Merge branch 'upstream' of git://git.linux-mips.org/pub/scm/ralf/upstream-linus

Pull MIPS updates from Ralf Baechle:
 "The most notable new addition inside this pull request is the support
  for MIPS's latest and greatest core called "inter/proAptiv".  The
  patch series describes this core as follows.

    "The interAptiv is a power-efficient multi-core microprocessor
     for use in system-on-chip (SoC) applications. The interAptiv combines
     a multi-threading pipeline with a coherence manager to deliver improved
     computational throughput and power efficiency. The interAptiv can
     contain one to four MIPS32R3 interAptiv cores, system level
     coherence manager with L2 cache, optional coherent I/O port,
     and optional floating point unit."

  The platform specific patches touch all 3 Broadcom families.  It adds
  support for the new Broadcom/Netlogix XLP9xx Soc, building a common
  BCM63XX SMP kernel for all BCM63XX SoCs regardless of core type/count
  and full gpio button/led descriptions for BCM47xx.

  The rest of the series are cleanups and bug fixes that are MIPS
  generic and consist largely of changes that Imgtec/MIPS had published
  in their linux-mti-3.10.git stable tree.  Random other cleanups and
  patches preparing code to be merged in 3.15"

* 'upstream' of git://git.linux-mips.org/pub/scm/ralf/upstream-linus: (139 commits)
  mips: select ARCH_MIGHT_HAVE_PC_SERIO
  mips: delete non-required instances of include <linux/init.h>
  MIPS: KVM: remove shadow_tlb code
  MIPS: KVM: use common EHINV aware UNIQUE_ENTRYHI
  mips/ide: flush dcache also if icache does not snoop dcache
  MIPS: BCM47XX: fix position of cpu_wait disabling
  MIPS: BCM63XX: select correct MIPS_L1_CACHE_SHIFT value
  MIPS: update MIPS_L1_CACHE_SHIFT based on MIPS_L1_CACHE_SHIFT_<N>
  MIPS: introduce MIPS_L1_CACHE_SHIFT_<N>
  MIPS: ZBOOT: gather string functions into string.c
  arch/mips/pci: don't check resource with devm_ioremap_resource
  arch/mips/lantiq/xway: don't check resource with devm_ioremap_resource
  bcma: gpio: don't cast u32 to unsigned long
  ssb: gpio: add own IRQ domain
  MIPS: BCM47XX: fix sparse warnings in board.c
  MIPS: BCM47XX: add board detection for Linksys WRT54GS V1
  MIPS: BCM47XX: fix detection for some boards
  MIPS: BCM47XX: Enable buttons support on SSB
  MIPS: BCM47XX: Convert WNDR4500 to new syntax
  MIPS: BCM47XX: Use "timer" trigger for status LEDs
  ...

16 files changed, 429 insertions, 116 deletions

diff --git a/arch/mips/netlogic/Kconfig b/arch/mips/netlogic/Kconfig
index 852a4ee0995..4eb683aef7d 100644
--- a/arch/mips/netlogic/Kconfig
+++ b/arch/mips/netlogic/Kconfig
@@ -28,6 +28,15 @@ config DT_XLP_FVP
 	  pointer to the kernel.  The corresponding DTS file is at
 	  arch/mips/netlogic/dts/xlp_fvp.dts
 
+config DT_XLP_GVP
+	bool "Built-in device tree for XLP GVP boards"
+	default y
+	help
+	  Add an FDT blob for XLP GVP board into the kernel.
+	  This DTB will be used if the firmware does not pass in a DTB
+	  pointer to the kernel.  The corresponding DTS file is at
+	  arch/mips/netlogic/dts/xlp_gvp.dts
+
 config NLM_MULTINODE
 	bool "Support for multi-chip boards"
 	depends on NLM_XLP_BOARD
diff --git a/arch/mips/netlogic/common/earlycons.c b/arch/mips/netlogic/common/earlycons.c
index 1902fa22d27..769f93032c5 100644
--- a/arch/mips/netlogic/common/earlycons.c
+++ b/arch/mips/netlogic/common/earlycons.c
@@ -37,9 +37,11 @@
 
 #include <asm/mipsregs.h>
 #include <asm/netlogic/haldefs.h>
+#include <asm/netlogic/common.h>
 
 #if defined(CONFIG_CPU_XLP)
 #include <asm/netlogic/xlp-hal/iomap.h>
+#include <asm/netlogic/xlp-hal/xlp.h>
 #include <asm/netlogic/xlp-hal/uart.h>
 #elif defined(CONFIG_CPU_XLR)
 #include <asm/netlogic/xlr/iomap.h>
diff --git a/arch/mips/netlogic/common/irq.c b/arch/mips/netlogic/common/irq.c
index 1c7e3a1b81a..5afc4b7fce0 100644
--- a/arch/mips/netlogic/common/irq.c
+++ b/arch/mips/netlogic/common/irq.c
@@ -180,6 +180,7 @@ static void __init nlm_init_percpu_irqs(void)
 #endif
 }
 
+
 void nlm_setup_pic_irq(int node, int picirq, int irq, int irt)
 {
 	struct nlm_pic_irq *pic_data;
@@ -207,32 +208,32 @@ void nlm_set_pic_extra_ack(int node, int irq, void (*xack)(struct irq_data *))
 
 static void nlm_init_node_irqs(int node)
 {
-	int i, irt;
-	uint64_t irqmask;
 	struct nlm_soc_info *nodep;
+	int i, irt;
 
 	pr_info("Init IRQ for node %d\n", node);
 	nodep = nlm_get_node(node);
-	irqmask = PERCPU_IRQ_MASK;
+	nodep->irqmask = PERCPU_IRQ_MASK;
 	for (i = PIC_IRT_FIRST_IRQ; i <= PIC_IRT_LAST_IRQ; i++) {
 		irt = nlm_irq_to_irt(i);
-		if (irt == -1)
+		if (irt == -1)		/* unused irq */
 			continue;
-		nlm_setup_pic_irq(node, i, i, irt);
-		/* set interrupts to first cpu in node */
+		nodep->irqmask |= 1ull << i;
+		if (irt == -2)		/* not a direct PIC irq */
+			continue;
+
 		nlm_pic_init_irt(nodep->picbase, irt, i,
-					node * NLM_CPUS_PER_NODE, 0);
-		irqmask |= (1ull << i);
+				node * nlm_threads_per_node(), 0);
+		nlm_setup_pic_irq(node, i, i, irt);
 	}
-	nodep->irqmask = irqmask;
 }
 
 void nlm_smp_irq_init(int hwcpuid)
 {
 	int node, cpu;
 
-	node = hwcpuid / NLM_CPUS_PER_NODE;
-	cpu  = hwcpuid % NLM_CPUS_PER_NODE;
+	node = nlm_cpuid_to_node(hwcpuid);
+	cpu  = hwcpuid % nlm_threads_per_node();
 
 	if (cpu == 0 && node != 0)
 		nlm_init_node_irqs(node);
@@ -256,13 +257,23 @@ asmlinkage void plat_irq_dispatch(void)
 		return;
 	}
 
+#if defined(CONFIG_PCI_MSI) && defined(CONFIG_CPU_XLP)
+	/* PCI interrupts need a second level dispatch for MSI bits */
+	if (i >= PIC_PCIE_LINK_MSI_IRQ(0) && i <= PIC_PCIE_LINK_MSI_IRQ(3)) {
+		nlm_dispatch_msi(node, i);
+		return;
+	}
+	if (i >= PIC_PCIE_MSIX_IRQ(0) && i <= PIC_PCIE_MSIX_IRQ(3)) {
+		nlm_dispatch_msix(node, i);
+		return;
+	}
+
+#endif
 	/* top level irq handling */
 	do_IRQ(nlm_irq_to_xirq(node, i));
 }
 
 #ifdef CONFIG_OF
-static struct irq_domain *xlp_pic_domain;
-
 static const struct irq_domain_ops xlp_pic_irq_domain_ops = {
 	.xlate = irq_domain_xlate_onetwocell,
 };
@@ -271,8 +282,9 @@ static int __init xlp_of_pic_init(struct device_node *node,
 					struct device_node *parent)
 {
 	const int n_picirqs = PIC_IRT_LAST_IRQ - PIC_IRQ_BASE + 1;
+	struct irq_domain *xlp_pic_domain;
 	struct resource res;
-	int socid, ret;
+	int socid, ret, bus;
 
 	/* we need a hack to get the PIC's SoC chip id */
 	ret = of_address_to_resource(node, 0, &res);
@@ -280,7 +292,34 @@ static int __init xlp_of_pic_init(struct device_node *node,
 		pr_err("PIC %s: reg property not found!\n", node->name);
 		return -EINVAL;
 	}
-	socid = (res.start >> 18) & 0x3;
+
+	if (cpu_is_xlp9xx()) {
+		bus = (res.start >> 20) & 0xf;
+		for (socid = 0; socid < NLM_NR_NODES; socid++) {
+			if (!nlm_node_present(socid))
+				continue;
+			if (nlm_get_node(socid)->socbus == bus)
+				break;
+		}
+		if (socid == NLM_NR_NODES) {
+			pr_err("PIC %s: Node mapping for bus %d not found!\n",
+					node->name, bus);
+			return -EINVAL;
+		}
+	} else {
+		socid = (res.start >> 18) & 0x3;
+		if (!nlm_node_present(socid)) {
+			pr_err("PIC %s: node %d does not exist!\n",
+							node->name, socid);
+			return -EINVAL;
+		}
+	}
+
+	if (!nlm_node_present(socid)) {
+		pr_err("PIC %s: node %d does not exist!\n", node->name, socid);
+		return -EINVAL;
+	}
+
 	xlp_pic_domain = irq_domain_add_legacy(node, n_picirqs,
 		nlm_irq_to_xirq(socid, PIC_IRQ_BASE), PIC_IRQ_BASE,
 		&xlp_pic_irq_domain_ops, NULL);
@@ -288,8 +327,7 @@ static int __init xlp_of_pic_init(struct device_node *node,
 		pr_err("PIC %s: Creating legacy domain failed!\n", node->name);
 		return -EINVAL;
 	}
-	pr_info("Node %d: IRQ domain created for PIC@%pa\n", socid,
-							&res.start);
+	pr_info("Node %d: IRQ domain created for PIC@%pR\n", socid, &res);
 	return 0;
 }
 
diff --git a/arch/mips/netlogic/common/reset.S b/arch/mips/netlogic/common/reset.S
index adb18288a6c..b231fe1e7a0 100644
--- a/arch/mips/netlogic/common/reset.S
+++ b/arch/mips/netlogic/common/reset.S
@@ -32,10 +32,10 @@
  * IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 
-#include <linux/init.h>
 
 #include <asm/asm.h>
 #include <asm/asm-offsets.h>
+#include <asm/cacheops.h>
 #include <asm/regdef.h>
 #include <asm/mipsregs.h>
 #include <asm/stackframe.h>
@@ -50,8 +50,8 @@
 #include <asm/netlogic/xlp-hal/cpucontrol.h>
 
 #define CP0_EBASE	$15
-#define SYS_CPU_COHERENT_BASE(node)	CKSEG1ADDR(XLP_DEFAULT_IO_BASE) + \
-			XLP_IO_SYS_OFFSET(node) + XLP_IO_PCI_HDRSZ + \
+#define SYS_CPU_COHERENT_BASE	CKSEG1ADDR(XLP_DEFAULT_IO_BASE) + \
+			XLP_IO_SYS_OFFSET(0) + XLP_IO_PCI_HDRSZ + \
 			SYS_CPU_NONCOHERENT_MODE * 4
 
 /* Enable XLP features and workarounds in the LSU */
@@ -74,35 +74,55 @@
 .endm
 
 /*
- * Low level flush for L1D cache on XLP, the normal cache ops does
- * not do the complete and correct cache flush.
+ * L1D cache has to be flushed before enabling threads in XLP.
+ * On XLP8xx/XLP3xx, we do a low level flush using processor control
+ * registers. On XLPII CPUs, usual cache instructions work.
  */
 .macro	xlp_flush_l1_dcache
+	mfc0	t0, CP0_EBASE, 0
+	andi	t0, t0, 0xff00
+	slt	t1, t0, 0x1200
+	beqz	t1, 15f
+	nop
+
+	/* XLP8xx low level cache flush */
 	li	t0, LSU_DEBUG_DATA0
 	li	t1, LSU_DEBUG_ADDR
 	li	t2, 0		/* index */
 	li	t3, 0x1000	/* loop count */
-1:
+11:
 	sll	v0, t2, 5
 	mtcr	zero, t0
 	ori	v1, v0, 0x3	/* way0 | write_enable | write_active */
 	mtcr	v1, t1
-2:
+12:
 	mfcr	v1, t1
 	andi	v1, 0x1		/* wait for write_active == 0 */
-	bnez	v1, 2b
+	bnez	v1, 12b
 	nop
 	mtcr	zero, t0
 	ori	v1, v0, 0x7	/* way1 | write_enable | write_active */
 	mtcr	v1, t1
-3:
+13:
 	mfcr	v1, t1
 	andi	v1, 0x1		/* wait for write_active == 0 */
-	bnez	v1, 3b
+	bnez	v1, 13b
 	nop
 	addi	t2, 1
-	bne	t3, t2, 1b
+	bne	t3, t2, 11b
+	nop
+	b	17f
+	nop
+
+	/* XLPII CPUs, Invalidate all 64k of L1 D-cache */
+15:
+	li	t0, 0x80000000
+	li	t1, 0x80010000
+16:	cache	Index_Writeback_Inv_D, 0(t0)
+	addiu	t0, t0, 32
+	bne	t0, t1, 16b
 	nop
+17:
 .endm
 
 /*
@@ -138,6 +158,13 @@ FEXPORT(nlm_reset_entry)
 	nop
 
 1:	/* Entry point on core wakeup */
+	mfc0	t0, CP0_EBASE, 0	/* processor ID */
+	andi	t0, 0xff00
+	li	t1, 0x1500		/* XLP 9xx */
+	beq	t0, t1, 2f		/* does not need to set coherent */
+	nop
+
+	/* set bit in SYS coherent register for the core */
 	mfc0	t0, CP0_EBASE, 1
 	mfc0	t1, CP0_EBASE, 1
 	srl	t1, 5
@@ -149,7 +176,7 @@ FEXPORT(nlm_reset_entry)
 	li	t1, 0x1
 	sll	t0, t1, t0
 	nor	t0, t0, zero		/* t0 <- ~(1 << core) */
-	li	t2, SYS_CPU_COHERENT_BASE(0)
+	li	t2, SYS_CPU_COHERENT_BASE
 	add	t2, t2, t3		/* t2 <- SYS offset for node */
 	lw	t1, 0(t2)
 	and	t1, t1, t0
@@ -159,13 +186,13 @@ FEXPORT(nlm_reset_entry)
 	lw	t1, 0(t2)
 	sync
 
+2:
 	/* Configure LSU on Non-0 Cores. */
 	xlp_config_lsu
 	/* FALL THROUGH */
 
 /*
- * Wake up sibling threads from the initial thread in
- * a core.
+ * Wake up sibling threads from the initial thread in a core.
  */
 EXPORT(nlm_boot_siblings)
 	/* core L1D flush before enable threads */
@@ -181,8 +208,10 @@ EXPORT(nlm_boot_siblings)
 	/*
 	 * The new hardware thread starts at the next instruction
 	 * For all the cases other than core 0 thread 0, we will
-	* jump to the secondary wait function.
-	*/
+	 * jump to the secondary wait function.
+
+	 * NOTE: All GPR contents are lost after the mtcr above!
+	 */
 	mfc0	v0, CP0_EBASE, 1
 	andi	v0, 0x3ff		/* v0 <- node/core */
 
@@ -196,7 +225,7 @@ EXPORT(nlm_boot_siblings)
 #endif
 	mtc0	t1, CP0_STATUS
 
-	/* mark CPU ready, careful here, previous mtcr trashed registers */
+	/* mark CPU ready */
 	li	t3, CKSEG1ADDR(RESET_DATA_PHYS)
 	ADDIU	t1, t3, BOOT_CPU_READY
 	sll	v1, v0, 2
diff --git a/arch/mips/netlogic/common/smp.c b/arch/mips/netlogic/common/smp.c
index c0eded01fde..6baae15cc7b 100644
--- a/arch/mips/netlogic/common/smp.c
+++ b/arch/mips/netlogic/common/smp.c
@@ -63,7 +63,7 @@ void nlm_send_ipi_single(int logical_cpu, unsigned int action)
 	uint64_t picbase;
 
 	cpu = cpu_logical_map(logical_cpu);
-	node = cpu / NLM_CPUS_PER_NODE;
+	node = nlm_cpuid_to_node(cpu);
 	picbase = nlm_get_node(node)->picbase;
 
 	if (action & SMP_CALL_FUNCTION)
@@ -152,7 +152,7 @@ void nlm_boot_secondary(int logical_cpu, struct task_struct *idle)
 	int cpu, node;
 
 	cpu = cpu_logical_map(logical_cpu);
-	node = cpu / NLM_CPUS_PER_NODE;
+	node = nlm_cpuid_to_node(logical_cpu);
 	nlm_next_sp = (unsigned long)__KSTK_TOS(idle);
 	nlm_next_gp = (unsigned long)task_thread_info(idle);
 
@@ -164,7 +164,7 @@ void nlm_boot_secondary(int logical_cpu, struct task_struct *idle)
 void __init nlm_smp_setup(void)
 {
 	unsigned int boot_cpu;
-	int num_cpus, i, ncore;
+	int num_cpus, i, ncore, node;
 	volatile u32 *cpu_ready = nlm_get_boot_data(BOOT_CPU_READY);
 	char buf[64];
 
@@ -187,6 +187,8 @@ void __init nlm_smp_setup(void)
 			__cpu_number_map[i] = num_cpus;
 			__cpu_logical_map[num_cpus] = i;
 			set_cpu_possible(num_cpus, true);
+			node = nlm_cpuid_to_node(i);
+			cpumask_set_cpu(num_cpus, &nlm_get_node(node)->cpumask);
 			++num_cpus;
 		}
 	}
diff --git a/arch/mips/netlogic/common/smpboot.S b/arch/mips/netlogic/common/smpboot.S
index aa6cff0a229..8597657c27f 100644
--- a/arch/mips/netlogic/common/smpboot.S
+++ b/arch/mips/netlogic/common/smpboot.S
@@ -32,7 +32,6 @@
  * IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 
-#include <linux/init.h>
 
 #include <asm/asm.h>
 #include <asm/asm-offsets.h>
@@ -98,7 +97,7 @@ END(nlm_boot_secondary_cpus)
  * In case of RMIboot bootloader which is used on XLR boards, the CPUs
  * be already woken up and waiting in bootloader code.
  * This will get them out of the bootloader code and into linux. Needed
- *  because the bootloader area will be taken and initialized by linux.
+ * because the bootloader area will be taken and initialized by linux.
  */
 NESTED(nlm_rmiboot_preboot, 16, sp)
 	mfc0	t0, $15, 1	/* read ebase */
@@ -133,6 +132,7 @@ NESTED(nlm_rmiboot_preboot, 16, sp)
 	or	t1, t2, v1	/* put in new value */
 	mtcr	t1, t0		/* update core control */
 
+	/* wait for NMI to hit */
 1:	wait
 	b	1b
 	nop
diff --git a/arch/mips/netlogic/dts/Makefile b/arch/mips/netlogic/dts/Makefile
index 0b9be5fd2e4..25c8e873ee2 100644
--- a/arch/mips/netlogic/dts/Makefile
+++ b/arch/mips/netlogic/dts/Makefile
@@ -1,3 +1,4 @@
 obj-$(CONFIG_DT_XLP_EVP) := xlp_evp.dtb.o
 obj-$(CONFIG_DT_XLP_SVP) += xlp_svp.dtb.o
 obj-$(CONFIG_DT_XLP_FVP) += xlp_fvp.dtb.o
+obj-$(CONFIG_DT_XLP_GVP) += xlp_gvp.dtb.o
diff --git a/arch/mips/netlogic/dts/xlp_gvp.dts b/arch/mips/netlogic/dts/xlp_gvp.dts
new file mode 100644
index 00000000000..047d27f5448
--- /dev/null
+++ b/arch/mips/netlogic/dts/xlp_gvp.dts
@@ -0,0 +1,76 @@
+/*
+ * XLP9XX Device Tree Source for GVP boards
+ */
+
+/dts-v1/;
+/ {
+	model = "netlogic,XLP-GVP";
+	compatible = "netlogic,xlp";
+	#address-cells = <2>;
+	#size-cells = <2>;
+
+	soc {
+		#address-cells = <2>;
+		#size-cells = <1>;
+		compatible = "simple-bus";
+		ranges = <0 0  0 0x18000000  0x04000000   // PCIe CFG
+			  1 0  0 0x16000000  0x02000000>; // GBU chipselects
+
+		serial0: serial@30000 {
+			device_type = "serial";
+			compatible = "ns16550";
+			reg = <0 0x112100 0xa00>;
+			reg-shift = <2>;
+			reg-io-width = <4>;
+			clock-frequency = <125000000>;
+			interrupt-parent = <&pic>;
+			interrupts = <17>;
+		};
+		pic: pic@4000 {
+			interrupt-controller;
+			#address-cells = <0>;
+			#interrupt-cells = <1>;
+			reg = <0 0x110000 0x200>;
+		};
+
+		nor_flash@1,0 {
+			compatible = "cfi-flash";
+			#address-cells = <1>;
+			#size-cells = <1>;
+			bank-width = <2>;
+			reg = <1 0 0x1000000>;
+
+			partition@0 {
+				label = "x-loader";
+				reg = <0x0 0x100000>; /* 1M */
+				read-only;
+			};
+
+			partition@100000 {
+				label = "u-boot";
+				reg = <0x100000 0x100000>; /* 1M */
+			};
+
+			partition@200000 {
+				label = "kernel";
+				reg = <0x200000 0x500000>; /* 5M */
+			};
+
+			partition@700000 {
+				label = "rootfs";
+				reg = <0x700000 0x800000>; /* 8M */
+			};
+
+			partition@f00000 {
+				label = "env";
+				reg = <0xf00000 0x100000>; /* 1M */
+				read-only;
+			};
+		};
+
+	};
+
+	chosen {
+		bootargs = "console=ttyS0,115200 rdinit=/sbin/init";
+	};
+};
diff --git a/arch/mips/netlogic/xlp/dt.c b/arch/mips/netlogic/xlp/dt.c
index 8316d5454b1..5754097b9cd 100644
--- a/arch/mips/netlogic/xlp/dt.c
+++ b/arch/mips/netlogic/xlp/dt.c
@@ -42,13 +42,18 @@
 #include <asm/prom.h>
 
 extern u32 __dtb_xlp_evp_begin[], __dtb_xlp_svp_begin[],
-	__dtb_xlp_fvp_begin[], __dtb_start[];
+	__dtb_xlp_fvp_begin[], __dtb_xlp_gvp_begin[], __dtb_start[];
 static void *xlp_fdt_blob;
 
 void __init *xlp_dt_init(void *fdtp)
 {
 	if (!fdtp) {
 		switch (current_cpu_data.processor_id & 0xff00) {
+#ifdef CONFIG_DT_XLP_GVP
+		case PRID_IMP_NETLOGIC_XLP9XX:
+			fdtp = __dtb_xlp_gvp_begin;
+			break;
+#endif
 #ifdef CONFIG_DT_XLP_FVP
 		case PRID_IMP_NETLOGIC_XLP2XX:
 			fdtp = __dtb_xlp_fvp_begin;
diff --git a/arch/mips/netlogic/xlp/nlm_hal.c b/arch/mips/netlogic/xlp/nlm_hal.c
index 56c50ba43c9..997cd9ee10d 100644
--- a/arch/mips/netlogic/xlp/nlm_hal.c
+++ b/arch/mips/netlogic/xlp/nlm_hal.c
@@ -57,6 +57,10 @@ void nlm_node_init(int node)
 	nodep->sysbase = nlm_get_sys_regbase(node);
 	nodep->picbase = nlm_get_pic_regbase(node);
 	nodep->ebase = read_c0_ebase() & (~((1 << 12) - 1));
+	if (cpu_is_xlp9xx())
+		nodep->socbus = xlp9xx_get_socbus(node);
+	else
+		nodep->socbus = 0;
 	spin_lock_init(&nodep->piclock);
 }
 
@@ -65,6 +69,26 @@ int nlm_irq_to_irt(int irq)
 	uint64_t pcibase;
 	int devoff, irt;
 
+	/* bypass for 9xx */
+	if (cpu_is_xlp9xx()) {
+		switch (irq) {
+		case PIC_9XX_XHCI_0_IRQ:
+			return 114;
+		case PIC_9XX_XHCI_1_IRQ:
+			return 115;
+		case PIC_UART_0_IRQ:
+			return 133;
+		case PIC_UART_1_IRQ:
+			return 134;
+		case PIC_PCIE_LINK_LEGACY_IRQ(0):
+		case PIC_PCIE_LINK_LEGACY_IRQ(1):
+		case PIC_PCIE_LINK_LEGACY_IRQ(2):
+		case PIC_PCIE_LINK_LEGACY_IRQ(3):
+			return 191 + irq - PIC_PCIE_LINK_LEGACY_IRQ_BASE;
+		}
+		return -1;
+	}
+
 	devoff = 0;
 	switch (irq) {
 	case PIC_UART_0_IRQ:
@@ -135,9 +159,17 @@ int nlm_irq_to_irt(int irq)
 		case PIC_I2C_3_IRQ:
 			irt = irt + 3; break;
 		}
-	} else if (irq >= PIC_PCIE_LINK_0_IRQ && irq <= PIC_PCIE_LINK_3_IRQ) {
+	} else if (irq >= PIC_PCIE_LINK_LEGACY_IRQ(0) &&
+			irq <= PIC_PCIE_LINK_LEGACY_IRQ(3)) {
 		/* HW bug, PCI IRT entries are bad on early silicon, fix */
-		irt = PIC_IRT_PCIE_LINK_INDEX(irq - PIC_PCIE_LINK_0_IRQ);
+		irt = PIC_IRT_PCIE_LINK_INDEX(irq -
+					PIC_PCIE_LINK_LEGACY_IRQ_BASE);
+	} else if (irq >= PIC_PCIE_LINK_MSI_IRQ(0) &&
+			irq <= PIC_PCIE_LINK_MSI_IRQ(3)) {
+		irt = -2;
+	} else if (irq >= PIC_PCIE_MSIX_IRQ(0) &&
+			irq <= PIC_PCIE_MSIX_IRQ(3)) {
+		irt = -2;
 	} else {
 		irt = -1;
 	}
@@ -151,7 +183,10 @@ unsigned int nlm_get_core_frequency(int node, int core)
 	uint64_t num, sysbase;
 
 	sysbase = nlm_get_node(node)->sysbase;
-	rstval = nlm_read_sys_reg(sysbase, SYS_POWER_ON_RESET_CFG);
+	if (cpu_is_xlp9xx())
+		rstval = nlm_read_sys_reg(sysbase, SYS_9XX_POWER_ON_RESET_CFG);
+	else
+		rstval = nlm_read_sys_reg(sysbase, SYS_POWER_ON_RESET_CFG);
 	if (cpu_is_xlpii()) {
 		num = 1000000ULL * (400 * 3 + 100 * (rstval >> 26));
 		denom = 3;
@@ -265,6 +300,10 @@ static unsigned int nlm_2xx_get_pic_frequency(int node)
 
 unsigned int nlm_get_pic_frequency(int node)
 {
+	/* TODO Has to calculate freq as like 2xx */
+	if (cpu_is_xlp9xx())
+		return 250000000;
+
 	if (cpu_is_xlpii())
 		return nlm_2xx_get_pic_frequency(node);
 	else
@@ -284,21 +323,33 @@ int xlp_get_dram_map(int n, uint64_t *dram_map)
 {
 	uint64_t bridgebase, base, lim;
 	uint32_t val;
+	unsigned int barreg, limreg, xlatreg;
 	int i, node, rv;
 
 	/* Look only at mapping on Node 0, we don't handle crazy configs */
 	bridgebase = nlm_get_bridge_regbase(0);
 	rv = 0;
 	for (i = 0; i < 8; i++) {
-		val = nlm_read_bridge_reg(bridgebase,
-					BRIDGE_DRAM_NODE_TRANSLN(i));
-		node = (val >> 1) & 0x3;
-		if (n >= 0 && n != node)
-			continue;
-		val = nlm_read_bridge_reg(bridgebase, BRIDGE_DRAM_BAR(i));
+		if (cpu_is_xlp9xx()) {
+			barreg = BRIDGE_9XX_DRAM_BAR(i);
+			limreg = BRIDGE_9XX_DRAM_LIMIT(i);
+			xlatreg = BRIDGE_9XX_DRAM_NODE_TRANSLN(i);
+		} else {
+			barreg = BRIDGE_DRAM_BAR(i);
+			limreg = BRIDGE_DRAM_LIMIT(i);
+			xlatreg = BRIDGE_DRAM_NODE_TRANSLN(i);
+		}
+		if (n >= 0) {
+			/* node specified, get node mapping of BAR */
+			val = nlm_read_bridge_reg(bridgebase, xlatreg);
+			node = (val >> 1) & 0x3;
+			if (n != node)
+				continue;
+		}
+		val = nlm_read_bridge_reg(bridgebase, barreg);
 		val = (val >>  12) & 0xfffff;
 		base = (uint64_t) val << 20;
-		val = nlm_read_bridge_reg(bridgebase, BRIDGE_DRAM_LIMIT(i));
+		val = nlm_read_bridge_reg(bridgebase, limreg);
 		val = (val >>  12) & 0xfffff;
 		if (val == 0)   /* BAR not used */
 			continue;
diff --git a/arch/mips/netlogic/xlp/setup.c b/arch/mips/netlogic/xlp/setup.c
index 54e75c77184..8c60a2dd9ef 100644
--- a/arch/mips/netlogic/xlp/setup.c
+++ b/arch/mips/netlogic/xlp/setup.c
@@ -51,12 +51,16 @@ uint64_t nlm_io_base;
 struct nlm_soc_info nlm_nodes[NLM_NR_NODES];
 cpumask_t nlm_cpumask = CPU_MASK_CPU0;
 unsigned int nlm_threads_per_core;
+unsigned int xlp_cores_per_node;
 
 static void nlm_linux_exit(void)
 {
 	uint64_t sysbase = nlm_get_node(0)->sysbase;
 
-	nlm_write_sys_reg(sysbase, SYS_CHIP_RESET, 1);
+	if (cpu_is_xlp9xx())
+		nlm_write_sys_reg(sysbase, SYS_9XX_CHIP_RESET, 1);
+	else
+		nlm_write_sys_reg(sysbase, SYS_CHIP_RESET, 1);
 	for ( ; ; )
 		cpu_wait();
 }
@@ -92,6 +96,14 @@ static void __init xlp_init_mem_from_bars(void)
 
 void __init plat_mem_setup(void)
 {
+#ifdef CONFIG_SMP
+	nlm_wakeup_secondary_cpus();
+
+	/* update TLB size after waking up threads */
+	current_cpu_data.tlbsize = ((read_c0_config6() >> 16) & 0xffff) + 1;
+
+	register_smp_ops(&nlm_smp_ops);
+#endif
 	_machine_restart = (void (*)(char *))nlm_linux_exit;
 	_machine_halt	= nlm_linux_exit;
 	pm_power_off	= nlm_linux_exit;
@@ -110,6 +122,7 @@ void __init plat_mem_setup(void)
 const char *get_system_type(void)
 {
 	switch (read_c0_prid() & 0xff00) {
+	case PRID_IMP_NETLOGIC_XLP9XX:
 	case PRID_IMP_NETLOGIC_XLP2XX:
 		return "Broadcom XLPII Series";
 	default:
@@ -149,6 +162,10 @@ void __init prom_init(void)
 	void *reset_vec;
 
 	nlm_io_base = CKSEG1ADDR(XLP_DEFAULT_IO_BASE);
+	if (cpu_is_xlp9xx())
+		xlp_cores_per_node = 32;
+	else
+		xlp_cores_per_node = 8;
 	nlm_init_boot_cpu();
 	xlp_mmu_init();
 	nlm_node_init(0);
@@ -162,11 +179,5 @@ void __init prom_init(void)
 
 #ifdef CONFIG_SMP
 	cpumask_setall(&nlm_cpumask);
-	nlm_wakeup_secondary_cpus();
-
-	/* update TLB size after waking up threads */
-	current_cpu_data.tlbsize = ((read_c0_config6() >> 16) & 0xffff) + 1;
-
-	register_smp_ops(&nlm_smp_ops);
 #endif
 }
diff --git a/arch/mips/netlogic/xlp/usb-init-xlp2.c b/arch/mips/netlogic/xlp/usb-init-xlp2.c
index 36e9c22afc4..17ade1ce5df 100644
--- a/arch/mips/netlogic/xlp/usb-init-xlp2.c
+++ b/arch/mips/netlogic/xlp/usb-init-xlp2.c
@@ -37,6 +37,7 @@
 #include <linux/delay.h>
 #include <linux/init.h>
 #include <linux/pci.h>
+#include <linux/pci_ids.h>
 #include <linux/platform_device.h>
 #include <linux/irq.h>
 
@@ -83,12 +84,14 @@
 #define nlm_read_usb_reg(b, r)		nlm_read_reg(b, r)
 #define nlm_write_usb_reg(b, r, v)	nlm_write_reg(b, r, v)
 
-#define nlm_xlpii_get_usb_pcibase(node, inst)		\
-	nlm_pcicfg_base(XLP2XX_IO_USB_OFFSET(node, inst))
+#define nlm_xlpii_get_usb_pcibase(node, inst)			\
+			nlm_pcicfg_base(cpu_is_xlp9xx() ?	\
+			XLP9XX_IO_USB_OFFSET(node, inst) :	\
+			XLP2XX_IO_USB_OFFSET(node, inst))
 #define nlm_xlpii_get_usb_regbase(node, inst)		\
 	(nlm_xlpii_get_usb_pcibase(node, inst) + XLP_IO_PCI_HDRSZ)
 
-static void xlpii_usb_ack(struct irq_data *data)
+static void xlp2xx_usb_ack(struct irq_data *data)
 {
 	u64 port_addr;
 
@@ -109,6 +112,29 @@ static void xlpii_usb_ack(struct irq_data *data)
 	nlm_write_usb_reg(port_addr, XLPII_USB3_INT_REG, 0xffffffff);
 }
 
+static void xlp9xx_usb_ack(struct irq_data *data)
+{
+	u64 port_addr;
+	int node, irq;
+
+	/* Find the node and irq on the node */
+	irq = data->irq % NLM_IRQS_PER_NODE;
+	node = data->irq / NLM_IRQS_PER_NODE;
+
+	switch (irq) {
+	case PIC_9XX_XHCI_0_IRQ:
+		port_addr = nlm_xlpii_get_usb_regbase(node, 1);
+		break;
+	case PIC_9XX_XHCI_1_IRQ:
+		port_addr = nlm_xlpii_get_usb_regbase(node, 2);
+		break;
+	default:
+		pr_err("No matching USB irq %d node  %d!\n", irq, node);
+		return;
+	}
+	nlm_write_usb_reg(port_addr, XLPII_USB3_INT_REG, 0xffffffff);
+}
+
 static void nlm_xlpii_usb_hw_reset(int node, int port)
 {
 	u64 port_addr, xhci_base, pci_base;
@@ -178,17 +204,33 @@ static void nlm_xlpii_usb_hw_reset(int node, int port)
 
 static int __init nlm_platform_xlpii_usb_init(void)
 {
+	int node;
+
 	if (!cpu_is_xlpii())
 		return 0;
 
-	pr_info("Initializing 2XX USB Interface\n");
-	nlm_xlpii_usb_hw_reset(0, 1);
-	nlm_xlpii_usb_hw_reset(0, 2);
-	nlm_xlpii_usb_hw_reset(0, 3);
-	nlm_set_pic_extra_ack(0, PIC_2XX_XHCI_0_IRQ, xlpii_usb_ack);
-	nlm_set_pic_extra_ack(0, PIC_2XX_XHCI_1_IRQ, xlpii_usb_ack);
-	nlm_set_pic_extra_ack(0, PIC_2XX_XHCI_2_IRQ, xlpii_usb_ack);
+	if (!cpu_is_xlp9xx()) {
+		/* XLP 2XX single node */
+		pr_info("Initializing 2XX USB Interface\n");
+		nlm_xlpii_usb_hw_reset(0, 1);
+		nlm_xlpii_usb_hw_reset(0, 2);
+		nlm_xlpii_usb_hw_reset(0, 3);
+		nlm_set_pic_extra_ack(0, PIC_2XX_XHCI_0_IRQ, xlp2xx_usb_ack);
+		nlm_set_pic_extra_ack(0, PIC_2XX_XHCI_1_IRQ, xlp2xx_usb_ack);
+		nlm_set_pic_extra_ack(0, PIC_2XX_XHCI_2_IRQ, xlp2xx_usb_ack);
+		return 0;
+	}
 
+	/* XLP 9XX, multi-node */
+	pr_info("Initializing 9XX USB Interface\n");
+	for (node = 0; node < NLM_NR_NODES; node++) {
+		if (!nlm_node_present(node))
+			continue;
+		nlm_xlpii_usb_hw_reset(node, 1);
+		nlm_xlpii_usb_hw_reset(node, 2);
+		nlm_set_pic_extra_ack(node, PIC_9XX_XHCI_0_IRQ, xlp9xx_usb_ack);
+		nlm_set_pic_extra_ack(node, PIC_9XX_XHCI_1_IRQ, xlp9xx_usb_ack);
+	}
 	return 0;
 }
 
@@ -196,8 +238,26 @@ arch_initcall(nlm_platform_xlpii_usb_init);
 
 static u64 xlp_usb_dmamask = ~(u32)0;
 
-/* Fixup IRQ for USB devices on XLP the SoC PCIe bus */
-static void nlm_usb_fixup_final(struct pci_dev *dev)
+/* Fixup the IRQ for USB devices which is exist on XLP9XX SOC PCIE bus */
+static void nlm_xlp9xx_usb_fixup_final(struct pci_dev *dev)
+{
+	int node;
+
+	node = xlp_socdev_to_node(dev);
+	dev->dev.dma_mask		= &xlp_usb_dmamask;
+	dev->dev.coherent_dma_mask	= DMA_BIT_MASK(32);
+	switch (dev->devfn) {
+	case 0x21:
+		dev->irq = nlm_irq_to_xirq(node, PIC_9XX_XHCI_0_IRQ);
+		break;
+	case 0x22:
+		dev->irq = nlm_irq_to_xirq(node, PIC_9XX_XHCI_1_IRQ);
+		break;
+	}
+}
+
+/* Fixup the IRQ for USB devices which is exist on XLP2XX SOC PCIE bus */
+static void nlm_xlp2xx_usb_fixup_final(struct pci_dev *dev)
 {
 	dev->dev.dma_mask		= &xlp_usb_dmamask;
 	dev->dev.coherent_dma_mask	= DMA_BIT_MASK(32);
@@ -214,5 +274,7 @@ static void nlm_usb_fixup_final(struct pci_dev *dev)
 	}
 }
 
+DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_ID_BROADCOM, PCI_DEVICE_ID_XLP9XX_XHCI,
+		nlm_xlp9xx_usb_fixup_final);
 DECLARE_PCI_FIXUP_FINAL(PCI_VENDOR_NETLOGIC, PCI_DEVICE_ID_NLM_XHCI,
-		nlm_usb_fixup_final);
+		nlm_xlp2xx_usb_fixup_final);
diff --git a/arch/mips/netlogic/xlp/wakeup.c b/arch/mips/netlogic/xlp/wakeup.c
index 682d5638dc0..9a92617a2af 100644
--- a/arch/mips/netlogic/xlp/wakeup.c
+++ b/arch/mips/netlogic/xlp/wakeup.c
@@ -32,7 +32,6 @@
  * IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 
-#include <linux/init.h>
 #include <linux/kernel.h>
 #include <linux/threads.h>
 
@@ -47,14 +46,14 @@
 #include <asm/netlogic/mips-extns.h>
 
 #include <asm/netlogic/xlp-hal/iomap.h>
-#include <asm/netlogic/xlp-hal/pic.h>
 #include <asm/netlogic/xlp-hal/xlp.h>
+#include <asm/netlogic/xlp-hal/pic.h>
 #include <asm/netlogic/xlp-hal/sys.h>
 
 static int xlp_wakeup_core(uint64_t sysbase, int node, int core)
 {
 	uint32_t coremask, value;
-	int count;
+	int count, resetreg;
 
 	coremask = (1 << core);
 
@@ -65,12 +64,24 @@ static int xlp_wakeup_core(uint64_t sysbase, int node, int core)
 		nlm_write_sys_reg(sysbase, SYS_CORE_DFS_DIS_CTRL, value);
 	}
 
+	/* On 9XX, mark coherent first */
+	if (cpu_is_xlp9xx()) {
+		value = nlm_read_sys_reg(sysbase, SYS_9XX_CPU_NONCOHERENT_MODE);
+		value &= ~coremask;
+		nlm_write_sys_reg(sysbase, SYS_9XX_CPU_NONCOHERENT_MODE, value);
+	}
+
 	/* Remove CPU Reset */
-	value = nlm_read_sys_reg(sysbase, SYS_CPU_RESET);
+	resetreg = cpu_is_xlp9xx() ? SYS_9XX_CPU_RESET : SYS_CPU_RESET;
+	value = nlm_read_sys_reg(sysbase, resetreg);
 	value &= ~coremask;
-	nlm_write_sys_reg(sysbase, SYS_CPU_RESET, value);
+	nlm_write_sys_reg(sysbase, resetreg, value);
+
+	/* We are done on 9XX */
+	if (cpu_is_xlp9xx())
+		return 1;
 
-	/* Poll for CPU to mark itself coherent */
+	/* Poll for CPU to mark itself coherent on other type of XLP */
 	count = 100000;
 	do {
 		value = nlm_read_sys_reg(sysbase, SYS_CPU_NONCOHERENT_MODE);
@@ -84,7 +95,7 @@ static int wait_for_cpus(int cpu, int bootcpu)
 	volatile uint32_t *cpu_ready = nlm_get_boot_data(BOOT_CPU_READY);
 	int i, count, notready;
 
-	count = 0x20000000;
+	count = 0x800000;
 	do {
 		notready = nlm_threads_per_core;
 		for (i = 0; i < nlm_threads_per_core; i++)
@@ -98,27 +109,62 @@ static int wait_for_cpus(int cpu, int bootcpu)
 static void xlp_enable_secondary_cores(const cpumask_t *wakeup_mask)
 {
 	struct nlm_soc_info *nodep;
-	uint64_t syspcibase;
-	uint32_t syscoremask;
+	uint64_t syspcibase, fusebase;
+	uint32_t syscoremask, mask, fusemask;
 	int core, n, cpu;
 
 	for (n = 0; n < NLM_NR_NODES; n++) {
-		syspcibase = nlm_get_sys_pcibase(n);
-		if (nlm_read_reg(syspcibase, 0) == 0xffffffff)
-			break;
+		if (n != 0) {
+			/* check if node exists and is online */
+			if (cpu_is_xlp9xx()) {
+				int b = xlp9xx_get_socbus(n);
+				pr_info("Node %d SoC PCI bus %d.\n", n, b);
+				if (b == 0)
+					break;
+			} else {
+				syspcibase = nlm_get_sys_pcibase(n);
+				if (nlm_read_reg(syspcibase, 0) == 0xffffffff)
+					break;
+			}
+			nlm_node_init(n);
+		}
 
 		/* read cores in reset from SYS */
-		if (n != 0)
-			nlm_node_init(n);
 		nodep = nlm_get_node(n);
-		syscoremask = nlm_read_sys_reg(nodep->sysbase, SYS_CPU_RESET);
+
+		if (cpu_is_xlp9xx()) {
+			fusebase = nlm_get_fuse_regbase(n);
+			fusemask = nlm_read_reg(fusebase, FUSE_9XX_DEVCFG6);
+			mask = 0xfffff;
+		} else {
+			fusemask = nlm_read_sys_reg(nodep->sysbase,
+						SYS_EFUSE_DEVICE_CFG_STATUS0);
+			switch (read_c0_prid() & 0xff00) {
+			case PRID_IMP_NETLOGIC_XLP3XX:
+				mask = 0xf;
+				break;
+			case PRID_IMP_NETLOGIC_XLP2XX:
+				mask = 0x3;
+				break;
+			case PRID_IMP_NETLOGIC_XLP8XX:
+			default:
+				mask = 0xff;
+				break;
+			}
+		}
+
+		/*
+		 * Fused out cores are set in the fusemask, and the remaining
+		 * cores are renumbered to range 0 .. nactive-1
+		 */
+		syscoremask = (1 << hweight32(~fusemask & mask)) - 1;
+
 		/* The boot cpu */
-		if (n == 0) {
-			syscoremask |= 1;
+		if (n == 0)
 			nodep->coremask = 1;
-		}
 
-		for (core = 0; core < NLM_CORES_PER_NODE; core++) {
+		pr_info("Node %d - SYS/FUSE coremask %x\n", n, syscoremask);
+		for (core = 0; core < nlm_cores_per_node(); core++) {
 			/* we will be on node 0 core 0 */
 			if (n == 0 && core == 0)
 				continue;
@@ -128,7 +174,7 @@ static void xlp_enable_secondary_cores(const cpumask_t *wakeup_mask)
 				continue;
 
 			/* see if at least the first hw thread is enabled */
-			cpu = (n * NLM_CORES_PER_NODE + core)
+			cpu = (n * nlm_cores_per_node() + core)
 						* NLM_THREADS_PER_CORE;
 			if (!cpumask_test_cpu(cpu, wakeup_mask))
 				continue;
@@ -141,7 +187,8 @@ static void xlp_enable_secondary_cores(const cpumask_t *wakeup_mask)
 			nodep->coremask |= 1u << core;
 
 			/* spin until the hw threads sets their ready */
-			wait_for_cpus(cpu, 0);
+			if (!wait_for_cpus(cpu, 0))
+				pr_err("Node %d : timeout core %d\n", n, core);
 		}
 	}
 }
@@ -153,7 +200,8 @@ void xlp_wakeup_secondary_cpus()
 	 * first wakeup core 0 threads
 	 */
 	xlp_boot_core0_siblings();
-	wait_for_cpus(0, 0);
+	if (!wait_for_cpus(0, 0))
+		pr_err("Node 0 : timeout core 0\n");
 
 	/* now get other cores out of reset */
 	xlp_enable_secondary_cores(&nlm_cpumask);
diff --git a/arch/mips/netlogic/xlr/platform.c b/arch/mips/netlogic/xlr/platform.c
index 7b96a91f477..4785932af24 100644
--- a/arch/mips/netlogic/xlr/platform.c
+++ b/arch/mips/netlogic/xlr/platform.c
@@ -23,7 +23,7 @@
 #include <asm/netlogic/xlr/pic.h>
 #include <asm/netlogic/xlr/xlr.h>
 
-unsigned int nlm_xlr_uart_in(struct uart_port *p, int offset)
+static unsigned int nlm_xlr_uart_in(struct uart_port *p, int offset)
 {
 	uint64_t uartbase;
 	unsigned int value;
@@ -41,7 +41,7 @@ unsigned int nlm_xlr_uart_in(struct uart_port *p, int offset)
 	return value;
 }
 
-void nlm_xlr_uart_out(struct uart_port *p, int offset, int value)
+static void nlm_xlr_uart_out(struct uart_port *p, int offset, int value)
 {
 	uint64_t uartbase;
 
diff --git a/arch/mips/netlogic/xlr/setup.c b/arch/mips/netlogic/xlr/setup.c
index 921be5f7779..d118b9aa764 100644
--- a/arch/mips/netlogic/xlr/setup.c
+++ b/arch/mips/netlogic/xlr/setup.c
@@ -60,25 +60,6 @@ unsigned int  nlm_threads_per_core = 1;
 struct nlm_soc_info nlm_nodes[NLM_NR_NODES];
 cpumask_t nlm_cpumask = CPU_MASK_CPU0;
 
-static void __init nlm_early_serial_setup(void)
-{
-	struct uart_port s;
-	unsigned long uart_base;
-
-	uart_base = (unsigned long)nlm_mmio_base(NETLOGIC_IO_UART_0_OFFSET);
-	memset(&s, 0, sizeof(s));
-	s.flags		= ASYNC_BOOT_AUTOCONF | ASYNC_SKIP_TEST;
-	s.iotype	= UPIO_MEM32;
-	s.regshift	= 2;
-	s.irq		= PIC_UART_0_IRQ;
-	s.uartclk	= PIC_CLK_HZ;
-	s.serial_in	= nlm_xlr_uart_in;
-	s.serial_out	= nlm_xlr_uart_out;
-	s.mapbase	= uart_base;
-	s.membase	= (unsigned char __iomem *)uart_base;
-	early_serial_setup(&s);
-}
-
 static void nlm_linux_exit(void)
 {
 	uint64_t gpiobase;
@@ -214,7 +195,6 @@ void __init prom_init(void)
 	memcpy(reset_vec, (void *)nlm_reset_entry,
 			(nlm_reset_entry_end - nlm_reset_entry));
 
-	nlm_early_serial_setup();
 	build_arcs_cmdline(argv);
 	prom_add_memory();
 
diff --git a/arch/mips/netlogic/xlr/wakeup.c b/arch/mips/netlogic/xlr/wakeup.c
index 9fb81fa6272..d61cba1e9c6 100644
--- a/arch/mips/netlogic/xlr/wakeup.c
+++ b/arch/mips/netlogic/xlr/wakeup.c
@@ -32,7 +32,6 @@
  * IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  */
 
-#include <linux/init.h>
 #include <linux/delay.h>
 #include <linux/threads.h>
 
@@ -70,7 +69,7 @@ int xlr_wakeup_secondary_cpus(void)
 
 	/* Fill up the coremask early */
 	nodep->coremask = 1;
-	for (i = 1; i < NLM_CORES_PER_NODE; i++) {
+	for (i = 1; i < nlm_cores_per_node(); i++) {
 		for (j = 1000000; j > 0; j--) {
 			if (cpu_ready[i * NLM_THREADS_PER_CORE])
 				break;