user_exp_rcv.c 29.7 KB
Newer Older
1
/*
J
Jubin John 已提交
2
 * Copyright(c) 2015, 2016 Intel Corporation.
3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50
 *
 * This file is provided under a dual BSD/GPLv2 license.  When using or
 * redistributing this file, you may do so under either license.
 *
 * GPL LICENSE SUMMARY
 *
 * This program is free software; you can redistribute it and/or modify
 * it under the terms of version 2 of the GNU General Public License as
 * published by the Free Software Foundation.
 *
 * This program is distributed in the hope that it will be useful, but
 * WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * General Public License for more details.
 *
 * BSD LICENSE
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 *
 *  - Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 *  - Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in
 *    the documentation and/or other materials provided with the
 *    distribution.
 *  - Neither the name of Intel Corporation nor the names of its
 *    contributors may be used to endorse or promote products derived
 *    from this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 *
 */
#include <asm/page.h>

#include "user_exp_rcv.h"
#include "trace.h"
51
#include "mmu_rb.h"
52

53 54 55 56 57 58 59 60
struct tid_group {
	struct list_head list;
	unsigned base;
	u8 size;
	u8 used;
	u8 map;
};

61 62
struct tid_rb_node {
	struct mmu_rb_node mmu;
63 64 65 66 67 68 69 70 71
	unsigned long phys;
	struct tid_group *grp;
	u32 rcventry;
	dma_addr_t dma_addr;
	bool freed;
	unsigned npages;
	struct page *pages[0];
};

72 73 74 75 76
struct tid_pageset {
	u16 idx;
	u16 count;
};

77 78
#define EXP_TID_SET_EMPTY(set) (set.count == 0 && list_empty(&set.list))

79 80 81 82 83
#define num_user_pages(vaddr, len)				       \
	(1 + (((((unsigned long)(vaddr) +			       \
		 (unsigned long)(len) - 1) & PAGE_MASK) -	       \
	       ((unsigned long)vaddr & PAGE_MASK)) >> PAGE_SHIFT))

84
static void unlock_exp_tids(struct hfi1_ctxtdata *, struct exp_tid_set *,
85
			    struct hfi1_filedata *);
86
static u32 find_phys_blocks(struct page **, unsigned, struct tid_pageset *);
87
static int set_rcvarray_entry(struct file *, unsigned long, u32,
88
			      struct tid_group *, struct page **, unsigned);
89 90
static int tid_rb_insert(void *, struct mmu_rb_node *);
static void tid_rb_remove(void *, struct mmu_rb_node *,
91
			  struct mm_struct *);
92
static int tid_rb_invalidate(void *, struct mmu_rb_node *);
93 94
static int program_rcvarray(struct file *, unsigned long, struct tid_group *,
			    struct tid_pageset *, unsigned, u16, struct page **,
95
			    u32 *, unsigned *, unsigned *);
96
static int unprogram_rcvarray(struct file *, u32, struct tid_group **);
97
static void clear_tid_node(struct hfi1_filedata *, struct tid_rb_node *);
98 99

static struct mmu_rb_ops tid_rb_ops = {
100 101 102
	.insert = tid_rb_insert,
	.remove = tid_rb_remove,
	.invalidate = tid_rb_invalidate
103
};
104 105 106 107 108 109 110 111

static inline u32 rcventry2tidinfo(u32 rcventry)
{
	u32 pair = rcventry & ~0x1;

	return EXP_TID_SET(IDX, pair >> 1) |
		EXP_TID_SET(CTRL, 1 << (rcventry - pair));
}
112

113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149
static inline void exp_tid_group_init(struct exp_tid_set *set)
{
	INIT_LIST_HEAD(&set->list);
	set->count = 0;
}

static inline void tid_group_remove(struct tid_group *grp,
				    struct exp_tid_set *set)
{
	list_del_init(&grp->list);
	set->count--;
}

static inline void tid_group_add_tail(struct tid_group *grp,
				      struct exp_tid_set *set)
{
	list_add_tail(&grp->list, &set->list);
	set->count++;
}

static inline struct tid_group *tid_group_pop(struct exp_tid_set *set)
{
	struct tid_group *grp =
		list_first_entry(&set->list, struct tid_group, list);
	list_del_init(&grp->list);
	set->count--;
	return grp;
}

static inline void tid_group_move(struct tid_group *group,
				  struct exp_tid_set *s1,
				  struct exp_tid_set *s2)
{
	tid_group_remove(group, s1);
	tid_group_add_tail(group, s2);
}

150 151 152 153 154 155 156
/*
 * Initialize context and file private data needed for Expected
 * receive caching. This needs to be done after the context has
 * been configured with the eager/expected RcvEntry counts.
 */
int hfi1_user_exp_rcv_init(struct file *fp)
{
157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192
	struct hfi1_filedata *fd = fp->private_data;
	struct hfi1_ctxtdata *uctxt = fd->uctxt;
	struct hfi1_devdata *dd = uctxt->dd;
	unsigned tidbase;
	int i, ret = 0;

	spin_lock_init(&fd->tid_lock);
	spin_lock_init(&fd->invalid_lock);

	if (!uctxt->subctxt_cnt || !fd->subctxt) {
		exp_tid_group_init(&uctxt->tid_group_list);
		exp_tid_group_init(&uctxt->tid_used_list);
		exp_tid_group_init(&uctxt->tid_full_list);

		tidbase = uctxt->expected_base;
		for (i = 0; i < uctxt->expected_count /
			     dd->rcv_entries.group_size; i++) {
			struct tid_group *grp;

			grp = kzalloc(sizeof(*grp), GFP_KERNEL);
			if (!grp) {
				/*
				 * If we fail here, the groups already
				 * allocated will be freed by the close
				 * call.
				 */
				ret = -ENOMEM;
				goto done;
			}
			grp->size = dd->rcv_entries.group_size;
			grp->base = tidbase;
			tid_group_add_tail(grp, &uctxt->tid_group_list);
			tidbase += dd->rcv_entries.group_size;
		}
	}

193 194 195 196 197 198
	fd->entry_to_rb = kcalloc(uctxt->expected_count,
				     sizeof(struct rb_node *),
				     GFP_KERNEL);
	if (!fd->entry_to_rb)
		return -ENOMEM;

D
Dean Luick 已提交
199
	if (!HFI1_CAP_UGET_MASK(uctxt->flags, TID_UNMAP)) {
200 201 202 203 204 205
		fd->invalid_tid_idx = 0;
		fd->invalid_tids = kzalloc(uctxt->expected_count *
					   sizeof(u32), GFP_KERNEL);
		if (!fd->invalid_tids) {
			ret = -ENOMEM;
			goto done;
206 207 208 209
		}

		/*
		 * Register MMU notifier callbacks. If the registration
D
Dean Luick 已提交
210
		 * fails, continue without TID caching for this context.
211
		 */
212
		ret = hfi1_mmu_rb_register(fd, fd->mm, &tid_rb_ops, &fd->handler);
213 214 215 216 217
		if (ret) {
			dd_dev_info(dd,
				    "Failed MMU notifier registration %d\n",
				    ret);
			ret = 0;
218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233
		}
	}

	/*
	 * PSM does not have a good way to separate, count, and
	 * effectively enforce a limit on RcvArray entries used by
	 * subctxts (when context sharing is used) when TID caching
	 * is enabled. To help with that, we calculate a per-process
	 * RcvArray entry share and enforce that.
	 * If TID caching is not in use, PSM deals with usage on its
	 * own. In that case, we allow any subctxt to take all of the
	 * entries.
	 *
	 * Make sure that we set the tid counts only after successful
	 * init.
	 */
234
	spin_lock(&fd->tid_lock);
D
Dean Luick 已提交
235
	if (uctxt->subctxt_cnt && fd->handler) {
236 237 238 239 240 241 242 243 244
		u16 remainder;

		fd->tid_limit = uctxt->expected_count / uctxt->subctxt_cnt;
		remainder = uctxt->expected_count % uctxt->subctxt_cnt;
		if (remainder && fd->subctxt < remainder)
			fd->tid_limit++;
	} else {
		fd->tid_limit = uctxt->expected_count;
	}
245
	spin_unlock(&fd->tid_lock);
246 247
done:
	return ret;
248 249 250 251
}

int hfi1_user_exp_rcv_free(struct hfi1_filedata *fd)
{
252 253 254
	struct hfi1_ctxtdata *uctxt = fd->uctxt;
	struct tid_group *grp, *gptr;

255 256
	if (!test_bit(HFI1_CTXT_SETUP_DONE, &uctxt->event_flags))
		return 0;
257 258 259 260
	/*
	 * The notifier would have been removed when the process'es mm
	 * was freed.
	 */
D
Dean Luick 已提交
261
	if (fd->handler)
262
		hfi1_mmu_rb_unregister(fd->handler);
263 264 265 266 267

	kfree(fd->invalid_tids);

	if (!uctxt->cnt) {
		if (!EXP_TID_SET_EMPTY(uctxt->tid_full_list))
268
			unlock_exp_tids(uctxt, &uctxt->tid_full_list, fd);
269
		if (!EXP_TID_SET_EMPTY(uctxt->tid_used_list))
270
			unlock_exp_tids(uctxt, &uctxt->tid_used_list, fd);
271 272 273 274 275 276 277
		list_for_each_entry_safe(grp, gptr, &uctxt->tid_group_list.list,
					 list) {
			list_del_init(&grp->list);
			kfree(grp);
		}
		hfi1_clear_tids(uctxt);
	}
278 279

	kfree(fd->entry_to_rb);
280
	return 0;
281 282
}

283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299
/*
 * Write an "empty" RcvArray entry.
 * This function exists so the TID registaration code can use it
 * to write to unused/unneeded entries and still take advantage
 * of the WC performance improvements. The HFI will ignore this
 * write to the RcvArray entry.
 */
static inline void rcv_array_wc_fill(struct hfi1_devdata *dd, u32 index)
{
	/*
	 * Doing the WC fill writes only makes sense if the device is
	 * present and the RcvArray has been mapped as WC memory.
	 */
	if ((dd->flags & HFI1_PRESENT) && dd->rcvarray_wc)
		writeq(0, dd->rcvarray_wc + (index * 8));
}

300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348
/*
 * RcvArray entry allocation for Expected Receives is done by the
 * following algorithm:
 *
 * The context keeps 3 lists of groups of RcvArray entries:
 *   1. List of empty groups - tid_group_list
 *      This list is created during user context creation and
 *      contains elements which describe sets (of 8) of empty
 *      RcvArray entries.
 *   2. List of partially used groups - tid_used_list
 *      This list contains sets of RcvArray entries which are
 *      not completely used up. Another mapping request could
 *      use some of all of the remaining entries.
 *   3. List of full groups - tid_full_list
 *      This is the list where sets that are completely used
 *      up go.
 *
 * An attempt to optimize the usage of RcvArray entries is
 * made by finding all sets of physically contiguous pages in a
 * user's buffer.
 * These physically contiguous sets are further split into
 * sizes supported by the receive engine of the HFI. The
 * resulting sets of pages are stored in struct tid_pageset,
 * which describes the sets as:
 *    * .count - number of pages in this set
 *    * .idx - starting index into struct page ** array
 *                    of this set
 *
 * From this point on, the algorithm deals with the page sets
 * described above. The number of pagesets is divided by the
 * RcvArray group size to produce the number of full groups
 * needed.
 *
 * Groups from the 3 lists are manipulated using the following
 * rules:
 *   1. For each set of 8 pagesets, a complete group from
 *      tid_group_list is taken, programmed, and moved to
 *      the tid_full_list list.
 *   2. For all remaining pagesets:
 *      2.1 If the tid_used_list is empty and the tid_group_list
 *          is empty, stop processing pageset and return only
 *          what has been programmed up to this point.
 *      2.2 If the tid_used_list is empty and the tid_group_list
 *          is not empty, move a group from tid_group_list to
 *          tid_used_list.
 *      2.3 For each group is tid_used_group, program as much as
 *          can fit into the group. If the group becomes fully
 *          used, move it to tid_full_list.
 */
349 350
int hfi1_user_exp_rcv_setup(struct file *fp, struct hfi1_tid_info *tinfo)
{
351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396
	int ret = 0, need_group = 0, pinned;
	struct hfi1_filedata *fd = fp->private_data;
	struct hfi1_ctxtdata *uctxt = fd->uctxt;
	struct hfi1_devdata *dd = uctxt->dd;
	unsigned npages, ngroups, pageidx = 0, pageset_count, npagesets,
		tididx = 0, mapped, mapped_pages = 0;
	unsigned long vaddr = tinfo->vaddr;
	struct page **pages = NULL;
	u32 *tidlist = NULL;
	struct tid_pageset *pagesets = NULL;

	/* Get the number of pages the user buffer spans */
	npages = num_user_pages(vaddr, tinfo->length);
	if (!npages)
		return -EINVAL;

	if (npages > uctxt->expected_count) {
		dd_dev_err(dd, "Expected buffer too big\n");
		return -EINVAL;
	}

	/* Verify that access is OK for the user buffer */
	if (!access_ok(VERIFY_WRITE, (void __user *)vaddr,
		       npages * PAGE_SIZE)) {
		dd_dev_err(dd, "Fail vaddr %p, %u pages, !access_ok\n",
			   (void *)vaddr, npages);
		return -EFAULT;
	}

	pagesets = kcalloc(uctxt->expected_count, sizeof(*pagesets),
			   GFP_KERNEL);
	if (!pagesets)
		return -ENOMEM;

	/* Allocate the array of struct page pointers needed for pinning */
	pages = kcalloc(npages, sizeof(*pages), GFP_KERNEL);
	if (!pages) {
		ret = -ENOMEM;
		goto bail;
	}

	/*
	 * Pin all the pages of the user buffer. If we can't pin all the
	 * pages, accept the amount pinned so far and program only that.
	 * User space knows how to deal with partially programmed buffers.
	 */
I
Ira Weiny 已提交
397
	if (!hfi1_can_pin_pages(dd, fd->mm, fd->tid_n_pinned, npages)) {
398 399 400 401
		ret = -ENOMEM;
		goto bail;
	}

I
Ira Weiny 已提交
402
	pinned = hfi1_acquire_user_pages(fd->mm, vaddr, npages, true, pages);
403 404 405 406
	if (pinned <= 0) {
		ret = pinned;
		goto bail;
	}
407
	fd->tid_n_pinned += npages;
408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555

	/* Find sets of physically contiguous pages */
	npagesets = find_phys_blocks(pages, pinned, pagesets);

	/*
	 * We don't need to access this under a lock since tid_used is per
	 * process and the same process cannot be in hfi1_user_exp_rcv_clear()
	 * and hfi1_user_exp_rcv_setup() at the same time.
	 */
	spin_lock(&fd->tid_lock);
	if (fd->tid_used + npagesets > fd->tid_limit)
		pageset_count = fd->tid_limit - fd->tid_used;
	else
		pageset_count = npagesets;
	spin_unlock(&fd->tid_lock);

	if (!pageset_count)
		goto bail;

	ngroups = pageset_count / dd->rcv_entries.group_size;
	tidlist = kcalloc(pageset_count, sizeof(*tidlist), GFP_KERNEL);
	if (!tidlist) {
		ret = -ENOMEM;
		goto nomem;
	}

	tididx = 0;

	/*
	 * From this point on, we are going to be using shared (between master
	 * and subcontexts) context resources. We need to take the lock.
	 */
	mutex_lock(&uctxt->exp_lock);
	/*
	 * The first step is to program the RcvArray entries which are complete
	 * groups.
	 */
	while (ngroups && uctxt->tid_group_list.count) {
		struct tid_group *grp =
			tid_group_pop(&uctxt->tid_group_list);

		ret = program_rcvarray(fp, vaddr, grp, pagesets,
				       pageidx, dd->rcv_entries.group_size,
				       pages, tidlist, &tididx, &mapped);
		/*
		 * If there was a failure to program the RcvArray
		 * entries for the entire group, reset the grp fields
		 * and add the grp back to the free group list.
		 */
		if (ret <= 0) {
			tid_group_add_tail(grp, &uctxt->tid_group_list);
			hfi1_cdbg(TID,
				  "Failed to program RcvArray group %d", ret);
			goto unlock;
		}

		tid_group_add_tail(grp, &uctxt->tid_full_list);
		ngroups--;
		pageidx += ret;
		mapped_pages += mapped;
	}

	while (pageidx < pageset_count) {
		struct tid_group *grp, *ptr;
		/*
		 * If we don't have any partially used tid groups, check
		 * if we have empty groups. If so, take one from there and
		 * put in the partially used list.
		 */
		if (!uctxt->tid_used_list.count || need_group) {
			if (!uctxt->tid_group_list.count)
				goto unlock;

			grp = tid_group_pop(&uctxt->tid_group_list);
			tid_group_add_tail(grp, &uctxt->tid_used_list);
			need_group = 0;
		}
		/*
		 * There is an optimization opportunity here - instead of
		 * fitting as many page sets as we can, check for a group
		 * later on in the list that could fit all of them.
		 */
		list_for_each_entry_safe(grp, ptr, &uctxt->tid_used_list.list,
					 list) {
			unsigned use = min_t(unsigned, pageset_count - pageidx,
					     grp->size - grp->used);

			ret = program_rcvarray(fp, vaddr, grp, pagesets,
					       pageidx, use, pages, tidlist,
					       &tididx, &mapped);
			if (ret < 0) {
				hfi1_cdbg(TID,
					  "Failed to program RcvArray entries %d",
					  ret);
				ret = -EFAULT;
				goto unlock;
			} else if (ret > 0) {
				if (grp->used == grp->size)
					tid_group_move(grp,
						       &uctxt->tid_used_list,
						       &uctxt->tid_full_list);
				pageidx += ret;
				mapped_pages += mapped;
				need_group = 0;
				/* Check if we are done so we break out early */
				if (pageidx >= pageset_count)
					break;
			} else if (WARN_ON(ret == 0)) {
				/*
				 * If ret is 0, we did not program any entries
				 * into this group, which can only happen if
				 * we've screwed up the accounting somewhere.
				 * Warn and try to continue.
				 */
				need_group = 1;
			}
		}
	}
unlock:
	mutex_unlock(&uctxt->exp_lock);
nomem:
	hfi1_cdbg(TID, "total mapped: tidpairs:%u pages:%u (%d)", tididx,
		  mapped_pages, ret);
	if (tididx) {
		spin_lock(&fd->tid_lock);
		fd->tid_used += tididx;
		spin_unlock(&fd->tid_lock);
		tinfo->tidcnt = tididx;
		tinfo->length = mapped_pages * PAGE_SIZE;

		if (copy_to_user((void __user *)(unsigned long)tinfo->tidlist,
				 tidlist, sizeof(tidlist[0]) * tididx)) {
			/*
			 * On failure to copy to the user level, we need to undo
			 * everything done so far so we don't leak resources.
			 */
			tinfo->tidlist = (unsigned long)&tidlist;
			hfi1_user_exp_rcv_clear(fp, tinfo);
			tinfo->tidlist = 0;
			ret = -EFAULT;
			goto bail;
		}
	}

	/*
	 * If not everything was mapped (due to insufficient RcvArray entries,
	 * for example), unpin all unmapped pages so we can pin them nex time.
	 */
556
	if (mapped_pages != pinned) {
I
Ira Weiny 已提交
557
		hfi1_release_user_pages(fd->mm, &pages[mapped_pages],
558 559
					pinned - mapped_pages,
					false);
560 561
		fd->tid_n_pinned -= pinned - mapped_pages;
	}
562 563 564 565 566
bail:
	kfree(pagesets);
	kfree(pages);
	kfree(tidlist);
	return ret > 0 ? 0 : ret;
567 568 569 570
}

int hfi1_user_exp_rcv_clear(struct file *fp, struct hfi1_tid_info *tinfo)
{
571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604
	int ret = 0;
	struct hfi1_filedata *fd = fp->private_data;
	struct hfi1_ctxtdata *uctxt = fd->uctxt;
	u32 *tidinfo;
	unsigned tididx;

	tidinfo = kcalloc(tinfo->tidcnt, sizeof(*tidinfo), GFP_KERNEL);
	if (!tidinfo)
		return -ENOMEM;

	if (copy_from_user(tidinfo, (void __user *)(unsigned long)
			   tinfo->tidlist, sizeof(tidinfo[0]) *
			   tinfo->tidcnt)) {
		ret = -EFAULT;
		goto done;
	}

	mutex_lock(&uctxt->exp_lock);
	for (tididx = 0; tididx < tinfo->tidcnt; tididx++) {
		ret = unprogram_rcvarray(fp, tidinfo[tididx], NULL);
		if (ret) {
			hfi1_cdbg(TID, "Failed to unprogram rcv array %d",
				  ret);
			break;
		}
	}
	spin_lock(&fd->tid_lock);
	fd->tid_used -= tididx;
	spin_unlock(&fd->tid_lock);
	tinfo->tidcnt = tididx;
	mutex_unlock(&uctxt->exp_lock);
done:
	kfree(tidinfo);
	return ret;
605 606 607 608
}

int hfi1_user_exp_rcv_invalid(struct file *fp, struct hfi1_tid_info *tinfo)
{
609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655
	struct hfi1_filedata *fd = fp->private_data;
	struct hfi1_ctxtdata *uctxt = fd->uctxt;
	unsigned long *ev = uctxt->dd->events +
		(((uctxt->ctxt - uctxt->dd->first_user_ctxt) *
		  HFI1_MAX_SHARED_CTXTS) + fd->subctxt);
	u32 *array;
	int ret = 0;

	if (!fd->invalid_tids)
		return -EINVAL;

	/*
	 * copy_to_user() can sleep, which will leave the invalid_lock
	 * locked and cause the MMU notifier to be blocked on the lock
	 * for a long time.
	 * Copy the data to a local buffer so we can release the lock.
	 */
	array = kcalloc(uctxt->expected_count, sizeof(*array), GFP_KERNEL);
	if (!array)
		return -EFAULT;

	spin_lock(&fd->invalid_lock);
	if (fd->invalid_tid_idx) {
		memcpy(array, fd->invalid_tids, sizeof(*array) *
		       fd->invalid_tid_idx);
		memset(fd->invalid_tids, 0, sizeof(*fd->invalid_tids) *
		       fd->invalid_tid_idx);
		tinfo->tidcnt = fd->invalid_tid_idx;
		fd->invalid_tid_idx = 0;
		/*
		 * Reset the user flag while still holding the lock.
		 * Otherwise, PSM can miss events.
		 */
		clear_bit(_HFI1_EVENT_TID_MMU_NOTIFY_BIT, ev);
	} else {
		tinfo->tidcnt = 0;
	}
	spin_unlock(&fd->invalid_lock);

	if (tinfo->tidcnt) {
		if (copy_to_user((void __user *)tinfo->tidlist,
				 array, sizeof(*array) * tinfo->tidcnt))
			ret = -EFAULT;
	}
	kfree(array);

	return ret;
656 657
}

658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824
static u32 find_phys_blocks(struct page **pages, unsigned npages,
			    struct tid_pageset *list)
{
	unsigned pagecount, pageidx, setcount = 0, i;
	unsigned long pfn, this_pfn;

	if (!npages)
		return 0;

	/*
	 * Look for sets of physically contiguous pages in the user buffer.
	 * This will allow us to optimize Expected RcvArray entry usage by
	 * using the bigger supported sizes.
	 */
	pfn = page_to_pfn(pages[0]);
	for (pageidx = 0, pagecount = 1, i = 1; i <= npages; i++) {
		this_pfn = i < npages ? page_to_pfn(pages[i]) : 0;

		/*
		 * If the pfn's are not sequential, pages are not physically
		 * contiguous.
		 */
		if (this_pfn != ++pfn) {
			/*
			 * At this point we have to loop over the set of
			 * physically contiguous pages and break them down it
			 * sizes supported by the HW.
			 * There are two main constraints:
			 *     1. The max buffer size is MAX_EXPECTED_BUFFER.
			 *        If the total set size is bigger than that
			 *        program only a MAX_EXPECTED_BUFFER chunk.
			 *     2. The buffer size has to be a power of two. If
			 *        it is not, round down to the closes power of
			 *        2 and program that size.
			 */
			while (pagecount) {
				int maxpages = pagecount;
				u32 bufsize = pagecount * PAGE_SIZE;

				if (bufsize > MAX_EXPECTED_BUFFER)
					maxpages =
						MAX_EXPECTED_BUFFER >>
						PAGE_SHIFT;
				else if (!is_power_of_2(bufsize))
					maxpages =
						rounddown_pow_of_two(bufsize) >>
						PAGE_SHIFT;

				list[setcount].idx = pageidx;
				list[setcount].count = maxpages;
				pagecount -= maxpages;
				pageidx += maxpages;
				setcount++;
			}
			pageidx = i;
			pagecount = 1;
			pfn = this_pfn;
		} else {
			pagecount++;
		}
	}
	return setcount;
}

/**
 * program_rcvarray() - program an RcvArray group with receive buffers
 * @fp: file pointer
 * @vaddr: starting user virtual address
 * @grp: RcvArray group
 * @sets: array of struct tid_pageset holding information on physically
 *        contiguous chunks from the user buffer
 * @start: starting index into sets array
 * @count: number of struct tid_pageset's to program
 * @pages: an array of struct page * for the user buffer
 * @tidlist: the array of u32 elements when the information about the
 *           programmed RcvArray entries is to be encoded.
 * @tididx: starting offset into tidlist
 * @pmapped: (output parameter) number of pages programmed into the RcvArray
 *           entries.
 *
 * This function will program up to 'count' number of RcvArray entries from the
 * group 'grp'. To make best use of write-combining writes, the function will
 * perform writes to the unused RcvArray entries which will be ignored by the
 * HW. Each RcvArray entry will be programmed with a physically contiguous
 * buffer chunk from the user's virtual buffer.
 *
 * Return:
 * -EINVAL if the requested count is larger than the size of the group,
 * -ENOMEM or -EFAULT on error from set_rcvarray_entry(), or
 * number of RcvArray entries programmed.
 */
static int program_rcvarray(struct file *fp, unsigned long vaddr,
			    struct tid_group *grp,
			    struct tid_pageset *sets,
			    unsigned start, u16 count, struct page **pages,
			    u32 *tidlist, unsigned *tididx, unsigned *pmapped)
{
	struct hfi1_filedata *fd = fp->private_data;
	struct hfi1_ctxtdata *uctxt = fd->uctxt;
	struct hfi1_devdata *dd = uctxt->dd;
	u16 idx;
	u32 tidinfo = 0, rcventry, useidx = 0;
	int mapped = 0;

	/* Count should never be larger than the group size */
	if (count > grp->size)
		return -EINVAL;

	/* Find the first unused entry in the group */
	for (idx = 0; idx < grp->size; idx++) {
		if (!(grp->map & (1 << idx))) {
			useidx = idx;
			break;
		}
		rcv_array_wc_fill(dd, grp->base + idx);
	}

	idx = 0;
	while (idx < count) {
		u16 npages, pageidx, setidx = start + idx;
		int ret = 0;

		/*
		 * If this entry in the group is used, move to the next one.
		 * If we go past the end of the group, exit the loop.
		 */
		if (useidx >= grp->size) {
			break;
		} else if (grp->map & (1 << useidx)) {
			rcv_array_wc_fill(dd, grp->base + useidx);
			useidx++;
			continue;
		}

		rcventry = grp->base + useidx;
		npages = sets[setidx].count;
		pageidx = sets[setidx].idx;

		ret = set_rcvarray_entry(fp, vaddr + (pageidx * PAGE_SIZE),
					 rcventry, grp, pages + pageidx,
					 npages);
		if (ret)
			return ret;
		mapped += npages;

		tidinfo = rcventry2tidinfo(rcventry - uctxt->expected_base) |
			EXP_TID_SET(LEN, npages);
		tidlist[(*tididx)++] = tidinfo;
		grp->used++;
		grp->map |= 1 << useidx++;
		idx++;
	}

	/* Fill the rest of the group with "blank" writes */
	for (; useidx < grp->size; useidx++)
		rcv_array_wc_fill(dd, grp->base + useidx);
	*pmapped = mapped;
	return idx;
}

static int set_rcvarray_entry(struct file *fp, unsigned long vaddr,
			      u32 rcventry, struct tid_group *grp,
			      struct page **pages, unsigned npages)
{
	int ret;
	struct hfi1_filedata *fd = fp->private_data;
	struct hfi1_ctxtdata *uctxt = fd->uctxt;
825
	struct tid_rb_node *node;
826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847
	struct hfi1_devdata *dd = uctxt->dd;
	dma_addr_t phys;

	/*
	 * Allocate the node first so we can handle a potential
	 * failure before we've programmed anything.
	 */
	node = kzalloc(sizeof(*node) + (sizeof(struct page *) * npages),
		       GFP_KERNEL);
	if (!node)
		return -ENOMEM;

	phys = pci_map_single(dd->pcidev,
			      __va(page_to_phys(pages[0])),
			      npages * PAGE_SIZE, PCI_DMA_FROMDEVICE);
	if (dma_mapping_error(&dd->pcidev->dev, phys)) {
		dd_dev_err(dd, "Failed to DMA map Exp Rcv pages 0x%llx\n",
			   phys);
		kfree(node);
		return -EFAULT;
	}

848 849
	node->mmu.addr = vaddr;
	node->mmu.len = npages * PAGE_SIZE;
850 851 852 853 854 855 856 857
	node->phys = page_to_phys(pages[0]);
	node->npages = npages;
	node->rcventry = rcventry;
	node->dma_addr = phys;
	node->grp = grp;
	node->freed = false;
	memcpy(node->pages, pages, sizeof(struct page *) * npages);

D
Dean Luick 已提交
858
	if (!fd->handler)
859
		ret = tid_rb_insert(fd, &node->mmu);
860
	else
861
		ret = hfi1_mmu_rb_insert(fd->handler, &node->mmu);
862 863 864

	if (ret) {
		hfi1_cdbg(TID, "Failed to insert RB node %u 0x%lx, 0x%lx %d",
865
			  node->rcventry, node->mmu.addr, node->phys, ret);
866 867 868 869 870 871
		pci_unmap_single(dd->pcidev, phys, npages * PAGE_SIZE,
				 PCI_DMA_FROMDEVICE);
		kfree(node);
		return -EFAULT;
	}
	hfi1_put_tid(dd, rcventry, PT_EXPECTED, phys, ilog2(npages) + 1);
872 873
	trace_hfi1_exp_tid_reg(uctxt->ctxt, fd->subctxt, rcventry, npages,
			       node->mmu.addr, node->phys, phys);
874 875 876 877 878 879 880 881 882
	return 0;
}

static int unprogram_rcvarray(struct file *fp, u32 tidinfo,
			      struct tid_group **grp)
{
	struct hfi1_filedata *fd = fp->private_data;
	struct hfi1_ctxtdata *uctxt = fd->uctxt;
	struct hfi1_devdata *dd = uctxt->dd;
883
	struct tid_rb_node *node;
884
	u8 tidctrl = EXP_TID_GET(tidinfo, CTRL);
885
	u32 tididx = EXP_TID_GET(tidinfo, IDX) << 1, rcventry;
886 887 888 889 890 891 892 893 894 895

	if (tididx >= uctxt->expected_count) {
		dd_dev_err(dd, "Invalid RcvArray entry (%u) index for ctxt %u\n",
			   tididx, uctxt->ctxt);
		return -EINVAL;
	}

	if (tidctrl == 0x3)
		return -EINVAL;

896
	rcventry = tididx + (tidctrl - 1);
897

898
	node = fd->entry_to_rb[rcventry];
899
	if (!node || node->rcventry != (uctxt->expected_base + rcventry))
900
		return -EBADF;
D
Dean Luick 已提交
901
	if (!fd->handler)
902
		tid_rb_remove(fd, &node->mmu, fd->mm);
903
	else
904
		hfi1_mmu_rb_remove(fd->handler, &node->mmu);
905

906 907
	if (grp)
		*grp = node->grp;
908
	clear_tid_node(fd, node);
909 910 911
	return 0;
}

912
static void clear_tid_node(struct hfi1_filedata *fd, struct tid_rb_node *node)
913 914 915 916
{
	struct hfi1_ctxtdata *uctxt = fd->uctxt;
	struct hfi1_devdata *dd = uctxt->dd;

917
	trace_hfi1_exp_tid_unreg(uctxt->ctxt, fd->subctxt, node->rcventry,
918
				 node->npages, node->mmu.addr, node->phys,
919 920
				 node->dma_addr);

921 922 923 924 925 926 927
	hfi1_put_tid(dd, node->rcventry, PT_INVALID, 0, 0);
	/*
	 * Make sure device has seen the write before we unpin the
	 * pages.
	 */
	flush_wc();

928
	pci_unmap_single(dd->pcidev, node->dma_addr, node->mmu.len,
929
			 PCI_DMA_FROMDEVICE);
I
Ira Weiny 已提交
930
	hfi1_release_user_pages(fd->mm, node->pages, node->npages, true);
931
	fd->tid_n_pinned -= node->npages;
932 933 934 935 936 937 938 939 940 941 942 943 944 945

	node->grp->used--;
	node->grp->map &= ~(1 << (node->rcventry - node->grp->base));

	if (node->grp->used == node->grp->size - 1)
		tid_group_move(node->grp, &uctxt->tid_full_list,
			       &uctxt->tid_used_list);
	else if (!node->grp->used)
		tid_group_move(node->grp, &uctxt->tid_used_list,
			       &uctxt->tid_group_list);
	kfree(node);
}

static void unlock_exp_tids(struct hfi1_ctxtdata *uctxt,
946 947
			    struct exp_tid_set *set,
			    struct hfi1_filedata *fd)
948 949 950 951 952 953 954 955 956 957
{
	struct tid_group *grp, *ptr;
	int i;

	list_for_each_entry_safe(grp, ptr, &set->list, list) {
		list_del_init(&grp->list);

		for (i = 0; i < grp->size; i++) {
			if (grp->map & (1 << i)) {
				u16 rcventry = grp->base + i;
958
				struct tid_rb_node *node;
959

960 961 962
				node = fd->entry_to_rb[rcventry -
							  uctxt->expected_base];
				if (!node || node->rcventry != rcventry)
963
					continue;
D
Dean Luick 已提交
964
				if (!fd->handler)
965
					tid_rb_remove(fd, &node->mmu, fd->mm);
966
				else
967
					hfi1_mmu_rb_remove(fd->handler,
968
							   &node->mmu);
969
				clear_tid_node(fd, node);
970 971 972 973 974
			}
		}
	}
}

975
static int tid_rb_invalidate(void *arg, struct mmu_rb_node *mnode)
976
{
977
	struct hfi1_filedata *fdata = arg;
978 979 980
	struct hfi1_ctxtdata *uctxt = fdata->uctxt;
	struct tid_rb_node *node =
		container_of(mnode, struct tid_rb_node, mmu);
981

982 983
	if (node->freed)
		return 0;
984

985 986 987
	trace_hfi1_exp_tid_inval(uctxt->ctxt, fdata->subctxt, node->mmu.addr,
				 node->rcventry, node->npages, node->dma_addr);
	node->freed = true;
988

989 990 991 992 993 994 995 996
	spin_lock(&fdata->invalid_lock);
	if (fdata->invalid_tid_idx < uctxt->expected_count) {
		fdata->invalid_tids[fdata->invalid_tid_idx] =
			rcventry2tidinfo(node->rcventry - uctxt->expected_base);
		fdata->invalid_tids[fdata->invalid_tid_idx] |=
			EXP_TID_SET(LEN, node->npages);
		if (!fdata->invalid_tid_idx) {
			unsigned long *ev;
997 998

			/*
999 1000 1001 1002 1003 1004 1005
			 * hfi1_set_uevent_bits() sets a user event flag
			 * for all processes. Because calling into the
			 * driver to process TID cache invalidations is
			 * expensive and TID cache invalidations are
			 * handled on a per-process basis, we can
			 * optimize this to set the flag only for the
			 * process in question.
1006
			 */
1007 1008 1009 1010
			ev = uctxt->dd->events +
				(((uctxt->ctxt - uctxt->dd->first_user_ctxt) *
				  HFI1_MAX_SHARED_CTXTS) + fdata->subctxt);
			set_bit(_HFI1_EVENT_TID_MMU_NOTIFY_BIT, ev);
1011
		}
1012
		fdata->invalid_tid_idx++;
1013
	}
1014 1015
	spin_unlock(&fdata->invalid_lock);
	return 0;
1016 1017
}

1018
static int tid_rb_insert(void *arg, struct mmu_rb_node *node)
1019
{
1020
	struct hfi1_filedata *fdata = arg;
1021 1022
	struct tid_rb_node *tnode =
		container_of(node, struct tid_rb_node, mmu);
1023
	u32 base = fdata->uctxt->expected_base;
1024

1025
	fdata->entry_to_rb[tnode->rcventry - base] = tnode;
1026 1027 1028
	return 0;
}

1029
static void tid_rb_remove(void *arg, struct mmu_rb_node *node,
1030
			  struct mm_struct *mm)
1031
{
1032
	struct hfi1_filedata *fdata = arg;
1033 1034
	struct tid_rb_node *tnode =
		container_of(node, struct tid_rb_node, mmu);
1035 1036
	u32 base = fdata->uctxt->expected_base;

1037
	fdata->entry_to_rb[tnode->rcventry - base] = NULL;
1038
}