1 /*-
2 * Copyright (c) 2010 Isilon Systems, Inc.
3 * Copyright (c) 2016 Matthew Macy (mmacy@mattmacy.io)
4 * Copyright (c) 2017 Mellanox Technologies, Ltd.
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice unmodified, this list of conditions, and the following
12 * disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
20 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
21 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
22 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
23 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
24 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
25 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
26 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
27 */
28
29 #include <sys/param.h>
30 #include <sys/systm.h>
31 #include <sys/malloc.h>
32 #include <sys/kernel.h>
33 #include <sys/sysctl.h>
34 #include <sys/lock.h>
35 #include <sys/mutex.h>
36 #include <sys/rwlock.h>
37 #include <sys/proc.h>
38 #include <sys/sched.h>
39 #include <sys/memrange.h>
40
41 #include <machine/bus.h>
42
43 #include <vm/vm.h>
44 #include <vm/pmap.h>
45 #include <vm/vm_param.h>
46 #include <vm/vm_kern.h>
47 #include <vm/vm_object.h>
48 #include <vm/vm_map.h>
49 #include <vm/vm_page.h>
50 #include <vm/vm_pageout.h>
51 #include <vm/vm_pager.h>
52 #include <vm/vm_radix.h>
53 #include <vm/vm_reserv.h>
54 #include <vm/vm_extern.h>
55
56 #include <vm/uma.h>
57 #include <vm/uma_int.h>
58
59 #include <linux/gfp.h>
60 #include <linux/mm.h>
61 #include <linux/preempt.h>
62 #include <linux/fs.h>
63 #include <linux/shmem_fs.h>
64 #include <linux/kernel.h>
65 #include <linux/idr.h>
66 #include <linux/io.h>
67 #include <linux/io-mapping.h>
68
69 #ifdef __i386__
70 DEFINE_IDR(mtrr_idr);
71 static MALLOC_DEFINE(M_LKMTRR, "idr", "Linux MTRR compat");
72 extern int pat_works;
73 #endif
74
75 void
si_meminfo(struct sysinfo * si)76 si_meminfo(struct sysinfo *si)
77 {
78 si->totalram = physmem;
79 si->freeram = vm_free_count();
80 si->totalhigh = 0;
81 si->freehigh = 0;
82 si->mem_unit = PAGE_SIZE;
83 }
84
85 void *
linux_page_address(struct page * page)86 linux_page_address(struct page *page)
87 {
88
89 if (page->object != kernel_object) {
90 return (PMAP_HAS_DMAP ?
91 ((void *)(uintptr_t)PHYS_TO_DMAP(page_to_phys(page))) :
92 NULL);
93 }
94 return ((void *)(uintptr_t)(VM_MIN_KERNEL_ADDRESS +
95 IDX_TO_OFF(page->pindex)));
96 }
97
98 struct page *
linux_alloc_pages(gfp_t flags,unsigned int order)99 linux_alloc_pages(gfp_t flags, unsigned int order)
100 {
101 struct page *page;
102
103 if (PMAP_HAS_DMAP) {
104 unsigned long npages = 1UL << order;
105 int req = VM_ALLOC_WIRED;
106
107 if ((flags & M_ZERO) != 0)
108 req |= VM_ALLOC_ZERO;
109 if (order == 0 && (flags & GFP_DMA32) == 0) {
110 page = vm_page_alloc_noobj(req);
111 if (page == NULL)
112 return (NULL);
113 } else {
114 vm_paddr_t pmax = (flags & GFP_DMA32) ?
115 BUS_SPACE_MAXADDR_32BIT : BUS_SPACE_MAXADDR;
116 retry:
117 page = vm_page_alloc_noobj_contig(req, npages, 0, pmax,
118 PAGE_SIZE, 0, VM_MEMATTR_DEFAULT);
119 if (page == NULL) {
120 if ((flags & (M_WAITOK | __GFP_NORETRY)) ==
121 M_WAITOK) {
122 int err = vm_page_reclaim_contig(req,
123 npages, 0, pmax, PAGE_SIZE, 0);
124 if (err == ENOMEM)
125 vm_wait(NULL);
126 else if (err != 0)
127 return (NULL);
128 flags &= ~M_WAITOK;
129 goto retry;
130 }
131 return (NULL);
132 }
133 }
134 } else {
135 vm_offset_t vaddr;
136
137 vaddr = linux_alloc_kmem(flags, order);
138 if (vaddr == 0)
139 return (NULL);
140
141 page = virt_to_page((void *)vaddr);
142
143 KASSERT(vaddr == (vm_offset_t)page_address(page),
144 ("Page address mismatch"));
145 }
146
147 return (page);
148 }
149
150 static void
_linux_free_kmem(vm_offset_t addr,unsigned int order)151 _linux_free_kmem(vm_offset_t addr, unsigned int order)
152 {
153 size_t size = ((size_t)PAGE_SIZE) << order;
154
155 kmem_free((void *)addr, size);
156 }
157
158 void
linux_free_pages(struct page * page,unsigned int order)159 linux_free_pages(struct page *page, unsigned int order)
160 {
161 if (PMAP_HAS_DMAP) {
162 unsigned long npages = 1UL << order;
163 unsigned long x;
164
165 for (x = 0; x != npages; x++) {
166 vm_page_t pgo = page + x;
167
168 if (vm_page_unwire_noq(pgo))
169 vm_page_free(pgo);
170 }
171 } else {
172 vm_offset_t vaddr;
173
174 vaddr = (vm_offset_t)page_address(page);
175
176 _linux_free_kmem(vaddr, order);
177 }
178 }
179
180 vm_offset_t
linux_alloc_kmem(gfp_t flags,unsigned int order)181 linux_alloc_kmem(gfp_t flags, unsigned int order)
182 {
183 size_t size = ((size_t)PAGE_SIZE) << order;
184 void *addr;
185
186 addr = kmem_alloc_contig(size, flags & GFP_NATIVE_MASK, 0,
187 ((flags & GFP_DMA32) == 0) ? -1UL : BUS_SPACE_MAXADDR_32BIT,
188 PAGE_SIZE, 0, VM_MEMATTR_DEFAULT);
189
190 return ((vm_offset_t)addr);
191 }
192
193 void
linux_free_kmem(vm_offset_t addr,unsigned int order)194 linux_free_kmem(vm_offset_t addr, unsigned int order)
195 {
196 KASSERT((addr & ~PAGE_MASK) == 0,
197 ("%s: addr %p is not page aligned", __func__, (void *)addr));
198
199 if (addr >= VM_MIN_KERNEL_ADDRESS && addr < VM_MAX_KERNEL_ADDRESS) {
200 _linux_free_kmem(addr, order);
201 } else {
202 vm_page_t page;
203
204 page = PHYS_TO_VM_PAGE(DMAP_TO_PHYS(addr));
205 linux_free_pages(page, order);
206 }
207 }
208
209 static int
linux_get_user_pages_internal(vm_map_t map,unsigned long start,int nr_pages,int write,struct page ** pages)210 linux_get_user_pages_internal(vm_map_t map, unsigned long start, int nr_pages,
211 int write, struct page **pages)
212 {
213 vm_prot_t prot;
214 size_t len;
215 int count;
216
217 prot = write ? (VM_PROT_READ | VM_PROT_WRITE) : VM_PROT_READ;
218 len = ptoa((vm_offset_t)nr_pages);
219 count = vm_fault_quick_hold_pages(map, start, len, prot, pages, nr_pages);
220 return (count == -1 ? -EFAULT : nr_pages);
221 }
222
223 int
__get_user_pages_fast(unsigned long start,int nr_pages,int write,struct page ** pages)224 __get_user_pages_fast(unsigned long start, int nr_pages, int write,
225 struct page **pages)
226 {
227 vm_map_t map;
228 vm_page_t *mp;
229 vm_offset_t va;
230 vm_offset_t end;
231 vm_prot_t prot;
232 int count;
233
234 if (nr_pages == 0 || in_interrupt())
235 return (0);
236
237 MPASS(pages != NULL);
238 map = &curthread->td_proc->p_vmspace->vm_map;
239 end = start + ptoa((vm_offset_t)nr_pages);
240 if (!vm_map_range_valid(map, start, end))
241 return (-EINVAL);
242 prot = write ? (VM_PROT_READ | VM_PROT_WRITE) : VM_PROT_READ;
243 for (count = 0, mp = pages, va = start; va < end;
244 mp++, va += PAGE_SIZE, count++) {
245 *mp = pmap_extract_and_hold(map->pmap, va, prot);
246 if (*mp == NULL)
247 break;
248
249 if ((prot & VM_PROT_WRITE) != 0 &&
250 (*mp)->dirty != VM_PAGE_BITS_ALL) {
251 /*
252 * Explicitly dirty the physical page. Otherwise, the
253 * caller's changes may go unnoticed because they are
254 * performed through an unmanaged mapping or by a DMA
255 * operation.
256 *
257 * The object lock is not held here.
258 * See vm_page_clear_dirty_mask().
259 */
260 vm_page_dirty(*mp);
261 }
262 }
263 return (count);
264 }
265
266 long
get_user_pages_remote(struct task_struct * task,struct mm_struct * mm,unsigned long start,unsigned long nr_pages,unsigned int gup_flags,struct page ** pages,struct vm_area_struct ** vmas)267 get_user_pages_remote(struct task_struct *task, struct mm_struct *mm,
268 unsigned long start, unsigned long nr_pages, unsigned int gup_flags,
269 struct page **pages, struct vm_area_struct **vmas)
270 {
271 vm_map_t map;
272
273 map = &task->task_thread->td_proc->p_vmspace->vm_map;
274 return (linux_get_user_pages_internal(map, start, nr_pages,
275 !!(gup_flags & FOLL_WRITE), pages));
276 }
277
278 long
lkpi_get_user_pages(unsigned long start,unsigned long nr_pages,unsigned int gup_flags,struct page ** pages)279 lkpi_get_user_pages(unsigned long start, unsigned long nr_pages,
280 unsigned int gup_flags, struct page **pages)
281 {
282 vm_map_t map;
283
284 map = &curthread->td_proc->p_vmspace->vm_map;
285 return (linux_get_user_pages_internal(map, start, nr_pages,
286 !!(gup_flags & FOLL_WRITE), pages));
287 }
288
289 int
is_vmalloc_addr(const void * addr)290 is_vmalloc_addr(const void *addr)
291 {
292 return (vtoslab((vm_offset_t)addr & ~UMA_SLAB_MASK) != NULL);
293 }
294
295 vm_fault_t
lkpi_vmf_insert_pfn_prot_locked(struct vm_area_struct * vma,unsigned long addr,unsigned long pfn,pgprot_t prot)296 lkpi_vmf_insert_pfn_prot_locked(struct vm_area_struct *vma, unsigned long addr,
297 unsigned long pfn, pgprot_t prot)
298 {
299 vm_object_t vm_obj = vma->vm_obj;
300 vm_object_t tmp_obj;
301 vm_page_t page;
302 vm_pindex_t pindex;
303
304 VM_OBJECT_ASSERT_WLOCKED(vm_obj);
305 pindex = OFF_TO_IDX(addr - vma->vm_start);
306 if (vma->vm_pfn_count == 0)
307 vma->vm_pfn_first = pindex;
308 MPASS(pindex <= OFF_TO_IDX(vma->vm_end));
309
310 retry:
311 page = vm_page_grab(vm_obj, pindex, VM_ALLOC_NOCREAT);
312 if (page == NULL) {
313 page = PHYS_TO_VM_PAGE(IDX_TO_OFF(pfn));
314 if (!vm_page_busy_acquire(page, VM_ALLOC_WAITFAIL))
315 goto retry;
316 if (page->object != NULL) {
317 tmp_obj = page->object;
318 vm_page_xunbusy(page);
319 VM_OBJECT_WUNLOCK(vm_obj);
320 VM_OBJECT_WLOCK(tmp_obj);
321 if (page->object == tmp_obj &&
322 vm_page_busy_acquire(page, VM_ALLOC_WAITFAIL)) {
323 KASSERT(page->object == tmp_obj,
324 ("page has changed identity"));
325 KASSERT((page->oflags & VPO_UNMANAGED) == 0,
326 ("page does not belong to shmem"));
327 vm_pager_page_unswapped(page);
328 if (pmap_page_is_mapped(page)) {
329 vm_page_xunbusy(page);
330 VM_OBJECT_WUNLOCK(tmp_obj);
331 printf("%s: page rename failed: page "
332 "is mapped\n", __func__);
333 VM_OBJECT_WLOCK(vm_obj);
334 return (VM_FAULT_NOPAGE);
335 }
336 vm_page_remove(page);
337 }
338 VM_OBJECT_WUNLOCK(tmp_obj);
339 VM_OBJECT_WLOCK(vm_obj);
340 goto retry;
341 }
342 if (vm_page_insert(page, vm_obj, pindex)) {
343 vm_page_xunbusy(page);
344 return (VM_FAULT_OOM);
345 }
346 vm_page_valid(page);
347 }
348 pmap_page_set_memattr(page, pgprot2cachemode(prot));
349 vma->vm_pfn_count++;
350
351 return (VM_FAULT_NOPAGE);
352 }
353
354 int
lkpi_remap_pfn_range(struct vm_area_struct * vma,unsigned long start_addr,unsigned long start_pfn,unsigned long size,pgprot_t prot)355 lkpi_remap_pfn_range(struct vm_area_struct *vma, unsigned long start_addr,
356 unsigned long start_pfn, unsigned long size, pgprot_t prot)
357 {
358 vm_object_t vm_obj;
359 unsigned long addr, pfn;
360 int err = 0;
361
362 vm_obj = vma->vm_obj;
363
364 VM_OBJECT_WLOCK(vm_obj);
365 for (addr = start_addr, pfn = start_pfn;
366 addr < start_addr + size;
367 addr += PAGE_SIZE) {
368 vm_fault_t ret;
369 retry:
370 ret = lkpi_vmf_insert_pfn_prot_locked(vma, addr, pfn, prot);
371
372 if ((ret & VM_FAULT_OOM) != 0) {
373 VM_OBJECT_WUNLOCK(vm_obj);
374 vm_wait(NULL);
375 VM_OBJECT_WLOCK(vm_obj);
376 goto retry;
377 }
378
379 if ((ret & VM_FAULT_ERROR) != 0) {
380 err = -EFAULT;
381 break;
382 }
383
384 pfn++;
385 }
386 VM_OBJECT_WUNLOCK(vm_obj);
387
388 if (unlikely(err)) {
389 zap_vma_ptes(vma, start_addr,
390 (pfn - start_pfn) << PAGE_SHIFT);
391 return (err);
392 }
393
394 return (0);
395 }
396
397 int
lkpi_io_mapping_map_user(struct io_mapping * iomap,struct vm_area_struct * vma,unsigned long addr,unsigned long pfn,unsigned long size)398 lkpi_io_mapping_map_user(struct io_mapping *iomap,
399 struct vm_area_struct *vma, unsigned long addr,
400 unsigned long pfn, unsigned long size)
401 {
402 pgprot_t prot;
403 int ret;
404
405 prot = cachemode2protval(iomap->attr);
406 ret = lkpi_remap_pfn_range(vma, addr, pfn, size, prot);
407
408 return (ret);
409 }
410
411 /*
412 * Although FreeBSD version of unmap_mapping_range has semantics and types of
413 * parameters compatible with Linux version, the values passed in are different
414 * @obj should match to vm_private_data field of vm_area_struct returned by
415 * mmap file operation handler, see linux_file_mmap_single() sources
416 * @holelen should match to size of area to be munmapped.
417 */
418 void
lkpi_unmap_mapping_range(void * obj,loff_t const holebegin __unused,loff_t const holelen __unused,int even_cows __unused)419 lkpi_unmap_mapping_range(void *obj, loff_t const holebegin __unused,
420 loff_t const holelen __unused, int even_cows __unused)
421 {
422 vm_object_t devobj;
423
424 devobj = cdev_pager_lookup(obj);
425 if (devobj != NULL) {
426 cdev_mgtdev_pager_free_pages(devobj);
427 vm_object_deallocate(devobj);
428 }
429 }
430
431 int
lkpi_arch_phys_wc_add(unsigned long base,unsigned long size)432 lkpi_arch_phys_wc_add(unsigned long base, unsigned long size)
433 {
434 #ifdef __i386__
435 struct mem_range_desc *mrdesc;
436 int error, id, act;
437
438 /* If PAT is available, do nothing */
439 if (pat_works)
440 return (0);
441
442 mrdesc = malloc(sizeof(*mrdesc), M_LKMTRR, M_WAITOK);
443 mrdesc->mr_base = base;
444 mrdesc->mr_len = size;
445 mrdesc->mr_flags = MDF_WRITECOMBINE;
446 strlcpy(mrdesc->mr_owner, "drm", sizeof(mrdesc->mr_owner));
447 act = MEMRANGE_SET_UPDATE;
448 error = mem_range_attr_set(mrdesc, &act);
449 if (error == 0) {
450 error = idr_get_new(&mtrr_idr, mrdesc, &id);
451 MPASS(idr_find(&mtrr_idr, id) == mrdesc);
452 if (error != 0) {
453 act = MEMRANGE_SET_REMOVE;
454 mem_range_attr_set(mrdesc, &act);
455 }
456 }
457 if (error != 0) {
458 free(mrdesc, M_LKMTRR);
459 pr_warn(
460 "Failed to add WC MTRR for [%p-%p]: %d; "
461 "performance may suffer\n",
462 (void *)base, (void *)(base + size - 1), error);
463 } else
464 pr_warn("Successfully added WC MTRR for [%p-%p]\n",
465 (void *)base, (void *)(base + size - 1));
466
467 return (error != 0 ? -error : id + __MTRR_ID_BASE);
468 #else
469 return (0);
470 #endif
471 }
472
473 void
lkpi_arch_phys_wc_del(int reg)474 lkpi_arch_phys_wc_del(int reg)
475 {
476 #ifdef __i386__
477 struct mem_range_desc *mrdesc;
478 int act;
479
480 /* Check if arch_phys_wc_add() failed. */
481 if (reg < __MTRR_ID_BASE)
482 return;
483
484 mrdesc = idr_find(&mtrr_idr, reg - __MTRR_ID_BASE);
485 MPASS(mrdesc != NULL);
486 idr_remove(&mtrr_idr, reg - __MTRR_ID_BASE);
487 act = MEMRANGE_SET_REMOVE;
488 mem_range_attr_set(mrdesc, &act);
489 free(mrdesc, M_LKMTRR);
490 #endif
491 }
492
493 /*
494 * This is a highly simplified version of the Linux page_frag_cache.
495 * We only support up-to 1 single page as fragment size and we will
496 * always return a full page. This may be wasteful on small objects
497 * but the only known consumer (mt76) is either asking for a half-page
498 * or a full page. If this was to become a problem we can implement
499 * a more elaborate version.
500 */
501 void *
linuxkpi_page_frag_alloc(struct page_frag_cache * pfc,size_t fragsz,gfp_t gfp)502 linuxkpi_page_frag_alloc(struct page_frag_cache *pfc,
503 size_t fragsz, gfp_t gfp)
504 {
505 vm_page_t pages;
506
507 if (fragsz == 0)
508 return (NULL);
509
510 KASSERT(fragsz <= PAGE_SIZE, ("%s: fragsz %zu > PAGE_SIZE not yet "
511 "supported", __func__, fragsz));
512
513 pages = alloc_pages(gfp, flsl(howmany(fragsz, PAGE_SIZE) - 1));
514 if (pages == NULL)
515 return (NULL);
516 pfc->va = linux_page_address(pages);
517
518 /* Passed in as "count" to __page_frag_cache_drain(). Unused by us. */
519 pfc->pagecnt_bias = 0;
520
521 return (pfc->va);
522 }
523
524 void
linuxkpi_page_frag_free(void * addr)525 linuxkpi_page_frag_free(void *addr)
526 {
527 vm_page_t page;
528
529 page = virt_to_page(addr);
530 linux_free_pages(page, 0);
531 }
532
533 void
linuxkpi__page_frag_cache_drain(struct page * page,size_t count __unused)534 linuxkpi__page_frag_cache_drain(struct page *page, size_t count __unused)
535 {
536
537 linux_free_pages(page, 0);
538 }
539