xtensa: add SMP support

This is largely based on SMP code from the xtensa-2.6.29-smp tree by
Piet Delaney, Marc Gauthier, Joe Taylor, Christian Zankel (and possibly
other Tensilica folks).

Signed-off-by: Max Filippov <jcmvbkbc@gmail.com>
Signed-off-by: Chris Zankel <chris@zankel.net>

arch/xtensa/Kconfig

@@ -9,7 +9,6 @@ config XTENSA
 	select GENERIC_CLOCKEVENTS
 	select VIRT_TO_BUS
 	select GENERIC_IRQ_SHOW
-	select GENERIC_CPU_DEVICES
 	select GENERIC_SCHED_CLOCK
 	select MODULES_USE_ELF_RELA
 	select GENERIC_PCI_IOMAP
@@ -65,6 +64,9 @@ config MMU
 config VARIANT_IRQ_SWITCH
 	def_bool n
 
+config MAY_HAVE_SMP
+	def_bool n
+
 menu "Processor type and features"
 
 choice
@@ -105,6 +107,39 @@ config XTENSA_UNALIGNED_USER
 
 source "kernel/Kconfig.preempt"
 
+config HAVE_SMP
+	bool "System Supports SMP (MX)"
+	depends on MAY_HAVE_SMP
+	select XTENSA_MX
+	help
+	  This option is use to indicate that the system-on-a-chip (SOC)
+	  supports Multiprocessing. Multiprocessor support implemented above
+	  the CPU core definition and currently needs to be selected manually.
+
+	  Multiprocessor support in implemented with external cache and
+	  interrupt controlers.
+
+	  The MX interrupt distributer adds Interprocessor Interrupts
+	  and causes the IRQ numbers to be increased by 4 for devices
+	  like the open cores ethernet driver and the serial interface.
+
+	  You still have to select "Enable SMP" to enable SMP on this SOC.
+
+config SMP
+	bool "Enable Symmetric multi-processing support"
+	depends on HAVE_SMP
+	select USE_GENERIC_SMP_HELPERS
+	select GENERIC_SMP_IDLE_THREAD
+	help
+	  Enabled SMP Software; allows more than one CPU/CORE
+	  to be activated during startup.
+
+config NR_CPUS
+	depends on SMP
+	int "Maximum number of CPUs (2-32)"
+	range 2 32
+	default "4"
+
 config MATH_EMULATION
 	bool "Math emulation"
 	help

arch/xtensa/include/asm/barrier.h

@@ -17,7 +17,9 @@
 #define wmb() mb()
 
 #ifdef CONFIG_SMP
-#error smp_* not defined
+#define smp_mb()	mb()
+#define smp_rmb()	rmb()
+#define smp_wmb()	wmb()
 #else
 #define smp_mb()	barrier()
 #define smp_rmb()	barrier()

arch/xtensa/include/asm/bitops.h

@@ -22,12 +22,8 @@
 #include <asm/processor.h>
 #include <asm/byteorder.h>
 
-#ifdef CONFIG_SMP
-# error SMP not supported on this architecture
-#endif
-
-#define smp_mb__before_clear_bit()	barrier()
-#define smp_mb__after_clear_bit()	barrier()
+#define smp_mb__before_clear_bit()	smp_mb()
+#define smp_mb__after_clear_bit()	smp_mb()
 
 #include <asm-generic/bitops/non-atomic.h>
 

arch/xtensa/include/asm/cacheflush.h

@@ -1,18 +1,14 @@
 /*
- * include/asm-xtensa/cacheflush.h
- *
  * 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.
  *
- * (C) 2001 - 2007 Tensilica Inc.
+ * (C) 2001 - 2013 Tensilica Inc.
  */
 
 #ifndef _XTENSA_CACHEFLUSH_H
 #define _XTENSA_CACHEFLUSH_H
 
-#ifdef __KERNEL__
-
 #include <linux/mm.h>
 #include <asm/processor.h>
 #include <asm/page.h>
@@ -51,7 +47,6 @@ extern void __invalidate_icache_page(unsigned long);
 extern void __invalidate_icache_range(unsigned long, unsigned long);
 extern void __invalidate_dcache_range(unsigned long, unsigned long);
 
-
 #if XCHAL_DCACHE_IS_WRITEBACK
 extern void __flush_invalidate_dcache_all(void);
 extern void __flush_dcache_page(unsigned long);
@@ -87,9 +82,22 @@ static inline void __invalidate_icache_page_alias(unsigned long virt,
  * (see also Documentation/cachetlb.txt)
  */
 
-#if (DCACHE_WAY_SIZE > PAGE_SIZE)
+#if (DCACHE_WAY_SIZE > PAGE_SIZE) || defined(CONFIG_SMP)
+
+#ifdef CONFIG_SMP
+void flush_cache_all(void);
+void flush_cache_range(struct vm_area_struct*, ulong, ulong);
+void flush_icache_range(unsigned long start, unsigned long end);
+void flush_cache_page(struct vm_area_struct*,
+			     unsigned long, unsigned long);
+#else
+#define flush_cache_all local_flush_cache_all
+#define flush_cache_range local_flush_cache_range
+#define flush_icache_range local_flush_icache_range
+#define flush_cache_page  local_flush_cache_page
+#endif
 
-#define flush_cache_all()						\
+#define local_flush_cache_all()						\
 	do {								\
 		__flush_invalidate_dcache_all();			\
 		__invalidate_icache_all();				\
@@ -103,9 +111,11 @@ static inline void __invalidate_icache_page_alias(unsigned long virt,
 
 #define ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE 1
 extern void flush_dcache_page(struct page*);
-extern void flush_cache_range(struct vm_area_struct*, ulong, ulong);
-extern void flush_cache_page(struct vm_area_struct*,
-			     unsigned long, unsigned long);
+
+void local_flush_cache_range(struct vm_area_struct *vma,
+		unsigned long start, unsigned long end);
+void local_flush_cache_page(struct vm_area_struct *vma,
+		unsigned long address, unsigned long pfn);
 
 #else
 
@@ -119,13 +129,14 @@ extern void flush_cache_page(struct vm_area_struct*,
 #define ARCH_IMPLEMENTS_FLUSH_DCACHE_PAGE 0
 #define flush_dcache_page(page)				do { } while (0)
 
-#define flush_cache_page(vma,addr,pfn)			do { } while (0)
-#define flush_cache_range(vma,start,end)		do { } while (0)
+#define flush_icache_range local_flush_icache_range
+#define flush_cache_page(vma, addr, pfn)		do { } while (0)
+#define flush_cache_range(vma, start, end)		do { } while (0)
 
 #endif
 
 /* Ensure consistency between data and instruction cache. */
-#define flush_icache_range(start,end) 					\
+#define local_flush_icache_range(start, end)				\
 	do {								\
 		__flush_dcache_range(start, (end) - (start));		\
 		__invalidate_icache_range(start,(end) - (start));	\
@@ -253,5 +264,4 @@ static inline void flush_invalidate_dcache_unaligned(u32 addr, u32 size)
 	}
 }
 
-#endif /* __KERNEL__ */
 #endif /* _XTENSA_CACHEFLUSH_H */

arch/xtensa/include/asm/mmu.h

@@ -1,11 +1,9 @@
 /*
- * include/asm-xtensa/mmu.h
- *
  * 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.
  *
- * Copyright (C) 2001 - 2005 Tensilica Inc.
+ * Copyright (C) 2001 - 2013 Tensilica Inc.
  */
 
 #ifndef _XTENSA_MMU_H
@@ -15,8 +13,10 @@
 #include <asm-generic/mmu.h>
 #else
 
-/* Default "unsigned long" context */
-typedef unsigned long mm_context_t;
+typedef struct {
+	unsigned long asid[NR_CPUS];
+	unsigned int cpu;
+} mm_context_t;
 
 #endif /* CONFIG_MMU */
 #endif	/* _XTENSA_MMU_H */

arch/xtensa/include/asm/mmu_context.h

@@ -1,13 +1,11 @@
 /*
- * include/asm-xtensa/mmu_context.h
- *
  * Switch an MMU context.
  *
  * 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.
  *
- * Copyright (C) 2001 - 2005 Tensilica Inc.
+ * Copyright (C) 2001 - 2013 Tensilica Inc.
  */
 
 #ifndef _XTENSA_MMU_CONTEXT_H
@@ -20,22 +18,25 @@
 #include <linux/stringify.h>
 #include <linux/sched.h>
 
-#include <variant/core.h>
+#include <asm/vectors.h>
 
 #include <asm/pgtable.h>
 #include <asm/cacheflush.h>
 #include <asm/tlbflush.h>
 #include <asm-generic/mm_hooks.h>
+#include <asm-generic/percpu.h>
 
 #if (XCHAL_HAVE_TLBS != 1)
 # error "Linux must have an MMU!"
 #endif
 
-extern unsigned long asid_cache;
+DECLARE_PER_CPU(unsigned long, asid_cache);
+#define cpu_asid_cache(cpu) per_cpu(asid_cache, cpu)
 
 /*
  * NO_CONTEXT is the invalid ASID value that we don't ever assign to
- * any user or kernel context.
+ * any user or kernel context.  We use the reserved values in the
+ * ASID_INSERT macro below.
  *
  * 0 invalid
  * 1 kernel
@@ -68,64 +69,77 @@ static inline unsigned long get_rasid_register (void)
 	return tmp;
 }
 
-static inline void
-__get_new_mmu_context(struct mm_struct *mm)
+static inline void get_new_mmu_context(struct mm_struct *mm, unsigned int cpu)
+{
+	unsigned long asid = cpu_asid_cache(cpu);
+	if ((++asid & ASID_MASK) == 0) {
+		/*
+		 * Start new asid cycle; continue counting with next
+		 * incarnation bits; skipping over 0, 1, 2, 3.
+		 */
+		local_flush_tlb_all();
+		asid += ASID_USER_FIRST;
+	}
+	cpu_asid_cache(cpu) = asid;
+	mm->context.asid[cpu] = asid;
+	mm->context.cpu = cpu;
+}
+
+static inline void get_mmu_context(struct mm_struct *mm, unsigned int cpu)
 {
-	extern void flush_tlb_all(void);
-	if (! (++asid_cache & ASID_MASK) ) {
-		flush_tlb_all(); /* start new asid cycle */
-		asid_cache += ASID_USER_FIRST;
+	/*
+	 * Check if our ASID is of an older version and thus invalid.
+	 */
+
+	if (mm) {
+		unsigned long asid = mm->context.asid[cpu];
+
+		if (asid == NO_CONTEXT ||
+				((asid ^ cpu_asid_cache(cpu)) & ~ASID_MASK))
+			get_new_mmu_context(mm, cpu);
 	}
-	mm->context = asid_cache;
 }
 
-static inline void
-__load_mmu_context(struct mm_struct *mm)
+static inline void activate_context(struct mm_struct *mm, unsigned int cpu)
 {
-	set_rasid_register(ASID_INSERT(mm->context));
+	get_mmu_context(mm, cpu);
+	set_rasid_register(ASID_INSERT(mm->context.asid[cpu]));
 	invalidate_page_directory();
 }
 
 /*
  * Initialize the context related info for a new mm_struct
- * instance.
+ * instance.  Valid cpu values are 0..(NR_CPUS-1), so initializing
+ * to -1 says the process has never run on any core.
  */
 
-static inline int
-init_new_context(struct task_struct *tsk, struct mm_struct *mm)
+static inline int init_new_context(struct task_struct *tsk,
+		struct mm_struct *mm)
 {
-	mm->context = NO_CONTEXT;
+	int cpu;
+	for_each_possible_cpu(cpu) {
+		mm->context.asid[cpu] = NO_CONTEXT;
+	}
+	mm->context.cpu = -1;
 	return 0;
 }
 
-/*
- * After we have set current->mm to a new value, this activates
- * the context for the new mm so we see the new mappings.
- */
-static inline void
-activate_mm(struct mm_struct *prev, struct mm_struct *next)
-{
-	/* Unconditionally get a new ASID.  */
-
-	__get_new_mmu_context(next);
-	__load_mmu_context(next);
-}
-
-
 static inline void switch_mm(struct mm_struct *prev, struct mm_struct *next,
 			     struct task_struct *tsk)
 {
-	unsigned long asid = asid_cache;
-
-	/* Check if our ASID is of an older version and thus invalid */
-
-	if (next->context == NO_CONTEXT || ((next->context^asid) & ~ASID_MASK))
-		__get_new_mmu_context(next);
-
-	__load_mmu_context(next);
+	unsigned int cpu = smp_processor_id();
+	int migrated = next->context.cpu != cpu;
+	/* Flush the icache if we migrated to a new core. */
+	if (migrated) {
+		__invalidate_icache_all();
+		next->context.cpu = cpu;
+	}
+	if (migrated || prev != next)
+		activate_context(next, cpu);
 }
 
-#define deactivate_mm(tsk, mm)	do { } while(0)
+#define activate_mm(prev, next)	switch_mm((prev), (next), NULL)
+#define deactivate_mm(tsk, mm)	do { } while (0)
 
 /*
  * Destroy context related info for an mm_struct that is about

arch/xtensa/include/asm/ptrace.h

@@ -59,9 +59,17 @@ struct pt_regs {
 	(task_stack_page(tsk) + KERNEL_STACK_SIZE - (XCHAL_NUM_AREGS-16)*4) - 1)
 # define user_mode(regs) (((regs)->ps & 0x00000020)!=0)
 # define instruction_pointer(regs) ((regs)->pc)
+# define return_pointer(regs) (MAKE_PC_FROM_RA((regs)->areg[0], \
+					       (regs)->areg[1]))
 
 # ifndef CONFIG_SMP
 #  define profile_pc(regs) instruction_pointer(regs)
+# else
+#  define profile_pc(regs)						\
+	({								\
+		in_lock_functions(instruction_pointer(regs)) ?		\
+		return_pointer(regs) : instruction_pointer(regs);	\
+	})
 # endif
 
 #define user_stack_pointer(regs) ((regs)->areg[1])