Home | History | Annotate | Line # | Download | only in dev
mm.c revision 1.15.2.3 
      1 /*	$NetBSD: mm.c,v 1.15.2.3 2014/05/22 11:40:19 yamt 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 1.15.2.3 2014/05/22 11:40:19 yamt 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 	.d_open = mm_md_open,
     64 #else
     65 	.d_open = nullopen,
     66 #endif
     67 	.d_close = nullclose,
     68 	.d_read = mm_readwrite,
     69 	.d_write = mm_readwrite,
     70 	.d_ioctl = mm_ioctl,
     71 	.d_stop = nostop,
     72 	.d_tty = notty,
     73 	.d_poll = nopoll,
     74 	.d_mmap = mm_mmap,
     75 	.d_kqfilter = nokqfilter,
     76 	.d_flag = D_MPSAFE
     77 };
     78 
     79 #ifdef pmax	/* XXX */
     80 const struct cdevsw mem_ultrix_cdevsw = {
     81 	.d_open = nullopen,
     82 	.d_close = nullclose,
     83 	.d_read = mm_readwrite,
     84 	.d_write = mm_readwrite,
     85 	.d_ioctl = mm_ioctl,
     86 	.d_stop = nostop,
     87 	.d_tty = notty,
     88 	.d_poll = nopoll,
     89 	.d_mmap = mm_mmap,
     90 	.d_kqfilter = nokqfilter,
     91 	.d_flag = D_MPSAFE
     92 };
     93 #endif
     94 
     95 /*
     96  * mm_init: initialize memory device driver.
     97  */
     98 void
     99 mm_init(void)
    100 {
    101 	vaddr_t pg;
    102 
    103 	mutex_init(&dev_mem_lock, MUTEX_DEFAULT, IPL_NONE);
    104 
    105 	/* Read-only zero-page. */
    106 	pg = uvm_km_alloc(kernel_map, PAGE_SIZE, 0, UVM_KMF_WIRED|UVM_KMF_ZERO);
    107 	KASSERT(pg != 0);
    108 	pmap_protect(pmap_kernel(), pg, pg + PAGE_SIZE, VM_PROT_READ);
    109 	pmap_update(pmap_kernel());
    110 	dev_zero_page = (void *)pg;
    111 
    112 #ifndef __HAVE_MM_MD_CACHE_ALIASING
    113 	/* KVA for mappings during I/O. */
    114 	dev_mem_addr = uvm_km_alloc(kernel_map, PAGE_SIZE, 0,
    115 	    UVM_KMF_VAONLY|UVM_KMF_WAITVA);
    116 	KASSERT(dev_mem_addr != 0);
    117 #else
    118 	dev_mem_addr = 0;
    119 #endif
    120 }
    121 
    122 
    123 /*
    124  * dev_mem_getva: get a special virtual address.  If architecture requires,
    125  * allocate VA according to PA, which avoids cache-aliasing issues.  Use a
    126  * constant, general mapping address otherwise.
    127  */
    128 static inline vaddr_t
    129 dev_mem_getva(paddr_t pa)
    130 {
    131 #ifdef __HAVE_MM_MD_CACHE_ALIASING
    132 	return uvm_km_alloc(kernel_map, PAGE_SIZE,
    133 	    atop(pa) & uvmexp.colormask,
    134 	    UVM_KMF_VAONLY | UVM_KMF_WAITVA | UVM_KMF_COLORMATCH);
    135 #else
    136 	return dev_mem_addr;
    137 #endif
    138 }
    139 
    140 static inline void
    141 dev_mem_relva(paddr_t pa, vaddr_t va)
    142 {
    143 #ifdef __HAVE_MM_MD_CACHE_ALIASING
    144 	uvm_km_free(kernel_map, va, PAGE_SIZE, UVM_KMF_VAONLY);
    145 #else
    146 	KASSERT(dev_mem_addr == va);
    147 #endif
    148 }
    149 
    150 /*
    151  * dev_kmem_readwrite: helper for DEV_MEM (/dev/mem) case of R/W.
    152  */
    153 static int
    154 dev_mem_readwrite(struct uio *uio, struct iovec *iov)
    155 {
    156 	paddr_t paddr;
    157 	vaddr_t vaddr;
    158 	vm_prot_t prot;
    159 	size_t len, offset;
    160 	bool have_direct;
    161 	int error;
    162 
    163 	/* Check for wrap around. */
    164 	if ((intptr_t)uio->uio_offset != uio->uio_offset) {
    165 		return EFAULT;
    166 	}
    167 	paddr = uio->uio_offset & ~PAGE_MASK;
    168 	prot = (uio->uio_rw == UIO_WRITE) ? VM_PROT_WRITE : VM_PROT_READ;
    169 	error = mm_md_physacc(paddr, prot);
    170 	if (error) {
    171 		return error;
    172 	}
    173 	offset = uio->uio_offset & PAGE_MASK;
    174 	len = MIN(uio->uio_resid, PAGE_SIZE - offset);
    175 
    176 #ifdef __HAVE_MM_MD_DIRECT_MAPPED_PHYS
    177 	/* Is physical address directly mapped?  Return VA. */
    178 	have_direct = mm_md_direct_mapped_phys(paddr, &vaddr);
    179 #else
    180 	vaddr = 0;
    181 	have_direct = false;
    182 #endif
    183 	if (!have_direct) {
    184 		/* Get a special virtual address. */
    185 		const vaddr_t va = dev_mem_getva(paddr);
    186 
    187 		/* Map selected KVA to physical address. */
    188 		mutex_enter(&dev_mem_lock);
    189 		pmap_kenter_pa(va, paddr, prot, 0);
    190 		pmap_update(pmap_kernel());
    191 
    192 		/* Perform I/O. */
    193 		vaddr = va + offset;
    194 		error = uiomove((void *)vaddr, len, uio);
    195 
    196 		/* Unmap, flush before unlock. */
    197 		pmap_kremove(va, PAGE_SIZE);
    198 		pmap_update(pmap_kernel());
    199 		mutex_exit(&dev_mem_lock);
    200 
    201 		/* "Release" the virtual address. */
    202 		dev_mem_relva(paddr, va);
    203 	} else {
    204 		/* Direct map, just perform I/O. */
    205 		vaddr += offset;
    206 		error = uiomove((void *)vaddr, len, uio);
    207 	}
    208 	return error;
    209 }
    210 
    211 /*
    212  * dev_kmem_readwrite: helper for DEV_KMEM (/dev/kmem) case of R/W.
    213  */
    214 static int
    215 dev_kmem_readwrite(struct uio *uio, struct iovec *iov)
    216 {
    217 	void *addr;
    218 	size_t len, offset;
    219 	vm_prot_t prot;
    220 	int error;
    221 	bool md_kva;
    222 
    223 	/* Check for wrap around. */
    224 	addr = (void *)(intptr_t)uio->uio_offset;
    225 	if ((uintptr_t)addr != uio->uio_offset) {
    226 		return EFAULT;
    227 	}
    228 	/*
    229 	 * Handle non-page aligned offset.
    230 	 * Otherwise, we operate in page-by-page basis.
    231 	 */
    232 	offset = uio->uio_offset & PAGE_MASK;
    233 	len = MIN(uio->uio_resid, PAGE_SIZE - offset);
    234 	prot = (uio->uio_rw == UIO_WRITE) ? VM_PROT_WRITE : VM_PROT_READ;
    235 
    236 	md_kva = false;
    237 
    238 #ifdef __HAVE_MM_MD_DIRECT_MAPPED_IO
    239 	paddr_t paddr;
    240 	/* MD case: is this is a directly mapped address? */
    241 	if (mm_md_direct_mapped_io(addr, &paddr)) {
    242 		/* If so, validate physical address. */
    243 		error = mm_md_physacc(paddr, prot);
    244 		if (error) {
    245 			return error;
    246 		}
    247 		md_kva = true;
    248 	}
    249 #endif
    250 	if (!md_kva) {
    251 		bool checked = false;
    252 
    253 #ifdef __HAVE_MM_MD_KERNACC
    254 		/* MD check for the address. */
    255 		error = mm_md_kernacc(addr, prot, &checked);
    256 		if (error) {
    257 			return error;
    258 		}
    259 #endif
    260 		/* UVM check for the address (unless MD indicated to not). */
    261 		if (!checked && !uvm_kernacc(addr, len, prot)) {
    262 			return EFAULT;
    263 		}
    264 	}
    265 	error = uiomove(addr, len, uio);
    266 	return error;
    267 }
    268 
    269 /*
    270  * dev_zero_readwrite: helper for DEV_ZERO (/dev/null) case of R/W.
    271  */
    272 static inline int
    273 dev_zero_readwrite(struct uio *uio, struct iovec *iov)
    274 {
    275 	size_t len;
    276 
    277 	/* Nothing to do for the write case. */
    278 	if (uio->uio_rw == UIO_WRITE) {
    279 		uio->uio_resid = 0;
    280 		return 0;
    281 	}
    282 	/*
    283 	 * Read in page-by-page basis, caller will continue.
    284 	 * Cut appropriately for a single/last-iteration cases.
    285 	 */
    286 	len = MIN(iov->iov_len, PAGE_SIZE);
    287 	return uiomove(dev_zero_page, len, uio);
    288 }
    289 
    290 /*
    291  * mm_readwrite: general memory R/W function.
    292  */
    293 static int
    294 mm_readwrite(dev_t dev, struct uio *uio, int flags)
    295 {
    296 	struct iovec *iov;
    297 	int error;
    298 
    299 #ifdef __HAVE_MM_MD_READWRITE
    300 	/* If defined - there are extra MD cases. */
    301 	switch (minor(dev)) {
    302 	case DEV_MEM:
    303 	case DEV_KMEM:
    304 	case DEV_NULL:
    305 	case DEV_ZERO:
    306 #if defined(COMPAT_16) && defined(__arm)
    307 	case _DEV_ZERO_oARM:
    308 #endif
    309 		break;
    310 	default:
    311 		return mm_md_readwrite(dev, uio);
    312 	}
    313 #endif
    314 	error = 0;
    315 	while (uio->uio_resid > 0 && error == 0) {
    316 		iov = uio->uio_iov;
    317 		if (iov->iov_len == 0) {
    318 			/* Processed; next I/O vector. */
    319 			uio->uio_iov++;
    320 			uio->uio_iovcnt--;
    321 			KASSERT(uio->uio_iovcnt >= 0);
    322 			continue;
    323 		}
    324 		/* Helper functions will process in page-by-page basis. */
    325 		switch (minor(dev)) {
    326 		case DEV_MEM:
    327 			error = dev_mem_readwrite(uio, iov);
    328 			break;
    329 		case DEV_KMEM:
    330 			error = dev_kmem_readwrite(uio, iov);
    331 			break;
    332 		case DEV_NULL:
    333 			if (uio->uio_rw == UIO_WRITE) {
    334 				uio->uio_resid = 0;
    335 			}
    336 			/* Break directly out of the loop. */
    337 			return 0;
    338 #if defined(COMPAT_16) && defined(__arm)
    339 		case _DEV_ZERO_oARM:
    340 #endif
    341 		case DEV_ZERO:
    342 			error = dev_zero_readwrite(uio, iov);
    343 			break;
    344 		default:
    345 			error = ENXIO;
    346 			break;
    347 		}
    348 	}
    349 	return error;
    350 }
    351 
    352 /*
    353  * mm_mmap: general mmap() handler.
    354  */
    355 static paddr_t
    356 mm_mmap(dev_t dev, off_t off, int acc)
    357 {
    358 	vm_prot_t prot;
    359 
    360 #ifdef __HAVE_MM_MD_MMAP
    361 	/* If defined - there are extra mmap() MD cases. */
    362 	switch (minor(dev)) {
    363 	case DEV_MEM:
    364 	case DEV_KMEM:
    365 	case DEV_NULL:
    366 #if defined(COMPAT_16) && defined(__arm)
    367 	case _DEV_ZERO_oARM:
    368 #endif
    369 	case DEV_ZERO:
    370 		break;
    371 	default:
    372 		return mm_md_mmap(dev, off, acc);
    373 	}
    374 #endif
    375 	/*
    376 	 * /dev/null does not make sense, /dev/kmem is volatile and
    377 	 * /dev/zero is handled in mmap already.
    378 	 */
    379 	if (minor(dev) != DEV_MEM) {
    380 		return -1;
    381 	}
    382 
    383 	prot = 0;
    384 	if (acc & PROT_EXEC)
    385 		prot |= VM_PROT_EXECUTE;
    386 	if (acc & PROT_READ)
    387 		prot |= VM_PROT_READ;
    388 	if (acc & PROT_WRITE)
    389 		prot |= VM_PROT_WRITE;
    390 
    391 	/* Validate the physical address. */
    392 	if (mm_md_physacc(off, prot) != 0) {
    393 		return -1;
    394 	}
    395 	return off >> PGSHIFT;
    396 }
    397 
    398 static int
    399 mm_ioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
    400 {
    401 
    402 	switch (cmd) {
    403 	case FIONBIO:
    404 		/* We never block anyway. */
    405 		return 0;
    406 
    407 	case FIOSETOWN:
    408 	case FIOGETOWN:
    409 	case TIOCGPGRP:
    410 	case TIOCSPGRP:
    411 	case TIOCGETA:
    412 		return ENOTTY;
    413 
    414 	case FIOASYNC:
    415 		if ((*(int *)data) == 0) {
    416 			return 0;
    417 		}
    418 		/* FALLTHROUGH */
    419 	default:
    420 		return EOPNOTSUPP;
    421 	}
    422 }
    423