sh: Use a per-cpu ASID cache.

Previously this was implemented using a global cache, cache
this per-CPU instead and bump up the number of context IDs to
match NR_CPUS.

Signed-off-by: Paul Mundt <lethal@linux-sh.org>

7 files changed, 97 insertions, 93 deletions

diff --git a/arch/sh/kernel/cpu/init.c b/arch/sh/kernel/cpu/init.c
index 48121766e8d2..6c3c7687e81f 100644
--- a/arch/sh/kernel/cpu/init.c
+++ b/arch/sh/kernel/cpu/init.c
@@ -3,7 +3,7 @@
  *
  * CPU init code
  *
- * Copyright (C) 2002, 2003  Paul Mundt
+ * Copyright (C) 2002 - 2006  Paul Mundt
  * Copyright (C) 2003  Richard Curnow
  *
  * This file is subject to the terms and conditions of the GNU General Public
@@ -12,6 +12,8 @@
  */
 #include <linux/init.h>
 #include <linux/kernel.h>
+#include <linux/mm.h>
+#include <asm/mmu_context.h>
 #include <asm/processor.h>
 #include <asm/uaccess.h>
 #include <asm/page.h>
@@ -218,6 +220,12 @@ asmlinkage void __init sh_cpu_init(void)
 		clear_used_math();
 	}
 
+	/*
+	 * Initialize the per-CPU ASID cache very early, since the
+	 * TLB flushing routines depend on this being setup.
+	 */
+	current_cpu_data.asid_cache = NO_CONTEXT;
+
 #ifdef CONFIG_SH_DSP
 	/* Probe for DSP */
 	dsp_init();
@@ -240,4 +248,3 @@ asmlinkage void __init sh_cpu_init(void)
 	ubc_wakeup();
 #endif
 }
-
diff --git a/arch/sh/kernel/process.c b/arch/sh/kernel/process.c
index cc8f306fd682..0298f0faa6e6 100644
--- a/arch/sh/kernel/process.c
+++ b/arch/sh/kernel/process.c
@@ -1,42 +1,30 @@
-/* $Id: process.c,v 1.28 2004/05/05 16:54:23 lethal Exp $
+/*
+ * arch/sh/kernel/process.c
  *
- *  linux/arch/sh/kernel/process.c
+ * This file handles the architecture-dependent parts of process handling..
  *
  *  Copyright (C) 1995  Linus Torvalds
  *
  *  SuperH version:  Copyright (C) 1999, 2000  Niibe Yutaka & Kaz Kojima
  *		     Copyright (C) 2006 Lineo Solutions Inc. support SH4A UBC
+ *		     Copyright (C) 2002 - 2006  Paul Mundt
  */
-
-/*
- * This file handles the architecture-dependent parts of process handling..
- */
-
 #include <linux/module.h>
-#include <linux/unistd.h>
 #include <linux/mm.h>
 #include <linux/elfcore.h>
-#include <linux/a.out.h>
-#include <linux/slab.h>
 #include <linux/pm.h>
-#include <linux/ptrace.h>
 #include <linux/kallsyms.h>
 #include <linux/kexec.h>
-
-#include <asm/io.h>
 #include <asm/uaccess.h>
 #include <asm/mmu_context.h>
-#include <asm/elf.h>
 #include <asm/ubc.h>
 
-static int hlt_counter=0;
-
+static int hlt_counter;
 int ubc_usercnt = 0;
 
 #define HARD_IDLE_TIMEOUT (HZ / 3)
 
 void (*pm_idle)(void);
-
 void (*pm_power_off)(void);
 EXPORT_SYMBOL(pm_power_off);
 
@@ -44,14 +32,12 @@ void disable_hlt(void)
 {
 	hlt_counter++;
 }
-
 EXPORT_SYMBOL(disable_hlt);
 
 void enable_hlt(void)
 {
 	hlt_counter--;
 }
-
 EXPORT_SYMBOL(enable_hlt);
 
 void default_idle(void)
@@ -152,19 +138,21 @@ __asm__(".align 5\n"
 	".align 2\n\t"
 	"1:.long do_exit");
 
+/* Don't use this in BL=1(cli).  Or else, CPU resets! */
 int kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
-{	/* Don't use this in BL=1(cli).  Or else, CPU resets! */
+{
 	struct pt_regs regs;
 
 	memset(&regs, 0, sizeof(regs));
-	regs.regs[4] = (unsigned long) arg;
-	regs.regs[5] = (unsigned long) fn;
+	regs.regs[4] = (unsigned long)arg;
+	regs.regs[5] = (unsigned long)fn;
 
-	regs.pc = (unsigned long) kernel_thread_helper;
+	regs.pc = (unsigned long)kernel_thread_helper;
 	regs.sr = (1 << 30);
 
 	/* Ok, create the new process.. */
-	return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, &regs, 0, NULL, NULL);
+	return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0,
+		       &regs, 0, NULL, NULL);
 }
 
 /*
@@ -211,21 +199,20 @@ int dump_fpu(struct pt_regs *regs, elf_fpregset_t *fpu)
 	return fpvalid;
 }
 
-/* 
+/*
  * Capture the user space registers if the task is not running (in user space)
  */
 int dump_task_regs(struct task_struct *tsk, elf_gregset_t *regs)
 {
 	struct pt_regs ptregs;
-	
+
 	ptregs = *task_pt_regs(tsk);
 	elf_core_copy_regs(regs, &ptregs);
 
 	return 1;
 }
 
-int
-dump_task_fpu (struct task_struct *tsk, elf_fpregset_t *fpu)
+int dump_task_fpu(struct task_struct *tsk, elf_fpregset_t *fpu)
 {
 	int fpvalid = 0;
 
@@ -263,12 +250,14 @@ int copy_thread(int nr, unsigned long clone_flags, unsigned long usp,
 		childregs->regs[15] = usp;
 		ti->addr_limit = USER_DS;
 	} else {
-		childregs->regs[15] = (unsigned long)task_stack_page(p) + THREAD_SIZE;
+		childregs->regs[15] = (unsigned long)task_stack_page(p) +
+							THREAD_SIZE;
 		ti->addr_limit = KERNEL_DS;
 	}
-        if (clone_flags & CLONE_SETTLS) {
+
+        if (clone_flags & CLONE_SETTLS)
 		childregs->gbr = childregs->regs[0];
-	}
+
 	childregs->regs[0] = 0; /* Set return value for child */
 
 	p->thread.sp = (unsigned long) childregs;
@@ -280,8 +269,7 @@ int copy_thread(int nr, unsigned long clone_flags, unsigned long usp,
 }
 
 /* Tracing by user break controller.  */
-static void
-ubc_set_tracing(int asid, unsigned long pc)
+static void ubc_set_tracing(int asid, unsigned long pc)
 {
 #if defined(CONFIG_CPU_SH4A)
 	unsigned long val;
@@ -297,7 +285,7 @@ ubc_set_tracing(int asid, unsigned long pc)
 	val = (UBC_CRR_RES | UBC_CRR_PCB | UBC_CRR_BIE);
 	ctrl_outl(val, UBC_CRR0);
 
-	/* Read UBC register that we writed last. For chekking UBC Register changed */
+	/* Read UBC register that we wrote last, for checking update */
 	val = ctrl_inl(UBC_CRR0);
 
 #else	/* CONFIG_CPU_SH4A */
@@ -325,7 +313,8 @@ ubc_set_tracing(int asid, unsigned long pc)
  *	switch_to(x,y) should switch tasks from x to y.
  *
  */
-struct task_struct *__switch_to(struct task_struct *prev, struct task_struct *next)
+struct task_struct *__switch_to(struct task_struct *prev,
+				struct task_struct *next)
 {
 #if defined(CONFIG_SH_FPU)
 	unlazy_fpu(prev, task_pt_regs(prev));
@@ -354,7 +343,7 @@ struct task_struct *__switch_to(struct task_struct *prev, struct task_struct *ne
 #ifdef CONFIG_MMU
 	/*
 	 * Restore the kernel mode register
-	 *   	k7 (r7_bank1)
+	 *	k7 (r7_bank1)
 	 */
 	asm volatile("ldc	%0, r7_bank"
 		     : /* no output */
@@ -367,7 +356,7 @@ struct task_struct *__switch_to(struct task_struct *prev, struct task_struct *ne
 	else if (next->thread.ubc_pc && next->mm) {
 		int asid = 0;
 #ifdef CONFIG_MMU
-		asid |= next->mm->context.id & MMU_CONTEXT_ASID_MASK;
+		asid |= cpu_asid(smp_processor_id(), next->mm);
 #endif
 		ubc_set_tracing(asid, next->thread.ubc_pc);
 	} else {
@@ -405,7 +394,8 @@ asmlinkage int sys_clone(unsigned long clone_flags, unsigned long newsp,
 	if (!newsp)
 		newsp = regs->regs[15];
 	return do_fork(clone_flags, newsp, regs, 0,
-			(int __user *)parent_tidptr, (int __user *)child_tidptr);
+			(int __user *)parent_tidptr,
+			(int __user *)child_tidptr);
 }
 
 /*
diff --git a/arch/sh/mm/init.c b/arch/sh/mm/init.c
index bf0c263cb6fd..d172065182fb 100644
--- a/arch/sh/mm/init.c
+++ b/arch/sh/mm/init.c
@@ -39,11 +39,6 @@
 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
 pgd_t swapper_pg_dir[PTRS_PER_PGD];
 
-/*
- * Cache of MMU context last used.
- */
-unsigned long mmu_context_cache = NO_CONTEXT;
-
 #ifdef CONFIG_MMU
 /* It'd be good if these lines were in the standard header file. */
 #define START_PFN	(NODE_DATA(0)->bdata->node_boot_start >> PAGE_SHIFT)
diff --git a/arch/sh/mm/tlb-flush.c b/arch/sh/mm/tlb-flush.c
index ef3e4d477864..b829c17c1d17 100644
--- a/arch/sh/mm/tlb-flush.c
+++ b/arch/sh/mm/tlb-flush.c
@@ -16,12 +16,14 @@
 
 void flush_tlb_page(struct vm_area_struct *vma, unsigned long page)
 {
-	if (vma->vm_mm && vma->vm_mm->context.id != NO_CONTEXT) {
+	unsigned int cpu = smp_processor_id();
+
+	if (vma->vm_mm && cpu_context(cpu, vma->vm_mm) != NO_CONTEXT) {
 		unsigned long flags;
 		unsigned long asid;
 		unsigned long saved_asid = MMU_NO_ASID;
 
-		asid = vma->vm_mm->context.id & MMU_CONTEXT_ASID_MASK;
+		asid = cpu_asid(cpu, vma->vm_mm);
 		page &= PAGE_MASK;
 
 		local_irq_save(flags);
@@ -40,22 +42,23 @@ void flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
 		     unsigned long end)
 {
 	struct mm_struct *mm = vma->vm_mm;
+	unsigned int cpu = smp_processor_id();
 
-	if (mm->context.id != NO_CONTEXT) {
+	if (cpu_context(cpu, mm) != NO_CONTEXT) {
 		unsigned long flags;
 		int size;
 
 		local_irq_save(flags);
 		size = (end - start + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
 		if (size > (MMU_NTLB_ENTRIES/4)) { /* Too many TLB to flush */
-			mm->context.id = NO_CONTEXT;
+			cpu_context(cpu, mm) = NO_CONTEXT;
 			if (mm == current->mm)
-				activate_context(mm);
+				activate_context(mm, cpu);
 		} else {
 			unsigned long asid;
 			unsigned long saved_asid = MMU_NO_ASID;
 
-			asid = mm->context.id & MMU_CONTEXT_ASID_MASK;
+			asid = cpu_asid(cpu, mm);
 			start &= PAGE_MASK;
 			end += (PAGE_SIZE - 1);
 			end &= PAGE_MASK;
@@ -76,6 +79,7 @@ void flush_tlb_range(struct vm_area_struct *vma, unsigned long start,
 
 void flush_tlb_kernel_range(unsigned long start, unsigned long end)
 {
+	unsigned int cpu = smp_processor_id();
 	unsigned long flags;
 	int size;
 
@@ -87,7 +91,7 @@ void flush_tlb_kernel_range(unsigned long start, unsigned long end)
 		unsigned long asid;
 		unsigned long saved_asid = get_asid();
 
-		asid = init_mm.context.id & MMU_CONTEXT_ASID_MASK;
+		asid = cpu_asid(cpu, &init_mm);
 		start &= PAGE_MASK;
 		end += (PAGE_SIZE - 1);
 		end &= PAGE_MASK;
@@ -103,15 +107,17 @@ void flush_tlb_kernel_range(unsigned long start, unsigned long end)
 
 void flush_tlb_mm(struct mm_struct *mm)
 {
+	unsigned int cpu = smp_processor_id();
+
 	/* Invalidate all TLB of this process. */
 	/* Instead of invalidating each TLB, we get new MMU context. */
-	if (mm->context.id != NO_CONTEXT) {
+	if (cpu_context(cpu, mm) != NO_CONTEXT) {
 		unsigned long flags;
 
 		local_irq_save(flags);
-		mm->context.id = NO_CONTEXT;
+		cpu_context(cpu, mm) = NO_CONTEXT;
 		if (mm == current->mm)
-			activate_context(mm);
+			activate_context(mm, cpu);
 		local_irq_restore(flags);
 	}
 }
diff --git a/include/asm-sh/mmu.h b/include/asm-sh/mmu.h
index cf47df79bb94..eb0358c097d0 100644
--- a/include/asm-sh/mmu.h
+++ b/include/asm-sh/mmu.h
@@ -1,25 +1,19 @@
 #ifndef __MMU_H
 #define __MMU_H
 
-#if !defined(CONFIG_MMU)
+/* Default "unsigned long" context */
+typedef unsigned long mm_context_id_t[NR_CPUS];
 
 typedef struct {
+#ifdef CONFIG_MMU
+	mm_context_id_t		id;
+	void			*vdso;
+#else
 	struct vm_list_struct	*vmlist;
 	unsigned long		end_brk;
+#endif
 } mm_context_t;
 
-#else
-
-/* Default "unsigned long" context */
-typedef unsigned long mm_context_id_t;
-
-typedef struct {
-	mm_context_id_t id;
-	void *vdso;
-} mm_context_t;
-
-#endif /* CONFIG_MMU */
-
 /*
  * Privileged Space Mapping Buffer (PMB) definitions
  */
diff --git a/include/asm-sh/mmu_context.h b/include/asm-sh/mmu_context.h
index 46f04e23bd45..342024425b7d 100644
--- a/include/asm-sh/mmu_context.h
+++ b/include/asm-sh/mmu_context.h
@@ -1,6 +1,6 @@
 /*
  * Copyright (C) 1999 Niibe Yutaka
- * Copyright (C) 2003 Paul Mundt
+ * Copyright (C) 2003 - 2006 Paul Mundt
  *
  * ASID handling idea taken from MIPS implementation.
  */
@@ -19,11 +19,6 @@
  *    (b) ASID (Address Space IDentifier)
  */
 
-/*
- * Cache of MMU context last used.
- */
-extern unsigned long mmu_context_cache;
-
 #define MMU_CONTEXT_ASID_MASK		0x000000ff
 #define MMU_CONTEXT_VERSION_MASK	0xffffff00
 #define MMU_CONTEXT_FIRST_VERSION	0x00000100
@@ -32,6 +27,11 @@ extern unsigned long mmu_context_cache;
 /* ASID is 8-bit value, so it can't be 0x100 */
 #define MMU_NO_ASID			0x100
 
+#define cpu_context(cpu, mm)	((mm)->context.id[cpu])
+#define cpu_asid(cpu, mm)	(cpu_context((cpu), (mm)) & \
+				 MMU_CONTEXT_ASID_MASK)
+#define asid_cache(cpu)		(cpu_data[cpu].asid_cache)
+
 /*
  * Virtual Page Number mask
  */
@@ -41,18 +41,17 @@ extern unsigned long mmu_context_cache;
 /*
  * Get MMU context if needed.
  */
-static inline void get_mmu_context(struct mm_struct *mm)
+static inline void get_mmu_context(struct mm_struct *mm, unsigned int cpu)
 {
-	unsigned long mc = mmu_context_cache;
+	unsigned long asid = asid_cache(cpu);
 
 	/* Check if we have old version of context. */
-	if (((mm->context.id ^ mc) & MMU_CONTEXT_VERSION_MASK) == 0)
+	if (((cpu_context(cpu, mm) ^ asid) & MMU_CONTEXT_VERSION_MASK) == 0)
 		/* It's up to date, do nothing */
 		return;
 
 	/* It's old, we need to get new context with new version. */
-	mc = ++mmu_context_cache;
-	if (!(mc & MMU_CONTEXT_ASID_MASK)) {
+	if (!(++asid & MMU_CONTEXT_ASID_MASK)) {
 		/*
 		 * We exhaust ASID of this version.
 		 * Flush all TLB and start new cycle.
@@ -63,10 +62,11 @@ static inline void get_mmu_context(struct mm_struct *mm)
 		 * Fix version; Note that we avoid version #0
 		 * to distingush NO_CONTEXT.
 		 */
-		if (!mc)
-			mmu_context_cache = mc = MMU_CONTEXT_FIRST_VERSION;
+		if (!asid)
+			asid = MMU_CONTEXT_FIRST_VERSION;
 	}
-	mm->context.id = mc;
+
+	cpu_context(cpu, mm) = asid_cache(cpu) = asid;
 }
 
 /*
@@ -74,9 +74,13 @@ static inline void get_mmu_context(struct mm_struct *mm)
  * instance.
  */
 static inline int init_new_context(struct task_struct *tsk,
-				       struct mm_struct *mm)
+				   struct mm_struct *mm)
 {
-	mm->context.id = NO_CONTEXT;
+	int i;
+
+	for (i = 0; i < num_online_cpus(); i++)
+		cpu_context(i, mm) = NO_CONTEXT;
+
 	return 0;
 }
 
@@ -117,10 +121,10 @@ static inline unsigned long get_asid(void)
  * 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_context(struct mm_struct *mm)
+static inline void activate_context(struct mm_struct *mm, unsigned int cpu)
 {
-	get_mmu_context(mm);
-	set_asid(mm->context.id & MMU_CONTEXT_ASID_MASK);
+	get_mmu_context(mm, cpu);
+	set_asid(cpu_asid(cpu, mm));
 }
 
 /* MMU_TTB is used for optimizing the fault handling. */
@@ -138,10 +142,15 @@ static inline void switch_mm(struct mm_struct *prev,
 			     struct mm_struct *next,
 			     struct task_struct *tsk)
 {
+	unsigned int cpu = smp_processor_id();
+
 	if (likely(prev != next)) {
+		cpu_set(cpu, next->cpu_vm_mask);
 		set_TTB(next->pgd);
-		activate_context(next);
-	}
+		activate_context(next, cpu);
+	} else
+		if (!cpu_test_and_set(cpu, next->cpu_vm_mask))
+			activate_context(next, cpu);
 }
 
 #define deactivate_mm(tsk,mm)	do { } while (0)
@@ -159,7 +168,7 @@ enter_lazy_tlb(struct mm_struct *mm, struct task_struct *tsk)
 #define destroy_context(mm)		do { } while (0)
 #define set_asid(asid)			do { } while (0)
 #define get_asid()			(0)
-#define activate_context(mm)		do { } while (0)
+#define activate_context(mm,cpu)	do { } while (0)
 #define switch_mm(prev,next,tsk)	do { } while (0)
 #define deactivate_mm(tsk,mm)		do { } while (0)
 #define activate_mm(prev,next)		do { } while (0)
@@ -174,14 +183,16 @@ enter_lazy_tlb(struct mm_struct *mm, struct task_struct *tsk)
  */
 static inline void enable_mmu(void)
 {
+	unsigned int cpu = smp_processor_id();
+
 	/* Enable MMU */
 	ctrl_outl(MMU_CONTROL_INIT, MMUCR);
 	ctrl_barrier();
 
-	if (mmu_context_cache == NO_CONTEXT)
-		mmu_context_cache = MMU_CONTEXT_FIRST_VERSION;
+	if (asid_cache(cpu) == NO_CONTEXT)
+		asid_cache(cpu) = MMU_CONTEXT_FIRST_VERSION;
 
-	set_asid(mmu_context_cache & MMU_CONTEXT_ASID_MASK);
+	set_asid(asid_cache(cpu) & MMU_CONTEXT_ASID_MASK);
 }
 
 static inline void disable_mmu(void)
diff --git a/include/asm-sh/processor.h b/include/asm-sh/processor.h
index e29f2abb92de..da229aae8e0f 100644
--- a/include/asm-sh/processor.h
+++ b/include/asm-sh/processor.h
@@ -66,6 +66,7 @@ enum cpu_type {
 struct sh_cpuinfo {
 	unsigned int type;
 	unsigned long loops_per_jiffy;
+	unsigned long asid_cache;
 
 	struct cache_info icache;	/* Primary I-cache */
 	struct cache_info dcache;	/* Primary D-cache */