1 /* $NetBSD: mm.c,v 1.25 2024/08/25 11:29:38 uwe 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 1.25 2024/08/25 11:29:38 uwe 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_open(mm_open); 57 static dev_type_read(mm_readwrite); 58 static dev_type_mmap(mm_mmap); 59 static dev_type_ioctl(mm_ioctl); 60 61 const struct cdevsw mem_cdevsw = { 62 .d_open = mm_open, 63 .d_close = nullclose, 64 .d_read = mm_readwrite, 65 .d_write = mm_readwrite, 66 .d_ioctl = mm_ioctl, 67 .d_stop = nostop, 68 .d_tty = notty, 69 .d_poll = nopoll, 70 .d_mmap = mm_mmap, 71 .d_kqfilter = nokqfilter, 72 .d_discard = nodiscard, 73 .d_flag = D_MPSAFE 74 }; 75 76 #ifdef pmax /* XXX */ 77 const struct cdevsw mem_ultrix_cdevsw = { 78 .d_open = nullopen, 79 .d_close = nullclose, 80 .d_read = mm_readwrite, 81 .d_write = mm_readwrite, 82 .d_ioctl = mm_ioctl, 83 .d_stop = nostop, 84 .d_tty = notty, 85 .d_poll = nopoll, 86 .d_mmap = mm_mmap, 87 .d_kqfilter = nokqfilter, 88 .d_discard = nodiscard, 89 .d_flag = D_MPSAFE 90 }; 91 #endif 92 93 static int 94 mm_open(dev_t dev, int flag, int mode, struct lwp *l) 95 { 96 #ifdef __HAVE_MM_MD_OPEN 97 int error; 98 if ((error = mm_md_open(dev, flag, mode, l)) != 0) 99 return error; 100 #endif 101 l->l_proc->p_flag |= PK_KMEM; 102 return 0; 103 } 104 105 /* 106 * mm_init: initialize memory device driver. 107 */ 108 void 109 mm_init(void) 110 { 111 vaddr_t pg; 112 113 mutex_init(&dev_mem_lock, MUTEX_DEFAULT, IPL_NONE); 114 115 /* Read-only zero-page. */ 116 pg = uvm_km_alloc(kernel_map, PAGE_SIZE, 0, UVM_KMF_WIRED|UVM_KMF_ZERO); 117 KASSERT(pg != 0); 118 pmap_protect(pmap_kernel(), pg, pg + PAGE_SIZE, VM_PROT_READ); 119 pmap_update(pmap_kernel()); 120 dev_zero_page = (void *)pg; 121 122 #ifndef __HAVE_MM_MD_CACHE_ALIASING 123 /* KVA for mappings during I/O. */ 124 dev_mem_addr = uvm_km_alloc(kernel_map, PAGE_SIZE, 0, 125 UVM_KMF_VAONLY|UVM_KMF_WAITVA); 126 KASSERT(dev_mem_addr != 0); 127 #else 128 dev_mem_addr = 0; 129 #endif 130 } 131 132 133 /* 134 * dev_mem_getva: get a special virtual address. If architecture requires, 135 * allocate VA according to PA, which avoids cache-aliasing issues. Use a 136 * constant, general mapping address otherwise. 137 */ 138 static inline vaddr_t 139 dev_mem_getva(paddr_t pa, int color) 140 { 141 #ifdef __HAVE_MM_MD_CACHE_ALIASING 142 return uvm_km_alloc(kernel_map, PAGE_SIZE, 143 color & uvmexp.colormask, 144 UVM_KMF_VAONLY | UVM_KMF_WAITVA | UVM_KMF_COLORMATCH); 145 #else 146 return dev_mem_addr; 147 #endif 148 } 149 150 static inline void 151 dev_mem_relva(paddr_t pa, vaddr_t va) 152 { 153 #ifdef __HAVE_MM_MD_CACHE_ALIASING 154 uvm_km_free(kernel_map, va, PAGE_SIZE, UVM_KMF_VAONLY); 155 #else 156 KASSERT(dev_mem_addr == va); 157 #endif 158 } 159 160 /* 161 * dev_kmem_readwrite: helper for DEV_MEM (/dev/mem) case of R/W. 162 */ 163 static int 164 dev_mem_readwrite(struct uio *uio, struct iovec *iov) 165 { 166 paddr_t paddr; 167 vaddr_t vaddr; 168 vm_prot_t prot; 169 size_t len, offset; 170 bool have_direct; 171 int error; 172 int color = 0; 173 174 /* Check for wrap around. */ 175 if ((uintptr_t)uio->uio_offset != uio->uio_offset) { 176 return EFAULT; 177 } 178 paddr = uio->uio_offset & ~PAGE_MASK; 179 prot = (uio->uio_rw == UIO_WRITE) ? VM_PROT_WRITE : VM_PROT_READ; 180 error = mm_md_physacc(paddr, prot); 181 if (error) { 182 return error; 183 } 184 offset = uio->uio_offset & PAGE_MASK; 185 len = MIN(uio->uio_resid, PAGE_SIZE - offset); 186 187 #ifdef __HAVE_MM_MD_CACHE_ALIASING 188 have_direct = mm_md_page_color(paddr, &color); 189 #else 190 have_direct = true; 191 color = 0; 192 #endif 193 194 #ifdef __HAVE_MM_MD_DIRECT_MAPPED_PHYS 195 /* Is physical address directly mapped? Return VA. */ 196 if (have_direct) 197 have_direct = mm_md_direct_mapped_phys(paddr, &vaddr); 198 #else 199 vaddr = 0; 200 have_direct = false; 201 #endif 202 if (!have_direct) { 203 /* Get a special virtual address. */ 204 const vaddr_t va = dev_mem_getva(paddr, color); 205 206 /* Map selected KVA to physical address. */ 207 mutex_enter(&dev_mem_lock); 208 pmap_kenter_pa(va, paddr, prot, 0); 209 pmap_update(pmap_kernel()); 210 211 /* Perform I/O. */ 212 vaddr = va + offset; 213 error = uiomove((void *)vaddr, len, uio); 214 215 /* Unmap, flush before unlock. */ 216 pmap_kremove(va, PAGE_SIZE); 217 pmap_update(pmap_kernel()); 218 mutex_exit(&dev_mem_lock); 219 220 /* "Release" the virtual address. */ 221 dev_mem_relva(paddr, va); 222 } else { 223 /* Direct map, just perform I/O. */ 224 vaddr += offset; 225 error = uiomove((void *)vaddr, len, uio); 226 } 227 return error; 228 } 229 230 /* 231 * dev_kmem_readwrite: helper for DEV_KMEM (/dev/kmem) case of R/W. 232 */ 233 static int 234 dev_kmem_readwrite(struct uio *uio, struct iovec *iov) 235 { 236 void *addr; 237 size_t len, offset; 238 vm_prot_t prot; 239 int error; 240 bool md_kva; 241 242 /* Check for wrap around. */ 243 addr = (void *)(intptr_t)uio->uio_offset; 244 if ((uintptr_t)addr != uio->uio_offset) { 245 return EFAULT; 246 } 247 /* 248 * Handle non-page aligned offset. 249 * Otherwise, we operate in page-by-page basis. 250 */ 251 offset = uio->uio_offset & PAGE_MASK; 252 len = MIN(uio->uio_resid, PAGE_SIZE - offset); 253 prot = (uio->uio_rw == UIO_WRITE) ? VM_PROT_WRITE : VM_PROT_READ; 254 255 md_kva = false; 256 257 #ifdef __HAVE_MM_MD_DIRECT_MAPPED_IO 258 paddr_t paddr; 259 /* MD case: is this is a directly mapped address? */ 260 if (mm_md_direct_mapped_io(addr, &paddr)) { 261 /* If so, validate physical address. */ 262 error = mm_md_physacc(paddr, prot); 263 if (error) { 264 return error; 265 } 266 md_kva = true; 267 } 268 #endif 269 if (!md_kva) { 270 bool checked = false; 271 272 #ifdef __HAVE_MM_MD_KERNACC 273 /* MD check for the address. */ 274 error = mm_md_kernacc(addr, prot, &checked); 275 if (error) { 276 return error; 277 } 278 #endif 279 /* UVM check for the address (unless MD indicated to not). */ 280 if (!checked && !uvm_kernacc(addr, len, prot)) { 281 return EFAULT; 282 } 283 } 284 error = uiomove(addr, len, uio); 285 return error; 286 } 287 288 /* 289 * dev_zero_readwrite: helper for DEV_ZERO (/dev/null) case of R/W. 290 */ 291 static inline int 292 dev_zero_readwrite(struct uio *uio, struct iovec *iov) 293 { 294 size_t len; 295 296 /* Nothing to do for the write case. */ 297 if (uio->uio_rw == UIO_WRITE) { 298 uio->uio_resid = 0; 299 return 0; 300 } 301 /* 302 * Read in page-by-page basis, caller will continue. 303 * Cut appropriately for a single/last-iteration cases. 304 */ 305 len = MIN(iov->iov_len, PAGE_SIZE); 306 return uiomove(dev_zero_page, len, uio); 307 } 308 309 /* 310 * mm_readwrite: general memory R/W function. 311 */ 312 static int 313 mm_readwrite(dev_t dev, struct uio *uio, int flags) 314 { 315 struct iovec *iov; 316 int error; 317 318 #ifdef __HAVE_MM_MD_READWRITE 319 /* If defined - there are extra MD cases. */ 320 switch (minor(dev)) { 321 case DEV_MEM: 322 case DEV_KMEM: 323 case DEV_NULL: 324 case DEV_ZERO: 325 #if defined(COMPAT_16) && defined(__arm) 326 case _DEV_ZERO_oARM: 327 #endif 328 break; 329 default: 330 return mm_md_readwrite(dev, uio); 331 } 332 #endif 333 error = 0; 334 while (uio->uio_resid > 0 && error == 0) { 335 iov = uio->uio_iov; 336 if (iov->iov_len == 0) { 337 /* Processed; next I/O vector. */ 338 uio->uio_iov++; 339 uio->uio_iovcnt--; 340 KASSERT(uio->uio_iovcnt >= 0); 341 continue; 342 } 343 /* Helper functions will process in page-by-page basis. */ 344 switch (minor(dev)) { 345 case DEV_MEM: 346 error = dev_mem_readwrite(uio, iov); 347 break; 348 case DEV_KMEM: 349 error = dev_kmem_readwrite(uio, iov); 350 break; 351 case DEV_NULL: 352 if (uio->uio_rw == UIO_WRITE) { 353 uio->uio_resid = 0; 354 } 355 /* Break directly out of the loop. */ 356 return 0; 357 case DEV_FULL: 358 if (uio->uio_rw == UIO_WRITE) { 359 return ENOSPC; 360 } 361 #if defined(COMPAT_16) && defined(__arm) 362 /* FALLTHROUGH */ 363 case _DEV_ZERO_oARM: 364 #endif 365 /* FALLTHROUGH */ 366 case DEV_ZERO: 367 error = dev_zero_readwrite(uio, iov); 368 break; 369 default: 370 error = ENXIO; 371 break; 372 } 373 } 374 return error; 375 } 376 377 /* 378 * mm_mmap: general mmap() handler. 379 */ 380 static paddr_t 381 mm_mmap(dev_t dev, off_t off, int acc) 382 { 383 vm_prot_t prot; 384 385 #ifdef __HAVE_MM_MD_MMAP 386 /* If defined - there are extra mmap() MD cases. */ 387 switch (minor(dev)) { 388 case DEV_MEM: 389 case DEV_KMEM: 390 case DEV_NULL: 391 #if defined(COMPAT_16) && defined(__arm) 392 case _DEV_ZERO_oARM: 393 #endif 394 case DEV_ZERO: 395 break; 396 default: 397 return mm_md_mmap(dev, off, acc); 398 } 399 #endif 400 /* 401 * /dev/null does not make sense, /dev/kmem is volatile and 402 * /dev/zero is handled in mmap already. 403 */ 404 if (minor(dev) != DEV_MEM) { 405 return -1; 406 } 407 408 prot = 0; 409 if (acc & PROT_EXEC) 410 prot |= VM_PROT_EXECUTE; 411 if (acc & PROT_READ) 412 prot |= VM_PROT_READ; 413 if (acc & PROT_WRITE) 414 prot |= VM_PROT_WRITE; 415 416 /* Validate the physical address. */ 417 if (mm_md_physacc(off, prot) != 0) { 418 return -1; 419 } 420 return off >> PGSHIFT; 421 } 422 423 static int 424 mm_ioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l) 425 { 426 427 switch (cmd) { 428 case FIONBIO: 429 /* We never block anyway. */ 430 return 0; 431 432 case FIOSETOWN: 433 case FIOGETOWN: 434 case TIOCGPGRP: 435 case TIOCSPGRP: 436 case TIOCGETA: 437 return ENOTTY; 438 439 case FIOASYNC: 440 if ((*(int *)data) == 0) { 441 return 0; 442 } 443 /* FALLTHROUGH */ 444 default: 445 return EOPNOTSUPP; 446 } 447 } 448