2 * Copyright (c) 2004, 2005 Topspin Communications. All rights reserved.
3 * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
4 * Copyright (c) 2005, 2006 Cisco Systems. All rights reserved.
5 * Copyright (c) 2005 Mellanox Technologies. All rights reserved.
6 * Copyright (c) 2004 Voltaire, Inc. All rights reserved.
8 * This software is available to you under a choice of one of two
9 * licenses. You may choose to be licensed under the terms of the GNU
10 * General Public License (GPL) Version 2, available from the file
11 * COPYING in the main directory of this source tree, or the
12 * OpenIB.org BSD license below:
14 * Redistribution and use in source and binary forms, with or
15 * without modification, are permitted provided that the following
18 * - Redistributions of source code must retain the above
19 * copyright notice, this list of conditions and the following
22 * - Redistributions in binary form must reproduce the above
23 * copyright notice, this list of conditions and the following
24 * disclaimer in the documentation and/or other materials
25 * provided with the distribution.
27 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
28 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
29 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
30 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
31 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
32 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
33 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
37 #include <rdma/ib_smi.h>
38 #include <rdma/ib_umem.h>
39 #include <rdma/ib_user_verbs.h>
40 #include <rdma/uverbs_ioctl.h>
42 #include <linux/sched.h>
43 #include <linux/slab.h>
47 #include "mthca_dev.h"
48 #include "mthca_cmd.h"
49 #include <rdma/mthca-abi.h>
50 #include "mthca_memfree.h"
52 static void init_query_mad(struct ib_smp *mad)
54 mad->base_version = 1;
55 mad->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED;
56 mad->class_version = 1;
57 mad->method = IB_MGMT_METHOD_GET;
60 static int mthca_query_device(struct ib_device *ibdev, struct ib_device_attr *props,
63 struct ib_smp *in_mad = NULL;
64 struct ib_smp *out_mad = NULL;
66 struct mthca_dev *mdev = to_mdev(ibdev);
68 if (uhw->inlen || uhw->outlen)
71 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
72 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
73 if (!in_mad || !out_mad)
76 memset(props, 0, sizeof *props);
78 props->fw_ver = mdev->fw_ver;
80 init_query_mad(in_mad);
81 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
83 err = mthca_MAD_IFC(mdev, 1, 1,
84 1, NULL, NULL, in_mad, out_mad);
88 props->device_cap_flags = mdev->device_cap_flags;
89 props->vendor_id = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
91 props->vendor_part_id = be16_to_cpup((__be16 *) (out_mad->data + 30));
92 props->hw_ver = be32_to_cpup((__be32 *) (out_mad->data + 32));
93 memcpy(&props->sys_image_guid, out_mad->data + 4, 8);
95 props->max_mr_size = ~0ull;
96 props->page_size_cap = mdev->limits.page_size_cap;
97 props->max_qp = mdev->limits.num_qps - mdev->limits.reserved_qps;
98 props->max_qp_wr = mdev->limits.max_wqes;
99 props->max_sge = mdev->limits.max_sg;
100 props->max_sge_rd = props->max_sge;
101 props->max_cq = mdev->limits.num_cqs - mdev->limits.reserved_cqs;
102 props->max_cqe = mdev->limits.max_cqes;
103 props->max_mr = mdev->limits.num_mpts - mdev->limits.reserved_mrws;
104 props->max_pd = mdev->limits.num_pds - mdev->limits.reserved_pds;
105 props->max_qp_rd_atom = 1 << mdev->qp_table.rdb_shift;
106 props->max_qp_init_rd_atom = mdev->limits.max_qp_init_rdma;
107 props->max_res_rd_atom = props->max_qp_rd_atom * props->max_qp;
108 props->max_srq = mdev->limits.num_srqs - mdev->limits.reserved_srqs;
109 props->max_srq_wr = mdev->limits.max_srq_wqes;
110 props->max_srq_sge = mdev->limits.max_srq_sge;
111 props->local_ca_ack_delay = mdev->limits.local_ca_ack_delay;
112 props->atomic_cap = mdev->limits.flags & DEV_LIM_FLAG_ATOMIC ?
113 IB_ATOMIC_HCA : IB_ATOMIC_NONE;
114 props->max_pkeys = mdev->limits.pkey_table_len;
115 props->max_mcast_grp = mdev->limits.num_mgms + mdev->limits.num_amgms;
116 props->max_mcast_qp_attach = MTHCA_QP_PER_MGM;
117 props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
118 props->max_mcast_grp;
120 * If Sinai memory key optimization is being used, then only
121 * the 8-bit key portion will change. For other HCAs, the
122 * unused index bits will also be used for FMR remapping.
124 if (mdev->mthca_flags & MTHCA_FLAG_SINAI_OPT)
125 props->max_map_per_fmr = 255;
127 props->max_map_per_fmr =
128 (1 << (32 - ilog2(mdev->limits.num_mpts))) - 1;
137 static int mthca_query_port(struct ib_device *ibdev,
138 u8 port, struct ib_port_attr *props)
140 struct ib_smp *in_mad = NULL;
141 struct ib_smp *out_mad = NULL;
144 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
145 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
146 if (!in_mad || !out_mad)
149 memset(props, 0, sizeof *props);
151 init_query_mad(in_mad);
152 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
153 in_mad->attr_mod = cpu_to_be32(port);
155 err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
156 port, NULL, NULL, in_mad, out_mad);
160 props->lid = be16_to_cpup((__be16 *) (out_mad->data + 16));
161 props->lmc = out_mad->data[34] & 0x7;
162 props->sm_lid = be16_to_cpup((__be16 *) (out_mad->data + 18));
163 props->sm_sl = out_mad->data[36] & 0xf;
164 props->state = out_mad->data[32] & 0xf;
165 props->phys_state = out_mad->data[33] >> 4;
166 props->port_cap_flags = be32_to_cpup((__be32 *) (out_mad->data + 20));
167 props->gid_tbl_len = to_mdev(ibdev)->limits.gid_table_len;
168 props->max_msg_sz = 0x80000000;
169 props->pkey_tbl_len = to_mdev(ibdev)->limits.pkey_table_len;
170 props->bad_pkey_cntr = be16_to_cpup((__be16 *) (out_mad->data + 46));
171 props->qkey_viol_cntr = be16_to_cpup((__be16 *) (out_mad->data + 48));
172 props->active_width = out_mad->data[31] & 0xf;
173 props->active_speed = out_mad->data[35] >> 4;
174 props->max_mtu = out_mad->data[41] & 0xf;
175 props->active_mtu = out_mad->data[36] >> 4;
176 props->subnet_timeout = out_mad->data[51] & 0x1f;
177 props->max_vl_num = out_mad->data[37] >> 4;
178 props->init_type_reply = out_mad->data[41] >> 4;
186 static int mthca_modify_device(struct ib_device *ibdev,
188 struct ib_device_modify *props)
190 if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
193 if (mask & IB_DEVICE_MODIFY_NODE_DESC) {
194 if (mutex_lock_interruptible(&to_mdev(ibdev)->cap_mask_mutex))
196 memcpy(ibdev->node_desc, props->node_desc,
197 IB_DEVICE_NODE_DESC_MAX);
198 mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
204 static int mthca_modify_port(struct ib_device *ibdev,
205 u8 port, int port_modify_mask,
206 struct ib_port_modify *props)
208 struct mthca_set_ib_param set_ib;
209 struct ib_port_attr attr;
212 if (mutex_lock_interruptible(&to_mdev(ibdev)->cap_mask_mutex))
215 err = mthca_query_port(ibdev, port, &attr);
219 set_ib.set_si_guid = 0;
220 set_ib.reset_qkey_viol = !!(port_modify_mask & IB_PORT_RESET_QKEY_CNTR);
222 set_ib.cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
223 ~props->clr_port_cap_mask;
225 err = mthca_SET_IB(to_mdev(ibdev), &set_ib, port);
229 mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
233 static int mthca_query_pkey(struct ib_device *ibdev,
234 u8 port, u16 index, u16 *pkey)
236 struct ib_smp *in_mad = NULL;
237 struct ib_smp *out_mad = NULL;
240 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
241 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
242 if (!in_mad || !out_mad)
245 init_query_mad(in_mad);
246 in_mad->attr_id = IB_SMP_ATTR_PKEY_TABLE;
247 in_mad->attr_mod = cpu_to_be32(index / 32);
249 err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
250 port, NULL, NULL, in_mad, out_mad);
254 *pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
262 static int mthca_query_gid(struct ib_device *ibdev, u8 port,
263 int index, union ib_gid *gid)
265 struct ib_smp *in_mad = NULL;
266 struct ib_smp *out_mad = NULL;
269 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
270 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
271 if (!in_mad || !out_mad)
274 init_query_mad(in_mad);
275 in_mad->attr_id = IB_SMP_ATTR_PORT_INFO;
276 in_mad->attr_mod = cpu_to_be32(port);
278 err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
279 port, NULL, NULL, in_mad, out_mad);
283 memcpy(gid->raw, out_mad->data + 8, 8);
285 init_query_mad(in_mad);
286 in_mad->attr_id = IB_SMP_ATTR_GUID_INFO;
287 in_mad->attr_mod = cpu_to_be32(index / 8);
289 err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
290 port, NULL, NULL, in_mad, out_mad);
294 memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
302 static int mthca_alloc_ucontext(struct ib_ucontext *uctx,
303 struct ib_udata *udata)
305 struct ib_device *ibdev = uctx->device;
306 struct mthca_alloc_ucontext_resp uresp = {};
307 struct mthca_ucontext *context = to_mucontext(uctx);
310 if (!(to_mdev(ibdev)->active))
313 uresp.qp_tab_size = to_mdev(ibdev)->limits.num_qps;
314 if (mthca_is_memfree(to_mdev(ibdev)))
315 uresp.uarc_size = to_mdev(ibdev)->uar_table.uarc_size;
319 err = mthca_uar_alloc(to_mdev(ibdev), &context->uar);
323 context->db_tab = mthca_init_user_db_tab(to_mdev(ibdev));
324 if (IS_ERR(context->db_tab)) {
325 err = PTR_ERR(context->db_tab);
326 mthca_uar_free(to_mdev(ibdev), &context->uar);
330 if (ib_copy_to_udata(udata, &uresp, sizeof(uresp))) {
331 mthca_cleanup_user_db_tab(to_mdev(ibdev), &context->uar, context->db_tab);
332 mthca_uar_free(to_mdev(ibdev), &context->uar);
336 context->reg_mr_warned = 0;
341 static void mthca_dealloc_ucontext(struct ib_ucontext *context)
343 mthca_cleanup_user_db_tab(to_mdev(context->device), &to_mucontext(context)->uar,
344 to_mucontext(context)->db_tab);
345 mthca_uar_free(to_mdev(context->device), &to_mucontext(context)->uar);
348 static int mthca_mmap_uar(struct ib_ucontext *context,
349 struct vm_area_struct *vma)
351 if (vma->vm_end - vma->vm_start != PAGE_SIZE)
354 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
356 if (io_remap_pfn_range(vma, vma->vm_start,
357 to_mucontext(context)->uar.pfn,
358 PAGE_SIZE, vma->vm_page_prot))
364 static int mthca_alloc_pd(struct ib_pd *ibpd, struct ib_udata *udata)
366 struct ib_device *ibdev = ibpd->device;
367 struct mthca_pd *pd = to_mpd(ibpd);
370 err = mthca_pd_alloc(to_mdev(ibdev), !udata, pd);
375 if (ib_copy_to_udata(udata, &pd->pd_num, sizeof (__u32))) {
376 mthca_pd_free(to_mdev(ibdev), pd);
384 static void mthca_dealloc_pd(struct ib_pd *pd, struct ib_udata *udata)
386 mthca_pd_free(to_mdev(pd->device), to_mpd(pd));
389 static int mthca_ah_create(struct ib_ah *ibah,
390 struct ib_ah_attr *init_attr, u32 flags,
391 struct ib_udata *udata)
394 struct mthca_ah *ah = to_mah(ibah);
396 return mthca_create_ah(to_mdev(ibah->device), to_mpd(ibah->pd),
400 static void mthca_ah_destroy(struct ib_ah *ah, u32 flags)
402 mthca_destroy_ah(to_mdev(ah->device), to_mah(ah));
405 static int mthca_create_srq(struct ib_srq *ibsrq,
406 struct ib_srq_init_attr *init_attr,
407 struct ib_udata *udata)
409 struct mthca_create_srq ucmd;
410 struct mthca_ucontext *context = rdma_udata_to_drv_context(
411 udata, struct mthca_ucontext, ibucontext);
412 struct mthca_srq *srq = to_msrq(ibsrq);
415 if (init_attr->srq_type != IB_SRQT_BASIC)
419 if (ib_copy_from_udata(&ucmd, udata, sizeof(ucmd)))
422 err = mthca_map_user_db(to_mdev(ibsrq->device), &context->uar,
423 context->db_tab, ucmd.db_index,
429 srq->mr.ibmr.lkey = ucmd.lkey;
430 srq->db_index = ucmd.db_index;
433 err = mthca_alloc_srq(to_mdev(ibsrq->device), to_mpd(ibsrq->pd),
434 &init_attr->attr, srq, udata);
437 mthca_unmap_user_db(to_mdev(ibsrq->device), &context->uar,
438 context->db_tab, ucmd.db_index);
443 if (context && ib_copy_to_udata(udata, &srq->srqn, sizeof(__u32))) {
444 mthca_free_srq(to_mdev(ibsrq->device), srq);
451 static void mthca_destroy_srq(struct ib_srq *srq, struct ib_udata *udata)
454 struct mthca_ucontext *context =
455 rdma_udata_to_drv_context(
457 struct mthca_ucontext,
460 mthca_unmap_user_db(to_mdev(srq->device), &context->uar,
461 context->db_tab, to_msrq(srq)->db_index);
464 mthca_free_srq(to_mdev(srq->device), to_msrq(srq));
467 static struct ib_qp *mthca_create_qp(struct ib_pd *pd,
468 struct ib_qp_init_attr *init_attr,
469 struct ib_udata *udata)
471 struct mthca_ucontext *context = rdma_udata_to_drv_context(
472 udata, struct mthca_ucontext, ibucontext);
473 struct mthca_create_qp ucmd;
477 if (init_attr->create_flags)
478 return ERR_PTR(-EINVAL);
480 switch (init_attr->qp_type) {
485 qp = kzalloc(sizeof(*qp), GFP_KERNEL);
487 return ERR_PTR(-ENOMEM);
490 if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd)) {
492 return ERR_PTR(-EFAULT);
495 err = mthca_map_user_db(to_mdev(pd->device), &context->uar,
497 ucmd.sq_db_index, ucmd.sq_db_page);
503 err = mthca_map_user_db(to_mdev(pd->device), &context->uar,
505 ucmd.rq_db_index, ucmd.rq_db_page);
507 mthca_unmap_user_db(to_mdev(pd->device),
515 qp->mr.ibmr.lkey = ucmd.lkey;
516 qp->sq.db_index = ucmd.sq_db_index;
517 qp->rq.db_index = ucmd.rq_db_index;
520 err = mthca_alloc_qp(to_mdev(pd->device), to_mpd(pd),
521 to_mcq(init_attr->send_cq),
522 to_mcq(init_attr->recv_cq),
523 init_attr->qp_type, init_attr->sq_sig_type,
524 &init_attr->cap, qp, udata);
527 mthca_unmap_user_db(to_mdev(pd->device),
531 mthca_unmap_user_db(to_mdev(pd->device),
537 qp->ibqp.qp_num = qp->qpn;
543 qp = kzalloc(sizeof(*qp), GFP_KERNEL);
545 return ERR_PTR(-ENOMEM);
547 qp->ibqp.qp_num = init_attr->qp_type == IB_QPT_SMI ? 0 : 1;
549 err = mthca_alloc_sqp(to_mdev(pd->device), to_mpd(pd),
550 to_mcq(init_attr->send_cq),
551 to_mcq(init_attr->recv_cq),
552 init_attr->sq_sig_type, &init_attr->cap,
553 qp->ibqp.qp_num, init_attr->port_num,
558 /* Don't support raw QPs */
559 return ERR_PTR(-ENOSYS);
567 init_attr->cap.max_send_wr = qp->sq.max;
568 init_attr->cap.max_recv_wr = qp->rq.max;
569 init_attr->cap.max_send_sge = qp->sq.max_gs;
570 init_attr->cap.max_recv_sge = qp->rq.max_gs;
571 init_attr->cap.max_inline_data = qp->max_inline_data;
576 static int mthca_destroy_qp(struct ib_qp *qp, struct ib_udata *udata)
579 struct mthca_ucontext *context =
580 rdma_udata_to_drv_context(
582 struct mthca_ucontext,
585 mthca_unmap_user_db(to_mdev(qp->device),
588 to_mqp(qp)->sq.db_index);
589 mthca_unmap_user_db(to_mdev(qp->device),
592 to_mqp(qp)->rq.db_index);
594 mthca_free_qp(to_mdev(qp->device), to_mqp(qp));
599 static int mthca_create_cq(struct ib_cq *ibcq,
600 const struct ib_cq_init_attr *attr,
601 struct ib_udata *udata)
603 struct ib_device *ibdev = ibcq->device;
604 int entries = attr->cqe;
605 struct mthca_create_cq ucmd;
609 struct mthca_ucontext *context = rdma_udata_to_drv_context(
610 udata, struct mthca_ucontext, ibucontext);
615 if (entries < 1 || entries > to_mdev(ibdev)->limits.max_cqes)
619 if (ib_copy_from_udata(&ucmd, udata, sizeof(ucmd)))
622 err = mthca_map_user_db(to_mdev(ibdev), &context->uar,
623 context->db_tab, ucmd.set_db_index,
628 err = mthca_map_user_db(to_mdev(ibdev), &context->uar,
629 context->db_tab, ucmd.arm_db_index,
638 cq->buf.mr.ibmr.lkey = ucmd.lkey;
639 cq->set_ci_db_index = ucmd.set_db_index;
640 cq->arm_db_index = ucmd.arm_db_index;
643 for (nent = 1; nent <= entries; nent <<= 1)
646 err = mthca_init_cq(to_mdev(ibdev), nent, context,
647 udata ? ucmd.pdn : to_mdev(ibdev)->driver_pd.pd_num,
652 if (udata && ib_copy_to_udata(udata, &cq->cqn, sizeof(__u32))) {
653 mthca_free_cq(to_mdev(ibdev), cq);
658 cq->resize_buf = NULL;
664 mthca_unmap_user_db(to_mdev(ibdev), &context->uar,
665 context->db_tab, ucmd.arm_db_index);
669 mthca_unmap_user_db(to_mdev(ibdev), &context->uar,
670 context->db_tab, ucmd.set_db_index);
675 static int mthca_alloc_resize_buf(struct mthca_dev *dev, struct mthca_cq *cq,
680 spin_lock_irq(&cq->lock);
681 if (cq->resize_buf) {
686 cq->resize_buf = kmalloc(sizeof *cq->resize_buf, GFP_ATOMIC);
687 if (!cq->resize_buf) {
692 cq->resize_buf->state = CQ_RESIZE_ALLOC;
697 spin_unlock_irq(&cq->lock);
702 ret = mthca_alloc_cq_buf(dev, &cq->resize_buf->buf, entries);
704 spin_lock_irq(&cq->lock);
705 kfree(cq->resize_buf);
706 cq->resize_buf = NULL;
707 spin_unlock_irq(&cq->lock);
711 cq->resize_buf->cqe = entries - 1;
713 spin_lock_irq(&cq->lock);
714 cq->resize_buf->state = CQ_RESIZE_READY;
715 spin_unlock_irq(&cq->lock);
720 static int mthca_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *udata)
722 struct mthca_dev *dev = to_mdev(ibcq->device);
723 struct mthca_cq *cq = to_mcq(ibcq);
724 struct mthca_resize_cq ucmd;
728 if (entries < 1 || entries > dev->limits.max_cqes)
731 mutex_lock(&cq->mutex);
733 entries = roundup_pow_of_two(entries + 1);
734 if (entries == ibcq->cqe + 1) {
740 ret = mthca_alloc_resize_buf(dev, cq, entries);
743 lkey = cq->resize_buf->buf.mr.ibmr.lkey;
745 if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd)) {
752 ret = mthca_RESIZE_CQ(dev, cq->cqn, lkey, ilog2(entries));
755 if (cq->resize_buf) {
756 mthca_free_cq_buf(dev, &cq->resize_buf->buf,
757 cq->resize_buf->cqe);
758 kfree(cq->resize_buf);
759 spin_lock_irq(&cq->lock);
760 cq->resize_buf = NULL;
761 spin_unlock_irq(&cq->lock);
767 struct mthca_cq_buf tbuf;
770 spin_lock_irq(&cq->lock);
771 if (cq->resize_buf->state == CQ_RESIZE_READY) {
772 mthca_cq_resize_copy_cqes(cq);
775 cq->buf = cq->resize_buf->buf;
776 cq->ibcq.cqe = cq->resize_buf->cqe;
778 tbuf = cq->resize_buf->buf;
779 tcqe = cq->resize_buf->cqe;
782 kfree(cq->resize_buf);
783 cq->resize_buf = NULL;
784 spin_unlock_irq(&cq->lock);
786 mthca_free_cq_buf(dev, &tbuf, tcqe);
788 ibcq->cqe = entries - 1;
791 mutex_unlock(&cq->mutex);
796 static void mthca_destroy_cq(struct ib_cq *cq, struct ib_udata *udata)
799 struct mthca_ucontext *context =
800 rdma_udata_to_drv_context(
802 struct mthca_ucontext,
805 mthca_unmap_user_db(to_mdev(cq->device),
808 to_mcq(cq)->arm_db_index);
809 mthca_unmap_user_db(to_mdev(cq->device),
812 to_mcq(cq)->set_ci_db_index);
814 mthca_free_cq(to_mdev(cq->device), to_mcq(cq));
817 static inline u32 convert_access(int acc)
819 return (acc & IB_ACCESS_REMOTE_ATOMIC ? MTHCA_MPT_FLAG_ATOMIC : 0) |
820 (acc & IB_ACCESS_REMOTE_WRITE ? MTHCA_MPT_FLAG_REMOTE_WRITE : 0) |
821 (acc & IB_ACCESS_REMOTE_READ ? MTHCA_MPT_FLAG_REMOTE_READ : 0) |
822 (acc & IB_ACCESS_LOCAL_WRITE ? MTHCA_MPT_FLAG_LOCAL_WRITE : 0) |
823 MTHCA_MPT_FLAG_LOCAL_READ;
826 static struct ib_mr *mthca_get_dma_mr(struct ib_pd *pd, int acc)
831 mr = kmalloc(sizeof *mr, GFP_KERNEL);
833 return ERR_PTR(-ENOMEM);
835 err = mthca_mr_alloc_notrans(to_mdev(pd->device),
837 convert_access(acc), mr);
849 static struct ib_mr *mthca_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
850 u64 virt, int acc, struct ib_udata *udata)
852 struct mthca_dev *dev = to_mdev(pd->device);
853 struct scatterlist *sg;
855 struct mthca_reg_mr ucmd;
862 if (udata->inlen - sizeof (struct ib_uverbs_cmd_hdr) < sizeof ucmd) {
863 if (!to_mucontext(pd->uobject->context)->reg_mr_warned) {
864 mthca_warn(dev, "Process '%s' did not pass in MR attrs.\n",
866 mthca_warn(dev, " Update libmthca to fix this.\n");
868 ++to_mucontext(pd->uobject->context)->reg_mr_warned;
870 } else if (ib_copy_from_udata(&ucmd, udata, sizeof ucmd))
871 return ERR_PTR(-EFAULT);
873 mr = kmalloc(sizeof *mr, GFP_KERNEL);
875 return ERR_PTR(-ENOMEM);
877 mr->umem = ib_umem_get(pd->uobject->context, start, length, acc,
878 ucmd.mr_attrs & MTHCA_MR_DMASYNC);
880 if (IS_ERR(mr->umem)) {
881 err = PTR_ERR(mr->umem);
885 shift = ffs(mr->umem->page_size) - 1;
888 mr->mtt = mthca_alloc_mtt(dev, n);
889 if (IS_ERR(mr->mtt)) {
890 err = PTR_ERR(mr->mtt);
894 pages = (u64 *) __get_free_page(GFP_KERNEL);
902 write_mtt_size = min(mthca_write_mtt_size(dev), (int) (PAGE_SIZE / sizeof *pages));
904 for_each_sg(mr->umem->sg_head.sgl, sg, mr->umem->nmap, entry) {
905 len = sg_dma_len(sg) >> shift;
906 for (k = 0; k < len; ++k) {
907 pages[i++] = sg_dma_address(sg) +
908 mr->umem->page_size * k;
910 * Be friendly to write_mtt and pass it chunks
911 * of appropriate size.
913 if (i == write_mtt_size) {
914 err = mthca_write_mtt(dev, mr->mtt, n, pages, i);
924 err = mthca_write_mtt(dev, mr->mtt, n, pages, i);
926 free_page((unsigned long) pages);
930 err = mthca_mr_alloc(dev, to_mpd(pd)->pd_num, shift, virt, length,
931 convert_access(acc), mr);
939 mthca_free_mtt(dev, mr->mtt);
942 ib_umem_release(mr->umem);
949 static int mthca_dereg_mr(struct ib_mr *mr, struct ib_udata *udata)
951 struct mthca_mr *mmr = to_mmr(mr);
953 mthca_free_mr(to_mdev(mr->device), mmr);
954 ib_umem_release(mmr->umem);
960 static struct ib_fmr *mthca_alloc_fmr(struct ib_pd *pd, int mr_access_flags,
961 struct ib_fmr_attr *fmr_attr)
963 struct mthca_fmr *fmr;
966 fmr = kmalloc(sizeof *fmr, GFP_KERNEL);
968 return ERR_PTR(-ENOMEM);
970 memcpy(&fmr->attr, fmr_attr, sizeof *fmr_attr);
971 err = mthca_fmr_alloc(to_mdev(pd->device), to_mpd(pd)->pd_num,
972 convert_access(mr_access_flags), fmr);
982 static int mthca_dealloc_fmr(struct ib_fmr *fmr)
984 struct mthca_fmr *mfmr = to_mfmr(fmr);
987 err = mthca_free_fmr(to_mdev(fmr->device), mfmr);
995 static int mthca_unmap_fmr(struct list_head *fmr_list)
999 struct mthca_dev *mdev = NULL;
1001 list_for_each_entry(fmr, fmr_list, list) {
1002 if (mdev && to_mdev(fmr->device) != mdev)
1004 mdev = to_mdev(fmr->device);
1010 if (mthca_is_memfree(mdev)) {
1011 list_for_each_entry(fmr, fmr_list, list)
1012 mthca_arbel_fmr_unmap(mdev, to_mfmr(fmr));
1016 list_for_each_entry(fmr, fmr_list, list)
1017 mthca_tavor_fmr_unmap(mdev, to_mfmr(fmr));
1019 err = mthca_SYNC_TPT(mdev);
1023 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
1026 struct mthca_dev *dev =
1027 container_of(device, struct mthca_dev, ib_dev.dev);
1028 return sprintf(buf, "%x\n", dev->rev_id);
1031 static ssize_t show_hca(struct device *device, struct device_attribute *attr,
1034 struct mthca_dev *dev =
1035 container_of(device, struct mthca_dev, ib_dev.dev);
1036 switch (dev->pdev->device) {
1037 case PCI_DEVICE_ID_MELLANOX_TAVOR:
1038 return sprintf(buf, "MT23108\n");
1039 case PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT:
1040 return sprintf(buf, "MT25208 (MT23108 compat mode)\n");
1041 case PCI_DEVICE_ID_MELLANOX_ARBEL:
1042 return sprintf(buf, "MT25208\n");
1043 case PCI_DEVICE_ID_MELLANOX_SINAI:
1044 case PCI_DEVICE_ID_MELLANOX_SINAI_OLD:
1045 return sprintf(buf, "MT25204\n");
1047 return sprintf(buf, "unknown\n");
1051 static ssize_t show_board(struct device *device, struct device_attribute *attr,
1054 struct mthca_dev *dev =
1055 container_of(device, struct mthca_dev, ib_dev.dev);
1056 return sprintf(buf, "%.*s\n", MTHCA_BOARD_ID_LEN, dev->board_id);
1059 static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
1060 static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
1061 static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
1063 static struct device_attribute *mthca_dev_attributes[] = {
1069 static int mthca_init_node_data(struct mthca_dev *dev)
1071 struct ib_smp *in_mad = NULL;
1072 struct ib_smp *out_mad = NULL;
1075 in_mad = kzalloc(sizeof *in_mad, GFP_KERNEL);
1076 out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
1077 if (!in_mad || !out_mad)
1080 init_query_mad(in_mad);
1081 in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
1083 err = mthca_MAD_IFC(dev, 1, 1,
1084 1, NULL, NULL, in_mad, out_mad);
1088 memcpy(dev->ib_dev.node_desc, out_mad->data, IB_DEVICE_NODE_DESC_MAX);
1090 in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
1092 err = mthca_MAD_IFC(dev, 1, 1,
1093 1, NULL, NULL, in_mad, out_mad);
1097 if (mthca_is_memfree(dev))
1098 dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
1099 memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
1107 static int mthca_port_immutable(struct ib_device *ibdev, u8 port_num,
1108 struct ib_port_immutable *immutable)
1110 struct ib_port_attr attr;
1113 err = mthca_query_port(ibdev, port_num, &attr);
1117 immutable->pkey_tbl_len = attr.pkey_tbl_len;
1118 immutable->gid_tbl_len = attr.gid_tbl_len;
1119 immutable->core_cap_flags = RDMA_CORE_PORT_IBA_IB;
1120 immutable->max_mad_size = IB_MGMT_MAD_SIZE;
1125 static void get_dev_fw_str(struct ib_device *device, char *str,
1128 struct mthca_dev *dev =
1129 container_of(device, struct mthca_dev, ib_dev);
1130 snprintf(str, str_len, "%d.%d.%d",
1131 (int) (dev->fw_ver >> 32),
1132 (int) (dev->fw_ver >> 16) & 0xffff,
1133 (int) dev->fw_ver & 0xffff);
1136 int mthca_register_device(struct mthca_dev *dev)
1141 ret = mthca_init_node_data(dev);
1145 #define mthca_ib_ah mthca_ah
1146 #define mthca_ib_cq mthca_cq
1147 #define mthca_ib_pd mthca_pd
1148 #define mthca_ib_qp mthca_qp
1149 #define mthca_ib_srq mthca_srq
1150 #define mthca_ib_ucontext mthca_ucontext
1151 INIT_IB_DEVICE_OPS(&dev->ib_dev.ops, mthca, MTHCA);
1152 strlcpy(dev->ib_dev.name, "mthca%d", IB_DEVICE_NAME_MAX);
1153 dev->ib_dev.owner = THIS_MODULE;
1155 dev->ib_dev.uverbs_abi_ver = MTHCA_UVERBS_ABI_VERSION;
1156 dev->ib_dev.uverbs_cmd_mask =
1157 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
1158 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
1159 (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
1160 (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
1161 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
1162 (1ull << IB_USER_VERBS_CMD_REG_MR) |
1163 (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
1164 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
1165 (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
1166 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ) |
1167 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
1168 (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
1169 (1ull << IB_USER_VERBS_CMD_QUERY_QP) |
1170 (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
1171 (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
1172 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) |
1173 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST);
1174 dev->ib_dev.node_type = RDMA_NODE_IB_CA;
1175 dev->ib_dev.phys_port_cnt = dev->limits.num_ports;
1176 dev->ib_dev.num_comp_vectors = 1;
1177 dev->ib_dev.dma_device = &dev->pdev->dev;
1178 dev->ib_dev.query_device = mthca_query_device;
1179 dev->ib_dev.query_port = mthca_query_port;
1180 dev->ib_dev.modify_device = mthca_modify_device;
1181 dev->ib_dev.modify_port = mthca_modify_port;
1182 dev->ib_dev.query_pkey = mthca_query_pkey;
1183 dev->ib_dev.query_gid = mthca_query_gid;
1184 dev->ib_dev.alloc_ucontext = mthca_alloc_ucontext;
1185 dev->ib_dev.dealloc_ucontext = mthca_dealloc_ucontext;
1186 dev->ib_dev.mmap = mthca_mmap_uar;
1187 dev->ib_dev.alloc_pd = mthca_alloc_pd;
1188 dev->ib_dev.dealloc_pd = mthca_dealloc_pd;
1189 dev->ib_dev.create_ah = mthca_ah_create;
1190 dev->ib_dev.query_ah = mthca_ah_query;
1191 dev->ib_dev.destroy_ah = mthca_ah_destroy;
1193 if (dev->mthca_flags & MTHCA_FLAG_SRQ) {
1194 dev->ib_dev.create_srq = mthca_create_srq;
1195 dev->ib_dev.modify_srq = mthca_modify_srq;
1196 dev->ib_dev.query_srq = mthca_query_srq;
1197 dev->ib_dev.destroy_srq = mthca_destroy_srq;
1198 dev->ib_dev.uverbs_cmd_mask |=
1199 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ) |
1200 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ) |
1201 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ) |
1202 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ);
1204 if (mthca_is_memfree(dev))
1205 dev->ib_dev.post_srq_recv = mthca_arbel_post_srq_recv;
1207 dev->ib_dev.post_srq_recv = mthca_tavor_post_srq_recv;
1210 dev->ib_dev.create_qp = mthca_create_qp;
1211 dev->ib_dev.modify_qp = mthca_modify_qp;
1212 dev->ib_dev.query_qp = mthca_query_qp;
1213 dev->ib_dev.destroy_qp = mthca_destroy_qp;
1214 dev->ib_dev.create_cq = mthca_create_cq;
1215 dev->ib_dev.resize_cq = mthca_resize_cq;
1216 dev->ib_dev.destroy_cq = mthca_destroy_cq;
1217 dev->ib_dev.poll_cq = mthca_poll_cq;
1218 dev->ib_dev.get_dma_mr = mthca_get_dma_mr;
1219 dev->ib_dev.reg_user_mr = mthca_reg_user_mr;
1220 dev->ib_dev.dereg_mr = mthca_dereg_mr;
1221 dev->ib_dev.get_port_immutable = mthca_port_immutable;
1222 dev->ib_dev.get_dev_fw_str = get_dev_fw_str;
1224 if (dev->mthca_flags & MTHCA_FLAG_FMR) {
1225 dev->ib_dev.alloc_fmr = mthca_alloc_fmr;
1226 dev->ib_dev.unmap_fmr = mthca_unmap_fmr;
1227 dev->ib_dev.dealloc_fmr = mthca_dealloc_fmr;
1228 if (mthca_is_memfree(dev))
1229 dev->ib_dev.map_phys_fmr = mthca_arbel_map_phys_fmr;
1231 dev->ib_dev.map_phys_fmr = mthca_tavor_map_phys_fmr;
1234 dev->ib_dev.attach_mcast = mthca_multicast_attach;
1235 dev->ib_dev.detach_mcast = mthca_multicast_detach;
1236 dev->ib_dev.process_mad = mthca_process_mad;
1238 if (mthca_is_memfree(dev)) {
1239 dev->ib_dev.req_notify_cq = mthca_arbel_arm_cq;
1240 dev->ib_dev.post_send = mthca_arbel_post_send;
1241 dev->ib_dev.post_recv = mthca_arbel_post_receive;
1243 dev->ib_dev.req_notify_cq = mthca_tavor_arm_cq;
1244 dev->ib_dev.post_send = mthca_tavor_post_send;
1245 dev->ib_dev.post_recv = mthca_tavor_post_receive;
1248 mutex_init(&dev->cap_mask_mutex);
1250 ret = ib_register_device(&dev->ib_dev, NULL);
1254 for (i = 0; i < ARRAY_SIZE(mthca_dev_attributes); ++i) {
1255 ret = device_create_file(&dev->ib_dev.dev,
1256 mthca_dev_attributes[i]);
1258 ib_unregister_device(&dev->ib_dev);
1263 mthca_start_catas_poll(dev);
1268 void mthca_unregister_device(struct mthca_dev *dev)
1270 mthca_stop_catas_poll(dev);
1271 ib_unregister_device(&dev->ib_dev);