m68k: Add NPTL support
[safe/jmp/linux-2.6] / arch / m68k / kernel / sys_m68k.c
1 /*
2  * linux/arch/m68k/kernel/sys_m68k.c
3  *
4  * This file contains various random system calls that
5  * have a non-standard calling sequence on the Linux/m68k
6  * platform.
7  */
8
9 #include <linux/capability.h>
10 #include <linux/errno.h>
11 #include <linux/sched.h>
12 #include <linux/mm.h>
13 #include <linux/fs.h>
14 #include <linux/smp.h>
15 #include <linux/smp_lock.h>
16 #include <linux/sem.h>
17 #include <linux/msg.h>
18 #include <linux/shm.h>
19 #include <linux/stat.h>
20 #include <linux/syscalls.h>
21 #include <linux/mman.h>
22 #include <linux/file.h>
23 #include <linux/ipc.h>
24
25 #include <asm/setup.h>
26 #include <asm/uaccess.h>
27 #include <asm/cachectl.h>
28 #include <asm/traps.h>
29 #include <asm/page.h>
30 #include <asm/unistd.h>
31 #include <linux/elf.h>
32 #include <asm/tlb.h>
33
34 asmlinkage int do_page_fault(struct pt_regs *regs, unsigned long address,
35                              unsigned long error_code);
36
37 asmlinkage long sys_mmap2(unsigned long addr, unsigned long len,
38         unsigned long prot, unsigned long flags,
39         unsigned long fd, unsigned long pgoff)
40 {
41         /*
42          * This is wrong for sun3 - there PAGE_SIZE is 8Kb,
43          * so we need to shift the argument down by 1; m68k mmap64(3)
44          * (in libc) expects the last argument of mmap2 in 4Kb units.
45          */
46         return sys_mmap_pgoff(addr, len, prot, flags, fd, pgoff);
47 }
48
49 /*
50  * Perform the select(nd, in, out, ex, tv) and mmap() system
51  * calls. Linux/m68k cloned Linux/i386, which didn't use to be able to
52  * handle more than 4 system call parameters, so these system calls
53  * used a memory block for parameter passing..
54  */
55
56 struct mmap_arg_struct {
57         unsigned long addr;
58         unsigned long len;
59         unsigned long prot;
60         unsigned long flags;
61         unsigned long fd;
62         unsigned long offset;
63 };
64
65 asmlinkage int old_mmap(struct mmap_arg_struct __user *arg)
66 {
67         struct mmap_arg_struct a;
68         int error = -EFAULT;
69
70         if (copy_from_user(&a, arg, sizeof(a)))
71                 goto out;
72
73         error = -EINVAL;
74         if (a.offset & ~PAGE_MASK)
75                 goto out;
76
77         error = sys_mmap_pgoff(a.addr, a.len, a.prot, a.flags, a.fd,
78                                a.offset >> PAGE_SHIFT);
79 out:
80         return error;
81 }
82
83 struct sel_arg_struct {
84         unsigned long n;
85         fd_set __user *inp, *outp, *exp;
86         struct timeval __user *tvp;
87 };
88
89 asmlinkage int old_select(struct sel_arg_struct __user *arg)
90 {
91         struct sel_arg_struct a;
92
93         if (copy_from_user(&a, arg, sizeof(a)))
94                 return -EFAULT;
95         /* sys_select() does the appropriate kernel locking */
96         return sys_select(a.n, a.inp, a.outp, a.exp, a.tvp);
97 }
98
99 /*
100  * sys_ipc() is the de-multiplexer for the SysV IPC calls..
101  *
102  * This is really horribly ugly.
103  */
104 asmlinkage int sys_ipc (uint call, int first, int second,
105                         int third, void __user *ptr, long fifth)
106 {
107         int version, ret;
108
109         version = call >> 16; /* hack for backward compatibility */
110         call &= 0xffff;
111
112         if (call <= SEMCTL)
113                 switch (call) {
114                 case SEMOP:
115                         return sys_semop (first, ptr, second);
116                 case SEMGET:
117                         return sys_semget (first, second, third);
118                 case SEMCTL: {
119                         union semun fourth;
120                         if (!ptr)
121                                 return -EINVAL;
122                         if (get_user(fourth.__pad, (void __user *__user *) ptr))
123                                 return -EFAULT;
124                         return sys_semctl (first, second, third, fourth);
125                         }
126                 default:
127                         return -ENOSYS;
128                 }
129         if (call <= MSGCTL)
130                 switch (call) {
131                 case MSGSND:
132                         return sys_msgsnd (first, ptr, second, third);
133                 case MSGRCV:
134                         switch (version) {
135                         case 0: {
136                                 struct ipc_kludge tmp;
137                                 if (!ptr)
138                                         return -EINVAL;
139                                 if (copy_from_user (&tmp, ptr, sizeof (tmp)))
140                                         return -EFAULT;
141                                 return sys_msgrcv (first, tmp.msgp, second,
142                                                    tmp.msgtyp, third);
143                                 }
144                         default:
145                                 return sys_msgrcv (first, ptr,
146                                                    second, fifth, third);
147                         }
148                 case MSGGET:
149                         return sys_msgget ((key_t) first, second);
150                 case MSGCTL:
151                         return sys_msgctl (first, second, ptr);
152                 default:
153                         return -ENOSYS;
154                 }
155         if (call <= SHMCTL)
156                 switch (call) {
157                 case SHMAT:
158                         switch (version) {
159                         default: {
160                                 ulong raddr;
161                                 ret = do_shmat (first, ptr, second, &raddr);
162                                 if (ret)
163                                         return ret;
164                                 return put_user (raddr, (ulong __user *) third);
165                         }
166                         }
167                 case SHMDT:
168                         return sys_shmdt (ptr);
169                 case SHMGET:
170                         return sys_shmget (first, second, third);
171                 case SHMCTL:
172                         return sys_shmctl (first, second, ptr);
173                 default:
174                         return -ENOSYS;
175                 }
176
177         return -EINVAL;
178 }
179
180 /* Convert virtual (user) address VADDR to physical address PADDR */
181 #define virt_to_phys_040(vaddr)                                         \
182 ({                                                                      \
183   unsigned long _mmusr, _paddr;                                         \
184                                                                         \
185   __asm__ __volatile__ (".chip 68040\n\t"                               \
186                         "ptestr (%1)\n\t"                               \
187                         "movec %%mmusr,%0\n\t"                          \
188                         ".chip 68k"                                     \
189                         : "=r" (_mmusr)                                 \
190                         : "a" (vaddr));                                 \
191   _paddr = (_mmusr & MMU_R_040) ? (_mmusr & PAGE_MASK) : 0;             \
192   _paddr;                                                               \
193 })
194
195 static inline int
196 cache_flush_040 (unsigned long addr, int scope, int cache, unsigned long len)
197 {
198   unsigned long paddr, i;
199
200   switch (scope)
201     {
202     case FLUSH_SCOPE_ALL:
203       switch (cache)
204         {
205         case FLUSH_CACHE_DATA:
206           /* This nop is needed for some broken versions of the 68040.  */
207           __asm__ __volatile__ ("nop\n\t"
208                                 ".chip 68040\n\t"
209                                 "cpusha %dc\n\t"
210                                 ".chip 68k");
211           break;
212         case FLUSH_CACHE_INSN:
213           __asm__ __volatile__ ("nop\n\t"
214                                 ".chip 68040\n\t"
215                                 "cpusha %ic\n\t"
216                                 ".chip 68k");
217           break;
218         default:
219         case FLUSH_CACHE_BOTH:
220           __asm__ __volatile__ ("nop\n\t"
221                                 ".chip 68040\n\t"
222                                 "cpusha %bc\n\t"
223                                 ".chip 68k");
224           break;
225         }
226       break;
227
228     case FLUSH_SCOPE_LINE:
229       /* Find the physical address of the first mapped page in the
230          address range.  */
231       if ((paddr = virt_to_phys_040(addr))) {
232         paddr += addr & ~(PAGE_MASK | 15);
233         len = (len + (addr & 15) + 15) >> 4;
234       } else {
235         unsigned long tmp = PAGE_SIZE - (addr & ~PAGE_MASK);
236
237         if (len <= tmp)
238           return 0;
239         addr += tmp;
240         len -= tmp;
241         tmp = PAGE_SIZE;
242         for (;;)
243           {
244             if ((paddr = virt_to_phys_040(addr)))
245               break;
246             if (len <= tmp)
247               return 0;
248             addr += tmp;
249             len -= tmp;
250           }
251         len = (len + 15) >> 4;
252       }
253       i = (PAGE_SIZE - (paddr & ~PAGE_MASK)) >> 4;
254       while (len--)
255         {
256           switch (cache)
257             {
258             case FLUSH_CACHE_DATA:
259               __asm__ __volatile__ ("nop\n\t"
260                                     ".chip 68040\n\t"
261                                     "cpushl %%dc,(%0)\n\t"
262                                     ".chip 68k"
263                                     : : "a" (paddr));
264               break;
265             case FLUSH_CACHE_INSN:
266               __asm__ __volatile__ ("nop\n\t"
267                                     ".chip 68040\n\t"
268                                     "cpushl %%ic,(%0)\n\t"
269                                     ".chip 68k"
270                                     : : "a" (paddr));
271               break;
272             default:
273             case FLUSH_CACHE_BOTH:
274               __asm__ __volatile__ ("nop\n\t"
275                                     ".chip 68040\n\t"
276                                     "cpushl %%bc,(%0)\n\t"
277                                     ".chip 68k"
278                                     : : "a" (paddr));
279               break;
280             }
281           if (!--i && len)
282             {
283               /*
284                * No need to page align here since it is done by
285                * virt_to_phys_040().
286                */
287               addr += PAGE_SIZE;
288               i = PAGE_SIZE / 16;
289               /* Recompute physical address when crossing a page
290                  boundary. */
291               for (;;)
292                 {
293                   if ((paddr = virt_to_phys_040(addr)))
294                     break;
295                   if (len <= i)
296                     return 0;
297                   len -= i;
298                   addr += PAGE_SIZE;
299                 }
300             }
301           else
302             paddr += 16;
303         }
304       break;
305
306     default:
307     case FLUSH_SCOPE_PAGE:
308       len += (addr & ~PAGE_MASK) + (PAGE_SIZE - 1);
309       for (len >>= PAGE_SHIFT; len--; addr += PAGE_SIZE)
310         {
311           if (!(paddr = virt_to_phys_040(addr)))
312             continue;
313           switch (cache)
314             {
315             case FLUSH_CACHE_DATA:
316               __asm__ __volatile__ ("nop\n\t"
317                                     ".chip 68040\n\t"
318                                     "cpushp %%dc,(%0)\n\t"
319                                     ".chip 68k"
320                                     : : "a" (paddr));
321               break;
322             case FLUSH_CACHE_INSN:
323               __asm__ __volatile__ ("nop\n\t"
324                                     ".chip 68040\n\t"
325                                     "cpushp %%ic,(%0)\n\t"
326                                     ".chip 68k"
327                                     : : "a" (paddr));
328               break;
329             default:
330             case FLUSH_CACHE_BOTH:
331               __asm__ __volatile__ ("nop\n\t"
332                                     ".chip 68040\n\t"
333                                     "cpushp %%bc,(%0)\n\t"
334                                     ".chip 68k"
335                                     : : "a" (paddr));
336               break;
337             }
338         }
339       break;
340     }
341   return 0;
342 }
343
344 #define virt_to_phys_060(vaddr)                         \
345 ({                                                      \
346   unsigned long paddr;                                  \
347   __asm__ __volatile__ (".chip 68060\n\t"               \
348                         "plpar (%0)\n\t"                \
349                         ".chip 68k"                     \
350                         : "=a" (paddr)                  \
351                         : "0" (vaddr));                 \
352   (paddr); /* XXX */                                    \
353 })
354
355 static inline int
356 cache_flush_060 (unsigned long addr, int scope, int cache, unsigned long len)
357 {
358   unsigned long paddr, i;
359
360   /*
361    * 68060 manual says:
362    *  cpush %dc : flush DC, remains valid (with our %cacr setup)
363    *  cpush %ic : invalidate IC
364    *  cpush %bc : flush DC + invalidate IC
365    */
366   switch (scope)
367     {
368     case FLUSH_SCOPE_ALL:
369       switch (cache)
370         {
371         case FLUSH_CACHE_DATA:
372           __asm__ __volatile__ (".chip 68060\n\t"
373                                 "cpusha %dc\n\t"
374                                 ".chip 68k");
375           break;
376         case FLUSH_CACHE_INSN:
377           __asm__ __volatile__ (".chip 68060\n\t"
378                                 "cpusha %ic\n\t"
379                                 ".chip 68k");
380           break;
381         default:
382         case FLUSH_CACHE_BOTH:
383           __asm__ __volatile__ (".chip 68060\n\t"
384                                 "cpusha %bc\n\t"
385                                 ".chip 68k");
386           break;
387         }
388       break;
389
390     case FLUSH_SCOPE_LINE:
391       /* Find the physical address of the first mapped page in the
392          address range.  */
393       len += addr & 15;
394       addr &= -16;
395       if (!(paddr = virt_to_phys_060(addr))) {
396         unsigned long tmp = PAGE_SIZE - (addr & ~PAGE_MASK);
397
398         if (len <= tmp)
399           return 0;
400         addr += tmp;
401         len -= tmp;
402         tmp = PAGE_SIZE;
403         for (;;)
404           {
405             if ((paddr = virt_to_phys_060(addr)))
406               break;
407             if (len <= tmp)
408               return 0;
409             addr += tmp;
410             len -= tmp;
411           }
412       }
413       len = (len + 15) >> 4;
414       i = (PAGE_SIZE - (paddr & ~PAGE_MASK)) >> 4;
415       while (len--)
416         {
417           switch (cache)
418             {
419             case FLUSH_CACHE_DATA:
420               __asm__ __volatile__ (".chip 68060\n\t"
421                                     "cpushl %%dc,(%0)\n\t"
422                                     ".chip 68k"
423                                     : : "a" (paddr));
424               break;
425             case FLUSH_CACHE_INSN:
426               __asm__ __volatile__ (".chip 68060\n\t"
427                                     "cpushl %%ic,(%0)\n\t"
428                                     ".chip 68k"
429                                     : : "a" (paddr));
430               break;
431             default:
432             case FLUSH_CACHE_BOTH:
433               __asm__ __volatile__ (".chip 68060\n\t"
434                                     "cpushl %%bc,(%0)\n\t"
435                                     ".chip 68k"
436                                     : : "a" (paddr));
437               break;
438             }
439           if (!--i && len)
440             {
441
442               /*
443                * We just want to jump to the first cache line
444                * in the next page.
445                */
446               addr += PAGE_SIZE;
447               addr &= PAGE_MASK;
448
449               i = PAGE_SIZE / 16;
450               /* Recompute physical address when crossing a page
451                  boundary. */
452               for (;;)
453                 {
454                   if ((paddr = virt_to_phys_060(addr)))
455                     break;
456                   if (len <= i)
457                     return 0;
458                   len -= i;
459                   addr += PAGE_SIZE;
460                 }
461             }
462           else
463             paddr += 16;
464         }
465       break;
466
467     default:
468     case FLUSH_SCOPE_PAGE:
469       len += (addr & ~PAGE_MASK) + (PAGE_SIZE - 1);
470       addr &= PAGE_MASK;        /* Workaround for bug in some
471                                    revisions of the 68060 */
472       for (len >>= PAGE_SHIFT; len--; addr += PAGE_SIZE)
473         {
474           if (!(paddr = virt_to_phys_060(addr)))
475             continue;
476           switch (cache)
477             {
478             case FLUSH_CACHE_DATA:
479               __asm__ __volatile__ (".chip 68060\n\t"
480                                     "cpushp %%dc,(%0)\n\t"
481                                     ".chip 68k"
482                                     : : "a" (paddr));
483               break;
484             case FLUSH_CACHE_INSN:
485               __asm__ __volatile__ (".chip 68060\n\t"
486                                     "cpushp %%ic,(%0)\n\t"
487                                     ".chip 68k"
488                                     : : "a" (paddr));
489               break;
490             default:
491             case FLUSH_CACHE_BOTH:
492               __asm__ __volatile__ (".chip 68060\n\t"
493                                     "cpushp %%bc,(%0)\n\t"
494                                     ".chip 68k"
495                                     : : "a" (paddr));
496               break;
497             }
498         }
499       break;
500     }
501   return 0;
502 }
503
504 /* sys_cacheflush -- flush (part of) the processor cache.  */
505 asmlinkage int
506 sys_cacheflush (unsigned long addr, int scope, int cache, unsigned long len)
507 {
508         struct vm_area_struct *vma;
509         int ret = -EINVAL;
510
511         lock_kernel();
512         if (scope < FLUSH_SCOPE_LINE || scope > FLUSH_SCOPE_ALL ||
513             cache & ~FLUSH_CACHE_BOTH)
514                 goto out;
515
516         if (scope == FLUSH_SCOPE_ALL) {
517                 /* Only the superuser may explicitly flush the whole cache. */
518                 ret = -EPERM;
519                 if (!capable(CAP_SYS_ADMIN))
520                         goto out;
521         } else {
522                 /*
523                  * Verify that the specified address region actually belongs
524                  * to this process.
525                  */
526                 vma = find_vma (current->mm, addr);
527                 ret = -EINVAL;
528                 /* Check for overflow.  */
529                 if (addr + len < addr)
530                         goto out;
531                 if (vma == NULL || addr < vma->vm_start || addr + len > vma->vm_end)
532                         goto out;
533         }
534
535         if (CPU_IS_020_OR_030) {
536                 if (scope == FLUSH_SCOPE_LINE && len < 256) {
537                         unsigned long cacr;
538                         __asm__ ("movec %%cacr, %0" : "=r" (cacr));
539                         if (cache & FLUSH_CACHE_INSN)
540                                 cacr |= 4;
541                         if (cache & FLUSH_CACHE_DATA)
542                                 cacr |= 0x400;
543                         len >>= 2;
544                         while (len--) {
545                                 __asm__ __volatile__ ("movec %1, %%caar\n\t"
546                                                       "movec %0, %%cacr"
547                                                       : /* no outputs */
548                                                       : "r" (cacr), "r" (addr));
549                                 addr += 4;
550                         }
551                 } else {
552                         /* Flush the whole cache, even if page granularity requested. */
553                         unsigned long cacr;
554                         __asm__ ("movec %%cacr, %0" : "=r" (cacr));
555                         if (cache & FLUSH_CACHE_INSN)
556                                 cacr |= 8;
557                         if (cache & FLUSH_CACHE_DATA)
558                                 cacr |= 0x800;
559                         __asm__ __volatile__ ("movec %0, %%cacr" : : "r" (cacr));
560                 }
561                 ret = 0;
562                 goto out;
563         } else {
564             /*
565              * 040 or 060: don't blindly trust 'scope', someone could
566              * try to flush a few megs of memory.
567              */
568
569             if (len>=3*PAGE_SIZE && scope<FLUSH_SCOPE_PAGE)
570                 scope=FLUSH_SCOPE_PAGE;
571             if (len>=10*PAGE_SIZE && scope<FLUSH_SCOPE_ALL)
572                 scope=FLUSH_SCOPE_ALL;
573             if (CPU_IS_040) {
574                 ret = cache_flush_040 (addr, scope, cache, len);
575             } else if (CPU_IS_060) {
576                 ret = cache_flush_060 (addr, scope, cache, len);
577             }
578         }
579 out:
580         unlock_kernel();
581         return ret;
582 }
583
584 asmlinkage int sys_getpagesize(void)
585 {
586         return PAGE_SIZE;
587 }
588
589 /*
590  * Do a system call from kernel instead of calling sys_execve so we
591  * end up with proper pt_regs.
592  */
593 int kernel_execve(const char *filename, char *const argv[], char *const envp[])
594 {
595         register long __res asm ("%d0") = __NR_execve;
596         register long __a asm ("%d1") = (long)(filename);
597         register long __b asm ("%d2") = (long)(argv);
598         register long __c asm ("%d3") = (long)(envp);
599         asm volatile ("trap  #0" : "+d" (__res)
600                         : "d" (__a), "d" (__b), "d" (__c));
601         return __res;
602 }
603
604 asmlinkage unsigned long sys_get_thread_area(void)
605 {
606         return current_thread_info()->tp_value;
607 }
608
609 asmlinkage int sys_set_thread_area(unsigned long tp)
610 {
611         current_thread_info()->tp_value = tp;
612         return 0;
613 }
614
615 /* This syscall gets its arguments in A0 (mem), D2 (oldval) and
616    D1 (newval).  */
617 asmlinkage int
618 sys_atomic_cmpxchg_32(unsigned long newval, int oldval, int d3, int d4, int d5,
619                       unsigned long __user * mem)
620 {
621         /* This was borrowed from ARM's implementation.  */
622         for (;;) {
623                 struct mm_struct *mm = current->mm;
624                 pgd_t *pgd;
625                 pmd_t *pmd;
626                 pte_t *pte;
627                 spinlock_t *ptl;
628                 unsigned long mem_value;
629
630                 down_read(&mm->mmap_sem);
631                 pgd = pgd_offset(mm, (unsigned long)mem);
632                 if (!pgd_present(*pgd))
633                         goto bad_access;
634                 pmd = pmd_offset(pgd, (unsigned long)mem);
635                 if (!pmd_present(*pmd))
636                         goto bad_access;
637                 pte = pte_offset_map_lock(mm, pmd, (unsigned long)mem, &ptl);
638                 if (!pte_present(*pte) || !pte_dirty(*pte)
639                     || !pte_write(*pte)) {
640                         pte_unmap_unlock(pte, ptl);
641                         goto bad_access;
642                 }
643
644                 mem_value = *mem;
645                 if (mem_value == oldval)
646                         *mem = newval;
647
648                 pte_unmap_unlock(pte, ptl);
649                 up_read(&mm->mmap_sem);
650                 return mem_value;
651
652               bad_access:
653                 up_read(&mm->mmap_sem);
654                 /* This is not necessarily a bad access, we can get here if
655                    a memory we're trying to write to should be copied-on-write.
656                    Make the kernel do the necessary page stuff, then re-iterate.
657                    Simulate a write access fault to do that.  */
658                 {
659                         /* The first argument of the function corresponds to
660                            D1, which is the first field of struct pt_regs.  */
661                         struct pt_regs *fp = (struct pt_regs *)&newval;
662
663                         /* '3' is an RMW flag.  */
664                         if (do_page_fault(fp, (unsigned long)mem, 3))
665                                 /* If the do_page_fault() failed, we don't
666                                    have anything meaningful to return.
667                                    There should be a SIGSEGV pending for
668                                    the process.  */
669                                 return 0xdeadbeef;
670                 }
671         }
672 }
673
674 asmlinkage int sys_atomic_barrier(void)
675 {
676         /* no code needed for uniprocs */
677         return 0;
678 }