1 /*
2  *  IBM eServer eHCA Infiniband device driver for Linux on POWER
3  *
4  *  MR/MW functions
5  *
6  *  Authors: Dietmar Decker <ddecker@de.ibm.com>
7  *           Christoph Raisch <raisch@de.ibm.com>
8  *           Hoang-Nam Nguyen <hnguyen@de.ibm.com>
9  *
10  *  Copyright (c) 2005 IBM Corporation
11  *
12  *  All rights reserved.
13  *
14  *  This source code is distributed under a dual license of GPL v2.0 and OpenIB
15  *  BSD.
16  *
17  * OpenIB BSD License
18  *
19  * Redistribution and use in source and binary forms, with or without
20  * modification, are permitted provided that the following conditions are met:
21  *
22  * Redistributions of source code must retain the above copyright notice, this
23  * list of conditions and the following disclaimer.
24  *
25  * Redistributions in binary form must reproduce the above copyright notice,
26  * this list of conditions and the following disclaimer in the documentation
27  * and/or other materials
28  * provided with the distribution.
29  *
30  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
31  * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
32  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
33  * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
34  * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
35  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
36  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
37  * BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER
38  * IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
39  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
40  * POSSIBILITY OF SUCH DAMAGE.
41  */
42 
43 #include <linux/slab.h>
44 #include <rdma/ib_umem.h>
45 
46 #include "ehca_iverbs.h"
47 #include "ehca_mrmw.h"
48 #include "hcp_if.h"
49 #include "hipz_hw.h"
50 
51 #define NUM_CHUNKS(length, chunk_size) \
52 	(((length) + (chunk_size - 1)) / (chunk_size))
53 
54 /* max number of rpages (per hcall register_rpages) */
55 #define MAX_RPAGES 512
56 
57 /* DMEM toleration management */
58 #define EHCA_SECTSHIFT        SECTION_SIZE_BITS
59 #define EHCA_SECTSIZE          (1UL << EHCA_SECTSHIFT)
60 #define EHCA_HUGEPAGESHIFT     34
61 #define EHCA_HUGEPAGE_SIZE     (1UL << EHCA_HUGEPAGESHIFT)
62 #define EHCA_HUGEPAGE_PFN_MASK ((EHCA_HUGEPAGE_SIZE - 1) >> PAGE_SHIFT)
63 #define EHCA_INVAL_ADDR        0xFFFFFFFFFFFFFFFFULL
64 #define EHCA_DIR_INDEX_SHIFT 13                   /* 8k Entries in 64k block */
65 #define EHCA_TOP_INDEX_SHIFT (EHCA_DIR_INDEX_SHIFT * 2)
66 #define EHCA_MAP_ENTRIES (1 << EHCA_DIR_INDEX_SHIFT)
67 #define EHCA_TOP_MAP_SIZE (0x10000)               /* currently fixed map size */
68 #define EHCA_DIR_MAP_SIZE (0x10000)
69 #define EHCA_ENT_MAP_SIZE (0x10000)
70 #define EHCA_INDEX_MASK (EHCA_MAP_ENTRIES - 1)
71 
72 static unsigned long ehca_mr_len;
73 
74 /*
75  * Memory map data structures
76  */
77 struct ehca_dir_bmap {
78 	u64 ent[EHCA_MAP_ENTRIES];
79 };
80 struct ehca_top_bmap {
81 	struct ehca_dir_bmap *dir[EHCA_MAP_ENTRIES];
82 };
83 struct ehca_bmap {
84 	struct ehca_top_bmap *top[EHCA_MAP_ENTRIES];
85 };
86 
87 static struct ehca_bmap *ehca_bmap;
88 
89 static struct kmem_cache *mr_cache;
90 static struct kmem_cache *mw_cache;
91 
92 enum ehca_mr_pgsize {
93 	EHCA_MR_PGSIZE4K  = 0x1000L,
94 	EHCA_MR_PGSIZE64K = 0x10000L,
95 	EHCA_MR_PGSIZE1M  = 0x100000L,
96 	EHCA_MR_PGSIZE16M = 0x1000000L
97 };
98 
99 #define EHCA_MR_PGSHIFT4K  12
100 #define EHCA_MR_PGSHIFT64K 16
101 #define EHCA_MR_PGSHIFT1M  20
102 #define EHCA_MR_PGSHIFT16M 24
103 
104 static u64 ehca_map_vaddr(void *caddr);
105 
ehca_encode_hwpage_size(u32 pgsize)106 static u32 ehca_encode_hwpage_size(u32 pgsize)
107 {
108 	int log = ilog2(pgsize);
109 	WARN_ON(log < 12 || log > 24 || log & 3);
110 	return (log - 12) / 4;
111 }
112 
ehca_get_max_hwpage_size(struct ehca_shca * shca)113 static u64 ehca_get_max_hwpage_size(struct ehca_shca *shca)
114 {
115 	return rounddown_pow_of_two(shca->hca_cap_mr_pgsize);
116 }
117 
ehca_mr_new(void)118 static struct ehca_mr *ehca_mr_new(void)
119 {
120 	struct ehca_mr *me;
121 
122 	me = kmem_cache_zalloc(mr_cache, GFP_KERNEL);
123 	if (me)
124 		spin_lock_init(&me->mrlock);
125 	else
126 		ehca_gen_err("alloc failed");
127 
128 	return me;
129 }
130 
ehca_mr_delete(struct ehca_mr * me)131 static void ehca_mr_delete(struct ehca_mr *me)
132 {
133 	kmem_cache_free(mr_cache, me);
134 }
135 
ehca_mw_new(void)136 static struct ehca_mw *ehca_mw_new(void)
137 {
138 	struct ehca_mw *me;
139 
140 	me = kmem_cache_zalloc(mw_cache, GFP_KERNEL);
141 	if (me)
142 		spin_lock_init(&me->mwlock);
143 	else
144 		ehca_gen_err("alloc failed");
145 
146 	return me;
147 }
148 
ehca_mw_delete(struct ehca_mw * me)149 static void ehca_mw_delete(struct ehca_mw *me)
150 {
151 	kmem_cache_free(mw_cache, me);
152 }
153 
154 /*----------------------------------------------------------------------*/
155 
ehca_get_dma_mr(struct ib_pd * pd,int mr_access_flags)156 struct ib_mr *ehca_get_dma_mr(struct ib_pd *pd, int mr_access_flags)
157 {
158 	struct ib_mr *ib_mr;
159 	int ret;
160 	struct ehca_mr *e_maxmr;
161 	struct ehca_pd *e_pd = container_of(pd, struct ehca_pd, ib_pd);
162 	struct ehca_shca *shca =
163 		container_of(pd->device, struct ehca_shca, ib_device);
164 
165 	if (shca->maxmr) {
166 		e_maxmr = ehca_mr_new();
167 		if (!e_maxmr) {
168 			ehca_err(&shca->ib_device, "out of memory");
169 			ib_mr = ERR_PTR(-ENOMEM);
170 			goto get_dma_mr_exit0;
171 		}
172 
173 		ret = ehca_reg_maxmr(shca, e_maxmr,
174 				     (void *)ehca_map_vaddr((void *)(KERNELBASE + PHYSICAL_START)),
175 				     mr_access_flags, e_pd,
176 				     &e_maxmr->ib.ib_mr.lkey,
177 				     &e_maxmr->ib.ib_mr.rkey);
178 		if (ret) {
179 			ehca_mr_delete(e_maxmr);
180 			ib_mr = ERR_PTR(ret);
181 			goto get_dma_mr_exit0;
182 		}
183 		ib_mr = &e_maxmr->ib.ib_mr;
184 	} else {
185 		ehca_err(&shca->ib_device, "no internal max-MR exist!");
186 		ib_mr = ERR_PTR(-EINVAL);
187 		goto get_dma_mr_exit0;
188 	}
189 
190 get_dma_mr_exit0:
191 	if (IS_ERR(ib_mr))
192 		ehca_err(&shca->ib_device, "h_ret=%li pd=%p mr_access_flags=%x",
193 			 PTR_ERR(ib_mr), pd, mr_access_flags);
194 	return ib_mr;
195 } /* end ehca_get_dma_mr() */
196 
197 /*----------------------------------------------------------------------*/
198 
ehca_reg_phys_mr(struct ib_pd * pd,struct ib_phys_buf * phys_buf_array,int num_phys_buf,int mr_access_flags,u64 * iova_start)199 struct ib_mr *ehca_reg_phys_mr(struct ib_pd *pd,
200 			       struct ib_phys_buf *phys_buf_array,
201 			       int num_phys_buf,
202 			       int mr_access_flags,
203 			       u64 *iova_start)
204 {
205 	struct ib_mr *ib_mr;
206 	int ret;
207 	struct ehca_mr *e_mr;
208 	struct ehca_shca *shca =
209 		container_of(pd->device, struct ehca_shca, ib_device);
210 	struct ehca_pd *e_pd = container_of(pd, struct ehca_pd, ib_pd);
211 
212 	u64 size;
213 
214 	if ((num_phys_buf <= 0) || !phys_buf_array) {
215 		ehca_err(pd->device, "bad input values: num_phys_buf=%x "
216 			 "phys_buf_array=%p", num_phys_buf, phys_buf_array);
217 		ib_mr = ERR_PTR(-EINVAL);
218 		goto reg_phys_mr_exit0;
219 	}
220 	if (((mr_access_flags & IB_ACCESS_REMOTE_WRITE) &&
221 	     !(mr_access_flags & IB_ACCESS_LOCAL_WRITE)) ||
222 	    ((mr_access_flags & IB_ACCESS_REMOTE_ATOMIC) &&
223 	     !(mr_access_flags & IB_ACCESS_LOCAL_WRITE))) {
224 		/*
225 		 * Remote Write Access requires Local Write Access
226 		 * Remote Atomic Access requires Local Write Access
227 		 */
228 		ehca_err(pd->device, "bad input values: mr_access_flags=%x",
229 			 mr_access_flags);
230 		ib_mr = ERR_PTR(-EINVAL);
231 		goto reg_phys_mr_exit0;
232 	}
233 
234 	/* check physical buffer list and calculate size */
235 	ret = ehca_mr_chk_buf_and_calc_size(phys_buf_array, num_phys_buf,
236 					    iova_start, &size);
237 	if (ret) {
238 		ib_mr = ERR_PTR(ret);
239 		goto reg_phys_mr_exit0;
240 	}
241 	if ((size == 0) ||
242 	    (((u64)iova_start + size) < (u64)iova_start)) {
243 		ehca_err(pd->device, "bad input values: size=%llx iova_start=%p",
244 			 size, iova_start);
245 		ib_mr = ERR_PTR(-EINVAL);
246 		goto reg_phys_mr_exit0;
247 	}
248 
249 	e_mr = ehca_mr_new();
250 	if (!e_mr) {
251 		ehca_err(pd->device, "out of memory");
252 		ib_mr = ERR_PTR(-ENOMEM);
253 		goto reg_phys_mr_exit0;
254 	}
255 
256 	/* register MR on HCA */
257 	if (ehca_mr_is_maxmr(size, iova_start)) {
258 		e_mr->flags |= EHCA_MR_FLAG_MAXMR;
259 		ret = ehca_reg_maxmr(shca, e_mr, iova_start, mr_access_flags,
260 				     e_pd, &e_mr->ib.ib_mr.lkey,
261 				     &e_mr->ib.ib_mr.rkey);
262 		if (ret) {
263 			ib_mr = ERR_PTR(ret);
264 			goto reg_phys_mr_exit1;
265 		}
266 	} else {
267 		struct ehca_mr_pginfo pginfo;
268 		u32 num_kpages;
269 		u32 num_hwpages;
270 		u64 hw_pgsize;
271 
272 		num_kpages = NUM_CHUNKS(((u64)iova_start % PAGE_SIZE) + size,
273 					PAGE_SIZE);
274 		/* for kernel space we try most possible pgsize */
275 		hw_pgsize = ehca_get_max_hwpage_size(shca);
276 		num_hwpages = NUM_CHUNKS(((u64)iova_start % hw_pgsize) + size,
277 					 hw_pgsize);
278 		memset(&pginfo, 0, sizeof(pginfo));
279 		pginfo.type = EHCA_MR_PGI_PHYS;
280 		pginfo.num_kpages = num_kpages;
281 		pginfo.hwpage_size = hw_pgsize;
282 		pginfo.num_hwpages = num_hwpages;
283 		pginfo.u.phy.num_phys_buf = num_phys_buf;
284 		pginfo.u.phy.phys_buf_array = phys_buf_array;
285 		pginfo.next_hwpage =
286 			((u64)iova_start & ~PAGE_MASK) / hw_pgsize;
287 
288 		ret = ehca_reg_mr(shca, e_mr, iova_start, size, mr_access_flags,
289 				  e_pd, &pginfo, &e_mr->ib.ib_mr.lkey,
290 				  &e_mr->ib.ib_mr.rkey, EHCA_REG_MR);
291 		if (ret) {
292 			ib_mr = ERR_PTR(ret);
293 			goto reg_phys_mr_exit1;
294 		}
295 	}
296 
297 	/* successful registration of all pages */
298 	return &e_mr->ib.ib_mr;
299 
300 reg_phys_mr_exit1:
301 	ehca_mr_delete(e_mr);
302 reg_phys_mr_exit0:
303 	if (IS_ERR(ib_mr))
304 		ehca_err(pd->device, "h_ret=%li pd=%p phys_buf_array=%p "
305 			 "num_phys_buf=%x mr_access_flags=%x iova_start=%p",
306 			 PTR_ERR(ib_mr), pd, phys_buf_array,
307 			 num_phys_buf, mr_access_flags, iova_start);
308 	return ib_mr;
309 } /* end ehca_reg_phys_mr() */
310 
311 /*----------------------------------------------------------------------*/
312 
ehca_reg_user_mr(struct ib_pd * pd,u64 start,u64 length,u64 virt,int mr_access_flags,struct ib_udata * udata)313 struct ib_mr *ehca_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
314 			       u64 virt, int mr_access_flags,
315 			       struct ib_udata *udata)
316 {
317 	struct ib_mr *ib_mr;
318 	struct ehca_mr *e_mr;
319 	struct ehca_shca *shca =
320 		container_of(pd->device, struct ehca_shca, ib_device);
321 	struct ehca_pd *e_pd = container_of(pd, struct ehca_pd, ib_pd);
322 	struct ehca_mr_pginfo pginfo;
323 	int ret, page_shift;
324 	u32 num_kpages;
325 	u32 num_hwpages;
326 	u64 hwpage_size;
327 
328 	if (!pd) {
329 		ehca_gen_err("bad pd=%p", pd);
330 		return ERR_PTR(-EFAULT);
331 	}
332 
333 	if (((mr_access_flags & IB_ACCESS_REMOTE_WRITE) &&
334 	     !(mr_access_flags & IB_ACCESS_LOCAL_WRITE)) ||
335 	    ((mr_access_flags & IB_ACCESS_REMOTE_ATOMIC) &&
336 	     !(mr_access_flags & IB_ACCESS_LOCAL_WRITE))) {
337 		/*
338 		 * Remote Write Access requires Local Write Access
339 		 * Remote Atomic Access requires Local Write Access
340 		 */
341 		ehca_err(pd->device, "bad input values: mr_access_flags=%x",
342 			 mr_access_flags);
343 		ib_mr = ERR_PTR(-EINVAL);
344 		goto reg_user_mr_exit0;
345 	}
346 
347 	if (length == 0 || virt + length < virt) {
348 		ehca_err(pd->device, "bad input values: length=%llx "
349 			 "virt_base=%llx", length, virt);
350 		ib_mr = ERR_PTR(-EINVAL);
351 		goto reg_user_mr_exit0;
352 	}
353 
354 	e_mr = ehca_mr_new();
355 	if (!e_mr) {
356 		ehca_err(pd->device, "out of memory");
357 		ib_mr = ERR_PTR(-ENOMEM);
358 		goto reg_user_mr_exit0;
359 	}
360 
361 	e_mr->umem = ib_umem_get(pd->uobject->context, start, length,
362 				 mr_access_flags, 0);
363 	if (IS_ERR(e_mr->umem)) {
364 		ib_mr = (void *)e_mr->umem;
365 		goto reg_user_mr_exit1;
366 	}
367 
368 	if (e_mr->umem->page_size != PAGE_SIZE) {
369 		ehca_err(pd->device, "page size not supported, "
370 			 "e_mr->umem->page_size=%x", e_mr->umem->page_size);
371 		ib_mr = ERR_PTR(-EINVAL);
372 		goto reg_user_mr_exit2;
373 	}
374 
375 	/* determine number of MR pages */
376 	num_kpages = NUM_CHUNKS((virt % PAGE_SIZE) + length, PAGE_SIZE);
377 	/* select proper hw_pgsize */
378 	page_shift = PAGE_SHIFT;
379 	if (e_mr->umem->hugetlb) {
380 		/* determine page_shift, clamp between 4K and 16M */
381 		page_shift = (fls64(length - 1) + 3) & ~3;
382 		page_shift = min(max(page_shift, EHCA_MR_PGSHIFT4K),
383 				 EHCA_MR_PGSHIFT16M);
384 	}
385 	hwpage_size = 1UL << page_shift;
386 
387 	/* now that we have the desired page size, shift until it's
388 	 * supported, too. 4K is always supported, so this terminates.
389 	 */
390 	while (!(hwpage_size & shca->hca_cap_mr_pgsize))
391 		hwpage_size >>= 4;
392 
393 reg_user_mr_fallback:
394 	num_hwpages = NUM_CHUNKS((virt % hwpage_size) + length, hwpage_size);
395 	/* register MR on HCA */
396 	memset(&pginfo, 0, sizeof(pginfo));
397 	pginfo.type = EHCA_MR_PGI_USER;
398 	pginfo.hwpage_size = hwpage_size;
399 	pginfo.num_kpages = num_kpages;
400 	pginfo.num_hwpages = num_hwpages;
401 	pginfo.u.usr.region = e_mr->umem;
402 	pginfo.next_hwpage = e_mr->umem->offset / hwpage_size;
403 	pginfo.u.usr.next_chunk = list_prepare_entry(pginfo.u.usr.next_chunk,
404 						     (&e_mr->umem->chunk_list),
405 						     list);
406 
407 	ret = ehca_reg_mr(shca, e_mr, (u64 *)virt, length, mr_access_flags,
408 			  e_pd, &pginfo, &e_mr->ib.ib_mr.lkey,
409 			  &e_mr->ib.ib_mr.rkey, EHCA_REG_MR);
410 	if (ret == -EINVAL && pginfo.hwpage_size > PAGE_SIZE) {
411 		ehca_warn(pd->device, "failed to register mr "
412 			  "with hwpage_size=%llx", hwpage_size);
413 		ehca_info(pd->device, "try to register mr with "
414 			  "kpage_size=%lx", PAGE_SIZE);
415 		/*
416 		 * this means kpages are not contiguous for a hw page
417 		 * try kernel page size as fallback solution
418 		 */
419 		hwpage_size = PAGE_SIZE;
420 		goto reg_user_mr_fallback;
421 	}
422 	if (ret) {
423 		ib_mr = ERR_PTR(ret);
424 		goto reg_user_mr_exit2;
425 	}
426 
427 	/* successful registration of all pages */
428 	return &e_mr->ib.ib_mr;
429 
430 reg_user_mr_exit2:
431 	ib_umem_release(e_mr->umem);
432 reg_user_mr_exit1:
433 	ehca_mr_delete(e_mr);
434 reg_user_mr_exit0:
435 	if (IS_ERR(ib_mr))
436 		ehca_err(pd->device, "rc=%li pd=%p mr_access_flags=%x udata=%p",
437 			 PTR_ERR(ib_mr), pd, mr_access_flags, udata);
438 	return ib_mr;
439 } /* end ehca_reg_user_mr() */
440 
441 /*----------------------------------------------------------------------*/
442 
ehca_rereg_phys_mr(struct ib_mr * mr,int mr_rereg_mask,struct ib_pd * pd,struct ib_phys_buf * phys_buf_array,int num_phys_buf,int mr_access_flags,u64 * iova_start)443 int ehca_rereg_phys_mr(struct ib_mr *mr,
444 		       int mr_rereg_mask,
445 		       struct ib_pd *pd,
446 		       struct ib_phys_buf *phys_buf_array,
447 		       int num_phys_buf,
448 		       int mr_access_flags,
449 		       u64 *iova_start)
450 {
451 	int ret;
452 
453 	struct ehca_shca *shca =
454 		container_of(mr->device, struct ehca_shca, ib_device);
455 	struct ehca_mr *e_mr = container_of(mr, struct ehca_mr, ib.ib_mr);
456 	u64 new_size;
457 	u64 *new_start;
458 	u32 new_acl;
459 	struct ehca_pd *new_pd;
460 	u32 tmp_lkey, tmp_rkey;
461 	unsigned long sl_flags;
462 	u32 num_kpages = 0;
463 	u32 num_hwpages = 0;
464 	struct ehca_mr_pginfo pginfo;
465 
466 	if (!(mr_rereg_mask & IB_MR_REREG_TRANS)) {
467 		/* TODO not supported, because PHYP rereg hCall needs pages */
468 		ehca_err(mr->device, "rereg without IB_MR_REREG_TRANS not "
469 			 "supported yet, mr_rereg_mask=%x", mr_rereg_mask);
470 		ret = -EINVAL;
471 		goto rereg_phys_mr_exit0;
472 	}
473 
474 	if (mr_rereg_mask & IB_MR_REREG_PD) {
475 		if (!pd) {
476 			ehca_err(mr->device, "rereg with bad pd, pd=%p "
477 				 "mr_rereg_mask=%x", pd, mr_rereg_mask);
478 			ret = -EINVAL;
479 			goto rereg_phys_mr_exit0;
480 		}
481 	}
482 
483 	if ((mr_rereg_mask &
484 	     ~(IB_MR_REREG_TRANS | IB_MR_REREG_PD | IB_MR_REREG_ACCESS)) ||
485 	    (mr_rereg_mask == 0)) {
486 		ret = -EINVAL;
487 		goto rereg_phys_mr_exit0;
488 	}
489 
490 	/* check other parameters */
491 	if (e_mr == shca->maxmr) {
492 		/* should be impossible, however reject to be sure */
493 		ehca_err(mr->device, "rereg internal max-MR impossible, mr=%p "
494 			 "shca->maxmr=%p mr->lkey=%x",
495 			 mr, shca->maxmr, mr->lkey);
496 		ret = -EINVAL;
497 		goto rereg_phys_mr_exit0;
498 	}
499 	if (mr_rereg_mask & IB_MR_REREG_TRANS) { /* transl., i.e. addr/size */
500 		if (e_mr->flags & EHCA_MR_FLAG_FMR) {
501 			ehca_err(mr->device, "not supported for FMR, mr=%p "
502 				 "flags=%x", mr, e_mr->flags);
503 			ret = -EINVAL;
504 			goto rereg_phys_mr_exit0;
505 		}
506 		if (!phys_buf_array || num_phys_buf <= 0) {
507 			ehca_err(mr->device, "bad input values mr_rereg_mask=%x"
508 				 " phys_buf_array=%p num_phys_buf=%x",
509 				 mr_rereg_mask, phys_buf_array, num_phys_buf);
510 			ret = -EINVAL;
511 			goto rereg_phys_mr_exit0;
512 		}
513 	}
514 	if ((mr_rereg_mask & IB_MR_REREG_ACCESS) &&	/* change ACL */
515 	    (((mr_access_flags & IB_ACCESS_REMOTE_WRITE) &&
516 	      !(mr_access_flags & IB_ACCESS_LOCAL_WRITE)) ||
517 	     ((mr_access_flags & IB_ACCESS_REMOTE_ATOMIC) &&
518 	      !(mr_access_flags & IB_ACCESS_LOCAL_WRITE)))) {
519 		/*
520 		 * Remote Write Access requires Local Write Access
521 		 * Remote Atomic Access requires Local Write Access
522 		 */
523 		ehca_err(mr->device, "bad input values: mr_rereg_mask=%x "
524 			 "mr_access_flags=%x", mr_rereg_mask, mr_access_flags);
525 		ret = -EINVAL;
526 		goto rereg_phys_mr_exit0;
527 	}
528 
529 	/* set requested values dependent on rereg request */
530 	spin_lock_irqsave(&e_mr->mrlock, sl_flags);
531 	new_start = e_mr->start;
532 	new_size = e_mr->size;
533 	new_acl = e_mr->acl;
534 	new_pd = container_of(mr->pd, struct ehca_pd, ib_pd);
535 
536 	if (mr_rereg_mask & IB_MR_REREG_TRANS) {
537 		u64 hw_pgsize = ehca_get_max_hwpage_size(shca);
538 
539 		new_start = iova_start;	/* change address */
540 		/* check physical buffer list and calculate size */
541 		ret = ehca_mr_chk_buf_and_calc_size(phys_buf_array,
542 						    num_phys_buf, iova_start,
543 						    &new_size);
544 		if (ret)
545 			goto rereg_phys_mr_exit1;
546 		if ((new_size == 0) ||
547 		    (((u64)iova_start + new_size) < (u64)iova_start)) {
548 			ehca_err(mr->device, "bad input values: new_size=%llx "
549 				 "iova_start=%p", new_size, iova_start);
550 			ret = -EINVAL;
551 			goto rereg_phys_mr_exit1;
552 		}
553 		num_kpages = NUM_CHUNKS(((u64)new_start % PAGE_SIZE) +
554 					new_size, PAGE_SIZE);
555 		num_hwpages = NUM_CHUNKS(((u64)new_start % hw_pgsize) +
556 					 new_size, hw_pgsize);
557 		memset(&pginfo, 0, sizeof(pginfo));
558 		pginfo.type = EHCA_MR_PGI_PHYS;
559 		pginfo.num_kpages = num_kpages;
560 		pginfo.hwpage_size = hw_pgsize;
561 		pginfo.num_hwpages = num_hwpages;
562 		pginfo.u.phy.num_phys_buf = num_phys_buf;
563 		pginfo.u.phy.phys_buf_array = phys_buf_array;
564 		pginfo.next_hwpage =
565 			((u64)iova_start & ~PAGE_MASK) / hw_pgsize;
566 	}
567 	if (mr_rereg_mask & IB_MR_REREG_ACCESS)
568 		new_acl = mr_access_flags;
569 	if (mr_rereg_mask & IB_MR_REREG_PD)
570 		new_pd = container_of(pd, struct ehca_pd, ib_pd);
571 
572 	ret = ehca_rereg_mr(shca, e_mr, new_start, new_size, new_acl,
573 			    new_pd, &pginfo, &tmp_lkey, &tmp_rkey);
574 	if (ret)
575 		goto rereg_phys_mr_exit1;
576 
577 	/* successful reregistration */
578 	if (mr_rereg_mask & IB_MR_REREG_PD)
579 		mr->pd = pd;
580 	mr->lkey = tmp_lkey;
581 	mr->rkey = tmp_rkey;
582 
583 rereg_phys_mr_exit1:
584 	spin_unlock_irqrestore(&e_mr->mrlock, sl_flags);
585 rereg_phys_mr_exit0:
586 	if (ret)
587 		ehca_err(mr->device, "ret=%i mr=%p mr_rereg_mask=%x pd=%p "
588 			 "phys_buf_array=%p num_phys_buf=%x mr_access_flags=%x "
589 			 "iova_start=%p",
590 			 ret, mr, mr_rereg_mask, pd, phys_buf_array,
591 			 num_phys_buf, mr_access_flags, iova_start);
592 	return ret;
593 } /* end ehca_rereg_phys_mr() */
594 
595 /*----------------------------------------------------------------------*/
596 
ehca_query_mr(struct ib_mr * mr,struct ib_mr_attr * mr_attr)597 int ehca_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr)
598 {
599 	int ret = 0;
600 	u64 h_ret;
601 	struct ehca_shca *shca =
602 		container_of(mr->device, struct ehca_shca, ib_device);
603 	struct ehca_mr *e_mr = container_of(mr, struct ehca_mr, ib.ib_mr);
604 	unsigned long sl_flags;
605 	struct ehca_mr_hipzout_parms hipzout;
606 
607 	if ((e_mr->flags & EHCA_MR_FLAG_FMR)) {
608 		ehca_err(mr->device, "not supported for FMR, mr=%p e_mr=%p "
609 			 "e_mr->flags=%x", mr, e_mr, e_mr->flags);
610 		ret = -EINVAL;
611 		goto query_mr_exit0;
612 	}
613 
614 	memset(mr_attr, 0, sizeof(struct ib_mr_attr));
615 	spin_lock_irqsave(&e_mr->mrlock, sl_flags);
616 
617 	h_ret = hipz_h_query_mr(shca->ipz_hca_handle, e_mr, &hipzout);
618 	if (h_ret != H_SUCCESS) {
619 		ehca_err(mr->device, "hipz_mr_query failed, h_ret=%lli mr=%p "
620 			 "hca_hndl=%llx mr_hndl=%llx lkey=%x",
621 			 h_ret, mr, shca->ipz_hca_handle.handle,
622 			 e_mr->ipz_mr_handle.handle, mr->lkey);
623 		ret = ehca2ib_return_code(h_ret);
624 		goto query_mr_exit1;
625 	}
626 	mr_attr->pd = mr->pd;
627 	mr_attr->device_virt_addr = hipzout.vaddr;
628 	mr_attr->size = hipzout.len;
629 	mr_attr->lkey = hipzout.lkey;
630 	mr_attr->rkey = hipzout.rkey;
631 	ehca_mrmw_reverse_map_acl(&hipzout.acl, &mr_attr->mr_access_flags);
632 
633 query_mr_exit1:
634 	spin_unlock_irqrestore(&e_mr->mrlock, sl_flags);
635 query_mr_exit0:
636 	if (ret)
637 		ehca_err(mr->device, "ret=%i mr=%p mr_attr=%p",
638 			 ret, mr, mr_attr);
639 	return ret;
640 } /* end ehca_query_mr() */
641 
642 /*----------------------------------------------------------------------*/
643 
ehca_dereg_mr(struct ib_mr * mr)644 int ehca_dereg_mr(struct ib_mr *mr)
645 {
646 	int ret = 0;
647 	u64 h_ret;
648 	struct ehca_shca *shca =
649 		container_of(mr->device, struct ehca_shca, ib_device);
650 	struct ehca_mr *e_mr = container_of(mr, struct ehca_mr, ib.ib_mr);
651 
652 	if ((e_mr->flags & EHCA_MR_FLAG_FMR)) {
653 		ehca_err(mr->device, "not supported for FMR, mr=%p e_mr=%p "
654 			 "e_mr->flags=%x", mr, e_mr, e_mr->flags);
655 		ret = -EINVAL;
656 		goto dereg_mr_exit0;
657 	} else if (e_mr == shca->maxmr) {
658 		/* should be impossible, however reject to be sure */
659 		ehca_err(mr->device, "dereg internal max-MR impossible, mr=%p "
660 			 "shca->maxmr=%p mr->lkey=%x",
661 			 mr, shca->maxmr, mr->lkey);
662 		ret = -EINVAL;
663 		goto dereg_mr_exit0;
664 	}
665 
666 	/* TODO: BUSY: MR still has bound window(s) */
667 	h_ret = hipz_h_free_resource_mr(shca->ipz_hca_handle, e_mr);
668 	if (h_ret != H_SUCCESS) {
669 		ehca_err(mr->device, "hipz_free_mr failed, h_ret=%lli shca=%p "
670 			 "e_mr=%p hca_hndl=%llx mr_hndl=%llx mr->lkey=%x",
671 			 h_ret, shca, e_mr, shca->ipz_hca_handle.handle,
672 			 e_mr->ipz_mr_handle.handle, mr->lkey);
673 		ret = ehca2ib_return_code(h_ret);
674 		goto dereg_mr_exit0;
675 	}
676 
677 	if (e_mr->umem)
678 		ib_umem_release(e_mr->umem);
679 
680 	/* successful deregistration */
681 	ehca_mr_delete(e_mr);
682 
683 dereg_mr_exit0:
684 	if (ret)
685 		ehca_err(mr->device, "ret=%i mr=%p", ret, mr);
686 	return ret;
687 } /* end ehca_dereg_mr() */
688 
689 /*----------------------------------------------------------------------*/
690 
ehca_alloc_mw(struct ib_pd * pd)691 struct ib_mw *ehca_alloc_mw(struct ib_pd *pd)
692 {
693 	struct ib_mw *ib_mw;
694 	u64 h_ret;
695 	struct ehca_mw *e_mw;
696 	struct ehca_pd *e_pd = container_of(pd, struct ehca_pd, ib_pd);
697 	struct ehca_shca *shca =
698 		container_of(pd->device, struct ehca_shca, ib_device);
699 	struct ehca_mw_hipzout_parms hipzout;
700 
701 	e_mw = ehca_mw_new();
702 	if (!e_mw) {
703 		ib_mw = ERR_PTR(-ENOMEM);
704 		goto alloc_mw_exit0;
705 	}
706 
707 	h_ret = hipz_h_alloc_resource_mw(shca->ipz_hca_handle, e_mw,
708 					 e_pd->fw_pd, &hipzout);
709 	if (h_ret != H_SUCCESS) {
710 		ehca_err(pd->device, "hipz_mw_allocate failed, h_ret=%lli "
711 			 "shca=%p hca_hndl=%llx mw=%p",
712 			 h_ret, shca, shca->ipz_hca_handle.handle, e_mw);
713 		ib_mw = ERR_PTR(ehca2ib_return_code(h_ret));
714 		goto alloc_mw_exit1;
715 	}
716 	/* successful MW allocation */
717 	e_mw->ipz_mw_handle = hipzout.handle;
718 	e_mw->ib_mw.rkey    = hipzout.rkey;
719 	return &e_mw->ib_mw;
720 
721 alloc_mw_exit1:
722 	ehca_mw_delete(e_mw);
723 alloc_mw_exit0:
724 	if (IS_ERR(ib_mw))
725 		ehca_err(pd->device, "h_ret=%li pd=%p", PTR_ERR(ib_mw), pd);
726 	return ib_mw;
727 } /* end ehca_alloc_mw() */
728 
729 /*----------------------------------------------------------------------*/
730 
ehca_bind_mw(struct ib_qp * qp,struct ib_mw * mw,struct ib_mw_bind * mw_bind)731 int ehca_bind_mw(struct ib_qp *qp,
732 		 struct ib_mw *mw,
733 		 struct ib_mw_bind *mw_bind)
734 {
735 	/* TODO: not supported up to now */
736 	ehca_gen_err("bind MW currently not supported by HCAD");
737 
738 	return -EPERM;
739 } /* end ehca_bind_mw() */
740 
741 /*----------------------------------------------------------------------*/
742 
ehca_dealloc_mw(struct ib_mw * mw)743 int ehca_dealloc_mw(struct ib_mw *mw)
744 {
745 	u64 h_ret;
746 	struct ehca_shca *shca =
747 		container_of(mw->device, struct ehca_shca, ib_device);
748 	struct ehca_mw *e_mw = container_of(mw, struct ehca_mw, ib_mw);
749 
750 	h_ret = hipz_h_free_resource_mw(shca->ipz_hca_handle, e_mw);
751 	if (h_ret != H_SUCCESS) {
752 		ehca_err(mw->device, "hipz_free_mw failed, h_ret=%lli shca=%p "
753 			 "mw=%p rkey=%x hca_hndl=%llx mw_hndl=%llx",
754 			 h_ret, shca, mw, mw->rkey, shca->ipz_hca_handle.handle,
755 			 e_mw->ipz_mw_handle.handle);
756 		return ehca2ib_return_code(h_ret);
757 	}
758 	/* successful deallocation */
759 	ehca_mw_delete(e_mw);
760 	return 0;
761 } /* end ehca_dealloc_mw() */
762 
763 /*----------------------------------------------------------------------*/
764 
ehca_alloc_fmr(struct ib_pd * pd,int mr_access_flags,struct ib_fmr_attr * fmr_attr)765 struct ib_fmr *ehca_alloc_fmr(struct ib_pd *pd,
766 			      int mr_access_flags,
767 			      struct ib_fmr_attr *fmr_attr)
768 {
769 	struct ib_fmr *ib_fmr;
770 	struct ehca_shca *shca =
771 		container_of(pd->device, struct ehca_shca, ib_device);
772 	struct ehca_pd *e_pd = container_of(pd, struct ehca_pd, ib_pd);
773 	struct ehca_mr *e_fmr;
774 	int ret;
775 	u32 tmp_lkey, tmp_rkey;
776 	struct ehca_mr_pginfo pginfo;
777 	u64 hw_pgsize;
778 
779 	/* check other parameters */
780 	if (((mr_access_flags & IB_ACCESS_REMOTE_WRITE) &&
781 	     !(mr_access_flags & IB_ACCESS_LOCAL_WRITE)) ||
782 	    ((mr_access_flags & IB_ACCESS_REMOTE_ATOMIC) &&
783 	     !(mr_access_flags & IB_ACCESS_LOCAL_WRITE))) {
784 		/*
785 		 * Remote Write Access requires Local Write Access
786 		 * Remote Atomic Access requires Local Write Access
787 		 */
788 		ehca_err(pd->device, "bad input values: mr_access_flags=%x",
789 			 mr_access_flags);
790 		ib_fmr = ERR_PTR(-EINVAL);
791 		goto alloc_fmr_exit0;
792 	}
793 	if (mr_access_flags & IB_ACCESS_MW_BIND) {
794 		ehca_err(pd->device, "bad input values: mr_access_flags=%x",
795 			 mr_access_flags);
796 		ib_fmr = ERR_PTR(-EINVAL);
797 		goto alloc_fmr_exit0;
798 	}
799 	if ((fmr_attr->max_pages == 0) || (fmr_attr->max_maps == 0)) {
800 		ehca_err(pd->device, "bad input values: fmr_attr->max_pages=%x "
801 			 "fmr_attr->max_maps=%x fmr_attr->page_shift=%x",
802 			 fmr_attr->max_pages, fmr_attr->max_maps,
803 			 fmr_attr->page_shift);
804 		ib_fmr = ERR_PTR(-EINVAL);
805 		goto alloc_fmr_exit0;
806 	}
807 
808 	hw_pgsize = 1 << fmr_attr->page_shift;
809 	if (!(hw_pgsize & shca->hca_cap_mr_pgsize)) {
810 		ehca_err(pd->device, "unsupported fmr_attr->page_shift=%x",
811 			 fmr_attr->page_shift);
812 		ib_fmr = ERR_PTR(-EINVAL);
813 		goto alloc_fmr_exit0;
814 	}
815 
816 	e_fmr = ehca_mr_new();
817 	if (!e_fmr) {
818 		ib_fmr = ERR_PTR(-ENOMEM);
819 		goto alloc_fmr_exit0;
820 	}
821 	e_fmr->flags |= EHCA_MR_FLAG_FMR;
822 
823 	/* register MR on HCA */
824 	memset(&pginfo, 0, sizeof(pginfo));
825 	pginfo.hwpage_size = hw_pgsize;
826 	/*
827 	 * pginfo.num_hwpages==0, ie register_rpages() will not be called
828 	 * but deferred to map_phys_fmr()
829 	 */
830 	ret = ehca_reg_mr(shca, e_fmr, NULL,
831 			  fmr_attr->max_pages * (1 << fmr_attr->page_shift),
832 			  mr_access_flags, e_pd, &pginfo,
833 			  &tmp_lkey, &tmp_rkey, EHCA_REG_MR);
834 	if (ret) {
835 		ib_fmr = ERR_PTR(ret);
836 		goto alloc_fmr_exit1;
837 	}
838 
839 	/* successful */
840 	e_fmr->hwpage_size = hw_pgsize;
841 	e_fmr->fmr_page_size = 1 << fmr_attr->page_shift;
842 	e_fmr->fmr_max_pages = fmr_attr->max_pages;
843 	e_fmr->fmr_max_maps = fmr_attr->max_maps;
844 	e_fmr->fmr_map_cnt = 0;
845 	return &e_fmr->ib.ib_fmr;
846 
847 alloc_fmr_exit1:
848 	ehca_mr_delete(e_fmr);
849 alloc_fmr_exit0:
850 	return ib_fmr;
851 } /* end ehca_alloc_fmr() */
852 
853 /*----------------------------------------------------------------------*/
854 
ehca_map_phys_fmr(struct ib_fmr * fmr,u64 * page_list,int list_len,u64 iova)855 int ehca_map_phys_fmr(struct ib_fmr *fmr,
856 		      u64 *page_list,
857 		      int list_len,
858 		      u64 iova)
859 {
860 	int ret;
861 	struct ehca_shca *shca =
862 		container_of(fmr->device, struct ehca_shca, ib_device);
863 	struct ehca_mr *e_fmr = container_of(fmr, struct ehca_mr, ib.ib_fmr);
864 	struct ehca_pd *e_pd = container_of(fmr->pd, struct ehca_pd, ib_pd);
865 	struct ehca_mr_pginfo pginfo;
866 	u32 tmp_lkey, tmp_rkey;
867 
868 	if (!(e_fmr->flags & EHCA_MR_FLAG_FMR)) {
869 		ehca_err(fmr->device, "not a FMR, e_fmr=%p e_fmr->flags=%x",
870 			 e_fmr, e_fmr->flags);
871 		ret = -EINVAL;
872 		goto map_phys_fmr_exit0;
873 	}
874 	ret = ehca_fmr_check_page_list(e_fmr, page_list, list_len);
875 	if (ret)
876 		goto map_phys_fmr_exit0;
877 	if (iova % e_fmr->fmr_page_size) {
878 		/* only whole-numbered pages */
879 		ehca_err(fmr->device, "bad iova, iova=%llx fmr_page_size=%x",
880 			 iova, e_fmr->fmr_page_size);
881 		ret = -EINVAL;
882 		goto map_phys_fmr_exit0;
883 	}
884 	if (e_fmr->fmr_map_cnt >= e_fmr->fmr_max_maps) {
885 		/* HCAD does not limit the maps, however trace this anyway */
886 		ehca_info(fmr->device, "map limit exceeded, fmr=%p "
887 			  "e_fmr->fmr_map_cnt=%x e_fmr->fmr_max_maps=%x",
888 			  fmr, e_fmr->fmr_map_cnt, e_fmr->fmr_max_maps);
889 	}
890 
891 	memset(&pginfo, 0, sizeof(pginfo));
892 	pginfo.type = EHCA_MR_PGI_FMR;
893 	pginfo.num_kpages = list_len;
894 	pginfo.hwpage_size = e_fmr->hwpage_size;
895 	pginfo.num_hwpages =
896 		list_len * e_fmr->fmr_page_size / pginfo.hwpage_size;
897 	pginfo.u.fmr.page_list = page_list;
898 	pginfo.next_hwpage =
899 		(iova & (e_fmr->fmr_page_size-1)) / pginfo.hwpage_size;
900 	pginfo.u.fmr.fmr_pgsize = e_fmr->fmr_page_size;
901 
902 	ret = ehca_rereg_mr(shca, e_fmr, (u64 *)iova,
903 			    list_len * e_fmr->fmr_page_size,
904 			    e_fmr->acl, e_pd, &pginfo, &tmp_lkey, &tmp_rkey);
905 	if (ret)
906 		goto map_phys_fmr_exit0;
907 
908 	/* successful reregistration */
909 	e_fmr->fmr_map_cnt++;
910 	e_fmr->ib.ib_fmr.lkey = tmp_lkey;
911 	e_fmr->ib.ib_fmr.rkey = tmp_rkey;
912 	return 0;
913 
914 map_phys_fmr_exit0:
915 	if (ret)
916 		ehca_err(fmr->device, "ret=%i fmr=%p page_list=%p list_len=%x "
917 			 "iova=%llx", ret, fmr, page_list, list_len, iova);
918 	return ret;
919 } /* end ehca_map_phys_fmr() */
920 
921 /*----------------------------------------------------------------------*/
922 
ehca_unmap_fmr(struct list_head * fmr_list)923 int ehca_unmap_fmr(struct list_head *fmr_list)
924 {
925 	int ret = 0;
926 	struct ib_fmr *ib_fmr;
927 	struct ehca_shca *shca = NULL;
928 	struct ehca_shca *prev_shca;
929 	struct ehca_mr *e_fmr;
930 	u32 num_fmr = 0;
931 	u32 unmap_fmr_cnt = 0;
932 
933 	/* check all FMR belong to same SHCA, and check internal flag */
934 	list_for_each_entry(ib_fmr, fmr_list, list) {
935 		prev_shca = shca;
936 		shca = container_of(ib_fmr->device, struct ehca_shca,
937 				    ib_device);
938 		e_fmr = container_of(ib_fmr, struct ehca_mr, ib.ib_fmr);
939 		if ((shca != prev_shca) && prev_shca) {
940 			ehca_err(&shca->ib_device, "SHCA mismatch, shca=%p "
941 				 "prev_shca=%p e_fmr=%p",
942 				 shca, prev_shca, e_fmr);
943 			ret = -EINVAL;
944 			goto unmap_fmr_exit0;
945 		}
946 		if (!(e_fmr->flags & EHCA_MR_FLAG_FMR)) {
947 			ehca_err(&shca->ib_device, "not a FMR, e_fmr=%p "
948 				 "e_fmr->flags=%x", e_fmr, e_fmr->flags);
949 			ret = -EINVAL;
950 			goto unmap_fmr_exit0;
951 		}
952 		num_fmr++;
953 	}
954 
955 	/* loop over all FMRs to unmap */
956 	list_for_each_entry(ib_fmr, fmr_list, list) {
957 		unmap_fmr_cnt++;
958 		e_fmr = container_of(ib_fmr, struct ehca_mr, ib.ib_fmr);
959 		shca = container_of(ib_fmr->device, struct ehca_shca,
960 				    ib_device);
961 		ret = ehca_unmap_one_fmr(shca, e_fmr);
962 		if (ret) {
963 			/* unmap failed, stop unmapping of rest of FMRs */
964 			ehca_err(&shca->ib_device, "unmap of one FMR failed, "
965 				 "stop rest, e_fmr=%p num_fmr=%x "
966 				 "unmap_fmr_cnt=%x lkey=%x", e_fmr, num_fmr,
967 				 unmap_fmr_cnt, e_fmr->ib.ib_fmr.lkey);
968 			goto unmap_fmr_exit0;
969 		}
970 	}
971 
972 unmap_fmr_exit0:
973 	if (ret)
974 		ehca_gen_err("ret=%i fmr_list=%p num_fmr=%x unmap_fmr_cnt=%x",
975 			     ret, fmr_list, num_fmr, unmap_fmr_cnt);
976 	return ret;
977 } /* end ehca_unmap_fmr() */
978 
979 /*----------------------------------------------------------------------*/
980 
ehca_dealloc_fmr(struct ib_fmr * fmr)981 int ehca_dealloc_fmr(struct ib_fmr *fmr)
982 {
983 	int ret;
984 	u64 h_ret;
985 	struct ehca_shca *shca =
986 		container_of(fmr->device, struct ehca_shca, ib_device);
987 	struct ehca_mr *e_fmr = container_of(fmr, struct ehca_mr, ib.ib_fmr);
988 
989 	if (!(e_fmr->flags & EHCA_MR_FLAG_FMR)) {
990 		ehca_err(fmr->device, "not a FMR, e_fmr=%p e_fmr->flags=%x",
991 			 e_fmr, e_fmr->flags);
992 		ret = -EINVAL;
993 		goto free_fmr_exit0;
994 	}
995 
996 	h_ret = hipz_h_free_resource_mr(shca->ipz_hca_handle, e_fmr);
997 	if (h_ret != H_SUCCESS) {
998 		ehca_err(fmr->device, "hipz_free_mr failed, h_ret=%lli e_fmr=%p "
999 			 "hca_hndl=%llx fmr_hndl=%llx fmr->lkey=%x",
1000 			 h_ret, e_fmr, shca->ipz_hca_handle.handle,
1001 			 e_fmr->ipz_mr_handle.handle, fmr->lkey);
1002 		ret = ehca2ib_return_code(h_ret);
1003 		goto free_fmr_exit0;
1004 	}
1005 	/* successful deregistration */
1006 	ehca_mr_delete(e_fmr);
1007 	return 0;
1008 
1009 free_fmr_exit0:
1010 	if (ret)
1011 		ehca_err(&shca->ib_device, "ret=%i fmr=%p", ret, fmr);
1012 	return ret;
1013 } /* end ehca_dealloc_fmr() */
1014 
1015 /*----------------------------------------------------------------------*/
1016 
1017 static int ehca_reg_bmap_mr_rpages(struct ehca_shca *shca,
1018 				   struct ehca_mr *e_mr,
1019 				   struct ehca_mr_pginfo *pginfo);
1020 
ehca_reg_mr(struct ehca_shca * shca,struct ehca_mr * e_mr,u64 * iova_start,u64 size,int acl,struct ehca_pd * e_pd,struct ehca_mr_pginfo * pginfo,u32 * lkey,u32 * rkey,enum ehca_reg_type reg_type)1021 int ehca_reg_mr(struct ehca_shca *shca,
1022 		struct ehca_mr *e_mr,
1023 		u64 *iova_start,
1024 		u64 size,
1025 		int acl,
1026 		struct ehca_pd *e_pd,
1027 		struct ehca_mr_pginfo *pginfo,
1028 		u32 *lkey, /*OUT*/
1029 		u32 *rkey, /*OUT*/
1030 		enum ehca_reg_type reg_type)
1031 {
1032 	int ret;
1033 	u64 h_ret;
1034 	u32 hipz_acl;
1035 	struct ehca_mr_hipzout_parms hipzout;
1036 
1037 	ehca_mrmw_map_acl(acl, &hipz_acl);
1038 	ehca_mrmw_set_pgsize_hipz_acl(pginfo->hwpage_size, &hipz_acl);
1039 	if (ehca_use_hp_mr == 1)
1040 		hipz_acl |= 0x00000001;
1041 
1042 	h_ret = hipz_h_alloc_resource_mr(shca->ipz_hca_handle, e_mr,
1043 					 (u64)iova_start, size, hipz_acl,
1044 					 e_pd->fw_pd, &hipzout);
1045 	if (h_ret != H_SUCCESS) {
1046 		ehca_err(&shca->ib_device, "hipz_alloc_mr failed, h_ret=%lli "
1047 			 "hca_hndl=%llx", h_ret, shca->ipz_hca_handle.handle);
1048 		ret = ehca2ib_return_code(h_ret);
1049 		goto ehca_reg_mr_exit0;
1050 	}
1051 
1052 	e_mr->ipz_mr_handle = hipzout.handle;
1053 
1054 	if (reg_type == EHCA_REG_BUSMAP_MR)
1055 		ret = ehca_reg_bmap_mr_rpages(shca, e_mr, pginfo);
1056 	else if (reg_type == EHCA_REG_MR)
1057 		ret = ehca_reg_mr_rpages(shca, e_mr, pginfo);
1058 	else
1059 		ret = -EINVAL;
1060 
1061 	if (ret)
1062 		goto ehca_reg_mr_exit1;
1063 
1064 	/* successful registration */
1065 	e_mr->num_kpages = pginfo->num_kpages;
1066 	e_mr->num_hwpages = pginfo->num_hwpages;
1067 	e_mr->hwpage_size = pginfo->hwpage_size;
1068 	e_mr->start = iova_start;
1069 	e_mr->size = size;
1070 	e_mr->acl = acl;
1071 	*lkey = hipzout.lkey;
1072 	*rkey = hipzout.rkey;
1073 	return 0;
1074 
1075 ehca_reg_mr_exit1:
1076 	h_ret = hipz_h_free_resource_mr(shca->ipz_hca_handle, e_mr);
1077 	if (h_ret != H_SUCCESS) {
1078 		ehca_err(&shca->ib_device, "h_ret=%lli shca=%p e_mr=%p "
1079 			 "iova_start=%p size=%llx acl=%x e_pd=%p lkey=%x "
1080 			 "pginfo=%p num_kpages=%llx num_hwpages=%llx ret=%i",
1081 			 h_ret, shca, e_mr, iova_start, size, acl, e_pd,
1082 			 hipzout.lkey, pginfo, pginfo->num_kpages,
1083 			 pginfo->num_hwpages, ret);
1084 		ehca_err(&shca->ib_device, "internal error in ehca_reg_mr, "
1085 			 "not recoverable");
1086 	}
1087 ehca_reg_mr_exit0:
1088 	if (ret)
1089 		ehca_err(&shca->ib_device, "ret=%i shca=%p e_mr=%p "
1090 			 "iova_start=%p size=%llx acl=%x e_pd=%p pginfo=%p "
1091 			 "num_kpages=%llx num_hwpages=%llx",
1092 			 ret, shca, e_mr, iova_start, size, acl, e_pd, pginfo,
1093 			 pginfo->num_kpages, pginfo->num_hwpages);
1094 	return ret;
1095 } /* end ehca_reg_mr() */
1096 
1097 /*----------------------------------------------------------------------*/
1098 
ehca_reg_mr_rpages(struct ehca_shca * shca,struct ehca_mr * e_mr,struct ehca_mr_pginfo * pginfo)1099 int ehca_reg_mr_rpages(struct ehca_shca *shca,
1100 		       struct ehca_mr *e_mr,
1101 		       struct ehca_mr_pginfo *pginfo)
1102 {
1103 	int ret = 0;
1104 	u64 h_ret;
1105 	u32 rnum;
1106 	u64 rpage;
1107 	u32 i;
1108 	u64 *kpage;
1109 
1110 	if (!pginfo->num_hwpages) /* in case of fmr */
1111 		return 0;
1112 
1113 	kpage = ehca_alloc_fw_ctrlblock(GFP_KERNEL);
1114 	if (!kpage) {
1115 		ehca_err(&shca->ib_device, "kpage alloc failed");
1116 		ret = -ENOMEM;
1117 		goto ehca_reg_mr_rpages_exit0;
1118 	}
1119 
1120 	/* max MAX_RPAGES ehca mr pages per register call */
1121 	for (i = 0; i < NUM_CHUNKS(pginfo->num_hwpages, MAX_RPAGES); i++) {
1122 
1123 		if (i == NUM_CHUNKS(pginfo->num_hwpages, MAX_RPAGES) - 1) {
1124 			rnum = pginfo->num_hwpages % MAX_RPAGES; /* last shot */
1125 			if (rnum == 0)
1126 				rnum = MAX_RPAGES;      /* last shot is full */
1127 		} else
1128 			rnum = MAX_RPAGES;
1129 
1130 		ret = ehca_set_pagebuf(pginfo, rnum, kpage);
1131 		if (ret) {
1132 			ehca_err(&shca->ib_device, "ehca_set_pagebuf "
1133 				 "bad rc, ret=%i rnum=%x kpage=%p",
1134 				 ret, rnum, kpage);
1135 			goto ehca_reg_mr_rpages_exit1;
1136 		}
1137 
1138 		if (rnum > 1) {
1139 			rpage = virt_to_abs(kpage);
1140 			if (!rpage) {
1141 				ehca_err(&shca->ib_device, "kpage=%p i=%x",
1142 					 kpage, i);
1143 				ret = -EFAULT;
1144 				goto ehca_reg_mr_rpages_exit1;
1145 			}
1146 		} else
1147 			rpage = *kpage;
1148 
1149 		h_ret = hipz_h_register_rpage_mr(
1150 			shca->ipz_hca_handle, e_mr,
1151 			ehca_encode_hwpage_size(pginfo->hwpage_size),
1152 			0, rpage, rnum);
1153 
1154 		if (i == NUM_CHUNKS(pginfo->num_hwpages, MAX_RPAGES) - 1) {
1155 			/*
1156 			 * check for 'registration complete'==H_SUCCESS
1157 			 * and for 'page registered'==H_PAGE_REGISTERED
1158 			 */
1159 			if (h_ret != H_SUCCESS) {
1160 				ehca_err(&shca->ib_device, "last "
1161 					 "hipz_reg_rpage_mr failed, h_ret=%lli "
1162 					 "e_mr=%p i=%x hca_hndl=%llx mr_hndl=%llx"
1163 					 " lkey=%x", h_ret, e_mr, i,
1164 					 shca->ipz_hca_handle.handle,
1165 					 e_mr->ipz_mr_handle.handle,
1166 					 e_mr->ib.ib_mr.lkey);
1167 				ret = ehca2ib_return_code(h_ret);
1168 				break;
1169 			} else
1170 				ret = 0;
1171 		} else if (h_ret != H_PAGE_REGISTERED) {
1172 			ehca_err(&shca->ib_device, "hipz_reg_rpage_mr failed, "
1173 				 "h_ret=%lli e_mr=%p i=%x lkey=%x hca_hndl=%llx "
1174 				 "mr_hndl=%llx", h_ret, e_mr, i,
1175 				 e_mr->ib.ib_mr.lkey,
1176 				 shca->ipz_hca_handle.handle,
1177 				 e_mr->ipz_mr_handle.handle);
1178 			ret = ehca2ib_return_code(h_ret);
1179 			break;
1180 		} else
1181 			ret = 0;
1182 	} /* end for(i) */
1183 
1184 
1185 ehca_reg_mr_rpages_exit1:
1186 	ehca_free_fw_ctrlblock(kpage);
1187 ehca_reg_mr_rpages_exit0:
1188 	if (ret)
1189 		ehca_err(&shca->ib_device, "ret=%i shca=%p e_mr=%p pginfo=%p "
1190 			 "num_kpages=%llx num_hwpages=%llx", ret, shca, e_mr,
1191 			 pginfo, pginfo->num_kpages, pginfo->num_hwpages);
1192 	return ret;
1193 } /* end ehca_reg_mr_rpages() */
1194 
1195 /*----------------------------------------------------------------------*/
1196 
ehca_rereg_mr_rereg1(struct ehca_shca * shca,struct ehca_mr * e_mr,u64 * iova_start,u64 size,u32 acl,struct ehca_pd * e_pd,struct ehca_mr_pginfo * pginfo,u32 * lkey,u32 * rkey)1197 inline int ehca_rereg_mr_rereg1(struct ehca_shca *shca,
1198 				struct ehca_mr *e_mr,
1199 				u64 *iova_start,
1200 				u64 size,
1201 				u32 acl,
1202 				struct ehca_pd *e_pd,
1203 				struct ehca_mr_pginfo *pginfo,
1204 				u32 *lkey, /*OUT*/
1205 				u32 *rkey) /*OUT*/
1206 {
1207 	int ret;
1208 	u64 h_ret;
1209 	u32 hipz_acl;
1210 	u64 *kpage;
1211 	u64 rpage;
1212 	struct ehca_mr_pginfo pginfo_save;
1213 	struct ehca_mr_hipzout_parms hipzout;
1214 
1215 	ehca_mrmw_map_acl(acl, &hipz_acl);
1216 	ehca_mrmw_set_pgsize_hipz_acl(pginfo->hwpage_size, &hipz_acl);
1217 
1218 	kpage = ehca_alloc_fw_ctrlblock(GFP_KERNEL);
1219 	if (!kpage) {
1220 		ehca_err(&shca->ib_device, "kpage alloc failed");
1221 		ret = -ENOMEM;
1222 		goto ehca_rereg_mr_rereg1_exit0;
1223 	}
1224 
1225 	pginfo_save = *pginfo;
1226 	ret = ehca_set_pagebuf(pginfo, pginfo->num_hwpages, kpage);
1227 	if (ret) {
1228 		ehca_err(&shca->ib_device, "set pagebuf failed, e_mr=%p "
1229 			 "pginfo=%p type=%x num_kpages=%llx num_hwpages=%llx "
1230 			 "kpage=%p", e_mr, pginfo, pginfo->type,
1231 			 pginfo->num_kpages, pginfo->num_hwpages, kpage);
1232 		goto ehca_rereg_mr_rereg1_exit1;
1233 	}
1234 	rpage = virt_to_abs(kpage);
1235 	if (!rpage) {
1236 		ehca_err(&shca->ib_device, "kpage=%p", kpage);
1237 		ret = -EFAULT;
1238 		goto ehca_rereg_mr_rereg1_exit1;
1239 	}
1240 	h_ret = hipz_h_reregister_pmr(shca->ipz_hca_handle, e_mr,
1241 				      (u64)iova_start, size, hipz_acl,
1242 				      e_pd->fw_pd, rpage, &hipzout);
1243 	if (h_ret != H_SUCCESS) {
1244 		/*
1245 		 * reregistration unsuccessful, try it again with the 3 hCalls,
1246 		 * e.g. this is required in case H_MR_CONDITION
1247 		 * (MW bound or MR is shared)
1248 		 */
1249 		ehca_warn(&shca->ib_device, "hipz_h_reregister_pmr failed "
1250 			  "(Rereg1), h_ret=%lli e_mr=%p", h_ret, e_mr);
1251 		*pginfo = pginfo_save;
1252 		ret = -EAGAIN;
1253 	} else if ((u64 *)hipzout.vaddr != iova_start) {
1254 		ehca_err(&shca->ib_device, "PHYP changed iova_start in "
1255 			 "rereg_pmr, iova_start=%p iova_start_out=%llx e_mr=%p "
1256 			 "mr_handle=%llx lkey=%x lkey_out=%x", iova_start,
1257 			 hipzout.vaddr, e_mr, e_mr->ipz_mr_handle.handle,
1258 			 e_mr->ib.ib_mr.lkey, hipzout.lkey);
1259 		ret = -EFAULT;
1260 	} else {
1261 		/*
1262 		 * successful reregistration
1263 		 * note: start and start_out are identical for eServer HCAs
1264 		 */
1265 		e_mr->num_kpages = pginfo->num_kpages;
1266 		e_mr->num_hwpages = pginfo->num_hwpages;
1267 		e_mr->hwpage_size = pginfo->hwpage_size;
1268 		e_mr->start = iova_start;
1269 		e_mr->size = size;
1270 		e_mr->acl = acl;
1271 		*lkey = hipzout.lkey;
1272 		*rkey = hipzout.rkey;
1273 	}
1274 
1275 ehca_rereg_mr_rereg1_exit1:
1276 	ehca_free_fw_ctrlblock(kpage);
1277 ehca_rereg_mr_rereg1_exit0:
1278 	if ( ret && (ret != -EAGAIN) )
1279 		ehca_err(&shca->ib_device, "ret=%i lkey=%x rkey=%x "
1280 			 "pginfo=%p num_kpages=%llx num_hwpages=%llx",
1281 			 ret, *lkey, *rkey, pginfo, pginfo->num_kpages,
1282 			 pginfo->num_hwpages);
1283 	return ret;
1284 } /* end ehca_rereg_mr_rereg1() */
1285 
1286 /*----------------------------------------------------------------------*/
1287 
ehca_rereg_mr(struct ehca_shca * shca,struct ehca_mr * e_mr,u64 * iova_start,u64 size,int acl,struct ehca_pd * e_pd,struct ehca_mr_pginfo * pginfo,u32 * lkey,u32 * rkey)1288 int ehca_rereg_mr(struct ehca_shca *shca,
1289 		  struct ehca_mr *e_mr,
1290 		  u64 *iova_start,
1291 		  u64 size,
1292 		  int acl,
1293 		  struct ehca_pd *e_pd,
1294 		  struct ehca_mr_pginfo *pginfo,
1295 		  u32 *lkey,
1296 		  u32 *rkey)
1297 {
1298 	int ret = 0;
1299 	u64 h_ret;
1300 	int rereg_1_hcall = 1; /* 1: use hipz_h_reregister_pmr directly */
1301 	int rereg_3_hcall = 0; /* 1: use 3 hipz calls for reregistration */
1302 
1303 	/* first determine reregistration hCall(s) */
1304 	if ((pginfo->num_hwpages > MAX_RPAGES) ||
1305 	    (e_mr->num_hwpages > MAX_RPAGES) ||
1306 	    (pginfo->num_hwpages > e_mr->num_hwpages)) {
1307 		ehca_dbg(&shca->ib_device, "Rereg3 case, "
1308 			 "pginfo->num_hwpages=%llx e_mr->num_hwpages=%x",
1309 			 pginfo->num_hwpages, e_mr->num_hwpages);
1310 		rereg_1_hcall = 0;
1311 		rereg_3_hcall = 1;
1312 	}
1313 
1314 	if (e_mr->flags & EHCA_MR_FLAG_MAXMR) {	/* check for max-MR */
1315 		rereg_1_hcall = 0;
1316 		rereg_3_hcall = 1;
1317 		e_mr->flags &= ~EHCA_MR_FLAG_MAXMR;
1318 		ehca_err(&shca->ib_device, "Rereg MR for max-MR! e_mr=%p",
1319 			 e_mr);
1320 	}
1321 
1322 	if (rereg_1_hcall) {
1323 		ret = ehca_rereg_mr_rereg1(shca, e_mr, iova_start, size,
1324 					   acl, e_pd, pginfo, lkey, rkey);
1325 		if (ret) {
1326 			if (ret == -EAGAIN)
1327 				rereg_3_hcall = 1;
1328 			else
1329 				goto ehca_rereg_mr_exit0;
1330 		}
1331 	}
1332 
1333 	if (rereg_3_hcall) {
1334 		struct ehca_mr save_mr;
1335 
1336 		/* first deregister old MR */
1337 		h_ret = hipz_h_free_resource_mr(shca->ipz_hca_handle, e_mr);
1338 		if (h_ret != H_SUCCESS) {
1339 			ehca_err(&shca->ib_device, "hipz_free_mr failed, "
1340 				 "h_ret=%lli e_mr=%p hca_hndl=%llx mr_hndl=%llx "
1341 				 "mr->lkey=%x",
1342 				 h_ret, e_mr, shca->ipz_hca_handle.handle,
1343 				 e_mr->ipz_mr_handle.handle,
1344 				 e_mr->ib.ib_mr.lkey);
1345 			ret = ehca2ib_return_code(h_ret);
1346 			goto ehca_rereg_mr_exit0;
1347 		}
1348 		/* clean ehca_mr_t, without changing struct ib_mr and lock */
1349 		save_mr = *e_mr;
1350 		ehca_mr_deletenew(e_mr);
1351 
1352 		/* set some MR values */
1353 		e_mr->flags = save_mr.flags;
1354 		e_mr->hwpage_size = save_mr.hwpage_size;
1355 		e_mr->fmr_page_size = save_mr.fmr_page_size;
1356 		e_mr->fmr_max_pages = save_mr.fmr_max_pages;
1357 		e_mr->fmr_max_maps = save_mr.fmr_max_maps;
1358 		e_mr->fmr_map_cnt = save_mr.fmr_map_cnt;
1359 
1360 		ret = ehca_reg_mr(shca, e_mr, iova_start, size, acl,
1361 				  e_pd, pginfo, lkey, rkey, EHCA_REG_MR);
1362 		if (ret) {
1363 			u32 offset = (u64)(&e_mr->flags) - (u64)e_mr;
1364 			memcpy(&e_mr->flags, &(save_mr.flags),
1365 			       sizeof(struct ehca_mr) - offset);
1366 			goto ehca_rereg_mr_exit0;
1367 		}
1368 	}
1369 
1370 ehca_rereg_mr_exit0:
1371 	if (ret)
1372 		ehca_err(&shca->ib_device, "ret=%i shca=%p e_mr=%p "
1373 			 "iova_start=%p size=%llx acl=%x e_pd=%p pginfo=%p "
1374 			 "num_kpages=%llx lkey=%x rkey=%x rereg_1_hcall=%x "
1375 			 "rereg_3_hcall=%x", ret, shca, e_mr, iova_start, size,
1376 			 acl, e_pd, pginfo, pginfo->num_kpages, *lkey, *rkey,
1377 			 rereg_1_hcall, rereg_3_hcall);
1378 	return ret;
1379 } /* end ehca_rereg_mr() */
1380 
1381 /*----------------------------------------------------------------------*/
1382 
ehca_unmap_one_fmr(struct ehca_shca * shca,struct ehca_mr * e_fmr)1383 int ehca_unmap_one_fmr(struct ehca_shca *shca,
1384 		       struct ehca_mr *e_fmr)
1385 {
1386 	int ret = 0;
1387 	u64 h_ret;
1388 	struct ehca_pd *e_pd =
1389 		container_of(e_fmr->ib.ib_fmr.pd, struct ehca_pd, ib_pd);
1390 	struct ehca_mr save_fmr;
1391 	u32 tmp_lkey, tmp_rkey;
1392 	struct ehca_mr_pginfo pginfo;
1393 	struct ehca_mr_hipzout_parms hipzout;
1394 	struct ehca_mr save_mr;
1395 
1396 	if (e_fmr->fmr_max_pages <= MAX_RPAGES) {
1397 		/*
1398 		 * note: after using rereg hcall with len=0,
1399 		 * rereg hcall must be used again for registering pages
1400 		 */
1401 		h_ret = hipz_h_reregister_pmr(shca->ipz_hca_handle, e_fmr, 0,
1402 					      0, 0, e_pd->fw_pd, 0, &hipzout);
1403 		if (h_ret == H_SUCCESS) {
1404 			/* successful reregistration */
1405 			e_fmr->start = NULL;
1406 			e_fmr->size = 0;
1407 			tmp_lkey = hipzout.lkey;
1408 			tmp_rkey = hipzout.rkey;
1409 			return 0;
1410 		}
1411 		/*
1412 		 * should not happen, because length checked above,
1413 		 * FMRs are not shared and no MW bound to FMRs
1414 		 */
1415 		ehca_err(&shca->ib_device, "hipz_reregister_pmr failed "
1416 			 "(Rereg1), h_ret=%lli e_fmr=%p hca_hndl=%llx "
1417 			 "mr_hndl=%llx lkey=%x lkey_out=%x",
1418 			 h_ret, e_fmr, shca->ipz_hca_handle.handle,
1419 			 e_fmr->ipz_mr_handle.handle,
1420 			 e_fmr->ib.ib_fmr.lkey, hipzout.lkey);
1421 		/* try free and rereg */
1422 	}
1423 
1424 	/* first free old FMR */
1425 	h_ret = hipz_h_free_resource_mr(shca->ipz_hca_handle, e_fmr);
1426 	if (h_ret != H_SUCCESS) {
1427 		ehca_err(&shca->ib_device, "hipz_free_mr failed, "
1428 			 "h_ret=%lli e_fmr=%p hca_hndl=%llx mr_hndl=%llx "
1429 			 "lkey=%x",
1430 			 h_ret, e_fmr, shca->ipz_hca_handle.handle,
1431 			 e_fmr->ipz_mr_handle.handle,
1432 			 e_fmr->ib.ib_fmr.lkey);
1433 		ret = ehca2ib_return_code(h_ret);
1434 		goto ehca_unmap_one_fmr_exit0;
1435 	}
1436 	/* clean ehca_mr_t, without changing lock */
1437 	save_fmr = *e_fmr;
1438 	ehca_mr_deletenew(e_fmr);
1439 
1440 	/* set some MR values */
1441 	e_fmr->flags = save_fmr.flags;
1442 	e_fmr->hwpage_size = save_fmr.hwpage_size;
1443 	e_fmr->fmr_page_size = save_fmr.fmr_page_size;
1444 	e_fmr->fmr_max_pages = save_fmr.fmr_max_pages;
1445 	e_fmr->fmr_max_maps = save_fmr.fmr_max_maps;
1446 	e_fmr->fmr_map_cnt = save_fmr.fmr_map_cnt;
1447 	e_fmr->acl = save_fmr.acl;
1448 
1449 	memset(&pginfo, 0, sizeof(pginfo));
1450 	pginfo.type = EHCA_MR_PGI_FMR;
1451 	ret = ehca_reg_mr(shca, e_fmr, NULL,
1452 			  (e_fmr->fmr_max_pages * e_fmr->fmr_page_size),
1453 			  e_fmr->acl, e_pd, &pginfo, &tmp_lkey,
1454 			  &tmp_rkey, EHCA_REG_MR);
1455 	if (ret) {
1456 		u32 offset = (u64)(&e_fmr->flags) - (u64)e_fmr;
1457 		memcpy(&e_fmr->flags, &(save_mr.flags),
1458 		       sizeof(struct ehca_mr) - offset);
1459 	}
1460 
1461 ehca_unmap_one_fmr_exit0:
1462 	if (ret)
1463 		ehca_err(&shca->ib_device, "ret=%i tmp_lkey=%x tmp_rkey=%x "
1464 			 "fmr_max_pages=%x",
1465 			 ret, tmp_lkey, tmp_rkey, e_fmr->fmr_max_pages);
1466 	return ret;
1467 } /* end ehca_unmap_one_fmr() */
1468 
1469 /*----------------------------------------------------------------------*/
1470 
ehca_reg_smr(struct ehca_shca * shca,struct ehca_mr * e_origmr,struct ehca_mr * e_newmr,u64 * iova_start,int acl,struct ehca_pd * e_pd,u32 * lkey,u32 * rkey)1471 int ehca_reg_smr(struct ehca_shca *shca,
1472 		 struct ehca_mr *e_origmr,
1473 		 struct ehca_mr *e_newmr,
1474 		 u64 *iova_start,
1475 		 int acl,
1476 		 struct ehca_pd *e_pd,
1477 		 u32 *lkey, /*OUT*/
1478 		 u32 *rkey) /*OUT*/
1479 {
1480 	int ret = 0;
1481 	u64 h_ret;
1482 	u32 hipz_acl;
1483 	struct ehca_mr_hipzout_parms hipzout;
1484 
1485 	ehca_mrmw_map_acl(acl, &hipz_acl);
1486 	ehca_mrmw_set_pgsize_hipz_acl(e_origmr->hwpage_size, &hipz_acl);
1487 
1488 	h_ret = hipz_h_register_smr(shca->ipz_hca_handle, e_newmr, e_origmr,
1489 				    (u64)iova_start, hipz_acl, e_pd->fw_pd,
1490 				    &hipzout);
1491 	if (h_ret != H_SUCCESS) {
1492 		ehca_err(&shca->ib_device, "hipz_reg_smr failed, h_ret=%lli "
1493 			 "shca=%p e_origmr=%p e_newmr=%p iova_start=%p acl=%x "
1494 			 "e_pd=%p hca_hndl=%llx mr_hndl=%llx lkey=%x",
1495 			 h_ret, shca, e_origmr, e_newmr, iova_start, acl, e_pd,
1496 			 shca->ipz_hca_handle.handle,
1497 			 e_origmr->ipz_mr_handle.handle,
1498 			 e_origmr->ib.ib_mr.lkey);
1499 		ret = ehca2ib_return_code(h_ret);
1500 		goto ehca_reg_smr_exit0;
1501 	}
1502 	/* successful registration */
1503 	e_newmr->num_kpages = e_origmr->num_kpages;
1504 	e_newmr->num_hwpages = e_origmr->num_hwpages;
1505 	e_newmr->hwpage_size   = e_origmr->hwpage_size;
1506 	e_newmr->start = iova_start;
1507 	e_newmr->size = e_origmr->size;
1508 	e_newmr->acl = acl;
1509 	e_newmr->ipz_mr_handle = hipzout.handle;
1510 	*lkey = hipzout.lkey;
1511 	*rkey = hipzout.rkey;
1512 	return 0;
1513 
1514 ehca_reg_smr_exit0:
1515 	if (ret)
1516 		ehca_err(&shca->ib_device, "ret=%i shca=%p e_origmr=%p "
1517 			 "e_newmr=%p iova_start=%p acl=%x e_pd=%p",
1518 			 ret, shca, e_origmr, e_newmr, iova_start, acl, e_pd);
1519 	return ret;
1520 } /* end ehca_reg_smr() */
1521 
1522 /*----------------------------------------------------------------------*/
ehca_calc_sectbase(int top,int dir,int idx)1523 static inline void *ehca_calc_sectbase(int top, int dir, int idx)
1524 {
1525 	unsigned long ret = idx;
1526 	ret |= dir << EHCA_DIR_INDEX_SHIFT;
1527 	ret |= top << EHCA_TOP_INDEX_SHIFT;
1528 	return abs_to_virt(ret << SECTION_SIZE_BITS);
1529 }
1530 
1531 #define ehca_bmap_valid(entry) \
1532 	((u64)entry != (u64)EHCA_INVAL_ADDR)
1533 
ehca_reg_mr_section(int top,int dir,int idx,u64 * kpage,struct ehca_shca * shca,struct ehca_mr * mr,struct ehca_mr_pginfo * pginfo)1534 static u64 ehca_reg_mr_section(int top, int dir, int idx, u64 *kpage,
1535 			       struct ehca_shca *shca, struct ehca_mr *mr,
1536 			       struct ehca_mr_pginfo *pginfo)
1537 {
1538 	u64 h_ret = 0;
1539 	unsigned long page = 0;
1540 	u64 rpage = virt_to_abs(kpage);
1541 	int page_count;
1542 
1543 	void *sectbase = ehca_calc_sectbase(top, dir, idx);
1544 	if ((unsigned long)sectbase & (pginfo->hwpage_size - 1)) {
1545 		ehca_err(&shca->ib_device, "reg_mr_section will probably fail:"
1546 					   "hwpage_size does not fit to "
1547 					   "section start address");
1548 	}
1549 	page_count = EHCA_SECTSIZE / pginfo->hwpage_size;
1550 
1551 	while (page < page_count) {
1552 		u64 rnum;
1553 		for (rnum = 0; (rnum < MAX_RPAGES) && (page < page_count);
1554 		     rnum++) {
1555 			void *pg = sectbase + ((page++) * pginfo->hwpage_size);
1556 			kpage[rnum] = virt_to_abs(pg);
1557 		}
1558 
1559 		h_ret = hipz_h_register_rpage_mr(shca->ipz_hca_handle, mr,
1560 			ehca_encode_hwpage_size(pginfo->hwpage_size),
1561 			0, rpage, rnum);
1562 
1563 		if ((h_ret != H_SUCCESS) && (h_ret != H_PAGE_REGISTERED)) {
1564 			ehca_err(&shca->ib_device, "register_rpage_mr failed");
1565 			return h_ret;
1566 		}
1567 	}
1568 	return h_ret;
1569 }
1570 
ehca_reg_mr_sections(int top,int dir,u64 * kpage,struct ehca_shca * shca,struct ehca_mr * mr,struct ehca_mr_pginfo * pginfo)1571 static u64 ehca_reg_mr_sections(int top, int dir, u64 *kpage,
1572 				struct ehca_shca *shca, struct ehca_mr *mr,
1573 				struct ehca_mr_pginfo *pginfo)
1574 {
1575 	u64 hret = H_SUCCESS;
1576 	int idx;
1577 
1578 	for (idx = 0; idx < EHCA_MAP_ENTRIES; idx++) {
1579 		if (!ehca_bmap_valid(ehca_bmap->top[top]->dir[dir]->ent[idx]))
1580 			continue;
1581 
1582 		hret = ehca_reg_mr_section(top, dir, idx, kpage, shca, mr,
1583 					   pginfo);
1584 		if ((hret != H_SUCCESS) && (hret != H_PAGE_REGISTERED))
1585 				return hret;
1586 	}
1587 	return hret;
1588 }
1589 
ehca_reg_mr_dir_sections(int top,u64 * kpage,struct ehca_shca * shca,struct ehca_mr * mr,struct ehca_mr_pginfo * pginfo)1590 static u64 ehca_reg_mr_dir_sections(int top, u64 *kpage, struct ehca_shca *shca,
1591 				    struct ehca_mr *mr,
1592 				    struct ehca_mr_pginfo *pginfo)
1593 {
1594 	u64 hret = H_SUCCESS;
1595 	int dir;
1596 
1597 	for (dir = 0; dir < EHCA_MAP_ENTRIES; dir++) {
1598 		if (!ehca_bmap_valid(ehca_bmap->top[top]->dir[dir]))
1599 			continue;
1600 
1601 		hret = ehca_reg_mr_sections(top, dir, kpage, shca, mr, pginfo);
1602 		if ((hret != H_SUCCESS) && (hret != H_PAGE_REGISTERED))
1603 				return hret;
1604 	}
1605 	return hret;
1606 }
1607 
1608 /* register internal max-MR to internal SHCA */
ehca_reg_internal_maxmr(struct ehca_shca * shca,struct ehca_pd * e_pd,struct ehca_mr ** e_maxmr)1609 int ehca_reg_internal_maxmr(
1610 	struct ehca_shca *shca,
1611 	struct ehca_pd *e_pd,
1612 	struct ehca_mr **e_maxmr)  /*OUT*/
1613 {
1614 	int ret;
1615 	struct ehca_mr *e_mr;
1616 	u64 *iova_start;
1617 	u64 size_maxmr;
1618 	struct ehca_mr_pginfo pginfo;
1619 	struct ib_phys_buf ib_pbuf;
1620 	u32 num_kpages;
1621 	u32 num_hwpages;
1622 	u64 hw_pgsize;
1623 
1624 	if (!ehca_bmap) {
1625 		ret = -EFAULT;
1626 		goto ehca_reg_internal_maxmr_exit0;
1627 	}
1628 
1629 	e_mr = ehca_mr_new();
1630 	if (!e_mr) {
1631 		ehca_err(&shca->ib_device, "out of memory");
1632 		ret = -ENOMEM;
1633 		goto ehca_reg_internal_maxmr_exit0;
1634 	}
1635 	e_mr->flags |= EHCA_MR_FLAG_MAXMR;
1636 
1637 	/* register internal max-MR on HCA */
1638 	size_maxmr = ehca_mr_len;
1639 	iova_start = (u64 *)ehca_map_vaddr((void *)(KERNELBASE + PHYSICAL_START));
1640 	ib_pbuf.addr = 0;
1641 	ib_pbuf.size = size_maxmr;
1642 	num_kpages = NUM_CHUNKS(((u64)iova_start % PAGE_SIZE) + size_maxmr,
1643 				PAGE_SIZE);
1644 	hw_pgsize = ehca_get_max_hwpage_size(shca);
1645 	num_hwpages = NUM_CHUNKS(((u64)iova_start % hw_pgsize) + size_maxmr,
1646 				 hw_pgsize);
1647 
1648 	memset(&pginfo, 0, sizeof(pginfo));
1649 	pginfo.type = EHCA_MR_PGI_PHYS;
1650 	pginfo.num_kpages = num_kpages;
1651 	pginfo.num_hwpages = num_hwpages;
1652 	pginfo.hwpage_size = hw_pgsize;
1653 	pginfo.u.phy.num_phys_buf = 1;
1654 	pginfo.u.phy.phys_buf_array = &ib_pbuf;
1655 
1656 	ret = ehca_reg_mr(shca, e_mr, iova_start, size_maxmr, 0, e_pd,
1657 			  &pginfo, &e_mr->ib.ib_mr.lkey,
1658 			  &e_mr->ib.ib_mr.rkey, EHCA_REG_BUSMAP_MR);
1659 	if (ret) {
1660 		ehca_err(&shca->ib_device, "reg of internal max MR failed, "
1661 			 "e_mr=%p iova_start=%p size_maxmr=%llx num_kpages=%x "
1662 			 "num_hwpages=%x", e_mr, iova_start, size_maxmr,
1663 			 num_kpages, num_hwpages);
1664 		goto ehca_reg_internal_maxmr_exit1;
1665 	}
1666 
1667 	/* successful registration of all pages */
1668 	e_mr->ib.ib_mr.device = e_pd->ib_pd.device;
1669 	e_mr->ib.ib_mr.pd = &e_pd->ib_pd;
1670 	e_mr->ib.ib_mr.uobject = NULL;
1671 	atomic_inc(&(e_pd->ib_pd.usecnt));
1672 	atomic_set(&(e_mr->ib.ib_mr.usecnt), 0);
1673 	*e_maxmr = e_mr;
1674 	return 0;
1675 
1676 ehca_reg_internal_maxmr_exit1:
1677 	ehca_mr_delete(e_mr);
1678 ehca_reg_internal_maxmr_exit0:
1679 	if (ret)
1680 		ehca_err(&shca->ib_device, "ret=%i shca=%p e_pd=%p e_maxmr=%p",
1681 			 ret, shca, e_pd, e_maxmr);
1682 	return ret;
1683 } /* end ehca_reg_internal_maxmr() */
1684 
1685 /*----------------------------------------------------------------------*/
1686 
ehca_reg_maxmr(struct ehca_shca * shca,struct ehca_mr * e_newmr,u64 * iova_start,int acl,struct ehca_pd * e_pd,u32 * lkey,u32 * rkey)1687 int ehca_reg_maxmr(struct ehca_shca *shca,
1688 		   struct ehca_mr *e_newmr,
1689 		   u64 *iova_start,
1690 		   int acl,
1691 		   struct ehca_pd *e_pd,
1692 		   u32 *lkey,
1693 		   u32 *rkey)
1694 {
1695 	u64 h_ret;
1696 	struct ehca_mr *e_origmr = shca->maxmr;
1697 	u32 hipz_acl;
1698 	struct ehca_mr_hipzout_parms hipzout;
1699 
1700 	ehca_mrmw_map_acl(acl, &hipz_acl);
1701 	ehca_mrmw_set_pgsize_hipz_acl(e_origmr->hwpage_size, &hipz_acl);
1702 
1703 	h_ret = hipz_h_register_smr(shca->ipz_hca_handle, e_newmr, e_origmr,
1704 				    (u64)iova_start, hipz_acl, e_pd->fw_pd,
1705 				    &hipzout);
1706 	if (h_ret != H_SUCCESS) {
1707 		ehca_err(&shca->ib_device, "hipz_reg_smr failed, h_ret=%lli "
1708 			 "e_origmr=%p hca_hndl=%llx mr_hndl=%llx lkey=%x",
1709 			 h_ret, e_origmr, shca->ipz_hca_handle.handle,
1710 			 e_origmr->ipz_mr_handle.handle,
1711 			 e_origmr->ib.ib_mr.lkey);
1712 		return ehca2ib_return_code(h_ret);
1713 	}
1714 	/* successful registration */
1715 	e_newmr->num_kpages = e_origmr->num_kpages;
1716 	e_newmr->num_hwpages = e_origmr->num_hwpages;
1717 	e_newmr->hwpage_size = e_origmr->hwpage_size;
1718 	e_newmr->start = iova_start;
1719 	e_newmr->size = e_origmr->size;
1720 	e_newmr->acl = acl;
1721 	e_newmr->ipz_mr_handle = hipzout.handle;
1722 	*lkey = hipzout.lkey;
1723 	*rkey = hipzout.rkey;
1724 	return 0;
1725 } /* end ehca_reg_maxmr() */
1726 
1727 /*----------------------------------------------------------------------*/
1728 
ehca_dereg_internal_maxmr(struct ehca_shca * shca)1729 int ehca_dereg_internal_maxmr(struct ehca_shca *shca)
1730 {
1731 	int ret;
1732 	struct ehca_mr *e_maxmr;
1733 	struct ib_pd *ib_pd;
1734 
1735 	if (!shca->maxmr) {
1736 		ehca_err(&shca->ib_device, "bad call, shca=%p", shca);
1737 		ret = -EINVAL;
1738 		goto ehca_dereg_internal_maxmr_exit0;
1739 	}
1740 
1741 	e_maxmr = shca->maxmr;
1742 	ib_pd = e_maxmr->ib.ib_mr.pd;
1743 	shca->maxmr = NULL; /* remove internal max-MR indication from SHCA */
1744 
1745 	ret = ehca_dereg_mr(&e_maxmr->ib.ib_mr);
1746 	if (ret) {
1747 		ehca_err(&shca->ib_device, "dereg internal max-MR failed, "
1748 			 "ret=%i e_maxmr=%p shca=%p lkey=%x",
1749 			 ret, e_maxmr, shca, e_maxmr->ib.ib_mr.lkey);
1750 		shca->maxmr = e_maxmr;
1751 		goto ehca_dereg_internal_maxmr_exit0;
1752 	}
1753 
1754 	atomic_dec(&ib_pd->usecnt);
1755 
1756 ehca_dereg_internal_maxmr_exit0:
1757 	if (ret)
1758 		ehca_err(&shca->ib_device, "ret=%i shca=%p shca->maxmr=%p",
1759 			 ret, shca, shca->maxmr);
1760 	return ret;
1761 } /* end ehca_dereg_internal_maxmr() */
1762 
1763 /*----------------------------------------------------------------------*/
1764 
1765 /*
1766  * check physical buffer array of MR verbs for validness and
1767  * calculates MR size
1768  */
ehca_mr_chk_buf_and_calc_size(struct ib_phys_buf * phys_buf_array,int num_phys_buf,u64 * iova_start,u64 * size)1769 int ehca_mr_chk_buf_and_calc_size(struct ib_phys_buf *phys_buf_array,
1770 				  int num_phys_buf,
1771 				  u64 *iova_start,
1772 				  u64 *size)
1773 {
1774 	struct ib_phys_buf *pbuf = phys_buf_array;
1775 	u64 size_count = 0;
1776 	u32 i;
1777 
1778 	if (num_phys_buf == 0) {
1779 		ehca_gen_err("bad phys buf array len, num_phys_buf=0");
1780 		return -EINVAL;
1781 	}
1782 	/* check first buffer */
1783 	if (((u64)iova_start & ~PAGE_MASK) != (pbuf->addr & ~PAGE_MASK)) {
1784 		ehca_gen_err("iova_start/addr mismatch, iova_start=%p "
1785 			     "pbuf->addr=%llx pbuf->size=%llx",
1786 			     iova_start, pbuf->addr, pbuf->size);
1787 		return -EINVAL;
1788 	}
1789 	if (((pbuf->addr + pbuf->size) % PAGE_SIZE) &&
1790 	    (num_phys_buf > 1)) {
1791 		ehca_gen_err("addr/size mismatch in 1st buf, pbuf->addr=%llx "
1792 			     "pbuf->size=%llx", pbuf->addr, pbuf->size);
1793 		return -EINVAL;
1794 	}
1795 
1796 	for (i = 0; i < num_phys_buf; i++) {
1797 		if ((i > 0) && (pbuf->addr % PAGE_SIZE)) {
1798 			ehca_gen_err("bad address, i=%x pbuf->addr=%llx "
1799 				     "pbuf->size=%llx",
1800 				     i, pbuf->addr, pbuf->size);
1801 			return -EINVAL;
1802 		}
1803 		if (((i > 0) &&	/* not 1st */
1804 		     (i < (num_phys_buf - 1)) &&	/* not last */
1805 		     (pbuf->size % PAGE_SIZE)) || (pbuf->size == 0)) {
1806 			ehca_gen_err("bad size, i=%x pbuf->size=%llx",
1807 				     i, pbuf->size);
1808 			return -EINVAL;
1809 		}
1810 		size_count += pbuf->size;
1811 		pbuf++;
1812 	}
1813 
1814 	*size = size_count;
1815 	return 0;
1816 } /* end ehca_mr_chk_buf_and_calc_size() */
1817 
1818 /*----------------------------------------------------------------------*/
1819 
1820 /* check page list of map FMR verb for validness */
ehca_fmr_check_page_list(struct ehca_mr * e_fmr,u64 * page_list,int list_len)1821 int ehca_fmr_check_page_list(struct ehca_mr *e_fmr,
1822 			     u64 *page_list,
1823 			     int list_len)
1824 {
1825 	u32 i;
1826 	u64 *page;
1827 
1828 	if ((list_len == 0) || (list_len > e_fmr->fmr_max_pages)) {
1829 		ehca_gen_err("bad list_len, list_len=%x "
1830 			     "e_fmr->fmr_max_pages=%x fmr=%p",
1831 			     list_len, e_fmr->fmr_max_pages, e_fmr);
1832 		return -EINVAL;
1833 	}
1834 
1835 	/* each page must be aligned */
1836 	page = page_list;
1837 	for (i = 0; i < list_len; i++) {
1838 		if (*page % e_fmr->fmr_page_size) {
1839 			ehca_gen_err("bad page, i=%x *page=%llx page=%p fmr=%p "
1840 				     "fmr_page_size=%x", i, *page, page, e_fmr,
1841 				     e_fmr->fmr_page_size);
1842 			return -EINVAL;
1843 		}
1844 		page++;
1845 	}
1846 
1847 	return 0;
1848 } /* end ehca_fmr_check_page_list() */
1849 
1850 /*----------------------------------------------------------------------*/
1851 
1852 /* PAGE_SIZE >= pginfo->hwpage_size */
ehca_set_pagebuf_user1(struct ehca_mr_pginfo * pginfo,u32 number,u64 * kpage)1853 static int ehca_set_pagebuf_user1(struct ehca_mr_pginfo *pginfo,
1854 				  u32 number,
1855 				  u64 *kpage)
1856 {
1857 	int ret = 0;
1858 	struct ib_umem_chunk *prev_chunk;
1859 	struct ib_umem_chunk *chunk;
1860 	u64 pgaddr;
1861 	u32 i = 0;
1862 	u32 j = 0;
1863 	int hwpages_per_kpage = PAGE_SIZE / pginfo->hwpage_size;
1864 
1865 	/* loop over desired chunk entries */
1866 	chunk      = pginfo->u.usr.next_chunk;
1867 	prev_chunk = pginfo->u.usr.next_chunk;
1868 	list_for_each_entry_continue(
1869 		chunk, (&(pginfo->u.usr.region->chunk_list)), list) {
1870 		for (i = pginfo->u.usr.next_nmap; i < chunk->nmap; ) {
1871 			pgaddr = page_to_pfn(sg_page(&chunk->page_list[i]))
1872 				<< PAGE_SHIFT ;
1873 			*kpage = phys_to_abs(pgaddr +
1874 					     (pginfo->next_hwpage *
1875 					      pginfo->hwpage_size));
1876 			if ( !(*kpage) ) {
1877 				ehca_gen_err("pgaddr=%llx "
1878 					     "chunk->page_list[i]=%llx "
1879 					     "i=%x next_hwpage=%llx",
1880 					     pgaddr, (u64)sg_dma_address(
1881 						     &chunk->page_list[i]),
1882 					     i, pginfo->next_hwpage);
1883 				return -EFAULT;
1884 			}
1885 			(pginfo->hwpage_cnt)++;
1886 			(pginfo->next_hwpage)++;
1887 			kpage++;
1888 			if (pginfo->next_hwpage % hwpages_per_kpage == 0) {
1889 				(pginfo->kpage_cnt)++;
1890 				(pginfo->u.usr.next_nmap)++;
1891 				pginfo->next_hwpage = 0;
1892 				i++;
1893 			}
1894 			j++;
1895 			if (j >= number) break;
1896 		}
1897 		if ((pginfo->u.usr.next_nmap >= chunk->nmap) &&
1898 		    (j >= number)) {
1899 			pginfo->u.usr.next_nmap = 0;
1900 			prev_chunk = chunk;
1901 			break;
1902 		} else if (pginfo->u.usr.next_nmap >= chunk->nmap) {
1903 			pginfo->u.usr.next_nmap = 0;
1904 			prev_chunk = chunk;
1905 		} else if (j >= number)
1906 			break;
1907 		else
1908 			prev_chunk = chunk;
1909 	}
1910 	pginfo->u.usr.next_chunk =
1911 		list_prepare_entry(prev_chunk,
1912 				   (&(pginfo->u.usr.region->chunk_list)),
1913 				   list);
1914 	return ret;
1915 }
1916 
1917 /*
1918  * check given pages for contiguous layout
1919  * last page addr is returned in prev_pgaddr for further check
1920  */
ehca_check_kpages_per_ate(struct scatterlist * page_list,int start_idx,int end_idx,u64 * prev_pgaddr)1921 static int ehca_check_kpages_per_ate(struct scatterlist *page_list,
1922 				     int start_idx, int end_idx,
1923 				     u64 *prev_pgaddr)
1924 {
1925 	int t;
1926 	for (t = start_idx; t <= end_idx; t++) {
1927 		u64 pgaddr = page_to_pfn(sg_page(&page_list[t])) << PAGE_SHIFT;
1928 		if (ehca_debug_level >= 3)
1929 			ehca_gen_dbg("chunk_page=%llx value=%016llx", pgaddr,
1930 				     *(u64 *)abs_to_virt(phys_to_abs(pgaddr)));
1931 		if (pgaddr - PAGE_SIZE != *prev_pgaddr) {
1932 			ehca_gen_err("uncontiguous page found pgaddr=%llx "
1933 				     "prev_pgaddr=%llx page_list_i=%x",
1934 				     pgaddr, *prev_pgaddr, t);
1935 			return -EINVAL;
1936 		}
1937 		*prev_pgaddr = pgaddr;
1938 	}
1939 	return 0;
1940 }
1941 
1942 /* PAGE_SIZE < pginfo->hwpage_size */
ehca_set_pagebuf_user2(struct ehca_mr_pginfo * pginfo,u32 number,u64 * kpage)1943 static int ehca_set_pagebuf_user2(struct ehca_mr_pginfo *pginfo,
1944 				  u32 number,
1945 				  u64 *kpage)
1946 {
1947 	int ret = 0;
1948 	struct ib_umem_chunk *prev_chunk;
1949 	struct ib_umem_chunk *chunk;
1950 	u64 pgaddr, prev_pgaddr;
1951 	u32 i = 0;
1952 	u32 j = 0;
1953 	int kpages_per_hwpage = pginfo->hwpage_size / PAGE_SIZE;
1954 	int nr_kpages = kpages_per_hwpage;
1955 
1956 	/* loop over desired chunk entries */
1957 	chunk      = pginfo->u.usr.next_chunk;
1958 	prev_chunk = pginfo->u.usr.next_chunk;
1959 	list_for_each_entry_continue(
1960 		chunk, (&(pginfo->u.usr.region->chunk_list)), list) {
1961 		for (i = pginfo->u.usr.next_nmap; i < chunk->nmap; ) {
1962 			if (nr_kpages == kpages_per_hwpage) {
1963 				pgaddr = ( page_to_pfn(sg_page(&chunk->page_list[i]))
1964 					   << PAGE_SHIFT );
1965 				*kpage = phys_to_abs(pgaddr);
1966 				if ( !(*kpage) ) {
1967 					ehca_gen_err("pgaddr=%llx i=%x",
1968 						     pgaddr, i);
1969 					ret = -EFAULT;
1970 					return ret;
1971 				}
1972 				/*
1973 				 * The first page in a hwpage must be aligned;
1974 				 * the first MR page is exempt from this rule.
1975 				 */
1976 				if (pgaddr & (pginfo->hwpage_size - 1)) {
1977 					if (pginfo->hwpage_cnt) {
1978 						ehca_gen_err(
1979 							"invalid alignment "
1980 							"pgaddr=%llx i=%x "
1981 							"mr_pgsize=%llx",
1982 							pgaddr, i,
1983 							pginfo->hwpage_size);
1984 						ret = -EFAULT;
1985 						return ret;
1986 					}
1987 					/* first MR page */
1988 					pginfo->kpage_cnt =
1989 						(pgaddr &
1990 						 (pginfo->hwpage_size - 1)) >>
1991 						PAGE_SHIFT;
1992 					nr_kpages -= pginfo->kpage_cnt;
1993 					*kpage = phys_to_abs(
1994 						pgaddr &
1995 						~(pginfo->hwpage_size - 1));
1996 				}
1997 				if (ehca_debug_level >= 3) {
1998 					u64 val = *(u64 *)abs_to_virt(
1999 						phys_to_abs(pgaddr));
2000 					ehca_gen_dbg("kpage=%llx chunk_page=%llx "
2001 						     "value=%016llx",
2002 						     *kpage, pgaddr, val);
2003 				}
2004 				prev_pgaddr = pgaddr;
2005 				i++;
2006 				pginfo->kpage_cnt++;
2007 				pginfo->u.usr.next_nmap++;
2008 				nr_kpages--;
2009 				if (!nr_kpages)
2010 					goto next_kpage;
2011 				continue;
2012 			}
2013 			if (i + nr_kpages > chunk->nmap) {
2014 				ret = ehca_check_kpages_per_ate(
2015 					chunk->page_list, i,
2016 					chunk->nmap - 1, &prev_pgaddr);
2017 				if (ret) return ret;
2018 				pginfo->kpage_cnt += chunk->nmap - i;
2019 				pginfo->u.usr.next_nmap += chunk->nmap - i;
2020 				nr_kpages -= chunk->nmap - i;
2021 				break;
2022 			}
2023 
2024 			ret = ehca_check_kpages_per_ate(chunk->page_list, i,
2025 							i + nr_kpages - 1,
2026 							&prev_pgaddr);
2027 			if (ret) return ret;
2028 			i += nr_kpages;
2029 			pginfo->kpage_cnt += nr_kpages;
2030 			pginfo->u.usr.next_nmap += nr_kpages;
2031 next_kpage:
2032 			nr_kpages = kpages_per_hwpage;
2033 			(pginfo->hwpage_cnt)++;
2034 			kpage++;
2035 			j++;
2036 			if (j >= number) break;
2037 		}
2038 		if ((pginfo->u.usr.next_nmap >= chunk->nmap) &&
2039 		    (j >= number)) {
2040 			pginfo->u.usr.next_nmap = 0;
2041 			prev_chunk = chunk;
2042 			break;
2043 		} else if (pginfo->u.usr.next_nmap >= chunk->nmap) {
2044 			pginfo->u.usr.next_nmap = 0;
2045 			prev_chunk = chunk;
2046 		} else if (j >= number)
2047 			break;
2048 		else
2049 			prev_chunk = chunk;
2050 	}
2051 	pginfo->u.usr.next_chunk =
2052 		list_prepare_entry(prev_chunk,
2053 				   (&(pginfo->u.usr.region->chunk_list)),
2054 				   list);
2055 	return ret;
2056 }
2057 
ehca_set_pagebuf_phys(struct ehca_mr_pginfo * pginfo,u32 number,u64 * kpage)2058 static int ehca_set_pagebuf_phys(struct ehca_mr_pginfo *pginfo,
2059 				 u32 number, u64 *kpage)
2060 {
2061 	int ret = 0;
2062 	struct ib_phys_buf *pbuf;
2063 	u64 num_hw, offs_hw;
2064 	u32 i = 0;
2065 
2066 	/* loop over desired phys_buf_array entries */
2067 	while (i < number) {
2068 		pbuf   = pginfo->u.phy.phys_buf_array + pginfo->u.phy.next_buf;
2069 		num_hw  = NUM_CHUNKS((pbuf->addr % pginfo->hwpage_size) +
2070 				     pbuf->size, pginfo->hwpage_size);
2071 		offs_hw = (pbuf->addr & ~(pginfo->hwpage_size - 1)) /
2072 			pginfo->hwpage_size;
2073 		while (pginfo->next_hwpage < offs_hw + num_hw) {
2074 			/* sanity check */
2075 			if ((pginfo->kpage_cnt >= pginfo->num_kpages) ||
2076 			    (pginfo->hwpage_cnt >= pginfo->num_hwpages)) {
2077 				ehca_gen_err("kpage_cnt >= num_kpages, "
2078 					     "kpage_cnt=%llx num_kpages=%llx "
2079 					     "hwpage_cnt=%llx "
2080 					     "num_hwpages=%llx i=%x",
2081 					     pginfo->kpage_cnt,
2082 					     pginfo->num_kpages,
2083 					     pginfo->hwpage_cnt,
2084 					     pginfo->num_hwpages, i);
2085 				return -EFAULT;
2086 			}
2087 			*kpage = phys_to_abs(
2088 				(pbuf->addr & ~(pginfo->hwpage_size - 1)) +
2089 				(pginfo->next_hwpage * pginfo->hwpage_size));
2090 			if ( !(*kpage) && pbuf->addr ) {
2091 				ehca_gen_err("pbuf->addr=%llx pbuf->size=%llx "
2092 					     "next_hwpage=%llx", pbuf->addr,
2093 					     pbuf->size, pginfo->next_hwpage);
2094 				return -EFAULT;
2095 			}
2096 			(pginfo->hwpage_cnt)++;
2097 			(pginfo->next_hwpage)++;
2098 			if (PAGE_SIZE >= pginfo->hwpage_size) {
2099 				if (pginfo->next_hwpage %
2100 				    (PAGE_SIZE / pginfo->hwpage_size) == 0)
2101 					(pginfo->kpage_cnt)++;
2102 			} else
2103 				pginfo->kpage_cnt += pginfo->hwpage_size /
2104 					PAGE_SIZE;
2105 			kpage++;
2106 			i++;
2107 			if (i >= number) break;
2108 		}
2109 		if (pginfo->next_hwpage >= offs_hw + num_hw) {
2110 			(pginfo->u.phy.next_buf)++;
2111 			pginfo->next_hwpage = 0;
2112 		}
2113 	}
2114 	return ret;
2115 }
2116 
ehca_set_pagebuf_fmr(struct ehca_mr_pginfo * pginfo,u32 number,u64 * kpage)2117 static int ehca_set_pagebuf_fmr(struct ehca_mr_pginfo *pginfo,
2118 				u32 number, u64 *kpage)
2119 {
2120 	int ret = 0;
2121 	u64 *fmrlist;
2122 	u32 i;
2123 
2124 	/* loop over desired page_list entries */
2125 	fmrlist = pginfo->u.fmr.page_list + pginfo->u.fmr.next_listelem;
2126 	for (i = 0; i < number; i++) {
2127 		*kpage = phys_to_abs((*fmrlist & ~(pginfo->hwpage_size - 1)) +
2128 				     pginfo->next_hwpage * pginfo->hwpage_size);
2129 		if ( !(*kpage) ) {
2130 			ehca_gen_err("*fmrlist=%llx fmrlist=%p "
2131 				     "next_listelem=%llx next_hwpage=%llx",
2132 				     *fmrlist, fmrlist,
2133 				     pginfo->u.fmr.next_listelem,
2134 				     pginfo->next_hwpage);
2135 			return -EFAULT;
2136 		}
2137 		(pginfo->hwpage_cnt)++;
2138 		if (pginfo->u.fmr.fmr_pgsize >= pginfo->hwpage_size) {
2139 			if (pginfo->next_hwpage %
2140 			    (pginfo->u.fmr.fmr_pgsize /
2141 			     pginfo->hwpage_size) == 0) {
2142 				(pginfo->kpage_cnt)++;
2143 				(pginfo->u.fmr.next_listelem)++;
2144 				fmrlist++;
2145 				pginfo->next_hwpage = 0;
2146 			} else
2147 				(pginfo->next_hwpage)++;
2148 		} else {
2149 			unsigned int cnt_per_hwpage = pginfo->hwpage_size /
2150 				pginfo->u.fmr.fmr_pgsize;
2151 			unsigned int j;
2152 			u64 prev = *kpage;
2153 			/* check if adrs are contiguous */
2154 			for (j = 1; j < cnt_per_hwpage; j++) {
2155 				u64 p = phys_to_abs(fmrlist[j] &
2156 						    ~(pginfo->hwpage_size - 1));
2157 				if (prev + pginfo->u.fmr.fmr_pgsize != p) {
2158 					ehca_gen_err("uncontiguous fmr pages "
2159 						     "found prev=%llx p=%llx "
2160 						     "idx=%x", prev, p, i + j);
2161 					return -EINVAL;
2162 				}
2163 				prev = p;
2164 			}
2165 			pginfo->kpage_cnt += cnt_per_hwpage;
2166 			pginfo->u.fmr.next_listelem += cnt_per_hwpage;
2167 			fmrlist += cnt_per_hwpage;
2168 		}
2169 		kpage++;
2170 	}
2171 	return ret;
2172 }
2173 
2174 /* setup page buffer from page info */
ehca_set_pagebuf(struct ehca_mr_pginfo * pginfo,u32 number,u64 * kpage)2175 int ehca_set_pagebuf(struct ehca_mr_pginfo *pginfo,
2176 		     u32 number,
2177 		     u64 *kpage)
2178 {
2179 	int ret;
2180 
2181 	switch (pginfo->type) {
2182 	case EHCA_MR_PGI_PHYS:
2183 		ret = ehca_set_pagebuf_phys(pginfo, number, kpage);
2184 		break;
2185 	case EHCA_MR_PGI_USER:
2186 		ret = PAGE_SIZE >= pginfo->hwpage_size ?
2187 			ehca_set_pagebuf_user1(pginfo, number, kpage) :
2188 			ehca_set_pagebuf_user2(pginfo, number, kpage);
2189 		break;
2190 	case EHCA_MR_PGI_FMR:
2191 		ret = ehca_set_pagebuf_fmr(pginfo, number, kpage);
2192 		break;
2193 	default:
2194 		ehca_gen_err("bad pginfo->type=%x", pginfo->type);
2195 		ret = -EFAULT;
2196 		break;
2197 	}
2198 	return ret;
2199 } /* end ehca_set_pagebuf() */
2200 
2201 /*----------------------------------------------------------------------*/
2202 
2203 /*
2204  * check MR if it is a max-MR, i.e. uses whole memory
2205  * in case it's a max-MR 1 is returned, else 0
2206  */
ehca_mr_is_maxmr(u64 size,u64 * iova_start)2207 int ehca_mr_is_maxmr(u64 size,
2208 		     u64 *iova_start)
2209 {
2210 	/* a MR is treated as max-MR only if it fits following: */
2211 	if ((size == ehca_mr_len) &&
2212 	    (iova_start == (void *)ehca_map_vaddr((void *)(KERNELBASE + PHYSICAL_START)))) {
2213 		ehca_gen_dbg("this is a max-MR");
2214 		return 1;
2215 	} else
2216 		return 0;
2217 } /* end ehca_mr_is_maxmr() */
2218 
2219 /*----------------------------------------------------------------------*/
2220 
2221 /* map access control for MR/MW. This routine is used for MR and MW. */
ehca_mrmw_map_acl(int ib_acl,u32 * hipz_acl)2222 void ehca_mrmw_map_acl(int ib_acl,
2223 		       u32 *hipz_acl)
2224 {
2225 	*hipz_acl = 0;
2226 	if (ib_acl & IB_ACCESS_REMOTE_READ)
2227 		*hipz_acl |= HIPZ_ACCESSCTRL_R_READ;
2228 	if (ib_acl & IB_ACCESS_REMOTE_WRITE)
2229 		*hipz_acl |= HIPZ_ACCESSCTRL_R_WRITE;
2230 	if (ib_acl & IB_ACCESS_REMOTE_ATOMIC)
2231 		*hipz_acl |= HIPZ_ACCESSCTRL_R_ATOMIC;
2232 	if (ib_acl & IB_ACCESS_LOCAL_WRITE)
2233 		*hipz_acl |= HIPZ_ACCESSCTRL_L_WRITE;
2234 	if (ib_acl & IB_ACCESS_MW_BIND)
2235 		*hipz_acl |= HIPZ_ACCESSCTRL_MW_BIND;
2236 } /* end ehca_mrmw_map_acl() */
2237 
2238 /*----------------------------------------------------------------------*/
2239 
2240 /* sets page size in hipz access control for MR/MW. */
ehca_mrmw_set_pgsize_hipz_acl(u32 pgsize,u32 * hipz_acl)2241 void ehca_mrmw_set_pgsize_hipz_acl(u32 pgsize, u32 *hipz_acl) /*INOUT*/
2242 {
2243 	*hipz_acl |= (ehca_encode_hwpage_size(pgsize) << 24);
2244 } /* end ehca_mrmw_set_pgsize_hipz_acl() */
2245 
2246 /*----------------------------------------------------------------------*/
2247 
2248 /*
2249  * reverse map access control for MR/MW.
2250  * This routine is used for MR and MW.
2251  */
ehca_mrmw_reverse_map_acl(const u32 * hipz_acl,int * ib_acl)2252 void ehca_mrmw_reverse_map_acl(const u32 *hipz_acl,
2253 			       int *ib_acl) /*OUT*/
2254 {
2255 	*ib_acl = 0;
2256 	if (*hipz_acl & HIPZ_ACCESSCTRL_R_READ)
2257 		*ib_acl |= IB_ACCESS_REMOTE_READ;
2258 	if (*hipz_acl & HIPZ_ACCESSCTRL_R_WRITE)
2259 		*ib_acl |= IB_ACCESS_REMOTE_WRITE;
2260 	if (*hipz_acl & HIPZ_ACCESSCTRL_R_ATOMIC)
2261 		*ib_acl |= IB_ACCESS_REMOTE_ATOMIC;
2262 	if (*hipz_acl & HIPZ_ACCESSCTRL_L_WRITE)
2263 		*ib_acl |= IB_ACCESS_LOCAL_WRITE;
2264 	if (*hipz_acl & HIPZ_ACCESSCTRL_MW_BIND)
2265 		*ib_acl |= IB_ACCESS_MW_BIND;
2266 } /* end ehca_mrmw_reverse_map_acl() */
2267 
2268 
2269 /*----------------------------------------------------------------------*/
2270 
2271 /*
2272  * MR destructor and constructor
2273  * used in Reregister MR verb, sets all fields in ehca_mr_t to 0,
2274  * except struct ib_mr and spinlock
2275  */
ehca_mr_deletenew(struct ehca_mr * mr)2276 void ehca_mr_deletenew(struct ehca_mr *mr)
2277 {
2278 	mr->flags = 0;
2279 	mr->num_kpages = 0;
2280 	mr->num_hwpages = 0;
2281 	mr->acl = 0;
2282 	mr->start = NULL;
2283 	mr->fmr_page_size = 0;
2284 	mr->fmr_max_pages = 0;
2285 	mr->fmr_max_maps = 0;
2286 	mr->fmr_map_cnt = 0;
2287 	memset(&mr->ipz_mr_handle, 0, sizeof(mr->ipz_mr_handle));
2288 	memset(&mr->galpas, 0, sizeof(mr->galpas));
2289 } /* end ehca_mr_deletenew() */
2290 
ehca_init_mrmw_cache(void)2291 int ehca_init_mrmw_cache(void)
2292 {
2293 	mr_cache = kmem_cache_create("ehca_cache_mr",
2294 				     sizeof(struct ehca_mr), 0,
2295 				     SLAB_HWCACHE_ALIGN,
2296 				     NULL);
2297 	if (!mr_cache)
2298 		return -ENOMEM;
2299 	mw_cache = kmem_cache_create("ehca_cache_mw",
2300 				     sizeof(struct ehca_mw), 0,
2301 				     SLAB_HWCACHE_ALIGN,
2302 				     NULL);
2303 	if (!mw_cache) {
2304 		kmem_cache_destroy(mr_cache);
2305 		mr_cache = NULL;
2306 		return -ENOMEM;
2307 	}
2308 	return 0;
2309 }
2310 
ehca_cleanup_mrmw_cache(void)2311 void ehca_cleanup_mrmw_cache(void)
2312 {
2313 	if (mr_cache)
2314 		kmem_cache_destroy(mr_cache);
2315 	if (mw_cache)
2316 		kmem_cache_destroy(mw_cache);
2317 }
2318 
ehca_init_top_bmap(struct ehca_top_bmap * ehca_top_bmap,int dir)2319 static inline int ehca_init_top_bmap(struct ehca_top_bmap *ehca_top_bmap,
2320 				     int dir)
2321 {
2322 	if (!ehca_bmap_valid(ehca_top_bmap->dir[dir])) {
2323 		ehca_top_bmap->dir[dir] =
2324 			kmalloc(sizeof(struct ehca_dir_bmap), GFP_KERNEL);
2325 		if (!ehca_top_bmap->dir[dir])
2326 			return -ENOMEM;
2327 		/* Set map block to 0xFF according to EHCA_INVAL_ADDR */
2328 		memset(ehca_top_bmap->dir[dir], 0xFF, EHCA_ENT_MAP_SIZE);
2329 	}
2330 	return 0;
2331 }
2332 
ehca_init_bmap(struct ehca_bmap * ehca_bmap,int top,int dir)2333 static inline int ehca_init_bmap(struct ehca_bmap *ehca_bmap, int top, int dir)
2334 {
2335 	if (!ehca_bmap_valid(ehca_bmap->top[top])) {
2336 		ehca_bmap->top[top] =
2337 			kmalloc(sizeof(struct ehca_top_bmap), GFP_KERNEL);
2338 		if (!ehca_bmap->top[top])
2339 			return -ENOMEM;
2340 		/* Set map block to 0xFF according to EHCA_INVAL_ADDR */
2341 		memset(ehca_bmap->top[top], 0xFF, EHCA_DIR_MAP_SIZE);
2342 	}
2343 	return ehca_init_top_bmap(ehca_bmap->top[top], dir);
2344 }
2345 
ehca_calc_index(unsigned long i,unsigned long s)2346 static inline int ehca_calc_index(unsigned long i, unsigned long s)
2347 {
2348 	return (i >> s) & EHCA_INDEX_MASK;
2349 }
2350 
ehca_destroy_busmap(void)2351 void ehca_destroy_busmap(void)
2352 {
2353 	int top, dir;
2354 
2355 	if (!ehca_bmap)
2356 		return;
2357 
2358 	for (top = 0; top < EHCA_MAP_ENTRIES; top++) {
2359 		if (!ehca_bmap_valid(ehca_bmap->top[top]))
2360 			continue;
2361 		for (dir = 0; dir < EHCA_MAP_ENTRIES; dir++) {
2362 			if (!ehca_bmap_valid(ehca_bmap->top[top]->dir[dir]))
2363 				continue;
2364 
2365 			kfree(ehca_bmap->top[top]->dir[dir]);
2366 		}
2367 
2368 		kfree(ehca_bmap->top[top]);
2369 	}
2370 
2371 	kfree(ehca_bmap);
2372 	ehca_bmap = NULL;
2373 }
2374 
ehca_update_busmap(unsigned long pfn,unsigned long nr_pages)2375 static int ehca_update_busmap(unsigned long pfn, unsigned long nr_pages)
2376 {
2377 	unsigned long i, start_section, end_section;
2378 	int top, dir, idx;
2379 
2380 	if (!nr_pages)
2381 		return 0;
2382 
2383 	if (!ehca_bmap) {
2384 		ehca_bmap = kmalloc(sizeof(struct ehca_bmap), GFP_KERNEL);
2385 		if (!ehca_bmap)
2386 			return -ENOMEM;
2387 		/* Set map block to 0xFF according to EHCA_INVAL_ADDR */
2388 		memset(ehca_bmap, 0xFF, EHCA_TOP_MAP_SIZE);
2389 	}
2390 
2391 	start_section = phys_to_abs(pfn * PAGE_SIZE) / EHCA_SECTSIZE;
2392 	end_section = phys_to_abs((pfn + nr_pages) * PAGE_SIZE) / EHCA_SECTSIZE;
2393 	for (i = start_section; i < end_section; i++) {
2394 		int ret;
2395 		top = ehca_calc_index(i, EHCA_TOP_INDEX_SHIFT);
2396 		dir = ehca_calc_index(i, EHCA_DIR_INDEX_SHIFT);
2397 		idx = i & EHCA_INDEX_MASK;
2398 
2399 		ret = ehca_init_bmap(ehca_bmap, top, dir);
2400 		if (ret) {
2401 			ehca_destroy_busmap();
2402 			return ret;
2403 		}
2404 		ehca_bmap->top[top]->dir[dir]->ent[idx] = ehca_mr_len;
2405 		ehca_mr_len += EHCA_SECTSIZE;
2406 	}
2407 	return 0;
2408 }
2409 
ehca_is_hugepage(unsigned long pfn)2410 static int ehca_is_hugepage(unsigned long pfn)
2411 {
2412 	int page_order;
2413 
2414 	if (pfn & EHCA_HUGEPAGE_PFN_MASK)
2415 		return 0;
2416 
2417 	page_order = compound_order(pfn_to_page(pfn));
2418 	if (page_order + PAGE_SHIFT != EHCA_HUGEPAGESHIFT)
2419 		return 0;
2420 
2421 	return 1;
2422 }
2423 
ehca_create_busmap_callback(unsigned long initial_pfn,unsigned long total_nr_pages,void * arg)2424 static int ehca_create_busmap_callback(unsigned long initial_pfn,
2425 				       unsigned long total_nr_pages, void *arg)
2426 {
2427 	int ret;
2428 	unsigned long pfn, start_pfn, end_pfn, nr_pages;
2429 
2430 	if ((total_nr_pages * PAGE_SIZE) < EHCA_HUGEPAGE_SIZE)
2431 		return ehca_update_busmap(initial_pfn, total_nr_pages);
2432 
2433 	/* Given chunk is >= 16GB -> check for hugepages */
2434 	start_pfn = initial_pfn;
2435 	end_pfn = initial_pfn + total_nr_pages;
2436 	pfn = start_pfn;
2437 
2438 	while (pfn < end_pfn) {
2439 		if (ehca_is_hugepage(pfn)) {
2440 			/* Add mem found in front of the hugepage */
2441 			nr_pages = pfn - start_pfn;
2442 			ret = ehca_update_busmap(start_pfn, nr_pages);
2443 			if (ret)
2444 				return ret;
2445 			/* Skip the hugepage */
2446 			pfn += (EHCA_HUGEPAGE_SIZE / PAGE_SIZE);
2447 			start_pfn = pfn;
2448 		} else
2449 			pfn += (EHCA_SECTSIZE / PAGE_SIZE);
2450 	}
2451 
2452 	/* Add mem found behind the hugepage(s)  */
2453 	nr_pages = pfn - start_pfn;
2454 	return ehca_update_busmap(start_pfn, nr_pages);
2455 }
2456 
ehca_create_busmap(void)2457 int ehca_create_busmap(void)
2458 {
2459 	int ret;
2460 
2461 	ehca_mr_len = 0;
2462 	ret = walk_system_ram_range(0, 1ULL << MAX_PHYSMEM_BITS, NULL,
2463 				   ehca_create_busmap_callback);
2464 	return ret;
2465 }
2466 
ehca_reg_bmap_mr_rpages(struct ehca_shca * shca,struct ehca_mr * e_mr,struct ehca_mr_pginfo * pginfo)2467 static int ehca_reg_bmap_mr_rpages(struct ehca_shca *shca,
2468 				   struct ehca_mr *e_mr,
2469 				   struct ehca_mr_pginfo *pginfo)
2470 {
2471 	int top;
2472 	u64 hret, *kpage;
2473 
2474 	kpage = ehca_alloc_fw_ctrlblock(GFP_KERNEL);
2475 	if (!kpage) {
2476 		ehca_err(&shca->ib_device, "kpage alloc failed");
2477 		return -ENOMEM;
2478 	}
2479 	for (top = 0; top < EHCA_MAP_ENTRIES; top++) {
2480 		if (!ehca_bmap_valid(ehca_bmap->top[top]))
2481 			continue;
2482 		hret = ehca_reg_mr_dir_sections(top, kpage, shca, e_mr, pginfo);
2483 		if ((hret != H_PAGE_REGISTERED) && (hret != H_SUCCESS))
2484 			break;
2485 	}
2486 
2487 	ehca_free_fw_ctrlblock(kpage);
2488 
2489 	if (hret == H_SUCCESS)
2490 		return 0; /* Everything is fine */
2491 	else {
2492 		ehca_err(&shca->ib_device, "ehca_reg_bmap_mr_rpages failed, "
2493 				 "h_ret=%lli e_mr=%p top=%x lkey=%x "
2494 				 "hca_hndl=%llx mr_hndl=%llx", hret, e_mr, top,
2495 				 e_mr->ib.ib_mr.lkey,
2496 				 shca->ipz_hca_handle.handle,
2497 				 e_mr->ipz_mr_handle.handle);
2498 		return ehca2ib_return_code(hret);
2499 	}
2500 }
2501 
ehca_map_vaddr(void * caddr)2502 static u64 ehca_map_vaddr(void *caddr)
2503 {
2504 	int top, dir, idx;
2505 	unsigned long abs_addr, offset;
2506 	u64 entry;
2507 
2508 	if (!ehca_bmap)
2509 		return EHCA_INVAL_ADDR;
2510 
2511 	abs_addr = virt_to_abs(caddr);
2512 	top = ehca_calc_index(abs_addr, EHCA_TOP_INDEX_SHIFT + EHCA_SECTSHIFT);
2513 	if (!ehca_bmap_valid(ehca_bmap->top[top]))
2514 		return EHCA_INVAL_ADDR;
2515 
2516 	dir = ehca_calc_index(abs_addr, EHCA_DIR_INDEX_SHIFT + EHCA_SECTSHIFT);
2517 	if (!ehca_bmap_valid(ehca_bmap->top[top]->dir[dir]))
2518 		return EHCA_INVAL_ADDR;
2519 
2520 	idx = ehca_calc_index(abs_addr, EHCA_SECTSHIFT);
2521 
2522 	entry = ehca_bmap->top[top]->dir[dir]->ent[idx];
2523 	if (ehca_bmap_valid(entry)) {
2524 		offset = (unsigned long)caddr & (EHCA_SECTSIZE - 1);
2525 		return entry | offset;
2526 	} else
2527 		return EHCA_INVAL_ADDR;
2528 }
2529 
ehca_dma_mapping_error(struct ib_device * dev,u64 dma_addr)2530 static int ehca_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
2531 {
2532 	return dma_addr == EHCA_INVAL_ADDR;
2533 }
2534 
ehca_dma_map_single(struct ib_device * dev,void * cpu_addr,size_t size,enum dma_data_direction direction)2535 static u64 ehca_dma_map_single(struct ib_device *dev, void *cpu_addr,
2536 			       size_t size, enum dma_data_direction direction)
2537 {
2538 	if (cpu_addr)
2539 		return ehca_map_vaddr(cpu_addr);
2540 	else
2541 		return EHCA_INVAL_ADDR;
2542 }
2543 
ehca_dma_unmap_single(struct ib_device * dev,u64 addr,size_t size,enum dma_data_direction direction)2544 static void ehca_dma_unmap_single(struct ib_device *dev, u64 addr, size_t size,
2545 				  enum dma_data_direction direction)
2546 {
2547 	/* This is only a stub; nothing to be done here */
2548 }
2549 
ehca_dma_map_page(struct ib_device * dev,struct page * page,unsigned long offset,size_t size,enum dma_data_direction direction)2550 static u64 ehca_dma_map_page(struct ib_device *dev, struct page *page,
2551 			     unsigned long offset, size_t size,
2552 			     enum dma_data_direction direction)
2553 {
2554 	u64 addr;
2555 
2556 	if (offset + size > PAGE_SIZE)
2557 		return EHCA_INVAL_ADDR;
2558 
2559 	addr = ehca_map_vaddr(page_address(page));
2560 	if (!ehca_dma_mapping_error(dev, addr))
2561 		addr += offset;
2562 
2563 	return addr;
2564 }
2565 
ehca_dma_unmap_page(struct ib_device * dev,u64 addr,size_t size,enum dma_data_direction direction)2566 static void ehca_dma_unmap_page(struct ib_device *dev, u64 addr, size_t size,
2567 				enum dma_data_direction direction)
2568 {
2569 	/* This is only a stub; nothing to be done here */
2570 }
2571 
ehca_dma_map_sg(struct ib_device * dev,struct scatterlist * sgl,int nents,enum dma_data_direction direction)2572 static int ehca_dma_map_sg(struct ib_device *dev, struct scatterlist *sgl,
2573 			   int nents, enum dma_data_direction direction)
2574 {
2575 	struct scatterlist *sg;
2576 	int i;
2577 
2578 	for_each_sg(sgl, sg, nents, i) {
2579 		u64 addr;
2580 		addr = ehca_map_vaddr(sg_virt(sg));
2581 		if (ehca_dma_mapping_error(dev, addr))
2582 			return 0;
2583 
2584 		sg->dma_address = addr;
2585 		sg->dma_length = sg->length;
2586 	}
2587 	return nents;
2588 }
2589 
ehca_dma_unmap_sg(struct ib_device * dev,struct scatterlist * sg,int nents,enum dma_data_direction direction)2590 static void ehca_dma_unmap_sg(struct ib_device *dev, struct scatterlist *sg,
2591 			      int nents, enum dma_data_direction direction)
2592 {
2593 	/* This is only a stub; nothing to be done here */
2594 }
2595 
ehca_dma_address(struct ib_device * dev,struct scatterlist * sg)2596 static u64 ehca_dma_address(struct ib_device *dev, struct scatterlist *sg)
2597 {
2598 	return sg->dma_address;
2599 }
2600 
ehca_dma_len(struct ib_device * dev,struct scatterlist * sg)2601 static unsigned int ehca_dma_len(struct ib_device *dev, struct scatterlist *sg)
2602 {
2603 	return sg->length;
2604 }
2605 
ehca_dma_sync_single_for_cpu(struct ib_device * dev,u64 addr,size_t size,enum dma_data_direction dir)2606 static void ehca_dma_sync_single_for_cpu(struct ib_device *dev, u64 addr,
2607 					 size_t size,
2608 					 enum dma_data_direction dir)
2609 {
2610 	dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
2611 }
2612 
ehca_dma_sync_single_for_device(struct ib_device * dev,u64 addr,size_t size,enum dma_data_direction dir)2613 static void ehca_dma_sync_single_for_device(struct ib_device *dev, u64 addr,
2614 					    size_t size,
2615 					    enum dma_data_direction dir)
2616 {
2617 	dma_sync_single_for_device(dev->dma_device, addr, size, dir);
2618 }
2619 
ehca_dma_alloc_coherent(struct ib_device * dev,size_t size,u64 * dma_handle,gfp_t flag)2620 static void *ehca_dma_alloc_coherent(struct ib_device *dev, size_t size,
2621 				     u64 *dma_handle, gfp_t flag)
2622 {
2623 	struct page *p;
2624 	void *addr = NULL;
2625 	u64 dma_addr;
2626 
2627 	p = alloc_pages(flag, get_order(size));
2628 	if (p) {
2629 		addr = page_address(p);
2630 		dma_addr = ehca_map_vaddr(addr);
2631 		if (ehca_dma_mapping_error(dev, dma_addr)) {
2632 			free_pages((unsigned long)addr,	get_order(size));
2633 			return NULL;
2634 		}
2635 		if (dma_handle)
2636 			*dma_handle = dma_addr;
2637 		return addr;
2638 	}
2639 	return NULL;
2640 }
2641 
ehca_dma_free_coherent(struct ib_device * dev,size_t size,void * cpu_addr,u64 dma_handle)2642 static void ehca_dma_free_coherent(struct ib_device *dev, size_t size,
2643 				   void *cpu_addr, u64 dma_handle)
2644 {
2645 	if (cpu_addr && size)
2646 		free_pages((unsigned long)cpu_addr, get_order(size));
2647 }
2648 
2649 
2650 struct ib_dma_mapping_ops ehca_dma_mapping_ops = {
2651 	.mapping_error          = ehca_dma_mapping_error,
2652 	.map_single             = ehca_dma_map_single,
2653 	.unmap_single           = ehca_dma_unmap_single,
2654 	.map_page               = ehca_dma_map_page,
2655 	.unmap_page             = ehca_dma_unmap_page,
2656 	.map_sg                 = ehca_dma_map_sg,
2657 	.unmap_sg               = ehca_dma_unmap_sg,
2658 	.dma_address            = ehca_dma_address,
2659 	.dma_len                = ehca_dma_len,
2660 	.sync_single_for_cpu    = ehca_dma_sync_single_for_cpu,
2661 	.sync_single_for_device = ehca_dma_sync_single_for_device,
2662 	.alloc_coherent         = ehca_dma_alloc_coherent,
2663 	.free_coherent          = ehca_dma_free_coherent,
2664 };
2665