Merge branch 'wip-mips-pm' of https://github.com/paulburton/linux into mips-for-linux-next

14 files changed, 1608 insertions, 183 deletions

diff --git a/arch/mips/kernel/Makefile b/arch/mips/kernel/Makefile
index 8f8b531..a61d108 100644
--- a/arch/mips/kernel/Makefile
+++ b/arch/mips/kernel/Makefile
@@ -105,6 +105,9 @@ obj-$(CONFIG_JUMP_LABEL)	+= jump_label.o
 obj-$(CONFIG_MIPS_CM)		+= mips-cm.o
 obj-$(CONFIG_MIPS_CPC)		+= mips-cpc.o
 
+obj-$(CONFIG_CPU_PM)		+= pm.o
+obj-$(CONFIG_MIPS_CPS_PM)	+= pm-cps.o
+
 #
 # DSP ASE supported for MIPS32 or MIPS64 Release 2 cores only. It is not
 # safe to unconditionnaly use the assembler -mdsp / -mdspr2 switches
diff --git a/arch/mips/kernel/asm-offsets.c b/arch/mips/kernel/asm-offsets.c
index 08f897e..02f075d 100644
--- a/arch/mips/kernel/asm-offsets.c
+++ b/arch/mips/kernel/asm-offsets.c
@@ -14,6 +14,7 @@
 #include <linux/mm.h>
 #include <linux/kbuild.h>
 #include <linux/suspend.h>
+#include <asm/pm.h>
 #include <asm/ptrace.h>
 #include <asm/processor.h>
 #include <asm/smp-cps.h>
@@ -401,6 +402,20 @@ void output_pbe_defines(void)
 }
 #endif
 
+#ifdef CONFIG_CPU_PM
+void output_pm_defines(void)
+{
+	COMMENT(" PM offsets. ");
+#ifdef CONFIG_EVA
+	OFFSET(SSS_SEGCTL0,	mips_static_suspend_state, segctl[0]);
+	OFFSET(SSS_SEGCTL1,	mips_static_suspend_state, segctl[1]);
+	OFFSET(SSS_SEGCTL2,	mips_static_suspend_state, segctl[2]);
+#endif
+	OFFSET(SSS_SP,		mips_static_suspend_state, sp);
+	BLANK();
+}
+#endif
+
 void output_kvm_defines(void)
 {
 	COMMENT(" KVM/MIPS Specfic offsets. ");
@@ -469,10 +484,14 @@ void output_kvm_defines(void)
 void output_cps_defines(void)
 {
 	COMMENT(" MIPS CPS offsets. ");
-	OFFSET(BOOTCFG_CORE, boot_config, core);
-	OFFSET(BOOTCFG_VPE, boot_config, vpe);
-	OFFSET(BOOTCFG_PC, boot_config, pc);
-	OFFSET(BOOTCFG_SP, boot_config, sp);
-	OFFSET(BOOTCFG_GP, boot_config, gp);
+
+	OFFSET(COREBOOTCFG_VPEMASK, core_boot_config, vpe_mask);
+	OFFSET(COREBOOTCFG_VPECONFIG, core_boot_config, vpe_config);
+	DEFINE(COREBOOTCFG_SIZE, sizeof(struct core_boot_config));
+
+	OFFSET(VPEBOOTCFG_PC, vpe_boot_config, pc);
+	OFFSET(VPEBOOTCFG_SP, vpe_boot_config, sp);
+	OFFSET(VPEBOOTCFG_GP, vpe_boot_config, gp);
+	DEFINE(VPEBOOTCFG_SIZE, sizeof(struct vpe_boot_config));
 }
 #endif
diff --git a/arch/mips/kernel/cevt-gic.c b/arch/mips/kernel/cevt-gic.c
index 594cbbf..6093716 100644
--- a/arch/mips/kernel/cevt-gic.c
+++ b/arch/mips/kernel/cevt-gic.c
@@ -26,7 +26,7 @@ static int gic_next_event(unsigned long delta, struct clock_event_device *evt)
 
 	cnt = gic_read_count();
 	cnt += (u64)delta;
-	gic_write_compare(cnt);
+	gic_write_cpu_compare(cnt, cpumask_first(evt->cpumask));
 	res = ((int)(gic_read_count() - cnt) >= 0) ? -ETIME : 0;
 	return res;
 }
@@ -73,7 +73,8 @@ int gic_clockevent_init(void)
 	cd = &per_cpu(gic_clockevent_device, cpu);
 
 	cd->name		= "MIPS GIC";
-	cd->features		= CLOCK_EVT_FEAT_ONESHOT;
+	cd->features		= CLOCK_EVT_FEAT_ONESHOT |
+				  CLOCK_EVT_FEAT_C3STOP;
 
 	clockevent_set_clock(cd, gic_frequency);
 
diff --git a/arch/mips/kernel/cevt-r4k.c b/arch/mips/kernel/cevt-r4k.c
index bff124a..bc127e2 100644
--- a/arch/mips/kernel/cevt-r4k.c
+++ b/arch/mips/kernel/cevt-r4k.c
@@ -62,9 +62,6 @@ irqreturn_t c0_compare_interrupt(int irq, void *dev_id)
 		/* Clear Count/Compare Interrupt */
 		write_c0_compare(read_c0_compare());
 		cd = &per_cpu(mips_clockevent_device, cpu);
-#ifdef CONFIG_CEVT_GIC
-		if (!gic_present)
-#endif
 		cd->event_handler(cd);
 	}
 
@@ -182,7 +179,9 @@ int r4k_clockevent_init(void)
 	cd = &per_cpu(mips_clockevent_device, cpu);
 
 	cd->name		= "MIPS";
-	cd->features		= CLOCK_EVT_FEAT_ONESHOT;
+	cd->features		= CLOCK_EVT_FEAT_ONESHOT |
+				  CLOCK_EVT_FEAT_C3STOP |
+				  CLOCK_EVT_FEAT_PERCPU;
 
 	clockevent_set_clock(cd, mips_hpt_frequency);
 
@@ -197,9 +196,6 @@ int r4k_clockevent_init(void)
 	cd->set_mode		= mips_set_clock_mode;
 	cd->event_handler	= mips_event_handler;
 
-#ifdef CONFIG_CEVT_GIC
-	if (!gic_present)
-#endif
 	clockevents_register_device(cd);
 
 	if (cp0_timer_irq_installed)
diff --git a/arch/mips/kernel/cps-vec.S b/arch/mips/kernel/cps-vec.S
index f7a46db..6f4f739 100644
--- a/arch/mips/kernel/cps-vec.S
+++ b/arch/mips/kernel/cps-vec.S
@@ -14,19 +14,43 @@
 #include <asm/asmmacro.h>
 #include <asm/cacheops.h>
 #include <asm/mipsregs.h>
+#include <asm/mipsmtregs.h>
+#include <asm/pm.h>
 
-#define GCR_CL_COHERENCE_OFS 0x2008
+#define GCR_CL_COHERENCE_OFS	0x2008
+#define GCR_CL_ID_OFS		0x2028
+
+.extern mips_cm_base
+
+.set noreorder
+
+	/*
+	 * Set dest to non-zero if the core supports the MT ASE, else zero. If
+	 * MT is not supported then branch to nomt.
+	 */
+	.macro	has_mt	dest, nomt
+	mfc0	\dest, CP0_CONFIG
+	bgez	\dest, \nomt
+	 mfc0	\dest, CP0_CONFIG, 1
+	bgez	\dest, \nomt
+	 mfc0	\dest, CP0_CONFIG, 2
+	bgez	\dest, \nomt
+	 mfc0	\dest, CP0_CONFIG, 3
+	andi	\dest, \dest, MIPS_CONF3_MT
+	beqz	\dest, \nomt
+	.endm
 
 .section .text.cps-vec
 .balign 0x1000
-.set noreorder
 
 LEAF(mips_cps_core_entry)
 	/*
-	 * These first 8 bytes will be patched by cps_smp_setup to load the
-	 * base address of the CM GCRs into register v1.
+	 * These first 12 bytes will be patched by cps_smp_setup to load the
+	 * base address of the CM GCRs into register v1 and the CCA to use into
+	 * register s0.
 	 */
 	.quad	0
+	.word	0
 
 	/* Check whether we're here due to an NMI */
 	mfc0	k0, CP0_STATUS
@@ -117,10 +141,11 @@ icache_done:
 	 add	a0, a0, t0
 dcache_done:
 
-	/* Set Kseg0 cacheable, coherent, write-back, write-allocate */
+	/* Set Kseg0 CCA to that in s0 */
 	mfc0	t0, CP0_CONFIG
 	ori	t0, 0x7
-	xori	t0, 0x2
+	xori	t0, 0x7
+	or	t0, t0, s0
 	mtc0	t0, CP0_CONFIG
 	ehb
 
@@ -134,21 +159,24 @@ dcache_done:
 	jr	t0
 	 nop
 
-1:	/* We're up, cached & coherent */
+	/*
+	 * We're up, cached & coherent. Perform any further required core-level
+	 * initialisation.
+	 */
+1:	jal	mips_cps_core_init
+	 nop
 
 	/*
-	 * TODO: We should check the VPE number we intended to boot here, and
-	 *       if non-zero we should start that VPE and stop this one. For
-	 *       the moment this doesn't matter since CPUs are brought up
-	 *       sequentially and in order, but once hotplug is implemented
-	 *       this will need revisiting.
+	 * Boot any other VPEs within this core that should be online, and
+	 * deactivate this VPE if it should be offline.
 	 */
+	jal	mips_cps_boot_vpes
+	 nop
 
 	/* Off we go! */
-	la	t0, mips_cps_bootcfg
-	lw	t1, BOOTCFG_PC(t0)
-	lw	gp, BOOTCFG_GP(t0)
-	lw	sp, BOOTCFG_SP(t0)
+	lw	t1, VPEBOOTCFG_PC(v0)
+	lw	gp, VPEBOOTCFG_GP(v0)
+	lw	sp, VPEBOOTCFG_SP(v0)
 	jr	t1
 	 nop
 	END(mips_cps_core_entry)
@@ -189,3 +217,271 @@ LEAF(excep_ejtag)
 	jr	k0
 	 nop
 	END(excep_ejtag)
+
+LEAF(mips_cps_core_init)
+#ifdef CONFIG_MIPS_MT
+	/* Check that the core implements the MT ASE */
+	has_mt	t0, 3f
+	 nop
+
+	.set	push
+	.set	mt
+
+	/* Only allow 1 TC per VPE to execute... */
+	dmt
+
+	/* ...and for the moment only 1 VPE */
+	dvpe
+	la	t1, 1f
+	jr.hb	t1
+	 nop
+
+	/* Enter VPE configuration state */
+1:	mfc0	t0, CP0_MVPCONTROL
+	ori	t0, t0, MVPCONTROL_VPC
+	mtc0	t0, CP0_MVPCONTROL
+
+	/* Retrieve the number of VPEs within the core */
+	mfc0	t0, CP0_MVPCONF0
+	srl	t0, t0, MVPCONF0_PVPE_SHIFT
+	andi	t0, t0, (MVPCONF0_PVPE >> MVPCONF0_PVPE_SHIFT)
+	addi	t7, t0, 1
+
+	/* If there's only 1, we're done */
+	beqz	t0, 2f
+	 nop
+
+	/* Loop through each VPE within this core */
+	li	t5, 1
+
+1:	/* Operate on the appropriate TC */
+	mtc0	t5, CP0_VPECONTROL
+	ehb
+
+	/* Bind TC to VPE (1:1 TC:VPE mapping) */
+	mttc0	t5, CP0_TCBIND
+
+	/* Set exclusive TC, non-active, master */
+	li	t0, VPECONF0_MVP
+	sll	t1, t5, VPECONF0_XTC_SHIFT
+	or	t0, t0, t1
+	mttc0	t0, CP0_VPECONF0
+
+	/* Set TC non-active, non-allocatable */
+	mttc0	zero, CP0_TCSTATUS
+
+	/* Set TC halted */
+	li	t0, TCHALT_H
+	mttc0	t0, CP0_TCHALT
+
+	/* Next VPE */
+	addi	t5, t5, 1
+	slt	t0, t5, t7
+	bnez	t0, 1b
+	 nop
+
+	/* Leave VPE configuration state */
+2:	mfc0	t0, CP0_MVPCONTROL
+	xori	t0, t0, MVPCONTROL_VPC
+	mtc0	t0, CP0_MVPCONTROL
+
+3:	.set	pop
+#endif
+	jr	ra
+	 nop
+	END(mips_cps_core_init)
+
+LEAF(mips_cps_boot_vpes)
+	/* Retrieve CM base address */
+	la	t0, mips_cm_base
+	lw	t0, 0(t0)
+
+	/* Calculate a pointer to this cores struct core_boot_config */
+	lw	t0, GCR_CL_ID_OFS(t0)
+	li	t1, COREBOOTCFG_SIZE
+	mul	t0, t0, t1
+	la	t1, mips_cps_core_bootcfg
+	lw	t1, 0(t1)
+	addu	t0, t0, t1
+
+	/* Calculate this VPEs ID. If the core doesn't support MT use 0 */
+	has_mt	t6, 1f
+	 li	t9, 0
+
+	/* Find the number of VPEs present in the core */
+	mfc0	t1, CP0_MVPCONF0
+	srl	t1, t1, MVPCONF0_PVPE_SHIFT
+	andi	t1, t1, MVPCONF0_PVPE >> MVPCONF0_PVPE_SHIFT
+	addi	t1, t1, 1
+
+	/* Calculate a mask for the VPE ID from EBase.CPUNum */
+	clz	t1, t1
+	li	t2, 31
+	subu	t1, t2, t1
+	li	t2, 1
+	sll	t1, t2, t1
+	addiu	t1, t1, -1
+
+	/* Retrieve the VPE ID from EBase.CPUNum */
+	mfc0	t9, $15, 1
+	and	t9, t9, t1
+
+1:	/* Calculate a pointer to this VPEs struct vpe_boot_config */
+	li	t1, VPEBOOTCFG_SIZE
+	mul	v0, t9, t1
+	lw	t7, COREBOOTCFG_VPECONFIG(t0)
+	addu	v0, v0, t7
+
+#ifdef CONFIG_MIPS_MT
+
+	/* If the core doesn't support MT then return */
+	bnez	t6, 1f
+	 nop
+	jr	ra
+	 nop
+
+	.set	push
+	.set	mt
+
+1:	/* Enter VPE configuration state */
+	dvpe
+	la	t1, 1f
+	jr.hb	t1
+	 nop
+1:	mfc0	t1, CP0_MVPCONTROL
+	ori	t1, t1, MVPCONTROL_VPC
+	mtc0	t1, CP0_MVPCONTROL
+	ehb
+
+	/* Loop through each VPE */
+	lw	t6, COREBOOTCFG_VPEMASK(t0)
+	move	t8, t6
+	li	t5, 0
+
+	/* Check whether the VPE should be running. If not, skip it */
+1:	andi	t0, t6, 1
+	beqz	t0, 2f
+	 nop
+
+	/* Operate on the appropriate TC */
+	mfc0	t0, CP0_VPECONTROL
+	ori	t0, t0, VPECONTROL_TARGTC
+	xori	t0, t0, VPECONTROL_TARGTC
+	or	t0, t0, t5
+	mtc0	t0, CP0_VPECONTROL
+	ehb
+
+	/* Skip the VPE if its TC is not halted */
+	mftc0	t0, CP0_TCHALT
+	beqz	t0, 2f
+	 nop
+
+	/* Calculate a pointer to the VPEs struct vpe_boot_config */
+	li	t0, VPEBOOTCFG_SIZE
+	mul	t0, t0, t5
+	addu	t0, t0, t7
+
+	/* Set the TC restart PC */
+	lw	t1, VPEBOOTCFG_PC(t0)
+	mttc0	t1, CP0_TCRESTART
+
+	/* Set the TC stack pointer */
+	lw	t1, VPEBOOTCFG_SP(t0)
+	mttgpr	t1, sp
+
+	/* Set the TC global pointer */
+	lw	t1, VPEBOOTCFG_GP(t0)
+	mttgpr	t1, gp
+
+	/* Copy config from this VPE */
+	mfc0	t0, CP0_CONFIG
+	mttc0	t0, CP0_CONFIG
+
+	/* Ensure no software interrupts are pending */
+	mttc0	zero, CP0_CAUSE
+	mttc0	zero, CP0_STATUS
+
+	/* Set TC active, not interrupt exempt */
+	mftc0	t0, CP0_TCSTATUS
+	li	t1, ~TCSTATUS_IXMT
+	and	t0, t0, t1
+	ori	t0, t0, TCSTATUS_A
+	mttc0	t0, CP0_TCSTATUS
+
+	/* Clear the TC halt bit */
+	mttc0	zero, CP0_TCHALT
+
+	/* Set VPE active */
+	mftc0	t0, CP0_VPECONF0
+	ori	t0, t0, VPECONF0_VPA
+	mttc0	t0, CP0_VPECONF0
+
+	/* Next VPE */
+2:	srl	t6, t6, 1
+	addi	t5, t5, 1
+	bnez	t6, 1b
+	 nop
+
+	/* Leave VPE configuration state */
+	mfc0	t1, CP0_MVPCONTROL
+	xori	t1, t1, MVPCONTROL_VPC
+	mtc0	t1, CP0_MVPCONTROL
+	ehb
+	evpe
+
+	/* Check whether this VPE is meant to be running */
+	li	t0, 1
+	sll	t0, t0, t9
+	and	t0, t0, t8
+	bnez	t0, 2f
+	 nop
+
+	/* This VPE should be offline, halt the TC */
+	li	t0, TCHALT_H
+	mtc0	t0, CP0_TCHALT
+	la	t0, 1f
+1:	jr.hb	t0
+	 nop
+
+2:	.set	pop
+
+#endif /* CONFIG_MIPS_MT */
+
+	/* Return */
+	jr	ra
+	 nop
+	END(mips_cps_boot_vpes)
+
+#if defined(CONFIG_MIPS_CPS_PM) && defined(CONFIG_CPU_PM)
+
+	/* Calculate a pointer to this CPUs struct mips_static_suspend_state */
+	.macro	psstate	dest
+	.set	push
+	.set	noat
+	lw	$1, TI_CPU(gp)
+	sll	$1, $1, LONGLOG
+	la	\dest, __per_cpu_offset
+	addu	$1, $1, \dest
+	lw	$1, 0($1)
+	la	\dest, cps_cpu_state
+	addu	\dest, \dest, $1
+	.set	pop
+	.endm
+
+LEAF(mips_cps_pm_save)
+	/* Save CPU state */
+	SUSPEND_SAVE_REGS
+	psstate	t1
+	SUSPEND_SAVE_STATIC
+	jr	v0
+	 nop
+	END(mips_cps_pm_save)
+
+LEAF(mips_cps_pm_restore)
+	/* Restore CPU state */
+	psstate	t1
+	RESUME_RESTORE_STATIC
+	RESUME_RESTORE_REGS_RETURN
+	END(mips_cps_pm_restore)
+
+#endif /* CONFIG_MIPS_CPS_PM && CONFIG_CPU_PM */
diff --git a/arch/mips/kernel/idle.c b/arch/mips/kernel/idle.c
index c4ceccf..09ce459 100644
--- a/arch/mips/kernel/idle.c
+++ b/arch/mips/kernel/idle.c
@@ -236,3 +236,14 @@ void arch_cpu_idle(void)
 	else
 		local_irq_enable();
 }
+
+#ifdef CONFIG_CPU_IDLE
+
+int mips_cpuidle_wait_enter(struct cpuidle_device *dev,
+			    struct cpuidle_driver *drv, int index)
+{
+	arch_cpu_idle();
+	return index;
+}
+
+#endif
diff --git a/arch/mips/kernel/irq-gic.c b/arch/mips/kernel/irq-gic.c
index 8520dad..88e4c32 100644
--- a/arch/mips/kernel/irq-gic.c
+++ b/arch/mips/kernel/irq-gic.c
@@ -54,6 +54,21 @@ void gic_write_compare(cycle_t cnt)
 				(int)(cnt & 0xffffffff));
 }
 
+void gic_write_cpu_compare(cycle_t cnt, int cpu)
+{
+	unsigned long flags;
+
+	local_irq_save(flags);
+
+	GICWRITE(GIC_REG(VPE_LOCAL, GIC_VPE_OTHER_ADDR), cpu);
+	GICWRITE(GIC_REG(VPE_OTHER, GIC_VPE_COMPARE_HI),
+				(int)(cnt >> 32));
+	GICWRITE(GIC_REG(VPE_OTHER, GIC_VPE_COMPARE_LO),
+				(int)(cnt & 0xffffffff));
+
+	local_irq_restore(flags);
+}
+
 cycle_t gic_read_compare(void)
 {
 	unsigned int hi, lo;
diff --git a/arch/mips/kernel/mips-cpc.c b/arch/mips/kernel/mips-cpc.c
index c9dc674..ba47360 100644
--- a/arch/mips/kernel/mips-cpc.c
+++ b/arch/mips/kernel/mips-cpc.c
@@ -9,12 +9,18 @@
  */
 
 #include <linux/errno.h>
+#include <linux/percpu.h>
+#include <linux/spinlock.h>
 
 #include <asm/mips-cm.h>
 #include <asm/mips-cpc.h>
 
 void __iomem *mips_cpc_base;
 
+static DEFINE_PER_CPU_ALIGNED(spinlock_t, cpc_core_lock);
+
+static DEFINE_PER_CPU_ALIGNED(unsigned long, cpc_core_lock_flags);
+
 phys_t __weak mips_cpc_phys_base(void)
 {
 	u32 cpc_base;
@@ -39,6 +45,10 @@ phys_t __weak mips_cpc_phys_base(void)
 int mips_cpc_probe(void)
 {
 	phys_t addr;
+	unsigned cpu;
+
+	for_each_possible_cpu(cpu)
+		spin_lock_init(&per_cpu(cpc_core_lock, cpu));
 
 	addr = mips_cpc_phys_base();
 	if (!addr)
@@ -50,3 +60,21 @@ int mips_cpc_probe(void)
 
 	return 0;
 }
+
+void mips_cpc_lock_other(unsigned int core)
+{
+	unsigned curr_core;
+	preempt_disable();
+	curr_core = current_cpu_data.core;
+	spin_lock_irqsave(&per_cpu(cpc_core_lock, curr_core),
+			  per_cpu(cpc_core_lock_flags, curr_core));
+	write_cpc_cl_other(core << CPC_Cx_OTHER_CORENUM_SHF);
+}
+
+void mips_cpc_unlock_other(void)
+{
+	unsigned curr_core = current_cpu_data.core;
+	spin_unlock_irqrestore(&per_cpu(cpc_core_lock, curr_core),
+			       per_cpu(cpc_core_lock_flags, curr_core));
+	preempt_enable();
+}
diff --git a/arch/mips/kernel/pm-cps.c b/arch/mips/kernel/pm-cps.c
new file mode 100644
index 0000000..5aa4c6f
--- /dev/null
+++ b/arch/mips/kernel/pm-cps.c
@@ -0,0 +1,716 @@
+/*
+ * Copyright (C) 2014 Imagination Technologies
+ * Author: Paul Burton <paul.burton@imgtec.com>
+ *
+ * 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.
+ */
+
+#include <linux/init.h>
+#include <linux/percpu.h>
+#include <linux/slab.h>
+
+#include <asm/asm-offsets.h>
+#include <asm/cacheflush.h>
+#include <asm/cacheops.h>
+#include <asm/idle.h>
+#include <asm/mips-cm.h>
+#include <asm/mips-cpc.h>
+#include <asm/mipsmtregs.h>
+#include <asm/pm.h>
+#include <asm/pm-cps.h>
+#include <asm/smp-cps.h>
+#include <asm/uasm.h>
+
+/*
+ * cps_nc_entry_fn - type of a generated non-coherent state entry function
+ * @online: the count of online coupled VPEs
+ * @nc_ready_count: pointer to a non-coherent mapping of the core ready_count
+ *
+ * The code entering & exiting non-coherent states is generated at runtime
+ * using uasm, in order to ensure that the compiler cannot insert a stray
+ * memory access at an unfortunate time and to allow the generation of optimal
+ * core-specific code particularly for cache routines. If coupled_coherence
+ * is non-zero and this is the entry function for the CPS_PM_NC_WAIT state,
+ * returns the number of VPEs that were in the wait state at the point this
+ * VPE left it. Returns garbage if coupled_coherence is zero or this is not
+ * the entry function for CPS_PM_NC_WAIT.
+ */
+typedef unsigned (*cps_nc_entry_fn)(unsigned online, u32 *nc_ready_count);
+
+/*
+ * The entry point of the generated non-coherent idle state entry/exit
+ * functions. Actually per-core rather than per-CPU.
+ */
+static DEFINE_PER_CPU_READ_MOSTLY(cps_nc_entry_fn[CPS_PM_STATE_COUNT],
+				  nc_asm_enter);
+
+/* Bitmap indicating which states are supported by the system */
+DECLARE_BITMAP(state_support, CPS_PM_STATE_COUNT);
+
+/*
+ * Indicates the number of coupled VPEs ready to operate in a non-coherent
+ * state. Actually per-core rather than per-CPU.
+ */
+static DEFINE_PER_CPU_ALIGNED(u32*, ready_count);
+static DEFINE_PER_CPU_ALIGNED(void*, ready_count_alloc);
+
+/* Indicates online CPUs coupled with the current CPU */
+static DEFINE_PER_CPU_ALIGNED(cpumask_t, online_coupled);
+
+/*
+ * Used to synchronize entry to deep idle states. Actually per-core rather
+ * than per-CPU.
+ */
+static DEFINE_PER_CPU_ALIGNED(atomic_t, pm_barrier);
+
+/* Saved CPU state across the CPS_PM_POWER_GATED state */
+DEFINE_PER_CPU_ALIGNED(struct mips_static_suspend_state, cps_cpu_state);
+
+/* A somewhat arbitrary number of labels & relocs for uasm */
+static struct uasm_label labels[32] __initdata;
+static struct uasm_reloc relocs[32] __initdata;
+
+/* CPU dependant sync types */
+static unsigned stype_intervention;
+static unsigned stype_memory;
+static unsigned stype_ordering;
+
+enum mips_reg {
+	zero, at, v0, v1, a0, a1, a2, a3,
+	t0, t1, t2, t3, t4, t5, t6, t7,
+	s0, s1, s2, s3, s4, s5, s6, s7,
+	t8, t9, k0, k1, gp, sp, fp, ra,
+};
+
+bool cps_pm_support_state(enum cps_pm_state state)
+{
+	return test_bit(state, state_support);
+}
+
+static void coupled_barrier(atomic_t *a, unsigned online)
+{
+	/*
+	 * This function is effectively the same as
+	 * cpuidle_coupled_parallel_barrier, which can't be used here since
+	 * there's no cpuidle device.
+	 */
+
+	if (!coupled_coherence)
+		return;
+
+	smp_mb__before_atomic_inc();
+	atomic_inc(a);
+
+	while (atomic_read(a) < online)
+		cpu_relax();
+
+	if (atomic_inc_return(a) == online * 2) {
+		atomic_set(a, 0);
+		return;
+	}
+
+	while (atomic_read(a) > online)
+		cpu_relax();
+}
+
+int cps_pm_enter_state(enum cps_pm_state state)
+{
+	unsigned cpu = smp_processor_id();
+	unsigned core = current_cpu_data.core;
+	unsigned online, left;
+	cpumask_t *coupled_mask = this_cpu_ptr(&online_coupled);
+	u32 *core_ready_count, *nc_core_ready_count;
+	void *nc_addr;
+	cps_nc_entry_fn entry;
+	struct core_boot_config *core_cfg;
+	struct vpe_boot_config *vpe_cfg;
+
+	/* Check that there is an entry function for this state */
+	entry = per_cpu(nc_asm_enter, core)[state];
+	if (!entry)
+		return -EINVAL;
+
+	/* Calculate which coupled CPUs (VPEs) are online */
+#ifdef CONFIG_MIPS_MT
+	if (cpu_online(cpu)) {
+		cpumask_and(coupled_mask, cpu_online_mask,
+			    &cpu_sibling_map[cpu]);
+		online = cpumask_weight(coupled_mask);
+		cpumask_clear_cpu(cpu, coupled_mask);
+	} else
+#endif
+	{
+		cpumask_clear(coupled_mask);
+		online = 1;
+	}
+
+	/* Setup the VPE to run mips_cps_pm_restore when started again */
+	if (config_enabled(CONFIG_CPU_PM) && state == CPS_PM_POWER_GATED) {
+		core_cfg = &mips_cps_core_bootcfg[core];
+		vpe_cfg = &core_cfg->vpe_config[current_cpu_data.vpe_id];
+		vpe_cfg->pc = (unsigned long)mips_cps_pm_restore;
+		vpe_cfg->gp = (unsigned long)current_thread_info();
+		vpe_cfg->sp = 0;
+	}
+
+	/* Indicate that this CPU might not be coherent */
+	cpumask_clear_cpu(cpu, &cpu_coherent_mask);
+	smp_mb__after_clear_bit();
+
+	/* Create a non-coherent mapping of the core ready_count */
+	core_ready_count = per_cpu(ready_count, core);
+	nc_addr = kmap_noncoherent(virt_to_page(core_ready_count),
+				   (unsigned long)core_ready_count);
+	nc_addr += ((unsigned long)core_ready_count & ~PAGE_MASK);
+	nc_core_ready_count = nc_addr;
+
+	/* Ensure ready_count is zero-initialised before the assembly runs */
+	ACCESS_ONCE(*nc_core_ready_count) = 0;
+	coupled_barrier(&per_cpu(pm_barrier, core), online);
+
+	/* Run the generated entry code */
+	left = entry(online, nc_core_ready_count);
+
+	/* Remove the non-coherent mapping of ready_count */
+	kunmap_noncoherent();
+
+	/* Indicate that this CPU is definitely coherent */
+	cpumask_set_cpu(cpu, &cpu_coherent_mask);
+
+	/*
+	 * If this VPE is the first to leave the non-coherent wait state then
+	 * it needs to wake up any coupled VPEs still running their wait
+	 * instruction so that they return to cpuidle, which can then complete
+	 * coordination between the coupled VPEs & provide the governor with
+	 * a chance to reflect on the length of time the VPEs were in the
+	 * idle state.
+	 */
+	if (coupled_coherence && (state == CPS_PM_NC_WAIT) && (left == online))
+		arch_send_call_function_ipi_mask(coupled_mask);
+
+	return 0;
+}
+
+static void __init cps_gen_cache_routine(u32 **pp, struct uasm_label **pl,
+					 struct uasm_reloc **pr,
+					 const struct cache_desc *cache,
+					 unsigned op, int lbl)
+{
+	unsigned cache_size = cache->ways << cache->waybit;
+	unsigned i;
+	const unsigned unroll_lines = 32;
+
+	/* If the cache isn't present this function has it easy */
+	if (cache->flags & MIPS_CACHE_NOT_PRESENT)
+		return;
+
+	/* Load base address */
+	UASM_i_LA(pp, t0, (long)CKSEG0);
+
+	/* Calculate end address */
+	if (cache_size < 0x8000)
+		uasm_i_addiu(pp, t1, t0, cache_size);
+	else
+		UASM_i_LA(pp, t1, (long)(CKSEG0 + cache_size));
+
+	/* Start of cache op loop */
+	uasm_build_label(pl, *pp, lbl);
+
+	/* Generate the cache ops */
+	for (i = 0; i < unroll_lines; i++)
+		uasm_i_cache(pp, op, i * cache->linesz, t0);
+
+	/* Update the base address */
+	uasm_i_addiu(pp, t0, t0, unroll_lines * cache->linesz);
+
+	/* Loop if we haven't reached the end address yet */
+	uasm_il_bne(pp, pr, t0, t1, lbl);
+	uasm_i_nop(pp);
+}
+
+static int __init cps_gen_flush_fsb(u32 **pp, struct uasm_label **pl,
+				    struct uasm_reloc **pr,
+				    const struct cpuinfo_mips *cpu_info,
+				    int lbl)
+{
+	unsigned i, fsb_size = 8;
+	unsigned num_loads = (fsb_size * 3) / 2;
+	unsigned line_stride = 2;
+	unsigned line_size = cpu_info->dcache.linesz;
+	unsigned perf_counter, perf_event;
+	unsigned revision = cpu_info->processor_id & PRID_REV_MASK;
+
+	/*
+	 * Determine whether this CPU requires an FSB flush, and if so which
+	 * performance counter/event reflect stalls due to a full FSB.
+	 */
+	switch (__get_cpu_type(cpu_info->cputype)) {
+	case CPU_INTERAPTIV:
+		perf_counter = 1;
+		perf_event = 51;
+		break;
+
+	case CPU_PROAPTIV:
+		/* Newer proAptiv cores don't require this workaround */
+		if (revision >= PRID_REV_ENCODE_332(1, 1, 0))
+			return 0;
+
+		/* On older ones it's unavailable */
+		return -1;
+
+	/* CPUs which do not require the workaround */
+	case CPU_P5600:
+		return 0;
+
+	default:
+		WARN_ONCE(1, "pm-cps: FSB flush unsupported for this CPU\n");
+		return -1;
+	}
+
+	/*
+	 * Ensure that the fill/store buffer (FSB) is not holding the results
+	 * of a prefetch, since if it is then the CPC sequencer may become
+	 * stuck in the D3 (ClrBus) state whilst entering a low power state.
+	 */
+
+	/* Preserve perf counter setup */
+	uasm_i_mfc0(pp, t2, 25, (perf_counter * 2) + 0); /* PerfCtlN */
+	uasm_i_mfc0(pp, t3, 25, (perf_counter * 2) + 1); /* PerfCntN */
+
+	/* Setup perf counter to count FSB full pipeline stalls */
+	uasm_i_addiu(pp, t0, zero, (perf_event << 5) | 0xf);
+	uasm_i_mtc0(pp, t0, 25, (perf_counter * 2) + 0); /* PerfCtlN */
+	uasm_i_ehb(pp);
+	uasm_i_mtc0(pp, zero, 25, (perf_counter * 2) + 1); /* PerfCntN */
+	uasm_i_ehb(pp);
+
+	/* Base address for loads */
+	UASM_i_LA(pp, t0, (long)CKSEG0);
+
+	/* Start of clear loop */
+	uasm_build_label(pl, *pp, lbl);
+
+	/* Perform some loads to fill the FSB */
+	for (i = 0; i < num_loads; i++)
+		uasm_i_lw(pp, zero, i * line_size * line_stride, t0);
+
+	/*
+	 * Invalidate the new D-cache entries so that the cache will need
+	 * refilling (via the FSB) if the loop is executed again.
+	 */
+	for (i = 0; i < num_loads; i++) {
+		uasm_i_cache(pp, Hit_Invalidate_D,
+			     i * line_size * line_stride, t0);
+		uasm_i_cache(pp, Hit_Writeback_Inv_SD,
+			     i * line_size * line_stride, t0);
+	}
+
+	/* Completion barrier */
+	uasm_i_sync(pp, stype_memory);
+	uasm_i_ehb(pp);
+
+	/* Check whether the pipeline stalled due to the FSB being full */
+	uasm_i_mfc0(pp, t1, 25, (perf_counter * 2) + 1); /* PerfCntN */
+
+	/* Loop if it didn't */
+	uasm_il_beqz(pp, pr, t1, lbl);
+	uasm_i_nop(pp);
+
+	/* Restore perf counter 1. The count may well now be wrong... */
+	uasm_i_mtc0(pp, t2, 25, (perf_counter * 2) + 0); /* PerfCtlN */
+	uasm_i_ehb(pp);
+	uasm_i_mtc0(pp, t3, 25, (perf_counter * 2) + 1); /* PerfCntN */
+	uasm_i_ehb(pp);
+
+	return 0;
+}
+
+static void __init cps_gen_set_top_bit(u32 **pp, struct uasm_label **pl,
+				       struct uasm_reloc **pr,
+				       unsigned r_addr, int lbl)
+{
+	uasm_i_lui(pp, t0, uasm_rel_hi(0x80000000));
+	uasm_build_label(pl, *pp, lbl);
+	uasm_i_ll(pp, t1, 0, r_addr);
+	uasm_i_or(pp, t1, t1, t0);
+	uasm_i_sc(pp, t1, 0, r_addr);
+	uasm_il_beqz(pp, pr, t1, lbl);
+	uasm_i_nop(pp);
+}
+
+static void * __init cps_gen_entry_code(unsigned cpu, enum cps_pm_state state)
+{
+	struct uasm_label *l = labels;
+	struct uasm_reloc *r = relocs;
+	u32 *buf, *p;
+	const unsigned r_online = a0;
+	const unsigned r_nc_count = a1;
+	const unsigned r_pcohctl = t7;
+	const unsigned max_instrs = 256;
+	unsigned cpc_cmd;
+	int err;
+	enum {
+		lbl_incready = 1,
+		lbl_poll_cont,
+		lbl_secondary_hang,
+		lbl_disable_coherence,
+		lbl_flush_fsb,
+		lbl_invicache,
+		lbl_flushdcache,
+		lbl_hang,
+		lbl_set_cont,
+		lbl_secondary_cont,
+		lbl_decready,
+	};
+
+	/* Allocate a buffer to hold the generated code */
+	p = buf = kcalloc(max_instrs, sizeof(u32), GFP_KERNEL);
+	if (!buf)
+		return NULL;
+
+	/* Clear labels & relocs ready for (re)use */
+	memset(labels, 0, sizeof(labels));
+	memset(relocs, 0, sizeof(relocs));
+
+	if (config_enabled(CONFIG_CPU_PM) && state == CPS_PM_POWER_GATED) {
+		/*
+		 * Save CPU state. Note the non-standard calling convention
+		 * with the return address placed in v0 to avoid clobbering
+		 * the ra register before it is saved.
+		 */
+		UASM_i_LA(&p, t0, (long)mips_cps_pm_save);
+		uasm_i_jalr(&p, v0, t0);
+		uasm_i_nop(&p);
+	}
+
+	/*
+	 * Load addresses of required CM & CPC registers. This is done early
+	 * because they're needed in both the enable & disable coherence steps
+	 * but in the coupled case the enable step will only run on one VPE.
+	 */
+	UASM_i_LA(&p, r_pcohctl, (long)addr_gcr_cl_coherence());
+
+	if (coupled_coherence) {
+		/* Increment ready_count */
+		uasm_i_sync(&p, stype_ordering);
+		uasm_build_label(&l, p, lbl_incready);
+		uasm_i_ll(&p, t1, 0, r_nc_count);
+		uasm_i_addiu(&p, t2, t1, 1);
+		uasm_i_sc(&p, t2, 0, r_nc_count);
+		uasm_il_beqz(&p, &r, t2, lbl_incready);
+		uasm_i_addiu(&p, t1, t1, 1);
+
+		/* Ordering barrier */
+		uasm_i_sync(&p, stype_ordering);
+
+		/*
+		 * If this is the last VPE to become ready for non-coherence
+		 * then it should branch below.
+		 */
+		uasm_il_beq(&p, &r, t1, r_online, lbl_disable_coherence);
+		uasm_i_nop(&p);
+
+		if (state < CPS_PM_POWER_GATED) {
+			/*
+			 * Otherwise this is not the last VPE to become ready
+			 * for non-coherence. It needs to wait until coherence
+			 * has been disabled before proceeding, which it will do
+			 * by polling for the top bit of ready_count being set.
+			 */
+			uasm_i_addiu(&p, t1, zero, -1);
+			uasm_build_label(&l, p, lbl_poll_cont);
+			uasm_i_lw(&p, t0, 0, r_nc_count);
+			uasm_il_bltz(&p, &r, t0, lbl_secondary_cont);
+			uasm_i_ehb(&p);
+			uasm_i_yield(&p, zero, t1);
+			uasm_il_b(&p, &r, lbl_poll_cont);
+			uasm_i_nop(&p);
+		} else {
+			/*
+			 * The core will lose power & this VPE will not continue
+			 * so it can simply halt here.
+			 */
+			uasm_i_addiu(&p, t0, zero, TCHALT_H);
+			uasm_i_mtc0(&p, t0, 2, 4);
+			uasm_build_label(&l, p, lbl_secondary_hang);
+			uasm_il_b(&p, &r, lbl_secondary_hang);
+			uasm_i_nop(&p);
+		}
+	}
+
+	/*
+	 * This is the point of no return - this VPE will now proceed to
+	 * disable coherence. At this point we *must* be sure that no other
+	 * VPE within the core will interfere with the L1 dcache.
+	 */
+	uasm_build_label(&l, p, lbl_disable_coherence);
+
+	/* Invalidate the L1 icache */
+	cps_gen_cache_routine(&p, &l, &r, &cpu_data[cpu].icache,
+			      Index_Invalidate_I, lbl_invicache);
+
+	/* Writeback & invalidate the L1 dcache */
+	cps_gen_cache_routine(&p, &l, &r, &cpu_data[cpu].dcache,
+			      Index_Writeback_Inv_D, lbl_flushdcache);
+
+	/* Completion barrier */
+	uasm_i_sync(&p, stype_memory);
+	uasm_i_ehb(&p);
+
+	/*
+	 * Disable all but self interventions. The load from COHCTL is defined
+	 * by the interAptiv & proAptiv SUMs as ensuring that the operation
+	 * resulting from the preceeding store is complete.
+	 */
+	uasm_i_addiu(&p, t0, zero, 1 << cpu_data[cpu].core);
+	uasm_i_sw(&p, t0, 0, r_pcohctl);
+	uasm_i_lw(&p, t0, 0, r_pcohctl);
+
+	/* Sync to ensure previous interventions are complete */
+	uasm_i_sync(&p, stype_intervention);
+	uasm_i_ehb(&p);
+
+	/* Disable coherence */
+	uasm_i_sw(&p, zero, 0, r_pcohctl);
+	uasm_i_lw(&p, t0, 0, r_pcohctl);
+
+	if (state >= CPS_PM_CLOCK_GATED) {
+		err = cps_gen_flush_fsb(&p, &l, &r, &cpu_data[cpu],
+					lbl_flush_fsb);
+		if (err)
+			goto out_err;
+
+		/* Determine the CPC command to issue */
+		switch (state) {
+		case CPS_PM_CLOCK_GATED:
+			cpc_cmd = CPC_Cx_CMD_CLOCKOFF;
+			break;
+		case CPS_PM_POWER_GATED:
+			cpc_cmd = CPC_Cx_CMD_PWRDOWN;
+			break;
+		default:
+			BUG();
+			goto out_err;
+		}
+
+		/* Issue the CPC command */
+		UASM_i_LA(&p, t0, (long)addr_cpc_cl_cmd());
+		uasm_i_addiu(&p, t1, zero, cpc_cmd);
+		uasm_i_sw(&p, t1, 0, t0);
+
+		if (state == CPS_PM_POWER_GATED) {
+			/* If anything goes wrong just hang */
+			uasm_build_label(&l, p, lbl_hang);
+			uasm_il_b(&p, &r, lbl_hang);
+			uasm_i_nop(&p);
+
+			/*
+			 * There's no point generating more code, the core is
+			 * powered down & if powered back up will run from the
+			 * reset vector not from here.
+			 */
+			goto gen_done;
+		}
+
+		/* Completion barrier */
+		uasm_i_sync(&p, stype_memory);
+		uasm_i_ehb(&p);
+	}
+
+	if (state == CPS_PM_NC_WAIT) {
+		/*
+		 * At this point it is safe for all VPEs to proceed with
+		 * execution. This VPE will set the top bit of ready_count
+		 * to indicate to the other VPEs that they may continue.
+		 */
+		if (coupled_coherence)
+			cps_gen_set_top_bit(&p, &l, &r, r_nc_count,
+					    lbl_set_cont);
+
+		/*
+		 * VPEs which did not disable coherence will continue
+		 * executing, after coherence has been disabled, from this
+		 * point.
+		 */
+		uasm_build_label(&l, p, lbl_secondary_cont);
+
+		/* Now perform our wait */
+		uasm_i_wait(&p, 0);
+	}
+
+	/*
+	 * Re-enable coherence. Note that for CPS_PM_NC_WAIT all coupled VPEs
+	 * will run this. The first will actually re-enable coherence & the
+	 * rest will just be performing a rather unusual nop.
+	 */
+	uasm_i_addiu(&p, t0, zero, CM_GCR_Cx_COHERENCE_COHDOMAINEN_MSK);
+	uasm_i_sw(&p, t0, 0, r_pcohctl);
+	uasm_i_lw(&p, t0, 0, r_pcohctl);
+
+	/* Completion barrier */
+	uasm_i_sync(&p, stype_memory);
+	uasm_i_ehb(&p);
+
+	if (coupled_coherence && (state == CPS_PM_NC_WAIT)) {
+		/* Decrement ready_count */
+		uasm_build_label(&l, p, lbl_decready);
+		uasm_i_sync(&p, stype_ordering);
+		uasm_i_ll(&p, t1, 0, r_nc_count);
+		uasm_i_addiu(&p, t2, t1, -1);
+		uasm_i_sc(&p, t2, 0, r_nc_count);
+		uasm_il_beqz(&p, &r, t2, lbl_decready);
+		uasm_i_andi(&p, v0, t1, (1 << fls(smp_num_siblings)) - 1);
+
+		/* Ordering barrier */
+		uasm_i_sync(&p, stype_ordering);
+	}
+
+	if (coupled_coherence && (state == CPS_PM_CLOCK_GATED)) {
+		/*
+		 * At this point it is safe for all VPEs to proceed with
+		 * execution. This VPE will set the top bit of ready_count
+		 * to indicate to the other VPEs that they may continue.
+		 */
+		cps_gen_set_top_bit(&p, &l, &r, r_nc_count, lbl_set_cont);
+
+		/*
+		 * This core will be reliant upon another core sending a
+		 * power-up command to the CPC in order to resume operation.
+		 * Thus an arbitrary VPE can't trigger the core leaving the
+		 * idle state and the one that disables coherence might as well
+		 * be the one to re-enable it. The rest will continue from here
+		 * after that has been done.
+		 */
+		uasm_build_label(&l, p, lbl_secondary_cont);
+
+		/* Ordering barrier */
+		uasm_i_sync(&p, stype_ordering);
+	}
+
+	/* The core is coherent, time to return to C code */
+	uasm_i_jr(&p, ra);
+	uasm_i_nop(&p);
+
+gen_done:
+	/* Ensure the code didn't exceed the resources allocated for it */
+	BUG_ON((p - buf) > max_instrs);
+	BUG_ON((l - labels) > ARRAY_SIZE(labels));
+	BUG_ON((r - relocs) > ARRAY_SIZE(relocs));
+
+	/* Patch branch offsets */
+	uasm_resolve_relocs(relocs, labels);
+
+	/* Flush the icache */
+	local_flush_icache_range((unsigned long)buf, (unsigned long)p);
+
+	return buf;
+out_err:
+	kfree(buf);
+	return NULL;
+}
+
+static int __init cps_gen_core_entries(unsigned cpu)
+{
+	enum cps_pm_state state;
+	unsigned core = cpu_data[cpu].core;
+	unsigned dlinesz = cpu_data[cpu].dcache.linesz;
+	void *entry_fn, *core_rc;
+
+	for (state = CPS_PM_NC_WAIT; state < CPS_PM_STATE_COUNT; state++) {
+		if (per_cpu(nc_asm_enter, core)[state])
+			continue;
+		if (!test_bit(state, state_support))
+			continue;
+
+		entry_fn = cps_gen_entry_code(cpu, state);
+		if (!entry_fn) {
+			pr_err("Failed to generate core %u state %u entry\n",
+			       core, state);
+			clear_bit(state, state_support);
+		}
+
+		per_cpu(nc_asm_enter, core)[state] = entry_fn;
+	}
+
+	if (!per_cpu(ready_count, core)) {
+		core_rc = kmalloc(dlinesz * 2, GFP_KERNEL);
+		if (!core_rc) {
+			pr_err("Failed allocate core %u ready_count\n", core);
+			return -ENOMEM;
+		}
+		per_cpu(ready_count_alloc, core) = core_rc;
+
+		/* Ensure ready_count is aligned to a cacheline boundary */
+		core_rc += dlinesz - 1;
+		core_rc = (void *)((unsigned long)core_rc & ~(dlinesz - 1));
+		per_cpu(ready_count, core) = core_rc;
+	}
+
+	return 0;
+}
+
+static int __init cps_pm_init(void)
+{
+	unsigned cpu;
+	int err;
+
+	/* Detect appropriate sync types for the system */
+	switch (current_cpu_data.cputype) {
+	case CPU_INTERAPTIV:
+	case CPU_PROAPTIV:
+	case CPU_M5150:
+	case CPU_P5600:
+		stype_intervention = 0x2;
+		stype_memory = 0x3;
+		stype_ordering = 0x10;
+		break;
+
+	default:
+		pr_warn("Power management is using heavyweight sync 0\n");
+	}
+
+	/* A CM is required for all non-coherent states */
+	if (!mips_cm_present()) {
+		pr_warn("pm-cps: no CM, non-coherent states unavailable\n");
+		goto out;
+	}
+
+	/*
+	 * If interrupts were enabled whilst running a wait instruction on a
+	 * non-coherent core then the VPE may end up processing interrupts
+	 * whilst non-coherent. That would be bad.
+	 */
+	if (cpu_wait == r4k_wait_irqoff)
+		set_bit(CPS_PM_NC_WAIT, state_support);
+	else
+		pr_warn("pm-cps: non-coherent wait unavailable\n");
+
+	/* Detect whether a CPC is present */
+	if (mips_cpc_present()) {
+		/* Detect whether clock gating is implemented */
+		if (read_cpc_cl_stat_conf() & CPC_Cx_STAT_CONF_CLKGAT_IMPL_MSK)
+			set_bit(CPS_PM_CLOCK_GATED, state_support);
+		else
+			pr_warn("pm-cps: CPC does not support clock gating\n");
+
+		/* Power gating is available with CPS SMP & any CPC */
+		if (mips_cps_smp_in_use())
+			set_bit(CPS_PM_POWER_GATED, state_support);
+		else
+			pr_warn("pm-cps: CPS SMP not in use, power gating unavailable\n");
+	} else {
+		pr_warn("pm-cps: no CPC, clock & power gating unavailable\n");
+	}
+
+	for_each_present_cpu(cpu) {
+		err = cps_gen_core_entries(cpu);
+		if (err)
+			return err;
+	}
+out:
+	return 0;
+}
+arch_initcall(cps_pm_init);
diff --git a/arch/mips/kernel/pm.c b/arch/mips/kernel/pm.c
new file mode 100644
index 0000000..fefdf39
--- /dev/null
+++ b/arch/mips/kernel/pm.c
@@ -0,0 +1,99 @@
+/*
+ * Copyright (C) 2014 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.
+ *
+ * CPU PM notifiers for saving/restoring general CPU state.
+ */
+
+#include <linux/cpu_pm.h>
+#include <linux/init.h>
+
+#include <asm/dsp.h>
+#include <asm/fpu.h>
+#include <asm/mmu_context.h>
+#include <asm/pm.h>
+#include <asm/watch.h>
+
+/* Used by PM helper macros in asm/pm.h */
+struct mips_static_suspend_state mips_static_suspend_state;
+
+/**
+ * mips_cpu_save() - Save general CPU state.
+ * Ensures that general CPU context is saved, notably FPU and DSP.
+ */
+static int mips_cpu_save(void)
+{
+	/* Save FPU state */
+	lose_fpu(1);
+
+	/* Save DSP state */
+	save_dsp(current);
+
+	return 0;
+}
+
+/**
+ * mips_cpu_restore() - Restore general CPU state.
+ * Restores important CPU context.
+ */
+static void mips_cpu_restore(void)
+{
+	unsigned int cpu = smp_processor_id();
+
+	/* Restore ASID */
+	if (current->mm)
+		write_c0_entryhi(cpu_asid(cpu, current->mm));
+
+	/* Restore DSP state */
+	restore_dsp(current);
+
+	/* Restore UserLocal */
+	if (cpu_has_userlocal)
+		write_c0_userlocal(current_thread_info()->tp_value);
+
+	/* Restore watch registers */
+	__restore_watch();
+}
+
+/**
+ * mips_pm_notifier() - Notifier for preserving general CPU context.
+ * @self:	Notifier block.
+ * @cmd:	CPU PM event.
+ * @v:		Private data (unused).
+ *
+ * This is called when a CPU power management event occurs, and is used to
+ * ensure that important CPU context is preserved across a CPU power down.
+ */
+static int mips_pm_notifier(struct notifier_block *self, unsigned long cmd,
+			    void *v)
+{
+	int ret;
+
+	switch (cmd) {
+	case CPU_PM_ENTER:
+		ret = mips_cpu_save();
+		if (ret)
+			return NOTIFY_STOP;
+		break;
+	case CPU_PM_ENTER_FAILED:
+	case CPU_PM_EXIT:
+		mips_cpu_restore();
+		break;
+	}
+
+	return NOTIFY_OK;
+}
+
+static struct notifier_block mips_pm_notifier_block = {
+	.notifier_call = mips_pm_notifier,
+};
+
+static int __init mips_pm_init(void)
+{
+	return cpu_pm_register_notifier(&mips_pm_notifier_block);
+}
+arch_initcall(mips_pm_init);
diff --git a/arch/mips/kernel/smp-cps.c b/arch/mips/kernel/smp-cps.c
index bb36b4e..df0598d 100644
--- a/arch/mips/kernel/smp-cps.c
+++ b/arch/mips/kernel/smp-cps.c
@@ -20,104 +20,43 @@
 #include <asm/mips-cpc.h>
 #include <asm/mips_mt.h>
 #include <asm/mipsregs.h>
+#include <asm/pm-cps.h>
 #include <asm/smp-cps.h>
 #include <asm/time.h>
 #include <asm/uasm.h>
 
 static DECLARE_BITMAP(core_power, NR_CPUS);
 
-struct boot_config mips_cps_bootcfg;
+struct core_boot_config *mips_cps_core_bootcfg;
 
-static void init_core(void)
+static unsigned core_vpe_count(unsigned core)
 {
-	unsigned int nvpes, t;
-	u32 mvpconf0, vpeconf0, vpecontrol, tcstatus, tcbind, status;
+	unsigned cfg;
 
-	if (!cpu_has_mipsmt)
-		return;
-
-	/* Enter VPE configuration state */
-	dvpe();
-	set_c0_mvpcontrol(MVPCONTROL_VPC);
-
-	/* Retrieve the count of VPEs in this core */
-	mvpconf0 = read_c0_mvpconf0();
-	nvpes = ((mvpconf0 & MVPCONF0_PVPE) >> MVPCONF0_PVPE_SHIFT) + 1;
-	smp_num_siblings = nvpes;
-
-	for (t = 1; t < nvpes; t++) {
-		/* Use a 1:1 mapping of TC index to VPE index */
-		settc(t);
-
-		/* Bind 1 TC to this VPE */
-		tcbind = read_tc_c0_tcbind();
-		tcbind &= ~TCBIND_CURVPE;
-		tcbind |= t << TCBIND_CURVPE_SHIFT;
-		write_tc_c0_tcbind(tcbind);
-
-		/* Set exclusive TC, non-active, master */
-		vpeconf0 = read_vpe_c0_vpeconf0();
-		vpeconf0 &= ~(VPECONF0_XTC | VPECONF0_VPA);
-		vpeconf0 |= t << VPECONF0_XTC_SHIFT;
-		vpeconf0 |= VPECONF0_MVP;
-		write_vpe_c0_vpeconf0(vpeconf0);
-
-		/* Declare TC non-active, non-allocatable & interrupt exempt */
-		tcstatus = read_tc_c0_tcstatus();
-		tcstatus &= ~(TCSTATUS_A | TCSTATUS_DA);
-		tcstatus |= TCSTATUS_IXMT;
-		write_tc_c0_tcstatus(tcstatus);
-
-		/* Halt the TC */
-		write_tc_c0_tchalt(TCHALT_H);
-
-		/* Allow only 1 TC to execute */
-		vpecontrol = read_vpe_c0_vpecontrol();
-		vpecontrol &= ~VPECONTROL_TE;
-		write_vpe_c0_vpecontrol(vpecontrol);
-
-		/* Copy (most of) Status from VPE 0 */
-		status = read_c0_status();
-		status &= ~(ST0_IM | ST0_IE | ST0_KSU);
-		status |= ST0_CU0;
-		write_vpe_c0_status(status);
-
-		/* Copy Config from VPE 0 */
-		write_vpe_c0_config(read_c0_config());
-		write_vpe_c0_config7(read_c0_config7());
-
-		/* Ensure no software interrupts are pending */
-		write_vpe_c0_cause(0);
-
-		/* Sync Count */
-		write_vpe_c0_count(read_c0_count());
-	}
+	if (!config_enabled(CONFIG_MIPS_MT_SMP) || !cpu_has_mipsmt)
+		return 1;
 
-	/* Leave VPE configuration state */
-	clear_c0_mvpcontrol(MVPCONTROL_VPC);
+	write_gcr_cl_other(core << CM_GCR_Cx_OTHER_CORENUM_SHF);
+	cfg = read_gcr_co_config() & CM_GCR_Cx_CONFIG_PVPE_MSK;
+	return (cfg >> CM_GCR_Cx_CONFIG_PVPE_SHF) + 1;
 }
 
 static void __init cps_smp_setup(void)
 {
 	unsigned int ncores, nvpes, core_vpes;
 	int c, v;
-	u32 core_cfg, *entry_code;
 
 	/* Detect & record VPE topology */
 	ncores = mips_cm_numcores();
 	pr_info("VPE topology ");
 	for (c = nvpes = 0; c < ncores; c++) {
-		if (cpu_has_mipsmt && config_enabled(CONFIG_MIPS_MT_SMP)) {
-			write_gcr_cl_other(c << CM_GCR_Cx_OTHER_CORENUM_SHF);
-			core_cfg = read_gcr_co_config();
-			core_vpes = ((core_cfg & CM_GCR_Cx_CONFIG_PVPE_MSK) >>
-				     CM_GCR_Cx_CONFIG_PVPE_SHF) + 1;
-		} else {
-			core_vpes = 1;
-		}
-
+		core_vpes = core_vpe_count(c);
 		pr_cont("%c%u", c ? ',' : '{', core_vpes);
 
+		/* Use the number of VPEs in core 0 for smp_num_siblings */
+		if (!c)
+			smp_num_siblings = core_vpes;
+
 		for (v = 0; v < min_t(int, core_vpes, NR_CPUS - nvpes); v++) {
 			cpu_data[nvpes + v].core = c;
 #ifdef CONFIG_MIPS_MT_SMP
@@ -137,19 +76,14 @@ static void __init cps_smp_setup(void)
 		__cpu_logical_map[v] = v;
 	}
 
+	/* Set a coherent default CCA (CWB) */
+	change_c0_config(CONF_CM_CMASK, 0x5);
+
 	/* Core 0 is powered up (we're running on it) */
 	bitmap_set(core_power, 0, 1);
 
-	/* Disable MT - we only want to run 1 TC per VPE */
-	if (cpu_has_mipsmt)
-		dmt();
-
 	/* Initialise core 0 */
-	init_core();
-
-	/* Patch the start of mips_cps_core_entry to provide the CM base */
-	entry_code = (u32 *)&mips_cps_core_entry;
-	UASM_i_LA(&entry_code, 3, (long)mips_cm_base);
+	mips_cps_core_init();
 
 	/* Make core 0 coherent with everything */
 	write_gcr_cl_coherence(0xff);
@@ -157,15 +91,99 @@ static void __init cps_smp_setup(void)
 
 static void __init cps_prepare_cpus(unsigned int max_cpus)
 {
+	unsigned ncores, core_vpes, c, cca;
+	bool cca_unsuitable;
+	u32 *entry_code;
+
 	mips_mt_set_cpuoptions();
+
+	/* Detect whether the CCA is unsuited to multi-core SMP */
+	cca = read_c0_config() & CONF_CM_CMASK;
+	switch (cca) {
+	case 0x4: /* CWBE */
+	case 0x5: /* CWB */
+		/* The CCA is coherent, multi-core is fine */
+		cca_unsuitable = false;
+		break;
+
+	default:
+		/* CCA is not coherent, multi-core is not usable */
+		cca_unsuitable = true;
+	}
+
+	/* Warn the user if the CCA prevents multi-core */
+	ncores = mips_cm_numcores();
+	if (cca_unsuitable && ncores > 1) {
+		pr_warn("Using only one core due to unsuitable CCA 0x%x\n",
+			cca);
+
+		for_each_present_cpu(c) {
+			if (cpu_data[c].core)
+				set_cpu_present(c, false);
+		}
+	}
+
+	/*
+	 * Patch the start of mips_cps_core_entry to provide:
+	 *
+	 * v0 = CM base address
+	 * s0 = kseg0 CCA
+	 */
+	entry_code = (u32 *)&mips_cps_core_entry;
+	UASM_i_LA(&entry_code, 3, (long)mips_cm_base);
+	uasm_i_addiu(&entry_code, 16, 0, cca);
+	dma_cache_wback_inv((unsigned long)&mips_cps_core_entry,
+			    (void *)entry_code - (void *)&mips_cps_core_entry);
+
+	/* Allocate core boot configuration structs */
+	mips_cps_core_bootcfg = kcalloc(ncores, sizeof(*mips_cps_core_bootcfg),
+					GFP_KERNEL);
+	if (!mips_cps_core_bootcfg) {
+		pr_err("Failed to allocate boot config for %u cores\n", ncores);
+		goto err_out;
+	}
+
+	/* Allocate VPE boot configuration structs */
+	for (c = 0; c < ncores; c++) {
+		core_vpes = core_vpe_count(c);
+		mips_cps_core_bootcfg[c].vpe_config = kcalloc(core_vpes,
+				sizeof(*mips_cps_core_bootcfg[c].vpe_config),
+				GFP_KERNEL);
+		if (!mips_cps_core_bootcfg[c].vpe_config) {
+			pr_err("Failed to allocate %u VPE boot configs\n",
+			       core_vpes);
+			goto err_out;
+		}
+	}
+
+	/* Mark this CPU as booted */
+	atomic_set(&mips_cps_core_bootcfg[current_cpu_data.core].vpe_mask,
+		   1 << cpu_vpe_id(&current_cpu_data));
+
+	return;
+err_out:
+	/* Clean up allocations */
+	if (mips_cps_core_bootcfg) {
+		for (c = 0; c < ncores; c++)
+			kfree(mips_cps_core_bootcfg[c].vpe_config);
+		kfree(mips_cps_core_bootcfg);
+		mips_cps_core_bootcfg = NULL;
+	}
+
+	/* Effectively disable SMP by declaring CPUs not present */
+	for_each_possible_cpu(c) {
+		if (c == 0)
+			continue;
+		set_cpu_present(c, false);
+	}
 }
 
-static void boot_core(struct boot_config *cfg)
+static void boot_core(unsigned core)
 {
 	u32 access;
 
 	/* Select the appropriate core */
-	write_gcr_cl_other(cfg->core << CM_GCR_Cx_OTHER_CORENUM_SHF);
+	write_gcr_cl_other(core << CM_GCR_Cx_OTHER_CORENUM_SHF);
 
 	/* Set its reset vector */
 	write_gcr_co_reset_base(CKSEG1ADDR((unsigned long)mips_cps_core_entry));
@@ -175,104 +193,74 @@ static void boot_core(struct boot_config *cfg)
 
 	/* Ensure the core can access the GCRs */
 	access = read_gcr_access();
-	access |= 1 << (CM_GCR_ACCESS_ACCESSEN_SHF + cfg->core);
+	access |= 1 << (CM_GCR_ACCESS_ACCESSEN_SHF + core);
 	write_gcr_access(access);
 
-	/* Copy cfg */
-	mips_cps_bootcfg = *cfg;
-
 	if (mips_cpc_present()) {
-		/* Select the appropriate core */
-		write_cpc_cl_other(cfg->core << CPC_Cx_OTHER_CORENUM_SHF);
-
 		/* Reset the core */
+		mips_cpc_lock_other(core);
 		write_cpc_co_cmd(CPC_Cx_CMD_RESET);
+		mips_cpc_unlock_other();
 	} else {
 		/* Take the core out of reset */
 		write_gcr_co_reset_release(0);
 	}
 
 	/* The core is now powered up */
-	bitmap_set(core_power, cfg->core, 1);
+	bitmap_set(core_power, core, 1);
 }
 
-static void boot_vpe(void *info)
+static void remote_vpe_boot(void *dummy)
 {
-	struct boot_config *cfg = info;
-	u32 tcstatus, vpeconf0;
-
-	/* Enter VPE configuration state */
-	dvpe();
-	set_c0_mvpcontrol(MVPCONTROL_VPC);
-
-	settc(cfg->vpe);
-
-	/* Set the TC restart PC */
-	write_tc_c0_tcrestart((unsigned long)&smp_bootstrap);
-
-	/* Activate the TC, allow interrupts */
-	tcstatus = read_tc_c0_tcstatus();
-	tcstatus &= ~TCSTATUS_IXMT;
-	tcstatus |= TCSTATUS_A;
-	write_tc_c0_tcstatus(tcstatus);
-
-	/* Clear the TC halt bit */
-	write_tc_c0_tchalt(0);
-
-	/* Activate the VPE */
-	vpeconf0 = read_vpe_c0_vpeconf0();
-	vpeconf0 |= VPECONF0_VPA;
-	write_vpe_c0_vpeconf0(vpeconf0);
-
-	/* Set the stack & global pointer registers */
-	write_tc_gpr_sp(cfg->sp);
-	write_tc_gpr_gp(cfg->gp);
-
-	/* Leave VPE configuration state */
-	clear_c0_mvpcontrol(MVPCONTROL_VPC);
-
-	/* Enable other VPEs to execute */
-	evpe(EVPE_ENABLE);
+	mips_cps_boot_vpes();
 }
 
 static void cps_boot_secondary(int cpu, struct task_struct *idle)
 {
-	struct boot_config cfg;
+	unsigned core = cpu_data[cpu].core;
+	unsigned vpe_id = cpu_vpe_id(&cpu_data[cpu]);
+	struct core_boot_config *core_cfg = &mips_cps_core_bootcfg[core];
+	struct vpe_boot_config *vpe_cfg = &core_cfg->vpe_config[vpe_id];
 	unsigned int remote;
 	int err;
 
-	cfg.core = cpu_data[cpu].core;
-	cfg.vpe = cpu_vpe_id(&cpu_data[cpu]);
-	cfg.pc = (unsigned long)&smp_bootstrap;
-	cfg.sp = __KSTK_TOS(idle);
-	cfg.gp = (unsigned long)task_thread_info(idle);
+	vpe_cfg->pc = (unsigned long)&smp_bootstrap;
+	vpe_cfg->sp = __KSTK_TOS(idle);
+	vpe_cfg->gp = (unsigned long)task_thread_info(idle);
+
+	atomic_or(1 << cpu_vpe_id(&cpu_data[cpu]), &core_cfg->vpe_mask);
+
+	preempt_disable();
 
-	if (!test_bit(cfg.core, core_power)) {
+	if (!test_bit(core, core_power)) {
 		/* Boot a VPE on a powered down core */
-		boot_core(&cfg);
-		return;
+		boot_core(core);
+		goto out;
 	}
 
-	if (cfg.core != current_cpu_data.core) {
+	if (core != current_cpu_data.core) {
 		/* Boot a VPE on another powered up core */
 		for (remote = 0; remote < NR_CPUS; remote++) {
-			if (cpu_data[remote].core != cfg.core)
+			if (cpu_data[remote].core != core)
 				continue;
 			if (cpu_online(remote))
 				break;
 		}
 		BUG_ON(remote >= NR_CPUS);
 
-		err = smp_call_function_single(remote, boot_vpe, &cfg, 1);
+		err = smp_call_function_single(remote, remote_vpe_boot,
+					       NULL, 1);
 		if (err)
 			panic("Failed to call remote CPU\n");
-		return;
+		goto out;
 	}
 
 	BUG_ON(!cpu_has_mipsmt);
 
 	/* Boot a VPE on this core */
-	boot_vpe(&cfg);
+	mips_cps_boot_vpes();
+out:
+	preempt_enable();
 }
 
 static void cps_init_secondary(void)
@@ -281,10 +269,6 @@ static void cps_init_secondary(void)
 	if (cpu_has_mipsmt)
 		dmt();
 
-	/* TODO: revisit this assumption once hotplug is implemented */
-	if (cpu_vpe_id(&current_cpu_data) == 0)
-		init_core();
-
 	change_c0_status(ST0_IM, STATUSF_IP3 | STATUSF_IP4 |
 				 STATUSF_IP6 | STATUSF_IP7);
 }
@@ -302,6 +286,148 @@ static void cps_smp_finish(void)
 	local_irq_enable();
 }
 
+#ifdef CONFIG_HOTPLUG_CPU
+
+static int cps_cpu_disable(void)
+{
+	unsigned cpu = smp_processor_id();
+	struct core_boot_config *core_cfg;
+
+	if (!cpu)
+		return -EBUSY;
+
+	if (!cps_pm_support_state(CPS_PM_POWER_GATED))
+		return -EINVAL;
+
+	core_cfg = &mips_cps_core_bootcfg[current_cpu_data.core];
+	atomic_sub(1 << cpu_vpe_id(&current_cpu_data), &core_cfg->vpe_mask);
+	smp_mb__after_atomic_dec();
+	set_cpu_online(cpu, false);
+	cpu_clear(cpu, cpu_callin_map);
+
+	return 0;
+}
+
+static DECLARE_COMPLETION(cpu_death_chosen);
+static unsigned cpu_death_sibling;
+static enum {
+	CPU_DEATH_HALT,
+	CPU_DEATH_POWER,
+} cpu_death;
+
+void play_dead(void)
+{
+	unsigned cpu, core;
+
+	local_irq_disable();
+	idle_task_exit();
+	cpu = smp_processor_id();
+	cpu_death = CPU_DEATH_POWER;
+
+	if (cpu_has_mipsmt) {
+		core = cpu_data[cpu].core;
+
+		/* Look for another online VPE within the core */
+		for_each_online_cpu(cpu_death_sibling) {
+			if (cpu_data[cpu_death_sibling].core != core)
+				continue;
+
+			/*
+			 * There is an online VPE within the core. Just halt
+			 * this TC and leave the core alone.
+			 */
+			cpu_death = CPU_DEATH_HALT;
+			break;
+		}
+	}
+
+	/* This CPU has chosen its way out */
+	complete(&cpu_death_chosen);
+
+	if (cpu_death == CPU_DEATH_HALT) {
+		/* Halt this TC */
+		write_c0_tchalt(TCHALT_H);
+		instruction_hazard();
+	} else {
+		/* Power down the core */
+		cps_pm_enter_state(CPS_PM_POWER_GATED);
+	}
+
+	/* This should never be reached */
+	panic("Failed to offline CPU %u", cpu);
+}
+
+static void wait_for_sibling_halt(void *ptr_cpu)
+{
+	unsigned cpu = (unsigned)ptr_cpu;
+	unsigned vpe_id = cpu_data[cpu].vpe_id;
+	unsigned halted;
+	unsigned long flags;
+
+	do {
+		local_irq_save(flags);
+		settc(vpe_id);
+		halted = read_tc_c0_tchalt();
+		local_irq_restore(flags);
+	} while (!(halted & TCHALT_H));
+}
+
+static void cps_cpu_die(unsigned int cpu)
+{
+	unsigned core = cpu_data[cpu].core;
+	unsigned stat;
+	int err;
+
+	/* Wait for the cpu to choose its way out */
+	if (!wait_for_completion_timeout(&cpu_death_chosen,
+					 msecs_to_jiffies(5000))) {
+		pr_err("CPU%u: didn't offline\n", cpu);
+		return;
+	}
+
+	/*
+	 * Now wait for the CPU to actually offline. Without doing this that
+	 * offlining may race with one or more of:
+	 *
+	 *   - Onlining the CPU again.
+	 *   - Powering down the core if another VPE within it is offlined.
+	 *   - A sibling VPE entering a non-coherent state.
+	 *
+	 * In the non-MT halt case (ie. infinite loop) the CPU is doing nothing
+	 * with which we could race, so do nothing.
+	 */
+	if (cpu_death == CPU_DEATH_POWER) {
+		/*
+		 * Wait for the core to enter a powered down or clock gated
+		 * state, the latter happening when a JTAG probe is connected
+		 * in which case the CPC will refuse to power down the core.
+		 */
+		do {
+			mips_cpc_lock_other(core);
+			stat = read_cpc_co_stat_conf();
+			stat &= CPC_Cx_STAT_CONF_SEQSTATE_MSK;
+			mips_cpc_unlock_other();
+		} while (stat != CPC_Cx_STAT_CONF_SEQSTATE_D0 &&
+			 stat != CPC_Cx_STAT_CONF_SEQSTATE_D2 &&
+			 stat != CPC_Cx_STAT_CONF_SEQSTATE_U2);
+
+		/* Indicate the core is powered off */
+		bitmap_clear(core_power, core, 1);
+	} else if (cpu_has_mipsmt) {
+		/*
+		 * Have a CPU with access to the offlined CPUs registers wait
+		 * for its TC to halt.
+		 */
+		err = smp_call_function_single(cpu_death_sibling,
+					       wait_for_sibling_halt,
+					       (void *)cpu, 1);
+		if (err)
+			panic("Failed to call remote sibling CPU\n");
+	}
+}
+
+#endif /* CONFIG_HOTPLUG_CPU */
+
 static struct plat_smp_ops cps_smp_ops = {
 	.smp_setup		= cps_smp_setup,
 	.prepare_cpus		= cps_prepare_cpus,
@@ -310,8 +436,18 @@ static struct plat_smp_ops cps_smp_ops = {
 	.smp_finish		= cps_smp_finish,
 	.send_ipi_single	= gic_send_ipi_single,
 	.send_ipi_mask		= gic_send_ipi_mask,
+#ifdef CONFIG_HOTPLUG_CPU
+	.cpu_disable		= cps_cpu_disable,
+	.cpu_die		= cps_cpu_die,
+#endif
 };
 
+bool mips_cps_smp_in_use(void)
+{
+	extern struct plat_smp_ops *mp_ops;
+	return mp_ops == &cps_smp_ops;
+}
+
 int register_cps_smp_ops(void)
 {
 	if (!mips_cm_present()) {
diff --git a/arch/mips/kernel/smp-gic.c b/arch/mips/kernel/smp-gic.c
index 3bb1f92..3b21a96 100644
--- a/arch/mips/kernel/smp-gic.c
+++ b/arch/mips/kernel/smp-gic.c
@@ -15,12 +15,14 @@
 #include <linux/printk.h>
 
 #include <asm/gic.h>
+#include <asm/mips-cpc.h>
 #include <asm/smp-ops.h>
 
 void gic_send_ipi_single(int cpu, unsigned int action)
 {
 	unsigned long flags;
 	unsigned int intr;
+	unsigned int core = cpu_data[cpu].core;
 
 	pr_debug("CPU%d: %s cpu %d action %u status %08x\n",
 		 smp_processor_id(), __func__, cpu, action, read_c0_status());
@@ -41,6 +43,15 @@ void gic_send_ipi_single(int cpu, unsigned int action)
 	}
 
 	gic_send_ipi(intr);
+
+	if (mips_cpc_present() && (core != current_cpu_data.core)) {
+		while (!cpumask_test_cpu(cpu, &cpu_coherent_mask)) {
+			mips_cpc_lock_other(core);
+			write_cpc_co_cmd(CPC_Cx_CMD_PWRUP);
+			mips_cpc_unlock_other();
+		}
+	}
+
 	local_irq_restore(flags);
 }
 
diff --git a/arch/mips/kernel/smp.c b/arch/mips/kernel/smp.c
index ce76775..9bad52e 100644
--- a/arch/mips/kernel/smp.c
+++ b/arch/mips/kernel/smp.c
@@ -62,6 +62,8 @@ EXPORT_SYMBOL(cpu_sibling_map);
 /* representing cpus for which sibling maps can be computed */
 static cpumask_t cpu_sibling_setup_map;
 
+cpumask_t cpu_coherent_mask;
+
 static inline void set_cpu_sibling_map(int cpu)
 {
 	int i;
@@ -114,6 +116,7 @@ asmlinkage void start_secondary(void)
 	cpu = smp_processor_id();
 	cpu_data[cpu].udelay_val = loops_per_jiffy;
 
+	cpu_set(cpu, cpu_coherent_mask);
 	notify_cpu_starting(cpu);
 
 	set_cpu_online(cpu, true);
@@ -175,6 +178,7 @@ void __init smp_prepare_cpus(unsigned int max_cpus)
 #ifndef CONFIG_HOTPLUG_CPU
 	init_cpu_present(cpu_possible_mask);
 #endif
+	cpumask_copy(&cpu_coherent_mask, cpu_possible_mask);
 }
 
 /* preload SMP state for boot cpu */
@@ -390,3 +394,46 @@ void dump_send_ipi(void (*dump_ipi_callback)(void *))
 }
 EXPORT_SYMBOL(dump_send_ipi);
 #endif
+
+#ifdef CONFIG_GENERIC_CLOCKEVENTS_BROADCAST
+
+static DEFINE_PER_CPU(atomic_t, tick_broadcast_count);
+static DEFINE_PER_CPU(struct call_single_data, tick_broadcast_csd);
+
+void tick_broadcast(const struct cpumask *mask)
+{
+	atomic_t *count;
+	struct call_single_data *csd;
+	int cpu;
+
+	for_each_cpu(cpu, mask) {
+		count = &per_cpu(tick_broadcast_count, cpu);
+		csd = &per_cpu(tick_broadcast_csd, cpu);
+
+		if (atomic_inc_return(count) == 1)
+			smp_call_function_single_async(cpu, csd);
+	}
+}
+
+static void tick_broadcast_callee(void *info)
+{
+	int cpu = smp_processor_id();
+	tick_receive_broadcast();
+	atomic_set(&per_cpu(tick_broadcast_count, cpu), 0);
+}
+
+static int __init tick_broadcast_init(void)
+{
+	struct call_single_data *csd;
+	int cpu;
+
+	for (cpu = 0; cpu < NR_CPUS; cpu++) {
+		csd = &per_cpu(tick_broadcast_csd, cpu);
+		csd->func = tick_broadcast_callee;
+	}
+
+	return 0;
+}
+early_initcall(tick_broadcast_init);
+
+#endif /* CONFIG_GENERIC_CLOCKEVENTS_BROADCAST */
diff --git a/arch/mips/kernel/traps.c b/arch/mips/kernel/traps.c
index 3a26729..1fd1a0c 100644
--- a/arch/mips/kernel/traps.c
+++ b/arch/mips/kernel/traps.c
@@ -15,6 +15,7 @@
 #include <linux/bug.h>
 #include <linux/compiler.h>
 #include <linux/context_tracking.h>
+#include <linux/cpu_pm.h>
 #include <linux/kexec.h>
 #include <linux/init.h>
 #include <linux/kernel.h>
@@ -1837,18 +1838,16 @@ static int __init ulri_disable(char *s)
 }
 __setup("noulri", ulri_disable);
 
-void per_cpu_trap_init(bool is_boot_cpu)
+/* configure STATUS register */
+static void configure_status(void)
 {
-	unsigned int cpu = smp_processor_id();
-	unsigned int status_set = ST0_CU0;
-	unsigned int hwrena = cpu_hwrena_impl_bits;
-
 	/*
 	 * Disable coprocessors and select 32-bit or 64-bit addressing
 	 * and the 16/32 or 32/32 FPR register model.  Reset the BEV
 	 * flag that some firmware may have left set and the TS bit (for
 	 * IP27).  Set XX for ISA IV code to work.
 	 */
+	unsigned int status_set = ST0_CU0;
 #ifdef CONFIG_64BIT
 	status_set |= ST0_FR|ST0_KX|ST0_SX|ST0_UX;
 #endif
@@ -1859,6 +1858,12 @@ void per_cpu_trap_init(bool is_boot_cpu)
 
 	change_c0_status(ST0_CU|ST0_MX|ST0_RE|ST0_FR|ST0_BEV|ST0_TS|ST0_KX|ST0_SX|ST0_UX,
 			 status_set);
+}
+
+/* configure HWRENA register */
+static void configure_hwrena(void)
+{
+	unsigned int hwrena = cpu_hwrena_impl_bits;
 
 	if (cpu_has_mips_r2)
 		hwrena |= 0x0000000f;
@@ -1868,7 +1873,10 @@ void per_cpu_trap_init(bool is_boot_cpu)
 
 	if (hwrena)
 		write_c0_hwrena(hwrena);
+}
 
+static void configure_exception_vector(void)
+{
 	if (cpu_has_veic || cpu_has_vint) {
 		unsigned long sr = set_c0_status(ST0_BEV);
 		write_c0_ebase(ebase);
@@ -1884,6 +1892,16 @@ void per_cpu_trap_init(bool is_boot_cpu)
 		} else
 			set_c0_cause(CAUSEF_IV);
 	}
+}
+
+void per_cpu_trap_init(bool is_boot_cpu)
+{
+	unsigned int cpu = smp_processor_id();
+
+	configure_status();
+	configure_hwrena();
+
+	configure_exception_vector();
 
 	/*
 	 * Before R2 both interrupt numbers were fixed to 7, so on R2 only:
@@ -2122,3 +2140,32 @@ void __init trap_init(void)
 
 	cu2_notifier(default_cu2_call, 0x80000000);	/* Run last  */
 }
+
+static int trap_pm_notifier(struct notifier_block *self, unsigned long cmd,
+			    void *v)
+{
+	switch (cmd) {
+	case CPU_PM_ENTER_FAILED:
+	case CPU_PM_EXIT:
+		configure_status();
+		configure_hwrena();
+		configure_exception_vector();
+
+		/* Restore register with CPU number for TLB handlers */
+		TLBMISS_HANDLER_RESTORE();
+
+		break;
+	}
+
+	return NOTIFY_OK;
+}
+
+static struct notifier_block trap_pm_notifier_block = {
+	.notifier_call = trap_pm_notifier,
+};
+
+static int __init trap_pm_init(void)
+{
+	return cpu_pm_register_notifier(&trap_pm_notifier_block);
+}
+arch_initcall(trap_pm_init);