1 /*- 2 * Copyright (c) 2010 Isilon Systems, Inc. 3 * Copyright (c) 2016 Matt Macy (mmacy@nextbsd.org) 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_phys.h> 55 #include <vm/vm_radix.h> 56 #include <vm/vm_reserv.h> 57 #include <vm/vm_extern.h> 58 59 #include <vm/uma.h> 60 #include <vm/uma_int.h> 61 62 #include <linux/gfp.h> 63 #include <linux/mm.h> 64 #include <linux/preempt.h> 65 66 #if defined(__amd64__) || defined(__aarch64__) || defined(__riscv__) 67 #define LINUXKPI_HAVE_DMAP 68 #else 69 #undef LINUXKPI_HAVE_DMAP 70 #endif 71 72 void * 73 linux_page_address(struct page *page) 74 { 75 #ifdef LINUXKPI_HAVE_DMAP 76 return ((void *)PHYS_TO_DMAP(VM_PAGE_TO_PHYS(page))); 77 #else 78 if (page->object != kmem_object && page->object != kernel_object) 79 return (NULL); 80 return ((void *)(uintptr_t)(VM_MIN_KERNEL_ADDRESS + 81 IDX_TO_OFF(page->pindex))); 82 #endif 83 } 84 85 vm_page_t 86 linux_alloc_pages(gfp_t flags, unsigned int order) 87 { 88 #ifdef LINUXKPI_HAVE_DMAP 89 unsigned long npages = 1UL << order; 90 int req = (flags & M_ZERO) ? (VM_ALLOC_ZERO | VM_ALLOC_NOOBJ | 91 VM_ALLOC_NORMAL) : (VM_ALLOC_NOOBJ | VM_ALLOC_NORMAL); 92 vm_page_t page; 93 94 if (order == 0 && (flags & GFP_DMA32) == 0) { 95 page = vm_page_alloc(NULL, 0, req); 96 if (page == NULL) 97 return (NULL); 98 } else { 99 vm_paddr_t pmax = (flags & GFP_DMA32) ? 100 BUS_SPACE_MAXADDR_32BIT : BUS_SPACE_MAXADDR; 101 retry: 102 page = vm_page_alloc_contig(NULL, 0, req, 103 npages, 0, pmax, PAGE_SIZE, 0, VM_MEMATTR_DEFAULT); 104 105 if (page == NULL) { 106 if (flags & M_WAITOK) { 107 if (!vm_page_reclaim_contig(req, 108 npages, 0, pmax, PAGE_SIZE, 0)) { 109 VM_WAIT; 110 } 111 flags &= ~M_WAITOK; 112 goto retry; 113 } 114 return (NULL); 115 } 116 } 117 if (flags & M_ZERO) { 118 unsigned long x; 119 120 for (x = 0; x != npages; x++) { 121 vm_page_t pgo = page + x; 122 123 if ((pgo->flags & PG_ZERO) == 0) 124 pmap_zero_page(pgo); 125 } 126 } 127 #else 128 vm_offset_t vaddr; 129 vm_page_t page; 130 131 vaddr = linux_alloc_kmem(flags, order); 132 if (vaddr == 0) 133 return (NULL); 134 135 page = PHYS_TO_VM_PAGE(vtophys((void *)vaddr)); 136 137 KASSERT(vaddr == (vm_offset_t)page_address(page), 138 ("Page address mismatch")); 139 #endif 140 return (page); 141 } 142 143 void 144 linux_free_pages(vm_page_t page, unsigned int order) 145 { 146 #ifdef LINUXKPI_HAVE_DMAP 147 unsigned long npages = 1UL << order; 148 unsigned long x; 149 150 for (x = 0; x != npages; x++) { 151 vm_page_t pgo = page + x; 152 153 vm_page_lock(pgo); 154 vm_page_free(pgo); 155 vm_page_unlock(pgo); 156 } 157 #else 158 vm_offset_t vaddr; 159 160 vaddr = (vm_offset_t)page_address(page); 161 162 linux_free_kmem(vaddr, order); 163 #endif 164 } 165 166 vm_offset_t 167 linux_alloc_kmem(gfp_t flags, unsigned int order) 168 { 169 size_t size = ((size_t)PAGE_SIZE) << order; 170 vm_offset_t addr; 171 172 if ((flags & GFP_DMA32) == 0) { 173 addr = kmem_malloc(kmem_arena, size, flags & GFP_NATIVE_MASK); 174 } else { 175 addr = kmem_alloc_contig(kmem_arena, size, 176 flags & GFP_NATIVE_MASK, 0, BUS_SPACE_MAXADDR_32BIT, 177 PAGE_SIZE, 0, VM_MEMATTR_DEFAULT); 178 } 179 return (addr); 180 } 181 182 void 183 linux_free_kmem(vm_offset_t addr, unsigned int order) 184 { 185 size_t size = ((size_t)PAGE_SIZE) << order; 186 187 kmem_free(kmem_arena, addr, size); 188 } 189 190 static int 191 linux_get_user_pages_internal(vm_map_t map, unsigned long start, int nr_pages, 192 int write, struct page **pages) 193 { 194 vm_prot_t prot; 195 size_t len; 196 int count; 197 int i; 198 199 prot = write ? (VM_PROT_READ | VM_PROT_WRITE) : VM_PROT_READ; 200 len = ((size_t)nr_pages) << PAGE_SHIFT; 201 count = vm_fault_quick_hold_pages(map, start, len, prot, pages, nr_pages); 202 if (count == -1) 203 return (-EFAULT); 204 205 for (i = 0; i != nr_pages; i++) { 206 struct page *pg = pages[i]; 207 208 vm_page_lock(pg); 209 vm_page_wire(pg); 210 vm_page_unlock(pg); 211 } 212 return (nr_pages); 213 } 214 215 int 216 __get_user_pages_fast(unsigned long start, int nr_pages, int write, 217 struct page **pages) 218 { 219 vm_map_t map; 220 vm_page_t *mp; 221 vm_offset_t va; 222 vm_offset_t end; 223 vm_prot_t prot; 224 int count; 225 226 if (nr_pages == 0 || in_interrupt()) 227 return (0); 228 229 MPASS(pages != NULL); 230 va = start; 231 map = &curthread->td_proc->p_vmspace->vm_map; 232 end = start + (((size_t)nr_pages) << PAGE_SHIFT); 233 if (start < vm_map_min(map) || end > vm_map_max(map)) 234 return (-EINVAL); 235 prot = write ? (VM_PROT_READ | VM_PROT_WRITE) : VM_PROT_READ; 236 for (count = 0, mp = pages, va = start; va < end; 237 mp++, va += PAGE_SIZE, count++) { 238 *mp = pmap_extract_and_hold(map->pmap, va, prot); 239 if (*mp == NULL) 240 break; 241 242 vm_page_lock(*mp); 243 vm_page_wire(*mp); 244 vm_page_unlock(*mp); 245 246 if ((prot & VM_PROT_WRITE) != 0 && 247 (*mp)->dirty != VM_PAGE_BITS_ALL) { 248 /* 249 * Explicitly dirty the physical page. Otherwise, the 250 * caller's changes may go unnoticed because they are 251 * performed through an unmanaged mapping or by a DMA 252 * operation. 253 * 254 * The object lock is not held here. 255 * See vm_page_clear_dirty_mask(). 256 */ 257 vm_page_dirty(*mp); 258 } 259 } 260 return (count); 261 } 262 263 long 264 get_user_pages_remote(struct task_struct *task, struct mm_struct *mm, 265 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 = &mm->vmspace->vm_map; 271 return (linux_get_user_pages_internal(map, start, nr_pages, 272 !!(gup_flags & FOLL_WRITE), pages)); 273 } 274 275 long 276 get_user_pages(unsigned long start, unsigned long nr_pages, int gup_flags, 277 struct page **pages, struct vm_area_struct **vmas) 278 { 279 vm_map_t map; 280 281 map = &curthread->td_proc->p_vmspace->vm_map; 282 return (linux_get_user_pages_internal(map, start, nr_pages, 283 !!(gup_flags & FOLL_WRITE), pages)); 284 } 285 286 int 287 is_vmalloc_addr(const void *addr) 288 { 289 return (vtoslab((vm_offset_t)addr & ~UMA_SLAB_MASK) != NULL); 290 } 291