xref: /illumos-gate/usr/src/uts/common/io/ib/ibtl/ibtl_mem.c (revision 89b2a9fbeabf42fa54594df0e5927bcc50a07cc9)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 
27 #include <sys/ib/ibtl/impl/ibtl.h>
28 
29 /*
30  * ibtl_mem.c
31  *    These routines implement all of the Memory Region verbs and the alloc/
32  *    query/free Memory Window verbs at the TI interface.
33  */
34 
35 static char ibtl_mem[] = "ibtl_mem";
36 
37 /*
38  * Function:
39  *	ibt_register_mr()
40  * Input:
41  *	hca_hdl   - HCA Handle.
42  *	pd        - Protection Domain Handle.
43  *	mem_attr  - Requested memory region attributes.
44  * Output:
45  *	mr_hdl_p  - The returned IBT memory region handle.
46  *	mem_desc  - Returned memory descriptor.
47  * Returns:
48  *      IBT_SUCCESS
49  *	IBT_CHAN_HDL_INVALID
50  *	IBT_MR_VA_INVALID
51  *	IBT_MR_LEN_INVALID
52  *	IBT_MR_ACCESS_REQ_INVALID
53  *	IBT_PD_HDL_INVALID
54  *	IBT_INSUFF_RESOURCE
55  * Description:
56  *    Prepares a virtually addressed memory region for use by a HCA. A
57  *    description of the registered memory suitable for use in Work Requests
58  *    (WRs) is returned in the ibt_mr_desc_t parameter.
59  */
60 ibt_status_t
61 ibt_register_mr(ibt_hca_hdl_t hca_hdl, ibt_pd_hdl_t pd, ibt_mr_attr_t *mem_attr,
62     ibt_mr_hdl_t *mr_hdl_p, ibt_mr_desc_t *mem_desc)
63 {
64 	ib_vaddr_t 	vaddr;
65 	ibt_status_t 	status;
66 
67 	IBTF_DPRINTF_L3(ibtl_mem, "ibt_register_mr(%p, %p, %p)",
68 	    hca_hdl, pd, mem_attr);
69 
70 	vaddr = mem_attr->mr_vaddr;
71 
72 	status = IBTL_HCA2CIHCAOPS_P(hca_hdl)->ibc_register_mr(
73 	    IBTL_HCA2CIHCA(hca_hdl), pd, mem_attr, IBTL_HCA2CLNT(hca_hdl),
74 	    mr_hdl_p, mem_desc);
75 	if (status == IBT_SUCCESS) {
76 		mem_desc->md_vaddr = vaddr;
77 		mutex_enter(&hca_hdl->ha_mutex);
78 		hca_hdl->ha_mr_cnt++;
79 		mutex_exit(&hca_hdl->ha_mutex);
80 	}
81 
82 	return (status);
83 }
84 
85 
86 /*
87  * Function:
88  *	ibt_register_buf()
89  * Input:
90  *	hca_hdl		HCA Handle.
91  *	pd		Protection Domain Handle.
92  *	mem_bpattr	Memory Registration attributes (IOVA and flags).
93  *	bp		A pointer to a buf(9S) struct.
94  * Output:
95  *	mr_hdl_p	The returned IBT memory region handle.
96  *	mem_desc	Returned memory descriptor.
97  * Returns:
98  *      IBT_SUCCESS
99  *	IBT_CHAN_HDL_INVALID
100  *	IBT_MR_VA_INVALID
101  *	IBT_MR_LEN_INVALID
102  *	IBT_MR_ACCESS_REQ_INVALID
103  *	IBT_PD_HDL_INVALID
104  *	IBT_INSUFF_RESOURCE
105  * Description:
106  *	Prepares a memory region described by a buf(9S) struct for use by a HCA.
107  *	A description of the registered memory suitable for use in
108  *	Work Requests (WRs) is returned in the ibt_mr_desc_t parameter.
109  */
110 ibt_status_t
111 ibt_register_buf(ibt_hca_hdl_t hca_hdl, ibt_pd_hdl_t pd,
112     ibt_smr_attr_t *mem_bpattr, struct buf *bp, ibt_mr_hdl_t *mr_hdl_p,
113     ibt_mr_desc_t *mem_desc)
114 {
115 	ibt_status_t 	status;
116 
117 	IBTF_DPRINTF_L3(ibtl_mem, "ibt_register_buf(%p, %p, %p, %p)",
118 	    hca_hdl, pd, mem_bpattr, bp);
119 
120 	status = IBTL_HCA2CIHCAOPS_P(hca_hdl)->ibc_register_buf(
121 	    IBTL_HCA2CIHCA(hca_hdl), pd, mem_bpattr, bp, IBTL_HCA2CLNT(hca_hdl),
122 	    mr_hdl_p, mem_desc);
123 	if (status == IBT_SUCCESS) {
124 		mutex_enter(&hca_hdl->ha_mutex);
125 		hca_hdl->ha_mr_cnt++;
126 		mutex_exit(&hca_hdl->ha_mutex);
127 	}
128 
129 	return (status);
130 }
131 
132 
133 /*
134  * Function:
135  *	ibt_query_mr()
136  * Input:
137  *	hca_hdl   - HCA Handle.
138  *	mr_hdl    - The IBT Memory Region handle.
139  * Output:
140  *      attr      - The pointer to Memory region attributes structure.
141  * Returns:
142  *      IBT_SUCCESS
143  *	IBT_CHAN_HDL_INVALID
144  *	IBT_MR_HDL_INVALID
145  * Description:
146  *    Retrieves information about a specified memory region.
147  */
148 ibt_status_t
149 ibt_query_mr(ibt_hca_hdl_t hca_hdl, ibt_mr_hdl_t mr_hdl,
150     ibt_mr_query_attr_t *attr)
151 {
152 	IBTF_DPRINTF_L3(ibtl_mem, "ibt_query_mr(%p, %p)", hca_hdl, mr_hdl);
153 
154 	return (IBTL_HCA2CIHCAOPS_P(hca_hdl)->ibc_query_mr(
155 	    IBTL_HCA2CIHCA(hca_hdl), mr_hdl, attr));
156 }
157 
158 
159 /*
160  * Function:
161  *	ibt_deregister_mr()
162  * Input:
163  *	hca_hdl   - HCA Handle.
164  *	mr_hdl    - The IBT Memory Region handle.
165  * Output:
166  *      none.
167  * Returns:
168  *      IBT_SUCCESS
169  *	IBT_CHAN_HDL_INVALID
170  *	IBT_MR_HDL_INVALID
171  *	IBT_MR_IN_USE
172  * Description:
173  *    De-register the registered memory region. Remove a memory region from a
174  *    HCA translation table, and free all resources associated with the
175  *    memory region.
176  */
177 ibt_status_t
178 ibt_deregister_mr(ibt_hca_hdl_t hca_hdl, ibt_mr_hdl_t mr_hdl)
179 {
180 	ibt_status_t 	status;
181 
182 	IBTF_DPRINTF_L3(ibtl_mem, "ibt_deregister_mr(%p, %p)", hca_hdl, mr_hdl);
183 
184 	status = IBTL_HCA2CIHCAOPS_P(hca_hdl)->ibc_deregister_mr(
185 	    IBTL_HCA2CIHCA(hca_hdl), mr_hdl);
186 	if (status == IBT_SUCCESS) {
187 		mutex_enter(&hca_hdl->ha_mutex);
188 		hca_hdl->ha_mr_cnt--;
189 		mutex_exit(&hca_hdl->ha_mutex);
190 	}
191 	return (status);
192 }
193 
194 
195 /*
196  * Function:
197  *	ibt_reregister_mr()
198  * Input:
199  *	hca_hdl   - HCA Handle.
200  *	mr_hdl    - The IBT Memory Region handle.
201  *	pd        - Optional Protection Domain Handle.
202  *	mem_attr  - Requested memory region attributes.
203  * Output:
204  *	mr_hdl_p  - The reregistered IBT memory region handle.
205  *	mem_desc  - Returned memory descriptor for the new memory region.
206  * Returns:
207  *      IBT_SUCCESS
208  *	IBT_CHAN_HDL_INVALID
209  *	IBT_MR_HDL_INVALID
210  *	IBT_MR_VA_INVALID
211  *	IBT_MR_LEN_INVALID
212  *	IBT_MR_ACCESS_REQ_INVALID
213  *	IBT_PD_HDL_INVALID
214  *	IBT_INSUFF_RESOURCE
215  *	IBT_MR_IN_USE
216  * Description:
217  *    Modify the attributes of an existing memory region.
218  */
219 ibt_status_t
220 ibt_reregister_mr(ibt_hca_hdl_t hca_hdl, ibt_mr_hdl_t mr_hdl, ibt_pd_hdl_t pd,
221     ibt_mr_attr_t *mem_attr, ibt_mr_hdl_t *mr_hdl_p, ibt_mr_desc_t *mem_desc)
222 {
223 	ibt_status_t 	status;
224 	ib_vaddr_t 	vaddr = mem_attr->mr_vaddr;
225 
226 	IBTF_DPRINTF_L3(ibtl_mem, "ibt_reregister_mr(%p, %p, %p, %p)",
227 	    hca_hdl, mr_hdl, pd, mem_attr);
228 
229 	status = IBTL_HCA2CIHCAOPS_P(hca_hdl)->ibc_reregister_mr(
230 	    IBTL_HCA2CIHCA(hca_hdl), mr_hdl, pd, mem_attr,
231 	    IBTL_HCA2CLNT(hca_hdl), mr_hdl_p, mem_desc);
232 
233 	if (status == IBT_SUCCESS)
234 		mem_desc->md_vaddr = vaddr;
235 	else if (!(status == IBT_MR_IN_USE || status == IBT_HCA_HDL_INVALID ||
236 	    status == IBT_MR_HDL_INVALID)) {
237 
238 		IBTF_DPRINTF_L2(ibtl_mem, "ibt_reregister_mr: "
239 		    "Re-registration Failed: %d", status);
240 
241 		/* we lost one memory region resource */
242 		mutex_enter(&hca_hdl->ha_mutex);
243 		hca_hdl->ha_mr_cnt--;
244 		mutex_exit(&hca_hdl->ha_mutex);
245 	}
246 
247 	return (status);
248 }
249 
250 
251 /*
252  * Function:
253  *	ibt_reregister_buf()
254  * Input:
255  *	hca_hdl		HCA Handle.
256  *	mr_hdl		The IBT Memory Region handle.
257  *	pd		Optional Protection Domain Handle.
258  *	mem_bpattr	Memory Registration attributes (IOVA and flags).
259  *	bp		A pointer to a buf(9S) struct.
260  * Output:
261  *	mr_hdl_p	The reregistered IBT memory region handle.
262  *	mem_desc	Returned memory descriptor for the new memory region.
263  * Returns:
264  *      IBT_SUCCESS
265  *	IBT_CHAN_HDL_INVALID
266  *	IBT_MR_HDL_INVALID
267  *	IBT_MR_VA_INVALID
268  *	IBT_MR_LEN_INVALID
269  *	IBT_MR_ACCESS_REQ_INVALID
270  *	IBT_PD_HDL_INVALID
271  *	IBT_INSUFF_RESOURCE
272  *	IBT_MR_IN_USE
273  * Description:
274  *	Modify the attributes of an existing memory region as described by a
275  *	buf(9S) struct for use by a HCA.  A description of the registered
276  *	memory suitable for use in Work Requests (WRs) is returned in the
277  *	ibt_mr_desc_t parameter.
278  */
279 ibt_status_t
280 ibt_reregister_buf(ibt_hca_hdl_t hca_hdl, ibt_mr_hdl_t mr_hdl,
281     ibt_pd_hdl_t pd, ibt_smr_attr_t *mem_bpattr, struct buf *bp,
282     ibt_mr_hdl_t *mr_hdl_p, ibt_mr_desc_t *mem_desc)
283 {
284 	ibt_status_t 		status;
285 
286 	IBTF_DPRINTF_L3(ibtl_mem, "ibt_reregister_buf(%p, %p, %p, %p, %p)",
287 	    hca_hdl, mr_hdl, pd, mem_bpattr, bp);
288 
289 	status = IBTL_HCA2CIHCAOPS_P(hca_hdl)->ibc_reregister_buf(
290 	    IBTL_HCA2CIHCA(hca_hdl), mr_hdl, pd, mem_bpattr, bp,
291 	    IBTL_HCA2CLNT(hca_hdl), mr_hdl_p, mem_desc);
292 
293 	if (!(status == IBT_SUCCESS || status == IBT_MR_IN_USE ||
294 	    status == IBT_HCA_HDL_INVALID || status == IBT_MR_HDL_INVALID)) {
295 
296 		IBTF_DPRINTF_L2(ibtl_mem, "ibt_reregister_buf: "
297 		    "Re-registration Mem Failed: %d", status);
298 
299 		/* we lost one memory region resource */
300 		mutex_enter(&hca_hdl->ha_mutex);
301 		hca_hdl->ha_mr_cnt--;
302 		mutex_exit(&hca_hdl->ha_mutex);
303 	}
304 	return (status);
305 }
306 
307 
308 /*
309  * Function:
310  *	ibt_register_shared_mr()
311  * Input:
312  *	hca_hdl   - HCA Handle.
313  *	mr_hdl    - The IBT Memory Region handle.
314  *	pd        - Protection Domain Handle.
315  *	mem_sattr - Requested memory region shared attributes.
316  * Output:
317  *	mr_hdl_p  - The reregistered IBT memory region handle.
318  *	mem_desc  - Returned memory descriptor for the new memory region.
319  * Returns:
320  *      IBT_SUCCESS
321  *	IBT_INSUFF_RESOURCE
322  *	IBT_CHAN_HDL_INVALID
323  *	IBT_MR_HDL_INVALID
324  *	IBT_PD_HDL_INVALID
325  *	IBT_MR_ACCESS_REQ_INVALID
326  * Description:
327  *    Given an existing memory region, a new memory region associated with
328  *    the same physical locations is created.
329  */
330 ibt_status_t
331 ibt_register_shared_mr(ibt_hca_hdl_t hca_hdl, ibt_mr_hdl_t mr_hdl,
332     ibt_pd_hdl_t pd, ibt_smr_attr_t *mem_sattr, ibt_mr_hdl_t *mr_hdl_p,
333     ibt_mr_desc_t *mem_desc)
334 {
335 	ibt_status_t		status;
336 
337 	IBTF_DPRINTF_L3(ibtl_mem, "ibt_register_shared_mr(%p, %p, %p, %p)",
338 	    hca_hdl, mr_hdl, pd, mem_sattr);
339 
340 	status = IBTL_HCA2CIHCAOPS_P(hca_hdl)->ibc_register_shared_mr(
341 	    IBTL_HCA2CIHCA(hca_hdl), mr_hdl, pd, mem_sattr,
342 	    IBTL_HCA2CLNT(hca_hdl), mr_hdl_p, mem_desc);
343 	if (status == IBT_SUCCESS) {
344 		mutex_enter(&hca_hdl->ha_mutex);
345 		hca_hdl->ha_mr_cnt++;
346 		mutex_exit(&hca_hdl->ha_mutex);
347 	}
348 	return (status);
349 }
350 
351 /*
352  * Function:
353  *	ibt_sync_mr()
354  * Input:
355  *	hca_hdl		- HCA Handle.
356  *	mr_segments	- A pointer to an array of ibt_mr_sync_t that describes
357  *			  the memory regions to sync.
358  *	num_segments	- The length of the mr_segments array.
359  * Output:
360  *	NONE
361  * Returns:
362  *      IBT_SUCCESS
363  *	IBT_HCA_HDL_INVALID
364  *	IBT_MR_HDL_INVALID
365  *	IBT_INVALID_PARAM
366  *	IBT_MR_VA_INVALID
367  *	IBT_MR_LEN_INVALID
368  * Description:
369  *	Make memory changes visible to incoming RDMA reads, or make the affects
370  *	of an incoming RDMA writes visible to the consumer.
371  */
372 ibt_status_t
373 ibt_sync_mr(ibt_hca_hdl_t hca_hdl, ibt_mr_sync_t *mr_segments,
374     size_t num_segments)
375 
376 {
377 	IBTF_DPRINTF_L3(ibtl_mem, "ibt_sync_mr(%p, %p, %d)", hca_hdl,
378 	    mr_segments, num_segments);
379 
380 	return (IBTL_HCA2CIHCAOPS_P(hca_hdl)->ibc_sync_mr(
381 	    IBTL_HCA2CIHCA(hca_hdl), mr_segments, num_segments));
382 }
383 
384 
385 /*
386  * Function:
387  *	ibt_alloc_mw()
388  * Input:
389  *	hca_hdl   - HCA Handle.
390  *	pd        - Protection Domain Handle.
391  *	flags     - Memory Window alloc flags.
392  * Output:
393  *	mw_hdl_p  - The returned IBT Memory Window handle.
394  *	rkey      - The IBT R_Key handle.
395  * Returns:
396  *      IBT_SUCCESS
397  *	IBT_INSUFF_RESOURCE
398  *	IBT_CHAN_HDL_INVALID
399  *	IBT_PD_HDL_INVALID
400  * Description:
401  *    Allocate a memory window from the HCA.
402  */
403 ibt_status_t
404 ibt_alloc_mw(ibt_hca_hdl_t hca_hdl, ibt_pd_hdl_t pd, ibt_mw_flags_t flags,
405     ibt_mw_hdl_t *mw_hdl_p, ibt_rkey_t *rkey)
406 {
407 	ibt_status_t		status;
408 
409 	IBTF_DPRINTF_L3(ibtl_mem, "ibt_alloc_mw(%p, %p, 0x%x)",
410 	    hca_hdl, pd, flags);
411 
412 	status = IBTL_HCA2CIHCAOPS_P(hca_hdl)->ibc_alloc_mw(
413 	    IBTL_HCA2CIHCA(hca_hdl), pd, flags, mw_hdl_p, rkey);
414 
415 	/*
416 	 * XXX - We should be able to allocate state and have a IBTF Memory
417 	 * Window Handle. Memory Windows are meant to be rebound on the fly
418 	 * (using a post) to make them fast. It is expected that alloc memory
419 	 * window will be done in a relatively static manner. But, we don't have
420 	 * a good reason to have local MW state at this point, so we won't.
421 	 */
422 	if (status == IBT_SUCCESS) {
423 		mutex_enter(&hca_hdl->ha_mutex);
424 		hca_hdl->ha_mw_cnt++;
425 		mutex_exit(&hca_hdl->ha_mutex);
426 	}
427 	return (status);
428 }
429 
430 
431 /*
432  * Function:
433  *	ibt_query_mw()
434  * Input:
435  *	hca_hdl   - HCA Handle.
436  *	mw_hdl    - The IBT Memory Window handle.
437  * Output:
438  *	pd        - Protection Domain Handle.
439  *	rkey      - The IBT R_Key handle.
440  * Returns:
441  *      IBT_SUCCESS
442  *	IBT_CHAN_HDL_INVALID
443  *	IBT_MW_HDL_INVALID
444  * Description:
445  *    Retrieves information about a specified memory region.
446  */
447 ibt_status_t
448 ibt_query_mw(ibt_hca_hdl_t hca_hdl, ibt_mw_hdl_t mw_hdl,
449     ibt_mw_query_attr_t *mw_attr_p)
450 {
451 	IBTF_DPRINTF_L3(ibtl_mem, "ibt_query_mw(%p, %p)", hca_hdl, mw_hdl);
452 
453 	return (IBTL_HCA2CIHCAOPS_P(hca_hdl)->ibc_query_mw(
454 	    IBTL_HCA2CIHCA(hca_hdl), mw_hdl, mw_attr_p));
455 }
456 
457 
458 /*
459  * Function:
460  *	ibt_free_mw()
461  * Input:
462  *      hca_hdl   - HCA Handle
463  *	mw_hdl    - The IBT Memory Window handle.
464  * Output:
465  *	none.
466  * Returns:
467  *      IBT_SUCCESS
468  *	IBT_CHAN_HDL_INVALID
469  *	IBT_MW_HDL_INVALID
470  * Description:
471  *    De-allocate the Memory Window.
472  */
473 ibt_status_t
474 ibt_free_mw(ibt_hca_hdl_t hca_hdl, ibt_mw_hdl_t mw_hdl)
475 {
476 	ibt_status_t		status;
477 
478 	IBTF_DPRINTF_L3(ibtl_mem, "ibt_free_mw(%p, %p)", hca_hdl, mw_hdl);
479 
480 	status = IBTL_HCA2CIHCAOPS_P(hca_hdl)->ibc_free_mw(
481 	    IBTL_HCA2CIHCA(hca_hdl), mw_hdl);
482 
483 	if (status == IBT_SUCCESS) {
484 		mutex_enter(&hca_hdl->ha_mutex);
485 		hca_hdl->ha_mw_cnt--;
486 		mutex_exit(&hca_hdl->ha_mutex);
487 	}
488 	return (status);
489 }
490 
491 
492 /*
493  * Function:
494  *	ibt_map_mem_area()
495  * Input:
496  *      hca_hdl		HCA Handle
497  *	va_attrs	A pointer to an ibt_va_attr_t that describes the
498  *			VA to be translated.
499  *	paddr_list_len	The number of entries in the 'paddr_list_p' array.
500  * Output:
501  *	paddr_list_p	Array of ibt_phys_buf_t (allocated by the caller),
502  *			in which the physical buffers that map the virtual
503  *			buffer are returned.
504  *	num_paddr_p	The actual number of ibt_phys_buf_t that were
505  *			returned in the 'paddr_list_p' array.
506  *	ma_hdl_p	Memory Area Handle.
507  * Returns:
508  *      IBT_SUCCESS
509  * Description:
510  * 	Translate a kernel virtual address range into HCA physical addresses.
511  *	A set of physical addresses, that can be used with "Reserved L_Key",
512  *	register physical,  and "Fast Registration Work Request" operations
513  *	is returned.
514  */
515 ibt_status_t
516 ibt_map_mem_area(ibt_hca_hdl_t hca_hdl, ibt_va_attr_t *va_attrs,
517     uint_t paddr_list_len, ibt_phys_buf_t *paddr_list_p, uint_t *num_paddr_p,
518     size_t *paddr_bufsz_p, ib_memlen_t *paddr_offset_p, ibt_ma_hdl_t *ma_hdl_p)
519 {
520 	ibt_status_t 	status;
521 
522 	IBTF_DPRINTF_L3(ibtl_mem, "ibt_map_mem_area(%p, %p, %d)",
523 	    hca_hdl, va_attrs, paddr_list_len);
524 
525 	status = IBTL_HCA2CIHCAOPS_P(hca_hdl)->ibc_map_mem_area(
526 	    IBTL_HCA2CIHCA(hca_hdl), va_attrs,
527 	    NULL, /* IBTL_HCA2MODI_P(hca_hdl)->mi_reserved */
528 	    paddr_list_len, paddr_list_p, num_paddr_p,  paddr_bufsz_p,
529 	    paddr_offset_p, ma_hdl_p);
530 	if (status == IBT_SUCCESS) {
531 		mutex_enter(&hca_hdl->ha_mutex);
532 		hca_hdl->ha_ma_cnt++;
533 		mutex_exit(&hca_hdl->ha_mutex);
534 	}
535 
536 	return (status);
537 }
538 
539 
540 /*
541  * Function:
542  *	ibt_unmap_mem_area()
543  * Input:
544  *      hca_hdl		HCA Handle
545  *	ma_hdl		Memory Area Handle.
546  * Output:
547  *	None.
548  * Returns:
549  *      IBT_SUCCESS
550  * Description:
551  * 	Un pin physical pages pinned during an ibt_map_mem_area() call.
552  */
553 ibt_status_t
554 ibt_unmap_mem_area(ibt_hca_hdl_t hca_hdl, ibt_ma_hdl_t ma_hdl)
555 {
556 	ibt_status_t 	status;
557 
558 	IBTF_DPRINTF_L3(ibtl_mem, "ibt_unmap_mem_area(%p, %p)",
559 	    hca_hdl, ma_hdl);
560 
561 	status = (IBTL_HCA2CIHCAOPS_P(hca_hdl)->ibc_unmap_mem_area(
562 	    IBTL_HCA2CIHCA(hca_hdl), ma_hdl));
563 	if (status == IBT_SUCCESS) {
564 		mutex_enter(&hca_hdl->ha_mutex);
565 		hca_hdl->ha_ma_cnt--;
566 		mutex_exit(&hca_hdl->ha_mutex);
567 	}
568 
569 	return (status);
570 }
571 
572 /*
573  * Function:
574  *	ibt_map_mem_iov()
575  * Input:
576  *      hca_hdl		HCA Handle
577  *	iov_attr	A pointer to an ibt_iov_attr_t that describes the
578  *			virtual ranges to be translated.
579  * Output:
580  *	wr		A pointer to the work request where the output
581  *			sgl (reserved_lkey, size, paddr) will be written.
582  *	mi_hdl_p	Memory IOV Handle.
583  * Returns:
584  *      IBT_SUCCESS
585  * Description:
586  * 	Translate an array of virtual address ranges into HCA physical
587  *	addresses, sizes, and reserved_lkey.
588  */
589 ibt_status_t
590 ibt_map_mem_iov(ibt_hca_hdl_t hca_hdl, ibt_iov_attr_t *iov_attr,
591     ibt_all_wr_t *wr, ibt_mi_hdl_t *mi_hdl_p)
592 {
593 	ibt_status_t 	status;
594 
595 	IBTF_DPRINTF_L3(ibtl_mem, "ibt_map_mem_iov(%p, %p, %p)",
596 	    hca_hdl, iov_attr, wr);
597 
598 	status = IBTL_HCA2CIHCAOPS_P(hca_hdl)->ibc_map_mem_iov(
599 	    IBTL_HCA2CIHCA(hca_hdl), iov_attr, wr, mi_hdl_p);
600 	if (status == IBT_SUCCESS) {
601 		mutex_enter(&hca_hdl->ha_mutex);
602 		hca_hdl->ha_ma_cnt++;
603 		mutex_exit(&hca_hdl->ha_mutex);
604 	}
605 
606 	return (status);
607 }
608 
609 
610 /*
611  * Function:
612  *	ibt_unmap_mem_iov()
613  * Input:
614  *      hca_hdl		HCA Handle
615  *	mi_hdl		Memory IOV Handle.
616  * Output:
617  *	None.
618  * Returns:
619  *      IBT_SUCCESS
620  * Description:
621  * 	Un pin physical pages pinned during an ibt_map_mem_iov() call.
622  */
623 ibt_status_t
624 ibt_unmap_mem_iov(ibt_hca_hdl_t hca_hdl, ibt_mi_hdl_t mi_hdl)
625 {
626 	ibt_status_t 	status;
627 
628 	IBTF_DPRINTF_L3(ibtl_mem, "ibt_unmap_mem_iov(%p, %p)",
629 	    hca_hdl, mi_hdl);
630 
631 	status = (IBTL_HCA2CIHCAOPS_P(hca_hdl)->ibc_unmap_mem_iov(
632 	    IBTL_HCA2CIHCA(hca_hdl), mi_hdl));
633 	if (status == IBT_SUCCESS) {
634 		mutex_enter(&hca_hdl->ha_mutex);
635 		hca_hdl->ha_ma_cnt--;
636 		mutex_exit(&hca_hdl->ha_mutex);
637 	}
638 
639 	return (status);
640 }
641 
642 /*
643  * Function:
644  *	ibt_alloc_io_mem()
645  * Input:
646  *      hca_hdl		HCA Handle
647  *	size		Number of bytes to allocate
648  *	mr_flag		Possible values: IBT_MR_SLEEP, IBT_MR_NONCOHERENT
649  * Output:
650  *     	kaddrp 		Contains pointer to the virtual address of the
651  *			memory allocated by this call.  (Set to NULL if
652  *			memory allocation fails).
653  *	mem_alloc_hdl	Memory access handle returned by ibt_mem_alloc()
654  *
655  * Returns:
656  *      IBT_SUCCESS
657  *	IBT_INSUFF_RESOURCE
658  * 	IBT_HCA_HDL_INVALID
659  *	IBT_MR_ACCESS_REQ_INVALID
660  *	IBT_INVALID_PARAM
661  * Description:
662  *	Wrapper for ddi_dma_mem_alloc()
663  */
664 ibt_status_t
665 ibt_alloc_io_mem(ibt_hca_hdl_t hca_hdl, size_t size, ibt_mr_flags_t mr_flag,
666     caddr_t *kaddrp, ibt_mem_alloc_hdl_t *mem_alloc_hdl)
667 {
668 	return (IBTL_HCA2CIHCAOPS_P(hca_hdl)->ibc_alloc_io_mem(
669 	    IBTL_HCA2CIHCA(hca_hdl), size, mr_flag, kaddrp,
670 	    (ibc_mem_alloc_hdl_t *)mem_alloc_hdl));
671 }
672 
673 /*
674  * Function:
675  *	ibt_free_io_mem()
676  * Input:
677  *      hca_hdl		HCA Handle
678  *	mem_alloc_hdl	Memory access handle returned by ibt_mem_alloc()
679  * Output:
680  *	None
681  *
682  * Returns:
683  *      IBT_SUCCESS
684  * Description:
685  *	Wrapper for ddi_dma_mem_free()
686  */
687 ibt_status_t
688 ibt_free_io_mem(ibt_hca_hdl_t hca_hdl, ibt_mem_alloc_hdl_t mem_alloc_hdl)
689 {
690 	return (IBTL_HCA2CIHCAOPS_P(hca_hdl)->ibc_free_io_mem(
691 	    IBTL_HCA2CIHCA(hca_hdl), (ibc_mem_alloc_hdl_t)mem_alloc_hdl));
692 }
693 
694 /*
695  * Function:
696  *	ibt_alloc_lkey()
697  * Input:
698  *      hca_hdl			HCA Handle
699  *	pd			A protection domain handle.
700  *	flags			Access control.
701  *	phys_buf_list_sz	Requested size of Physical Buffer List (PBL)
702  *				resources to be allocated.
703  * Output:
704  *	mr_hdl_p		The returned IBT memory region handle.
705  *	mem_desc_p		Returned memory descriptor.
706  * Returns:
707  *      IBT_SUCCESS
708  * Description:
709  * 	Allocates physical buffer list resources for use in memory
710  *	registrations.
711  */
712 ibt_status_t
713 ibt_alloc_lkey(ibt_hca_hdl_t hca_hdl, ibt_pd_hdl_t pd, ibt_lkey_flags_t flags,
714     uint_t phys_buf_list_sz, ibt_mr_hdl_t *mr_hdl_p,
715     ibt_pmr_desc_t *mem_desc_p)
716 {
717 	ibt_status_t 	status;
718 
719 	IBTF_DPRINTF_L3(ibtl_mem, "ibt_alloc_lkey(%p, %p, 0x%X, %d)",
720 	    hca_hdl, pd, flags, phys_buf_list_sz);
721 
722 	status = IBTL_HCA2CIHCAOPS_P(hca_hdl)->ibc_alloc_lkey(
723 	    IBTL_HCA2CIHCA(hca_hdl), pd, flags, phys_buf_list_sz, mr_hdl_p,
724 	    mem_desc_p);
725 	if (status == IBT_SUCCESS) {
726 		mutex_enter(&hca_hdl->ha_mutex);
727 		hca_hdl->ha_mr_cnt++;
728 		mutex_exit(&hca_hdl->ha_mutex);
729 	}
730 
731 	return (status);
732 }
733 
734 
735 /*
736  * Function:
737  *	ibt_register_phys_mr()
738  * Input:
739  *      hca_hdl		HCA Handle
740  *	pd		A protection domain handle.
741  *	mem_pattr	Requested memory region physical attributes.
742  * Output:
743  *	mr_hdl_p	The returned IBT memory region handle.
744  *	mem_desc_p	Returned memory descriptor.
745  * Returns:
746  *      IBT_SUCCESS
747  * Description:
748  * 	Prepares a physically addressed memory region for use by a HCA.
749  */
750 ibt_status_t
751 ibt_register_phys_mr(ibt_hca_hdl_t hca_hdl, ibt_pd_hdl_t pd,
752     ibt_pmr_attr_t *mem_pattr, ibt_mr_hdl_t *mr_hdl_p,
753     ibt_pmr_desc_t *mem_desc_p)
754 {
755 	ibt_status_t 	status;
756 
757 	IBTF_DPRINTF_L3(ibtl_mem, "ibt_register_phys_mr(%p, %p, %p)",
758 	    hca_hdl, pd, mem_pattr);
759 
760 	status = IBTL_HCA2CIHCAOPS_P(hca_hdl)->ibc_register_physical_mr(
761 	    IBTL_HCA2CIHCA(hca_hdl), pd, mem_pattr,
762 	    NULL, /* IBTL_HCA2MODI_P(hca_hdl)->mi_reserved */
763 	    mr_hdl_p, mem_desc_p);
764 	if (status == IBT_SUCCESS) {
765 		mutex_enter(&hca_hdl->ha_mutex);
766 		hca_hdl->ha_mr_cnt++;
767 		mutex_exit(&hca_hdl->ha_mutex);
768 	}
769 
770 	return (status);
771 }
772 
773 
774 /*
775  * Function:
776  *	ibt_reregister_phys_mr()
777  * Input:
778  *      hca_hdl		HCA Handle
779  *	mr_hdl		The IBT memory region handle.
780  *	pd		A protection domain handle.
781  *	mem_pattr	Requested memory region physical attributes.
782  * Output:
783  *	mr_hdl_p	The returned IBT memory region handle.
784  *	mem_desc_p	Returned memory descriptor.
785  * Returns:
786  *      IBT_SUCCESS
787  * Description:
788  * 	Prepares a physically addressed memory region for use by a HCA.
789  */
790 ibt_status_t
791 ibt_reregister_phys_mr(ibt_hca_hdl_t hca_hdl, ibt_mr_hdl_t mr_hdl,
792     ibt_pd_hdl_t pd, ibt_pmr_attr_t *mem_pattr, ibt_mr_hdl_t *mr_hdl_p,
793     ibt_pmr_desc_t *mem_desc_p)
794 {
795 	ibt_status_t 	status;
796 
797 	IBTF_DPRINTF_L3(ibtl_mem, "ibt_reregister_phys_mr(%p, %p, %p, %p)",
798 	    hca_hdl, mr_hdl, pd, mem_pattr);
799 
800 	status = IBTL_HCA2CIHCAOPS_P(hca_hdl)->ibc_reregister_physical_mr(
801 	    IBTL_HCA2CIHCA(hca_hdl), mr_hdl, pd, mem_pattr,
802 	    NULL, /* IBTL_HCA2MODI_P(hca_hdl)->mi_reserved */
803 	    mr_hdl_p, mem_desc_p);
804 
805 	if (!(status == IBT_SUCCESS || status == IBT_MR_IN_USE ||
806 	    status == IBT_HCA_HDL_INVALID || status == IBT_MR_HDL_INVALID)) {
807 		IBTF_DPRINTF_L2(ibtl_mem, "ibt_reregister_phys_mr: "
808 		    "Re-registration Mem Failed: %d", status);
809 
810 		/* we lost one memory region resource */
811 		mutex_enter(&hca_hdl->ha_mutex);
812 		hca_hdl->ha_mr_cnt--;
813 		mutex_exit(&hca_hdl->ha_mutex);
814 
815 	}
816 	return (status);
817 }
818 
819 
820 /*
821  * Fast Memory Registration (FMR).
822  *
823  * ibt_create_fmr_pool
824  *      Not fast-path.
825  *      ibt_create_fmr_pool() verifies that the HCA supports FMR and allocates
826  *      and initializes an "FMR pool".  This pool contains state specific to
827  *      this registration, including the watermark setting to determine when
828  *      to sync, and the total number of FMR regions available within this pool.
829  *
830  */
831 ibt_status_t
832 ibt_create_fmr_pool(ibt_hca_hdl_t hca_hdl, ibt_pd_hdl_t pd,
833     ibt_fmr_pool_attr_t *fmr_params, ibt_fmr_pool_hdl_t *fmr_pool_p)
834 {
835 	ibt_status_t 		status;
836 
837 	IBTF_DPRINTF_L3(ibtl_mem, "ibt_create_fmr_pool(%p, %p, %p)",
838 	    hca_hdl, pd, fmr_params);
839 
840 	status = IBTL_HCA2CIHCAOPS_P(hca_hdl)->ibc_create_fmr_pool(
841 	    IBTL_HCA2CIHCA(hca_hdl), pd, fmr_params, fmr_pool_p);
842 	if (status != IBT_SUCCESS) {
843 		*fmr_pool_p = NULL;
844 		return (status);
845 	}
846 
847 	/* Update the FMR resource count */
848 	mutex_enter(&hca_hdl->ha_mutex);
849 	hca_hdl->ha_fmr_pool_cnt++;
850 	mutex_exit(&hca_hdl->ha_mutex);
851 
852 	return (status);
853 }
854 
855 
856 /*
857  * ibt_destroy_fmr_pool
858  *      ibt_destroy_fmr_pool() deallocates all of the FMR regions in a specific
859  *      pool.  All state and information regarding the pool are destroyed and
860  *      returned as free space once again.  No more use of FMR regions in this
861  *      pool are possible without a subsequent call to ibt_create_fmr_pool().
862  */
863 ibt_status_t
864 ibt_destroy_fmr_pool(ibt_hca_hdl_t hca_hdl, ibt_fmr_pool_hdl_t fmr_pool)
865 {
866 	ibt_status_t 	status;
867 
868 	IBTF_DPRINTF_L3(ibtl_mem, "ibt_destroy_fmr_pool(%p, %p)",
869 	    hca_hdl, fmr_pool);
870 
871 	status = IBTL_HCA2CIHCAOPS_P(hca_hdl)->ibc_destroy_fmr_pool(
872 	    IBTL_HCA2CIHCA(hca_hdl), fmr_pool);
873 	if (status != IBT_SUCCESS) {
874 		IBTF_DPRINTF_L2(ibtl_mem, "ibt_destroy_fmr_pool: "
875 		    "CI FMR Pool destroy failed (%d)", status);
876 		return (status);
877 	}
878 
879 	mutex_enter(&hca_hdl->ha_mutex);
880 	hca_hdl->ha_fmr_pool_cnt--;
881 	mutex_exit(&hca_hdl->ha_mutex);
882 
883 	return (status);
884 }
885 
886 /*
887  * ibt_flush_fmr_pool
888  *      ibt_flush_fmr_pool forces a flush to occur.  At the client's request,
889  *      any unmapped FMR regions (See 'ibt_deregister_mr())') are returned to
890  *      a free state.  This function allows for an asynchronous cleanup of
891  *      formerly used FMR regions.  Sync operation is also performed internally
892  *      by HCA driver, when 'watermark' settings for the number of free FMR
893  *      regions left in the "pool" is reached.
894  */
895 ibt_status_t
896 ibt_flush_fmr_pool(ibt_hca_hdl_t hca_hdl, ibt_fmr_pool_hdl_t fmr_pool)
897 {
898 	IBTF_DPRINTF_L3(ibtl_mem, "ibt_flush_fmr_pool(%p, %p)",
899 	    hca_hdl, fmr_pool);
900 
901 	return (IBTL_HCA2CIHCAOPS_P(hca_hdl)->ibc_flush_fmr_pool(
902 	    IBTL_HCA2CIHCA(hca_hdl), fmr_pool));
903 }
904 
905 /*
906  * ibt_register_physical_fmr
907  *      ibt_register_physical_fmr() assigns a "free" entry from the FMR Pool.
908  *      It first consults the "FMR cache" to see if this is a duplicate memory
909  *      registration to something already in use.  If not, then a free entry
910  *      in the "pool" is marked used.
911  */
912 ibt_status_t
913 ibt_register_physical_fmr(ibt_hca_hdl_t hca_hdl, ibt_fmr_pool_hdl_t fmr_pool,
914     ibt_pmr_attr_t *mem_pattr, ibt_mr_hdl_t *mr_hdl_p,
915     ibt_pmr_desc_t *mem_desc_p)
916 {
917 	IBTF_DPRINTF_L3(ibtl_mem, "ibt_register_physical_fmr(%p, %p, %p, %p)",
918 	    hca_hdl, fmr_pool, mem_pattr, mem_desc_p);
919 
920 	return (IBTL_HCA2CIHCAOPS_P(hca_hdl)->ibc_register_physical_fmr(
921 	    IBTL_HCA2CIHCA(hca_hdl), fmr_pool, mem_pattr,
922 	    NULL, /* IBTL_HCA2MODI_P(hca_hdl)->mi_reserved */
923 	    mr_hdl_p, mem_desc_p));
924 }
925 
926 /*
927  * ibt_deregister_fmr
928  *	The ibt_deregister_fmr un-maps the resources reserved from the FMR
929  *	pool by ibt_register_physical_fmr().   The ibt_deregister_fmr() will
930  *	mark the region as free in the FMR Pool.
931  */
932 ibt_status_t
933 ibt_deregister_fmr(ibt_hca_hdl_t hca, ibt_mr_hdl_t mr_hdl)
934 {
935 	IBTF_DPRINTF_L3(ibtl_mem, "ibt_deregister_fmr(%p, %p)", hca, mr_hdl);
936 
937 	return (IBTL_HCA2CIHCAOPS_P(hca)->ibc_deregister_fmr(
938 	    IBTL_HCA2CIHCA(hca), mr_hdl));
939 }
940