Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
[pandora-kernel.git] / arch / m68k / kernel / sys_m68k.c
index 63013df..8623f8d 100644 (file)
@@ -1,5 +1,580 @@
+/*
+ * linux/arch/m68k/kernel/sys_m68k.c
+ *
+ * This file contains various random system calls that
+ * have a non-standard calling sequence on the Linux/m68k
+ * platform.
+ */
+
+#include <linux/capability.h>
+#include <linux/errno.h>
+#include <linux/sched.h>
+#include <linux/mm.h>
+#include <linux/fs.h>
+#include <linux/smp.h>
+#include <linux/sem.h>
+#include <linux/msg.h>
+#include <linux/shm.h>
+#include <linux/stat.h>
+#include <linux/syscalls.h>
+#include <linux/mman.h>
+#include <linux/file.h>
+#include <linux/ipc.h>
+
+#include <asm/setup.h>
+#include <asm/uaccess.h>
+#include <asm/cachectl.h>
+#include <asm/traps.h>
+#include <asm/page.h>
+#include <asm/unistd.h>
+#include <asm/cacheflush.h>
+
 #ifdef CONFIG_MMU
-#include "sys_m68k_mm.c"
+
+#include <asm/tlb.h>
+
+asmlinkage int do_page_fault(struct pt_regs *regs, unsigned long address,
+                            unsigned long error_code);
+
+asmlinkage long sys_mmap2(unsigned long addr, unsigned long len,
+       unsigned long prot, unsigned long flags,
+       unsigned long fd, unsigned long pgoff)
+{
+       /*
+        * This is wrong for sun3 - there PAGE_SIZE is 8Kb,
+        * so we need to shift the argument down by 1; m68k mmap64(3)
+        * (in libc) expects the last argument of mmap2 in 4Kb units.
+        */
+       return sys_mmap_pgoff(addr, len, prot, flags, fd, pgoff);
+}
+
+/* Convert virtual (user) address VADDR to physical address PADDR */
+#define virt_to_phys_040(vaddr)                                                \
+({                                                                     \
+  unsigned long _mmusr, _paddr;                                                \
+                                                                       \
+  __asm__ __volatile__ (".chip 68040\n\t"                              \
+                       "ptestr (%1)\n\t"                               \
+                       "movec %%mmusr,%0\n\t"                          \
+                       ".chip 68k"                                     \
+                       : "=r" (_mmusr)                                 \
+                       : "a" (vaddr));                                 \
+  _paddr = (_mmusr & MMU_R_040) ? (_mmusr & PAGE_MASK) : 0;            \
+  _paddr;                                                              \
+})
+
+static inline int
+cache_flush_040 (unsigned long addr, int scope, int cache, unsigned long len)
+{
+  unsigned long paddr, i;
+
+  switch (scope)
+    {
+    case FLUSH_SCOPE_ALL:
+      switch (cache)
+       {
+       case FLUSH_CACHE_DATA:
+         /* This nop is needed for some broken versions of the 68040.  */
+         __asm__ __volatile__ ("nop\n\t"
+                               ".chip 68040\n\t"
+                               "cpusha %dc\n\t"
+                               ".chip 68k");
+         break;
+       case FLUSH_CACHE_INSN:
+         __asm__ __volatile__ ("nop\n\t"
+                               ".chip 68040\n\t"
+                               "cpusha %ic\n\t"
+                               ".chip 68k");
+         break;
+       default:
+       case FLUSH_CACHE_BOTH:
+         __asm__ __volatile__ ("nop\n\t"
+                               ".chip 68040\n\t"
+                               "cpusha %bc\n\t"
+                               ".chip 68k");
+         break;
+       }
+      break;
+
+    case FLUSH_SCOPE_LINE:
+      /* Find the physical address of the first mapped page in the
+        address range.  */
+      if ((paddr = virt_to_phys_040(addr))) {
+        paddr += addr & ~(PAGE_MASK | 15);
+        len = (len + (addr & 15) + 15) >> 4;
+      } else {
+       unsigned long tmp = PAGE_SIZE - (addr & ~PAGE_MASK);
+
+       if (len <= tmp)
+         return 0;
+       addr += tmp;
+       len -= tmp;
+       tmp = PAGE_SIZE;
+       for (;;)
+         {
+           if ((paddr = virt_to_phys_040(addr)))
+             break;
+           if (len <= tmp)
+             return 0;
+           addr += tmp;
+           len -= tmp;
+         }
+       len = (len + 15) >> 4;
+      }
+      i = (PAGE_SIZE - (paddr & ~PAGE_MASK)) >> 4;
+      while (len--)
+       {
+         switch (cache)
+           {
+           case FLUSH_CACHE_DATA:
+             __asm__ __volatile__ ("nop\n\t"
+                                   ".chip 68040\n\t"
+                                   "cpushl %%dc,(%0)\n\t"
+                                   ".chip 68k"
+                                   : : "a" (paddr));
+             break;
+           case FLUSH_CACHE_INSN:
+             __asm__ __volatile__ ("nop\n\t"
+                                   ".chip 68040\n\t"
+                                   "cpushl %%ic,(%0)\n\t"
+                                   ".chip 68k"
+                                   : : "a" (paddr));
+             break;
+           default:
+           case FLUSH_CACHE_BOTH:
+             __asm__ __volatile__ ("nop\n\t"
+                                   ".chip 68040\n\t"
+                                   "cpushl %%bc,(%0)\n\t"
+                                   ".chip 68k"
+                                   : : "a" (paddr));
+             break;
+           }
+         if (!--i && len)
+           {
+             /*
+              * No need to page align here since it is done by
+              * virt_to_phys_040().
+              */
+             addr += PAGE_SIZE;
+             i = PAGE_SIZE / 16;
+             /* Recompute physical address when crossing a page
+                boundary. */
+             for (;;)
+               {
+                 if ((paddr = virt_to_phys_040(addr)))
+                   break;
+                 if (len <= i)
+                   return 0;
+                 len -= i;
+                 addr += PAGE_SIZE;
+               }
+           }
+         else
+           paddr += 16;
+       }
+      break;
+
+    default:
+    case FLUSH_SCOPE_PAGE:
+      len += (addr & ~PAGE_MASK) + (PAGE_SIZE - 1);
+      for (len >>= PAGE_SHIFT; len--; addr += PAGE_SIZE)
+       {
+         if (!(paddr = virt_to_phys_040(addr)))
+           continue;
+         switch (cache)
+           {
+           case FLUSH_CACHE_DATA:
+             __asm__ __volatile__ ("nop\n\t"
+                                   ".chip 68040\n\t"
+                                   "cpushp %%dc,(%0)\n\t"
+                                   ".chip 68k"
+                                   : : "a" (paddr));
+             break;
+           case FLUSH_CACHE_INSN:
+             __asm__ __volatile__ ("nop\n\t"
+                                   ".chip 68040\n\t"
+                                   "cpushp %%ic,(%0)\n\t"
+                                   ".chip 68k"
+                                   : : "a" (paddr));
+             break;
+           default:
+           case FLUSH_CACHE_BOTH:
+             __asm__ __volatile__ ("nop\n\t"
+                                   ".chip 68040\n\t"
+                                   "cpushp %%bc,(%0)\n\t"
+                                   ".chip 68k"
+                                   : : "a" (paddr));
+             break;
+           }
+       }
+      break;
+    }
+  return 0;
+}
+
+#define virt_to_phys_060(vaddr)                                \
+({                                                     \
+  unsigned long paddr;                                 \
+  __asm__ __volatile__ (".chip 68060\n\t"              \
+                       "plpar (%0)\n\t"                \
+                       ".chip 68k"                     \
+                       : "=a" (paddr)                  \
+                       : "0" (vaddr));                 \
+  (paddr); /* XXX */                                   \
+})
+
+static inline int
+cache_flush_060 (unsigned long addr, int scope, int cache, unsigned long len)
+{
+  unsigned long paddr, i;
+
+  /*
+   * 68060 manual says:
+   *  cpush %dc : flush DC, remains valid (with our %cacr setup)
+   *  cpush %ic : invalidate IC
+   *  cpush %bc : flush DC + invalidate IC
+   */
+  switch (scope)
+    {
+    case FLUSH_SCOPE_ALL:
+      switch (cache)
+       {
+       case FLUSH_CACHE_DATA:
+         __asm__ __volatile__ (".chip 68060\n\t"
+                               "cpusha %dc\n\t"
+                               ".chip 68k");
+         break;
+       case FLUSH_CACHE_INSN:
+         __asm__ __volatile__ (".chip 68060\n\t"
+                               "cpusha %ic\n\t"
+                               ".chip 68k");
+         break;
+       default:
+       case FLUSH_CACHE_BOTH:
+         __asm__ __volatile__ (".chip 68060\n\t"
+                               "cpusha %bc\n\t"
+                               ".chip 68k");
+         break;
+       }
+      break;
+
+    case FLUSH_SCOPE_LINE:
+      /* Find the physical address of the first mapped page in the
+        address range.  */
+      len += addr & 15;
+      addr &= -16;
+      if (!(paddr = virt_to_phys_060(addr))) {
+       unsigned long tmp = PAGE_SIZE - (addr & ~PAGE_MASK);
+
+       if (len <= tmp)
+         return 0;
+       addr += tmp;
+       len -= tmp;
+       tmp = PAGE_SIZE;
+       for (;;)
+         {
+           if ((paddr = virt_to_phys_060(addr)))
+             break;
+           if (len <= tmp)
+             return 0;
+           addr += tmp;
+           len -= tmp;
+         }
+      }
+      len = (len + 15) >> 4;
+      i = (PAGE_SIZE - (paddr & ~PAGE_MASK)) >> 4;
+      while (len--)
+       {
+         switch (cache)
+           {
+           case FLUSH_CACHE_DATA:
+             __asm__ __volatile__ (".chip 68060\n\t"
+                                   "cpushl %%dc,(%0)\n\t"
+                                   ".chip 68k"
+                                   : : "a" (paddr));
+             break;
+           case FLUSH_CACHE_INSN:
+             __asm__ __volatile__ (".chip 68060\n\t"
+                                   "cpushl %%ic,(%0)\n\t"
+                                   ".chip 68k"
+                                   : : "a" (paddr));
+             break;
+           default:
+           case FLUSH_CACHE_BOTH:
+             __asm__ __volatile__ (".chip 68060\n\t"
+                                   "cpushl %%bc,(%0)\n\t"
+                                   ".chip 68k"
+                                   : : "a" (paddr));
+             break;
+           }
+         if (!--i && len)
+           {
+
+             /*
+              * We just want to jump to the first cache line
+              * in the next page.
+              */
+             addr += PAGE_SIZE;
+             addr &= PAGE_MASK;
+
+             i = PAGE_SIZE / 16;
+             /* Recompute physical address when crossing a page
+                boundary. */
+             for (;;)
+               {
+                 if ((paddr = virt_to_phys_060(addr)))
+                   break;
+                 if (len <= i)
+                   return 0;
+                 len -= i;
+                 addr += PAGE_SIZE;
+               }
+           }
+         else
+           paddr += 16;
+       }
+      break;
+
+    default:
+    case FLUSH_SCOPE_PAGE:
+      len += (addr & ~PAGE_MASK) + (PAGE_SIZE - 1);
+      addr &= PAGE_MASK;       /* Workaround for bug in some
+                                  revisions of the 68060 */
+      for (len >>= PAGE_SHIFT; len--; addr += PAGE_SIZE)
+       {
+         if (!(paddr = virt_to_phys_060(addr)))
+           continue;
+         switch (cache)
+           {
+           case FLUSH_CACHE_DATA:
+             __asm__ __volatile__ (".chip 68060\n\t"
+                                   "cpushp %%dc,(%0)\n\t"
+                                   ".chip 68k"
+                                   : : "a" (paddr));
+             break;
+           case FLUSH_CACHE_INSN:
+             __asm__ __volatile__ (".chip 68060\n\t"
+                                   "cpushp %%ic,(%0)\n\t"
+                                   ".chip 68k"
+                                   : : "a" (paddr));
+             break;
+           default:
+           case FLUSH_CACHE_BOTH:
+             __asm__ __volatile__ (".chip 68060\n\t"
+                                   "cpushp %%bc,(%0)\n\t"
+                                   ".chip 68k"
+                                   : : "a" (paddr));
+             break;
+           }
+       }
+      break;
+    }
+  return 0;
+}
+
+/* sys_cacheflush -- flush (part of) the processor cache.  */
+asmlinkage int
+sys_cacheflush (unsigned long addr, int scope, int cache, unsigned long len)
+{
+       struct vm_area_struct *vma;
+       int ret = -EINVAL;
+
+       if (scope < FLUSH_SCOPE_LINE || scope > FLUSH_SCOPE_ALL ||
+           cache & ~FLUSH_CACHE_BOTH)
+               goto out;
+
+       if (scope == FLUSH_SCOPE_ALL) {
+               /* Only the superuser may explicitly flush the whole cache. */
+               ret = -EPERM;
+               if (!capable(CAP_SYS_ADMIN))
+                       goto out;
+       } else {
+               /*
+                * Verify that the specified address region actually belongs
+                * to this process.
+                */
+               vma = find_vma (current->mm, addr);
+               ret = -EINVAL;
+               /* Check for overflow.  */
+               if (addr + len < addr)
+                       goto out;
+               if (vma == NULL || addr < vma->vm_start || addr + len > vma->vm_end)
+                       goto out;
+       }
+
+       if (CPU_IS_020_OR_030) {
+               if (scope == FLUSH_SCOPE_LINE && len < 256) {
+                       unsigned long cacr;
+                       __asm__ ("movec %%cacr, %0" : "=r" (cacr));
+                       if (cache & FLUSH_CACHE_INSN)
+                               cacr |= 4;
+                       if (cache & FLUSH_CACHE_DATA)
+                               cacr |= 0x400;
+                       len >>= 2;
+                       while (len--) {
+                               __asm__ __volatile__ ("movec %1, %%caar\n\t"
+                                                     "movec %0, %%cacr"
+                                                     : /* no outputs */
+                                                     : "r" (cacr), "r" (addr));
+                               addr += 4;
+                       }
+               } else {
+                       /* Flush the whole cache, even if page granularity requested. */
+                       unsigned long cacr;
+                       __asm__ ("movec %%cacr, %0" : "=r" (cacr));
+                       if (cache & FLUSH_CACHE_INSN)
+                               cacr |= 8;
+                       if (cache & FLUSH_CACHE_DATA)
+                               cacr |= 0x800;
+                       __asm__ __volatile__ ("movec %0, %%cacr" : : "r" (cacr));
+               }
+               ret = 0;
+               goto out;
+       } else {
+           /*
+            * 040 or 060: don't blindly trust 'scope', someone could
+            * try to flush a few megs of memory.
+            */
+
+           if (len>=3*PAGE_SIZE && scope<FLUSH_SCOPE_PAGE)
+               scope=FLUSH_SCOPE_PAGE;
+           if (len>=10*PAGE_SIZE && scope<FLUSH_SCOPE_ALL)
+               scope=FLUSH_SCOPE_ALL;
+           if (CPU_IS_040) {
+               ret = cache_flush_040 (addr, scope, cache, len);
+           } else if (CPU_IS_060) {
+               ret = cache_flush_060 (addr, scope, cache, len);
+           }
+       }
+out:
+       return ret;
+}
+
+/* This syscall gets its arguments in A0 (mem), D2 (oldval) and
+   D1 (newval).  */
+asmlinkage int
+sys_atomic_cmpxchg_32(unsigned long newval, int oldval, int d3, int d4, int d5,
+                     unsigned long __user * mem)
+{
+       /* This was borrowed from ARM's implementation.  */
+       for (;;) {
+               struct mm_struct *mm = current->mm;
+               pgd_t *pgd;
+               pmd_t *pmd;
+               pte_t *pte;
+               spinlock_t *ptl;
+               unsigned long mem_value;
+
+               down_read(&mm->mmap_sem);
+               pgd = pgd_offset(mm, (unsigned long)mem);
+               if (!pgd_present(*pgd))
+                       goto bad_access;
+               pmd = pmd_offset(pgd, (unsigned long)mem);
+               if (!pmd_present(*pmd))
+                       goto bad_access;
+               pte = pte_offset_map_lock(mm, pmd, (unsigned long)mem, &ptl);
+               if (!pte_present(*pte) || !pte_dirty(*pte)
+                   || !pte_write(*pte)) {
+                       pte_unmap_unlock(pte, ptl);
+                       goto bad_access;
+               }
+
+               mem_value = *mem;
+               if (mem_value == oldval)
+                       *mem = newval;
+
+               pte_unmap_unlock(pte, ptl);
+               up_read(&mm->mmap_sem);
+               return mem_value;
+
+             bad_access:
+               up_read(&mm->mmap_sem);
+               /* This is not necessarily a bad access, we can get here if
+                  a memory we're trying to write to should be copied-on-write.
+                  Make the kernel do the necessary page stuff, then re-iterate.
+                  Simulate a write access fault to do that.  */
+               {
+                       /* The first argument of the function corresponds to
+                          D1, which is the first field of struct pt_regs.  */
+                       struct pt_regs *fp = (struct pt_regs *)&newval;
+
+                       /* '3' is an RMW flag.  */
+                       if (do_page_fault(fp, (unsigned long)mem, 3))
+                               /* If the do_page_fault() failed, we don't
+                                  have anything meaningful to return.
+                                  There should be a SIGSEGV pending for
+                                  the process.  */
+                               return 0xdeadbeef;
+               }
+       }
+}
+
 #else
-#include "sys_m68k_no.c"
-#endif
+
+/* sys_cacheflush -- flush (part of) the processor cache.  */
+asmlinkage int
+sys_cacheflush (unsigned long addr, int scope, int cache, unsigned long len)
+{
+       flush_cache_all();
+       return 0;
+}
+
+/* This syscall gets its arguments in A0 (mem), D2 (oldval) and
+   D1 (newval).  */
+asmlinkage int
+sys_atomic_cmpxchg_32(unsigned long newval, int oldval, int d3, int d4, int d5,
+                     unsigned long __user * mem)
+{
+       struct mm_struct *mm = current->mm;
+       unsigned long mem_value;
+
+       down_read(&mm->mmap_sem);
+
+       mem_value = *mem;
+       if (mem_value == oldval)
+               *mem = newval;
+
+       up_read(&mm->mmap_sem);
+       return mem_value;
+}
+
+#endif /* CONFIG_MMU */
+
+asmlinkage int sys_getpagesize(void)
+{
+       return PAGE_SIZE;
+}
+
+/*
+ * Do a system call from kernel instead of calling sys_execve so we
+ * end up with proper pt_regs.
+ */
+int kernel_execve(const char *filename,
+                 const char *const argv[],
+                 const char *const envp[])
+{
+       register long __res asm ("%d0") = __NR_execve;
+       register long __a asm ("%d1") = (long)(filename);
+       register long __b asm ("%d2") = (long)(argv);
+       register long __c asm ("%d3") = (long)(envp);
+       asm volatile ("trap  #0" : "+d" (__res)
+                       : "d" (__a), "d" (__b), "d" (__c));
+       return __res;
+}
+
+asmlinkage unsigned long sys_get_thread_area(void)
+{
+       return current_thread_info()->tp_value;
+}
+
+asmlinkage int sys_set_thread_area(unsigned long tp)
+{
+       current_thread_info()->tp_value = tp;
+       return 0;
+}
+
+asmlinkage int sys_atomic_barrier(void)
+{
+       /* no code needed for uniprocs */
+       return 0;
+}