Merge branch 'x86/mce' into x86/urgent
[safe/jmp/linux-2.6] / arch / mips / kernel / traps.c
index 7b78d13..0a18b4c 100644 (file)
@@ -22,6 +22,9 @@
 #include <linux/kallsyms.h>
 #include <linux/bootmem.h>
 #include <linux/interrupt.h>
+#include <linux/ptrace.h>
+#include <linux/kgdb.h>
+#include <linux/kdebug.h>
 
 #include <asm/bootinfo.h>
 #include <asm/branch.h>
@@ -29,6 +32,7 @@
 #include <asm/cpu.h>
 #include <asm/dsp.h>
 #include <asm/fpu.h>
+#include <asm/fpu_emulator.h>
 #include <asm/mipsregs.h>
 #include <asm/mipsmtregs.h>
 #include <asm/module.h>
 #include <asm/tlbdebug.h>
 #include <asm/traps.h>
 #include <asm/uaccess.h>
+#include <asm/watch.h>
 #include <asm/mmu_context.h>
 #include <asm/types.h>
 #include <asm/stacktrace.h>
+#include <asm/irq.h>
 
+extern void check_wait(void);
+extern asmlinkage void r4k_wait(void);
+extern asmlinkage void rollback_handle_int(void);
 extern asmlinkage void handle_int(void);
 extern asmlinkage void handle_tlbm(void);
 extern asmlinkage void handle_tlbl(void);
@@ -70,7 +79,10 @@ extern asmlinkage void handle_reserved(void);
 extern int fpu_emulator_cop1Handler(struct pt_regs *xcp,
        struct mips_fpu_struct *ctx, int has_fpu);
 
-void (*board_watchpoint_handler)(struct pt_regs *regs);
+#ifdef CONFIG_CPU_CAVIUM_OCTEON
+extern asmlinkage void octeon_cop2_restore(struct octeon_cop2_state *task);
+#endif
+
 void (*board_be_init)(void);
 int (*board_be_handler)(struct pt_regs *regs, int is_fixup);
 void (*board_nmi_handler_setup)(void);
@@ -80,7 +92,7 @@ void (*board_bind_eic_interrupt)(int irq, int regset);
 
 static void show_raw_backtrace(unsigned long reg29)
 {
-       unsigned long *sp = (unsigned long *)reg29;
+       unsigned long *sp = (unsigned long *)(reg29 & ~3);
        unsigned long addr;
 
        printk("Call Trace:");
@@ -88,7 +100,12 @@ static void show_raw_backtrace(unsigned long reg29)
        printk("\n");
 #endif
        while (!kstack_end(sp)) {
-               addr = *sp++;
+               unsigned long __user *p =
+                       (unsigned long __user *)(unsigned long)sp++;
+               if (__get_user(addr, p)) {
+                       printk(" (Bad stack address)");
+                       break;
+               }
                if (__kernel_text_address(addr))
                        print_ip_sym(addr);
        }
@@ -192,16 +209,19 @@ EXPORT_SYMBOL(dump_stack);
 static void show_code(unsigned int __user *pc)
 {
        long i;
+       unsigned short __user *pc16 = NULL;
 
        printk("\nCode:");
 
+       if ((unsigned long)pc & 1)
+               pc16 = (unsigned short __user *)((unsigned long)pc & ~1);
        for(i = -3 ; i < 6 ; i++) {
                unsigned int insn;
-               if (__get_user(insn, pc + i)) {
+               if (pc16 ? __get_user(insn, pc16 + i) : __get_user(insn, pc + i)) {
                        printk(" (Bad address in epc)\n");
                        break;
                }
-               printk("%c%08x%c", (i?' ':'<'), insn, (i?' ':'>'));
+               printk("%c%0*x%c", (i?' ':'<'), pc16 ? 4 : 8, insn, (i?' ':'>'));
        }
 }
 
@@ -240,11 +260,11 @@ static void __show_regs(const struct pt_regs *regs)
        /*
         * Saved cp0 registers
         */
-       printk("epc   : %0*lx ", field, regs->cp0_epc);
-       print_symbol("%s ", regs->cp0_epc);
+       printk("epc   : %0*lx %pS\n", field, regs->cp0_epc,
+              (void *) regs->cp0_epc);
        printk("    %s\n", print_tainted());
-       printk("ra    : %0*lx ", field, regs->regs[31]);
-       print_symbol("%s\n", regs->regs[31]);
+       printk("ra    : %0*lx %pS\n", field, regs->regs[31],
+              (void *) regs->regs[31]);
 
        printk("Status: %08x    ", (uint32_t) regs->cp0_status);
 
@@ -311,10 +331,21 @@ void show_regs(struct pt_regs *regs)
 
 void show_registers(const struct pt_regs *regs)
 {
+       const int field = 2 * sizeof(unsigned long);
+
        __show_regs(regs);
        print_modules();
-       printk("Process %s (pid: %d, threadinfo=%p, task=%p)\n",
-               current->comm, current->pid, current_thread_info(), current);
+       printk("Process %s (pid: %d, threadinfo=%p, task=%p, tls=%0*lx)\n",
+              current->comm, current->pid, current_thread_info(), current,
+             field, current_thread_info()->tp_value);
+       if (cpu_has_userlocal) {
+               unsigned long tls;
+
+               tls = read_c0_userlocal();
+               if (tls != current_thread_info()->tp_value)
+                       printk("*HwTLS: %0*lx\n", field, tls);
+       }
+
        show_stacktrace(current, regs);
        show_code((unsigned int __user *) regs->cp0_epc);
        printk("\n");
@@ -352,8 +383,8 @@ void __noreturn die(const char * str, const struct pt_regs * regs)
        do_exit(SIGSEGV);
 }
 
-extern const struct exception_table_entry __start___dbe_table[];
-extern const struct exception_table_entry __stop___dbe_table[];
+extern struct exception_table_entry __start___dbe_table[];
+extern struct exception_table_entry __stop___dbe_table[];
 
 __asm__(
 "      .section        __dbe_table, \"a\"\n"
@@ -406,6 +437,10 @@ asmlinkage void do_be(struct pt_regs *regs)
        printk(KERN_ALERT "%s bus error, epc == %0*lx, ra == %0*lx\n",
               data ? "Data" : "Instruction",
               field, regs->cp0_epc, field, regs->regs[31]);
+       if (notify_die(DIE_OOPS, "bus error", regs, SIGBUS, 0, 0)
+           == NOTIFY_STOP)
+               return;
+
        die_if_kernel("Oops", regs);
        force_sig(SIGBUS, current);
 }
@@ -431,9 +466,8 @@ asmlinkage void do_be(struct pt_regs *regs)
  * The ll_bit is cleared by r*_switch.S
  */
 
-unsigned long ll_bit;
-
-static struct task_struct *ll_task = NULL;
+unsigned int ll_bit;
+struct task_struct *ll_task;
 
 static inline int simulate_ll(struct pt_regs *regs, unsigned int opcode)
 {
@@ -534,8 +568,7 @@ static int simulate_llsc(struct pt_regs *regs, unsigned int opcode)
 
 /*
  * Simulate trapping 'rdhwr' instructions to provide user accessible
- * registers not implemented in hardware.  The only current use of this
- * is the thread area pointer.
+ * registers not implemented in hardware.
  */
 static int simulate_rdhwr(struct pt_regs *regs, unsigned int opcode)
 {
@@ -545,11 +578,31 @@ static int simulate_rdhwr(struct pt_regs *regs, unsigned int opcode)
                int rd = (opcode & RD) >> 11;
                int rt = (opcode & RT) >> 16;
                switch (rd) {
-                       case 29:
-                               regs->regs[rt] = ti->tp_value;
-                               return 0;
+               case 0:         /* CPU number */
+                       regs->regs[rt] = smp_processor_id();
+                       return 0;
+               case 1:         /* SYNCI length */
+                       regs->regs[rt] = min(current_cpu_data.dcache.linesz,
+                                            current_cpu_data.icache.linesz);
+                       return 0;
+               case 2:         /* Read count register */
+                       regs->regs[rt] = read_c0_count();
+                       return 0;
+               case 3:         /* Count register resolution */
+                       switch (current_cpu_data.cputype) {
+                       case CPU_20KC:
+                       case CPU_25KF:
+                               regs->regs[rt] = 1;
+                               break;
                        default:
-                               return -1;
+                               regs->regs[rt] = 2;
+                       }
+                       return 0;
+               case 29:
+                       regs->regs[rt] = ti->tp_value;
+                       return 0;
+               default:
+                       return -1;
                }
        }
 
@@ -585,6 +638,9 @@ asmlinkage void do_fpe(struct pt_regs *regs, unsigned long fcr31)
 {
        siginfo_t info;
 
+       if (notify_die(DIE_FP, "FP exception", regs, SIGFPE, 0, 0)
+           == NOTIFY_STOP)
+               return;
        die_if_kernel("FP exception in kernel code", regs);
 
        if (fcr31 & FPU_CSR_UNI_X) {
@@ -638,35 +694,27 @@ asmlinkage void do_fpe(struct pt_regs *regs, unsigned long fcr31)
        force_sig_info(SIGFPE, &info, current);
 }
 
-asmlinkage void do_bp(struct pt_regs *regs)
+static void do_trap_or_bp(struct pt_regs *regs, unsigned int code,
+       const char *str)
 {
-       unsigned int opcode, bcode;
        siginfo_t info;
+       char b[40];
 
-       if (__get_user(opcode, (unsigned int __user *) exception_epc(regs)))
-               goto out_sigsegv;
-
-       /*
-        * There is the ancient bug in the MIPS assemblers that the break
-        * code starts left to bit 16 instead to bit 6 in the opcode.
-        * Gas is bug-compatible, but not always, grrr...
-        * We handle both cases with a simple heuristics.  --macro
-        */
-       bcode = ((opcode >> 6) & ((1 << 20) - 1));
-       if (bcode < (1 << 10))
-               bcode <<= 10;
+       if (notify_die(DIE_TRAP, str, regs, code, 0, 0) == NOTIFY_STOP)
+               return;
 
        /*
-        * (A short test says that IRIX 5.3 sends SIGTRAP for all break
-        * insns, even for break codes that indicate arithmetic failures.
-        * Weird ...)
+        * A short test says that IRIX 5.3 sends SIGTRAP for all trap
+        * insns, even for trap and break codes that indicate arithmetic
+        * failures.  Weird ...
         * But should we continue the brokenness???  --macro
         */
-       switch (bcode) {
-       case BRK_OVERFLOW << 10:
-       case BRK_DIVZERO << 10:
-               die_if_kernel("Break instruction in kernel code", regs);
-               if (bcode == (BRK_DIVZERO << 10))
+       switch (code) {
+       case BRK_OVERFLOW:
+       case BRK_DIVZERO:
+               scnprintf(b, sizeof(b), "%s instruction in kernel code", str);
+               die_if_kernel(b, regs);
+               if (code == BRK_DIVZERO)
                        info.si_code = FPE_INTDIV;
                else
                        info.si_code = FPE_INTOVF;
@@ -676,12 +724,49 @@ asmlinkage void do_bp(struct pt_regs *regs)
                force_sig_info(SIGFPE, &info, current);
                break;
        case BRK_BUG:
-               die("Kernel bug detected", regs);
+               die_if_kernel("Kernel bug detected", regs);
+               force_sig(SIGTRAP, current);
+               break;
+       case BRK_MEMU:
+               /*
+                * Address errors may be deliberately induced by the FPU
+                * emulator to retake control of the CPU after executing the
+                * instruction in the delay slot of an emulated branch.
+                *
+                * Terminate if exception was recognized as a delay slot return
+                * otherwise handle as normal.
+                */
+               if (do_dsemulret(regs))
+                       return;
+
+               die_if_kernel("Math emu break/trap", regs);
+               force_sig(SIGTRAP, current);
                break;
        default:
-               die_if_kernel("Break instruction in kernel code", regs);
+               scnprintf(b, sizeof(b), "%s instruction in kernel code", str);
+               die_if_kernel(b, regs);
                force_sig(SIGTRAP, current);
        }
+}
+
+asmlinkage void do_bp(struct pt_regs *regs)
+{
+       unsigned int opcode, bcode;
+
+       if (__get_user(opcode, (unsigned int __user *) exception_epc(regs)))
+               goto out_sigsegv;
+
+       /*
+        * There is the ancient bug in the MIPS assemblers that the break
+        * code starts left to bit 16 instead to bit 6 in the opcode.
+        * Gas is bug-compatible, but not always, grrr...
+        * We handle both cases with a simple heuristics.  --macro
+        */
+       bcode = ((opcode >> 6) & ((1 << 20) - 1));
+       if (bcode >= (1 << 10))
+               bcode >>= 10;
+
+       do_trap_or_bp(regs, bcode, "Break");
        return;
 
 out_sigsegv:
@@ -691,7 +776,6 @@ out_sigsegv:
 asmlinkage void do_tr(struct pt_regs *regs)
 {
        unsigned int opcode, tcode = 0;
-       siginfo_t info;
 
        if (__get_user(opcode, (unsigned int __user *) exception_epc(regs)))
                goto out_sigsegv;
@@ -700,32 +784,7 @@ asmlinkage void do_tr(struct pt_regs *regs)
        if (!(opcode & OPCODE))
                tcode = ((opcode >> 6) & ((1 << 10) - 1));
 
-       /*
-        * (A short test says that IRIX 5.3 sends SIGTRAP for all trap
-        * insns, even for trap codes that indicate arithmetic failures.
-        * Weird ...)
-        * But should we continue the brokenness???  --macro
-        */
-       switch (tcode) {
-       case BRK_OVERFLOW:
-       case BRK_DIVZERO:
-               die_if_kernel("Trap instruction in kernel code", regs);
-               if (tcode == BRK_DIVZERO)
-                       info.si_code = FPE_INTDIV;
-               else
-                       info.si_code = FPE_INTOVF;
-               info.si_signo = SIGFPE;
-               info.si_errno = 0;
-               info.si_addr = (void __user *) regs->cp0_epc;
-               force_sig_info(SIGFPE, &info, current);
-               break;
-       case BRK_BUG:
-               die("Kernel bug detected", regs);
-               break;
-       default:
-               die_if_kernel("Trap instruction in kernel code", regs);
-               force_sig(SIGTRAP, current);
-       }
+       do_trap_or_bp(regs, tcode, "Trap");
        return;
 
 out_sigsegv:
@@ -739,6 +798,10 @@ asmlinkage void do_ri(struct pt_regs *regs)
        unsigned int opcode = 0;
        int status = -1;
 
+       if (notify_die(DIE_RI, "RI Fault", regs, SIGSEGV, 0, 0)
+           == NOTIFY_STOP)
+               return;
+
        die_if_kernel("Reserved instruction in kernel code", regs);
 
        if (unlikely(compute_return_epc(regs) < 0))
@@ -783,8 +846,10 @@ static void mt_ase_fp_affinity(void)
                if (cpus_intersects(current->cpus_allowed, mt_fpu_cpumask)) {
                        cpumask_t tmask;
 
-                       cpus_and(tmask, current->thread.user_cpus_allowed,
-                                mt_fpu_cpumask);
+                       current->thread.user_cpus_allowed
+                               = current->cpus_allowed;
+                       cpus_and(tmask, current->cpus_allowed,
+                               mt_fpu_cpumask);
                        set_cpus_allowed(current, tmask);
                        set_thread_flag(TIF_FPUBOUND);
                }
@@ -799,6 +864,7 @@ asmlinkage void do_cpu(struct pt_regs *regs)
        unsigned int opcode;
        unsigned int cpid;
        int status;
+       unsigned long __maybe_unused flags;
 
        die_if_kernel("do_cpu invoked from kernel context!", regs);
 
@@ -854,6 +920,17 @@ asmlinkage void do_cpu(struct pt_regs *regs)
                return;
 
        case 2:
+#ifdef CONFIG_CPU_CAVIUM_OCTEON
+               prefetch(&current->thread.cp2);
+               local_irq_save(flags);
+               KSTK_STATUS(current) |= ST0_CU2;
+               status = read_c0_status();
+               write_c0_status(status | ST0_CU2);
+               octeon_cop2_restore(&(current->thread.cp2));
+               write_c0_status(status & ~ST0_CU2);
+               local_irq_restore(flags);
+               return;
+#endif
        case 3:
                break;
        }
@@ -866,20 +943,34 @@ asmlinkage void do_mdmx(struct pt_regs *regs)
        force_sig(SIGILL, current);
 }
 
+/*
+ * Called with interrupts disabled.
+ */
 asmlinkage void do_watch(struct pt_regs *regs)
 {
-       if (board_watchpoint_handler) {
-               (*board_watchpoint_handler)(regs);
-               return;
-       }
+       u32 cause;
 
        /*
-        * We use the watch exception where available to detect stack
-        * overflows.
+        * Clear WP (bit 22) bit of cause register so we don't loop
+        * forever.
         */
-       dump_tlb_all();
-       show_regs(regs);
-       panic("Caught WATCH exception - probably caused by stack overflow.");
+       cause = read_c0_cause();
+       cause &= ~(1 << 22);
+       write_c0_cause(cause);
+
+       /*
+        * If the current thread has the watch registers loaded, save
+        * their values and send SIGTRAP.  Otherwise another thread
+        * left the registers set, clear them and continue.
+        */
+       if (test_tsk_thread_flag(current, TIF_LOAD_WATCH)) {
+               mips_read_watch_registers();
+               local_irq_enable();
+               force_sig(SIGTRAP, current);
+       } else {
+               mips_clear_watch_registers();
+               local_irq_enable();
+       }
 }
 
 asmlinkage void do_mcheck(struct pt_regs *regs)
@@ -966,6 +1057,21 @@ asmlinkage void do_reserved(struct pt_regs *regs)
              (regs->cp0_cause & 0x7f) >> 2);
 }
 
+static int __initdata l1parity = 1;
+static int __init nol1parity(char *s)
+{
+       l1parity = 0;
+       return 1;
+}
+__setup("nol1par", nol1parity);
+static int __initdata l2parity = 1;
+static int __init nol2parity(char *s)
+{
+       l2parity = 0;
+       return 1;
+}
+__setup("nol2par", nol2parity);
+
 /*
  * Some MIPS CPUs can enable/disable for cache parity detection, but do
  * it different ways.
@@ -975,6 +1081,62 @@ static inline void parity_protection_init(void)
        switch (current_cpu_type()) {
        case CPU_24K:
        case CPU_34K:
+       case CPU_74K:
+       case CPU_1004K:
+               {
+#define ERRCTL_PE      0x80000000
+#define ERRCTL_L2P     0x00800000
+                       unsigned long errctl;
+                       unsigned int l1parity_present, l2parity_present;
+
+                       errctl = read_c0_ecc();
+                       errctl &= ~(ERRCTL_PE|ERRCTL_L2P);
+
+                       /* probe L1 parity support */
+                       write_c0_ecc(errctl | ERRCTL_PE);
+                       back_to_back_c0_hazard();
+                       l1parity_present = (read_c0_ecc() & ERRCTL_PE);
+
+                       /* probe L2 parity support */
+                       write_c0_ecc(errctl|ERRCTL_L2P);
+                       back_to_back_c0_hazard();
+                       l2parity_present = (read_c0_ecc() & ERRCTL_L2P);
+
+                       if (l1parity_present && l2parity_present) {
+                               if (l1parity)
+                                       errctl |= ERRCTL_PE;
+                               if (l1parity ^ l2parity)
+                                       errctl |= ERRCTL_L2P;
+                       } else if (l1parity_present) {
+                               if (l1parity)
+                                       errctl |= ERRCTL_PE;
+                       } else if (l2parity_present) {
+                               if (l2parity)
+                                       errctl |= ERRCTL_L2P;
+                       } else {
+                               /* No parity available */
+                       }
+
+                       printk(KERN_INFO "Writing ErrCtl register=%08lx\n", errctl);
+
+                       write_c0_ecc(errctl);
+                       back_to_back_c0_hazard();
+                       errctl = read_c0_ecc();
+                       printk(KERN_INFO "Readback ErrCtl register=%08lx\n", errctl);
+
+                       if (l1parity_present)
+                               printk(KERN_INFO "Cache parity protection %sabled\n",
+                                      (errctl & ERRCTL_PE) ? "en" : "dis");
+
+                       if (l2parity_present) {
+                               if (l1parity_present && l1parity)
+                                       errctl ^= ERRCTL_L2P;
+                               printk(KERN_INFO "L2 cache parity protection %sabled\n",
+                                      (errctl & ERRCTL_L2P) ? "en" : "dis");
+                       }
+               }
+               break;
+
        case CPU_5KC:
                write_c0_ecc(0x80000000);
                back_to_back_c0_hazard();
@@ -1095,64 +1257,11 @@ void *set_except_vector(int n, void *addr)
        if (n == 0 && cpu_has_divec) {
                *(u32 *)(ebase + 0x200) = 0x08000000 |
                                          (0x03ffffff & (handler >> 2));
-               flush_icache_range(ebase + 0x200, ebase + 0x204);
+               local_flush_icache_range(ebase + 0x200, ebase + 0x204);
        }
        return (void *)old_handler;
 }
 
-#ifdef CONFIG_CPU_MIPSR2_SRS
-/*
- * MIPSR2 shadow register set allocation
- * FIXME: SMP...
- */
-
-static struct shadow_registers {
-       /*
-        * Number of shadow register sets supported
-        */
-       unsigned long sr_supported;
-       /*
-        * Bitmap of allocated shadow registers
-        */
-       unsigned long sr_allocated;
-} shadow_registers;
-
-static void mips_srs_init(void)
-{
-       shadow_registers.sr_supported = ((read_c0_srsctl() >> 26) & 0x0f) + 1;
-       printk(KERN_INFO "%ld MIPSR2 register sets available\n",
-              shadow_registers.sr_supported);
-       shadow_registers.sr_allocated = 1;      /* Set 0 used by kernel */
-}
-
-int mips_srs_max(void)
-{
-       return shadow_registers.sr_supported;
-}
-
-int mips_srs_alloc(void)
-{
-       struct shadow_registers *sr = &shadow_registers;
-       int set;
-
-again:
-       set = find_first_zero_bit(&sr->sr_allocated, sr->sr_supported);
-       if (set >= sr->sr_supported)
-               return -1;
-
-       if (test_and_set_bit(set, &sr->sr_allocated))
-               goto again;
-
-       return set;
-}
-
-void mips_srs_free(int set)
-{
-       struct shadow_registers *sr = &shadow_registers;
-
-       clear_bit(set, &sr->sr_allocated);
-}
-
 static asmlinkage void do_default_vi(void)
 {
        show_regs(get_irq_regs());
@@ -1163,11 +1272,11 @@ static void *set_vi_srs_handler(int n, vi_handler_t addr, int srs)
 {
        unsigned long handler;
        unsigned long old_handler = vi_handlers[n];
+       int srssets = current_cpu_data.srsets;
        u32 *w;
        unsigned char *b;
 
-       if (!cpu_has_veic && !cpu_has_vint)
-               BUG();
+       BUG_ON(!cpu_has_veic && !cpu_has_vint);
 
        if (addr == NULL) {
                handler = (unsigned long) do_default_vi;
@@ -1178,7 +1287,7 @@ static void *set_vi_srs_handler(int n, vi_handler_t addr, int srs)
 
        b = (unsigned char *)(ebase + 0x200 + n*VECTORSPACING);
 
-       if (srs >= mips_srs_max())
+       if (srs >= srssets)
                panic("Shadow register set %d not supported", srs);
 
        if (cpu_has_veic) {
@@ -1186,7 +1295,7 @@ static void *set_vi_srs_handler(int n, vi_handler_t addr, int srs)
                        board_bind_eic_interrupt(n, srs);
        } else if (cpu_has_vint) {
                /* SRSMap is only defined if shadow sets are implemented */
-               if (mips_srs_max() > 1)
+               if (srssets > 1)
                        change_c0_srsmap(0xf << n*4, srs << n*4);
        }
 
@@ -1198,6 +1307,9 @@ static void *set_vi_srs_handler(int n, vi_handler_t addr, int srs)
 
                extern char except_vec_vi, except_vec_vi_lui;
                extern char except_vec_vi_ori, except_vec_vi_end;
+               extern char rollback_except_vec_vi;
+               char *vec_start = (cpu_wait == r4k_wait) ?
+                       &rollback_except_vec_vi : &except_vec_vi;
 #ifdef CONFIG_MIPS_MT_SMTC
                /*
                 * We need to provide the SMTC vectored interrupt handler
@@ -1205,11 +1317,11 @@ static void *set_vi_srs_handler(int n, vi_handler_t addr, int srs)
                 * Status.IM bit to be masked before going there.
                 */
                extern char except_vec_vi_mori;
-               const int mori_offset = &except_vec_vi_mori - &except_vec_vi;
+               const int mori_offset = &except_vec_vi_mori - vec_start;
 #endif /* CONFIG_MIPS_MT_SMTC */
-               const int handler_len = &except_vec_vi_end - &except_vec_vi;
-               const int lui_offset = &except_vec_vi_lui - &except_vec_vi;
-               const int ori_offset = &except_vec_vi_ori - &except_vec_vi;
+               const int handler_len = &except_vec_vi_end - vec_start;
+               const int lui_offset = &except_vec_vi_lui - vec_start;
+               const int ori_offset = &except_vec_vi_ori - vec_start;
 
                if (handler_len > VECTORSPACING) {
                        /*
@@ -1219,7 +1331,7 @@ static void *set_vi_srs_handler(int n, vi_handler_t addr, int srs)
                        panic("VECTORSPACING too small");
                }
 
-               memcpy(b, &except_vec_vi, handler_len);
+               memcpy(b, vec_start, handler_len);
 #ifdef CONFIG_MIPS_MT_SMTC
                BUG_ON(n > 7);  /* Vector index %d exceeds SMTC maximum. */
 
@@ -1230,7 +1342,8 @@ static void *set_vi_srs_handler(int n, vi_handler_t addr, int srs)
                *w = (*w & 0xffff0000) | (((u32)handler >> 16) & 0xffff);
                w = (u32 *)(b + ori_offset);
                *w = (*w & 0xffff0000) | ((u32)handler & 0xffff);
-               flush_icache_range((unsigned long)b, (unsigned long)(b+handler_len));
+               local_flush_icache_range((unsigned long)b,
+                                        (unsigned long)(b+handler_len));
        }
        else {
                /*
@@ -1242,7 +1355,8 @@ static void *set_vi_srs_handler(int n, vi_handler_t addr, int srs)
                w = (u32 *)b;
                *w++ = 0x08000000 | (((u32)handler >> 2) & 0x03fffff); /* j handler */
                *w = 0;
-               flush_icache_range((unsigned long)b, (unsigned long)(b+8));
+               local_flush_icache_range((unsigned long)b,
+                                        (unsigned long)(b+8));
        }
 
        return (void *)old_handler;
@@ -1253,14 +1367,6 @@ void *set_vi_handler(int n, vi_handler_t addr)
        return set_vi_srs_handler(n, addr, 0);
 }
 
-#else
-
-static inline void mips_srs_init(void)
-{
-}
-
-#endif /* CONFIG_CPU_MIPSR2_SRS */
-
 /*
  * This is used by native signal handling
  */
@@ -1347,7 +1453,18 @@ int cp0_compare_irq;
 int cp0_perfcount_irq;
 EXPORT_SYMBOL_GPL(cp0_perfcount_irq);
 
-void __init per_cpu_trap_init(void)
+static int __cpuinitdata noulri;
+
+static int __init ulri_disable(char *s)
+{
+       pr_info("Disabling ulri\n");
+       noulri = 1;
+
+       return 1;
+}
+__setup("noulri", ulri_disable);
+
+void __cpuinit per_cpu_trap_init(void)
 {
        unsigned int cpu = smp_processor_id();
        unsigned int status_set = ST0_CU0;
@@ -1377,29 +1494,29 @@ void __init per_cpu_trap_init(void)
 #endif
        if (current_cpu_data.isa_level == MIPS_CPU_ISA_IV)
                status_set |= ST0_XX;
+       if (cpu_has_dsp)
+               status_set |= ST0_MX;
+
        change_c0_status(ST0_CU|ST0_MX|ST0_RE|ST0_FR|ST0_BEV|ST0_TS|ST0_KX|ST0_SX|ST0_UX,
                         status_set);
 
-       if (cpu_has_dsp)
-               set_c0_status(ST0_MX);
-
-#ifdef CONFIG_CPU_MIPSR2
        if (cpu_has_mips_r2) {
-               unsigned int enable = 0x0000000f;
+               unsigned int enable = 0x0000000f | cpu_hwrena_impl_bits;
 
-               if (cpu_has_userlocal)
+               if (!noulri && cpu_has_userlocal)
                        enable |= (1 << 29);
 
                write_c0_hwrena(enable);
        }
-#endif
 
 #ifdef CONFIG_MIPS_MT_SMTC
        if (!secondaryTC) {
 #endif /* CONFIG_MIPS_MT_SMTC */
 
        if (cpu_has_veic || cpu_has_vint) {
+               unsigned long sr = set_c0_status(ST0_BEV);
                write_c0_ebase(ebase);
+               write_c0_status(sr);
                /* Setting vector spacing enables EI/VI mode  */
                change_c0_intctl(0x3e0, VECTORSPACING);
        }
@@ -1461,14 +1578,19 @@ void __init per_cpu_trap_init(void)
 void __init set_handler(unsigned long offset, void *addr, unsigned long size)
 {
        memcpy((void *)(ebase + offset), addr, size);
-       flush_icache_range(ebase + offset, ebase + offset + size);
+       local_flush_icache_range(ebase + offset, ebase + offset + size);
 }
 
-static char panic_null_cerr[] __initdata =
+static char panic_null_cerr[] __cpuinitdata =
        "Trying to set NULL cache error exception handler";
 
-/* Install uncached CPU exception handler */
-void __init set_uncached_handler(unsigned long offset, void *addr, unsigned long size)
+/*
+ * Install uncached CPU exception handler.
+ * This is suitable only for the cache error exception which is the only
+ * exception handler that is being run uncached.
+ */
+void __cpuinit set_uncached_handler(unsigned long offset, void *addr,
+       unsigned long size)
 {
 #ifdef CONFIG_32BIT
        unsigned long uncached_ebase = KSEG1ADDR(ebase);
@@ -1497,13 +1619,25 @@ void __init trap_init(void)
        extern char except_vec3_generic, except_vec3_r4000;
        extern char except_vec4;
        unsigned long i;
+       int rollback;
 
-       if (cpu_has_veic || cpu_has_vint)
-               ebase = (unsigned long) alloc_bootmem_low_pages(0x200 + VECTORSPACING*64);
-       else
-               ebase = CAC_BASE;
+       check_wait();
+       rollback = (cpu_wait == r4k_wait);
 
-       mips_srs_init();
+#if defined(CONFIG_KGDB)
+       if (kgdb_early_setup)
+               return; /* Already done */
+#endif
+
+       if (cpu_has_veic || cpu_has_vint) {
+               unsigned long size = 0x200 + VECTORSPACING*64;
+               ebase = (unsigned long)
+                       __alloc_bootmem(size, 1 << fls(size), 0);
+       } else {
+               ebase = CAC_BASE;
+               if (cpu_has_mips_r2)
+                       ebase += (read_c0_ebase() & 0x3ffff000);
+       }
 
        per_cpu_trap_init();
 
@@ -1558,7 +1692,7 @@ void __init trap_init(void)
        if (board_be_init)
                board_be_init();
 
-       set_except_vector(0, handle_int);
+       set_except_vector(0, rollback ? rollback_handle_int : handle_int);
        set_except_vector(1, handle_tlbm);
        set_except_vector(2, handle_tlbl);
        set_except_vector(3, handle_tlbs);
@@ -1611,17 +1745,19 @@ void __init trap_init(void)
 
        if (cpu_has_vce)
                /* Special exception: R4[04]00 uses also the divec space. */
-               memcpy((void *)(CAC_BASE + 0x180), &except_vec3_r4000, 0x100);
+               memcpy((void *)(ebase + 0x180), &except_vec3_r4000, 0x100);
        else if (cpu_has_4kex)
-               memcpy((void *)(CAC_BASE + 0x180), &except_vec3_generic, 0x80);
+               memcpy((void *)(ebase + 0x180), &except_vec3_generic, 0x80);
        else
-               memcpy((void *)(CAC_BASE + 0x080), &except_vec3_generic, 0x80);
+               memcpy((void *)(ebase + 0x080), &except_vec3_generic, 0x80);
 
        signal_init();
 #ifdef CONFIG_MIPS32_COMPAT
        signal32_init();
 #endif
 
-       flush_icache_range(ebase, ebase + 0x400);
+       local_flush_icache_range(ebase, ebase + 0x400);
        flush_tlb_handlers();
+
+       sort_extable(__start___dbe_table, __stop___dbe_table);
 }