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[] =
72 MODULE_AUTHOR("Eddie Wai <eddie.wai@broadcom.com>");
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_remove_device(struct bnxt_re_dev *rdev);
82 static void bnxt_re_dealloc_driver(struct ib_device *ib_dev);
83 static void bnxt_re_stop_irq(void *handle);
85 static void bnxt_re_set_drv_mode(struct bnxt_re_dev *rdev, u8 mode)
87 struct bnxt_qplib_chip_ctx *cctx;
89 cctx = rdev->chip_ctx;
90 cctx->modes.wqe_mode = bnxt_qplib_is_chip_gen_p5(rdev->chip_ctx) ?
91 mode : BNXT_QPLIB_WQE_MODE_STATIC;
94 static void bnxt_re_destroy_chip_ctx(struct bnxt_re_dev *rdev)
96 struct bnxt_qplib_chip_ctx *chip_ctx;
100 chip_ctx = rdev->chip_ctx;
101 rdev->chip_ctx = NULL;
102 rdev->rcfw.res = NULL;
103 rdev->qplib_res.cctx = NULL;
104 rdev->qplib_res.pdev = NULL;
105 rdev->qplib_res.netdev = NULL;
109 static int bnxt_re_setup_chip_ctx(struct bnxt_re_dev *rdev, u8 wqe_mode)
111 struct bnxt_qplib_chip_ctx *chip_ctx;
112 struct bnxt_en_dev *en_dev;
115 en_dev = rdev->en_dev;
116 bp = netdev_priv(en_dev->net);
118 chip_ctx = kzalloc(sizeof(*chip_ctx), GFP_KERNEL);
121 chip_ctx->chip_num = bp->chip_num;
123 rdev->chip_ctx = chip_ctx;
124 /* rest members to follow eventually */
126 rdev->qplib_res.cctx = rdev->chip_ctx;
127 rdev->rcfw.res = &rdev->qplib_res;
129 bnxt_re_set_drv_mode(rdev, wqe_mode);
133 /* SR-IOV helper functions */
135 static void bnxt_re_get_sriov_func_type(struct bnxt_re_dev *rdev)
139 bp = netdev_priv(rdev->en_dev->net);
144 /* Set the maximum number of each resource that the driver actually wants
145 * to allocate. This may be up to the maximum number the firmware has
146 * reserved for the function. The driver may choose to allocate fewer
147 * resources than the firmware maximum.
149 static void bnxt_re_limit_pf_res(struct bnxt_re_dev *rdev)
151 struct bnxt_qplib_dev_attr *attr;
152 struct bnxt_qplib_ctx *ctx;
155 attr = &rdev->dev_attr;
156 ctx = &rdev->qplib_ctx;
158 ctx->qpc_count = min_t(u32, BNXT_RE_MAX_QPC_COUNT,
160 ctx->mrw_count = BNXT_RE_MAX_MRW_COUNT_256K;
161 /* Use max_mr from fw since max_mrw does not get set */
162 ctx->mrw_count = min_t(u32, ctx->mrw_count, attr->max_mr);
163 ctx->srqc_count = min_t(u32, BNXT_RE_MAX_SRQC_COUNT,
165 ctx->cq_count = min_t(u32, BNXT_RE_MAX_CQ_COUNT, attr->max_cq);
166 if (!bnxt_qplib_is_chip_gen_p5(rdev->chip_ctx))
167 for (i = 0; i < MAX_TQM_ALLOC_REQ; i++)
168 rdev->qplib_ctx.tqm_ctx.qcount[i] =
169 rdev->dev_attr.tqm_alloc_reqs[i];
172 static void bnxt_re_limit_vf_res(struct bnxt_qplib_ctx *qplib_ctx, u32 num_vf)
174 struct bnxt_qplib_vf_res *vf_res;
179 vf_res = &qplib_ctx->vf_res;
181 * Reserve a set of resources for the PF. Divide the remaining
182 * resources among the VFs
184 vf_pct = 100 - BNXT_RE_PCT_RSVD_FOR_PF;
186 num_vf = 100 * num_vf;
187 vf_res->max_qp_per_vf = (qplib_ctx->qpc_count * vf_pct) / num_vf;
188 vf_res->max_srq_per_vf = (qplib_ctx->srqc_count * vf_pct) / num_vf;
189 vf_res->max_cq_per_vf = (qplib_ctx->cq_count * vf_pct) / num_vf;
191 * The driver allows many more MRs than other resources. If the
192 * firmware does also, then reserve a fixed amount for the PF and
193 * divide the rest among VFs. VFs may use many MRs for NFS
194 * mounts, ISER, NVME applications, etc. If the firmware severely
195 * restricts the number of MRs, then let PF have half and divide
196 * the rest among VFs, as for the other resource types.
198 if (qplib_ctx->mrw_count < BNXT_RE_MAX_MRW_COUNT_64K) {
199 mrws = qplib_ctx->mrw_count * vf_pct;
202 mrws = qplib_ctx->mrw_count - BNXT_RE_RESVD_MR_FOR_PF;
204 vf_res->max_mrw_per_vf = (mrws / nvfs);
205 vf_res->max_gid_per_vf = BNXT_RE_MAX_GID_PER_VF;
208 static void bnxt_re_set_resource_limits(struct bnxt_re_dev *rdev)
212 memset(&rdev->qplib_ctx.vf_res, 0, sizeof(struct bnxt_qplib_vf_res));
213 bnxt_re_limit_pf_res(rdev);
215 num_vfs = bnxt_qplib_is_chip_gen_p5(rdev->chip_ctx) ?
216 BNXT_RE_GEN_P5_MAX_VF : rdev->num_vfs;
218 bnxt_re_limit_vf_res(&rdev->qplib_ctx, num_vfs);
221 /* for handling bnxt_en callbacks later */
222 static void bnxt_re_stop(void *p)
226 static void bnxt_re_start(void *p)
230 static void bnxt_re_sriov_config(void *p, int num_vfs)
232 struct bnxt_re_dev *rdev = p;
237 rdev->num_vfs = num_vfs;
238 if (!bnxt_qplib_is_chip_gen_p5(rdev->chip_ctx)) {
239 bnxt_re_set_resource_limits(rdev);
240 bnxt_qplib_set_func_resources(&rdev->qplib_res, &rdev->rcfw,
245 static void bnxt_re_shutdown(void *p)
247 struct bnxt_re_dev *rdev = p;
252 /* Release the MSIx vectors before queuing unregister */
253 bnxt_re_stop_irq(rdev);
254 ib_unregister_device_queued(&rdev->ibdev);
257 static void bnxt_re_stop_irq(void *handle)
259 struct bnxt_re_dev *rdev = (struct bnxt_re_dev *)handle;
260 struct bnxt_qplib_rcfw *rcfw = &rdev->rcfw;
261 struct bnxt_qplib_nq *nq;
264 for (indx = BNXT_RE_NQ_IDX; indx < rdev->num_msix; indx++) {
265 nq = &rdev->nq[indx - 1];
266 bnxt_qplib_nq_stop_irq(nq, false);
269 bnxt_qplib_rcfw_stop_irq(rcfw, false);
272 static void bnxt_re_start_irq(void *handle, struct bnxt_msix_entry *ent)
274 struct bnxt_re_dev *rdev = (struct bnxt_re_dev *)handle;
275 struct bnxt_msix_entry *msix_ent = rdev->msix_entries;
276 struct bnxt_qplib_rcfw *rcfw = &rdev->rcfw;
277 struct bnxt_qplib_nq *nq;
281 /* Not setting the f/w timeout bit in rcfw.
282 * During the driver unload the first command
283 * to f/w will timeout and that will set the
286 ibdev_err(&rdev->ibdev, "Failed to re-start IRQs\n");
290 /* Vectors may change after restart, so update with new vectors
291 * in device sctructure.
293 for (indx = 0; indx < rdev->num_msix; indx++)
294 rdev->msix_entries[indx].vector = ent[indx].vector;
296 bnxt_qplib_rcfw_start_irq(rcfw, msix_ent[BNXT_RE_AEQ_IDX].vector,
298 for (indx = BNXT_RE_NQ_IDX ; indx < rdev->num_msix; indx++) {
299 nq = &rdev->nq[indx - 1];
300 rc = bnxt_qplib_nq_start_irq(nq, indx - 1,
301 msix_ent[indx].vector, false);
303 ibdev_warn(&rdev->ibdev, "Failed to reinit NQ index %d\n",
308 static struct bnxt_ulp_ops bnxt_re_ulp_ops = {
309 .ulp_async_notifier = NULL,
310 .ulp_stop = bnxt_re_stop,
311 .ulp_start = bnxt_re_start,
312 .ulp_sriov_config = bnxt_re_sriov_config,
313 .ulp_shutdown = bnxt_re_shutdown,
314 .ulp_irq_stop = bnxt_re_stop_irq,
315 .ulp_irq_restart = bnxt_re_start_irq
318 /* RoCE -> Net driver */
320 /* Driver registration routines used to let the networking driver (bnxt_en)
321 * to know that the RoCE driver is now installed
323 static int bnxt_re_unregister_netdev(struct bnxt_re_dev *rdev)
325 struct bnxt_en_dev *en_dev;
331 en_dev = rdev->en_dev;
333 rc = en_dev->en_ops->bnxt_unregister_device(rdev->en_dev,
338 static int bnxt_re_register_netdev(struct bnxt_re_dev *rdev)
340 struct bnxt_en_dev *en_dev;
346 en_dev = rdev->en_dev;
348 rc = en_dev->en_ops->bnxt_register_device(en_dev, BNXT_ROCE_ULP,
349 &bnxt_re_ulp_ops, rdev);
350 rdev->qplib_res.pdev = rdev->en_dev->pdev;
354 static int bnxt_re_free_msix(struct bnxt_re_dev *rdev)
356 struct bnxt_en_dev *en_dev;
362 en_dev = rdev->en_dev;
365 rc = en_dev->en_ops->bnxt_free_msix(rdev->en_dev, BNXT_ROCE_ULP);
370 static int bnxt_re_request_msix(struct bnxt_re_dev *rdev)
372 int rc = 0, num_msix_want = BNXT_RE_MAX_MSIX, num_msix_got;
373 struct bnxt_en_dev *en_dev;
378 en_dev = rdev->en_dev;
380 num_msix_want = min_t(u32, BNXT_RE_MAX_MSIX, num_online_cpus());
382 num_msix_got = en_dev->en_ops->bnxt_request_msix(en_dev, BNXT_ROCE_ULP,
385 if (num_msix_got < BNXT_RE_MIN_MSIX) {
389 if (num_msix_got != num_msix_want) {
390 ibdev_warn(&rdev->ibdev,
391 "Requested %d MSI-X vectors, got %d\n",
392 num_msix_want, num_msix_got);
394 rdev->num_msix = num_msix_got;
399 static void bnxt_re_init_hwrm_hdr(struct bnxt_re_dev *rdev, struct input *hdr,
400 u16 opcd, u16 crid, u16 trid)
402 hdr->req_type = cpu_to_le16(opcd);
403 hdr->cmpl_ring = cpu_to_le16(crid);
404 hdr->target_id = cpu_to_le16(trid);
407 static void bnxt_re_fill_fw_msg(struct bnxt_fw_msg *fw_msg, void *msg,
408 int msg_len, void *resp, int resp_max_len,
412 fw_msg->msg_len = msg_len;
414 fw_msg->resp_max_len = resp_max_len;
415 fw_msg->timeout = timeout;
418 static int bnxt_re_net_ring_free(struct bnxt_re_dev *rdev,
419 u16 fw_ring_id, int type)
421 struct bnxt_en_dev *en_dev = rdev->en_dev;
422 struct hwrm_ring_free_input req = {0};
423 struct hwrm_ring_free_output resp;
424 struct bnxt_fw_msg fw_msg;
430 memset(&fw_msg, 0, sizeof(fw_msg));
432 bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_RING_FREE, -1, -1);
433 req.ring_type = type;
434 req.ring_id = cpu_to_le16(fw_ring_id);
435 bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
436 sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
437 rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
439 ibdev_err(&rdev->ibdev, "Failed to free HW ring:%d :%#x",
444 static int bnxt_re_net_ring_alloc(struct bnxt_re_dev *rdev,
445 struct bnxt_re_ring_attr *ring_attr,
448 struct bnxt_en_dev *en_dev = rdev->en_dev;
449 struct hwrm_ring_alloc_input req = {0};
450 struct hwrm_ring_alloc_output resp;
451 struct bnxt_fw_msg fw_msg;
457 memset(&fw_msg, 0, sizeof(fw_msg));
458 bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_RING_ALLOC, -1, -1);
460 req.page_tbl_addr = cpu_to_le64(ring_attr->dma_arr[0]);
461 if (ring_attr->pages > 1) {
462 /* Page size is in log2 units */
463 req.page_size = BNXT_PAGE_SHIFT;
464 req.page_tbl_depth = 1;
467 /* Association of ring index with doorbell index and MSIX number */
468 req.logical_id = cpu_to_le16(ring_attr->lrid);
469 req.length = cpu_to_le32(ring_attr->depth + 1);
470 req.ring_type = ring_attr->type;
471 req.int_mode = ring_attr->mode;
472 bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
473 sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
474 rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
476 *fw_ring_id = le16_to_cpu(resp.ring_id);
481 static int bnxt_re_net_stats_ctx_free(struct bnxt_re_dev *rdev,
484 struct bnxt_en_dev *en_dev = rdev->en_dev;
485 struct hwrm_stat_ctx_free_input req = {0};
486 struct bnxt_fw_msg fw_msg;
492 memset(&fw_msg, 0, sizeof(fw_msg));
494 bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_STAT_CTX_FREE, -1, -1);
495 req.stat_ctx_id = cpu_to_le32(fw_stats_ctx_id);
496 bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&req,
497 sizeof(req), DFLT_HWRM_CMD_TIMEOUT);
498 rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
500 ibdev_err(&rdev->ibdev, "Failed to free HW stats context %#x",
506 static int bnxt_re_net_stats_ctx_alloc(struct bnxt_re_dev *rdev,
508 u32 *fw_stats_ctx_id)
510 struct hwrm_stat_ctx_alloc_output resp = {0};
511 struct hwrm_stat_ctx_alloc_input req = {0};
512 struct bnxt_en_dev *en_dev = rdev->en_dev;
513 struct bnxt_fw_msg fw_msg;
516 *fw_stats_ctx_id = INVALID_STATS_CTX_ID;
521 memset(&fw_msg, 0, sizeof(fw_msg));
523 bnxt_re_init_hwrm_hdr(rdev, (void *)&req, HWRM_STAT_CTX_ALLOC, -1, -1);
524 req.update_period_ms = cpu_to_le32(1000);
525 req.stats_dma_addr = cpu_to_le64(dma_map);
526 req.stats_dma_length = cpu_to_le16(sizeof(struct ctx_hw_stats_ext));
527 req.stat_ctx_flags = STAT_CTX_ALLOC_REQ_STAT_CTX_FLAGS_ROCE;
528 bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
529 sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
530 rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
532 *fw_stats_ctx_id = le32_to_cpu(resp.stat_ctx_id);
539 static bool is_bnxt_re_dev(struct net_device *netdev)
541 struct ethtool_drvinfo drvinfo;
543 if (netdev->ethtool_ops && netdev->ethtool_ops->get_drvinfo) {
544 memset(&drvinfo, 0, sizeof(drvinfo));
545 netdev->ethtool_ops->get_drvinfo(netdev, &drvinfo);
547 if (strcmp(drvinfo.driver, "bnxt_en"))
554 static struct bnxt_re_dev *bnxt_re_from_netdev(struct net_device *netdev)
556 struct ib_device *ibdev =
557 ib_device_get_by_netdev(netdev, RDMA_DRIVER_BNXT_RE);
561 return container_of(ibdev, struct bnxt_re_dev, ibdev);
564 static void bnxt_re_dev_unprobe(struct net_device *netdev,
565 struct bnxt_en_dev *en_dev)
568 module_put(en_dev->pdev->driver->driver.owner);
571 static struct bnxt_en_dev *bnxt_re_dev_probe(struct net_device *netdev)
573 struct bnxt *bp = netdev_priv(netdev);
574 struct bnxt_en_dev *en_dev;
575 struct pci_dev *pdev;
577 /* Call bnxt_en's RoCE probe via indirect API */
579 return ERR_PTR(-EINVAL);
581 en_dev = bp->ulp_probe(netdev);
587 return ERR_PTR(-EINVAL);
589 if (!(en_dev->flags & BNXT_EN_FLAG_ROCE_CAP)) {
591 "%s: probe error: RoCE is not supported on this device",
592 ROCE_DRV_MODULE_NAME);
593 return ERR_PTR(-ENODEV);
596 /* Bump net device reference count */
597 if (!try_module_get(pdev->driver->driver.owner))
598 return ERR_PTR(-ENODEV);
605 static ssize_t hw_rev_show(struct device *device, struct device_attribute *attr,
608 struct bnxt_re_dev *rdev =
609 rdma_device_to_drv_device(device, struct bnxt_re_dev, ibdev);
611 return scnprintf(buf, PAGE_SIZE, "0x%x\n", rdev->en_dev->pdev->vendor);
613 static DEVICE_ATTR_RO(hw_rev);
615 static ssize_t hca_type_show(struct device *device,
616 struct device_attribute *attr, char *buf)
618 struct bnxt_re_dev *rdev =
619 rdma_device_to_drv_device(device, struct bnxt_re_dev, ibdev);
621 return scnprintf(buf, PAGE_SIZE, "%s\n", rdev->ibdev.node_desc);
623 static DEVICE_ATTR_RO(hca_type);
625 static struct attribute *bnxt_re_attributes[] = {
626 &dev_attr_hw_rev.attr,
627 &dev_attr_hca_type.attr,
631 static const struct attribute_group bnxt_re_dev_attr_group = {
632 .attrs = bnxt_re_attributes,
635 static const struct ib_device_ops bnxt_re_dev_ops = {
636 .owner = THIS_MODULE,
637 .driver_id = RDMA_DRIVER_BNXT_RE,
638 .uverbs_abi_ver = BNXT_RE_ABI_VERSION,
640 .add_gid = bnxt_re_add_gid,
641 .alloc_hw_stats = bnxt_re_ib_alloc_hw_stats,
642 .alloc_mr = bnxt_re_alloc_mr,
643 .alloc_pd = bnxt_re_alloc_pd,
644 .alloc_ucontext = bnxt_re_alloc_ucontext,
645 .create_ah = bnxt_re_create_ah,
646 .create_cq = bnxt_re_create_cq,
647 .create_qp = bnxt_re_create_qp,
648 .create_srq = bnxt_re_create_srq,
649 .dealloc_driver = bnxt_re_dealloc_driver,
650 .dealloc_pd = bnxt_re_dealloc_pd,
651 .dealloc_ucontext = bnxt_re_dealloc_ucontext,
652 .del_gid = bnxt_re_del_gid,
653 .dereg_mr = bnxt_re_dereg_mr,
654 .destroy_ah = bnxt_re_destroy_ah,
655 .destroy_cq = bnxt_re_destroy_cq,
656 .destroy_qp = bnxt_re_destroy_qp,
657 .destroy_srq = bnxt_re_destroy_srq,
658 .get_dev_fw_str = bnxt_re_query_fw_str,
659 .get_dma_mr = bnxt_re_get_dma_mr,
660 .get_hw_stats = bnxt_re_ib_get_hw_stats,
661 .get_link_layer = bnxt_re_get_link_layer,
662 .get_port_immutable = bnxt_re_get_port_immutable,
663 .map_mr_sg = bnxt_re_map_mr_sg,
664 .mmap = bnxt_re_mmap,
665 .modify_ah = bnxt_re_modify_ah,
666 .modify_qp = bnxt_re_modify_qp,
667 .modify_srq = bnxt_re_modify_srq,
668 .poll_cq = bnxt_re_poll_cq,
669 .post_recv = bnxt_re_post_recv,
670 .post_send = bnxt_re_post_send,
671 .post_srq_recv = bnxt_re_post_srq_recv,
672 .query_ah = bnxt_re_query_ah,
673 .query_device = bnxt_re_query_device,
674 .query_pkey = bnxt_re_query_pkey,
675 .query_port = bnxt_re_query_port,
676 .query_qp = bnxt_re_query_qp,
677 .query_srq = bnxt_re_query_srq,
678 .reg_user_mr = bnxt_re_reg_user_mr,
679 .req_notify_cq = bnxt_re_req_notify_cq,
680 INIT_RDMA_OBJ_SIZE(ib_ah, bnxt_re_ah, ib_ah),
681 INIT_RDMA_OBJ_SIZE(ib_cq, bnxt_re_cq, ib_cq),
682 INIT_RDMA_OBJ_SIZE(ib_pd, bnxt_re_pd, ib_pd),
683 INIT_RDMA_OBJ_SIZE(ib_srq, bnxt_re_srq, ib_srq),
684 INIT_RDMA_OBJ_SIZE(ib_ucontext, bnxt_re_ucontext, ib_uctx),
687 static int bnxt_re_register_ib(struct bnxt_re_dev *rdev)
689 struct ib_device *ibdev = &rdev->ibdev;
693 ibdev->node_type = RDMA_NODE_IB_CA;
694 strlcpy(ibdev->node_desc, BNXT_RE_DESC " HCA",
695 strlen(BNXT_RE_DESC) + 5);
696 ibdev->phys_port_cnt = 1;
698 bnxt_qplib_get_guid(rdev->netdev->dev_addr, (u8 *)&ibdev->node_guid);
700 ibdev->num_comp_vectors = rdev->num_msix - 1;
701 ibdev->dev.parent = &rdev->en_dev->pdev->dev;
702 ibdev->local_dma_lkey = BNXT_QPLIB_RSVD_LKEY;
705 ibdev->uverbs_cmd_mask =
706 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
707 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
708 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
709 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
710 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
711 (1ull << IB_USER_VERBS_CMD_REG_MR) |
712 (1ull << IB_USER_VERBS_CMD_REREG_MR) |
713 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
714 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
715 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
716 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ) |
717 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
718 (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
719 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
720 (1ull << IB_USER_VERBS_CMD_QUERY_QP) |
721 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
722 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) |
723 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) |
724 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) |
725 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ) |
726 (1ull << IB_USER_VERBS_CMD_CREATE_AH) |
727 (1ull << IB_USER_VERBS_CMD_MODIFY_AH) |
728 (1ull << IB_USER_VERBS_CMD_QUERY_AH) |
729 (1ull << IB_USER_VERBS_CMD_DESTROY_AH);
730 /* POLL_CQ and REQ_NOTIFY_CQ is directly handled in libbnxt_re */
733 rdma_set_device_sysfs_group(ibdev, &bnxt_re_dev_attr_group);
734 ib_set_device_ops(ibdev, &bnxt_re_dev_ops);
735 ret = ib_device_set_netdev(&rdev->ibdev, rdev->netdev, 1);
739 return ib_register_device(ibdev, "bnxt_re%d");
742 static void bnxt_re_dev_remove(struct bnxt_re_dev *rdev)
744 dev_put(rdev->netdev);
746 mutex_lock(&bnxt_re_dev_lock);
747 list_del_rcu(&rdev->list);
748 mutex_unlock(&bnxt_re_dev_lock);
753 static struct bnxt_re_dev *bnxt_re_dev_add(struct net_device *netdev,
754 struct bnxt_en_dev *en_dev)
756 struct bnxt_re_dev *rdev;
758 /* Allocate bnxt_re_dev instance here */
759 rdev = ib_alloc_device(bnxt_re_dev, ibdev);
761 ibdev_err(NULL, "%s: bnxt_re_dev allocation failure!",
762 ROCE_DRV_MODULE_NAME);
766 rdev->netdev = netdev;
767 dev_hold(rdev->netdev);
768 rdev->en_dev = en_dev;
769 rdev->id = rdev->en_dev->pdev->devfn;
770 INIT_LIST_HEAD(&rdev->qp_list);
771 mutex_init(&rdev->qp_lock);
772 atomic_set(&rdev->qp_count, 0);
773 atomic_set(&rdev->cq_count, 0);
774 atomic_set(&rdev->srq_count, 0);
775 atomic_set(&rdev->mr_count, 0);
776 atomic_set(&rdev->mw_count, 0);
777 rdev->cosq[0] = 0xFFFF;
778 rdev->cosq[1] = 0xFFFF;
780 mutex_lock(&bnxt_re_dev_lock);
781 list_add_tail_rcu(&rdev->list, &bnxt_re_dev_list);
782 mutex_unlock(&bnxt_re_dev_lock);
786 static int bnxt_re_handle_unaffi_async_event(struct creq_func_event
789 switch (unaffi_async->event) {
790 case CREQ_FUNC_EVENT_EVENT_TX_WQE_ERROR:
792 case CREQ_FUNC_EVENT_EVENT_TX_DATA_ERROR:
794 case CREQ_FUNC_EVENT_EVENT_RX_WQE_ERROR:
796 case CREQ_FUNC_EVENT_EVENT_RX_DATA_ERROR:
798 case CREQ_FUNC_EVENT_EVENT_CQ_ERROR:
800 case CREQ_FUNC_EVENT_EVENT_TQM_ERROR:
802 case CREQ_FUNC_EVENT_EVENT_CFCQ_ERROR:
804 case CREQ_FUNC_EVENT_EVENT_CFCS_ERROR:
806 case CREQ_FUNC_EVENT_EVENT_CFCC_ERROR:
808 case CREQ_FUNC_EVENT_EVENT_CFCM_ERROR:
810 case CREQ_FUNC_EVENT_EVENT_TIM_ERROR:
818 static int bnxt_re_handle_qp_async_event(struct creq_qp_event *qp_event,
819 struct bnxt_re_qp *qp)
821 struct ib_event event;
824 if (qp->qplib_qp.state == CMDQ_MODIFY_QP_NEW_STATE_ERR &&
825 rdma_is_kernel_res(&qp->ib_qp.res)) {
826 flags = bnxt_re_lock_cqs(qp);
827 bnxt_qplib_add_flush_qp(&qp->qplib_qp);
828 bnxt_re_unlock_cqs(qp, flags);
831 memset(&event, 0, sizeof(event));
832 if (qp->qplib_qp.srq) {
833 event.device = &qp->rdev->ibdev;
834 event.element.qp = &qp->ib_qp;
835 event.event = IB_EVENT_QP_LAST_WQE_REACHED;
838 if (event.device && qp->ib_qp.event_handler)
839 qp->ib_qp.event_handler(&event, qp->ib_qp.qp_context);
844 static int bnxt_re_handle_affi_async_event(struct creq_qp_event *affi_async,
851 return rc; /* QP was already dead, still return success */
853 event = affi_async->event;
854 if (event == CREQ_QP_EVENT_EVENT_QP_ERROR_NOTIFICATION) {
855 struct bnxt_qplib_qp *lib_qp = obj;
856 struct bnxt_re_qp *qp = container_of(lib_qp, struct bnxt_re_qp,
858 rc = bnxt_re_handle_qp_async_event(affi_async, qp);
863 static int bnxt_re_aeq_handler(struct bnxt_qplib_rcfw *rcfw,
864 void *aeqe, void *obj)
866 struct creq_qp_event *affi_async;
867 struct creq_func_event *unaffi_async;
871 type = ((struct creq_base *)aeqe)->type;
872 if (type == CREQ_BASE_TYPE_FUNC_EVENT) {
874 rc = bnxt_re_handle_unaffi_async_event(unaffi_async);
877 rc = bnxt_re_handle_affi_async_event(affi_async, obj);
883 static int bnxt_re_srqn_handler(struct bnxt_qplib_nq *nq,
884 struct bnxt_qplib_srq *handle, u8 event)
886 struct bnxt_re_srq *srq = container_of(handle, struct bnxt_re_srq,
888 struct ib_event ib_event;
892 ibdev_err(NULL, "%s: SRQ is NULL, SRQN not handled",
893 ROCE_DRV_MODULE_NAME);
897 ib_event.device = &srq->rdev->ibdev;
898 ib_event.element.srq = &srq->ib_srq;
899 if (event == NQ_SRQ_EVENT_EVENT_SRQ_THRESHOLD_EVENT)
900 ib_event.event = IB_EVENT_SRQ_LIMIT_REACHED;
902 ib_event.event = IB_EVENT_SRQ_ERR;
904 if (srq->ib_srq.event_handler) {
905 /* Lock event_handler? */
906 (*srq->ib_srq.event_handler)(&ib_event,
907 srq->ib_srq.srq_context);
913 static int bnxt_re_cqn_handler(struct bnxt_qplib_nq *nq,
914 struct bnxt_qplib_cq *handle)
916 struct bnxt_re_cq *cq = container_of(handle, struct bnxt_re_cq,
920 ibdev_err(NULL, "%s: CQ is NULL, CQN not handled",
921 ROCE_DRV_MODULE_NAME);
924 if (cq->ib_cq.comp_handler) {
925 /* Lock comp_handler? */
926 (*cq->ib_cq.comp_handler)(&cq->ib_cq, cq->ib_cq.cq_context);
932 #define BNXT_RE_GEN_P5_PF_NQ_DB 0x10000
933 #define BNXT_RE_GEN_P5_VF_NQ_DB 0x4000
934 static u32 bnxt_re_get_nqdb_offset(struct bnxt_re_dev *rdev, u16 indx)
936 return bnxt_qplib_is_chip_gen_p5(rdev->chip_ctx) ?
937 (rdev->is_virtfn ? BNXT_RE_GEN_P5_VF_NQ_DB :
938 BNXT_RE_GEN_P5_PF_NQ_DB) :
939 rdev->msix_entries[indx].db_offset;
942 static void bnxt_re_cleanup_res(struct bnxt_re_dev *rdev)
946 for (i = 1; i < rdev->num_msix; i++)
947 bnxt_qplib_disable_nq(&rdev->nq[i - 1]);
949 if (rdev->qplib_res.rcfw)
950 bnxt_qplib_cleanup_res(&rdev->qplib_res);
953 static int bnxt_re_init_res(struct bnxt_re_dev *rdev)
955 int num_vec_enabled = 0;
959 bnxt_qplib_init_res(&rdev->qplib_res);
961 for (i = 1; i < rdev->num_msix ; i++) {
962 db_offt = bnxt_re_get_nqdb_offset(rdev, i);
963 rc = bnxt_qplib_enable_nq(rdev->en_dev->pdev, &rdev->nq[i - 1],
964 i - 1, rdev->msix_entries[i].vector,
965 db_offt, &bnxt_re_cqn_handler,
966 &bnxt_re_srqn_handler);
968 ibdev_err(&rdev->ibdev,
969 "Failed to enable NQ with rc = 0x%x", rc);
976 for (i = num_vec_enabled; i >= 0; i--)
977 bnxt_qplib_disable_nq(&rdev->nq[i]);
981 static void bnxt_re_free_nq_res(struct bnxt_re_dev *rdev)
986 for (i = 0; i < rdev->num_msix - 1; i++) {
987 type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
988 bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id, type);
989 bnxt_qplib_free_nq(&rdev->nq[i]);
990 rdev->nq[i].res = NULL;
994 static void bnxt_re_free_res(struct bnxt_re_dev *rdev)
996 bnxt_re_free_nq_res(rdev);
998 if (rdev->qplib_res.dpi_tbl.max) {
999 bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
1000 &rdev->qplib_res.dpi_tbl,
1001 &rdev->dpi_privileged);
1003 if (rdev->qplib_res.rcfw) {
1004 bnxt_qplib_free_res(&rdev->qplib_res);
1005 rdev->qplib_res.rcfw = NULL;
1009 static int bnxt_re_alloc_res(struct bnxt_re_dev *rdev)
1011 struct bnxt_re_ring_attr rattr = {};
1012 struct bnxt_qplib_ctx *qplib_ctx;
1013 int num_vec_created = 0;
1017 /* Configure and allocate resources for qplib */
1018 rdev->qplib_res.rcfw = &rdev->rcfw;
1019 rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr,
1024 rc = bnxt_qplib_alloc_res(&rdev->qplib_res, rdev->en_dev->pdev,
1025 rdev->netdev, &rdev->dev_attr);
1029 rc = bnxt_qplib_alloc_dpi(&rdev->qplib_res.dpi_tbl,
1030 &rdev->dpi_privileged,
1035 qplib_ctx = &rdev->qplib_ctx;
1036 for (i = 0; i < rdev->num_msix - 1; i++) {
1037 struct bnxt_qplib_nq *nq;
1040 nq->hwq.max_elements = (qplib_ctx->cq_count +
1041 qplib_ctx->srqc_count + 2);
1042 rc = bnxt_qplib_alloc_nq(&rdev->qplib_res, &rdev->nq[i]);
1044 ibdev_err(&rdev->ibdev, "Alloc Failed NQ%d rc:%#x",
1048 type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
1049 rattr.dma_arr = nq->hwq.pbl[PBL_LVL_0].pg_map_arr;
1050 rattr.pages = nq->hwq.pbl[rdev->nq[i].hwq.level].pg_count;
1052 rattr.mode = RING_ALLOC_REQ_INT_MODE_MSIX;
1053 rattr.depth = BNXT_QPLIB_NQE_MAX_CNT - 1;
1054 rattr.lrid = rdev->msix_entries[i + 1].ring_idx;
1055 rc = bnxt_re_net_ring_alloc(rdev, &rattr, &nq->ring_id);
1057 ibdev_err(&rdev->ibdev,
1058 "Failed to allocate NQ fw id with rc = 0x%x",
1060 bnxt_qplib_free_nq(&rdev->nq[i]);
1067 for (i = num_vec_created - 1; i >= 0; i--) {
1068 type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
1069 bnxt_re_net_ring_free(rdev, rdev->nq[i].ring_id, type);
1070 bnxt_qplib_free_nq(&rdev->nq[i]);
1072 bnxt_qplib_dealloc_dpi(&rdev->qplib_res,
1073 &rdev->qplib_res.dpi_tbl,
1074 &rdev->dpi_privileged);
1076 bnxt_qplib_free_res(&rdev->qplib_res);
1079 rdev->qplib_res.rcfw = NULL;
1083 static void bnxt_re_dispatch_event(struct ib_device *ibdev, struct ib_qp *qp,
1084 u8 port_num, enum ib_event_type event)
1086 struct ib_event ib_event;
1088 ib_event.device = ibdev;
1090 ib_event.element.qp = qp;
1091 ib_event.event = event;
1092 if (qp->event_handler)
1093 qp->event_handler(&ib_event, qp->qp_context);
1096 ib_event.element.port_num = port_num;
1097 ib_event.event = event;
1098 ib_dispatch_event(&ib_event);
1102 #define HWRM_QUEUE_PRI2COS_QCFG_INPUT_FLAGS_IVLAN 0x02
1103 static int bnxt_re_query_hwrm_pri2cos(struct bnxt_re_dev *rdev, u8 dir,
1106 struct hwrm_queue_pri2cos_qcfg_input req = {0};
1107 struct bnxt *bp = netdev_priv(rdev->netdev);
1108 struct hwrm_queue_pri2cos_qcfg_output resp;
1109 struct bnxt_en_dev *en_dev = rdev->en_dev;
1110 struct bnxt_fw_msg fw_msg;
1112 u8 *qcfgmap, *tmp_map;
1118 memset(&fw_msg, 0, sizeof(fw_msg));
1119 bnxt_re_init_hwrm_hdr(rdev, (void *)&req,
1120 HWRM_QUEUE_PRI2COS_QCFG, -1, -1);
1121 flags |= (dir & 0x01);
1122 flags |= HWRM_QUEUE_PRI2COS_QCFG_INPUT_FLAGS_IVLAN;
1123 req.flags = cpu_to_le32(flags);
1124 req.port_id = bp->pf.port_id;
1126 bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
1127 sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
1128 rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
1132 if (resp.queue_cfg_info) {
1133 ibdev_warn(&rdev->ibdev,
1134 "Asymmetric cos queue configuration detected");
1135 ibdev_warn(&rdev->ibdev,
1136 " on device, QoS may not be fully functional\n");
1138 qcfgmap = &resp.pri0_cos_queue_id;
1139 tmp_map = (u8 *)cid_map;
1140 for (i = 0; i < IEEE_8021QAZ_MAX_TCS; i++)
1141 tmp_map[i] = qcfgmap[i];
1146 static bool bnxt_re_is_qp1_or_shadow_qp(struct bnxt_re_dev *rdev,
1147 struct bnxt_re_qp *qp)
1149 return (qp->ib_qp.qp_type == IB_QPT_GSI) ||
1150 (qp == rdev->gsi_ctx.gsi_sqp);
1153 static void bnxt_re_dev_stop(struct bnxt_re_dev *rdev)
1155 int mask = IB_QP_STATE;
1156 struct ib_qp_attr qp_attr;
1157 struct bnxt_re_qp *qp;
1159 qp_attr.qp_state = IB_QPS_ERR;
1160 mutex_lock(&rdev->qp_lock);
1161 list_for_each_entry(qp, &rdev->qp_list, list) {
1162 /* Modify the state of all QPs except QP1/Shadow QP */
1163 if (!bnxt_re_is_qp1_or_shadow_qp(rdev, qp)) {
1164 if (qp->qplib_qp.state !=
1165 CMDQ_MODIFY_QP_NEW_STATE_RESET &&
1166 qp->qplib_qp.state !=
1167 CMDQ_MODIFY_QP_NEW_STATE_ERR) {
1168 bnxt_re_dispatch_event(&rdev->ibdev, &qp->ib_qp,
1169 1, IB_EVENT_QP_FATAL);
1170 bnxt_re_modify_qp(&qp->ib_qp, &qp_attr, mask,
1175 mutex_unlock(&rdev->qp_lock);
1178 static int bnxt_re_update_gid(struct bnxt_re_dev *rdev)
1180 struct bnxt_qplib_sgid_tbl *sgid_tbl = &rdev->qplib_res.sgid_tbl;
1181 struct bnxt_qplib_gid gid;
1185 if (!ib_device_try_get(&rdev->ibdev))
1189 ibdev_err(&rdev->ibdev, "QPLIB: SGID table not allocated");
1194 for (index = 0; index < sgid_tbl->active; index++) {
1195 gid_idx = sgid_tbl->hw_id[index];
1197 if (!memcmp(&sgid_tbl->tbl[index], &bnxt_qplib_gid_zero,
1198 sizeof(bnxt_qplib_gid_zero)))
1200 /* need to modify the VLAN enable setting of non VLAN GID only
1201 * as setting is done for VLAN GID while adding GID
1203 if (sgid_tbl->vlan[index])
1206 memcpy(&gid, &sgid_tbl->tbl[index], sizeof(gid));
1208 rc = bnxt_qplib_update_sgid(sgid_tbl, &gid, gid_idx,
1209 rdev->qplib_res.netdev->dev_addr);
1212 ib_device_put(&rdev->ibdev);
1216 static u32 bnxt_re_get_priority_mask(struct bnxt_re_dev *rdev)
1218 u32 prio_map = 0, tmp_map = 0;
1219 struct net_device *netdev;
1222 netdev = rdev->netdev;
1224 memset(&app, 0, sizeof(app));
1225 app.selector = IEEE_8021QAZ_APP_SEL_ETHERTYPE;
1226 app.protocol = ETH_P_IBOE;
1227 tmp_map = dcb_ieee_getapp_mask(netdev, &app);
1230 app.selector = IEEE_8021QAZ_APP_SEL_DGRAM;
1231 app.protocol = ROCE_V2_UDP_DPORT;
1232 tmp_map = dcb_ieee_getapp_mask(netdev, &app);
1233 prio_map |= tmp_map;
1238 static void bnxt_re_parse_cid_map(u8 prio_map, u8 *cid_map, u16 *cosq)
1243 for (prio = 0, id = 0; prio < 8; prio++) {
1244 if (prio_map & (1 << prio)) {
1245 cosq[id] = cid_map[prio];
1247 if (id == 2) /* Max 2 tcs supported */
1253 static int bnxt_re_setup_qos(struct bnxt_re_dev *rdev)
1259 /* Get priority for roce */
1260 prio_map = bnxt_re_get_priority_mask(rdev);
1262 if (prio_map == rdev->cur_prio_map)
1264 rdev->cur_prio_map = prio_map;
1265 /* Get cosq id for this priority */
1266 rc = bnxt_re_query_hwrm_pri2cos(rdev, 0, &cid_map);
1268 ibdev_warn(&rdev->ibdev, "no cos for p_mask %x\n", prio_map);
1271 /* Parse CoS IDs for app priority */
1272 bnxt_re_parse_cid_map(prio_map, (u8 *)&cid_map, rdev->cosq);
1275 rc = bnxt_qplib_map_tc2cos(&rdev->qplib_res, rdev->cosq);
1277 ibdev_warn(&rdev->ibdev, "no tc for cos{%x, %x}\n",
1278 rdev->cosq[0], rdev->cosq[1]);
1282 /* Actual priorities are not programmed as they are already
1283 * done by L2 driver; just enable or disable priority vlan tagging
1285 if ((prio_map == 0 && rdev->qplib_res.prio) ||
1286 (prio_map != 0 && !rdev->qplib_res.prio)) {
1287 rdev->qplib_res.prio = prio_map ? true : false;
1289 bnxt_re_update_gid(rdev);
1295 static void bnxt_re_query_hwrm_intf_version(struct bnxt_re_dev *rdev)
1297 struct bnxt_en_dev *en_dev = rdev->en_dev;
1298 struct hwrm_ver_get_output resp = {0};
1299 struct hwrm_ver_get_input req = {0};
1300 struct bnxt_fw_msg fw_msg;
1303 memset(&fw_msg, 0, sizeof(fw_msg));
1304 bnxt_re_init_hwrm_hdr(rdev, (void *)&req,
1305 HWRM_VER_GET, -1, -1);
1306 req.hwrm_intf_maj = HWRM_VERSION_MAJOR;
1307 req.hwrm_intf_min = HWRM_VERSION_MINOR;
1308 req.hwrm_intf_upd = HWRM_VERSION_UPDATE;
1309 bnxt_re_fill_fw_msg(&fw_msg, (void *)&req, sizeof(req), (void *)&resp,
1310 sizeof(resp), DFLT_HWRM_CMD_TIMEOUT);
1311 rc = en_dev->en_ops->bnxt_send_fw_msg(en_dev, BNXT_ROCE_ULP, &fw_msg);
1313 ibdev_err(&rdev->ibdev, "Failed to query HW version, rc = 0x%x",
1317 rdev->qplib_ctx.hwrm_intf_ver =
1318 (u64)le16_to_cpu(resp.hwrm_intf_major) << 48 |
1319 (u64)le16_to_cpu(resp.hwrm_intf_minor) << 32 |
1320 (u64)le16_to_cpu(resp.hwrm_intf_build) << 16 |
1321 le16_to_cpu(resp.hwrm_intf_patch);
1324 static int bnxt_re_ib_init(struct bnxt_re_dev *rdev)
1329 /* Register ib dev */
1330 rc = bnxt_re_register_ib(rdev);
1332 pr_err("Failed to register with IB: %#x\n", rc);
1335 dev_info(rdev_to_dev(rdev), "Device registered successfully");
1336 ib_get_eth_speed(&rdev->ibdev, 1, &rdev->active_speed,
1337 &rdev->active_width);
1338 set_bit(BNXT_RE_FLAG_ISSUE_ROCE_STATS, &rdev->flags);
1340 event = netif_running(rdev->netdev) && netif_carrier_ok(rdev->netdev) ?
1341 IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
1343 bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1, event);
1348 static void bnxt_re_dev_uninit(struct bnxt_re_dev *rdev)
1353 if (test_and_clear_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags))
1354 cancel_delayed_work_sync(&rdev->worker);
1356 if (test_and_clear_bit(BNXT_RE_FLAG_RESOURCES_INITIALIZED,
1358 bnxt_re_cleanup_res(rdev);
1359 if (test_and_clear_bit(BNXT_RE_FLAG_RESOURCES_ALLOCATED, &rdev->flags))
1360 bnxt_re_free_res(rdev);
1362 if (test_and_clear_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags)) {
1363 rc = bnxt_qplib_deinit_rcfw(&rdev->rcfw);
1365 ibdev_warn(&rdev->ibdev,
1366 "Failed to deinitialize RCFW: %#x", rc);
1367 bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id);
1368 bnxt_qplib_free_ctx(&rdev->qplib_res, &rdev->qplib_ctx);
1369 bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
1370 type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
1371 bnxt_re_net_ring_free(rdev, rdev->rcfw.creq.ring_id, type);
1372 bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
1374 if (test_and_clear_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags)) {
1375 rc = bnxt_re_free_msix(rdev);
1377 ibdev_warn(&rdev->ibdev,
1378 "Failed to free MSI-X vectors: %#x", rc);
1381 bnxt_re_destroy_chip_ctx(rdev);
1382 if (test_and_clear_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags)) {
1383 rc = bnxt_re_unregister_netdev(rdev);
1385 ibdev_warn(&rdev->ibdev,
1386 "Failed to unregister with netdev: %#x", rc);
1390 /* worker thread for polling periodic events. Now used for QoS programming*/
1391 static void bnxt_re_worker(struct work_struct *work)
1393 struct bnxt_re_dev *rdev = container_of(work, struct bnxt_re_dev,
1396 bnxt_re_setup_qos(rdev);
1397 schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
1400 static int bnxt_re_dev_init(struct bnxt_re_dev *rdev, u8 wqe_mode)
1402 struct bnxt_qplib_creq_ctx *creq;
1403 struct bnxt_re_ring_attr rattr;
1409 /* Registered a new RoCE device instance to netdev */
1410 memset(&rattr, 0, sizeof(rattr));
1411 rc = bnxt_re_register_netdev(rdev);
1414 ibdev_err(&rdev->ibdev,
1415 "Failed to register with netedev: %#x\n", rc);
1418 set_bit(BNXT_RE_FLAG_NETDEV_REGISTERED, &rdev->flags);
1420 rc = bnxt_re_setup_chip_ctx(rdev, wqe_mode);
1422 ibdev_err(&rdev->ibdev, "Failed to get chip context\n");
1426 /* Check whether VF or PF */
1427 bnxt_re_get_sriov_func_type(rdev);
1429 rc = bnxt_re_request_msix(rdev);
1431 ibdev_err(&rdev->ibdev,
1432 "Failed to get MSI-X vectors: %#x\n", rc);
1436 set_bit(BNXT_RE_FLAG_GOT_MSIX, &rdev->flags);
1438 bnxt_re_query_hwrm_intf_version(rdev);
1440 /* Establish RCFW Communication Channel to initialize the context
1441 * memory for the function and all child VFs
1443 rc = bnxt_qplib_alloc_rcfw_channel(&rdev->qplib_res, &rdev->rcfw,
1445 BNXT_RE_MAX_QPC_COUNT);
1447 ibdev_err(&rdev->ibdev,
1448 "Failed to allocate RCFW Channel: %#x\n", rc);
1452 type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
1453 creq = &rdev->rcfw.creq;
1454 rattr.dma_arr = creq->hwq.pbl[PBL_LVL_0].pg_map_arr;
1455 rattr.pages = creq->hwq.pbl[creq->hwq.level].pg_count;
1457 rattr.mode = RING_ALLOC_REQ_INT_MODE_MSIX;
1458 rattr.depth = BNXT_QPLIB_CREQE_MAX_CNT - 1;
1459 rattr.lrid = rdev->msix_entries[BNXT_RE_AEQ_IDX].ring_idx;
1460 rc = bnxt_re_net_ring_alloc(rdev, &rattr, &creq->ring_id);
1462 ibdev_err(&rdev->ibdev, "Failed to allocate CREQ: %#x\n", rc);
1465 db_offt = bnxt_re_get_nqdb_offset(rdev, BNXT_RE_AEQ_IDX);
1466 vid = rdev->msix_entries[BNXT_RE_AEQ_IDX].vector;
1467 rc = bnxt_qplib_enable_rcfw_channel(&rdev->rcfw,
1468 vid, db_offt, rdev->is_virtfn,
1469 &bnxt_re_aeq_handler);
1471 ibdev_err(&rdev->ibdev, "Failed to enable RCFW channel: %#x\n",
1476 rc = bnxt_qplib_get_dev_attr(&rdev->rcfw, &rdev->dev_attr,
1481 bnxt_re_set_resource_limits(rdev);
1483 rc = bnxt_qplib_alloc_ctx(&rdev->qplib_res, &rdev->qplib_ctx, 0,
1484 bnxt_qplib_is_chip_gen_p5(rdev->chip_ctx));
1486 ibdev_err(&rdev->ibdev,
1487 "Failed to allocate QPLIB context: %#x\n", rc);
1490 rc = bnxt_re_net_stats_ctx_alloc(rdev,
1491 rdev->qplib_ctx.stats.dma_map,
1492 &rdev->qplib_ctx.stats.fw_id);
1494 ibdev_err(&rdev->ibdev,
1495 "Failed to allocate stats context: %#x\n", rc);
1499 rc = bnxt_qplib_init_rcfw(&rdev->rcfw, &rdev->qplib_ctx,
1502 ibdev_err(&rdev->ibdev,
1503 "Failed to initialize RCFW: %#x\n", rc);
1506 set_bit(BNXT_RE_FLAG_RCFW_CHANNEL_EN, &rdev->flags);
1508 /* Resources based on the 'new' device caps */
1509 rc = bnxt_re_alloc_res(rdev);
1511 ibdev_err(&rdev->ibdev,
1512 "Failed to allocate resources: %#x\n", rc);
1515 set_bit(BNXT_RE_FLAG_RESOURCES_ALLOCATED, &rdev->flags);
1516 rc = bnxt_re_init_res(rdev);
1518 ibdev_err(&rdev->ibdev,
1519 "Failed to initialize resources: %#x\n", rc);
1523 set_bit(BNXT_RE_FLAG_RESOURCES_INITIALIZED, &rdev->flags);
1525 if (!rdev->is_virtfn) {
1526 rc = bnxt_re_setup_qos(rdev);
1528 ibdev_info(&rdev->ibdev,
1529 "RoCE priority not yet configured\n");
1531 INIT_DELAYED_WORK(&rdev->worker, bnxt_re_worker);
1532 set_bit(BNXT_RE_FLAG_QOS_WORK_REG, &rdev->flags);
1533 schedule_delayed_work(&rdev->worker, msecs_to_jiffies(30000));
1538 bnxt_re_net_stats_ctx_free(rdev, rdev->qplib_ctx.stats.fw_id);
1540 bnxt_qplib_free_ctx(&rdev->qplib_res, &rdev->qplib_ctx);
1542 bnxt_qplib_disable_rcfw_channel(&rdev->rcfw);
1544 type = bnxt_qplib_get_ring_type(rdev->chip_ctx);
1545 bnxt_re_net_ring_free(rdev, rdev->rcfw.creq.ring_id, type);
1547 bnxt_qplib_free_rcfw_channel(&rdev->rcfw);
1549 bnxt_re_dev_uninit(rdev);
1554 static void bnxt_re_dev_unreg(struct bnxt_re_dev *rdev)
1556 struct bnxt_en_dev *en_dev = rdev->en_dev;
1557 struct net_device *netdev = rdev->netdev;
1559 bnxt_re_dev_remove(rdev);
1562 bnxt_re_dev_unprobe(netdev, en_dev);
1565 static int bnxt_re_dev_reg(struct bnxt_re_dev **rdev, struct net_device *netdev)
1567 struct bnxt_en_dev *en_dev;
1570 if (!is_bnxt_re_dev(netdev))
1573 en_dev = bnxt_re_dev_probe(netdev);
1574 if (IS_ERR(en_dev)) {
1575 if (en_dev != ERR_PTR(-ENODEV))
1576 ibdev_err(&(*rdev)->ibdev, "%s: Failed to probe\n",
1577 ROCE_DRV_MODULE_NAME);
1578 rc = PTR_ERR(en_dev);
1581 *rdev = bnxt_re_dev_add(netdev, en_dev);
1584 bnxt_re_dev_unprobe(netdev, en_dev);
1591 static void bnxt_re_remove_device(struct bnxt_re_dev *rdev)
1593 bnxt_re_dev_uninit(rdev);
1594 pci_dev_put(rdev->en_dev->pdev);
1595 bnxt_re_dev_unreg(rdev);
1598 static int bnxt_re_add_device(struct bnxt_re_dev **rdev,
1599 struct net_device *netdev, u8 wqe_mode)
1603 rc = bnxt_re_dev_reg(rdev, netdev);
1607 pr_err("Failed to register with the device %s: %#x\n",
1612 pci_dev_get((*rdev)->en_dev->pdev);
1613 rc = bnxt_re_dev_init(*rdev, wqe_mode);
1615 pci_dev_put((*rdev)->en_dev->pdev);
1616 bnxt_re_dev_unreg(*rdev);
1622 static void bnxt_re_dealloc_driver(struct ib_device *ib_dev)
1624 struct bnxt_re_dev *rdev =
1625 container_of(ib_dev, struct bnxt_re_dev, ibdev);
1627 dev_info(rdev_to_dev(rdev), "Unregistering Device");
1630 bnxt_re_remove_device(rdev);
1634 /* Handle all deferred netevents tasks */
1635 static void bnxt_re_task(struct work_struct *work)
1637 struct bnxt_re_work *re_work;
1638 struct bnxt_re_dev *rdev;
1641 re_work = container_of(work, struct bnxt_re_work, work);
1642 rdev = re_work->rdev;
1644 if (re_work->event == NETDEV_REGISTER) {
1645 rc = bnxt_re_ib_init(rdev);
1647 ibdev_err(&rdev->ibdev,
1648 "Failed to register with IB: %#x", rc);
1650 bnxt_re_remove_device(rdev);
1657 if (!ib_device_try_get(&rdev->ibdev))
1660 switch (re_work->event) {
1662 bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1663 IB_EVENT_PORT_ACTIVE);
1666 bnxt_re_dev_stop(rdev);
1669 if (!netif_carrier_ok(rdev->netdev))
1670 bnxt_re_dev_stop(rdev);
1671 else if (netif_carrier_ok(rdev->netdev))
1672 bnxt_re_dispatch_event(&rdev->ibdev, NULL, 1,
1673 IB_EVENT_PORT_ACTIVE);
1674 ib_get_eth_speed(&rdev->ibdev, 1, &rdev->active_speed,
1675 &rdev->active_width);
1680 ib_device_put(&rdev->ibdev);
1682 put_device(&rdev->ibdev.dev);
1687 * "Notifier chain callback can be invoked for the same chain from
1688 * different CPUs at the same time".
1690 * For cases when the netdev is already present, our call to the
1691 * register_netdevice_notifier() will actually get the rtnl_lock()
1692 * before sending NETDEV_REGISTER and (if up) NETDEV_UP
1695 * But for cases when the netdev is not already present, the notifier
1696 * chain is subjected to be invoked from different CPUs simultaneously.
1698 * This is protected by the netdev_mutex.
1700 static int bnxt_re_netdev_event(struct notifier_block *notifier,
1701 unsigned long event, void *ptr)
1703 struct net_device *real_dev, *netdev = netdev_notifier_info_to_dev(ptr);
1704 struct bnxt_re_work *re_work;
1705 struct bnxt_re_dev *rdev;
1707 bool sch_work = false;
1708 bool release = true;
1710 real_dev = rdma_vlan_dev_real_dev(netdev);
1714 rdev = bnxt_re_from_netdev(real_dev);
1715 if (!rdev && event != NETDEV_REGISTER)
1718 if (real_dev != netdev)
1722 case NETDEV_REGISTER:
1725 rc = bnxt_re_add_device(&rdev, real_dev,
1726 BNXT_QPLIB_WQE_MODE_STATIC);
1732 case NETDEV_UNREGISTER:
1733 ib_unregister_device_queued(&rdev->ibdev);
1741 /* Allocate for the deferred task */
1742 re_work = kzalloc(sizeof(*re_work), GFP_ATOMIC);
1744 get_device(&rdev->ibdev.dev);
1745 re_work->rdev = rdev;
1746 re_work->event = event;
1747 re_work->vlan_dev = (real_dev == netdev ?
1749 INIT_WORK(&re_work->work, bnxt_re_task);
1750 queue_work(bnxt_re_wq, &re_work->work);
1755 if (rdev && release)
1756 ib_device_put(&rdev->ibdev);
1760 static struct notifier_block bnxt_re_netdev_notifier = {
1761 .notifier_call = bnxt_re_netdev_event
1764 static int __init bnxt_re_mod_init(void)
1768 pr_info("%s: %s", ROCE_DRV_MODULE_NAME, version);
1770 bnxt_re_wq = create_singlethread_workqueue("bnxt_re");
1774 INIT_LIST_HEAD(&bnxt_re_dev_list);
1776 rc = register_netdevice_notifier(&bnxt_re_netdev_notifier);
1778 pr_err("%s: Cannot register to netdevice_notifier",
1779 ROCE_DRV_MODULE_NAME);
1785 destroy_workqueue(bnxt_re_wq);
1790 static void __exit bnxt_re_mod_exit(void)
1792 struct bnxt_re_dev *rdev;
1794 unregister_netdevice_notifier(&bnxt_re_netdev_notifier);
1796 destroy_workqueue(bnxt_re_wq);
1797 list_for_each_entry(rdev, &bnxt_re_dev_list, list) {
1798 /* VF device removal should be called before the removal
1799 * of PF device. Queue VFs unregister first, so that VFs
1800 * shall be removed before the PF during the call of
1801 * ib_unregister_driver.
1803 if (rdev->is_virtfn)
1804 ib_unregister_device(&rdev->ibdev);
1806 ib_unregister_driver(RDMA_DRIVER_BNXT_RE);
1809 module_init(bnxt_re_mod_init);
1810 module_exit(bnxt_re_mod_exit);