xref: /illumos-gate/usr/src/uts/common/os/move.c (revision 5422785d352a2bb398daceab3d1898a8aa64d006)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
27 /*	  All Rights Reserved  	*/
28 
29 /*
30  * University Copyright- Copyright (c) 1982, 1986, 1988
31  * The Regents of the University of California
32  * All Rights Reserved
33  *
34  * University Acknowledgment- Portions of this document are derived from
35  * software developed by the University of California, Berkeley, and its
36  * contributors.
37  */
38 
39 #include <sys/types.h>
40 #include <sys/sysmacros.h>
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/uio.h>
44 #include <sys/errno.h>
45 #include <sys/vmsystm.h>
46 #include <sys/cmn_err.h>
47 #include <vm/as.h>
48 #include <vm/page.h>
49 
50 #include <sys/dcopy.h>
51 
52 int64_t uioa_maxpoll = -1;	/* <0 = noblock, 0 = block, >0 = block after */
53 #define	UIO_DCOPY_CHANNEL	0
54 #define	UIO_DCOPY_CMD		1
55 
56 /*
57  * Move "n" bytes at byte address "p"; "rw" indicates the direction
58  * of the move, and the I/O parameters are provided in "uio", which is
59  * update to reflect the data which was moved.  Returns 0 on success or
60  * a non-zero errno on failure.
61  */
62 int
63 uiomove(void *p, size_t n, enum uio_rw rw, struct uio *uio)
64 {
65 	struct iovec *iov;
66 	ulong_t cnt;
67 	int error;
68 
69 	while (n && uio->uio_resid) {
70 		iov = uio->uio_iov;
71 		cnt = MIN(iov->iov_len, n);
72 		if (cnt == 0l) {
73 			uio->uio_iov++;
74 			uio->uio_iovcnt--;
75 			continue;
76 		}
77 		switch (uio->uio_segflg) {
78 
79 		case UIO_USERSPACE:
80 		case UIO_USERISPACE:
81 			if (rw == UIO_READ) {
82 				error = xcopyout_nta(p, iov->iov_base, cnt,
83 				    (uio->uio_extflg & UIO_COPY_CACHED));
84 			} else {
85 				error = xcopyin_nta(iov->iov_base, p, cnt,
86 				    (uio->uio_extflg & UIO_COPY_CACHED));
87 			}
88 
89 			if (error)
90 				return (error);
91 			break;
92 
93 		case UIO_SYSSPACE:
94 			if (rw == UIO_READ)
95 				error = kcopy_nta(p, iov->iov_base, cnt,
96 				    (uio->uio_extflg & UIO_COPY_CACHED));
97 			else
98 				error = kcopy_nta(iov->iov_base, p, cnt,
99 				    (uio->uio_extflg & UIO_COPY_CACHED));
100 			if (error)
101 				return (error);
102 			break;
103 		}
104 		iov->iov_base += cnt;
105 		iov->iov_len -= cnt;
106 		uio->uio_resid -= cnt;
107 		uio->uio_loffset += cnt;
108 		p = (caddr_t)p + cnt;
109 		n -= cnt;
110 	}
111 	return (0);
112 }
113 
114 /*
115  * Fault in the pages of the first n bytes specified by the uio structure.
116  * 1 byte in each page is touched and the uio struct is unmodified. Any
117  * error will terminate the process as this is only a best attempt to get
118  * the pages resident.
119  */
120 void
121 uio_prefaultpages(ssize_t n, struct uio *uio)
122 {
123 	struct iovec *iov;
124 	ulong_t cnt, incr;
125 	caddr_t p;
126 	uint8_t tmp;
127 	int iovcnt;
128 
129 	iov = uio->uio_iov;
130 	iovcnt = uio->uio_iovcnt;
131 
132 	while ((n > 0) && (iovcnt > 0)) {
133 		cnt = MIN(iov->iov_len, n);
134 		if (cnt == 0) {
135 			/* empty iov entry */
136 			iov++;
137 			iovcnt--;
138 			continue;
139 		}
140 		n -= cnt;
141 		/*
142 		 * touch each page in this segment.
143 		 */
144 		p = iov->iov_base;
145 		while (cnt) {
146 			switch (uio->uio_segflg) {
147 			case UIO_USERSPACE:
148 			case UIO_USERISPACE:
149 				if (fuword8(p, &tmp))
150 					return;
151 				break;
152 			case UIO_SYSSPACE:
153 				if (kcopy(p, &tmp, 1))
154 					return;
155 				break;
156 			}
157 			incr = MIN(cnt, PAGESIZE);
158 			p += incr;
159 			cnt -= incr;
160 		}
161 		/*
162 		 * touch the last byte in case it straddles a page.
163 		 */
164 		p--;
165 		switch (uio->uio_segflg) {
166 		case UIO_USERSPACE:
167 		case UIO_USERISPACE:
168 			if (fuword8(p, &tmp))
169 				return;
170 			break;
171 		case UIO_SYSSPACE:
172 			if (kcopy(p, &tmp, 1))
173 				return;
174 			break;
175 		}
176 		iov++;
177 		iovcnt--;
178 	}
179 }
180 
181 /*
182  * same as uiomove() but doesn't modify uio structure.
183  * return in cbytes how many bytes were copied.
184  */
185 int
186 uiocopy(void *p, size_t n, enum uio_rw rw, struct uio *uio, size_t *cbytes)
187 {
188 	struct iovec *iov;
189 	ulong_t cnt;
190 	int error;
191 	int iovcnt;
192 
193 	iovcnt = uio->uio_iovcnt;
194 	*cbytes = 0;
195 
196 	for (iov = uio->uio_iov; n && iovcnt; iov++, iovcnt--) {
197 		cnt = MIN(iov->iov_len, n);
198 		if (cnt == 0)
199 			continue;
200 
201 		switch (uio->uio_segflg) {
202 
203 		case UIO_USERSPACE:
204 		case UIO_USERISPACE:
205 			if (rw == UIO_READ) {
206 				error = xcopyout_nta(p, iov->iov_base, cnt,
207 				    (uio->uio_extflg & UIO_COPY_CACHED));
208 			} else {
209 				error = xcopyin_nta(iov->iov_base, p, cnt,
210 				    (uio->uio_extflg & UIO_COPY_CACHED));
211 			}
212 
213 			if (error)
214 				return (error);
215 			break;
216 
217 		case UIO_SYSSPACE:
218 			if (rw == UIO_READ)
219 				error = kcopy_nta(p, iov->iov_base, cnt,
220 				    (uio->uio_extflg & UIO_COPY_CACHED));
221 			else
222 				error = kcopy_nta(iov->iov_base, p, cnt,
223 				    (uio->uio_extflg & UIO_COPY_CACHED));
224 			if (error)
225 				return (error);
226 			break;
227 		}
228 		p = (caddr_t)p + cnt;
229 		n -= cnt;
230 		*cbytes += cnt;
231 	}
232 	return (0);
233 }
234 
235 /*
236  * transfer a character value into the address space
237  * delineated by a uio and update fields within the
238  * uio for next character. Return 0 for success, EFAULT
239  * for error.
240  */
241 int
242 ureadc(int val, struct uio *uiop)
243 {
244 	struct iovec *iovp;
245 	unsigned char c;
246 
247 	/*
248 	 * first determine if uio is valid.  uiop should be
249 	 * non-NULL and the resid count > 0.
250 	 */
251 	if (!(uiop && uiop->uio_resid > 0))
252 		return (EFAULT);
253 
254 	/*
255 	 * scan through iovecs until one is found that is non-empty.
256 	 * Return EFAULT if none found.
257 	 */
258 	while (uiop->uio_iovcnt > 0) {
259 		iovp = uiop->uio_iov;
260 		if (iovp->iov_len <= 0) {
261 			uiop->uio_iovcnt--;
262 			uiop->uio_iov++;
263 		} else
264 			break;
265 	}
266 
267 	if (uiop->uio_iovcnt <= 0)
268 		return (EFAULT);
269 
270 	/*
271 	 * Transfer character to uio space.
272 	 */
273 
274 	c = (unsigned char) (val & 0xFF);
275 
276 	switch (uiop->uio_segflg) {
277 
278 	case UIO_USERISPACE:
279 	case UIO_USERSPACE:
280 		if (copyout(&c, iovp->iov_base, sizeof (unsigned char)))
281 			return (EFAULT);
282 		break;
283 
284 	case UIO_SYSSPACE: /* can do direct copy since kernel-kernel */
285 		*iovp->iov_base = c;
286 		break;
287 
288 	default:
289 		return (EFAULT); /* invalid segflg value */
290 	}
291 
292 	/*
293 	 * bump up/down iovec and uio members to reflect transfer.
294 	 */
295 	iovp->iov_base++;
296 	iovp->iov_len--;
297 	uiop->uio_resid--;
298 	uiop->uio_loffset++;
299 	return (0); /* success */
300 }
301 
302 /*
303  * return a character value from the address space
304  * delineated by a uio and update fields within the
305  * uio for next character. Return the character for success,
306  * -1 for error.
307  */
308 int
309 uwritec(struct uio *uiop)
310 {
311 	struct iovec *iovp;
312 	unsigned char c;
313 
314 	/*
315 	 * verify we were passed a valid uio structure.
316 	 * (1) non-NULL uiop, (2) positive resid count
317 	 * (3) there is an iovec with positive length
318 	 */
319 
320 	if (!(uiop && uiop->uio_resid > 0))
321 		return (-1);
322 
323 	while (uiop->uio_iovcnt > 0) {
324 		iovp = uiop->uio_iov;
325 		if (iovp->iov_len <= 0) {
326 			uiop->uio_iovcnt--;
327 			uiop->uio_iov++;
328 		} else
329 			break;
330 	}
331 
332 	if (uiop->uio_iovcnt <= 0)
333 		return (-1);
334 
335 	/*
336 	 * Get the character from the uio address space.
337 	 */
338 	switch (uiop->uio_segflg) {
339 
340 	case UIO_USERISPACE:
341 	case UIO_USERSPACE:
342 		if (copyin(iovp->iov_base, &c, sizeof (unsigned char)))
343 			return (-1);
344 		break;
345 
346 	case UIO_SYSSPACE:
347 		c = *iovp->iov_base;
348 		break;
349 
350 	default:
351 		return (-1); /* invalid segflg */
352 	}
353 
354 	/*
355 	 * Adjust fields of iovec and uio appropriately.
356 	 */
357 	iovp->iov_base++;
358 	iovp->iov_len--;
359 	uiop->uio_resid--;
360 	uiop->uio_loffset++;
361 	return ((int)c & 0xFF); /* success */
362 }
363 
364 /*
365  * Drop the next n chars out of *uiop.
366  */
367 void
368 uioskip(uio_t *uiop, size_t n)
369 {
370 	if (n > uiop->uio_resid)
371 		return;
372 	while (n != 0) {
373 		register iovec_t	*iovp = uiop->uio_iov;
374 		register size_t		niovb = MIN(iovp->iov_len, n);
375 
376 		if (niovb == 0) {
377 			uiop->uio_iov++;
378 			uiop->uio_iovcnt--;
379 			continue;
380 		}
381 		iovp->iov_base += niovb;
382 		uiop->uio_loffset += niovb;
383 		iovp->iov_len -= niovb;
384 		uiop->uio_resid -= niovb;
385 		n -= niovb;
386 	}
387 }
388 
389 /*
390  * Dup the suio into the duio and diovec of size diov_cnt. If diov
391  * is too small to dup suio then an error will be returned, else 0.
392  */
393 int
394 uiodup(uio_t *suio, uio_t *duio, iovec_t *diov, int diov_cnt)
395 {
396 	int ix;
397 	iovec_t *siov = suio->uio_iov;
398 
399 	*duio = *suio;
400 	for (ix = 0; ix < suio->uio_iovcnt; ix++) {
401 		diov[ix] = siov[ix];
402 		if (ix >= diov_cnt)
403 			return (1);
404 	}
405 	duio->uio_iov = diov;
406 	return (0);
407 }
408 
409 /*
410  * Shadow state for checking if a platform has hardware asynchronous
411  * copy capability and minimum copy size, e.g. Intel's I/OAT dma engine,
412  *
413  * Dcopy does a call-back to uioa_dcopy_enable() when a dma device calls
414  * into dcopy to register and uioa_dcopy_disable() when the device calls
415  * into dcopy to unregister.
416  */
417 uioasync_t uioasync = {B_FALSE, 1024};
418 
419 void
420 uioa_dcopy_enable()
421 {
422 	uioasync.enabled = B_TRUE;
423 }
424 
425 void
426 uioa_dcopy_disable()
427 {
428 	uioasync.enabled = B_FALSE;
429 }
430 
431 /*
432  * Schedule an asynchronous move of "n" bytes at byte address "p",
433  * "rw" indicates the direction of the move, I/O parameters and
434  * async state are provided in "uioa" which is update to reflect
435  * the data which is to be moved.
436  *
437  * Returns 0 on success or a non-zero errno on failure.
438  *
439  * Note, while the uioasync APIs are general purpose in design
440  * the current implementation is Intel I/OAT specific.
441  */
442 int
443 uioamove(void *p, size_t n, enum uio_rw rw, uioa_t *uioa)
444 {
445 	int		soff, doff;
446 	uint64_t	pa;
447 	int		cnt;
448 	iovec_t		*iov;
449 	dcopy_handle_t	channel;
450 	dcopy_cmd_t	cmd;
451 	int		ret = 0;
452 	int		dcopy_flags;
453 
454 	if (!(uioa->uioa_state & UIOA_ENABLED)) {
455 		/* The uioa_t isn't enabled */
456 		return (ENXIO);
457 	}
458 
459 	if (uioa->uio_segflg != UIO_USERSPACE || rw != UIO_READ) {
460 		/* Only support to user-land from kernel */
461 		return (ENOTSUP);
462 	}
463 
464 
465 	channel = uioa->uioa_hwst[UIO_DCOPY_CHANNEL];
466 	cmd = uioa->uioa_hwst[UIO_DCOPY_CMD];
467 	dcopy_flags = DCOPY_NOSLEEP;
468 
469 	/*
470 	 * While source bytes and destination bytes.
471 	 */
472 	while (n > 0 && uioa->uio_resid > 0) {
473 		iov = uioa->uio_iov;
474 		if (iov->iov_len == 0l) {
475 			uioa->uio_iov++;
476 			uioa->uio_iovcnt--;
477 			uioa->uioa_lcur++;
478 			uioa->uioa_lppp = uioa->uioa_lcur->uioa_ppp;
479 			continue;
480 		}
481 		/*
482 		 * While source bytes schedule an async
483 		 * dma for destination page by page.
484 		 */
485 		while (n > 0) {
486 			/* Addr offset in page src/dst */
487 			soff = (uintptr_t)p & PAGEOFFSET;
488 			doff = (uintptr_t)iov->iov_base & PAGEOFFSET;
489 			/* Min copy count src and dst and page sized */
490 			cnt = MIN(n, iov->iov_len);
491 			cnt = MIN(cnt, PAGESIZE - soff);
492 			cnt = MIN(cnt, PAGESIZE - doff);
493 			/* XXX if next page(s) contiguous could use multipage */
494 
495 			/*
496 			 * if we have an old command, we want to link all
497 			 * other commands to the next command we alloced so
498 			 * we only need to track the last command but can
499 			 * still free them all.
500 			 */
501 			if (cmd != NULL) {
502 				dcopy_flags |= DCOPY_ALLOC_LINK;
503 			}
504 			ret = dcopy_cmd_alloc(channel, dcopy_flags, &cmd);
505 			if (ret != DCOPY_SUCCESS) {
506 				/* Error of some sort */
507 				return (EIO);
508 			}
509 			uioa->uioa_hwst[UIO_DCOPY_CMD] = cmd;
510 
511 			ASSERT(cmd->dp_version == DCOPY_CMD_V0);
512 			if (uioa_maxpoll >= 0) {
513 				/* Blocking (>0 may be) used in uioafini() */
514 				cmd->dp_flags = DCOPY_CMD_INTR;
515 			} else {
516 				/* Non blocking uioafini() so no intr */
517 				cmd->dp_flags = DCOPY_CMD_NOFLAGS;
518 			}
519 			cmd->dp_cmd = DCOPY_CMD_COPY;
520 			pa = ptob((uint64_t)hat_getpfnum(kas.a_hat, p));
521 			cmd->dp.copy.cc_source = pa + soff;
522 			if (uioa->uioa_lcur->uioa_pfncnt == 0) {
523 				/* Have a (page_t **) */
524 				pa = ptob((uint64_t)(
525 				    *(page_t **)uioa->uioa_lppp)->p_pagenum);
526 			} else {
527 				/* Have a (pfn_t *) */
528 				pa = ptob((uint64_t)(
529 				    *(pfn_t *)uioa->uioa_lppp));
530 			}
531 			cmd->dp.copy.cc_dest = pa + doff;
532 			cmd->dp.copy.cc_size = cnt;
533 			ret = dcopy_cmd_post(cmd);
534 			if (ret != DCOPY_SUCCESS) {
535 				/* Error of some sort */
536 				return (EIO);
537 			}
538 			ret = 0;
539 
540 			/* If UIOA_POLL not set, set it */
541 			if (!(uioa->uioa_state & UIOA_POLL))
542 				uioa->uioa_state |= UIOA_POLL;
543 
544 			/* Update iov, uio, and local pointers/counters */
545 			iov->iov_base += cnt;
546 			iov->iov_len -= cnt;
547 			uioa->uio_resid -= cnt;
548 			uioa->uioa_mbytes += cnt;
549 			uioa->uio_loffset += cnt;
550 			p = (caddr_t)p + cnt;
551 			n -= cnt;
552 
553 			/* End of iovec? */
554 			if (iov->iov_len == 0) {
555 				/* Yup, next iovec */
556 				break;
557 			}
558 
559 			/* Next dst addr page? */
560 			if (doff + cnt == PAGESIZE) {
561 				/* Yup, next page_t */
562 				uioa->uioa_lppp++;
563 			}
564 		}
565 	}
566 
567 	return (ret);
568 }
569 
570 /*
571  * Initialize a uioa_t for a given uio_t for the current user context,
572  * copy the common uio_t to the uioa_t, walk the shared iovec_t and
573  * lock down the user-land page(s) containing iovec_t data, then mapin
574  * user-land pages using segkpm.
575  */
576 int
577 uioainit(uio_t *uiop, uioa_t *uioap)
578 {
579 	caddr_t	addr;
580 	page_t		**pages;
581 	int		off;
582 	int		len;
583 	proc_t		*procp = ttoproc(curthread);
584 	struct as	*as = procp->p_as;
585 	iovec_t		*iov = uiop->uio_iov;
586 	int32_t		iovcnt = uiop->uio_iovcnt;
587 	uioa_page_t	*locked = uioap->uioa_locked;
588 	dcopy_handle_t	channel;
589 	int		error;
590 
591 	if (! (uioap->uioa_state & UIOA_ALLOC)) {
592 		/* Can only init() a freshly allocated uioa_t */
593 		return (EINVAL);
594 	}
595 
596 	error = dcopy_alloc(DCOPY_NOSLEEP, &channel);
597 	if (error == DCOPY_NORESOURCES) {
598 		/* Turn off uioa */
599 		uioasync.enabled = B_FALSE;
600 		return (ENODEV);
601 	}
602 	if (error != DCOPY_SUCCESS) {
603 		/* Alloc failed */
604 		return (EIO);
605 	}
606 
607 	uioap->uioa_hwst[UIO_DCOPY_CHANNEL] = channel;
608 	uioap->uioa_hwst[UIO_DCOPY_CMD] = NULL;
609 
610 	/* Indicate uioa_t (will be) initialized */
611 	uioap->uioa_state = UIOA_INIT;
612 
613 	uioap->uioa_mbytes = 0;
614 
615 	/* uio_t/uioa_t uio_t common struct copy */
616 	*((uio_t *)uioap) = *uiop;
617 
618 	/* initialize *uiop->uio_iov */
619 	if (iovcnt > UIOA_IOV_MAX) {
620 		/* Too big? */
621 		return (E2BIG);
622 	}
623 	uioap->uio_iov = iov;
624 	uioap->uio_iovcnt = iovcnt;
625 
626 	/* Mark the uioap as such */
627 	uioap->uio_extflg |= UIO_ASYNC;
628 
629 	/*
630 	 * For each iovec_t, lock-down the page(s) backing the iovec_t
631 	 * and save the page_t list for phys addr use in uioamove().
632 	 */
633 	iov = uiop->uio_iov;
634 	iovcnt = uiop->uio_iovcnt;
635 	while (iovcnt > 0) {
636 		addr = iov->iov_base;
637 		off = (uintptr_t)addr & PAGEOFFSET;
638 		addr = (caddr_t)((uintptr_t)addr & (uintptr_t)PAGEMASK);
639 		len = iov->iov_len + off;
640 
641 		/* Lock down page(s) for the iov span */
642 		if ((error = as_pagelock(as, &pages,
643 		    iov->iov_base, iov->iov_len, S_WRITE)) != 0) {
644 			/* Error */
645 			goto cleanup;
646 		}
647 
648 		if (pages == NULL) {
649 			/*
650 			 * Need page_t list, really only need
651 			 * a pfn list so build one.
652 			 */
653 			pfn_t   *pfnp;
654 			int	pcnt = len >> PAGESHIFT;
655 
656 			if (off)
657 				pcnt++;
658 			if ((pfnp = kmem_alloc(pcnt * sizeof (pfnp),
659 			    KM_NOSLEEP)) == NULL) {
660 				error = ENOMEM;
661 				goto cleanup;
662 			}
663 			locked->uioa_ppp = (void **)pfnp;
664 			locked->uioa_pfncnt = pcnt;
665 			AS_LOCK_ENTER(as, &as->a_lock, RW_READER);
666 			while (pcnt-- > 0) {
667 				*pfnp++ = hat_getpfnum(as->a_hat, addr);
668 				addr += PAGESIZE;
669 			}
670 			AS_LOCK_EXIT(as, &as->a_lock);
671 		} else {
672 			/* Have a page_t list, save it */
673 			locked->uioa_ppp = (void **)pages;
674 			locked->uioa_pfncnt = 0;
675 		}
676 		/* Save for as_pageunlock() in uioafini() */
677 		locked->uioa_base = iov->iov_base;
678 		locked->uioa_len = iov->iov_len;
679 		locked++;
680 
681 		/* Next iovec_t */
682 		iov++;
683 		iovcnt--;
684 	}
685 	/* Initialize curret pointer into uioa_locked[] and it's uioa_ppp */
686 	uioap->uioa_lcur = uioap->uioa_locked;
687 	uioap->uioa_lppp = uioap->uioa_lcur->uioa_ppp;
688 	return (0);
689 
690 cleanup:
691 	/* Unlock any previously locked page_t(s) */
692 	while (locked > uioap->uioa_locked) {
693 		locked--;
694 		as_pageunlock(as, (page_t **)locked->uioa_ppp,
695 		    locked->uioa_base, locked->uioa_len, S_WRITE);
696 	}
697 
698 	/* Last indicate uioa_t still in alloc state */
699 	uioap->uioa_state = UIOA_ALLOC;
700 	uioap->uioa_mbytes = 0;
701 
702 	return (error);
703 }
704 
705 /*
706  * Finish processing of a uioa_t by cleanup any pending "uioap" actions.
707  */
708 int
709 uioafini(uio_t *uiop, uioa_t *uioap)
710 {
711 	int32_t		iovcnt = uiop->uio_iovcnt;
712 	uioa_page_t	*locked = uioap->uioa_locked;
713 	struct as	*as = ttoproc(curthread)->p_as;
714 	dcopy_handle_t	channel;
715 	dcopy_cmd_t	cmd;
716 	int		ret = 0;
717 
718 	ASSERT(uioap->uio_extflg & UIO_ASYNC);
719 
720 	if (!(uioap->uioa_state & (UIOA_ENABLED|UIOA_FINI))) {
721 		/* Must be an active uioa_t */
722 		return (EINVAL);
723 	}
724 
725 	channel = uioap->uioa_hwst[UIO_DCOPY_CHANNEL];
726 	cmd = uioap->uioa_hwst[UIO_DCOPY_CMD];
727 
728 	/* XXX - why do we get cmd == NULL sometimes? */
729 	if (cmd != NULL) {
730 		if (uioap->uioa_state & UIOA_POLL) {
731 			/* Wait for last dcopy() to finish */
732 			int64_t poll = 1;
733 			int poll_flag = DCOPY_POLL_NOFLAGS;
734 
735 			do {
736 				if (uioa_maxpoll == 0 ||
737 				    (uioa_maxpoll > 0 &&
738 				    poll >= uioa_maxpoll)) {
739 					/* Always block or after maxpoll */
740 					poll_flag = DCOPY_POLL_BLOCK;
741 				} else {
742 					/* No block, poll */
743 					poll++;
744 				}
745 				ret = dcopy_cmd_poll(cmd, poll_flag);
746 			} while (ret == DCOPY_PENDING);
747 
748 			if (ret == DCOPY_COMPLETED) {
749 				/* Poll/block succeeded */
750 				ret = 0;
751 			} else {
752 				/* Poll/block failed */
753 				ret = EIO;
754 			}
755 		}
756 		dcopy_cmd_free(&cmd);
757 	}
758 
759 	dcopy_free(&channel);
760 
761 	/* Unlock all page(s) iovec_t by iovec_t */
762 	while (iovcnt-- > 0) {
763 		page_t **pages;
764 
765 		if (locked->uioa_pfncnt == 0) {
766 			/* A as_pagelock() returned (page_t **) */
767 			pages = (page_t **)locked->uioa_ppp;
768 		} else {
769 			/* Our pfn_t array */
770 			pages = NULL;
771 			kmem_free(locked->uioa_ppp, locked->uioa_pfncnt *
772 			    sizeof (pfn_t *));
773 		}
774 		as_pageunlock(as, pages, locked->uioa_base, locked->uioa_len,
775 		    S_WRITE);
776 
777 		locked++;
778 	}
779 	/* uioa_t->uio_t common struct copy */
780 	*uiop = *((uio_t *)uioap);
781 
782 	/*
783 	 * Last, reset uioa state to alloc.
784 	 *
785 	 * Note, we only initialize the state here, all other members
786 	 * will be initialized in a subsequent uioainit().
787 	 */
788 	uioap->uioa_state = UIOA_ALLOC;
789 	uioap->uioa_mbytes = 0;
790 
791 	uioap->uioa_hwst[UIO_DCOPY_CMD] = NULL;
792 	uioap->uioa_hwst[UIO_DCOPY_CHANNEL] = NULL;
793 
794 	return (ret);
795 }
796