2 * linux/drivers/char/mem.c
4 * Copyright (C) 1991, 1992 Linus Torvalds
7 * Jan-11-1998, C. Scott Ananian <cananian@alumni.princeton.edu>
8 * Shared /dev/zero mmaping support, Feb 2000, Kanoj Sarcar <kanoj@sgi.com>
11 #include <linux/config.h>
13 #include <linux/miscdevice.h>
14 #include <linux/slab.h>
15 #include <linux/vmalloc.h>
16 #include <linux/mman.h>
17 #include <linux/random.h>
18 #include <linux/init.h>
19 #include <linux/raw.h>
20 #include <linux/tty.h>
21 #include <linux/capability.h>
22 #include <linux/smp_lock.h>
23 #include <linux/devfs_fs_kernel.h>
24 #include <linux/ptrace.h>
25 #include <linux/device.h>
27 #include <asm/uaccess.h>
31 # include <linux/efi.h>
35 extern void fbmem_init(void);
37 #if defined(CONFIG_S390_TAPE) && defined(CONFIG_S390_TAPE_CHAR)
38 extern void tapechar_init(void);
42 * Architectures vary in how they handle caching for addresses
43 * outside of main memory.
46 static inline int uncached_access(struct file *file, unsigned long addr)
50 * On the PPro and successors, the MTRRs are used to set
51 * memory types for physical addresses outside main memory,
52 * so blindly setting PCD or PWT on those pages is wrong.
53 * For Pentiums and earlier, the surround logic should disable
54 * caching for the high addresses through the KEN pin, but
55 * we maintain the tradition of paranoia in this code.
57 if (file->f_flags & O_SYNC)
59 return !( test_bit(X86_FEATURE_MTRR, boot_cpu_data.x86_capability) ||
60 test_bit(X86_FEATURE_K6_MTRR, boot_cpu_data.x86_capability) ||
61 test_bit(X86_FEATURE_CYRIX_ARR, boot_cpu_data.x86_capability) ||
62 test_bit(X86_FEATURE_CENTAUR_MCR, boot_cpu_data.x86_capability) )
63 && addr >= __pa(high_memory);
64 #elif defined(__x86_64__)
66 * This is broken because it can generate memory type aliases,
67 * which can cause cache corruptions
68 * But it is only available for root and we have to be bug-to-bug
69 * compatible with i386.
71 if (file->f_flags & O_SYNC)
73 /* same behaviour as i386. PAT always set to cached and MTRRs control the
75 Hopefully a full PAT implementation will fix that soon. */
77 #elif defined(CONFIG_IA64)
79 * On ia64, we ignore O_SYNC because we cannot tolerate memory attribute aliases.
81 return !(efi_mem_attributes(addr) & EFI_MEMORY_WB);
82 #elif defined(CONFIG_PPC64)
83 /* On PPC64, we always do non-cacheable access to the IO hole and
84 * cacheable elsewhere. Cache paradox can checkstop the CPU and
85 * the high_memory heuristic below is wrong on machines with memory
86 * above the IO hole... Ah, and of course, XFree86 doesn't pass
87 * O_SYNC when mapping us to tap IO space. Surprised ?
89 return !page_is_ram(addr);
92 * Accessing memory above the top the kernel knows about or through a file pointer
93 * that was marked O_SYNC will be done non-cached.
95 if (file->f_flags & O_SYNC)
97 return addr >= __pa(high_memory);
101 #ifndef ARCH_HAS_VALID_PHYS_ADDR_RANGE
102 static inline int valid_phys_addr_range(unsigned long addr, size_t *count)
104 unsigned long end_mem;
106 end_mem = __pa(high_memory);
110 if (*count > end_mem - addr)
111 *count = end_mem - addr;
117 static ssize_t do_write_mem(void *p, unsigned long realp,
118 const char __user * buf, size_t count, loff_t *ppos)
121 unsigned long copied;
124 #if defined(__sparc__) || (defined(__mc68000__) && defined(CONFIG_MMU))
125 /* we don't have page 0 mapped on sparc and m68k.. */
126 if (realp < PAGE_SIZE) {
127 unsigned long sz = PAGE_SIZE-realp;
128 if (sz > count) sz = count;
129 /* Hmm. Do something? */
136 copied = copy_from_user(p, buf, count);
138 ssize_t ret = written + (count - copied);
151 * This funcion reads the *physical* memory. The f_pos points directly to the
154 static ssize_t read_mem(struct file * file, char __user * buf,
155 size_t count, loff_t *ppos)
157 unsigned long p = *ppos;
160 if (!valid_phys_addr_range(p, &count))
163 #if defined(__sparc__) || (defined(__mc68000__) && defined(CONFIG_MMU))
164 /* we don't have page 0 mapped on sparc and m68k.. */
166 unsigned long sz = PAGE_SIZE-p;
170 if (clear_user(buf, sz))
179 if (copy_to_user(buf, __va(p), count))
186 static ssize_t write_mem(struct file * file, const char __user * buf,
187 size_t count, loff_t *ppos)
189 unsigned long p = *ppos;
191 if (!valid_phys_addr_range(p, &count))
193 return do_write_mem(__va(p), p, buf, count, ppos);
196 static int mmap_mem(struct file * file, struct vm_area_struct * vma)
198 unsigned long offset = vma->vm_pgoff << PAGE_SHIFT;
201 uncached = uncached_access(file, offset);
202 #ifdef pgprot_noncached
204 vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
207 /* Don't try to swap out physical pages.. */
208 vma->vm_flags |= VM_RESERVED;
211 * Don't dump addresses that are not real memory to a core file.
214 vma->vm_flags |= VM_IO;
216 if (remap_page_range(vma, vma->vm_start, offset, vma->vm_end-vma->vm_start,
222 extern long vread(char *buf, char *addr, unsigned long count);
223 extern long vwrite(char *buf, char *addr, unsigned long count);
226 * This function reads the *virtual* memory as seen by the kernel.
228 static ssize_t read_kmem(struct file *file, char __user *buf,
229 size_t count, loff_t *ppos)
231 unsigned long p = *ppos;
234 char * kbuf; /* k-addr because vread() takes vmlist_lock rwlock */
236 if (p < (unsigned long) high_memory) {
238 if (count > (unsigned long) high_memory - p)
239 read = (unsigned long) high_memory - p;
241 #if defined(__sparc__) || (defined(__mc68000__) && defined(CONFIG_MMU))
242 /* we don't have page 0 mapped on sparc and m68k.. */
243 if (p < PAGE_SIZE && read > 0) {
244 size_t tmp = PAGE_SIZE - p;
245 if (tmp > read) tmp = read;
246 if (clear_user(buf, tmp))
254 if (copy_to_user(buf, (char *)p, read))
262 kbuf = (char *)__get_free_page(GFP_KERNEL);
270 len = vread(kbuf, (char *)p, len);
273 if (copy_to_user(buf, kbuf, len)) {
274 free_page((unsigned long)kbuf);
282 free_page((unsigned long)kbuf);
289 * This function writes to the *virtual* memory as seen by the kernel.
291 static ssize_t write_kmem(struct file * file, const char __user * buf,
292 size_t count, loff_t *ppos)
294 unsigned long p = *ppos;
298 char * kbuf; /* k-addr because vwrite() takes vmlist_lock rwlock */
300 if (p < (unsigned long) high_memory) {
303 if (count > (unsigned long) high_memory - p)
304 wrote = (unsigned long) high_memory - p;
306 written = do_write_mem((void*)p, p, buf, wrote, ppos);
307 if (written != wrote)
316 kbuf = (char *)__get_free_page(GFP_KERNEL);
318 return wrote ? wrote : -ENOMEM;
325 written = copy_from_user(kbuf, buf, len);
329 free_page((unsigned long)kbuf);
330 ret = wrote + virtr + (len - written);
331 return ret ? ret : -EFAULT;
334 len = vwrite(kbuf, (char *)p, len);
340 free_page((unsigned long)kbuf);
344 return virtr + wrote;
347 #if defined(CONFIG_ISA) || !defined(__mc68000__)
348 static ssize_t read_port(struct file * file, char __user * buf,
349 size_t count, loff_t *ppos)
351 unsigned long i = *ppos;
352 char __user *tmp = buf;
354 if (verify_area(VERIFY_WRITE,buf,count))
356 while (count-- > 0 && i < 65536) {
357 if (__put_user(inb(i),tmp) < 0)
366 static ssize_t write_port(struct file * file, const char __user * buf,
367 size_t count, loff_t *ppos)
369 unsigned long i = *ppos;
370 const char __user * tmp = buf;
372 if (verify_area(VERIFY_READ,buf,count))
374 while (count-- > 0 && i < 65536) {
376 if (__get_user(c, tmp))
387 static ssize_t read_null(struct file * file, char __user * buf,
388 size_t count, loff_t *ppos)
393 static ssize_t write_null(struct file * file, const char __user * buf,
394 size_t count, loff_t *ppos)
401 * For fun, we are using the MMU for this.
403 static inline size_t read_zero_pagealigned(char __user * buf, size_t size)
405 struct mm_struct *mm;
406 struct vm_area_struct * vma;
407 unsigned long addr=(unsigned long)buf;
410 /* Oops, this was forgotten before. -ben */
411 down_read(&mm->mmap_sem);
413 /* For private mappings, just map in zero pages. */
414 for (vma = find_vma(mm, addr); vma; vma = vma->vm_next) {
417 if (vma->vm_start > addr || (vma->vm_flags & VM_WRITE) == 0)
419 if (vma->vm_flags & VM_SHARED)
421 count = vma->vm_end - addr;
425 zap_page_range(vma, addr, count, NULL);
426 zeromap_page_range(vma, addr, count, PAGE_COPY);
435 up_read(&mm->mmap_sem);
437 /* The shared case is hard. Let's do the conventional zeroing. */
439 unsigned long unwritten = clear_user(buf, PAGE_SIZE);
441 return size + unwritten - PAGE_SIZE;
449 up_read(&mm->mmap_sem);
453 static ssize_t read_zero(struct file * file, char __user * buf,
454 size_t count, loff_t *ppos)
456 unsigned long left, unwritten, written = 0;
461 if (!access_ok(VERIFY_WRITE, buf, count))
466 /* do we want to be clever? Arbitrary cut-off */
467 if (count >= PAGE_SIZE*4) {
468 unsigned long partial;
470 /* How much left of the page? */
471 partial = (PAGE_SIZE-1) & -(unsigned long) buf;
472 unwritten = clear_user(buf, partial);
473 written = partial - unwritten;
478 unwritten = read_zero_pagealigned(buf, left & PAGE_MASK);
479 written += (left & PAGE_MASK) - unwritten;
482 buf += left & PAGE_MASK;
485 unwritten = clear_user(buf, left);
486 written += left - unwritten;
488 return written ? written : -EFAULT;
491 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
493 if (vma->vm_flags & VM_SHARED)
494 return shmem_zero_setup(vma);
495 if (zeromap_page_range(vma, vma->vm_start, vma->vm_end - vma->vm_start, vma->vm_page_prot))
499 #else /* CONFIG_MMU */
500 static ssize_t read_zero(struct file * file, char * buf,
501 size_t count, loff_t *ppos)
509 chunk = 4096; /* Just for latency reasons */
510 if (clear_user(buf, chunk))
519 static int mmap_zero(struct file * file, struct vm_area_struct * vma)
523 #endif /* CONFIG_MMU */
525 static ssize_t write_full(struct file * file, const char __user * buf,
526 size_t count, loff_t *ppos)
532 * Special lseek() function for /dev/null and /dev/zero. Most notably, you
533 * can fopen() both devices with "a" now. This was previously impossible.
537 static loff_t null_lseek(struct file * file, loff_t offset, int orig)
539 return file->f_pos = 0;
543 * The memory devices use the full 32/64 bits of the offset, and so we cannot
544 * check against negative addresses: they are ok. The return value is weird,
545 * though, in that case (0).
547 * also note that seeking relative to the "end of file" isn't supported:
548 * it has no meaning, so it returns -EINVAL.
550 static loff_t memory_lseek(struct file * file, loff_t offset, int orig)
554 down(&file->f_dentry->d_inode->i_sem);
557 file->f_pos = offset;
559 force_successful_syscall_return();
562 file->f_pos += offset;
564 force_successful_syscall_return();
569 up(&file->f_dentry->d_inode->i_sem);
573 static int open_port(struct inode * inode, struct file * filp)
575 return capable(CAP_SYS_RAWIO) ? 0 : -EPERM;
578 #define mmap_kmem mmap_mem
579 #define zero_lseek null_lseek
580 #define full_lseek null_lseek
581 #define write_zero write_null
582 #define read_full read_zero
583 #define open_mem open_port
584 #define open_kmem open_mem
586 static struct file_operations mem_fops = {
587 .llseek = memory_lseek,
594 static struct file_operations kmem_fops = {
595 .llseek = memory_lseek,
602 static struct file_operations null_fops = {
603 .llseek = null_lseek,
608 #if defined(CONFIG_ISA) || !defined(__mc68000__)
609 static struct file_operations port_fops = {
610 .llseek = memory_lseek,
617 static struct file_operations zero_fops = {
618 .llseek = zero_lseek,
624 static struct file_operations full_fops = {
625 .llseek = full_lseek,
630 static ssize_t kmsg_write(struct file * file, const char __user * buf,
631 size_t count, loff_t *ppos)
636 tmp = kmalloc(count + 1, GFP_KERNEL);
640 if (!copy_from_user(tmp, buf, count)) {
642 ret = printk("%s", tmp);
648 static struct file_operations kmsg_fops = {
652 static int memory_open(struct inode * inode, struct file * filp)
654 switch (iminor(inode)) {
656 filp->f_op = &mem_fops;
659 filp->f_op = &kmem_fops;
662 filp->f_op = &null_fops;
664 #if defined(CONFIG_ISA) || !defined(__mc68000__)
666 filp->f_op = &port_fops;
670 filp->f_op = &zero_fops;
673 filp->f_op = &full_fops;
676 filp->f_op = &random_fops;
679 filp->f_op = &urandom_fops;
682 filp->f_op = &kmsg_fops;
687 if (filp->f_op && filp->f_op->open)
688 return filp->f_op->open(inode,filp);
692 static struct file_operations memory_fops = {
693 .open = memory_open, /* just a selector for the real open */
696 static const struct {
700 struct file_operations *fops;
701 } devlist[] = { /* list of minor devices */
702 {1, "mem", S_IRUSR | S_IWUSR | S_IRGRP, &mem_fops},
703 {2, "kmem", S_IRUSR | S_IWUSR | S_IRGRP, &kmem_fops},
704 {3, "null", S_IRUGO | S_IWUGO, &null_fops},
705 #if defined(CONFIG_ISA) || !defined(__mc68000__)
706 {4, "port", S_IRUSR | S_IWUSR | S_IRGRP, &port_fops},
708 {5, "zero", S_IRUGO | S_IWUGO, &zero_fops},
709 {7, "full", S_IRUGO | S_IWUGO, &full_fops},
710 {8, "random", S_IRUGO | S_IWUSR, &random_fops},
711 {9, "urandom", S_IRUGO | S_IWUSR, &urandom_fops},
712 {11,"kmsg", S_IRUGO | S_IWUSR, &kmsg_fops},
715 static struct class_simple *mem_class;
717 static int __init chr_dev_init(void)
721 if (register_chrdev(MEM_MAJOR,"mem",&memory_fops))
722 printk("unable to get major %d for memory devs\n", MEM_MAJOR);
724 mem_class = class_simple_create(THIS_MODULE, "mem");
725 for (i = 0; i < ARRAY_SIZE(devlist); i++) {
726 class_simple_device_add(mem_class,
727 MKDEV(MEM_MAJOR, devlist[i].minor),
728 NULL, devlist[i].name);
729 devfs_mk_cdev(MKDEV(MEM_MAJOR, devlist[i].minor),
730 S_IFCHR | devlist[i].mode, devlist[i].name);
733 #if defined (CONFIG_FB)
739 fs_initcall(chr_dev_init);