1 /*-
2 * Copyright (c) 2010 Isilon Systems, Inc.
3 * Copyright (c) 2016 Matthew Macy ([email protected])
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/cdefs.h>
30 __FBSDID("$FreeBSD$");
31
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/malloc.h>
35 #include <sys/kernel.h>
36 #include <sys/sysctl.h>
37 #include <sys/lock.h>
38 #include <sys/mutex.h>
39 #include <sys/rwlock.h>
40 #include <sys/proc.h>
41 #include <sys/sched.h>
42
43 #include <machine/bus.h>
44
45 #include <vm/vm.h>
46 #include <vm/pmap.h>
47 #include <vm/vm_param.h>
48 #include <vm/vm_kern.h>
49 #include <vm/vm_object.h>
50 #include <vm/vm_map.h>
51 #include <vm/vm_page.h>
52 #include <vm/vm_pageout.h>
53 #include <vm/vm_pager.h>
54 #include <vm/vm_radix.h>
55 #include <vm/vm_reserv.h>
56 #include <vm/vm_extern.h>
57
58 #include <vm/uma.h>
59 #include <vm/uma_int.h>
60
61 #include <linux/gfp.h>
62 #include <linux/mm.h>
63 #include <linux/preempt.h>
64 #include <linux/fs.h>
65
66 void
si_meminfo(struct sysinfo * si)67 si_meminfo(struct sysinfo *si)
68 {
69 si->totalram = physmem;
70 si->totalhigh = 0;
71 si->mem_unit = PAGE_SIZE;
72 }
73
74 void *
linux_page_address(struct page * page)75 linux_page_address(struct page *page)
76 {
77
78 if (page->object != kmem_object && page->object != kernel_object) {
79 return (PMAP_HAS_DMAP ?
80 ((void *)(uintptr_t)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(page))) :
81 NULL);
82 }
83 return ((void *)(uintptr_t)(VM_MIN_KERNEL_ADDRESS +
84 IDX_TO_OFF(page->pindex)));
85 }
86
87 vm_page_t
linux_alloc_pages(gfp_t flags,unsigned int order)88 linux_alloc_pages(gfp_t flags, unsigned int order)
89 {
90 vm_page_t page;
91
92 if (PMAP_HAS_DMAP) {
93 unsigned long npages = 1UL << order;
94 int req = VM_ALLOC_NOOBJ | VM_ALLOC_WIRED | VM_ALLOC_NORMAL;
95
96 if ((flags & M_ZERO) != 0)
97 req |= VM_ALLOC_ZERO;
98 if (order == 0 && (flags & GFP_DMA32) == 0) {
99 page = vm_page_alloc(NULL, 0, req);
100 if (page == NULL)
101 return (NULL);
102 } else {
103 vm_paddr_t pmax = (flags & GFP_DMA32) ?
104 BUS_SPACE_MAXADDR_32BIT : BUS_SPACE_MAXADDR;
105 retry:
106 page = vm_page_alloc_contig(NULL, 0, req,
107 npages, 0, pmax, PAGE_SIZE, 0, VM_MEMATTR_DEFAULT);
108
109 if (page == NULL) {
110 if (flags & M_WAITOK) {
111 if (!vm_page_reclaim_contig(req,
112 npages, 0, pmax, PAGE_SIZE, 0)) {
113 vm_wait(NULL);
114 }
115 flags &= ~M_WAITOK;
116 goto retry;
117 }
118 return (NULL);
119 }
120 }
121 if (flags & M_ZERO) {
122 unsigned long x;
123
124 for (x = 0; x != npages; x++) {
125 vm_page_t pgo = page + x;
126
127 if ((pgo->flags & PG_ZERO) == 0)
128 pmap_zero_page(pgo);
129 }
130 }
131 } else {
132 vm_offset_t vaddr;
133
134 vaddr = linux_alloc_kmem(flags, order);
135 if (vaddr == 0)
136 return (NULL);
137
138 page = PHYS_TO_VM_PAGE(vtophys((void *)vaddr));
139
140 KASSERT(vaddr == (vm_offset_t)page_address(page),
141 ("Page address mismatch"));
142 }
143
144 return (page);
145 }
146
147 void
linux_free_pages(vm_page_t page,unsigned int order)148 linux_free_pages(vm_page_t page, unsigned int order)
149 {
150 if (PMAP_HAS_DMAP) {
151 unsigned long npages = 1UL << order;
152 unsigned long x;
153
154 for (x = 0; x != npages; x++) {
155 vm_page_t pgo = page + x;
156
157 vm_page_lock(pgo);
158 if (vm_page_unwire_noq(pgo))
159 vm_page_free(pgo);
160 vm_page_unlock(pgo);
161 }
162 } else {
163 vm_offset_t vaddr;
164
165 vaddr = (vm_offset_t)page_address(page);
166
167 linux_free_kmem(vaddr, order);
168 }
169 }
170
171 vm_offset_t
linux_alloc_kmem(gfp_t flags,unsigned int order)172 linux_alloc_kmem(gfp_t flags, unsigned int order)
173 {
174 size_t size = ((size_t)PAGE_SIZE) << order;
175 vm_offset_t addr;
176
177 if ((flags & GFP_DMA32) == 0) {
178 addr = kmem_malloc(size, flags & GFP_NATIVE_MASK);
179 } else {
180 addr = kmem_alloc_contig(size, flags & GFP_NATIVE_MASK, 0,
181 BUS_SPACE_MAXADDR_32BIT, PAGE_SIZE, 0, VM_MEMATTR_DEFAULT);
182 }
183 return (addr);
184 }
185
186 void
linux_free_kmem(vm_offset_t addr,unsigned int order)187 linux_free_kmem(vm_offset_t addr, unsigned int order)
188 {
189 size_t size = ((size_t)PAGE_SIZE) << order;
190
191 kmem_free(addr, size);
192 }
193
194 static int
linux_get_user_pages_internal(vm_map_t map,unsigned long start,int nr_pages,int write,struct page ** pages)195 linux_get_user_pages_internal(vm_map_t map, unsigned long start, int nr_pages,
196 int write, struct page **pages)
197 {
198 vm_prot_t prot;
199 size_t len;
200 int count;
201 int i;
202
203 prot = write ? (VM_PROT_READ | VM_PROT_WRITE) : VM_PROT_READ;
204 len = ((size_t)nr_pages) << PAGE_SHIFT;
205 count = vm_fault_quick_hold_pages(map, start, len, prot, pages, nr_pages);
206 if (count == -1)
207 return (-EFAULT);
208
209 for (i = 0; i != nr_pages; i++) {
210 struct page *pg = pages[i];
211
212 vm_page_lock(pg);
213 vm_page_wire(pg);
214 vm_page_unhold(pg);
215 vm_page_unlock(pg);
216 }
217 return (nr_pages);
218 }
219
220 int
__get_user_pages_fast(unsigned long start,int nr_pages,int write,struct page ** pages)221 __get_user_pages_fast(unsigned long start, int nr_pages, int write,
222 struct page **pages)
223 {
224 vm_map_t map;
225 vm_page_t *mp;
226 vm_offset_t va;
227 vm_offset_t end;
228 vm_prot_t prot;
229 int count;
230
231 if (nr_pages == 0 || in_interrupt())
232 return (0);
233
234 MPASS(pages != NULL);
235 va = start;
236 map = &curthread->td_proc->p_vmspace->vm_map;
237 end = start + (((size_t)nr_pages) << PAGE_SHIFT);
238 if (start < vm_map_min(map) || end > vm_map_max(map))
239 return (-EINVAL);
240 prot = write ? (VM_PROT_READ | VM_PROT_WRITE) : VM_PROT_READ;
241 for (count = 0, mp = pages, va = start; va < end;
242 mp++, va += PAGE_SIZE, count++) {
243 *mp = pmap_extract_and_hold(map->pmap, va, prot);
244 if (*mp == NULL)
245 break;
246
247 vm_page_lock(*mp);
248 vm_page_wire(*mp);
249 vm_page_unhold(*mp);
250 vm_page_unlock(*mp);
251
252 if ((prot & VM_PROT_WRITE) != 0 &&
253 (*mp)->dirty != VM_PAGE_BITS_ALL) {
254 /*
255 * Explicitly dirty the physical page. Otherwise, the
256 * caller's changes may go unnoticed because they are
257 * performed through an unmanaged mapping or by a DMA
258 * operation.
259 *
260 * The object lock is not held here.
261 * See vm_page_clear_dirty_mask().
262 */
263 vm_page_dirty(*mp);
264 }
265 }
266 return (count);
267 }
268
269 long
get_user_pages_remote(struct task_struct * task,struct mm_struct * mm,unsigned long start,unsigned long nr_pages,int gup_flags,struct page ** pages,struct vm_area_struct ** vmas)270 get_user_pages_remote(struct task_struct *task, struct mm_struct *mm,
271 unsigned long start, unsigned long nr_pages, int gup_flags,
272 struct page **pages, struct vm_area_struct **vmas)
273 {
274 vm_map_t map;
275
276 map = &task->task_thread->td_proc->p_vmspace->vm_map;
277 return (linux_get_user_pages_internal(map, start, nr_pages,
278 !!(gup_flags & FOLL_WRITE), pages));
279 }
280
281 long
get_user_pages(unsigned long start,unsigned long nr_pages,int gup_flags,struct page ** pages,struct vm_area_struct ** vmas)282 get_user_pages(unsigned long start, unsigned long nr_pages, int gup_flags,
283 struct page **pages, struct vm_area_struct **vmas)
284 {
285 vm_map_t map;
286
287 map = &curthread->td_proc->p_vmspace->vm_map;
288 return (linux_get_user_pages_internal(map, start, nr_pages,
289 !!(gup_flags & FOLL_WRITE), pages));
290 }
291
292 int
is_vmalloc_addr(const void * addr)293 is_vmalloc_addr(const void *addr)
294 {
295 return (vtoslab((vm_offset_t)addr & ~UMA_SLAB_MASK) != NULL);
296 }
297
298 struct page *
linux_shmem_read_mapping_page_gfp(vm_object_t obj,int pindex,gfp_t gfp)299 linux_shmem_read_mapping_page_gfp(vm_object_t obj, int pindex, gfp_t gfp)
300 {
301 vm_page_t page;
302 int rv;
303
304 if ((gfp & GFP_NOWAIT) != 0)
305 panic("GFP_NOWAIT is unimplemented");
306
307 VM_OBJECT_WLOCK(obj);
308 page = vm_page_grab(obj, pindex, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY |
309 VM_ALLOC_WIRED);
310 if (page->valid != VM_PAGE_BITS_ALL) {
311 vm_page_xbusy(page);
312 if (vm_pager_has_page(obj, pindex, NULL, NULL)) {
313 rv = vm_pager_get_pages(obj, &page, 1, NULL, NULL);
314 if (rv != VM_PAGER_OK) {
315 vm_page_lock(page);
316 vm_page_unwire_noq(page);
317 vm_page_free(page);
318 vm_page_unlock(page);
319 VM_OBJECT_WUNLOCK(obj);
320 return (ERR_PTR(-EINVAL));
321 }
322 MPASS(page->valid == VM_PAGE_BITS_ALL);
323 } else {
324 pmap_zero_page(page);
325 page->valid = VM_PAGE_BITS_ALL;
326 page->dirty = 0;
327 }
328 vm_page_xunbusy(page);
329 }
330 VM_OBJECT_WUNLOCK(obj);
331 return (page);
332 }
333
334 struct linux_file *
linux_shmem_file_setup(const char * name,loff_t size,unsigned long flags)335 linux_shmem_file_setup(const char *name, loff_t size, unsigned long flags)
336 {
337 struct fileobj {
338 struct linux_file file __aligned(sizeof(void *));
339 struct vnode vnode __aligned(sizeof(void *));
340 };
341 struct fileobj *fileobj;
342 struct linux_file *filp;
343 struct vnode *vp;
344 int error;
345
346 fileobj = kzalloc(sizeof(*fileobj), GFP_KERNEL);
347 if (fileobj == NULL) {
348 error = -ENOMEM;
349 goto err_0;
350 }
351 filp = &fileobj->file;
352 vp = &fileobj->vnode;
353
354 filp->f_count = 1;
355 filp->f_vnode = vp;
356 filp->f_shmem = vm_pager_allocate(OBJT_DEFAULT, NULL, size,
357 VM_PROT_READ | VM_PROT_WRITE, 0, curthread->td_ucred);
358 if (filp->f_shmem == NULL) {
359 error = -ENOMEM;
360 goto err_1;
361 }
362 return (filp);
363 err_1:
364 kfree(filp);
365 err_0:
366 return (ERR_PTR(error));
367 }
368
369 static vm_ooffset_t
linux_invalidate_mapping_pages_sub(vm_object_t obj,vm_pindex_t start,vm_pindex_t end,int flags)370 linux_invalidate_mapping_pages_sub(vm_object_t obj, vm_pindex_t start,
371 vm_pindex_t end, int flags)
372 {
373 int start_count, end_count;
374
375 VM_OBJECT_WLOCK(obj);
376 start_count = obj->resident_page_count;
377 vm_object_page_remove(obj, start, end, flags);
378 end_count = obj->resident_page_count;
379 VM_OBJECT_WUNLOCK(obj);
380 return (start_count - end_count);
381 }
382
383 unsigned long
linux_invalidate_mapping_pages(vm_object_t obj,pgoff_t start,pgoff_t end)384 linux_invalidate_mapping_pages(vm_object_t obj, pgoff_t start, pgoff_t end)
385 {
386
387 return (linux_invalidate_mapping_pages_sub(obj, start, end, OBJPR_CLEANONLY));
388 }
389
390 void
linux_shmem_truncate_range(vm_object_t obj,loff_t lstart,loff_t lend)391 linux_shmem_truncate_range(vm_object_t obj, loff_t lstart, loff_t lend)
392 {
393 vm_pindex_t start = OFF_TO_IDX(lstart + PAGE_SIZE - 1);
394 vm_pindex_t end = OFF_TO_IDX(lend + 1);
395
396 (void) linux_invalidate_mapping_pages_sub(obj, start, end, 0);
397 }
398