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...) printk(fmt)
  32 #else
  33 #define DBG pr_debug
  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 #define slb_esid_mask(ssize)    \
  48         (((ssize) == MMU_SEGSIZE_256M)? ESID_MASK: ESID_MASK_1T)
  49
  50 static inline unsigned long mk_esid_data(unsigned long ea, int ssize,
  51                                          unsigned long slot)
  52 {
  53         return (ea & slb_esid_mask(ssize)) | 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         /*
 128          * We can't take a PMU exception in the following code, so hard
 129          * disable interrupts.
 130          */
 131         hard_irq_disable();
 132
 133         /* We need to do this all in asm, so we're sure we don't touch
 134          * the stack between the slbia and rebolting it. */
 135         asm volatile("isync\n"
 136                      "slbia\n"
 137                      /* Slot 1 - first VMALLOC segment */
 138                      "slbmte    %0,%1\n"
 139                      /* Slot 2 - kernel stack */
 140                      "slbmte    %2,%3\n"
 141                      "isync"
 142                      :: "r"(mk_vsid_data(VMALLOC_START, mmu_kernel_ssize, vflags)),
 143                         "r"(mk_esid_data(VMALLOC_START, mmu_kernel_ssize, 1)),
 144                         "r"(ksp_vsid_data),
 145                         "r"(ksp_esid_data)
 146                      : "memory");
 147 }
 148
 149 void slb_vmalloc_update(void)
 150 {
 151         unsigned long vflags;
 152
 153         vflags = SLB_VSID_KERNEL | mmu_psize_defs[mmu_vmalloc_psize].sllp;
 154         slb_shadow_update(VMALLOC_START, mmu_kernel_ssize, vflags, 1);
 155         slb_flush_and_rebolt();
 156 }
 157
 158 /* Helper function to compare esids.  There are four cases to handle.
 159  * 1. The system is not 1T segment size capable.  Use the GET_ESID compare.
 160  * 2. The system is 1T capable, both addresses are < 1T, use the GET_ESID compare.
 161  * 3. The system is 1T capable, only one of the two addresses is > 1T.  This is not a match.
 162  * 4. The system is 1T capable, both addresses are > 1T, use the GET_ESID_1T macro to compare.
 163  */
 164 static inline int esids_match(unsigned long addr1, unsigned long addr2)
 165 {
 166         int esid_1t_count;
 167
 168         /* System is not 1T segment size capable. */
 169         if (!cpu_has_feature(CPU_FTR_1T_SEGMENT))
 170                 return (GET_ESID(addr1) == GET_ESID(addr2));
 171
 172         esid_1t_count = (((addr1 >> SID_SHIFT_1T) != 0) +
 173                                 ((addr2 >> SID_SHIFT_1T) != 0));
 174
 175         /* both addresses are < 1T */
 176         if (esid_1t_count == 0)
 177                 return (GET_ESID(addr1) == GET_ESID(addr2));
 178
 179         /* One address < 1T, the other > 1T.  Not a match */
 180         if (esid_1t_count == 1)
 181                 return 0;
 182
 183         /* Both addresses are > 1T. */
 184         return (GET_ESID_1T(addr1) == GET_ESID_1T(addr2));
 185 }
 186
 187 /* Flush all user entries from the segment table of the current processor. */
 188 void switch_slb(struct task_struct *tsk, struct mm_struct *mm)
 189 {
 190         unsigned long offset = get_paca()->slb_cache_ptr;
 191         unsigned long slbie_data = 0;
 192         unsigned long pc = KSTK_EIP(tsk);
 193         unsigned long stack = KSTK_ESP(tsk);
 194         unsigned long unmapped_base;
 195
 196         if (!cpu_has_feature(CPU_FTR_NO_SLBIE_B) &&
 197             offset <= SLB_CACHE_ENTRIES) {
 198                 int i;
 199                 asm volatile("isync" : : : "memory");
 200                 for (i = 0; i < offset; i++) {
 201                         slbie_data = (unsigned long)get_paca()->slb_cache[i]
 202                                 << SID_SHIFT; /* EA */
 203                         slbie_data |= user_segment_size(slbie_data)
 204                                 << SLBIE_SSIZE_SHIFT;
 205                         slbie_data |= SLBIE_C; /* C set for user addresses */
 206                         asm volatile("slbie %0" : : "r" (slbie_data));
 207                 }
 208                 asm volatile("isync" : : : "memory");
 209         } else {
 210                 slb_flush_and_rebolt();
 211         }
 212
 213         /* Workaround POWER5 < DD2.1 issue */
 214         if (offset == 1 || offset > SLB_CACHE_ENTRIES)
 215                 asm volatile("slbie %0" : : "r" (slbie_data));
 216
 217         get_paca()->slb_cache_ptr = 0;
 218         get_paca()->context = mm->context;
 219
 220         /*
 221          * preload some userspace segments into the SLB.
 222          */
 223         if (test_tsk_thread_flag(tsk, TIF_32BIT))
 224                 unmapped_base = TASK_UNMAPPED_BASE_USER32;
 225         else
 226                 unmapped_base = TASK_UNMAPPED_BASE_USER64;
 227
 228         if (is_kernel_addr(pc))
 229                 return;
 230         slb_allocate(pc);
 231
 232         if (esids_match(pc,stack))
 233                 return;
 234
 235         if (is_kernel_addr(stack))
 236                 return;
 237         slb_allocate(stack);
 238
 239         if (esids_match(pc,unmapped_base) || esids_match(stack,unmapped_base))
 240                 return;
 241
 242         if (is_kernel_addr(unmapped_base))
 243                 return;
 244         slb_allocate(unmapped_base);
 245 }
 246
 247 static inline void patch_slb_encoding(unsigned int *insn_addr,
 248                                       unsigned int immed)
 249 {
 250         /* Assume the instruction had a "0" immediate value, just
 251          * "or" in the new value
 252          */
 253         *insn_addr |= immed;
 254         flush_icache_range((unsigned long)insn_addr, 4+
 255                            (unsigned long)insn_addr);
 256 }
 257
 258 void slb_initialize(void)
 259 {
 260         unsigned long linear_llp, vmalloc_llp, io_llp;
 261         unsigned long lflags, vflags;
 262         static int slb_encoding_inited;
 263         extern unsigned int *slb_miss_kernel_load_linear;
 264         extern unsigned int *slb_miss_kernel_load_io;
 265         extern unsigned int *slb_compare_rr_to_size;
 266 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 267         extern unsigned int *slb_miss_kernel_load_vmemmap;
 268         unsigned long vmemmap_llp;
 269 #endif
 270
 271         /* Prepare our SLB miss handler based on our page size */
 272         linear_llp = mmu_psize_defs[mmu_linear_psize].sllp;
 273         io_llp = mmu_psize_defs[mmu_io_psize].sllp;
 274         vmalloc_llp = mmu_psize_defs[mmu_vmalloc_psize].sllp;
 275         get_paca()->vmalloc_sllp = SLB_VSID_KERNEL | vmalloc_llp;
 276 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 277         vmemmap_llp = mmu_psize_defs[mmu_vmemmap_psize].sllp;
 278 #endif
 279         if (!slb_encoding_inited) {
 280                 slb_encoding_inited = 1;
 281                 patch_slb_encoding(slb_miss_kernel_load_linear,
 282                                    SLB_VSID_KERNEL | linear_llp);
 283                 patch_slb_encoding(slb_miss_kernel_load_io,
 284                                    SLB_VSID_KERNEL | io_llp);
 285                 patch_slb_encoding(slb_compare_rr_to_size,
 286                                    mmu_slb_size);
 287
 288                 DBG("SLB: linear  LLP = %04lx\n", linear_llp);
 289                 DBG("SLB: io      LLP = %04lx\n", io_llp);
 290
 291 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 292                 patch_slb_encoding(slb_miss_kernel_load_vmemmap,
 293                                    SLB_VSID_KERNEL | vmemmap_llp);
 294                 DBG("SLB: vmemmap LLP = %04lx\n", vmemmap_llp);
 295 #endif
 296         }
 297
 298         get_paca()->stab_rr = SLB_NUM_BOLTED;
 299
 300         /* On iSeries the bolted entries have already been set up by
 301          * the hypervisor from the lparMap data in head.S */
 302         if (firmware_has_feature(FW_FEATURE_ISERIES))
 303                 return;
 304
 305         lflags = SLB_VSID_KERNEL | linear_llp;
 306         vflags = SLB_VSID_KERNEL | vmalloc_llp;
 307
 308         /* Invalidate the entire SLB (even slot 0) & all the ERATS */
 309         asm volatile("isync":::"memory");
 310         asm volatile("slbmte  %0,%0"::"r" (0) : "memory");
 311         asm volatile("isync; slbia; isync":::"memory");
 312         create_shadowed_slbe(PAGE_OFFSET, mmu_kernel_ssize, lflags, 0);
 313
 314         create_shadowed_slbe(VMALLOC_START, mmu_kernel_ssize, vflags, 1);
 315
 316         /* For the boot cpu, we're running on the stack in init_thread_union,
 317          * which is in the first segment of the linear mapping, and also
 318          * get_paca()->kstack hasn't been initialized yet.
 319          * For secondary cpus, we need to bolt the kernel stack entry now.
 320          */
 321         slb_shadow_clear(2);
 322         if (raw_smp_processor_id() != boot_cpuid &&
 323             (get_paca()->kstack & slb_esid_mask(mmu_kernel_ssize)) > PAGE_OFFSET)
 324                 create_shadowed_slbe(get_paca()->kstack,
 325                                      mmu_kernel_ssize, lflags, 2);
 326
 327         asm volatile("isync":::"memory");
 328 }