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