xref: /netbsd-src/sys/dev/mm.c (revision aad9773e38ed2370a628a6416e098f9008fc10a7)
1 /*	$NetBSD: mm.c,v 1.19 2014/07/25 08:10:35 dholland 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 1.19 2014/07/25 08:10:35 dholland 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_discard = nodiscard,
77 	.d_flag = D_MPSAFE
78 };
79 
80 #ifdef pmax	/* XXX */
81 const struct cdevsw mem_ultrix_cdevsw = {
82 	.d_open = nullopen,
83 	.d_close = nullclose,
84 	.d_read = mm_readwrite,
85 	.d_write = mm_readwrite,
86 	.d_ioctl = mm_ioctl,
87 	.d_stop = nostop,
88 	.d_tty = notty,
89 	.d_poll = nopoll,
90 	.d_mmap = mm_mmap,
91 	.d_kqfilter = nokqfilter,
92 	.d_discard = nodiscard,
93 	.d_flag = D_MPSAFE
94 };
95 #endif
96 
97 /*
98  * mm_init: initialize memory device driver.
99  */
100 void
101 mm_init(void)
102 {
103 	vaddr_t pg;
104 
105 	mutex_init(&dev_mem_lock, MUTEX_DEFAULT, IPL_NONE);
106 
107 	/* Read-only zero-page. */
108 	pg = uvm_km_alloc(kernel_map, PAGE_SIZE, 0, UVM_KMF_WIRED|UVM_KMF_ZERO);
109 	KASSERT(pg != 0);
110 	pmap_protect(pmap_kernel(), pg, pg + PAGE_SIZE, VM_PROT_READ);
111 	pmap_update(pmap_kernel());
112 	dev_zero_page = (void *)pg;
113 
114 #ifndef __HAVE_MM_MD_CACHE_ALIASING
115 	/* KVA for mappings during I/O. */
116 	dev_mem_addr = uvm_km_alloc(kernel_map, PAGE_SIZE, 0,
117 	    UVM_KMF_VAONLY|UVM_KMF_WAITVA);
118 	KASSERT(dev_mem_addr != 0);
119 #else
120 	dev_mem_addr = 0;
121 #endif
122 }
123 
124 
125 /*
126  * dev_mem_getva: get a special virtual address.  If architecture requires,
127  * allocate VA according to PA, which avoids cache-aliasing issues.  Use a
128  * constant, general mapping address otherwise.
129  */
130 static inline vaddr_t
131 dev_mem_getva(paddr_t pa)
132 {
133 #ifdef __HAVE_MM_MD_CACHE_ALIASING
134 	return uvm_km_alloc(kernel_map, PAGE_SIZE,
135 	    atop(pa) & uvmexp.colormask,
136 	    UVM_KMF_VAONLY | UVM_KMF_WAITVA | UVM_KMF_COLORMATCH);
137 #else
138 	return dev_mem_addr;
139 #endif
140 }
141 
142 static inline void
143 dev_mem_relva(paddr_t pa, vaddr_t va)
144 {
145 #ifdef __HAVE_MM_MD_CACHE_ALIASING
146 	uvm_km_free(kernel_map, va, PAGE_SIZE, UVM_KMF_VAONLY);
147 #else
148 	KASSERT(dev_mem_addr == va);
149 #endif
150 }
151 
152 /*
153  * dev_kmem_readwrite: helper for DEV_MEM (/dev/mem) case of R/W.
154  */
155 static int
156 dev_mem_readwrite(struct uio *uio, struct iovec *iov)
157 {
158 	paddr_t paddr;
159 	vaddr_t vaddr;
160 	vm_prot_t prot;
161 	size_t len, offset;
162 	bool have_direct;
163 	int error;
164 
165 	/* Check for wrap around. */
166 	if ((intptr_t)uio->uio_offset != uio->uio_offset) {
167 		return EFAULT;
168 	}
169 	paddr = uio->uio_offset & ~PAGE_MASK;
170 	prot = (uio->uio_rw == UIO_WRITE) ? VM_PROT_WRITE : VM_PROT_READ;
171 	error = mm_md_physacc(paddr, prot);
172 	if (error) {
173 		return error;
174 	}
175 	offset = uio->uio_offset & PAGE_MASK;
176 	len = MIN(uio->uio_resid, PAGE_SIZE - offset);
177 
178 #ifdef __HAVE_MM_MD_DIRECT_MAPPED_PHYS
179 	/* Is physical address directly mapped?  Return VA. */
180 	have_direct = mm_md_direct_mapped_phys(paddr, &vaddr);
181 #else
182 	vaddr = 0;
183 	have_direct = false;
184 #endif
185 	if (!have_direct) {
186 		/* Get a special virtual address. */
187 		const vaddr_t va = dev_mem_getva(paddr);
188 
189 		/* Map selected KVA to physical address. */
190 		mutex_enter(&dev_mem_lock);
191 		pmap_kenter_pa(va, paddr, prot, 0);
192 		pmap_update(pmap_kernel());
193 
194 		/* Perform I/O. */
195 		vaddr = va + offset;
196 		error = uiomove((void *)vaddr, len, uio);
197 
198 		/* Unmap, flush before unlock. */
199 		pmap_kremove(va, PAGE_SIZE);
200 		pmap_update(pmap_kernel());
201 		mutex_exit(&dev_mem_lock);
202 
203 		/* "Release" the virtual address. */
204 		dev_mem_relva(paddr, va);
205 	} else {
206 		/* Direct map, just perform I/O. */
207 		vaddr += offset;
208 		error = uiomove((void *)vaddr, len, uio);
209 	}
210 	return error;
211 }
212 
213 /*
214  * dev_kmem_readwrite: helper for DEV_KMEM (/dev/kmem) case of R/W.
215  */
216 static int
217 dev_kmem_readwrite(struct uio *uio, struct iovec *iov)
218 {
219 	void *addr;
220 	size_t len, offset;
221 	vm_prot_t prot;
222 	int error;
223 	bool md_kva;
224 
225 	/* Check for wrap around. */
226 	addr = (void *)(intptr_t)uio->uio_offset;
227 	if ((uintptr_t)addr != uio->uio_offset) {
228 		return EFAULT;
229 	}
230 	/*
231 	 * Handle non-page aligned offset.
232 	 * Otherwise, we operate in page-by-page basis.
233 	 */
234 	offset = uio->uio_offset & PAGE_MASK;
235 	len = MIN(uio->uio_resid, PAGE_SIZE - offset);
236 	prot = (uio->uio_rw == UIO_WRITE) ? VM_PROT_WRITE : VM_PROT_READ;
237 
238 	md_kva = false;
239 
240 #ifdef __HAVE_MM_MD_DIRECT_MAPPED_IO
241 	paddr_t paddr;
242 	/* MD case: is this is a directly mapped address? */
243 	if (mm_md_direct_mapped_io(addr, &paddr)) {
244 		/* If so, validate physical address. */
245 		error = mm_md_physacc(paddr, prot);
246 		if (error) {
247 			return error;
248 		}
249 		md_kva = true;
250 	}
251 #endif
252 	if (!md_kva) {
253 		bool checked = false;
254 
255 #ifdef __HAVE_MM_MD_KERNACC
256 		/* MD check for the address. */
257 		error = mm_md_kernacc(addr, prot, &checked);
258 		if (error) {
259 			return error;
260 		}
261 #endif
262 		/* UVM check for the address (unless MD indicated to not). */
263 		if (!checked && !uvm_kernacc(addr, len, prot)) {
264 			return EFAULT;
265 		}
266 	}
267 	error = uiomove(addr, len, uio);
268 	return error;
269 }
270 
271 /*
272  * dev_zero_readwrite: helper for DEV_ZERO (/dev/null) case of R/W.
273  */
274 static inline int
275 dev_zero_readwrite(struct uio *uio, struct iovec *iov)
276 {
277 	size_t len;
278 
279 	/* Nothing to do for the write case. */
280 	if (uio->uio_rw == UIO_WRITE) {
281 		uio->uio_resid = 0;
282 		return 0;
283 	}
284 	/*
285 	 * Read in page-by-page basis, caller will continue.
286 	 * Cut appropriately for a single/last-iteration cases.
287 	 */
288 	len = MIN(iov->iov_len, PAGE_SIZE);
289 	return uiomove(dev_zero_page, len, uio);
290 }
291 
292 /*
293  * mm_readwrite: general memory R/W function.
294  */
295 static int
296 mm_readwrite(dev_t dev, struct uio *uio, int flags)
297 {
298 	struct iovec *iov;
299 	int error;
300 
301 #ifdef __HAVE_MM_MD_READWRITE
302 	/* If defined - there are extra MD cases. */
303 	switch (minor(dev)) {
304 	case DEV_MEM:
305 	case DEV_KMEM:
306 	case DEV_NULL:
307 	case DEV_ZERO:
308 #if defined(COMPAT_16) && defined(__arm)
309 	case _DEV_ZERO_oARM:
310 #endif
311 		break;
312 	default:
313 		return mm_md_readwrite(dev, uio);
314 	}
315 #endif
316 	error = 0;
317 	while (uio->uio_resid > 0 && error == 0) {
318 		iov = uio->uio_iov;
319 		if (iov->iov_len == 0) {
320 			/* Processed; next I/O vector. */
321 			uio->uio_iov++;
322 			uio->uio_iovcnt--;
323 			KASSERT(uio->uio_iovcnt >= 0);
324 			continue;
325 		}
326 		/* Helper functions will process in page-by-page basis. */
327 		switch (minor(dev)) {
328 		case DEV_MEM:
329 			error = dev_mem_readwrite(uio, iov);
330 			break;
331 		case DEV_KMEM:
332 			error = dev_kmem_readwrite(uio, iov);
333 			break;
334 		case DEV_NULL:
335 			if (uio->uio_rw == UIO_WRITE) {
336 				uio->uio_resid = 0;
337 			}
338 			/* Break directly out of the loop. */
339 			return 0;
340 #if defined(COMPAT_16) && defined(__arm)
341 		case _DEV_ZERO_oARM:
342 #endif
343 		case DEV_ZERO:
344 			error = dev_zero_readwrite(uio, iov);
345 			break;
346 		default:
347 			error = ENXIO;
348 			break;
349 		}
350 	}
351 	return error;
352 }
353 
354 /*
355  * mm_mmap: general mmap() handler.
356  */
357 static paddr_t
358 mm_mmap(dev_t dev, off_t off, int acc)
359 {
360 	vm_prot_t prot;
361 
362 #ifdef __HAVE_MM_MD_MMAP
363 	/* If defined - there are extra mmap() MD cases. */
364 	switch (minor(dev)) {
365 	case DEV_MEM:
366 	case DEV_KMEM:
367 	case DEV_NULL:
368 #if defined(COMPAT_16) && defined(__arm)
369 	case _DEV_ZERO_oARM:
370 #endif
371 	case DEV_ZERO:
372 		break;
373 	default:
374 		return mm_md_mmap(dev, off, acc);
375 	}
376 #endif
377 	/*
378 	 * /dev/null does not make sense, /dev/kmem is volatile and
379 	 * /dev/zero is handled in mmap already.
380 	 */
381 	if (minor(dev) != DEV_MEM) {
382 		return -1;
383 	}
384 
385 	prot = 0;
386 	if (acc & PROT_EXEC)
387 		prot |= VM_PROT_EXECUTE;
388 	if (acc & PROT_READ)
389 		prot |= VM_PROT_READ;
390 	if (acc & PROT_WRITE)
391 		prot |= VM_PROT_WRITE;
392 
393 	/* Validate the physical address. */
394 	if (mm_md_physacc(off, prot) != 0) {
395 		return -1;
396 	}
397 	return off >> PGSHIFT;
398 }
399 
400 static int
401 mm_ioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
402 {
403 
404 	switch (cmd) {
405 	case FIONBIO:
406 		/* We never block anyway. */
407 		return 0;
408 
409 	case FIOSETOWN:
410 	case FIOGETOWN:
411 	case TIOCGPGRP:
412 	case TIOCSPGRP:
413 	case TIOCGETA:
414 		return ENOTTY;
415 
416 	case FIOASYNC:
417 		if ((*(int *)data) == 0) {
418 			return 0;
419 		}
420 		/* FALLTHROUGH */
421 	default:
422 		return EOPNOTSUPP;
423 	}
424 }
425