MIPS: Netlogic: Add cpu to node mapping for XLP9XX

XLP9XX has 20 cores per node, opposed to 8 on earlier XLP8XX.
Update code that calculates node id from cpu id to handle this.
Signed-off-by: Jayachandran C <jchandra@broadcom.com>
Signed-off-by: John Crispin <blogic@openwrt.org>
Patchwork: http://patchwork.linux-mips.org/patch/6283/

arch/mips/include/asm/mach-netlogic/multi-node.h

@@ -47,9 +47,16 @@
 #endif
 #endif
 
-#define NLM_CORES_PER_NODE	8
 #define NLM_THREADS_PER_CORE	4
-#define NLM_CPUS_PER_NODE	(NLM_CORES_PER_NODE * NLM_THREADS_PER_CORE)
+#ifdef CONFIG_CPU_XLR
+#define nlm_cores_per_node()	8
+#else
+extern unsigned int xlp_cores_per_node;
+#define nlm_cores_per_node()	xlp_cores_per_node
+#endif
+
+#define nlm_threads_per_node()	(nlm_cores_per_node() * NLM_THREADS_PER_CORE)
+#define nlm_cpuid_to_node(c)	((c) / nlm_threads_per_node())
 
 struct nlm_soc_info {
 	unsigned long	coremask;	/* cores enabled on the soc */

arch/mips/include/asm/netlogic/mips-extns.h

@@ -146,7 +146,12 @@ static inline int hard_smp_processor_id(void)
 
 static inline int nlm_nodeid(void)
 {
-	return (__read_32bit_c0_register($15, 1) >> 5) & 0x3;
+	uint32_t prid = read_c0_prid();
+
+	if ((prid & 0xff00) == PRID_IMP_NETLOGIC_XLP9XX)
+		return (__read_32bit_c0_register($15, 1) >> 7) & 0x7;
+	else
+		return (__read_32bit_c0_register($15, 1) >> 5) & 0x3;
 }
 
 static inline unsigned int nlm_core_id(void)

arch/mips/netlogic/common/irq.c

@@ -223,7 +223,7 @@ static void nlm_init_node_irqs(int node)
 			continue;
 
 		nlm_pic_init_irt(nodep->picbase, irt, i,
-					node * NLM_CPUS_PER_NODE, 0);
+				node * nlm_threads_per_node(), 0);
 		nlm_setup_pic_irq(node, i, i, irt);
 	}
 }
@@ -232,8 +232,8 @@ 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);

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;
 		}
 	}

arch/mips/netlogic/xlp/setup.c

@@ -51,6 +51,7 @@ 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)
 {
@@ -154,6 +155,10 @@ 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);

arch/mips/netlogic/xlp/wakeup.c

@@ -165,7 +165,7 @@ static void xlp_enable_secondary_cores(const cpumask_t *wakeup_mask)
 			nodep->coremask = 1;
 
 		pr_info("Node %d - SYS/FUSE coremask %x\n", n, syscoremask);
-		for (core = 0; core < NLM_CORES_PER_NODE; core++) {
+		for (core = 0; core < nlm_cores_per_node(); core++) {
 			/* we will be on node 0 core 0 */
 			if (n == 0 && core == 0)
 				continue;
@@ -175,7 +175,7 @@ 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;

arch/mips/netlogic/xlr/wakeup.c

@@ -70,7 +70,7 @@ 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;

arch/mips/pci/msi-xlp.c

@@ -280,7 +280,7 @@ static int xlp_setup_msi(uint64_t lnkbase, int node, int link,
 		irt = PIC_IRT_PCIE_LINK_INDEX(link);
 		nlm_setup_pic_irq(node, lirq, lirq, irt);
 		nlm_pic_init_irt(nlm_get_node(node)->picbase, irt, lirq,
-				 node * NLM_CPUS_PER_NODE, 1 /*en */);
+				 node * nlm_threads_per_node(), 1 /*en */);
 	}
 
 	/* allocate a MSI vec, and tell the bridge about it */
@@ -443,7 +443,7 @@ void __init xlp_init_node_msi_irqs(int node, int link)
 		msixvec = link * XLP_MSIXVEC_PER_LINK + i;
 		irt = PIC_IRT_PCIE_MSIX_INDEX(msixvec);
 		nlm_pic_init_irt(nodep->picbase, irt, PIC_PCIE_MSIX_IRQ(link),
-			node * NLM_CPUS_PER_NODE, 1 /* enable */);
+			node * nlm_threads_per_node(), 1 /* enable */);
 
 		/* Initialize MSI-X extended irq space for the link  */
 		irq = nlm_irq_to_xirq(node, nlm_link_msixirq(link, i));