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>
41 #include <linux/sched.h>
44 #include "mthca_dev.h"
45 #include "mthca_cmd.h"
46 #include "mthca_user.h"
47 #include "mthca_memfree.h"
49 static void init_query_mad(struct ib_smp
*mad
)
51 mad
->base_version
= 1;
52 mad
->mgmt_class
= IB_MGMT_CLASS_SUBN_LID_ROUTED
;
53 mad
->class_version
= 1;
54 mad
->method
= IB_MGMT_METHOD_GET
;
57 static int mthca_query_device(struct ib_device
*ibdev
,
58 struct ib_device_attr
*props
)
60 struct ib_smp
*in_mad
= NULL
;
61 struct ib_smp
*out_mad
= NULL
;
63 struct mthca_dev
*mdev
= to_mdev(ibdev
);
67 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
68 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
69 if (!in_mad
|| !out_mad
)
72 memset(props
, 0, sizeof *props
);
74 props
->fw_ver
= mdev
->fw_ver
;
76 init_query_mad(in_mad
);
77 in_mad
->attr_id
= IB_SMP_ATTR_NODE_INFO
;
79 err
= mthca_MAD_IFC(mdev
, 1, 1,
80 1, NULL
, NULL
, in_mad
, out_mad
,
89 props
->device_cap_flags
= mdev
->device_cap_flags
;
90 props
->vendor_id
= be32_to_cpup((__be32
*) (out_mad
->data
+ 36)) &
92 props
->vendor_part_id
= be16_to_cpup((__be16
*) (out_mad
->data
+ 30));
93 props
->hw_ver
= be32_to_cpup((__be32
*) (out_mad
->data
+ 32));
94 memcpy(&props
->sys_image_guid
, out_mad
->data
+ 4, 8);
96 props
->max_mr_size
= ~0ull;
97 props
->page_size_cap
= mdev
->limits
.page_size_cap
;
98 props
->max_qp
= mdev
->limits
.num_qps
- mdev
->limits
.reserved_qps
;
99 props
->max_qp_wr
= mdev
->limits
.max_wqes
;
100 props
->max_sge
= mdev
->limits
.max_sg
;
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
;
145 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
146 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
147 if (!in_mad
|| !out_mad
)
150 memset(props
, 0, sizeof *props
);
152 init_query_mad(in_mad
);
153 in_mad
->attr_id
= IB_SMP_ATTR_PORT_INFO
;
154 in_mad
->attr_mod
= cpu_to_be32(port
);
156 err
= mthca_MAD_IFC(to_mdev(ibdev
), 1, 1,
157 port
, NULL
, NULL
, in_mad
, out_mad
,
166 props
->lid
= be16_to_cpup((__be16
*) (out_mad
->data
+ 16));
167 props
->lmc
= out_mad
->data
[34] & 0x7;
168 props
->sm_lid
= be16_to_cpup((__be16
*) (out_mad
->data
+ 18));
169 props
->sm_sl
= out_mad
->data
[36] & 0xf;
170 props
->state
= out_mad
->data
[32] & 0xf;
171 props
->phys_state
= out_mad
->data
[33] >> 4;
172 props
->port_cap_flags
= be32_to_cpup((__be32
*) (out_mad
->data
+ 20));
173 props
->gid_tbl_len
= to_mdev(ibdev
)->limits
.gid_table_len
;
174 props
->max_msg_sz
= 0x80000000;
175 props
->pkey_tbl_len
= to_mdev(ibdev
)->limits
.pkey_table_len
;
176 props
->bad_pkey_cntr
= be16_to_cpup((__be16
*) (out_mad
->data
+ 46));
177 props
->qkey_viol_cntr
= be16_to_cpup((__be16
*) (out_mad
->data
+ 48));
178 props
->active_width
= out_mad
->data
[31] & 0xf;
179 props
->active_speed
= out_mad
->data
[35] >> 4;
180 props
->max_mtu
= out_mad
->data
[41] & 0xf;
181 props
->active_mtu
= out_mad
->data
[36] >> 4;
182 props
->subnet_timeout
= out_mad
->data
[51] & 0x1f;
183 props
->max_vl_num
= out_mad
->data
[37] >> 4;
184 props
->init_type_reply
= out_mad
->data
[41] >> 4;
192 static int mthca_modify_device(struct ib_device
*ibdev
,
194 struct ib_device_modify
*props
)
196 if (mask
& ~IB_DEVICE_MODIFY_NODE_DESC
)
199 if (mask
& IB_DEVICE_MODIFY_NODE_DESC
) {
200 if (mutex_lock_interruptible(&to_mdev(ibdev
)->cap_mask_mutex
))
202 memcpy(ibdev
->node_desc
, props
->node_desc
, 64);
203 mutex_unlock(&to_mdev(ibdev
)->cap_mask_mutex
);
209 static int mthca_modify_port(struct ib_device
*ibdev
,
210 u8 port
, int port_modify_mask
,
211 struct ib_port_modify
*props
)
213 struct mthca_set_ib_param set_ib
;
214 struct ib_port_attr attr
;
218 if (mutex_lock_interruptible(&to_mdev(ibdev
)->cap_mask_mutex
))
221 err
= mthca_query_port(ibdev
, port
, &attr
);
225 set_ib
.set_si_guid
= 0;
226 set_ib
.reset_qkey_viol
= !!(port_modify_mask
& IB_PORT_RESET_QKEY_CNTR
);
228 set_ib
.cap_mask
= (attr
.port_cap_flags
| props
->set_port_cap_mask
) &
229 ~props
->clr_port_cap_mask
;
231 err
= mthca_SET_IB(to_mdev(ibdev
), &set_ib
, port
, &status
);
240 mutex_unlock(&to_mdev(ibdev
)->cap_mask_mutex
);
244 static int mthca_query_pkey(struct ib_device
*ibdev
,
245 u8 port
, u16 index
, u16
*pkey
)
247 struct ib_smp
*in_mad
= NULL
;
248 struct ib_smp
*out_mad
= NULL
;
252 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
253 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
254 if (!in_mad
|| !out_mad
)
257 init_query_mad(in_mad
);
258 in_mad
->attr_id
= IB_SMP_ATTR_PKEY_TABLE
;
259 in_mad
->attr_mod
= cpu_to_be32(index
/ 32);
261 err
= mthca_MAD_IFC(to_mdev(ibdev
), 1, 1,
262 port
, NULL
, NULL
, in_mad
, out_mad
,
271 *pkey
= be16_to_cpu(((__be16
*) out_mad
->data
)[index
% 32]);
279 static int mthca_query_gid(struct ib_device
*ibdev
, u8 port
,
280 int index
, union ib_gid
*gid
)
282 struct ib_smp
*in_mad
= NULL
;
283 struct ib_smp
*out_mad
= NULL
;
287 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
288 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
289 if (!in_mad
|| !out_mad
)
292 init_query_mad(in_mad
);
293 in_mad
->attr_id
= IB_SMP_ATTR_PORT_INFO
;
294 in_mad
->attr_mod
= cpu_to_be32(port
);
296 err
= mthca_MAD_IFC(to_mdev(ibdev
), 1, 1,
297 port
, NULL
, NULL
, in_mad
, out_mad
,
306 memcpy(gid
->raw
, out_mad
->data
+ 8, 8);
308 init_query_mad(in_mad
);
309 in_mad
->attr_id
= IB_SMP_ATTR_GUID_INFO
;
310 in_mad
->attr_mod
= cpu_to_be32(index
/ 8);
312 err
= mthca_MAD_IFC(to_mdev(ibdev
), 1, 1,
313 port
, NULL
, NULL
, in_mad
, out_mad
,
322 memcpy(gid
->raw
+ 8, out_mad
->data
+ (index
% 8) * 8, 8);
330 static struct ib_ucontext
*mthca_alloc_ucontext(struct ib_device
*ibdev
,
331 struct ib_udata
*udata
)
333 struct mthca_alloc_ucontext_resp uresp
;
334 struct mthca_ucontext
*context
;
337 if (!(to_mdev(ibdev
)->active
))
338 return ERR_PTR(-EAGAIN
);
340 memset(&uresp
, 0, sizeof uresp
);
342 uresp
.qp_tab_size
= to_mdev(ibdev
)->limits
.num_qps
;
343 if (mthca_is_memfree(to_mdev(ibdev
)))
344 uresp
.uarc_size
= to_mdev(ibdev
)->uar_table
.uarc_size
;
348 context
= kmalloc(sizeof *context
, GFP_KERNEL
);
350 return ERR_PTR(-ENOMEM
);
352 err
= mthca_uar_alloc(to_mdev(ibdev
), &context
->uar
);
358 context
->db_tab
= mthca_init_user_db_tab(to_mdev(ibdev
));
359 if (IS_ERR(context
->db_tab
)) {
360 err
= PTR_ERR(context
->db_tab
);
361 mthca_uar_free(to_mdev(ibdev
), &context
->uar
);
366 if (ib_copy_to_udata(udata
, &uresp
, sizeof uresp
)) {
367 mthca_cleanup_user_db_tab(to_mdev(ibdev
), &context
->uar
, context
->db_tab
);
368 mthca_uar_free(to_mdev(ibdev
), &context
->uar
);
370 return ERR_PTR(-EFAULT
);
373 context
->reg_mr_warned
= 0;
375 return &context
->ibucontext
;
378 static int mthca_dealloc_ucontext(struct ib_ucontext
*context
)
380 mthca_cleanup_user_db_tab(to_mdev(context
->device
), &to_mucontext(context
)->uar
,
381 to_mucontext(context
)->db_tab
);
382 mthca_uar_free(to_mdev(context
->device
), &to_mucontext(context
)->uar
);
383 kfree(to_mucontext(context
));
388 static int mthca_mmap_uar(struct ib_ucontext
*context
,
389 struct vm_area_struct
*vma
)
391 if (vma
->vm_end
- vma
->vm_start
!= PAGE_SIZE
)
394 vma
->vm_page_prot
= pgprot_noncached(vma
->vm_page_prot
);
396 if (io_remap_pfn_range(vma
, vma
->vm_start
,
397 to_mucontext(context
)->uar
.pfn
,
398 PAGE_SIZE
, vma
->vm_page_prot
))
404 static struct ib_pd
*mthca_alloc_pd(struct ib_device
*ibdev
,
405 struct ib_ucontext
*context
,
406 struct ib_udata
*udata
)
411 pd
= kmalloc(sizeof *pd
, GFP_KERNEL
);
413 return ERR_PTR(-ENOMEM
);
415 err
= mthca_pd_alloc(to_mdev(ibdev
), !context
, pd
);
422 if (ib_copy_to_udata(udata
, &pd
->pd_num
, sizeof (__u32
))) {
423 mthca_pd_free(to_mdev(ibdev
), pd
);
425 return ERR_PTR(-EFAULT
);
432 static int mthca_dealloc_pd(struct ib_pd
*pd
)
434 mthca_pd_free(to_mdev(pd
->device
), to_mpd(pd
));
440 static struct ib_ah
*mthca_ah_create(struct ib_pd
*pd
,
441 struct ib_ah_attr
*ah_attr
)
446 ah
= kmalloc(sizeof *ah
, GFP_ATOMIC
);
448 return ERR_PTR(-ENOMEM
);
450 err
= mthca_create_ah(to_mdev(pd
->device
), to_mpd(pd
), ah_attr
, ah
);
459 static int mthca_ah_destroy(struct ib_ah
*ah
)
461 mthca_destroy_ah(to_mdev(ah
->device
), to_mah(ah
));
467 static struct ib_srq
*mthca_create_srq(struct ib_pd
*pd
,
468 struct ib_srq_init_attr
*init_attr
,
469 struct ib_udata
*udata
)
471 struct mthca_create_srq ucmd
;
472 struct mthca_ucontext
*context
= NULL
;
473 struct mthca_srq
*srq
;
476 srq
= kmalloc(sizeof *srq
, GFP_KERNEL
);
478 return ERR_PTR(-ENOMEM
);
481 context
= to_mucontext(pd
->uobject
->context
);
483 if (ib_copy_from_udata(&ucmd
, udata
, sizeof ucmd
)) {
488 err
= mthca_map_user_db(to_mdev(pd
->device
), &context
->uar
,
489 context
->db_tab
, ucmd
.db_index
,
495 srq
->mr
.ibmr
.lkey
= ucmd
.lkey
;
496 srq
->db_index
= ucmd
.db_index
;
499 err
= mthca_alloc_srq(to_mdev(pd
->device
), to_mpd(pd
),
500 &init_attr
->attr
, srq
);
502 if (err
&& pd
->uobject
)
503 mthca_unmap_user_db(to_mdev(pd
->device
), &context
->uar
,
504 context
->db_tab
, ucmd
.db_index
);
509 if (context
&& ib_copy_to_udata(udata
, &srq
->srqn
, sizeof (__u32
))) {
510 mthca_free_srq(to_mdev(pd
->device
), srq
);
523 static int mthca_destroy_srq(struct ib_srq
*srq
)
525 struct mthca_ucontext
*context
;
528 context
= to_mucontext(srq
->uobject
->context
);
530 mthca_unmap_user_db(to_mdev(srq
->device
), &context
->uar
,
531 context
->db_tab
, to_msrq(srq
)->db_index
);
534 mthca_free_srq(to_mdev(srq
->device
), to_msrq(srq
));
540 static struct ib_qp
*mthca_create_qp(struct ib_pd
*pd
,
541 struct ib_qp_init_attr
*init_attr
,
542 struct ib_udata
*udata
)
544 struct mthca_create_qp ucmd
;
548 if (init_attr
->create_flags
)
549 return ERR_PTR(-EINVAL
);
551 switch (init_attr
->qp_type
) {
556 struct mthca_ucontext
*context
;
558 qp
= kmalloc(sizeof *qp
, GFP_KERNEL
);
560 return ERR_PTR(-ENOMEM
);
563 context
= to_mucontext(pd
->uobject
->context
);
565 if (ib_copy_from_udata(&ucmd
, udata
, sizeof ucmd
)) {
567 return ERR_PTR(-EFAULT
);
570 err
= mthca_map_user_db(to_mdev(pd
->device
), &context
->uar
,
572 ucmd
.sq_db_index
, ucmd
.sq_db_page
);
578 err
= mthca_map_user_db(to_mdev(pd
->device
), &context
->uar
,
580 ucmd
.rq_db_index
, ucmd
.rq_db_page
);
582 mthca_unmap_user_db(to_mdev(pd
->device
),
590 qp
->mr
.ibmr
.lkey
= ucmd
.lkey
;
591 qp
->sq
.db_index
= ucmd
.sq_db_index
;
592 qp
->rq
.db_index
= ucmd
.rq_db_index
;
595 err
= mthca_alloc_qp(to_mdev(pd
->device
), to_mpd(pd
),
596 to_mcq(init_attr
->send_cq
),
597 to_mcq(init_attr
->recv_cq
),
598 init_attr
->qp_type
, init_attr
->sq_sig_type
,
599 &init_attr
->cap
, qp
);
601 if (err
&& pd
->uobject
) {
602 context
= to_mucontext(pd
->uobject
->context
);
604 mthca_unmap_user_db(to_mdev(pd
->device
),
608 mthca_unmap_user_db(to_mdev(pd
->device
),
614 qp
->ibqp
.qp_num
= qp
->qpn
;
620 /* Don't allow userspace to create special QPs */
622 return ERR_PTR(-EINVAL
);
624 qp
= kmalloc(sizeof (struct mthca_sqp
), GFP_KERNEL
);
626 return ERR_PTR(-ENOMEM
);
628 qp
->ibqp
.qp_num
= init_attr
->qp_type
== IB_QPT_SMI
? 0 : 1;
630 err
= mthca_alloc_sqp(to_mdev(pd
->device
), to_mpd(pd
),
631 to_mcq(init_attr
->send_cq
),
632 to_mcq(init_attr
->recv_cq
),
633 init_attr
->sq_sig_type
, &init_attr
->cap
,
634 qp
->ibqp
.qp_num
, init_attr
->port_num
,
639 /* Don't support raw QPs */
640 return ERR_PTR(-ENOSYS
);
648 init_attr
->cap
.max_send_wr
= qp
->sq
.max
;
649 init_attr
->cap
.max_recv_wr
= qp
->rq
.max
;
650 init_attr
->cap
.max_send_sge
= qp
->sq
.max_gs
;
651 init_attr
->cap
.max_recv_sge
= qp
->rq
.max_gs
;
652 init_attr
->cap
.max_inline_data
= qp
->max_inline_data
;
657 static int mthca_destroy_qp(struct ib_qp
*qp
)
660 mthca_unmap_user_db(to_mdev(qp
->device
),
661 &to_mucontext(qp
->uobject
->context
)->uar
,
662 to_mucontext(qp
->uobject
->context
)->db_tab
,
663 to_mqp(qp
)->sq
.db_index
);
664 mthca_unmap_user_db(to_mdev(qp
->device
),
665 &to_mucontext(qp
->uobject
->context
)->uar
,
666 to_mucontext(qp
->uobject
->context
)->db_tab
,
667 to_mqp(qp
)->rq
.db_index
);
669 mthca_free_qp(to_mdev(qp
->device
), to_mqp(qp
));
674 static struct ib_cq
*mthca_create_cq(struct ib_device
*ibdev
, int entries
,
676 struct ib_ucontext
*context
,
677 struct ib_udata
*udata
)
679 struct mthca_create_cq ucmd
;
684 if (entries
< 1 || entries
> to_mdev(ibdev
)->limits
.max_cqes
)
685 return ERR_PTR(-EINVAL
);
688 if (ib_copy_from_udata(&ucmd
, udata
, sizeof ucmd
))
689 return ERR_PTR(-EFAULT
);
691 err
= mthca_map_user_db(to_mdev(ibdev
), &to_mucontext(context
)->uar
,
692 to_mucontext(context
)->db_tab
,
693 ucmd
.set_db_index
, ucmd
.set_db_page
);
697 err
= mthca_map_user_db(to_mdev(ibdev
), &to_mucontext(context
)->uar
,
698 to_mucontext(context
)->db_tab
,
699 ucmd
.arm_db_index
, ucmd
.arm_db_page
);
704 cq
= kmalloc(sizeof *cq
, GFP_KERNEL
);
711 cq
->buf
.mr
.ibmr
.lkey
= ucmd
.lkey
;
712 cq
->set_ci_db_index
= ucmd
.set_db_index
;
713 cq
->arm_db_index
= ucmd
.arm_db_index
;
716 for (nent
= 1; nent
<= entries
; nent
<<= 1)
719 err
= mthca_init_cq(to_mdev(ibdev
), nent
,
720 context
? to_mucontext(context
) : NULL
,
721 context
? ucmd
.pdn
: to_mdev(ibdev
)->driver_pd
.pd_num
,
726 if (context
&& ib_copy_to_udata(udata
, &cq
->cqn
, sizeof (__u32
))) {
727 mthca_free_cq(to_mdev(ibdev
), cq
);
731 cq
->resize_buf
= NULL
;
740 mthca_unmap_user_db(to_mdev(ibdev
), &to_mucontext(context
)->uar
,
741 to_mucontext(context
)->db_tab
, ucmd
.arm_db_index
);
745 mthca_unmap_user_db(to_mdev(ibdev
), &to_mucontext(context
)->uar
,
746 to_mucontext(context
)->db_tab
, ucmd
.set_db_index
);
751 static int mthca_alloc_resize_buf(struct mthca_dev
*dev
, struct mthca_cq
*cq
,
756 spin_lock_irq(&cq
->lock
);
757 if (cq
->resize_buf
) {
762 cq
->resize_buf
= kmalloc(sizeof *cq
->resize_buf
, GFP_ATOMIC
);
763 if (!cq
->resize_buf
) {
768 cq
->resize_buf
->state
= CQ_RESIZE_ALLOC
;
773 spin_unlock_irq(&cq
->lock
);
778 ret
= mthca_alloc_cq_buf(dev
, &cq
->resize_buf
->buf
, entries
);
780 spin_lock_irq(&cq
->lock
);
781 kfree(cq
->resize_buf
);
782 cq
->resize_buf
= NULL
;
783 spin_unlock_irq(&cq
->lock
);
787 cq
->resize_buf
->cqe
= entries
- 1;
789 spin_lock_irq(&cq
->lock
);
790 cq
->resize_buf
->state
= CQ_RESIZE_READY
;
791 spin_unlock_irq(&cq
->lock
);
796 static int mthca_resize_cq(struct ib_cq
*ibcq
, int entries
, struct ib_udata
*udata
)
798 struct mthca_dev
*dev
= to_mdev(ibcq
->device
);
799 struct mthca_cq
*cq
= to_mcq(ibcq
);
800 struct mthca_resize_cq ucmd
;
805 if (entries
< 1 || entries
> dev
->limits
.max_cqes
)
808 mutex_lock(&cq
->mutex
);
810 entries
= roundup_pow_of_two(entries
+ 1);
811 if (entries
== ibcq
->cqe
+ 1) {
817 ret
= mthca_alloc_resize_buf(dev
, cq
, entries
);
820 lkey
= cq
->resize_buf
->buf
.mr
.ibmr
.lkey
;
822 if (ib_copy_from_udata(&ucmd
, udata
, sizeof ucmd
)) {
829 ret
= mthca_RESIZE_CQ(dev
, cq
->cqn
, lkey
, ilog2(entries
), &status
);
834 if (cq
->resize_buf
) {
835 mthca_free_cq_buf(dev
, &cq
->resize_buf
->buf
,
836 cq
->resize_buf
->cqe
);
837 kfree(cq
->resize_buf
);
838 spin_lock_irq(&cq
->lock
);
839 cq
->resize_buf
= NULL
;
840 spin_unlock_irq(&cq
->lock
);
846 struct mthca_cq_buf tbuf
;
849 spin_lock_irq(&cq
->lock
);
850 if (cq
->resize_buf
->state
== CQ_RESIZE_READY
) {
851 mthca_cq_resize_copy_cqes(cq
);
854 cq
->buf
= cq
->resize_buf
->buf
;
855 cq
->ibcq
.cqe
= cq
->resize_buf
->cqe
;
857 tbuf
= cq
->resize_buf
->buf
;
858 tcqe
= cq
->resize_buf
->cqe
;
861 kfree(cq
->resize_buf
);
862 cq
->resize_buf
= NULL
;
863 spin_unlock_irq(&cq
->lock
);
865 mthca_free_cq_buf(dev
, &tbuf
, tcqe
);
867 ibcq
->cqe
= entries
- 1;
870 mutex_unlock(&cq
->mutex
);
875 static int mthca_destroy_cq(struct ib_cq
*cq
)
878 mthca_unmap_user_db(to_mdev(cq
->device
),
879 &to_mucontext(cq
->uobject
->context
)->uar
,
880 to_mucontext(cq
->uobject
->context
)->db_tab
,
881 to_mcq(cq
)->arm_db_index
);
882 mthca_unmap_user_db(to_mdev(cq
->device
),
883 &to_mucontext(cq
->uobject
->context
)->uar
,
884 to_mucontext(cq
->uobject
->context
)->db_tab
,
885 to_mcq(cq
)->set_ci_db_index
);
887 mthca_free_cq(to_mdev(cq
->device
), to_mcq(cq
));
893 static inline u32
convert_access(int acc
)
895 return (acc
& IB_ACCESS_REMOTE_ATOMIC
? MTHCA_MPT_FLAG_ATOMIC
: 0) |
896 (acc
& IB_ACCESS_REMOTE_WRITE
? MTHCA_MPT_FLAG_REMOTE_WRITE
: 0) |
897 (acc
& IB_ACCESS_REMOTE_READ
? MTHCA_MPT_FLAG_REMOTE_READ
: 0) |
898 (acc
& IB_ACCESS_LOCAL_WRITE
? MTHCA_MPT_FLAG_LOCAL_WRITE
: 0) |
899 MTHCA_MPT_FLAG_LOCAL_READ
;
902 static struct ib_mr
*mthca_get_dma_mr(struct ib_pd
*pd
, int acc
)
907 mr
= kmalloc(sizeof *mr
, GFP_KERNEL
);
909 return ERR_PTR(-ENOMEM
);
911 err
= mthca_mr_alloc_notrans(to_mdev(pd
->device
),
913 convert_access(acc
), mr
);
925 static struct ib_mr
*mthca_reg_phys_mr(struct ib_pd
*pd
,
926 struct ib_phys_buf
*buffer_list
,
940 mask
= buffer_list
[0].addr
^ *iova_start
;
942 for (i
= 0; i
< num_phys_buf
; ++i
) {
944 mask
|= buffer_list
[i
].addr
;
945 if (i
!= num_phys_buf
- 1)
946 mask
|= buffer_list
[i
].addr
+ buffer_list
[i
].size
;
948 total_size
+= buffer_list
[i
].size
;
951 if (mask
& ~PAGE_MASK
)
952 return ERR_PTR(-EINVAL
);
954 shift
= __ffs(mask
| 1 << 31);
956 buffer_list
[0].size
+= buffer_list
[0].addr
& ((1ULL << shift
) - 1);
957 buffer_list
[0].addr
&= ~0ull << shift
;
959 mr
= kmalloc(sizeof *mr
, GFP_KERNEL
);
961 return ERR_PTR(-ENOMEM
);
964 for (i
= 0; i
< num_phys_buf
; ++i
)
965 npages
+= (buffer_list
[i
].size
+ (1ULL << shift
) - 1) >> shift
;
970 page_list
= kmalloc(npages
* sizeof *page_list
, GFP_KERNEL
);
973 return ERR_PTR(-ENOMEM
);
977 for (i
= 0; i
< num_phys_buf
; ++i
)
979 j
< (buffer_list
[i
].size
+ (1ULL << shift
) - 1) >> shift
;
981 page_list
[n
++] = buffer_list
[i
].addr
+ ((u64
) j
<< shift
);
983 mthca_dbg(to_mdev(pd
->device
), "Registering memory at %llx (iova %llx) "
984 "in PD %x; shift %d, npages %d.\n",
985 (unsigned long long) buffer_list
[0].addr
,
986 (unsigned long long) *iova_start
,
990 err
= mthca_mr_alloc_phys(to_mdev(pd
->device
),
992 page_list
, shift
, npages
,
993 *iova_start
, total_size
,
994 convert_access(acc
), mr
);
1008 static struct ib_mr
*mthca_reg_user_mr(struct ib_pd
*pd
, u64 start
, u64 length
,
1009 u64 virt
, int acc
, struct ib_udata
*udata
)
1011 struct mthca_dev
*dev
= to_mdev(pd
->device
);
1012 struct ib_umem_chunk
*chunk
;
1013 struct mthca_mr
*mr
;
1014 struct mthca_reg_mr ucmd
;
1021 if (udata
->inlen
- sizeof (struct ib_uverbs_cmd_hdr
) < sizeof ucmd
) {
1022 if (!to_mucontext(pd
->uobject
->context
)->reg_mr_warned
) {
1023 mthca_warn(dev
, "Process '%s' did not pass in MR attrs.\n",
1025 mthca_warn(dev
, " Update libmthca to fix this.\n");
1027 ++to_mucontext(pd
->uobject
->context
)->reg_mr_warned
;
1029 } else if (ib_copy_from_udata(&ucmd
, udata
, sizeof ucmd
))
1030 return ERR_PTR(-EFAULT
);
1032 mr
= kmalloc(sizeof *mr
, GFP_KERNEL
);
1034 return ERR_PTR(-ENOMEM
);
1036 mr
->umem
= ib_umem_get(pd
->uobject
->context
, start
, length
, acc
,
1037 ucmd
.mr_attrs
& MTHCA_MR_DMASYNC
);
1039 if (IS_ERR(mr
->umem
)) {
1040 err
= PTR_ERR(mr
->umem
);
1044 shift
= ffs(mr
->umem
->page_size
) - 1;
1047 list_for_each_entry(chunk
, &mr
->umem
->chunk_list
, list
)
1050 mr
->mtt
= mthca_alloc_mtt(dev
, n
);
1051 if (IS_ERR(mr
->mtt
)) {
1052 err
= PTR_ERR(mr
->mtt
);
1056 pages
= (u64
*) __get_free_page(GFP_KERNEL
);
1064 write_mtt_size
= min(mthca_write_mtt_size(dev
), (int) (PAGE_SIZE
/ sizeof *pages
));
1066 list_for_each_entry(chunk
, &mr
->umem
->chunk_list
, list
)
1067 for (j
= 0; j
< chunk
->nmap
; ++j
) {
1068 len
= sg_dma_len(&chunk
->page_list
[j
]) >> shift
;
1069 for (k
= 0; k
< len
; ++k
) {
1070 pages
[i
++] = sg_dma_address(&chunk
->page_list
[j
]) +
1071 mr
->umem
->page_size
* k
;
1073 * Be friendly to write_mtt and pass it chunks
1074 * of appropriate size.
1076 if (i
== write_mtt_size
) {
1077 err
= mthca_write_mtt(dev
, mr
->mtt
, n
, pages
, i
);
1087 err
= mthca_write_mtt(dev
, mr
->mtt
, n
, pages
, i
);
1089 free_page((unsigned long) pages
);
1093 err
= mthca_mr_alloc(dev
, to_mpd(pd
)->pd_num
, shift
, virt
, length
,
1094 convert_access(acc
), mr
);
1102 mthca_free_mtt(dev
, mr
->mtt
);
1105 ib_umem_release(mr
->umem
);
1109 return ERR_PTR(err
);
1112 static int mthca_dereg_mr(struct ib_mr
*mr
)
1114 struct mthca_mr
*mmr
= to_mmr(mr
);
1116 mthca_free_mr(to_mdev(mr
->device
), mmr
);
1118 ib_umem_release(mmr
->umem
);
1124 static struct ib_fmr
*mthca_alloc_fmr(struct ib_pd
*pd
, int mr_access_flags
,
1125 struct ib_fmr_attr
*fmr_attr
)
1127 struct mthca_fmr
*fmr
;
1130 fmr
= kmalloc(sizeof *fmr
, GFP_KERNEL
);
1132 return ERR_PTR(-ENOMEM
);
1134 memcpy(&fmr
->attr
, fmr_attr
, sizeof *fmr_attr
);
1135 err
= mthca_fmr_alloc(to_mdev(pd
->device
), to_mpd(pd
)->pd_num
,
1136 convert_access(mr_access_flags
), fmr
);
1140 return ERR_PTR(err
);
1146 static int mthca_dealloc_fmr(struct ib_fmr
*fmr
)
1148 struct mthca_fmr
*mfmr
= to_mfmr(fmr
);
1151 err
= mthca_free_fmr(to_mdev(fmr
->device
), mfmr
);
1159 static int mthca_unmap_fmr(struct list_head
*fmr_list
)
1164 struct mthca_dev
*mdev
= NULL
;
1166 list_for_each_entry(fmr
, fmr_list
, list
) {
1167 if (mdev
&& to_mdev(fmr
->device
) != mdev
)
1169 mdev
= to_mdev(fmr
->device
);
1175 if (mthca_is_memfree(mdev
)) {
1176 list_for_each_entry(fmr
, fmr_list
, list
)
1177 mthca_arbel_fmr_unmap(mdev
, to_mfmr(fmr
));
1181 list_for_each_entry(fmr
, fmr_list
, list
)
1182 mthca_tavor_fmr_unmap(mdev
, to_mfmr(fmr
));
1184 err
= mthca_SYNC_TPT(mdev
, &status
);
1192 static ssize_t
show_rev(struct device
*device
, struct device_attribute
*attr
,
1195 struct mthca_dev
*dev
=
1196 container_of(device
, struct mthca_dev
, ib_dev
.dev
);
1197 return sprintf(buf
, "%x\n", dev
->rev_id
);
1200 static ssize_t
show_fw_ver(struct device
*device
, struct device_attribute
*attr
,
1203 struct mthca_dev
*dev
=
1204 container_of(device
, struct mthca_dev
, ib_dev
.dev
);
1205 return sprintf(buf
, "%d.%d.%d\n", (int) (dev
->fw_ver
>> 32),
1206 (int) (dev
->fw_ver
>> 16) & 0xffff,
1207 (int) dev
->fw_ver
& 0xffff);
1210 static ssize_t
show_hca(struct device
*device
, struct device_attribute
*attr
,
1213 struct mthca_dev
*dev
=
1214 container_of(device
, struct mthca_dev
, ib_dev
.dev
);
1215 switch (dev
->pdev
->device
) {
1216 case PCI_DEVICE_ID_MELLANOX_TAVOR
:
1217 return sprintf(buf
, "MT23108\n");
1218 case PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT
:
1219 return sprintf(buf
, "MT25208 (MT23108 compat mode)\n");
1220 case PCI_DEVICE_ID_MELLANOX_ARBEL
:
1221 return sprintf(buf
, "MT25208\n");
1222 case PCI_DEVICE_ID_MELLANOX_SINAI
:
1223 case PCI_DEVICE_ID_MELLANOX_SINAI_OLD
:
1224 return sprintf(buf
, "MT25204\n");
1226 return sprintf(buf
, "unknown\n");
1230 static ssize_t
show_board(struct device
*device
, struct device_attribute
*attr
,
1233 struct mthca_dev
*dev
=
1234 container_of(device
, struct mthca_dev
, ib_dev
.dev
);
1235 return sprintf(buf
, "%.*s\n", MTHCA_BOARD_ID_LEN
, dev
->board_id
);
1238 static DEVICE_ATTR(hw_rev
, S_IRUGO
, show_rev
, NULL
);
1239 static DEVICE_ATTR(fw_ver
, S_IRUGO
, show_fw_ver
, NULL
);
1240 static DEVICE_ATTR(hca_type
, S_IRUGO
, show_hca
, NULL
);
1241 static DEVICE_ATTR(board_id
, S_IRUGO
, show_board
, NULL
);
1243 static struct device_attribute
*mthca_dev_attributes
[] = {
1250 static int mthca_init_node_data(struct mthca_dev
*dev
)
1252 struct ib_smp
*in_mad
= NULL
;
1253 struct ib_smp
*out_mad
= NULL
;
1257 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
1258 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
1259 if (!in_mad
|| !out_mad
)
1262 init_query_mad(in_mad
);
1263 in_mad
->attr_id
= IB_SMP_ATTR_NODE_DESC
;
1265 err
= mthca_MAD_IFC(dev
, 1, 1,
1266 1, NULL
, NULL
, in_mad
, out_mad
,
1275 memcpy(dev
->ib_dev
.node_desc
, out_mad
->data
, 64);
1277 in_mad
->attr_id
= IB_SMP_ATTR_NODE_INFO
;
1279 err
= mthca_MAD_IFC(dev
, 1, 1,
1280 1, NULL
, NULL
, in_mad
, out_mad
,
1289 if (mthca_is_memfree(dev
))
1290 dev
->rev_id
= be32_to_cpup((__be32
*) (out_mad
->data
+ 32));
1291 memcpy(&dev
->ib_dev
.node_guid
, out_mad
->data
+ 12, 8);
1299 int mthca_register_device(struct mthca_dev
*dev
)
1304 ret
= mthca_init_node_data(dev
);
1308 strlcpy(dev
->ib_dev
.name
, "mthca%d", IB_DEVICE_NAME_MAX
);
1309 dev
->ib_dev
.owner
= THIS_MODULE
;
1311 dev
->ib_dev
.uverbs_abi_ver
= MTHCA_UVERBS_ABI_VERSION
;
1312 dev
->ib_dev
.uverbs_cmd_mask
=
1313 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT
) |
1314 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE
) |
1315 (1ull << IB_USER_VERBS_CMD_QUERY_PORT
) |
1316 (1ull << IB_USER_VERBS_CMD_ALLOC_PD
) |
1317 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD
) |
1318 (1ull << IB_USER_VERBS_CMD_REG_MR
) |
1319 (1ull << IB_USER_VERBS_CMD_DEREG_MR
) |
1320 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL
) |
1321 (1ull << IB_USER_VERBS_CMD_CREATE_CQ
) |
1322 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ
) |
1323 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ
) |
1324 (1ull << IB_USER_VERBS_CMD_CREATE_QP
) |
1325 (1ull << IB_USER_VERBS_CMD_QUERY_QP
) |
1326 (1ull << IB_USER_VERBS_CMD_MODIFY_QP
) |
1327 (1ull << IB_USER_VERBS_CMD_DESTROY_QP
) |
1328 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST
) |
1329 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST
);
1330 dev
->ib_dev
.node_type
= RDMA_NODE_IB_CA
;
1331 dev
->ib_dev
.phys_port_cnt
= dev
->limits
.num_ports
;
1332 dev
->ib_dev
.num_comp_vectors
= 1;
1333 dev
->ib_dev
.dma_device
= &dev
->pdev
->dev
;
1334 dev
->ib_dev
.query_device
= mthca_query_device
;
1335 dev
->ib_dev
.query_port
= mthca_query_port
;
1336 dev
->ib_dev
.modify_device
= mthca_modify_device
;
1337 dev
->ib_dev
.modify_port
= mthca_modify_port
;
1338 dev
->ib_dev
.query_pkey
= mthca_query_pkey
;
1339 dev
->ib_dev
.query_gid
= mthca_query_gid
;
1340 dev
->ib_dev
.alloc_ucontext
= mthca_alloc_ucontext
;
1341 dev
->ib_dev
.dealloc_ucontext
= mthca_dealloc_ucontext
;
1342 dev
->ib_dev
.mmap
= mthca_mmap_uar
;
1343 dev
->ib_dev
.alloc_pd
= mthca_alloc_pd
;
1344 dev
->ib_dev
.dealloc_pd
= mthca_dealloc_pd
;
1345 dev
->ib_dev
.create_ah
= mthca_ah_create
;
1346 dev
->ib_dev
.query_ah
= mthca_ah_query
;
1347 dev
->ib_dev
.destroy_ah
= mthca_ah_destroy
;
1349 if (dev
->mthca_flags
& MTHCA_FLAG_SRQ
) {
1350 dev
->ib_dev
.create_srq
= mthca_create_srq
;
1351 dev
->ib_dev
.modify_srq
= mthca_modify_srq
;
1352 dev
->ib_dev
.query_srq
= mthca_query_srq
;
1353 dev
->ib_dev
.destroy_srq
= mthca_destroy_srq
;
1354 dev
->ib_dev
.uverbs_cmd_mask
|=
1355 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ
) |
1356 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ
) |
1357 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ
) |
1358 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ
);
1360 if (mthca_is_memfree(dev
))
1361 dev
->ib_dev
.post_srq_recv
= mthca_arbel_post_srq_recv
;
1363 dev
->ib_dev
.post_srq_recv
= mthca_tavor_post_srq_recv
;
1366 dev
->ib_dev
.create_qp
= mthca_create_qp
;
1367 dev
->ib_dev
.modify_qp
= mthca_modify_qp
;
1368 dev
->ib_dev
.query_qp
= mthca_query_qp
;
1369 dev
->ib_dev
.destroy_qp
= mthca_destroy_qp
;
1370 dev
->ib_dev
.create_cq
= mthca_create_cq
;
1371 dev
->ib_dev
.resize_cq
= mthca_resize_cq
;
1372 dev
->ib_dev
.destroy_cq
= mthca_destroy_cq
;
1373 dev
->ib_dev
.poll_cq
= mthca_poll_cq
;
1374 dev
->ib_dev
.get_dma_mr
= mthca_get_dma_mr
;
1375 dev
->ib_dev
.reg_phys_mr
= mthca_reg_phys_mr
;
1376 dev
->ib_dev
.reg_user_mr
= mthca_reg_user_mr
;
1377 dev
->ib_dev
.dereg_mr
= mthca_dereg_mr
;
1379 if (dev
->mthca_flags
& MTHCA_FLAG_FMR
) {
1380 dev
->ib_dev
.alloc_fmr
= mthca_alloc_fmr
;
1381 dev
->ib_dev
.unmap_fmr
= mthca_unmap_fmr
;
1382 dev
->ib_dev
.dealloc_fmr
= mthca_dealloc_fmr
;
1383 if (mthca_is_memfree(dev
))
1384 dev
->ib_dev
.map_phys_fmr
= mthca_arbel_map_phys_fmr
;
1386 dev
->ib_dev
.map_phys_fmr
= mthca_tavor_map_phys_fmr
;
1389 dev
->ib_dev
.attach_mcast
= mthca_multicast_attach
;
1390 dev
->ib_dev
.detach_mcast
= mthca_multicast_detach
;
1391 dev
->ib_dev
.process_mad
= mthca_process_mad
;
1393 if (mthca_is_memfree(dev
)) {
1394 dev
->ib_dev
.req_notify_cq
= mthca_arbel_arm_cq
;
1395 dev
->ib_dev
.post_send
= mthca_arbel_post_send
;
1396 dev
->ib_dev
.post_recv
= mthca_arbel_post_receive
;
1398 dev
->ib_dev
.req_notify_cq
= mthca_tavor_arm_cq
;
1399 dev
->ib_dev
.post_send
= mthca_tavor_post_send
;
1400 dev
->ib_dev
.post_recv
= mthca_tavor_post_receive
;
1403 mutex_init(&dev
->cap_mask_mutex
);
1405 ret
= ib_register_device(&dev
->ib_dev
);
1409 for (i
= 0; i
< ARRAY_SIZE(mthca_dev_attributes
); ++i
) {
1410 ret
= device_create_file(&dev
->ib_dev
.dev
,
1411 mthca_dev_attributes
[i
]);
1413 ib_unregister_device(&dev
->ib_dev
);
1418 mthca_start_catas_poll(dev
);
1423 void mthca_unregister_device(struct mthca_dev
*dev
)
1425 mthca_stop_catas_poll(dev
);
1426 ib_unregister_device(&dev
->ib_dev
);