sys/compat/linuxkpi/common/src/linux_page.c

   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/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
  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 *
  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
  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
 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
 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
 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
 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
 202         prot = write ? (VM_PROT_READ | VM_PROT_WRITE) : VM_PROT_READ;
 203         len = ((size_t)nr_pages) << PAGE_SHIFT;
 204         count = vm_fault_quick_hold_pages(map, start, len, prot, pages, nr_pages);
 205         return (count == -1 ? -EFAULT : nr_pages);
 206 }
 207
 208 int
 209 __get_user_pages_fast(unsigned long start, int nr_pages, int write,
 210     struct page **pages)
 211 {
 212         vm_map_t map;
 213         vm_page_t *mp;
 214         vm_offset_t va;
 215         vm_offset_t end;
 216         vm_prot_t prot;
 217         int count;
 218
 219         if (nr_pages == 0 || in_interrupt())
 220                 return (0);
 221
 222         MPASS(pages != NULL);
 223         va = start;
 224         map = &curthread->td_proc->p_vmspace->vm_map;
 225         end = start + (((size_t)nr_pages) << PAGE_SHIFT);
 226         if (start < vm_map_min(map) || end > vm_map_max(map))
 227                 return (-EINVAL);
 228         prot = write ? (VM_PROT_READ | VM_PROT_WRITE) : VM_PROT_READ;
 229         for (count = 0, mp = pages, va = start; va < end;
 230             mp++, va += PAGE_SIZE, count++) {
 231                 *mp = pmap_extract_and_hold(map->pmap, va, prot);
 232                 if (*mp == NULL)
 233                         break;
 234
 235                 if ((prot & VM_PROT_WRITE) != 0 &&
 236                     (*mp)->dirty != VM_PAGE_BITS_ALL) {
 237                         /*
 238                          * Explicitly dirty the physical page.  Otherwise, the
 239                          * caller's changes may go unnoticed because they are
 240                          * performed through an unmanaged mapping or by a DMA
 241                          * operation.
 242                          *
 243                          * The object lock is not held here.
 244                          * See vm_page_clear_dirty_mask().
 245                          */
 246                         vm_page_dirty(*mp);
 247                 }
 248         }
 249         return (count);
 250 }
 251
 252 long
 253 get_user_pages_remote(struct task_struct *task, struct mm_struct *mm,
 254     unsigned long start, unsigned long nr_pages, int gup_flags,
 255     struct page **pages, struct vm_area_struct **vmas)
 256 {
 257         vm_map_t map;
 258
 259         map = &task->task_thread->td_proc->p_vmspace->vm_map;
 260         return (linux_get_user_pages_internal(map, start, nr_pages,
 261             !!(gup_flags & FOLL_WRITE), pages));
 262 }
 263
 264 long
 265 get_user_pages(unsigned long start, unsigned long nr_pages, int gup_flags,
 266     struct page **pages, struct vm_area_struct **vmas)
 267 {
 268         vm_map_t map;
 269
 270         map = &curthread->td_proc->p_vmspace->vm_map;
 271         return (linux_get_user_pages_internal(map, start, nr_pages,
 272             !!(gup_flags & FOLL_WRITE), pages));
 273 }
 274
 275 int
 276 is_vmalloc_addr(const void *addr)
 277 {
 278         return (vtoslab((vm_offset_t)addr & ~UMA_SLAB_MASK) != NULL);
 279 }
 280
 281 struct page *
 282 linux_shmem_read_mapping_page_gfp(vm_object_t obj, int pindex, gfp_t gfp)
 283 {
 284         vm_page_t page;
 285         int rv;
 286
 287         if ((gfp & GFP_NOWAIT) != 0)
 288                 panic("GFP_NOWAIT is unimplemented");
 289
 290         VM_OBJECT_WLOCK(obj);
 291         page = vm_page_grab(obj, pindex, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY |
 292             VM_ALLOC_WIRED);
 293         if (page->valid != VM_PAGE_BITS_ALL) {
 294                 vm_page_xbusy(page);
 295                 if (vm_pager_has_page(obj, pindex, NULL, NULL)) {
 296                         rv = vm_pager_get_pages(obj, &page, 1, NULL, NULL);
 297                         if (rv != VM_PAGER_OK) {
 298                                 vm_page_lock(page);
 299                                 vm_page_unwire_noq(page);
 300                                 vm_page_free(page);
 301                                 vm_page_unlock(page);
 302                                 VM_OBJECT_WUNLOCK(obj);
 303                                 return (ERR_PTR(-EINVAL));
 304                         }
 305                         MPASS(page->valid == VM_PAGE_BITS_ALL);
 306                 } else {
 307                         pmap_zero_page(page);
 308                         page->valid = VM_PAGE_BITS_ALL;
 309                         page->dirty = 0;
 310                 }
 311                 vm_page_xunbusy(page);
 312         }
 313         VM_OBJECT_WUNLOCK(obj);
 314         return (page);
 315 }
 316
 317 struct linux_file *
 318 linux_shmem_file_setup(const char *name, loff_t size, unsigned long flags)
 319 {
 320         struct fileobj {
 321                 struct linux_file file __aligned(sizeof(void *));
 322                 struct vnode vnode __aligned(sizeof(void *));
 323         };
 324         struct fileobj *fileobj;
 325         struct linux_file *filp;
 326         struct vnode *vp;
 327         int error;
 328
 329         fileobj = kzalloc(sizeof(*fileobj), GFP_KERNEL);
 330         if (fileobj == NULL) {
 331                 error = -ENOMEM;
 332                 goto err_0;
 333         }
 334         filp = &fileobj->file;
 335         vp = &fileobj->vnode;
 336
 337         filp->f_count = 1;
 338         filp->f_vnode = vp;
 339         filp->f_shmem = vm_pager_allocate(OBJT_DEFAULT, NULL, size,
 340             VM_PROT_READ | VM_PROT_WRITE, 0, curthread->td_ucred);
 341         if (filp->f_shmem == NULL) {
 342                 error = -ENOMEM;
 343                 goto err_1;
 344         }
 345         return (filp);
 346 err_1:
 347         kfree(filp);
 348 err_0:
 349         return (ERR_PTR(error));
 350 }
 351
 352 static vm_ooffset_t
 353 linux_invalidate_mapping_pages_sub(vm_object_t obj, vm_pindex_t start,
 354     vm_pindex_t end, int flags)
 355 {
 356         int start_count, end_count;
 357
 358         VM_OBJECT_WLOCK(obj);
 359         start_count = obj->resident_page_count;
 360         vm_object_page_remove(obj, start, end, flags);
 361         end_count = obj->resident_page_count;
 362         VM_OBJECT_WUNLOCK(obj);
 363         return (start_count - end_count);
 364 }
 365
 366 unsigned long
 367 linux_invalidate_mapping_pages(vm_object_t obj, pgoff_t start, pgoff_t end)
 368 {
 369
 370         return (linux_invalidate_mapping_pages_sub(obj, start, end, OBJPR_CLEANONLY));
 371 }
 372
 373 void
 374 linux_shmem_truncate_range(vm_object_t obj, loff_t lstart, loff_t lend)
 375 {
 376         vm_pindex_t start = OFF_TO_IDX(lstart + PAGE_SIZE - 1);
 377         vm_pindex_t end = OFF_TO_IDX(lend + 1);
 378
 379         (void) linux_invalidate_mapping_pages_sub(obj, start, end, 0);
 380 }