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 memset(&uresp
, 0, sizeof uresp
);
339 uresp
.qp_tab_size
= to_mdev(ibdev
)->limits
.num_qps
;
340 if (mthca_is_memfree(to_mdev(ibdev
)))
341 uresp
.uarc_size
= to_mdev(ibdev
)->uar_table
.uarc_size
;
345 context
= kmalloc(sizeof *context
, GFP_KERNEL
);
347 return ERR_PTR(-ENOMEM
);
349 err
= mthca_uar_alloc(to_mdev(ibdev
), &context
->uar
);
355 context
->db_tab
= mthca_init_user_db_tab(to_mdev(ibdev
));
356 if (IS_ERR(context
->db_tab
)) {
357 err
= PTR_ERR(context
->db_tab
);
358 mthca_uar_free(to_mdev(ibdev
), &context
->uar
);
363 if (ib_copy_to_udata(udata
, &uresp
, sizeof uresp
)) {
364 mthca_cleanup_user_db_tab(to_mdev(ibdev
), &context
->uar
, context
->db_tab
);
365 mthca_uar_free(to_mdev(ibdev
), &context
->uar
);
367 return ERR_PTR(-EFAULT
);
370 context
->reg_mr_warned
= 0;
372 return &context
->ibucontext
;
375 static int mthca_dealloc_ucontext(struct ib_ucontext
*context
)
377 mthca_cleanup_user_db_tab(to_mdev(context
->device
), &to_mucontext(context
)->uar
,
378 to_mucontext(context
)->db_tab
);
379 mthca_uar_free(to_mdev(context
->device
), &to_mucontext(context
)->uar
);
380 kfree(to_mucontext(context
));
385 static int mthca_mmap_uar(struct ib_ucontext
*context
,
386 struct vm_area_struct
*vma
)
388 if (vma
->vm_end
- vma
->vm_start
!= PAGE_SIZE
)
391 vma
->vm_page_prot
= pgprot_noncached(vma
->vm_page_prot
);
393 if (io_remap_pfn_range(vma
, vma
->vm_start
,
394 to_mucontext(context
)->uar
.pfn
,
395 PAGE_SIZE
, vma
->vm_page_prot
))
401 static struct ib_pd
*mthca_alloc_pd(struct ib_device
*ibdev
,
402 struct ib_ucontext
*context
,
403 struct ib_udata
*udata
)
408 pd
= kmalloc(sizeof *pd
, GFP_KERNEL
);
410 return ERR_PTR(-ENOMEM
);
412 err
= mthca_pd_alloc(to_mdev(ibdev
), !context
, pd
);
419 if (ib_copy_to_udata(udata
, &pd
->pd_num
, sizeof (__u32
))) {
420 mthca_pd_free(to_mdev(ibdev
), pd
);
422 return ERR_PTR(-EFAULT
);
429 static int mthca_dealloc_pd(struct ib_pd
*pd
)
431 mthca_pd_free(to_mdev(pd
->device
), to_mpd(pd
));
437 static struct ib_ah
*mthca_ah_create(struct ib_pd
*pd
,
438 struct ib_ah_attr
*ah_attr
)
443 ah
= kmalloc(sizeof *ah
, GFP_ATOMIC
);
445 return ERR_PTR(-ENOMEM
);
447 err
= mthca_create_ah(to_mdev(pd
->device
), to_mpd(pd
), ah_attr
, ah
);
456 static int mthca_ah_destroy(struct ib_ah
*ah
)
458 mthca_destroy_ah(to_mdev(ah
->device
), to_mah(ah
));
464 static struct ib_srq
*mthca_create_srq(struct ib_pd
*pd
,
465 struct ib_srq_init_attr
*init_attr
,
466 struct ib_udata
*udata
)
468 struct mthca_create_srq ucmd
;
469 struct mthca_ucontext
*context
= NULL
;
470 struct mthca_srq
*srq
;
473 srq
= kmalloc(sizeof *srq
, GFP_KERNEL
);
475 return ERR_PTR(-ENOMEM
);
478 context
= to_mucontext(pd
->uobject
->context
);
480 if (ib_copy_from_udata(&ucmd
, udata
, sizeof ucmd
)) {
485 err
= mthca_map_user_db(to_mdev(pd
->device
), &context
->uar
,
486 context
->db_tab
, ucmd
.db_index
,
492 srq
->mr
.ibmr
.lkey
= ucmd
.lkey
;
493 srq
->db_index
= ucmd
.db_index
;
496 err
= mthca_alloc_srq(to_mdev(pd
->device
), to_mpd(pd
),
497 &init_attr
->attr
, srq
);
499 if (err
&& pd
->uobject
)
500 mthca_unmap_user_db(to_mdev(pd
->device
), &context
->uar
,
501 context
->db_tab
, ucmd
.db_index
);
506 if (context
&& ib_copy_to_udata(udata
, &srq
->srqn
, sizeof (__u32
))) {
507 mthca_free_srq(to_mdev(pd
->device
), srq
);
520 static int mthca_destroy_srq(struct ib_srq
*srq
)
522 struct mthca_ucontext
*context
;
525 context
= to_mucontext(srq
->uobject
->context
);
527 mthca_unmap_user_db(to_mdev(srq
->device
), &context
->uar
,
528 context
->db_tab
, to_msrq(srq
)->db_index
);
531 mthca_free_srq(to_mdev(srq
->device
), to_msrq(srq
));
537 static struct ib_qp
*mthca_create_qp(struct ib_pd
*pd
,
538 struct ib_qp_init_attr
*init_attr
,
539 struct ib_udata
*udata
)
541 struct mthca_create_qp ucmd
;
545 if (init_attr
->create_flags
)
546 return ERR_PTR(-EINVAL
);
548 switch (init_attr
->qp_type
) {
553 struct mthca_ucontext
*context
;
555 qp
= kmalloc(sizeof *qp
, GFP_KERNEL
);
557 return ERR_PTR(-ENOMEM
);
560 context
= to_mucontext(pd
->uobject
->context
);
562 if (ib_copy_from_udata(&ucmd
, udata
, sizeof ucmd
)) {
564 return ERR_PTR(-EFAULT
);
567 err
= mthca_map_user_db(to_mdev(pd
->device
), &context
->uar
,
569 ucmd
.sq_db_index
, ucmd
.sq_db_page
);
575 err
= mthca_map_user_db(to_mdev(pd
->device
), &context
->uar
,
577 ucmd
.rq_db_index
, ucmd
.rq_db_page
);
579 mthca_unmap_user_db(to_mdev(pd
->device
),
587 qp
->mr
.ibmr
.lkey
= ucmd
.lkey
;
588 qp
->sq
.db_index
= ucmd
.sq_db_index
;
589 qp
->rq
.db_index
= ucmd
.rq_db_index
;
592 err
= mthca_alloc_qp(to_mdev(pd
->device
), to_mpd(pd
),
593 to_mcq(init_attr
->send_cq
),
594 to_mcq(init_attr
->recv_cq
),
595 init_attr
->qp_type
, init_attr
->sq_sig_type
,
596 &init_attr
->cap
, qp
);
598 if (err
&& pd
->uobject
) {
599 context
= to_mucontext(pd
->uobject
->context
);
601 mthca_unmap_user_db(to_mdev(pd
->device
),
605 mthca_unmap_user_db(to_mdev(pd
->device
),
611 qp
->ibqp
.qp_num
= qp
->qpn
;
617 /* Don't allow userspace to create special QPs */
619 return ERR_PTR(-EINVAL
);
621 qp
= kmalloc(sizeof (struct mthca_sqp
), GFP_KERNEL
);
623 return ERR_PTR(-ENOMEM
);
625 qp
->ibqp
.qp_num
= init_attr
->qp_type
== IB_QPT_SMI
? 0 : 1;
627 err
= mthca_alloc_sqp(to_mdev(pd
->device
), to_mpd(pd
),
628 to_mcq(init_attr
->send_cq
),
629 to_mcq(init_attr
->recv_cq
),
630 init_attr
->sq_sig_type
, &init_attr
->cap
,
631 qp
->ibqp
.qp_num
, init_attr
->port_num
,
636 /* Don't support raw QPs */
637 return ERR_PTR(-ENOSYS
);
645 init_attr
->cap
.max_send_wr
= qp
->sq
.max
;
646 init_attr
->cap
.max_recv_wr
= qp
->rq
.max
;
647 init_attr
->cap
.max_send_sge
= qp
->sq
.max_gs
;
648 init_attr
->cap
.max_recv_sge
= qp
->rq
.max_gs
;
649 init_attr
->cap
.max_inline_data
= qp
->max_inline_data
;
654 static int mthca_destroy_qp(struct ib_qp
*qp
)
657 mthca_unmap_user_db(to_mdev(qp
->device
),
658 &to_mucontext(qp
->uobject
->context
)->uar
,
659 to_mucontext(qp
->uobject
->context
)->db_tab
,
660 to_mqp(qp
)->sq
.db_index
);
661 mthca_unmap_user_db(to_mdev(qp
->device
),
662 &to_mucontext(qp
->uobject
->context
)->uar
,
663 to_mucontext(qp
->uobject
->context
)->db_tab
,
664 to_mqp(qp
)->rq
.db_index
);
666 mthca_free_qp(to_mdev(qp
->device
), to_mqp(qp
));
671 static struct ib_cq
*mthca_create_cq(struct ib_device
*ibdev
, int entries
,
673 struct ib_ucontext
*context
,
674 struct ib_udata
*udata
)
676 struct mthca_create_cq ucmd
;
681 if (entries
< 1 || entries
> to_mdev(ibdev
)->limits
.max_cqes
)
682 return ERR_PTR(-EINVAL
);
685 if (ib_copy_from_udata(&ucmd
, udata
, sizeof ucmd
))
686 return ERR_PTR(-EFAULT
);
688 err
= mthca_map_user_db(to_mdev(ibdev
), &to_mucontext(context
)->uar
,
689 to_mucontext(context
)->db_tab
,
690 ucmd
.set_db_index
, ucmd
.set_db_page
);
694 err
= mthca_map_user_db(to_mdev(ibdev
), &to_mucontext(context
)->uar
,
695 to_mucontext(context
)->db_tab
,
696 ucmd
.arm_db_index
, ucmd
.arm_db_page
);
701 cq
= kmalloc(sizeof *cq
, GFP_KERNEL
);
708 cq
->buf
.mr
.ibmr
.lkey
= ucmd
.lkey
;
709 cq
->set_ci_db_index
= ucmd
.set_db_index
;
710 cq
->arm_db_index
= ucmd
.arm_db_index
;
713 for (nent
= 1; nent
<= entries
; nent
<<= 1)
716 err
= mthca_init_cq(to_mdev(ibdev
), nent
,
717 context
? to_mucontext(context
) : NULL
,
718 context
? ucmd
.pdn
: to_mdev(ibdev
)->driver_pd
.pd_num
,
723 if (context
&& ib_copy_to_udata(udata
, &cq
->cqn
, sizeof (__u32
))) {
724 mthca_free_cq(to_mdev(ibdev
), cq
);
728 cq
->resize_buf
= NULL
;
737 mthca_unmap_user_db(to_mdev(ibdev
), &to_mucontext(context
)->uar
,
738 to_mucontext(context
)->db_tab
, ucmd
.arm_db_index
);
742 mthca_unmap_user_db(to_mdev(ibdev
), &to_mucontext(context
)->uar
,
743 to_mucontext(context
)->db_tab
, ucmd
.set_db_index
);
748 static int mthca_alloc_resize_buf(struct mthca_dev
*dev
, struct mthca_cq
*cq
,
753 spin_lock_irq(&cq
->lock
);
754 if (cq
->resize_buf
) {
759 cq
->resize_buf
= kmalloc(sizeof *cq
->resize_buf
, GFP_ATOMIC
);
760 if (!cq
->resize_buf
) {
765 cq
->resize_buf
->state
= CQ_RESIZE_ALLOC
;
770 spin_unlock_irq(&cq
->lock
);
775 ret
= mthca_alloc_cq_buf(dev
, &cq
->resize_buf
->buf
, entries
);
777 spin_lock_irq(&cq
->lock
);
778 kfree(cq
->resize_buf
);
779 cq
->resize_buf
= NULL
;
780 spin_unlock_irq(&cq
->lock
);
784 cq
->resize_buf
->cqe
= entries
- 1;
786 spin_lock_irq(&cq
->lock
);
787 cq
->resize_buf
->state
= CQ_RESIZE_READY
;
788 spin_unlock_irq(&cq
->lock
);
793 static int mthca_resize_cq(struct ib_cq
*ibcq
, int entries
, struct ib_udata
*udata
)
795 struct mthca_dev
*dev
= to_mdev(ibcq
->device
);
796 struct mthca_cq
*cq
= to_mcq(ibcq
);
797 struct mthca_resize_cq ucmd
;
802 if (entries
< 1 || entries
> dev
->limits
.max_cqes
)
805 mutex_lock(&cq
->mutex
);
807 entries
= roundup_pow_of_two(entries
+ 1);
808 if (entries
== ibcq
->cqe
+ 1) {
814 ret
= mthca_alloc_resize_buf(dev
, cq
, entries
);
817 lkey
= cq
->resize_buf
->buf
.mr
.ibmr
.lkey
;
819 if (ib_copy_from_udata(&ucmd
, udata
, sizeof ucmd
)) {
826 ret
= mthca_RESIZE_CQ(dev
, cq
->cqn
, lkey
, ilog2(entries
), &status
);
831 if (cq
->resize_buf
) {
832 mthca_free_cq_buf(dev
, &cq
->resize_buf
->buf
,
833 cq
->resize_buf
->cqe
);
834 kfree(cq
->resize_buf
);
835 spin_lock_irq(&cq
->lock
);
836 cq
->resize_buf
= NULL
;
837 spin_unlock_irq(&cq
->lock
);
843 struct mthca_cq_buf tbuf
;
846 spin_lock_irq(&cq
->lock
);
847 if (cq
->resize_buf
->state
== CQ_RESIZE_READY
) {
848 mthca_cq_resize_copy_cqes(cq
);
851 cq
->buf
= cq
->resize_buf
->buf
;
852 cq
->ibcq
.cqe
= cq
->resize_buf
->cqe
;
854 tbuf
= cq
->resize_buf
->buf
;
855 tcqe
= cq
->resize_buf
->cqe
;
858 kfree(cq
->resize_buf
);
859 cq
->resize_buf
= NULL
;
860 spin_unlock_irq(&cq
->lock
);
862 mthca_free_cq_buf(dev
, &tbuf
, tcqe
);
864 ibcq
->cqe
= entries
- 1;
867 mutex_unlock(&cq
->mutex
);
872 static int mthca_destroy_cq(struct ib_cq
*cq
)
875 mthca_unmap_user_db(to_mdev(cq
->device
),
876 &to_mucontext(cq
->uobject
->context
)->uar
,
877 to_mucontext(cq
->uobject
->context
)->db_tab
,
878 to_mcq(cq
)->arm_db_index
);
879 mthca_unmap_user_db(to_mdev(cq
->device
),
880 &to_mucontext(cq
->uobject
->context
)->uar
,
881 to_mucontext(cq
->uobject
->context
)->db_tab
,
882 to_mcq(cq
)->set_ci_db_index
);
884 mthca_free_cq(to_mdev(cq
->device
), to_mcq(cq
));
890 static inline u32
convert_access(int acc
)
892 return (acc
& IB_ACCESS_REMOTE_ATOMIC
? MTHCA_MPT_FLAG_ATOMIC
: 0) |
893 (acc
& IB_ACCESS_REMOTE_WRITE
? MTHCA_MPT_FLAG_REMOTE_WRITE
: 0) |
894 (acc
& IB_ACCESS_REMOTE_READ
? MTHCA_MPT_FLAG_REMOTE_READ
: 0) |
895 (acc
& IB_ACCESS_LOCAL_WRITE
? MTHCA_MPT_FLAG_LOCAL_WRITE
: 0) |
896 MTHCA_MPT_FLAG_LOCAL_READ
;
899 static struct ib_mr
*mthca_get_dma_mr(struct ib_pd
*pd
, int acc
)
904 mr
= kmalloc(sizeof *mr
, GFP_KERNEL
);
906 return ERR_PTR(-ENOMEM
);
908 err
= mthca_mr_alloc_notrans(to_mdev(pd
->device
),
910 convert_access(acc
), mr
);
922 static struct ib_mr
*mthca_reg_phys_mr(struct ib_pd
*pd
,
923 struct ib_phys_buf
*buffer_list
,
937 mask
= buffer_list
[0].addr
^ *iova_start
;
939 for (i
= 0; i
< num_phys_buf
; ++i
) {
941 mask
|= buffer_list
[i
].addr
;
942 if (i
!= num_phys_buf
- 1)
943 mask
|= buffer_list
[i
].addr
+ buffer_list
[i
].size
;
945 total_size
+= buffer_list
[i
].size
;
948 if (mask
& ~PAGE_MASK
)
949 return ERR_PTR(-EINVAL
);
951 shift
= __ffs(mask
| 1 << 31);
953 buffer_list
[0].size
+= buffer_list
[0].addr
& ((1ULL << shift
) - 1);
954 buffer_list
[0].addr
&= ~0ull << shift
;
956 mr
= kmalloc(sizeof *mr
, GFP_KERNEL
);
958 return ERR_PTR(-ENOMEM
);
961 for (i
= 0; i
< num_phys_buf
; ++i
)
962 npages
+= (buffer_list
[i
].size
+ (1ULL << shift
) - 1) >> shift
;
967 page_list
= kmalloc(npages
* sizeof *page_list
, GFP_KERNEL
);
970 return ERR_PTR(-ENOMEM
);
974 for (i
= 0; i
< num_phys_buf
; ++i
)
976 j
< (buffer_list
[i
].size
+ (1ULL << shift
) - 1) >> shift
;
978 page_list
[n
++] = buffer_list
[i
].addr
+ ((u64
) j
<< shift
);
980 mthca_dbg(to_mdev(pd
->device
), "Registering memory at %llx (iova %llx) "
981 "in PD %x; shift %d, npages %d.\n",
982 (unsigned long long) buffer_list
[0].addr
,
983 (unsigned long long) *iova_start
,
987 err
= mthca_mr_alloc_phys(to_mdev(pd
->device
),
989 page_list
, shift
, npages
,
990 *iova_start
, total_size
,
991 convert_access(acc
), mr
);
1005 static struct ib_mr
*mthca_reg_user_mr(struct ib_pd
*pd
, u64 start
, u64 length
,
1006 u64 virt
, int acc
, struct ib_udata
*udata
)
1008 struct mthca_dev
*dev
= to_mdev(pd
->device
);
1009 struct ib_umem_chunk
*chunk
;
1010 struct mthca_mr
*mr
;
1011 struct mthca_reg_mr ucmd
;
1018 if (udata
->inlen
- sizeof (struct ib_uverbs_cmd_hdr
) < sizeof ucmd
) {
1019 if (!to_mucontext(pd
->uobject
->context
)->reg_mr_warned
) {
1020 mthca_warn(dev
, "Process '%s' did not pass in MR attrs.\n",
1022 mthca_warn(dev
, " Update libmthca to fix this.\n");
1024 ++to_mucontext(pd
->uobject
->context
)->reg_mr_warned
;
1026 } else if (ib_copy_from_udata(&ucmd
, udata
, sizeof ucmd
))
1027 return ERR_PTR(-EFAULT
);
1029 mr
= kmalloc(sizeof *mr
, GFP_KERNEL
);
1031 return ERR_PTR(-ENOMEM
);
1033 mr
->umem
= ib_umem_get(pd
->uobject
->context
, start
, length
, acc
,
1034 ucmd
.mr_attrs
& MTHCA_MR_DMASYNC
);
1036 if (IS_ERR(mr
->umem
)) {
1037 err
= PTR_ERR(mr
->umem
);
1041 shift
= ffs(mr
->umem
->page_size
) - 1;
1044 list_for_each_entry(chunk
, &mr
->umem
->chunk_list
, list
)
1047 mr
->mtt
= mthca_alloc_mtt(dev
, n
);
1048 if (IS_ERR(mr
->mtt
)) {
1049 err
= PTR_ERR(mr
->mtt
);
1053 pages
= (u64
*) __get_free_page(GFP_KERNEL
);
1061 write_mtt_size
= min(mthca_write_mtt_size(dev
), (int) (PAGE_SIZE
/ sizeof *pages
));
1063 list_for_each_entry(chunk
, &mr
->umem
->chunk_list
, list
)
1064 for (j
= 0; j
< chunk
->nmap
; ++j
) {
1065 len
= sg_dma_len(&chunk
->page_list
[j
]) >> shift
;
1066 for (k
= 0; k
< len
; ++k
) {
1067 pages
[i
++] = sg_dma_address(&chunk
->page_list
[j
]) +
1068 mr
->umem
->page_size
* k
;
1070 * Be friendly to write_mtt and pass it chunks
1071 * of appropriate size.
1073 if (i
== write_mtt_size
) {
1074 err
= mthca_write_mtt(dev
, mr
->mtt
, n
, pages
, i
);
1084 err
= mthca_write_mtt(dev
, mr
->mtt
, n
, pages
, i
);
1086 free_page((unsigned long) pages
);
1090 err
= mthca_mr_alloc(dev
, to_mpd(pd
)->pd_num
, shift
, virt
, length
,
1091 convert_access(acc
), mr
);
1099 mthca_free_mtt(dev
, mr
->mtt
);
1102 ib_umem_release(mr
->umem
);
1106 return ERR_PTR(err
);
1109 static int mthca_dereg_mr(struct ib_mr
*mr
)
1111 struct mthca_mr
*mmr
= to_mmr(mr
);
1113 mthca_free_mr(to_mdev(mr
->device
), mmr
);
1115 ib_umem_release(mmr
->umem
);
1121 static struct ib_fmr
*mthca_alloc_fmr(struct ib_pd
*pd
, int mr_access_flags
,
1122 struct ib_fmr_attr
*fmr_attr
)
1124 struct mthca_fmr
*fmr
;
1127 fmr
= kmalloc(sizeof *fmr
, GFP_KERNEL
);
1129 return ERR_PTR(-ENOMEM
);
1131 memcpy(&fmr
->attr
, fmr_attr
, sizeof *fmr_attr
);
1132 err
= mthca_fmr_alloc(to_mdev(pd
->device
), to_mpd(pd
)->pd_num
,
1133 convert_access(mr_access_flags
), fmr
);
1137 return ERR_PTR(err
);
1143 static int mthca_dealloc_fmr(struct ib_fmr
*fmr
)
1145 struct mthca_fmr
*mfmr
= to_mfmr(fmr
);
1148 err
= mthca_free_fmr(to_mdev(fmr
->device
), mfmr
);
1156 static int mthca_unmap_fmr(struct list_head
*fmr_list
)
1161 struct mthca_dev
*mdev
= NULL
;
1163 list_for_each_entry(fmr
, fmr_list
, list
) {
1164 if (mdev
&& to_mdev(fmr
->device
) != mdev
)
1166 mdev
= to_mdev(fmr
->device
);
1172 if (mthca_is_memfree(mdev
)) {
1173 list_for_each_entry(fmr
, fmr_list
, list
)
1174 mthca_arbel_fmr_unmap(mdev
, to_mfmr(fmr
));
1178 list_for_each_entry(fmr
, fmr_list
, list
)
1179 mthca_tavor_fmr_unmap(mdev
, to_mfmr(fmr
));
1181 err
= mthca_SYNC_TPT(mdev
, &status
);
1189 static ssize_t
show_rev(struct device
*device
, struct device_attribute
*attr
,
1192 struct mthca_dev
*dev
=
1193 container_of(device
, struct mthca_dev
, ib_dev
.dev
);
1194 return sprintf(buf
, "%x\n", dev
->rev_id
);
1197 static ssize_t
show_fw_ver(struct device
*device
, struct device_attribute
*attr
,
1200 struct mthca_dev
*dev
=
1201 container_of(device
, struct mthca_dev
, ib_dev
.dev
);
1202 return sprintf(buf
, "%d.%d.%d\n", (int) (dev
->fw_ver
>> 32),
1203 (int) (dev
->fw_ver
>> 16) & 0xffff,
1204 (int) dev
->fw_ver
& 0xffff);
1207 static ssize_t
show_hca(struct device
*device
, struct device_attribute
*attr
,
1210 struct mthca_dev
*dev
=
1211 container_of(device
, struct mthca_dev
, ib_dev
.dev
);
1212 switch (dev
->pdev
->device
) {
1213 case PCI_DEVICE_ID_MELLANOX_TAVOR
:
1214 return sprintf(buf
, "MT23108\n");
1215 case PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT
:
1216 return sprintf(buf
, "MT25208 (MT23108 compat mode)\n");
1217 case PCI_DEVICE_ID_MELLANOX_ARBEL
:
1218 return sprintf(buf
, "MT25208\n");
1219 case PCI_DEVICE_ID_MELLANOX_SINAI
:
1220 case PCI_DEVICE_ID_MELLANOX_SINAI_OLD
:
1221 return sprintf(buf
, "MT25204\n");
1223 return sprintf(buf
, "unknown\n");
1227 static ssize_t
show_board(struct device
*device
, struct device_attribute
*attr
,
1230 struct mthca_dev
*dev
=
1231 container_of(device
, struct mthca_dev
, ib_dev
.dev
);
1232 return sprintf(buf
, "%.*s\n", MTHCA_BOARD_ID_LEN
, dev
->board_id
);
1235 static DEVICE_ATTR(hw_rev
, S_IRUGO
, show_rev
, NULL
);
1236 static DEVICE_ATTR(fw_ver
, S_IRUGO
, show_fw_ver
, NULL
);
1237 static DEVICE_ATTR(hca_type
, S_IRUGO
, show_hca
, NULL
);
1238 static DEVICE_ATTR(board_id
, S_IRUGO
, show_board
, NULL
);
1240 static struct device_attribute
*mthca_dev_attributes
[] = {
1247 static int mthca_init_node_data(struct mthca_dev
*dev
)
1249 struct ib_smp
*in_mad
= NULL
;
1250 struct ib_smp
*out_mad
= NULL
;
1254 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
1255 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
1256 if (!in_mad
|| !out_mad
)
1259 init_query_mad(in_mad
);
1260 in_mad
->attr_id
= IB_SMP_ATTR_NODE_DESC
;
1262 err
= mthca_MAD_IFC(dev
, 1, 1,
1263 1, NULL
, NULL
, in_mad
, out_mad
,
1272 memcpy(dev
->ib_dev
.node_desc
, out_mad
->data
, 64);
1274 in_mad
->attr_id
= IB_SMP_ATTR_NODE_INFO
;
1276 err
= mthca_MAD_IFC(dev
, 1, 1,
1277 1, NULL
, NULL
, in_mad
, out_mad
,
1286 if (mthca_is_memfree(dev
))
1287 dev
->rev_id
= be32_to_cpup((__be32
*) (out_mad
->data
+ 32));
1288 memcpy(&dev
->ib_dev
.node_guid
, out_mad
->data
+ 12, 8);
1296 int mthca_register_device(struct mthca_dev
*dev
)
1301 ret
= mthca_init_node_data(dev
);
1305 strlcpy(dev
->ib_dev
.name
, "mthca%d", IB_DEVICE_NAME_MAX
);
1306 dev
->ib_dev
.owner
= THIS_MODULE
;
1308 dev
->ib_dev
.uverbs_abi_ver
= MTHCA_UVERBS_ABI_VERSION
;
1309 dev
->ib_dev
.uverbs_cmd_mask
=
1310 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT
) |
1311 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE
) |
1312 (1ull << IB_USER_VERBS_CMD_QUERY_PORT
) |
1313 (1ull << IB_USER_VERBS_CMD_ALLOC_PD
) |
1314 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD
) |
1315 (1ull << IB_USER_VERBS_CMD_REG_MR
) |
1316 (1ull << IB_USER_VERBS_CMD_DEREG_MR
) |
1317 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL
) |
1318 (1ull << IB_USER_VERBS_CMD_CREATE_CQ
) |
1319 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ
) |
1320 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ
) |
1321 (1ull << IB_USER_VERBS_CMD_CREATE_QP
) |
1322 (1ull << IB_USER_VERBS_CMD_QUERY_QP
) |
1323 (1ull << IB_USER_VERBS_CMD_MODIFY_QP
) |
1324 (1ull << IB_USER_VERBS_CMD_DESTROY_QP
) |
1325 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST
) |
1326 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST
);
1327 dev
->ib_dev
.node_type
= RDMA_NODE_IB_CA
;
1328 dev
->ib_dev
.phys_port_cnt
= dev
->limits
.num_ports
;
1329 dev
->ib_dev
.num_comp_vectors
= 1;
1330 dev
->ib_dev
.dma_device
= &dev
->pdev
->dev
;
1331 dev
->ib_dev
.query_device
= mthca_query_device
;
1332 dev
->ib_dev
.query_port
= mthca_query_port
;
1333 dev
->ib_dev
.modify_device
= mthca_modify_device
;
1334 dev
->ib_dev
.modify_port
= mthca_modify_port
;
1335 dev
->ib_dev
.query_pkey
= mthca_query_pkey
;
1336 dev
->ib_dev
.query_gid
= mthca_query_gid
;
1337 dev
->ib_dev
.alloc_ucontext
= mthca_alloc_ucontext
;
1338 dev
->ib_dev
.dealloc_ucontext
= mthca_dealloc_ucontext
;
1339 dev
->ib_dev
.mmap
= mthca_mmap_uar
;
1340 dev
->ib_dev
.alloc_pd
= mthca_alloc_pd
;
1341 dev
->ib_dev
.dealloc_pd
= mthca_dealloc_pd
;
1342 dev
->ib_dev
.create_ah
= mthca_ah_create
;
1343 dev
->ib_dev
.query_ah
= mthca_ah_query
;
1344 dev
->ib_dev
.destroy_ah
= mthca_ah_destroy
;
1346 if (dev
->mthca_flags
& MTHCA_FLAG_SRQ
) {
1347 dev
->ib_dev
.create_srq
= mthca_create_srq
;
1348 dev
->ib_dev
.modify_srq
= mthca_modify_srq
;
1349 dev
->ib_dev
.query_srq
= mthca_query_srq
;
1350 dev
->ib_dev
.destroy_srq
= mthca_destroy_srq
;
1351 dev
->ib_dev
.uverbs_cmd_mask
|=
1352 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ
) |
1353 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ
) |
1354 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ
) |
1355 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ
);
1357 if (mthca_is_memfree(dev
))
1358 dev
->ib_dev
.post_srq_recv
= mthca_arbel_post_srq_recv
;
1360 dev
->ib_dev
.post_srq_recv
= mthca_tavor_post_srq_recv
;
1363 dev
->ib_dev
.create_qp
= mthca_create_qp
;
1364 dev
->ib_dev
.modify_qp
= mthca_modify_qp
;
1365 dev
->ib_dev
.query_qp
= mthca_query_qp
;
1366 dev
->ib_dev
.destroy_qp
= mthca_destroy_qp
;
1367 dev
->ib_dev
.create_cq
= mthca_create_cq
;
1368 dev
->ib_dev
.resize_cq
= mthca_resize_cq
;
1369 dev
->ib_dev
.destroy_cq
= mthca_destroy_cq
;
1370 dev
->ib_dev
.poll_cq
= mthca_poll_cq
;
1371 dev
->ib_dev
.get_dma_mr
= mthca_get_dma_mr
;
1372 dev
->ib_dev
.reg_phys_mr
= mthca_reg_phys_mr
;
1373 dev
->ib_dev
.reg_user_mr
= mthca_reg_user_mr
;
1374 dev
->ib_dev
.dereg_mr
= mthca_dereg_mr
;
1376 if (dev
->mthca_flags
& MTHCA_FLAG_FMR
) {
1377 dev
->ib_dev
.alloc_fmr
= mthca_alloc_fmr
;
1378 dev
->ib_dev
.unmap_fmr
= mthca_unmap_fmr
;
1379 dev
->ib_dev
.dealloc_fmr
= mthca_dealloc_fmr
;
1380 if (mthca_is_memfree(dev
))
1381 dev
->ib_dev
.map_phys_fmr
= mthca_arbel_map_phys_fmr
;
1383 dev
->ib_dev
.map_phys_fmr
= mthca_tavor_map_phys_fmr
;
1386 dev
->ib_dev
.attach_mcast
= mthca_multicast_attach
;
1387 dev
->ib_dev
.detach_mcast
= mthca_multicast_detach
;
1388 dev
->ib_dev
.process_mad
= mthca_process_mad
;
1390 if (mthca_is_memfree(dev
)) {
1391 dev
->ib_dev
.req_notify_cq
= mthca_arbel_arm_cq
;
1392 dev
->ib_dev
.post_send
= mthca_arbel_post_send
;
1393 dev
->ib_dev
.post_recv
= mthca_arbel_post_receive
;
1395 dev
->ib_dev
.req_notify_cq
= mthca_tavor_arm_cq
;
1396 dev
->ib_dev
.post_send
= mthca_tavor_post_send
;
1397 dev
->ib_dev
.post_recv
= mthca_tavor_post_receive
;
1400 mutex_init(&dev
->cap_mask_mutex
);
1402 ret
= ib_register_device(&dev
->ib_dev
);
1406 for (i
= 0; i
< ARRAY_SIZE(mthca_dev_attributes
); ++i
) {
1407 ret
= device_create_file(&dev
->ib_dev
.dev
,
1408 mthca_dev_attributes
[i
]);
1410 ib_unregister_device(&dev
->ib_dev
);
1415 mthca_start_catas_poll(dev
);
1420 void mthca_unregister_device(struct mthca_dev
*dev
)
1422 mthca_stop_catas_poll(dev
);
1423 ib_unregister_device(&dev
->ib_dev
);