From dcc17d1baef3721d1574e5b2f4f2d4607514bcff Mon Sep 17 00:00:00 2001 From: Peter Keilty Date: Mon, 31 Oct 2005 16:44:47 -0500 Subject: [IA64] Use bitmaps for efficient context allocation/free Corrects the very inefficent method of finding free context_ids in get_mmu_context(). Instead of walking the task_list of all processes, 2 bitmaps are used to efficently store and lookup state, inuse and needs flushing. The entire rid address space is now used before calling wrap_mmu_context and global tlb flushing. Special thanks to Ken and Rohit for their review and modifications in using a bit flushmap. Signed-off-by: Peter Keilty Signed-off-by: Tony Luck --- arch/ia64/kernel/setup.c | 1 + arch/ia64/mm/tlb.c | 56 ++++++++++++++++++++---------------------- include/asm-ia64/mmu_context.h | 19 +++++++++++--- include/asm-ia64/tlbflush.h | 1 + 4 files changed, 44 insertions(+), 33 deletions(-) diff --git a/arch/ia64/kernel/setup.c b/arch/ia64/kernel/setup.c index fc56ca2..c9388a9 100644 --- a/arch/ia64/kernel/setup.c +++ b/arch/ia64/kernel/setup.c @@ -454,6 +454,7 @@ setup_arch (char **cmdline_p) #endif cpu_init(); /* initialize the bootstrap CPU */ + mmu_context_init(); /* initialize context_id bitmap */ #ifdef CONFIG_ACPI acpi_boot_init(); diff --git a/arch/ia64/mm/tlb.c b/arch/ia64/mm/tlb.c index c79a9b9..39628fc 100644 --- a/arch/ia64/mm/tlb.c +++ b/arch/ia64/mm/tlb.c @@ -8,6 +8,8 @@ * Modified RID allocation for SMP * Goutham Rao * IPI based ptc implementation and A-step IPI implementation. + * Rohit Seth + * Ken Chen */ #include #include @@ -16,12 +18,14 @@ #include #include #include +#include #include #include #include #include #include +#include static struct { unsigned long mask; /* mask of supported purge page-sizes */ @@ -31,49 +35,43 @@ static struct { struct ia64_ctx ia64_ctx = { .lock = SPIN_LOCK_UNLOCKED, .next = 1, - .limit = (1 << 15) - 1, /* start out with the safe (architected) limit */ .max_ctx = ~0U }; DEFINE_PER_CPU(u8, ia64_need_tlb_flush); /* + * Initializes the ia64_ctx.bitmap array based on max_ctx+1. + * Called after cpu_init() has setup ia64_ctx.max_ctx based on + * maximum RID that is supported by boot CPU. + */ +void __init +mmu_context_init (void) +{ + ia64_ctx.bitmap = alloc_bootmem((ia64_ctx.max_ctx+1)>>3); + ia64_ctx.flushmap = alloc_bootmem((ia64_ctx.max_ctx+1)>>3); +} + +/* * Acquire the ia64_ctx.lock before calling this function! */ void wrap_mmu_context (struct mm_struct *mm) { - unsigned long tsk_context, max_ctx = ia64_ctx.max_ctx; - struct task_struct *tsk; int i; + unsigned long flush_bit; - if (ia64_ctx.next > max_ctx) - ia64_ctx.next = 300; /* skip daemons */ - ia64_ctx.limit = max_ctx + 1; - - /* - * Scan all the task's mm->context and set proper safe range - */ - - read_lock(&tasklist_lock); - repeat: - for_each_process(tsk) { - if (!tsk->mm) - continue; - tsk_context = tsk->mm->context; - if (tsk_context == ia64_ctx.next) { - if (++ia64_ctx.next >= ia64_ctx.limit) { - /* empty range: reset the range limit and start over */ - if (ia64_ctx.next > max_ctx) - ia64_ctx.next = 300; - ia64_ctx.limit = max_ctx + 1; - goto repeat; - } - } - if ((tsk_context > ia64_ctx.next) && (tsk_context < ia64_ctx.limit)) - ia64_ctx.limit = tsk_context; + for (i=0; i <= ia64_ctx.max_ctx / BITS_PER_LONG; i++) { + flush_bit = xchg(&ia64_ctx.flushmap[i], 0); + ia64_ctx.bitmap[i] ^= flush_bit; } - read_unlock(&tasklist_lock); + + /* use offset at 300 to skip daemons */ + ia64_ctx.next = find_next_zero_bit(ia64_ctx.bitmap, + ia64_ctx.max_ctx, 300); + ia64_ctx.limit = find_next_bit(ia64_ctx.bitmap, + ia64_ctx.max_ctx, ia64_ctx.next); + /* can't call flush_tlb_all() here because of race condition with O(1) scheduler [EF] */ { int cpu = get_cpu(); /* prevent preemption/migration */ diff --git a/include/asm-ia64/mmu_context.h b/include/asm-ia64/mmu_context.h index 8d6e72f..8d9b30b 100644 --- a/include/asm-ia64/mmu_context.h +++ b/include/asm-ia64/mmu_context.h @@ -32,13 +32,17 @@ struct ia64_ctx { spinlock_t lock; unsigned int next; /* next context number to use */ - unsigned int limit; /* next >= limit => must call wrap_mmu_context() */ - unsigned int max_ctx; /* max. context value supported by all CPUs */ + unsigned int limit; /* available free range */ + unsigned int max_ctx; /* max. context value supported by all CPUs */ + /* call wrap_mmu_context when next >= max */ + unsigned long *bitmap; /* bitmap size is max_ctx+1 */ + unsigned long *flushmap;/* pending rid to be flushed */ }; extern struct ia64_ctx ia64_ctx; DECLARE_PER_CPU(u8, ia64_need_tlb_flush); +extern void mmu_context_init (void); extern void wrap_mmu_context (struct mm_struct *mm); static inline void @@ -83,9 +87,16 @@ get_mmu_context (struct mm_struct *mm) context = mm->context; if (context == 0) { cpus_clear(mm->cpu_vm_mask); - if (ia64_ctx.next >= ia64_ctx.limit) - wrap_mmu_context(mm); + if (ia64_ctx.next >= ia64_ctx.limit) { + ia64_ctx.next = find_next_zero_bit(ia64_ctx.bitmap, + ia64_ctx.max_ctx, ia64_ctx.next); + ia64_ctx.limit = find_next_bit(ia64_ctx.bitmap, + ia64_ctx.max_ctx, ia64_ctx.next); + if (ia64_ctx.next >= ia64_ctx.max_ctx) + wrap_mmu_context(mm); + } mm->context = context = ia64_ctx.next++; + __set_bit(context, ia64_ctx.bitmap); } } spin_unlock_irqrestore(&ia64_ctx.lock, flags); diff --git a/include/asm-ia64/tlbflush.h b/include/asm-ia64/tlbflush.h index b65c627..a35b323 100644 --- a/include/asm-ia64/tlbflush.h +++ b/include/asm-ia64/tlbflush.h @@ -51,6 +51,7 @@ flush_tlb_mm (struct mm_struct *mm) if (!mm) return; + set_bit(mm->context, ia64_ctx.flushmap); mm->context = 0; if (atomic_read(&mm->mm_users) == 0) -- cgit v1.1 From 58cd90829918dabbd81a453de676d41fb7b628ad Mon Sep 17 00:00:00 2001 From: "Chen, Kenneth W" Date: Sat, 29 Oct 2005 18:47:04 -0700 Subject: [IA64] make mmu_context.h and tlb.c 80-column friendly wrap_mmu_context(), delayed_tlb_flush(), get_mmu_context() all have an extra { } block which cause one extra indentation. get_mmu_context() is particularly bad with 5 indentations to the most inner "if". It finally gets on my nerve that I can't keep the code within 80 columns. Remove the extra { } block and while I'm at it, reformat all the comments to 80-column friendly. No functional change at all with this patch. Signed-off-by: Ken Chen Signed-off-by: Tony Luck --- arch/ia64/mm/tlb.c | 33 +++++++++-------- include/asm-ia64/mmu_context.h | 80 ++++++++++++++++++++++-------------------- 2 files changed, 59 insertions(+), 54 deletions(-) diff --git a/arch/ia64/mm/tlb.c b/arch/ia64/mm/tlb.c index 39628fc..41105d4 100644 --- a/arch/ia64/mm/tlb.c +++ b/arch/ia64/mm/tlb.c @@ -29,7 +29,7 @@ static struct { unsigned long mask; /* mask of supported purge page-sizes */ - unsigned long max_bits; /* log2() of largest supported purge page-size */ + unsigned long max_bits; /* log2 of largest supported purge page-size */ } purge; struct ia64_ctx ia64_ctx = { @@ -58,7 +58,7 @@ mmu_context_init (void) void wrap_mmu_context (struct mm_struct *mm) { - int i; + int i, cpu; unsigned long flush_bit; for (i=0; i <= ia64_ctx.max_ctx / BITS_PER_LONG; i++) { @@ -72,20 +72,21 @@ wrap_mmu_context (struct mm_struct *mm) ia64_ctx.limit = find_next_bit(ia64_ctx.bitmap, ia64_ctx.max_ctx, ia64_ctx.next); - /* can't call flush_tlb_all() here because of race condition with O(1) scheduler [EF] */ - { - int cpu = get_cpu(); /* prevent preemption/migration */ - for_each_online_cpu(i) { - if (i != cpu) - per_cpu(ia64_need_tlb_flush, i) = 1; - } - put_cpu(); - } + /* + * can't call flush_tlb_all() here because of race condition + * with O(1) scheduler [EF] + */ + cpu = get_cpu(); /* prevent preemption/migration */ + for_each_online_cpu(i) + if (i != cpu) + per_cpu(ia64_need_tlb_flush, i) = 1; + put_cpu(); local_flush_tlb_all(); } void -ia64_global_tlb_purge (struct mm_struct *mm, unsigned long start, unsigned long end, unsigned long nbits) +ia64_global_tlb_purge (struct mm_struct *mm, unsigned long start, + unsigned long end, unsigned long nbits) { static DEFINE_SPINLOCK(ptcg_lock); @@ -133,7 +134,8 @@ local_flush_tlb_all (void) } void -flush_tlb_range (struct vm_area_struct *vma, unsigned long start, unsigned long end) +flush_tlb_range (struct vm_area_struct *vma, unsigned long start, + unsigned long end) { struct mm_struct *mm = vma->vm_mm; unsigned long size = end - start; @@ -147,7 +149,8 @@ flush_tlb_range (struct vm_area_struct *vma, unsigned long start, unsigned long #endif nbits = ia64_fls(size + 0xfff); - while (unlikely (((1UL << nbits) & purge.mask) == 0) && (nbits < purge.max_bits)) + while (unlikely (((1UL << nbits) & purge.mask) == 0) && + (nbits < purge.max_bits)) ++nbits; if (nbits > purge.max_bits) nbits = purge.max_bits; @@ -189,5 +192,5 @@ ia64_tlb_init (void) local_cpu_data->ptce_stride[0] = ptce_info.stride[0]; local_cpu_data->ptce_stride[1] = ptce_info.stride[1]; - local_flush_tlb_all(); /* nuke left overs from bootstrapping... */ + local_flush_tlb_all(); /* nuke left overs from bootstrapping... */ } diff --git a/include/asm-ia64/mmu_context.h b/include/asm-ia64/mmu_context.h index 8d9b30b..b5c6508 100644 --- a/include/asm-ia64/mmu_context.h +++ b/include/asm-ia64/mmu_context.h @@ -7,12 +7,13 @@ */ /* - * Routines to manage the allocation of task context numbers. Task context numbers are - * used to reduce or eliminate the need to perform TLB flushes due to context switches. - * Context numbers are implemented using ia-64 region ids. Since the IA-64 TLB does not - * consider the region number when performing a TLB lookup, we need to assign a unique - * region id to each region in a process. We use the least significant three bits in a - * region id for this purpose. + * Routines to manage the allocation of task context numbers. Task context + * numbers are used to reduce or eliminate the need to perform TLB flushes + * due to context switches. Context numbers are implemented using ia-64 + * region ids. Since the IA-64 TLB does not consider the region number when + * performing a TLB lookup, we need to assign a unique region id to each + * region in a process. We use the least significant three bits in aregion + * id for this purpose. */ #define IA64_REGION_ID_KERNEL 0 /* the kernel's region id (tlb.c depends on this being 0) */ @@ -51,10 +52,10 @@ enter_lazy_tlb (struct mm_struct *mm, struct task_struct *tsk) } /* - * When the context counter wraps around all TLBs need to be flushed because an old - * context number might have been reused. This is signalled by the ia64_need_tlb_flush - * per-CPU variable, which is checked in the routine below. Called by activate_mm(). - * + * When the context counter wraps around all TLBs need to be flushed because + * an old context number might have been reused. This is signalled by the + * ia64_need_tlb_flush per-CPU variable, which is checked in the routine + * below. Called by activate_mm(). */ static inline void delayed_tlb_flush (void) @@ -64,11 +65,9 @@ delayed_tlb_flush (void) if (unlikely(__ia64_per_cpu_var(ia64_need_tlb_flush))) { spin_lock_irqsave(&ia64_ctx.lock, flags); - { - if (__ia64_per_cpu_var(ia64_need_tlb_flush)) { - local_flush_tlb_all(); - __ia64_per_cpu_var(ia64_need_tlb_flush) = 0; - } + if (__ia64_per_cpu_var(ia64_need_tlb_flush)) { + local_flush_tlb_all(); + __ia64_per_cpu_var(ia64_need_tlb_flush) = 0; } spin_unlock_irqrestore(&ia64_ctx.lock, flags); } @@ -80,27 +79,27 @@ get_mmu_context (struct mm_struct *mm) unsigned long flags; nv_mm_context_t context = mm->context; - if (unlikely(!context)) { - spin_lock_irqsave(&ia64_ctx.lock, flags); - { - /* re-check, now that we've got the lock: */ - context = mm->context; - if (context == 0) { - cpus_clear(mm->cpu_vm_mask); - if (ia64_ctx.next >= ia64_ctx.limit) { - ia64_ctx.next = find_next_zero_bit(ia64_ctx.bitmap, - ia64_ctx.max_ctx, ia64_ctx.next); - ia64_ctx.limit = find_next_bit(ia64_ctx.bitmap, - ia64_ctx.max_ctx, ia64_ctx.next); - if (ia64_ctx.next >= ia64_ctx.max_ctx) - wrap_mmu_context(mm); - } - mm->context = context = ia64_ctx.next++; - __set_bit(context, ia64_ctx.bitmap); - } + if (likely(context)) + goto out; + + spin_lock_irqsave(&ia64_ctx.lock, flags); + /* re-check, now that we've got the lock: */ + context = mm->context; + if (context == 0) { + cpus_clear(mm->cpu_vm_mask); + if (ia64_ctx.next >= ia64_ctx.limit) { + ia64_ctx.next = find_next_zero_bit(ia64_ctx.bitmap, + ia64_ctx.max_ctx, ia64_ctx.next); + ia64_ctx.limit = find_next_bit(ia64_ctx.bitmap, + ia64_ctx.max_ctx, ia64_ctx.next); + if (ia64_ctx.next >= ia64_ctx.max_ctx) + wrap_mmu_context(mm); } - spin_unlock_irqrestore(&ia64_ctx.lock, flags); + mm->context = context = ia64_ctx.next++; + __set_bit(context, ia64_ctx.bitmap); } + spin_unlock_irqrestore(&ia64_ctx.lock, flags); +out: /* * Ensure we're not starting to use "context" before any old * uses of it are gone from our TLB. @@ -111,8 +110,8 @@ get_mmu_context (struct mm_struct *mm) } /* - * Initialize context number to some sane value. MM is guaranteed to be a brand-new - * address-space, so no TLB flushing is needed, ever. + * Initialize context number to some sane value. MM is guaranteed to be a + * brand-new address-space, so no TLB flushing is needed, ever. */ static inline int init_new_context (struct task_struct *p, struct mm_struct *mm) @@ -173,7 +172,10 @@ activate_context (struct mm_struct *mm) if (!cpu_isset(smp_processor_id(), mm->cpu_vm_mask)) cpu_set(smp_processor_id(), mm->cpu_vm_mask); reload_context(context); - /* in the unlikely event of a TLB-flush by another thread, redo the load: */ + /* + * in the unlikely event of a TLB-flush by another thread, + * redo the load. + */ } while (unlikely(context != mm->context)); } @@ -186,8 +188,8 @@ static inline void activate_mm (struct mm_struct *prev, struct mm_struct *next) { /* - * We may get interrupts here, but that's OK because interrupt handlers cannot - * touch user-space. + * We may get interrupts here, but that's OK because interrupt + * handlers cannot touch user-space. */ ia64_set_kr(IA64_KR_PT_BASE, __pa(next->pgd)); activate_context(next); -- cgit v1.1