arch/powerpc/mm/slb.c

   1 /*
   2  * PowerPC64 SLB support.
   3  *
   4  * Copyright (C) 2004 David Gibson <dwg@au.ibm.com>, IBM
   5  * Based on earlier code writteh by:
   6  * Dave Engebretsen and Mike Corrigan {engebret|mikejc}@us.ibm.com
   7  *    Copyright (c) 2001 Dave Engebretsen
   8  * Copyright (C) 2002 Anton Blanchard <anton@au.ibm.com>, IBM
   9  *
  10  *
  11  *      This program is free software; you can redistribute it and/or
  12  *      modify it under the terms of the GNU General Public License
  13  *      as published by the Free Software Foundation; either version
  14  *      2 of the License, or (at your option) any later version.
  15  */
  16
  17 #undef DEBUG
  18
  19 #include <asm/pgtable.h>
  20 #include <asm/mmu.h>
  21 #include <asm/mmu_context.h>
  22 #include <asm/paca.h>
  23 #include <asm/cputable.h>
  24 #include <asm/cacheflush.h>
  25 #include <asm/smp.h>
  26 #include <asm/firmware.h>
  27 #include <linux/compiler.h>
  28 #include <asm/udbg.h>
  29
  30 #ifdef DEBUG
  31 #define DBG(fmt...) udbg_printf(fmt)
  32 #else
  33 #define DBG(fmt...)
  34 #endif
  35
  36 extern void slb_allocate_realmode(unsigned long ea);
  37 extern void slb_allocate_user(unsigned long ea);
  38
  39 static void slb_allocate(unsigned long ea)
  40 {
  41         /* Currently, we do real mode for all SLBs including user, but
  42          * that will change if we bring back dynamic VSIDs
  43          */
  44         slb_allocate_realmode(ea);
  45 }
  46
  47 static inline unsigned long mk_esid_data(unsigned long ea, int ssize,
  48                                          unsigned long slot)
  49 {
  50         unsigned long mask;
  51
  52         mask = (ssize == MMU_SEGSIZE_256M)? ESID_MASK: ESID_MASK_1T;
  53         return (ea & mask) | SLB_ESID_V | slot;
  54 }
  55
  56 #define slb_vsid_shift(ssize)   \
  57         ((ssize) == MMU_SEGSIZE_256M? SLB_VSID_SHIFT: SLB_VSID_SHIFT_1T)
  58
  59 static inline unsigned long mk_vsid_data(unsigned long ea, int ssize,
  60                                          unsigned long flags)
  61 {
  62         return (get_kernel_vsid(ea, ssize) << slb_vsid_shift(ssize)) | flags |
  63                 ((unsigned long) ssize << SLB_VSID_SSIZE_SHIFT);
  64 }
  65
  66 static inline void slb_shadow_update(unsigned long ea, int ssize,
  67                                      unsigned long flags,
  68                                      unsigned long entry)
  69 {
  70         /*
  71          * Clear the ESID first so the entry is not valid while we are
  72          * updating it.  No write barriers are needed here, provided
  73          * we only update the current CPU's SLB shadow buffer.
  74          */
  75         get_slb_shadow()->save_area[entry].esid = 0;
  76         get_slb_shadow()->save_area[entry].vsid = mk_vsid_data(ea, ssize, flags);
  77         get_slb_shadow()->save_area[entry].esid = mk_esid_data(ea, ssize, entry);
  78 }
  79
  80 static inline void slb_shadow_clear(unsigned long entry)
  81 {
  82         get_slb_shadow()->save_area[entry].esid = 0;
  83 }
  84
  85 static inline void create_shadowed_slbe(unsigned long ea, int ssize,
  86                                         unsigned long flags,
  87                                         unsigned long entry)
  88 {
  89         /*
  90          * Updating the shadow buffer before writing the SLB ensures
  91          * we don't get a stale entry here if we get preempted by PHYP
  92          * between these two statements.
  93          */
  94         slb_shadow_update(ea, ssize, flags, entry);
  95
  96         asm volatile("slbmte  %0,%1" :
  97                      : "r" (mk_vsid_data(ea, ssize, flags)),
  98                        "r" (mk_esid_data(ea, ssize, entry))
  99                      : "memory" );
 100 }
 101
 102 void slb_flush_and_rebolt(void)
 103 {
 104         /* If you change this make sure you change SLB_NUM_BOLTED
 105          * appropriately too. */
 106         unsigned long linear_llp, vmalloc_llp, lflags, vflags;
 107         unsigned long ksp_esid_data, ksp_vsid_data;
 108
 109         WARN_ON(!irqs_disabled());
 110
 111         linear_llp = mmu_psize_defs[mmu_linear_psize].sllp;
 112         vmalloc_llp = mmu_psize_defs[mmu_vmalloc_psize].sllp;
 113         lflags = SLB_VSID_KERNEL | linear_llp;
 114         vflags = SLB_VSID_KERNEL | vmalloc_llp;
 115
 116         ksp_esid_data = mk_esid_data(get_paca()->kstack, mmu_kernel_ssize, 2);
 117         if ((ksp_esid_data & ~0xfffffffUL) <= PAGE_OFFSET) {
 118                 ksp_esid_data &= ~SLB_ESID_V;
 119                 ksp_vsid_data = 0;
 120                 slb_shadow_clear(2);
 121         } else {
 122                 /* Update stack entry; others don't change */
 123                 slb_shadow_update(get_paca()->kstack, mmu_kernel_ssize, lflags, 2);
 124                 ksp_vsid_data = get_slb_shadow()->save_area[2].vsid;
 125         }
 126
 127         /* We need to do this all in asm, so we're sure we don't touch
 128          * the stack between the slbia and rebolting it. */
 129         asm volatile("isync\n"
 130                      "slbia\n"
 131                      /* Slot 1 - first VMALLOC segment */
 132                      "slbmte    %0,%1\n"
 133                      /* Slot 2 - kernel stack */
 134                      "slbmte    %2,%3\n"
 135                      "isync"
 136                      :: "r"(mk_vsid_data(VMALLOC_START, mmu_kernel_ssize, vflags)),
 137                         "r"(mk_esid_data(VMALLOC_START, mmu_kernel_ssize, 1)),
 138                         "r"(ksp_vsid_data),
 139                         "r"(ksp_esid_data)
 140                      : "memory");
 141 }
 142
 143 void slb_vmalloc_update(void)
 144 {
 145         unsigned long vflags;
 146
 147         vflags = SLB_VSID_KERNEL | mmu_psize_defs[mmu_vmalloc_psize].sllp;
 148         slb_shadow_update(VMALLOC_START, mmu_kernel_ssize, vflags, 1);
 149         slb_flush_and_rebolt();
 150 }
 151
 152 /* Helper function to compare esids.  There are four cases to handle.
 153  * 1. The system is not 1T segment size capable.  Use the GET_ESID compare.
 154  * 2. The system is 1T capable, both addresses are < 1T, use the GET_ESID compare.
 155  * 3. The system is 1T capable, only one of the two addresses is > 1T.  This is not a match.
 156  * 4. The system is 1T capable, both addresses are > 1T, use the GET_ESID_1T macro to compare.
 157  */
 158 static inline int esids_match(unsigned long addr1, unsigned long addr2)
 159 {
 160         int esid_1t_count;
 161
 162         /* System is not 1T segment size capable. */
 163         if (!cpu_has_feature(CPU_FTR_1T_SEGMENT))
 164                 return (GET_ESID(addr1) == GET_ESID(addr2));
 165
 166         esid_1t_count = (((addr1 >> SID_SHIFT_1T) != 0) +
 167                                 ((addr2 >> SID_SHIFT_1T) != 0));
 168
 169         /* both addresses are < 1T */
 170         if (esid_1t_count == 0)
 171                 return (GET_ESID(addr1) == GET_ESID(addr2));
 172
 173         /* One address < 1T, the other > 1T.  Not a match */
 174         if (esid_1t_count == 1)
 175                 return 0;
 176
 177         /* Both addresses are > 1T. */
 178         return (GET_ESID_1T(addr1) == GET_ESID_1T(addr2));
 179 }
 180
 181 /* Flush all user entries from the segment table of the current processor. */
 182 void switch_slb(struct task_struct *tsk, struct mm_struct *mm)
 183 {
 184         unsigned long offset = get_paca()->slb_cache_ptr;
 185         unsigned long slbie_data = 0;
 186         unsigned long pc = KSTK_EIP(tsk);
 187         unsigned long stack = KSTK_ESP(tsk);
 188         unsigned long unmapped_base;
 189
 190         if (!cpu_has_feature(CPU_FTR_NO_SLBIE_B) &&
 191             offset <= SLB_CACHE_ENTRIES) {
 192                 int i;
 193                 asm volatile("isync" : : : "memory");
 194                 for (i = 0; i < offset; i++) {
 195                         slbie_data = (unsigned long)get_paca()->slb_cache[i]
 196                                 << SID_SHIFT; /* EA */
 197                         slbie_data |= user_segment_size(slbie_data)
 198                                 << SLBIE_SSIZE_SHIFT;
 199                         slbie_data |= SLBIE_C; /* C set for user addresses */
 200                         asm volatile("slbie %0" : : "r" (slbie_data));
 201                 }
 202                 asm volatile("isync" : : : "memory");
 203         } else {
 204                 slb_flush_and_rebolt();
 205         }
 206
 207         /* Workaround POWER5 < DD2.1 issue */
 208         if (offset == 1 || offset > SLB_CACHE_ENTRIES)
 209                 asm volatile("slbie %0" : : "r" (slbie_data));
 210
 211         get_paca()->slb_cache_ptr = 0;
 212         get_paca()->context = mm->context;
 213
 214         /*
 215          * preload some userspace segments into the SLB.
 216          */
 217         if (test_tsk_thread_flag(tsk, TIF_32BIT))
 218                 unmapped_base = TASK_UNMAPPED_BASE_USER32;
 219         else
 220                 unmapped_base = TASK_UNMAPPED_BASE_USER64;
 221
 222         if (is_kernel_addr(pc))
 223                 return;
 224         slb_allocate(pc);
 225
 226         if (esids_match(pc,stack))
 227                 return;
 228
 229         if (is_kernel_addr(stack))
 230                 return;
 231         slb_allocate(stack);
 232
 233         if (esids_match(pc,unmapped_base) || esids_match(stack,unmapped_base))
 234                 return;
 235
 236         if (is_kernel_addr(unmapped_base))
 237                 return;
 238         slb_allocate(unmapped_base);
 239 }
 240
 241 static inline void patch_slb_encoding(unsigned int *insn_addr,
 242                                       unsigned int immed)
 243 {
 244         /* Assume the instruction had a "0" immediate value, just
 245          * "or" in the new value
 246          */
 247         *insn_addr |= immed;
 248         flush_icache_range((unsigned long)insn_addr, 4+
 249                            (unsigned long)insn_addr);
 250 }
 251
 252 void slb_initialize(void)
 253 {
 254         unsigned long linear_llp, vmalloc_llp, io_llp;
 255         unsigned long lflags, vflags;
 256         static int slb_encoding_inited;
 257         extern unsigned int *slb_miss_kernel_load_linear;
 258         extern unsigned int *slb_miss_kernel_load_io;
 259
 260         /* Prepare our SLB miss handler based on our page size */
 261         linear_llp = mmu_psize_defs[mmu_linear_psize].sllp;
 262         io_llp = mmu_psize_defs[mmu_io_psize].sllp;
 263         vmalloc_llp = mmu_psize_defs[mmu_vmalloc_psize].sllp;
 264         get_paca()->vmalloc_sllp = SLB_VSID_KERNEL | vmalloc_llp;
 265
 266         if (!slb_encoding_inited) {
 267                 slb_encoding_inited = 1;
 268                 patch_slb_encoding(slb_miss_kernel_load_linear,
 269                                    SLB_VSID_KERNEL | linear_llp);
 270                 patch_slb_encoding(slb_miss_kernel_load_io,
 271                                    SLB_VSID_KERNEL | io_llp);
 272
 273                 DBG("SLB: linear  LLP = %04x\n", linear_llp);
 274                 DBG("SLB: io      LLP = %04x\n", io_llp);
 275         }
 276
 277         get_paca()->stab_rr = SLB_NUM_BOLTED;
 278
 279         /* On iSeries the bolted entries have already been set up by
 280          * the hypervisor from the lparMap data in head.S */
 281         if (firmware_has_feature(FW_FEATURE_ISERIES))
 282                 return;
 283
 284         lflags = SLB_VSID_KERNEL | linear_llp;
 285         vflags = SLB_VSID_KERNEL | vmalloc_llp;
 286
 287         /* Invalidate the entire SLB (even slot 0) & all the ERATS */
 288         asm volatile("isync":::"memory");
 289         asm volatile("slbmte  %0,%0"::"r" (0) : "memory");
 290         asm volatile("isync; slbia; isync":::"memory");
 291         create_shadowed_slbe(PAGE_OFFSET, mmu_kernel_ssize, lflags, 0);
 292
 293         create_shadowed_slbe(VMALLOC_START, mmu_kernel_ssize, vflags, 1);
 294
 295         slb_shadow_clear(2);
 296
 297         /* We don't bolt the stack for the time being - we're in boot,
 298          * so the stack is in the bolted segment.  By the time it goes
 299          * elsewhere, we'll call _switch() which will bolt in the new
 300          * one. */
 301         asm volatile("isync":::"memory");
 302 }