metag: SMP support

Add SMP support for metag. This allows Linux to take control of multiple
hardware threads on a single Meta core, treating them as separate Linux
CPUs.
Signed-off-by: James Hogan <james.hogan@imgtec.com>

arch/metag/include/asm/cachepart.h

+/*
+ * Meta cache partition manipulation.
+ *
+ * Copyright 2010 Imagination Technologies Ltd.
+ */
+
+#ifndef _METAG_CACHEPART_H_
+#define _METAG_CACHEPART_H_
+
+/**
+ * get_dcache_size() - Get size of data cache.
+ */
+unsigned int get_dcache_size(void);
+
+/**
+ * get_icache_size() - Get size of code cache.
+ */
+unsigned int get_icache_size(void);
+
+/**
+ * get_global_dcache_size() - Get the thread's global dcache.
+ *
+ * Returns the size of the current thread's global dcache partition.
+ */
+unsigned int get_global_dcache_size(void);
+
+/**
+ * get_global_icache_size() - Get the thread's global icache.
+ *
+ * Returns the size of the current thread's global icache partition.
+ */
+unsigned int get_global_icache_size(void);
+
+/**
+ * check_for_dache_aliasing() - Ensure that the bootloader has configured the
+ * dache and icache properly to avoid aliasing
+ * @thread_id: Hardware thread ID
+ *
+ */
+void check_for_cache_aliasing(int thread_id);
+
+#endif

arch/metag/include/asm/core_reg.h

+#ifndef __ASM_METAG_CORE_REG_H_
+#define __ASM_METAG_CORE_REG_H_
+
+#include <asm/metag_regs.h>
+
+extern void core_reg_write(int unit, int reg, int thread, unsigned int val);
+extern unsigned int core_reg_read(int unit, int reg, int thread);
+
+/*
+ * These macros allow direct access from C to any register known to the
+ * assembler. Example candidates are TXTACTCYC, TXIDLECYC, and TXPRIVEXT.
+ */
+
+#define __core_reg_get(reg) ({						\
+	unsigned int __grvalue;						\
+	asm volatile("MOV	%0," #reg				\
+		     : "=r" (__grvalue));				\
+	__grvalue;							\
+})
+
+#define __core_reg_set(reg, value) do {					\
+	unsigned int __srvalue = (value);				\
+	asm volatile("MOV	" #reg ",%0"				\
+		     :							\
+		     : "r" (__srvalue));				\
+} while (0)
+
+#define __core_reg_swap(reg, value) do {				\
+	unsigned int __srvalue = (value);				\
+	asm volatile("SWAP	" #reg ",%0"				\
+		     : "+r" (__srvalue));				\
+	(value) = __srvalue;						\
+} while (0)
+
+#endif

arch/metag/include/asm/smp.h

+#ifndef __ASM_SMP_H
+#define __ASM_SMP_H
+
+#include <linux/cpumask.h>
+
+#define raw_smp_processor_id() (current_thread_info()->cpu)
+
+enum ipi_msg_type {
+	IPI_CALL_FUNC,
+	IPI_CALL_FUNC_SINGLE,
+	IPI_RESCHEDULE,
+};
+
+extern void arch_send_call_function_single_ipi(int cpu);
+extern void arch_send_call_function_ipi_mask(const struct cpumask *mask);
+#define arch_send_call_function_ipi_mask arch_send_call_function_ipi_mask
+
+asmlinkage void secondary_start_kernel(void);
+
+extern void secondary_startup(void);
+
+#ifdef CONFIG_HOTPLUG_CPU
+extern void __cpu_die(unsigned int cpu);
+extern int __cpu_disable(void);
+extern void cpu_die(void);
+#endif
+
+extern void smp_init_cpus(void);
+#endif /* __ASM_SMP_H */

arch/metag/include/asm/topology.h

+#ifndef _ASM_METAG_TOPOLOGY_H
+#define _ASM_METAG_TOPOLOGY_H
+
+#ifdef CONFIG_NUMA
+
+/* sched_domains SD_NODE_INIT for Meta machines */
+#define SD_NODE_INIT (struct sched_domain) {		\
+	.parent			= NULL,			\
+	.child			= NULL,			\
+	.groups			= NULL,			\
+	.min_interval		= 8,			\
+	.max_interval		= 32,			\
+	.busy_factor		= 32,			\
+	.imbalance_pct		= 125,			\
+	.cache_nice_tries	= 2,			\
+	.busy_idx		= 3,			\
+	.idle_idx		= 2,			\
+	.newidle_idx		= 0,			\
+	.wake_idx		= 0,			\
+	.forkexec_idx		= 0,			\
+	.flags			= SD_LOAD_BALANCE	\
+				| SD_BALANCE_FORK	\
+				| SD_BALANCE_EXEC	\
+				| SD_BALANCE_NEWIDLE	\
+				| SD_SERIALIZE,		\
+	.last_balance		= jiffies,		\
+	.balance_interval	= 1,			\
+	.nr_balance_failed	= 0,			\
+}
+
+#define cpu_to_node(cpu)	((void)(cpu), 0)
+#define parent_node(node)	((void)(node), 0)
+
+#define cpumask_of_node(node)	((void)node, cpu_online_mask)
+
+#define pcibus_to_node(bus)	((void)(bus), -1)
+#define cpumask_of_pcibus(bus)	(pcibus_to_node(bus) == -1 ? \
+					cpu_all_mask : \
+					cpumask_of_node(pcibus_to_node(bus)))
+
+#endif
+
+#define mc_capable()    (1)
+
+const struct cpumask *cpu_coregroup_mask(unsigned int cpu);
+
+extern cpumask_t cpu_core_map[NR_CPUS];
+
+#define topology_core_cpumask(cpu)	(&cpu_core_map[cpu])
+
+#include <asm-generic/topology.h>
+
+#endif /* _ASM_METAG_TOPOLOGY_H */

arch/metag/kernel/cachepart.c

+/*
+ * Meta cache partition manipulation.
+ *
+ * Copyright 2010 Imagination Technologies Ltd.
+ */
+
+#include <linux/kernel.h>
+#include <linux/io.h>
+#include <linux/errno.h>
+#include <asm/processor.h>
+#include <asm/cachepart.h>
+#include <asm/metag_isa.h>
+#include <asm/metag_mem.h>
+
+#define SYSC_DCPART(n)	(SYSC_DCPART0 + SYSC_xCPARTn_STRIDE * (n))
+#define SYSC_ICPART(n)	(SYSC_ICPART0 + SYSC_xCPARTn_STRIDE * (n))
+
+#define CACHE_ASSOCIATIVITY 4 /* 4 way set-assosiative */
+#define ICACHE 0
+#define DCACHE 1
+
+/* The CORE_CONFIG2 register is not available on Meta 1 */
+#ifdef CONFIG_METAG_META21
+unsigned int get_dcache_size(void)
+{
+	unsigned int config2 = metag_in32(METAC_CORE_CONFIG2);
+	return 0x1000 << ((config2 & METAC_CORECFG2_DCSZ_BITS)
+				>> METAC_CORECFG2_DCSZ_S);
+}
+
+unsigned int get_icache_size(void)
+{
+	unsigned int config2 = metag_in32(METAC_CORE_CONFIG2);
+	return 0x1000 << ((config2 & METAC_CORE_C2ICSZ_BITS)
+				>> METAC_CORE_C2ICSZ_S);
+}
+
+unsigned int get_global_dcache_size(void)
+{
+	unsigned int cpart = metag_in32(SYSC_DCPART(hard_processor_id()));
+	unsigned int temp = cpart & SYSC_xCPARTG_AND_BITS;
+	return (get_dcache_size() * ((temp >> SYSC_xCPARTG_AND_S) + 1)) >> 4;
+}
+
+unsigned int get_global_icache_size(void)
+{
+	unsigned int cpart = metag_in32(SYSC_ICPART(hard_processor_id()));
+	unsigned int temp = cpart & SYSC_xCPARTG_AND_BITS;
+	return (get_icache_size() * ((temp >> SYSC_xCPARTG_AND_S) + 1)) >> 4;
+}
+
+static unsigned int get_thread_cache_size(unsigned int cache, int thread_id)
+{
+	unsigned int cache_size;
+	unsigned int t_cache_part;
+	unsigned int isEnabled;
+	unsigned int offset = 0;
+	isEnabled = (cache == DCACHE ? metag_in32(MMCU_DCACHE_CTRL_ADDR) & 0x1 :
+		metag_in32(MMCU_ICACHE_CTRL_ADDR) & 0x1);
+	if (!isEnabled)
+		return 0;
+#if PAGE_OFFSET >= LINGLOBAL_BASE
+	/* Checking for global cache */
+	cache_size = (cache == DCACHE ? get_global_dache_size() :
+		get_global_icache_size());
+	offset = 8;
+#else
+	cache_size = (cache == DCACHE ? get_dcache_size() :
+		get_icache_size());
+#endif
+	t_cache_part = (cache == DCACHE ?
+		(metag_in32(SYSC_DCPART(thread_id)) >> offset) & 0xF :
+		(metag_in32(SYSC_ICPART(thread_id)) >> offset) & 0xF);
+	switch (t_cache_part) {
+	case 0xF:
+		return cache_size;
+	case 0x7:
+		return cache_size / 2;
+	case 0x3:
+		return cache_size / 4;
+	case 0x1:
+		return cache_size / 8;
+	case 0:
+		return cache_size / 16;
+	}
+	return -1;
+}
+
+void check_for_cache_aliasing(int thread_id)
+{
+	unsigned int thread_cache_size;
+	unsigned int cache_type;
+	for (cache_type = ICACHE; cache_type <= DCACHE; cache_type++) {
+		thread_cache_size =
+				get_thread_cache_size(cache_type, thread_id);
+		if (thread_cache_size < 0)
+			pr_emerg("Can't read %s cache size", \
+				 cache_type ? "DCACHE" : "ICACHE");
+		else if (thread_cache_size == 0)
+			/* Cache is off. No need to check for aliasing */
+			continue;
+		if (thread_cache_size / CACHE_ASSOCIATIVITY > PAGE_SIZE) {
+			pr_emerg("Cache aliasing detected in %s on Thread %d",
+				 cache_type ? "DCACHE" : "ICACHE", thread_id);
+			pr_warn("Total %s size: %u bytes",
+				cache_type ? "DCACHE" : "ICACHE ",
+				cache_type ? get_dcache_size()
+				: get_icache_size());
+			pr_warn("Thread %s size: %d bytes",
+				cache_type ? "CACHE" : "ICACHE",
+				thread_cache_size);
+			pr_warn("Page Size: %lu bytes", PAGE_SIZE);
+		}
+	}
+}
+
+#else
+
+void check_for_cache_aliasing(int thread_id)
+{
+	return;
+}
+
+#endif

arch/metag/kernel/core_reg.c

+/*
+ *  Support for reading and writing Meta core internal registers.
+ *
+ *  Copyright (C) 2011 Imagination Technologies Ltd.
+ *
+ */
+
+#include <linux/delay.h>
+#include <linux/export.h>
+
+#include <asm/core_reg.h>
+#include <asm/global_lock.h>
+#include <asm/hwthread.h>
+#include <asm/io.h>
+#include <asm/metag_mem.h>
+#include <asm/metag_regs.h>
+
+#define UNIT_BIT_MASK		TXUXXRXRQ_UXX_BITS
+#define REG_BIT_MASK		TXUXXRXRQ_RX_BITS
+#define THREAD_BIT_MASK		TXUXXRXRQ_TX_BITS
+
+#define UNIT_SHIFTS		TXUXXRXRQ_UXX_S
+#define REG_SHIFTS		TXUXXRXRQ_RX_S
+#define THREAD_SHIFTS		TXUXXRXRQ_TX_S
+
+#define UNIT_VAL(x)		(((x) << UNIT_SHIFTS) & UNIT_BIT_MASK)
+#define REG_VAL(x)		(((x) << REG_SHIFTS) & REG_BIT_MASK)
+#define THREAD_VAL(x)		(((x) << THREAD_SHIFTS) & THREAD_BIT_MASK)
+
+/*
+ * core_reg_write() - modify the content of a register in a core unit.
+ * @unit:	The unit to be modified.
+ * @reg:	Register number within the unit.
+ * @thread:	The thread we want to access.
+ * @val:	The new value to write.
+ *
+ * Check asm/metag_regs.h for a list/defines of supported units (ie: TXUPC_ID,
+ * TXUTR_ID, etc), and regnums within the units (ie: TXMASKI_REGNUM,
+ * TXPOLLI_REGNUM, etc).
+ */
+void core_reg_write(int unit, int reg, int thread, unsigned int val)
+{
+	unsigned long flags;
+
+	/* TXUCT_ID has its own memory mapped registers */
+	if (unit == TXUCT_ID) {
+		void __iomem *cu_reg = __CU_addr(thread, reg);
+		metag_out32(val, cu_reg);
+		return;
+	}
+
+	__global_lock2(flags);
+
+	/* wait for ready */
+	while (!(metag_in32(TXUXXRXRQ) & TXUXXRXRQ_DREADY_BIT))
+		udelay(10);
+
+	/* set the value to write */
+	metag_out32(val, TXUXXRXDT);
+
+	/* set the register to write */
+	val = UNIT_VAL(unit) | REG_VAL(reg) | THREAD_VAL(thread);
+	metag_out32(val, TXUXXRXRQ);
+
+	/* wait for finish */
+	while (!(metag_in32(TXUXXRXRQ) & TXUXXRXRQ_DREADY_BIT))
+		udelay(10);
+
+	__global_unlock2(flags);
+}
+EXPORT_SYMBOL(core_reg_write);
+
+/*
+ * core_reg_read() - read the content of a register in a core unit.
+ * @unit:	The unit to be modified.
+ * @reg:	Register number within the unit.
+ * @thread:	The thread we want to access.
+ *
+ * Check asm/metag_regs.h for a list/defines of supported units (ie: TXUPC_ID,
+ * TXUTR_ID, etc), and regnums within the units (ie: TXMASKI_REGNUM,
+ * TXPOLLI_REGNUM, etc).
+ */
+unsigned int core_reg_read(int unit, int reg, int thread)
+{
+	unsigned long flags;
+	unsigned int val;
+
+	/* TXUCT_ID has its own memory mapped registers */
+	if (unit == TXUCT_ID) {
+		void __iomem *cu_reg = __CU_addr(thread, reg);
+		val = metag_in32(cu_reg);
+		return val;
+	}
+
+	__global_lock2(flags);
+
+	/* wait for ready */
+	while (!(metag_in32(TXUXXRXRQ) & TXUXXRXRQ_DREADY_BIT))
+		udelay(10);
+
+	/* set the register to read */
+	val = (UNIT_VAL(unit) | REG_VAL(reg) | THREAD_VAL(thread) |
+							TXUXXRXRQ_RDnWR_BIT);
+	metag_out32(val, TXUXXRXRQ);
+
+	/* wait for finish */
+	while (!(metag_in32(TXUXXRXRQ) & TXUXXRXRQ_DREADY_BIT))
+		udelay(10);
+
+	/* read the register value */
+	val = metag_in32(TXUXXRXDT);
+
+	__global_unlock2(flags);
+
+	return val;
+}
+EXPORT_SYMBOL(core_reg_read);

arch/metag/kernel/head.S

@@ -43,3 +43,15 @@ __start:
 __exit:
 	XOR     TXENABLE,D0Re0,D0Re0
 	.size	__exit,.-__exit
+
+#ifdef CONFIG_SMP
+	.global _secondary_startup
+	.type _secondary_startup,function
+_secondary_startup:
+	MOVT	A0StP,#HI(_secondary_data_stack)
+	ADD	A0StP,A0StP,#LO(_secondary_data_stack)
+	GETD	A0StP,[A0StP]
+	ADD	A0StP,A0StP,#THREAD_INFO_SIZE
+	B	_secondary_start_kernel
+	.size	_secondary_startup,.-_secondary_startup
+#endif

arch/metag/kernel/topology.c

+/*
+ *  Copyright (C) 2007  Paul Mundt
+ *  Copyright (C) 2010  Imagination Technolohies Ltd.
+ *
+ * This file is subject to the terms and conditions of the GNU General Public
+ * License.  See the file "COPYING" in the main directory of this archive
+ * for more details.
+ */
+#include <linux/cpu.h>
+#include <linux/cpumask.h>
+#include <linux/init.h>
+#include <linux/percpu.h>
+#include <linux/node.h>
+#include <linux/nodemask.h>
+#include <linux/topology.h>
+
+#include <asm/cpu.h>
+
+DEFINE_PER_CPU(struct cpuinfo_metag, cpu_data);
+
+cpumask_t cpu_core_map[NR_CPUS];
+
+static cpumask_t cpu_coregroup_map(unsigned int cpu)
+{
+	return *cpu_possible_mask;
+}
+
+const struct cpumask *cpu_coregroup_mask(unsigned int cpu)
+{
+	return &cpu_core_map[cpu];
+}
+
+int arch_update_cpu_topology(void)
+{
+	unsigned int cpu;
+
+	for_each_possible_cpu(cpu)
+		cpu_core_map[cpu] = cpu_coregroup_map(cpu);
+
+	return 0;
+}
+
+static int __init topology_init(void)
+{
+	int i, ret;
+
+#ifdef CONFIG_NEED_MULTIPLE_NODES
+	for_each_online_node(i)
+		register_one_node(i);
+#endif
+
+	for_each_present_cpu(i) {
+		struct cpuinfo_metag *cpuinfo = &per_cpu(cpu_data, i);
+#ifdef CONFIG_HOTPLUG_CPU
+		cpuinfo->cpu.hotpluggable = 1;
+#endif
+		ret = register_cpu(&cpuinfo->cpu, i);
+		if (unlikely(ret))
+			pr_warn("%s: register_cpu %d failed (%d)\n",
+				__func__, i, ret);
+	}
+
+#if defined(CONFIG_NUMA) && !defined(CONFIG_SMP)
+	/*
+	 * In the UP case, make sure the CPU association is still
+	 * registered under each node. Without this, sysfs fails
+	 * to make the connection between nodes other than node0
+	 * and cpu0.
+	 */
+	for_each_online_node(i)
+		if (i != numa_node_id())
+			register_cpu_under_node(raw_smp_processor_id(), i);
+#endif
+
+	return 0;
+}
+subsys_initcall(topology_init);