1 /* $NetBSD: mm.c,v 1.17 2013/01/19 00:27:34 matt Exp $ */ 2 3 /*- 4 * Copyright (c) 2002, 2008, 2010 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Christos Zoulas, Joerg Sonnenberger and Mindaugas Rasiukevicius. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32 /* 33 * Special /dev/{mem,kmem,zero,null} memory devices. 34 */ 35 36 #include <sys/cdefs.h> 37 __KERNEL_RCSID(0, "$NetBSD: mm.c,v 1.17 2013/01/19 00:27:34 matt Exp $"); 38 39 #include "opt_compat_netbsd.h" 40 41 #include <sys/param.h> 42 #include <sys/conf.h> 43 #include <sys/ioctl.h> 44 #include <sys/mman.h> 45 #include <sys/uio.h> 46 #include <sys/termios.h> 47 48 #include <dev/mm.h> 49 50 #include <uvm/uvm_extern.h> 51 52 static void * dev_zero_page __read_mostly; 53 static kmutex_t dev_mem_lock __cacheline_aligned; 54 static vaddr_t dev_mem_addr __read_mostly; 55 56 static dev_type_read(mm_readwrite); 57 static dev_type_ioctl(mm_ioctl); 58 static dev_type_mmap(mm_mmap); 59 static dev_type_ioctl(mm_ioctl); 60 61 const struct cdevsw mem_cdevsw = { 62 #ifdef __HAVE_MM_MD_OPEN 63 mm_md_open, 64 #else 65 nullopen, 66 #endif 67 nullclose, mm_readwrite, mm_readwrite, 68 mm_ioctl, nostop, notty, nopoll, mm_mmap, nokqfilter, 69 D_MPSAFE 70 }; 71 72 #ifdef pmax /* XXX */ 73 const struct cdevsw mem_ultrix_cdevsw = { 74 nullopen, nullclose, mm_readwrite, mm_readwrite, mm_ioctl, 75 nostop, notty, nopoll, mm_mmap, nokqfilter, D_MPSAFE 76 }; 77 #endif 78 79 /* 80 * mm_init: initialize memory device driver. 81 */ 82 void 83 mm_init(void) 84 { 85 vaddr_t pg; 86 87 mutex_init(&dev_mem_lock, MUTEX_DEFAULT, IPL_NONE); 88 89 /* Read-only zero-page. */ 90 pg = uvm_km_alloc(kernel_map, PAGE_SIZE, 0, UVM_KMF_WIRED|UVM_KMF_ZERO); 91 KASSERT(pg != 0); 92 pmap_protect(pmap_kernel(), pg, pg + PAGE_SIZE, VM_PROT_READ); 93 pmap_update(pmap_kernel()); 94 dev_zero_page = (void *)pg; 95 96 #ifndef __HAVE_MM_MD_CACHE_ALIASING 97 /* KVA for mappings during I/O. */ 98 dev_mem_addr = uvm_km_alloc(kernel_map, PAGE_SIZE, 0, 99 UVM_KMF_VAONLY|UVM_KMF_WAITVA); 100 KASSERT(dev_mem_addr != 0); 101 #else 102 dev_mem_addr = 0; 103 #endif 104 } 105 106 107 /* 108 * dev_mem_getva: get a special virtual address. If architecture requires, 109 * allocate VA according to PA, which avoids cache-aliasing issues. Use a 110 * constant, general mapping address otherwise. 111 */ 112 static inline vaddr_t 113 dev_mem_getva(paddr_t pa) 114 { 115 #ifdef __HAVE_MM_MD_CACHE_ALIASING 116 return uvm_km_alloc(kernel_map, PAGE_SIZE, 117 atop(pa) & uvmexp.colormask, 118 UVM_KMF_VAONLY | UVM_KMF_WAITVA | UVM_KMF_COLORMATCH); 119 #else 120 return dev_mem_addr; 121 #endif 122 } 123 124 static inline void 125 dev_mem_relva(paddr_t pa, vaddr_t va) 126 { 127 #ifdef __HAVE_MM_MD_CACHE_ALIASING 128 uvm_km_free(kernel_map, va, PAGE_SIZE, UVM_KMF_VAONLY); 129 #else 130 KASSERT(dev_mem_addr == va); 131 #endif 132 } 133 134 /* 135 * dev_kmem_readwrite: helper for DEV_MEM (/dev/mem) case of R/W. 136 */ 137 static int 138 dev_mem_readwrite(struct uio *uio, struct iovec *iov) 139 { 140 paddr_t paddr; 141 vaddr_t vaddr; 142 vm_prot_t prot; 143 size_t len, offset; 144 bool have_direct; 145 int error; 146 147 /* Check for wrap around. */ 148 if ((intptr_t)uio->uio_offset != uio->uio_offset) { 149 return EFAULT; 150 } 151 paddr = uio->uio_offset & ~PAGE_MASK; 152 prot = (uio->uio_rw == UIO_WRITE) ? VM_PROT_WRITE : VM_PROT_READ; 153 error = mm_md_physacc(paddr, prot); 154 if (error) { 155 return error; 156 } 157 offset = uio->uio_offset & PAGE_MASK; 158 len = MIN(uio->uio_resid, PAGE_SIZE - offset); 159 160 #ifdef __HAVE_MM_MD_DIRECT_MAPPED_PHYS 161 /* Is physical address directly mapped? Return VA. */ 162 have_direct = mm_md_direct_mapped_phys(paddr, &vaddr); 163 #else 164 vaddr = 0; 165 have_direct = false; 166 #endif 167 if (!have_direct) { 168 /* Get a special virtual address. */ 169 const vaddr_t va = dev_mem_getva(paddr); 170 171 /* Map selected KVA to physical address. */ 172 mutex_enter(&dev_mem_lock); 173 pmap_kenter_pa(va, paddr, prot, 0); 174 pmap_update(pmap_kernel()); 175 176 /* Perform I/O. */ 177 vaddr = va + offset; 178 error = uiomove((void *)vaddr, len, uio); 179 180 /* Unmap, flush before unlock. */ 181 pmap_kremove(va, PAGE_SIZE); 182 pmap_update(pmap_kernel()); 183 mutex_exit(&dev_mem_lock); 184 185 /* "Release" the virtual address. */ 186 dev_mem_relva(paddr, va); 187 } else { 188 /* Direct map, just perform I/O. */ 189 vaddr += offset; 190 error = uiomove((void *)vaddr, len, uio); 191 } 192 return error; 193 } 194 195 /* 196 * dev_kmem_readwrite: helper for DEV_KMEM (/dev/kmem) case of R/W. 197 */ 198 static int 199 dev_kmem_readwrite(struct uio *uio, struct iovec *iov) 200 { 201 void *addr; 202 size_t len, offset; 203 vm_prot_t prot; 204 int error; 205 bool md_kva; 206 207 /* Check for wrap around. */ 208 addr = (void *)(intptr_t)uio->uio_offset; 209 if ((uintptr_t)addr != uio->uio_offset) { 210 return EFAULT; 211 } 212 /* 213 * Handle non-page aligned offset. 214 * Otherwise, we operate in page-by-page basis. 215 */ 216 offset = uio->uio_offset & PAGE_MASK; 217 len = MIN(uio->uio_resid, PAGE_SIZE - offset); 218 prot = (uio->uio_rw == UIO_WRITE) ? VM_PROT_WRITE : VM_PROT_READ; 219 220 md_kva = false; 221 222 #ifdef __HAVE_MM_MD_DIRECT_MAPPED_IO 223 paddr_t paddr; 224 /* MD case: is this is a directly mapped address? */ 225 if (mm_md_direct_mapped_io(addr, &paddr)) { 226 /* If so, validate physical address. */ 227 error = mm_md_physacc(paddr, prot); 228 if (error) { 229 return error; 230 } 231 md_kva = true; 232 } 233 #endif 234 if (!md_kva) { 235 bool checked = false; 236 237 #ifdef __HAVE_MM_MD_KERNACC 238 /* MD check for the address. */ 239 error = mm_md_kernacc(addr, prot, &checked); 240 if (error) { 241 return error; 242 } 243 #endif 244 /* UVM check for the address (unless MD indicated to not). */ 245 if (!checked && !uvm_kernacc(addr, len, prot)) { 246 return EFAULT; 247 } 248 } 249 error = uiomove(addr, len, uio); 250 return error; 251 } 252 253 /* 254 * dev_zero_readwrite: helper for DEV_ZERO (/dev/null) case of R/W. 255 */ 256 static inline int 257 dev_zero_readwrite(struct uio *uio, struct iovec *iov) 258 { 259 size_t len; 260 261 /* Nothing to do for the write case. */ 262 if (uio->uio_rw == UIO_WRITE) { 263 uio->uio_resid = 0; 264 return 0; 265 } 266 /* 267 * Read in page-by-page basis, caller will continue. 268 * Cut appropriately for a single/last-iteration cases. 269 */ 270 len = MIN(iov->iov_len, PAGE_SIZE); 271 return uiomove(dev_zero_page, len, uio); 272 } 273 274 /* 275 * mm_readwrite: general memory R/W function. 276 */ 277 static int 278 mm_readwrite(dev_t dev, struct uio *uio, int flags) 279 { 280 struct iovec *iov; 281 int error; 282 283 #ifdef __HAVE_MM_MD_READWRITE 284 /* If defined - there are extra MD cases. */ 285 switch (minor(dev)) { 286 case DEV_MEM: 287 case DEV_KMEM: 288 case DEV_NULL: 289 case DEV_ZERO: 290 #if defined(COMPAT_16) && defined(__arm) 291 case _DEV_ZERO_oARM: 292 #endif 293 break; 294 default: 295 return mm_md_readwrite(dev, uio); 296 } 297 #endif 298 error = 0; 299 while (uio->uio_resid > 0 && error == 0) { 300 iov = uio->uio_iov; 301 if (iov->iov_len == 0) { 302 /* Processed; next I/O vector. */ 303 uio->uio_iov++; 304 uio->uio_iovcnt--; 305 KASSERT(uio->uio_iovcnt >= 0); 306 continue; 307 } 308 /* Helper functions will process in page-by-page basis. */ 309 switch (minor(dev)) { 310 case DEV_MEM: 311 error = dev_mem_readwrite(uio, iov); 312 break; 313 case DEV_KMEM: 314 error = dev_kmem_readwrite(uio, iov); 315 break; 316 case DEV_NULL: 317 if (uio->uio_rw == UIO_WRITE) { 318 uio->uio_resid = 0; 319 } 320 /* Break directly out of the loop. */ 321 return 0; 322 #if defined(COMPAT_16) && defined(__arm) 323 case _DEV_ZERO_oARM: 324 #endif 325 case DEV_ZERO: 326 error = dev_zero_readwrite(uio, iov); 327 break; 328 default: 329 error = ENXIO; 330 break; 331 } 332 } 333 return error; 334 } 335 336 /* 337 * mm_mmap: general mmap() handler. 338 */ 339 static paddr_t 340 mm_mmap(dev_t dev, off_t off, int acc) 341 { 342 vm_prot_t prot; 343 344 #ifdef __HAVE_MM_MD_MMAP 345 /* If defined - there are extra mmap() MD cases. */ 346 switch (minor(dev)) { 347 case DEV_MEM: 348 case DEV_KMEM: 349 case DEV_NULL: 350 #if defined(COMPAT_16) && defined(__arm) 351 case _DEV_ZERO_oARM: 352 #endif 353 case DEV_ZERO: 354 break; 355 default: 356 return mm_md_mmap(dev, off, acc); 357 } 358 #endif 359 /* 360 * /dev/null does not make sense, /dev/kmem is volatile and 361 * /dev/zero is handled in mmap already. 362 */ 363 if (minor(dev) != DEV_MEM) { 364 return -1; 365 } 366 367 prot = 0; 368 if (acc & PROT_EXEC) 369 prot |= VM_PROT_EXECUTE; 370 if (acc & PROT_READ) 371 prot |= VM_PROT_READ; 372 if (acc & PROT_WRITE) 373 prot |= VM_PROT_WRITE; 374 375 /* Validate the physical address. */ 376 if (mm_md_physacc(off, prot) != 0) { 377 return -1; 378 } 379 return off >> PGSHIFT; 380 } 381 382 static int 383 mm_ioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 384 { 385 386 switch (cmd) { 387 case FIONBIO: 388 /* We never block anyway. */ 389 return 0; 390 391 case FIOSETOWN: 392 case FIOGETOWN: 393 case TIOCGPGRP: 394 case TIOCSPGRP: 395 case TIOCGETA: 396 return ENOTTY; 397 398 case FIOASYNC: 399 if ((*(int *)data) == 0) { 400 return 0; 401 } 402 /* FALLTHROUGH */ 403 default: 404 return EOPNOTSUPP; 405 } 406 } 407