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