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_user_verbs.h>
43 #include "mthca_dev.h"
44 #include "mthca_cmd.h"
45 #include "mthca_user.h"
46 #include "mthca_memfree.h"
48 static void init_query_mad(struct ib_smp
*mad
)
50 mad
->base_version
= 1;
51 mad
->mgmt_class
= IB_MGMT_CLASS_SUBN_LID_ROUTED
;
52 mad
->class_version
= 1;
53 mad
->method
= IB_MGMT_METHOD_GET
;
56 static int mthca_query_device(struct ib_device
*ibdev
,
57 struct ib_device_attr
*props
)
59 struct ib_smp
*in_mad
= NULL
;
60 struct ib_smp
*out_mad
= NULL
;
62 struct mthca_dev
* mdev
= to_mdev(ibdev
);
66 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
67 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
68 if (!in_mad
|| !out_mad
)
71 memset(props
, 0, sizeof *props
);
73 props
->fw_ver
= mdev
->fw_ver
;
75 init_query_mad(in_mad
);
76 in_mad
->attr_id
= IB_SMP_ATTR_NODE_INFO
;
78 err
= mthca_MAD_IFC(mdev
, 1, 1,
79 1, NULL
, NULL
, in_mad
, out_mad
,
88 props
->device_cap_flags
= mdev
->device_cap_flags
;
89 props
->vendor_id
= be32_to_cpup((__be32
*) (out_mad
->data
+ 36)) &
91 props
->vendor_part_id
= be16_to_cpup((__be16
*) (out_mad
->data
+ 30));
92 props
->hw_ver
= be32_to_cpup((__be32
*) (out_mad
->data
+ 32));
93 memcpy(&props
->sys_image_guid
, out_mad
->data
+ 4, 8);
95 props
->max_mr_size
= ~0ull;
96 props
->page_size_cap
= mdev
->limits
.page_size_cap
;
97 props
->max_qp
= mdev
->limits
.num_qps
- mdev
->limits
.reserved_qps
;
98 props
->max_qp_wr
= mdev
->limits
.max_wqes
;
99 props
->max_sge
= mdev
->limits
.max_sg
;
100 props
->max_cq
= mdev
->limits
.num_cqs
- mdev
->limits
.reserved_cqs
;
101 props
->max_cqe
= mdev
->limits
.max_cqes
;
102 props
->max_mr
= mdev
->limits
.num_mpts
- mdev
->limits
.reserved_mrws
;
103 props
->max_pd
= mdev
->limits
.num_pds
- mdev
->limits
.reserved_pds
;
104 props
->max_qp_rd_atom
= 1 << mdev
->qp_table
.rdb_shift
;
105 props
->max_qp_init_rd_atom
= mdev
->limits
.max_qp_init_rdma
;
106 props
->max_res_rd_atom
= props
->max_qp_rd_atom
* props
->max_qp
;
107 props
->max_srq
= mdev
->limits
.num_srqs
- mdev
->limits
.reserved_srqs
;
108 props
->max_srq_wr
= mdev
->limits
.max_srq_wqes
;
109 props
->max_srq_sge
= mdev
->limits
.max_srq_sge
;
110 props
->local_ca_ack_delay
= mdev
->limits
.local_ca_ack_delay
;
111 props
->atomic_cap
= mdev
->limits
.flags
& DEV_LIM_FLAG_ATOMIC
?
112 IB_ATOMIC_HCA
: IB_ATOMIC_NONE
;
113 props
->max_pkeys
= mdev
->limits
.pkey_table_len
;
114 props
->max_mcast_grp
= mdev
->limits
.num_mgms
+ mdev
->limits
.num_amgms
;
115 props
->max_mcast_qp_attach
= MTHCA_QP_PER_MGM
;
116 props
->max_total_mcast_qp_attach
= props
->max_mcast_qp_attach
*
117 props
->max_mcast_grp
;
126 static int mthca_query_port(struct ib_device
*ibdev
,
127 u8 port
, struct ib_port_attr
*props
)
129 struct ib_smp
*in_mad
= NULL
;
130 struct ib_smp
*out_mad
= NULL
;
134 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
135 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
136 if (!in_mad
|| !out_mad
)
139 memset(props
, 0, sizeof *props
);
141 init_query_mad(in_mad
);
142 in_mad
->attr_id
= IB_SMP_ATTR_PORT_INFO
;
143 in_mad
->attr_mod
= cpu_to_be32(port
);
145 err
= mthca_MAD_IFC(to_mdev(ibdev
), 1, 1,
146 port
, NULL
, NULL
, in_mad
, out_mad
,
155 props
->lid
= be16_to_cpup((__be16
*) (out_mad
->data
+ 16));
156 props
->lmc
= out_mad
->data
[34] & 0x7;
157 props
->sm_lid
= be16_to_cpup((__be16
*) (out_mad
->data
+ 18));
158 props
->sm_sl
= out_mad
->data
[36] & 0xf;
159 props
->state
= out_mad
->data
[32] & 0xf;
160 props
->phys_state
= out_mad
->data
[33] >> 4;
161 props
->port_cap_flags
= be32_to_cpup((__be32
*) (out_mad
->data
+ 20));
162 props
->gid_tbl_len
= to_mdev(ibdev
)->limits
.gid_table_len
;
163 props
->max_msg_sz
= 0x80000000;
164 props
->pkey_tbl_len
= to_mdev(ibdev
)->limits
.pkey_table_len
;
165 props
->bad_pkey_cntr
= be16_to_cpup((__be16
*) (out_mad
->data
+ 46));
166 props
->qkey_viol_cntr
= be16_to_cpup((__be16
*) (out_mad
->data
+ 48));
167 props
->active_width
= out_mad
->data
[31] & 0xf;
168 props
->active_speed
= out_mad
->data
[35] >> 4;
169 props
->max_mtu
= out_mad
->data
[41] & 0xf;
170 props
->active_mtu
= out_mad
->data
[36] >> 4;
171 props
->subnet_timeout
= out_mad
->data
[51] & 0x1f;
179 static int mthca_modify_device(struct ib_device
*ibdev
,
181 struct ib_device_modify
*props
)
183 if (mask
& ~IB_DEVICE_MODIFY_NODE_DESC
)
186 if (mask
& IB_DEVICE_MODIFY_NODE_DESC
) {
187 if (mutex_lock_interruptible(&to_mdev(ibdev
)->cap_mask_mutex
))
189 memcpy(ibdev
->node_desc
, props
->node_desc
, 64);
190 mutex_unlock(&to_mdev(ibdev
)->cap_mask_mutex
);
196 static int mthca_modify_port(struct ib_device
*ibdev
,
197 u8 port
, int port_modify_mask
,
198 struct ib_port_modify
*props
)
200 struct mthca_set_ib_param set_ib
;
201 struct ib_port_attr attr
;
205 if (mutex_lock_interruptible(&to_mdev(ibdev
)->cap_mask_mutex
))
208 err
= mthca_query_port(ibdev
, port
, &attr
);
212 set_ib
.set_si_guid
= 0;
213 set_ib
.reset_qkey_viol
= !!(port_modify_mask
& IB_PORT_RESET_QKEY_CNTR
);
215 set_ib
.cap_mask
= (attr
.port_cap_flags
| props
->set_port_cap_mask
) &
216 ~props
->clr_port_cap_mask
;
218 err
= mthca_SET_IB(to_mdev(ibdev
), &set_ib
, port
, &status
);
227 mutex_unlock(&to_mdev(ibdev
)->cap_mask_mutex
);
231 static int mthca_query_pkey(struct ib_device
*ibdev
,
232 u8 port
, u16 index
, u16
*pkey
)
234 struct ib_smp
*in_mad
= NULL
;
235 struct ib_smp
*out_mad
= NULL
;
239 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
240 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
241 if (!in_mad
|| !out_mad
)
244 init_query_mad(in_mad
);
245 in_mad
->attr_id
= IB_SMP_ATTR_PKEY_TABLE
;
246 in_mad
->attr_mod
= cpu_to_be32(index
/ 32);
248 err
= mthca_MAD_IFC(to_mdev(ibdev
), 1, 1,
249 port
, NULL
, NULL
, in_mad
, out_mad
,
258 *pkey
= be16_to_cpu(((__be16
*) out_mad
->data
)[index
% 32]);
266 static int mthca_query_gid(struct ib_device
*ibdev
, u8 port
,
267 int index
, union ib_gid
*gid
)
269 struct ib_smp
*in_mad
= NULL
;
270 struct ib_smp
*out_mad
= NULL
;
274 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
275 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
276 if (!in_mad
|| !out_mad
)
279 init_query_mad(in_mad
);
280 in_mad
->attr_id
= IB_SMP_ATTR_PORT_INFO
;
281 in_mad
->attr_mod
= cpu_to_be32(port
);
283 err
= mthca_MAD_IFC(to_mdev(ibdev
), 1, 1,
284 port
, NULL
, NULL
, in_mad
, out_mad
,
293 memcpy(gid
->raw
, out_mad
->data
+ 8, 8);
295 init_query_mad(in_mad
);
296 in_mad
->attr_id
= IB_SMP_ATTR_GUID_INFO
;
297 in_mad
->attr_mod
= cpu_to_be32(index
/ 8);
299 err
= mthca_MAD_IFC(to_mdev(ibdev
), 1, 1,
300 port
, NULL
, NULL
, in_mad
, out_mad
,
309 memcpy(gid
->raw
+ 8, out_mad
->data
+ (index
% 8) * 8, 8);
317 static struct ib_ucontext
*mthca_alloc_ucontext(struct ib_device
*ibdev
,
318 struct ib_udata
*udata
)
320 struct mthca_alloc_ucontext_resp uresp
;
321 struct mthca_ucontext
*context
;
324 memset(&uresp
, 0, sizeof uresp
);
326 uresp
.qp_tab_size
= to_mdev(ibdev
)->limits
.num_qps
;
327 if (mthca_is_memfree(to_mdev(ibdev
)))
328 uresp
.uarc_size
= to_mdev(ibdev
)->uar_table
.uarc_size
;
332 context
= kmalloc(sizeof *context
, GFP_KERNEL
);
334 return ERR_PTR(-ENOMEM
);
336 err
= mthca_uar_alloc(to_mdev(ibdev
), &context
->uar
);
342 context
->db_tab
= mthca_init_user_db_tab(to_mdev(ibdev
));
343 if (IS_ERR(context
->db_tab
)) {
344 err
= PTR_ERR(context
->db_tab
);
345 mthca_uar_free(to_mdev(ibdev
), &context
->uar
);
350 if (ib_copy_to_udata(udata
, &uresp
, sizeof uresp
)) {
351 mthca_cleanup_user_db_tab(to_mdev(ibdev
), &context
->uar
, context
->db_tab
);
352 mthca_uar_free(to_mdev(ibdev
), &context
->uar
);
354 return ERR_PTR(-EFAULT
);
357 return &context
->ibucontext
;
360 static int mthca_dealloc_ucontext(struct ib_ucontext
*context
)
362 mthca_cleanup_user_db_tab(to_mdev(context
->device
), &to_mucontext(context
)->uar
,
363 to_mucontext(context
)->db_tab
);
364 mthca_uar_free(to_mdev(context
->device
), &to_mucontext(context
)->uar
);
365 kfree(to_mucontext(context
));
370 static int mthca_mmap_uar(struct ib_ucontext
*context
,
371 struct vm_area_struct
*vma
)
373 if (vma
->vm_end
- vma
->vm_start
!= PAGE_SIZE
)
376 vma
->vm_page_prot
= pgprot_noncached(vma
->vm_page_prot
);
378 if (io_remap_pfn_range(vma
, vma
->vm_start
,
379 to_mucontext(context
)->uar
.pfn
,
380 PAGE_SIZE
, vma
->vm_page_prot
))
386 static struct ib_pd
*mthca_alloc_pd(struct ib_device
*ibdev
,
387 struct ib_ucontext
*context
,
388 struct ib_udata
*udata
)
393 pd
= kmalloc(sizeof *pd
, GFP_KERNEL
);
395 return ERR_PTR(-ENOMEM
);
397 err
= mthca_pd_alloc(to_mdev(ibdev
), !context
, pd
);
404 if (ib_copy_to_udata(udata
, &pd
->pd_num
, sizeof (__u32
))) {
405 mthca_pd_free(to_mdev(ibdev
), pd
);
407 return ERR_PTR(-EFAULT
);
414 static int mthca_dealloc_pd(struct ib_pd
*pd
)
416 mthca_pd_free(to_mdev(pd
->device
), to_mpd(pd
));
422 static struct ib_ah
*mthca_ah_create(struct ib_pd
*pd
,
423 struct ib_ah_attr
*ah_attr
)
428 ah
= kmalloc(sizeof *ah
, GFP_ATOMIC
);
430 return ERR_PTR(-ENOMEM
);
432 err
= mthca_create_ah(to_mdev(pd
->device
), to_mpd(pd
), ah_attr
, ah
);
441 static int mthca_ah_destroy(struct ib_ah
*ah
)
443 mthca_destroy_ah(to_mdev(ah
->device
), to_mah(ah
));
449 static struct ib_srq
*mthca_create_srq(struct ib_pd
*pd
,
450 struct ib_srq_init_attr
*init_attr
,
451 struct ib_udata
*udata
)
453 struct mthca_create_srq ucmd
;
454 struct mthca_ucontext
*context
= NULL
;
455 struct mthca_srq
*srq
;
458 srq
= kmalloc(sizeof *srq
, GFP_KERNEL
);
460 return ERR_PTR(-ENOMEM
);
463 context
= to_mucontext(pd
->uobject
->context
);
465 if (ib_copy_from_udata(&ucmd
, udata
, sizeof ucmd
)) {
470 err
= mthca_map_user_db(to_mdev(pd
->device
), &context
->uar
,
471 context
->db_tab
, ucmd
.db_index
,
477 srq
->mr
.ibmr
.lkey
= ucmd
.lkey
;
478 srq
->db_index
= ucmd
.db_index
;
481 err
= mthca_alloc_srq(to_mdev(pd
->device
), to_mpd(pd
),
482 &init_attr
->attr
, srq
);
484 if (err
&& pd
->uobject
)
485 mthca_unmap_user_db(to_mdev(pd
->device
), &context
->uar
,
486 context
->db_tab
, ucmd
.db_index
);
491 if (context
&& ib_copy_to_udata(udata
, &srq
->srqn
, sizeof (__u32
))) {
492 mthca_free_srq(to_mdev(pd
->device
), srq
);
505 static int mthca_destroy_srq(struct ib_srq
*srq
)
507 struct mthca_ucontext
*context
;
510 context
= to_mucontext(srq
->uobject
->context
);
512 mthca_unmap_user_db(to_mdev(srq
->device
), &context
->uar
,
513 context
->db_tab
, to_msrq(srq
)->db_index
);
516 mthca_free_srq(to_mdev(srq
->device
), to_msrq(srq
));
522 static struct ib_qp
*mthca_create_qp(struct ib_pd
*pd
,
523 struct ib_qp_init_attr
*init_attr
,
524 struct ib_udata
*udata
)
526 struct mthca_create_qp ucmd
;
530 switch (init_attr
->qp_type
) {
535 struct mthca_ucontext
*context
;
537 qp
= kmalloc(sizeof *qp
, GFP_KERNEL
);
539 return ERR_PTR(-ENOMEM
);
542 context
= to_mucontext(pd
->uobject
->context
);
544 if (ib_copy_from_udata(&ucmd
, udata
, sizeof ucmd
)) {
546 return ERR_PTR(-EFAULT
);
549 err
= mthca_map_user_db(to_mdev(pd
->device
), &context
->uar
,
551 ucmd
.sq_db_index
, ucmd
.sq_db_page
);
557 err
= mthca_map_user_db(to_mdev(pd
->device
), &context
->uar
,
559 ucmd
.rq_db_index
, ucmd
.rq_db_page
);
561 mthca_unmap_user_db(to_mdev(pd
->device
),
569 qp
->mr
.ibmr
.lkey
= ucmd
.lkey
;
570 qp
->sq
.db_index
= ucmd
.sq_db_index
;
571 qp
->rq
.db_index
= ucmd
.rq_db_index
;
574 err
= mthca_alloc_qp(to_mdev(pd
->device
), to_mpd(pd
),
575 to_mcq(init_attr
->send_cq
),
576 to_mcq(init_attr
->recv_cq
),
577 init_attr
->qp_type
, init_attr
->sq_sig_type
,
578 &init_attr
->cap
, qp
);
580 if (err
&& pd
->uobject
) {
581 context
= to_mucontext(pd
->uobject
->context
);
583 mthca_unmap_user_db(to_mdev(pd
->device
),
587 mthca_unmap_user_db(to_mdev(pd
->device
),
593 qp
->ibqp
.qp_num
= qp
->qpn
;
599 /* Don't allow userspace to create special QPs */
601 return ERR_PTR(-EINVAL
);
603 qp
= kmalloc(sizeof (struct mthca_sqp
), GFP_KERNEL
);
605 return ERR_PTR(-ENOMEM
);
607 qp
->ibqp
.qp_num
= init_attr
->qp_type
== IB_QPT_SMI
? 0 : 1;
609 err
= mthca_alloc_sqp(to_mdev(pd
->device
), to_mpd(pd
),
610 to_mcq(init_attr
->send_cq
),
611 to_mcq(init_attr
->recv_cq
),
612 init_attr
->sq_sig_type
, &init_attr
->cap
,
613 qp
->ibqp
.qp_num
, init_attr
->port_num
,
618 /* Don't support raw QPs */
619 return ERR_PTR(-ENOSYS
);
627 init_attr
->cap
.max_send_wr
= qp
->sq
.max
;
628 init_attr
->cap
.max_recv_wr
= qp
->rq
.max
;
629 init_attr
->cap
.max_send_sge
= qp
->sq
.max_gs
;
630 init_attr
->cap
.max_recv_sge
= qp
->rq
.max_gs
;
631 init_attr
->cap
.max_inline_data
= qp
->max_inline_data
;
636 static int mthca_destroy_qp(struct ib_qp
*qp
)
639 mthca_unmap_user_db(to_mdev(qp
->device
),
640 &to_mucontext(qp
->uobject
->context
)->uar
,
641 to_mucontext(qp
->uobject
->context
)->db_tab
,
642 to_mqp(qp
)->sq
.db_index
);
643 mthca_unmap_user_db(to_mdev(qp
->device
),
644 &to_mucontext(qp
->uobject
->context
)->uar
,
645 to_mucontext(qp
->uobject
->context
)->db_tab
,
646 to_mqp(qp
)->rq
.db_index
);
648 mthca_free_qp(to_mdev(qp
->device
), to_mqp(qp
));
653 static struct ib_cq
*mthca_create_cq(struct ib_device
*ibdev
, int entries
,
654 struct ib_ucontext
*context
,
655 struct ib_udata
*udata
)
657 struct mthca_create_cq ucmd
;
662 if (entries
< 1 || entries
> to_mdev(ibdev
)->limits
.max_cqes
)
663 return ERR_PTR(-EINVAL
);
666 if (ib_copy_from_udata(&ucmd
, udata
, sizeof ucmd
))
667 return ERR_PTR(-EFAULT
);
669 err
= mthca_map_user_db(to_mdev(ibdev
), &to_mucontext(context
)->uar
,
670 to_mucontext(context
)->db_tab
,
671 ucmd
.set_db_index
, ucmd
.set_db_page
);
675 err
= mthca_map_user_db(to_mdev(ibdev
), &to_mucontext(context
)->uar
,
676 to_mucontext(context
)->db_tab
,
677 ucmd
.arm_db_index
, ucmd
.arm_db_page
);
682 cq
= kmalloc(sizeof *cq
, GFP_KERNEL
);
689 cq
->buf
.mr
.ibmr
.lkey
= ucmd
.lkey
;
690 cq
->set_ci_db_index
= ucmd
.set_db_index
;
691 cq
->arm_db_index
= ucmd
.arm_db_index
;
694 for (nent
= 1; nent
<= entries
; nent
<<= 1)
697 err
= mthca_init_cq(to_mdev(ibdev
), nent
,
698 context
? to_mucontext(context
) : NULL
,
699 context
? ucmd
.pdn
: to_mdev(ibdev
)->driver_pd
.pd_num
,
704 if (context
&& ib_copy_to_udata(udata
, &cq
->cqn
, sizeof (__u32
))) {
705 mthca_free_cq(to_mdev(ibdev
), cq
);
709 cq
->resize_buf
= NULL
;
718 mthca_unmap_user_db(to_mdev(ibdev
), &to_mucontext(context
)->uar
,
719 to_mucontext(context
)->db_tab
, ucmd
.arm_db_index
);
723 mthca_unmap_user_db(to_mdev(ibdev
), &to_mucontext(context
)->uar
,
724 to_mucontext(context
)->db_tab
, ucmd
.set_db_index
);
729 static int mthca_alloc_resize_buf(struct mthca_dev
*dev
, struct mthca_cq
*cq
,
734 spin_lock_irq(&cq
->lock
);
735 if (cq
->resize_buf
) {
740 cq
->resize_buf
= kmalloc(sizeof *cq
->resize_buf
, GFP_ATOMIC
);
741 if (!cq
->resize_buf
) {
746 cq
->resize_buf
->state
= CQ_RESIZE_ALLOC
;
751 spin_unlock_irq(&cq
->lock
);
756 ret
= mthca_alloc_cq_buf(dev
, &cq
->resize_buf
->buf
, entries
);
758 spin_lock_irq(&cq
->lock
);
759 kfree(cq
->resize_buf
);
760 cq
->resize_buf
= NULL
;
761 spin_unlock_irq(&cq
->lock
);
765 cq
->resize_buf
->cqe
= entries
- 1;
767 spin_lock_irq(&cq
->lock
);
768 cq
->resize_buf
->state
= CQ_RESIZE_READY
;
769 spin_unlock_irq(&cq
->lock
);
774 static int mthca_resize_cq(struct ib_cq
*ibcq
, int entries
, struct ib_udata
*udata
)
776 struct mthca_dev
*dev
= to_mdev(ibcq
->device
);
777 struct mthca_cq
*cq
= to_mcq(ibcq
);
778 struct mthca_resize_cq ucmd
;
783 if (entries
< 1 || entries
> dev
->limits
.max_cqes
)
786 entries
= roundup_pow_of_two(entries
+ 1);
787 if (entries
== ibcq
->cqe
+ 1)
791 ret
= mthca_alloc_resize_buf(dev
, cq
, entries
);
794 lkey
= cq
->resize_buf
->buf
.mr
.ibmr
.lkey
;
796 if (ib_copy_from_udata(&ucmd
, udata
, sizeof ucmd
))
801 ret
= mthca_RESIZE_CQ(dev
, cq
->cqn
, lkey
, long_log2(entries
), &status
);
806 if (cq
->resize_buf
) {
807 mthca_free_cq_buf(dev
, &cq
->resize_buf
->buf
,
808 cq
->resize_buf
->cqe
);
809 kfree(cq
->resize_buf
);
810 spin_lock_irq(&cq
->lock
);
811 cq
->resize_buf
= NULL
;
812 spin_unlock_irq(&cq
->lock
);
818 struct mthca_cq_buf tbuf
;
821 spin_lock_irq(&cq
->lock
);
822 if (cq
->resize_buf
->state
== CQ_RESIZE_READY
) {
823 mthca_cq_resize_copy_cqes(cq
);
826 cq
->buf
= cq
->resize_buf
->buf
;
827 cq
->ibcq
.cqe
= cq
->resize_buf
->cqe
;
829 tbuf
= cq
->resize_buf
->buf
;
830 tcqe
= cq
->resize_buf
->cqe
;
833 kfree(cq
->resize_buf
);
834 cq
->resize_buf
= NULL
;
835 spin_unlock_irq(&cq
->lock
);
837 mthca_free_cq_buf(dev
, &tbuf
, tcqe
);
839 ibcq
->cqe
= entries
- 1;
844 static int mthca_destroy_cq(struct ib_cq
*cq
)
847 mthca_unmap_user_db(to_mdev(cq
->device
),
848 &to_mucontext(cq
->uobject
->context
)->uar
,
849 to_mucontext(cq
->uobject
->context
)->db_tab
,
850 to_mcq(cq
)->arm_db_index
);
851 mthca_unmap_user_db(to_mdev(cq
->device
),
852 &to_mucontext(cq
->uobject
->context
)->uar
,
853 to_mucontext(cq
->uobject
->context
)->db_tab
,
854 to_mcq(cq
)->set_ci_db_index
);
856 mthca_free_cq(to_mdev(cq
->device
), to_mcq(cq
));
862 static inline u32
convert_access(int acc
)
864 return (acc
& IB_ACCESS_REMOTE_ATOMIC
? MTHCA_MPT_FLAG_ATOMIC
: 0) |
865 (acc
& IB_ACCESS_REMOTE_WRITE
? MTHCA_MPT_FLAG_REMOTE_WRITE
: 0) |
866 (acc
& IB_ACCESS_REMOTE_READ
? MTHCA_MPT_FLAG_REMOTE_READ
: 0) |
867 (acc
& IB_ACCESS_LOCAL_WRITE
? MTHCA_MPT_FLAG_LOCAL_WRITE
: 0) |
868 MTHCA_MPT_FLAG_LOCAL_READ
;
871 static struct ib_mr
*mthca_get_dma_mr(struct ib_pd
*pd
, int acc
)
876 mr
= kmalloc(sizeof *mr
, GFP_KERNEL
);
878 return ERR_PTR(-ENOMEM
);
880 err
= mthca_mr_alloc_notrans(to_mdev(pd
->device
),
882 convert_access(acc
), mr
);
892 static struct ib_mr
*mthca_reg_phys_mr(struct ib_pd
*pd
,
893 struct ib_phys_buf
*buffer_list
,
907 /* First check that we have enough alignment */
908 if ((*iova_start
& ~PAGE_MASK
) != (buffer_list
[0].addr
& ~PAGE_MASK
))
909 return ERR_PTR(-EINVAL
);
913 for (i
= 0; i
< num_phys_buf
; ++i
) {
915 mask
|= buffer_list
[i
].addr
;
916 if (i
!= num_phys_buf
- 1)
917 mask
|= buffer_list
[i
].addr
+ buffer_list
[i
].size
;
919 total_size
+= buffer_list
[i
].size
;
922 if (mask
& ~PAGE_MASK
)
923 return ERR_PTR(-EINVAL
);
925 /* Find largest page shift we can use to cover buffers */
926 for (shift
= PAGE_SHIFT
; shift
< 31; ++shift
)
927 if (num_phys_buf
> 1) {
928 if ((1ULL << shift
) & mask
)
932 buffer_list
[0].size
+
933 (buffer_list
[0].addr
& ((1ULL << shift
) - 1)))
937 buffer_list
[0].size
+= buffer_list
[0].addr
& ((1ULL << shift
) - 1);
938 buffer_list
[0].addr
&= ~0ull << shift
;
940 mr
= kmalloc(sizeof *mr
, GFP_KERNEL
);
942 return ERR_PTR(-ENOMEM
);
945 for (i
= 0; i
< num_phys_buf
; ++i
)
946 npages
+= (buffer_list
[i
].size
+ (1ULL << shift
) - 1) >> shift
;
951 page_list
= kmalloc(npages
* sizeof *page_list
, GFP_KERNEL
);
954 return ERR_PTR(-ENOMEM
);
958 for (i
= 0; i
< num_phys_buf
; ++i
)
960 j
< (buffer_list
[i
].size
+ (1ULL << shift
) - 1) >> shift
;
962 page_list
[n
++] = buffer_list
[i
].addr
+ ((u64
) j
<< shift
);
964 mthca_dbg(to_mdev(pd
->device
), "Registering memory at %llx (iova %llx) "
965 "in PD %x; shift %d, npages %d.\n",
966 (unsigned long long) buffer_list
[0].addr
,
967 (unsigned long long) *iova_start
,
971 err
= mthca_mr_alloc_phys(to_mdev(pd
->device
),
973 page_list
, shift
, npages
,
974 *iova_start
, total_size
,
975 convert_access(acc
), mr
);
987 static struct ib_mr
*mthca_reg_user_mr(struct ib_pd
*pd
, struct ib_umem
*region
,
988 int acc
, struct ib_udata
*udata
)
990 struct mthca_dev
*dev
= to_mdev(pd
->device
);
991 struct ib_umem_chunk
*chunk
;
998 shift
= ffs(region
->page_size
) - 1;
1000 mr
= kmalloc(sizeof *mr
, GFP_KERNEL
);
1002 return ERR_PTR(-ENOMEM
);
1005 list_for_each_entry(chunk
, ®ion
->chunk_list
, list
)
1008 mr
->mtt
= mthca_alloc_mtt(dev
, n
);
1009 if (IS_ERR(mr
->mtt
)) {
1010 err
= PTR_ERR(mr
->mtt
);
1014 pages
= (u64
*) __get_free_page(GFP_KERNEL
);
1022 list_for_each_entry(chunk
, ®ion
->chunk_list
, list
)
1023 for (j
= 0; j
< chunk
->nmap
; ++j
) {
1024 len
= sg_dma_len(&chunk
->page_list
[j
]) >> shift
;
1025 for (k
= 0; k
< len
; ++k
) {
1026 pages
[i
++] = sg_dma_address(&chunk
->page_list
[j
]) +
1027 region
->page_size
* k
;
1029 * Be friendly to WRITE_MTT command
1030 * and leave two empty slots for the
1031 * index and reserved fields of the
1034 if (i
== PAGE_SIZE
/ sizeof (u64
) - 2) {
1035 err
= mthca_write_mtt(dev
, mr
->mtt
,
1046 err
= mthca_write_mtt(dev
, mr
->mtt
, n
, pages
, i
);
1048 free_page((unsigned long) pages
);
1052 err
= mthca_mr_alloc(dev
, to_mpd(pd
)->pd_num
, shift
, region
->virt_base
,
1053 region
->length
, convert_access(acc
), mr
);
1061 mthca_free_mtt(dev
, mr
->mtt
);
1065 return ERR_PTR(err
);
1068 static int mthca_dereg_mr(struct ib_mr
*mr
)
1070 struct mthca_mr
*mmr
= to_mmr(mr
);
1071 mthca_free_mr(to_mdev(mr
->device
), mmr
);
1076 static struct ib_fmr
*mthca_alloc_fmr(struct ib_pd
*pd
, int mr_access_flags
,
1077 struct ib_fmr_attr
*fmr_attr
)
1079 struct mthca_fmr
*fmr
;
1082 fmr
= kmalloc(sizeof *fmr
, GFP_KERNEL
);
1084 return ERR_PTR(-ENOMEM
);
1086 memcpy(&fmr
->attr
, fmr_attr
, sizeof *fmr_attr
);
1087 err
= mthca_fmr_alloc(to_mdev(pd
->device
), to_mpd(pd
)->pd_num
,
1088 convert_access(mr_access_flags
), fmr
);
1092 return ERR_PTR(err
);
1098 static int mthca_dealloc_fmr(struct ib_fmr
*fmr
)
1100 struct mthca_fmr
*mfmr
= to_mfmr(fmr
);
1103 err
= mthca_free_fmr(to_mdev(fmr
->device
), mfmr
);
1111 static int mthca_unmap_fmr(struct list_head
*fmr_list
)
1116 struct mthca_dev
*mdev
= NULL
;
1118 list_for_each_entry(fmr
, fmr_list
, list
) {
1119 if (mdev
&& to_mdev(fmr
->device
) != mdev
)
1121 mdev
= to_mdev(fmr
->device
);
1127 if (mthca_is_memfree(mdev
)) {
1128 list_for_each_entry(fmr
, fmr_list
, list
)
1129 mthca_arbel_fmr_unmap(mdev
, to_mfmr(fmr
));
1133 list_for_each_entry(fmr
, fmr_list
, list
)
1134 mthca_tavor_fmr_unmap(mdev
, to_mfmr(fmr
));
1136 err
= mthca_SYNC_TPT(mdev
, &status
);
1144 static ssize_t
show_rev(struct class_device
*cdev
, char *buf
)
1146 struct mthca_dev
*dev
= container_of(cdev
, struct mthca_dev
, ib_dev
.class_dev
);
1147 return sprintf(buf
, "%x\n", dev
->rev_id
);
1150 static ssize_t
show_fw_ver(struct class_device
*cdev
, char *buf
)
1152 struct mthca_dev
*dev
= container_of(cdev
, struct mthca_dev
, ib_dev
.class_dev
);
1153 return sprintf(buf
, "%d.%d.%d\n", (int) (dev
->fw_ver
>> 32),
1154 (int) (dev
->fw_ver
>> 16) & 0xffff,
1155 (int) dev
->fw_ver
& 0xffff);
1158 static ssize_t
show_hca(struct class_device
*cdev
, char *buf
)
1160 struct mthca_dev
*dev
= container_of(cdev
, struct mthca_dev
, ib_dev
.class_dev
);
1161 switch (dev
->pdev
->device
) {
1162 case PCI_DEVICE_ID_MELLANOX_TAVOR
:
1163 return sprintf(buf
, "MT23108\n");
1164 case PCI_DEVICE_ID_MELLANOX_ARBEL_COMPAT
:
1165 return sprintf(buf
, "MT25208 (MT23108 compat mode)\n");
1166 case PCI_DEVICE_ID_MELLANOX_ARBEL
:
1167 return sprintf(buf
, "MT25208\n");
1168 case PCI_DEVICE_ID_MELLANOX_SINAI
:
1169 case PCI_DEVICE_ID_MELLANOX_SINAI_OLD
:
1170 return sprintf(buf
, "MT25204\n");
1172 return sprintf(buf
, "unknown\n");
1176 static ssize_t
show_board(struct class_device
*cdev
, char *buf
)
1178 struct mthca_dev
*dev
= container_of(cdev
, struct mthca_dev
, ib_dev
.class_dev
);
1179 return sprintf(buf
, "%.*s\n", MTHCA_BOARD_ID_LEN
, dev
->board_id
);
1182 static CLASS_DEVICE_ATTR(hw_rev
, S_IRUGO
, show_rev
, NULL
);
1183 static CLASS_DEVICE_ATTR(fw_ver
, S_IRUGO
, show_fw_ver
, NULL
);
1184 static CLASS_DEVICE_ATTR(hca_type
, S_IRUGO
, show_hca
, NULL
);
1185 static CLASS_DEVICE_ATTR(board_id
, S_IRUGO
, show_board
, NULL
);
1187 static struct class_device_attribute
*mthca_class_attributes
[] = {
1188 &class_device_attr_hw_rev
,
1189 &class_device_attr_fw_ver
,
1190 &class_device_attr_hca_type
,
1191 &class_device_attr_board_id
1194 static int mthca_init_node_data(struct mthca_dev
*dev
)
1196 struct ib_smp
*in_mad
= NULL
;
1197 struct ib_smp
*out_mad
= NULL
;
1201 in_mad
= kzalloc(sizeof *in_mad
, GFP_KERNEL
);
1202 out_mad
= kmalloc(sizeof *out_mad
, GFP_KERNEL
);
1203 if (!in_mad
|| !out_mad
)
1206 init_query_mad(in_mad
);
1207 in_mad
->attr_id
= IB_SMP_ATTR_NODE_DESC
;
1209 err
= mthca_MAD_IFC(dev
, 1, 1,
1210 1, NULL
, NULL
, in_mad
, out_mad
,
1219 memcpy(dev
->ib_dev
.node_desc
, out_mad
->data
, 64);
1221 in_mad
->attr_id
= IB_SMP_ATTR_NODE_INFO
;
1223 err
= mthca_MAD_IFC(dev
, 1, 1,
1224 1, NULL
, NULL
, in_mad
, out_mad
,
1233 memcpy(&dev
->ib_dev
.node_guid
, out_mad
->data
+ 12, 8);
1241 int mthca_register_device(struct mthca_dev
*dev
)
1246 ret
= mthca_init_node_data(dev
);
1250 strlcpy(dev
->ib_dev
.name
, "mthca%d", IB_DEVICE_NAME_MAX
);
1251 dev
->ib_dev
.owner
= THIS_MODULE
;
1253 dev
->ib_dev
.uverbs_abi_ver
= MTHCA_UVERBS_ABI_VERSION
;
1254 dev
->ib_dev
.uverbs_cmd_mask
=
1255 (1ull << IB_USER_VERBS_CMD_GET_CONTEXT
) |
1256 (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE
) |
1257 (1ull << IB_USER_VERBS_CMD_QUERY_PORT
) |
1258 (1ull << IB_USER_VERBS_CMD_ALLOC_PD
) |
1259 (1ull << IB_USER_VERBS_CMD_DEALLOC_PD
) |
1260 (1ull << IB_USER_VERBS_CMD_REG_MR
) |
1261 (1ull << IB_USER_VERBS_CMD_DEREG_MR
) |
1262 (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL
) |
1263 (1ull << IB_USER_VERBS_CMD_CREATE_CQ
) |
1264 (1ull << IB_USER_VERBS_CMD_RESIZE_CQ
) |
1265 (1ull << IB_USER_VERBS_CMD_DESTROY_CQ
) |
1266 (1ull << IB_USER_VERBS_CMD_CREATE_QP
) |
1267 (1ull << IB_USER_VERBS_CMD_QUERY_QP
) |
1268 (1ull << IB_USER_VERBS_CMD_MODIFY_QP
) |
1269 (1ull << IB_USER_VERBS_CMD_DESTROY_QP
) |
1270 (1ull << IB_USER_VERBS_CMD_ATTACH_MCAST
) |
1271 (1ull << IB_USER_VERBS_CMD_DETACH_MCAST
) |
1272 (1ull << IB_USER_VERBS_CMD_CREATE_SRQ
) |
1273 (1ull << IB_USER_VERBS_CMD_MODIFY_SRQ
) |
1274 (1ull << IB_USER_VERBS_CMD_QUERY_SRQ
) |
1275 (1ull << IB_USER_VERBS_CMD_DESTROY_SRQ
);
1276 dev
->ib_dev
.node_type
= IB_NODE_CA
;
1277 dev
->ib_dev
.phys_port_cnt
= dev
->limits
.num_ports
;
1278 dev
->ib_dev
.dma_device
= &dev
->pdev
->dev
;
1279 dev
->ib_dev
.class_dev
.dev
= &dev
->pdev
->dev
;
1280 dev
->ib_dev
.query_device
= mthca_query_device
;
1281 dev
->ib_dev
.query_port
= mthca_query_port
;
1282 dev
->ib_dev
.modify_device
= mthca_modify_device
;
1283 dev
->ib_dev
.modify_port
= mthca_modify_port
;
1284 dev
->ib_dev
.query_pkey
= mthca_query_pkey
;
1285 dev
->ib_dev
.query_gid
= mthca_query_gid
;
1286 dev
->ib_dev
.alloc_ucontext
= mthca_alloc_ucontext
;
1287 dev
->ib_dev
.dealloc_ucontext
= mthca_dealloc_ucontext
;
1288 dev
->ib_dev
.mmap
= mthca_mmap_uar
;
1289 dev
->ib_dev
.alloc_pd
= mthca_alloc_pd
;
1290 dev
->ib_dev
.dealloc_pd
= mthca_dealloc_pd
;
1291 dev
->ib_dev
.create_ah
= mthca_ah_create
;
1292 dev
->ib_dev
.query_ah
= mthca_ah_query
;
1293 dev
->ib_dev
.destroy_ah
= mthca_ah_destroy
;
1295 if (dev
->mthca_flags
& MTHCA_FLAG_SRQ
) {
1296 dev
->ib_dev
.create_srq
= mthca_create_srq
;
1297 dev
->ib_dev
.modify_srq
= mthca_modify_srq
;
1298 dev
->ib_dev
.query_srq
= mthca_query_srq
;
1299 dev
->ib_dev
.destroy_srq
= mthca_destroy_srq
;
1301 if (mthca_is_memfree(dev
))
1302 dev
->ib_dev
.post_srq_recv
= mthca_arbel_post_srq_recv
;
1304 dev
->ib_dev
.post_srq_recv
= mthca_tavor_post_srq_recv
;
1307 dev
->ib_dev
.create_qp
= mthca_create_qp
;
1308 dev
->ib_dev
.modify_qp
= mthca_modify_qp
;
1309 dev
->ib_dev
.query_qp
= mthca_query_qp
;
1310 dev
->ib_dev
.destroy_qp
= mthca_destroy_qp
;
1311 dev
->ib_dev
.create_cq
= mthca_create_cq
;
1312 dev
->ib_dev
.resize_cq
= mthca_resize_cq
;
1313 dev
->ib_dev
.destroy_cq
= mthca_destroy_cq
;
1314 dev
->ib_dev
.poll_cq
= mthca_poll_cq
;
1315 dev
->ib_dev
.get_dma_mr
= mthca_get_dma_mr
;
1316 dev
->ib_dev
.reg_phys_mr
= mthca_reg_phys_mr
;
1317 dev
->ib_dev
.reg_user_mr
= mthca_reg_user_mr
;
1318 dev
->ib_dev
.dereg_mr
= mthca_dereg_mr
;
1320 if (dev
->mthca_flags
& MTHCA_FLAG_FMR
) {
1321 dev
->ib_dev
.alloc_fmr
= mthca_alloc_fmr
;
1322 dev
->ib_dev
.unmap_fmr
= mthca_unmap_fmr
;
1323 dev
->ib_dev
.dealloc_fmr
= mthca_dealloc_fmr
;
1324 if (mthca_is_memfree(dev
))
1325 dev
->ib_dev
.map_phys_fmr
= mthca_arbel_map_phys_fmr
;
1327 dev
->ib_dev
.map_phys_fmr
= mthca_tavor_map_phys_fmr
;
1330 dev
->ib_dev
.attach_mcast
= mthca_multicast_attach
;
1331 dev
->ib_dev
.detach_mcast
= mthca_multicast_detach
;
1332 dev
->ib_dev
.process_mad
= mthca_process_mad
;
1334 if (mthca_is_memfree(dev
)) {
1335 dev
->ib_dev
.req_notify_cq
= mthca_arbel_arm_cq
;
1336 dev
->ib_dev
.post_send
= mthca_arbel_post_send
;
1337 dev
->ib_dev
.post_recv
= mthca_arbel_post_receive
;
1339 dev
->ib_dev
.req_notify_cq
= mthca_tavor_arm_cq
;
1340 dev
->ib_dev
.post_send
= mthca_tavor_post_send
;
1341 dev
->ib_dev
.post_recv
= mthca_tavor_post_receive
;
1344 mutex_init(&dev
->cap_mask_mutex
);
1346 ret
= ib_register_device(&dev
->ib_dev
);
1350 for (i
= 0; i
< ARRAY_SIZE(mthca_class_attributes
); ++i
) {
1351 ret
= class_device_create_file(&dev
->ib_dev
.class_dev
,
1352 mthca_class_attributes
[i
]);
1354 ib_unregister_device(&dev
->ib_dev
);
1359 mthca_start_catas_poll(dev
);
1364 void mthca_unregister_device(struct mthca_dev
*dev
)
1366 mthca_stop_catas_poll(dev
);
1367 ib_unregister_device(&dev
->ib_dev
);