kernel - Optimize the x86-64 lwbuf API
[dragonfly.git] / sys / kern / kern_xio.c
CommitLineData
81ee925d 1/*
8c10bfcf
MD
2 * Copyright (c) 2003,2004 The DragonFly Project. All rights reserved.
3 *
4 * This code is derived from software contributed to The DragonFly Project
5 * by Matthew Dillon <dillon@backplane.com>
6 *
81ee925d
MD
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
8c10bfcf 10 *
81ee925d
MD
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
8c10bfcf
MD
14 * notice, this list of conditions and the following disclaimer in
15 * the documentation and/or other materials provided with the
16 * distribution.
17 * 3. Neither the name of The DragonFly Project nor the names of its
18 * contributors may be used to endorse or promote products derived
19 * from this software without specific, prior written permission.
20 *
21 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
22 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
23 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
24 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE
25 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
26 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
27 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
28 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
29 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
30 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
31 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
81ee925d 32 * SUCH DAMAGE.
8c10bfcf 33 *
17cde63e 34 * $DragonFly: src/sys/kern/kern_xio.c,v 1.16 2008/05/09 07:24:45 dillon Exp $
81ee925d
MD
35 */
36/*
37 * Kernel XIO interface. An initialized XIO is basically a collection of
38 * appropriately held vm_page_t's. XIO buffers are vmspace agnostic and
39 * can represent userspace or kernelspace buffers, and can be passed to
40 * foreign threads outside of the originating vmspace. XIO buffers are
41 * not mapped into KVM and thus can be manipulated and passed around with
42 * very low overheads.
43 *
44 * The intent is for XIO to be used in the I/O path, VFS, CAPS, and other
45 * places that need to pass (possibly userspace) data between threads.
46 *
47 * TODO: check for busy page when modifying, check writeable.
48 */
49
50#include <sys/param.h>
51#include <sys/systm.h>
52#include <sys/malloc.h>
53#include <sys/proc.h>
54#include <sys/vmmeter.h>
55#include <sys/vnode.h>
56#include <sys/xio.h>
5c5185ae
SG
57
58#include <cpu/lwbuf.h>
81ee925d
MD
59
60#include <vm/vm.h>
61#include <vm/vm_param.h>
62#include <sys/lock.h>
63#include <vm/vm_kern.h>
64#include <vm/pmap.h>
65#include <vm/vm_map.h>
66#include <vm/vm_object.h>
67#include <vm/vm_page.h>
68#include <vm/vm_pageout.h>
69#include <vm/vm_pager.h>
70#include <vm/vm_extern.h>
71#include <vm/vm_page2.h>
72
73/*
d6a46bb7
MD
74 * Just do basic initialization of an empty XIO
75 */
76void
77xio_init(xio_t xio)
78{
79 xio->xio_flags = 0;
80 xio->xio_bytes = 0;
81 xio->xio_error = 0;
82 xio->xio_offset = 0;
83 xio->xio_npages = 0;
84 xio->xio_pages = xio->xio_internal_pages;
85}
86
87/*
81ee925d
MD
88 * Initialize an XIO given a userspace buffer. 0 is returned on success,
89 * an error code on failure. The actual number of bytes that could be
03aa69bd
MD
90 * accomodated in the XIO will be stored in xio_bytes and the page offset
91 * will be stored in xio_offset.
81ee925d
MD
92 */
93int
94xio_init_ubuf(xio_t xio, void *ubase, size_t ubytes, int flags)
95{
96 vm_offset_t addr;
81ee925d 97 vm_page_t m;
c4734fe7 98 vm_page_t m0;
06c5a8d6 99 int error;
81ee925d
MD
100 int i;
101 int n;
102 int vmprot;
103
104 addr = trunc_page((vm_offset_t)ubase);
105 xio->xio_flags = flags;
106 xio->xio_bytes = 0;
107 xio->xio_error = 0;
108 if (ubytes == 0) {
109 xio->xio_offset = 0;
110 xio->xio_npages = 0;
111 } else {
112 vmprot = (flags & XIOF_WRITE) ? VM_PROT_WRITE : VM_PROT_READ;
113 xio->xio_offset = (vm_offset_t)ubase & PAGE_MASK;
114 xio->xio_pages = xio->xio_internal_pages;
115 if ((n = PAGE_SIZE - xio->xio_offset) > ubytes)
116 n = ubytes;
c4734fe7 117 m0 = NULL;
81ee925d 118 for (i = 0; n && i < XIO_INTERNAL_PAGES; ++i) {
06c5a8d6
MD
119 m = vm_fault_page_quick(addr, vmprot, &error);
120 if (m == NULL)
81ee925d 121 break;
81ee925d
MD
122 xio->xio_pages[i] = m;
123 ubytes -= n;
124 xio->xio_bytes += n;
125 if ((n = ubytes) > PAGE_SIZE)
126 n = PAGE_SIZE;
127 addr += PAGE_SIZE;
c4734fe7
MD
128
129 /*
130 * Check linearity, used by syslink to memory map DMA buffers.
131 */
132 if (flags & XIOF_VMLINEAR) {
133 if (i == 0) {
134 m0 = m;
135 } else
136 if (m->object != m0->object || m->pindex != m0->pindex + i) {
137 error = EINVAL;
138 break;
139 }
140 }
81ee925d
MD
141 }
142 xio->xio_npages = i;
143
144 /*
145 * If a failure occured clean out what we loaded and return EFAULT.
17cde63e 146 * Return 0 on success. Do not dirty the pages.
81ee925d
MD
147 */
148 if (i < XIO_INTERNAL_PAGES && n) {
17cde63e 149 xio->xio_flags &= ~XIOF_WRITE;
81ee925d
MD
150 xio_release(xio);
151 xio->xio_error = EFAULT;
152 }
153 }
154 return(xio->xio_error);
155}
156
157/*
158 * Initialize an XIO given a kernelspace buffer. 0 is returned on success,
159 * an error code on failure. The actual number of bytes that could be
03aa69bd
MD
160 * accomodated in the XIO will be stored in xio_bytes and the page offset
161 * will be stored in xio_offset.
81ee925d
MD
162 */
163int
164xio_init_kbuf(xio_t xio, void *kbase, size_t kbytes)
165{
166 vm_offset_t addr;
167 vm_paddr_t paddr;
168 vm_page_t m;
169 int i;
170 int n;
171
172 addr = trunc_page((vm_offset_t)kbase);
173 xio->xio_flags = 0;
174 xio->xio_offset = (vm_offset_t)kbase & PAGE_MASK;
175 xio->xio_bytes = 0;
176 xio->xio_pages = xio->xio_internal_pages;
177 xio->xio_error = 0;
178 if ((n = PAGE_SIZE - xio->xio_offset) > kbytes)
179 n = kbytes;
573fb415
MD
180 lwkt_gettoken(&vm_token);
181 crit_enter();
81ee925d
MD
182 for (i = 0; n && i < XIO_INTERNAL_PAGES; ++i) {
183 if ((paddr = pmap_kextract(addr)) == 0)
184 break;
185 m = PHYS_TO_VM_PAGE(paddr);
186 vm_page_hold(m);
187 xio->xio_pages[i] = m;
188 kbytes -= n;
189 xio->xio_bytes += n;
190 if ((n = kbytes) > PAGE_SIZE)
191 n = PAGE_SIZE;
192 addr += PAGE_SIZE;
193 }
573fb415
MD
194 crit_exit();
195 lwkt_reltoken(&vm_token);
81ee925d
MD
196 xio->xio_npages = i;
197
198 /*
199 * If a failure occured clean out what we loaded and return EFAULT.
200 * Return 0 on success.
201 */
202 if (i < XIO_INTERNAL_PAGES && n) {
203 xio_release(xio);
204 xio->xio_error = EFAULT;
205 }
206 return(xio->xio_error);
207}
208
06ecca5a 209/*
17cde63e
MD
210 * Initialize an XIO given an array of vm_page pointers. The caller is
211 * responsible for any modified state changes for the pages.
83269e7d
MD
212 */
213int
214xio_init_pages(xio_t xio, struct vm_page **mbase, int npages, int xflags)
215{
216 int i;
217
218 KKASSERT(npages <= XIO_INTERNAL_PAGES);
219
220 xio->xio_flags = xflags;
221 xio->xio_offset = 0;
255b6068 222 xio->xio_bytes = npages * PAGE_SIZE;
83269e7d
MD
223 xio->xio_pages = xio->xio_internal_pages;
224 xio->xio_npages = npages;
225 xio->xio_error = 0;
573fb415 226 lwkt_gettoken(&vm_token);
83269e7d
MD
227 crit_enter();
228 for (i = 0; i < npages; ++i) {
229 vm_page_hold(mbase[i]);
230 xio->xio_pages[i] = mbase[i];
231 }
232 crit_exit();
573fb415 233 lwkt_reltoken(&vm_token);
83269e7d
MD
234 return(0);
235}
236
237/*
06ecca5a 238 * Cleanup an XIO so it can be destroyed. The pages associated with the
03aa69bd 239 * XIO are released.
06ecca5a 240 */
81ee925d
MD
241void
242xio_release(xio_t xio)
243{
244 int i;
245 vm_page_t m;
246
573fb415 247 lwkt_gettoken(&vm_token);
e43a034f 248 crit_enter();
81ee925d
MD
249 for (i = 0; i < xio->xio_npages; ++i) {
250 m = xio->xio_pages[i];
17cde63e
MD
251 if (xio->xio_flags & XIOF_WRITE)
252 vm_page_dirty(m);
81ee925d
MD
253 vm_page_unhold(m);
254 }
e43a034f 255 crit_exit();
573fb415 256 lwkt_reltoken(&vm_token);
81ee925d
MD
257 xio->xio_offset = 0;
258 xio->xio_npages = 0;
259 xio->xio_bytes = 0;
260 xio->xio_error = ENOBUFS;
261}
262
263/*
264 * Copy data between an XIO and a UIO. If the UIO represents userspace it
03aa69bd
MD
265 * must be relative to the current context.
266 *
267 * uoffset is the abstracted starting offset in the XIO, not the actual
268 * offset, and usually starts at 0.
269 *
270 * The XIO is not modified. The UIO is updated to reflect the copy.
81ee925d
MD
271 *
272 * UIO_READ xio -> uio
273 * UIO_WRITE uio -> xio
274 */
275int
e54488bb 276xio_uio_copy(xio_t xio, int uoffset, struct uio *uio, size_t *sizep)
81ee925d 277{
e54488bb 278 size_t bytes;
81ee925d 279 int error;
81ee925d 280
03aa69bd
MD
281 bytes = xio->xio_bytes - uoffset;
282 if (bytes > uio->uio_resid)
81ee925d 283 bytes = uio->uio_resid;
03aa69bd
MD
284 KKASSERT(bytes >= 0);
285 error = uiomove_fromphys(xio->xio_pages, xio->xio_offset + uoffset,
286 bytes, uio);
287 if (error == 0)
81ee925d 288 *sizep = bytes;
03aa69bd 289 else
81ee925d 290 *sizep = 0;
81ee925d
MD
291 return(error);
292}
293
294/*
295 * Copy the specified number of bytes from the xio to a userland
03aa69bd 296 * buffer. Return an error code or 0 on success.
81ee925d 297 *
03aa69bd
MD
298 * uoffset is the abstracted starting offset in the XIO, not the actual
299 * offset, and usually starts at 0.
300 *
301 * The XIO is not modified.
81ee925d
MD
302 */
303int
03aa69bd 304xio_copy_xtou(xio_t xio, int uoffset, void *uptr, int bytes)
81ee925d
MD
305{
306 int i;
307 int n;
308 int error;
309 int offset;
310 vm_page_t m;
5c5185ae 311 struct lwbuf *lwb;
7a683a24 312 struct lwbuf lwb_cache;
81ee925d 313
0a7648b9 314 if (uoffset + bytes > xio->xio_bytes)
81ee925d
MD
315 return(EFAULT);
316
03aa69bd 317 offset = (xio->xio_offset + uoffset) & PAGE_MASK;
82f4c82a 318 if ((n = PAGE_SIZE - offset) > bytes)
81ee925d
MD
319 n = bytes;
320
321 error = 0;
03aa69bd
MD
322 for (i = (xio->xio_offset + uoffset) >> PAGE_SHIFT;
323 i < xio->xio_npages;
324 ++i
325 ) {
81ee925d 326 m = xio->xio_pages[i];
7a683a24 327 lwb = lwbuf_alloc(m, &lwb_cache);
5c5185ae
SG
328 error = copyout((char *)lwbuf_kva(lwb) + offset, uptr, n);
329 lwbuf_free(lwb);
81ee925d
MD
330 if (error)
331 break;
332 bytes -= n;
81ee925d
MD
333 uptr = (char *)uptr + n;
334 if (bytes == 0)
335 break;
336 if ((n = bytes) > PAGE_SIZE)
337 n = PAGE_SIZE;
338 offset = 0;
339 }
340 return(error);
341}
342
343/*
344 * Copy the specified number of bytes from the xio to a kernel
345 * buffer. Return an error code or 0 on success.
346 *
03aa69bd
MD
347 * uoffset is the abstracted starting offset in the XIO, not the actual
348 * offset, and usually starts at 0.
349 *
350 * The XIO is not modified.
81ee925d
MD
351 */
352int
03aa69bd 353xio_copy_xtok(xio_t xio, int uoffset, void *kptr, int bytes)
81ee925d
MD
354{
355 int i;
356 int n;
357 int error;
358 int offset;
359 vm_page_t m;
5c5185ae 360 struct lwbuf *lwb;
7a683a24 361 struct lwbuf lwb_cache;
81ee925d 362
03aa69bd 363 if (bytes + uoffset > xio->xio_bytes)
81ee925d
MD
364 return(EFAULT);
365
03aa69bd 366 offset = (xio->xio_offset + uoffset) & PAGE_MASK;
82f4c82a 367 if ((n = PAGE_SIZE - offset) > bytes)
81ee925d
MD
368 n = bytes;
369
370 error = 0;
03aa69bd
MD
371 for (i = (xio->xio_offset + uoffset) >> PAGE_SHIFT;
372 i < xio->xio_npages;
373 ++i
374 ) {
81ee925d 375 m = xio->xio_pages[i];
7a683a24 376 lwb = lwbuf_alloc(m, &lwb_cache);
5c5185ae
SG
377 bcopy((char *)lwbuf_kva(lwb) + offset, kptr, n);
378 lwbuf_free(lwb);
81ee925d 379 bytes -= n;
81ee925d
MD
380 kptr = (char *)kptr + n;
381 if (bytes == 0)
382 break;
383 if ((n = bytes) > PAGE_SIZE)
384 n = PAGE_SIZE;
385 offset = 0;
386 }
387 return(error);
388}
389
0a7648b9
MD
390/*
391 * Copy the specified number of bytes from userland to the xio.
392 * Return an error code or 0 on success.
393 *
394 * uoffset is the abstracted starting offset in the XIO, not the actual
395 * offset, and usually starts at 0.
396 *
397 * Data in pages backing the XIO will be modified.
398 */
399int
400xio_copy_utox(xio_t xio, int uoffset, const void *uptr, int bytes)
401{
402 int i;
403 int n;
404 int error;
405 int offset;
406 vm_page_t m;
5c5185ae 407 struct lwbuf *lwb;
7a683a24 408 struct lwbuf lwb_cache;
0a7648b9
MD
409
410 if (uoffset + bytes > xio->xio_bytes)
411 return(EFAULT);
412
413 offset = (xio->xio_offset + uoffset) & PAGE_MASK;
414 if ((n = PAGE_SIZE - offset) > bytes)
415 n = bytes;
416
417 error = 0;
418 for (i = (xio->xio_offset + uoffset) >> PAGE_SHIFT;
419 i < xio->xio_npages;
420 ++i
421 ) {
422 m = xio->xio_pages[i];
7a683a24 423 lwb = lwbuf_alloc(m, &lwb_cache);
5c5185ae
SG
424 error = copyin(uptr, (char *)lwbuf_kva(lwb) + offset, n);
425 lwbuf_free(lwb);
0a7648b9
MD
426 if (error)
427 break;
428 bytes -= n;
429 uptr = (const char *)uptr + n;
430 if (bytes == 0)
431 break;
432 if ((n = bytes) > PAGE_SIZE)
433 n = PAGE_SIZE;
434 offset = 0;
435 }
436 return(error);
437}
438
439/*
440 * Copy the specified number of bytes from the kernel to the xio.
441 * Return an error code or 0 on success.
442 *
443 * uoffset is the abstracted starting offset in the XIO, not the actual
444 * offset, and usually starts at 0.
445 *
446 * Data in pages backing the XIO will be modified.
447 */
448int
449xio_copy_ktox(xio_t xio, int uoffset, const void *kptr, int bytes)
450{
451 int i;
452 int n;
453 int error;
454 int offset;
455 vm_page_t m;
5c5185ae 456 struct lwbuf *lwb;
7a683a24 457 struct lwbuf lwb_cache;
0a7648b9
MD
458
459 if (uoffset + bytes > xio->xio_bytes)
460 return(EFAULT);
461
462 offset = (xio->xio_offset + uoffset) & PAGE_MASK;
463 if ((n = PAGE_SIZE - offset) > bytes)
464 n = bytes;
465
466 error = 0;
467 for (i = (xio->xio_offset + uoffset) >> PAGE_SHIFT;
468 i < xio->xio_npages;
469 ++i
470 ) {
471 m = xio->xio_pages[i];
7a683a24 472 lwb = lwbuf_alloc(m, &lwb_cache);
5c5185ae
SG
473 bcopy(kptr, (char *)lwbuf_kva(lwb) + offset, n);
474 lwbuf_free(lwb);
0a7648b9
MD
475 bytes -= n;
476 kptr = (const char *)kptr + n;
477 if (bytes == 0)
478 break;
479 if ((n = bytes) > PAGE_SIZE)
480 n = PAGE_SIZE;
481 offset = 0;
482 }
483 return(error);
484}