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 #ifdef __GFP_NOTWIRED
  97                 if ((flags & __GFP_NOTWIRED) != 0)
  98                         req &= ~VM_ALLOC_WIRED;
  99 #endif
 100                 if ((flags & M_ZERO) != 0)
 101                         req |= VM_ALLOC_ZERO;
 102                 if (order == 0 && (flags & GFP_DMA32) == 0) {
 103                         page = vm_page_alloc(NULL, 0, req);
 104                         if (page == NULL)
 105                                 return (NULL);
 106                 } else {
 107                         vm_paddr_t pmax = (flags & GFP_DMA32) ?
 108                             BUS_SPACE_MAXADDR_32BIT : BUS_SPACE_MAXADDR;
 109                 retry:
 110                         page = vm_page_alloc_contig(NULL, 0, req,
 111                             npages, 0, pmax, PAGE_SIZE, 0, VM_MEMATTR_DEFAULT);
 112
 113                         if (page == NULL) {
 114                                 if (flags & M_WAITOK) {
 115                                         if (!vm_page_reclaim_contig(req,
 116                                             npages, 0, pmax, PAGE_SIZE, 0)) {
 117                                                 vm_wait(NULL);
 118                                         }
 119                                         flags &= ~M_WAITOK;
 120                                         goto retry;
 121                                 }
 122                                 return (NULL);
 123                         }
 124                 }
 125                 if (flags & M_ZERO) {
 126                         unsigned long x;
 127
 128                         for (x = 0; x != npages; x++) {
 129                                 vm_page_t pgo = page + x;
 130
 131                                 if ((pgo->flags & PG_ZERO) == 0)
 132                                         pmap_zero_page(pgo);
 133                         }
 134                 }
 135         } else {
 136                 vm_offset_t vaddr;
 137
 138                 vaddr = linux_alloc_kmem(flags, order);
 139                 if (vaddr == 0)
 140                         return (NULL);
 141
 142                 page = PHYS_TO_VM_PAGE(vtophys((void *)vaddr));
 143
 144                 KASSERT(vaddr == (vm_offset_t)page_address(page),
 145                     ("Page address mismatch"));
 146         }
 147
 148         return (page);
 149 }
 150
 151 void
 152 linux_free_pages(vm_page_t page, unsigned int order)
 153 {
 154         if (PMAP_HAS_DMAP) {
 155                 unsigned long npages = 1UL << order;
 156                 unsigned long x;
 157
 158                 for (x = 0; x != npages; x++) {
 159                         vm_page_t pgo = page + x;
 160
 161                         vm_page_lock(pgo);
 162                         if (vm_page_unwire_noq(pgo))
 163                                 vm_page_free(pgo);
 164                         vm_page_unlock(pgo);
 165                 }
 166         } else {
 167                 vm_offset_t vaddr;
 168
 169                 vaddr = (vm_offset_t)page_address(page);
 170
 171                 linux_free_kmem(vaddr, order);
 172         }
 173 }
 174
 175 vm_offset_t
 176 linux_alloc_kmem(gfp_t flags, unsigned int order)
 177 {
 178         size_t size = ((size_t)PAGE_SIZE) << order;
 179         vm_offset_t addr;
 180
 181         if ((flags & GFP_DMA32) == 0) {
 182                 addr = kmem_malloc(size, flags & GFP_NATIVE_MASK);
 183         } else {
 184                 addr = kmem_alloc_contig(size, flags & GFP_NATIVE_MASK, 0,
 185                     BUS_SPACE_MAXADDR_32BIT, PAGE_SIZE, 0, VM_MEMATTR_DEFAULT);
 186         }
 187         return (addr);
 188 }
 189
 190 void
 191 linux_free_kmem(vm_offset_t addr, unsigned int order)
 192 {
 193         size_t size = ((size_t)PAGE_SIZE) << order;
 194
 195         kmem_free(addr, size);
 196 }
 197
 198 static int
 199 linux_get_user_pages_internal(vm_map_t map, unsigned long start, int nr_pages,
 200     int write, struct page **pages)
 201 {
 202         vm_prot_t prot;
 203         size_t len;
 204         int count;
 205         int i;
 206
 207         prot = write ? (VM_PROT_READ | VM_PROT_WRITE) : VM_PROT_READ;
 208         len = ((size_t)nr_pages) << PAGE_SHIFT;
 209         count = vm_fault_quick_hold_pages(map, start, len, prot, pages, nr_pages);
 210         if (count == -1)
 211                 return (-EFAULT);
 212
 213         for (i = 0; i != nr_pages; i++) {
 214                 struct page *pg = pages[i];
 215
 216                 vm_page_lock(pg);
 217                 vm_page_wire(pg);
 218                 vm_page_unhold(pg);
 219                 vm_page_unlock(pg);
 220         }
 221         return (nr_pages);
 222 }
 223
 224 int
 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         va = start;
 240         map = &curthread->td_proc->p_vmspace->vm_map;
 241         end = start + (((size_t)nr_pages) << PAGE_SHIFT);
 242         if (start < vm_map_min(map) || end > vm_map_max(map))
 243                 return (-EINVAL);
 244         prot = write ? (VM_PROT_READ | VM_PROT_WRITE) : VM_PROT_READ;
 245         for (count = 0, mp = pages, va = start; va < end;
 246             mp++, va += PAGE_SIZE, count++) {
 247                 *mp = pmap_extract_and_hold(map->pmap, va, prot);
 248                 if (*mp == NULL)
 249                         break;
 250
 251                 vm_page_lock(*mp);
 252                 vm_page_wire(*mp);
 253                 vm_page_unhold(*mp);
 254                 vm_page_unlock(*mp);
 255
 256                 if ((prot & VM_PROT_WRITE) != 0 &&
 257                     (*mp)->dirty != VM_PAGE_BITS_ALL) {
 258                         /*
 259                          * Explicitly dirty the physical page.  Otherwise, the
 260                          * caller's changes may go unnoticed because they are
 261                          * performed through an unmanaged mapping or by a DMA
 262                          * operation.
 263                          *
 264                          * The object lock is not held here.
 265                          * See vm_page_clear_dirty_mask().
 266                          */
 267                         vm_page_dirty(*mp);
 268                 }
 269         }
 270         return (count);
 271 }
 272
 273 long
 274 get_user_pages_remote(struct task_struct *task, struct mm_struct *mm,
 275     unsigned long start, unsigned long nr_pages, int gup_flags,
 276     struct page **pages, struct vm_area_struct **vmas)
 277 {
 278         vm_map_t map;
 279
 280         map = &task->task_thread->td_proc->p_vmspace->vm_map;
 281         return (linux_get_user_pages_internal(map, start, nr_pages,
 282             !!(gup_flags & FOLL_WRITE), pages));
 283 }
 284
 285 long
 286 get_user_pages(unsigned long start, unsigned long nr_pages, int gup_flags,
 287     struct page **pages, struct vm_area_struct **vmas)
 288 {
 289         vm_map_t map;
 290
 291         map = &curthread->td_proc->p_vmspace->vm_map;
 292         return (linux_get_user_pages_internal(map, start, nr_pages,
 293             !!(gup_flags & FOLL_WRITE), pages));
 294 }
 295
 296 int
 297 is_vmalloc_addr(const void *addr)
 298 {
 299         return (vtoslab((vm_offset_t)addr & ~UMA_SLAB_MASK) != NULL);
 300 }
 301
 302 struct page *
 303 linux_shmem_read_mapping_page_gfp(vm_object_t obj, int pindex, gfp_t gfp)
 304 {
 305         vm_page_t page;
 306         int rv;
 307
 308         if ((gfp & GFP_NOWAIT) != 0)
 309                 panic("GFP_NOWAIT is unimplemented");
 310
 311         VM_OBJECT_WLOCK(obj);
 312         page = vm_page_grab(obj, pindex, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY |
 313             VM_ALLOC_WIRED);
 314         if (page->valid != VM_PAGE_BITS_ALL) {
 315                 vm_page_xbusy(page);
 316                 if (vm_pager_has_page(obj, pindex, NULL, NULL)) {
 317                         rv = vm_pager_get_pages(obj, &page, 1, NULL, NULL);
 318                         if (rv != VM_PAGER_OK) {
 319                                 vm_page_lock(page);
 320                                 vm_page_unwire_noq(page);
 321                                 vm_page_free(page);
 322                                 vm_page_unlock(page);
 323                                 VM_OBJECT_WUNLOCK(obj);
 324                                 return (ERR_PTR(-EINVAL));
 325                         }
 326                         MPASS(page->valid == VM_PAGE_BITS_ALL);
 327                 } else {
 328                         pmap_zero_page(page);
 329                         page->valid = VM_PAGE_BITS_ALL;
 330                         page->dirty = 0;
 331                 }
 332                 vm_page_xunbusy(page);
 333         }
 334         VM_OBJECT_WUNLOCK(obj);
 335         return (page);
 336 }
 337
 338 struct linux_file *
 339 linux_shmem_file_setup(const char *name, loff_t size, unsigned long flags)
 340 {
 341         struct fileobj {
 342                 struct linux_file file __aligned(sizeof(void *));
 343                 struct vnode vnode __aligned(sizeof(void *));
 344         };
 345         struct fileobj *fileobj;
 346         struct linux_file *filp;
 347         struct vnode *vp;
 348         int error;
 349
 350         fileobj = kzalloc(sizeof(*fileobj), GFP_KERNEL);
 351         if (fileobj == NULL) {
 352                 error = -ENOMEM;
 353                 goto err_0;
 354         }
 355         filp = &fileobj->file;
 356         vp = &fileobj->vnode;
 357
 358         filp->f_count = 1;
 359         filp->f_vnode = vp;
 360         filp->f_shmem = vm_pager_allocate(OBJT_DEFAULT, NULL, size,
 361             VM_PROT_READ | VM_PROT_WRITE, 0, curthread->td_ucred);
 362         if (filp->f_shmem == NULL) {
 363                 error = -ENOMEM;
 364                 goto err_1;
 365         }
 366         return (filp);
 367 err_1:
 368         kfree(filp);
 369 err_0:
 370         return (ERR_PTR(error));
 371 }
 372
 373 static vm_ooffset_t
 374 linux_invalidate_mapping_pages_sub(vm_object_t obj, vm_pindex_t start,
 375     vm_pindex_t end, int flags)
 376 {
 377         int start_count, end_count;
 378
 379         VM_OBJECT_WLOCK(obj);
 380         start_count = obj->resident_page_count;
 381         vm_object_page_remove(obj, start, end, flags);
 382         end_count = obj->resident_page_count;
 383         VM_OBJECT_WUNLOCK(obj);
 384         return (start_count - end_count);
 385 }
 386
 387 unsigned long
 388 linux_invalidate_mapping_pages(vm_object_t obj, pgoff_t start, pgoff_t end)
 389 {
 390
 391         return (linux_invalidate_mapping_pages_sub(obj, start, end, OBJPR_CLEANONLY));
 392 }
 393
 394 void
 395 linux_shmem_truncate_range(vm_object_t obj, loff_t lstart, loff_t lend)
 396 {
 397         vm_pindex_t start = OFF_TO_IDX(lstart + PAGE_SIZE - 1);
 398         vm_pindex_t end = OFF_TO_IDX(lend + 1);
 399
 400         (void) linux_invalidate_mapping_pages_sub(obj, start, end, 0);
 401 }