metag: Traps
authorJames Hogan <james.hogan@imgtec.com>
Tue, 9 Oct 2012 09:54:43 +0000 (10:54 +0100)
committerJames Hogan <james.hogan@imgtec.com>
Sat, 2 Mar 2013 20:09:45 +0000 (20:09 +0000)
Add trap code for metag. At the lowest level Meta traps (and return from
interrupt instruction - RTI) simply swap the PC and PCX registers and
optionally toggle the interrupt status bit (ISTAT). Low level TBX code
in tbipcx.S handles the core context save, determine the TBX signal
number based on the core trigger that fired (using the TXSTATI status
register), and call TBX signal handlers (mostly in traps.c) via a vector
table.

Signed-off-by: James Hogan <james.hogan@imgtec.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
arch/metag/include/asm/switch.h [new file with mode: 0644]
arch/metag/include/asm/traps.h [new file with mode: 0644]
arch/metag/kernel/kick.c [new file with mode: 0644]
arch/metag/kernel/tbiunexp.S [new file with mode: 0644]
arch/metag/kernel/traps.c [new file with mode: 0644]

diff --git a/arch/metag/include/asm/switch.h b/arch/metag/include/asm/switch.h
new file mode 100644 (file)
index 0000000..1fd6a58
--- /dev/null
@@ -0,0 +1,21 @@
+/*
+ * Copyright (C) 2012 Imagination Technologies Ltd.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; either version 2 of the License, or
+ * (at your option) any later version.
+ */
+
+#ifndef _ASM_METAG_SWITCH_H
+#define _ASM_METAG_SWITCH_H
+
+/* metag SWITCH codes */
+#define __METAG_SW_PERM_BREAK  0x400002        /* compiled in breakpoint */
+#define __METAG_SW_SYS_LEGACY  0x440000        /* legacy system calls */
+#define __METAG_SW_SYS         0x440001        /* system calls */
+
+/* metag SWITCH instruction encoding */
+#define __METAG_SW_ENCODING(TYPE)      (0xaf000000 | (__METAG_SW_##TYPE))
+
+#endif /* _ASM_METAG_SWITCH_H */
diff --git a/arch/metag/include/asm/traps.h b/arch/metag/include/asm/traps.h
new file mode 100644 (file)
index 0000000..ac80874
--- /dev/null
@@ -0,0 +1,48 @@
+/*
+ *  Copyright (C) 2005,2008 Imagination Technologies
+ *
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file COPYING in the main directory of this archive
+ * for more details.
+ */
+
+#ifndef _METAG_TBIVECTORS_H
+#define _METAG_TBIVECTORS_H
+
+#ifndef __ASSEMBLY__
+
+#include <asm/tbx.h>
+
+typedef TBIRES (*kick_irq_func_t)(TBIRES, int, int, int, PTBI, int *);
+
+extern TBIRES kick_handler(TBIRES, int, int, int, PTBI);
+struct kick_irq_handler {
+       struct list_head list;
+       kick_irq_func_t func;
+};
+
+extern void kick_register_func(struct kick_irq_handler *);
+extern void kick_unregister_func(struct kick_irq_handler *);
+
+extern void head_end(TBIRES, unsigned long);
+extern void restart_critical_section(TBIRES State);
+extern TBIRES tail_end_sys(TBIRES, int, int *);
+static inline TBIRES tail_end(TBIRES state)
+{
+       return tail_end_sys(state, -1, NULL);
+}
+
+DECLARE_PER_CPU(PTBI, pTBI);
+extern PTBI pTBI_get(unsigned int);
+
+extern int ret_from_fork(TBIRES arg);
+
+extern int do_page_fault(struct pt_regs *regs, unsigned long address,
+                        unsigned int write_access, unsigned int trapno);
+
+extern TBIRES __TBIUnExpXXX(TBIRES State, int SigNum, int Triggers, int Inst,
+                           PTBI pTBI);
+
+#endif
+
+#endif /* _METAG_TBIVECTORS_H */
diff --git a/arch/metag/kernel/kick.c b/arch/metag/kernel/kick.c
new file mode 100644 (file)
index 0000000..c309096
--- /dev/null
@@ -0,0 +1,98 @@
+/*
+ *  Copyright (C) 2009 Imagination Technologies
+ *
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file COPYING in the main directory of this archive
+ * for more details.
+ *
+ * The Meta KICK interrupt mechanism is generally a useful feature, so
+ * we provide an interface for registering multiple interrupt
+ * handlers. All the registered interrupt handlers are "chained". When
+ * a KICK interrupt is received the first function in the list is
+ * called. If that interrupt handler cannot handle the KICK the next
+ * one is called, then the next until someone handles it (or we run
+ * out of functions). As soon as one function handles the interrupt no
+ * other handlers are called.
+ *
+ * The only downside of chaining interrupt handlers is that each
+ * handler must be able to detect whether the KICK was intended for it
+ * or not.  For example, when the IPI handler runs and it sees that
+ * there are no IPI messages it must not signal that the KICK was
+ * handled, thereby giving the other handlers a chance to run.
+ *
+ * The reason that we provide our own interface for calling KICK
+ * handlers instead of using the generic kernel infrastructure is that
+ * the KICK handlers require access to a CPU's pTBI structure. So we
+ * pass it as an argument.
+ */
+#include <linux/kernel.h>
+#include <linux/mm.h>
+#include <linux/types.h>
+
+#include <asm/traps.h>
+
+/*
+ * All accesses/manipulations of kick_handlers_list should be
+ * performed while holding kick_handlers_lock.
+ */
+static DEFINE_SPINLOCK(kick_handlers_lock);
+static LIST_HEAD(kick_handlers_list);
+
+void kick_register_func(struct kick_irq_handler *kh)
+{
+       unsigned long flags;
+
+       spin_lock_irqsave(&kick_handlers_lock, flags);
+
+       list_add_tail(&kh->list, &kick_handlers_list);
+
+       spin_unlock_irqrestore(&kick_handlers_lock, flags);
+}
+
+void kick_unregister_func(struct kick_irq_handler *kh)
+{
+       unsigned long flags;
+
+       spin_lock_irqsave(&kick_handlers_lock, flags);
+
+       list_del(&kh->list);
+
+       spin_unlock_irqrestore(&kick_handlers_lock, flags);
+}
+
+TBIRES
+kick_handler(TBIRES State, int SigNum, int Triggers, int Inst, PTBI pTBI)
+{
+       struct kick_irq_handler *kh;
+       struct list_head *lh;
+       int handled = 0;
+       TBIRES ret;
+
+       head_end(State, ~INTS_OFF_MASK);
+
+       /* If we interrupted user code handle any critical sections. */
+       if (State.Sig.SaveMask & TBICTX_PRIV_BIT)
+               restart_critical_section(State);
+
+       trace_hardirqs_off();
+
+       /*
+        * There is no need to disable interrupts here because we
+        * can't nest KICK interrupts in a KICK interrupt handler.
+        */
+       spin_lock(&kick_handlers_lock);
+
+       list_for_each(lh, &kick_handlers_list) {
+               kh = list_entry(lh, struct kick_irq_handler, list);
+
+               ret = kh->func(State, SigNum, Triggers, Inst, pTBI, &handled);
+               if (handled)
+                       break;
+       }
+
+       spin_unlock(&kick_handlers_lock);
+
+       WARN_ON(!handled);
+
+       return tail_end(ret);
+}
diff --git a/arch/metag/kernel/tbiunexp.S b/arch/metag/kernel/tbiunexp.S
new file mode 100644 (file)
index 0000000..907bbe0
--- /dev/null
@@ -0,0 +1,22 @@
+/* Pass a breakpoint through to Codescape */
+
+#include <asm/tbx.h>
+
+       .text
+        .global        ___TBIUnExpXXX
+        .type   ___TBIUnExpXXX,function
+___TBIUnExpXXX:
+       TSTT    D0Ar2,#TBICTX_CRIT_BIT  ! Result of nestable int call?
+       BZ      $LTBINormCase           ! UnExpXXX at background level
+       MOV     D0Re0,TXMASKI           ! Read TXMASKI
+       XOR     TXMASKI,D1Re0,D1Re0     ! Turn off BGNDHALT handling!
+       OR      D0Ar2,D0Ar2,D0Re0       ! Preserve bits cleared
+$LTBINormCase:
+       MSETL   [A0StP],D0Ar6,D0Ar4,D0Ar2       ! Save args on stack
+       SETL    [A0StP++],D0Ar2,D1Ar1   ! Init area for returned values
+       SWITCH  #0xC20208               ! Total stack frame size 8 Dwords
+                                       !            write back size 2 Dwords
+       GETL    D0Re0,D1Re0,[--A0StP]   ! Get result
+       SUB     A0StP,A0StP,#(8*3)      ! Recover stack frame
+       MOV     PC,D1RtP
+        .size          ___TBIUnExpXXX,.-___TBIUnExpXXX
diff --git a/arch/metag/kernel/traps.c b/arch/metag/kernel/traps.c
new file mode 100644 (file)
index 0000000..1ad363c
--- /dev/null
@@ -0,0 +1,978 @@
+/*
+ *  Meta exception handling.
+ *
+ *  Copyright (C) 2005,2006,2007,2008,2009,2012 Imagination Technologies Ltd.
+ *
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file COPYING in the main directory of this archive
+ * for more details.
+ */
+
+#include <linux/sched.h>
+#include <linux/signal.h>
+#include <linux/kernel.h>
+#include <linux/mm.h>
+#include <linux/types.h>
+#include <linux/init.h>
+#include <linux/interrupt.h>
+#include <linux/ptrace.h>
+#include <linux/module.h>
+#include <linux/kallsyms.h>
+#include <linux/kdebug.h>
+#include <linux/kexec.h>
+#include <linux/unistd.h>
+#include <linux/smp.h>
+#include <linux/slab.h>
+#include <linux/syscalls.h>
+
+#include <asm/bug.h>
+#include <asm/core_reg.h>
+#include <asm/irqflags.h>
+#include <asm/siginfo.h>
+#include <asm/traps.h>
+#include <asm/hwthread.h>
+#include <asm/switch.h>
+#include <asm/user_gateway.h>
+#include <asm/syscall.h>
+#include <asm/syscalls.h>
+
+/* Passing syscall arguments as long long is quicker. */
+typedef unsigned int (*LPSYSCALL) (unsigned long long,
+                                  unsigned long long,
+                                  unsigned long long);
+
+/*
+ * Users of LNKSET should compare the bus error bits obtained from DEFR
+ * against TXDEFR_LNKSET_SUCCESS only as the failure code will vary between
+ * different cores revisions.
+ */
+#define TXDEFR_LNKSET_SUCCESS 0x02000000
+#define TXDEFR_LNKSET_FAILURE 0x04000000
+
+/*
+ * Our global TBI handle.  Initialised from setup.c/setup_arch.
+ */
+DECLARE_PER_CPU(PTBI, pTBI);
+
+#ifdef CONFIG_SMP
+static DEFINE_PER_CPU(unsigned int, trigger_mask);
+#else
+unsigned int global_trigger_mask;
+#endif
+
+unsigned long per_cpu__stack_save[NR_CPUS];
+
+static const char * const trap_names[] = {
+       [TBIXXF_SIGNUM_IIF] = "Illegal instruction fault",
+       [TBIXXF_SIGNUM_PGF] = "Privilege violation",
+       [TBIXXF_SIGNUM_DHF] = "Unaligned data access fault",
+       [TBIXXF_SIGNUM_IGF] = "Code fetch general read failure",
+       [TBIXXF_SIGNUM_DGF] = "Data access general read/write fault",
+       [TBIXXF_SIGNUM_IPF] = "Code fetch page fault",
+       [TBIXXF_SIGNUM_DPF] = "Data access page fault",
+       [TBIXXF_SIGNUM_IHF] = "Instruction breakpoint",
+       [TBIXXF_SIGNUM_DWF] = "Read-only data access fault",
+};
+
+const char *trap_name(int trapno)
+{
+       if (trapno >= 0 && trapno < ARRAY_SIZE(trap_names)
+                       && trap_names[trapno])
+               return trap_names[trapno];
+       return "Unknown fault";
+}
+
+static DEFINE_SPINLOCK(die_lock);
+
+void die(const char *str, struct pt_regs *regs, long err,
+        unsigned long addr)
+{
+       static int die_counter;
+
+       oops_enter();
+
+       spin_lock_irq(&die_lock);
+       console_verbose();
+       bust_spinlocks(1);
+       pr_err("%s: err %04lx (%s) addr %08lx [#%d]\n", str, err & 0xffff,
+              trap_name(err & 0xffff), addr, ++die_counter);
+
+       print_modules();
+       show_regs(regs);
+
+       pr_err("Process: %s (pid: %d, stack limit = %p)\n", current->comm,
+              task_pid_nr(current), task_stack_page(current) + THREAD_SIZE);
+
+       bust_spinlocks(0);
+       add_taint(TAINT_DIE);
+       if (kexec_should_crash(current))
+               crash_kexec(regs);
+
+       if (in_interrupt())
+               panic("Fatal exception in interrupt");
+
+       if (panic_on_oops)
+               panic("Fatal exception");
+
+       spin_unlock_irq(&die_lock);
+       oops_exit();
+       do_exit(SIGSEGV);
+}
+
+#ifdef CONFIG_METAG_DSP
+/*
+ * The ECH encoding specifies the size of a DSPRAM as,
+ *
+ *             "slots" / 4
+ *
+ * A "slot" is the size of two DSPRAM bank entries; an entry from
+ * DSPRAM bank A and an entry from DSPRAM bank B. One DSPRAM bank
+ * entry is 4 bytes.
+ */
+#define SLOT_SZ        8
+static inline unsigned int decode_dspram_size(unsigned int size)
+{
+       unsigned int _sz = size & 0x7f;
+
+       return _sz * SLOT_SZ * 4;
+}
+
+static void dspram_save(struct meta_ext_context *dsp_ctx,
+                       unsigned int ramA_sz, unsigned int ramB_sz)
+{
+       unsigned int ram_sz[2];
+       int i;
+
+       ram_sz[0] = ramA_sz;
+       ram_sz[1] = ramB_sz;
+
+       for (i = 0; i < 2; i++) {
+               if (ram_sz[i] != 0) {
+                       unsigned int sz;
+
+                       if (i == 0)
+                               sz = decode_dspram_size(ram_sz[i] >> 8);
+                       else
+                               sz = decode_dspram_size(ram_sz[i]);
+
+                       if (dsp_ctx->ram[i] == NULL) {
+                               dsp_ctx->ram[i] = kmalloc(sz, GFP_KERNEL);
+
+                               if (dsp_ctx->ram[i] == NULL)
+                                       panic("couldn't save DSP context");
+                       } else {
+                               if (ram_sz[i] > dsp_ctx->ram_sz[i]) {
+                                       kfree(dsp_ctx->ram[i]);
+
+                                       dsp_ctx->ram[i] = kmalloc(sz,
+                                                                 GFP_KERNEL);
+
+                                       if (dsp_ctx->ram[i] == NULL)
+                                               panic("couldn't save DSP context");
+                               }
+                       }
+
+                       if (i == 0)
+                               __TBIDspramSaveA(ram_sz[i], dsp_ctx->ram[i]);
+                       else
+                               __TBIDspramSaveB(ram_sz[i], dsp_ctx->ram[i]);
+
+                       dsp_ctx->ram_sz[i] = ram_sz[i];
+               }
+       }
+}
+#endif /* CONFIG_METAG_DSP */
+
+/*
+ * Allow interrupts to be nested and save any "extended" register
+ * context state, e.g. DSP regs and RAMs.
+ */
+static void nest_interrupts(TBIRES State, unsigned long mask)
+{
+#ifdef CONFIG_METAG_DSP
+       struct meta_ext_context *dsp_ctx;
+       unsigned int D0_8;
+
+       /*
+        * D0.8 may contain an ECH encoding. The upper 16 bits
+        * tell us what DSP resources the current process is
+        * using. OR the bits into the SaveMask so that
+        * __TBINestInts() knows what resources to save as
+        * part of this context.
+        *
+        * Don't save the context if we're nesting interrupts in the
+        * kernel because the kernel doesn't use DSP hardware.
+        */
+       D0_8 = __core_reg_get(D0.8);
+
+       if (D0_8 && (State.Sig.SaveMask & TBICTX_PRIV_BIT)) {
+               State.Sig.SaveMask |= (D0_8 >> 16);
+
+               dsp_ctx = current->thread.dsp_context;
+               if (dsp_ctx == NULL) {
+                       dsp_ctx = kzalloc(sizeof(*dsp_ctx), GFP_KERNEL);
+                       if (dsp_ctx == NULL)
+                               panic("couldn't save DSP context: ENOMEM");
+
+                       current->thread.dsp_context = dsp_ctx;
+               }
+
+               current->thread.user_flags |= (D0_8 & 0xffff0000);
+               __TBINestInts(State, &dsp_ctx->regs, mask);
+               dspram_save(dsp_ctx, D0_8 & 0x7f00, D0_8 & 0x007f);
+       } else
+               __TBINestInts(State, NULL, mask);
+#else
+       __TBINestInts(State, NULL, mask);
+#endif
+}
+
+void head_end(TBIRES State, unsigned long mask)
+{
+       unsigned int savemask = (unsigned short)State.Sig.SaveMask;
+       unsigned int ctx_savemask = (unsigned short)State.Sig.pCtx->SaveMask;
+
+       if (savemask & TBICTX_PRIV_BIT) {
+               ctx_savemask |= TBICTX_PRIV_BIT;
+               current->thread.user_flags = savemask;
+       }
+
+       /* Always undo the sleep bit */
+       ctx_savemask &= ~TBICTX_WAIT_BIT;
+
+       /* Always save the catch buffer and RD pipe if they are dirty */
+       savemask |= TBICTX_XCBF_BIT;
+
+       /* Only save the catch and RD if we have not already done so.
+        * Note - the RD bits are in the pCtx only, and not in the
+        * State.SaveMask.
+        */
+       if ((savemask & TBICTX_CBUF_BIT) ||
+           (ctx_savemask & TBICTX_CBRP_BIT)) {
+               /* Have we already saved the buffers though?
+                * - See TestTrack 5071 */
+               if (ctx_savemask & TBICTX_XCBF_BIT) {
+                       /* Strip off the bits so the call to __TBINestInts
+                        * won't save the buffers again. */
+                       savemask &= ~TBICTX_CBUF_BIT;
+                       ctx_savemask &= ~TBICTX_CBRP_BIT;
+               }
+       }
+
+#ifdef CONFIG_METAG_META21
+       {
+               unsigned int depth, txdefr;
+
+               /*
+                * Save TXDEFR state.
+                *
+                * The process may have been interrupted after a LNKSET, but
+                * before it could read the DEFR state, so we mustn't lose that
+                * state or it could end up retrying an atomic operation that
+                * succeeded.
+                *
+                * All interrupts are disabled at this point so we
+                * don't need to perform any locking. We must do this
+                * dance before we use LNKGET or LNKSET.
+                */
+               BUG_ON(current->thread.int_depth > HARDIRQ_BITS);
+
+               depth = current->thread.int_depth++;
+
+               txdefr = __core_reg_get(TXDEFR);
+
+               txdefr &= TXDEFR_BUS_STATE_BITS;
+               if (txdefr & TXDEFR_LNKSET_SUCCESS)
+                       current->thread.txdefr_failure &= ~(1 << depth);
+               else
+                       current->thread.txdefr_failure |= (1 << depth);
+       }
+#endif
+
+       State.Sig.SaveMask = savemask;
+       State.Sig.pCtx->SaveMask = ctx_savemask;
+
+       nest_interrupts(State, mask);
+
+#ifdef CONFIG_METAG_POISON_CATCH_BUFFERS
+       /* Poison the catch registers.  This shows up any mistakes we have
+        * made in their handling MUCH quicker.
+        */
+       __core_reg_set(TXCATCH0, 0x87650021);
+       __core_reg_set(TXCATCH1, 0x87654322);
+       __core_reg_set(TXCATCH2, 0x87654323);
+       __core_reg_set(TXCATCH3, 0x87654324);
+#endif /* CONFIG_METAG_POISON_CATCH_BUFFERS */
+}
+
+TBIRES tail_end_sys(TBIRES State, int syscall, int *restart)
+{
+       struct pt_regs *regs = (struct pt_regs *)State.Sig.pCtx;
+       unsigned long flags;
+
+       local_irq_disable();
+
+       if (user_mode(regs)) {
+               flags = current_thread_info()->flags;
+               if (flags & _TIF_WORK_MASK &&
+                   do_work_pending(regs, flags, syscall)) {
+                       *restart = 1;
+                       return State;
+               }
+
+#ifdef CONFIG_METAG_FPU
+               if (current->thread.fpu_context &&
+                   current->thread.fpu_context->needs_restore) {
+                       __TBICtxFPURestore(State, current->thread.fpu_context);
+                       /*
+                        * Clearing this bit ensures the FP unit is not made
+                        * active again unless it is used.
+                        */
+                       State.Sig.SaveMask &= ~TBICTX_FPAC_BIT;
+                       current->thread.fpu_context->needs_restore = false;
+               }
+               State.Sig.TrigMask |= TBI_TRIG_BIT(TBID_SIGNUM_DFR);
+#endif
+       }
+
+       /* TBI will turn interrupts back on at some point. */
+       if (!irqs_disabled_flags((unsigned long)State.Sig.TrigMask))
+               trace_hardirqs_on();
+
+#ifdef CONFIG_METAG_DSP
+       /*
+        * If we previously saved an extended context then restore it
+        * now. Otherwise, clear D0.8 because this process is not
+        * using DSP hardware.
+        */
+       if (State.Sig.pCtx->SaveMask & TBICTX_XEXT_BIT) {
+               unsigned int D0_8;
+               struct meta_ext_context *dsp_ctx = current->thread.dsp_context;
+
+               /* Make sure we're going to return to userland. */
+               BUG_ON(current->thread.int_depth != 1);
+
+               if (dsp_ctx->ram_sz[0] > 0)
+                       __TBIDspramRestoreA(dsp_ctx->ram_sz[0],
+                                           dsp_ctx->ram[0]);
+               if (dsp_ctx->ram_sz[1] > 0)
+                       __TBIDspramRestoreB(dsp_ctx->ram_sz[1],
+                                           dsp_ctx->ram[1]);
+
+               State.Sig.SaveMask |= State.Sig.pCtx->SaveMask;
+               __TBICtxRestore(State, current->thread.dsp_context);
+               D0_8 = __core_reg_get(D0.8);
+               D0_8 |= current->thread.user_flags & 0xffff0000;
+               D0_8 |= (dsp_ctx->ram_sz[1] | dsp_ctx->ram_sz[0]) & 0xffff;
+               __core_reg_set(D0.8, D0_8);
+       } else
+               __core_reg_set(D0.8, 0);
+#endif /* CONFIG_METAG_DSP */
+
+#ifdef CONFIG_METAG_META21
+       {
+               unsigned int depth, txdefr;
+
+               /*
+                * If there hasn't been a LNKSET since the last LNKGET then the
+                * link flag will be set, causing the next LNKSET to succeed if
+                * the addresses match. The two LNK operations may not be a pair
+                * (e.g. see atomic_read()), so the LNKSET should fail.
+                * We use a conditional-never LNKSET to clear the link flag
+                * without side effects.
+                */
+               asm volatile("LNKSETDNV [D0Re0],D0Re0");
+
+               depth = --current->thread.int_depth;
+
+               BUG_ON(user_mode(regs) && depth);
+
+               txdefr = __core_reg_get(TXDEFR);
+
+               txdefr &= ~TXDEFR_BUS_STATE_BITS;
+
+               /* Do we need to restore a failure code into TXDEFR? */
+               if (current->thread.txdefr_failure & (1 << depth))
+                       txdefr |= (TXDEFR_LNKSET_FAILURE | TXDEFR_BUS_TRIG_BIT);
+               else
+                       txdefr |= (TXDEFR_LNKSET_SUCCESS | TXDEFR_BUS_TRIG_BIT);
+
+               __core_reg_set(TXDEFR, txdefr);
+       }
+#endif
+       return State;
+}
+
+#ifdef CONFIG_SMP
+/*
+ * If we took an interrupt in the middle of __kuser_get_tls then we need
+ * to rewind the PC to the start of the function in case the process
+ * gets migrated to another thread (SMP only) and it reads the wrong tls
+ * data.
+ */
+static inline void _restart_critical_section(TBIRES State)
+{
+       unsigned long get_tls_start;
+       unsigned long get_tls_end;
+
+       get_tls_start = (unsigned long)__kuser_get_tls -
+               (unsigned long)&__user_gateway_start;
+
+       get_tls_start += USER_GATEWAY_PAGE;
+
+       get_tls_end = (unsigned long)__kuser_get_tls_end -
+               (unsigned long)&__user_gateway_start;
+
+       get_tls_end += USER_GATEWAY_PAGE;
+
+       if ((State.Sig.pCtx->CurrPC >= get_tls_start) &&
+           (State.Sig.pCtx->CurrPC < get_tls_end))
+               State.Sig.pCtx->CurrPC = get_tls_start;
+}
+#else
+/*
+ * If we took an interrupt in the middle of
+ * __kuser_cmpxchg then we need to rewind the PC to the
+ * start of the function.
+ */
+static inline void _restart_critical_section(TBIRES State)
+{
+       unsigned long cmpxchg_start;
+       unsigned long cmpxchg_end;
+
+       cmpxchg_start = (unsigned long)__kuser_cmpxchg -
+               (unsigned long)&__user_gateway_start;
+
+       cmpxchg_start += USER_GATEWAY_PAGE;
+
+       cmpxchg_end = (unsigned long)__kuser_cmpxchg_end -
+               (unsigned long)&__user_gateway_start;
+
+       cmpxchg_end += USER_GATEWAY_PAGE;
+
+       if ((State.Sig.pCtx->CurrPC >= cmpxchg_start) &&
+           (State.Sig.pCtx->CurrPC < cmpxchg_end))
+               State.Sig.pCtx->CurrPC = cmpxchg_start;
+}
+#endif
+
+/* Used by kick_handler() */
+void restart_critical_section(TBIRES State)
+{
+       _restart_critical_section(State);
+}
+
+TBIRES trigger_handler(TBIRES State, int SigNum, int Triggers, int Inst,
+                      PTBI pTBI)
+{
+       head_end(State, ~INTS_OFF_MASK);
+
+       /* If we interrupted user code handle any critical sections. */
+       if (State.Sig.SaveMask & TBICTX_PRIV_BIT)
+               _restart_critical_section(State);
+
+       trace_hardirqs_off();
+
+       do_IRQ(SigNum, (struct pt_regs *)State.Sig.pCtx);
+
+       return tail_end(State);
+}
+
+static unsigned int load_fault(PTBICTXEXTCB0 pbuf)
+{
+       return pbuf->CBFlags & TXCATCH0_READ_BIT;
+}
+
+static unsigned long fault_address(PTBICTXEXTCB0 pbuf)
+{
+       return pbuf->CBAddr;
+}
+
+static void unhandled_fault(struct pt_regs *regs, unsigned long addr,
+                           int signo, int code, int trapno)
+{
+       if (user_mode(regs)) {
+               siginfo_t info;
+
+               if (show_unhandled_signals && unhandled_signal(current, signo)
+                   && printk_ratelimit()) {
+
+                       pr_info("pid %d unhandled fault: pc 0x%08x, addr 0x%08lx, trap %d (%s)\n",
+                               current->pid, regs->ctx.CurrPC, addr,
+                               trapno, trap_name(trapno));
+                       print_vma_addr(" in ", regs->ctx.CurrPC);
+                       print_vma_addr(" rtp in ", regs->ctx.DX[4].U1);
+                       printk("\n");
+                       show_regs(regs);
+               }
+
+               info.si_signo = signo;
+               info.si_errno = 0;
+               info.si_code = code;
+               info.si_addr = (__force void __user *)addr;
+               info.si_trapno = trapno;
+               force_sig_info(signo, &info, current);
+       } else {
+               die("Oops", regs, trapno, addr);
+       }
+}
+
+static int handle_data_fault(PTBICTXEXTCB0 pcbuf, struct pt_regs *regs,
+                            unsigned int data_address, int trapno)
+{
+       int ret;
+
+       ret = do_page_fault(regs, data_address, !load_fault(pcbuf), trapno);
+
+       return ret;
+}
+
+static unsigned long get_inst_fault_address(struct pt_regs *regs)
+{
+       return regs->ctx.CurrPC;
+}
+
+TBIRES fault_handler(TBIRES State, int SigNum, int Triggers,
+                    int Inst, PTBI pTBI)
+{
+       struct pt_regs *regs = (struct pt_regs *)State.Sig.pCtx;
+       PTBICTXEXTCB0 pcbuf = (PTBICTXEXTCB0)&regs->extcb0;
+       unsigned long data_address;
+
+       head_end(State, ~INTS_OFF_MASK);
+
+       /* Hardware breakpoint or data watch */
+       if ((SigNum == TBIXXF_SIGNUM_IHF) ||
+           ((SigNum == TBIXXF_SIGNUM_DHF) &&
+            (pcbuf[0].CBFlags & (TXCATCH0_WATCH1_BIT |
+                                 TXCATCH0_WATCH0_BIT)))) {
+               State = __TBIUnExpXXX(State, SigNum, Triggers, Inst,
+                                     pTBI);
+               return tail_end(State);
+       }
+
+       local_irq_enable();
+
+       data_address = fault_address(pcbuf);
+
+       switch (SigNum) {
+       case TBIXXF_SIGNUM_IGF:
+               /* 1st-level entry invalid (instruction fetch) */
+       case TBIXXF_SIGNUM_IPF: {
+               /* 2nd-level entry invalid (instruction fetch) */
+               unsigned long addr = get_inst_fault_address(regs);
+               do_page_fault(regs, addr, 0, SigNum);
+               break;
+       }
+
+       case TBIXXF_SIGNUM_DGF:
+               /* 1st-level entry invalid (data access) */
+       case TBIXXF_SIGNUM_DPF:
+               /* 2nd-level entry invalid (data access) */
+       case TBIXXF_SIGNUM_DWF:
+               /* Write to read only page */
+               handle_data_fault(pcbuf, regs, data_address, SigNum);
+               break;
+
+       case TBIXXF_SIGNUM_IIF:
+               /* Illegal instruction */
+               unhandled_fault(regs, regs->ctx.CurrPC, SIGILL, ILL_ILLOPC,
+                               SigNum);
+               break;
+
+       case TBIXXF_SIGNUM_DHF:
+               /* Unaligned access */
+               unhandled_fault(regs, data_address, SIGBUS, BUS_ADRALN,
+                               SigNum);
+               break;
+       case TBIXXF_SIGNUM_PGF:
+               /* Privilege violation */
+               unhandled_fault(regs, data_address, SIGSEGV, SEGV_ACCERR,
+                               SigNum);
+               break;
+       default:
+               BUG();
+               break;
+       }
+
+       return tail_end(State);
+}
+
+static bool switch_is_syscall(unsigned int inst)
+{
+       return inst == __METAG_SW_ENCODING(SYS);
+}
+
+static bool switch_is_legacy_syscall(unsigned int inst)
+{
+       return inst == __METAG_SW_ENCODING(SYS_LEGACY);
+}
+
+static inline void step_over_switch(struct pt_regs *regs, unsigned int inst)
+{
+       regs->ctx.CurrPC += 4;
+}
+
+static inline int test_syscall_work(void)
+{
+       return current_thread_info()->flags & _TIF_WORK_SYSCALL_MASK;
+}
+
+TBIRES switch1_handler(TBIRES State, int SigNum, int Triggers,
+                      int Inst, PTBI pTBI)
+{
+       struct pt_regs *regs = (struct pt_regs *)State.Sig.pCtx;
+       unsigned int sysnumber;
+       unsigned long long a1_a2, a3_a4, a5_a6;
+       LPSYSCALL syscall_entry;
+       int restart;
+
+       head_end(State, ~INTS_OFF_MASK);
+
+       /*
+        * If this is not a syscall SWITCH it could be a breakpoint.
+        */
+       if (!switch_is_syscall(Inst)) {
+               /*
+                * Alert the user if they're trying to use legacy system
+                * calls. This suggests they need to update their C
+                * library and build against up to date kernel headers.
+                */
+               if (switch_is_legacy_syscall(Inst))
+                       pr_warn_once("WARNING: A legacy syscall was made. Your userland needs updating.\n");
+               /*
+                * We don't know how to handle the SWITCH and cannot
+                * safely ignore it, so treat all unknown switches
+                * (including breakpoints) as traps.
+                */
+               force_sig(SIGTRAP, current);
+               return tail_end(State);
+       }
+
+       local_irq_enable();
+
+restart_syscall:
+       restart = 0;
+       sysnumber = regs->ctx.DX[0].U1;
+
+       if (test_syscall_work())
+               sysnumber = syscall_trace_enter(regs);
+
+       /* Skip over the SWITCH instruction - or you just get 'stuck' on it! */
+       step_over_switch(regs, Inst);
+
+       if (sysnumber >= __NR_syscalls) {
+               pr_debug("unknown syscall number: %d\n", sysnumber);
+               syscall_entry = (LPSYSCALL) sys_ni_syscall;
+       } else {
+               syscall_entry = (LPSYSCALL) sys_call_table[sysnumber];
+       }
+
+       /* Use 64bit loads for speed. */
+       a5_a6 = *(unsigned long long *)&regs->ctx.DX[1];
+       a3_a4 = *(unsigned long long *)&regs->ctx.DX[2];
+       a1_a2 = *(unsigned long long *)&regs->ctx.DX[3];
+
+       /* here is the actual call to the syscall handler functions */
+       regs->ctx.DX[0].U0 = syscall_entry(a1_a2, a3_a4, a5_a6);
+
+       if (test_syscall_work())
+               syscall_trace_leave(regs);
+
+       State = tail_end_sys(State, sysnumber, &restart);
+       /* Handlerless restarts shouldn't go via userland */
+       if (restart)
+               goto restart_syscall;
+       return State;
+}
+
+TBIRES switchx_handler(TBIRES State, int SigNum, int Triggers,
+                      int Inst, PTBI pTBI)
+{
+       struct pt_regs *regs = (struct pt_regs *)State.Sig.pCtx;
+
+       /*
+        * This can be caused by any user process simply executing an unusual
+        * SWITCH instruction. If there's no DA, __TBIUnExpXXX will cause the
+        * thread to stop, so signal a SIGTRAP instead.
+        */
+       head_end(State, ~INTS_OFF_MASK);
+       if (user_mode(regs))
+               force_sig(SIGTRAP, current);
+       else
+               State = __TBIUnExpXXX(State, SigNum, Triggers, Inst, pTBI);
+       return tail_end(State);
+}
+
+#ifdef CONFIG_METAG_META21
+TBIRES fpe_handler(TBIRES State, int SigNum, int Triggers, int Inst, PTBI pTBI)
+{
+       struct pt_regs *regs = (struct pt_regs *)State.Sig.pCtx;
+       unsigned int error_state = Triggers;
+       siginfo_t info;
+
+       head_end(State, ~INTS_OFF_MASK);
+
+       local_irq_enable();
+
+       info.si_signo = SIGFPE;
+
+       if (error_state & TXSTAT_FPE_INVALID_BIT)
+               info.si_code = FPE_FLTINV;
+       else if (error_state & TXSTAT_FPE_DIVBYZERO_BIT)
+               info.si_code = FPE_FLTDIV;
+       else if (error_state & TXSTAT_FPE_OVERFLOW_BIT)
+               info.si_code = FPE_FLTOVF;
+       else if (error_state & TXSTAT_FPE_UNDERFLOW_BIT)
+               info.si_code = FPE_FLTUND;
+       else if (error_state & TXSTAT_FPE_INEXACT_BIT)
+               info.si_code = FPE_FLTRES;
+       else
+               info.si_code = 0;
+       info.si_errno = 0;
+       info.si_addr = (__force void __user *)regs->ctx.CurrPC;
+       force_sig_info(SIGFPE, &info, current);
+
+       return tail_end(State);
+}
+#endif
+
+#ifdef CONFIG_METAG_SUSPEND_MEM
+struct traps_context {
+       PTBIAPIFN fnSigs[TBID_SIGNUM_MAX + 1];
+};
+
+static struct traps_context *metag_traps_context;
+
+int traps_save_context(void)
+{
+       unsigned long cpu = smp_processor_id();
+       PTBI _pTBI = per_cpu(pTBI, cpu);
+       struct traps_context *context;
+
+       context = kzalloc(sizeof(*context), GFP_ATOMIC);
+       if (!context)
+               return -ENOMEM;
+
+       memcpy(context->fnSigs, (void *)_pTBI->fnSigs, sizeof(context->fnSigs));
+
+       metag_traps_context = context;
+       return 0;
+}
+
+int traps_restore_context(void)
+{
+       unsigned long cpu = smp_processor_id();
+       PTBI _pTBI = per_cpu(pTBI, cpu);
+       struct traps_context *context = metag_traps_context;
+
+       metag_traps_context = NULL;
+
+       memcpy((void *)_pTBI->fnSigs, context->fnSigs, sizeof(context->fnSigs));
+
+       kfree(context);
+       return 0;
+}
+#endif
+
+#ifdef CONFIG_SMP
+unsigned int get_trigger_mask(void)
+{
+       unsigned long cpu = smp_processor_id();
+       return per_cpu(trigger_mask, cpu);
+}
+
+static void set_trigger_mask(unsigned int mask)
+{
+       unsigned long cpu = smp_processor_id();
+       per_cpu(trigger_mask, cpu) = mask;
+}
+#else
+static void set_trigger_mask(unsigned int mask)
+{
+       global_trigger_mask = mask;
+}
+#endif
+
+void __cpuinit per_cpu_trap_init(unsigned long cpu)
+{
+       TBIRES int_context;
+       unsigned int thread = cpu_2_hwthread_id[cpu];
+
+       set_trigger_mask(TBI_INTS_INIT(thread) | /* interrupts */
+                        TBI_TRIG_BIT(TBID_SIGNUM_LWK) | /* low level kick */
+                        TBI_TRIG_BIT(TBID_SIGNUM_SW1) |
+                        TBI_TRIG_BIT(TBID_SIGNUM_SWS));
+
+       /* non-priv - use current stack */
+       int_context.Sig.pCtx = NULL;
+       /* Start with interrupts off */
+       int_context.Sig.TrigMask = INTS_OFF_MASK;
+       int_context.Sig.SaveMask = 0;
+
+       /* And call __TBIASyncTrigger() */
+       __TBIASyncTrigger(int_context);
+}
+
+void __init trap_init(void)
+{
+       unsigned long cpu = smp_processor_id();
+       PTBI _pTBI = per_cpu(pTBI, cpu);
+
+       _pTBI->fnSigs[TBID_SIGNUM_XXF] = fault_handler;
+       _pTBI->fnSigs[TBID_SIGNUM_SW0] = switchx_handler;
+       _pTBI->fnSigs[TBID_SIGNUM_SW1] = switch1_handler;
+       _pTBI->fnSigs[TBID_SIGNUM_SW2] = switchx_handler;
+       _pTBI->fnSigs[TBID_SIGNUM_SW3] = switchx_handler;
+       _pTBI->fnSigs[TBID_SIGNUM_SWK] = kick_handler;
+
+#ifdef CONFIG_METAG_META21
+       _pTBI->fnSigs[TBID_SIGNUM_DFR] = __TBIHandleDFR;
+       _pTBI->fnSigs[TBID_SIGNUM_FPE] = fpe_handler;
+#endif
+
+       per_cpu_trap_init(cpu);
+}
+
+void tbi_startup_interrupt(int irq)
+{
+       unsigned long cpu = smp_processor_id();
+       PTBI _pTBI = per_cpu(pTBI, cpu);
+
+       BUG_ON(irq > TBID_SIGNUM_MAX);
+
+       /* For TR1 and TR2, the thread id is encoded in the irq number */
+       if (irq >= TBID_SIGNUM_T10 && irq < TBID_SIGNUM_TR3)
+               cpu = hwthread_id_2_cpu[(irq - TBID_SIGNUM_T10) % 4];
+
+       set_trigger_mask(get_trigger_mask() | TBI_TRIG_BIT(irq));
+
+       _pTBI->fnSigs[irq] = trigger_handler;
+}
+
+void tbi_shutdown_interrupt(int irq)
+{
+       unsigned long cpu = smp_processor_id();
+       PTBI _pTBI = per_cpu(pTBI, cpu);
+
+       BUG_ON(irq > TBID_SIGNUM_MAX);
+
+       set_trigger_mask(get_trigger_mask() & ~TBI_TRIG_BIT(irq));
+
+       _pTBI->fnSigs[irq] = __TBIUnExpXXX;
+}
+
+int ret_from_fork(TBIRES arg)
+{
+       struct task_struct *prev = arg.Switch.pPara;
+       struct task_struct *tsk = current;
+       struct pt_regs *regs = task_pt_regs(tsk);
+       int (*fn)(void *);
+       TBIRES Next;
+
+       schedule_tail(prev);
+
+       if (tsk->flags & PF_KTHREAD) {
+               fn = (void *)regs->ctx.DX[4].U1;
+               BUG_ON(!fn);
+
+               fn((void *)regs->ctx.DX[3].U1);
+       }
+
+       if (test_syscall_work())
+               syscall_trace_leave(regs);
+
+       preempt_disable();
+
+       Next.Sig.TrigMask = get_trigger_mask();
+       Next.Sig.SaveMask = 0;
+       Next.Sig.pCtx = &regs->ctx;
+
+       set_gateway_tls(current->thread.tls_ptr);
+
+       preempt_enable_no_resched();
+
+       /* And interrupts should come back on when we resume the real usermode
+        * code. Call __TBIASyncResume()
+        */
+       __TBIASyncResume(tail_end(Next));
+       /* ASyncResume should NEVER return */
+       BUG();
+       return 0;
+}
+
+void show_trace(struct task_struct *tsk, unsigned long *sp,
+               struct pt_regs *regs)
+{
+       unsigned long addr;
+#ifdef CONFIG_FRAME_POINTER
+       unsigned long fp, fpnew;
+       unsigned long stack;
+#endif
+
+       if (regs && user_mode(regs))
+               return;
+
+       printk("\nCall trace: ");
+#ifdef CONFIG_KALLSYMS
+       printk("\n");
+#endif
+
+       if (!tsk)
+               tsk = current;
+
+#ifdef CONFIG_FRAME_POINTER
+       if (regs) {
+               print_ip_sym(regs->ctx.CurrPC);
+               fp = regs->ctx.AX[1].U0;
+       } else {
+               fp = __core_reg_get(A0FrP);
+       }
+
+       /* detect when the frame pointer has been used for other purposes and
+        * doesn't point to the stack (it may point completely elsewhere which
+        * kstack_end may not detect).
+        */
+       stack = (unsigned long)task_stack_page(tsk);
+       while (fp >= stack && fp + 8 <= stack + THREAD_SIZE) {
+               addr = __raw_readl((unsigned long *)(fp + 4)) - 4;
+               if (kernel_text_address(addr))
+                       print_ip_sym(addr);
+               else
+                       break;
+               /* stack grows up, so frame pointers must decrease */
+               fpnew = __raw_readl((unsigned long *)(fp + 0));
+               if (fpnew >= fp)
+                       break;
+               fp = fpnew;
+       }
+#else
+       while (!kstack_end(sp)) {
+               addr = (*sp--) - 4;
+               if (kernel_text_address(addr))
+                       print_ip_sym(addr);
+       }
+#endif
+
+       printk("\n");
+
+       debug_show_held_locks(tsk);
+}
+
+void show_stack(struct task_struct *tsk, unsigned long *sp)
+{
+       if (!tsk)
+               tsk = current;
+       if (tsk == current)
+               sp = (unsigned long *)current_stack_pointer;
+       else
+               sp = (unsigned long *)tsk->thread.kernel_context->AX[0].U0;
+
+       show_trace(tsk, sp, NULL);
+}
+
+void dump_stack(void)
+{
+       show_stack(NULL, NULL);
+}
+EXPORT_SYMBOL(dump_stack);