Home | History | Annotate | Line # | Download | only in dev
mm.c revision 1.16 
      1 /*	$NetBSD: mm.c,v 1.16 2012/02/21 21:57:06 rmind 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 1.16 2012/02/21 21:57:06 rmind 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 	const vsize_t coloroff = trunc_page(pa) & ptoa(uvmexp.colormask);
    117 	const vaddr_t kva = uvm_km_alloc(kernel_map, PAGE_SIZE + coloroff,
    118 	    ptoa(uvmexp.ncolors), UVM_KMF_VAONLY | UVM_KMF_WAITVA);
    119 
    120 	return kva + coloroff;
    121 #else
    122 	return dev_mem_addr;
    123 #endif
    124 }
    125 
    126 static inline void
    127 dev_mem_relva(paddr_t pa, vaddr_t va)
    128 {
    129 #ifdef __HAVE_MM_MD_CACHE_ALIASING
    130 	const vsize_t coloroff = trunc_page(pa) & ptoa(uvmexp.colormask);
    131 	const vaddr_t origva = va - coloroff;
    132 
    133 	uvm_km_free(kernel_map, origva, PAGE_SIZE + coloroff, UVM_KMF_VAONLY);
    134 #else
    135 	KASSERT(dev_mem_addr == va);
    136 #endif
    137 }
    138 
    139 /*
    140  * dev_kmem_readwrite: helper for DEV_MEM (/dev/mem) case of R/W.
    141  */
    142 static int
    143 dev_mem_readwrite(struct uio *uio, struct iovec *iov)
    144 {
    145 	paddr_t paddr;
    146 	vaddr_t vaddr;
    147 	vm_prot_t prot;
    148 	size_t len, offset;
    149 	bool have_direct;
    150 	int error;
    151 
    152 	/* Check for wrap around. */
    153 	if ((intptr_t)uio->uio_offset != uio->uio_offset) {
    154 		return EFAULT;
    155 	}
    156 	paddr = uio->uio_offset & ~PAGE_MASK;
    157 	prot = (uio->uio_rw == UIO_WRITE) ? VM_PROT_WRITE : VM_PROT_READ;
    158 	error = mm_md_physacc(paddr, prot);
    159 	if (error) {
    160 		return error;
    161 	}
    162 	offset = uio->uio_offset & PAGE_MASK;
    163 	len = MIN(uio->uio_resid, PAGE_SIZE - offset);
    164 
    165 #ifdef __HAVE_MM_MD_DIRECT_MAPPED_PHYS
    166 	/* Is physical address directly mapped?  Return VA. */
    167 	have_direct = mm_md_direct_mapped_phys(paddr, &vaddr);
    168 #else
    169 	vaddr = 0;
    170 	have_direct = false;
    171 #endif
    172 	if (!have_direct) {
    173 		/* Get a special virtual address. */
    174 		const vaddr_t va = dev_mem_getva(paddr);
    175 
    176 		/* Map selected KVA to physical address. */
    177 		mutex_enter(&dev_mem_lock);
    178 		pmap_kenter_pa(va, paddr, prot, 0);
    179 		pmap_update(pmap_kernel());
    180 
    181 		/* Perform I/O. */
    182 		vaddr = va + offset;
    183 		error = uiomove((void *)vaddr, len, uio);
    184 
    185 		/* Unmap, flush before unlock. */
    186 		pmap_kremove(va, PAGE_SIZE);
    187 		pmap_update(pmap_kernel());
    188 		mutex_exit(&dev_mem_lock);
    189 
    190 		/* "Release" the virtual address. */
    191 		dev_mem_relva(paddr, va);
    192 	} else {
    193 		/* Direct map, just perform I/O. */
    194 		vaddr += offset;
    195 		error = uiomove((void *)vaddr, len, uio);
    196 	}
    197 	return error;
    198 }
    199 
    200 /*
    201  * dev_kmem_readwrite: helper for DEV_KMEM (/dev/kmem) case of R/W.
    202  */
    203 static int
    204 dev_kmem_readwrite(struct uio *uio, struct iovec *iov)
    205 {
    206 	void *addr;
    207 	size_t len, offset;
    208 	vm_prot_t prot;
    209 	int error;
    210 	bool md_kva;
    211 
    212 	/* Check for wrap around. */
    213 	addr = (void *)(intptr_t)uio->uio_offset;
    214 	if ((uintptr_t)addr != uio->uio_offset) {
    215 		return EFAULT;
    216 	}
    217 	/*
    218 	 * Handle non-page aligned offset.
    219 	 * Otherwise, we operate in page-by-page basis.
    220 	 */
    221 	offset = uio->uio_offset & PAGE_MASK;
    222 	len = MIN(uio->uio_resid, PAGE_SIZE - offset);
    223 	prot = (uio->uio_rw == UIO_WRITE) ? VM_PROT_WRITE : VM_PROT_READ;
    224 
    225 	md_kva = false;
    226 
    227 #ifdef __HAVE_MM_MD_DIRECT_MAPPED_IO
    228 	paddr_t paddr;
    229 	/* MD case: is this is a directly mapped address? */
    230 	if (mm_md_direct_mapped_io(addr, &paddr)) {
    231 		/* If so, validate physical address. */
    232 		error = mm_md_physacc(paddr, prot);
    233 		if (error) {
    234 			return error;
    235 		}
    236 		md_kva = true;
    237 	}
    238 #endif
    239 	if (!md_kva) {
    240 		bool checked = false;
    241 
    242 #ifdef __HAVE_MM_MD_KERNACC
    243 		/* MD check for the address. */
    244 		error = mm_md_kernacc(addr, prot, &checked);
    245 		if (error) {
    246 			return error;
    247 		}
    248 #endif
    249 		/* UVM check for the address (unless MD indicated to not). */
    250 		if (!checked && !uvm_kernacc(addr, len, prot)) {
    251 			return EFAULT;
    252 		}
    253 	}
    254 	error = uiomove(addr, len, uio);
    255 	return error;
    256 }
    257 
    258 /*
    259  * dev_zero_readwrite: helper for DEV_ZERO (/dev/null) case of R/W.
    260  */
    261 static inline int
    262 dev_zero_readwrite(struct uio *uio, struct iovec *iov)
    263 {
    264 	size_t len;
    265 
    266 	/* Nothing to do for the write case. */
    267 	if (uio->uio_rw == UIO_WRITE) {
    268 		uio->uio_resid = 0;
    269 		return 0;
    270 	}
    271 	/*
    272 	 * Read in page-by-page basis, caller will continue.
    273 	 * Cut appropriately for a single/last-iteration cases.
    274 	 */
    275 	len = MIN(iov->iov_len, PAGE_SIZE);
    276 	return uiomove(dev_zero_page, len, uio);
    277 }
    278 
    279 /*
    280  * mm_readwrite: general memory R/W function.
    281  */
    282 static int
    283 mm_readwrite(dev_t dev, struct uio *uio, int flags)
    284 {
    285 	struct iovec *iov;
    286 	int error;
    287 
    288 #ifdef __HAVE_MM_MD_READWRITE
    289 	/* If defined - there are extra MD cases. */
    290 	switch (minor(dev)) {
    291 	case DEV_MEM:
    292 	case DEV_KMEM:
    293 	case DEV_NULL:
    294 	case DEV_ZERO:
    295 #if defined(COMPAT_16) && defined(__arm)
    296 	case _DEV_ZERO_oARM:
    297 #endif
    298 		break;
    299 	default:
    300 		return mm_md_readwrite(dev, uio);
    301 	}
    302 #endif
    303 	error = 0;
    304 	while (uio->uio_resid > 0 && error == 0) {
    305 		iov = uio->uio_iov;
    306 		if (iov->iov_len == 0) {
    307 			/* Processed; next I/O vector. */
    308 			uio->uio_iov++;
    309 			uio->uio_iovcnt--;
    310 			KASSERT(uio->uio_iovcnt >= 0);
    311 			continue;
    312 		}
    313 		/* Helper functions will process in page-by-page basis. */
    314 		switch (minor(dev)) {
    315 		case DEV_MEM:
    316 			error = dev_mem_readwrite(uio, iov);
    317 			break;
    318 		case DEV_KMEM:
    319 			error = dev_kmem_readwrite(uio, iov);
    320 			break;
    321 		case DEV_NULL:
    322 			if (uio->uio_rw == UIO_WRITE) {
    323 				uio->uio_resid = 0;
    324 			}
    325 			/* Break directly out of the loop. */
    326 			return 0;
    327 #if defined(COMPAT_16) && defined(__arm)
    328 		case _DEV_ZERO_oARM:
    329 #endif
    330 		case DEV_ZERO:
    331 			error = dev_zero_readwrite(uio, iov);
    332 			break;
    333 		default:
    334 			error = ENXIO;
    335 			break;
    336 		}
    337 	}
    338 	return error;
    339 }
    340 
    341 /*
    342  * mm_mmap: general mmap() handler.
    343  */
    344 static paddr_t
    345 mm_mmap(dev_t dev, off_t off, int acc)
    346 {
    347 	vm_prot_t prot;
    348 
    349 #ifdef __HAVE_MM_MD_MMAP
    350 	/* If defined - there are extra mmap() MD cases. */
    351 	switch (minor(dev)) {
    352 	case DEV_MEM:
    353 	case DEV_KMEM:
    354 	case DEV_NULL:
    355 #if defined(COMPAT_16) && defined(__arm)
    356 	case _DEV_ZERO_oARM:
    357 #endif
    358 	case DEV_ZERO:
    359 		break;
    360 	default:
    361 		return mm_md_mmap(dev, off, acc);
    362 	}
    363 #endif
    364 	/*
    365 	 * /dev/null does not make sense, /dev/kmem is volatile and
    366 	 * /dev/zero is handled in mmap already.
    367 	 */
    368 	if (minor(dev) != DEV_MEM) {
    369 		return -1;
    370 	}
    371 
    372 	prot = 0;
    373 	if (acc & PROT_EXEC)
    374 		prot |= VM_PROT_EXECUTE;
    375 	if (acc & PROT_READ)
    376 		prot |= VM_PROT_READ;
    377 	if (acc & PROT_WRITE)
    378 		prot |= VM_PROT_WRITE;
    379 
    380 	/* Validate the physical address. */
    381 	if (mm_md_physacc(off, prot) != 0) {
    382 		return -1;
    383 	}
    384 	return off >> PGSHIFT;
    385 }
    386 
    387 static int
    388 mm_ioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
    389 {
    390 
    391 	switch (cmd) {
    392 	case FIONBIO:
    393 		/* We never block anyway. */
    394 		return 0;
    395 
    396 	case FIOSETOWN:
    397 	case FIOGETOWN:
    398 	case TIOCGPGRP:
    399 	case TIOCSPGRP:
    400 	case TIOCGETA:
    401 		return ENOTTY;
    402 
    403 	case FIOASYNC:
    404 		if ((*(int *)data) == 0) {
    405 			return 0;
    406 		}
    407 		/* FALLTHROUGH */
    408 	default:
    409 		return EOPNOTSUPP;
    410 	}
    411 }
    412