arch/ia64/mm/tlb.c

   1 /*
   2  * TLB support routines.
   3  *
   4  * Copyright (C) 1998-2001, 2003 Hewlett-Packard Co
   5  *      David Mosberger-Tang <davidm@hpl.hp.com>
   6  *
   7  * 08/02/00 A. Mallick <asit.k.mallick@intel.com>
   8  *              Modified RID allocation for SMP
   9  *          Goutham Rao <goutham.rao@intel.com>
  10  *              IPI based ptc implementation and A-step IPI implementation.
  11  */
  12 #include <linux/config.h>
  13 #include <linux/module.h>
  14 #include <linux/init.h>
  15 #include <linux/kernel.h>
  16 #include <linux/sched.h>
  17 #include <linux/smp.h>
  18 #include <linux/mm.h>
  19
  20 #include <asm/delay.h>
  21 #include <asm/mmu_context.h>
  22 #include <asm/pgalloc.h>
  23 #include <asm/pal.h>
  24 #include <asm/tlbflush.h>
  25
  26 static struct {
  27         unsigned long mask;     /* mask of supported purge page-sizes */
  28         unsigned long max_bits; /* log2() of largest supported purge page-size */
  29 } purge;
  30
  31 struct ia64_ctx ia64_ctx = {
  32         .lock =         SPIN_LOCK_UNLOCKED,
  33         .next =         1,
  34         .limit =        (1 << 15) - 1,          /* start out with the safe (architected) limit */
  35         .max_ctx =      ~0U
  36 };
  37
  38 DEFINE_PER_CPU(u8, ia64_need_tlb_flush);
  39
  40 /*
  41  * Acquire the ia64_ctx.lock before calling this function!
  42  */
  43 void
  44 wrap_mmu_context (struct mm_struct *mm)
  45 {
  46         unsigned long tsk_context, max_ctx = ia64_ctx.max_ctx;
  47         struct task_struct *tsk;
  48         int i;
  49
  50         if (ia64_ctx.next > max_ctx)
  51                 ia64_ctx.next = 300;    /* skip daemons */
  52         ia64_ctx.limit = max_ctx + 1;
  53
  54         /*
  55          * Scan all the task's mm->context and set proper safe range
  56          */
  57
  58         read_lock(&tasklist_lock);
  59   repeat:
  60         for_each_process(tsk) {
  61                 if (!tsk->mm)
  62                         continue;
  63                 tsk_context = tsk->mm->context;
  64                 if (tsk_context == ia64_ctx.next) {
  65                         if (++ia64_ctx.next >= ia64_ctx.limit) {
  66                                 /* empty range: reset the range limit and start over */
  67                                 if (ia64_ctx.next > max_ctx)
  68                                         ia64_ctx.next = 300;
  69                                 ia64_ctx.limit = max_ctx + 1;
  70                                 goto repeat;
  71                         }
  72                 }
  73                 if ((tsk_context > ia64_ctx.next) && (tsk_context < ia64_ctx.limit))
  74                         ia64_ctx.limit = tsk_context;
  75         }
  76         read_unlock(&tasklist_lock);
  77         /* can't call flush_tlb_all() here because of race condition with O(1) scheduler [EF] */
  78         {
  79                 int cpu = get_cpu(); /* prevent preemption/migration */
  80                 for (i = 0; i < NR_CPUS; ++i)
  81                         if (cpu_online(i) && (i != cpu))
  82                                 per_cpu(ia64_need_tlb_flush, i) = 1;
  83                 put_cpu();
  84         }
  85         local_flush_tlb_all();
  86 }
  87
  88 void
  89 ia64_global_tlb_purge (unsigned long start, unsigned long end, unsigned long nbits)
  90 {
  91         static spinlock_t ptcg_lock = SPIN_LOCK_UNLOCKED;
  92
  93         /* HW requires global serialization of ptc.ga.  */
  94         spin_lock(&ptcg_lock);
  95         {
  96                 do {
  97                         /*
  98                          * Flush ALAT entries also.
  99                          */
 100                         ia64_ptcga(start, (nbits<<2));
 101                         ia64_srlz_i();
 102                         start += (1UL << nbits);
 103                 } while (start < end);
 104         }
 105         spin_unlock(&ptcg_lock);
 106 }
 107
 108 void
 109 local_flush_tlb_all (void)
 110 {
 111         unsigned long i, j, flags, count0, count1, stride0, stride1, addr;
 112
 113         addr    = local_cpu_data->ptce_base;
 114         count0  = local_cpu_data->ptce_count[0];
 115         count1  = local_cpu_data->ptce_count[1];
 116         stride0 = local_cpu_data->ptce_stride[0];
 117         stride1 = local_cpu_data->ptce_stride[1];
 118
 119         local_irq_save(flags);
 120         for (i = 0; i < count0; ++i) {
 121                 for (j = 0; j < count1; ++j) {
 122                         ia64_ptce(addr);
 123                         addr += stride1;
 124                 }
 125                 addr += stride0;
 126         }
 127         local_irq_restore(flags);
 128         ia64_srlz_i();                  /* srlz.i implies srlz.d */
 129 }
 130 EXPORT_SYMBOL(local_flush_tlb_all);
 131
 132 void
 133 flush_tlb_range (struct vm_area_struct *vma, unsigned long start, unsigned long end)
 134 {
 135         struct mm_struct *mm = vma->vm_mm;
 136         unsigned long size = end - start;
 137         unsigned long nbits;
 138
 139         if (mm != current->active_mm) {
 140                 /* this does happen, but perhaps it's not worth optimizing for? */
 141 #ifdef CONFIG_SMP
 142                 flush_tlb_all();
 143 #else
 144                 mm->context = 0;
 145 #endif
 146                 return;
 147         }
 148
 149         nbits = ia64_fls(size + 0xfff);
 150         while (unlikely (((1UL << nbits) & purge.mask) == 0) && (nbits < purge.max_bits))
 151                 ++nbits;
 152         if (nbits > purge.max_bits)
 153                 nbits = purge.max_bits;
 154         start &= ~((1UL << nbits) - 1);
 155
 156 # ifdef CONFIG_SMP
 157         platform_global_tlb_purge(start, end, nbits);
 158 # else
 159         do {
 160                 ia64_ptcl(start, (nbits<<2));
 161                 start += (1UL << nbits);
 162         } while (start < end);
 163 # endif
 164
 165         ia64_srlz_i();                  /* srlz.i implies srlz.d */
 166 }
 167 EXPORT_SYMBOL(flush_tlb_range);
 168
 169 void __devinit
 170 ia64_tlb_init (void)
 171 {
 172         ia64_ptce_info_t ptce_info;
 173         unsigned long tr_pgbits;
 174         long status;
 175
 176         if ((status = ia64_pal_vm_page_size(&tr_pgbits, &purge.mask)) != 0) {
 177                 printk(KERN_ERR "PAL_VM_PAGE_SIZE failed with status=%ld;"
 178                        "defaulting to architected purge page-sizes.\n", status);
 179                 purge.mask = 0x115557000UL;
 180         }
 181         purge.max_bits = ia64_fls(purge.mask);
 182
 183         ia64_get_ptce(&ptce_info);
 184         local_cpu_data->ptce_base = ptce_info.base;
 185         local_cpu_data->ptce_count[0] = ptce_info.count[0];
 186         local_cpu_data->ptce_count[1] = ptce_info.count[1];
 187         local_cpu_data->ptce_stride[0] = ptce_info.stride[0];
 188         local_cpu_data->ptce_stride[1] = ptce_info.stride[1];
 189
 190         local_flush_tlb_all();          /* nuke left overs from bootstrapping... */
 191 }