From aec5e0e1c179fac4bbca4007a3f0d3107275a73c Mon Sep 17 00:00:00 2001 From: Paul Mundt Date: Mon, 25 Dec 2006 09:51:47 +0900 Subject: 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 --- arch/sh/kernel/cpu/init.c | 11 ++++++-- arch/sh/kernel/process.c | 66 ++++++++++++++++++++--------------------------- arch/sh/mm/init.c | 5 ---- arch/sh/mm/tlb-flush.c | 26 ++++++++++++------- 4 files changed, 53 insertions(+), 55 deletions(-) (limited to 'arch') diff --git a/arch/sh/kernel/cpu/init.c b/arch/sh/kernel/cpu/init.c index 48121766e8d..6c3c7687e81 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 #include +#include +#include #include #include #include @@ -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 cc8f306fd68..0298f0faa6e 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 -#include #include #include -#include -#include #include -#include #include #include - -#include #include #include -#include #include -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(®s, 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, ®s, 0, NULL, NULL); + return do_fork(flags | CLONE_VM | CLONE_UNTRACED, 0, + ®s, 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 bf0c263cb6f..d172065182f 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 ef3e4d47786..b829c17c1d1 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); } } -- cgit v1.2.3