mm: move bootmem descriptors definition to a single place
[safe/jmp/linux-2.6] / arch / powerpc / mm / fault.c
index 5aea090..1707d00 100644 (file)
@@ -15,7 +15,6 @@
  *  2 of the License, or (at your option) any later version.
  */
 
-#include <linux/config.h>
 #include <linux/signal.h>
 #include <linux/sched.h>
 #include <linux/kernel.h>
@@ -29,6 +28,7 @@
 #include <linux/highmem.h>
 #include <linux/module.h>
 #include <linux/kprobes.h>
+#include <linux/kdebug.h>
 
 #include <asm/page.h>
 #include <asm/pgtable.h>
 #include <asm/system.h>
 #include <asm/uaccess.h>
 #include <asm/tlbflush.h>
-#include <asm/kdebug.h>
 #include <asm/siginfo.h>
 
+
+#ifdef CONFIG_KPROBES
+static inline int notify_page_fault(struct pt_regs *regs)
+{
+       int ret = 0;
+
+       /* kprobe_running() needs smp_processor_id() */
+       if (!user_mode(regs)) {
+               preempt_disable();
+               if (kprobe_running() && kprobe_fault_handler(regs, 11))
+                       ret = 1;
+               preempt_enable();
+       }
+
+       return ret;
+}
+#else
+static inline int notify_page_fault(struct pt_regs *regs)
+{
+       return 0;
+}
+#endif
+
 /*
  * Check whether the instruction at regs->nip is a store using
  * an update addressing form which will update r1.
@@ -123,7 +145,7 @@ int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address,
        struct mm_struct *mm = current->mm;
        siginfo_t info;
        int code = SEGV_MAPERR;
-       int is_write = 0;
+       int is_write = 0, ret;
        int trap = TRAP(regs);
        int is_exec = trap == 0x400;
 
@@ -142,14 +164,11 @@ int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address,
        is_write = error_code & ESR_DST;
 #endif /* CONFIG_4xx || CONFIG_BOOKE */
 
-       if (notify_die(DIE_PAGE_FAULT, "page_fault", regs, error_code,
-                               11, SIGSEGV) == NOTIFY_STOP)
+       if (notify_page_fault(regs))
                return 0;
 
-       if (trap == 0x300) {
-               if (debugger_fault_handler(regs))
-                       return 0;
-       }
+       if (unlikely(debugger_fault_handler(regs)))
+               return 0;
 
        /* On a kernel SLB miss we can only check for a valid exception entry */
        if (!user_mode(regs) && (address >= TASK_SIZE))
@@ -168,7 +187,7 @@ int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address,
                        return SIGSEGV;
                /* in_atomic() in user mode is really bad,
                   as is current->mm == NULL. */
-               printk(KERN_EMERG "Page fault in user mode with"
+               printk(KERN_EMERG "Page fault in user mode with "
                       "in_atomic() = %d mm = %p\n", in_atomic(), mm);
                printk(KERN_EMERG "NIP = %lx  MSR = %lx\n",
                       regs->nip, regs->msr);
@@ -177,15 +196,15 @@ int __kprobes do_page_fault(struct pt_regs *regs, unsigned long address,
 
        /* When running in the kernel we expect faults to occur only to
         * addresses in user space.  All other faults represent errors in the
-        * kernel and should generate an OOPS.  Unfortunatly, in the case of an
-        * erroneous fault occuring in a code path which already holds mmap_sem
+        * kernel and should generate an OOPS.  Unfortunately, in the case of an
+        * erroneous fault occurring in a code path which already holds mmap_sem
         * we will deadlock attempting to validate the fault against the
         * address space.  Luckily the kernel only validly references user
         * space from well defined areas of code, which are listed in the
         * exceptions table.
         *
         * As the vast majority of faults will be valid we will only perform
-        * the source reference check when there is a possibilty of a deadlock.
+        * the source reference check when there is a possibility of a deadlock.
         * Attempt to lock the address space, if we cannot we then validate the
         * source.  If this is invalid we can skip the address space check,
         * thus avoiding the deadlock.
@@ -258,14 +277,19 @@ good_area:
 #endif /* CONFIG_8xx */
 
        if (is_exec) {
-#ifdef CONFIG_PPC64
+#if !(defined(CONFIG_4xx) || defined(CONFIG_BOOKE))
                /* protection fault */
                if (error_code & DSISR_PROTFAULT)
                        goto bad_area;
-               if (!(vma->vm_flags & VM_EXEC))
+               /*
+                * Allow execution from readable areas if the MMU does not
+                * provide separate controls over reading and executing.
+                */
+               if (!(vma->vm_flags & VM_EXEC) &&
+                   (cpu_has_feature(CPU_FTR_NOEXECUTE) ||
+                    !(vma->vm_flags & (VM_READ | VM_WRITE))))
                        goto bad_area;
-#endif
-#if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
+#else
                pte_t *ptep;
                pmd_t *pmdp;
 
@@ -282,8 +306,9 @@ good_area:
                                        flush_dcache_icache_page(page);
                                        set_bit(PG_arch_1, &page->flags);
                                }
-                               pte_update(ptep, 0, _PAGE_HWEXEC);
-                               _tlbie(address);
+                               pte_update(ptep, 0, _PAGE_HWEXEC |
+                                          _PAGE_ACCESSED);
+                               _tlbie(address, mm->context.id);
                                pte_unmap_unlock(ptep, ptl);
                                up_read(&mm->mmap_sem);
                                return 0;
@@ -300,7 +325,7 @@ good_area:
                /* protection fault */
                if (error_code & 0x08000000)
                        goto bad_area;
-               if (!(vma->vm_flags & (VM_READ | VM_EXEC)))
+               if (!(vma->vm_flags & (VM_READ | VM_EXEC | VM_WRITE)))
                        goto bad_area;
        }
 
@@ -310,22 +335,18 @@ good_area:
         * the fault.
         */
  survive:
-       switch (handle_mm_fault(mm, vma, address, is_write)) {
-
-       case VM_FAULT_MINOR:
-               current->min_flt++;
-               break;
-       case VM_FAULT_MAJOR:
-               current->maj_flt++;
-               break;
-       case VM_FAULT_SIGBUS:
-               goto do_sigbus;
-       case VM_FAULT_OOM:
-               goto out_of_memory;
-       default:
+       ret = handle_mm_fault(mm, vma, address, is_write);
+       if (unlikely(ret & VM_FAULT_ERROR)) {
+               if (ret & VM_FAULT_OOM)
+                       goto out_of_memory;
+               else if (ret & VM_FAULT_SIGBUS)
+                       goto do_sigbus;
                BUG();
        }
-
+       if (ret & VM_FAULT_MAJOR)
+               current->maj_flt++;
+       else
+               current->min_flt++;
        up_read(&mm->mmap_sem);
        return 0;
 
@@ -353,14 +374,14 @@ bad_area_nosemaphore:
  */
 out_of_memory:
        up_read(&mm->mmap_sem);
-       if (current->pid == 1) {
+       if (is_global_init(current)) {
                yield();
                down_read(&mm->mmap_sem);
                goto survive;
        }
        printk("VM: killing process %s\n", current->comm);
        if (user_mode(regs))
-               do_exit(SIGKILL);
+               do_group_exit(SIGKILL);
        return SIGKILL;
 
 do_sigbus:
@@ -393,18 +414,21 @@ void bad_page_fault(struct pt_regs *regs, unsigned long address, int sig)
 
        /* kernel has accessed a bad area */
 
-       printk(KERN_ALERT "Unable to handle kernel paging request for ");
        switch (regs->trap) {
-               case 0x300:
-               case 0x380:
-                       printk("data at address 0x%08lx\n", regs->dar);
-                       break;
-               case 0x400:
-               case 0x480:
-                       printk("instruction fetch\n");
-                       break;
-               default:
-                       printk("unknown fault\n");
+       case 0x300:
+       case 0x380:
+               printk(KERN_ALERT "Unable to handle kernel paging request for "
+                       "data at address 0x%08lx\n", regs->dar);
+               break;
+       case 0x400:
+       case 0x480:
+               printk(KERN_ALERT "Unable to handle kernel paging request for "
+                       "instruction fetch\n");
+               break;
+       default:
+               printk(KERN_ALERT "Unable to handle kernel paging request for "
+                       "unknown fault\n");
+               break;
        }
        printk(KERN_ALERT "Faulting instruction address: 0x%08lx\n",
                regs->nip);