pvrdma_pci_table,
1082 .probe = pvrdma_pci_probe,
1083 .remove = pvrdma_pci_remove,
1086 static int __init pvrdma_init(void)
1090 event_wq = alloc_ordered_workqueue("pvrdma_event_wq", WQ_MEM_RECLAIM);
1094 err = pci_register_driver(&pvrdma_driver);
1096 destroy_workqueue(event_wq);
1101 static void __exit pvrdma_cleanup(void)
1103 pci_unregister_driver(&pvrdma_driver);
1105 destroy_workqueue(event_wq);
1108 module_init(pvrdma_init);
1109 module_exit(pvrdma_cleanup);
1111 MODULE_AUTHOR("VMware, Inc");
1112 MODULE_DESCRIPTION("VMware Paravirtual RDMA driver");
1113 MODULE_VERSION(DRV_VERSION);
1114 MODULE_LICENSE("Dual BSD/GPL");