2 * Broadcom NetXtreme-E RoCE driver.
4 * Copyright (c) 2016 - 2017, Broadcom. All rights reserved. The term
5 * Broadcom refers to Broadcom Limited and/or its subsidiaries.
7 * This software is available to you under a choice of one of two
8 * licenses. You may choose to be licensed under the terms of the GNU
9 * General Public License (GPL) Version 2, available from the file
10 * COPYING in the main directory of this source tree, or the
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
17 * 1. Redistributions of source code must retain the above copyright
18 * notice, this list of conditions and the following disclaimer.
19 * 2. Redistributions in binary form must reproduce the above copyright
20 * notice, this list of conditions and the following disclaimer in
21 * the documentation and/or other materials provided with the
24 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS''
25 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO,
26 * THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
27 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS
28 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
29 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
30 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
31 * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
32 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE
33 * OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN
34 * IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
36 * Description: Main component of the bnxt_re driver
39 #include <linux/module.h>
40 #include <linux/netdevice.h>
41 #include <linux/ethtool.h>
42 #include <linux/mutex.h>
43 #include <linux/list.h>
44 #include <linux/rculist.h>
45 #include <linux/spinlock.h>
46 #include <linux/pci.h>
47 #include <net/dcbnl.h>
49 #include <net/addrconf.h>
50 #include <linux/if_ether.h>
52 #include <rdma/ib_verbs.h>
53 #include <rdma/ib_user_verbs.h>
54 #include <rdma/ib_umem.h>
55 #include <rdma/ib_addr.h>
59 #include "qplib_res.h"
62 #include "qplib_rcfw.h"
65 #include <rdma/bnxt_re-abi.h>
67 #include "hw_counters.h"
69 static char version[] =
73 MODULE_DESCRIPTION(BNXT_RE_DESC " Driver");
74 MODULE_LICENSE("Dual BSD/GPL");
77 static struct list_head bnxt_re_dev_list = LIST_HEAD_INIT(bnxt_re_dev_list);
78 /* Mutex to protect the list of bnxt_re devices added */
79 static DEFINE_MUTEX(bnxt_re_dev_lock);
80 static struct workqueue_struct *bnxt_re_wq;
81 static void bnxt_re_ib_unreg(struct bnxt_re_dev *rdev);
83 /* SR-IOV helper functions */
85 static void bnxt_re_get_sriov_func_type(struct bnxt_re_dev *rdev)
89 bp = netdev_priv(rdev->en_dev->net);
94 /* Set the maximum number of each resource that the driver actually wants
95 * to allocate. This may be up to the maximum number the firmware has
96 * reserved for the function. The driver may choose to allocate fewer
97 * resources than the firmware maximum.
99 static void bnxt_re_set_resource_limits(struct bnxt_re_dev *rdev)
101 u32 vf_qps = 0, vf_srqs = 0, vf_cqs = 0, vf_mrws = 0, vf_gids = 0;
105 struct bnxt_qplib_dev_attr *dev_attr = &rdev->dev_attr;
107 rdev->qplib_ctx.qpc_count = min_t(u32, BNXT_RE_MAX_QPC_COUNT,
110 rdev->qplib_ctx.mrw_count = BNXT_RE_MAX_MRW_COUNT_256K;
111 /* Use max_mr from fw since max_mrw does not get set */
112 rdev->qplib_ctx.mrw_count = min_t(u32, rdev->qplib_ctx.mrw_count,
114 rdev->qplib_ctx.srqc_count = min_t(u32, BNXT_RE_MAX_SRQC_COUNT,
116 rdev->qplib_ctx.cq_count = min_t(u32, BNXT_RE_MAX_CQ_COUNT,
119 for (i = 0; i < MAX_TQM_ALLOC_REQ; i++)
120 rdev->qplib_ctx.tqm_count[i] =
121 rdev->dev_attr.tqm_alloc_reqs[i];
125 * Reserve a set of resources for the PF. Divide the remaining
126 * resources among the VFs
128 vf_pct = 100 - BNXT_RE_PCT_RSVD_FOR_PF;
129 num_vfs = 100 * rdev->num_vfs;
130 vf_qps = (rdev->qplib_ctx.qpc_count * vf_pct) / num_vfs;
131 vf_srqs = (rdev->qplib_ctx.srqc_count * vf_pct) / num_vfs;
132 vf_cqs = (rdev->qplib_ctx.cq_count * vf_pct) / num_vfs;
134 * The driver allows many more MRs than other resources. If the
135 * firmware does also, then reserve a fixed amount for the PF
136 * and divide the rest among VFs. VFs may use many MRs for NFS
137 * mounts, ISER, NVME applications, etc. If the firmware
138 * severely restricts the number of MRs, then let PF have
139 * half and divide the rest among VFs, as for the other
142 if (rdev->qplib_ctx.mrw_count < BNXT_RE_MAX_MRW_COUNT_64K)
143 vf_mrws = rdev->qplib_ctx.mrw_count * vf_pct / num_vfs;
145 vf_mrws = (rdev->qplib_ctx.mrw_count -
146 BNXT_RE_RESVD_MR_FOR_PF) / rdev->num_vfs;
147 vf_gids = BNXT_RE_MAX_GID_PER_VF;
149 rdev->qplib_ctx.vf_res.max_mrw_per_vf = vf_mrws;
150 rdev->qplib_ctx.vf_res.max_gid_per_vf = vf_gids;
151 rdev->qplib_ctx.vf_res.max_qp_per_vf = vf_qps;
152 rdev->qplib_ctx.vf_res.max_srq_per_vf = vf_srqs;
153 rdev->qplib_ctx.vf_res.max_cq_per_vf = vf_cqs;
156 /* for handling bnxt_en callbacks later */
157 static void bnxt_re_stop(void *p)
161 static void bnxt_re_start(void *p)
165 static void bnxt_re_sriov_config(void *p, int num_vfs)
167 struct bnxt_re_dev *rdev = p;
172 rdev->num_vfs = num_vfs;
173 bnxt_re_set_resource_limits(rdev);
174 bnxt_qplib_set_func_resources(&rdev->qplib_res, &rdev->rcfw,
178 static void bnxt_re_shutdown(void *p)
180 struct bnxt_re_dev *rdev = p;
185 bnxt_re_ib_unreg(rdev);
188 static void bnxt_re_stop_irq(void *handle)
190 struct bnxt_re_dev *rdev = (struct bnxt_re_dev *)handle;
191 struct bnxt_qplib_rcfw *rcfw = &rdev->rcfw;
192 struct bnxt_qplib_nq *nq;
195 for (indx = BNXT_RE_NQ_IDX; indx < rdev->num_msix; indx++) {
196 nq = &rdev->nq[indx - 1];
197 bnxt_qplib_nq_stop_irq(nq, false);
200 bnxt_qplib_rcfw_stop_irq(rcfw, false);
203 static void bnxt_re_start_irq(void *handle, struct bnxt_msix_entry *ent)
205 struct bnxt_re_dev *rdev = (struct bnxt_re_dev *)handle;
206 struct bnxt_msix_entry *msix_ent = rdev->msix_entries;
207 struct bnxt_qplib_rcfw *rcfw = &rdev->rcfw;
208 struct bnxt_qplib_nq *nq;
212 /* Not setting the f/w timeout bit in rcfw.
213 * During the driver unload the first command
214 * to f/w will timeout and that will set the
217 dev_err(rdev_to_dev(rdev), "Failed to re-start IRQs\n");
221 /* Vectors may change after restart, so update with new vectors
222 * in device sctructure.
224 for (indx = 0; indx < rdev->num_msix; indx++)
225 rdev->msix_entries[indx].vector = ent[indx].vector;
227 bnxt_qplib_rcfw_start_irq(rcfw, msix_ent[BNXT_RE_AEQ_IDX].vector,
229 for (indx = BNXT_RE_NQ_IDX ; indx < rdev->num_msix; indx++) {
230 nq = &rdev->nq[indx - 1];
231 rc = bnxt_qplib_nq_start_irq(nq, indx - 1,
232 msix_ent[indx].vector, false);
234 dev_warn(rdev_to_dev(rdev),
235 "Failed to reinit NQ index %d\n", indx - 1);
239 static struct bnxt_ulp_ops bnxt_re_ulp_ops = {
240 .ulp_async_notifier = NULL,
241 .ulp_stop = bnxt_re_stop,
242 .ulp_start = bnxt_re_start,
243 .ulp_sriov_config = bnxt_re_sriov_config,
244 .ulp_shutdown = bnxt_re_shutdown,
245 .ulp_irq_stop = bnxt_re_stop_irq,
246 .ulp_irq_restart = bnxt_re_start_irq
249 /* RoCE -> Net driver */
251 /* Driver registration routines used to let the networking driver (bnxt_en)
252 * to know that the RoCE driver is now installed
254 static int bnxt_re_unregister_netdev(struct bnxt_re_dev *rdev)
256 struct bnxt_en_dev *en_dev;
262 en_dev = rdev->en_dev;
264 rc = en_dev->en_ops->bnxt_unregister_device(rdev->en_dev,
269 static int bnxt_re_register_netdev(struct bnxt_re_dev *rdev)
271 struct bnxt_en_dev *en_dev;
277 en_dev = rdev->en_dev;
279 rc = en_dev->en_ops->bnxt_register_device(en_dev, BNXT_ROCE_ULP,
280 &bnxt_re_ulp_ops, rdev);
284 static int bnxt_re_free_msix(struct bnxt_re_dev *rdev)
286 struct bnxt_en_dev *en_dev;
292 en_dev = rdev->en_dev;
295 rc = en_dev->en_ops->bnxt_free_msix(rdev->en_dev, BNXT_ROCE_ULP);
300 static int bnxt_re_request_msix(struct bnxt_re_dev *rdev)
302 int rc = 0, num_msix_want = BNXT_RE_MAX_MSIX, num_msix_got;
303 struct bnxt_en_dev *en_dev;
308 en_dev = rdev->en_dev;
310 num_msix_want = min_t(u32, BNXT_RE_MAX_MSIX, num_online_cpus());
312 num_msix_got = en_dev->en_ops->bnxt_request_msix(en_dev, BNXT_ROCE_ULP,
315 if (num_msix_got < BNXT_RE_MIN_MSIX) {
319 if (num_msix_got != num_msix_want) {
320 dev_warn(rdev_to_dev(rdev),
321 "Requested %d MSI-X vectors, got %d\n",
322 num_msix_want, num_msix_got);
324 rdev->num_msix = num_msix_got;
329 static void bnxt_re_init_hwrm_hdr(struct bnxt_re_dev *rdev, struct input *hdr,
330 u16 opcd, u16 crid, u16 trid)
332 hdr->req_type = cpu_to_le16(opcd);
333 hdr->cmpl_ring = cpu_to_le16(crid);
334 hdr->target_id = cpu_to_le16(trid);
337 static void bnxt_re_fill_fw_msg(struct bnxt_fw_msg *fw_msg, void *msg,
338 int msg_len, void *resp, int resp_max_len,
342 fw_msg->msg_len = msg_len;
344 fw_msg->resp_max_len = resp_max_len;
345 fw_msg->timeout = timeout;
348 static int bnxt_re_net_ring_free(struct bnxt_re_dev *rdev, u16 fw_ring_id)
350 struct bnxt_en_dev *en_dev = rdev->en_dev;
351 struct hwrm_ring_free_input req = {0};
352 struct hwrm_ring_free_output resp;
353 struct bnxt_fw_msg fw_msg;
359 memset(&fw_msg, 0, sizeof(fw_msg));
361 bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_RING_FREE, -1, -1);
362 req.ring_type = RING_ALLOC_REQ_RING_TYPE_L2_CMPL;
363 req.ring_id = cpu_to_le16(fw_ring_id);
364 bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
365 sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
366 rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
368 dev_err(rdev_to_dev(rdev),
369 "Failed to free HW ring:%d :%#x", req.ring_id, rc);
373 static int bnxt_re_net_ring_alloc(struct bnxt_re_dev *rdev, dma_addr_t *dma_arr,
374 int pages, int type, u32 ring_mask,
375 u32 map_index, u16 *fw_ring_id)
377 struct bnxt_en_dev *en_dev = rdev->en_dev;
378 struct hwrm_ring_alloc_input req = {0};
379 struct hwrm_ring_alloc_output resp;
380 struct bnxt_fw_msg fw_msg;
386 memset(&fw_msg, 0, sizeof(fw_msg));
387 bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_RING_ALLOC, -1, -1);
389 req.page_tbl_addr = cpu_to_le64(dma_arr[0]);
391 /* Page size is in log2 units */
392 req.page_size = BNXT_PAGE_SHIFT;
393 req.page_tbl_depth = 1;
396 /* Association of ring index with doorbell index and MSIX number */
397 req.logical_id = cpu_to_le16(map_index);
398 req.length = cpu_to_le32(ring_mask + 1);
399 req.ring_type = RING_ALLOC_REQ_RING_TYPE_L2_CMPL;
400 req.int_mode = RING_ALLOC_REQ_INT_MODE_MSIX;
401 bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
402 sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
403 rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
405 *fw_ring_id = le16_to_cpu(resp.ring_id);
410 static int bnxt_re_net_stats_ctx_free(struct bnxt_re_dev *rdev,
413 struct bnxt_en_dev *en_dev = rdev->en_dev;
414 struct hwrm_stat_ctx_free_input req = {0};
415 struct bnxt_fw_msg fw_msg;
421 memset(&fw_msg, 0, sizeof(fw_msg));
423 bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_STAT_CTX_FREE, -1, -1);
424 req.stat_ctx_id = cpu_to_le32(fw_stats_ctx_id);
425 bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&req,
426 sizeof(req), DFLT_HWRM_CMD_TIMEOUT);
427 rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
429 dev_err(rdev_to_dev(rdev),
430 "Failed to free HW stats context %#x", rc);
435 static int bnxt_re_net_stats_ctx_alloc(struct bnxt_re_dev *rdev,
437 u32 *fw_stats_ctx_id)
439 struct hwrm_stat_ctx_alloc_output resp = {0};
440 struct hwrm_stat_ctx_alloc_input req = {0};
441 struct bnxt_en_dev *en_dev = rdev->en_dev;
442 struct bnxt_fw_msg fw_msg;
445 *fw_stats_ctx_id = INVALID_STATS_CTX_ID;
450 memset(&fw_msg, 0, sizeof(fw_msg));
452 bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_STAT_CTX_ALLOC, -1, -1);
453 req.update_period_ms = cpu_to_le32(1000);
454 req.stats_dma_addr = cpu_to_le64(dma_map);
455 req.stat_ctx_flags = STAT_CTX_ALLOC_REQ_STAT_CTX_FLAGS_ROCE;
456 bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
457 sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
458 rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
460 *fw_stats_ctx_id = le32_to_cpu(resp.stat_ctx_id);
467 static bool is_bnxt_re_dev(struct net_device *netdev)
469 struct ethtool_drvinfo drvinfo;
471 if (netdev->ethtool_ops && netdev->ethtool_ops->get_drvinfo) {
472 memset(&drvinfo, 0, sizeof(drvinfo));
473 netdev->ethtool_ops->get_drvinfo(netdev, &drvinfo);
475 if (strcmp(drvinfo.driver, "bnxt_en"))
482 static struct bnxt_re_dev *bnxt_re_from_netdev(struct net_device *netdev)
484 struct bnxt_re_dev *rdev;
487 list_for_each_entry_rcu(rdev, &bnxt_re_dev_list, list) {
488 if (rdev->netdev == netdev) {
497 static void bnxt_re_dev_unprobe(struct net_device *netdev,
498 struct bnxt_en_dev *en_dev)
501 module_put(en_dev->pdev->driver->driver.owner);
504 static struct bnxt_en_dev *bnxt_re_dev_probe(struct net_device *netdev)
506 struct bnxt *bp = netdev_priv(netdev);
507 struct bnxt_en_dev *en_dev;
508 struct pci_dev *pdev;
510 /* Call bnxt_en's RoCE probe via indirect API */
512 return ERR_PTR(-EINVAL);
514 en_dev = bp->ulp_probe(netdev);
520 return ERR_PTR(-EINVAL);
522 if (!(en_dev->flags & BNXT_EN_FLAG_ROCE_CAP)) {
524 "%s: probe error: RoCE is not supported on this device",
525 ROCE_DRV_MODULE_NAME);
526 return ERR_PTR(-ENODEV);
529 /* Bump net device reference count */
530 if (!try_module_get(pdev->driver->driver.owner))
531 return ERR_PTR(-ENODEV);
538 static ssize_t hw_rev_show(struct device *device, struct device_attribute *attr,
541 struct bnxt_re_dev *rdev = to_bnxt_re_dev(device, ibdev.dev);
543 return scnprintf(buf, PAGE_SIZE, "0x%x\n", rdev->en_dev->pdev->vendor);
545 static DEVICE_ATTR_RO(hw_rev);
547 static ssize_t hca_type_show(struct device *device,
548 struct device_attribute *attr, char *buf)
550 struct bnxt_re_dev *rdev = to_bnxt_re_dev(device, ibdev.dev);
552 return scnprintf(buf, PAGE_SIZE, "%s\n", rdev->ibdev.node_desc);
554 static DEVICE_ATTR_RO(hca_type);
556 static struct attribute *bnxt_re_attributes[] = {
557 &dev_attr_hw_rev.attr,
558 &dev_attr_hca_type.attr,
562 static const struct attribute_group bnxt_re_dev_attr_group = {
563 .attrs = bnxt_re_attributes,
566 static void bnxt_re_unregister_ib(struct bnxt_re_dev *rdev)
568 ib_unregister_device(&rdev->ibdev);
571 static int bnxt_re_register_ib(struct bnxt_re_dev *rdev)
573 struct ib_device *ibdev = &rdev->ibdev;
576 ibdev->owner = THIS_MODULE;
577 ibdev->node_type = RDMA_NODE_IB_CA;
578 strlcpy(ibdev->node_desc, BNXT_RE_DESC " HCA",
579 strlen(BNXT_RE_DESC) + 5);
580 ibdev->phys_port_cnt = 1;
582 bnxt_qplib_get_guid(rdev->netdev->dev_addr, (u8 *)&ibdev->node_guid);
584 ibdev->num_comp_vectors = 1;
585 ibdev->dev.parent = &rdev->en_dev->pdev->dev;
586 ibdev->local_dma_lkey = BNXT_QPLIB_RSVD_LKEY;
589 ibdev->uverbs_abi_ver = BNXT_RE_ABI_VERSION;
590 ibdev->uverbs_cmd_mask =
591 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
592 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
593 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
594 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
595 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
596 (1ull << IB_USER_VERBS_CMD_REG_MR) |
597 (1ull << IB_USER_VERBS_CMD_REREG_MR) |
598 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
599 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
600 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
601 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ) |
602 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
603 (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
604 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
605 (1ull << IB_USER_VERBS_CMD_QUERY_QP) |
606 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
607 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) |
608 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) |
609 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) |
610 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) |
611 (1ull << IB_USER_VERBS_CMD_CREATE_AH) |
612 (1ull << IB_USER_VERBS_CMD_MODIFY_AH) |
613 (1ull << IB_USER_VERBS_CMD_QUERY_AH) |
614 (1ull << IB_USER_VERBS_CMD_DESTROY_AH);
615 /* POLL_CQ and REQ_NOTIFY_CQ is directly handled in libbnxt_re */
618 ibdev->query_device = bnxt_re_query_device;
619 ibdev->modify_device = bnxt_re_modify_device;
621 ibdev->query_port = bnxt_re_query_port;
622 ibdev->get_port_immutable = bnxt_re_get_port_immutable;
623 ibdev->get_dev_fw_str = bnxt_re_query_fw_str;
624 ibdev->query_pkey = bnxt_re_query_pkey;
625 ibdev->get_netdev = bnxt_re_get_netdev;
626 ibdev->add_gid = bnxt_re_add_gid;
627 ibdev->del_gid = bnxt_re_del_gid;
628 ibdev->get_link_layer = bnxt_re_get_link_layer;
630 ibdev->alloc_pd = bnxt_re_alloc_pd;
631 ibdev->dealloc_pd = bnxt_re_dealloc_pd;
633 ibdev->create_ah = bnxt_re_create_ah;
634 ibdev->modify_ah = bnxt_re_modify_ah;
635 ibdev->query_ah = bnxt_re_query_ah;
636 ibdev->destroy_ah = bnxt_re_destroy_ah;
638 ibdev->create_srq = bnxt_re_create_srq;
639 ibdev->modify_srq = bnxt_re_modify_srq;
640 ibdev->query_srq = bnxt_re_query_srq;
641 ibdev->destroy_srq = bnxt_re_destroy_srq;
642 ibdev->post_srq_recv = bnxt_re_post_srq_recv;
644 ibdev->create_qp = bnxt_re_create_qp;
645 ibdev->modify_qp = bnxt_re_modify_qp;
646 ibdev->query_qp = bnxt_re_query_qp;
647 ibdev->destroy_qp = bnxt_re_destroy_qp;
649 ibdev->post_send = bnxt_re_post_send;
650 ibdev->post_recv = bnxt_re_post_recv;
652 ibdev->create_cq = bnxt_re_create_cq;
653 ibdev->destroy_cq = bnxt_re_destroy_cq;
654 ibdev->poll_cq = bnxt_re_poll_cq;
655 ibdev->req_notify_cq = bnxt_re_req_notify_cq;
657 ibdev->get_dma_mr = bnxt_re_get_dma_mr;
658 ibdev->dereg_mr = bnxt_re_dereg_mr;
659 ibdev->alloc_mr = bnxt_re_alloc_mr;
660 ibdev->map_mr_sg = bnxt_re_map_mr_sg;
662 ibdev->reg_user_mr = bnxt_re_reg_user_mr;
663 ibdev->alloc_ucontext = bnxt_re_alloc_ucontext;
664 ibdev->dealloc_ucontext = bnxt_re_dealloc_ucontext;
665 ibdev->mmap = bnxt_re_mmap;
666 ibdev->get_hw_stats = bnxt_re_ib_get_hw_stats;
667 ibdev->alloc_hw_stats = bnxt_re_ib_alloc_hw_stats;
669 rdma_set_device_sysfs_group(ibdev, &bnxt_re_dev_attr_group);
670 ibdev->driver_id = RDMA_DRIVER_BNXT_RE;
671 return ib_register_device(ibdev, "bnxt_re%d", NULL);
674 static void bnxt_re_dev_remove(struct bnxt_re_dev *rdev)
676 dev_put(rdev->netdev);
679 mutex_lock(&bnxt_re_dev_lock);
680 list_del_rcu(&rdev->list);
681 mutex_unlock(&bnxt_re_dev_lock);
685 ib_dealloc_device(&rdev->ibdev);
689 static struct bnxt_re_dev *bnxt_re_dev_add(struct net_device *netdev,
690 struct bnxt_en_dev *en_dev)
692 struct bnxt_re_dev *rdev;
694 /* Allocate bnxt_re_dev instance here */
695 rdev = (struct bnxt_re_dev *)ib_alloc_device(sizeof(*rdev));
697 dev_err(NULL, "%s: bnxt_re_dev allocation failure!",
698 ROCE_DRV_MODULE_NAME);
702 rdev->netdev = netdev;
703 dev_hold(rdev->netdev);
704 rdev->en_dev = en_dev;
705 rdev->id = rdev->en_dev->pdev->devfn;
706 INIT_LIST_HEAD(&rdev->qp_list);
707 mutex_init(&rdev->qp_lock);
708 atomic_set(&rdev->qp_count, 0);
709 atomic_set(&rdev->cq_count, 0);
710 atomic_set(&rdev->srq_count, 0);
711 atomic_set(&rdev->mr_count, 0);
712 atomic_set(&rdev->mw_count, 0);
713 rdev->cosq[0] = 0xFFFF;
714 rdev->cosq[1] = 0xFFFF;
716 mutex_lock(&bnxt_re_dev_lock);
717 list_add_tail_rcu(&rdev->list, &bnxt_re_dev_list);
718 mutex_unlock(&bnxt_re_dev_lock);
722 static int bnxt_re_handle_unaffi_async_event(struct creq_func_event
725 switch (unaffi_async->event) {
726 case CREQ_FUNC_EVENT_EVENT_TX_WQE_ERROR:
728 case CREQ_FUNC_EVENT_EVENT_TX_DATA_ERROR:
730 case CREQ_FUNC_EVENT_EVENT_RX_WQE_ERROR:
732 case CREQ_FUNC_EVENT_EVENT_RX_DATA_ERROR:
734 case CREQ_FUNC_EVENT_EVENT_CQ_ERROR:
736 case CREQ_FUNC_EVENT_EVENT_TQM_ERROR:
738 case CREQ_FUNC_EVENT_EVENT_CFCQ_ERROR:
740 case CREQ_FUNC_EVENT_EVENT_CFCS_ERROR:
742 case CREQ_FUNC_EVENT_EVENT_CFCC_ERROR:
744 case CREQ_FUNC_EVENT_EVENT_CFCM_ERROR:
746 case CREQ_FUNC_EVENT_EVENT_TIM_ERROR:
754 static int bnxt_re_handle_qp_async_event(struct creq_qp_event *qp_event,
755 struct bnxt_re_qp *qp)
757 struct ib_event event;
760 if (qp->qplib_qp.state == CMDQ_MODIFY_QP_NEW_STATE_ERR) {
761 flags = bnxt_re_lock_cqs(qp);
762 bnxt_qplib_add_flush_qp(&qp->qplib_qp);
763 bnxt_re_unlock_cqs(qp, flags);
766 memset(&event, 0, sizeof(event));
767 if (qp->qplib_qp.srq) {
768 event.device = &qp->rdev->ibdev;
769 event.element.qp = &qp->ib_qp;
770 event.event = IB_EVENT_QP_LAST_WQE_REACHED;
773 if (event.device && qp->ib_qp.event_handler)
774 qp->ib_qp.event_handler(&event, qp->ib_qp.qp_context);
779 static int bnxt_re_handle_affi_async_event(struct creq_qp_event *affi_async,
786 return rc; /* QP was already dead, still return success */
788 event = affi_async->event;
789 if (event == CREQ_QP_EVENT_EVENT_QP_ERROR_NOTIFICATION) {
790 struct bnxt_qplib_qp *lib_qp = obj;
791 struct bnxt_re_qp *qp = container_of(lib_qp, struct bnxt_re_qp,
793 rc = bnxt_re_handle_qp_async_event(affi_async, qp);
798 static int bnxt_re_aeq_handler(struct bnxt_qplib_rcfw *rcfw,
799 void *aeqe, void *obj)
801 struct creq_qp_event *affi_async;
802 struct creq_func_event *unaffi_async;
806 type = ((struct creq_base *)aeqe)->type;
807 if (type == CREQ_BASE_TYPE_FUNC_EVENT) {
809 rc = bnxt_re_handle_unaffi_async_event(unaffi_async);
812 rc = bnxt_re_handle_affi_async_event(affi_async, obj);
818 static int bnxt_re_srqn_handler(struct bnxt_qplib_nq *nq,
819 struct bnxt_qplib_srq *handle, u8 event)
821 struct bnxt_re_srq *srq = container_of(handle, struct bnxt_re_srq,
823 struct ib_event ib_event;
827 dev_err(NULL, "%s: SRQ is NULL, SRQN not handled",
828 ROCE_DRV_MODULE_NAME);
832 ib_event.device = &srq->rdev->ibdev;
833 ib_event.element.srq = &srq->ib_srq;
834 if (event == NQ_SRQ_EVENT_EVENT_SRQ_THRESHOLD_EVENT)
835 ib_event.event = IB_EVENT_SRQ_LIMIT_REACHED;
837 ib_event.event = IB_EVENT_SRQ_ERR;
839 if (srq->ib_srq.event_handler) {
840 /* Lock event_handler? */
841 (*srq->ib_srq.event_handler)(&ib_event,
842 srq->ib_srq.srq_context);
848 static int bnxt_re_cqn_handler(struct bnxt_qplib_nq *nq,
849 struct bnxt_qplib_cq *handle)
851 struct bnxt_re_cq *cq = container_of(handle, struct bnxt_re_cq,
855 dev_err(NULL, "%s: CQ is NULL, CQN not handled",
856 ROCE_DRV_MODULE_NAME);
859 if (cq->ib_cq.comp_handler) {
860 /* Lock comp_handler? */
861 (*cq->ib_cq.comp_handler)(&cq->ib_cq, cq->ib_cq.cq_context);
867 static void bnxt_re_cleanup_res(struct bnxt_re_dev *rdev)
871 for (i = 1; i < rdev->num_msix; i++)
872 bnxt_qplib_disable_nq(&rdev->nq[i - 1]);
874 if (rdev->qplib_res.rcfw)
875 bnxt_qplib_cleanup_res(&rdev->qplib_res);
878 static int bnxt_re_init_res(struct bnxt_re_dev *rdev)
881 int num_vec_enabled = 0;
883 bnxt_qplib_init_res(&rdev->qplib_res);
885 for (i = 1; i < rdev->num_msix ; i++) {
886 rc = bnxt_qplib_enable_nq(rdev->en_dev->pdev, &rdev->nq[i - 1],
887 i - 1, rdev->msix_entries[i].vector,
888 rdev->msix_entries[i].db_offset,
889 &bnxt_re_cqn_handler,
890 &bnxt_re_srqn_handler);
893 dev_err(rdev_to_dev(rdev),
894 "Failed to enable NQ with rc = 0x%x", rc);
901 for (i = num_vec_enabled; i >= 0; i--)
902 bnxt_qplib_disable_nq(&rdev->nq[i]);
907 static void bnxt_re_free_nq_res(struct bnxt_re_dev *rdev)
911 for (i = 0; i < rdev->num_msix - 1; i++) {
912 bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id);
913 bnxt_qplib_free_nq(&rdev->nq[i]);
917 static void bnxt_re_free_res(struct bnxt_re_dev *rdev)
919 bnxt_re_free_nq_res(rdev);
921 if (rdev->qplib_res.dpi_tbl.max) {
922 bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
923 &rdev->qplib_res.dpi_tbl,
924 &rdev->dpi_privileged);
926 if (rdev->qplib_res.rcfw) {
927 bnxt_qplib_free_res(&rdev->qplib_res);
928 rdev->qplib_res.rcfw = NULL;
932 static int bnxt_re_alloc_res(struct bnxt_re_dev *rdev)
935 int num_vec_created = 0;
937 /* Configure and allocate resources for qplib */
938 rdev->qplib_res.rcfw = &rdev->rcfw;
939 rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr,
944 rc = bnxt_qplib_alloc_res(&rdev->qplib_res, rdev->en_dev->pdev,
945 rdev->netdev, &rdev->dev_attr);
949 rc = bnxt_qplib_alloc_dpi(&rdev->qplib_res.dpi_tbl,
950 &rdev->dpi_privileged,
955 for (i = 0; i < rdev->num_msix - 1; i++) {
956 rdev->nq[i].hwq.max_elements = BNXT_RE_MAX_CQ_COUNT +
957 BNXT_RE_MAX_SRQC_COUNT + 2;
958 rc = bnxt_qplib_alloc_nq(rdev->en_dev->pdev, &rdev->nq[i]);
960 dev_err(rdev_to_dev(rdev), "Alloc Failed NQ%d rc:%#x",
964 rc = bnxt_re_net_ring_alloc
965 (rdev, rdev->nq[i].hwq.pbl[PBL_LVL_0].pg_map_arr,
966 rdev->nq[i].hwq.pbl[rdev->nq[i].hwq.level].pg_count,
967 HWRM_RING_ALLOC_CMPL,
968 BNXT_QPLIB_NQE_MAX_CNT - 1,
969 rdev->msix_entries[i + 1].ring_idx,
970 &rdev->nq[i].ring_id);
972 dev_err(rdev_to_dev(rdev),
973 "Failed to allocate NQ fw id with rc = 0x%x",
975 bnxt_qplib_free_nq(&rdev->nq[i]);
982 for (i = num_vec_created; i >= 0; i--) {
983 bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id);
984 bnxt_qplib_free_nq(&rdev->nq[i]);
986 bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
987 &rdev->qplib_res.dpi_tbl,
988 &rdev->dpi_privileged);
990 bnxt_qplib_free_res(&rdev->qplib_res);
993 rdev->qplib_res.rcfw = NULL;
997 static void bnxt_re_dispatch_event(struct ib_device *ibdev, struct ib_qp *qp,
998 u8 port_num, enum ib_event_type event)
1000 struct ib_event ib_event;
1002 ib_event.device = ibdev;
1004 ib_event.element.qp = qp;
1005 ib_event.event = event;
1006 if (qp->event_handler)
1007 qp->event_handler(&ib_event, qp->qp_context);
1010 ib_event.element.port_num = port_num;
1011 ib_event.event = event;
1012 ib_dispatch_event(&ib_event);
1016 #define HWRM_QUEUE_PRI2COS_QCFG_INPUT_FLAGS_IVLAN 0x02
1017 static int bnxt_re_query_hwrm_pri2cos(struct bnxt_re_dev *rdev, u8 dir,
1020 struct hwrm_queue_pri2cos_qcfg_input req = {0};
1021 struct bnxt *bp = netdev_priv(rdev->netdev);
1022 struct hwrm_queue_pri2cos_qcfg_output resp;
1023 struct bnxt_en_dev *en_dev = rdev->en_dev;
1024 struct bnxt_fw_msg fw_msg;
1026 u8 *qcfgmap, *tmp_map;
1032 memset(&fw_msg, 0, sizeof(fw_msg));
1033 bnxt_re_init_hwrm_hdr(rdev, (void *)&req,
1034 HWRM_QUEUE_PRI2COS_QCFG, -1, -1);
1035 flags |= (dir & 0x01);
1036 flags |= HWRM_QUEUE_PRI2COS_QCFG_INPUT_FLAGS_IVLAN;
1037 req.flags = cpu_to_le32(flags);
1038 req.port_id = bp->pf.port_id;
1040 bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
1041 sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
1042 rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
1046 if (resp.queue_cfg_info) {
1047 dev_warn(rdev_to_dev(rdev),
1048 "Asymmetric cos queue configuration detected");
1049 dev_warn(rdev_to_dev(rdev),
1050 " on device, QoS may not be fully functional\n");
1052 qcfgmap = &resp.pri0_cos_queue_id;
1053 tmp_map = (u8 *)cid_map;
1054 for (i = 0; i < IEEE_8021QAZ_MAX_TCS; i++)
1055 tmp_map[i] = qcfgmap[i];
1060 static bool bnxt_re_is_qp1_or_shadow_qp(struct bnxt_re_dev *rdev,
1061 struct bnxt_re_qp *qp)
1063 return (qp->ib_qp.qp_type == IB_QPT_GSI) || (qp == rdev->qp1_sqp);
1066 static void bnxt_re_dev_stop(struct bnxt_re_dev *rdev)
1068 int mask = IB_QP_STATE;
1069 struct ib_qp_attr qp_attr;
1070 struct bnxt_re_qp *qp;
1072 qp_attr.qp_state = IB_QPS_ERR;
1073 mutex_lock(&rdev->qp_lock);
1074 list_for_each_entry(qp, &rdev->qp_list, list) {
1075 /* Modify the state of all QPs except QP1/Shadow QP */
1076 if (!bnxt_re_is_qp1_or_shadow_qp(rdev, qp)) {
1077 if (qp->qplib_qp.state !=
1078 CMDQ_MODIFY_QP_NEW_STATE_RESET &&
1079 qp->qplib_qp.state !=
1080 CMDQ_MODIFY_QP_NEW_STATE_ERR) {
1081 bnxt_re_dispatch_event(&rdev->ibdev, &qp->ib_qp,
1082 1, IB_EVENT_QP_FATAL);
1083 bnxt_re_modify_qp(&qp->ib_qp, &qp_attr, mask,
1088 mutex_unlock(&rdev->qp_lock);
1091 static int bnxt_re_update_gid(struct bnxt_re_dev *rdev)
1093 struct bnxt_qplib_sgid_tbl *sgid_tbl = &rdev->qplib_res.sgid_tbl;
1094 struct bnxt_qplib_gid gid;
1098 if (!test_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags))
1102 dev_err(rdev_to_dev(rdev), "QPLIB: SGID table not allocated");
1106 for (index = 0; index < sgid_tbl->active; index++) {
1107 gid_idx = sgid_tbl->hw_id[index];
1109 if (!memcmp(&sgid_tbl->tbl[index], &bnxt_qplib_gid_zero,
1110 sizeof(bnxt_qplib_gid_zero)))
1112 /* need to modify the VLAN enable setting of non VLAN GID only
1113 * as setting is done for VLAN GID while adding GID
1115 if (sgid_tbl->vlan[index])
1118 memcpy(&gid, &sgid_tbl->tbl[index], sizeof(gid));
1120 rc = bnxt_qplib_update_sgid(sgid_tbl, &gid, gid_idx,
1121 rdev->qplib_res.netdev->dev_addr);
1127 static u32 bnxt_re_get_priority_mask(struct bnxt_re_dev *rdev)
1129 u32 prio_map = 0, tmp_map = 0;
1130 struct net_device *netdev;
1133 netdev = rdev->netdev;
1135 memset(&app, 0, sizeof(app));
1136 app.selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE;
1137 app.protocol = ETH_P_IBOE;
1138 tmp_map = dcb_ieee_getapp_mask(netdev, &app);
1141 app.selector = IEEE_8021QAZ_APP_SEL_DGRAM;
1142 app.protocol = ROCE_V2_UDP_DPORT;
1143 tmp_map = dcb_ieee_getapp_mask(netdev, &app);
1144 prio_map |= tmp_map;
1149 static void bnxt_re_parse_cid_map(u8 prio_map, u8 *cid_map, u16 *cosq)
1154 for (prio = 0, id = 0; prio < 8; prio++) {
1155 if (prio_map & (1 << prio)) {
1156 cosq[id] = cid_map[prio];
1158 if (id == 2) /* Max 2 tcs supported */
1164 static int bnxt_re_setup_qos(struct bnxt_re_dev *rdev)
1170 /* Get priority for roce */
1171 prio_map = bnxt_re_get_priority_mask(rdev);
1173 if (prio_map == rdev->cur_prio_map)
1175 rdev->cur_prio_map = prio_map;
1176 /* Get cosq id for this priority */
1177 rc = bnxt_re_query_hwrm_pri2cos(rdev, 0, &cid_map);
1179 dev_warn(rdev_to_dev(rdev), "no cos for p_mask %x\n", prio_map);
1182 /* Parse CoS IDs for app priority */
1183 bnxt_re_parse_cid_map(prio_map, (u8 *)&cid_map, rdev->cosq);
1186 rc = bnxt_qplib_map_tc2cos(&rdev->qplib_res, rdev->cosq);
1188 dev_warn(rdev_to_dev(rdev), "no tc for cos{%x, %x}\n",
1189 rdev->cosq[0], rdev->cosq[1]);
1193 /* Actual priorities are not programmed as they are already
1194 * done by L2 driver; just enable or disable priority vlan tagging
1196 if ((prio_map == 0 && rdev->qplib_res.prio) ||
1197 (prio_map != 0 && !rdev->qplib_res.prio)) {
1198 rdev->qplib_res.prio = prio_map ? true : false;
1200 bnxt_re_update_gid(rdev);
1206 static void bnxt_re_ib_unreg(struct bnxt_re_dev *rdev)
1210 if (test_and_clear_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags)) {
1211 /* Cleanup ib dev */
1212 bnxt_re_unregister_ib(rdev);
1214 if (test_and_clear_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags))
1215 cancel_delayed_work_sync(&rdev->worker);
1217 if (test_and_clear_bit(BNXT_RE_FLAG_RESOURCES_INITIALIZED,
1219 bnxt_re_cleanup_res(rdev);
1220 if (test_and_clear_bit(BNXT_RE_FLAG_RESOURCES_ALLOCATED, &rdev->flags))
1221 bnxt_re_free_res(rdev);
1223 if (test_and_clear_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags)) {
1224 rc = bnxt_qplib_deinit_rcfw(&rdev->rcfw);
1226 dev_warn(rdev_to_dev(rdev),
1227 "Failed to deinitialize RCFW: %#x", rc);
1228 bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id);
1229 bnxt_qplib_free_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx);
1230 bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
1231 bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id);
1232 bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
1234 if (test_and_clear_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags)) {
1235 rc = bnxt_re_free_msix(rdev);
1237 dev_warn(rdev_to_dev(rdev),
1238 "Failed to free MSI-X vectors: %#x", rc);
1240 if (test_and_clear_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags)) {
1241 rc = bnxt_re_unregister_netdev(rdev);
1243 dev_warn(rdev_to_dev(rdev),
1244 "Failed to unregister with netdev: %#x", rc);
1248 /* worker thread for polling periodic events. Now used for QoS programming*/
1249 static void bnxt_re_worker(struct work_struct *work)
1251 struct bnxt_re_dev *rdev = container_of(work, struct bnxt_re_dev,
1254 bnxt_re_setup_qos(rdev);
1255 schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
1258 static int bnxt_re_ib_reg(struct bnxt_re_dev *rdev)
1264 /* Acquire rtnl lock through out this function */
1268 /* Registered a new RoCE device instance to netdev */
1269 rc = bnxt_re_register_netdev(rdev);
1272 pr_err("Failed to register with netedev: %#x\n", rc);
1275 set_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
1277 /* Check whether VF or PF */
1278 bnxt_re_get_sriov_func_type(rdev);
1280 rc = bnxt_re_request_msix(rdev);
1282 pr_err("Failed to get MSI-X vectors: %#x\n", rc);
1286 set_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags);
1288 /* Establish RCFW Communication Channel to initialize the context
1289 * memory for the function and all child VFs
1291 rc = bnxt_qplib_alloc_rcfw_channel(rdev->en_dev->pdev, &rdev->rcfw,
1292 BNXT_RE_MAX_QPC_COUNT);
1294 pr_err("Failed to allocate RCFW Channel: %#x\n", rc);
1297 rc = bnxt_re_net_ring_alloc
1298 (rdev, rdev->rcfw.creq.pbl[PBL_LVL_0].pg_map_arr,
1299 rdev->rcfw.creq.pbl[rdev->rcfw.creq.level].pg_count,
1300 HWRM_RING_ALLOC_CMPL, BNXT_QPLIB_CREQE_MAX_CNT - 1,
1301 rdev->msix_entries[BNXT_RE_AEQ_IDX].ring_idx,
1302 &rdev->rcfw.creq_ring_id);
1304 pr_err("Failed to allocate CREQ: %#x\n", rc);
1307 rc = bnxt_qplib_enable_rcfw_channel
1308 (rdev->en_dev->pdev, &rdev->rcfw,
1309 rdev->msix_entries[BNXT_RE_AEQ_IDX].vector,
1310 rdev->msix_entries[BNXT_RE_AEQ_IDX].db_offset,
1311 rdev->is_virtfn, &bnxt_re_aeq_handler);
1313 pr_err("Failed to enable RCFW channel: %#x\n", rc);
1317 rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr,
1321 if (!rdev->is_virtfn)
1322 bnxt_re_set_resource_limits(rdev);
1324 rc = bnxt_qplib_alloc_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx, 0);
1326 pr_err("Failed to allocate QPLIB context: %#x\n", rc);
1329 rc = bnxt_re_net_stats_ctx_alloc(rdev,
1330 rdev->qplib_ctx.stats.dma_map,
1331 &rdev->qplib_ctx.stats.fw_id);
1333 pr_err("Failed to allocate stats context: %#x\n", rc);
1337 rc = bnxt_qplib_init_rcfw(&rdev->rcfw, &rdev->qplib_ctx,
1340 pr_err("Failed to initialize RCFW: %#x\n", rc);
1343 set_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags);
1345 /* Resources based on the 'new' device caps */
1346 rc = bnxt_re_alloc_res(rdev);
1348 pr_err("Failed to allocate resources: %#x\n", rc);
1351 set_bit(BNXT_RE_FLAG_RESOURCES_ALLOCATED, &rdev->flags);
1352 rc = bnxt_re_init_res(rdev);
1354 pr_err("Failed to initialize resources: %#x\n", rc);
1358 set_bit(BNXT_RE_FLAG_RESOURCES_INITIALIZED, &rdev->flags);
1360 if (!rdev->is_virtfn) {
1361 rc = bnxt_re_setup_qos(rdev);
1363 pr_info("RoCE priority not yet configured\n");
1365 INIT_DELAYED_WORK(&rdev->worker, bnxt_re_worker);
1366 set_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags);
1367 schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
1373 /* Register ib dev */
1374 rc = bnxt_re_register_ib(rdev);
1376 pr_err("Failed to register with IB: %#x\n", rc);
1379 set_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags);
1380 dev_info(rdev_to_dev(rdev), "Device registered successfully");
1381 ib_get_eth_speed(&rdev->ibdev, 1, &rdev->active_speed,
1382 &rdev->active_width);
1383 set_bit(BNXT_RE_FLAG_ISSUE_ROCE_STATS, &rdev->flags);
1384 bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1, IB_EVENT_PORT_ACTIVE);
1385 bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1, IB_EVENT_GID_CHANGE);
1389 bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id);
1391 bnxt_qplib_free_ctx(rdev->en_dev->pdev, &rdev->qplib_ctx);
1393 bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
1395 bnxt_re_net_ring_free(rdev, rdev->rcfw.creq_ring_id);
1397 bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
1401 bnxt_re_ib_unreg(rdev);
1407 static void bnxt_re_dev_unreg(struct bnxt_re_dev *rdev)
1409 struct bnxt_en_dev *en_dev = rdev->en_dev;
1410 struct net_device *netdev = rdev->netdev;
1412 bnxt_re_dev_remove(rdev);
1415 bnxt_re_dev_unprobe(netdev, en_dev);
1418 static int bnxt_re_dev_reg(struct bnxt_re_dev **rdev, struct net_device *netdev)
1420 struct bnxt_en_dev *en_dev;
1423 if (!is_bnxt_re_dev(netdev))
1426 en_dev = bnxt_re_dev_probe(netdev);
1427 if (IS_ERR(en_dev)) {
1428 if (en_dev != ERR_PTR(-ENODEV))
1429 pr_err("%s: Failed to probe\n", ROCE_DRV_MODULE_NAME);
1430 rc = PTR_ERR(en_dev);
1433 *rdev = bnxt_re_dev_add(netdev, en_dev);
1436 bnxt_re_dev_unprobe(netdev, en_dev);
1443 static void bnxt_re_remove_one(struct bnxt_re_dev *rdev)
1445 pci_dev_put(rdev->en_dev->pdev);
1448 /* Handle all deferred netevents tasks */
1449 static void bnxt_re_task(struct work_struct *work)
1451 struct bnxt_re_work *re_work;
1452 struct bnxt_re_dev *rdev;
1455 re_work = container_of(work, struct bnxt_re_work, work);
1456 rdev = re_work->rdev;
1458 if (re_work->event != NETDEV_REGISTER &&
1459 !test_bit(BNXT_RE_FLAG_IBDEV_REGISTERED, &rdev->flags))
1462 switch (re_work->event) {
1463 case NETDEV_REGISTER:
1464 rc = bnxt_re_ib_reg(rdev);
1466 dev_err(rdev_to_dev(rdev),
1467 "Failed to register with IB: %#x", rc);
1468 bnxt_re_remove_one(rdev);
1469 bnxt_re_dev_unreg(rdev);
1474 bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1475 IB_EVENT_PORT_ACTIVE);
1478 bnxt_re_dev_stop(rdev);
1481 if (!netif_carrier_ok(rdev->netdev))
1482 bnxt_re_dev_stop(rdev);
1483 else if (netif_carrier_ok(rdev->netdev))
1484 bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1485 IB_EVENT_PORT_ACTIVE);
1486 ib_get_eth_speed(&rdev->ibdev, 1, &rdev->active_speed,
1487 &rdev->active_width);
1492 smp_mb__before_atomic();
1493 atomic_dec(&rdev->sched_count);
1498 static void bnxt_re_init_one(struct bnxt_re_dev *rdev)
1500 pci_dev_get(rdev->en_dev->pdev);
1504 * "Notifier chain callback can be invoked for the same chain from
1505 * different CPUs at the same time".
1507 * For cases when the netdev is already present, our call to the
1508 * register_netdevice_notifier() will actually get the rtnl_lock()
1509 * before sending NETDEV_REGISTER and (if up) NETDEV_UP
1512 * But for cases when the netdev is not already present, the notifier
1513 * chain is subjected to be invoked from different CPUs simultaneously.
1515 * This is protected by the netdev_mutex.
1517 static int bnxt_re_netdev_event(struct notifier_block *notifier,
1518 unsigned long event, void *ptr)
1520 struct net_device *real_dev, *netdev = netdev_notifier_info_to_dev(ptr);
1521 struct bnxt_re_work *re_work;
1522 struct bnxt_re_dev *rdev;
1524 bool sch_work = false;
1526 real_dev = rdma_vlan_dev_real_dev(netdev);
1530 rdev = bnxt_re_from_netdev(real_dev);
1531 if (!rdev && event != NETDEV_REGISTER)
1533 if (real_dev != netdev)
1537 case NETDEV_REGISTER:
1540 rc = bnxt_re_dev_reg(&rdev, real_dev);
1544 pr_err("Failed to register with the device %s: %#x\n",
1545 real_dev->name, rc);
1548 bnxt_re_init_one(rdev);
1552 case NETDEV_UNREGISTER:
1553 /* netdev notifier will call NETDEV_UNREGISTER again later since
1554 * we are still holding the reference to the netdev
1556 if (atomic_read(&rdev->sched_count) > 0)
1558 bnxt_re_ib_unreg(rdev);
1559 bnxt_re_remove_one(rdev);
1560 bnxt_re_dev_unreg(rdev);
1568 /* Allocate for the deferred task */
1569 re_work = kzalloc(sizeof(*re_work), GFP_ATOMIC);
1571 re_work->rdev = rdev;
1572 re_work->event = event;
1573 re_work->vlan_dev = (real_dev == netdev ?
1575 INIT_WORK(&re_work->work, bnxt_re_task);
1576 atomic_inc(&rdev->sched_count);
1577 queue_work(bnxt_re_wq, &re_work->work);
1585 static struct notifier_block bnxt_re_netdev_notifier = {
1586 .notifier_call = bnxt_re_netdev_event
1589 static int __init bnxt_re_mod_init(void)
1593 pr_info("%s: %s", ROCE_DRV_MODULE_NAME, version);
1595 bnxt_re_wq = create_singlethread_workqueue("bnxt_re");
1599 INIT_LIST_HEAD(&bnxt_re_dev_list);
1601 rc = register_netdevice_notifier(&bnxt_re_netdev_notifier);
1603 pr_err("%s: Cannot register to netdevice_notifier",
1604 ROCE_DRV_MODULE_NAME);
1610 destroy_workqueue(bnxt_re_wq);
1615 static void __exit bnxt_re_mod_exit(void)
1617 struct bnxt_re_dev *rdev, *next;
1618 LIST_HEAD(to_be_deleted);
1620 mutex_lock(&bnxt_re_dev_lock);
1621 /* Free all adapter allocated resources */
1622 if (!list_empty(&bnxt_re_dev_list))
1623 list_splice_init(&bnxt_re_dev_list, &to_be_deleted);
1624 mutex_unlock(&bnxt_re_dev_lock);
1626 * Cleanup the devices in reverse order so that the VF device
1627 * cleanup is done before PF cleanup
1629 list_for_each_entry_safe_reverse(rdev, next, &to_be_deleted, list) {
1630 dev_info(rdev_to_dev(rdev), "Unregistering Device");
1632 * Flush out any scheduled tasks before destroying the
1635 flush_workqueue(bnxt_re_wq);
1636 bnxt_re_dev_stop(rdev);
1637 /* Acquire the rtnl_lock as the L2 resources are freed here */
1639 bnxt_re_ib_unreg(rdev);
1641 bnxt_re_remove_one(rdev);
1642 bnxt_re_dev_unreg(rdev);
1644 unregister_netdevice_notifier(&bnxt_re_netdev_notifier);
1646 destroy_workqueue(bnxt_re_wq);
1649 module_init(bnxt_re_mod_init);
1650 module_exit(bnxt_re_mod_exit);