fedora core 6 1.2949 + vserver 2.2.0
[linux-2.6.git] / arch / powerpc / mm / fault.c
index a4815d3..2d9e651 100644 (file)
@@ -1,6 +1,4 @@
 /*
- *  arch/ppc/mm/fault.c
- *
  *  PowerPC version
  *    Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
  *
@@ -17,7 +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>
 #include <asm/kdebug.h>
 #include <asm/siginfo.h>
 
+#ifdef CONFIG_KPROBES
+ATOMIC_NOTIFIER_HEAD(notify_page_fault_chain);
+
+/* Hook to register for page fault notifications */
+int register_page_fault_notifier(struct notifier_block *nb)
+{
+       return atomic_notifier_chain_register(&notify_page_fault_chain, nb);
+}
+
+int unregister_page_fault_notifier(struct notifier_block *nb)
+{
+       return atomic_notifier_chain_unregister(&notify_page_fault_chain, nb);
+}
+
+static inline int notify_page_fault(enum die_val val, const char *str,
+                       struct pt_regs *regs, long err, int trap, int sig)
+{
+       struct die_args args = {
+               .regs = regs,
+               .str = str,
+               .err = err,
+               .trapnr = trap,
+               .signr = sig
+       };
+       return atomic_notifier_call_chain(&notify_page_fault_chain, val, &args);
+}
+#else
+static inline int notify_page_fault(enum die_val val, const char *str,
+                       struct pt_regs *regs, long err, int trap, int sig)
+{
+       return NOTIFY_DONE;
+}
+#endif
+
 /*
  * Check whether the instruction at regs->nip is a store using
  * an update addressing form which will update r1.
@@ -144,7 +175,7 @@ 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,
+       if (notify_page_fault(DIE_PAGE_FAULT, "page_fault", regs, error_code,
                                11, SIGSEGV) == NOTIFY_STOP)
                return 0;
 
@@ -179,15 +210,15 @@ 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.
@@ -269,25 +300,29 @@ good_area:
 #endif
 #if defined(CONFIG_4xx) || defined(CONFIG_BOOKE)
                pte_t *ptep;
+               pmd_t *pmdp;
 
                /* Since 4xx/Book-E supports per-page execute permission,
                 * we lazily flush dcache to icache. */
                ptep = NULL;
-               if (get_pteptr(mm, address, &ptep) && pte_present(*ptep)) {
-                       struct page *page = pte_page(*ptep);
-
-                       if (! test_bit(PG_arch_1, &page->flags)) {
-                               flush_dcache_icache_page(page);
-                               set_bit(PG_arch_1, &page->flags);
+               if (get_pteptr(mm, address, &ptep, &pmdp)) {
+                       spinlock_t *ptl = pte_lockptr(mm, pmdp);
+                       spin_lock(ptl);
+                       if (pte_present(*ptep)) {
+                               struct page *page = pte_page(*ptep);
+
+                               if (!test_bit(PG_arch_1, &page->flags)) {
+                                       flush_dcache_icache_page(page);
+                                       set_bit(PG_arch_1, &page->flags);
+                               }
+                               pte_update(ptep, 0, _PAGE_HWEXEC);
+                               _tlbie(address);
+                               pte_unmap_unlock(ptep, ptl);
+                               up_read(&mm->mmap_sem);
+                               return 0;
                        }
-                       pte_update(ptep, 0, _PAGE_HWEXEC);
-                       _tlbie(address);
-                       pte_unmap(ptep);
-                       up_read(&mm->mmap_sem);
-                       return 0;
+                       pte_unmap_unlock(ptep, ptl);
                }
-               if (ptep != NULL)
-                       pte_unmap(ptep);
 #endif
        /* a write */
        } else if (is_write) {
@@ -298,7 +333,7 @@ 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;
        }
 
@@ -351,12 +386,13 @@ bad_area_nosemaphore:
  */
 out_of_memory:
        up_read(&mm->mmap_sem);
-       if (current->pid == 1) {
+       if (is_init(current)) {
                yield();
                down_read(&mm->mmap_sem);
                goto survive;
        }
-       printk("VM: killing process %s\n", current->comm);
+       printk("VM: killing process %s(%d:#%u)\n",
+               current->comm, current->pid, current->xid);
        if (user_mode(regs))
                do_exit(SIGKILL);
        return SIGKILL;
@@ -391,18 +427,21 @@ 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);