Merge branch 'for_paulus' of master.kernel.org:/pub/scm/linux/kernel/git/galak/powerpc
[pandora-kernel.git] / arch / x86_64 / mm / fault.c
index 2e7c3c8..08dc696 100644 (file)
 #define PF_RSVD        (1<<3)
 #define PF_INSTR       (1<<4)
 
+#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)
+{
+       vmalloc_sync_all();
+       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
+
 void bust_spinlocks(int yes)
 {
        int loglevel_save = console_loglevel;
@@ -160,7 +195,7 @@ void dump_pagetable(unsigned long address)
        printk("PGD %lx ", pgd_val(*pgd));
        if (!pgd_present(*pgd)) goto ret; 
 
-       pud = __pud_offset_k((pud_t *)pgd_page(*pgd), address);
+       pud = pud_offset(pgd, address);
        if (bad_address(pud)) goto bad;
        printk("PUD %lx ", pud_val(*pud));
        if (!pud_present(*pud)) goto ret;
@@ -264,6 +299,8 @@ static int vmalloc_fault(unsigned long address)
                return -1;
        if (pgd_none(*pgd))
                set_pgd(pgd, *pgd_ref);
+       else
+               BUG_ON(pgd_page(*pgd) != pgd_page(*pgd_ref));
 
        /* Below here mismatches are bugs because these lower tables
           are shared */
@@ -312,21 +349,13 @@ asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
        unsigned long flags;
        siginfo_t info;
 
+       tsk = current;
+       mm = tsk->mm;
+       prefetchw(&mm->mmap_sem);
+
        /* get the address */
        __asm__("movq %%cr2,%0":"=r" (address));
-       if (notify_die(DIE_PAGE_FAULT, "page fault", regs, error_code, 14,
-                                       SIGSEGV) == NOTIFY_STOP)
-               return;
 
-       if (likely(regs->eflags & X86_EFLAGS_IF))
-               local_irq_enable();
-
-       if (unlikely(page_fault_trace))
-               printk("pagefault rip:%lx rsp:%lx cs:%lu ss:%lu address %lx error %lx\n",
-                      regs->rip,regs->rsp,regs->cs,regs->ss,address,error_code); 
-
-       tsk = current;
-       mm = tsk->mm;
        info.si_code = SEGV_MAPERR;
 
 
@@ -351,10 +380,12 @@ asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
                 */
                if (!(error_code & (PF_RSVD|PF_USER|PF_PROT)) &&
                      ((address >= VMALLOC_START && address < VMALLOC_END))) {
-                       if (vmalloc_fault(address) < 0)
-                               goto bad_area_nosemaphore;
-                       return;
+                       if (vmalloc_fault(address) >= 0)
+                               return;
                }
+               if (notify_page_fault(DIE_PAGE_FAULT, "page fault", regs, error_code, 14,
+                                               SIGSEGV) == NOTIFY_STOP)
+                       return;
                /*
                 * Don't take the mm semaphore here. If we fixup a prefetch
                 * fault we could otherwise deadlock.
@@ -362,6 +393,17 @@ asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
                goto bad_area_nosemaphore;
        }
 
+       if (notify_page_fault(DIE_PAGE_FAULT, "page fault", regs, error_code, 14,
+                                       SIGSEGV) == NOTIFY_STOP)
+               return;
+
+       if (likely(regs->eflags & X86_EFLAGS_IF))
+               local_irq_enable();
+
+       if (unlikely(page_fault_trace))
+               printk("pagefault rip:%lx rsp:%lx cs:%lu ss:%lu address %lx error %lx\n",
+                      regs->rip,regs->rsp,regs->cs,regs->ss,address,error_code); 
+
        if (unlikely(error_code & PF_RSVD))
                pgtable_bad(address, regs, error_code);
 
@@ -403,8 +445,10 @@ asmlinkage void __kprobes do_page_fault(struct pt_regs *regs,
        if (!(vma->vm_flags & VM_GROWSDOWN))
                goto bad_area;
        if (error_code & 4) {
-               // XXX: align red zone size with ABI 
-               if (address + 128 < regs->rsp)
+               /* Allow userspace just enough access below the stack pointer
+                * to let the 'enter' instruction work.
+                */
+               if (address + 65536 + 32 * sizeof(unsigned long) < regs->rsp)
                        goto bad_area;
        }
        if (expand_stack(vma, address))
@@ -571,9 +615,51 @@ do_sigbus:
        return;
 }
 
+DEFINE_SPINLOCK(pgd_lock);
+struct page *pgd_list;
+
+void vmalloc_sync_all(void)
+{
+       /* Note that races in the updates of insync and start aren't 
+          problematic:
+          insync can only get set bits added, and updates to start are only
+          improving performance (without affecting correctness if undone). */
+       static DECLARE_BITMAP(insync, PTRS_PER_PGD);
+       static unsigned long start = VMALLOC_START & PGDIR_MASK;
+       unsigned long address;
+
+       for (address = start; address <= VMALLOC_END; address += PGDIR_SIZE) {
+               if (!test_bit(pgd_index(address), insync)) {
+                       const pgd_t *pgd_ref = pgd_offset_k(address);
+                       struct page *page;
+
+                       if (pgd_none(*pgd_ref))
+                               continue;
+                       spin_lock(&pgd_lock);
+                       for (page = pgd_list; page;
+                            page = (struct page *)page->index) {
+                               pgd_t *pgd;
+                               pgd = (pgd_t *)page_address(page) + pgd_index(address);
+                               if (pgd_none(*pgd))
+                                       set_pgd(pgd, *pgd_ref);
+                               else
+                                       BUG_ON(pgd_page(*pgd) != pgd_page(*pgd_ref));
+                       }
+                       spin_unlock(&pgd_lock);
+                       set_bit(pgd_index(address), insync);
+               }
+               if (address == start)
+                       start = address + PGDIR_SIZE;
+       }
+       /* Check that there is no need to do the same for the modules area. */
+       BUILD_BUG_ON(!(MODULES_VADDR > __START_KERNEL));
+       BUILD_BUG_ON(!(((MODULES_END - 1) & PGDIR_MASK) == 
+                               (__START_KERNEL & PGDIR_MASK)));
+}
+
 static int __init enable_pagefaulttrace(char *str)
 {
        page_fault_trace = 1;
-       return 0;
+       return 1;
 }
 __setup("pagefaulttrace", enable_pagefaulttrace);