xref: /freebsd/sys/compat/linuxkpi/common/src/linux_page.c (revision d48524e21f1a49752485418324538755571ed13f)
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) {
121 					int err = vm_page_reclaim_contig(req,
122 					    npages, 0, pmax, PAGE_SIZE, 0);
123 					if (err == ENOMEM)
124 						vm_wait(NULL);
125 					else if (err != 0)
126 						return (NULL);
127 					flags &= ~M_WAITOK;
128 					goto retry;
129 				}
130 				return (NULL);
131 			}
132 		}
133 	} else {
134 		vm_offset_t vaddr;
135 
136 		vaddr = linux_alloc_kmem(flags, order);
137 		if (vaddr == 0)
138 			return (NULL);
139 
140 		page = virt_to_page((void *)vaddr);
141 
142 		KASSERT(vaddr == (vm_offset_t)page_address(page),
143 		    ("Page address mismatch"));
144 	}
145 
146 	return (page);
147 }
148 
149 static void
_linux_free_kmem(vm_offset_t addr,unsigned int order)150 _linux_free_kmem(vm_offset_t addr, unsigned int order)
151 {
152 	size_t size = ((size_t)PAGE_SIZE) << order;
153 
154 	kmem_free((void *)addr, size);
155 }
156 
157 void
linux_free_pages(struct page * page,unsigned int order)158 linux_free_pages(struct page *page, unsigned int order)
159 {
160 	if (PMAP_HAS_DMAP) {
161 		unsigned long npages = 1UL << order;
162 		unsigned long x;
163 
164 		for (x = 0; x != npages; x++) {
165 			vm_page_t pgo = page + x;
166 
167 			if (vm_page_unwire_noq(pgo))
168 				vm_page_free(pgo);
169 		}
170 	} else {
171 		vm_offset_t vaddr;
172 
173 		vaddr = (vm_offset_t)page_address(page);
174 
175 		_linux_free_kmem(vaddr, order);
176 	}
177 }
178 
179 vm_offset_t
linux_alloc_kmem(gfp_t flags,unsigned int order)180 linux_alloc_kmem(gfp_t flags, unsigned int order)
181 {
182 	size_t size = ((size_t)PAGE_SIZE) << order;
183 	void *addr;
184 
185 	if ((flags & GFP_DMA32) == 0) {
186 		addr = kmem_malloc(size, flags & GFP_NATIVE_MASK);
187 	} else {
188 		addr = kmem_alloc_contig(size, flags & GFP_NATIVE_MASK, 0,
189 		    BUS_SPACE_MAXADDR_32BIT, PAGE_SIZE, 0, VM_MEMATTR_DEFAULT);
190 	}
191 	return ((vm_offset_t)addr);
192 }
193 
194 void
linux_free_kmem(vm_offset_t addr,unsigned int order)195 linux_free_kmem(vm_offset_t addr, unsigned int order)
196 {
197 	KASSERT((addr & ~PAGE_MASK) == 0,
198 	    ("%s: addr %p is not page aligned", __func__, (void *)addr));
199 
200 	if (addr >= VM_MIN_KERNEL_ADDRESS && addr < VM_MAX_KERNEL_ADDRESS) {
201 		_linux_free_kmem(addr, order);
202 	} else {
203 		vm_page_t page;
204 
205 		page = PHYS_TO_VM_PAGE(DMAP_TO_PHYS(addr));
206 		linux_free_pages(page, order);
207 	}
208 }
209 
210 static int
linux_get_user_pages_internal(vm_map_t map,unsigned long start,int nr_pages,int write,struct page ** pages)211 linux_get_user_pages_internal(vm_map_t map, unsigned long start, int nr_pages,
212     int write, struct page **pages)
213 {
214 	vm_prot_t prot;
215 	size_t len;
216 	int count;
217 
218 	prot = write ? (VM_PROT_READ | VM_PROT_WRITE) : VM_PROT_READ;
219 	len = ptoa((vm_offset_t)nr_pages);
220 	count = vm_fault_quick_hold_pages(map, start, len, prot, pages, nr_pages);
221 	return (count == -1 ? -EFAULT : nr_pages);
222 }
223 
224 int
__get_user_pages_fast(unsigned long start,int nr_pages,int write,struct page ** pages)225 __get_user_pages_fast(unsigned long start, int nr_pages, int write,
226     struct page **pages)
227 {
228 	vm_map_t map;
229 	vm_page_t *mp;
230 	vm_offset_t va;
231 	vm_offset_t end;
232 	vm_prot_t prot;
233 	int count;
234 
235 	if (nr_pages == 0 || in_interrupt())
236 		return (0);
237 
238 	MPASS(pages != NULL);
239 	map = &curthread->td_proc->p_vmspace->vm_map;
240 	end = start + ptoa((vm_offset_t)nr_pages);
241 	if (!vm_map_range_valid(map, start, end))
242 		return (-EINVAL);
243 	prot = write ? (VM_PROT_READ | VM_PROT_WRITE) : VM_PROT_READ;
244 	for (count = 0, mp = pages, va = start; va < end;
245 	    mp++, va += PAGE_SIZE, count++) {
246 		*mp = pmap_extract_and_hold(map->pmap, va, prot);
247 		if (*mp == NULL)
248 			break;
249 
250 		if ((prot & VM_PROT_WRITE) != 0 &&
251 		    (*mp)->dirty != VM_PAGE_BITS_ALL) {
252 			/*
253 			 * Explicitly dirty the physical page.  Otherwise, the
254 			 * caller's changes may go unnoticed because they are
255 			 * performed through an unmanaged mapping or by a DMA
256 			 * operation.
257 			 *
258 			 * The object lock is not held here.
259 			 * See vm_page_clear_dirty_mask().
260 			 */
261 			vm_page_dirty(*mp);
262 		}
263 	}
264 	return (count);
265 }
266 
267 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)268 get_user_pages_remote(struct task_struct *task, struct mm_struct *mm,
269     unsigned long start, unsigned long nr_pages, unsigned int gup_flags,
270     struct page **pages, struct vm_area_struct **vmas)
271 {
272 	vm_map_t map;
273 
274 	map = &task->task_thread->td_proc->p_vmspace->vm_map;
275 	return (linux_get_user_pages_internal(map, start, nr_pages,
276 	    !!(gup_flags & FOLL_WRITE), pages));
277 }
278 
279 long
lkpi_get_user_pages(unsigned long start,unsigned long nr_pages,unsigned int gup_flags,struct page ** pages)280 lkpi_get_user_pages(unsigned long start, unsigned long nr_pages,
281     unsigned int gup_flags, struct page **pages)
282 {
283 	vm_map_t map;
284 
285 	map = &curthread->td_proc->p_vmspace->vm_map;
286 	return (linux_get_user_pages_internal(map, start, nr_pages,
287 	    !!(gup_flags & FOLL_WRITE), pages));
288 }
289 
290 int
is_vmalloc_addr(const void * addr)291 is_vmalloc_addr(const void *addr)
292 {
293 	return (vtoslab((vm_offset_t)addr & ~UMA_SLAB_MASK) != NULL);
294 }
295 
296 vm_fault_t
lkpi_vmf_insert_pfn_prot_locked(struct vm_area_struct * vma,unsigned long addr,unsigned long pfn,pgprot_t prot)297 lkpi_vmf_insert_pfn_prot_locked(struct vm_area_struct *vma, unsigned long addr,
298     unsigned long pfn, pgprot_t prot)
299 {
300 	vm_object_t vm_obj = vma->vm_obj;
301 	vm_object_t tmp_obj;
302 	vm_page_t page;
303 	vm_pindex_t pindex;
304 
305 	VM_OBJECT_ASSERT_WLOCKED(vm_obj);
306 	pindex = OFF_TO_IDX(addr - vma->vm_start);
307 	if (vma->vm_pfn_count == 0)
308 		vma->vm_pfn_first = pindex;
309 	MPASS(pindex <= OFF_TO_IDX(vma->vm_end));
310 
311 retry:
312 	page = vm_page_grab(vm_obj, pindex, VM_ALLOC_NOCREAT);
313 	if (page == NULL) {
314 		page = PHYS_TO_VM_PAGE(IDX_TO_OFF(pfn));
315 		if (!vm_page_busy_acquire(page, VM_ALLOC_WAITFAIL))
316 			goto retry;
317 		if (page->object != NULL) {
318 			tmp_obj = page->object;
319 			vm_page_xunbusy(page);
320 			VM_OBJECT_WUNLOCK(vm_obj);
321 			VM_OBJECT_WLOCK(tmp_obj);
322 			if (page->object == tmp_obj &&
323 			    vm_page_busy_acquire(page, VM_ALLOC_WAITFAIL)) {
324 				KASSERT(page->object == tmp_obj,
325 				    ("page has changed identity"));
326 				KASSERT((page->oflags & VPO_UNMANAGED) == 0,
327 				    ("page does not belong to shmem"));
328 				vm_pager_page_unswapped(page);
329 				if (pmap_page_is_mapped(page)) {
330 					vm_page_xunbusy(page);
331 					VM_OBJECT_WUNLOCK(tmp_obj);
332 					printf("%s: page rename failed: page "
333 					    "is mapped\n", __func__);
334 					VM_OBJECT_WLOCK(vm_obj);
335 					return (VM_FAULT_NOPAGE);
336 				}
337 				vm_page_remove(page);
338 			}
339 			VM_OBJECT_WUNLOCK(tmp_obj);
340 			VM_OBJECT_WLOCK(vm_obj);
341 			goto retry;
342 		}
343 		if (vm_page_insert(page, vm_obj, pindex)) {
344 			vm_page_xunbusy(page);
345 			return (VM_FAULT_OOM);
346 		}
347 		vm_page_valid(page);
348 	}
349 	pmap_page_set_memattr(page, pgprot2cachemode(prot));
350 	vma->vm_pfn_count++;
351 
352 	return (VM_FAULT_NOPAGE);
353 }
354 
355 int
lkpi_remap_pfn_range(struct vm_area_struct * vma,unsigned long start_addr,unsigned long start_pfn,unsigned long size,pgprot_t prot)356 lkpi_remap_pfn_range(struct vm_area_struct *vma, unsigned long start_addr,
357     unsigned long start_pfn, unsigned long size, pgprot_t prot)
358 {
359 	vm_object_t vm_obj;
360 	unsigned long addr, pfn;
361 	int err = 0;
362 
363 	vm_obj = vma->vm_obj;
364 
365 	VM_OBJECT_WLOCK(vm_obj);
366 	for (addr = start_addr, pfn = start_pfn;
367 	    addr < start_addr + size;
368 	    addr += PAGE_SIZE) {
369 		vm_fault_t ret;
370 retry:
371 		ret = lkpi_vmf_insert_pfn_prot_locked(vma, addr, pfn, prot);
372 
373 		if ((ret & VM_FAULT_OOM) != 0) {
374 			VM_OBJECT_WUNLOCK(vm_obj);
375 			vm_wait(NULL);
376 			VM_OBJECT_WLOCK(vm_obj);
377 			goto retry;
378 		}
379 
380 		if ((ret & VM_FAULT_ERROR) != 0) {
381 			err = -EFAULT;
382 			break;
383 		}
384 
385 		pfn++;
386 	}
387 	VM_OBJECT_WUNLOCK(vm_obj);
388 
389 	if (unlikely(err)) {
390 		zap_vma_ptes(vma, start_addr,
391 		    (pfn - start_pfn) << PAGE_SHIFT);
392 		return (err);
393 	}
394 
395 	return (0);
396 }
397 
398 int
lkpi_io_mapping_map_user(struct io_mapping * iomap,struct vm_area_struct * vma,unsigned long addr,unsigned long pfn,unsigned long size)399 lkpi_io_mapping_map_user(struct io_mapping *iomap,
400     struct vm_area_struct *vma, unsigned long addr,
401     unsigned long pfn, unsigned long size)
402 {
403 	pgprot_t prot;
404 	int ret;
405 
406 	prot = cachemode2protval(iomap->attr);
407 	ret = lkpi_remap_pfn_range(vma, addr, pfn, size, prot);
408 
409 	return (ret);
410 }
411 
412 /*
413  * Although FreeBSD version of unmap_mapping_range has semantics and types of
414  * parameters compatible with Linux version, the values passed in are different
415  * @obj should match to vm_private_data field of vm_area_struct returned by
416  *      mmap file operation handler, see linux_file_mmap_single() sources
417  * @holelen should match to size of area to be munmapped.
418  */
419 void
lkpi_unmap_mapping_range(void * obj,loff_t const holebegin __unused,loff_t const holelen,int even_cows __unused)420 lkpi_unmap_mapping_range(void *obj, loff_t const holebegin __unused,
421     loff_t const holelen, int even_cows __unused)
422 {
423 	vm_object_t devobj;
424 	vm_page_t page;
425 	int i, page_count;
426 
427 	devobj = cdev_pager_lookup(obj);
428 	if (devobj != NULL) {
429 		page_count = OFF_TO_IDX(holelen);
430 
431 		VM_OBJECT_WLOCK(devobj);
432 retry:
433 		for (i = 0; i < page_count; i++) {
434 			page = vm_page_lookup(devobj, i);
435 			if (page == NULL)
436 				continue;
437 			if (!vm_page_busy_acquire(page, VM_ALLOC_WAITFAIL))
438 				goto retry;
439 			cdev_mgtdev_pager_free_page(devobj, page);
440 		}
441 		VM_OBJECT_WUNLOCK(devobj);
442 		vm_object_deallocate(devobj);
443 	}
444 }
445 
446 int
lkpi_arch_phys_wc_add(unsigned long base,unsigned long size)447 lkpi_arch_phys_wc_add(unsigned long base, unsigned long size)
448 {
449 #ifdef __i386__
450 	struct mem_range_desc *mrdesc;
451 	int error, id, act;
452 
453 	/* If PAT is available, do nothing */
454 	if (pat_works)
455 		return (0);
456 
457 	mrdesc = malloc(sizeof(*mrdesc), M_LKMTRR, M_WAITOK);
458 	mrdesc->mr_base = base;
459 	mrdesc->mr_len = size;
460 	mrdesc->mr_flags = MDF_WRITECOMBINE;
461 	strlcpy(mrdesc->mr_owner, "drm", sizeof(mrdesc->mr_owner));
462 	act = MEMRANGE_SET_UPDATE;
463 	error = mem_range_attr_set(mrdesc, &act);
464 	if (error == 0) {
465 		error = idr_get_new(&mtrr_idr, mrdesc, &id);
466 		MPASS(idr_find(&mtrr_idr, id) == mrdesc);
467 		if (error != 0) {
468 			act = MEMRANGE_SET_REMOVE;
469 			mem_range_attr_set(mrdesc, &act);
470 		}
471 	}
472 	if (error != 0) {
473 		free(mrdesc, M_LKMTRR);
474 		pr_warn(
475 		    "Failed to add WC MTRR for [%p-%p]: %d; "
476 		    "performance may suffer\n",
477 		    (void *)base, (void *)(base + size - 1), error);
478 	} else
479 		pr_warn("Successfully added WC MTRR for [%p-%p]\n",
480 		    (void *)base, (void *)(base + size - 1));
481 
482 	return (error != 0 ? -error : id + __MTRR_ID_BASE);
483 #else
484 	return (0);
485 #endif
486 }
487 
488 void
lkpi_arch_phys_wc_del(int reg)489 lkpi_arch_phys_wc_del(int reg)
490 {
491 #ifdef __i386__
492 	struct mem_range_desc *mrdesc;
493 	int act;
494 
495 	/* Check if arch_phys_wc_add() failed. */
496 	if (reg < __MTRR_ID_BASE)
497 		return;
498 
499 	mrdesc = idr_find(&mtrr_idr, reg - __MTRR_ID_BASE);
500 	MPASS(mrdesc != NULL);
501 	idr_remove(&mtrr_idr, reg - __MTRR_ID_BASE);
502 	act = MEMRANGE_SET_REMOVE;
503 	mem_range_attr_set(mrdesc, &act);
504 	free(mrdesc, M_LKMTRR);
505 #endif
506 }
507 
508 /*
509  * This is a highly simplified version of the Linux page_frag_cache.
510  * We only support up-to 1 single page as fragment size and we will
511  * always return a full page.  This may be wasteful on small objects
512  * but the only known consumer (mt76) is either asking for a half-page
513  * or a full page.  If this was to become a problem we can implement
514  * a more elaborate version.
515  */
516 void *
linuxkpi_page_frag_alloc(struct page_frag_cache * pfc,size_t fragsz,gfp_t gfp)517 linuxkpi_page_frag_alloc(struct page_frag_cache *pfc,
518     size_t fragsz, gfp_t gfp)
519 {
520 	vm_page_t pages;
521 
522 	if (fragsz == 0)
523 		return (NULL);
524 
525 	KASSERT(fragsz <= PAGE_SIZE, ("%s: fragsz %zu > PAGE_SIZE not yet "
526 	    "supported", __func__, fragsz));
527 
528 	pages = alloc_pages(gfp, flsl(howmany(fragsz, PAGE_SIZE) - 1));
529 	if (pages == NULL)
530 		return (NULL);
531 	pfc->va = linux_page_address(pages);
532 
533 	/* Passed in as "count" to __page_frag_cache_drain(). Unused by us. */
534 	pfc->pagecnt_bias = 0;
535 
536 	return (pfc->va);
537 }
538 
539 void
linuxkpi_page_frag_free(void * addr)540 linuxkpi_page_frag_free(void *addr)
541 {
542 	vm_page_t page;
543 
544 	page = virt_to_page(addr);
545 	linux_free_pages(page, 0);
546 }
547 
548 void
linuxkpi__page_frag_cache_drain(struct page * page,size_t count __unused)549 linuxkpi__page_frag_cache_drain(struct page *page, size_t count __unused)
550 {
551 
552 	linux_free_pages(page, 0);
553 }
554