1 // SPDX-License-Identifier: GPL-2.0 or Linux-OpenIB
2 /* Copyright (c) 2015 - 2021 Intel Corporation */
6 * irdma_query_device - get device attributes
7 * @ibdev: device pointer from stack
8 * @props: returning device attributes
11 static int irdma_query_device(struct ib_device *ibdev,
12 struct ib_device_attr *props,
13 struct ib_udata *udata)
15 struct irdma_device *iwdev = to_iwdev(ibdev);
16 struct irdma_pci_f *rf = iwdev->rf;
17 struct pci_dev *pcidev = iwdev->rf->pcidev;
18 struct irdma_hw_attrs *hw_attrs = &rf->sc_dev.hw_attrs;
20 if (udata->inlen || udata->outlen)
23 memset(props, 0, sizeof(*props));
24 ether_addr_copy((u8 *)&props->sys_image_guid, iwdev->netdev->dev_addr);
25 props->fw_ver = (u64)irdma_fw_major_ver(&rf->sc_dev) << 32 |
26 irdma_fw_minor_ver(&rf->sc_dev);
27 props->device_cap_flags = iwdev->device_cap_flags;
28 props->vendor_id = pcidev->vendor;
29 props->vendor_part_id = pcidev->device;
31 props->hw_ver = rf->pcidev->revision;
32 props->page_size_cap = SZ_4K | SZ_2M | SZ_1G;
33 props->max_mr_size = hw_attrs->max_mr_size;
34 props->max_qp = rf->max_qp - rf->used_qps;
35 props->max_qp_wr = hw_attrs->max_qp_wr;
36 props->max_send_sge = hw_attrs->uk_attrs.max_hw_wq_frags;
37 props->max_recv_sge = hw_attrs->uk_attrs.max_hw_wq_frags;
38 props->max_cq = rf->max_cq - rf->used_cqs;
39 props->max_cqe = rf->max_cqe;
40 props->max_mr = rf->max_mr - rf->used_mrs;
41 props->max_mw = props->max_mr;
42 props->max_pd = rf->max_pd - rf->used_pds;
43 props->max_sge_rd = hw_attrs->uk_attrs.max_hw_read_sges;
44 props->max_qp_rd_atom = hw_attrs->max_hw_ird;
45 props->max_qp_init_rd_atom = hw_attrs->max_hw_ord;
46 if (rdma_protocol_roce(ibdev, 1))
47 props->max_pkeys = IRDMA_PKEY_TBL_SZ;
48 props->max_ah = rf->max_ah;
49 props->max_mcast_grp = rf->max_mcg;
50 props->max_mcast_qp_attach = IRDMA_MAX_MGS_PER_CTX;
51 props->max_total_mcast_qp_attach = rf->max_qp * IRDMA_MAX_MGS_PER_CTX;
52 props->max_fast_reg_page_list_len = IRDMA_MAX_PAGES_PER_FMR;
53 #define HCA_CLOCK_TIMESTAMP_MASK 0x1ffff
54 if (hw_attrs->uk_attrs.hw_rev >= IRDMA_GEN_2)
55 props->timestamp_mask = HCA_CLOCK_TIMESTAMP_MASK;
61 * irdma_get_eth_speed_and_width - Get IB port speed and width from netdev speed
62 * @link_speed: netdev phy link speed
63 * @active_speed: IB port speed
64 * @active_width: IB port width
66 static void irdma_get_eth_speed_and_width(u32 link_speed, u16 *active_speed,
69 if (link_speed <= SPEED_1000) {
70 *active_width = IB_WIDTH_1X;
71 *active_speed = IB_SPEED_SDR;
72 } else if (link_speed <= SPEED_10000) {
73 *active_width = IB_WIDTH_1X;
74 *active_speed = IB_SPEED_FDR10;
75 } else if (link_speed <= SPEED_20000) {
76 *active_width = IB_WIDTH_4X;
77 *active_speed = IB_SPEED_DDR;
78 } else if (link_speed <= SPEED_25000) {
79 *active_width = IB_WIDTH_1X;
80 *active_speed = IB_SPEED_EDR;
81 } else if (link_speed <= SPEED_40000) {
82 *active_width = IB_WIDTH_4X;
83 *active_speed = IB_SPEED_FDR10;
85 *active_width = IB_WIDTH_4X;
86 *active_speed = IB_SPEED_EDR;
91 * irdma_query_port - get port attributes
92 * @ibdev: device pointer from stack
93 * @port: port number for query
94 * @props: returning device attributes
96 static int irdma_query_port(struct ib_device *ibdev, u32 port,
97 struct ib_port_attr *props)
99 struct irdma_device *iwdev = to_iwdev(ibdev);
100 struct net_device *netdev = iwdev->netdev;
102 /* no need to zero out pros here. done by caller */
104 props->max_mtu = IB_MTU_4096;
105 props->active_mtu = ib_mtu_int_to_enum(netdev->mtu);
110 if (netif_carrier_ok(netdev) && netif_running(netdev)) {
111 props->state = IB_PORT_ACTIVE;
112 props->phys_state = IB_PORT_PHYS_STATE_LINK_UP;
114 props->state = IB_PORT_DOWN;
115 props->phys_state = IB_PORT_PHYS_STATE_DISABLED;
117 irdma_get_eth_speed_and_width(SPEED_100000, &props->active_speed,
118 &props->active_width);
120 if (rdma_protocol_roce(ibdev, 1)) {
121 props->gid_tbl_len = 32;
122 props->ip_gids = true;
123 props->pkey_tbl_len = IRDMA_PKEY_TBL_SZ;
125 props->gid_tbl_len = 1;
127 props->qkey_viol_cntr = 0;
128 props->port_cap_flags |= IB_PORT_CM_SUP | IB_PORT_REINIT_SUP;
129 props->max_msg_sz = iwdev->rf->sc_dev.hw_attrs.max_hw_outbound_msg_size;
135 * irdma_disassociate_ucontext - Disassociate user context
136 * @context: ib user context
138 static void irdma_disassociate_ucontext(struct ib_ucontext *context)
142 static int irdma_mmap_legacy(struct irdma_ucontext *ucontext,
143 struct vm_area_struct *vma)
147 if (vma->vm_pgoff || vma->vm_end - vma->vm_start != PAGE_SIZE)
150 vma->vm_private_data = ucontext;
151 pfn = ((uintptr_t)ucontext->iwdev->rf->sc_dev.hw_regs[IRDMA_DB_ADDR_OFFSET] +
152 pci_resource_start(ucontext->iwdev->rf->pcidev, 0)) >> PAGE_SHIFT;
154 return rdma_user_mmap_io(&ucontext->ibucontext, vma, pfn, PAGE_SIZE,
155 pgprot_noncached(vma->vm_page_prot), NULL);
158 static void irdma_mmap_free(struct rdma_user_mmap_entry *rdma_entry)
160 struct irdma_user_mmap_entry *entry = to_irdma_mmap_entry(rdma_entry);
165 static struct rdma_user_mmap_entry*
166 irdma_user_mmap_entry_insert(struct irdma_ucontext *ucontext, u64 bar_offset,
167 enum irdma_mmap_flag mmap_flag, u64 *mmap_offset)
169 struct irdma_user_mmap_entry *entry = kzalloc(sizeof(*entry), GFP_KERNEL);
175 entry->bar_offset = bar_offset;
176 entry->mmap_flag = mmap_flag;
178 ret = rdma_user_mmap_entry_insert(&ucontext->ibucontext,
179 &entry->rdma_entry, PAGE_SIZE);
184 *mmap_offset = rdma_user_mmap_get_offset(&entry->rdma_entry);
186 return &entry->rdma_entry;
190 * irdma_mmap - user memory map
191 * @context: context created during alloc
192 * @vma: kernel info for user memory map
194 static int irdma_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
196 struct rdma_user_mmap_entry *rdma_entry;
197 struct irdma_user_mmap_entry *entry;
198 struct irdma_ucontext *ucontext;
202 ucontext = to_ucontext(context);
204 /* Legacy support for libi40iw with hard-coded mmap key */
205 if (ucontext->legacy_mode)
206 return irdma_mmap_legacy(ucontext, vma);
208 rdma_entry = rdma_user_mmap_entry_get(&ucontext->ibucontext, vma);
210 ibdev_dbg(&ucontext->iwdev->ibdev,
211 "VERBS: pgoff[0x%lx] does not have valid entry\n",
216 entry = to_irdma_mmap_entry(rdma_entry);
217 ibdev_dbg(&ucontext->iwdev->ibdev,
218 "VERBS: bar_offset [0x%llx] mmap_flag [%d]\n",
219 entry->bar_offset, entry->mmap_flag);
221 pfn = (entry->bar_offset +
222 pci_resource_start(ucontext->iwdev->rf->pcidev, 0)) >> PAGE_SHIFT;
224 switch (entry->mmap_flag) {
225 case IRDMA_MMAP_IO_NC:
226 ret = rdma_user_mmap_io(context, vma, pfn, PAGE_SIZE,
227 pgprot_noncached(vma->vm_page_prot),
230 case IRDMA_MMAP_IO_WC:
231 ret = rdma_user_mmap_io(context, vma, pfn, PAGE_SIZE,
232 pgprot_writecombine(vma->vm_page_prot),
240 ibdev_dbg(&ucontext->iwdev->ibdev,
241 "VERBS: bar_offset [0x%llx] mmap_flag[%d] err[%d]\n",
242 entry->bar_offset, entry->mmap_flag, ret);
243 rdma_user_mmap_entry_put(rdma_entry);
249 * irdma_alloc_push_page - allocate a push page for qp
252 static void irdma_alloc_push_page(struct irdma_qp *iwqp)
254 struct irdma_cqp_request *cqp_request;
255 struct cqp_cmds_info *cqp_info;
256 struct irdma_device *iwdev = iwqp->iwdev;
257 struct irdma_sc_qp *qp = &iwqp->sc_qp;
258 enum irdma_status_code status;
260 cqp_request = irdma_alloc_and_get_cqp_request(&iwdev->rf->cqp, true);
264 cqp_info = &cqp_request->info;
265 cqp_info->cqp_cmd = IRDMA_OP_MANAGE_PUSH_PAGE;
266 cqp_info->post_sq = 1;
267 cqp_info->in.u.manage_push_page.info.push_idx = 0;
268 cqp_info->in.u.manage_push_page.info.qs_handle =
269 qp->vsi->qos[qp->user_pri].qs_handle;
270 cqp_info->in.u.manage_push_page.info.free_page = 0;
271 cqp_info->in.u.manage_push_page.info.push_page_type = 0;
272 cqp_info->in.u.manage_push_page.cqp = &iwdev->rf->cqp.sc_cqp;
273 cqp_info->in.u.manage_push_page.scratch = (uintptr_t)cqp_request;
275 status = irdma_handle_cqp_op(iwdev->rf, cqp_request);
276 if (!status && cqp_request->compl_info.op_ret_val <
277 iwdev->rf->sc_dev.hw_attrs.max_hw_device_pages) {
278 qp->push_idx = cqp_request->compl_info.op_ret_val;
282 irdma_put_cqp_request(&iwdev->rf->cqp, cqp_request);
286 * irdma_alloc_ucontext - Allocate the user context data structure
287 * @uctx: uverbs context pointer
290 * This keeps track of all objects associated with a particular
293 static int irdma_alloc_ucontext(struct ib_ucontext *uctx,
294 struct ib_udata *udata)
296 struct ib_device *ibdev = uctx->device;
297 struct irdma_device *iwdev = to_iwdev(ibdev);
298 struct irdma_alloc_ucontext_req req;
299 struct irdma_alloc_ucontext_resp uresp = {};
300 struct irdma_ucontext *ucontext = to_ucontext(uctx);
301 struct irdma_uk_attrs *uk_attrs;
303 if (ib_copy_from_udata(&req, udata, min(sizeof(req), udata->inlen)))
306 if (req.userspace_ver < 4 || req.userspace_ver > IRDMA_ABI_VER)
309 ucontext->iwdev = iwdev;
310 ucontext->abi_ver = req.userspace_ver;
312 uk_attrs = &iwdev->rf->sc_dev.hw_attrs.uk_attrs;
313 /* GEN_1 legacy support with libi40iw */
314 if (udata->outlen < sizeof(uresp)) {
315 if (uk_attrs->hw_rev != IRDMA_GEN_1)
318 ucontext->legacy_mode = true;
319 uresp.max_qps = iwdev->rf->max_qp;
320 uresp.max_pds = iwdev->rf->sc_dev.hw_attrs.max_hw_pds;
321 uresp.wq_size = iwdev->rf->sc_dev.hw_attrs.max_qp_wr * 2;
322 uresp.kernel_ver = req.userspace_ver;
323 if (ib_copy_to_udata(udata, &uresp,
324 min(sizeof(uresp), udata->outlen)))
327 u64 bar_off = (uintptr_t)iwdev->rf->sc_dev.hw_regs[IRDMA_DB_ADDR_OFFSET];
329 ucontext->db_mmap_entry =
330 irdma_user_mmap_entry_insert(ucontext, bar_off,
333 if (!ucontext->db_mmap_entry)
336 uresp.kernel_ver = IRDMA_ABI_VER;
337 uresp.feature_flags = uk_attrs->feature_flags;
338 uresp.max_hw_wq_frags = uk_attrs->max_hw_wq_frags;
339 uresp.max_hw_read_sges = uk_attrs->max_hw_read_sges;
340 uresp.max_hw_inline = uk_attrs->max_hw_inline;
341 uresp.max_hw_rq_quanta = uk_attrs->max_hw_rq_quanta;
342 uresp.max_hw_wq_quanta = uk_attrs->max_hw_wq_quanta;
343 uresp.max_hw_sq_chunk = uk_attrs->max_hw_sq_chunk;
344 uresp.max_hw_cq_size = uk_attrs->max_hw_cq_size;
345 uresp.min_hw_cq_size = uk_attrs->min_hw_cq_size;
346 uresp.hw_rev = uk_attrs->hw_rev;
347 if (ib_copy_to_udata(udata, &uresp,
348 min(sizeof(uresp), udata->outlen))) {
349 rdma_user_mmap_entry_remove(ucontext->db_mmap_entry);
354 INIT_LIST_HEAD(&ucontext->cq_reg_mem_list);
355 spin_lock_init(&ucontext->cq_reg_mem_list_lock);
356 INIT_LIST_HEAD(&ucontext->qp_reg_mem_list);
357 spin_lock_init(&ucontext->qp_reg_mem_list_lock);
362 ibdev_err(&iwdev->ibdev,
363 "Invalid userspace driver version detected. Detected version %d, should be %d\n",
364 req.userspace_ver, IRDMA_ABI_VER);
369 * irdma_dealloc_ucontext - deallocate the user context data structure
370 * @context: user context created during alloc
372 static void irdma_dealloc_ucontext(struct ib_ucontext *context)
374 struct irdma_ucontext *ucontext = to_ucontext(context);
376 rdma_user_mmap_entry_remove(ucontext->db_mmap_entry);
380 * irdma_alloc_pd - allocate protection domain
384 static int irdma_alloc_pd(struct ib_pd *pd, struct ib_udata *udata)
386 struct irdma_pd *iwpd = to_iwpd(pd);
387 struct irdma_device *iwdev = to_iwdev(pd->device);
388 struct irdma_sc_dev *dev = &iwdev->rf->sc_dev;
389 struct irdma_pci_f *rf = iwdev->rf;
390 struct irdma_alloc_pd_resp uresp = {};
391 struct irdma_sc_pd *sc_pd;
395 err = irdma_alloc_rsrc(rf, rf->allocated_pds, rf->max_pd, &pd_id,
400 sc_pd = &iwpd->sc_pd;
402 struct irdma_ucontext *ucontext =
403 rdma_udata_to_drv_context(udata, struct irdma_ucontext,
405 irdma_sc_pd_init(dev, sc_pd, pd_id, ucontext->abi_ver);
407 if (ib_copy_to_udata(udata, &uresp,
408 min(sizeof(uresp), udata->outlen))) {
413 irdma_sc_pd_init(dev, sc_pd, pd_id, IRDMA_ABI_VER);
418 irdma_free_rsrc(rf, rf->allocated_pds, pd_id);
424 * irdma_dealloc_pd - deallocate pd
425 * @ibpd: ptr of pd to be deallocated
428 static int irdma_dealloc_pd(struct ib_pd *ibpd, struct ib_udata *udata)
430 struct irdma_pd *iwpd = to_iwpd(ibpd);
431 struct irdma_device *iwdev = to_iwdev(ibpd->device);
433 irdma_free_rsrc(iwdev->rf, iwdev->rf->allocated_pds, iwpd->sc_pd.pd_id);
439 * irdma_get_pbl - Retrieve pbl from a list given a virtual
441 * @va: user virtual address
442 * @pbl_list: pbl list to search in (QP's or CQ's)
444 static struct irdma_pbl *irdma_get_pbl(unsigned long va,
445 struct list_head *pbl_list)
447 struct irdma_pbl *iwpbl;
449 list_for_each_entry (iwpbl, pbl_list, list) {
450 if (iwpbl->user_base == va) {
451 list_del(&iwpbl->list);
452 iwpbl->on_list = false;
461 * irdma_clean_cqes - clean cq entries for qp
462 * @iwqp: qp ptr (user or kernel)
465 static void irdma_clean_cqes(struct irdma_qp *iwqp, struct irdma_cq *iwcq)
467 struct irdma_cq_uk *ukcq = &iwcq->sc_cq.cq_uk;
470 spin_lock_irqsave(&iwcq->lock, flags);
471 irdma_uk_clean_cq(&iwqp->sc_qp.qp_uk, ukcq);
472 spin_unlock_irqrestore(&iwcq->lock, flags);
475 static void irdma_remove_push_mmap_entries(struct irdma_qp *iwqp)
477 if (iwqp->push_db_mmap_entry) {
478 rdma_user_mmap_entry_remove(iwqp->push_db_mmap_entry);
479 iwqp->push_db_mmap_entry = NULL;
481 if (iwqp->push_wqe_mmap_entry) {
482 rdma_user_mmap_entry_remove(iwqp->push_wqe_mmap_entry);
483 iwqp->push_wqe_mmap_entry = NULL;
487 static int irdma_setup_push_mmap_entries(struct irdma_ucontext *ucontext,
488 struct irdma_qp *iwqp,
489 u64 *push_wqe_mmap_key,
490 u64 *push_db_mmap_key)
492 struct irdma_device *iwdev = ucontext->iwdev;
495 rsvd = IRDMA_PF_BAR_RSVD;
496 bar_off = (uintptr_t)iwdev->rf->sc_dev.hw_regs[IRDMA_DB_ADDR_OFFSET];
497 /* skip over db page */
498 bar_off += IRDMA_HW_PAGE_SIZE;
500 bar_off += rsvd + iwqp->sc_qp.push_idx * IRDMA_HW_PAGE_SIZE;
501 iwqp->push_wqe_mmap_entry = irdma_user_mmap_entry_insert(ucontext,
502 bar_off, IRDMA_MMAP_IO_WC,
504 if (!iwqp->push_wqe_mmap_entry)
507 /* push doorbell page */
508 bar_off += IRDMA_HW_PAGE_SIZE;
509 iwqp->push_db_mmap_entry = irdma_user_mmap_entry_insert(ucontext,
510 bar_off, IRDMA_MMAP_IO_NC,
512 if (!iwqp->push_db_mmap_entry) {
513 rdma_user_mmap_entry_remove(iwqp->push_wqe_mmap_entry);
521 * irdma_destroy_qp - destroy qp
522 * @ibqp: qp's ib pointer also to get to device's qp address
525 static int irdma_destroy_qp(struct ib_qp *ibqp, struct ib_udata *udata)
527 struct irdma_qp *iwqp = to_iwqp(ibqp);
528 struct irdma_device *iwdev = iwqp->iwdev;
530 iwqp->sc_qp.qp_uk.destroy_pending = true;
532 if (iwqp->iwarp_state == IRDMA_QP_STATE_RTS)
533 irdma_modify_qp_to_err(&iwqp->sc_qp);
535 irdma_qp_rem_ref(&iwqp->ibqp);
536 wait_for_completion(&iwqp->free_qp);
537 irdma_free_lsmm_rsrc(iwqp);
539 irdma_cqp_qp_destroy_cmd(&iwdev->rf->sc_dev, &iwqp->sc_qp);
541 if (!iwqp->user_mode) {
543 irdma_clean_cqes(iwqp, iwqp->iwscq);
544 if (iwqp->iwrcq != iwqp->iwscq)
545 irdma_clean_cqes(iwqp, iwqp->iwrcq);
548 irdma_remove_push_mmap_entries(iwqp);
549 irdma_free_qp_rsrc(iwqp);
555 * irdma_setup_virt_qp - setup for allocation of virtual qp
556 * @iwdev: irdma device
558 * @init_info: initialize info to return
560 static int irdma_setup_virt_qp(struct irdma_device *iwdev,
561 struct irdma_qp *iwqp,
562 struct irdma_qp_init_info *init_info)
564 struct irdma_pbl *iwpbl = iwqp->iwpbl;
565 struct irdma_qp_mr *qpmr = &iwpbl->qp_mr;
567 iwqp->page = qpmr->sq_page;
568 init_info->shadow_area_pa = qpmr->shadow;
569 if (iwpbl->pbl_allocated) {
570 init_info->virtual_map = true;
571 init_info->sq_pa = qpmr->sq_pbl.idx;
572 init_info->rq_pa = qpmr->rq_pbl.idx;
574 init_info->sq_pa = qpmr->sq_pbl.addr;
575 init_info->rq_pa = qpmr->rq_pbl.addr;
582 * irdma_setup_kmode_qp - setup initialization for kernel mode qp
583 * @iwdev: iwarp device
584 * @iwqp: qp ptr (user or kernel)
585 * @info: initialize info to return
586 * @init_attr: Initial QP create attributes
588 static int irdma_setup_kmode_qp(struct irdma_device *iwdev,
589 struct irdma_qp *iwqp,
590 struct irdma_qp_init_info *info,
591 struct ib_qp_init_attr *init_attr)
593 struct irdma_dma_mem *mem = &iwqp->kqp.dma_mem;
594 u32 sqdepth, rqdepth;
597 enum irdma_status_code status;
598 struct irdma_qp_uk_init_info *ukinfo = &info->qp_uk_init_info;
599 struct irdma_uk_attrs *uk_attrs = &iwdev->rf->sc_dev.hw_attrs.uk_attrs;
601 irdma_get_wqe_shift(uk_attrs,
602 uk_attrs->hw_rev >= IRDMA_GEN_2 ? ukinfo->max_sq_frag_cnt + 1 :
603 ukinfo->max_sq_frag_cnt,
604 ukinfo->max_inline_data, &sqshift);
605 status = irdma_get_sqdepth(uk_attrs, ukinfo->sq_size, sqshift,
610 if (uk_attrs->hw_rev == IRDMA_GEN_1)
611 rqshift = IRDMA_MAX_RQ_WQE_SHIFT_GEN1;
613 irdma_get_wqe_shift(uk_attrs, ukinfo->max_rq_frag_cnt, 0,
616 status = irdma_get_rqdepth(uk_attrs, ukinfo->rq_size, rqshift,
621 iwqp->kqp.sq_wrid_mem =
622 kcalloc(sqdepth, sizeof(*iwqp->kqp.sq_wrid_mem), GFP_KERNEL);
623 if (!iwqp->kqp.sq_wrid_mem)
626 iwqp->kqp.rq_wrid_mem =
627 kcalloc(rqdepth, sizeof(*iwqp->kqp.rq_wrid_mem), GFP_KERNEL);
628 if (!iwqp->kqp.rq_wrid_mem) {
629 kfree(iwqp->kqp.sq_wrid_mem);
630 iwqp->kqp.sq_wrid_mem = NULL;
634 ukinfo->sq_wrtrk_array = iwqp->kqp.sq_wrid_mem;
635 ukinfo->rq_wrid_array = iwqp->kqp.rq_wrid_mem;
637 size = (sqdepth + rqdepth) * IRDMA_QP_WQE_MIN_SIZE;
638 size += (IRDMA_SHADOW_AREA_SIZE << 3);
640 mem->size = ALIGN(size, 256);
641 mem->va = dma_alloc_coherent(iwdev->rf->hw.device, mem->size,
642 &mem->pa, GFP_KERNEL);
644 kfree(iwqp->kqp.sq_wrid_mem);
645 iwqp->kqp.sq_wrid_mem = NULL;
646 kfree(iwqp->kqp.rq_wrid_mem);
647 iwqp->kqp.rq_wrid_mem = NULL;
651 ukinfo->sq = mem->va;
652 info->sq_pa = mem->pa;
653 ukinfo->rq = &ukinfo->sq[sqdepth];
654 info->rq_pa = info->sq_pa + (sqdepth * IRDMA_QP_WQE_MIN_SIZE);
655 ukinfo->shadow_area = ukinfo->rq[rqdepth].elem;
656 info->shadow_area_pa = info->rq_pa + (rqdepth * IRDMA_QP_WQE_MIN_SIZE);
657 ukinfo->sq_size = sqdepth >> sqshift;
658 ukinfo->rq_size = rqdepth >> rqshift;
659 ukinfo->qp_id = iwqp->ibqp.qp_num;
661 init_attr->cap.max_send_wr = (sqdepth - IRDMA_SQ_RSVD) >> sqshift;
662 init_attr->cap.max_recv_wr = (rqdepth - IRDMA_RQ_RSVD) >> rqshift;
667 static int irdma_cqp_create_qp_cmd(struct irdma_qp *iwqp)
669 struct irdma_pci_f *rf = iwqp->iwdev->rf;
670 struct irdma_cqp_request *cqp_request;
671 struct cqp_cmds_info *cqp_info;
672 struct irdma_create_qp_info *qp_info;
673 enum irdma_status_code status;
675 cqp_request = irdma_alloc_and_get_cqp_request(&rf->cqp, true);
679 cqp_info = &cqp_request->info;
680 qp_info = &cqp_request->info.in.u.qp_create.info;
681 memset(qp_info, 0, sizeof(*qp_info));
682 qp_info->mac_valid = true;
683 qp_info->cq_num_valid = true;
684 qp_info->next_iwarp_state = IRDMA_QP_STATE_IDLE;
686 cqp_info->cqp_cmd = IRDMA_OP_QP_CREATE;
687 cqp_info->post_sq = 1;
688 cqp_info->in.u.qp_create.qp = &iwqp->sc_qp;
689 cqp_info->in.u.qp_create.scratch = (uintptr_t)cqp_request;
690 status = irdma_handle_cqp_op(rf, cqp_request);
691 irdma_put_cqp_request(&rf->cqp, cqp_request);
693 return status ? -ENOMEM : 0;
696 static void irdma_roce_fill_and_set_qpctx_info(struct irdma_qp *iwqp,
697 struct irdma_qp_host_ctx_info *ctx_info)
699 struct irdma_device *iwdev = iwqp->iwdev;
700 struct irdma_sc_dev *dev = &iwdev->rf->sc_dev;
701 struct irdma_roce_offload_info *roce_info;
702 struct irdma_udp_offload_info *udp_info;
704 udp_info = &iwqp->udp_info;
705 udp_info->snd_mss = ib_mtu_enum_to_int(ib_mtu_int_to_enum(iwdev->vsi.mtu));
706 udp_info->cwnd = iwdev->roce_cwnd;
707 udp_info->rexmit_thresh = 2;
708 udp_info->rnr_nak_thresh = 2;
709 udp_info->src_port = 0xc000;
710 udp_info->dst_port = ROCE_V2_UDP_DPORT;
711 roce_info = &iwqp->roce_info;
712 ether_addr_copy(roce_info->mac_addr, iwdev->netdev->dev_addr);
714 roce_info->rd_en = true;
715 roce_info->wr_rdresp_en = true;
716 roce_info->bind_en = true;
717 roce_info->dcqcn_en = false;
718 roce_info->rtomin = 5;
720 roce_info->ack_credits = iwdev->roce_ackcreds;
721 roce_info->ird_size = dev->hw_attrs.max_hw_ird;
722 roce_info->ord_size = dev->hw_attrs.max_hw_ord;
724 if (!iwqp->user_mode) {
725 roce_info->priv_mode_en = true;
726 roce_info->fast_reg_en = true;
727 roce_info->udprivcq_en = true;
729 roce_info->roce_tver = 0;
731 ctx_info->roce_info = &iwqp->roce_info;
732 ctx_info->udp_info = &iwqp->udp_info;
733 irdma_sc_qp_setctx_roce(&iwqp->sc_qp, iwqp->host_ctx.va, ctx_info);
736 static void irdma_iw_fill_and_set_qpctx_info(struct irdma_qp *iwqp,
737 struct irdma_qp_host_ctx_info *ctx_info)
739 struct irdma_device *iwdev = iwqp->iwdev;
740 struct irdma_sc_dev *dev = &iwdev->rf->sc_dev;
741 struct irdma_iwarp_offload_info *iwarp_info;
743 iwarp_info = &iwqp->iwarp_info;
744 ether_addr_copy(iwarp_info->mac_addr, iwdev->netdev->dev_addr);
745 iwarp_info->rd_en = true;
746 iwarp_info->wr_rdresp_en = true;
747 iwarp_info->bind_en = true;
748 iwarp_info->ecn_en = true;
749 iwarp_info->rtomin = 5;
751 if (dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2)
752 iwarp_info->ib_rd_en = true;
753 if (!iwqp->user_mode) {
754 iwarp_info->priv_mode_en = true;
755 iwarp_info->fast_reg_en = true;
757 iwarp_info->ddp_ver = 1;
758 iwarp_info->rdmap_ver = 1;
760 ctx_info->iwarp_info = &iwqp->iwarp_info;
761 ctx_info->iwarp_info_valid = true;
762 irdma_sc_qp_setctx(&iwqp->sc_qp, iwqp->host_ctx.va, ctx_info);
763 ctx_info->iwarp_info_valid = false;
766 static int irdma_validate_qp_attrs(struct ib_qp_init_attr *init_attr,
767 struct irdma_device *iwdev)
769 struct irdma_sc_dev *dev = &iwdev->rf->sc_dev;
770 struct irdma_uk_attrs *uk_attrs = &dev->hw_attrs.uk_attrs;
772 if (init_attr->create_flags)
775 if (init_attr->cap.max_inline_data > uk_attrs->max_hw_inline ||
776 init_attr->cap.max_send_sge > uk_attrs->max_hw_wq_frags ||
777 init_attr->cap.max_recv_sge > uk_attrs->max_hw_wq_frags)
780 if (rdma_protocol_roce(&iwdev->ibdev, 1)) {
781 if (init_attr->qp_type != IB_QPT_RC &&
782 init_attr->qp_type != IB_QPT_UD &&
783 init_attr->qp_type != IB_QPT_GSI)
786 if (init_attr->qp_type != IB_QPT_RC)
794 * irdma_create_qp - create qp
796 * @init_attr: attributes for qp
797 * @udata: user data for create qp
799 static struct ib_qp *irdma_create_qp(struct ib_pd *ibpd,
800 struct ib_qp_init_attr *init_attr,
801 struct ib_udata *udata)
803 struct irdma_pd *iwpd = to_iwpd(ibpd);
804 struct irdma_device *iwdev = to_iwdev(ibpd->device);
805 struct irdma_pci_f *rf = iwdev->rf;
806 struct irdma_qp *iwqp;
807 struct irdma_create_qp_req req;
808 struct irdma_create_qp_resp uresp = {};
810 enum irdma_status_code ret;
814 struct irdma_sc_qp *qp;
815 struct irdma_sc_dev *dev = &rf->sc_dev;
816 struct irdma_uk_attrs *uk_attrs = &dev->hw_attrs.uk_attrs;
817 struct irdma_qp_init_info init_info = {};
818 struct irdma_qp_host_ctx_info *ctx_info;
821 err_code = irdma_validate_qp_attrs(init_attr, iwdev);
823 return ERR_PTR(err_code);
825 sq_size = init_attr->cap.max_send_wr;
826 rq_size = init_attr->cap.max_recv_wr;
828 init_info.vsi = &iwdev->vsi;
829 init_info.qp_uk_init_info.uk_attrs = uk_attrs;
830 init_info.qp_uk_init_info.sq_size = sq_size;
831 init_info.qp_uk_init_info.rq_size = rq_size;
832 init_info.qp_uk_init_info.max_sq_frag_cnt = init_attr->cap.max_send_sge;
833 init_info.qp_uk_init_info.max_rq_frag_cnt = init_attr->cap.max_recv_sge;
834 init_info.qp_uk_init_info.max_inline_data = init_attr->cap.max_inline_data;
836 iwqp = kzalloc(sizeof(*iwqp), GFP_KERNEL);
838 return ERR_PTR(-ENOMEM);
841 qp->qp_uk.back_qp = iwqp;
842 qp->qp_uk.lock = &iwqp->lock;
843 qp->push_idx = IRDMA_INVALID_PUSH_PAGE_INDEX;
846 iwqp->q2_ctx_mem.size = ALIGN(IRDMA_Q2_BUF_SIZE + IRDMA_QP_CTX_SIZE,
848 iwqp->q2_ctx_mem.va = dma_alloc_coherent(dev->hw->device,
849 iwqp->q2_ctx_mem.size,
850 &iwqp->q2_ctx_mem.pa,
852 if (!iwqp->q2_ctx_mem.va) {
857 init_info.q2 = iwqp->q2_ctx_mem.va;
858 init_info.q2_pa = iwqp->q2_ctx_mem.pa;
859 init_info.host_ctx = (__le64 *)(init_info.q2 + IRDMA_Q2_BUF_SIZE);
860 init_info.host_ctx_pa = init_info.q2_pa + IRDMA_Q2_BUF_SIZE;
862 if (init_attr->qp_type == IB_QPT_GSI)
865 err_code = irdma_alloc_rsrc(rf, rf->allocated_qps, rf->max_qp,
866 &qp_num, &rf->next_qp);
871 iwqp->ibqp.qp_num = qp_num;
873 iwqp->iwscq = to_iwcq(init_attr->send_cq);
874 iwqp->iwrcq = to_iwcq(init_attr->recv_cq);
875 iwqp->host_ctx.va = init_info.host_ctx;
876 iwqp->host_ctx.pa = init_info.host_ctx_pa;
877 iwqp->host_ctx.size = IRDMA_QP_CTX_SIZE;
879 init_info.pd = &iwpd->sc_pd;
880 init_info.qp_uk_init_info.qp_id = iwqp->ibqp.qp_num;
881 if (!rdma_protocol_roce(&iwdev->ibdev, 1))
882 init_info.qp_uk_init_info.first_sq_wq = 1;
883 iwqp->ctx_info.qp_compl_ctx = (uintptr_t)qp;
884 init_waitqueue_head(&iwqp->waitq);
885 init_waitqueue_head(&iwqp->mod_qp_waitq);
888 err_code = ib_copy_from_udata(&req, udata,
889 min(sizeof(req), udata->inlen));
891 ibdev_dbg(&iwdev->ibdev,
892 "VERBS: ib_copy_from_data fail\n");
896 iwqp->ctx_info.qp_compl_ctx = req.user_compl_ctx;
898 if (req.user_wqe_bufs) {
899 struct irdma_ucontext *ucontext =
900 rdma_udata_to_drv_context(udata,
901 struct irdma_ucontext,
904 init_info.qp_uk_init_info.legacy_mode = ucontext->legacy_mode;
905 spin_lock_irqsave(&ucontext->qp_reg_mem_list_lock, flags);
906 iwqp->iwpbl = irdma_get_pbl((unsigned long)req.user_wqe_bufs,
907 &ucontext->qp_reg_mem_list);
908 spin_unlock_irqrestore(&ucontext->qp_reg_mem_list_lock, flags);
912 ibdev_dbg(&iwdev->ibdev, "VERBS: no pbl info\n");
916 init_info.qp_uk_init_info.abi_ver = iwpd->sc_pd.abi_ver;
917 err_code = irdma_setup_virt_qp(iwdev, iwqp, &init_info);
919 init_info.qp_uk_init_info.abi_ver = IRDMA_ABI_VER;
920 err_code = irdma_setup_kmode_qp(iwdev, iwqp, &init_info, init_attr);
924 ibdev_dbg(&iwdev->ibdev, "VERBS: setup qp failed\n");
928 if (rdma_protocol_roce(&iwdev->ibdev, 1)) {
929 if (init_attr->qp_type == IB_QPT_RC) {
930 init_info.qp_uk_init_info.type = IRDMA_QP_TYPE_ROCE_RC;
931 init_info.qp_uk_init_info.qp_caps = IRDMA_SEND_WITH_IMM |
932 IRDMA_WRITE_WITH_IMM |
935 init_info.qp_uk_init_info.type = IRDMA_QP_TYPE_ROCE_UD;
936 init_info.qp_uk_init_info.qp_caps = IRDMA_SEND_WITH_IMM |
940 init_info.qp_uk_init_info.type = IRDMA_QP_TYPE_IWARP;
941 init_info.qp_uk_init_info.qp_caps = IRDMA_WRITE_WITH_IMM;
944 if (dev->hw_attrs.uk_attrs.hw_rev > IRDMA_GEN_1)
945 init_info.qp_uk_init_info.qp_caps |= IRDMA_PUSH_MODE;
947 ret = irdma_sc_qp_init(qp, &init_info);
950 ibdev_dbg(&iwdev->ibdev, "VERBS: qp_init fail\n");
954 ctx_info = &iwqp->ctx_info;
955 ctx_info->send_cq_num = iwqp->iwscq->sc_cq.cq_uk.cq_id;
956 ctx_info->rcv_cq_num = iwqp->iwrcq->sc_cq.cq_uk.cq_id;
958 if (rdma_protocol_roce(&iwdev->ibdev, 1))
959 irdma_roce_fill_and_set_qpctx_info(iwqp, ctx_info);
961 irdma_iw_fill_and_set_qpctx_info(iwqp, ctx_info);
963 err_code = irdma_cqp_create_qp_cmd(iwqp);
967 refcount_set(&iwqp->refcnt, 1);
968 spin_lock_init(&iwqp->lock);
969 spin_lock_init(&iwqp->sc_qp.pfpdu.lock);
970 iwqp->sig_all = (init_attr->sq_sig_type == IB_SIGNAL_ALL_WR) ? 1 : 0;
971 rf->qp_table[qp_num] = iwqp;
972 iwqp->max_send_wr = sq_size;
973 iwqp->max_recv_wr = rq_size;
975 if (rdma_protocol_roce(&iwdev->ibdev, 1)) {
976 if (dev->ws_add(&iwdev->vsi, 0)) {
977 irdma_cqp_qp_destroy_cmd(&rf->sc_dev, &iwqp->sc_qp);
982 irdma_qp_add_qos(&iwqp->sc_qp);
986 /* GEN_1 legacy support with libi40iw does not have expanded uresp struct */
987 if (udata->outlen < sizeof(uresp)) {
989 uresp.push_idx = IRDMA_INVALID_PUSH_PAGE_INDEX_GEN_1;
991 if (rdma_protocol_iwarp(&iwdev->ibdev, 1))
994 uresp.actual_sq_size = sq_size;
995 uresp.actual_rq_size = rq_size;
996 uresp.qp_id = qp_num;
997 uresp.qp_caps = qp->qp_uk.qp_caps;
999 err_code = ib_copy_to_udata(udata, &uresp,
1000 min(sizeof(uresp), udata->outlen));
1002 ibdev_dbg(&iwdev->ibdev, "VERBS: copy_to_udata failed\n");
1003 irdma_destroy_qp(&iwqp->ibqp, udata);
1004 return ERR_PTR(err_code);
1008 init_completion(&iwqp->free_qp);
1012 irdma_free_qp_rsrc(iwqp);
1014 return ERR_PTR(err_code);
1017 static int irdma_get_ib_acc_flags(struct irdma_qp *iwqp)
1021 if (rdma_protocol_roce(iwqp->ibqp.device, 1)) {
1022 if (iwqp->roce_info.wr_rdresp_en) {
1023 acc_flags |= IB_ACCESS_LOCAL_WRITE;
1024 acc_flags |= IB_ACCESS_REMOTE_WRITE;
1026 if (iwqp->roce_info.rd_en)
1027 acc_flags |= IB_ACCESS_REMOTE_READ;
1028 if (iwqp->roce_info.bind_en)
1029 acc_flags |= IB_ACCESS_MW_BIND;
1031 if (iwqp->iwarp_info.wr_rdresp_en) {
1032 acc_flags |= IB_ACCESS_LOCAL_WRITE;
1033 acc_flags |= IB_ACCESS_REMOTE_WRITE;
1035 if (iwqp->iwarp_info.rd_en)
1036 acc_flags |= IB_ACCESS_REMOTE_READ;
1037 if (iwqp->iwarp_info.bind_en)
1038 acc_flags |= IB_ACCESS_MW_BIND;
1044 * irdma_query_qp - query qp attributes
1046 * @attr: attributes pointer
1047 * @attr_mask: Not used
1048 * @init_attr: qp attributes to return
1050 static int irdma_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
1051 int attr_mask, struct ib_qp_init_attr *init_attr)
1053 struct irdma_qp *iwqp = to_iwqp(ibqp);
1054 struct irdma_sc_qp *qp = &iwqp->sc_qp;
1056 memset(attr, 0, sizeof(*attr));
1057 memset(init_attr, 0, sizeof(*init_attr));
1059 attr->qp_state = iwqp->ibqp_state;
1060 attr->cur_qp_state = iwqp->ibqp_state;
1061 attr->cap.max_send_wr = iwqp->max_send_wr;
1062 attr->cap.max_recv_wr = iwqp->max_recv_wr;
1063 attr->cap.max_inline_data = qp->qp_uk.max_inline_data;
1064 attr->cap.max_send_sge = qp->qp_uk.max_sq_frag_cnt;
1065 attr->cap.max_recv_sge = qp->qp_uk.max_rq_frag_cnt;
1066 attr->qp_access_flags = irdma_get_ib_acc_flags(iwqp);
1068 if (rdma_protocol_roce(ibqp->device, 1)) {
1069 attr->path_mtu = ib_mtu_int_to_enum(iwqp->udp_info.snd_mss);
1070 attr->qkey = iwqp->roce_info.qkey;
1071 attr->rq_psn = iwqp->udp_info.epsn;
1072 attr->sq_psn = iwqp->udp_info.psn_nxt;
1073 attr->dest_qp_num = iwqp->roce_info.dest_qp;
1074 attr->pkey_index = iwqp->roce_info.p_key;
1075 attr->retry_cnt = iwqp->udp_info.rexmit_thresh;
1076 attr->rnr_retry = iwqp->udp_info.rnr_nak_thresh;
1077 attr->max_rd_atomic = iwqp->roce_info.ord_size;
1078 attr->max_dest_rd_atomic = iwqp->roce_info.ird_size;
1081 init_attr->event_handler = iwqp->ibqp.event_handler;
1082 init_attr->qp_context = iwqp->ibqp.qp_context;
1083 init_attr->send_cq = iwqp->ibqp.send_cq;
1084 init_attr->recv_cq = iwqp->ibqp.recv_cq;
1085 init_attr->cap = attr->cap;
1091 * irdma_query_pkey - Query partition key
1092 * @ibdev: device pointer from stack
1093 * @port: port number
1094 * @index: index of pkey
1095 * @pkey: pointer to store the pkey
1097 static int irdma_query_pkey(struct ib_device *ibdev, u32 port, u16 index,
1100 if (index >= IRDMA_PKEY_TBL_SZ)
1103 *pkey = IRDMA_DEFAULT_PKEY;
1108 * irdma_modify_qp_roce - modify qp request
1109 * @ibqp: qp's pointer for modify
1110 * @attr: access attributes
1111 * @attr_mask: state mask
1114 int irdma_modify_qp_roce(struct ib_qp *ibqp, struct ib_qp_attr *attr,
1115 int attr_mask, struct ib_udata *udata)
1117 struct irdma_pd *iwpd = to_iwpd(ibqp->pd);
1118 struct irdma_qp *iwqp = to_iwqp(ibqp);
1119 struct irdma_device *iwdev = iwqp->iwdev;
1120 struct irdma_sc_dev *dev = &iwdev->rf->sc_dev;
1121 struct irdma_qp_host_ctx_info *ctx_info;
1122 struct irdma_roce_offload_info *roce_info;
1123 struct irdma_udp_offload_info *udp_info;
1124 struct irdma_modify_qp_info info = {};
1125 struct irdma_modify_qp_resp uresp = {};
1126 struct irdma_modify_qp_req ureq = {};
1127 unsigned long flags;
1128 u8 issue_modify_qp = 0;
1131 ctx_info = &iwqp->ctx_info;
1132 roce_info = &iwqp->roce_info;
1133 udp_info = &iwqp->udp_info;
1135 if (attr_mask & ~IB_QP_ATTR_STANDARD_BITS)
1138 if (attr_mask & IB_QP_DEST_QPN)
1139 roce_info->dest_qp = attr->dest_qp_num;
1141 if (attr_mask & IB_QP_PKEY_INDEX) {
1142 ret = irdma_query_pkey(ibqp->device, 0, attr->pkey_index,
1148 if (attr_mask & IB_QP_QKEY)
1149 roce_info->qkey = attr->qkey;
1151 if (attr_mask & IB_QP_PATH_MTU)
1152 udp_info->snd_mss = ib_mtu_enum_to_int(attr->path_mtu);
1154 if (attr_mask & IB_QP_SQ_PSN) {
1155 udp_info->psn_nxt = attr->sq_psn;
1156 udp_info->lsn = 0xffff;
1157 udp_info->psn_una = attr->sq_psn;
1158 udp_info->psn_max = attr->sq_psn;
1161 if (attr_mask & IB_QP_RQ_PSN)
1162 udp_info->epsn = attr->rq_psn;
1164 if (attr_mask & IB_QP_RNR_RETRY)
1165 udp_info->rnr_nak_thresh = attr->rnr_retry;
1167 if (attr_mask & IB_QP_RETRY_CNT)
1168 udp_info->rexmit_thresh = attr->retry_cnt;
1170 ctx_info->roce_info->pd_id = iwpd->sc_pd.pd_id;
1172 if (attr_mask & IB_QP_AV) {
1173 struct irdma_av *av = &iwqp->roce_ah.av;
1174 const struct ib_gid_attr *sgid_attr;
1175 u16 vlan_id = VLAN_N_VID;
1178 memset(&iwqp->roce_ah, 0, sizeof(iwqp->roce_ah));
1179 if (attr->ah_attr.ah_flags & IB_AH_GRH) {
1180 udp_info->ttl = attr->ah_attr.grh.hop_limit;
1181 udp_info->flow_label = attr->ah_attr.grh.flow_label;
1182 udp_info->tos = attr->ah_attr.grh.traffic_class;
1183 irdma_qp_rem_qos(&iwqp->sc_qp);
1184 dev->ws_remove(iwqp->sc_qp.vsi, ctx_info->user_pri);
1185 ctx_info->user_pri = rt_tos2priority(udp_info->tos);
1186 iwqp->sc_qp.user_pri = ctx_info->user_pri;
1187 if (dev->ws_add(iwqp->sc_qp.vsi, ctx_info->user_pri))
1189 irdma_qp_add_qos(&iwqp->sc_qp);
1191 sgid_attr = attr->ah_attr.grh.sgid_attr;
1192 ret = rdma_read_gid_l2_fields(sgid_attr, &vlan_id,
1193 ctx_info->roce_info->mac_addr);
1197 if (vlan_id >= VLAN_N_VID && iwdev->dcb)
1199 if (vlan_id < VLAN_N_VID) {
1200 udp_info->insert_vlan_tag = true;
1201 udp_info->vlan_tag = vlan_id |
1202 ctx_info->user_pri << VLAN_PRIO_SHIFT;
1204 udp_info->insert_vlan_tag = false;
1207 av->attrs = attr->ah_attr;
1208 rdma_gid2ip((struct sockaddr *)&av->sgid_addr, &sgid_attr->gid);
1209 rdma_gid2ip((struct sockaddr *)&av->dgid_addr, &attr->ah_attr.grh.dgid);
1210 roce_info->local_qp = ibqp->qp_num;
1211 if (av->sgid_addr.saddr.sa_family == AF_INET6) {
1213 av->dgid_addr.saddr_in6.sin6_addr.in6_u.u6_addr32;
1215 av->sgid_addr.saddr_in6.sin6_addr.in6_u.u6_addr32;
1217 irdma_copy_ip_ntohl(&udp_info->dest_ip_addr[0], daddr);
1218 irdma_copy_ip_ntohl(&udp_info->local_ipaddr[0], saddr);
1220 udp_info->ipv4 = false;
1221 irdma_copy_ip_ntohl(local_ip, daddr);
1223 udp_info->arp_idx = irdma_arp_table(iwdev->rf,
1228 __be32 saddr = av->sgid_addr.saddr_in.sin_addr.s_addr;
1229 __be32 daddr = av->dgid_addr.saddr_in.sin_addr.s_addr;
1231 local_ip[0] = ntohl(daddr);
1233 udp_info->ipv4 = true;
1234 udp_info->dest_ip_addr[0] = 0;
1235 udp_info->dest_ip_addr[1] = 0;
1236 udp_info->dest_ip_addr[2] = 0;
1237 udp_info->dest_ip_addr[3] = local_ip[0];
1239 udp_info->local_ipaddr[0] = 0;
1240 udp_info->local_ipaddr[1] = 0;
1241 udp_info->local_ipaddr[2] = 0;
1242 udp_info->local_ipaddr[3] = ntohl(saddr);
1245 irdma_add_arp(iwdev->rf, local_ip, udp_info->ipv4,
1246 attr->ah_attr.roce.dmac);
1249 if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC) {
1250 if (attr->max_rd_atomic > dev->hw_attrs.max_hw_ord) {
1251 ibdev_err(&iwdev->ibdev,
1252 "rd_atomic = %d, above max_hw_ord=%d\n",
1253 attr->max_rd_atomic,
1254 dev->hw_attrs.max_hw_ord);
1257 if (attr->max_rd_atomic)
1258 roce_info->ord_size = attr->max_rd_atomic;
1259 info.ord_valid = true;
1262 if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) {
1263 if (attr->max_dest_rd_atomic > dev->hw_attrs.max_hw_ird) {
1264 ibdev_err(&iwdev->ibdev,
1265 "rd_atomic = %d, above max_hw_ird=%d\n",
1266 attr->max_rd_atomic,
1267 dev->hw_attrs.max_hw_ird);
1270 if (attr->max_dest_rd_atomic)
1271 roce_info->ird_size = attr->max_dest_rd_atomic;
1274 if (attr_mask & IB_QP_ACCESS_FLAGS) {
1275 if (attr->qp_access_flags & IB_ACCESS_LOCAL_WRITE)
1276 roce_info->wr_rdresp_en = true;
1277 if (attr->qp_access_flags & IB_ACCESS_REMOTE_WRITE)
1278 roce_info->wr_rdresp_en = true;
1279 if (attr->qp_access_flags & IB_ACCESS_REMOTE_READ)
1280 roce_info->rd_en = true;
1283 wait_event(iwqp->mod_qp_waitq, !atomic_read(&iwqp->hw_mod_qp_pend));
1285 ibdev_dbg(&iwdev->ibdev,
1286 "VERBS: caller: %pS qp_id=%d to_ibqpstate=%d ibqpstate=%d irdma_qpstate=%d attr_mask=0x%x\n",
1287 __builtin_return_address(0), ibqp->qp_num, attr->qp_state,
1288 iwqp->ibqp_state, iwqp->iwarp_state, attr_mask);
1290 spin_lock_irqsave(&iwqp->lock, flags);
1291 if (attr_mask & IB_QP_STATE) {
1292 if (!ib_modify_qp_is_ok(iwqp->ibqp_state, attr->qp_state,
1293 iwqp->ibqp.qp_type, attr_mask)) {
1294 ibdev_warn(&iwdev->ibdev, "modify_qp invalid for qp_id=%d, old_state=0x%x, new_state=0x%x\n",
1295 iwqp->ibqp.qp_num, iwqp->ibqp_state,
1300 info.curr_iwarp_state = iwqp->iwarp_state;
1302 switch (attr->qp_state) {
1304 if (iwqp->iwarp_state > IRDMA_QP_STATE_IDLE) {
1309 if (iwqp->iwarp_state == IRDMA_QP_STATE_INVALID) {
1310 info.next_iwarp_state = IRDMA_QP_STATE_IDLE;
1311 issue_modify_qp = 1;
1315 if (iwqp->iwarp_state > IRDMA_QP_STATE_IDLE) {
1319 info.arp_cache_idx_valid = true;
1320 info.cq_num_valid = true;
1321 info.next_iwarp_state = IRDMA_QP_STATE_RTR;
1322 issue_modify_qp = 1;
1325 if (iwqp->ibqp_state < IB_QPS_RTR ||
1326 iwqp->ibqp_state == IB_QPS_ERR) {
1331 info.arp_cache_idx_valid = true;
1332 info.cq_num_valid = true;
1333 info.ord_valid = true;
1334 info.next_iwarp_state = IRDMA_QP_STATE_RTS;
1335 issue_modify_qp = 1;
1336 if (iwdev->push_mode && udata &&
1337 iwqp->sc_qp.push_idx == IRDMA_INVALID_PUSH_PAGE_INDEX &&
1338 dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2) {
1339 spin_unlock_irqrestore(&iwqp->lock, flags);
1340 irdma_alloc_push_page(iwqp);
1341 spin_lock_irqsave(&iwqp->lock, flags);
1345 if (iwqp->iwarp_state == IRDMA_QP_STATE_SQD)
1348 if (iwqp->iwarp_state != IRDMA_QP_STATE_RTS) {
1353 info.next_iwarp_state = IRDMA_QP_STATE_SQD;
1354 issue_modify_qp = 1;
1359 if (iwqp->iwarp_state == IRDMA_QP_STATE_RTS) {
1360 spin_unlock_irqrestore(&iwqp->lock, flags);
1361 info.next_iwarp_state = IRDMA_QP_STATE_SQD;
1362 irdma_hw_modify_qp(iwdev, iwqp, &info, true);
1363 spin_lock_irqsave(&iwqp->lock, flags);
1366 if (iwqp->iwarp_state == IRDMA_QP_STATE_ERROR) {
1367 spin_unlock_irqrestore(&iwqp->lock, flags);
1369 if (ib_copy_from_udata(&ureq, udata,
1370 min(sizeof(ureq), udata->inlen)))
1373 irdma_flush_wqes(iwqp,
1374 (ureq.sq_flush ? IRDMA_FLUSH_SQ : 0) |
1375 (ureq.rq_flush ? IRDMA_FLUSH_RQ : 0) |
1381 info.next_iwarp_state = IRDMA_QP_STATE_ERROR;
1382 issue_modify_qp = 1;
1389 iwqp->ibqp_state = attr->qp_state;
1392 ctx_info->send_cq_num = iwqp->iwscq->sc_cq.cq_uk.cq_id;
1393 ctx_info->rcv_cq_num = iwqp->iwrcq->sc_cq.cq_uk.cq_id;
1394 irdma_sc_qp_setctx_roce(&iwqp->sc_qp, iwqp->host_ctx.va, ctx_info);
1395 spin_unlock_irqrestore(&iwqp->lock, flags);
1397 if (attr_mask & IB_QP_STATE) {
1398 if (issue_modify_qp) {
1399 ctx_info->rem_endpoint_idx = udp_info->arp_idx;
1400 if (irdma_hw_modify_qp(iwdev, iwqp, &info, true))
1402 spin_lock_irqsave(&iwqp->lock, flags);
1403 if (iwqp->iwarp_state == info.curr_iwarp_state) {
1404 iwqp->iwarp_state = info.next_iwarp_state;
1405 iwqp->ibqp_state = attr->qp_state;
1407 if (iwqp->ibqp_state > IB_QPS_RTS &&
1408 !iwqp->flush_issued) {
1409 iwqp->flush_issued = 1;
1410 spin_unlock_irqrestore(&iwqp->lock, flags);
1411 irdma_flush_wqes(iwqp, IRDMA_FLUSH_SQ |
1415 spin_unlock_irqrestore(&iwqp->lock, flags);
1418 iwqp->ibqp_state = attr->qp_state;
1420 if (udata && dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2) {
1421 struct irdma_ucontext *ucontext;
1423 ucontext = rdma_udata_to_drv_context(udata,
1424 struct irdma_ucontext, ibucontext);
1425 if (iwqp->sc_qp.push_idx != IRDMA_INVALID_PUSH_PAGE_INDEX &&
1426 !iwqp->push_wqe_mmap_entry &&
1427 !irdma_setup_push_mmap_entries(ucontext, iwqp,
1428 &uresp.push_wqe_mmap_key, &uresp.push_db_mmap_key)) {
1429 uresp.push_valid = 1;
1430 uresp.push_offset = iwqp->sc_qp.push_offset;
1432 ret = ib_copy_to_udata(udata, &uresp, min(sizeof(uresp),
1435 irdma_remove_push_mmap_entries(iwqp);
1436 ibdev_dbg(&iwdev->ibdev,
1437 "VERBS: copy_to_udata failed\n");
1445 spin_unlock_irqrestore(&iwqp->lock, flags);
1451 * irdma_modify_qp - modify qp request
1452 * @ibqp: qp's pointer for modify
1453 * @attr: access attributes
1454 * @attr_mask: state mask
1457 int irdma_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask,
1458 struct ib_udata *udata)
1460 struct irdma_qp *iwqp = to_iwqp(ibqp);
1461 struct irdma_device *iwdev = iwqp->iwdev;
1462 struct irdma_sc_dev *dev = &iwdev->rf->sc_dev;
1463 struct irdma_qp_host_ctx_info *ctx_info;
1464 struct irdma_tcp_offload_info *tcp_info;
1465 struct irdma_iwarp_offload_info *offload_info;
1466 struct irdma_modify_qp_info info = {};
1467 struct irdma_modify_qp_resp uresp = {};
1468 struct irdma_modify_qp_req ureq = {};
1469 u8 issue_modify_qp = 0;
1472 unsigned long flags;
1474 if (attr_mask & ~IB_QP_ATTR_STANDARD_BITS)
1477 ctx_info = &iwqp->ctx_info;
1478 offload_info = &iwqp->iwarp_info;
1479 tcp_info = &iwqp->tcp_info;
1480 wait_event(iwqp->mod_qp_waitq, !atomic_read(&iwqp->hw_mod_qp_pend));
1481 ibdev_dbg(&iwdev->ibdev,
1482 "VERBS: caller: %pS qp_id=%d to_ibqpstate=%d ibqpstate=%d irdma_qpstate=%d last_aeq=%d hw_tcp_state=%d hw_iwarp_state=%d attr_mask=0x%x\n",
1483 __builtin_return_address(0), ibqp->qp_num, attr->qp_state,
1484 iwqp->ibqp_state, iwqp->iwarp_state, iwqp->last_aeq,
1485 iwqp->hw_tcp_state, iwqp->hw_iwarp_state, attr_mask);
1487 spin_lock_irqsave(&iwqp->lock, flags);
1488 if (attr_mask & IB_QP_STATE) {
1489 info.curr_iwarp_state = iwqp->iwarp_state;
1490 switch (attr->qp_state) {
1493 if (iwqp->iwarp_state > IRDMA_QP_STATE_IDLE) {
1498 if (iwqp->iwarp_state == IRDMA_QP_STATE_INVALID) {
1499 info.next_iwarp_state = IRDMA_QP_STATE_IDLE;
1500 issue_modify_qp = 1;
1502 if (iwdev->push_mode && udata &&
1503 iwqp->sc_qp.push_idx == IRDMA_INVALID_PUSH_PAGE_INDEX &&
1504 dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2) {
1505 spin_unlock_irqrestore(&iwqp->lock, flags);
1506 irdma_alloc_push_page(iwqp);
1507 spin_lock_irqsave(&iwqp->lock, flags);
1511 if (iwqp->iwarp_state > IRDMA_QP_STATE_RTS ||
1517 issue_modify_qp = 1;
1518 iwqp->hw_tcp_state = IRDMA_TCP_STATE_ESTABLISHED;
1519 iwqp->hte_added = 1;
1520 info.next_iwarp_state = IRDMA_QP_STATE_RTS;
1521 info.tcp_ctx_valid = true;
1522 info.ord_valid = true;
1523 info.arp_cache_idx_valid = true;
1524 info.cq_num_valid = true;
1527 if (iwqp->hw_iwarp_state > IRDMA_QP_STATE_RTS) {
1532 if (iwqp->iwarp_state == IRDMA_QP_STATE_CLOSING ||
1533 iwqp->iwarp_state < IRDMA_QP_STATE_RTS) {
1538 if (iwqp->iwarp_state > IRDMA_QP_STATE_CLOSING) {
1543 info.next_iwarp_state = IRDMA_QP_STATE_CLOSING;
1544 issue_modify_qp = 1;
1547 if (iwqp->iwarp_state >= IRDMA_QP_STATE_TERMINATE) {
1552 info.next_iwarp_state = IRDMA_QP_STATE_TERMINATE;
1553 issue_modify_qp = 1;
1557 if (iwqp->iwarp_state == IRDMA_QP_STATE_ERROR) {
1558 spin_unlock_irqrestore(&iwqp->lock, flags);
1560 if (ib_copy_from_udata(&ureq, udata,
1561 min(sizeof(ureq), udata->inlen)))
1564 irdma_flush_wqes(iwqp,
1565 (ureq.sq_flush ? IRDMA_FLUSH_SQ : 0) |
1566 (ureq.rq_flush ? IRDMA_FLUSH_RQ : 0) |
1572 if (iwqp->sc_qp.term_flags) {
1573 spin_unlock_irqrestore(&iwqp->lock, flags);
1574 irdma_terminate_del_timer(&iwqp->sc_qp);
1575 spin_lock_irqsave(&iwqp->lock, flags);
1577 info.next_iwarp_state = IRDMA_QP_STATE_ERROR;
1578 if (iwqp->hw_tcp_state > IRDMA_TCP_STATE_CLOSED &&
1580 iwqp->hw_tcp_state != IRDMA_TCP_STATE_TIME_WAIT)
1581 info.reset_tcp_conn = true;
1585 issue_modify_qp = 1;
1586 info.next_iwarp_state = IRDMA_QP_STATE_ERROR;
1593 iwqp->ibqp_state = attr->qp_state;
1595 if (attr_mask & IB_QP_ACCESS_FLAGS) {
1596 ctx_info->iwarp_info_valid = true;
1597 if (attr->qp_access_flags & IB_ACCESS_LOCAL_WRITE)
1598 offload_info->wr_rdresp_en = true;
1599 if (attr->qp_access_flags & IB_ACCESS_REMOTE_WRITE)
1600 offload_info->wr_rdresp_en = true;
1601 if (attr->qp_access_flags & IB_ACCESS_REMOTE_READ)
1602 offload_info->rd_en = true;
1605 if (ctx_info->iwarp_info_valid) {
1606 ctx_info->send_cq_num = iwqp->iwscq->sc_cq.cq_uk.cq_id;
1607 ctx_info->rcv_cq_num = iwqp->iwrcq->sc_cq.cq_uk.cq_id;
1608 irdma_sc_qp_setctx(&iwqp->sc_qp, iwqp->host_ctx.va, ctx_info);
1610 spin_unlock_irqrestore(&iwqp->lock, flags);
1612 if (attr_mask & IB_QP_STATE) {
1613 if (issue_modify_qp) {
1614 ctx_info->rem_endpoint_idx = tcp_info->arp_idx;
1615 if (irdma_hw_modify_qp(iwdev, iwqp, &info, true))
1619 spin_lock_irqsave(&iwqp->lock, flags);
1620 if (iwqp->iwarp_state == info.curr_iwarp_state) {
1621 iwqp->iwarp_state = info.next_iwarp_state;
1622 iwqp->ibqp_state = attr->qp_state;
1624 spin_unlock_irqrestore(&iwqp->lock, flags);
1627 if (issue_modify_qp && iwqp->ibqp_state > IB_QPS_RTS) {
1629 if (iwqp->cm_id && iwqp->hw_tcp_state) {
1630 spin_lock_irqsave(&iwqp->lock, flags);
1631 iwqp->hw_tcp_state = IRDMA_TCP_STATE_CLOSED;
1632 iwqp->last_aeq = IRDMA_AE_RESET_SENT;
1633 spin_unlock_irqrestore(&iwqp->lock, flags);
1634 irdma_cm_disconn(iwqp);
1637 int close_timer_started;
1639 spin_lock_irqsave(&iwdev->cm_core.ht_lock, flags);
1641 if (iwqp->cm_node) {
1642 refcount_inc(&iwqp->cm_node->refcnt);
1643 spin_unlock_irqrestore(&iwdev->cm_core.ht_lock, flags);
1644 close_timer_started = atomic_inc_return(&iwqp->close_timer_started);
1645 if (iwqp->cm_id && close_timer_started == 1)
1646 irdma_schedule_cm_timer(iwqp->cm_node,
1647 (struct irdma_puda_buf *)iwqp,
1648 IRDMA_TIMER_TYPE_CLOSE, 1, 0);
1650 irdma_rem_ref_cm_node(iwqp->cm_node);
1652 spin_unlock_irqrestore(&iwdev->cm_core.ht_lock, flags);
1656 if (attr_mask & IB_QP_STATE && udata &&
1657 dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2) {
1658 struct irdma_ucontext *ucontext;
1660 ucontext = rdma_udata_to_drv_context(udata,
1661 struct irdma_ucontext, ibucontext);
1662 if (iwqp->sc_qp.push_idx != IRDMA_INVALID_PUSH_PAGE_INDEX &&
1663 !iwqp->push_wqe_mmap_entry &&
1664 !irdma_setup_push_mmap_entries(ucontext, iwqp,
1665 &uresp.push_wqe_mmap_key, &uresp.push_db_mmap_key)) {
1666 uresp.push_valid = 1;
1667 uresp.push_offset = iwqp->sc_qp.push_offset;
1670 err = ib_copy_to_udata(udata, &uresp, min(sizeof(uresp),
1673 irdma_remove_push_mmap_entries(iwqp);
1674 ibdev_dbg(&iwdev->ibdev,
1675 "VERBS: copy_to_udata failed\n");
1682 spin_unlock_irqrestore(&iwqp->lock, flags);
1688 * irdma_cq_free_rsrc - free up resources for cq
1689 * @rf: RDMA PCI function
1692 static void irdma_cq_free_rsrc(struct irdma_pci_f *rf, struct irdma_cq *iwcq)
1694 struct irdma_sc_cq *cq = &iwcq->sc_cq;
1696 if (!iwcq->user_mode) {
1697 dma_free_coherent(rf->sc_dev.hw->device, iwcq->kmem.size,
1698 iwcq->kmem.va, iwcq->kmem.pa);
1699 iwcq->kmem.va = NULL;
1700 dma_free_coherent(rf->sc_dev.hw->device,
1701 iwcq->kmem_shadow.size,
1702 iwcq->kmem_shadow.va, iwcq->kmem_shadow.pa);
1703 iwcq->kmem_shadow.va = NULL;
1706 irdma_free_rsrc(rf, rf->allocated_cqs, cq->cq_uk.cq_id);
1710 * irdma_free_cqbuf - worker to free a cq buffer
1711 * @work: provides access to the cq buffer to free
1713 static void irdma_free_cqbuf(struct work_struct *work)
1715 struct irdma_cq_buf *cq_buf = container_of(work, struct irdma_cq_buf, work);
1717 dma_free_coherent(cq_buf->hw->device, cq_buf->kmem_buf.size,
1718 cq_buf->kmem_buf.va, cq_buf->kmem_buf.pa);
1719 cq_buf->kmem_buf.va = NULL;
1724 * irdma_process_resize_list - remove resized cq buffers from the resize_list
1725 * @iwcq: cq which owns the resize_list
1726 * @iwdev: irdma device
1727 * @lcqe_buf: the buffer where the last cqe is received
1729 static int irdma_process_resize_list(struct irdma_cq *iwcq,
1730 struct irdma_device *iwdev,
1731 struct irdma_cq_buf *lcqe_buf)
1733 struct list_head *tmp_node, *list_node;
1734 struct irdma_cq_buf *cq_buf;
1737 list_for_each_safe(list_node, tmp_node, &iwcq->resize_list) {
1738 cq_buf = list_entry(list_node, struct irdma_cq_buf, list);
1739 if (cq_buf == lcqe_buf)
1742 list_del(&cq_buf->list);
1743 queue_work(iwdev->cleanup_wq, &cq_buf->work);
1751 * irdma_destroy_cq - destroy cq
1752 * @ib_cq: cq pointer
1755 static int irdma_destroy_cq(struct ib_cq *ib_cq, struct ib_udata *udata)
1757 struct irdma_device *iwdev = to_iwdev(ib_cq->device);
1758 struct irdma_cq *iwcq = to_iwcq(ib_cq);
1759 struct irdma_sc_cq *cq = &iwcq->sc_cq;
1760 struct irdma_sc_dev *dev = cq->dev;
1761 struct irdma_sc_ceq *ceq = dev->ceq[cq->ceq_id];
1762 struct irdma_ceq *iwceq = container_of(ceq, struct irdma_ceq, sc_ceq);
1763 unsigned long flags;
1765 spin_lock_irqsave(&iwcq->lock, flags);
1766 if (!list_empty(&iwcq->resize_list))
1767 irdma_process_resize_list(iwcq, iwdev, NULL);
1768 spin_unlock_irqrestore(&iwcq->lock, flags);
1770 irdma_cq_wq_destroy(iwdev->rf, cq);
1771 irdma_cq_free_rsrc(iwdev->rf, iwcq);
1773 spin_lock_irqsave(&iwceq->ce_lock, flags);
1774 irdma_sc_cleanup_ceqes(cq, ceq);
1775 spin_unlock_irqrestore(&iwceq->ce_lock, flags);
1781 * irdma_resize_cq - resize cq
1782 * @ibcq: cq to be resized
1783 * @entries: desired cq size
1786 static int irdma_resize_cq(struct ib_cq *ibcq, int entries,
1787 struct ib_udata *udata)
1789 struct irdma_cq *iwcq = to_iwcq(ibcq);
1790 struct irdma_sc_dev *dev = iwcq->sc_cq.dev;
1791 struct irdma_cqp_request *cqp_request;
1792 struct cqp_cmds_info *cqp_info;
1793 struct irdma_modify_cq_info *m_info;
1794 struct irdma_modify_cq_info info = {};
1795 struct irdma_dma_mem kmem_buf;
1796 struct irdma_cq_mr *cqmr_buf;
1797 struct irdma_pbl *iwpbl_buf;
1798 struct irdma_device *iwdev;
1799 struct irdma_pci_f *rf;
1800 struct irdma_cq_buf *cq_buf = NULL;
1801 enum irdma_status_code status = 0;
1802 unsigned long flags;
1805 iwdev = to_iwdev(ibcq->device);
1808 if (!(rf->sc_dev.hw_attrs.uk_attrs.feature_flags &
1809 IRDMA_FEATURE_CQ_RESIZE))
1812 if (entries > rf->max_cqe)
1815 if (!iwcq->user_mode) {
1817 if (rf->sc_dev.hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2)
1821 info.cq_size = max(entries, 4);
1823 if (info.cq_size == iwcq->sc_cq.cq_uk.cq_size - 1)
1827 struct irdma_resize_cq_req req = {};
1828 struct irdma_ucontext *ucontext =
1829 rdma_udata_to_drv_context(udata, struct irdma_ucontext,
1832 /* CQ resize not supported with legacy GEN_1 libi40iw */
1833 if (ucontext->legacy_mode)
1836 if (ib_copy_from_udata(&req, udata,
1837 min(sizeof(req), udata->inlen)))
1840 spin_lock_irqsave(&ucontext->cq_reg_mem_list_lock, flags);
1841 iwpbl_buf = irdma_get_pbl((unsigned long)req.user_cq_buffer,
1842 &ucontext->cq_reg_mem_list);
1843 spin_unlock_irqrestore(&ucontext->cq_reg_mem_list_lock, flags);
1848 cqmr_buf = &iwpbl_buf->cq_mr;
1849 if (iwpbl_buf->pbl_allocated) {
1850 info.virtual_map = true;
1851 info.pbl_chunk_size = 1;
1852 info.first_pm_pbl_idx = cqmr_buf->cq_pbl.idx;
1854 info.cq_pa = cqmr_buf->cq_pbl.addr;
1857 /* Kmode CQ resize */
1860 rsize = info.cq_size * sizeof(struct irdma_cqe);
1861 kmem_buf.size = ALIGN(round_up(rsize, 256), 256);
1862 kmem_buf.va = dma_alloc_coherent(dev->hw->device,
1863 kmem_buf.size, &kmem_buf.pa,
1868 info.cq_base = kmem_buf.va;
1869 info.cq_pa = kmem_buf.pa;
1870 cq_buf = kzalloc(sizeof(*cq_buf), GFP_KERNEL);
1877 cqp_request = irdma_alloc_and_get_cqp_request(&rf->cqp, true);
1883 info.shadow_read_threshold = iwcq->sc_cq.shadow_read_threshold;
1884 info.cq_resize = true;
1886 cqp_info = &cqp_request->info;
1887 m_info = &cqp_info->in.u.cq_modify.info;
1888 memcpy(m_info, &info, sizeof(*m_info));
1890 cqp_info->cqp_cmd = IRDMA_OP_CQ_MODIFY;
1891 cqp_info->in.u.cq_modify.cq = &iwcq->sc_cq;
1892 cqp_info->in.u.cq_modify.scratch = (uintptr_t)cqp_request;
1893 cqp_info->post_sq = 1;
1894 status = irdma_handle_cqp_op(rf, cqp_request);
1895 irdma_put_cqp_request(&rf->cqp, cqp_request);
1901 spin_lock_irqsave(&iwcq->lock, flags);
1903 cq_buf->kmem_buf = iwcq->kmem;
1904 cq_buf->hw = dev->hw;
1905 memcpy(&cq_buf->cq_uk, &iwcq->sc_cq.cq_uk, sizeof(cq_buf->cq_uk));
1906 INIT_WORK(&cq_buf->work, irdma_free_cqbuf);
1907 list_add_tail(&cq_buf->list, &iwcq->resize_list);
1908 iwcq->kmem = kmem_buf;
1911 irdma_sc_cq_resize(&iwcq->sc_cq, &info);
1912 ibcq->cqe = info.cq_size - 1;
1913 spin_unlock_irqrestore(&iwcq->lock, flags);
1918 dma_free_coherent(dev->hw->device, kmem_buf.size, kmem_buf.va,
1927 static inline int cq_validate_flags(u32 flags, u8 hw_rev)
1929 /* GEN1 does not support CQ create flags */
1930 if (hw_rev == IRDMA_GEN_1)
1931 return flags ? -EOPNOTSUPP : 0;
1933 return flags & ~IB_UVERBS_CQ_FLAGS_TIMESTAMP_COMPLETION ? -EOPNOTSUPP : 0;
1937 * irdma_create_cq - create cq
1938 * @ibcq: CQ allocated
1939 * @attr: attributes for cq
1942 static int irdma_create_cq(struct ib_cq *ibcq,
1943 const struct ib_cq_init_attr *attr,
1944 struct ib_udata *udata)
1946 struct ib_device *ibdev = ibcq->device;
1947 struct irdma_device *iwdev = to_iwdev(ibdev);
1948 struct irdma_pci_f *rf = iwdev->rf;
1949 struct irdma_cq *iwcq = to_iwcq(ibcq);
1951 struct irdma_sc_cq *cq;
1952 struct irdma_sc_dev *dev = &rf->sc_dev;
1953 struct irdma_cq_init_info info = {};
1954 enum irdma_status_code status;
1955 struct irdma_cqp_request *cqp_request;
1956 struct cqp_cmds_info *cqp_info;
1957 struct irdma_cq_uk_init_info *ukinfo = &info.cq_uk_init_info;
1958 unsigned long flags;
1960 int entries = attr->cqe;
1962 err_code = cq_validate_flags(attr->flags, dev->hw_attrs.uk_attrs.hw_rev);
1965 err_code = irdma_alloc_rsrc(rf, rf->allocated_cqs, rf->max_cq, &cq_num,
1972 spin_lock_init(&iwcq->lock);
1973 INIT_LIST_HEAD(&iwcq->resize_list);
1975 ukinfo->cq_size = max(entries, 4);
1976 ukinfo->cq_id = cq_num;
1977 iwcq->ibcq.cqe = info.cq_uk_init_info.cq_size;
1978 if (attr->comp_vector < rf->ceqs_count)
1979 info.ceq_id = attr->comp_vector;
1980 info.ceq_id_valid = true;
1982 info.type = IRDMA_CQ_TYPE_IWARP;
1983 info.vsi = &iwdev->vsi;
1986 struct irdma_ucontext *ucontext;
1987 struct irdma_create_cq_req req = {};
1988 struct irdma_cq_mr *cqmr;
1989 struct irdma_pbl *iwpbl;
1990 struct irdma_pbl *iwpbl_shadow;
1991 struct irdma_cq_mr *cqmr_shadow;
1993 iwcq->user_mode = true;
1995 rdma_udata_to_drv_context(udata, struct irdma_ucontext,
1997 if (ib_copy_from_udata(&req, udata,
1998 min(sizeof(req), udata->inlen))) {
2003 spin_lock_irqsave(&ucontext->cq_reg_mem_list_lock, flags);
2004 iwpbl = irdma_get_pbl((unsigned long)req.user_cq_buf,
2005 &ucontext->cq_reg_mem_list);
2006 spin_unlock_irqrestore(&ucontext->cq_reg_mem_list_lock, flags);
2012 iwcq->iwpbl = iwpbl;
2013 iwcq->cq_mem_size = 0;
2014 cqmr = &iwpbl->cq_mr;
2016 if (rf->sc_dev.hw_attrs.uk_attrs.feature_flags &
2017 IRDMA_FEATURE_CQ_RESIZE && !ucontext->legacy_mode) {
2018 spin_lock_irqsave(&ucontext->cq_reg_mem_list_lock, flags);
2019 iwpbl_shadow = irdma_get_pbl(
2020 (unsigned long)req.user_shadow_area,
2021 &ucontext->cq_reg_mem_list);
2022 spin_unlock_irqrestore(&ucontext->cq_reg_mem_list_lock, flags);
2024 if (!iwpbl_shadow) {
2028 iwcq->iwpbl_shadow = iwpbl_shadow;
2029 cqmr_shadow = &iwpbl_shadow->cq_mr;
2030 info.shadow_area_pa = cqmr_shadow->cq_pbl.addr;
2033 info.shadow_area_pa = cqmr->shadow;
2035 if (iwpbl->pbl_allocated) {
2036 info.virtual_map = true;
2037 info.pbl_chunk_size = 1;
2038 info.first_pm_pbl_idx = cqmr->cq_pbl.idx;
2040 info.cq_base_pa = cqmr->cq_pbl.addr;
2043 /* Kmode allocations */
2046 if (entries > rf->max_cqe) {
2052 if (dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2)
2054 ukinfo->cq_size = entries;
2056 rsize = info.cq_uk_init_info.cq_size * sizeof(struct irdma_cqe);
2057 iwcq->kmem.size = ALIGN(round_up(rsize, 256), 256);
2058 iwcq->kmem.va = dma_alloc_coherent(dev->hw->device,
2060 &iwcq->kmem.pa, GFP_KERNEL);
2061 if (!iwcq->kmem.va) {
2066 iwcq->kmem_shadow.size = ALIGN(IRDMA_SHADOW_AREA_SIZE << 3,
2068 iwcq->kmem_shadow.va = dma_alloc_coherent(dev->hw->device,
2069 iwcq->kmem_shadow.size,
2070 &iwcq->kmem_shadow.pa,
2072 if (!iwcq->kmem_shadow.va) {
2076 info.shadow_area_pa = iwcq->kmem_shadow.pa;
2077 ukinfo->shadow_area = iwcq->kmem_shadow.va;
2078 ukinfo->cq_base = iwcq->kmem.va;
2079 info.cq_base_pa = iwcq->kmem.pa;
2082 if (dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2)
2083 info.shadow_read_threshold = min(info.cq_uk_init_info.cq_size / 2,
2084 (u32)IRDMA_MAX_CQ_READ_THRESH);
2086 if (irdma_sc_cq_init(cq, &info)) {
2087 ibdev_dbg(&iwdev->ibdev, "VERBS: init cq fail\n");
2092 cqp_request = irdma_alloc_and_get_cqp_request(&rf->cqp, true);
2098 cqp_info = &cqp_request->info;
2099 cqp_info->cqp_cmd = IRDMA_OP_CQ_CREATE;
2100 cqp_info->post_sq = 1;
2101 cqp_info->in.u.cq_create.cq = cq;
2102 cqp_info->in.u.cq_create.check_overflow = true;
2103 cqp_info->in.u.cq_create.scratch = (uintptr_t)cqp_request;
2104 status = irdma_handle_cqp_op(rf, cqp_request);
2105 irdma_put_cqp_request(&rf->cqp, cqp_request);
2112 struct irdma_create_cq_resp resp = {};
2114 resp.cq_id = info.cq_uk_init_info.cq_id;
2115 resp.cq_size = info.cq_uk_init_info.cq_size;
2116 if (ib_copy_to_udata(udata, &resp,
2117 min(sizeof(resp), udata->outlen))) {
2118 ibdev_dbg(&iwdev->ibdev,
2119 "VERBS: copy to user data\n");
2126 irdma_cq_wq_destroy(rf, cq);
2128 irdma_cq_free_rsrc(rf, iwcq);
2134 * irdma_get_mr_access - get hw MR access permissions from IB access flags
2135 * @access: IB access flags
2137 static inline u16 irdma_get_mr_access(int access)
2141 hw_access |= (access & IB_ACCESS_LOCAL_WRITE) ?
2142 IRDMA_ACCESS_FLAGS_LOCALWRITE : 0;
2143 hw_access |= (access & IB_ACCESS_REMOTE_WRITE) ?
2144 IRDMA_ACCESS_FLAGS_REMOTEWRITE : 0;
2145 hw_access |= (access & IB_ACCESS_REMOTE_READ) ?
2146 IRDMA_ACCESS_FLAGS_REMOTEREAD : 0;
2147 hw_access |= (access & IB_ACCESS_MW_BIND) ?
2148 IRDMA_ACCESS_FLAGS_BIND_WINDOW : 0;
2149 hw_access |= (access & IB_ZERO_BASED) ?
2150 IRDMA_ACCESS_FLAGS_ZERO_BASED : 0;
2151 hw_access |= IRDMA_ACCESS_FLAGS_LOCALREAD;
2157 * irdma_free_stag - free stag resource
2158 * @iwdev: irdma device
2159 * @stag: stag to free
2161 static void irdma_free_stag(struct irdma_device *iwdev, u32 stag)
2165 stag_idx = (stag & iwdev->rf->mr_stagmask) >> IRDMA_CQPSQ_STAG_IDX_S;
2166 irdma_free_rsrc(iwdev->rf, iwdev->rf->allocated_mrs, stag_idx);
2170 * irdma_create_stag - create random stag
2171 * @iwdev: irdma device
2173 static u32 irdma_create_stag(struct irdma_device *iwdev)
2177 u32 next_stag_index;
2183 get_random_bytes(&random, sizeof(random));
2184 consumer_key = (u8)random;
2186 driver_key = random & ~iwdev->rf->mr_stagmask;
2187 next_stag_index = (random & iwdev->rf->mr_stagmask) >> 8;
2188 next_stag_index %= iwdev->rf->max_mr;
2190 ret = irdma_alloc_rsrc(iwdev->rf, iwdev->rf->allocated_mrs,
2191 iwdev->rf->max_mr, &stag_index,
2195 stag = stag_index << IRDMA_CQPSQ_STAG_IDX_S;
2197 stag += (u32)consumer_key;
2203 * irdma_next_pbl_addr - Get next pbl address
2204 * @pbl: pointer to a pble
2205 * @pinfo: info pointer
2208 static inline u64 *irdma_next_pbl_addr(u64 *pbl, struct irdma_pble_info **pinfo,
2212 if (!(*pinfo) || *idx != (*pinfo)->cnt)
2217 return (*pinfo)->addr;
2221 * irdma_copy_user_pgaddrs - copy user page address to pble's os locally
2222 * @iwmr: iwmr for IB's user page addresses
2223 * @pbl: ple pointer to save 1 level or 0 level pble
2224 * @level: indicated level 0, 1 or 2
2226 static void irdma_copy_user_pgaddrs(struct irdma_mr *iwmr, u64 *pbl,
2227 enum irdma_pble_level level)
2229 struct ib_umem *region = iwmr->region;
2230 struct irdma_pbl *iwpbl = &iwmr->iwpbl;
2231 struct irdma_pble_alloc *palloc = &iwpbl->pble_alloc;
2232 struct irdma_pble_info *pinfo;
2233 struct ib_block_iter biter;
2237 pinfo = (level == PBLE_LEVEL_1) ? NULL : palloc->level2.leaf;
2239 if (iwmr->type == IRDMA_MEMREG_TYPE_QP)
2240 iwpbl->qp_mr.sq_page = sg_page(region->sg_head.sgl);
2242 rdma_umem_for_each_dma_block(region, &biter, iwmr->page_size) {
2243 *pbl = rdma_block_iter_dma_address(&biter);
2244 if (++pbl_cnt == palloc->total_cnt)
2246 pbl = irdma_next_pbl_addr(pbl, &pinfo, &idx);
2251 * irdma_check_mem_contiguous - check if pbls stored in arr are contiguous
2252 * @arr: lvl1 pbl array
2253 * @npages: page count
2254 * @pg_size: page size
2257 static bool irdma_check_mem_contiguous(u64 *arr, u32 npages, u32 pg_size)
2261 for (pg_idx = 0; pg_idx < npages; pg_idx++) {
2262 if ((*arr + (pg_size * pg_idx)) != arr[pg_idx])
2270 * irdma_check_mr_contiguous - check if MR is physically contiguous
2271 * @palloc: pbl allocation struct
2272 * @pg_size: page size
2274 static bool irdma_check_mr_contiguous(struct irdma_pble_alloc *palloc,
2277 struct irdma_pble_level2 *lvl2 = &palloc->level2;
2278 struct irdma_pble_info *leaf = lvl2->leaf;
2280 u64 *start_addr = NULL;
2284 if (palloc->level == PBLE_LEVEL_1) {
2285 arr = palloc->level1.addr;
2286 ret = irdma_check_mem_contiguous(arr, palloc->total_cnt,
2291 start_addr = leaf->addr;
2293 for (i = 0; i < lvl2->leaf_cnt; i++, leaf++) {
2295 if ((*start_addr + (i * pg_size * PBLE_PER_PAGE)) != *arr)
2297 ret = irdma_check_mem_contiguous(arr, leaf->cnt, pg_size);
2306 * irdma_setup_pbles - copy user pg address to pble's
2307 * @rf: RDMA PCI function
2308 * @iwmr: mr pointer for this memory registration
2309 * @use_pbles: flag if to use pble's
2311 static int irdma_setup_pbles(struct irdma_pci_f *rf, struct irdma_mr *iwmr,
2314 struct irdma_pbl *iwpbl = &iwmr->iwpbl;
2315 struct irdma_pble_alloc *palloc = &iwpbl->pble_alloc;
2316 struct irdma_pble_info *pinfo;
2318 enum irdma_status_code status;
2319 enum irdma_pble_level level = PBLE_LEVEL_1;
2322 status = irdma_get_pble(rf->pble_rsrc, palloc, iwmr->page_cnt,
2327 iwpbl->pbl_allocated = true;
2328 level = palloc->level;
2329 pinfo = (level == PBLE_LEVEL_1) ? &palloc->level1 :
2330 palloc->level2.leaf;
2333 pbl = iwmr->pgaddrmem;
2336 irdma_copy_user_pgaddrs(iwmr, pbl, level);
2339 iwmr->pgaddrmem[0] = *pbl;
2345 * irdma_handle_q_mem - handle memory for qp and cq
2346 * @iwdev: irdma device
2347 * @req: information for q memory management
2348 * @iwpbl: pble struct
2349 * @use_pbles: flag to use pble
2351 static int irdma_handle_q_mem(struct irdma_device *iwdev,
2352 struct irdma_mem_reg_req *req,
2353 struct irdma_pbl *iwpbl, bool use_pbles)
2355 struct irdma_pble_alloc *palloc = &iwpbl->pble_alloc;
2356 struct irdma_mr *iwmr = iwpbl->iwmr;
2357 struct irdma_qp_mr *qpmr = &iwpbl->qp_mr;
2358 struct irdma_cq_mr *cqmr = &iwpbl->cq_mr;
2359 struct irdma_hmc_pble *hmc_p;
2360 u64 *arr = iwmr->pgaddrmem;
2365 pg_size = iwmr->page_size;
2366 err = irdma_setup_pbles(iwdev->rf, iwmr, use_pbles);
2370 if (use_pbles && palloc->level != PBLE_LEVEL_1) {
2371 irdma_free_pble(iwdev->rf->pble_rsrc, palloc);
2372 iwpbl->pbl_allocated = false;
2377 arr = palloc->level1.addr;
2379 switch (iwmr->type) {
2380 case IRDMA_MEMREG_TYPE_QP:
2381 total = req->sq_pages + req->rq_pages;
2382 hmc_p = &qpmr->sq_pbl;
2383 qpmr->shadow = (dma_addr_t)arr[total];
2386 ret = irdma_check_mem_contiguous(arr, req->sq_pages,
2389 ret = irdma_check_mem_contiguous(&arr[req->sq_pages],
2395 hmc_p->idx = palloc->level1.idx;
2396 hmc_p = &qpmr->rq_pbl;
2397 hmc_p->idx = palloc->level1.idx + req->sq_pages;
2399 hmc_p->addr = arr[0];
2400 hmc_p = &qpmr->rq_pbl;
2401 hmc_p->addr = arr[req->sq_pages];
2404 case IRDMA_MEMREG_TYPE_CQ:
2405 hmc_p = &cqmr->cq_pbl;
2408 cqmr->shadow = (dma_addr_t)arr[req->cq_pages];
2411 ret = irdma_check_mem_contiguous(arr, req->cq_pages,
2415 hmc_p->idx = palloc->level1.idx;
2417 hmc_p->addr = arr[0];
2420 ibdev_dbg(&iwdev->ibdev, "VERBS: MR type error\n");
2424 if (use_pbles && ret) {
2425 irdma_free_pble(iwdev->rf->pble_rsrc, palloc);
2426 iwpbl->pbl_allocated = false;
2433 * irdma_hw_alloc_mw - create the hw memory window
2434 * @iwdev: irdma device
2435 * @iwmr: pointer to memory window info
2437 static int irdma_hw_alloc_mw(struct irdma_device *iwdev, struct irdma_mr *iwmr)
2439 struct irdma_mw_alloc_info *info;
2440 struct irdma_pd *iwpd = to_iwpd(iwmr->ibmr.pd);
2441 struct irdma_cqp_request *cqp_request;
2442 struct cqp_cmds_info *cqp_info;
2443 enum irdma_status_code status;
2445 cqp_request = irdma_alloc_and_get_cqp_request(&iwdev->rf->cqp, true);
2449 cqp_info = &cqp_request->info;
2450 info = &cqp_info->in.u.mw_alloc.info;
2451 memset(info, 0, sizeof(*info));
2452 if (iwmr->ibmw.type == IB_MW_TYPE_1)
2453 info->mw_wide = true;
2455 info->page_size = PAGE_SIZE;
2456 info->mw_stag_index = iwmr->stag >> IRDMA_CQPSQ_STAG_IDX_S;
2457 info->pd_id = iwpd->sc_pd.pd_id;
2458 info->remote_access = true;
2459 cqp_info->cqp_cmd = IRDMA_OP_MW_ALLOC;
2460 cqp_info->post_sq = 1;
2461 cqp_info->in.u.mw_alloc.dev = &iwdev->rf->sc_dev;
2462 cqp_info->in.u.mw_alloc.scratch = (uintptr_t)cqp_request;
2463 status = irdma_handle_cqp_op(iwdev->rf, cqp_request);
2464 irdma_put_cqp_request(&iwdev->rf->cqp, cqp_request);
2466 return status ? -ENOMEM : 0;
2470 * irdma_alloc_mw - Allocate memory window
2471 * @ibmw: Memory Window
2472 * @udata: user data pointer
2474 static int irdma_alloc_mw(struct ib_mw *ibmw, struct ib_udata *udata)
2476 struct irdma_device *iwdev = to_iwdev(ibmw->device);
2477 struct irdma_mr *iwmr = to_iwmw(ibmw);
2481 stag = irdma_create_stag(iwdev);
2488 err_code = irdma_hw_alloc_mw(iwdev, iwmr);
2490 irdma_free_stag(iwdev, stag);
2498 * irdma_dealloc_mw - Dealloc memory window
2499 * @ibmw: memory window structure.
2501 static int irdma_dealloc_mw(struct ib_mw *ibmw)
2503 struct ib_pd *ibpd = ibmw->pd;
2504 struct irdma_pd *iwpd = to_iwpd(ibpd);
2505 struct irdma_mr *iwmr = to_iwmr((struct ib_mr *)ibmw);
2506 struct irdma_device *iwdev = to_iwdev(ibmw->device);
2507 struct irdma_cqp_request *cqp_request;
2508 struct cqp_cmds_info *cqp_info;
2509 struct irdma_dealloc_stag_info *info;
2511 cqp_request = irdma_alloc_and_get_cqp_request(&iwdev->rf->cqp, true);
2515 cqp_info = &cqp_request->info;
2516 info = &cqp_info->in.u.dealloc_stag.info;
2517 memset(info, 0, sizeof(*info));
2518 info->pd_id = iwpd->sc_pd.pd_id & 0x00007fff;
2519 info->stag_idx = ibmw->rkey >> IRDMA_CQPSQ_STAG_IDX_S;
2521 cqp_info->cqp_cmd = IRDMA_OP_DEALLOC_STAG;
2522 cqp_info->post_sq = 1;
2523 cqp_info->in.u.dealloc_stag.dev = &iwdev->rf->sc_dev;
2524 cqp_info->in.u.dealloc_stag.scratch = (uintptr_t)cqp_request;
2525 irdma_handle_cqp_op(iwdev->rf, cqp_request);
2526 irdma_put_cqp_request(&iwdev->rf->cqp, cqp_request);
2527 irdma_free_stag(iwdev, iwmr->stag);
2533 * irdma_hw_alloc_stag - cqp command to allocate stag
2534 * @iwdev: irdma device
2535 * @iwmr: irdma mr pointer
2537 static int irdma_hw_alloc_stag(struct irdma_device *iwdev,
2538 struct irdma_mr *iwmr)
2540 struct irdma_allocate_stag_info *info;
2541 struct irdma_pd *iwpd = to_iwpd(iwmr->ibmr.pd);
2542 enum irdma_status_code status;
2544 struct irdma_cqp_request *cqp_request;
2545 struct cqp_cmds_info *cqp_info;
2547 cqp_request = irdma_alloc_and_get_cqp_request(&iwdev->rf->cqp, true);
2551 cqp_info = &cqp_request->info;
2552 info = &cqp_info->in.u.alloc_stag.info;
2553 memset(info, 0, sizeof(*info));
2554 info->page_size = PAGE_SIZE;
2555 info->stag_idx = iwmr->stag >> IRDMA_CQPSQ_STAG_IDX_S;
2556 info->pd_id = iwpd->sc_pd.pd_id;
2557 info->total_len = iwmr->len;
2558 info->remote_access = true;
2559 cqp_info->cqp_cmd = IRDMA_OP_ALLOC_STAG;
2560 cqp_info->post_sq = 1;
2561 cqp_info->in.u.alloc_stag.dev = &iwdev->rf->sc_dev;
2562 cqp_info->in.u.alloc_stag.scratch = (uintptr_t)cqp_request;
2563 status = irdma_handle_cqp_op(iwdev->rf, cqp_request);
2564 irdma_put_cqp_request(&iwdev->rf->cqp, cqp_request);
2572 * irdma_alloc_mr - register stag for fast memory registration
2574 * @mr_type: memory for stag registrion
2575 * @max_num_sg: man number of pages
2577 static struct ib_mr *irdma_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
2580 struct irdma_device *iwdev = to_iwdev(pd->device);
2581 struct irdma_pble_alloc *palloc;
2582 struct irdma_pbl *iwpbl;
2583 struct irdma_mr *iwmr;
2584 enum irdma_status_code status;
2586 int err_code = -ENOMEM;
2588 iwmr = kzalloc(sizeof(*iwmr), GFP_KERNEL);
2590 return ERR_PTR(-ENOMEM);
2592 stag = irdma_create_stag(iwdev);
2599 iwmr->ibmr.rkey = stag;
2600 iwmr->ibmr.lkey = stag;
2602 iwmr->ibmr.device = pd->device;
2603 iwpbl = &iwmr->iwpbl;
2605 iwmr->type = IRDMA_MEMREG_TYPE_MEM;
2606 palloc = &iwpbl->pble_alloc;
2607 iwmr->page_cnt = max_num_sg;
2608 status = irdma_get_pble(iwdev->rf->pble_rsrc, palloc, iwmr->page_cnt,
2613 err_code = irdma_hw_alloc_stag(iwdev, iwmr);
2615 goto err_alloc_stag;
2617 iwpbl->pbl_allocated = true;
2621 irdma_free_pble(iwdev->rf->pble_rsrc, palloc);
2623 irdma_free_stag(iwdev, stag);
2627 return ERR_PTR(err_code);
2631 * irdma_set_page - populate pbl list for fmr
2632 * @ibmr: ib mem to access iwarp mr pointer
2633 * @addr: page dma address fro pbl list
2635 static int irdma_set_page(struct ib_mr *ibmr, u64 addr)
2637 struct irdma_mr *iwmr = to_iwmr(ibmr);
2638 struct irdma_pbl *iwpbl = &iwmr->iwpbl;
2639 struct irdma_pble_alloc *palloc = &iwpbl->pble_alloc;
2642 if (unlikely(iwmr->npages == iwmr->page_cnt))
2645 pbl = palloc->level1.addr;
2646 pbl[iwmr->npages++] = addr;
2652 * irdma_map_mr_sg - map of sg list for fmr
2653 * @ibmr: ib mem to access iwarp mr pointer
2654 * @sg: scatter gather list
2655 * @sg_nents: number of sg pages
2656 * @sg_offset: scatter gather list for fmr
2658 static int irdma_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg,
2659 int sg_nents, unsigned int *sg_offset)
2661 struct irdma_mr *iwmr = to_iwmr(ibmr);
2665 return ib_sg_to_pages(ibmr, sg, sg_nents, sg_offset, irdma_set_page);
2669 * irdma_hwreg_mr - send cqp command for memory registration
2670 * @iwdev: irdma device
2671 * @iwmr: irdma mr pointer
2672 * @access: access for MR
2674 static int irdma_hwreg_mr(struct irdma_device *iwdev, struct irdma_mr *iwmr,
2677 struct irdma_pbl *iwpbl = &iwmr->iwpbl;
2678 struct irdma_reg_ns_stag_info *stag_info;
2679 struct irdma_pd *iwpd = to_iwpd(iwmr->ibmr.pd);
2680 struct irdma_pble_alloc *palloc = &iwpbl->pble_alloc;
2681 enum irdma_status_code status;
2683 struct irdma_cqp_request *cqp_request;
2684 struct cqp_cmds_info *cqp_info;
2686 cqp_request = irdma_alloc_and_get_cqp_request(&iwdev->rf->cqp, true);
2690 cqp_info = &cqp_request->info;
2691 stag_info = &cqp_info->in.u.mr_reg_non_shared.info;
2692 memset(stag_info, 0, sizeof(*stag_info));
2693 stag_info->va = iwpbl->user_base;
2694 stag_info->stag_idx = iwmr->stag >> IRDMA_CQPSQ_STAG_IDX_S;
2695 stag_info->stag_key = (u8)iwmr->stag;
2696 stag_info->total_len = iwmr->len;
2697 stag_info->access_rights = irdma_get_mr_access(access);
2698 stag_info->pd_id = iwpd->sc_pd.pd_id;
2699 if (stag_info->access_rights & IRDMA_ACCESS_FLAGS_ZERO_BASED)
2700 stag_info->addr_type = IRDMA_ADDR_TYPE_ZERO_BASED;
2702 stag_info->addr_type = IRDMA_ADDR_TYPE_VA_BASED;
2703 stag_info->page_size = iwmr->page_size;
2705 if (iwpbl->pbl_allocated) {
2706 if (palloc->level == PBLE_LEVEL_1) {
2707 stag_info->first_pm_pbl_index = palloc->level1.idx;
2708 stag_info->chunk_size = 1;
2710 stag_info->first_pm_pbl_index = palloc->level2.root.idx;
2711 stag_info->chunk_size = 3;
2714 stag_info->reg_addr_pa = iwmr->pgaddrmem[0];
2717 cqp_info->cqp_cmd = IRDMA_OP_MR_REG_NON_SHARED;
2718 cqp_info->post_sq = 1;
2719 cqp_info->in.u.mr_reg_non_shared.dev = &iwdev->rf->sc_dev;
2720 cqp_info->in.u.mr_reg_non_shared.scratch = (uintptr_t)cqp_request;
2721 status = irdma_handle_cqp_op(iwdev->rf, cqp_request);
2722 irdma_put_cqp_request(&iwdev->rf->cqp, cqp_request);
2730 * irdma_reg_user_mr - Register a user memory region
2732 * @start: virtual start address
2733 * @len: length of mr
2734 * @virt: virtual address
2735 * @access: access of mr
2738 static struct ib_mr *irdma_reg_user_mr(struct ib_pd *pd, u64 start, u64 len,
2739 u64 virt, int access,
2740 struct ib_udata *udata)
2742 struct irdma_device *iwdev = to_iwdev(pd->device);
2743 struct irdma_ucontext *ucontext;
2744 struct irdma_pble_alloc *palloc;
2745 struct irdma_pbl *iwpbl;
2746 struct irdma_mr *iwmr;
2747 struct ib_umem *region;
2748 struct irdma_mem_reg_req req;
2749 u32 total, stag = 0;
2750 u8 shadow_pgcnt = 1;
2751 bool use_pbles = false;
2752 unsigned long flags;
2756 if (len > iwdev->rf->sc_dev.hw_attrs.max_mr_size)
2757 return ERR_PTR(-EINVAL);
2759 region = ib_umem_get(pd->device, start, len, access);
2761 if (IS_ERR(region)) {
2762 ibdev_dbg(&iwdev->ibdev,
2763 "VERBS: Failed to create ib_umem region\n");
2764 return (struct ib_mr *)region;
2767 if (ib_copy_from_udata(&req, udata, min(sizeof(req), udata->inlen))) {
2768 ib_umem_release(region);
2769 return ERR_PTR(-EFAULT);
2772 iwmr = kzalloc(sizeof(*iwmr), GFP_KERNEL);
2774 ib_umem_release(region);
2775 return ERR_PTR(-ENOMEM);
2778 iwpbl = &iwmr->iwpbl;
2780 iwmr->region = region;
2782 iwmr->ibmr.device = pd->device;
2783 iwmr->ibmr.iova = virt;
2784 iwmr->page_size = PAGE_SIZE;
2786 if (req.reg_type == IRDMA_MEMREG_TYPE_MEM) {
2787 iwmr->page_size = ib_umem_find_best_pgsz(region,
2788 SZ_4K | SZ_2M | SZ_1G,
2790 if (unlikely(!iwmr->page_size)) {
2792 ib_umem_release(region);
2793 return ERR_PTR(-EOPNOTSUPP);
2796 iwmr->len = region->length;
2797 iwpbl->user_base = virt;
2798 palloc = &iwpbl->pble_alloc;
2799 iwmr->type = req.reg_type;
2800 iwmr->page_cnt = ib_umem_num_dma_blocks(region, iwmr->page_size);
2802 switch (req.reg_type) {
2803 case IRDMA_MEMREG_TYPE_QP:
2804 total = req.sq_pages + req.rq_pages + shadow_pgcnt;
2805 if (total > iwmr->page_cnt) {
2809 total = req.sq_pages + req.rq_pages;
2810 use_pbles = (total > 2);
2811 err = irdma_handle_q_mem(iwdev, &req, iwpbl, use_pbles);
2815 ucontext = rdma_udata_to_drv_context(udata, struct irdma_ucontext,
2817 spin_lock_irqsave(&ucontext->qp_reg_mem_list_lock, flags);
2818 list_add_tail(&iwpbl->list, &ucontext->qp_reg_mem_list);
2819 iwpbl->on_list = true;
2820 spin_unlock_irqrestore(&ucontext->qp_reg_mem_list_lock, flags);
2822 case IRDMA_MEMREG_TYPE_CQ:
2823 if (iwdev->rf->sc_dev.hw_attrs.uk_attrs.feature_flags & IRDMA_FEATURE_CQ_RESIZE)
2825 total = req.cq_pages + shadow_pgcnt;
2826 if (total > iwmr->page_cnt) {
2831 use_pbles = (req.cq_pages > 1);
2832 err = irdma_handle_q_mem(iwdev, &req, iwpbl, use_pbles);
2836 ucontext = rdma_udata_to_drv_context(udata, struct irdma_ucontext,
2838 spin_lock_irqsave(&ucontext->cq_reg_mem_list_lock, flags);
2839 list_add_tail(&iwpbl->list, &ucontext->cq_reg_mem_list);
2840 iwpbl->on_list = true;
2841 spin_unlock_irqrestore(&ucontext->cq_reg_mem_list_lock, flags);
2843 case IRDMA_MEMREG_TYPE_MEM:
2844 use_pbles = (iwmr->page_cnt != 1);
2846 err = irdma_setup_pbles(iwdev->rf, iwmr, use_pbles);
2851 ret = irdma_check_mr_contiguous(palloc,
2854 irdma_free_pble(iwdev->rf->pble_rsrc, palloc);
2855 iwpbl->pbl_allocated = false;
2859 stag = irdma_create_stag(iwdev);
2866 iwmr->ibmr.rkey = stag;
2867 iwmr->ibmr.lkey = stag;
2868 err = irdma_hwreg_mr(iwdev, iwmr, access);
2870 irdma_free_stag(iwdev, stag);
2879 iwmr->type = req.reg_type;
2884 if (palloc->level != PBLE_LEVEL_0 && iwpbl->pbl_allocated)
2885 irdma_free_pble(iwdev->rf->pble_rsrc, palloc);
2886 ib_umem_release(region);
2889 return ERR_PTR(err);
2893 * irdma_reg_phys_mr - register kernel physical memory
2895 * @addr: physical address of memory to register
2896 * @size: size of memory to register
2897 * @access: Access rights
2898 * @iova_start: start of virtual address for physical buffers
2900 struct ib_mr *irdma_reg_phys_mr(struct ib_pd *pd, u64 addr, u64 size, int access,
2903 struct irdma_device *iwdev = to_iwdev(pd->device);
2904 struct irdma_pbl *iwpbl;
2905 struct irdma_mr *iwmr;
2906 enum irdma_status_code status;
2910 iwmr = kzalloc(sizeof(*iwmr), GFP_KERNEL);
2912 return ERR_PTR(-ENOMEM);
2915 iwmr->ibmr.device = pd->device;
2916 iwpbl = &iwmr->iwpbl;
2918 iwmr->type = IRDMA_MEMREG_TYPE_MEM;
2919 iwpbl->user_base = *iova_start;
2920 stag = irdma_create_stag(iwdev);
2927 iwmr->ibmr.iova = *iova_start;
2928 iwmr->ibmr.rkey = stag;
2929 iwmr->ibmr.lkey = stag;
2931 iwmr->pgaddrmem[0] = addr;
2933 iwmr->page_size = SZ_4K;
2934 status = irdma_hwreg_mr(iwdev, iwmr, access);
2936 irdma_free_stag(iwdev, stag);
2946 return ERR_PTR(ret);
2950 * irdma_get_dma_mr - register physical mem
2952 * @acc: access for memory
2954 static struct ib_mr *irdma_get_dma_mr(struct ib_pd *pd, int acc)
2958 return irdma_reg_phys_mr(pd, 0, 0, acc, &kva);
2962 * irdma_del_memlist - Deleting pbl list entries for CQ/QP
2963 * @iwmr: iwmr for IB's user page addresses
2964 * @ucontext: ptr to user context
2966 static void irdma_del_memlist(struct irdma_mr *iwmr,
2967 struct irdma_ucontext *ucontext)
2969 struct irdma_pbl *iwpbl = &iwmr->iwpbl;
2970 unsigned long flags;
2972 switch (iwmr->type) {
2973 case IRDMA_MEMREG_TYPE_CQ:
2974 spin_lock_irqsave(&ucontext->cq_reg_mem_list_lock, flags);
2975 if (iwpbl->on_list) {
2976 iwpbl->on_list = false;
2977 list_del(&iwpbl->list);
2979 spin_unlock_irqrestore(&ucontext->cq_reg_mem_list_lock, flags);
2981 case IRDMA_MEMREG_TYPE_QP:
2982 spin_lock_irqsave(&ucontext->qp_reg_mem_list_lock, flags);
2983 if (iwpbl->on_list) {
2984 iwpbl->on_list = false;
2985 list_del(&iwpbl->list);
2987 spin_unlock_irqrestore(&ucontext->qp_reg_mem_list_lock, flags);
2995 * irdma_dereg_mr - deregister mr
2996 * @ib_mr: mr ptr for dereg
2999 static int irdma_dereg_mr(struct ib_mr *ib_mr, struct ib_udata *udata)
3001 struct ib_pd *ibpd = ib_mr->pd;
3002 struct irdma_pd *iwpd = to_iwpd(ibpd);
3003 struct irdma_mr *iwmr = to_iwmr(ib_mr);
3004 struct irdma_device *iwdev = to_iwdev(ib_mr->device);
3005 struct irdma_dealloc_stag_info *info;
3006 struct irdma_pbl *iwpbl = &iwmr->iwpbl;
3007 struct irdma_pble_alloc *palloc = &iwpbl->pble_alloc;
3008 struct irdma_cqp_request *cqp_request;
3009 struct cqp_cmds_info *cqp_info;
3011 if (iwmr->type != IRDMA_MEMREG_TYPE_MEM) {
3013 struct irdma_ucontext *ucontext;
3015 ucontext = rdma_udata_to_drv_context(udata,
3016 struct irdma_ucontext,
3018 irdma_del_memlist(iwmr, ucontext);
3023 cqp_request = irdma_alloc_and_get_cqp_request(&iwdev->rf->cqp, true);
3027 cqp_info = &cqp_request->info;
3028 info = &cqp_info->in.u.dealloc_stag.info;
3029 memset(info, 0, sizeof(*info));
3030 info->pd_id = iwpd->sc_pd.pd_id & 0x00007fff;
3031 info->stag_idx = ib_mr->rkey >> IRDMA_CQPSQ_STAG_IDX_S;
3033 if (iwpbl->pbl_allocated)
3034 info->dealloc_pbl = true;
3036 cqp_info->cqp_cmd = IRDMA_OP_DEALLOC_STAG;
3037 cqp_info->post_sq = 1;
3038 cqp_info->in.u.dealloc_stag.dev = &iwdev->rf->sc_dev;
3039 cqp_info->in.u.dealloc_stag.scratch = (uintptr_t)cqp_request;
3040 irdma_handle_cqp_op(iwdev->rf, cqp_request);
3041 irdma_put_cqp_request(&iwdev->rf->cqp, cqp_request);
3042 irdma_free_stag(iwdev, iwmr->stag);
3044 if (iwpbl->pbl_allocated)
3045 irdma_free_pble(iwdev->rf->pble_rsrc, palloc);
3046 ib_umem_release(iwmr->region);
3053 * irdma_copy_sg_list - copy sg list for qp
3054 * @sg_list: copied into sg_list
3055 * @sgl: copy from sgl
3056 * @num_sges: count of sg entries
3058 static void irdma_copy_sg_list(struct irdma_sge *sg_list, struct ib_sge *sgl,
3063 for (i = 0; (i < num_sges) && (i < IRDMA_MAX_WQ_FRAGMENT_COUNT); i++) {
3064 sg_list[i].tag_off = sgl[i].addr;
3065 sg_list[i].len = sgl[i].length;
3066 sg_list[i].stag = sgl[i].lkey;
3071 * irdma_post_send - kernel application wr
3072 * @ibqp: qp ptr for wr
3073 * @ib_wr: work request ptr
3074 * @bad_wr: return of bad wr if err
3076 static int irdma_post_send(struct ib_qp *ibqp,
3077 const struct ib_send_wr *ib_wr,
3078 const struct ib_send_wr **bad_wr)
3080 struct irdma_qp *iwqp;
3081 struct irdma_qp_uk *ukqp;
3082 struct irdma_sc_dev *dev;
3083 struct irdma_post_sq_info info;
3084 enum irdma_status_code ret;
3086 unsigned long flags;
3088 struct irdma_ah *ah;
3089 bool reflush = false;
3091 iwqp = to_iwqp(ibqp);
3092 ukqp = &iwqp->sc_qp.qp_uk;
3093 dev = &iwqp->iwdev->rf->sc_dev;
3095 spin_lock_irqsave(&iwqp->lock, flags);
3096 if (iwqp->flush_issued && ukqp->sq_flush_complete)
3099 memset(&info, 0, sizeof(info));
3101 info.wr_id = (ib_wr->wr_id);
3102 if ((ib_wr->send_flags & IB_SEND_SIGNALED) || iwqp->sig_all)
3103 info.signaled = true;
3104 if (ib_wr->send_flags & IB_SEND_FENCE)
3105 info.read_fence = true;
3106 switch (ib_wr->opcode) {
3107 case IB_WR_SEND_WITH_IMM:
3108 if (ukqp->qp_caps & IRDMA_SEND_WITH_IMM) {
3109 info.imm_data_valid = true;
3110 info.imm_data = ntohl(ib_wr->ex.imm_data);
3117 case IB_WR_SEND_WITH_INV:
3118 if (ib_wr->opcode == IB_WR_SEND ||
3119 ib_wr->opcode == IB_WR_SEND_WITH_IMM) {
3120 if (ib_wr->send_flags & IB_SEND_SOLICITED)
3121 info.op_type = IRDMA_OP_TYPE_SEND_SOL;
3123 info.op_type = IRDMA_OP_TYPE_SEND;
3125 if (ib_wr->send_flags & IB_SEND_SOLICITED)
3126 info.op_type = IRDMA_OP_TYPE_SEND_SOL_INV;
3128 info.op_type = IRDMA_OP_TYPE_SEND_INV;
3129 info.stag_to_inv = ib_wr->ex.invalidate_rkey;
3132 if (ib_wr->send_flags & IB_SEND_INLINE) {
3133 info.op.inline_send.data = (void *)(unsigned long)
3134 ib_wr->sg_list[0].addr;
3135 info.op.inline_send.len = ib_wr->sg_list[0].length;
3136 if (iwqp->ibqp.qp_type == IB_QPT_UD ||
3137 iwqp->ibqp.qp_type == IB_QPT_GSI) {
3138 ah = to_iwah(ud_wr(ib_wr)->ah);
3139 info.op.inline_send.ah_id = ah->sc_ah.ah_info.ah_idx;
3140 info.op.inline_send.qkey = ud_wr(ib_wr)->remote_qkey;
3141 info.op.inline_send.dest_qp = ud_wr(ib_wr)->remote_qpn;
3143 ret = irdma_uk_inline_send(ukqp, &info, false);
3145 info.op.send.num_sges = ib_wr->num_sge;
3146 info.op.send.sg_list = (struct irdma_sge *)
3148 if (iwqp->ibqp.qp_type == IB_QPT_UD ||
3149 iwqp->ibqp.qp_type == IB_QPT_GSI) {
3150 ah = to_iwah(ud_wr(ib_wr)->ah);
3151 info.op.send.ah_id = ah->sc_ah.ah_info.ah_idx;
3152 info.op.send.qkey = ud_wr(ib_wr)->remote_qkey;
3153 info.op.send.dest_qp = ud_wr(ib_wr)->remote_qpn;
3155 ret = irdma_uk_send(ukqp, &info, false);
3159 if (ret == IRDMA_ERR_QP_TOOMANY_WRS_POSTED)
3165 case IB_WR_RDMA_WRITE_WITH_IMM:
3166 if (ukqp->qp_caps & IRDMA_WRITE_WITH_IMM) {
3167 info.imm_data_valid = true;
3168 info.imm_data = ntohl(ib_wr->ex.imm_data);
3174 case IB_WR_RDMA_WRITE:
3175 if (ib_wr->send_flags & IB_SEND_SOLICITED)
3176 info.op_type = IRDMA_OP_TYPE_RDMA_WRITE_SOL;
3178 info.op_type = IRDMA_OP_TYPE_RDMA_WRITE;
3180 if (ib_wr->send_flags & IB_SEND_INLINE) {
3181 info.op.inline_rdma_write.data = (void *)(uintptr_t)ib_wr->sg_list[0].addr;
3182 info.op.inline_rdma_write.len = ib_wr->sg_list[0].length;
3183 info.op.inline_rdma_write.rem_addr.tag_off = rdma_wr(ib_wr)->remote_addr;
3184 info.op.inline_rdma_write.rem_addr.stag = rdma_wr(ib_wr)->rkey;
3185 ret = irdma_uk_inline_rdma_write(ukqp, &info, false);
3187 info.op.rdma_write.lo_sg_list = (void *)ib_wr->sg_list;
3188 info.op.rdma_write.num_lo_sges = ib_wr->num_sge;
3189 info.op.rdma_write.rem_addr.tag_off = rdma_wr(ib_wr)->remote_addr;
3190 info.op.rdma_write.rem_addr.stag = rdma_wr(ib_wr)->rkey;
3191 ret = irdma_uk_rdma_write(ukqp, &info, false);
3195 if (ret == IRDMA_ERR_QP_TOOMANY_WRS_POSTED)
3201 case IB_WR_RDMA_READ_WITH_INV:
3204 case IB_WR_RDMA_READ:
3205 if (ib_wr->num_sge >
3206 dev->hw_attrs.uk_attrs.max_hw_read_sges) {
3210 info.op_type = IRDMA_OP_TYPE_RDMA_READ;
3211 info.op.rdma_read.rem_addr.tag_off = rdma_wr(ib_wr)->remote_addr;
3212 info.op.rdma_read.rem_addr.stag = rdma_wr(ib_wr)->rkey;
3213 info.op.rdma_read.lo_sg_list = (void *)ib_wr->sg_list;
3214 info.op.rdma_read.num_lo_sges = ib_wr->num_sge;
3216 ret = irdma_uk_rdma_read(ukqp, &info, inv_stag, false);
3218 if (ret == IRDMA_ERR_QP_TOOMANY_WRS_POSTED)
3224 case IB_WR_LOCAL_INV:
3225 info.op_type = IRDMA_OP_TYPE_INV_STAG;
3226 info.op.inv_local_stag.target_stag = ib_wr->ex.invalidate_rkey;
3227 ret = irdma_uk_stag_local_invalidate(ukqp, &info, true);
3231 case IB_WR_REG_MR: {
3232 struct irdma_mr *iwmr = to_iwmr(reg_wr(ib_wr)->mr);
3233 struct irdma_pble_alloc *palloc = &iwmr->iwpbl.pble_alloc;
3234 struct irdma_fast_reg_stag_info stag_info = {};
3236 stag_info.signaled = info.signaled;
3237 stag_info.read_fence = info.read_fence;
3238 stag_info.access_rights = irdma_get_mr_access(reg_wr(ib_wr)->access);
3239 stag_info.stag_key = reg_wr(ib_wr)->key & 0xff;
3240 stag_info.stag_idx = reg_wr(ib_wr)->key >> 8;
3241 stag_info.page_size = reg_wr(ib_wr)->mr->page_size;
3242 stag_info.wr_id = ib_wr->wr_id;
3243 stag_info.addr_type = IRDMA_ADDR_TYPE_VA_BASED;
3244 stag_info.va = (void *)(uintptr_t)iwmr->ibmr.iova;
3245 stag_info.total_len = iwmr->ibmr.length;
3246 stag_info.reg_addr_pa = *palloc->level1.addr;
3247 stag_info.first_pm_pbl_index = palloc->level1.idx;
3248 stag_info.local_fence = ib_wr->send_flags & IB_SEND_FENCE;
3249 if (iwmr->npages > IRDMA_MIN_PAGES_PER_FMR)
3250 stag_info.chunk_size = 1;
3251 ret = irdma_sc_mr_fast_register(&iwqp->sc_qp, &stag_info,
3259 ibdev_dbg(&iwqp->iwdev->ibdev,
3260 "VERBS: upost_send bad opcode = 0x%x\n",
3267 ib_wr = ib_wr->next;
3270 if (!iwqp->flush_issued && iwqp->hw_iwarp_state <= IRDMA_QP_STATE_RTS) {
3271 irdma_uk_qp_post_wr(ukqp);
3272 spin_unlock_irqrestore(&iwqp->lock, flags);
3273 } else if (reflush) {
3274 ukqp->sq_flush_complete = false;
3275 spin_unlock_irqrestore(&iwqp->lock, flags);
3276 irdma_flush_wqes(iwqp, IRDMA_FLUSH_SQ | IRDMA_REFLUSH);
3278 spin_unlock_irqrestore(&iwqp->lock, flags);
3287 * irdma_post_recv - post receive wr for kernel application
3288 * @ibqp: ib qp pointer
3289 * @ib_wr: work request for receive
3290 * @bad_wr: bad wr caused an error
3292 static int irdma_post_recv(struct ib_qp *ibqp,
3293 const struct ib_recv_wr *ib_wr,
3294 const struct ib_recv_wr **bad_wr)
3296 struct irdma_qp *iwqp;
3297 struct irdma_qp_uk *ukqp;
3298 struct irdma_post_rq_info post_recv = {};
3299 struct irdma_sge sg_list[IRDMA_MAX_WQ_FRAGMENT_COUNT];
3300 enum irdma_status_code ret = 0;
3301 unsigned long flags;
3303 bool reflush = false;
3305 iwqp = to_iwqp(ibqp);
3306 ukqp = &iwqp->sc_qp.qp_uk;
3308 spin_lock_irqsave(&iwqp->lock, flags);
3309 if (iwqp->flush_issued && ukqp->rq_flush_complete)
3312 post_recv.num_sges = ib_wr->num_sge;
3313 post_recv.wr_id = ib_wr->wr_id;
3314 irdma_copy_sg_list(sg_list, ib_wr->sg_list, ib_wr->num_sge);
3315 post_recv.sg_list = sg_list;
3316 ret = irdma_uk_post_receive(ukqp, &post_recv);
3318 ibdev_dbg(&iwqp->iwdev->ibdev,
3319 "VERBS: post_recv err %d\n", ret);
3320 if (ret == IRDMA_ERR_QP_TOOMANY_WRS_POSTED)
3327 ib_wr = ib_wr->next;
3332 ukqp->rq_flush_complete = false;
3333 spin_unlock_irqrestore(&iwqp->lock, flags);
3334 irdma_flush_wqes(iwqp, IRDMA_FLUSH_RQ | IRDMA_REFLUSH);
3336 spin_unlock_irqrestore(&iwqp->lock, flags);
3346 * irdma_flush_err_to_ib_wc_status - return change flush error code to IB status
3347 * @opcode: iwarp flush code
3349 static enum ib_wc_status irdma_flush_err_to_ib_wc_status(enum irdma_flush_opcode opcode)
3352 case FLUSH_PROT_ERR:
3353 return IB_WC_LOC_PROT_ERR;
3354 case FLUSH_REM_ACCESS_ERR:
3355 return IB_WC_REM_ACCESS_ERR;
3356 case FLUSH_LOC_QP_OP_ERR:
3357 return IB_WC_LOC_QP_OP_ERR;
3358 case FLUSH_REM_OP_ERR:
3359 return IB_WC_REM_OP_ERR;
3360 case FLUSH_LOC_LEN_ERR:
3361 return IB_WC_LOC_LEN_ERR;
3362 case FLUSH_GENERAL_ERR:
3363 return IB_WC_WR_FLUSH_ERR;
3364 case FLUSH_FATAL_ERR:
3366 return IB_WC_FATAL_ERR;
3371 * irdma_process_cqe - process cqe info
3372 * @entry: processed cqe
3373 * @cq_poll_info: cqe info
3375 static void irdma_process_cqe(struct ib_wc *entry,
3376 struct irdma_cq_poll_info *cq_poll_info)
3378 struct irdma_qp *iwqp;
3379 struct irdma_sc_qp *qp;
3381 entry->wc_flags = 0;
3382 entry->pkey_index = 0;
3383 entry->wr_id = cq_poll_info->wr_id;
3385 qp = cq_poll_info->qp_handle;
3386 iwqp = qp->qp_uk.back_qp;
3387 entry->qp = qp->qp_uk.back_qp;
3389 if (cq_poll_info->error) {
3390 entry->status = (cq_poll_info->comp_status == IRDMA_COMPL_STATUS_FLUSHED) ?
3391 irdma_flush_err_to_ib_wc_status(cq_poll_info->minor_err) : IB_WC_GENERAL_ERR;
3393 entry->vendor_err = cq_poll_info->major_err << 16 |
3394 cq_poll_info->minor_err;
3396 entry->status = IB_WC_SUCCESS;
3397 if (cq_poll_info->imm_valid) {
3398 entry->ex.imm_data = htonl(cq_poll_info->imm_data);
3399 entry->wc_flags |= IB_WC_WITH_IMM;
3401 if (cq_poll_info->ud_smac_valid) {
3402 ether_addr_copy(entry->smac, cq_poll_info->ud_smac);
3403 entry->wc_flags |= IB_WC_WITH_SMAC;
3406 if (cq_poll_info->ud_vlan_valid) {
3407 entry->vlan_id = cq_poll_info->ud_vlan & VLAN_VID_MASK;
3408 entry->wc_flags |= IB_WC_WITH_VLAN;
3409 entry->sl = cq_poll_info->ud_vlan >> VLAN_PRIO_SHIFT;
3415 switch (cq_poll_info->op_type) {
3416 case IRDMA_OP_TYPE_RDMA_WRITE:
3417 case IRDMA_OP_TYPE_RDMA_WRITE_SOL:
3418 entry->opcode = IB_WC_RDMA_WRITE;
3420 case IRDMA_OP_TYPE_RDMA_READ_INV_STAG:
3421 case IRDMA_OP_TYPE_RDMA_READ:
3422 entry->opcode = IB_WC_RDMA_READ;
3424 case IRDMA_OP_TYPE_SEND_INV:
3425 case IRDMA_OP_TYPE_SEND_SOL:
3426 case IRDMA_OP_TYPE_SEND_SOL_INV:
3427 case IRDMA_OP_TYPE_SEND:
3428 entry->opcode = IB_WC_SEND;
3430 case IRDMA_OP_TYPE_FAST_REG_NSMR:
3431 entry->opcode = IB_WC_REG_MR;
3433 case IRDMA_OP_TYPE_INV_STAG:
3434 entry->opcode = IB_WC_LOCAL_INV;
3436 case IRDMA_OP_TYPE_REC_IMM:
3437 case IRDMA_OP_TYPE_REC:
3438 entry->opcode = cq_poll_info->op_type == IRDMA_OP_TYPE_REC_IMM ?
3439 IB_WC_RECV_RDMA_WITH_IMM : IB_WC_RECV;
3440 if (qp->qp_uk.qp_type != IRDMA_QP_TYPE_ROCE_UD &&
3441 cq_poll_info->stag_invalid_set) {
3442 entry->ex.invalidate_rkey = cq_poll_info->inv_stag;
3443 entry->wc_flags |= IB_WC_WITH_INVALIDATE;
3447 ibdev_err(&iwqp->iwdev->ibdev,
3448 "Invalid opcode = %d in CQE\n", cq_poll_info->op_type);
3449 entry->status = IB_WC_GENERAL_ERR;
3453 if (qp->qp_uk.qp_type == IRDMA_QP_TYPE_ROCE_UD) {
3454 entry->src_qp = cq_poll_info->ud_src_qpn;
3457 (IB_WC_GRH | IB_WC_WITH_NETWORK_HDR_TYPE);
3458 entry->network_hdr_type = cq_poll_info->ipv4 ?
3462 entry->src_qp = cq_poll_info->qp_id;
3465 entry->byte_len = cq_poll_info->bytes_xfered;
3469 * irdma_poll_one - poll one entry of the CQ
3470 * @ukcq: ukcq to poll
3471 * @cur_cqe: current CQE info to be filled in
3472 * @entry: ibv_wc object to be filled for non-extended CQ or NULL for extended CQ
3474 * Returns the internal irdma device error code or 0 on success
3476 static inline int irdma_poll_one(struct irdma_cq_uk *ukcq,
3477 struct irdma_cq_poll_info *cur_cqe,
3478 struct ib_wc *entry)
3480 int ret = irdma_uk_cq_poll_cmpl(ukcq, cur_cqe);
3485 irdma_process_cqe(entry, cur_cqe);
3491 * __irdma_poll_cq - poll cq for completion (kernel apps)
3493 * @num_entries: number of entries to poll
3494 * @entry: wr of a completed entry
3496 static int __irdma_poll_cq(struct irdma_cq *iwcq, int num_entries, struct ib_wc *entry)
3498 struct list_head *tmp_node, *list_node;
3499 struct irdma_cq_buf *last_buf = NULL;
3500 struct irdma_cq_poll_info *cur_cqe = &iwcq->cur_cqe;
3501 struct irdma_cq_buf *cq_buf;
3502 enum irdma_status_code ret;
3503 struct irdma_device *iwdev;
3504 struct irdma_cq_uk *ukcq;
3505 bool cq_new_cqe = false;
3506 int resized_bufs = 0;
3509 iwdev = to_iwdev(iwcq->ibcq.device);
3510 ukcq = &iwcq->sc_cq.cq_uk;
3512 /* go through the list of previously resized CQ buffers */
3513 list_for_each_safe(list_node, tmp_node, &iwcq->resize_list) {
3514 cq_buf = container_of(list_node, struct irdma_cq_buf, list);
3515 while (npolled < num_entries) {
3516 ret = irdma_poll_one(&cq_buf->cq_uk, cur_cqe, entry + npolled);
3522 if (ret == IRDMA_ERR_Q_EMPTY)
3524 /* QP using the CQ is destroyed. Skip reporting this CQE */
3525 if (ret == IRDMA_ERR_Q_DESTROYED) {
3532 /* save the resized CQ buffer which received the last cqe */
3538 /* check the current CQ for new cqes */
3539 while (npolled < num_entries) {
3540 ret = irdma_poll_one(ukcq, cur_cqe, entry + npolled);
3547 if (ret == IRDMA_ERR_Q_EMPTY)
3549 /* QP using the CQ is destroyed. Skip reporting this CQE */
3550 if (ret == IRDMA_ERR_Q_DESTROYED) {
3558 /* all previous CQ resizes are complete */
3559 resized_bufs = irdma_process_resize_list(iwcq, iwdev, NULL);
3561 /* only CQ resizes up to the last_buf are complete */
3562 resized_bufs = irdma_process_resize_list(iwcq, iwdev, last_buf);
3564 /* report to the HW the number of complete CQ resizes */
3565 irdma_uk_cq_set_resized_cnt(ukcq, resized_bufs);
3569 ibdev_dbg(&iwdev->ibdev, "%s: Error polling CQ, irdma_err: %d\n",
3576 * irdma_poll_cq - poll cq for completion (kernel apps)
3578 * @num_entries: number of entries to poll
3579 * @entry: wr of a completed entry
3581 static int irdma_poll_cq(struct ib_cq *ibcq, int num_entries,
3582 struct ib_wc *entry)
3584 struct irdma_cq *iwcq;
3585 unsigned long flags;
3588 iwcq = to_iwcq(ibcq);
3590 spin_lock_irqsave(&iwcq->lock, flags);
3591 ret = __irdma_poll_cq(iwcq, num_entries, entry);
3592 spin_unlock_irqrestore(&iwcq->lock, flags);
3598 * irdma_req_notify_cq - arm cq kernel application
3600 * @notify_flags: notofication flags
3602 static int irdma_req_notify_cq(struct ib_cq *ibcq,
3603 enum ib_cq_notify_flags notify_flags)
3605 struct irdma_cq *iwcq;
3606 struct irdma_cq_uk *ukcq;
3607 unsigned long flags;
3608 enum irdma_cmpl_notify cq_notify = IRDMA_CQ_COMPL_EVENT;
3610 iwcq = to_iwcq(ibcq);
3611 ukcq = &iwcq->sc_cq.cq_uk;
3612 if (notify_flags == IB_CQ_SOLICITED)
3613 cq_notify = IRDMA_CQ_COMPL_SOLICITED;
3615 spin_lock_irqsave(&iwcq->lock, flags);
3616 irdma_uk_cq_request_notification(ukcq, cq_notify);
3617 spin_unlock_irqrestore(&iwcq->lock, flags);
3622 static int irdma_roce_port_immutable(struct ib_device *ibdev, u32 port_num,
3623 struct ib_port_immutable *immutable)
3625 struct ib_port_attr attr;
3628 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
3629 err = ib_query_port(ibdev, port_num, &attr);
3633 immutable->max_mad_size = IB_MGMT_MAD_SIZE;
3634 immutable->pkey_tbl_len = attr.pkey_tbl_len;
3635 immutable->gid_tbl_len = attr.gid_tbl_len;
3640 static int irdma_iw_port_immutable(struct ib_device *ibdev, u32 port_num,
3641 struct ib_port_immutable *immutable)
3643 struct ib_port_attr attr;
3646 immutable->core_cap_flags = RDMA_CORE_PORT_IWARP;
3647 err = ib_query_port(ibdev, port_num, &attr);
3650 immutable->gid_tbl_len = attr.gid_tbl_len;
3655 static const char *const irdma_hw_stat_names[] = {
3657 [IRDMA_HW_STAT_INDEX_RXVLANERR] = "rxVlanErrors",
3658 [IRDMA_HW_STAT_INDEX_IP4RXDISCARD] = "ip4InDiscards",
3659 [IRDMA_HW_STAT_INDEX_IP4RXTRUNC] = "ip4InTruncatedPkts",
3660 [IRDMA_HW_STAT_INDEX_IP4TXNOROUTE] = "ip4OutNoRoutes",
3661 [IRDMA_HW_STAT_INDEX_IP6RXDISCARD] = "ip6InDiscards",
3662 [IRDMA_HW_STAT_INDEX_IP6RXTRUNC] = "ip6InTruncatedPkts",
3663 [IRDMA_HW_STAT_INDEX_IP6TXNOROUTE] = "ip6OutNoRoutes",
3664 [IRDMA_HW_STAT_INDEX_TCPRTXSEG] = "tcpRetransSegs",
3665 [IRDMA_HW_STAT_INDEX_TCPRXOPTERR] = "tcpInOptErrors",
3666 [IRDMA_HW_STAT_INDEX_TCPRXPROTOERR] = "tcpInProtoErrors",
3667 [IRDMA_HW_STAT_INDEX_RXRPCNPHANDLED] = "cnpHandled",
3668 [IRDMA_HW_STAT_INDEX_RXRPCNPIGNORED] = "cnpIgnored",
3669 [IRDMA_HW_STAT_INDEX_TXNPCNPSENT] = "cnpSent",
3672 [IRDMA_HW_STAT_INDEX_IP4RXOCTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3674 [IRDMA_HW_STAT_INDEX_IP4RXPKTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3676 [IRDMA_HW_STAT_INDEX_IP4RXFRAGS + IRDMA_HW_STAT_INDEX_MAX_32] =
3678 [IRDMA_HW_STAT_INDEX_IP4RXMCOCTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3680 [IRDMA_HW_STAT_INDEX_IP4RXMCPKTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3682 [IRDMA_HW_STAT_INDEX_IP4TXOCTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3684 [IRDMA_HW_STAT_INDEX_IP4TXPKTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3686 [IRDMA_HW_STAT_INDEX_IP4TXFRAGS + IRDMA_HW_STAT_INDEX_MAX_32] =
3688 [IRDMA_HW_STAT_INDEX_IP4TXMCOCTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3689 "ip4OutMcastOctets",
3690 [IRDMA_HW_STAT_INDEX_IP4TXMCPKTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3692 [IRDMA_HW_STAT_INDEX_IP6RXOCTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3694 [IRDMA_HW_STAT_INDEX_IP6RXPKTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3696 [IRDMA_HW_STAT_INDEX_IP6RXFRAGS + IRDMA_HW_STAT_INDEX_MAX_32] =
3698 [IRDMA_HW_STAT_INDEX_IP6RXMCOCTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3700 [IRDMA_HW_STAT_INDEX_IP6RXMCPKTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3702 [IRDMA_HW_STAT_INDEX_IP6TXOCTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3704 [IRDMA_HW_STAT_INDEX_IP6TXPKTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3706 [IRDMA_HW_STAT_INDEX_IP6TXFRAGS + IRDMA_HW_STAT_INDEX_MAX_32] =
3708 [IRDMA_HW_STAT_INDEX_IP6TXMCOCTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3709 "ip6OutMcastOctets",
3710 [IRDMA_HW_STAT_INDEX_IP6TXMCPKTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3712 [IRDMA_HW_STAT_INDEX_TCPRXSEGS + IRDMA_HW_STAT_INDEX_MAX_32] =
3714 [IRDMA_HW_STAT_INDEX_TCPTXSEG + IRDMA_HW_STAT_INDEX_MAX_32] =
3716 [IRDMA_HW_STAT_INDEX_RDMARXRDS + IRDMA_HW_STAT_INDEX_MAX_32] =
3718 [IRDMA_HW_STAT_INDEX_RDMARXSNDS + IRDMA_HW_STAT_INDEX_MAX_32] =
3720 [IRDMA_HW_STAT_INDEX_RDMARXWRS + IRDMA_HW_STAT_INDEX_MAX_32] =
3722 [IRDMA_HW_STAT_INDEX_RDMATXRDS + IRDMA_HW_STAT_INDEX_MAX_32] =
3724 [IRDMA_HW_STAT_INDEX_RDMATXSNDS + IRDMA_HW_STAT_INDEX_MAX_32] =
3726 [IRDMA_HW_STAT_INDEX_RDMATXWRS + IRDMA_HW_STAT_INDEX_MAX_32] =
3728 [IRDMA_HW_STAT_INDEX_RDMAVBND + IRDMA_HW_STAT_INDEX_MAX_32] =
3730 [IRDMA_HW_STAT_INDEX_RDMAVINV + IRDMA_HW_STAT_INDEX_MAX_32] =
3732 [IRDMA_HW_STAT_INDEX_UDPRXPKTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3734 [IRDMA_HW_STAT_INDEX_UDPTXPKTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3736 [IRDMA_HW_STAT_INDEX_RXNPECNMARKEDPKTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3740 static void irdma_get_dev_fw_str(struct ib_device *dev, char *str)
3742 struct irdma_device *iwdev = to_iwdev(dev);
3744 snprintf(str, IB_FW_VERSION_NAME_MAX, "%u.%u",
3745 irdma_fw_major_ver(&iwdev->rf->sc_dev),
3746 irdma_fw_minor_ver(&iwdev->rf->sc_dev));
3750 * irdma_alloc_hw_port_stats - Allocate a hw stats structure
3751 * @ibdev: device pointer from stack
3752 * @port_num: port number
3754 static struct rdma_hw_stats *irdma_alloc_hw_port_stats(struct ib_device *ibdev,
3757 int num_counters = IRDMA_HW_STAT_INDEX_MAX_32 +
3758 IRDMA_HW_STAT_INDEX_MAX_64;
3759 unsigned long lifespan = RDMA_HW_STATS_DEFAULT_LIFESPAN;
3761 BUILD_BUG_ON(ARRAY_SIZE(irdma_hw_stat_names) !=
3762 (IRDMA_HW_STAT_INDEX_MAX_32 + IRDMA_HW_STAT_INDEX_MAX_64));
3764 return rdma_alloc_hw_stats_struct(irdma_hw_stat_names, num_counters,
3769 * irdma_get_hw_stats - Populates the rdma_hw_stats structure
3770 * @ibdev: device pointer from stack
3771 * @stats: stats pointer from stack
3772 * @port_num: port number
3773 * @index: which hw counter the stack is requesting we update
3775 static int irdma_get_hw_stats(struct ib_device *ibdev,
3776 struct rdma_hw_stats *stats, u32 port_num,
3779 struct irdma_device *iwdev = to_iwdev(ibdev);
3780 struct irdma_dev_hw_stats *hw_stats = &iwdev->vsi.pestat->hw_stats;
3782 if (iwdev->rf->rdma_ver >= IRDMA_GEN_2)
3783 irdma_cqp_gather_stats_cmd(&iwdev->rf->sc_dev, iwdev->vsi.pestat, true);
3785 irdma_cqp_gather_stats_gen1(&iwdev->rf->sc_dev, iwdev->vsi.pestat);
3787 memcpy(&stats->value[0], hw_stats, sizeof(*hw_stats));
3789 return stats->num_counters;
3793 * irdma_query_gid - Query port GID
3794 * @ibdev: device pointer from stack
3795 * @port: port number
3796 * @index: Entry index
3799 static int irdma_query_gid(struct ib_device *ibdev, u32 port, int index,
3802 struct irdma_device *iwdev = to_iwdev(ibdev);
3804 memset(gid->raw, 0, sizeof(gid->raw));
3805 ether_addr_copy(gid->raw, iwdev->netdev->dev_addr);
3811 * mcast_list_add - Add a new mcast item to list
3812 * @rf: RDMA PCI function
3813 * @new_elem: pointer to element to add
3815 static void mcast_list_add(struct irdma_pci_f *rf,
3816 struct mc_table_list *new_elem)
3818 list_add(&new_elem->list, &rf->mc_qht_list.list);
3822 * mcast_list_del - Remove an mcast item from list
3823 * @mc_qht_elem: pointer to mcast table list element
3825 static void mcast_list_del(struct mc_table_list *mc_qht_elem)
3828 list_del(&mc_qht_elem->list);
3832 * mcast_list_lookup_ip - Search mcast list for address
3833 * @rf: RDMA PCI function
3834 * @ip_mcast: pointer to mcast IP address
3836 static struct mc_table_list *mcast_list_lookup_ip(struct irdma_pci_f *rf,
3839 struct mc_table_list *mc_qht_el;
3840 struct list_head *pos, *q;
3842 list_for_each_safe (pos, q, &rf->mc_qht_list.list) {
3843 mc_qht_el = list_entry(pos, struct mc_table_list, list);
3844 if (!memcmp(mc_qht_el->mc_info.dest_ip, ip_mcast,
3845 sizeof(mc_qht_el->mc_info.dest_ip)))
3853 * irdma_mcast_cqp_op - perform a mcast cqp operation
3854 * @iwdev: irdma device
3855 * @mc_grp_ctx: mcast group info
3858 * returns error status
3860 static int irdma_mcast_cqp_op(struct irdma_device *iwdev,
3861 struct irdma_mcast_grp_info *mc_grp_ctx, u8 op)
3863 struct cqp_cmds_info *cqp_info;
3864 struct irdma_cqp_request *cqp_request;
3865 enum irdma_status_code status;
3867 cqp_request = irdma_alloc_and_get_cqp_request(&iwdev->rf->cqp, true);
3871 cqp_request->info.in.u.mc_create.info = *mc_grp_ctx;
3872 cqp_info = &cqp_request->info;
3873 cqp_info->cqp_cmd = op;
3874 cqp_info->post_sq = 1;
3875 cqp_info->in.u.mc_create.scratch = (uintptr_t)cqp_request;
3876 cqp_info->in.u.mc_create.cqp = &iwdev->rf->cqp.sc_cqp;
3877 status = irdma_handle_cqp_op(iwdev->rf, cqp_request);
3878 irdma_put_cqp_request(&iwdev->rf->cqp, cqp_request);
3886 * irdma_mcast_mac - Get the multicast MAC for an IP address
3887 * @ip_addr: IPv4 or IPv6 address
3888 * @mac: pointer to result MAC address
3889 * @ipv4: flag indicating IPv4 or IPv6
3892 void irdma_mcast_mac(u32 *ip_addr, u8 *mac, bool ipv4)
3894 u8 *ip = (u8 *)ip_addr;
3897 unsigned char mac4[ETH_ALEN] = {0x01, 0x00, 0x5E, 0x00,
3900 mac4[3] = ip[2] & 0x7F;
3903 ether_addr_copy(mac, mac4);
3905 unsigned char mac6[ETH_ALEN] = {0x33, 0x33, 0x00, 0x00,
3912 ether_addr_copy(mac, mac6);
3917 * irdma_attach_mcast - attach a qp to a multicast group
3919 * @ibgid: pointer to global ID
3922 * returns error status
3924 static int irdma_attach_mcast(struct ib_qp *ibqp, union ib_gid *ibgid, u16 lid)
3926 struct irdma_qp *iwqp = to_iwqp(ibqp);
3927 struct irdma_device *iwdev = iwqp->iwdev;
3928 struct irdma_pci_f *rf = iwdev->rf;
3929 struct mc_table_list *mc_qht_elem;
3930 struct irdma_mcast_grp_ctx_entry_info mcg_info = {};
3931 unsigned long flags;
3932 u32 ip_addr[4] = {};
3939 struct sockaddr saddr;
3940 struct sockaddr_in saddr_in;
3941 struct sockaddr_in6 saddr_in6;
3943 unsigned char dmac[ETH_ALEN];
3945 rdma_gid2ip((struct sockaddr *)&sgid_addr, ibgid);
3947 if (!ipv6_addr_v4mapped((struct in6_addr *)ibgid)) {
3948 irdma_copy_ip_ntohl(ip_addr,
3949 sgid_addr.saddr_in6.sin6_addr.in6_u.u6_addr32);
3950 irdma_netdev_vlan_ipv6(ip_addr, &vlan_id, NULL);
3952 ibdev_dbg(&iwdev->ibdev,
3953 "VERBS: qp_id=%d, IP6address=%pI6\n", ibqp->qp_num,
3955 irdma_mcast_mac(ip_addr, dmac, false);
3957 ip_addr[0] = ntohl(sgid_addr.saddr_in.sin_addr.s_addr);
3959 vlan_id = irdma_get_vlan_ipv4(ip_addr);
3960 irdma_mcast_mac(ip_addr, dmac, true);
3961 ibdev_dbg(&iwdev->ibdev,
3962 "VERBS: qp_id=%d, IP4address=%pI4, MAC=%pM\n",
3963 ibqp->qp_num, ip_addr, dmac);
3966 spin_lock_irqsave(&rf->qh_list_lock, flags);
3967 mc_qht_elem = mcast_list_lookup_ip(rf, ip_addr);
3969 struct irdma_dma_mem *dma_mem_mc;
3971 spin_unlock_irqrestore(&rf->qh_list_lock, flags);
3972 mc_qht_elem = kzalloc(sizeof(*mc_qht_elem), GFP_KERNEL);
3976 mc_qht_elem->mc_info.ipv4_valid = ipv4;
3977 memcpy(mc_qht_elem->mc_info.dest_ip, ip_addr,
3978 sizeof(mc_qht_elem->mc_info.dest_ip));
3979 ret = irdma_alloc_rsrc(rf, rf->allocated_mcgs, rf->max_mcg,
3980 &mgn, &rf->next_mcg);
3986 mc_qht_elem->mc_info.mgn = mgn;
3987 dma_mem_mc = &mc_qht_elem->mc_grp_ctx.dma_mem_mc;
3988 dma_mem_mc->size = ALIGN(sizeof(u64) * IRDMA_MAX_MGS_PER_CTX,
3989 IRDMA_HW_PAGE_SIZE);
3990 dma_mem_mc->va = dma_alloc_coherent(rf->hw.device,
3994 if (!dma_mem_mc->va) {
3995 irdma_free_rsrc(rf, rf->allocated_mcgs, mgn);
4000 mc_qht_elem->mc_grp_ctx.mg_id = (u16)mgn;
4001 memcpy(mc_qht_elem->mc_grp_ctx.dest_ip_addr, ip_addr,
4002 sizeof(mc_qht_elem->mc_grp_ctx.dest_ip_addr));
4003 mc_qht_elem->mc_grp_ctx.ipv4_valid = ipv4;
4004 mc_qht_elem->mc_grp_ctx.vlan_id = vlan_id;
4005 if (vlan_id < VLAN_N_VID)
4006 mc_qht_elem->mc_grp_ctx.vlan_valid = true;
4007 mc_qht_elem->mc_grp_ctx.hmc_fcn_id = iwdev->vsi.fcn_id;
4008 mc_qht_elem->mc_grp_ctx.qs_handle =
4009 iwqp->sc_qp.vsi->qos[iwqp->sc_qp.user_pri].qs_handle;
4010 ether_addr_copy(mc_qht_elem->mc_grp_ctx.dest_mac_addr, dmac);
4012 spin_lock_irqsave(&rf->qh_list_lock, flags);
4013 mcast_list_add(rf, mc_qht_elem);
4015 if (mc_qht_elem->mc_grp_ctx.no_of_mgs ==
4016 IRDMA_MAX_MGS_PER_CTX) {
4017 spin_unlock_irqrestore(&rf->qh_list_lock, flags);
4022 mcg_info.qp_id = iwqp->ibqp.qp_num;
4023 no_mgs = mc_qht_elem->mc_grp_ctx.no_of_mgs;
4024 irdma_sc_add_mcast_grp(&mc_qht_elem->mc_grp_ctx, &mcg_info);
4025 spin_unlock_irqrestore(&rf->qh_list_lock, flags);
4027 /* Only if there is a change do we need to modify or create */
4029 ret = irdma_mcast_cqp_op(iwdev, &mc_qht_elem->mc_grp_ctx,
4030 IRDMA_OP_MC_CREATE);
4031 } else if (no_mgs != mc_qht_elem->mc_grp_ctx.no_of_mgs) {
4032 ret = irdma_mcast_cqp_op(iwdev, &mc_qht_elem->mc_grp_ctx,
4033 IRDMA_OP_MC_MODIFY);
4044 irdma_sc_del_mcast_grp(&mc_qht_elem->mc_grp_ctx, &mcg_info);
4045 if (!mc_qht_elem->mc_grp_ctx.no_of_mgs) {
4046 mcast_list_del(mc_qht_elem);
4047 dma_free_coherent(rf->hw.device,
4048 mc_qht_elem->mc_grp_ctx.dma_mem_mc.size,
4049 mc_qht_elem->mc_grp_ctx.dma_mem_mc.va,
4050 mc_qht_elem->mc_grp_ctx.dma_mem_mc.pa);
4051 mc_qht_elem->mc_grp_ctx.dma_mem_mc.va = NULL;
4052 irdma_free_rsrc(rf, rf->allocated_mcgs,
4053 mc_qht_elem->mc_grp_ctx.mg_id);
4061 * irdma_detach_mcast - detach a qp from a multicast group
4063 * @ibgid: pointer to global ID
4066 * returns error status
4068 static int irdma_detach_mcast(struct ib_qp *ibqp, union ib_gid *ibgid, u16 lid)
4070 struct irdma_qp *iwqp = to_iwqp(ibqp);
4071 struct irdma_device *iwdev = iwqp->iwdev;
4072 struct irdma_pci_f *rf = iwdev->rf;
4073 u32 ip_addr[4] = {};
4074 struct mc_table_list *mc_qht_elem;
4075 struct irdma_mcast_grp_ctx_entry_info mcg_info = {};
4077 unsigned long flags;
4079 struct sockaddr saddr;
4080 struct sockaddr_in saddr_in;
4081 struct sockaddr_in6 saddr_in6;
4084 rdma_gid2ip((struct sockaddr *)&sgid_addr, ibgid);
4085 if (!ipv6_addr_v4mapped((struct in6_addr *)ibgid))
4086 irdma_copy_ip_ntohl(ip_addr,
4087 sgid_addr.saddr_in6.sin6_addr.in6_u.u6_addr32);
4089 ip_addr[0] = ntohl(sgid_addr.saddr_in.sin_addr.s_addr);
4091 spin_lock_irqsave(&rf->qh_list_lock, flags);
4092 mc_qht_elem = mcast_list_lookup_ip(rf, ip_addr);
4094 spin_unlock_irqrestore(&rf->qh_list_lock, flags);
4095 ibdev_dbg(&iwdev->ibdev,
4096 "VERBS: address not found MCG\n");
4100 mcg_info.qp_id = iwqp->ibqp.qp_num;
4101 irdma_sc_del_mcast_grp(&mc_qht_elem->mc_grp_ctx, &mcg_info);
4102 if (!mc_qht_elem->mc_grp_ctx.no_of_mgs) {
4103 mcast_list_del(mc_qht_elem);
4104 spin_unlock_irqrestore(&rf->qh_list_lock, flags);
4105 ret = irdma_mcast_cqp_op(iwdev, &mc_qht_elem->mc_grp_ctx,
4106 IRDMA_OP_MC_DESTROY);
4108 ibdev_dbg(&iwdev->ibdev,
4109 "VERBS: failed MC_DESTROY MCG\n");
4110 spin_lock_irqsave(&rf->qh_list_lock, flags);
4111 mcast_list_add(rf, mc_qht_elem);
4112 spin_unlock_irqrestore(&rf->qh_list_lock, flags);
4116 dma_free_coherent(rf->hw.device,
4117 mc_qht_elem->mc_grp_ctx.dma_mem_mc.size,
4118 mc_qht_elem->mc_grp_ctx.dma_mem_mc.va,
4119 mc_qht_elem->mc_grp_ctx.dma_mem_mc.pa);
4120 mc_qht_elem->mc_grp_ctx.dma_mem_mc.va = NULL;
4121 irdma_free_rsrc(rf, rf->allocated_mcgs,
4122 mc_qht_elem->mc_grp_ctx.mg_id);
4125 spin_unlock_irqrestore(&rf->qh_list_lock, flags);
4126 ret = irdma_mcast_cqp_op(iwdev, &mc_qht_elem->mc_grp_ctx,
4127 IRDMA_OP_MC_MODIFY);
4129 ibdev_dbg(&iwdev->ibdev,
4130 "VERBS: failed Modify MCG\n");
4139 * irdma_create_ah - create address handle
4140 * @ibah: address handle
4141 * @attr: address handle attributes
4144 * returns 0 on success, error otherwise
4146 static int irdma_create_ah(struct ib_ah *ibah,
4147 struct rdma_ah_init_attr *attr,
4148 struct ib_udata *udata)
4150 struct irdma_pd *pd = to_iwpd(ibah->pd);
4151 struct irdma_ah *ah = container_of(ibah, struct irdma_ah, ibah);
4152 struct rdma_ah_attr *ah_attr = attr->ah_attr;
4153 const struct ib_gid_attr *sgid_attr;
4154 struct irdma_device *iwdev = to_iwdev(ibah->pd->device);
4155 struct irdma_pci_f *rf = iwdev->rf;
4156 struct irdma_sc_ah *sc_ah;
4158 struct irdma_ah_info *ah_info;
4159 struct irdma_create_ah_resp uresp;
4161 struct sockaddr saddr;
4162 struct sockaddr_in saddr_in;
4163 struct sockaddr_in6 saddr_in6;
4164 } sgid_addr, dgid_addr;
4168 err = irdma_alloc_rsrc(rf, rf->allocated_ahs, rf->max_ah, &ah_id,
4175 sc_ah->ah_info.ah_idx = ah_id;
4176 sc_ah->ah_info.vsi = &iwdev->vsi;
4177 irdma_sc_init_ah(&rf->sc_dev, sc_ah);
4178 ah->sgid_index = ah_attr->grh.sgid_index;
4179 sgid_attr = ah_attr->grh.sgid_attr;
4180 memcpy(&ah->dgid, &ah_attr->grh.dgid, sizeof(ah->dgid));
4181 rdma_gid2ip((struct sockaddr *)&sgid_addr, &sgid_attr->gid);
4182 rdma_gid2ip((struct sockaddr *)&dgid_addr, &ah_attr->grh.dgid);
4183 ah->av.attrs = *ah_attr;
4184 ah->av.net_type = rdma_gid_attr_network_type(sgid_attr);
4185 ah->av.sgid_addr.saddr = sgid_addr.saddr;
4186 ah->av.dgid_addr.saddr = dgid_addr.saddr;
4187 ah_info = &sc_ah->ah_info;
4188 ah_info->ah_idx = ah_id;
4189 ah_info->pd_idx = pd->sc_pd.pd_id;
4190 if (ah_attr->ah_flags & IB_AH_GRH) {
4191 ah_info->flow_label = ah_attr->grh.flow_label;
4192 ah_info->hop_ttl = ah_attr->grh.hop_limit;
4193 ah_info->tc_tos = ah_attr->grh.traffic_class;
4196 ether_addr_copy(dmac, ah_attr->roce.dmac);
4197 if (rdma_gid_attr_network_type(sgid_attr) == RDMA_NETWORK_IPV4) {
4198 ah_info->ipv4_valid = true;
4199 ah_info->dest_ip_addr[0] =
4200 ntohl(dgid_addr.saddr_in.sin_addr.s_addr);
4201 ah_info->src_ip_addr[0] =
4202 ntohl(sgid_addr.saddr_in.sin_addr.s_addr);
4203 ah_info->do_lpbk = irdma_ipv4_is_lpb(ah_info->src_ip_addr[0],
4204 ah_info->dest_ip_addr[0]);
4205 if (ipv4_is_multicast(dgid_addr.saddr_in.sin_addr.s_addr)) {
4206 ah_info->do_lpbk = true;
4207 irdma_mcast_mac(ah_info->dest_ip_addr, dmac, true);
4210 irdma_copy_ip_ntohl(ah_info->dest_ip_addr,
4211 dgid_addr.saddr_in6.sin6_addr.in6_u.u6_addr32);
4212 irdma_copy_ip_ntohl(ah_info->src_ip_addr,
4213 sgid_addr.saddr_in6.sin6_addr.in6_u.u6_addr32);
4214 ah_info->do_lpbk = irdma_ipv6_is_lpb(ah_info->src_ip_addr,
4215 ah_info->dest_ip_addr);
4216 if (rdma_is_multicast_addr(&dgid_addr.saddr_in6.sin6_addr)) {
4217 ah_info->do_lpbk = true;
4218 irdma_mcast_mac(ah_info->dest_ip_addr, dmac, false);
4222 err = rdma_read_gid_l2_fields(sgid_attr, &ah_info->vlan_tag,
4227 ah_info->dst_arpindex = irdma_add_arp(iwdev->rf, ah_info->dest_ip_addr,
4228 ah_info->ipv4_valid, dmac);
4230 if (ah_info->dst_arpindex == -1) {
4235 if (ah_info->vlan_tag >= VLAN_N_VID && iwdev->dcb)
4236 ah_info->vlan_tag = 0;
4238 if (ah_info->vlan_tag < VLAN_N_VID) {
4239 ah_info->insert_vlan_tag = true;
4240 ah_info->vlan_tag |=
4241 rt_tos2priority(ah_info->tc_tos) << VLAN_PRIO_SHIFT;
4244 err = irdma_ah_cqp_op(iwdev->rf, sc_ah, IRDMA_OP_AH_CREATE,
4245 attr->flags & RDMA_CREATE_AH_SLEEPABLE,
4246 irdma_gsi_ud_qp_ah_cb, sc_ah);
4249 ibdev_dbg(&iwdev->ibdev,
4250 "VERBS: CQP-OP Create AH fail");
4254 if (!(attr->flags & RDMA_CREATE_AH_SLEEPABLE)) {
4255 int cnt = CQP_COMPL_WAIT_TIME_MS * CQP_TIMEOUT_THRESHOLD;
4258 irdma_cqp_ce_handler(rf, &rf->ccq.sc_cq);
4260 } while (!sc_ah->ah_info.ah_valid && --cnt);
4263 ibdev_dbg(&iwdev->ibdev,
4264 "VERBS: CQP create AH timed out");
4271 uresp.ah_id = ah->sc_ah.ah_info.ah_idx;
4272 err = ib_copy_to_udata(udata, &uresp,
4273 min(sizeof(uresp), udata->outlen));
4278 irdma_free_rsrc(iwdev->rf, iwdev->rf->allocated_ahs, ah_id);
4284 * irdma_destroy_ah - Destroy address handle
4285 * @ibah: pointer to address handle
4286 * @ah_flags: flags for sleepable
4288 static int irdma_destroy_ah(struct ib_ah *ibah, u32 ah_flags)
4290 struct irdma_device *iwdev = to_iwdev(ibah->device);
4291 struct irdma_ah *ah = to_iwah(ibah);
4293 irdma_ah_cqp_op(iwdev->rf, &ah->sc_ah, IRDMA_OP_AH_DESTROY,
4296 irdma_free_rsrc(iwdev->rf, iwdev->rf->allocated_ahs,
4297 ah->sc_ah.ah_info.ah_idx);
4303 * irdma_query_ah - Query address handle
4304 * @ibah: pointer to address handle
4305 * @ah_attr: address handle attributes
4307 static int irdma_query_ah(struct ib_ah *ibah, struct rdma_ah_attr *ah_attr)
4309 struct irdma_ah *ah = to_iwah(ibah);
4311 memset(ah_attr, 0, sizeof(*ah_attr));
4312 if (ah->av.attrs.ah_flags & IB_AH_GRH) {
4313 ah_attr->ah_flags = IB_AH_GRH;
4314 ah_attr->grh.flow_label = ah->sc_ah.ah_info.flow_label;
4315 ah_attr->grh.traffic_class = ah->sc_ah.ah_info.tc_tos;
4316 ah_attr->grh.hop_limit = ah->sc_ah.ah_info.hop_ttl;
4317 ah_attr->grh.sgid_index = ah->sgid_index;
4318 ah_attr->grh.sgid_index = ah->sgid_index;
4319 memcpy(&ah_attr->grh.dgid, &ah->dgid,
4320 sizeof(ah_attr->grh.dgid));
4326 static enum rdma_link_layer irdma_get_link_layer(struct ib_device *ibdev,
4329 return IB_LINK_LAYER_ETHERNET;
4332 static __be64 irdma_mac_to_guid(struct net_device *ndev)
4334 unsigned char *mac = ndev->dev_addr;
4336 unsigned char *dst = (unsigned char *)&guid;
4338 dst[0] = mac[0] ^ 2;
4350 static const struct ib_device_ops irdma_roce_dev_ops = {
4351 .attach_mcast = irdma_attach_mcast,
4352 .create_ah = irdma_create_ah,
4353 .create_user_ah = irdma_create_ah,
4354 .destroy_ah = irdma_destroy_ah,
4355 .detach_mcast = irdma_detach_mcast,
4356 .get_link_layer = irdma_get_link_layer,
4357 .get_port_immutable = irdma_roce_port_immutable,
4358 .modify_qp = irdma_modify_qp_roce,
4359 .query_ah = irdma_query_ah,
4360 .query_pkey = irdma_query_pkey,
4363 static const struct ib_device_ops irdma_iw_dev_ops = {
4364 .modify_qp = irdma_modify_qp,
4365 .get_port_immutable = irdma_iw_port_immutable,
4366 .query_gid = irdma_query_gid,
4369 static const struct ib_device_ops irdma_dev_ops = {
4370 .owner = THIS_MODULE,
4371 .driver_id = RDMA_DRIVER_IRDMA,
4372 .uverbs_abi_ver = IRDMA_ABI_VER,
4374 .alloc_hw_port_stats = irdma_alloc_hw_port_stats,
4375 .alloc_mr = irdma_alloc_mr,
4376 .alloc_mw = irdma_alloc_mw,
4377 .alloc_pd = irdma_alloc_pd,
4378 .alloc_ucontext = irdma_alloc_ucontext,
4379 .create_cq = irdma_create_cq,
4380 .create_qp = irdma_create_qp,
4381 .dealloc_driver = irdma_ib_dealloc_device,
4382 .dealloc_mw = irdma_dealloc_mw,
4383 .dealloc_pd = irdma_dealloc_pd,
4384 .dealloc_ucontext = irdma_dealloc_ucontext,
4385 .dereg_mr = irdma_dereg_mr,
4386 .destroy_cq = irdma_destroy_cq,
4387 .destroy_qp = irdma_destroy_qp,
4388 .disassociate_ucontext = irdma_disassociate_ucontext,
4389 .get_dev_fw_str = irdma_get_dev_fw_str,
4390 .get_dma_mr = irdma_get_dma_mr,
4391 .get_hw_stats = irdma_get_hw_stats,
4392 .map_mr_sg = irdma_map_mr_sg,
4394 .mmap_free = irdma_mmap_free,
4395 .poll_cq = irdma_poll_cq,
4396 .post_recv = irdma_post_recv,
4397 .post_send = irdma_post_send,
4398 .query_device = irdma_query_device,
4399 .query_port = irdma_query_port,
4400 .query_qp = irdma_query_qp,
4401 .reg_user_mr = irdma_reg_user_mr,
4402 .req_notify_cq = irdma_req_notify_cq,
4403 .resize_cq = irdma_resize_cq,
4404 INIT_RDMA_OBJ_SIZE(ib_pd, irdma_pd, ibpd),
4405 INIT_RDMA_OBJ_SIZE(ib_ucontext, irdma_ucontext, ibucontext),
4406 INIT_RDMA_OBJ_SIZE(ib_ah, irdma_ah, ibah),
4407 INIT_RDMA_OBJ_SIZE(ib_cq, irdma_cq, ibcq),
4408 INIT_RDMA_OBJ_SIZE(ib_mw, irdma_mr, ibmw),
4412 * irdma_init_roce_device - initialization of roce rdma device
4413 * @iwdev: irdma device
4415 static void irdma_init_roce_device(struct irdma_device *iwdev)
4417 iwdev->ibdev.node_type = RDMA_NODE_IB_CA;
4418 iwdev->ibdev.node_guid = irdma_mac_to_guid(iwdev->netdev);
4419 ib_set_device_ops(&iwdev->ibdev, &irdma_roce_dev_ops);
4423 * irdma_init_iw_device - initialization of iwarp rdma device
4424 * @iwdev: irdma device
4426 static int irdma_init_iw_device(struct irdma_device *iwdev)
4428 struct net_device *netdev = iwdev->netdev;
4430 iwdev->ibdev.node_type = RDMA_NODE_RNIC;
4431 ether_addr_copy((u8 *)&iwdev->ibdev.node_guid, netdev->dev_addr);
4432 iwdev->ibdev.ops.iw_add_ref = irdma_qp_add_ref;
4433 iwdev->ibdev.ops.iw_rem_ref = irdma_qp_rem_ref;
4434 iwdev->ibdev.ops.iw_get_qp = irdma_get_qp;
4435 iwdev->ibdev.ops.iw_connect = irdma_connect;
4436 iwdev->ibdev.ops.iw_accept = irdma_accept;
4437 iwdev->ibdev.ops.iw_reject = irdma_reject;
4438 iwdev->ibdev.ops.iw_create_listen = irdma_create_listen;
4439 iwdev->ibdev.ops.iw_destroy_listen = irdma_destroy_listen;
4440 memcpy(iwdev->ibdev.iw_ifname, netdev->name,
4441 sizeof(iwdev->ibdev.iw_ifname));
4442 ib_set_device_ops(&iwdev->ibdev, &irdma_iw_dev_ops);
4448 * irdma_init_rdma_device - initialization of rdma device
4449 * @iwdev: irdma device
4451 static int irdma_init_rdma_device(struct irdma_device *iwdev)
4453 struct pci_dev *pcidev = iwdev->rf->pcidev;
4456 if (iwdev->roce_mode) {
4457 irdma_init_roce_device(iwdev);
4459 ret = irdma_init_iw_device(iwdev);
4463 iwdev->ibdev.phys_port_cnt = 1;
4464 iwdev->ibdev.num_comp_vectors = iwdev->rf->ceqs_count;
4465 iwdev->ibdev.dev.parent = &pcidev->dev;
4466 ib_set_device_ops(&iwdev->ibdev, &irdma_dev_ops);
4472 * irdma_port_ibevent - indicate port event
4473 * @iwdev: irdma device
4475 void irdma_port_ibevent(struct irdma_device *iwdev)
4477 struct ib_event event;
4479 event.device = &iwdev->ibdev;
4480 event.element.port_num = 1;
4482 iwdev->iw_status ? IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
4483 ib_dispatch_event(&event);
4487 * irdma_ib_unregister_device - unregister rdma device from IB
4489 * @iwdev: irdma device
4491 void irdma_ib_unregister_device(struct irdma_device *iwdev)
4493 iwdev->iw_status = 0;
4494 irdma_port_ibevent(iwdev);
4495 ib_unregister_device(&iwdev->ibdev);
4499 * irdma_ib_register_device - register irdma device to IB core
4500 * @iwdev: irdma device
4502 int irdma_ib_register_device(struct irdma_device *iwdev)
4506 ret = irdma_init_rdma_device(iwdev);
4510 ret = ib_device_set_netdev(&iwdev->ibdev, iwdev->netdev, 1);
4513 dma_set_max_seg_size(iwdev->rf->hw.device, UINT_MAX);
4514 ret = ib_register_device(&iwdev->ibdev, "irdma%d", iwdev->rf->hw.device);
4518 iwdev->iw_status = 1;
4519 irdma_port_ibevent(iwdev);
4525 ibdev_dbg(&iwdev->ibdev, "VERBS: Register RDMA device fail\n");
4531 * irdma_ib_dealloc_device
4534 * callback from ibdev dealloc_driver to deallocate resources
4535 * unber irdma device
4537 void irdma_ib_dealloc_device(struct ib_device *ibdev)
4539 struct irdma_device *iwdev = to_iwdev(ibdev);
4541 irdma_rt_deinit_hw(iwdev);
4542 irdma_ctrl_deinit_hw(iwdev->rf);