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 void 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;
580 * irdma_setup_kmode_qp - setup initialization for kernel mode qp
581 * @iwdev: iwarp device
582 * @iwqp: qp ptr (user or kernel)
583 * @info: initialize info to return
584 * @init_attr: Initial QP create attributes
586 static int irdma_setup_kmode_qp(struct irdma_device *iwdev,
587 struct irdma_qp *iwqp,
588 struct irdma_qp_init_info *info,
589 struct ib_qp_init_attr *init_attr)
591 struct irdma_dma_mem *mem = &iwqp->kqp.dma_mem;
592 u32 sqdepth, rqdepth;
595 enum irdma_status_code status;
596 struct irdma_qp_uk_init_info *ukinfo = &info->qp_uk_init_info;
597 struct irdma_uk_attrs *uk_attrs = &iwdev->rf->sc_dev.hw_attrs.uk_attrs;
599 irdma_get_wqe_shift(uk_attrs,
600 uk_attrs->hw_rev >= IRDMA_GEN_2 ? ukinfo->max_sq_frag_cnt + 1 :
601 ukinfo->max_sq_frag_cnt,
602 ukinfo->max_inline_data, &sqshift);
603 status = irdma_get_sqdepth(uk_attrs, ukinfo->sq_size, sqshift,
608 if (uk_attrs->hw_rev == IRDMA_GEN_1)
609 rqshift = IRDMA_MAX_RQ_WQE_SHIFT_GEN1;
611 irdma_get_wqe_shift(uk_attrs, ukinfo->max_rq_frag_cnt, 0,
614 status = irdma_get_rqdepth(uk_attrs, ukinfo->rq_size, rqshift,
619 iwqp->kqp.sq_wrid_mem =
620 kcalloc(sqdepth, sizeof(*iwqp->kqp.sq_wrid_mem), GFP_KERNEL);
621 if (!iwqp->kqp.sq_wrid_mem)
624 iwqp->kqp.rq_wrid_mem =
625 kcalloc(rqdepth, sizeof(*iwqp->kqp.rq_wrid_mem), GFP_KERNEL);
626 if (!iwqp->kqp.rq_wrid_mem) {
627 kfree(iwqp->kqp.sq_wrid_mem);
628 iwqp->kqp.sq_wrid_mem = NULL;
632 ukinfo->sq_wrtrk_array = iwqp->kqp.sq_wrid_mem;
633 ukinfo->rq_wrid_array = iwqp->kqp.rq_wrid_mem;
635 size = (sqdepth + rqdepth) * IRDMA_QP_WQE_MIN_SIZE;
636 size += (IRDMA_SHADOW_AREA_SIZE << 3);
638 mem->size = ALIGN(size, 256);
639 mem->va = dma_alloc_coherent(iwdev->rf->hw.device, mem->size,
640 &mem->pa, GFP_KERNEL);
642 kfree(iwqp->kqp.sq_wrid_mem);
643 iwqp->kqp.sq_wrid_mem = NULL;
644 kfree(iwqp->kqp.rq_wrid_mem);
645 iwqp->kqp.rq_wrid_mem = NULL;
649 ukinfo->sq = mem->va;
650 info->sq_pa = mem->pa;
651 ukinfo->rq = &ukinfo->sq[sqdepth];
652 info->rq_pa = info->sq_pa + (sqdepth * IRDMA_QP_WQE_MIN_SIZE);
653 ukinfo->shadow_area = ukinfo->rq[rqdepth].elem;
654 info->shadow_area_pa = info->rq_pa + (rqdepth * IRDMA_QP_WQE_MIN_SIZE);
655 ukinfo->sq_size = sqdepth >> sqshift;
656 ukinfo->rq_size = rqdepth >> rqshift;
657 ukinfo->qp_id = iwqp->ibqp.qp_num;
659 init_attr->cap.max_send_wr = (sqdepth - IRDMA_SQ_RSVD) >> sqshift;
660 init_attr->cap.max_recv_wr = (rqdepth - IRDMA_RQ_RSVD) >> rqshift;
665 static int irdma_cqp_create_qp_cmd(struct irdma_qp *iwqp)
667 struct irdma_pci_f *rf = iwqp->iwdev->rf;
668 struct irdma_cqp_request *cqp_request;
669 struct cqp_cmds_info *cqp_info;
670 struct irdma_create_qp_info *qp_info;
671 enum irdma_status_code status;
673 cqp_request = irdma_alloc_and_get_cqp_request(&rf->cqp, true);
677 cqp_info = &cqp_request->info;
678 qp_info = &cqp_request->info.in.u.qp_create.info;
679 memset(qp_info, 0, sizeof(*qp_info));
680 qp_info->mac_valid = true;
681 qp_info->cq_num_valid = true;
682 qp_info->next_iwarp_state = IRDMA_QP_STATE_IDLE;
684 cqp_info->cqp_cmd = IRDMA_OP_QP_CREATE;
685 cqp_info->post_sq = 1;
686 cqp_info->in.u.qp_create.qp = &iwqp->sc_qp;
687 cqp_info->in.u.qp_create.scratch = (uintptr_t)cqp_request;
688 status = irdma_handle_cqp_op(rf, cqp_request);
689 irdma_put_cqp_request(&rf->cqp, cqp_request);
691 return status ? -ENOMEM : 0;
694 static void irdma_roce_fill_and_set_qpctx_info(struct irdma_qp *iwqp,
695 struct irdma_qp_host_ctx_info *ctx_info)
697 struct irdma_device *iwdev = iwqp->iwdev;
698 struct irdma_sc_dev *dev = &iwdev->rf->sc_dev;
699 struct irdma_roce_offload_info *roce_info;
700 struct irdma_udp_offload_info *udp_info;
702 udp_info = &iwqp->udp_info;
703 udp_info->snd_mss = ib_mtu_enum_to_int(ib_mtu_int_to_enum(iwdev->vsi.mtu));
704 udp_info->cwnd = iwdev->roce_cwnd;
705 udp_info->rexmit_thresh = 2;
706 udp_info->rnr_nak_thresh = 2;
707 udp_info->src_port = 0xc000;
708 udp_info->dst_port = ROCE_V2_UDP_DPORT;
709 roce_info = &iwqp->roce_info;
710 ether_addr_copy(roce_info->mac_addr, iwdev->netdev->dev_addr);
712 roce_info->rd_en = true;
713 roce_info->wr_rdresp_en = true;
714 roce_info->bind_en = true;
715 roce_info->dcqcn_en = false;
716 roce_info->rtomin = 5;
718 roce_info->ack_credits = iwdev->roce_ackcreds;
719 roce_info->ird_size = dev->hw_attrs.max_hw_ird;
720 roce_info->ord_size = dev->hw_attrs.max_hw_ord;
722 if (!iwqp->user_mode) {
723 roce_info->priv_mode_en = true;
724 roce_info->fast_reg_en = true;
725 roce_info->udprivcq_en = true;
727 roce_info->roce_tver = 0;
729 ctx_info->roce_info = &iwqp->roce_info;
730 ctx_info->udp_info = &iwqp->udp_info;
731 irdma_sc_qp_setctx_roce(&iwqp->sc_qp, iwqp->host_ctx.va, ctx_info);
734 static void irdma_iw_fill_and_set_qpctx_info(struct irdma_qp *iwqp,
735 struct irdma_qp_host_ctx_info *ctx_info)
737 struct irdma_device *iwdev = iwqp->iwdev;
738 struct irdma_sc_dev *dev = &iwdev->rf->sc_dev;
739 struct irdma_iwarp_offload_info *iwarp_info;
741 iwarp_info = &iwqp->iwarp_info;
742 ether_addr_copy(iwarp_info->mac_addr, iwdev->netdev->dev_addr);
743 iwarp_info->rd_en = true;
744 iwarp_info->wr_rdresp_en = true;
745 iwarp_info->bind_en = true;
746 iwarp_info->ecn_en = true;
747 iwarp_info->rtomin = 5;
749 if (dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2)
750 iwarp_info->ib_rd_en = true;
751 if (!iwqp->user_mode) {
752 iwarp_info->priv_mode_en = true;
753 iwarp_info->fast_reg_en = true;
755 iwarp_info->ddp_ver = 1;
756 iwarp_info->rdmap_ver = 1;
758 ctx_info->iwarp_info = &iwqp->iwarp_info;
759 ctx_info->iwarp_info_valid = true;
760 irdma_sc_qp_setctx(&iwqp->sc_qp, iwqp->host_ctx.va, ctx_info);
761 ctx_info->iwarp_info_valid = false;
764 static int irdma_validate_qp_attrs(struct ib_qp_init_attr *init_attr,
765 struct irdma_device *iwdev)
767 struct irdma_sc_dev *dev = &iwdev->rf->sc_dev;
768 struct irdma_uk_attrs *uk_attrs = &dev->hw_attrs.uk_attrs;
770 if (init_attr->create_flags)
773 if (init_attr->cap.max_inline_data > uk_attrs->max_hw_inline ||
774 init_attr->cap.max_send_sge > uk_attrs->max_hw_wq_frags ||
775 init_attr->cap.max_recv_sge > uk_attrs->max_hw_wq_frags)
778 if (rdma_protocol_roce(&iwdev->ibdev, 1)) {
779 if (init_attr->qp_type != IB_QPT_RC &&
780 init_attr->qp_type != IB_QPT_UD &&
781 init_attr->qp_type != IB_QPT_GSI)
784 if (init_attr->qp_type != IB_QPT_RC)
792 * irdma_create_qp - create qp
794 * @init_attr: attributes for qp
795 * @udata: user data for create qp
797 static struct ib_qp *irdma_create_qp(struct ib_pd *ibpd,
798 struct ib_qp_init_attr *init_attr,
799 struct ib_udata *udata)
801 struct irdma_pd *iwpd = to_iwpd(ibpd);
802 struct irdma_device *iwdev = to_iwdev(ibpd->device);
803 struct irdma_pci_f *rf = iwdev->rf;
804 struct irdma_qp *iwqp;
805 struct irdma_create_qp_req req;
806 struct irdma_create_qp_resp uresp = {};
808 enum irdma_status_code ret;
812 struct irdma_sc_qp *qp;
813 struct irdma_sc_dev *dev = &rf->sc_dev;
814 struct irdma_uk_attrs *uk_attrs = &dev->hw_attrs.uk_attrs;
815 struct irdma_qp_init_info init_info = {};
816 struct irdma_qp_host_ctx_info *ctx_info;
819 err_code = irdma_validate_qp_attrs(init_attr, iwdev);
821 return ERR_PTR(err_code);
823 sq_size = init_attr->cap.max_send_wr;
824 rq_size = init_attr->cap.max_recv_wr;
826 init_info.vsi = &iwdev->vsi;
827 init_info.qp_uk_init_info.uk_attrs = uk_attrs;
828 init_info.qp_uk_init_info.sq_size = sq_size;
829 init_info.qp_uk_init_info.rq_size = rq_size;
830 init_info.qp_uk_init_info.max_sq_frag_cnt = init_attr->cap.max_send_sge;
831 init_info.qp_uk_init_info.max_rq_frag_cnt = init_attr->cap.max_recv_sge;
832 init_info.qp_uk_init_info.max_inline_data = init_attr->cap.max_inline_data;
834 iwqp = kzalloc(sizeof(*iwqp), GFP_KERNEL);
836 return ERR_PTR(-ENOMEM);
839 qp->qp_uk.back_qp = iwqp;
840 qp->qp_uk.lock = &iwqp->lock;
841 qp->push_idx = IRDMA_INVALID_PUSH_PAGE_INDEX;
844 iwqp->q2_ctx_mem.size = ALIGN(IRDMA_Q2_BUF_SIZE + IRDMA_QP_CTX_SIZE,
846 iwqp->q2_ctx_mem.va = dma_alloc_coherent(dev->hw->device,
847 iwqp->q2_ctx_mem.size,
848 &iwqp->q2_ctx_mem.pa,
850 if (!iwqp->q2_ctx_mem.va) {
855 init_info.q2 = iwqp->q2_ctx_mem.va;
856 init_info.q2_pa = iwqp->q2_ctx_mem.pa;
857 init_info.host_ctx = (__le64 *)(init_info.q2 + IRDMA_Q2_BUF_SIZE);
858 init_info.host_ctx_pa = init_info.q2_pa + IRDMA_Q2_BUF_SIZE;
860 if (init_attr->qp_type == IB_QPT_GSI)
863 err_code = irdma_alloc_rsrc(rf, rf->allocated_qps, rf->max_qp,
864 &qp_num, &rf->next_qp);
869 iwqp->ibqp.qp_num = qp_num;
871 iwqp->iwscq = to_iwcq(init_attr->send_cq);
872 iwqp->iwrcq = to_iwcq(init_attr->recv_cq);
873 iwqp->host_ctx.va = init_info.host_ctx;
874 iwqp->host_ctx.pa = init_info.host_ctx_pa;
875 iwqp->host_ctx.size = IRDMA_QP_CTX_SIZE;
877 init_info.pd = &iwpd->sc_pd;
878 init_info.qp_uk_init_info.qp_id = iwqp->ibqp.qp_num;
879 if (!rdma_protocol_roce(&iwdev->ibdev, 1))
880 init_info.qp_uk_init_info.first_sq_wq = 1;
881 iwqp->ctx_info.qp_compl_ctx = (uintptr_t)qp;
882 init_waitqueue_head(&iwqp->waitq);
883 init_waitqueue_head(&iwqp->mod_qp_waitq);
886 err_code = ib_copy_from_udata(&req, udata,
887 min(sizeof(req), udata->inlen));
889 ibdev_dbg(&iwdev->ibdev,
890 "VERBS: ib_copy_from_data fail\n");
894 iwqp->ctx_info.qp_compl_ctx = req.user_compl_ctx;
896 if (req.user_wqe_bufs) {
897 struct irdma_ucontext *ucontext =
898 rdma_udata_to_drv_context(udata,
899 struct irdma_ucontext,
902 init_info.qp_uk_init_info.legacy_mode = ucontext->legacy_mode;
903 spin_lock_irqsave(&ucontext->qp_reg_mem_list_lock, flags);
904 iwqp->iwpbl = irdma_get_pbl((unsigned long)req.user_wqe_bufs,
905 &ucontext->qp_reg_mem_list);
906 spin_unlock_irqrestore(&ucontext->qp_reg_mem_list_lock, flags);
910 ibdev_dbg(&iwdev->ibdev, "VERBS: no pbl info\n");
914 init_info.qp_uk_init_info.abi_ver = iwpd->sc_pd.abi_ver;
915 irdma_setup_virt_qp(iwdev, iwqp, &init_info);
917 init_info.qp_uk_init_info.abi_ver = IRDMA_ABI_VER;
918 err_code = irdma_setup_kmode_qp(iwdev, iwqp, &init_info, init_attr);
922 ibdev_dbg(&iwdev->ibdev, "VERBS: setup qp failed\n");
926 if (rdma_protocol_roce(&iwdev->ibdev, 1)) {
927 if (init_attr->qp_type == IB_QPT_RC) {
928 init_info.qp_uk_init_info.type = IRDMA_QP_TYPE_ROCE_RC;
929 init_info.qp_uk_init_info.qp_caps = IRDMA_SEND_WITH_IMM |
930 IRDMA_WRITE_WITH_IMM |
933 init_info.qp_uk_init_info.type = IRDMA_QP_TYPE_ROCE_UD;
934 init_info.qp_uk_init_info.qp_caps = IRDMA_SEND_WITH_IMM |
938 init_info.qp_uk_init_info.type = IRDMA_QP_TYPE_IWARP;
939 init_info.qp_uk_init_info.qp_caps = IRDMA_WRITE_WITH_IMM;
942 if (dev->hw_attrs.uk_attrs.hw_rev > IRDMA_GEN_1)
943 init_info.qp_uk_init_info.qp_caps |= IRDMA_PUSH_MODE;
945 ret = irdma_sc_qp_init(qp, &init_info);
948 ibdev_dbg(&iwdev->ibdev, "VERBS: qp_init fail\n");
952 ctx_info = &iwqp->ctx_info;
953 ctx_info->send_cq_num = iwqp->iwscq->sc_cq.cq_uk.cq_id;
954 ctx_info->rcv_cq_num = iwqp->iwrcq->sc_cq.cq_uk.cq_id;
956 if (rdma_protocol_roce(&iwdev->ibdev, 1))
957 irdma_roce_fill_and_set_qpctx_info(iwqp, ctx_info);
959 irdma_iw_fill_and_set_qpctx_info(iwqp, ctx_info);
961 err_code = irdma_cqp_create_qp_cmd(iwqp);
965 refcount_set(&iwqp->refcnt, 1);
966 spin_lock_init(&iwqp->lock);
967 spin_lock_init(&iwqp->sc_qp.pfpdu.lock);
968 iwqp->sig_all = (init_attr->sq_sig_type == IB_SIGNAL_ALL_WR) ? 1 : 0;
969 rf->qp_table[qp_num] = iwqp;
970 iwqp->max_send_wr = sq_size;
971 iwqp->max_recv_wr = rq_size;
973 if (rdma_protocol_roce(&iwdev->ibdev, 1)) {
974 if (dev->ws_add(&iwdev->vsi, 0)) {
975 irdma_cqp_qp_destroy_cmd(&rf->sc_dev, &iwqp->sc_qp);
980 irdma_qp_add_qos(&iwqp->sc_qp);
984 /* GEN_1 legacy support with libi40iw does not have expanded uresp struct */
985 if (udata->outlen < sizeof(uresp)) {
987 uresp.push_idx = IRDMA_INVALID_PUSH_PAGE_INDEX_GEN_1;
989 if (rdma_protocol_iwarp(&iwdev->ibdev, 1))
992 uresp.actual_sq_size = sq_size;
993 uresp.actual_rq_size = rq_size;
994 uresp.qp_id = qp_num;
995 uresp.qp_caps = qp->qp_uk.qp_caps;
997 err_code = ib_copy_to_udata(udata, &uresp,
998 min(sizeof(uresp), udata->outlen));
1000 ibdev_dbg(&iwdev->ibdev, "VERBS: copy_to_udata failed\n");
1001 irdma_destroy_qp(&iwqp->ibqp, udata);
1002 return ERR_PTR(err_code);
1006 init_completion(&iwqp->free_qp);
1010 irdma_free_qp_rsrc(iwqp);
1012 return ERR_PTR(err_code);
1015 static int irdma_get_ib_acc_flags(struct irdma_qp *iwqp)
1019 if (rdma_protocol_roce(iwqp->ibqp.device, 1)) {
1020 if (iwqp->roce_info.wr_rdresp_en) {
1021 acc_flags |= IB_ACCESS_LOCAL_WRITE;
1022 acc_flags |= IB_ACCESS_REMOTE_WRITE;
1024 if (iwqp->roce_info.rd_en)
1025 acc_flags |= IB_ACCESS_REMOTE_READ;
1026 if (iwqp->roce_info.bind_en)
1027 acc_flags |= IB_ACCESS_MW_BIND;
1029 if (iwqp->iwarp_info.wr_rdresp_en) {
1030 acc_flags |= IB_ACCESS_LOCAL_WRITE;
1031 acc_flags |= IB_ACCESS_REMOTE_WRITE;
1033 if (iwqp->iwarp_info.rd_en)
1034 acc_flags |= IB_ACCESS_REMOTE_READ;
1035 if (iwqp->iwarp_info.bind_en)
1036 acc_flags |= IB_ACCESS_MW_BIND;
1042 * irdma_query_qp - query qp attributes
1044 * @attr: attributes pointer
1045 * @attr_mask: Not used
1046 * @init_attr: qp attributes to return
1048 static int irdma_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
1049 int attr_mask, struct ib_qp_init_attr *init_attr)
1051 struct irdma_qp *iwqp = to_iwqp(ibqp);
1052 struct irdma_sc_qp *qp = &iwqp->sc_qp;
1054 memset(attr, 0, sizeof(*attr));
1055 memset(init_attr, 0, sizeof(*init_attr));
1057 attr->qp_state = iwqp->ibqp_state;
1058 attr->cur_qp_state = iwqp->ibqp_state;
1059 attr->cap.max_send_wr = iwqp->max_send_wr;
1060 attr->cap.max_recv_wr = iwqp->max_recv_wr;
1061 attr->cap.max_inline_data = qp->qp_uk.max_inline_data;
1062 attr->cap.max_send_sge = qp->qp_uk.max_sq_frag_cnt;
1063 attr->cap.max_recv_sge = qp->qp_uk.max_rq_frag_cnt;
1064 attr->qp_access_flags = irdma_get_ib_acc_flags(iwqp);
1066 if (rdma_protocol_roce(ibqp->device, 1)) {
1067 attr->path_mtu = ib_mtu_int_to_enum(iwqp->udp_info.snd_mss);
1068 attr->qkey = iwqp->roce_info.qkey;
1069 attr->rq_psn = iwqp->udp_info.epsn;
1070 attr->sq_psn = iwqp->udp_info.psn_nxt;
1071 attr->dest_qp_num = iwqp->roce_info.dest_qp;
1072 attr->pkey_index = iwqp->roce_info.p_key;
1073 attr->retry_cnt = iwqp->udp_info.rexmit_thresh;
1074 attr->rnr_retry = iwqp->udp_info.rnr_nak_thresh;
1075 attr->max_rd_atomic = iwqp->roce_info.ord_size;
1076 attr->max_dest_rd_atomic = iwqp->roce_info.ird_size;
1079 init_attr->event_handler = iwqp->ibqp.event_handler;
1080 init_attr->qp_context = iwqp->ibqp.qp_context;
1081 init_attr->send_cq = iwqp->ibqp.send_cq;
1082 init_attr->recv_cq = iwqp->ibqp.recv_cq;
1083 init_attr->cap = attr->cap;
1089 * irdma_query_pkey - Query partition key
1090 * @ibdev: device pointer from stack
1091 * @port: port number
1092 * @index: index of pkey
1093 * @pkey: pointer to store the pkey
1095 static int irdma_query_pkey(struct ib_device *ibdev, u32 port, u16 index,
1098 if (index >= IRDMA_PKEY_TBL_SZ)
1101 *pkey = IRDMA_DEFAULT_PKEY;
1106 * irdma_modify_qp_roce - modify qp request
1107 * @ibqp: qp's pointer for modify
1108 * @attr: access attributes
1109 * @attr_mask: state mask
1112 int irdma_modify_qp_roce(struct ib_qp *ibqp, struct ib_qp_attr *attr,
1113 int attr_mask, struct ib_udata *udata)
1115 struct irdma_pd *iwpd = to_iwpd(ibqp->pd);
1116 struct irdma_qp *iwqp = to_iwqp(ibqp);
1117 struct irdma_device *iwdev = iwqp->iwdev;
1118 struct irdma_sc_dev *dev = &iwdev->rf->sc_dev;
1119 struct irdma_qp_host_ctx_info *ctx_info;
1120 struct irdma_roce_offload_info *roce_info;
1121 struct irdma_udp_offload_info *udp_info;
1122 struct irdma_modify_qp_info info = {};
1123 struct irdma_modify_qp_resp uresp = {};
1124 struct irdma_modify_qp_req ureq = {};
1125 unsigned long flags;
1126 u8 issue_modify_qp = 0;
1129 ctx_info = &iwqp->ctx_info;
1130 roce_info = &iwqp->roce_info;
1131 udp_info = &iwqp->udp_info;
1133 if (attr_mask & ~IB_QP_ATTR_STANDARD_BITS)
1136 if (attr_mask & IB_QP_DEST_QPN)
1137 roce_info->dest_qp = attr->dest_qp_num;
1139 if (attr_mask & IB_QP_PKEY_INDEX) {
1140 ret = irdma_query_pkey(ibqp->device, 0, attr->pkey_index,
1146 if (attr_mask & IB_QP_QKEY)
1147 roce_info->qkey = attr->qkey;
1149 if (attr_mask & IB_QP_PATH_MTU)
1150 udp_info->snd_mss = ib_mtu_enum_to_int(attr->path_mtu);
1152 if (attr_mask & IB_QP_SQ_PSN) {
1153 udp_info->psn_nxt = attr->sq_psn;
1154 udp_info->lsn = 0xffff;
1155 udp_info->psn_una = attr->sq_psn;
1156 udp_info->psn_max = attr->sq_psn;
1159 if (attr_mask & IB_QP_RQ_PSN)
1160 udp_info->epsn = attr->rq_psn;
1162 if (attr_mask & IB_QP_RNR_RETRY)
1163 udp_info->rnr_nak_thresh = attr->rnr_retry;
1165 if (attr_mask & IB_QP_RETRY_CNT)
1166 udp_info->rexmit_thresh = attr->retry_cnt;
1168 ctx_info->roce_info->pd_id = iwpd->sc_pd.pd_id;
1170 if (attr_mask & IB_QP_AV) {
1171 struct irdma_av *av = &iwqp->roce_ah.av;
1172 const struct ib_gid_attr *sgid_attr;
1173 u16 vlan_id = VLAN_N_VID;
1176 memset(&iwqp->roce_ah, 0, sizeof(iwqp->roce_ah));
1177 if (attr->ah_attr.ah_flags & IB_AH_GRH) {
1178 udp_info->ttl = attr->ah_attr.grh.hop_limit;
1179 udp_info->flow_label = attr->ah_attr.grh.flow_label;
1180 udp_info->tos = attr->ah_attr.grh.traffic_class;
1181 irdma_qp_rem_qos(&iwqp->sc_qp);
1182 dev->ws_remove(iwqp->sc_qp.vsi, ctx_info->user_pri);
1183 ctx_info->user_pri = rt_tos2priority(udp_info->tos);
1184 iwqp->sc_qp.user_pri = ctx_info->user_pri;
1185 if (dev->ws_add(iwqp->sc_qp.vsi, ctx_info->user_pri))
1187 irdma_qp_add_qos(&iwqp->sc_qp);
1189 sgid_attr = attr->ah_attr.grh.sgid_attr;
1190 ret = rdma_read_gid_l2_fields(sgid_attr, &vlan_id,
1191 ctx_info->roce_info->mac_addr);
1195 if (vlan_id >= VLAN_N_VID && iwdev->dcb)
1197 if (vlan_id < VLAN_N_VID) {
1198 udp_info->insert_vlan_tag = true;
1199 udp_info->vlan_tag = vlan_id |
1200 ctx_info->user_pri << VLAN_PRIO_SHIFT;
1202 udp_info->insert_vlan_tag = false;
1205 av->attrs = attr->ah_attr;
1206 rdma_gid2ip((struct sockaddr *)&av->sgid_addr, &sgid_attr->gid);
1207 rdma_gid2ip((struct sockaddr *)&av->dgid_addr, &attr->ah_attr.grh.dgid);
1208 roce_info->local_qp = ibqp->qp_num;
1209 if (av->sgid_addr.saddr.sa_family == AF_INET6) {
1211 av->dgid_addr.saddr_in6.sin6_addr.in6_u.u6_addr32;
1213 av->sgid_addr.saddr_in6.sin6_addr.in6_u.u6_addr32;
1215 irdma_copy_ip_ntohl(&udp_info->dest_ip_addr[0], daddr);
1216 irdma_copy_ip_ntohl(&udp_info->local_ipaddr[0], saddr);
1218 udp_info->ipv4 = false;
1219 irdma_copy_ip_ntohl(local_ip, daddr);
1221 udp_info->arp_idx = irdma_arp_table(iwdev->rf,
1226 __be32 saddr = av->sgid_addr.saddr_in.sin_addr.s_addr;
1227 __be32 daddr = av->dgid_addr.saddr_in.sin_addr.s_addr;
1229 local_ip[0] = ntohl(daddr);
1231 udp_info->ipv4 = true;
1232 udp_info->dest_ip_addr[0] = 0;
1233 udp_info->dest_ip_addr[1] = 0;
1234 udp_info->dest_ip_addr[2] = 0;
1235 udp_info->dest_ip_addr[3] = local_ip[0];
1237 udp_info->local_ipaddr[0] = 0;
1238 udp_info->local_ipaddr[1] = 0;
1239 udp_info->local_ipaddr[2] = 0;
1240 udp_info->local_ipaddr[3] = ntohl(saddr);
1243 irdma_add_arp(iwdev->rf, local_ip, udp_info->ipv4,
1244 attr->ah_attr.roce.dmac);
1247 if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC) {
1248 if (attr->max_rd_atomic > dev->hw_attrs.max_hw_ord) {
1249 ibdev_err(&iwdev->ibdev,
1250 "rd_atomic = %d, above max_hw_ord=%d\n",
1251 attr->max_rd_atomic,
1252 dev->hw_attrs.max_hw_ord);
1255 if (attr->max_rd_atomic)
1256 roce_info->ord_size = attr->max_rd_atomic;
1257 info.ord_valid = true;
1260 if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) {
1261 if (attr->max_dest_rd_atomic > dev->hw_attrs.max_hw_ird) {
1262 ibdev_err(&iwdev->ibdev,
1263 "rd_atomic = %d, above max_hw_ird=%d\n",
1264 attr->max_rd_atomic,
1265 dev->hw_attrs.max_hw_ird);
1268 if (attr->max_dest_rd_atomic)
1269 roce_info->ird_size = attr->max_dest_rd_atomic;
1272 if (attr_mask & IB_QP_ACCESS_FLAGS) {
1273 if (attr->qp_access_flags & IB_ACCESS_LOCAL_WRITE)
1274 roce_info->wr_rdresp_en = true;
1275 if (attr->qp_access_flags & IB_ACCESS_REMOTE_WRITE)
1276 roce_info->wr_rdresp_en = true;
1277 if (attr->qp_access_flags & IB_ACCESS_REMOTE_READ)
1278 roce_info->rd_en = true;
1281 wait_event(iwqp->mod_qp_waitq, !atomic_read(&iwqp->hw_mod_qp_pend));
1283 ibdev_dbg(&iwdev->ibdev,
1284 "VERBS: caller: %pS qp_id=%d to_ibqpstate=%d ibqpstate=%d irdma_qpstate=%d attr_mask=0x%x\n",
1285 __builtin_return_address(0), ibqp->qp_num, attr->qp_state,
1286 iwqp->ibqp_state, iwqp->iwarp_state, attr_mask);
1288 spin_lock_irqsave(&iwqp->lock, flags);
1289 if (attr_mask & IB_QP_STATE) {
1290 if (!ib_modify_qp_is_ok(iwqp->ibqp_state, attr->qp_state,
1291 iwqp->ibqp.qp_type, attr_mask)) {
1292 ibdev_warn(&iwdev->ibdev, "modify_qp invalid for qp_id=%d, old_state=0x%x, new_state=0x%x\n",
1293 iwqp->ibqp.qp_num, iwqp->ibqp_state,
1298 info.curr_iwarp_state = iwqp->iwarp_state;
1300 switch (attr->qp_state) {
1302 if (iwqp->iwarp_state > IRDMA_QP_STATE_IDLE) {
1307 if (iwqp->iwarp_state == IRDMA_QP_STATE_INVALID) {
1308 info.next_iwarp_state = IRDMA_QP_STATE_IDLE;
1309 issue_modify_qp = 1;
1313 if (iwqp->iwarp_state > IRDMA_QP_STATE_IDLE) {
1317 info.arp_cache_idx_valid = true;
1318 info.cq_num_valid = true;
1319 info.next_iwarp_state = IRDMA_QP_STATE_RTR;
1320 issue_modify_qp = 1;
1323 if (iwqp->ibqp_state < IB_QPS_RTR ||
1324 iwqp->ibqp_state == IB_QPS_ERR) {
1329 info.arp_cache_idx_valid = true;
1330 info.cq_num_valid = true;
1331 info.ord_valid = true;
1332 info.next_iwarp_state = IRDMA_QP_STATE_RTS;
1333 issue_modify_qp = 1;
1334 if (iwdev->push_mode && udata &&
1335 iwqp->sc_qp.push_idx == IRDMA_INVALID_PUSH_PAGE_INDEX &&
1336 dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2) {
1337 spin_unlock_irqrestore(&iwqp->lock, flags);
1338 irdma_alloc_push_page(iwqp);
1339 spin_lock_irqsave(&iwqp->lock, flags);
1343 if (iwqp->iwarp_state == IRDMA_QP_STATE_SQD)
1346 if (iwqp->iwarp_state != IRDMA_QP_STATE_RTS) {
1351 info.next_iwarp_state = IRDMA_QP_STATE_SQD;
1352 issue_modify_qp = 1;
1357 if (iwqp->iwarp_state == IRDMA_QP_STATE_RTS) {
1358 spin_unlock_irqrestore(&iwqp->lock, flags);
1359 info.next_iwarp_state = IRDMA_QP_STATE_SQD;
1360 irdma_hw_modify_qp(iwdev, iwqp, &info, true);
1361 spin_lock_irqsave(&iwqp->lock, flags);
1364 if (iwqp->iwarp_state == IRDMA_QP_STATE_ERROR) {
1365 spin_unlock_irqrestore(&iwqp->lock, flags);
1367 if (ib_copy_from_udata(&ureq, udata,
1368 min(sizeof(ureq), udata->inlen)))
1371 irdma_flush_wqes(iwqp,
1372 (ureq.sq_flush ? IRDMA_FLUSH_SQ : 0) |
1373 (ureq.rq_flush ? IRDMA_FLUSH_RQ : 0) |
1379 info.next_iwarp_state = IRDMA_QP_STATE_ERROR;
1380 issue_modify_qp = 1;
1387 iwqp->ibqp_state = attr->qp_state;
1390 ctx_info->send_cq_num = iwqp->iwscq->sc_cq.cq_uk.cq_id;
1391 ctx_info->rcv_cq_num = iwqp->iwrcq->sc_cq.cq_uk.cq_id;
1392 irdma_sc_qp_setctx_roce(&iwqp->sc_qp, iwqp->host_ctx.va, ctx_info);
1393 spin_unlock_irqrestore(&iwqp->lock, flags);
1395 if (attr_mask & IB_QP_STATE) {
1396 if (issue_modify_qp) {
1397 ctx_info->rem_endpoint_idx = udp_info->arp_idx;
1398 if (irdma_hw_modify_qp(iwdev, iwqp, &info, true))
1400 spin_lock_irqsave(&iwqp->lock, flags);
1401 if (iwqp->iwarp_state == info.curr_iwarp_state) {
1402 iwqp->iwarp_state = info.next_iwarp_state;
1403 iwqp->ibqp_state = attr->qp_state;
1405 if (iwqp->ibqp_state > IB_QPS_RTS &&
1406 !iwqp->flush_issued) {
1407 iwqp->flush_issued = 1;
1408 spin_unlock_irqrestore(&iwqp->lock, flags);
1409 irdma_flush_wqes(iwqp, IRDMA_FLUSH_SQ |
1413 spin_unlock_irqrestore(&iwqp->lock, flags);
1416 iwqp->ibqp_state = attr->qp_state;
1418 if (udata && dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2) {
1419 struct irdma_ucontext *ucontext;
1421 ucontext = rdma_udata_to_drv_context(udata,
1422 struct irdma_ucontext, ibucontext);
1423 if (iwqp->sc_qp.push_idx != IRDMA_INVALID_PUSH_PAGE_INDEX &&
1424 !iwqp->push_wqe_mmap_entry &&
1425 !irdma_setup_push_mmap_entries(ucontext, iwqp,
1426 &uresp.push_wqe_mmap_key, &uresp.push_db_mmap_key)) {
1427 uresp.push_valid = 1;
1428 uresp.push_offset = iwqp->sc_qp.push_offset;
1430 ret = ib_copy_to_udata(udata, &uresp, min(sizeof(uresp),
1433 irdma_remove_push_mmap_entries(iwqp);
1434 ibdev_dbg(&iwdev->ibdev,
1435 "VERBS: copy_to_udata failed\n");
1443 spin_unlock_irqrestore(&iwqp->lock, flags);
1449 * irdma_modify_qp - modify qp request
1450 * @ibqp: qp's pointer for modify
1451 * @attr: access attributes
1452 * @attr_mask: state mask
1455 int irdma_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask,
1456 struct ib_udata *udata)
1458 struct irdma_qp *iwqp = to_iwqp(ibqp);
1459 struct irdma_device *iwdev = iwqp->iwdev;
1460 struct irdma_sc_dev *dev = &iwdev->rf->sc_dev;
1461 struct irdma_qp_host_ctx_info *ctx_info;
1462 struct irdma_tcp_offload_info *tcp_info;
1463 struct irdma_iwarp_offload_info *offload_info;
1464 struct irdma_modify_qp_info info = {};
1465 struct irdma_modify_qp_resp uresp = {};
1466 struct irdma_modify_qp_req ureq = {};
1467 u8 issue_modify_qp = 0;
1470 unsigned long flags;
1472 if (attr_mask & ~IB_QP_ATTR_STANDARD_BITS)
1475 ctx_info = &iwqp->ctx_info;
1476 offload_info = &iwqp->iwarp_info;
1477 tcp_info = &iwqp->tcp_info;
1478 wait_event(iwqp->mod_qp_waitq, !atomic_read(&iwqp->hw_mod_qp_pend));
1479 ibdev_dbg(&iwdev->ibdev,
1480 "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",
1481 __builtin_return_address(0), ibqp->qp_num, attr->qp_state,
1482 iwqp->ibqp_state, iwqp->iwarp_state, iwqp->last_aeq,
1483 iwqp->hw_tcp_state, iwqp->hw_iwarp_state, attr_mask);
1485 spin_lock_irqsave(&iwqp->lock, flags);
1486 if (attr_mask & IB_QP_STATE) {
1487 info.curr_iwarp_state = iwqp->iwarp_state;
1488 switch (attr->qp_state) {
1491 if (iwqp->iwarp_state > IRDMA_QP_STATE_IDLE) {
1496 if (iwqp->iwarp_state == IRDMA_QP_STATE_INVALID) {
1497 info.next_iwarp_state = IRDMA_QP_STATE_IDLE;
1498 issue_modify_qp = 1;
1500 if (iwdev->push_mode && udata &&
1501 iwqp->sc_qp.push_idx == IRDMA_INVALID_PUSH_PAGE_INDEX &&
1502 dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2) {
1503 spin_unlock_irqrestore(&iwqp->lock, flags);
1504 irdma_alloc_push_page(iwqp);
1505 spin_lock_irqsave(&iwqp->lock, flags);
1509 if (iwqp->iwarp_state > IRDMA_QP_STATE_RTS ||
1515 issue_modify_qp = 1;
1516 iwqp->hw_tcp_state = IRDMA_TCP_STATE_ESTABLISHED;
1517 iwqp->hte_added = 1;
1518 info.next_iwarp_state = IRDMA_QP_STATE_RTS;
1519 info.tcp_ctx_valid = true;
1520 info.ord_valid = true;
1521 info.arp_cache_idx_valid = true;
1522 info.cq_num_valid = true;
1525 if (iwqp->hw_iwarp_state > IRDMA_QP_STATE_RTS) {
1530 if (iwqp->iwarp_state == IRDMA_QP_STATE_CLOSING ||
1531 iwqp->iwarp_state < IRDMA_QP_STATE_RTS) {
1536 if (iwqp->iwarp_state > IRDMA_QP_STATE_CLOSING) {
1541 info.next_iwarp_state = IRDMA_QP_STATE_CLOSING;
1542 issue_modify_qp = 1;
1545 if (iwqp->iwarp_state >= IRDMA_QP_STATE_TERMINATE) {
1550 info.next_iwarp_state = IRDMA_QP_STATE_TERMINATE;
1551 issue_modify_qp = 1;
1555 if (iwqp->iwarp_state == IRDMA_QP_STATE_ERROR) {
1556 spin_unlock_irqrestore(&iwqp->lock, flags);
1558 if (ib_copy_from_udata(&ureq, udata,
1559 min(sizeof(ureq), udata->inlen)))
1562 irdma_flush_wqes(iwqp,
1563 (ureq.sq_flush ? IRDMA_FLUSH_SQ : 0) |
1564 (ureq.rq_flush ? IRDMA_FLUSH_RQ : 0) |
1570 if (iwqp->sc_qp.term_flags) {
1571 spin_unlock_irqrestore(&iwqp->lock, flags);
1572 irdma_terminate_del_timer(&iwqp->sc_qp);
1573 spin_lock_irqsave(&iwqp->lock, flags);
1575 info.next_iwarp_state = IRDMA_QP_STATE_ERROR;
1576 if (iwqp->hw_tcp_state > IRDMA_TCP_STATE_CLOSED &&
1578 iwqp->hw_tcp_state != IRDMA_TCP_STATE_TIME_WAIT)
1579 info.reset_tcp_conn = true;
1583 issue_modify_qp = 1;
1584 info.next_iwarp_state = IRDMA_QP_STATE_ERROR;
1591 iwqp->ibqp_state = attr->qp_state;
1593 if (attr_mask & IB_QP_ACCESS_FLAGS) {
1594 ctx_info->iwarp_info_valid = true;
1595 if (attr->qp_access_flags & IB_ACCESS_LOCAL_WRITE)
1596 offload_info->wr_rdresp_en = true;
1597 if (attr->qp_access_flags & IB_ACCESS_REMOTE_WRITE)
1598 offload_info->wr_rdresp_en = true;
1599 if (attr->qp_access_flags & IB_ACCESS_REMOTE_READ)
1600 offload_info->rd_en = true;
1603 if (ctx_info->iwarp_info_valid) {
1604 ctx_info->send_cq_num = iwqp->iwscq->sc_cq.cq_uk.cq_id;
1605 ctx_info->rcv_cq_num = iwqp->iwrcq->sc_cq.cq_uk.cq_id;
1606 irdma_sc_qp_setctx(&iwqp->sc_qp, iwqp->host_ctx.va, ctx_info);
1608 spin_unlock_irqrestore(&iwqp->lock, flags);
1610 if (attr_mask & IB_QP_STATE) {
1611 if (issue_modify_qp) {
1612 ctx_info->rem_endpoint_idx = tcp_info->arp_idx;
1613 if (irdma_hw_modify_qp(iwdev, iwqp, &info, true))
1617 spin_lock_irqsave(&iwqp->lock, flags);
1618 if (iwqp->iwarp_state == info.curr_iwarp_state) {
1619 iwqp->iwarp_state = info.next_iwarp_state;
1620 iwqp->ibqp_state = attr->qp_state;
1622 spin_unlock_irqrestore(&iwqp->lock, flags);
1625 if (issue_modify_qp && iwqp->ibqp_state > IB_QPS_RTS) {
1627 if (iwqp->cm_id && iwqp->hw_tcp_state) {
1628 spin_lock_irqsave(&iwqp->lock, flags);
1629 iwqp->hw_tcp_state = IRDMA_TCP_STATE_CLOSED;
1630 iwqp->last_aeq = IRDMA_AE_RESET_SENT;
1631 spin_unlock_irqrestore(&iwqp->lock, flags);
1632 irdma_cm_disconn(iwqp);
1635 int close_timer_started;
1637 spin_lock_irqsave(&iwdev->cm_core.ht_lock, flags);
1639 if (iwqp->cm_node) {
1640 refcount_inc(&iwqp->cm_node->refcnt);
1641 spin_unlock_irqrestore(&iwdev->cm_core.ht_lock, flags);
1642 close_timer_started = atomic_inc_return(&iwqp->close_timer_started);
1643 if (iwqp->cm_id && close_timer_started == 1)
1644 irdma_schedule_cm_timer(iwqp->cm_node,
1645 (struct irdma_puda_buf *)iwqp,
1646 IRDMA_TIMER_TYPE_CLOSE, 1, 0);
1648 irdma_rem_ref_cm_node(iwqp->cm_node);
1650 spin_unlock_irqrestore(&iwdev->cm_core.ht_lock, flags);
1654 if (attr_mask & IB_QP_STATE && udata &&
1655 dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2) {
1656 struct irdma_ucontext *ucontext;
1658 ucontext = rdma_udata_to_drv_context(udata,
1659 struct irdma_ucontext, ibucontext);
1660 if (iwqp->sc_qp.push_idx != IRDMA_INVALID_PUSH_PAGE_INDEX &&
1661 !iwqp->push_wqe_mmap_entry &&
1662 !irdma_setup_push_mmap_entries(ucontext, iwqp,
1663 &uresp.push_wqe_mmap_key, &uresp.push_db_mmap_key)) {
1664 uresp.push_valid = 1;
1665 uresp.push_offset = iwqp->sc_qp.push_offset;
1668 err = ib_copy_to_udata(udata, &uresp, min(sizeof(uresp),
1671 irdma_remove_push_mmap_entries(iwqp);
1672 ibdev_dbg(&iwdev->ibdev,
1673 "VERBS: copy_to_udata failed\n");
1680 spin_unlock_irqrestore(&iwqp->lock, flags);
1686 * irdma_cq_free_rsrc - free up resources for cq
1687 * @rf: RDMA PCI function
1690 static void irdma_cq_free_rsrc(struct irdma_pci_f *rf, struct irdma_cq *iwcq)
1692 struct irdma_sc_cq *cq = &iwcq->sc_cq;
1694 if (!iwcq->user_mode) {
1695 dma_free_coherent(rf->sc_dev.hw->device, iwcq->kmem.size,
1696 iwcq->kmem.va, iwcq->kmem.pa);
1697 iwcq->kmem.va = NULL;
1698 dma_free_coherent(rf->sc_dev.hw->device,
1699 iwcq->kmem_shadow.size,
1700 iwcq->kmem_shadow.va, iwcq->kmem_shadow.pa);
1701 iwcq->kmem_shadow.va = NULL;
1704 irdma_free_rsrc(rf, rf->allocated_cqs, cq->cq_uk.cq_id);
1708 * irdma_free_cqbuf - worker to free a cq buffer
1709 * @work: provides access to the cq buffer to free
1711 static void irdma_free_cqbuf(struct work_struct *work)
1713 struct irdma_cq_buf *cq_buf = container_of(work, struct irdma_cq_buf, work);
1715 dma_free_coherent(cq_buf->hw->device, cq_buf->kmem_buf.size,
1716 cq_buf->kmem_buf.va, cq_buf->kmem_buf.pa);
1717 cq_buf->kmem_buf.va = NULL;
1722 * irdma_process_resize_list - remove resized cq buffers from the resize_list
1723 * @iwcq: cq which owns the resize_list
1724 * @iwdev: irdma device
1725 * @lcqe_buf: the buffer where the last cqe is received
1727 static int irdma_process_resize_list(struct irdma_cq *iwcq,
1728 struct irdma_device *iwdev,
1729 struct irdma_cq_buf *lcqe_buf)
1731 struct list_head *tmp_node, *list_node;
1732 struct irdma_cq_buf *cq_buf;
1735 list_for_each_safe(list_node, tmp_node, &iwcq->resize_list) {
1736 cq_buf = list_entry(list_node, struct irdma_cq_buf, list);
1737 if (cq_buf == lcqe_buf)
1740 list_del(&cq_buf->list);
1741 queue_work(iwdev->cleanup_wq, &cq_buf->work);
1749 * irdma_destroy_cq - destroy cq
1750 * @ib_cq: cq pointer
1753 static int irdma_destroy_cq(struct ib_cq *ib_cq, struct ib_udata *udata)
1755 struct irdma_device *iwdev = to_iwdev(ib_cq->device);
1756 struct irdma_cq *iwcq = to_iwcq(ib_cq);
1757 struct irdma_sc_cq *cq = &iwcq->sc_cq;
1758 struct irdma_sc_dev *dev = cq->dev;
1759 struct irdma_sc_ceq *ceq = dev->ceq[cq->ceq_id];
1760 struct irdma_ceq *iwceq = container_of(ceq, struct irdma_ceq, sc_ceq);
1761 unsigned long flags;
1763 spin_lock_irqsave(&iwcq->lock, flags);
1764 if (!list_empty(&iwcq->resize_list))
1765 irdma_process_resize_list(iwcq, iwdev, NULL);
1766 spin_unlock_irqrestore(&iwcq->lock, flags);
1768 irdma_cq_wq_destroy(iwdev->rf, cq);
1769 irdma_cq_free_rsrc(iwdev->rf, iwcq);
1771 spin_lock_irqsave(&iwceq->ce_lock, flags);
1772 irdma_sc_cleanup_ceqes(cq, ceq);
1773 spin_unlock_irqrestore(&iwceq->ce_lock, flags);
1779 * irdma_resize_cq - resize cq
1780 * @ibcq: cq to be resized
1781 * @entries: desired cq size
1784 static int irdma_resize_cq(struct ib_cq *ibcq, int entries,
1785 struct ib_udata *udata)
1787 struct irdma_cq *iwcq = to_iwcq(ibcq);
1788 struct irdma_sc_dev *dev = iwcq->sc_cq.dev;
1789 struct irdma_cqp_request *cqp_request;
1790 struct cqp_cmds_info *cqp_info;
1791 struct irdma_modify_cq_info *m_info;
1792 struct irdma_modify_cq_info info = {};
1793 struct irdma_dma_mem kmem_buf;
1794 struct irdma_cq_mr *cqmr_buf;
1795 struct irdma_pbl *iwpbl_buf;
1796 struct irdma_device *iwdev;
1797 struct irdma_pci_f *rf;
1798 struct irdma_cq_buf *cq_buf = NULL;
1799 enum irdma_status_code status = 0;
1800 unsigned long flags;
1803 iwdev = to_iwdev(ibcq->device);
1806 if (!(rf->sc_dev.hw_attrs.uk_attrs.feature_flags &
1807 IRDMA_FEATURE_CQ_RESIZE))
1810 if (entries > rf->max_cqe)
1813 if (!iwcq->user_mode) {
1815 if (rf->sc_dev.hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2)
1819 info.cq_size = max(entries, 4);
1821 if (info.cq_size == iwcq->sc_cq.cq_uk.cq_size - 1)
1825 struct irdma_resize_cq_req req = {};
1826 struct irdma_ucontext *ucontext =
1827 rdma_udata_to_drv_context(udata, struct irdma_ucontext,
1830 /* CQ resize not supported with legacy GEN_1 libi40iw */
1831 if (ucontext->legacy_mode)
1834 if (ib_copy_from_udata(&req, udata,
1835 min(sizeof(req), udata->inlen)))
1838 spin_lock_irqsave(&ucontext->cq_reg_mem_list_lock, flags);
1839 iwpbl_buf = irdma_get_pbl((unsigned long)req.user_cq_buffer,
1840 &ucontext->cq_reg_mem_list);
1841 spin_unlock_irqrestore(&ucontext->cq_reg_mem_list_lock, flags);
1846 cqmr_buf = &iwpbl_buf->cq_mr;
1847 if (iwpbl_buf->pbl_allocated) {
1848 info.virtual_map = true;
1849 info.pbl_chunk_size = 1;
1850 info.first_pm_pbl_idx = cqmr_buf->cq_pbl.idx;
1852 info.cq_pa = cqmr_buf->cq_pbl.addr;
1855 /* Kmode CQ resize */
1858 rsize = info.cq_size * sizeof(struct irdma_cqe);
1859 kmem_buf.size = ALIGN(round_up(rsize, 256), 256);
1860 kmem_buf.va = dma_alloc_coherent(dev->hw->device,
1861 kmem_buf.size, &kmem_buf.pa,
1866 info.cq_base = kmem_buf.va;
1867 info.cq_pa = kmem_buf.pa;
1868 cq_buf = kzalloc(sizeof(*cq_buf), GFP_KERNEL);
1875 cqp_request = irdma_alloc_and_get_cqp_request(&rf->cqp, true);
1881 info.shadow_read_threshold = iwcq->sc_cq.shadow_read_threshold;
1882 info.cq_resize = true;
1884 cqp_info = &cqp_request->info;
1885 m_info = &cqp_info->in.u.cq_modify.info;
1886 memcpy(m_info, &info, sizeof(*m_info));
1888 cqp_info->cqp_cmd = IRDMA_OP_CQ_MODIFY;
1889 cqp_info->in.u.cq_modify.cq = &iwcq->sc_cq;
1890 cqp_info->in.u.cq_modify.scratch = (uintptr_t)cqp_request;
1891 cqp_info->post_sq = 1;
1892 status = irdma_handle_cqp_op(rf, cqp_request);
1893 irdma_put_cqp_request(&rf->cqp, cqp_request);
1899 spin_lock_irqsave(&iwcq->lock, flags);
1901 cq_buf->kmem_buf = iwcq->kmem;
1902 cq_buf->hw = dev->hw;
1903 memcpy(&cq_buf->cq_uk, &iwcq->sc_cq.cq_uk, sizeof(cq_buf->cq_uk));
1904 INIT_WORK(&cq_buf->work, irdma_free_cqbuf);
1905 list_add_tail(&cq_buf->list, &iwcq->resize_list);
1906 iwcq->kmem = kmem_buf;
1909 irdma_sc_cq_resize(&iwcq->sc_cq, &info);
1910 ibcq->cqe = info.cq_size - 1;
1911 spin_unlock_irqrestore(&iwcq->lock, flags);
1916 dma_free_coherent(dev->hw->device, kmem_buf.size, kmem_buf.va,
1925 static inline int cq_validate_flags(u32 flags, u8 hw_rev)
1927 /* GEN1 does not support CQ create flags */
1928 if (hw_rev == IRDMA_GEN_1)
1929 return flags ? -EOPNOTSUPP : 0;
1931 return flags & ~IB_UVERBS_CQ_FLAGS_TIMESTAMP_COMPLETION ? -EOPNOTSUPP : 0;
1935 * irdma_create_cq - create cq
1936 * @ibcq: CQ allocated
1937 * @attr: attributes for cq
1940 static int irdma_create_cq(struct ib_cq *ibcq,
1941 const struct ib_cq_init_attr *attr,
1942 struct ib_udata *udata)
1944 struct ib_device *ibdev = ibcq->device;
1945 struct irdma_device *iwdev = to_iwdev(ibdev);
1946 struct irdma_pci_f *rf = iwdev->rf;
1947 struct irdma_cq *iwcq = to_iwcq(ibcq);
1949 struct irdma_sc_cq *cq;
1950 struct irdma_sc_dev *dev = &rf->sc_dev;
1951 struct irdma_cq_init_info info = {};
1952 enum irdma_status_code status;
1953 struct irdma_cqp_request *cqp_request;
1954 struct cqp_cmds_info *cqp_info;
1955 struct irdma_cq_uk_init_info *ukinfo = &info.cq_uk_init_info;
1956 unsigned long flags;
1958 int entries = attr->cqe;
1960 err_code = cq_validate_flags(attr->flags, dev->hw_attrs.uk_attrs.hw_rev);
1963 err_code = irdma_alloc_rsrc(rf, rf->allocated_cqs, rf->max_cq, &cq_num,
1970 spin_lock_init(&iwcq->lock);
1971 INIT_LIST_HEAD(&iwcq->resize_list);
1973 ukinfo->cq_size = max(entries, 4);
1974 ukinfo->cq_id = cq_num;
1975 iwcq->ibcq.cqe = info.cq_uk_init_info.cq_size;
1976 if (attr->comp_vector < rf->ceqs_count)
1977 info.ceq_id = attr->comp_vector;
1978 info.ceq_id_valid = true;
1980 info.type = IRDMA_CQ_TYPE_IWARP;
1981 info.vsi = &iwdev->vsi;
1984 struct irdma_ucontext *ucontext;
1985 struct irdma_create_cq_req req = {};
1986 struct irdma_cq_mr *cqmr;
1987 struct irdma_pbl *iwpbl;
1988 struct irdma_pbl *iwpbl_shadow;
1989 struct irdma_cq_mr *cqmr_shadow;
1991 iwcq->user_mode = true;
1993 rdma_udata_to_drv_context(udata, struct irdma_ucontext,
1995 if (ib_copy_from_udata(&req, udata,
1996 min(sizeof(req), udata->inlen))) {
2001 spin_lock_irqsave(&ucontext->cq_reg_mem_list_lock, flags);
2002 iwpbl = irdma_get_pbl((unsigned long)req.user_cq_buf,
2003 &ucontext->cq_reg_mem_list);
2004 spin_unlock_irqrestore(&ucontext->cq_reg_mem_list_lock, flags);
2010 iwcq->iwpbl = iwpbl;
2011 iwcq->cq_mem_size = 0;
2012 cqmr = &iwpbl->cq_mr;
2014 if (rf->sc_dev.hw_attrs.uk_attrs.feature_flags &
2015 IRDMA_FEATURE_CQ_RESIZE && !ucontext->legacy_mode) {
2016 spin_lock_irqsave(&ucontext->cq_reg_mem_list_lock, flags);
2017 iwpbl_shadow = irdma_get_pbl(
2018 (unsigned long)req.user_shadow_area,
2019 &ucontext->cq_reg_mem_list);
2020 spin_unlock_irqrestore(&ucontext->cq_reg_mem_list_lock, flags);
2022 if (!iwpbl_shadow) {
2026 iwcq->iwpbl_shadow = iwpbl_shadow;
2027 cqmr_shadow = &iwpbl_shadow->cq_mr;
2028 info.shadow_area_pa = cqmr_shadow->cq_pbl.addr;
2031 info.shadow_area_pa = cqmr->shadow;
2033 if (iwpbl->pbl_allocated) {
2034 info.virtual_map = true;
2035 info.pbl_chunk_size = 1;
2036 info.first_pm_pbl_idx = cqmr->cq_pbl.idx;
2038 info.cq_base_pa = cqmr->cq_pbl.addr;
2041 /* Kmode allocations */
2044 if (entries > rf->max_cqe) {
2050 if (dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2)
2052 ukinfo->cq_size = entries;
2054 rsize = info.cq_uk_init_info.cq_size * sizeof(struct irdma_cqe);
2055 iwcq->kmem.size = ALIGN(round_up(rsize, 256), 256);
2056 iwcq->kmem.va = dma_alloc_coherent(dev->hw->device,
2058 &iwcq->kmem.pa, GFP_KERNEL);
2059 if (!iwcq->kmem.va) {
2064 iwcq->kmem_shadow.size = ALIGN(IRDMA_SHADOW_AREA_SIZE << 3,
2066 iwcq->kmem_shadow.va = dma_alloc_coherent(dev->hw->device,
2067 iwcq->kmem_shadow.size,
2068 &iwcq->kmem_shadow.pa,
2070 if (!iwcq->kmem_shadow.va) {
2074 info.shadow_area_pa = iwcq->kmem_shadow.pa;
2075 ukinfo->shadow_area = iwcq->kmem_shadow.va;
2076 ukinfo->cq_base = iwcq->kmem.va;
2077 info.cq_base_pa = iwcq->kmem.pa;
2080 if (dev->hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2)
2081 info.shadow_read_threshold = min(info.cq_uk_init_info.cq_size / 2,
2082 (u32)IRDMA_MAX_CQ_READ_THRESH);
2084 if (irdma_sc_cq_init(cq, &info)) {
2085 ibdev_dbg(&iwdev->ibdev, "VERBS: init cq fail\n");
2090 cqp_request = irdma_alloc_and_get_cqp_request(&rf->cqp, true);
2096 cqp_info = &cqp_request->info;
2097 cqp_info->cqp_cmd = IRDMA_OP_CQ_CREATE;
2098 cqp_info->post_sq = 1;
2099 cqp_info->in.u.cq_create.cq = cq;
2100 cqp_info->in.u.cq_create.check_overflow = true;
2101 cqp_info->in.u.cq_create.scratch = (uintptr_t)cqp_request;
2102 status = irdma_handle_cqp_op(rf, cqp_request);
2103 irdma_put_cqp_request(&rf->cqp, cqp_request);
2110 struct irdma_create_cq_resp resp = {};
2112 resp.cq_id = info.cq_uk_init_info.cq_id;
2113 resp.cq_size = info.cq_uk_init_info.cq_size;
2114 if (ib_copy_to_udata(udata, &resp,
2115 min(sizeof(resp), udata->outlen))) {
2116 ibdev_dbg(&iwdev->ibdev,
2117 "VERBS: copy to user data\n");
2124 irdma_cq_wq_destroy(rf, cq);
2126 irdma_cq_free_rsrc(rf, iwcq);
2132 * irdma_get_mr_access - get hw MR access permissions from IB access flags
2133 * @access: IB access flags
2135 static inline u16 irdma_get_mr_access(int access)
2139 hw_access |= (access & IB_ACCESS_LOCAL_WRITE) ?
2140 IRDMA_ACCESS_FLAGS_LOCALWRITE : 0;
2141 hw_access |= (access & IB_ACCESS_REMOTE_WRITE) ?
2142 IRDMA_ACCESS_FLAGS_REMOTEWRITE : 0;
2143 hw_access |= (access & IB_ACCESS_REMOTE_READ) ?
2144 IRDMA_ACCESS_FLAGS_REMOTEREAD : 0;
2145 hw_access |= (access & IB_ACCESS_MW_BIND) ?
2146 IRDMA_ACCESS_FLAGS_BIND_WINDOW : 0;
2147 hw_access |= (access & IB_ZERO_BASED) ?
2148 IRDMA_ACCESS_FLAGS_ZERO_BASED : 0;
2149 hw_access |= IRDMA_ACCESS_FLAGS_LOCALREAD;
2155 * irdma_free_stag - free stag resource
2156 * @iwdev: irdma device
2157 * @stag: stag to free
2159 static void irdma_free_stag(struct irdma_device *iwdev, u32 stag)
2163 stag_idx = (stag & iwdev->rf->mr_stagmask) >> IRDMA_CQPSQ_STAG_IDX_S;
2164 irdma_free_rsrc(iwdev->rf, iwdev->rf->allocated_mrs, stag_idx);
2168 * irdma_create_stag - create random stag
2169 * @iwdev: irdma device
2171 static u32 irdma_create_stag(struct irdma_device *iwdev)
2175 u32 next_stag_index;
2181 get_random_bytes(&random, sizeof(random));
2182 consumer_key = (u8)random;
2184 driver_key = random & ~iwdev->rf->mr_stagmask;
2185 next_stag_index = (random & iwdev->rf->mr_stagmask) >> 8;
2186 next_stag_index %= iwdev->rf->max_mr;
2188 ret = irdma_alloc_rsrc(iwdev->rf, iwdev->rf->allocated_mrs,
2189 iwdev->rf->max_mr, &stag_index,
2193 stag = stag_index << IRDMA_CQPSQ_STAG_IDX_S;
2195 stag += (u32)consumer_key;
2201 * irdma_next_pbl_addr - Get next pbl address
2202 * @pbl: pointer to a pble
2203 * @pinfo: info pointer
2206 static inline u64 *irdma_next_pbl_addr(u64 *pbl, struct irdma_pble_info **pinfo,
2210 if (!(*pinfo) || *idx != (*pinfo)->cnt)
2215 return (*pinfo)->addr;
2219 * irdma_copy_user_pgaddrs - copy user page address to pble's os locally
2220 * @iwmr: iwmr for IB's user page addresses
2221 * @pbl: ple pointer to save 1 level or 0 level pble
2222 * @level: indicated level 0, 1 or 2
2224 static void irdma_copy_user_pgaddrs(struct irdma_mr *iwmr, u64 *pbl,
2225 enum irdma_pble_level level)
2227 struct ib_umem *region = iwmr->region;
2228 struct irdma_pbl *iwpbl = &iwmr->iwpbl;
2229 struct irdma_pble_alloc *palloc = &iwpbl->pble_alloc;
2230 struct irdma_pble_info *pinfo;
2231 struct ib_block_iter biter;
2235 pinfo = (level == PBLE_LEVEL_1) ? NULL : palloc->level2.leaf;
2237 if (iwmr->type == IRDMA_MEMREG_TYPE_QP)
2238 iwpbl->qp_mr.sq_page = sg_page(region->sg_head.sgl);
2240 rdma_umem_for_each_dma_block(region, &biter, iwmr->page_size) {
2241 *pbl = rdma_block_iter_dma_address(&biter);
2242 if (++pbl_cnt == palloc->total_cnt)
2244 pbl = irdma_next_pbl_addr(pbl, &pinfo, &idx);
2249 * irdma_check_mem_contiguous - check if pbls stored in arr are contiguous
2250 * @arr: lvl1 pbl array
2251 * @npages: page count
2252 * @pg_size: page size
2255 static bool irdma_check_mem_contiguous(u64 *arr, u32 npages, u32 pg_size)
2259 for (pg_idx = 0; pg_idx < npages; pg_idx++) {
2260 if ((*arr + (pg_size * pg_idx)) != arr[pg_idx])
2268 * irdma_check_mr_contiguous - check if MR is physically contiguous
2269 * @palloc: pbl allocation struct
2270 * @pg_size: page size
2272 static bool irdma_check_mr_contiguous(struct irdma_pble_alloc *palloc,
2275 struct irdma_pble_level2 *lvl2 = &palloc->level2;
2276 struct irdma_pble_info *leaf = lvl2->leaf;
2278 u64 *start_addr = NULL;
2282 if (palloc->level == PBLE_LEVEL_1) {
2283 arr = palloc->level1.addr;
2284 ret = irdma_check_mem_contiguous(arr, palloc->total_cnt,
2289 start_addr = leaf->addr;
2291 for (i = 0; i < lvl2->leaf_cnt; i++, leaf++) {
2293 if ((*start_addr + (i * pg_size * PBLE_PER_PAGE)) != *arr)
2295 ret = irdma_check_mem_contiguous(arr, leaf->cnt, pg_size);
2304 * irdma_setup_pbles - copy user pg address to pble's
2305 * @rf: RDMA PCI function
2306 * @iwmr: mr pointer for this memory registration
2307 * @use_pbles: flag if to use pble's
2309 static int irdma_setup_pbles(struct irdma_pci_f *rf, struct irdma_mr *iwmr,
2312 struct irdma_pbl *iwpbl = &iwmr->iwpbl;
2313 struct irdma_pble_alloc *palloc = &iwpbl->pble_alloc;
2314 struct irdma_pble_info *pinfo;
2316 enum irdma_status_code status;
2317 enum irdma_pble_level level = PBLE_LEVEL_1;
2320 status = irdma_get_pble(rf->pble_rsrc, palloc, iwmr->page_cnt,
2325 iwpbl->pbl_allocated = true;
2326 level = palloc->level;
2327 pinfo = (level == PBLE_LEVEL_1) ? &palloc->level1 :
2328 palloc->level2.leaf;
2331 pbl = iwmr->pgaddrmem;
2334 irdma_copy_user_pgaddrs(iwmr, pbl, level);
2337 iwmr->pgaddrmem[0] = *pbl;
2343 * irdma_handle_q_mem - handle memory for qp and cq
2344 * @iwdev: irdma device
2345 * @req: information for q memory management
2346 * @iwpbl: pble struct
2347 * @use_pbles: flag to use pble
2349 static int irdma_handle_q_mem(struct irdma_device *iwdev,
2350 struct irdma_mem_reg_req *req,
2351 struct irdma_pbl *iwpbl, bool use_pbles)
2353 struct irdma_pble_alloc *palloc = &iwpbl->pble_alloc;
2354 struct irdma_mr *iwmr = iwpbl->iwmr;
2355 struct irdma_qp_mr *qpmr = &iwpbl->qp_mr;
2356 struct irdma_cq_mr *cqmr = &iwpbl->cq_mr;
2357 struct irdma_hmc_pble *hmc_p;
2358 u64 *arr = iwmr->pgaddrmem;
2363 pg_size = iwmr->page_size;
2364 err = irdma_setup_pbles(iwdev->rf, iwmr, use_pbles);
2368 if (use_pbles && palloc->level != PBLE_LEVEL_1) {
2369 irdma_free_pble(iwdev->rf->pble_rsrc, palloc);
2370 iwpbl->pbl_allocated = false;
2375 arr = palloc->level1.addr;
2377 switch (iwmr->type) {
2378 case IRDMA_MEMREG_TYPE_QP:
2379 total = req->sq_pages + req->rq_pages;
2380 hmc_p = &qpmr->sq_pbl;
2381 qpmr->shadow = (dma_addr_t)arr[total];
2384 ret = irdma_check_mem_contiguous(arr, req->sq_pages,
2387 ret = irdma_check_mem_contiguous(&arr[req->sq_pages],
2393 hmc_p->idx = palloc->level1.idx;
2394 hmc_p = &qpmr->rq_pbl;
2395 hmc_p->idx = palloc->level1.idx + req->sq_pages;
2397 hmc_p->addr = arr[0];
2398 hmc_p = &qpmr->rq_pbl;
2399 hmc_p->addr = arr[req->sq_pages];
2402 case IRDMA_MEMREG_TYPE_CQ:
2403 hmc_p = &cqmr->cq_pbl;
2406 cqmr->shadow = (dma_addr_t)arr[req->cq_pages];
2409 ret = irdma_check_mem_contiguous(arr, req->cq_pages,
2413 hmc_p->idx = palloc->level1.idx;
2415 hmc_p->addr = arr[0];
2418 ibdev_dbg(&iwdev->ibdev, "VERBS: MR type error\n");
2422 if (use_pbles && ret) {
2423 irdma_free_pble(iwdev->rf->pble_rsrc, palloc);
2424 iwpbl->pbl_allocated = false;
2431 * irdma_hw_alloc_mw - create the hw memory window
2432 * @iwdev: irdma device
2433 * @iwmr: pointer to memory window info
2435 static int irdma_hw_alloc_mw(struct irdma_device *iwdev, struct irdma_mr *iwmr)
2437 struct irdma_mw_alloc_info *info;
2438 struct irdma_pd *iwpd = to_iwpd(iwmr->ibmr.pd);
2439 struct irdma_cqp_request *cqp_request;
2440 struct cqp_cmds_info *cqp_info;
2441 enum irdma_status_code status;
2443 cqp_request = irdma_alloc_and_get_cqp_request(&iwdev->rf->cqp, true);
2447 cqp_info = &cqp_request->info;
2448 info = &cqp_info->in.u.mw_alloc.info;
2449 memset(info, 0, sizeof(*info));
2450 if (iwmr->ibmw.type == IB_MW_TYPE_1)
2451 info->mw_wide = true;
2453 info->page_size = PAGE_SIZE;
2454 info->mw_stag_index = iwmr->stag >> IRDMA_CQPSQ_STAG_IDX_S;
2455 info->pd_id = iwpd->sc_pd.pd_id;
2456 info->remote_access = true;
2457 cqp_info->cqp_cmd = IRDMA_OP_MW_ALLOC;
2458 cqp_info->post_sq = 1;
2459 cqp_info->in.u.mw_alloc.dev = &iwdev->rf->sc_dev;
2460 cqp_info->in.u.mw_alloc.scratch = (uintptr_t)cqp_request;
2461 status = irdma_handle_cqp_op(iwdev->rf, cqp_request);
2462 irdma_put_cqp_request(&iwdev->rf->cqp, cqp_request);
2464 return status ? -ENOMEM : 0;
2468 * irdma_alloc_mw - Allocate memory window
2469 * @ibmw: Memory Window
2470 * @udata: user data pointer
2472 static int irdma_alloc_mw(struct ib_mw *ibmw, struct ib_udata *udata)
2474 struct irdma_device *iwdev = to_iwdev(ibmw->device);
2475 struct irdma_mr *iwmr = to_iwmw(ibmw);
2479 stag = irdma_create_stag(iwdev);
2486 err_code = irdma_hw_alloc_mw(iwdev, iwmr);
2488 irdma_free_stag(iwdev, stag);
2496 * irdma_dealloc_mw - Dealloc memory window
2497 * @ibmw: memory window structure.
2499 static int irdma_dealloc_mw(struct ib_mw *ibmw)
2501 struct ib_pd *ibpd = ibmw->pd;
2502 struct irdma_pd *iwpd = to_iwpd(ibpd);
2503 struct irdma_mr *iwmr = to_iwmr((struct ib_mr *)ibmw);
2504 struct irdma_device *iwdev = to_iwdev(ibmw->device);
2505 struct irdma_cqp_request *cqp_request;
2506 struct cqp_cmds_info *cqp_info;
2507 struct irdma_dealloc_stag_info *info;
2509 cqp_request = irdma_alloc_and_get_cqp_request(&iwdev->rf->cqp, true);
2513 cqp_info = &cqp_request->info;
2514 info = &cqp_info->in.u.dealloc_stag.info;
2515 memset(info, 0, sizeof(*info));
2516 info->pd_id = iwpd->sc_pd.pd_id & 0x00007fff;
2517 info->stag_idx = ibmw->rkey >> IRDMA_CQPSQ_STAG_IDX_S;
2519 cqp_info->cqp_cmd = IRDMA_OP_DEALLOC_STAG;
2520 cqp_info->post_sq = 1;
2521 cqp_info->in.u.dealloc_stag.dev = &iwdev->rf->sc_dev;
2522 cqp_info->in.u.dealloc_stag.scratch = (uintptr_t)cqp_request;
2523 irdma_handle_cqp_op(iwdev->rf, cqp_request);
2524 irdma_put_cqp_request(&iwdev->rf->cqp, cqp_request);
2525 irdma_free_stag(iwdev, iwmr->stag);
2531 * irdma_hw_alloc_stag - cqp command to allocate stag
2532 * @iwdev: irdma device
2533 * @iwmr: irdma mr pointer
2535 static int irdma_hw_alloc_stag(struct irdma_device *iwdev,
2536 struct irdma_mr *iwmr)
2538 struct irdma_allocate_stag_info *info;
2539 struct irdma_pd *iwpd = to_iwpd(iwmr->ibmr.pd);
2540 enum irdma_status_code status;
2542 struct irdma_cqp_request *cqp_request;
2543 struct cqp_cmds_info *cqp_info;
2545 cqp_request = irdma_alloc_and_get_cqp_request(&iwdev->rf->cqp, true);
2549 cqp_info = &cqp_request->info;
2550 info = &cqp_info->in.u.alloc_stag.info;
2551 memset(info, 0, sizeof(*info));
2552 info->page_size = PAGE_SIZE;
2553 info->stag_idx = iwmr->stag >> IRDMA_CQPSQ_STAG_IDX_S;
2554 info->pd_id = iwpd->sc_pd.pd_id;
2555 info->total_len = iwmr->len;
2556 info->remote_access = true;
2557 cqp_info->cqp_cmd = IRDMA_OP_ALLOC_STAG;
2558 cqp_info->post_sq = 1;
2559 cqp_info->in.u.alloc_stag.dev = &iwdev->rf->sc_dev;
2560 cqp_info->in.u.alloc_stag.scratch = (uintptr_t)cqp_request;
2561 status = irdma_handle_cqp_op(iwdev->rf, cqp_request);
2562 irdma_put_cqp_request(&iwdev->rf->cqp, cqp_request);
2570 * irdma_alloc_mr - register stag for fast memory registration
2572 * @mr_type: memory for stag registrion
2573 * @max_num_sg: man number of pages
2575 static struct ib_mr *irdma_alloc_mr(struct ib_pd *pd, enum ib_mr_type mr_type,
2578 struct irdma_device *iwdev = to_iwdev(pd->device);
2579 struct irdma_pble_alloc *palloc;
2580 struct irdma_pbl *iwpbl;
2581 struct irdma_mr *iwmr;
2582 enum irdma_status_code status;
2584 int err_code = -ENOMEM;
2586 iwmr = kzalloc(sizeof(*iwmr), GFP_KERNEL);
2588 return ERR_PTR(-ENOMEM);
2590 stag = irdma_create_stag(iwdev);
2597 iwmr->ibmr.rkey = stag;
2598 iwmr->ibmr.lkey = stag;
2600 iwmr->ibmr.device = pd->device;
2601 iwpbl = &iwmr->iwpbl;
2603 iwmr->type = IRDMA_MEMREG_TYPE_MEM;
2604 palloc = &iwpbl->pble_alloc;
2605 iwmr->page_cnt = max_num_sg;
2606 status = irdma_get_pble(iwdev->rf->pble_rsrc, palloc, iwmr->page_cnt,
2611 err_code = irdma_hw_alloc_stag(iwdev, iwmr);
2613 goto err_alloc_stag;
2615 iwpbl->pbl_allocated = true;
2619 irdma_free_pble(iwdev->rf->pble_rsrc, palloc);
2621 irdma_free_stag(iwdev, stag);
2625 return ERR_PTR(err_code);
2629 * irdma_set_page - populate pbl list for fmr
2630 * @ibmr: ib mem to access iwarp mr pointer
2631 * @addr: page dma address fro pbl list
2633 static int irdma_set_page(struct ib_mr *ibmr, u64 addr)
2635 struct irdma_mr *iwmr = to_iwmr(ibmr);
2636 struct irdma_pbl *iwpbl = &iwmr->iwpbl;
2637 struct irdma_pble_alloc *palloc = &iwpbl->pble_alloc;
2640 if (unlikely(iwmr->npages == iwmr->page_cnt))
2643 pbl = palloc->level1.addr;
2644 pbl[iwmr->npages++] = addr;
2650 * irdma_map_mr_sg - map of sg list for fmr
2651 * @ibmr: ib mem to access iwarp mr pointer
2652 * @sg: scatter gather list
2653 * @sg_nents: number of sg pages
2654 * @sg_offset: scatter gather list for fmr
2656 static int irdma_map_mr_sg(struct ib_mr *ibmr, struct scatterlist *sg,
2657 int sg_nents, unsigned int *sg_offset)
2659 struct irdma_mr *iwmr = to_iwmr(ibmr);
2663 return ib_sg_to_pages(ibmr, sg, sg_nents, sg_offset, irdma_set_page);
2667 * irdma_hwreg_mr - send cqp command for memory registration
2668 * @iwdev: irdma device
2669 * @iwmr: irdma mr pointer
2670 * @access: access for MR
2672 static int irdma_hwreg_mr(struct irdma_device *iwdev, struct irdma_mr *iwmr,
2675 struct irdma_pbl *iwpbl = &iwmr->iwpbl;
2676 struct irdma_reg_ns_stag_info *stag_info;
2677 struct irdma_pd *iwpd = to_iwpd(iwmr->ibmr.pd);
2678 struct irdma_pble_alloc *palloc = &iwpbl->pble_alloc;
2679 enum irdma_status_code status;
2681 struct irdma_cqp_request *cqp_request;
2682 struct cqp_cmds_info *cqp_info;
2684 cqp_request = irdma_alloc_and_get_cqp_request(&iwdev->rf->cqp, true);
2688 cqp_info = &cqp_request->info;
2689 stag_info = &cqp_info->in.u.mr_reg_non_shared.info;
2690 memset(stag_info, 0, sizeof(*stag_info));
2691 stag_info->va = iwpbl->user_base;
2692 stag_info->stag_idx = iwmr->stag >> IRDMA_CQPSQ_STAG_IDX_S;
2693 stag_info->stag_key = (u8)iwmr->stag;
2694 stag_info->total_len = iwmr->len;
2695 stag_info->access_rights = irdma_get_mr_access(access);
2696 stag_info->pd_id = iwpd->sc_pd.pd_id;
2697 if (stag_info->access_rights & IRDMA_ACCESS_FLAGS_ZERO_BASED)
2698 stag_info->addr_type = IRDMA_ADDR_TYPE_ZERO_BASED;
2700 stag_info->addr_type = IRDMA_ADDR_TYPE_VA_BASED;
2701 stag_info->page_size = iwmr->page_size;
2703 if (iwpbl->pbl_allocated) {
2704 if (palloc->level == PBLE_LEVEL_1) {
2705 stag_info->first_pm_pbl_index = palloc->level1.idx;
2706 stag_info->chunk_size = 1;
2708 stag_info->first_pm_pbl_index = palloc->level2.root.idx;
2709 stag_info->chunk_size = 3;
2712 stag_info->reg_addr_pa = iwmr->pgaddrmem[0];
2715 cqp_info->cqp_cmd = IRDMA_OP_MR_REG_NON_SHARED;
2716 cqp_info->post_sq = 1;
2717 cqp_info->in.u.mr_reg_non_shared.dev = &iwdev->rf->sc_dev;
2718 cqp_info->in.u.mr_reg_non_shared.scratch = (uintptr_t)cqp_request;
2719 status = irdma_handle_cqp_op(iwdev->rf, cqp_request);
2720 irdma_put_cqp_request(&iwdev->rf->cqp, cqp_request);
2728 * irdma_reg_user_mr - Register a user memory region
2730 * @start: virtual start address
2731 * @len: length of mr
2732 * @virt: virtual address
2733 * @access: access of mr
2736 static struct ib_mr *irdma_reg_user_mr(struct ib_pd *pd, u64 start, u64 len,
2737 u64 virt, int access,
2738 struct ib_udata *udata)
2740 struct irdma_device *iwdev = to_iwdev(pd->device);
2741 struct irdma_ucontext *ucontext;
2742 struct irdma_pble_alloc *palloc;
2743 struct irdma_pbl *iwpbl;
2744 struct irdma_mr *iwmr;
2745 struct ib_umem *region;
2746 struct irdma_mem_reg_req req;
2747 u32 total, stag = 0;
2748 u8 shadow_pgcnt = 1;
2749 bool use_pbles = false;
2750 unsigned long flags;
2754 if (len > iwdev->rf->sc_dev.hw_attrs.max_mr_size)
2755 return ERR_PTR(-EINVAL);
2757 region = ib_umem_get(pd->device, start, len, access);
2759 if (IS_ERR(region)) {
2760 ibdev_dbg(&iwdev->ibdev,
2761 "VERBS: Failed to create ib_umem region\n");
2762 return (struct ib_mr *)region;
2765 if (ib_copy_from_udata(&req, udata, min(sizeof(req), udata->inlen))) {
2766 ib_umem_release(region);
2767 return ERR_PTR(-EFAULT);
2770 iwmr = kzalloc(sizeof(*iwmr), GFP_KERNEL);
2772 ib_umem_release(region);
2773 return ERR_PTR(-ENOMEM);
2776 iwpbl = &iwmr->iwpbl;
2778 iwmr->region = region;
2780 iwmr->ibmr.device = pd->device;
2781 iwmr->ibmr.iova = virt;
2782 iwmr->page_size = PAGE_SIZE;
2784 if (req.reg_type == IRDMA_MEMREG_TYPE_MEM) {
2785 iwmr->page_size = ib_umem_find_best_pgsz(region,
2786 SZ_4K | SZ_2M | SZ_1G,
2788 if (unlikely(!iwmr->page_size)) {
2790 ib_umem_release(region);
2791 return ERR_PTR(-EOPNOTSUPP);
2794 iwmr->len = region->length;
2795 iwpbl->user_base = virt;
2796 palloc = &iwpbl->pble_alloc;
2797 iwmr->type = req.reg_type;
2798 iwmr->page_cnt = ib_umem_num_dma_blocks(region, iwmr->page_size);
2800 switch (req.reg_type) {
2801 case IRDMA_MEMREG_TYPE_QP:
2802 total = req.sq_pages + req.rq_pages + shadow_pgcnt;
2803 if (total > iwmr->page_cnt) {
2807 total = req.sq_pages + req.rq_pages;
2808 use_pbles = (total > 2);
2809 err = irdma_handle_q_mem(iwdev, &req, iwpbl, use_pbles);
2813 ucontext = rdma_udata_to_drv_context(udata, struct irdma_ucontext,
2815 spin_lock_irqsave(&ucontext->qp_reg_mem_list_lock, flags);
2816 list_add_tail(&iwpbl->list, &ucontext->qp_reg_mem_list);
2817 iwpbl->on_list = true;
2818 spin_unlock_irqrestore(&ucontext->qp_reg_mem_list_lock, flags);
2820 case IRDMA_MEMREG_TYPE_CQ:
2821 if (iwdev->rf->sc_dev.hw_attrs.uk_attrs.feature_flags & IRDMA_FEATURE_CQ_RESIZE)
2823 total = req.cq_pages + shadow_pgcnt;
2824 if (total > iwmr->page_cnt) {
2829 use_pbles = (req.cq_pages > 1);
2830 err = irdma_handle_q_mem(iwdev, &req, iwpbl, use_pbles);
2834 ucontext = rdma_udata_to_drv_context(udata, struct irdma_ucontext,
2836 spin_lock_irqsave(&ucontext->cq_reg_mem_list_lock, flags);
2837 list_add_tail(&iwpbl->list, &ucontext->cq_reg_mem_list);
2838 iwpbl->on_list = true;
2839 spin_unlock_irqrestore(&ucontext->cq_reg_mem_list_lock, flags);
2841 case IRDMA_MEMREG_TYPE_MEM:
2842 use_pbles = (iwmr->page_cnt != 1);
2844 err = irdma_setup_pbles(iwdev->rf, iwmr, use_pbles);
2849 ret = irdma_check_mr_contiguous(palloc,
2852 irdma_free_pble(iwdev->rf->pble_rsrc, palloc);
2853 iwpbl->pbl_allocated = false;
2857 stag = irdma_create_stag(iwdev);
2864 iwmr->ibmr.rkey = stag;
2865 iwmr->ibmr.lkey = stag;
2866 err = irdma_hwreg_mr(iwdev, iwmr, access);
2868 irdma_free_stag(iwdev, stag);
2877 iwmr->type = req.reg_type;
2882 if (palloc->level != PBLE_LEVEL_0 && iwpbl->pbl_allocated)
2883 irdma_free_pble(iwdev->rf->pble_rsrc, palloc);
2884 ib_umem_release(region);
2887 return ERR_PTR(err);
2891 * irdma_reg_phys_mr - register kernel physical memory
2893 * @addr: physical address of memory to register
2894 * @size: size of memory to register
2895 * @access: Access rights
2896 * @iova_start: start of virtual address for physical buffers
2898 struct ib_mr *irdma_reg_phys_mr(struct ib_pd *pd, u64 addr, u64 size, int access,
2901 struct irdma_device *iwdev = to_iwdev(pd->device);
2902 struct irdma_pbl *iwpbl;
2903 struct irdma_mr *iwmr;
2904 enum irdma_status_code status;
2908 iwmr = kzalloc(sizeof(*iwmr), GFP_KERNEL);
2910 return ERR_PTR(-ENOMEM);
2913 iwmr->ibmr.device = pd->device;
2914 iwpbl = &iwmr->iwpbl;
2916 iwmr->type = IRDMA_MEMREG_TYPE_MEM;
2917 iwpbl->user_base = *iova_start;
2918 stag = irdma_create_stag(iwdev);
2925 iwmr->ibmr.iova = *iova_start;
2926 iwmr->ibmr.rkey = stag;
2927 iwmr->ibmr.lkey = stag;
2929 iwmr->pgaddrmem[0] = addr;
2931 iwmr->page_size = SZ_4K;
2932 status = irdma_hwreg_mr(iwdev, iwmr, access);
2934 irdma_free_stag(iwdev, stag);
2944 return ERR_PTR(ret);
2948 * irdma_get_dma_mr - register physical mem
2950 * @acc: access for memory
2952 static struct ib_mr *irdma_get_dma_mr(struct ib_pd *pd, int acc)
2956 return irdma_reg_phys_mr(pd, 0, 0, acc, &kva);
2960 * irdma_del_memlist - Deleting pbl list entries for CQ/QP
2961 * @iwmr: iwmr for IB's user page addresses
2962 * @ucontext: ptr to user context
2964 static void irdma_del_memlist(struct irdma_mr *iwmr,
2965 struct irdma_ucontext *ucontext)
2967 struct irdma_pbl *iwpbl = &iwmr->iwpbl;
2968 unsigned long flags;
2970 switch (iwmr->type) {
2971 case IRDMA_MEMREG_TYPE_CQ:
2972 spin_lock_irqsave(&ucontext->cq_reg_mem_list_lock, flags);
2973 if (iwpbl->on_list) {
2974 iwpbl->on_list = false;
2975 list_del(&iwpbl->list);
2977 spin_unlock_irqrestore(&ucontext->cq_reg_mem_list_lock, flags);
2979 case IRDMA_MEMREG_TYPE_QP:
2980 spin_lock_irqsave(&ucontext->qp_reg_mem_list_lock, flags);
2981 if (iwpbl->on_list) {
2982 iwpbl->on_list = false;
2983 list_del(&iwpbl->list);
2985 spin_unlock_irqrestore(&ucontext->qp_reg_mem_list_lock, flags);
2993 * irdma_dereg_mr - deregister mr
2994 * @ib_mr: mr ptr for dereg
2997 static int irdma_dereg_mr(struct ib_mr *ib_mr, struct ib_udata *udata)
2999 struct ib_pd *ibpd = ib_mr->pd;
3000 struct irdma_pd *iwpd = to_iwpd(ibpd);
3001 struct irdma_mr *iwmr = to_iwmr(ib_mr);
3002 struct irdma_device *iwdev = to_iwdev(ib_mr->device);
3003 struct irdma_dealloc_stag_info *info;
3004 struct irdma_pbl *iwpbl = &iwmr->iwpbl;
3005 struct irdma_pble_alloc *palloc = &iwpbl->pble_alloc;
3006 struct irdma_cqp_request *cqp_request;
3007 struct cqp_cmds_info *cqp_info;
3009 if (iwmr->type != IRDMA_MEMREG_TYPE_MEM) {
3011 struct irdma_ucontext *ucontext;
3013 ucontext = rdma_udata_to_drv_context(udata,
3014 struct irdma_ucontext,
3016 irdma_del_memlist(iwmr, ucontext);
3021 cqp_request = irdma_alloc_and_get_cqp_request(&iwdev->rf->cqp, true);
3025 cqp_info = &cqp_request->info;
3026 info = &cqp_info->in.u.dealloc_stag.info;
3027 memset(info, 0, sizeof(*info));
3028 info->pd_id = iwpd->sc_pd.pd_id & 0x00007fff;
3029 info->stag_idx = ib_mr->rkey >> IRDMA_CQPSQ_STAG_IDX_S;
3031 if (iwpbl->pbl_allocated)
3032 info->dealloc_pbl = true;
3034 cqp_info->cqp_cmd = IRDMA_OP_DEALLOC_STAG;
3035 cqp_info->post_sq = 1;
3036 cqp_info->in.u.dealloc_stag.dev = &iwdev->rf->sc_dev;
3037 cqp_info->in.u.dealloc_stag.scratch = (uintptr_t)cqp_request;
3038 irdma_handle_cqp_op(iwdev->rf, cqp_request);
3039 irdma_put_cqp_request(&iwdev->rf->cqp, cqp_request);
3040 irdma_free_stag(iwdev, iwmr->stag);
3042 if (iwpbl->pbl_allocated)
3043 irdma_free_pble(iwdev->rf->pble_rsrc, palloc);
3044 ib_umem_release(iwmr->region);
3051 * irdma_copy_sg_list - copy sg list for qp
3052 * @sg_list: copied into sg_list
3053 * @sgl: copy from sgl
3054 * @num_sges: count of sg entries
3056 static void irdma_copy_sg_list(struct irdma_sge *sg_list, struct ib_sge *sgl,
3061 for (i = 0; (i < num_sges) && (i < IRDMA_MAX_WQ_FRAGMENT_COUNT); i++) {
3062 sg_list[i].tag_off = sgl[i].addr;
3063 sg_list[i].len = sgl[i].length;
3064 sg_list[i].stag = sgl[i].lkey;
3069 * irdma_post_send - kernel application wr
3070 * @ibqp: qp ptr for wr
3071 * @ib_wr: work request ptr
3072 * @bad_wr: return of bad wr if err
3074 static int irdma_post_send(struct ib_qp *ibqp,
3075 const struct ib_send_wr *ib_wr,
3076 const struct ib_send_wr **bad_wr)
3078 struct irdma_qp *iwqp;
3079 struct irdma_qp_uk *ukqp;
3080 struct irdma_sc_dev *dev;
3081 struct irdma_post_sq_info info;
3082 enum irdma_status_code ret;
3084 unsigned long flags;
3086 struct irdma_ah *ah;
3087 bool reflush = false;
3089 iwqp = to_iwqp(ibqp);
3090 ukqp = &iwqp->sc_qp.qp_uk;
3091 dev = &iwqp->iwdev->rf->sc_dev;
3093 spin_lock_irqsave(&iwqp->lock, flags);
3094 if (iwqp->flush_issued && ukqp->sq_flush_complete)
3097 memset(&info, 0, sizeof(info));
3099 info.wr_id = (ib_wr->wr_id);
3100 if ((ib_wr->send_flags & IB_SEND_SIGNALED) || iwqp->sig_all)
3101 info.signaled = true;
3102 if (ib_wr->send_flags & IB_SEND_FENCE)
3103 info.read_fence = true;
3104 switch (ib_wr->opcode) {
3105 case IB_WR_SEND_WITH_IMM:
3106 if (ukqp->qp_caps & IRDMA_SEND_WITH_IMM) {
3107 info.imm_data_valid = true;
3108 info.imm_data = ntohl(ib_wr->ex.imm_data);
3115 case IB_WR_SEND_WITH_INV:
3116 if (ib_wr->opcode == IB_WR_SEND ||
3117 ib_wr->opcode == IB_WR_SEND_WITH_IMM) {
3118 if (ib_wr->send_flags & IB_SEND_SOLICITED)
3119 info.op_type = IRDMA_OP_TYPE_SEND_SOL;
3121 info.op_type = IRDMA_OP_TYPE_SEND;
3123 if (ib_wr->send_flags & IB_SEND_SOLICITED)
3124 info.op_type = IRDMA_OP_TYPE_SEND_SOL_INV;
3126 info.op_type = IRDMA_OP_TYPE_SEND_INV;
3127 info.stag_to_inv = ib_wr->ex.invalidate_rkey;
3130 if (ib_wr->send_flags & IB_SEND_INLINE) {
3131 info.op.inline_send.data = (void *)(unsigned long)
3132 ib_wr->sg_list[0].addr;
3133 info.op.inline_send.len = ib_wr->sg_list[0].length;
3134 if (iwqp->ibqp.qp_type == IB_QPT_UD ||
3135 iwqp->ibqp.qp_type == IB_QPT_GSI) {
3136 ah = to_iwah(ud_wr(ib_wr)->ah);
3137 info.op.inline_send.ah_id = ah->sc_ah.ah_info.ah_idx;
3138 info.op.inline_send.qkey = ud_wr(ib_wr)->remote_qkey;
3139 info.op.inline_send.dest_qp = ud_wr(ib_wr)->remote_qpn;
3141 ret = irdma_uk_inline_send(ukqp, &info, false);
3143 info.op.send.num_sges = ib_wr->num_sge;
3144 info.op.send.sg_list = (struct irdma_sge *)
3146 if (iwqp->ibqp.qp_type == IB_QPT_UD ||
3147 iwqp->ibqp.qp_type == IB_QPT_GSI) {
3148 ah = to_iwah(ud_wr(ib_wr)->ah);
3149 info.op.send.ah_id = ah->sc_ah.ah_info.ah_idx;
3150 info.op.send.qkey = ud_wr(ib_wr)->remote_qkey;
3151 info.op.send.dest_qp = ud_wr(ib_wr)->remote_qpn;
3153 ret = irdma_uk_send(ukqp, &info, false);
3157 if (ret == IRDMA_ERR_QP_TOOMANY_WRS_POSTED)
3163 case IB_WR_RDMA_WRITE_WITH_IMM:
3164 if (ukqp->qp_caps & IRDMA_WRITE_WITH_IMM) {
3165 info.imm_data_valid = true;
3166 info.imm_data = ntohl(ib_wr->ex.imm_data);
3172 case IB_WR_RDMA_WRITE:
3173 if (ib_wr->send_flags & IB_SEND_SOLICITED)
3174 info.op_type = IRDMA_OP_TYPE_RDMA_WRITE_SOL;
3176 info.op_type = IRDMA_OP_TYPE_RDMA_WRITE;
3178 if (ib_wr->send_flags & IB_SEND_INLINE) {
3179 info.op.inline_rdma_write.data = (void *)(uintptr_t)ib_wr->sg_list[0].addr;
3180 info.op.inline_rdma_write.len = ib_wr->sg_list[0].length;
3181 info.op.inline_rdma_write.rem_addr.tag_off = rdma_wr(ib_wr)->remote_addr;
3182 info.op.inline_rdma_write.rem_addr.stag = rdma_wr(ib_wr)->rkey;
3183 ret = irdma_uk_inline_rdma_write(ukqp, &info, false);
3185 info.op.rdma_write.lo_sg_list = (void *)ib_wr->sg_list;
3186 info.op.rdma_write.num_lo_sges = ib_wr->num_sge;
3187 info.op.rdma_write.rem_addr.tag_off = rdma_wr(ib_wr)->remote_addr;
3188 info.op.rdma_write.rem_addr.stag = rdma_wr(ib_wr)->rkey;
3189 ret = irdma_uk_rdma_write(ukqp, &info, false);
3193 if (ret == IRDMA_ERR_QP_TOOMANY_WRS_POSTED)
3199 case IB_WR_RDMA_READ_WITH_INV:
3202 case IB_WR_RDMA_READ:
3203 if (ib_wr->num_sge >
3204 dev->hw_attrs.uk_attrs.max_hw_read_sges) {
3208 info.op_type = IRDMA_OP_TYPE_RDMA_READ;
3209 info.op.rdma_read.rem_addr.tag_off = rdma_wr(ib_wr)->remote_addr;
3210 info.op.rdma_read.rem_addr.stag = rdma_wr(ib_wr)->rkey;
3211 info.op.rdma_read.lo_sg_list = (void *)ib_wr->sg_list;
3212 info.op.rdma_read.num_lo_sges = ib_wr->num_sge;
3214 ret = irdma_uk_rdma_read(ukqp, &info, inv_stag, false);
3216 if (ret == IRDMA_ERR_QP_TOOMANY_WRS_POSTED)
3222 case IB_WR_LOCAL_INV:
3223 info.op_type = IRDMA_OP_TYPE_INV_STAG;
3224 info.op.inv_local_stag.target_stag = ib_wr->ex.invalidate_rkey;
3225 ret = irdma_uk_stag_local_invalidate(ukqp, &info, true);
3229 case IB_WR_REG_MR: {
3230 struct irdma_mr *iwmr = to_iwmr(reg_wr(ib_wr)->mr);
3231 struct irdma_pble_alloc *palloc = &iwmr->iwpbl.pble_alloc;
3232 struct irdma_fast_reg_stag_info stag_info = {};
3234 stag_info.signaled = info.signaled;
3235 stag_info.read_fence = info.read_fence;
3236 stag_info.access_rights = irdma_get_mr_access(reg_wr(ib_wr)->access);
3237 stag_info.stag_key = reg_wr(ib_wr)->key & 0xff;
3238 stag_info.stag_idx = reg_wr(ib_wr)->key >> 8;
3239 stag_info.page_size = reg_wr(ib_wr)->mr->page_size;
3240 stag_info.wr_id = ib_wr->wr_id;
3241 stag_info.addr_type = IRDMA_ADDR_TYPE_VA_BASED;
3242 stag_info.va = (void *)(uintptr_t)iwmr->ibmr.iova;
3243 stag_info.total_len = iwmr->ibmr.length;
3244 stag_info.reg_addr_pa = *palloc->level1.addr;
3245 stag_info.first_pm_pbl_index = palloc->level1.idx;
3246 stag_info.local_fence = ib_wr->send_flags & IB_SEND_FENCE;
3247 if (iwmr->npages > IRDMA_MIN_PAGES_PER_FMR)
3248 stag_info.chunk_size = 1;
3249 ret = irdma_sc_mr_fast_register(&iwqp->sc_qp, &stag_info,
3257 ibdev_dbg(&iwqp->iwdev->ibdev,
3258 "VERBS: upost_send bad opcode = 0x%x\n",
3265 ib_wr = ib_wr->next;
3268 if (!iwqp->flush_issued && iwqp->hw_iwarp_state <= IRDMA_QP_STATE_RTS) {
3269 irdma_uk_qp_post_wr(ukqp);
3270 spin_unlock_irqrestore(&iwqp->lock, flags);
3271 } else if (reflush) {
3272 ukqp->sq_flush_complete = false;
3273 spin_unlock_irqrestore(&iwqp->lock, flags);
3274 irdma_flush_wqes(iwqp, IRDMA_FLUSH_SQ | IRDMA_REFLUSH);
3276 spin_unlock_irqrestore(&iwqp->lock, flags);
3285 * irdma_post_recv - post receive wr for kernel application
3286 * @ibqp: ib qp pointer
3287 * @ib_wr: work request for receive
3288 * @bad_wr: bad wr caused an error
3290 static int irdma_post_recv(struct ib_qp *ibqp,
3291 const struct ib_recv_wr *ib_wr,
3292 const struct ib_recv_wr **bad_wr)
3294 struct irdma_qp *iwqp;
3295 struct irdma_qp_uk *ukqp;
3296 struct irdma_post_rq_info post_recv = {};
3297 struct irdma_sge sg_list[IRDMA_MAX_WQ_FRAGMENT_COUNT];
3298 enum irdma_status_code ret = 0;
3299 unsigned long flags;
3301 bool reflush = false;
3303 iwqp = to_iwqp(ibqp);
3304 ukqp = &iwqp->sc_qp.qp_uk;
3306 spin_lock_irqsave(&iwqp->lock, flags);
3307 if (iwqp->flush_issued && ukqp->rq_flush_complete)
3310 post_recv.num_sges = ib_wr->num_sge;
3311 post_recv.wr_id = ib_wr->wr_id;
3312 irdma_copy_sg_list(sg_list, ib_wr->sg_list, ib_wr->num_sge);
3313 post_recv.sg_list = sg_list;
3314 ret = irdma_uk_post_receive(ukqp, &post_recv);
3316 ibdev_dbg(&iwqp->iwdev->ibdev,
3317 "VERBS: post_recv err %d\n", ret);
3318 if (ret == IRDMA_ERR_QP_TOOMANY_WRS_POSTED)
3325 ib_wr = ib_wr->next;
3330 ukqp->rq_flush_complete = false;
3331 spin_unlock_irqrestore(&iwqp->lock, flags);
3332 irdma_flush_wqes(iwqp, IRDMA_FLUSH_RQ | IRDMA_REFLUSH);
3334 spin_unlock_irqrestore(&iwqp->lock, flags);
3344 * irdma_flush_err_to_ib_wc_status - return change flush error code to IB status
3345 * @opcode: iwarp flush code
3347 static enum ib_wc_status irdma_flush_err_to_ib_wc_status(enum irdma_flush_opcode opcode)
3350 case FLUSH_PROT_ERR:
3351 return IB_WC_LOC_PROT_ERR;
3352 case FLUSH_REM_ACCESS_ERR:
3353 return IB_WC_REM_ACCESS_ERR;
3354 case FLUSH_LOC_QP_OP_ERR:
3355 return IB_WC_LOC_QP_OP_ERR;
3356 case FLUSH_REM_OP_ERR:
3357 return IB_WC_REM_OP_ERR;
3358 case FLUSH_LOC_LEN_ERR:
3359 return IB_WC_LOC_LEN_ERR;
3360 case FLUSH_GENERAL_ERR:
3361 return IB_WC_WR_FLUSH_ERR;
3362 case FLUSH_FATAL_ERR:
3364 return IB_WC_FATAL_ERR;
3369 * irdma_process_cqe - process cqe info
3370 * @entry: processed cqe
3371 * @cq_poll_info: cqe info
3373 static void irdma_process_cqe(struct ib_wc *entry,
3374 struct irdma_cq_poll_info *cq_poll_info)
3376 struct irdma_qp *iwqp;
3377 struct irdma_sc_qp *qp;
3379 entry->wc_flags = 0;
3380 entry->pkey_index = 0;
3381 entry->wr_id = cq_poll_info->wr_id;
3383 qp = cq_poll_info->qp_handle;
3384 iwqp = qp->qp_uk.back_qp;
3385 entry->qp = qp->qp_uk.back_qp;
3387 if (cq_poll_info->error) {
3388 entry->status = (cq_poll_info->comp_status == IRDMA_COMPL_STATUS_FLUSHED) ?
3389 irdma_flush_err_to_ib_wc_status(cq_poll_info->minor_err) : IB_WC_GENERAL_ERR;
3391 entry->vendor_err = cq_poll_info->major_err << 16 |
3392 cq_poll_info->minor_err;
3394 entry->status = IB_WC_SUCCESS;
3395 if (cq_poll_info->imm_valid) {
3396 entry->ex.imm_data = htonl(cq_poll_info->imm_data);
3397 entry->wc_flags |= IB_WC_WITH_IMM;
3399 if (cq_poll_info->ud_smac_valid) {
3400 ether_addr_copy(entry->smac, cq_poll_info->ud_smac);
3401 entry->wc_flags |= IB_WC_WITH_SMAC;
3404 if (cq_poll_info->ud_vlan_valid) {
3405 entry->vlan_id = cq_poll_info->ud_vlan & VLAN_VID_MASK;
3406 entry->wc_flags |= IB_WC_WITH_VLAN;
3407 entry->sl = cq_poll_info->ud_vlan >> VLAN_PRIO_SHIFT;
3413 switch (cq_poll_info->op_type) {
3414 case IRDMA_OP_TYPE_RDMA_WRITE:
3415 case IRDMA_OP_TYPE_RDMA_WRITE_SOL:
3416 entry->opcode = IB_WC_RDMA_WRITE;
3418 case IRDMA_OP_TYPE_RDMA_READ_INV_STAG:
3419 case IRDMA_OP_TYPE_RDMA_READ:
3420 entry->opcode = IB_WC_RDMA_READ;
3422 case IRDMA_OP_TYPE_SEND_INV:
3423 case IRDMA_OP_TYPE_SEND_SOL:
3424 case IRDMA_OP_TYPE_SEND_SOL_INV:
3425 case IRDMA_OP_TYPE_SEND:
3426 entry->opcode = IB_WC_SEND;
3428 case IRDMA_OP_TYPE_FAST_REG_NSMR:
3429 entry->opcode = IB_WC_REG_MR;
3431 case IRDMA_OP_TYPE_INV_STAG:
3432 entry->opcode = IB_WC_LOCAL_INV;
3434 case IRDMA_OP_TYPE_REC_IMM:
3435 case IRDMA_OP_TYPE_REC:
3436 entry->opcode = cq_poll_info->op_type == IRDMA_OP_TYPE_REC_IMM ?
3437 IB_WC_RECV_RDMA_WITH_IMM : IB_WC_RECV;
3438 if (qp->qp_uk.qp_type != IRDMA_QP_TYPE_ROCE_UD &&
3439 cq_poll_info->stag_invalid_set) {
3440 entry->ex.invalidate_rkey = cq_poll_info->inv_stag;
3441 entry->wc_flags |= IB_WC_WITH_INVALIDATE;
3445 ibdev_err(&iwqp->iwdev->ibdev,
3446 "Invalid opcode = %d in CQE\n", cq_poll_info->op_type);
3447 entry->status = IB_WC_GENERAL_ERR;
3451 if (qp->qp_uk.qp_type == IRDMA_QP_TYPE_ROCE_UD) {
3452 entry->src_qp = cq_poll_info->ud_src_qpn;
3455 (IB_WC_GRH | IB_WC_WITH_NETWORK_HDR_TYPE);
3456 entry->network_hdr_type = cq_poll_info->ipv4 ?
3460 entry->src_qp = cq_poll_info->qp_id;
3463 entry->byte_len = cq_poll_info->bytes_xfered;
3467 * irdma_poll_one - poll one entry of the CQ
3468 * @ukcq: ukcq to poll
3469 * @cur_cqe: current CQE info to be filled in
3470 * @entry: ibv_wc object to be filled for non-extended CQ or NULL for extended CQ
3472 * Returns the internal irdma device error code or 0 on success
3474 static inline int irdma_poll_one(struct irdma_cq_uk *ukcq,
3475 struct irdma_cq_poll_info *cur_cqe,
3476 struct ib_wc *entry)
3478 int ret = irdma_uk_cq_poll_cmpl(ukcq, cur_cqe);
3483 irdma_process_cqe(entry, cur_cqe);
3489 * __irdma_poll_cq - poll cq for completion (kernel apps)
3491 * @num_entries: number of entries to poll
3492 * @entry: wr of a completed entry
3494 static int __irdma_poll_cq(struct irdma_cq *iwcq, int num_entries, struct ib_wc *entry)
3496 struct list_head *tmp_node, *list_node;
3497 struct irdma_cq_buf *last_buf = NULL;
3498 struct irdma_cq_poll_info *cur_cqe = &iwcq->cur_cqe;
3499 struct irdma_cq_buf *cq_buf;
3500 enum irdma_status_code ret;
3501 struct irdma_device *iwdev;
3502 struct irdma_cq_uk *ukcq;
3503 bool cq_new_cqe = false;
3504 int resized_bufs = 0;
3507 iwdev = to_iwdev(iwcq->ibcq.device);
3508 ukcq = &iwcq->sc_cq.cq_uk;
3510 /* go through the list of previously resized CQ buffers */
3511 list_for_each_safe(list_node, tmp_node, &iwcq->resize_list) {
3512 cq_buf = container_of(list_node, struct irdma_cq_buf, list);
3513 while (npolled < num_entries) {
3514 ret = irdma_poll_one(&cq_buf->cq_uk, cur_cqe, entry + npolled);
3520 if (ret == IRDMA_ERR_Q_EMPTY)
3522 /* QP using the CQ is destroyed. Skip reporting this CQE */
3523 if (ret == IRDMA_ERR_Q_DESTROYED) {
3530 /* save the resized CQ buffer which received the last cqe */
3536 /* check the current CQ for new cqes */
3537 while (npolled < num_entries) {
3538 ret = irdma_poll_one(ukcq, cur_cqe, entry + npolled);
3545 if (ret == IRDMA_ERR_Q_EMPTY)
3547 /* QP using the CQ is destroyed. Skip reporting this CQE */
3548 if (ret == IRDMA_ERR_Q_DESTROYED) {
3556 /* all previous CQ resizes are complete */
3557 resized_bufs = irdma_process_resize_list(iwcq, iwdev, NULL);
3559 /* only CQ resizes up to the last_buf are complete */
3560 resized_bufs = irdma_process_resize_list(iwcq, iwdev, last_buf);
3562 /* report to the HW the number of complete CQ resizes */
3563 irdma_uk_cq_set_resized_cnt(ukcq, resized_bufs);
3567 ibdev_dbg(&iwdev->ibdev, "%s: Error polling CQ, irdma_err: %d\n",
3574 * irdma_poll_cq - poll cq for completion (kernel apps)
3576 * @num_entries: number of entries to poll
3577 * @entry: wr of a completed entry
3579 static int irdma_poll_cq(struct ib_cq *ibcq, int num_entries,
3580 struct ib_wc *entry)
3582 struct irdma_cq *iwcq;
3583 unsigned long flags;
3586 iwcq = to_iwcq(ibcq);
3588 spin_lock_irqsave(&iwcq->lock, flags);
3589 ret = __irdma_poll_cq(iwcq, num_entries, entry);
3590 spin_unlock_irqrestore(&iwcq->lock, flags);
3596 * irdma_req_notify_cq - arm cq kernel application
3598 * @notify_flags: notofication flags
3600 static int irdma_req_notify_cq(struct ib_cq *ibcq,
3601 enum ib_cq_notify_flags notify_flags)
3603 struct irdma_cq *iwcq;
3604 struct irdma_cq_uk *ukcq;
3605 unsigned long flags;
3606 enum irdma_cmpl_notify cq_notify = IRDMA_CQ_COMPL_EVENT;
3608 iwcq = to_iwcq(ibcq);
3609 ukcq = &iwcq->sc_cq.cq_uk;
3610 if (notify_flags == IB_CQ_SOLICITED)
3611 cq_notify = IRDMA_CQ_COMPL_SOLICITED;
3613 spin_lock_irqsave(&iwcq->lock, flags);
3614 irdma_uk_cq_request_notification(ukcq, cq_notify);
3615 spin_unlock_irqrestore(&iwcq->lock, flags);
3620 static int irdma_roce_port_immutable(struct ib_device *ibdev, u32 port_num,
3621 struct ib_port_immutable *immutable)
3623 struct ib_port_attr attr;
3626 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_ROCE_UDP_ENCAP;
3627 err = ib_query_port(ibdev, port_num, &attr);
3631 immutable->max_mad_size = IB_MGMT_MAD_SIZE;
3632 immutable->pkey_tbl_len = attr.pkey_tbl_len;
3633 immutable->gid_tbl_len = attr.gid_tbl_len;
3638 static int irdma_iw_port_immutable(struct ib_device *ibdev, u32 port_num,
3639 struct ib_port_immutable *immutable)
3641 struct ib_port_attr attr;
3644 immutable->core_cap_flags = RDMA_CORE_PORT_IWARP;
3645 err = ib_query_port(ibdev, port_num, &attr);
3648 immutable->gid_tbl_len = attr.gid_tbl_len;
3653 static const char *const irdma_hw_stat_names[] = {
3655 [IRDMA_HW_STAT_INDEX_RXVLANERR] = "rxVlanErrors",
3656 [IRDMA_HW_STAT_INDEX_IP4RXDISCARD] = "ip4InDiscards",
3657 [IRDMA_HW_STAT_INDEX_IP4RXTRUNC] = "ip4InTruncatedPkts",
3658 [IRDMA_HW_STAT_INDEX_IP4TXNOROUTE] = "ip4OutNoRoutes",
3659 [IRDMA_HW_STAT_INDEX_IP6RXDISCARD] = "ip6InDiscards",
3660 [IRDMA_HW_STAT_INDEX_IP6RXTRUNC] = "ip6InTruncatedPkts",
3661 [IRDMA_HW_STAT_INDEX_IP6TXNOROUTE] = "ip6OutNoRoutes",
3662 [IRDMA_HW_STAT_INDEX_TCPRTXSEG] = "tcpRetransSegs",
3663 [IRDMA_HW_STAT_INDEX_TCPRXOPTERR] = "tcpInOptErrors",
3664 [IRDMA_HW_STAT_INDEX_TCPRXPROTOERR] = "tcpInProtoErrors",
3665 [IRDMA_HW_STAT_INDEX_RXRPCNPHANDLED] = "cnpHandled",
3666 [IRDMA_HW_STAT_INDEX_RXRPCNPIGNORED] = "cnpIgnored",
3667 [IRDMA_HW_STAT_INDEX_TXNPCNPSENT] = "cnpSent",
3670 [IRDMA_HW_STAT_INDEX_IP4RXOCTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3672 [IRDMA_HW_STAT_INDEX_IP4RXPKTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3674 [IRDMA_HW_STAT_INDEX_IP4RXFRAGS + IRDMA_HW_STAT_INDEX_MAX_32] =
3676 [IRDMA_HW_STAT_INDEX_IP4RXMCOCTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3678 [IRDMA_HW_STAT_INDEX_IP4RXMCPKTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3680 [IRDMA_HW_STAT_INDEX_IP4TXOCTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3682 [IRDMA_HW_STAT_INDEX_IP4TXPKTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3684 [IRDMA_HW_STAT_INDEX_IP4TXFRAGS + IRDMA_HW_STAT_INDEX_MAX_32] =
3686 [IRDMA_HW_STAT_INDEX_IP4TXMCOCTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3687 "ip4OutMcastOctets",
3688 [IRDMA_HW_STAT_INDEX_IP4TXMCPKTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3690 [IRDMA_HW_STAT_INDEX_IP6RXOCTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3692 [IRDMA_HW_STAT_INDEX_IP6RXPKTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3694 [IRDMA_HW_STAT_INDEX_IP6RXFRAGS + IRDMA_HW_STAT_INDEX_MAX_32] =
3696 [IRDMA_HW_STAT_INDEX_IP6RXMCOCTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3698 [IRDMA_HW_STAT_INDEX_IP6RXMCPKTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3700 [IRDMA_HW_STAT_INDEX_IP6TXOCTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3702 [IRDMA_HW_STAT_INDEX_IP6TXPKTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3704 [IRDMA_HW_STAT_INDEX_IP6TXFRAGS + IRDMA_HW_STAT_INDEX_MAX_32] =
3706 [IRDMA_HW_STAT_INDEX_IP6TXMCOCTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3707 "ip6OutMcastOctets",
3708 [IRDMA_HW_STAT_INDEX_IP6TXMCPKTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3710 [IRDMA_HW_STAT_INDEX_TCPRXSEGS + IRDMA_HW_STAT_INDEX_MAX_32] =
3712 [IRDMA_HW_STAT_INDEX_TCPTXSEG + IRDMA_HW_STAT_INDEX_MAX_32] =
3714 [IRDMA_HW_STAT_INDEX_RDMARXRDS + IRDMA_HW_STAT_INDEX_MAX_32] =
3716 [IRDMA_HW_STAT_INDEX_RDMARXSNDS + IRDMA_HW_STAT_INDEX_MAX_32] =
3718 [IRDMA_HW_STAT_INDEX_RDMARXWRS + IRDMA_HW_STAT_INDEX_MAX_32] =
3720 [IRDMA_HW_STAT_INDEX_RDMATXRDS + IRDMA_HW_STAT_INDEX_MAX_32] =
3722 [IRDMA_HW_STAT_INDEX_RDMATXSNDS + IRDMA_HW_STAT_INDEX_MAX_32] =
3724 [IRDMA_HW_STAT_INDEX_RDMATXWRS + IRDMA_HW_STAT_INDEX_MAX_32] =
3726 [IRDMA_HW_STAT_INDEX_RDMAVBND + IRDMA_HW_STAT_INDEX_MAX_32] =
3728 [IRDMA_HW_STAT_INDEX_RDMAVINV + IRDMA_HW_STAT_INDEX_MAX_32] =
3730 [IRDMA_HW_STAT_INDEX_UDPRXPKTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3732 [IRDMA_HW_STAT_INDEX_UDPTXPKTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3734 [IRDMA_HW_STAT_INDEX_RXNPECNMARKEDPKTS + IRDMA_HW_STAT_INDEX_MAX_32] =
3738 static void irdma_get_dev_fw_str(struct ib_device *dev, char *str)
3740 struct irdma_device *iwdev = to_iwdev(dev);
3742 snprintf(str, IB_FW_VERSION_NAME_MAX, "%u.%u",
3743 irdma_fw_major_ver(&iwdev->rf->sc_dev),
3744 irdma_fw_minor_ver(&iwdev->rf->sc_dev));
3748 * irdma_alloc_hw_port_stats - Allocate a hw stats structure
3749 * @ibdev: device pointer from stack
3750 * @port_num: port number
3752 static struct rdma_hw_stats *irdma_alloc_hw_port_stats(struct ib_device *ibdev,
3755 int num_counters = IRDMA_HW_STAT_INDEX_MAX_32 +
3756 IRDMA_HW_STAT_INDEX_MAX_64;
3757 unsigned long lifespan = RDMA_HW_STATS_DEFAULT_LIFESPAN;
3759 BUILD_BUG_ON(ARRAY_SIZE(irdma_hw_stat_names) !=
3760 (IRDMA_HW_STAT_INDEX_MAX_32 + IRDMA_HW_STAT_INDEX_MAX_64));
3762 return rdma_alloc_hw_stats_struct(irdma_hw_stat_names, num_counters,
3767 * irdma_get_hw_stats - Populates the rdma_hw_stats structure
3768 * @ibdev: device pointer from stack
3769 * @stats: stats pointer from stack
3770 * @port_num: port number
3771 * @index: which hw counter the stack is requesting we update
3773 static int irdma_get_hw_stats(struct ib_device *ibdev,
3774 struct rdma_hw_stats *stats, u32 port_num,
3777 struct irdma_device *iwdev = to_iwdev(ibdev);
3778 struct irdma_dev_hw_stats *hw_stats = &iwdev->vsi.pestat->hw_stats;
3780 if (iwdev->rf->rdma_ver >= IRDMA_GEN_2)
3781 irdma_cqp_gather_stats_cmd(&iwdev->rf->sc_dev, iwdev->vsi.pestat, true);
3783 irdma_cqp_gather_stats_gen1(&iwdev->rf->sc_dev, iwdev->vsi.pestat);
3785 memcpy(&stats->value[0], hw_stats, sizeof(*hw_stats));
3787 return stats->num_counters;
3791 * irdma_query_gid - Query port GID
3792 * @ibdev: device pointer from stack
3793 * @port: port number
3794 * @index: Entry index
3797 static int irdma_query_gid(struct ib_device *ibdev, u32 port, int index,
3800 struct irdma_device *iwdev = to_iwdev(ibdev);
3802 memset(gid->raw, 0, sizeof(gid->raw));
3803 ether_addr_copy(gid->raw, iwdev->netdev->dev_addr);
3809 * mcast_list_add - Add a new mcast item to list
3810 * @rf: RDMA PCI function
3811 * @new_elem: pointer to element to add
3813 static void mcast_list_add(struct irdma_pci_f *rf,
3814 struct mc_table_list *new_elem)
3816 list_add(&new_elem->list, &rf->mc_qht_list.list);
3820 * mcast_list_del - Remove an mcast item from list
3821 * @mc_qht_elem: pointer to mcast table list element
3823 static void mcast_list_del(struct mc_table_list *mc_qht_elem)
3826 list_del(&mc_qht_elem->list);
3830 * mcast_list_lookup_ip - Search mcast list for address
3831 * @rf: RDMA PCI function
3832 * @ip_mcast: pointer to mcast IP address
3834 static struct mc_table_list *mcast_list_lookup_ip(struct irdma_pci_f *rf,
3837 struct mc_table_list *mc_qht_el;
3838 struct list_head *pos, *q;
3840 list_for_each_safe (pos, q, &rf->mc_qht_list.list) {
3841 mc_qht_el = list_entry(pos, struct mc_table_list, list);
3842 if (!memcmp(mc_qht_el->mc_info.dest_ip, ip_mcast,
3843 sizeof(mc_qht_el->mc_info.dest_ip)))
3851 * irdma_mcast_cqp_op - perform a mcast cqp operation
3852 * @iwdev: irdma device
3853 * @mc_grp_ctx: mcast group info
3856 * returns error status
3858 static int irdma_mcast_cqp_op(struct irdma_device *iwdev,
3859 struct irdma_mcast_grp_info *mc_grp_ctx, u8 op)
3861 struct cqp_cmds_info *cqp_info;
3862 struct irdma_cqp_request *cqp_request;
3863 enum irdma_status_code status;
3865 cqp_request = irdma_alloc_and_get_cqp_request(&iwdev->rf->cqp, true);
3869 cqp_request->info.in.u.mc_create.info = *mc_grp_ctx;
3870 cqp_info = &cqp_request->info;
3871 cqp_info->cqp_cmd = op;
3872 cqp_info->post_sq = 1;
3873 cqp_info->in.u.mc_create.scratch = (uintptr_t)cqp_request;
3874 cqp_info->in.u.mc_create.cqp = &iwdev->rf->cqp.sc_cqp;
3875 status = irdma_handle_cqp_op(iwdev->rf, cqp_request);
3876 irdma_put_cqp_request(&iwdev->rf->cqp, cqp_request);
3884 * irdma_mcast_mac - Get the multicast MAC for an IP address
3885 * @ip_addr: IPv4 or IPv6 address
3886 * @mac: pointer to result MAC address
3887 * @ipv4: flag indicating IPv4 or IPv6
3890 void irdma_mcast_mac(u32 *ip_addr, u8 *mac, bool ipv4)
3892 u8 *ip = (u8 *)ip_addr;
3895 unsigned char mac4[ETH_ALEN] = {0x01, 0x00, 0x5E, 0x00,
3898 mac4[3] = ip[2] & 0x7F;
3901 ether_addr_copy(mac, mac4);
3903 unsigned char mac6[ETH_ALEN] = {0x33, 0x33, 0x00, 0x00,
3910 ether_addr_copy(mac, mac6);
3915 * irdma_attach_mcast - attach a qp to a multicast group
3917 * @ibgid: pointer to global ID
3920 * returns error status
3922 static int irdma_attach_mcast(struct ib_qp *ibqp, union ib_gid *ibgid, u16 lid)
3924 struct irdma_qp *iwqp = to_iwqp(ibqp);
3925 struct irdma_device *iwdev = iwqp->iwdev;
3926 struct irdma_pci_f *rf = iwdev->rf;
3927 struct mc_table_list *mc_qht_elem;
3928 struct irdma_mcast_grp_ctx_entry_info mcg_info = {};
3929 unsigned long flags;
3930 u32 ip_addr[4] = {};
3937 struct sockaddr saddr;
3938 struct sockaddr_in saddr_in;
3939 struct sockaddr_in6 saddr_in6;
3941 unsigned char dmac[ETH_ALEN];
3943 rdma_gid2ip((struct sockaddr *)&sgid_addr, ibgid);
3945 if (!ipv6_addr_v4mapped((struct in6_addr *)ibgid)) {
3946 irdma_copy_ip_ntohl(ip_addr,
3947 sgid_addr.saddr_in6.sin6_addr.in6_u.u6_addr32);
3948 irdma_netdev_vlan_ipv6(ip_addr, &vlan_id, NULL);
3950 ibdev_dbg(&iwdev->ibdev,
3951 "VERBS: qp_id=%d, IP6address=%pI6\n", ibqp->qp_num,
3953 irdma_mcast_mac(ip_addr, dmac, false);
3955 ip_addr[0] = ntohl(sgid_addr.saddr_in.sin_addr.s_addr);
3957 vlan_id = irdma_get_vlan_ipv4(ip_addr);
3958 irdma_mcast_mac(ip_addr, dmac, true);
3959 ibdev_dbg(&iwdev->ibdev,
3960 "VERBS: qp_id=%d, IP4address=%pI4, MAC=%pM\n",
3961 ibqp->qp_num, ip_addr, dmac);
3964 spin_lock_irqsave(&rf->qh_list_lock, flags);
3965 mc_qht_elem = mcast_list_lookup_ip(rf, ip_addr);
3967 struct irdma_dma_mem *dma_mem_mc;
3969 spin_unlock_irqrestore(&rf->qh_list_lock, flags);
3970 mc_qht_elem = kzalloc(sizeof(*mc_qht_elem), GFP_KERNEL);
3974 mc_qht_elem->mc_info.ipv4_valid = ipv4;
3975 memcpy(mc_qht_elem->mc_info.dest_ip, ip_addr,
3976 sizeof(mc_qht_elem->mc_info.dest_ip));
3977 ret = irdma_alloc_rsrc(rf, rf->allocated_mcgs, rf->max_mcg,
3978 &mgn, &rf->next_mcg);
3984 mc_qht_elem->mc_info.mgn = mgn;
3985 dma_mem_mc = &mc_qht_elem->mc_grp_ctx.dma_mem_mc;
3986 dma_mem_mc->size = ALIGN(sizeof(u64) * IRDMA_MAX_MGS_PER_CTX,
3987 IRDMA_HW_PAGE_SIZE);
3988 dma_mem_mc->va = dma_alloc_coherent(rf->hw.device,
3992 if (!dma_mem_mc->va) {
3993 irdma_free_rsrc(rf, rf->allocated_mcgs, mgn);
3998 mc_qht_elem->mc_grp_ctx.mg_id = (u16)mgn;
3999 memcpy(mc_qht_elem->mc_grp_ctx.dest_ip_addr, ip_addr,
4000 sizeof(mc_qht_elem->mc_grp_ctx.dest_ip_addr));
4001 mc_qht_elem->mc_grp_ctx.ipv4_valid = ipv4;
4002 mc_qht_elem->mc_grp_ctx.vlan_id = vlan_id;
4003 if (vlan_id < VLAN_N_VID)
4004 mc_qht_elem->mc_grp_ctx.vlan_valid = true;
4005 mc_qht_elem->mc_grp_ctx.hmc_fcn_id = iwdev->vsi.fcn_id;
4006 mc_qht_elem->mc_grp_ctx.qs_handle =
4007 iwqp->sc_qp.vsi->qos[iwqp->sc_qp.user_pri].qs_handle;
4008 ether_addr_copy(mc_qht_elem->mc_grp_ctx.dest_mac_addr, dmac);
4010 spin_lock_irqsave(&rf->qh_list_lock, flags);
4011 mcast_list_add(rf, mc_qht_elem);
4013 if (mc_qht_elem->mc_grp_ctx.no_of_mgs ==
4014 IRDMA_MAX_MGS_PER_CTX) {
4015 spin_unlock_irqrestore(&rf->qh_list_lock, flags);
4020 mcg_info.qp_id = iwqp->ibqp.qp_num;
4021 no_mgs = mc_qht_elem->mc_grp_ctx.no_of_mgs;
4022 irdma_sc_add_mcast_grp(&mc_qht_elem->mc_grp_ctx, &mcg_info);
4023 spin_unlock_irqrestore(&rf->qh_list_lock, flags);
4025 /* Only if there is a change do we need to modify or create */
4027 ret = irdma_mcast_cqp_op(iwdev, &mc_qht_elem->mc_grp_ctx,
4028 IRDMA_OP_MC_CREATE);
4029 } else if (no_mgs != mc_qht_elem->mc_grp_ctx.no_of_mgs) {
4030 ret = irdma_mcast_cqp_op(iwdev, &mc_qht_elem->mc_grp_ctx,
4031 IRDMA_OP_MC_MODIFY);
4042 irdma_sc_del_mcast_grp(&mc_qht_elem->mc_grp_ctx, &mcg_info);
4043 if (!mc_qht_elem->mc_grp_ctx.no_of_mgs) {
4044 mcast_list_del(mc_qht_elem);
4045 dma_free_coherent(rf->hw.device,
4046 mc_qht_elem->mc_grp_ctx.dma_mem_mc.size,
4047 mc_qht_elem->mc_grp_ctx.dma_mem_mc.va,
4048 mc_qht_elem->mc_grp_ctx.dma_mem_mc.pa);
4049 mc_qht_elem->mc_grp_ctx.dma_mem_mc.va = NULL;
4050 irdma_free_rsrc(rf, rf->allocated_mcgs,
4051 mc_qht_elem->mc_grp_ctx.mg_id);
4059 * irdma_detach_mcast - detach a qp from a multicast group
4061 * @ibgid: pointer to global ID
4064 * returns error status
4066 static int irdma_detach_mcast(struct ib_qp *ibqp, union ib_gid *ibgid, u16 lid)
4068 struct irdma_qp *iwqp = to_iwqp(ibqp);
4069 struct irdma_device *iwdev = iwqp->iwdev;
4070 struct irdma_pci_f *rf = iwdev->rf;
4071 u32 ip_addr[4] = {};
4072 struct mc_table_list *mc_qht_elem;
4073 struct irdma_mcast_grp_ctx_entry_info mcg_info = {};
4075 unsigned long flags;
4077 struct sockaddr saddr;
4078 struct sockaddr_in saddr_in;
4079 struct sockaddr_in6 saddr_in6;
4082 rdma_gid2ip((struct sockaddr *)&sgid_addr, ibgid);
4083 if (!ipv6_addr_v4mapped((struct in6_addr *)ibgid))
4084 irdma_copy_ip_ntohl(ip_addr,
4085 sgid_addr.saddr_in6.sin6_addr.in6_u.u6_addr32);
4087 ip_addr[0] = ntohl(sgid_addr.saddr_in.sin_addr.s_addr);
4089 spin_lock_irqsave(&rf->qh_list_lock, flags);
4090 mc_qht_elem = mcast_list_lookup_ip(rf, ip_addr);
4092 spin_unlock_irqrestore(&rf->qh_list_lock, flags);
4093 ibdev_dbg(&iwdev->ibdev,
4094 "VERBS: address not found MCG\n");
4098 mcg_info.qp_id = iwqp->ibqp.qp_num;
4099 irdma_sc_del_mcast_grp(&mc_qht_elem->mc_grp_ctx, &mcg_info);
4100 if (!mc_qht_elem->mc_grp_ctx.no_of_mgs) {
4101 mcast_list_del(mc_qht_elem);
4102 spin_unlock_irqrestore(&rf->qh_list_lock, flags);
4103 ret = irdma_mcast_cqp_op(iwdev, &mc_qht_elem->mc_grp_ctx,
4104 IRDMA_OP_MC_DESTROY);
4106 ibdev_dbg(&iwdev->ibdev,
4107 "VERBS: failed MC_DESTROY MCG\n");
4108 spin_lock_irqsave(&rf->qh_list_lock, flags);
4109 mcast_list_add(rf, mc_qht_elem);
4110 spin_unlock_irqrestore(&rf->qh_list_lock, flags);
4114 dma_free_coherent(rf->hw.device,
4115 mc_qht_elem->mc_grp_ctx.dma_mem_mc.size,
4116 mc_qht_elem->mc_grp_ctx.dma_mem_mc.va,
4117 mc_qht_elem->mc_grp_ctx.dma_mem_mc.pa);
4118 mc_qht_elem->mc_grp_ctx.dma_mem_mc.va = NULL;
4119 irdma_free_rsrc(rf, rf->allocated_mcgs,
4120 mc_qht_elem->mc_grp_ctx.mg_id);
4123 spin_unlock_irqrestore(&rf->qh_list_lock, flags);
4124 ret = irdma_mcast_cqp_op(iwdev, &mc_qht_elem->mc_grp_ctx,
4125 IRDMA_OP_MC_MODIFY);
4127 ibdev_dbg(&iwdev->ibdev,
4128 "VERBS: failed Modify MCG\n");
4137 * irdma_create_ah - create address handle
4138 * @ibah: address handle
4139 * @attr: address handle attributes
4142 * returns 0 on success, error otherwise
4144 static int irdma_create_ah(struct ib_ah *ibah,
4145 struct rdma_ah_init_attr *attr,
4146 struct ib_udata *udata)
4148 struct irdma_pd *pd = to_iwpd(ibah->pd);
4149 struct irdma_ah *ah = container_of(ibah, struct irdma_ah, ibah);
4150 struct rdma_ah_attr *ah_attr = attr->ah_attr;
4151 const struct ib_gid_attr *sgid_attr;
4152 struct irdma_device *iwdev = to_iwdev(ibah->pd->device);
4153 struct irdma_pci_f *rf = iwdev->rf;
4154 struct irdma_sc_ah *sc_ah;
4156 struct irdma_ah_info *ah_info;
4157 struct irdma_create_ah_resp uresp;
4159 struct sockaddr saddr;
4160 struct sockaddr_in saddr_in;
4161 struct sockaddr_in6 saddr_in6;
4162 } sgid_addr, dgid_addr;
4166 err = irdma_alloc_rsrc(rf, rf->allocated_ahs, rf->max_ah, &ah_id,
4173 sc_ah->ah_info.ah_idx = ah_id;
4174 sc_ah->ah_info.vsi = &iwdev->vsi;
4175 irdma_sc_init_ah(&rf->sc_dev, sc_ah);
4176 ah->sgid_index = ah_attr->grh.sgid_index;
4177 sgid_attr = ah_attr->grh.sgid_attr;
4178 memcpy(&ah->dgid, &ah_attr->grh.dgid, sizeof(ah->dgid));
4179 rdma_gid2ip((struct sockaddr *)&sgid_addr, &sgid_attr->gid);
4180 rdma_gid2ip((struct sockaddr *)&dgid_addr, &ah_attr->grh.dgid);
4181 ah->av.attrs = *ah_attr;
4182 ah->av.net_type = rdma_gid_attr_network_type(sgid_attr);
4183 ah->av.sgid_addr.saddr = sgid_addr.saddr;
4184 ah->av.dgid_addr.saddr = dgid_addr.saddr;
4185 ah_info = &sc_ah->ah_info;
4186 ah_info->ah_idx = ah_id;
4187 ah_info->pd_idx = pd->sc_pd.pd_id;
4188 if (ah_attr->ah_flags & IB_AH_GRH) {
4189 ah_info->flow_label = ah_attr->grh.flow_label;
4190 ah_info->hop_ttl = ah_attr->grh.hop_limit;
4191 ah_info->tc_tos = ah_attr->grh.traffic_class;
4194 ether_addr_copy(dmac, ah_attr->roce.dmac);
4195 if (rdma_gid_attr_network_type(sgid_attr) == RDMA_NETWORK_IPV4) {
4196 ah_info->ipv4_valid = true;
4197 ah_info->dest_ip_addr[0] =
4198 ntohl(dgid_addr.saddr_in.sin_addr.s_addr);
4199 ah_info->src_ip_addr[0] =
4200 ntohl(sgid_addr.saddr_in.sin_addr.s_addr);
4201 ah_info->do_lpbk = irdma_ipv4_is_lpb(ah_info->src_ip_addr[0],
4202 ah_info->dest_ip_addr[0]);
4203 if (ipv4_is_multicast(dgid_addr.saddr_in.sin_addr.s_addr)) {
4204 ah_info->do_lpbk = true;
4205 irdma_mcast_mac(ah_info->dest_ip_addr, dmac, true);
4208 irdma_copy_ip_ntohl(ah_info->dest_ip_addr,
4209 dgid_addr.saddr_in6.sin6_addr.in6_u.u6_addr32);
4210 irdma_copy_ip_ntohl(ah_info->src_ip_addr,
4211 sgid_addr.saddr_in6.sin6_addr.in6_u.u6_addr32);
4212 ah_info->do_lpbk = irdma_ipv6_is_lpb(ah_info->src_ip_addr,
4213 ah_info->dest_ip_addr);
4214 if (rdma_is_multicast_addr(&dgid_addr.saddr_in6.sin6_addr)) {
4215 ah_info->do_lpbk = true;
4216 irdma_mcast_mac(ah_info->dest_ip_addr, dmac, false);
4220 err = rdma_read_gid_l2_fields(sgid_attr, &ah_info->vlan_tag,
4225 ah_info->dst_arpindex = irdma_add_arp(iwdev->rf, ah_info->dest_ip_addr,
4226 ah_info->ipv4_valid, dmac);
4228 if (ah_info->dst_arpindex == -1) {
4233 if (ah_info->vlan_tag >= VLAN_N_VID && iwdev->dcb)
4234 ah_info->vlan_tag = 0;
4236 if (ah_info->vlan_tag < VLAN_N_VID) {
4237 ah_info->insert_vlan_tag = true;
4238 ah_info->vlan_tag |=
4239 rt_tos2priority(ah_info->tc_tos) << VLAN_PRIO_SHIFT;
4242 err = irdma_ah_cqp_op(iwdev->rf, sc_ah, IRDMA_OP_AH_CREATE,
4243 attr->flags & RDMA_CREATE_AH_SLEEPABLE,
4244 irdma_gsi_ud_qp_ah_cb, sc_ah);
4247 ibdev_dbg(&iwdev->ibdev,
4248 "VERBS: CQP-OP Create AH fail");
4252 if (!(attr->flags & RDMA_CREATE_AH_SLEEPABLE)) {
4253 int cnt = CQP_COMPL_WAIT_TIME_MS * CQP_TIMEOUT_THRESHOLD;
4256 irdma_cqp_ce_handler(rf, &rf->ccq.sc_cq);
4258 } while (!sc_ah->ah_info.ah_valid && --cnt);
4261 ibdev_dbg(&iwdev->ibdev,
4262 "VERBS: CQP create AH timed out");
4269 uresp.ah_id = ah->sc_ah.ah_info.ah_idx;
4270 err = ib_copy_to_udata(udata, &uresp,
4271 min(sizeof(uresp), udata->outlen));
4276 irdma_free_rsrc(iwdev->rf, iwdev->rf->allocated_ahs, ah_id);
4282 * irdma_destroy_ah - Destroy address handle
4283 * @ibah: pointer to address handle
4284 * @ah_flags: flags for sleepable
4286 static int irdma_destroy_ah(struct ib_ah *ibah, u32 ah_flags)
4288 struct irdma_device *iwdev = to_iwdev(ibah->device);
4289 struct irdma_ah *ah = to_iwah(ibah);
4291 irdma_ah_cqp_op(iwdev->rf, &ah->sc_ah, IRDMA_OP_AH_DESTROY,
4294 irdma_free_rsrc(iwdev->rf, iwdev->rf->allocated_ahs,
4295 ah->sc_ah.ah_info.ah_idx);
4301 * irdma_query_ah - Query address handle
4302 * @ibah: pointer to address handle
4303 * @ah_attr: address handle attributes
4305 static int irdma_query_ah(struct ib_ah *ibah, struct rdma_ah_attr *ah_attr)
4307 struct irdma_ah *ah = to_iwah(ibah);
4309 memset(ah_attr, 0, sizeof(*ah_attr));
4310 if (ah->av.attrs.ah_flags & IB_AH_GRH) {
4311 ah_attr->ah_flags = IB_AH_GRH;
4312 ah_attr->grh.flow_label = ah->sc_ah.ah_info.flow_label;
4313 ah_attr->grh.traffic_class = ah->sc_ah.ah_info.tc_tos;
4314 ah_attr->grh.hop_limit = ah->sc_ah.ah_info.hop_ttl;
4315 ah_attr->grh.sgid_index = ah->sgid_index;
4316 ah_attr->grh.sgid_index = ah->sgid_index;
4317 memcpy(&ah_attr->grh.dgid, &ah->dgid,
4318 sizeof(ah_attr->grh.dgid));
4324 static enum rdma_link_layer irdma_get_link_layer(struct ib_device *ibdev,
4327 return IB_LINK_LAYER_ETHERNET;
4330 static __be64 irdma_mac_to_guid(struct net_device *ndev)
4332 unsigned char *mac = ndev->dev_addr;
4334 unsigned char *dst = (unsigned char *)&guid;
4336 dst[0] = mac[0] ^ 2;
4348 static const struct ib_device_ops irdma_roce_dev_ops = {
4349 .attach_mcast = irdma_attach_mcast,
4350 .create_ah = irdma_create_ah,
4351 .create_user_ah = irdma_create_ah,
4352 .destroy_ah = irdma_destroy_ah,
4353 .detach_mcast = irdma_detach_mcast,
4354 .get_link_layer = irdma_get_link_layer,
4355 .get_port_immutable = irdma_roce_port_immutable,
4356 .modify_qp = irdma_modify_qp_roce,
4357 .query_ah = irdma_query_ah,
4358 .query_pkey = irdma_query_pkey,
4361 static const struct ib_device_ops irdma_iw_dev_ops = {
4362 .modify_qp = irdma_modify_qp,
4363 .get_port_immutable = irdma_iw_port_immutable,
4364 .query_gid = irdma_query_gid,
4367 static const struct ib_device_ops irdma_dev_ops = {
4368 .owner = THIS_MODULE,
4369 .driver_id = RDMA_DRIVER_IRDMA,
4370 .uverbs_abi_ver = IRDMA_ABI_VER,
4372 .alloc_hw_port_stats = irdma_alloc_hw_port_stats,
4373 .alloc_mr = irdma_alloc_mr,
4374 .alloc_mw = irdma_alloc_mw,
4375 .alloc_pd = irdma_alloc_pd,
4376 .alloc_ucontext = irdma_alloc_ucontext,
4377 .create_cq = irdma_create_cq,
4378 .create_qp = irdma_create_qp,
4379 .dealloc_driver = irdma_ib_dealloc_device,
4380 .dealloc_mw = irdma_dealloc_mw,
4381 .dealloc_pd = irdma_dealloc_pd,
4382 .dealloc_ucontext = irdma_dealloc_ucontext,
4383 .dereg_mr = irdma_dereg_mr,
4384 .destroy_cq = irdma_destroy_cq,
4385 .destroy_qp = irdma_destroy_qp,
4386 .disassociate_ucontext = irdma_disassociate_ucontext,
4387 .get_dev_fw_str = irdma_get_dev_fw_str,
4388 .get_dma_mr = irdma_get_dma_mr,
4389 .get_hw_stats = irdma_get_hw_stats,
4390 .map_mr_sg = irdma_map_mr_sg,
4392 .mmap_free = irdma_mmap_free,
4393 .poll_cq = irdma_poll_cq,
4394 .post_recv = irdma_post_recv,
4395 .post_send = irdma_post_send,
4396 .query_device = irdma_query_device,
4397 .query_port = irdma_query_port,
4398 .query_qp = irdma_query_qp,
4399 .reg_user_mr = irdma_reg_user_mr,
4400 .req_notify_cq = irdma_req_notify_cq,
4401 .resize_cq = irdma_resize_cq,
4402 INIT_RDMA_OBJ_SIZE(ib_pd, irdma_pd, ibpd),
4403 INIT_RDMA_OBJ_SIZE(ib_ucontext, irdma_ucontext, ibucontext),
4404 INIT_RDMA_OBJ_SIZE(ib_ah, irdma_ah, ibah),
4405 INIT_RDMA_OBJ_SIZE(ib_cq, irdma_cq, ibcq),
4406 INIT_RDMA_OBJ_SIZE(ib_mw, irdma_mr, ibmw),
4410 * irdma_init_roce_device - initialization of roce rdma device
4411 * @iwdev: irdma device
4413 static void irdma_init_roce_device(struct irdma_device *iwdev)
4415 iwdev->ibdev.node_type = RDMA_NODE_IB_CA;
4416 iwdev->ibdev.node_guid = irdma_mac_to_guid(iwdev->netdev);
4417 ib_set_device_ops(&iwdev->ibdev, &irdma_roce_dev_ops);
4421 * irdma_init_iw_device - initialization of iwarp rdma device
4422 * @iwdev: irdma device
4424 static int irdma_init_iw_device(struct irdma_device *iwdev)
4426 struct net_device *netdev = iwdev->netdev;
4428 iwdev->ibdev.node_type = RDMA_NODE_RNIC;
4429 ether_addr_copy((u8 *)&iwdev->ibdev.node_guid, netdev->dev_addr);
4430 iwdev->ibdev.ops.iw_add_ref = irdma_qp_add_ref;
4431 iwdev->ibdev.ops.iw_rem_ref = irdma_qp_rem_ref;
4432 iwdev->ibdev.ops.iw_get_qp = irdma_get_qp;
4433 iwdev->ibdev.ops.iw_connect = irdma_connect;
4434 iwdev->ibdev.ops.iw_accept = irdma_accept;
4435 iwdev->ibdev.ops.iw_reject = irdma_reject;
4436 iwdev->ibdev.ops.iw_create_listen = irdma_create_listen;
4437 iwdev->ibdev.ops.iw_destroy_listen = irdma_destroy_listen;
4438 memcpy(iwdev->ibdev.iw_ifname, netdev->name,
4439 sizeof(iwdev->ibdev.iw_ifname));
4440 ib_set_device_ops(&iwdev->ibdev, &irdma_iw_dev_ops);
4446 * irdma_init_rdma_device - initialization of rdma device
4447 * @iwdev: irdma device
4449 static int irdma_init_rdma_device(struct irdma_device *iwdev)
4451 struct pci_dev *pcidev = iwdev->rf->pcidev;
4454 if (iwdev->roce_mode) {
4455 irdma_init_roce_device(iwdev);
4457 ret = irdma_init_iw_device(iwdev);
4461 iwdev->ibdev.phys_port_cnt = 1;
4462 iwdev->ibdev.num_comp_vectors = iwdev->rf->ceqs_count;
4463 iwdev->ibdev.dev.parent = &pcidev->dev;
4464 ib_set_device_ops(&iwdev->ibdev, &irdma_dev_ops);
4470 * irdma_port_ibevent - indicate port event
4471 * @iwdev: irdma device
4473 void irdma_port_ibevent(struct irdma_device *iwdev)
4475 struct ib_event event;
4477 event.device = &iwdev->ibdev;
4478 event.element.port_num = 1;
4480 iwdev->iw_status ? IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
4481 ib_dispatch_event(&event);
4485 * irdma_ib_unregister_device - unregister rdma device from IB
4487 * @iwdev: irdma device
4489 void irdma_ib_unregister_device(struct irdma_device *iwdev)
4491 iwdev->iw_status = 0;
4492 irdma_port_ibevent(iwdev);
4493 ib_unregister_device(&iwdev->ibdev);
4497 * irdma_ib_register_device - register irdma device to IB core
4498 * @iwdev: irdma device
4500 int irdma_ib_register_device(struct irdma_device *iwdev)
4504 ret = irdma_init_rdma_device(iwdev);
4508 ret = ib_device_set_netdev(&iwdev->ibdev, iwdev->netdev, 1);
4511 dma_set_max_seg_size(iwdev->rf->hw.device, UINT_MAX);
4512 ret = ib_register_device(&iwdev->ibdev, "irdma%d", iwdev->rf->hw.device);
4516 iwdev->iw_status = 1;
4517 irdma_port_ibevent(iwdev);
4523 ibdev_dbg(&iwdev->ibdev, "VERBS: Register RDMA device fail\n");
4529 * irdma_ib_dealloc_device
4532 * callback from ibdev dealloc_driver to deallocate resources
4533 * unber irdma device
4535 void irdma_ib_dealloc_device(struct ib_device *ibdev)
4537 struct irdma_device *iwdev = to_iwdev(ibdev);
4539 irdma_rt_deinit_hw(iwdev);
4540 irdma_ctrl_deinit_hw(iwdev->rf);