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 #include <sys/memrange.h>
43
44 #include <machine/bus.h>
45
46 #include <vm/vm.h>
47 #include <vm/pmap.h>
48 #include <vm/vm_param.h>
49 #include <vm/vm_kern.h>
50 #include <vm/vm_object.h>
51 #include <vm/vm_map.h>
52 #include <vm/vm_page.h>
53 #include <vm/vm_pageout.h>
54 #include <vm/vm_pager.h>
55 #include <vm/vm_radix.h>
56 #include <vm/vm_reserv.h>
57 #include <vm/vm_extern.h>
58
59 #include <vm/uma.h>
60 #include <vm/uma_int.h>
61
62 #include <linux/gfp.h>
63 #include <linux/mm.h>
64 #include <linux/preempt.h>
65 #include <linux/fs.h>
66 #include <linux/shmem_fs.h>
67 #include <linux/kernel.h>
68 #include <linux/idr.h>
69 #include <linux/io.h>
70
71 #ifdef __i386__
72 DEFINE_IDR(mtrr_idr);
73 static MALLOC_DEFINE(M_LKMTRR, "idr", "Linux MTRR compat");
74 extern int pat_works;
75 #endif
76
77 void
si_meminfo(struct sysinfo * si)78 si_meminfo(struct sysinfo *si)
79 {
80 si->totalram = physmem;
81 si->totalhigh = 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(VM_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 vm_page_t
linux_alloc_pages(gfp_t flags,unsigned int order)99 linux_alloc_pages(gfp_t flags, unsigned int order)
100 {
101 vm_page_t 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) {
121 if (!vm_page_reclaim_contig(req,
122 npages, 0, pmax, PAGE_SIZE, 0)) {
123 vm_wait(NULL);
124 }
125 flags &= ~M_WAITOK;
126 goto retry;
127 }
128 return (NULL);
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 if (vm_page_unwire_noq(pgo))
158 vm_page_free(pgo);
159 }
160 } else {
161 vm_offset_t vaddr;
162
163 vaddr = (vm_offset_t)page_address(page);
164
165 linux_free_kmem(vaddr, order);
166 }
167 }
168
169 vm_offset_t
linux_alloc_kmem(gfp_t flags,unsigned int order)170 linux_alloc_kmem(gfp_t flags, unsigned int order)
171 {
172 size_t size = ((size_t)PAGE_SIZE) << order;
173 vm_offset_t addr;
174
175 if ((flags & GFP_DMA32) == 0) {
176 addr = kmem_malloc(size, flags & GFP_NATIVE_MASK);
177 } else {
178 addr = kmem_alloc_contig(size, flags & GFP_NATIVE_MASK, 0,
179 BUS_SPACE_MAXADDR_32BIT, PAGE_SIZE, 0, VM_MEMATTR_DEFAULT);
180 }
181 return (addr);
182 }
183
184 void
linux_free_kmem(vm_offset_t addr,unsigned int order)185 linux_free_kmem(vm_offset_t addr, unsigned int order)
186 {
187 size_t size = ((size_t)PAGE_SIZE) << order;
188
189 kmem_free(addr, size);
190 }
191
192 static int
linux_get_user_pages_internal(vm_map_t map,unsigned long start,int nr_pages,int write,struct page ** pages)193 linux_get_user_pages_internal(vm_map_t map, unsigned long start, int nr_pages,
194 int write, struct page **pages)
195 {
196 vm_prot_t prot;
197 size_t len;
198 int count;
199
200 prot = write ? (VM_PROT_READ | VM_PROT_WRITE) : VM_PROT_READ;
201 len = ptoa((vm_offset_t)nr_pages);
202 count = vm_fault_quick_hold_pages(map, start, len, prot, pages, nr_pages);
203 return (count == -1 ? -EFAULT : nr_pages);
204 }
205
206 int
__get_user_pages_fast(unsigned long start,int nr_pages,int write,struct page ** pages)207 __get_user_pages_fast(unsigned long start, int nr_pages, int write,
208 struct page **pages)
209 {
210 vm_map_t map;
211 vm_page_t *mp;
212 vm_offset_t va;
213 vm_offset_t end;
214 vm_prot_t prot;
215 int count;
216
217 if (nr_pages == 0 || in_interrupt())
218 return (0);
219
220 MPASS(pages != NULL);
221 map = &curthread->td_proc->p_vmspace->vm_map;
222 end = start + ptoa((vm_offset_t)nr_pages);
223 if (!vm_map_range_valid(map, start, end))
224 return (-EINVAL);
225 prot = write ? (VM_PROT_READ | VM_PROT_WRITE) : VM_PROT_READ;
226 for (count = 0, mp = pages, va = start; va < end;
227 mp++, va += PAGE_SIZE, count++) {
228 *mp = pmap_extract_and_hold(map->pmap, va, prot);
229 if (*mp == NULL)
230 break;
231
232 if ((prot & VM_PROT_WRITE) != 0 &&
233 (*mp)->dirty != VM_PAGE_BITS_ALL) {
234 /*
235 * Explicitly dirty the physical page. Otherwise, the
236 * caller's changes may go unnoticed because they are
237 * performed through an unmanaged mapping or by a DMA
238 * operation.
239 *
240 * The object lock is not held here.
241 * See vm_page_clear_dirty_mask().
242 */
243 vm_page_dirty(*mp);
244 }
245 }
246 return (count);
247 }
248
249 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)250 get_user_pages_remote(struct task_struct *task, struct mm_struct *mm,
251 unsigned long start, unsigned long nr_pages, int gup_flags,
252 struct page **pages, struct vm_area_struct **vmas)
253 {
254 vm_map_t map;
255
256 map = &task->task_thread->td_proc->p_vmspace->vm_map;
257 return (linux_get_user_pages_internal(map, start, nr_pages,
258 !!(gup_flags & FOLL_WRITE), pages));
259 }
260
261 long
get_user_pages(unsigned long start,unsigned long nr_pages,int gup_flags,struct page ** pages,struct vm_area_struct ** vmas)262 get_user_pages(unsigned long start, unsigned long nr_pages, int gup_flags,
263 struct page **pages, struct vm_area_struct **vmas)
264 {
265 vm_map_t map;
266
267 map = &curthread->td_proc->p_vmspace->vm_map;
268 return (linux_get_user_pages_internal(map, start, nr_pages,
269 !!(gup_flags & FOLL_WRITE), pages));
270 }
271
272 int
is_vmalloc_addr(const void * addr)273 is_vmalloc_addr(const void *addr)
274 {
275 return (vtoslab((vm_offset_t)addr & ~UMA_SLAB_MASK) != NULL);
276 }
277
278 vm_fault_t
lkpi_vmf_insert_pfn_prot_locked(struct vm_area_struct * vma,unsigned long addr,unsigned long pfn,pgprot_t prot)279 lkpi_vmf_insert_pfn_prot_locked(struct vm_area_struct *vma, unsigned long addr,
280 unsigned long pfn, pgprot_t prot)
281 {
282 vm_object_t vm_obj = vma->vm_obj;
283 vm_object_t tmp_obj;
284 vm_page_t page;
285 vm_pindex_t pindex;
286
287 VM_OBJECT_ASSERT_WLOCKED(vm_obj);
288 pindex = OFF_TO_IDX(addr - vma->vm_start);
289 if (vma->vm_pfn_count == 0)
290 vma->vm_pfn_first = pindex;
291 MPASS(pindex <= OFF_TO_IDX(vma->vm_end));
292
293 retry:
294 page = vm_page_grab(vm_obj, pindex, VM_ALLOC_NOCREAT);
295 if (page == NULL) {
296 page = PHYS_TO_VM_PAGE(IDX_TO_OFF(pfn));
297 if (!vm_page_busy_acquire(page, VM_ALLOC_WAITFAIL))
298 goto retry;
299 if (page->object != NULL) {
300 tmp_obj = page->object;
301 vm_page_xunbusy(page);
302 VM_OBJECT_WUNLOCK(vm_obj);
303 VM_OBJECT_WLOCK(tmp_obj);
304 if (page->object == tmp_obj &&
305 vm_page_busy_acquire(page, VM_ALLOC_WAITFAIL)) {
306 KASSERT(page->object == tmp_obj,
307 ("page has changed identity"));
308 KASSERT((page->oflags & VPO_UNMANAGED) == 0,
309 ("page does not belong to shmem"));
310 vm_pager_page_unswapped(page);
311 if (pmap_page_is_mapped(page)) {
312 vm_page_xunbusy(page);
313 VM_OBJECT_WUNLOCK(tmp_obj);
314 printf("%s: page rename failed: page "
315 "is mapped\n", __func__);
316 VM_OBJECT_WLOCK(vm_obj);
317 return (VM_FAULT_NOPAGE);
318 }
319 vm_page_remove(page);
320 }
321 VM_OBJECT_WUNLOCK(tmp_obj);
322 VM_OBJECT_WLOCK(vm_obj);
323 goto retry;
324 }
325 if (vm_page_insert(page, vm_obj, pindex)) {
326 vm_page_xunbusy(page);
327 return (VM_FAULT_OOM);
328 }
329 vm_page_valid(page);
330 }
331 pmap_page_set_memattr(page, pgprot2cachemode(prot));
332 vma->vm_pfn_count++;
333
334 return (VM_FAULT_NOPAGE);
335 }
336
337 /*
338 * Although FreeBSD version of unmap_mapping_range has semantics and types of
339 * parameters compatible with Linux version, the values passed in are different
340 * @obj should match to vm_private_data field of vm_area_struct returned by
341 * mmap file operation handler, see linux_file_mmap_single() sources
342 * @holelen should match to size of area to be munmapped.
343 */
344 void
lkpi_unmap_mapping_range(void * obj,loff_t const holebegin __unused,loff_t const holelen,int even_cows __unused)345 lkpi_unmap_mapping_range(void *obj, loff_t const holebegin __unused,
346 loff_t const holelen, int even_cows __unused)
347 {
348 vm_object_t devobj;
349 vm_page_t page;
350 int i, page_count;
351
352 devobj = cdev_pager_lookup(obj);
353 if (devobj != NULL) {
354 page_count = OFF_TO_IDX(holelen);
355
356 VM_OBJECT_WLOCK(devobj);
357 retry:
358 for (i = 0; i < page_count; i++) {
359 page = vm_page_lookup(devobj, i);
360 if (page == NULL)
361 continue;
362 if (!vm_page_busy_acquire(page, VM_ALLOC_WAITFAIL))
363 goto retry;
364 cdev_pager_free_page(devobj, page);
365 }
366 VM_OBJECT_WUNLOCK(devobj);
367 vm_object_deallocate(devobj);
368 }
369 }
370
371 int
lkpi_arch_phys_wc_add(unsigned long base,unsigned long size)372 lkpi_arch_phys_wc_add(unsigned long base, unsigned long size)
373 {
374 #ifdef __i386__
375 struct mem_range_desc *mrdesc;
376 int error, id, act;
377
378 /* If PAT is available, do nothing */
379 if (pat_works)
380 return (0);
381
382 mrdesc = malloc(sizeof(*mrdesc), M_LKMTRR, M_WAITOK);
383 mrdesc->mr_base = base;
384 mrdesc->mr_len = size;
385 mrdesc->mr_flags = MDF_WRITECOMBINE;
386 strlcpy(mrdesc->mr_owner, "drm", sizeof(mrdesc->mr_owner));
387 act = MEMRANGE_SET_UPDATE;
388 error = mem_range_attr_set(mrdesc, &act);
389 if (error == 0) {
390 error = idr_get_new(&mtrr_idr, mrdesc, &id);
391 MPASS(idr_find(&mtrr_idr, id) == mrdesc);
392 if (error != 0) {
393 act = MEMRANGE_SET_REMOVE;
394 mem_range_attr_set(mrdesc, &act);
395 }
396 }
397 if (error != 0) {
398 free(mrdesc, M_LKMTRR);
399 pr_warn(
400 "Failed to add WC MTRR for [%p-%p]: %d; "
401 "performance may suffer\n",
402 (void *)base, (void *)(base + size - 1), error);
403 } else
404 pr_warn("Successfully added WC MTRR for [%p-%p]\n",
405 (void *)base, (void *)(base + size - 1));
406
407 return (error != 0 ? -error : id + __MTRR_ID_BASE);
408 #else
409 return (0);
410 #endif
411 }
412
413 void
lkpi_arch_phys_wc_del(int reg)414 lkpi_arch_phys_wc_del(int reg)
415 {
416 #ifdef __i386__
417 struct mem_range_desc *mrdesc;
418 int act;
419
420 /* Check if arch_phys_wc_add() failed. */
421 if (reg < __MTRR_ID_BASE)
422 return;
423
424 mrdesc = idr_find(&mtrr_idr, reg - __MTRR_ID_BASE);
425 MPASS(mrdesc != NULL);
426 idr_remove(&mtrr_idr, reg - __MTRR_ID_BASE);
427 act = MEMRANGE_SET_REMOVE;
428 mem_range_attr_set(mrdesc, &act);
429 free(mrdesc, M_LKMTRR);
430 #endif
431 }
432