Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / arch / ia64 / kernel / process.c
diff --git a/kernel/arch/ia64/kernel/process.c b/kernel/arch/ia64/kernel/process.c
new file mode 100644 (file)
index 0000000..b515149
--- /dev/null
@@ -0,0 +1,682 @@
+/*
+ * Architecture-specific setup.
+ *
+ * Copyright (C) 1998-2003 Hewlett-Packard Co
+ *     David Mosberger-Tang <davidm@hpl.hp.com>
+ * 04/11/17 Ashok Raj  <ashok.raj@intel.com> Added CPU Hotplug Support
+ *
+ * 2005-10-07 Keith Owens <kaos@sgi.com>
+ *           Add notify_die() hooks.
+ */
+#include <linux/cpu.h>
+#include <linux/pm.h>
+#include <linux/elf.h>
+#include <linux/errno.h>
+#include <linux/kallsyms.h>
+#include <linux/kernel.h>
+#include <linux/mm.h>
+#include <linux/slab.h>
+#include <linux/module.h>
+#include <linux/notifier.h>
+#include <linux/personality.h>
+#include <linux/sched.h>
+#include <linux/stddef.h>
+#include <linux/thread_info.h>
+#include <linux/unistd.h>
+#include <linux/efi.h>
+#include <linux/interrupt.h>
+#include <linux/delay.h>
+#include <linux/kdebug.h>
+#include <linux/utsname.h>
+#include <linux/tracehook.h>
+#include <linux/rcupdate.h>
+
+#include <asm/cpu.h>
+#include <asm/delay.h>
+#include <asm/elf.h>
+#include <asm/irq.h>
+#include <asm/kexec.h>
+#include <asm/pgalloc.h>
+#include <asm/processor.h>
+#include <asm/sal.h>
+#include <asm/switch_to.h>
+#include <asm/tlbflush.h>
+#include <asm/uaccess.h>
+#include <asm/unwind.h>
+#include <asm/user.h>
+
+#include "entry.h"
+
+#ifdef CONFIG_PERFMON
+# include <asm/perfmon.h>
+#endif
+
+#include "sigframe.h"
+
+void (*ia64_mark_idle)(int);
+
+unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
+EXPORT_SYMBOL(boot_option_idle_override);
+void (*pm_power_off) (void);
+EXPORT_SYMBOL(pm_power_off);
+
+void
+ia64_do_show_stack (struct unw_frame_info *info, void *arg)
+{
+       unsigned long ip, sp, bsp;
+       char buf[128];                  /* don't make it so big that it overflows the stack! */
+
+       printk("\nCall Trace:\n");
+       do {
+               unw_get_ip(info, &ip);
+               if (ip == 0)
+                       break;
+
+               unw_get_sp(info, &sp);
+               unw_get_bsp(info, &bsp);
+               snprintf(buf, sizeof(buf),
+                        " [<%016lx>] %%s\n"
+                        "                                sp=%016lx bsp=%016lx\n",
+                        ip, sp, bsp);
+               print_symbol(buf, ip);
+       } while (unw_unwind(info) >= 0);
+}
+
+void
+show_stack (struct task_struct *task, unsigned long *sp)
+{
+       if (!task)
+               unw_init_running(ia64_do_show_stack, NULL);
+       else {
+               struct unw_frame_info info;
+
+               unw_init_from_blocked_task(&info, task);
+               ia64_do_show_stack(&info, NULL);
+       }
+}
+
+void
+show_regs (struct pt_regs *regs)
+{
+       unsigned long ip = regs->cr_iip + ia64_psr(regs)->ri;
+
+       print_modules();
+       printk("\n");
+       show_regs_print_info(KERN_DEFAULT);
+       printk("psr : %016lx ifs : %016lx ip  : [<%016lx>]    %s (%s)\n",
+              regs->cr_ipsr, regs->cr_ifs, ip, print_tainted(),
+              init_utsname()->release);
+       print_symbol("ip is at %s\n", ip);
+       printk("unat: %016lx pfs : %016lx rsc : %016lx\n",
+              regs->ar_unat, regs->ar_pfs, regs->ar_rsc);
+       printk("rnat: %016lx bsps: %016lx pr  : %016lx\n",
+              regs->ar_rnat, regs->ar_bspstore, regs->pr);
+       printk("ldrs: %016lx ccv : %016lx fpsr: %016lx\n",
+              regs->loadrs, regs->ar_ccv, regs->ar_fpsr);
+       printk("csd : %016lx ssd : %016lx\n", regs->ar_csd, regs->ar_ssd);
+       printk("b0  : %016lx b6  : %016lx b7  : %016lx\n", regs->b0, regs->b6, regs->b7);
+       printk("f6  : %05lx%016lx f7  : %05lx%016lx\n",
+              regs->f6.u.bits[1], regs->f6.u.bits[0],
+              regs->f7.u.bits[1], regs->f7.u.bits[0]);
+       printk("f8  : %05lx%016lx f9  : %05lx%016lx\n",
+              regs->f8.u.bits[1], regs->f8.u.bits[0],
+              regs->f9.u.bits[1], regs->f9.u.bits[0]);
+       printk("f10 : %05lx%016lx f11 : %05lx%016lx\n",
+              regs->f10.u.bits[1], regs->f10.u.bits[0],
+              regs->f11.u.bits[1], regs->f11.u.bits[0]);
+
+       printk("r1  : %016lx r2  : %016lx r3  : %016lx\n", regs->r1, regs->r2, regs->r3);
+       printk("r8  : %016lx r9  : %016lx r10 : %016lx\n", regs->r8, regs->r9, regs->r10);
+       printk("r11 : %016lx r12 : %016lx r13 : %016lx\n", regs->r11, regs->r12, regs->r13);
+       printk("r14 : %016lx r15 : %016lx r16 : %016lx\n", regs->r14, regs->r15, regs->r16);
+       printk("r17 : %016lx r18 : %016lx r19 : %016lx\n", regs->r17, regs->r18, regs->r19);
+       printk("r20 : %016lx r21 : %016lx r22 : %016lx\n", regs->r20, regs->r21, regs->r22);
+       printk("r23 : %016lx r24 : %016lx r25 : %016lx\n", regs->r23, regs->r24, regs->r25);
+       printk("r26 : %016lx r27 : %016lx r28 : %016lx\n", regs->r26, regs->r27, regs->r28);
+       printk("r29 : %016lx r30 : %016lx r31 : %016lx\n", regs->r29, regs->r30, regs->r31);
+
+       if (user_mode(regs)) {
+               /* print the stacked registers */
+               unsigned long val, *bsp, ndirty;
+               int i, sof, is_nat = 0;
+
+               sof = regs->cr_ifs & 0x7f;      /* size of frame */
+               ndirty = (regs->loadrs >> 19);
+               bsp = ia64_rse_skip_regs((unsigned long *) regs->ar_bspstore, ndirty);
+               for (i = 0; i < sof; ++i) {
+                       get_user(val, (unsigned long __user *) ia64_rse_skip_regs(bsp, i));
+                       printk("r%-3u:%c%016lx%s", 32 + i, is_nat ? '*' : ' ', val,
+                              ((i == sof - 1) || (i % 3) == 2) ? "\n" : " ");
+               }
+       } else
+               show_stack(NULL, NULL);
+}
+
+/* local support for deprecated console_print */
+void
+console_print(const char *s)
+{
+       printk(KERN_EMERG "%s", s);
+}
+
+void
+do_notify_resume_user(sigset_t *unused, struct sigscratch *scr, long in_syscall)
+{
+       if (fsys_mode(current, &scr->pt)) {
+               /*
+                * defer signal-handling etc. until we return to
+                * privilege-level 0.
+                */
+               if (!ia64_psr(&scr->pt)->lp)
+                       ia64_psr(&scr->pt)->lp = 1;
+               return;
+       }
+
+#ifdef CONFIG_PERFMON
+       if (current->thread.pfm_needs_checking)
+               /*
+                * Note: pfm_handle_work() allow us to call it with interrupts
+                * disabled, and may enable interrupts within the function.
+                */
+               pfm_handle_work();
+#endif
+
+       /* deal with pending signal delivery */
+       if (test_thread_flag(TIF_SIGPENDING)) {
+               local_irq_enable();     /* force interrupt enable */
+               ia64_do_signal(scr, in_syscall);
+       }
+
+       if (test_and_clear_thread_flag(TIF_NOTIFY_RESUME)) {
+               local_irq_enable();     /* force interrupt enable */
+               tracehook_notify_resume(&scr->pt);
+       }
+
+       /* copy user rbs to kernel rbs */
+       if (unlikely(test_thread_flag(TIF_RESTORE_RSE))) {
+               local_irq_enable();     /* force interrupt enable */
+               ia64_sync_krbs();
+       }
+
+       local_irq_disable();    /* force interrupt disable */
+}
+
+static int __init nohalt_setup(char * str)
+{
+       cpu_idle_poll_ctrl(true);
+       return 1;
+}
+__setup("nohalt", nohalt_setup);
+
+#ifdef CONFIG_HOTPLUG_CPU
+/* We don't actually take CPU down, just spin without interrupts. */
+static inline void play_dead(void)
+{
+       unsigned int this_cpu = smp_processor_id();
+
+       /* Ack it */
+       __this_cpu_write(cpu_state, CPU_DEAD);
+
+       max_xtp();
+       local_irq_disable();
+       idle_task_exit();
+       ia64_jump_to_sal(&sal_boot_rendez_state[this_cpu]);
+       /*
+        * The above is a point of no-return, the processor is
+        * expected to be in SAL loop now.
+        */
+       BUG();
+}
+#else
+static inline void play_dead(void)
+{
+       BUG();
+}
+#endif /* CONFIG_HOTPLUG_CPU */
+
+void arch_cpu_idle_dead(void)
+{
+       play_dead();
+}
+
+void arch_cpu_idle(void)
+{
+       void (*mark_idle)(int) = ia64_mark_idle;
+
+#ifdef CONFIG_SMP
+       min_xtp();
+#endif
+       rmb();
+       if (mark_idle)
+               (*mark_idle)(1);
+
+       safe_halt();
+
+       if (mark_idle)
+               (*mark_idle)(0);
+#ifdef CONFIG_SMP
+       normal_xtp();
+#endif
+}
+
+void
+ia64_save_extra (struct task_struct *task)
+{
+#ifdef CONFIG_PERFMON
+       unsigned long info;
+#endif
+
+       if ((task->thread.flags & IA64_THREAD_DBG_VALID) != 0)
+               ia64_save_debug_regs(&task->thread.dbr[0]);
+
+#ifdef CONFIG_PERFMON
+       if ((task->thread.flags & IA64_THREAD_PM_VALID) != 0)
+               pfm_save_regs(task);
+
+       info = __this_cpu_read(pfm_syst_info);
+       if (info & PFM_CPUINFO_SYST_WIDE)
+               pfm_syst_wide_update_task(task, info, 0);
+#endif
+}
+
+void
+ia64_load_extra (struct task_struct *task)
+{
+#ifdef CONFIG_PERFMON
+       unsigned long info;
+#endif
+
+       if ((task->thread.flags & IA64_THREAD_DBG_VALID) != 0)
+               ia64_load_debug_regs(&task->thread.dbr[0]);
+
+#ifdef CONFIG_PERFMON
+       if ((task->thread.flags & IA64_THREAD_PM_VALID) != 0)
+               pfm_load_regs(task);
+
+       info = __this_cpu_read(pfm_syst_info);
+       if (info & PFM_CPUINFO_SYST_WIDE) 
+               pfm_syst_wide_update_task(task, info, 1);
+#endif
+}
+
+/*
+ * Copy the state of an ia-64 thread.
+ *
+ * We get here through the following  call chain:
+ *
+ *     from user-level:        from kernel:
+ *
+ *     <clone syscall>         <some kernel call frames>
+ *     sys_clone                  :
+ *     do_fork                 do_fork
+ *     copy_thread             copy_thread
+ *
+ * This means that the stack layout is as follows:
+ *
+ *     +---------------------+ (highest addr)
+ *     |   struct pt_regs    |
+ *     +---------------------+
+ *     | struct switch_stack |
+ *     +---------------------+
+ *     |                     |
+ *     |    memory stack     |
+ *     |                     | <-- sp (lowest addr)
+ *     +---------------------+
+ *
+ * Observe that we copy the unat values that are in pt_regs and switch_stack.  Spilling an
+ * integer to address X causes bit N in ar.unat to be set to the NaT bit of the register,
+ * with N=(X & 0x1ff)/8.  Thus, copying the unat value preserves the NaT bits ONLY if the
+ * pt_regs structure in the parent is congruent to that of the child, modulo 512.  Since
+ * the stack is page aligned and the page size is at least 4KB, this is always the case,
+ * so there is nothing to worry about.
+ */
+int
+copy_thread(unsigned long clone_flags,
+            unsigned long user_stack_base, unsigned long user_stack_size,
+            struct task_struct *p)
+{
+       extern char ia64_ret_from_clone;
+       struct switch_stack *child_stack, *stack;
+       unsigned long rbs, child_rbs, rbs_size;
+       struct pt_regs *child_ptregs;
+       struct pt_regs *regs = current_pt_regs();
+       int retval = 0;
+
+       child_ptregs = (struct pt_regs *) ((unsigned long) p + IA64_STK_OFFSET) - 1;
+       child_stack = (struct switch_stack *) child_ptregs - 1;
+
+       rbs = (unsigned long) current + IA64_RBS_OFFSET;
+       child_rbs = (unsigned long) p + IA64_RBS_OFFSET;
+
+       /* copy parts of thread_struct: */
+       p->thread.ksp = (unsigned long) child_stack - 16;
+
+       /*
+        * NOTE: The calling convention considers all floating point
+        * registers in the high partition (fph) to be scratch.  Since
+        * the only way to get to this point is through a system call,
+        * we know that the values in fph are all dead.  Hence, there
+        * is no need to inherit the fph state from the parent to the
+        * child and all we have to do is to make sure that
+        * IA64_THREAD_FPH_VALID is cleared in the child.
+        *
+        * XXX We could push this optimization a bit further by
+        * clearing IA64_THREAD_FPH_VALID on ANY system call.
+        * However, it's not clear this is worth doing.  Also, it
+        * would be a slight deviation from the normal Linux system
+        * call behavior where scratch registers are preserved across
+        * system calls (unless used by the system call itself).
+        */
+#      define THREAD_FLAGS_TO_CLEAR    (IA64_THREAD_FPH_VALID | IA64_THREAD_DBG_VALID \
+                                        | IA64_THREAD_PM_VALID)
+#      define THREAD_FLAGS_TO_SET      0
+       p->thread.flags = ((current->thread.flags & ~THREAD_FLAGS_TO_CLEAR)
+                          | THREAD_FLAGS_TO_SET);
+
+       ia64_drop_fpu(p);       /* don't pick up stale state from a CPU's fph */
+
+       if (unlikely(p->flags & PF_KTHREAD)) {
+               if (unlikely(!user_stack_base)) {
+                       /* fork_idle() called us */
+                       return 0;
+               }
+               memset(child_stack, 0, sizeof(*child_ptregs) + sizeof(*child_stack));
+               child_stack->r4 = user_stack_base;      /* payload */
+               child_stack->r5 = user_stack_size;      /* argument */
+               /*
+                * Preserve PSR bits, except for bits 32-34 and 37-45,
+                * which we can't read.
+                */
+               child_ptregs->cr_ipsr = ia64_getreg(_IA64_REG_PSR) | IA64_PSR_BN;
+               /* mark as valid, empty frame */
+               child_ptregs->cr_ifs = 1UL << 63;
+               child_stack->ar_fpsr = child_ptregs->ar_fpsr
+                       = ia64_getreg(_IA64_REG_AR_FPSR);
+               child_stack->pr = (1 << PRED_KERNEL_STACK);
+               child_stack->ar_bspstore = child_rbs;
+               child_stack->b0 = (unsigned long) &ia64_ret_from_clone;
+
+               /* stop some PSR bits from being inherited.
+                * the psr.up/psr.pp bits must be cleared on fork but inherited on execve()
+                * therefore we must specify them explicitly here and not include them in
+                * IA64_PSR_BITS_TO_CLEAR.
+                */
+               child_ptregs->cr_ipsr = ((child_ptregs->cr_ipsr | IA64_PSR_BITS_TO_SET)
+                                & ~(IA64_PSR_BITS_TO_CLEAR | IA64_PSR_PP | IA64_PSR_UP));
+
+               return 0;
+       }
+       stack = ((struct switch_stack *) regs) - 1;
+       /* copy parent's switch_stack & pt_regs to child: */
+       memcpy(child_stack, stack, sizeof(*child_ptregs) + sizeof(*child_stack));
+
+       /* copy the parent's register backing store to the child: */
+       rbs_size = stack->ar_bspstore - rbs;
+       memcpy((void *) child_rbs, (void *) rbs, rbs_size);
+       if (clone_flags & CLONE_SETTLS)
+               child_ptregs->r13 = regs->r16;  /* see sys_clone2() in entry.S */
+       if (user_stack_base) {
+               child_ptregs->r12 = user_stack_base + user_stack_size - 16;
+               child_ptregs->ar_bspstore = user_stack_base;
+               child_ptregs->ar_rnat = 0;
+               child_ptregs->loadrs = 0;
+       }
+       child_stack->ar_bspstore = child_rbs + rbs_size;
+       child_stack->b0 = (unsigned long) &ia64_ret_from_clone;
+
+       /* stop some PSR bits from being inherited.
+        * the psr.up/psr.pp bits must be cleared on fork but inherited on execve()
+        * therefore we must specify them explicitly here and not include them in
+        * IA64_PSR_BITS_TO_CLEAR.
+        */
+       child_ptregs->cr_ipsr = ((child_ptregs->cr_ipsr | IA64_PSR_BITS_TO_SET)
+                                & ~(IA64_PSR_BITS_TO_CLEAR | IA64_PSR_PP | IA64_PSR_UP));
+
+#ifdef CONFIG_PERFMON
+       if (current->thread.pfm_context)
+               pfm_inherit(p, child_ptregs);
+#endif
+       return retval;
+}
+
+static void
+do_copy_task_regs (struct task_struct *task, struct unw_frame_info *info, void *arg)
+{
+       unsigned long mask, sp, nat_bits = 0, ar_rnat, urbs_end, cfm;
+       unsigned long uninitialized_var(ip);    /* GCC be quiet */
+       elf_greg_t *dst = arg;
+       struct pt_regs *pt;
+       char nat;
+       int i;
+
+       memset(dst, 0, sizeof(elf_gregset_t));  /* don't leak any kernel bits to user-level */
+
+       if (unw_unwind_to_user(info) < 0)
+               return;
+
+       unw_get_sp(info, &sp);
+       pt = (struct pt_regs *) (sp + 16);
+
+       urbs_end = ia64_get_user_rbs_end(task, pt, &cfm);
+
+       if (ia64_sync_user_rbs(task, info->sw, pt->ar_bspstore, urbs_end) < 0)
+               return;
+
+       ia64_peek(task, info->sw, urbs_end, (long) ia64_rse_rnat_addr((long *) urbs_end),
+                 &ar_rnat);
+
+       /*
+        * coredump format:
+        *      r0-r31
+        *      NaT bits (for r0-r31; bit N == 1 iff rN is a NaT)
+        *      predicate registers (p0-p63)
+        *      b0-b7
+        *      ip cfm user-mask
+        *      ar.rsc ar.bsp ar.bspstore ar.rnat
+        *      ar.ccv ar.unat ar.fpsr ar.pfs ar.lc ar.ec
+        */
+
+       /* r0 is zero */
+       for (i = 1, mask = (1UL << i); i < 32; ++i) {
+               unw_get_gr(info, i, &dst[i], &nat);
+               if (nat)
+                       nat_bits |= mask;
+               mask <<= 1;
+       }
+       dst[32] = nat_bits;
+       unw_get_pr(info, &dst[33]);
+
+       for (i = 0; i < 8; ++i)
+               unw_get_br(info, i, &dst[34 + i]);
+
+       unw_get_rp(info, &ip);
+       dst[42] = ip + ia64_psr(pt)->ri;
+       dst[43] = cfm;
+       dst[44] = pt->cr_ipsr & IA64_PSR_UM;
+
+       unw_get_ar(info, UNW_AR_RSC, &dst[45]);
+       /*
+        * For bsp and bspstore, unw_get_ar() would return the kernel
+        * addresses, but we need the user-level addresses instead:
+        */
+       dst[46] = urbs_end;     /* note: by convention PT_AR_BSP points to the end of the urbs! */
+       dst[47] = pt->ar_bspstore;
+       dst[48] = ar_rnat;
+       unw_get_ar(info, UNW_AR_CCV, &dst[49]);
+       unw_get_ar(info, UNW_AR_UNAT, &dst[50]);
+       unw_get_ar(info, UNW_AR_FPSR, &dst[51]);
+       dst[52] = pt->ar_pfs;   /* UNW_AR_PFS is == to pt->cr_ifs for interrupt frames */
+       unw_get_ar(info, UNW_AR_LC, &dst[53]);
+       unw_get_ar(info, UNW_AR_EC, &dst[54]);
+       unw_get_ar(info, UNW_AR_CSD, &dst[55]);
+       unw_get_ar(info, UNW_AR_SSD, &dst[56]);
+}
+
+void
+do_dump_task_fpu (struct task_struct *task, struct unw_frame_info *info, void *arg)
+{
+       elf_fpreg_t *dst = arg;
+       int i;
+
+       memset(dst, 0, sizeof(elf_fpregset_t)); /* don't leak any "random" bits */
+
+       if (unw_unwind_to_user(info) < 0)
+               return;
+
+       /* f0 is 0.0, f1 is 1.0 */
+
+       for (i = 2; i < 32; ++i)
+               unw_get_fr(info, i, dst + i);
+
+       ia64_flush_fph(task);
+       if ((task->thread.flags & IA64_THREAD_FPH_VALID) != 0)
+               memcpy(dst + 32, task->thread.fph, 96*16);
+}
+
+void
+do_copy_regs (struct unw_frame_info *info, void *arg)
+{
+       do_copy_task_regs(current, info, arg);
+}
+
+void
+do_dump_fpu (struct unw_frame_info *info, void *arg)
+{
+       do_dump_task_fpu(current, info, arg);
+}
+
+void
+ia64_elf_core_copy_regs (struct pt_regs *pt, elf_gregset_t dst)
+{
+       unw_init_running(do_copy_regs, dst);
+}
+
+int
+dump_fpu (struct pt_regs *pt, elf_fpregset_t dst)
+{
+       unw_init_running(do_dump_fpu, dst);
+       return 1;       /* f0-f31 are always valid so we always return 1 */
+}
+
+/*
+ * Flush thread state.  This is called when a thread does an execve().
+ */
+void
+flush_thread (void)
+{
+       /* drop floating-point and debug-register state if it exists: */
+       current->thread.flags &= ~(IA64_THREAD_FPH_VALID | IA64_THREAD_DBG_VALID);
+       ia64_drop_fpu(current);
+}
+
+/*
+ * Clean up state associated with current thread.  This is called when
+ * the thread calls exit().
+ */
+void
+exit_thread (void)
+{
+
+       ia64_drop_fpu(current);
+#ifdef CONFIG_PERFMON
+       /* if needed, stop monitoring and flush state to perfmon context */
+       if (current->thread.pfm_context)
+               pfm_exit_thread(current);
+
+       /* free debug register resources */
+       if (current->thread.flags & IA64_THREAD_DBG_VALID)
+               pfm_release_debug_registers(current);
+#endif
+}
+
+unsigned long
+get_wchan (struct task_struct *p)
+{
+       struct unw_frame_info info;
+       unsigned long ip;
+       int count = 0;
+
+       if (!p || p == current || p->state == TASK_RUNNING)
+               return 0;
+
+       /*
+        * Note: p may not be a blocked task (it could be current or
+        * another process running on some other CPU.  Rather than
+        * trying to determine if p is really blocked, we just assume
+        * it's blocked and rely on the unwind routines to fail
+        * gracefully if the process wasn't really blocked after all.
+        * --davidm 99/12/15
+        */
+       unw_init_from_blocked_task(&info, p);
+       do {
+               if (p->state == TASK_RUNNING)
+                       return 0;
+               if (unw_unwind(&info) < 0)
+                       return 0;
+               unw_get_ip(&info, &ip);
+               if (!in_sched_functions(ip))
+                       return ip;
+       } while (count++ < 16);
+       return 0;
+}
+
+void
+cpu_halt (void)
+{
+       pal_power_mgmt_info_u_t power_info[8];
+       unsigned long min_power;
+       int i, min_power_state;
+
+       if (ia64_pal_halt_info(power_info) != 0)
+               return;
+
+       min_power_state = 0;
+       min_power = power_info[0].pal_power_mgmt_info_s.power_consumption;
+       for (i = 1; i < 8; ++i)
+               if (power_info[i].pal_power_mgmt_info_s.im
+                   && power_info[i].pal_power_mgmt_info_s.power_consumption < min_power) {
+                       min_power = power_info[i].pal_power_mgmt_info_s.power_consumption;
+                       min_power_state = i;
+               }
+
+       while (1)
+               ia64_pal_halt(min_power_state);
+}
+
+void machine_shutdown(void)
+{
+#ifdef CONFIG_HOTPLUG_CPU
+       int cpu;
+
+       for_each_online_cpu(cpu) {
+               if (cpu != smp_processor_id())
+                       cpu_down(cpu);
+       }
+#endif
+#ifdef CONFIG_KEXEC
+       kexec_disable_iosapic();
+#endif
+}
+
+void
+machine_restart (char *restart_cmd)
+{
+       (void) notify_die(DIE_MACHINE_RESTART, restart_cmd, NULL, 0, 0, 0);
+       efi_reboot(REBOOT_WARM, NULL);
+}
+
+void
+machine_halt (void)
+{
+       (void) notify_die(DIE_MACHINE_HALT, "", NULL, 0, 0, 0);
+       cpu_halt();
+}
+
+void
+machine_power_off (void)
+{
+       if (pm_power_off)
+               pm_power_off();
+       machine_halt();
+}
+