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
36 * $Id: mthca_provider.c 4859 2006-01-09 21:55:10Z roland $
39 #include <rdma/ib_smi.h>
40 #include <rdma/ib_umem.h>
41 #include <rdma/ib_user_verbs.h>
43 #include <linux/sched.h>
46 #include "mthca_dev.h"
47 #include "mthca_cmd.h"
48 #include "mthca_user.h"
49 #include "mthca_memfree.h"
51 static void init_query_mad(struct ib_smp
*mad
)
53 mad
->base_version
= 1;
54 mad
->mgmt_class
= IB_MGMT_CLASS_SUBN_LID_ROUTED
;
55 mad
->class_version
= 1;
56 mad
->method
= IB_MGMT_METHOD_GET
;
59 static int mthca_query_device(struct ib_device
*ibdev
,
60 struct ib_device_attr
*props
)
62 struct ib_smp
*in_mad
= NULL
;
63 struct ib_smp
*out_mad
= NULL
;
65 struct mthca_dev
*mdev
= to_mdev(ibdev
);
69 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
70 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
71 if (!in_mad
|| !out_mad
)
74 memset(props
, 0, sizeof *props
);
76 props
->fw_ver
= mdev
->fw_ver
;
78 init_query_mad(in_mad
);
79 in_mad
->attr_id
= IB_SMP_ATTR_NODE_INFO
;
81 err
= mthca_MAD_IFC(mdev
, 1, 1,
82 1, NULL
, NULL
, in_mad
, out_mad
,
91 props
->device_cap_flags
= mdev
->device_cap_flags
;
92 props
->vendor_id
= be32_to_cpup((__be32
*) (out_mad
->data
+ 36)) &
94 props
->vendor_part_id
= be16_to_cpup((__be16
*) (out_mad
->data
+ 30));
95 props
->hw_ver
= be32_to_cpup((__be32
*) (out_mad
->data
+ 32));
96 memcpy(&props
->sys_image_guid
, out_mad
->data
+ 4, 8);
98 props
->max_mr_size
= ~0ull;
99 props
->page_size_cap
= mdev
->limits
.page_size_cap
;
100 props
->max_qp
= mdev
->limits
.num_qps
- mdev
->limits
.reserved_qps
;
101 props
->max_qp_wr
= mdev
->limits
.max_wqes
;
102 props
->max_sge
= mdev
->limits
.max_sg
;
103 props
->max_cq
= mdev
->limits
.num_cqs
- mdev
->limits
.reserved_cqs
;
104 props
->max_cqe
= mdev
->limits
.max_cqes
;
105 props
->max_mr
= mdev
->limits
.num_mpts
- mdev
->limits
.reserved_mrws
;
106 props
->max_pd
= mdev
->limits
.num_pds
- mdev
->limits
.reserved_pds
;
107 props
->max_qp_rd_atom
= 1 << mdev
->qp_table
.rdb_shift
;
108 props
->max_qp_init_rd_atom
= mdev
->limits
.max_qp_init_rdma
;
109 props
->max_res_rd_atom
= props
->max_qp_rd_atom
* props
->max_qp
;
110 props
->max_srq
= mdev
->limits
.num_srqs
- mdev
->limits
.reserved_srqs
;
111 props
->max_srq_wr
= mdev
->limits
.max_srq_wqes
;
112 props
->max_srq_sge
= mdev
->limits
.max_srq_sge
;
113 props
->local_ca_ack_delay
= mdev
->limits
.local_ca_ack_delay
;
114 props
->atomic_cap
= mdev
->limits
.flags
& DEV_LIM_FLAG_ATOMIC
?
115 IB_ATOMIC_HCA
: IB_ATOMIC_NONE
;
116 props
->max_pkeys
= mdev
->limits
.pkey_table_len
;
117 props
->max_mcast_grp
= mdev
->limits
.num_mgms
+ mdev
->limits
.num_amgms
;
118 props
->max_mcast_qp_attach
= MTHCA_QP_PER_MGM
;
119 props
->max_total_mcast_qp_attach
= props
->max_mcast_qp_attach
*
120 props
->max_mcast_grp
;
122 * If Sinai memory key optimization is being used, then only
123 * the 8-bit key portion will change. For other HCAs, the
124 * unused index bits will also be used for FMR remapping.
126 if (mdev
->mthca_flags
& MTHCA_FLAG_SINAI_OPT
)
127 props
->max_map_per_fmr
= 255;
129 props
->max_map_per_fmr
=
130 (1 << (32 - ilog2(mdev
->limits
.num_mpts
))) - 1;
139 static int mthca_query_port(struct ib_device
*ibdev
,
140 u8 port
, struct ib_port_attr
*props
)
142 struct ib_smp
*in_mad
= NULL
;
143 struct ib_smp
*out_mad
= NULL
;
147 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
148 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
149 if (!in_mad
|| !out_mad
)
152 memset(props
, 0, sizeof *props
);
154 init_query_mad(in_mad
);
155 in_mad
->attr_id
= IB_SMP_ATTR_PORT_INFO
;
156 in_mad
->attr_mod
= cpu_to_be32(port
);
158 err
= mthca_MAD_IFC(to_mdev(ibdev
), 1, 1,
159 port
, NULL
, NULL
, in_mad
, out_mad
,
168 props
->lid
= be16_to_cpup((__be16
*) (out_mad
->data
+ 16));
169 props
->lmc
= out_mad
->data
[34] & 0x7;
170 props
->sm_lid
= be16_to_cpup((__be16
*) (out_mad
->data
+ 18));
171 props
->sm_sl
= out_mad
->data
[36] & 0xf;
172 props
->state
= out_mad
->data
[32] & 0xf;
173 props
->phys_state
= out_mad
->data
[33] >> 4;
174 props
->port_cap_flags
= be32_to_cpup((__be32
*) (out_mad
->data
+ 20));
175 props
->gid_tbl_len
= to_mdev(ibdev
)->limits
.gid_table_len
;
176 props
->max_msg_sz
= 0x80000000;
177 props
->pkey_tbl_len
= to_mdev(ibdev
)->limits
.pkey_table_len
;
178 props
->bad_pkey_cntr
= be16_to_cpup((__be16
*) (out_mad
->data
+ 46));
179 props
->qkey_viol_cntr
= be16_to_cpup((__be16
*) (out_mad
->data
+ 48));
180 props
->active_width
= out_mad
->data
[31] & 0xf;
181 props
->active_speed
= out_mad
->data
[35] >> 4;
182 props
->max_mtu
= out_mad
->data
[41] & 0xf;
183 props
->active_mtu
= out_mad
->data
[36] >> 4;
184 props
->subnet_timeout
= out_mad
->data
[51] & 0x1f;
185 props
->max_vl_num
= out_mad
->data
[37] >> 4;
186 props
->init_type_reply
= out_mad
->data
[41] >> 4;
194 static int mthca_modify_device(struct ib_device
*ibdev
,
196 struct ib_device_modify
*props
)
198 if (mask
& ~IB_DEVICE_MODIFY_NODE_DESC
)
201 if (mask
& IB_DEVICE_MODIFY_NODE_DESC
) {
202 if (mutex_lock_interruptible(&to_mdev(ibdev
)->cap_mask_mutex
))
204 memcpy(ibdev
->node_desc
, props
->node_desc
, 64);
205 mutex_unlock(&to_mdev(ibdev
)->cap_mask_mutex
);
211 static int mthca_modify_port(struct ib_device
*ibdev
,
212 u8 port
, int port_modify_mask
,
213 struct ib_port_modify
*props
)
215 struct mthca_set_ib_param set_ib
;
216 struct ib_port_attr attr
;
220 if (mutex_lock_interruptible(&to_mdev(ibdev
)->cap_mask_mutex
))
223 err
= mthca_query_port(ibdev
, port
, &attr
);
227 set_ib
.set_si_guid
= 0;
228 set_ib
.reset_qkey_viol
= !!(port_modify_mask
& IB_PORT_RESET_QKEY_CNTR
);
230 set_ib
.cap_mask
= (attr
.port_cap_flags
| props
->set_port_cap_mask
) &
231 ~props
->clr_port_cap_mask
;
233 err
= mthca_SET_IB(to_mdev(ibdev
), &set_ib
, port
, &status
);
242 mutex_unlock(&to_mdev(ibdev
)->cap_mask_mutex
);
246 static int mthca_query_pkey(struct ib_device
*ibdev
,
247 u8 port
, u16 index
, u16
*pkey
)
249 struct ib_smp
*in_mad
= NULL
;
250 struct ib_smp
*out_mad
= NULL
;
254 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
255 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
256 if (!in_mad
|| !out_mad
)
259 init_query_mad(in_mad
);
260 in_mad
->attr_id
= IB_SMP_ATTR_PKEY_TABLE
;
261 in_mad
->attr_mod
= cpu_to_be32(index
/ 32);
263 err
= mthca_MAD_IFC(to_mdev(ibdev
), 1, 1,
264 port
, NULL
, NULL
, in_mad
, out_mad
,
273 *pkey
= be16_to_cpu(((__be16
*) out_mad
->data
)[index
% 32]);
281 static int mthca_query_gid(struct ib_device
*ibdev
, u8 port
,
282 int index
, union ib_gid
*gid
)
284 struct ib_smp
*in_mad
= NULL
;
285 struct ib_smp
*out_mad
= NULL
;
289 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
290 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
291 if (!in_mad
|| !out_mad
)
294 init_query_mad(in_mad
);
295 in_mad
->attr_id
= IB_SMP_ATTR_PORT_INFO
;
296 in_mad
->attr_mod
= cpu_to_be32(port
);
298 err
= mthca_MAD_IFC(to_mdev(ibdev
), 1, 1,
299 port
, NULL
, NULL
, in_mad
, out_mad
,
308 memcpy(gid
->raw
, out_mad
->data
+ 8, 8);
310 init_query_mad(in_mad
);
311 in_mad
->attr_id
= IB_SMP_ATTR_GUID_INFO
;
312 in_mad
->attr_mod
= cpu_to_be32(index
/ 8);
314 err
= mthca_MAD_IFC(to_mdev(ibdev
), 1, 1,
315 port
, NULL
, NULL
, in_mad
, out_mad
,
324 memcpy(gid
->raw
+ 8, out_mad
->data
+ (index
% 8) * 8, 8);
332 static struct ib_ucontext
*mthca_alloc_ucontext(struct ib_device
*ibdev
,
333 struct ib_udata
*udata
)
335 struct mthca_alloc_ucontext_resp uresp
;
336 struct mthca_ucontext
*context
;
339 memset(&uresp
, 0, sizeof uresp
);
341 uresp
.qp_tab_size
= to_mdev(ibdev
)->limits
.num_qps
;
342 if (mthca_is_memfree(to_mdev(ibdev
)))
343 uresp
.uarc_size
= to_mdev(ibdev
)->uar_table
.uarc_size
;
347 context
= kmalloc(sizeof *context
, GFP_KERNEL
);
349 return ERR_PTR(-ENOMEM
);
351 err
= mthca_uar_alloc(to_mdev(ibdev
), &context
->uar
);
357 context
->db_tab
= mthca_init_user_db_tab(to_mdev(ibdev
));
358 if (IS_ERR(context
->db_tab
)) {
359 err
= PTR_ERR(context
->db_tab
);
360 mthca_uar_free(to_mdev(ibdev
), &context
->uar
);
365 if (ib_copy_to_udata(udata
, &uresp
, sizeof uresp
)) {
366 mthca_cleanup_user_db_tab(to_mdev(ibdev
), &context
->uar
, context
->db_tab
);
367 mthca_uar_free(to_mdev(ibdev
), &context
->uar
);
369 return ERR_PTR(-EFAULT
);
372 context
->reg_mr_warned
= 0;
374 return &context
->ibucontext
;
377 static int mthca_dealloc_ucontext(struct ib_ucontext
*context
)
379 mthca_cleanup_user_db_tab(to_mdev(context
->device
), &to_mucontext(context
)->uar
,
380 to_mucontext(context
)->db_tab
);
381 mthca_uar_free(to_mdev(context
->device
), &to_mucontext(context
)->uar
);
382 kfree(to_mucontext(context
));
387 static int mthca_mmap_uar(struct ib_ucontext
*context
,
388 struct vm_area_struct
*vma
)
390 if (vma
->vm_end
- vma
->vm_start
!= PAGE_SIZE
)
393 vma
->vm_page_prot
= pgprot_noncached(vma
->vm_page_prot
);
395 if (io_remap_pfn_range(vma
, vma
->vm_start
,
396 to_mucontext(context
)->uar
.pfn
,
397 PAGE_SIZE
, vma
->vm_page_prot
))
403 static struct ib_pd
*mthca_alloc_pd(struct ib_device
*ibdev
,
404 struct ib_ucontext
*context
,
405 struct ib_udata
*udata
)
410 pd
= kmalloc(sizeof *pd
, GFP_KERNEL
);
412 return ERR_PTR(-ENOMEM
);
414 err
= mthca_pd_alloc(to_mdev(ibdev
), !context
, pd
);
421 if (ib_copy_to_udata(udata
, &pd
->pd_num
, sizeof (__u32
))) {
422 mthca_pd_free(to_mdev(ibdev
), pd
);
424 return ERR_PTR(-EFAULT
);
431 static int mthca_dealloc_pd(struct ib_pd
*pd
)
433 mthca_pd_free(to_mdev(pd
->device
), to_mpd(pd
));
439 static struct ib_ah
*mthca_ah_create(struct ib_pd
*pd
,
440 struct ib_ah_attr
*ah_attr
)
445 ah
= kmalloc(sizeof *ah
, GFP_ATOMIC
);
447 return ERR_PTR(-ENOMEM
);
449 err
= mthca_create_ah(to_mdev(pd
->device
), to_mpd(pd
), ah_attr
, ah
);
458 static int mthca_ah_destroy(struct ib_ah
*ah
)
460 mthca_destroy_ah(to_mdev(ah
->device
), to_mah(ah
));
466 static struct ib_srq
*mthca_create_srq(struct ib_pd
*pd
,
467 struct ib_srq_init_attr
*init_attr
,
468 struct ib_udata
*udata
)
470 struct mthca_create_srq ucmd
;
471 struct mthca_ucontext
*context
= NULL
;
472 struct mthca_srq
*srq
;
475 srq
= kmalloc(sizeof *srq
, GFP_KERNEL
);
477 return ERR_PTR(-ENOMEM
);
480 context
= to_mucontext(pd
->uobject
->context
);
482 if (ib_copy_from_udata(&ucmd
, udata
, sizeof ucmd
)) {
487 err
= mthca_map_user_db(to_mdev(pd
->device
), &context
->uar
,
488 context
->db_tab
, ucmd
.db_index
,
494 srq
->mr
.ibmr
.lkey
= ucmd
.lkey
;
495 srq
->db_index
= ucmd
.db_index
;
498 err
= mthca_alloc_srq(to_mdev(pd
->device
), to_mpd(pd
),
499 &init_attr
->attr
, srq
);
501 if (err
&& pd
->uobject
)
502 mthca_unmap_user_db(to_mdev(pd
->device
), &context
->uar
,
503 context
->db_tab
, ucmd
.db_index
);
508 if (context
&& ib_copy_to_udata(udata
, &srq
->srqn
, sizeof (__u32
))) {
509 mthca_free_srq(to_mdev(pd
->device
), srq
);
522 static int mthca_destroy_srq(struct ib_srq
*srq
)
524 struct mthca_ucontext
*context
;
527 context
= to_mucontext(srq
->uobject
->context
);
529 mthca_unmap_user_db(to_mdev(srq
->device
), &context
->uar
,
530 context
->db_tab
, to_msrq(srq
)->db_index
);
533 mthca_free_srq(to_mdev(srq
->device
), to_msrq(srq
));
539 static struct ib_qp
*mthca_create_qp(struct ib_pd
*pd
,
540 struct ib_qp_init_attr
*init_attr
,
541 struct ib_udata
*udata
)
543 struct mthca_create_qp ucmd
;
547 if (init_attr
->create_flags
)
548 return ERR_PTR(-EINVAL
);
550 switch (init_attr
->qp_type
) {
555 struct mthca_ucontext
*context
;
557 qp
= kmalloc(sizeof *qp
, GFP_KERNEL
);
559 return ERR_PTR(-ENOMEM
);
562 context
= to_mucontext(pd
->uobject
->context
);
564 if (ib_copy_from_udata(&ucmd
, udata
, sizeof ucmd
)) {
566 return ERR_PTR(-EFAULT
);
569 err
= mthca_map_user_db(to_mdev(pd
->device
), &context
->uar
,
571 ucmd
.sq_db_index
, ucmd
.sq_db_page
);
577 err
= mthca_map_user_db(to_mdev(pd
->device
), &context
->uar
,
579 ucmd
.rq_db_index
, ucmd
.rq_db_page
);
581 mthca_unmap_user_db(to_mdev(pd
->device
),
589 qp
->mr
.ibmr
.lkey
= ucmd
.lkey
;
590 qp
->sq
.db_index
= ucmd
.sq_db_index
;
591 qp
->rq
.db_index
= ucmd
.rq_db_index
;
594 err
= mthca_alloc_qp(to_mdev(pd
->device
), to_mpd(pd
),
595 to_mcq(init_attr
->send_cq
),
596 to_mcq(init_attr
->recv_cq
),
597 init_attr
->qp_type
, init_attr
->sq_sig_type
,
598 &init_attr
->cap
, qp
);
600 if (err
&& pd
->uobject
) {
601 context
= to_mucontext(pd
->uobject
->context
);
603 mthca_unmap_user_db(to_mdev(pd
->device
),
607 mthca_unmap_user_db(to_mdev(pd
->device
),
613 qp
->ibqp
.qp_num
= qp
->qpn
;
619 /* Don't allow userspace to create special QPs */
621 return ERR_PTR(-EINVAL
);
623 qp
= kmalloc(sizeof (struct mthca_sqp
), GFP_KERNEL
);
625 return ERR_PTR(-ENOMEM
);
627 qp
->ibqp
.qp_num
= init_attr
->qp_type
== IB_QPT_SMI
? 0 : 1;
629 err
= mthca_alloc_sqp(to_mdev(pd
->device
), to_mpd(pd
),
630 to_mcq(init_attr
->send_cq
),
631 to_mcq(init_attr
->recv_cq
),
632 init_attr
->sq_sig_type
, &init_attr
->cap
,
633 qp
->ibqp
.qp_num
, init_attr
->port_num
,
638 /* Don't support raw QPs */
639 return ERR_PTR(-ENOSYS
);
647 init_attr
->cap
.max_send_wr
= qp
->sq
.max
;
648 init_attr
->cap
.max_recv_wr
= qp
->rq
.max
;
649 init_attr
->cap
.max_send_sge
= qp
->sq
.max_gs
;
650 init_attr
->cap
.max_recv_sge
= qp
->rq
.max_gs
;
651 init_attr
->cap
.max_inline_data
= qp
->max_inline_data
;
656 static int mthca_destroy_qp(struct ib_qp
*qp
)
659 mthca_unmap_user_db(to_mdev(qp
->device
),
660 &to_mucontext(qp
->uobject
->context
)->uar
,
661 to_mucontext(qp
->uobject
->context
)->db_tab
,
662 to_mqp(qp
)->sq
.db_index
);
663 mthca_unmap_user_db(to_mdev(qp
->device
),
664 &to_mucontext(qp
->uobject
->context
)->uar
,
665 to_mucontext(qp
->uobject
->context
)->db_tab
,
666 to_mqp(qp
)->rq
.db_index
);
668 mthca_free_qp(to_mdev(qp
->device
), to_mqp(qp
));
673 static struct ib_cq
*mthca_create_cq(struct ib_device
*ibdev
, int entries
,
675 struct ib_ucontext
*context
,
676 struct ib_udata
*udata
)
678 struct mthca_create_cq ucmd
;
683 if (entries
< 1 || entries
> to_mdev(ibdev
)->limits
.max_cqes
)
684 return ERR_PTR(-EINVAL
);
687 if (ib_copy_from_udata(&ucmd
, udata
, sizeof ucmd
))
688 return ERR_PTR(-EFAULT
);
690 err
= mthca_map_user_db(to_mdev(ibdev
), &to_mucontext(context
)->uar
,
691 to_mucontext(context
)->db_tab
,
692 ucmd
.set_db_index
, ucmd
.set_db_page
);
696 err
= mthca_map_user_db(to_mdev(ibdev
), &to_mucontext(context
)->uar
,
697 to_mucontext(context
)->db_tab
,
698 ucmd
.arm_db_index
, ucmd
.arm_db_page
);
703 cq
= kmalloc(sizeof *cq
, GFP_KERNEL
);
710 cq
->buf
.mr
.ibmr
.lkey
= ucmd
.lkey
;
711 cq
->set_ci_db_index
= ucmd
.set_db_index
;
712 cq
->arm_db_index
= ucmd
.arm_db_index
;
715 for (nent
= 1; nent
<= entries
; nent
<<= 1)
718 err
= mthca_init_cq(to_mdev(ibdev
), nent
,
719 context
? to_mucontext(context
) : NULL
,
720 context
? ucmd
.pdn
: to_mdev(ibdev
)->driver_pd
.pd_num
,
725 if (context
&& ib_copy_to_udata(udata
, &cq
->cqn
, sizeof (__u32
))) {
726 mthca_free_cq(to_mdev(ibdev
), cq
);
730 cq
->resize_buf
= NULL
;
739 mthca_unmap_user_db(to_mdev(ibdev
), &to_mucontext(context
)->uar
,
740 to_mucontext(context
)->db_tab
, ucmd
.arm_db_index
);
744 mthca_unmap_user_db(to_mdev(ibdev
), &to_mucontext(context
)->uar
,
745 to_mucontext(context
)->db_tab
, ucmd
.set_db_index
);
750 static int mthca_alloc_resize_buf(struct mthca_dev
*dev
, struct mthca_cq
*cq
,
755 spin_lock_irq(&cq
->lock
);
756 if (cq
->resize_buf
) {
761 cq
->resize_buf
= kmalloc(sizeof *cq
->resize_buf
, GFP_ATOMIC
);
762 if (!cq
->resize_buf
) {
767 cq
->resize_buf
->state
= CQ_RESIZE_ALLOC
;
772 spin_unlock_irq(&cq
->lock
);
777 ret
= mthca_alloc_cq_buf(dev
, &cq
->resize_buf
->buf
, entries
);
779 spin_lock_irq(&cq
->lock
);
780 kfree(cq
->resize_buf
);
781 cq
->resize_buf
= NULL
;
782 spin_unlock_irq(&cq
->lock
);
786 cq
->resize_buf
->cqe
= entries
- 1;
788 spin_lock_irq(&cq
->lock
);
789 cq
->resize_buf
->state
= CQ_RESIZE_READY
;
790 spin_unlock_irq(&cq
->lock
);
795 static int mthca_resize_cq(struct ib_cq
*ibcq
, int entries
, struct ib_udata
*udata
)
797 struct mthca_dev
*dev
= to_mdev(ibcq
->device
);
798 struct mthca_cq
*cq
= to_mcq(ibcq
);
799 struct mthca_resize_cq ucmd
;
804 if (entries
< 1 || entries
> dev
->limits
.max_cqes
)
807 mutex_lock(&cq
->mutex
);
809 entries
= roundup_pow_of_two(entries
+ 1);
810 if (entries
== ibcq
->cqe
+ 1) {
816 ret
= mthca_alloc_resize_buf(dev
, cq
, entries
);
819 lkey
= cq
->resize_buf
->buf
.mr
.ibmr
.lkey
;
821 if (ib_copy_from_udata(&ucmd
, udata
, sizeof ucmd
)) {
828 ret
= mthca_RESIZE_CQ(dev
, cq
->cqn
, lkey
, ilog2(entries
), &status
);
833 if (cq
->resize_buf
) {
834 mthca_free_cq_buf(dev
, &cq
->resize_buf
->buf
,
835 cq
->resize_buf
->cqe
);
836 kfree(cq
->resize_buf
);
837 spin_lock_irq(&cq
->lock
);
838 cq
->resize_buf
= NULL
;
839 spin_unlock_irq(&cq
->lock
);
845 struct mthca_cq_buf tbuf
;
848 spin_lock_irq(&cq
->lock
);
849 if (cq
->resize_buf
->state
== CQ_RESIZE_READY
) {
850 mthca_cq_resize_copy_cqes(cq
);
853 cq
->buf
= cq
->resize_buf
->buf
;
854 cq
->ibcq
.cqe
= cq
->resize_buf
->cqe
;
856 tbuf
= cq
->resize_buf
->buf
;
857 tcqe
= cq
->resize_buf
->cqe
;
860 kfree(cq
->resize_buf
);
861 cq
->resize_buf
= NULL
;
862 spin_unlock_irq(&cq
->lock
);
864 mthca_free_cq_buf(dev
, &tbuf
, tcqe
);
866 ibcq
->cqe
= entries
- 1;
869 mutex_unlock(&cq
->mutex
);
874 static int mthca_destroy_cq(struct ib_cq
*cq
)
877 mthca_unmap_user_db(to_mdev(cq
->device
),
878 &to_mucontext(cq
->uobject
->context
)->uar
,
879 to_mucontext(cq
->uobject
->context
)->db_tab
,
880 to_mcq(cq
)->arm_db_index
);
881 mthca_unmap_user_db(to_mdev(cq
->device
),
882 &to_mucontext(cq
->uobject
->context
)->uar
,
883 to_mucontext(cq
->uobject
->context
)->db_tab
,
884 to_mcq(cq
)->set_ci_db_index
);
886 mthca_free_cq(to_mdev(cq
->device
), to_mcq(cq
));
892 static inline u32
convert_access(int acc
)
894 return (acc
& IB_ACCESS_REMOTE_ATOMIC
? MTHCA_MPT_FLAG_ATOMIC
: 0) |
895 (acc
& IB_ACCESS_REMOTE_WRITE
? MTHCA_MPT_FLAG_REMOTE_WRITE
: 0) |
896 (acc
& IB_ACCESS_REMOTE_READ
? MTHCA_MPT_FLAG_REMOTE_READ
: 0) |
897 (acc
& IB_ACCESS_LOCAL_WRITE
? MTHCA_MPT_FLAG_LOCAL_WRITE
: 0) |
898 MTHCA_MPT_FLAG_LOCAL_READ
;
901 static struct ib_mr
*mthca_get_dma_mr(struct ib_pd
*pd
, int acc
)
906 mr
= kmalloc(sizeof *mr
, GFP_KERNEL
);
908 return ERR_PTR(-ENOMEM
);
910 err
= mthca_mr_alloc_notrans(to_mdev(pd
->device
),
912 convert_access(acc
), mr
);
924 static struct ib_mr
*mthca_reg_phys_mr(struct ib_pd
*pd
,
925 struct ib_phys_buf
*buffer_list
,
939 mask
= buffer_list
[0].addr
^ *iova_start
;
941 for (i
= 0; i
< num_phys_buf
; ++i
) {
943 mask
|= buffer_list
[i
].addr
;
944 if (i
!= num_phys_buf
- 1)
945 mask
|= buffer_list
[i
].addr
+ buffer_list
[i
].size
;
947 total_size
+= buffer_list
[i
].size
;
950 if (mask
& ~PAGE_MASK
)
951 return ERR_PTR(-EINVAL
);
953 shift
= __ffs(mask
| 1 << 31);
955 buffer_list
[0].size
+= buffer_list
[0].addr
& ((1ULL << shift
) - 1);
956 buffer_list
[0].addr
&= ~0ull << shift
;
958 mr
= kmalloc(sizeof *mr
, GFP_KERNEL
);
960 return ERR_PTR(-ENOMEM
);
963 for (i
= 0; i
< num_phys_buf
; ++i
)
964 npages
+= (buffer_list
[i
].size
+ (1ULL << shift
) - 1) >> shift
;
969 page_list
= kmalloc(npages
* sizeof *page_list
, GFP_KERNEL
);
972 return ERR_PTR(-ENOMEM
);
976 for (i
= 0; i
< num_phys_buf
; ++i
)
978 j
< (buffer_list
[i
].size
+ (1ULL << shift
) - 1) >> shift
;
980 page_list
[n
++] = buffer_list
[i
].addr
+ ((u64
) j
<< shift
);
982 mthca_dbg(to_mdev(pd
->device
), "Registering memory at %llx (iova %llx) "
983 "in PD %x; shift %d, npages %d.\n",
984 (unsigned long long) buffer_list
[0].addr
,
985 (unsigned long long) *iova_start
,
989 err
= mthca_mr_alloc_phys(to_mdev(pd
->device
),
991 page_list
, shift
, npages
,
992 *iova_start
, total_size
,
993 convert_access(acc
), mr
);
1007 static struct ib_mr
*mthca_reg_user_mr(struct ib_pd
*pd
, u64 start
, u64 length
,
1008 u64 virt
, int acc
, struct ib_udata
*udata
)
1010 struct mthca_dev
*dev
= to_mdev(pd
->device
);
1011 struct ib_umem_chunk
*chunk
;
1012 struct mthca_mr
*mr
;
1013 struct mthca_reg_mr ucmd
;
1020 if (udata
->inlen
- sizeof (struct ib_uverbs_cmd_hdr
) < sizeof ucmd
) {
1021 if (!to_mucontext(pd
->uobject
->context
)->reg_mr_warned
) {
1022 mthca_warn(dev
, "Process '%s' did not pass in MR attrs.\n",
1024 mthca_warn(dev
, " Update libmthca to fix this.\n");
1026 ++to_mucontext(pd
->uobject
->context
)->reg_mr_warned
;
1028 } else if (ib_copy_from_udata(&ucmd
, udata
, sizeof ucmd
))
1029 return ERR_PTR(-EFAULT
);
1031 mr
= kmalloc(sizeof *mr
, GFP_KERNEL
);
1033 return ERR_PTR(-ENOMEM
);
1035 mr
->umem
= ib_umem_get(pd
->uobject
->context
, start
, length
, acc
,
1036 ucmd
.mr_attrs
& MTHCA_MR_DMASYNC
);
1038 if (IS_ERR(mr
->umem
)) {
1039 err
= PTR_ERR(mr
->umem
);
1043 shift
= ffs(mr
->umem
->page_size
) - 1;
1046 list_for_each_entry(chunk
, &mr
->umem
->chunk_list
, list
)
1049 mr
->mtt
= mthca_alloc_mtt(dev
, n
);
1050 if (IS_ERR(mr
->mtt
)) {
1051 err
= PTR_ERR(mr
->mtt
);
1055 pages
= (u64
*) __get_free_page(GFP_KERNEL
);
1063 write_mtt_size
= min(mthca_write_mtt_size(dev
), (int) (PAGE_SIZE
/ sizeof *pages
));
1065 list_for_each_entry(chunk
, &mr
->umem
->chunk_list
, list
)
1066 for (j
= 0; j
< chunk
->nmap
; ++j
) {
1067 len
= sg_dma_len(&chunk
->page_list
[j
]) >> shift
;
1068 for (k
= 0; k
< len
; ++k
) {
1069 pages
[i
++] = sg_dma_address(&chunk
->page_list
[j
]) +
1070 mr
->umem
->page_size
* k
;
1072 * Be friendly to write_mtt and pass it chunks
1073 * of appropriate size.
1075 if (i
== write_mtt_size
) {
1076 err
= mthca_write_mtt(dev
, mr
->mtt
, n
, pages
, i
);
1086 err
= mthca_write_mtt(dev
, mr
->mtt
, n
, pages
, i
);
1088 free_page((unsigned long) pages
);
1092 err
= mthca_mr_alloc(dev
, to_mpd(pd
)->pd_num
, shift
, virt
, length
,
1093 convert_access(acc
), mr
);
1101 mthca_free_mtt(dev
, mr
->mtt
);
1104 ib_umem_release(mr
->umem
);
1108 return ERR_PTR(err
);
1111 static int mthca_dereg_mr(struct ib_mr
*mr
)
1113 struct mthca_mr
*mmr
= to_mmr(mr
);
1115 mthca_free_mr(to_mdev(mr
->device
), mmr
);
1117 ib_umem_release(mmr
->umem
);
1123 static struct ib_fmr
*mthca_alloc_fmr(struct ib_pd
*pd
, int mr_access_flags
,
1124 struct ib_fmr_attr
*fmr_attr
)
1126 struct mthca_fmr
*fmr
;
1129 fmr
= kmalloc(sizeof *fmr
, GFP_KERNEL
);
1131 return ERR_PTR(-ENOMEM
);
1133 memcpy(&fmr
->attr
, fmr_attr
, sizeof *fmr_attr
);
1134 err
= mthca_fmr_alloc(to_mdev(pd
->device
), to_mpd(pd
)->pd_num
,
1135 convert_access(mr_access_flags
), fmr
);
1139 return ERR_PTR(err
);
1145 static int mthca_dealloc_fmr(struct ib_fmr
*fmr
)
1147 struct mthca_fmr
*mfmr
= to_mfmr(fmr
);
1150 err
= mthca_free_fmr(to_mdev(fmr
->device
), mfmr
);
1158 static int mthca_unmap_fmr(struct list_head
*fmr_list
)
1163 struct mthca_dev
*mdev
= NULL
;
1165 list_for_each_entry(fmr
, fmr_list
, list
) {
1166 if (mdev
&& to_mdev(fmr
->device
) != mdev
)
1168 mdev
= to_mdev(fmr
->device
);
1174 if (mthca_is_memfree(mdev
)) {
1175 list_for_each_entry(fmr
, fmr_list
, list
)
1176 mthca_arbel_fmr_unmap(mdev
, to_mfmr(fmr
));
1180 list_for_each_entry(fmr
, fmr_list
, list
)
1181 mthca_tavor_fmr_unmap(mdev
, to_mfmr(fmr
));
1183 err
= mthca_SYNC_TPT(mdev
, &status
);
1191 static ssize_t
show_rev(struct device
*device
, struct device_attribute
*attr
,
1194 struct mthca_dev
*dev
=
1195 container_of(device
, struct mthca_dev
, ib_dev
.dev
);
1196 return sprintf(buf
, "%x\n", dev
->rev_id
);
1199 static ssize_t
show_fw_ver(struct device
*device
, struct device_attribute
*attr
,
1202 struct mthca_dev
*dev
=
1203 container_of(device
, struct mthca_dev
, ib_dev
.dev
);
1204 return sprintf(buf
, "%d.%d.%d\n", (int) (dev
->fw_ver
>> 32),
1205 (int) (dev
->fw_ver
>> 16) & 0xffff,
1206 (int) dev
->fw_ver
& 0xffff);
1209 static ssize_t
show_hca(struct device
*device
, struct device_attribute
*attr
,
1212 struct mthca_dev
*dev
=
1213 container_of(device
, struct mthca_dev
, ib_dev
.dev
);
1214 switch (dev
->pdev
->device
) {
1215 case PCI_DEVICE_ID_MELLANOX_TAVOR
:
1216 return sprintf(buf
, "MT23108\n");
1217 case PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT
:
1218 return sprintf(buf
, "MT25208 (MT23108 compat mode)\n");
1219 case PCI_DEVICE_ID_MELLANOX_ARBEL
:
1220 return sprintf(buf
, "MT25208\n");
1221 case PCI_DEVICE_ID_MELLANOX_SINAI
:
1222 case PCI_DEVICE_ID_MELLANOX_SINAI_OLD
:
1223 return sprintf(buf
, "MT25204\n");
1225 return sprintf(buf
, "unknown\n");
1229 static ssize_t
show_board(struct device
*device
, struct device_attribute
*attr
,
1232 struct mthca_dev
*dev
=
1233 container_of(device
, struct mthca_dev
, ib_dev
.dev
);
1234 return sprintf(buf
, "%.*s\n", MTHCA_BOARD_ID_LEN
, dev
->board_id
);
1237 static DEVICE_ATTR(hw_rev
, S_IRUGO
, show_rev
, NULL
);
1238 static DEVICE_ATTR(fw_ver
, S_IRUGO
, show_fw_ver
, NULL
);
1239 static DEVICE_ATTR(hca_type
, S_IRUGO
, show_hca
, NULL
);
1240 static DEVICE_ATTR(board_id
, S_IRUGO
, show_board
, NULL
);
1242 static struct device_attribute
*mthca_dev_attributes
[] = {
1249 static int mthca_init_node_data(struct mthca_dev
*dev
)
1251 struct ib_smp
*in_mad
= NULL
;
1252 struct ib_smp
*out_mad
= NULL
;
1256 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
1257 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
1258 if (!in_mad
|| !out_mad
)
1261 init_query_mad(in_mad
);
1262 in_mad
->attr_id
= IB_SMP_ATTR_NODE_DESC
;
1264 err
= mthca_MAD_IFC(dev
, 1, 1,
1265 1, NULL
, NULL
, in_mad
, out_mad
,
1274 memcpy(dev
->ib_dev
.node_desc
, out_mad
->data
, 64);
1276 in_mad
->attr_id
= IB_SMP_ATTR_NODE_INFO
;
1278 err
= mthca_MAD_IFC(dev
, 1, 1,
1279 1, NULL
, NULL
, in_mad
, out_mad
,
1288 if (mthca_is_memfree(dev
))
1289 dev
->rev_id
= be32_to_cpup((__be32
*) (out_mad
->data
+ 32));
1290 memcpy(&dev
->ib_dev
.node_guid
, out_mad
->data
+ 12, 8);
1298 int mthca_register_device(struct mthca_dev
*dev
)
1303 ret
= mthca_init_node_data(dev
);
1307 strlcpy(dev
->ib_dev
.name
, "mthca%d", IB_DEVICE_NAME_MAX
);
1308 dev
->ib_dev
.owner
= THIS_MODULE
;
1310 dev
->ib_dev
.uverbs_abi_ver
= MTHCA_UVERBS_ABI_VERSION
;
1311 dev
->ib_dev
.uverbs_cmd_mask
=
1312 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT
) |
1313 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE
) |
1314 (1ull << IB_USER_VERBS_CMD_QUERY_PORT
) |
1315 (1ull << IB_USER_VERBS_CMD_ALLOC_PD
) |
1316 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD
) |
1317 (1ull << IB_USER_VERBS_CMD_REG_MR
) |
1318 (1ull << IB_USER_VERBS_CMD_DEREG_MR
) |
1319 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL
) |
1320 (1ull << IB_USER_VERBS_CMD_CREATE_CQ
) |
1321 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ
) |
1322 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ
) |
1323 (1ull << IB_USER_VERBS_CMD_CREATE_QP
) |
1324 (1ull << IB_USER_VERBS_CMD_QUERY_QP
) |
1325 (1ull << IB_USER_VERBS_CMD_MODIFY_QP
) |
1326 (1ull << IB_USER_VERBS_CMD_DESTROY_QP
) |
1327 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST
) |
1328 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST
);
1329 dev
->ib_dev
.node_type
= RDMA_NODE_IB_CA
;
1330 dev
->ib_dev
.phys_port_cnt
= dev
->limits
.num_ports
;
1331 dev
->ib_dev
.num_comp_vectors
= 1;
1332 dev
->ib_dev
.dma_device
= &dev
->pdev
->dev
;
1333 dev
->ib_dev
.query_device
= mthca_query_device
;
1334 dev
->ib_dev
.query_port
= mthca_query_port
;
1335 dev
->ib_dev
.modify_device
= mthca_modify_device
;
1336 dev
->ib_dev
.modify_port
= mthca_modify_port
;
1337 dev
->ib_dev
.query_pkey
= mthca_query_pkey
;
1338 dev
->ib_dev
.query_gid
= mthca_query_gid
;
1339 dev
->ib_dev
.alloc_ucontext
= mthca_alloc_ucontext
;
1340 dev
->ib_dev
.dealloc_ucontext
= mthca_dealloc_ucontext
;
1341 dev
->ib_dev
.mmap
= mthca_mmap_uar
;
1342 dev
->ib_dev
.alloc_pd
= mthca_alloc_pd
;
1343 dev
->ib_dev
.dealloc_pd
= mthca_dealloc_pd
;
1344 dev
->ib_dev
.create_ah
= mthca_ah_create
;
1345 dev
->ib_dev
.query_ah
= mthca_ah_query
;
1346 dev
->ib_dev
.destroy_ah
= mthca_ah_destroy
;
1348 if (dev
->mthca_flags
& MTHCA_FLAG_SRQ
) {
1349 dev
->ib_dev
.create_srq
= mthca_create_srq
;
1350 dev
->ib_dev
.modify_srq
= mthca_modify_srq
;
1351 dev
->ib_dev
.query_srq
= mthca_query_srq
;
1352 dev
->ib_dev
.destroy_srq
= mthca_destroy_srq
;
1353 dev
->ib_dev
.uverbs_cmd_mask
|=
1354 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ
) |
1355 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ
) |
1356 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ
) |
1357 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ
);
1359 if (mthca_is_memfree(dev
))
1360 dev
->ib_dev
.post_srq_recv
= mthca_arbel_post_srq_recv
;
1362 dev
->ib_dev
.post_srq_recv
= mthca_tavor_post_srq_recv
;
1365 dev
->ib_dev
.create_qp
= mthca_create_qp
;
1366 dev
->ib_dev
.modify_qp
= mthca_modify_qp
;
1367 dev
->ib_dev
.query_qp
= mthca_query_qp
;
1368 dev
->ib_dev
.destroy_qp
= mthca_destroy_qp
;
1369 dev
->ib_dev
.create_cq
= mthca_create_cq
;
1370 dev
->ib_dev
.resize_cq
= mthca_resize_cq
;
1371 dev
->ib_dev
.destroy_cq
= mthca_destroy_cq
;
1372 dev
->ib_dev
.poll_cq
= mthca_poll_cq
;
1373 dev
->ib_dev
.get_dma_mr
= mthca_get_dma_mr
;
1374 dev
->ib_dev
.reg_phys_mr
= mthca_reg_phys_mr
;
1375 dev
->ib_dev
.reg_user_mr
= mthca_reg_user_mr
;
1376 dev
->ib_dev
.dereg_mr
= mthca_dereg_mr
;
1378 if (dev
->mthca_flags
& MTHCA_FLAG_FMR
) {
1379 dev
->ib_dev
.alloc_fmr
= mthca_alloc_fmr
;
1380 dev
->ib_dev
.unmap_fmr
= mthca_unmap_fmr
;
1381 dev
->ib_dev
.dealloc_fmr
= mthca_dealloc_fmr
;
1382 if (mthca_is_memfree(dev
))
1383 dev
->ib_dev
.map_phys_fmr
= mthca_arbel_map_phys_fmr
;
1385 dev
->ib_dev
.map_phys_fmr
= mthca_tavor_map_phys_fmr
;
1388 dev
->ib_dev
.attach_mcast
= mthca_multicast_attach
;
1389 dev
->ib_dev
.detach_mcast
= mthca_multicast_detach
;
1390 dev
->ib_dev
.process_mad
= mthca_process_mad
;
1392 if (mthca_is_memfree(dev
)) {
1393 dev
->ib_dev
.req_notify_cq
= mthca_arbel_arm_cq
;
1394 dev
->ib_dev
.post_send
= mthca_arbel_post_send
;
1395 dev
->ib_dev
.post_recv
= mthca_arbel_post_receive
;
1397 dev
->ib_dev
.req_notify_cq
= mthca_tavor_arm_cq
;
1398 dev
->ib_dev
.post_send
= mthca_tavor_post_send
;
1399 dev
->ib_dev
.post_recv
= mthca_tavor_post_receive
;
1402 mutex_init(&dev
->cap_mask_mutex
);
1404 ret
= ib_register_device(&dev
->ib_dev
);
1408 for (i
= 0; i
< ARRAY_SIZE(mthca_dev_attributes
); ++i
) {
1409 ret
= device_create_file(&dev
->ib_dev
.dev
,
1410 mthca_dev_attributes
[i
]);
1412 ib_unregister_device(&dev
->ib_dev
);
1417 mthca_start_catas_poll(dev
);
1422 void mthca_unregister_device(struct mthca_dev
*dev
)
1424 mthca_stop_catas_poll(dev
);
1425 ib_unregister_device(&dev
->ib_dev
);