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>
42 #include <linux/slab.h>
43 #include <linux/stat.h>
45 #include <linux/export.h>
47 #include "mthca_dev.h"
48 #include "mthca_cmd.h"
49 #include "mthca_user.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
,
61 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 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
69 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
70 if (!in_mad
|| !out_mad
)
73 memset(props
, 0, sizeof *props
);
75 props
->fw_ver
= mdev
->fw_ver
;
77 init_query_mad(in_mad
);
78 in_mad
->attr_id
= IB_SMP_ATTR_NODE_INFO
;
80 err
= mthca_MAD_IFC(mdev
, 1, 1,
81 1, NULL
, NULL
, in_mad
, out_mad
);
85 props
->device_cap_flags
= mdev
->device_cap_flags
;
86 props
->vendor_id
= be32_to_cpup((__be32
*) (out_mad
->data
+ 36)) &
88 props
->vendor_part_id
= be16_to_cpup((__be16
*) (out_mad
->data
+ 30));
89 props
->hw_ver
= be32_to_cpup((__be32
*) (out_mad
->data
+ 32));
90 memcpy(&props
->sys_image_guid
, out_mad
->data
+ 4, 8);
92 props
->max_mr_size
= ~0ull;
93 props
->page_size_cap
= mdev
->limits
.page_size_cap
;
94 props
->max_qp
= mdev
->limits
.num_qps
- mdev
->limits
.reserved_qps
;
95 props
->max_qp_wr
= mdev
->limits
.max_wqes
;
96 props
->max_sge
= mdev
->limits
.max_sg
;
97 props
->max_cq
= mdev
->limits
.num_cqs
- mdev
->limits
.reserved_cqs
;
98 props
->max_cqe
= mdev
->limits
.max_cqes
;
99 props
->max_mr
= mdev
->limits
.num_mpts
- mdev
->limits
.reserved_mrws
;
100 props
->max_pd
= mdev
->limits
.num_pds
- mdev
->limits
.reserved_pds
;
101 props
->max_qp_rd_atom
= 1 << mdev
->qp_table
.rdb_shift
;
102 props
->max_qp_init_rd_atom
= mdev
->limits
.max_qp_init_rdma
;
103 props
->max_res_rd_atom
= props
->max_qp_rd_atom
* props
->max_qp
;
104 props
->max_srq
= mdev
->limits
.num_srqs
- mdev
->limits
.reserved_srqs
;
105 props
->max_srq_wr
= mdev
->limits
.max_srq_wqes
;
106 props
->max_srq_sge
= mdev
->limits
.max_srq_sge
;
107 props
->local_ca_ack_delay
= mdev
->limits
.local_ca_ack_delay
;
108 props
->atomic_cap
= mdev
->limits
.flags
& DEV_LIM_FLAG_ATOMIC
?
109 IB_ATOMIC_HCA
: IB_ATOMIC_NONE
;
110 props
->max_pkeys
= mdev
->limits
.pkey_table_len
;
111 props
->max_mcast_grp
= mdev
->limits
.num_mgms
+ mdev
->limits
.num_amgms
;
112 props
->max_mcast_qp_attach
= MTHCA_QP_PER_MGM
;
113 props
->max_total_mcast_qp_attach
= props
->max_mcast_qp_attach
*
114 props
->max_mcast_grp
;
116 * If Sinai memory key optimization is being used, then only
117 * the 8-bit key portion will change. For other HCAs, the
118 * unused index bits will also be used for FMR remapping.
120 if (mdev
->mthca_flags
& MTHCA_FLAG_SINAI_OPT
)
121 props
->max_map_per_fmr
= 255;
123 props
->max_map_per_fmr
=
124 (1 << (32 - ilog2(mdev
->limits
.num_mpts
))) - 1;
133 static int mthca_query_port(struct ib_device
*ibdev
,
134 u8 port
, struct ib_port_attr
*props
)
136 struct ib_smp
*in_mad
= NULL
;
137 struct ib_smp
*out_mad
= NULL
;
140 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
141 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
142 if (!in_mad
|| !out_mad
)
145 memset(props
, 0, sizeof *props
);
147 init_query_mad(in_mad
);
148 in_mad
->attr_id
= IB_SMP_ATTR_PORT_INFO
;
149 in_mad
->attr_mod
= cpu_to_be32(port
);
151 err
= mthca_MAD_IFC(to_mdev(ibdev
), 1, 1,
152 port
, NULL
, NULL
, in_mad
, out_mad
);
156 props
->lid
= be16_to_cpup((__be16
*) (out_mad
->data
+ 16));
157 props
->lmc
= out_mad
->data
[34] & 0x7;
158 props
->sm_lid
= be16_to_cpup((__be16
*) (out_mad
->data
+ 18));
159 props
->sm_sl
= out_mad
->data
[36] & 0xf;
160 props
->state
= out_mad
->data
[32] & 0xf;
161 props
->phys_state
= out_mad
->data
[33] >> 4;
162 props
->port_cap_flags
= be32_to_cpup((__be32
*) (out_mad
->data
+ 20));
163 props
->gid_tbl_len
= to_mdev(ibdev
)->limits
.gid_table_len
;
164 props
->max_msg_sz
= 0x80000000;
165 props
->pkey_tbl_len
= to_mdev(ibdev
)->limits
.pkey_table_len
;
166 props
->bad_pkey_cntr
= be16_to_cpup((__be16
*) (out_mad
->data
+ 46));
167 props
->qkey_viol_cntr
= be16_to_cpup((__be16
*) (out_mad
->data
+ 48));
168 props
->active_width
= out_mad
->data
[31] & 0xf;
169 props
->active_speed
= out_mad
->data
[35] >> 4;
170 props
->max_mtu
= out_mad
->data
[41] & 0xf;
171 props
->active_mtu
= out_mad
->data
[36] >> 4;
172 props
->subnet_timeout
= out_mad
->data
[51] & 0x1f;
173 props
->max_vl_num
= out_mad
->data
[37] >> 4;
174 props
->init_type_reply
= out_mad
->data
[41] >> 4;
182 static int mthca_modify_device(struct ib_device
*ibdev
,
184 struct ib_device_modify
*props
)
186 if (mask
& ~IB_DEVICE_MODIFY_NODE_DESC
)
189 if (mask
& IB_DEVICE_MODIFY_NODE_DESC
) {
190 if (mutex_lock_interruptible(&to_mdev(ibdev
)->cap_mask_mutex
))
192 memcpy(ibdev
->node_desc
, props
->node_desc
, 64);
193 mutex_unlock(&to_mdev(ibdev
)->cap_mask_mutex
);
199 static int mthca_modify_port(struct ib_device
*ibdev
,
200 u8 port
, int port_modify_mask
,
201 struct ib_port_modify
*props
)
203 struct mthca_set_ib_param set_ib
;
204 struct ib_port_attr attr
;
207 if (mutex_lock_interruptible(&to_mdev(ibdev
)->cap_mask_mutex
))
210 err
= mthca_query_port(ibdev
, port
, &attr
);
214 set_ib
.set_si_guid
= 0;
215 set_ib
.reset_qkey_viol
= !!(port_modify_mask
& IB_PORT_RESET_QKEY_CNTR
);
217 set_ib
.cap_mask
= (attr
.port_cap_flags
| props
->set_port_cap_mask
) &
218 ~props
->clr_port_cap_mask
;
220 err
= mthca_SET_IB(to_mdev(ibdev
), &set_ib
, port
);
224 mutex_unlock(&to_mdev(ibdev
)->cap_mask_mutex
);
228 static int mthca_query_pkey(struct ib_device
*ibdev
,
229 u8 port
, u16 index
, u16
*pkey
)
231 struct ib_smp
*in_mad
= NULL
;
232 struct ib_smp
*out_mad
= NULL
;
235 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
236 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
237 if (!in_mad
|| !out_mad
)
240 init_query_mad(in_mad
);
241 in_mad
->attr_id
= IB_SMP_ATTR_PKEY_TABLE
;
242 in_mad
->attr_mod
= cpu_to_be32(index
/ 32);
244 err
= mthca_MAD_IFC(to_mdev(ibdev
), 1, 1,
245 port
, NULL
, NULL
, in_mad
, out_mad
);
249 *pkey
= be16_to_cpu(((__be16
*) out_mad
->data
)[index
% 32]);
257 static int mthca_query_gid(struct ib_device
*ibdev
, u8 port
,
258 int index
, union ib_gid
*gid
)
260 struct ib_smp
*in_mad
= NULL
;
261 struct ib_smp
*out_mad
= NULL
;
264 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
265 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
266 if (!in_mad
|| !out_mad
)
269 init_query_mad(in_mad
);
270 in_mad
->attr_id
= IB_SMP_ATTR_PORT_INFO
;
271 in_mad
->attr_mod
= cpu_to_be32(port
);
273 err
= mthca_MAD_IFC(to_mdev(ibdev
), 1, 1,
274 port
, NULL
, NULL
, in_mad
, out_mad
);
278 memcpy(gid
->raw
, out_mad
->data
+ 8, 8);
280 init_query_mad(in_mad
);
281 in_mad
->attr_id
= IB_SMP_ATTR_GUID_INFO
;
282 in_mad
->attr_mod
= cpu_to_be32(index
/ 8);
284 err
= mthca_MAD_IFC(to_mdev(ibdev
), 1, 1,
285 port
, NULL
, NULL
, in_mad
, out_mad
);
289 memcpy(gid
->raw
+ 8, out_mad
->data
+ (index
% 8) * 8, 8);
297 static struct ib_ucontext
*mthca_alloc_ucontext(struct ib_device
*ibdev
,
298 struct ib_udata
*udata
)
300 struct mthca_alloc_ucontext_resp uresp
;
301 struct mthca_ucontext
*context
;
304 if (!(to_mdev(ibdev
)->active
))
305 return ERR_PTR(-EAGAIN
);
307 memset(&uresp
, 0, sizeof uresp
);
309 uresp
.qp_tab_size
= to_mdev(ibdev
)->limits
.num_qps
;
310 if (mthca_is_memfree(to_mdev(ibdev
)))
311 uresp
.uarc_size
= to_mdev(ibdev
)->uar_table
.uarc_size
;
315 context
= kmalloc(sizeof *context
, GFP_KERNEL
);
317 return ERR_PTR(-ENOMEM
);
319 err
= mthca_uar_alloc(to_mdev(ibdev
), &context
->uar
);
325 context
->db_tab
= mthca_init_user_db_tab(to_mdev(ibdev
));
326 if (IS_ERR(context
->db_tab
)) {
327 err
= PTR_ERR(context
->db_tab
);
328 mthca_uar_free(to_mdev(ibdev
), &context
->uar
);
333 if (ib_copy_to_udata(udata
, &uresp
, sizeof uresp
)) {
334 mthca_cleanup_user_db_tab(to_mdev(ibdev
), &context
->uar
, context
->db_tab
);
335 mthca_uar_free(to_mdev(ibdev
), &context
->uar
);
337 return ERR_PTR(-EFAULT
);
340 context
->reg_mr_warned
= 0;
342 return &context
->ibucontext
;
345 static int mthca_dealloc_ucontext(struct ib_ucontext
*context
)
347 mthca_cleanup_user_db_tab(to_mdev(context
->device
), &to_mucontext(context
)->uar
,
348 to_mucontext(context
)->db_tab
);
349 mthca_uar_free(to_mdev(context
->device
), &to_mucontext(context
)->uar
);
350 kfree(to_mucontext(context
));
355 static int mthca_mmap_uar(struct ib_ucontext
*context
,
356 struct vm_area_struct
*vma
)
358 if (vma
->vm_end
- vma
->vm_start
!= PAGE_SIZE
)
361 vma
->vm_page_prot
= pgprot_noncached(vma
->vm_page_prot
);
363 if (io_remap_pfn_range(vma
, vma
->vm_start
,
364 to_mucontext(context
)->uar
.pfn
,
365 PAGE_SIZE
, vma
->vm_page_prot
))
371 static struct ib_pd
*mthca_alloc_pd(struct ib_device
*ibdev
,
372 struct ib_ucontext
*context
,
373 struct ib_udata
*udata
)
378 pd
= kmalloc(sizeof *pd
, GFP_KERNEL
);
380 return ERR_PTR(-ENOMEM
);
382 err
= mthca_pd_alloc(to_mdev(ibdev
), !context
, pd
);
389 if (ib_copy_to_udata(udata
, &pd
->pd_num
, sizeof (__u32
))) {
390 mthca_pd_free(to_mdev(ibdev
), pd
);
392 return ERR_PTR(-EFAULT
);
399 static int mthca_dealloc_pd(struct ib_pd
*pd
)
401 mthca_pd_free(to_mdev(pd
->device
), to_mpd(pd
));
407 static struct ib_ah
*mthca_ah_create(struct ib_pd
*pd
,
408 struct ib_ah_attr
*ah_attr
)
413 ah
= kmalloc(sizeof *ah
, GFP_ATOMIC
);
415 return ERR_PTR(-ENOMEM
);
417 err
= mthca_create_ah(to_mdev(pd
->device
), to_mpd(pd
), ah_attr
, ah
);
426 static int mthca_ah_destroy(struct ib_ah
*ah
)
428 mthca_destroy_ah(to_mdev(ah
->device
), to_mah(ah
));
434 static struct ib_srq
*mthca_create_srq(struct ib_pd
*pd
,
435 struct ib_srq_init_attr
*init_attr
,
436 struct ib_udata
*udata
)
438 struct mthca_create_srq ucmd
;
439 struct mthca_ucontext
*context
= NULL
;
440 struct mthca_srq
*srq
;
443 if (init_attr
->srq_type
!= IB_SRQT_BASIC
)
444 return ERR_PTR(-ENOSYS
);
446 srq
= kmalloc(sizeof *srq
, GFP_KERNEL
);
448 return ERR_PTR(-ENOMEM
);
451 context
= to_mucontext(pd
->uobject
->context
);
453 if (ib_copy_from_udata(&ucmd
, udata
, sizeof ucmd
)) {
458 err
= mthca_map_user_db(to_mdev(pd
->device
), &context
->uar
,
459 context
->db_tab
, ucmd
.db_index
,
465 srq
->mr
.ibmr
.lkey
= ucmd
.lkey
;
466 srq
->db_index
= ucmd
.db_index
;
469 err
= mthca_alloc_srq(to_mdev(pd
->device
), to_mpd(pd
),
470 &init_attr
->attr
, srq
);
472 if (err
&& pd
->uobject
)
473 mthca_unmap_user_db(to_mdev(pd
->device
), &context
->uar
,
474 context
->db_tab
, ucmd
.db_index
);
479 if (context
&& ib_copy_to_udata(udata
, &srq
->srqn
, sizeof (__u32
))) {
480 mthca_free_srq(to_mdev(pd
->device
), srq
);
493 static int mthca_destroy_srq(struct ib_srq
*srq
)
495 struct mthca_ucontext
*context
;
498 context
= to_mucontext(srq
->uobject
->context
);
500 mthca_unmap_user_db(to_mdev(srq
->device
), &context
->uar
,
501 context
->db_tab
, to_msrq(srq
)->db_index
);
504 mthca_free_srq(to_mdev(srq
->device
), to_msrq(srq
));
510 static struct ib_qp
*mthca_create_qp(struct ib_pd
*pd
,
511 struct ib_qp_init_attr
*init_attr
,
512 struct ib_udata
*udata
)
514 struct mthca_create_qp ucmd
;
518 if (init_attr
->create_flags
)
519 return ERR_PTR(-EINVAL
);
521 switch (init_attr
->qp_type
) {
526 struct mthca_ucontext
*context
;
528 qp
= kmalloc(sizeof *qp
, GFP_KERNEL
);
530 return ERR_PTR(-ENOMEM
);
533 context
= to_mucontext(pd
->uobject
->context
);
535 if (ib_copy_from_udata(&ucmd
, udata
, sizeof ucmd
)) {
537 return ERR_PTR(-EFAULT
);
540 err
= mthca_map_user_db(to_mdev(pd
->device
), &context
->uar
,
542 ucmd
.sq_db_index
, ucmd
.sq_db_page
);
548 err
= mthca_map_user_db(to_mdev(pd
->device
), &context
->uar
,
550 ucmd
.rq_db_index
, ucmd
.rq_db_page
);
552 mthca_unmap_user_db(to_mdev(pd
->device
),
560 qp
->mr
.ibmr
.lkey
= ucmd
.lkey
;
561 qp
->sq
.db_index
= ucmd
.sq_db_index
;
562 qp
->rq
.db_index
= ucmd
.rq_db_index
;
565 err
= mthca_alloc_qp(to_mdev(pd
->device
), to_mpd(pd
),
566 to_mcq(init_attr
->send_cq
),
567 to_mcq(init_attr
->recv_cq
),
568 init_attr
->qp_type
, init_attr
->sq_sig_type
,
569 &init_attr
->cap
, qp
);
571 if (err
&& pd
->uobject
) {
572 context
= to_mucontext(pd
->uobject
->context
);
574 mthca_unmap_user_db(to_mdev(pd
->device
),
578 mthca_unmap_user_db(to_mdev(pd
->device
),
584 qp
->ibqp
.qp_num
= qp
->qpn
;
590 /* Don't allow userspace to create special QPs */
592 return ERR_PTR(-EINVAL
);
594 qp
= kmalloc(sizeof (struct mthca_sqp
), GFP_KERNEL
);
596 return ERR_PTR(-ENOMEM
);
598 qp
->ibqp
.qp_num
= init_attr
->qp_type
== IB_QPT_SMI
? 0 : 1;
600 err
= mthca_alloc_sqp(to_mdev(pd
->device
), to_mpd(pd
),
601 to_mcq(init_attr
->send_cq
),
602 to_mcq(init_attr
->recv_cq
),
603 init_attr
->sq_sig_type
, &init_attr
->cap
,
604 qp
->ibqp
.qp_num
, init_attr
->port_num
,
609 /* Don't support raw QPs */
610 return ERR_PTR(-ENOSYS
);
618 init_attr
->cap
.max_send_wr
= qp
->sq
.max
;
619 init_attr
->cap
.max_recv_wr
= qp
->rq
.max
;
620 init_attr
->cap
.max_send_sge
= qp
->sq
.max_gs
;
621 init_attr
->cap
.max_recv_sge
= qp
->rq
.max_gs
;
622 init_attr
->cap
.max_inline_data
= qp
->max_inline_data
;
627 static int mthca_destroy_qp(struct ib_qp
*qp
)
630 mthca_unmap_user_db(to_mdev(qp
->device
),
631 &to_mucontext(qp
->uobject
->context
)->uar
,
632 to_mucontext(qp
->uobject
->context
)->db_tab
,
633 to_mqp(qp
)->sq
.db_index
);
634 mthca_unmap_user_db(to_mdev(qp
->device
),
635 &to_mucontext(qp
->uobject
->context
)->uar
,
636 to_mucontext(qp
->uobject
->context
)->db_tab
,
637 to_mqp(qp
)->rq
.db_index
);
639 mthca_free_qp(to_mdev(qp
->device
), to_mqp(qp
));
644 static struct ib_cq
*mthca_create_cq(struct ib_device
*ibdev
, int entries
,
646 struct ib_ucontext
*context
,
647 struct ib_udata
*udata
)
649 struct mthca_create_cq ucmd
;
654 if (entries
< 1 || entries
> to_mdev(ibdev
)->limits
.max_cqes
)
655 return ERR_PTR(-EINVAL
);
658 if (ib_copy_from_udata(&ucmd
, udata
, sizeof ucmd
))
659 return ERR_PTR(-EFAULT
);
661 err
= mthca_map_user_db(to_mdev(ibdev
), &to_mucontext(context
)->uar
,
662 to_mucontext(context
)->db_tab
,
663 ucmd
.set_db_index
, ucmd
.set_db_page
);
667 err
= mthca_map_user_db(to_mdev(ibdev
), &to_mucontext(context
)->uar
,
668 to_mucontext(context
)->db_tab
,
669 ucmd
.arm_db_index
, ucmd
.arm_db_page
);
674 cq
= kmalloc(sizeof *cq
, GFP_KERNEL
);
681 cq
->buf
.mr
.ibmr
.lkey
= ucmd
.lkey
;
682 cq
->set_ci_db_index
= ucmd
.set_db_index
;
683 cq
->arm_db_index
= ucmd
.arm_db_index
;
686 for (nent
= 1; nent
<= entries
; nent
<<= 1)
689 err
= mthca_init_cq(to_mdev(ibdev
), nent
,
690 context
? to_mucontext(context
) : NULL
,
691 context
? ucmd
.pdn
: to_mdev(ibdev
)->driver_pd
.pd_num
,
696 if (context
&& ib_copy_to_udata(udata
, &cq
->cqn
, sizeof (__u32
))) {
697 mthca_free_cq(to_mdev(ibdev
), cq
);
702 cq
->resize_buf
= NULL
;
711 mthca_unmap_user_db(to_mdev(ibdev
), &to_mucontext(context
)->uar
,
712 to_mucontext(context
)->db_tab
, ucmd
.arm_db_index
);
716 mthca_unmap_user_db(to_mdev(ibdev
), &to_mucontext(context
)->uar
,
717 to_mucontext(context
)->db_tab
, ucmd
.set_db_index
);
722 static int mthca_alloc_resize_buf(struct mthca_dev
*dev
, struct mthca_cq
*cq
,
727 spin_lock_irq(&cq
->lock
);
728 if (cq
->resize_buf
) {
733 cq
->resize_buf
= kmalloc(sizeof *cq
->resize_buf
, GFP_ATOMIC
);
734 if (!cq
->resize_buf
) {
739 cq
->resize_buf
->state
= CQ_RESIZE_ALLOC
;
744 spin_unlock_irq(&cq
->lock
);
749 ret
= mthca_alloc_cq_buf(dev
, &cq
->resize_buf
->buf
, entries
);
751 spin_lock_irq(&cq
->lock
);
752 kfree(cq
->resize_buf
);
753 cq
->resize_buf
= NULL
;
754 spin_unlock_irq(&cq
->lock
);
758 cq
->resize_buf
->cqe
= entries
- 1;
760 spin_lock_irq(&cq
->lock
);
761 cq
->resize_buf
->state
= CQ_RESIZE_READY
;
762 spin_unlock_irq(&cq
->lock
);
767 static int mthca_resize_cq(struct ib_cq
*ibcq
, int entries
, struct ib_udata
*udata
)
769 struct mthca_dev
*dev
= to_mdev(ibcq
->device
);
770 struct mthca_cq
*cq
= to_mcq(ibcq
);
771 struct mthca_resize_cq ucmd
;
775 if (entries
< 1 || entries
> dev
->limits
.max_cqes
)
778 mutex_lock(&cq
->mutex
);
780 entries
= roundup_pow_of_two(entries
+ 1);
781 if (entries
== ibcq
->cqe
+ 1) {
787 ret
= mthca_alloc_resize_buf(dev
, cq
, entries
);
790 lkey
= cq
->resize_buf
->buf
.mr
.ibmr
.lkey
;
792 if (ib_copy_from_udata(&ucmd
, udata
, sizeof ucmd
)) {
799 ret
= mthca_RESIZE_CQ(dev
, cq
->cqn
, lkey
, ilog2(entries
));
802 if (cq
->resize_buf
) {
803 mthca_free_cq_buf(dev
, &cq
->resize_buf
->buf
,
804 cq
->resize_buf
->cqe
);
805 kfree(cq
->resize_buf
);
806 spin_lock_irq(&cq
->lock
);
807 cq
->resize_buf
= NULL
;
808 spin_unlock_irq(&cq
->lock
);
814 struct mthca_cq_buf tbuf
;
817 spin_lock_irq(&cq
->lock
);
818 if (cq
->resize_buf
->state
== CQ_RESIZE_READY
) {
819 mthca_cq_resize_copy_cqes(cq
);
822 cq
->buf
= cq
->resize_buf
->buf
;
823 cq
->ibcq
.cqe
= cq
->resize_buf
->cqe
;
825 tbuf
= cq
->resize_buf
->buf
;
826 tcqe
= cq
->resize_buf
->cqe
;
829 kfree(cq
->resize_buf
);
830 cq
->resize_buf
= NULL
;
831 spin_unlock_irq(&cq
->lock
);
833 mthca_free_cq_buf(dev
, &tbuf
, tcqe
);
835 ibcq
->cqe
= entries
- 1;
838 mutex_unlock(&cq
->mutex
);
843 static int mthca_destroy_cq(struct ib_cq
*cq
)
846 mthca_unmap_user_db(to_mdev(cq
->device
),
847 &to_mucontext(cq
->uobject
->context
)->uar
,
848 to_mucontext(cq
->uobject
->context
)->db_tab
,
849 to_mcq(cq
)->arm_db_index
);
850 mthca_unmap_user_db(to_mdev(cq
->device
),
851 &to_mucontext(cq
->uobject
->context
)->uar
,
852 to_mucontext(cq
->uobject
->context
)->db_tab
,
853 to_mcq(cq
)->set_ci_db_index
);
855 mthca_free_cq(to_mdev(cq
->device
), to_mcq(cq
));
861 static inline u32
convert_access(int acc
)
863 return (acc
& IB_ACCESS_REMOTE_ATOMIC
? MTHCA_MPT_FLAG_ATOMIC
: 0) |
864 (acc
& IB_ACCESS_REMOTE_WRITE
? MTHCA_MPT_FLAG_REMOTE_WRITE
: 0) |
865 (acc
& IB_ACCESS_REMOTE_READ
? MTHCA_MPT_FLAG_REMOTE_READ
: 0) |
866 (acc
& IB_ACCESS_LOCAL_WRITE
? MTHCA_MPT_FLAG_LOCAL_WRITE
: 0) |
867 MTHCA_MPT_FLAG_LOCAL_READ
;
870 static struct ib_mr
*mthca_get_dma_mr(struct ib_pd
*pd
, int acc
)
875 mr
= kmalloc(sizeof *mr
, GFP_KERNEL
);
877 return ERR_PTR(-ENOMEM
);
879 err
= mthca_mr_alloc_notrans(to_mdev(pd
->device
),
881 convert_access(acc
), mr
);
893 static struct ib_mr
*mthca_reg_phys_mr(struct ib_pd
*pd
,
894 struct ib_phys_buf
*buffer_list
,
908 mask
= buffer_list
[0].addr
^ *iova_start
;
910 for (i
= 0; i
< num_phys_buf
; ++i
) {
912 mask
|= buffer_list
[i
].addr
;
913 if (i
!= num_phys_buf
- 1)
914 mask
|= buffer_list
[i
].addr
+ buffer_list
[i
].size
;
916 total_size
+= buffer_list
[i
].size
;
919 if (mask
& ~PAGE_MASK
)
920 return ERR_PTR(-EINVAL
);
922 shift
= __ffs(mask
| 1 << 31);
924 buffer_list
[0].size
+= buffer_list
[0].addr
& ((1ULL << shift
) - 1);
925 buffer_list
[0].addr
&= ~0ull << shift
;
927 mr
= kmalloc(sizeof *mr
, GFP_KERNEL
);
929 return ERR_PTR(-ENOMEM
);
932 for (i
= 0; i
< num_phys_buf
; ++i
)
933 npages
+= (buffer_list
[i
].size
+ (1ULL << shift
) - 1) >> shift
;
938 page_list
= kmalloc(npages
* sizeof *page_list
, GFP_KERNEL
);
941 return ERR_PTR(-ENOMEM
);
945 for (i
= 0; i
< num_phys_buf
; ++i
)
947 j
< (buffer_list
[i
].size
+ (1ULL << shift
) - 1) >> shift
;
949 page_list
[n
++] = buffer_list
[i
].addr
+ ((u64
) j
<< shift
);
951 mthca_dbg(to_mdev(pd
->device
), "Registering memory at %llx (iova %llx) "
952 "in PD %x; shift %d, npages %d.\n",
953 (unsigned long long) buffer_list
[0].addr
,
954 (unsigned long long) *iova_start
,
958 err
= mthca_mr_alloc_phys(to_mdev(pd
->device
),
960 page_list
, shift
, npages
,
961 *iova_start
, total_size
,
962 convert_access(acc
), mr
);
976 static struct ib_mr
*mthca_reg_user_mr(struct ib_pd
*pd
, u64 start
, u64 length
,
977 u64 virt
, int acc
, struct ib_udata
*udata
)
979 struct mthca_dev
*dev
= to_mdev(pd
->device
);
980 struct ib_umem_chunk
*chunk
;
982 struct mthca_reg_mr ucmd
;
989 if (udata
->inlen
- sizeof (struct ib_uverbs_cmd_hdr
) < sizeof ucmd
) {
990 if (!to_mucontext(pd
->uobject
->context
)->reg_mr_warned
) {
991 mthca_warn(dev
, "Process '%s' did not pass in MR attrs.\n",
993 mthca_warn(dev
, " Update libmthca to fix this.\n");
995 ++to_mucontext(pd
->uobject
->context
)->reg_mr_warned
;
997 } else if (ib_copy_from_udata(&ucmd
, udata
, sizeof ucmd
))
998 return ERR_PTR(-EFAULT
);
1000 mr
= kmalloc(sizeof *mr
, GFP_KERNEL
);
1002 return ERR_PTR(-ENOMEM
);
1004 mr
->umem
= ib_umem_get(pd
->uobject
->context
, start
, length
, acc
,
1005 ucmd
.mr_attrs
& MTHCA_MR_DMASYNC
);
1007 if (IS_ERR(mr
->umem
)) {
1008 err
= PTR_ERR(mr
->umem
);
1012 shift
= ffs(mr
->umem
->page_size
) - 1;
1015 list_for_each_entry(chunk
, &mr
->umem
->chunk_list
, list
)
1018 mr
->mtt
= mthca_alloc_mtt(dev
, n
);
1019 if (IS_ERR(mr
->mtt
)) {
1020 err
= PTR_ERR(mr
->mtt
);
1024 pages
= (u64
*) __get_free_page(GFP_KERNEL
);
1032 write_mtt_size
= min(mthca_write_mtt_size(dev
), (int) (PAGE_SIZE
/ sizeof *pages
));
1034 list_for_each_entry(chunk
, &mr
->umem
->chunk_list
, list
)
1035 for (j
= 0; j
< chunk
->nmap
; ++j
) {
1036 len
= sg_dma_len(&chunk
->page_list
[j
]) >> shift
;
1037 for (k
= 0; k
< len
; ++k
) {
1038 pages
[i
++] = sg_dma_address(&chunk
->page_list
[j
]) +
1039 mr
->umem
->page_size
* k
;
1041 * Be friendly to write_mtt and pass it chunks
1042 * of appropriate size.
1044 if (i
== write_mtt_size
) {
1045 err
= mthca_write_mtt(dev
, mr
->mtt
, n
, pages
, i
);
1055 err
= mthca_write_mtt(dev
, mr
->mtt
, n
, pages
, i
);
1057 free_page((unsigned long) pages
);
1061 err
= mthca_mr_alloc(dev
, to_mpd(pd
)->pd_num
, shift
, virt
, length
,
1062 convert_access(acc
), mr
);
1070 mthca_free_mtt(dev
, mr
->mtt
);
1073 ib_umem_release(mr
->umem
);
1077 return ERR_PTR(err
);
1080 static int mthca_dereg_mr(struct ib_mr
*mr
)
1082 struct mthca_mr
*mmr
= to_mmr(mr
);
1084 mthca_free_mr(to_mdev(mr
->device
), mmr
);
1086 ib_umem_release(mmr
->umem
);
1092 static struct ib_fmr
*mthca_alloc_fmr(struct ib_pd
*pd
, int mr_access_flags
,
1093 struct ib_fmr_attr
*fmr_attr
)
1095 struct mthca_fmr
*fmr
;
1098 fmr
= kmalloc(sizeof *fmr
, GFP_KERNEL
);
1100 return ERR_PTR(-ENOMEM
);
1102 memcpy(&fmr
->attr
, fmr_attr
, sizeof *fmr_attr
);
1103 err
= mthca_fmr_alloc(to_mdev(pd
->device
), to_mpd(pd
)->pd_num
,
1104 convert_access(mr_access_flags
), fmr
);
1108 return ERR_PTR(err
);
1114 static int mthca_dealloc_fmr(struct ib_fmr
*fmr
)
1116 struct mthca_fmr
*mfmr
= to_mfmr(fmr
);
1119 err
= mthca_free_fmr(to_mdev(fmr
->device
), mfmr
);
1127 static int mthca_unmap_fmr(struct list_head
*fmr_list
)
1131 struct mthca_dev
*mdev
= NULL
;
1133 list_for_each_entry(fmr
, fmr_list
, list
) {
1134 if (mdev
&& to_mdev(fmr
->device
) != mdev
)
1136 mdev
= to_mdev(fmr
->device
);
1142 if (mthca_is_memfree(mdev
)) {
1143 list_for_each_entry(fmr
, fmr_list
, list
)
1144 mthca_arbel_fmr_unmap(mdev
, to_mfmr(fmr
));
1148 list_for_each_entry(fmr
, fmr_list
, list
)
1149 mthca_tavor_fmr_unmap(mdev
, to_mfmr(fmr
));
1151 err
= mthca_SYNC_TPT(mdev
);
1155 static ssize_t
show_rev(struct device
*device
, struct device_attribute
*attr
,
1158 struct mthca_dev
*dev
=
1159 container_of(device
, struct mthca_dev
, ib_dev
.dev
);
1160 return sprintf(buf
, "%x\n", dev
->rev_id
);
1163 static ssize_t
show_fw_ver(struct device
*device
, struct device_attribute
*attr
,
1166 struct mthca_dev
*dev
=
1167 container_of(device
, struct mthca_dev
, ib_dev
.dev
);
1168 return sprintf(buf
, "%d.%d.%d\n", (int) (dev
->fw_ver
>> 32),
1169 (int) (dev
->fw_ver
>> 16) & 0xffff,
1170 (int) dev
->fw_ver
& 0xffff);
1173 static ssize_t
show_hca(struct device
*device
, struct device_attribute
*attr
,
1176 struct mthca_dev
*dev
=
1177 container_of(device
, struct mthca_dev
, ib_dev
.dev
);
1178 switch (dev
->pdev
->device
) {
1179 case PCI_DEVICE_ID_MELLANOX_TAVOR
:
1180 return sprintf(buf
, "MT23108\n");
1181 case PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT
:
1182 return sprintf(buf
, "MT25208 (MT23108 compat mode)\n");
1183 case PCI_DEVICE_ID_MELLANOX_ARBEL
:
1184 return sprintf(buf
, "MT25208\n");
1185 case PCI_DEVICE_ID_MELLANOX_SINAI
:
1186 case PCI_DEVICE_ID_MELLANOX_SINAI_OLD
:
1187 return sprintf(buf
, "MT25204\n");
1189 return sprintf(buf
, "unknown\n");
1193 static ssize_t
show_board(struct device
*device
, struct device_attribute
*attr
,
1196 struct mthca_dev
*dev
=
1197 container_of(device
, struct mthca_dev
, ib_dev
.dev
);
1198 return sprintf(buf
, "%.*s\n", MTHCA_BOARD_ID_LEN
, dev
->board_id
);
1201 static DEVICE_ATTR(hw_rev
, S_IRUGO
, show_rev
, NULL
);
1202 static DEVICE_ATTR(fw_ver
, S_IRUGO
, show_fw_ver
, NULL
);
1203 static DEVICE_ATTR(hca_type
, S_IRUGO
, show_hca
, NULL
);
1204 static DEVICE_ATTR(board_id
, S_IRUGO
, show_board
, NULL
);
1206 static struct device_attribute
*mthca_dev_attributes
[] = {
1213 static int mthca_init_node_data(struct mthca_dev
*dev
)
1215 struct ib_smp
*in_mad
= NULL
;
1216 struct ib_smp
*out_mad
= NULL
;
1219 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
1220 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
1221 if (!in_mad
|| !out_mad
)
1224 init_query_mad(in_mad
);
1225 in_mad
->attr_id
= IB_SMP_ATTR_NODE_DESC
;
1227 err
= mthca_MAD_IFC(dev
, 1, 1,
1228 1, NULL
, NULL
, in_mad
, out_mad
);
1232 memcpy(dev
->ib_dev
.node_desc
, out_mad
->data
, 64);
1234 in_mad
->attr_id
= IB_SMP_ATTR_NODE_INFO
;
1236 err
= mthca_MAD_IFC(dev
, 1, 1,
1237 1, NULL
, NULL
, in_mad
, out_mad
);
1241 if (mthca_is_memfree(dev
))
1242 dev
->rev_id
= be32_to_cpup((__be32
*) (out_mad
->data
+ 32));
1243 memcpy(&dev
->ib_dev
.node_guid
, out_mad
->data
+ 12, 8);
1251 int mthca_register_device(struct mthca_dev
*dev
)
1256 ret
= mthca_init_node_data(dev
);
1260 strlcpy(dev
->ib_dev
.name
, "mthca%d", IB_DEVICE_NAME_MAX
);
1261 dev
->ib_dev
.owner
= THIS_MODULE
;
1263 dev
->ib_dev
.uverbs_abi_ver
= MTHCA_UVERBS_ABI_VERSION
;
1264 dev
->ib_dev
.uverbs_cmd_mask
=
1265 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT
) |
1266 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE
) |
1267 (1ull << IB_USER_VERBS_CMD_QUERY_PORT
) |
1268 (1ull << IB_USER_VERBS_CMD_ALLOC_PD
) |
1269 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD
) |
1270 (1ull << IB_USER_VERBS_CMD_REG_MR
) |
1271 (1ull << IB_USER_VERBS_CMD_DEREG_MR
) |
1272 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL
) |
1273 (1ull << IB_USER_VERBS_CMD_CREATE_CQ
) |
1274 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ
) |
1275 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ
) |
1276 (1ull << IB_USER_VERBS_CMD_CREATE_QP
) |
1277 (1ull << IB_USER_VERBS_CMD_QUERY_QP
) |
1278 (1ull << IB_USER_VERBS_CMD_MODIFY_QP
) |
1279 (1ull << IB_USER_VERBS_CMD_DESTROY_QP
) |
1280 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST
) |
1281 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST
);
1282 dev
->ib_dev
.node_type
= RDMA_NODE_IB_CA
;
1283 dev
->ib_dev
.phys_port_cnt
= dev
->limits
.num_ports
;
1284 dev
->ib_dev
.num_comp_vectors
= 1;
1285 dev
->ib_dev
.dma_device
= &dev
->pdev
->dev
;
1286 dev
->ib_dev
.query_device
= mthca_query_device
;
1287 dev
->ib_dev
.query_port
= mthca_query_port
;
1288 dev
->ib_dev
.modify_device
= mthca_modify_device
;
1289 dev
->ib_dev
.modify_port
= mthca_modify_port
;
1290 dev
->ib_dev
.query_pkey
= mthca_query_pkey
;
1291 dev
->ib_dev
.query_gid
= mthca_query_gid
;
1292 dev
->ib_dev
.alloc_ucontext
= mthca_alloc_ucontext
;
1293 dev
->ib_dev
.dealloc_ucontext
= mthca_dealloc_ucontext
;
1294 dev
->ib_dev
.mmap
= mthca_mmap_uar
;
1295 dev
->ib_dev
.alloc_pd
= mthca_alloc_pd
;
1296 dev
->ib_dev
.dealloc_pd
= mthca_dealloc_pd
;
1297 dev
->ib_dev
.create_ah
= mthca_ah_create
;
1298 dev
->ib_dev
.query_ah
= mthca_ah_query
;
1299 dev
->ib_dev
.destroy_ah
= mthca_ah_destroy
;
1301 if (dev
->mthca_flags
& MTHCA_FLAG_SRQ
) {
1302 dev
->ib_dev
.create_srq
= mthca_create_srq
;
1303 dev
->ib_dev
.modify_srq
= mthca_modify_srq
;
1304 dev
->ib_dev
.query_srq
= mthca_query_srq
;
1305 dev
->ib_dev
.destroy_srq
= mthca_destroy_srq
;
1306 dev
->ib_dev
.uverbs_cmd_mask
|=
1307 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ
) |
1308 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ
) |
1309 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ
) |
1310 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ
);
1312 if (mthca_is_memfree(dev
))
1313 dev
->ib_dev
.post_srq_recv
= mthca_arbel_post_srq_recv
;
1315 dev
->ib_dev
.post_srq_recv
= mthca_tavor_post_srq_recv
;
1318 dev
->ib_dev
.create_qp
= mthca_create_qp
;
1319 dev
->ib_dev
.modify_qp
= mthca_modify_qp
;
1320 dev
->ib_dev
.query_qp
= mthca_query_qp
;
1321 dev
->ib_dev
.destroy_qp
= mthca_destroy_qp
;
1322 dev
->ib_dev
.create_cq
= mthca_create_cq
;
1323 dev
->ib_dev
.resize_cq
= mthca_resize_cq
;
1324 dev
->ib_dev
.destroy_cq
= mthca_destroy_cq
;
1325 dev
->ib_dev
.poll_cq
= mthca_poll_cq
;
1326 dev
->ib_dev
.get_dma_mr
= mthca_get_dma_mr
;
1327 dev
->ib_dev
.reg_phys_mr
= mthca_reg_phys_mr
;
1328 dev
->ib_dev
.reg_user_mr
= mthca_reg_user_mr
;
1329 dev
->ib_dev
.dereg_mr
= mthca_dereg_mr
;
1331 if (dev
->mthca_flags
& MTHCA_FLAG_FMR
) {
1332 dev
->ib_dev
.alloc_fmr
= mthca_alloc_fmr
;
1333 dev
->ib_dev
.unmap_fmr
= mthca_unmap_fmr
;
1334 dev
->ib_dev
.dealloc_fmr
= mthca_dealloc_fmr
;
1335 if (mthca_is_memfree(dev
))
1336 dev
->ib_dev
.map_phys_fmr
= mthca_arbel_map_phys_fmr
;
1338 dev
->ib_dev
.map_phys_fmr
= mthca_tavor_map_phys_fmr
;
1341 dev
->ib_dev
.attach_mcast
= mthca_multicast_attach
;
1342 dev
->ib_dev
.detach_mcast
= mthca_multicast_detach
;
1343 dev
->ib_dev
.process_mad
= mthca_process_mad
;
1345 if (mthca_is_memfree(dev
)) {
1346 dev
->ib_dev
.req_notify_cq
= mthca_arbel_arm_cq
;
1347 dev
->ib_dev
.post_send
= mthca_arbel_post_send
;
1348 dev
->ib_dev
.post_recv
= mthca_arbel_post_receive
;
1350 dev
->ib_dev
.req_notify_cq
= mthca_tavor_arm_cq
;
1351 dev
->ib_dev
.post_send
= mthca_tavor_post_send
;
1352 dev
->ib_dev
.post_recv
= mthca_tavor_post_receive
;
1355 mutex_init(&dev
->cap_mask_mutex
);
1357 ret
= ib_register_device(&dev
->ib_dev
, NULL
);
1361 for (i
= 0; i
< ARRAY_SIZE(mthca_dev_attributes
); ++i
) {
1362 ret
= device_create_file(&dev
->ib_dev
.dev
,
1363 mthca_dev_attributes
[i
]);
1365 ib_unregister_device(&dev
->ib_dev
);
1370 mthca_start_catas_poll(dev
);
1375 void mthca_unregister_device(struct mthca_dev
*dev
)
1377 mthca_stop_catas_poll(dev
);
1378 ib_unregister_device(&dev
->ib_dev
);