Merge branch 'linus' into cont_syslog
[safe/jmp/linux-2.6] / arch / arm / mm / fault.c
index ee01cca..cbfb2ed 100644 (file)
@@ -16,6 +16,9 @@
 #include <linux/kprobes.h>
 #include <linux/uaccess.h>
 #include <linux/page-flags.h>
+#include <linux/sched.h>
+#include <linux/highmem.h>
+#include <linux/perf_event.h>
 
 #include <asm/system.h>
 #include <asm/pgtable.h>
 
 #include "fault.h"
 
+/*
+ * Fault status register encodings.  We steal bit 31 for our own purposes.
+ */
+#define FSR_LNX_PF             (1 << 31)
+#define FSR_WRITE              (1 << 11)
+#define FSR_FS4                        (1 << 10)
+#define FSR_FS3_0              (15)
+
+static inline int fsr_fs(unsigned int fsr)
+{
+       return (fsr & FSR_FS3_0) | (fsr & FSR_FS4) >> 6;
+}
+
 #ifdef CONFIG_MMU
 
 #ifdef CONFIG_KPROBES
@@ -180,18 +196,35 @@ void do_bad_area(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 #define VM_FAULT_BADMAP                0x010000
 #define VM_FAULT_BADACCESS     0x020000
 
-static int
+/*
+ * Check that the permissions on the VMA allow for the fault which occurred.
+ * If we encountered a write fault, we must have write permission, otherwise
+ * we allow any permission.
+ */
+static inline bool access_error(unsigned int fsr, struct vm_area_struct *vma)
+{
+       unsigned int mask = VM_READ | VM_WRITE | VM_EXEC;
+
+       if (fsr & FSR_WRITE)
+               mask = VM_WRITE;
+       if (fsr & FSR_LNX_PF)
+               mask = VM_EXEC;
+
+       return vma->vm_flags & mask ? false : true;
+}
+
+static int __kprobes
 __do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
                struct task_struct *tsk)
 {
        struct vm_area_struct *vma;
-       int fault, mask;
+       int fault;
 
        vma = find_vma(mm, addr);
        fault = VM_FAULT_BADMAP;
-       if (!vma)
+       if (unlikely(!vma))
                goto out;
-       if (vma->vm_start > addr)
+       if (unlikely(vma->vm_start > addr))
                goto check_stack;
 
        /*
@@ -199,47 +232,24 @@ __do_page_fault(struct mm_struct *mm, unsigned long addr, unsigned int fsr,
         * memory access, so we can handle it.
         */
 good_area:
-       if (fsr & (1 << 11)) /* write? */
-               mask = VM_WRITE;
-       else
-               mask = VM_READ|VM_EXEC|VM_WRITE;
-
-       fault = VM_FAULT_BADACCESS;
-       if (!(vma->vm_flags & mask))
+       if (access_error(fsr, vma)) {
+               fault = VM_FAULT_BADACCESS;
                goto out;
+       }
 
        /*
-        * If for any reason at all we couldn't handle
-        * the fault, make sure we exit gracefully rather
-        * than endlessly redo the fault.
+        * If for any reason at all we couldn't handle the fault, make
+        * sure we exit gracefully rather than endlessly redo the fault.
         */
-survive:
-       fault = handle_mm_fault(mm, vma, addr & PAGE_MASK, (fsr & (1 << 11)) ? FAULT_FLAG_WRITE : 0);
-       if (unlikely(fault & VM_FAULT_ERROR)) {
-               if (fault & VM_FAULT_OOM)
-                       goto out_of_memory;
-               else if (fault & VM_FAULT_SIGBUS)
-                       return fault;
-               BUG();
-       }
+       fault = handle_mm_fault(mm, vma, addr & PAGE_MASK, (fsr & FSR_WRITE) ? FAULT_FLAG_WRITE : 0);
+       if (unlikely(fault & VM_FAULT_ERROR))
+               return fault;
        if (fault & VM_FAULT_MAJOR)
                tsk->maj_flt++;
        else
                tsk->min_flt++;
        return fault;
 
-out_of_memory:
-       if (!is_global_init(tsk))
-               goto out;
-
-       /*
-        * If we are out of memory for pid1, sleep for a while and retry
-        */
-       up_read(&mm->mmap_sem);
-       yield();
-       down_read(&mm->mmap_sem);
-       goto survive;
-
 check_stack:
        if (vma->vm_flags & VM_GROWSDOWN && !expand_stack(vma, addr))
                goto good_area;
@@ -276,17 +286,45 @@ do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
                if (!user_mode(regs) && !search_exception_tables(regs->ARM_pc))
                        goto no_context;
                down_read(&mm->mmap_sem);
+       } else {
+               /*
+                * The above down_read_trylock() might have succeeded in
+                * which case, we'll have missed the might_sleep() from
+                * down_read()
+                */
+               might_sleep();
+#ifdef CONFIG_DEBUG_VM
+               if (!user_mode(regs) &&
+                   !search_exception_tables(regs->ARM_pc))
+                       goto no_context;
+#endif
        }
 
        fault = __do_page_fault(mm, addr, fsr, tsk);
        up_read(&mm->mmap_sem);
 
+       perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS, 1, 0, regs, addr);
+       if (fault & VM_FAULT_MAJOR)
+               perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MAJ, 1, 0, regs, addr);
+       else if (fault & VM_FAULT_MINOR)
+               perf_sw_event(PERF_COUNT_SW_PAGE_FAULTS_MIN, 1, 0, regs, addr);
+
        /*
         * Handle the "normal" case first - VM_FAULT_MAJOR / VM_FAULT_MINOR
         */
        if (likely(!(fault & (VM_FAULT_ERROR | VM_FAULT_BADMAP | VM_FAULT_BADACCESS))))
                return 0;
 
+       if (fault & VM_FAULT_OOM) {
+               /*
+                * We ran out of memory, call the OOM killer, and return to
+                * userspace (which will retry the fault, or kill us if we
+                * got oom-killed)
+                */
+               pagefault_out_of_memory();
+               return 0;
+       }
+
        /*
         * If we are in kernel mode at this point, we
         * have no context to handle this fault with.
@@ -294,16 +332,6 @@ do_page_fault(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
        if (!user_mode(regs))
                goto no_context;
 
-       if (fault & VM_FAULT_OOM) {
-               /*
-                * We ran out of memory, or some other thing
-                * happened to us that made us unable to handle
-                * the page fault gracefully.
-                */
-               printk("VM: killing process %s\n", tsk->comm);
-               do_group_exit(SIGKILL);
-               return 0;
-       }
        if (fault & VM_FAULT_SIGBUS) {
                /*
                 * We had some memory, but were unable to
@@ -365,6 +393,9 @@ do_translation_fault(unsigned long addr, unsigned int fsr,
        if (addr < TASK_SIZE)
                return do_page_fault(addr, fsr, regs);
 
+       if (user_mode(regs))
+               goto bad_area;
+
        index = pgd_index(addr);
 
        /*
@@ -435,7 +466,12 @@ static struct fsr_info {
        { do_bad,               SIGILL,  BUS_ADRALN,    "alignment exception"              },
        { do_bad,               SIGKILL, 0,             "terminal exception"               },
        { do_bad,               SIGILL,  BUS_ADRALN,    "alignment exception"              },
+/* Do we need runtime check ? */
+#if __LINUX_ARM_ARCH__ < 6
        { do_bad,               SIGBUS,  0,             "external abort on linefetch"      },
+#else
+       { do_translation_fault, SIGSEGV, SEGV_MAPERR,   "I-cache maintenance fault"        },
+#endif
        { do_translation_fault, SIGSEGV, SEGV_MAPERR,   "section translation fault"        },
        { do_bad,               SIGBUS,  0,             "external abort on linefetch"      },
        { do_page_fault,        SIGSEGV, SEGV_MAPERR,   "page translation fault"           },
@@ -487,10 +523,10 @@ hook_fault_code(int nr, int (*fn)(unsigned long, unsigned int, struct pt_regs *)
 asmlinkage void __exception
 do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
 {
-       const struct fsr_info *inf = fsr_info + (fsr & 15) + ((fsr & (1 << 10)) >> 6);
+       const struct fsr_info *inf = fsr_info + fsr_fs(fsr);
        struct siginfo info;
 
-       if (!inf->fn(addr, fsr, regs))
+       if (!inf->fn(addr, fsr & ~FSR_LNX_PF, regs))
                return;
 
        printk(KERN_ALERT "Unhandled fault: %s (0x%03x) at 0x%08lx\n",
@@ -503,9 +539,58 @@ do_DataAbort(unsigned long addr, unsigned int fsr, struct pt_regs *regs)
        arm_notify_die("", regs, &info, fsr, 0);
 }
 
+
+static struct fsr_info ifsr_info[] = {
+       { do_bad,               SIGBUS,  0,             "unknown 0"                        },
+       { do_bad,               SIGBUS,  0,             "unknown 1"                        },
+       { do_bad,               SIGBUS,  0,             "debug event"                      },
+       { do_bad,               SIGSEGV, SEGV_ACCERR,   "section access flag fault"        },
+       { do_bad,               SIGBUS,  0,             "unknown 4"                        },
+       { do_translation_fault, SIGSEGV, SEGV_MAPERR,   "section translation fault"        },
+       { do_bad,               SIGSEGV, SEGV_ACCERR,   "page access flag fault"           },
+       { do_page_fault,        SIGSEGV, SEGV_MAPERR,   "page translation fault"           },
+       { do_bad,               SIGBUS,  0,             "external abort on non-linefetch"  },
+       { do_bad,               SIGSEGV, SEGV_ACCERR,   "section domain fault"             },
+       { do_bad,               SIGBUS,  0,             "unknown 10"                       },
+       { do_bad,               SIGSEGV, SEGV_ACCERR,   "page domain fault"                },
+       { do_bad,               SIGBUS,  0,             "external abort on translation"    },
+       { do_sect_fault,        SIGSEGV, SEGV_ACCERR,   "section permission fault"         },
+       { do_bad,               SIGBUS,  0,             "external abort on translation"    },
+       { do_page_fault,        SIGSEGV, SEGV_ACCERR,   "page permission fault"            },
+       { do_bad,               SIGBUS,  0,             "unknown 16"                       },
+       { do_bad,               SIGBUS,  0,             "unknown 17"                       },
+       { do_bad,               SIGBUS,  0,             "unknown 18"                       },
+       { do_bad,               SIGBUS,  0,             "unknown 19"                       },
+       { do_bad,               SIGBUS,  0,             "unknown 20"                       },
+       { do_bad,               SIGBUS,  0,             "unknown 21"                       },
+       { do_bad,               SIGBUS,  0,             "unknown 22"                       },
+       { do_bad,               SIGBUS,  0,             "unknown 23"                       },
+       { do_bad,               SIGBUS,  0,             "unknown 24"                       },
+       { do_bad,               SIGBUS,  0,             "unknown 25"                       },
+       { do_bad,               SIGBUS,  0,             "unknown 26"                       },
+       { do_bad,               SIGBUS,  0,             "unknown 27"                       },
+       { do_bad,               SIGBUS,  0,             "unknown 28"                       },
+       { do_bad,               SIGBUS,  0,             "unknown 29"                       },
+       { do_bad,               SIGBUS,  0,             "unknown 30"                       },
+       { do_bad,               SIGBUS,  0,             "unknown 31"                       },
+};
+
 asmlinkage void __exception
-do_PrefetchAbort(unsigned long addr, struct pt_regs *regs)
+do_PrefetchAbort(unsigned long addr, unsigned int ifsr, struct pt_regs *regs)
 {
-       do_translation_fault(addr, 0, regs);
+       const struct fsr_info *inf = ifsr_info + fsr_fs(ifsr);
+       struct siginfo info;
+
+       if (!inf->fn(addr, ifsr | FSR_LNX_PF, regs))
+               return;
+
+       printk(KERN_ALERT "Unhandled prefetch abort: %s (0x%03x) at 0x%08lx\n",
+               inf->name, ifsr, addr);
+
+       info.si_signo = inf->sig;
+       info.si_errno = 0;
+       info.si_code  = inf->code;
+       info.si_addr  = (void __user *)addr;
+       arm_notify_die("", regs, &info, ifsr, 0);
 }