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 written 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 #include <asm/pgtable.h>
  18 #include <asm/mmu.h>
  19 #include <asm/mmu_context.h>
  20 #include <asm/paca.h>
  21 #include <asm/cputable.h>
  22 #include <asm/cacheflush.h>
  23 #include <asm/smp.h>
  24 #include <asm/firmware.h>
  25 #include <linux/compiler.h>
  26 #include <asm/udbg.h>
  27
  28
  29 extern void slb_allocate_realmode(unsigned long ea);
  30 extern void slb_allocate_user(unsigned long ea);
  31
  32 static void slb_allocate(unsigned long ea)
  33 {
  34         /* Currently, we do real mode for all SLBs including user, but
  35          * that will change if we bring back dynamic VSIDs
  36          */
  37         slb_allocate_realmode(ea);
  38 }
  39
  40 #define slb_esid_mask(ssize)    \
  41         (((ssize) == MMU_SEGSIZE_256M)? ESID_MASK: ESID_MASK_1T)
  42
  43 static inline unsigned long mk_esid_data(unsigned long ea, int ssize,
  44                                          unsigned long slot)
  45 {
  46         return (ea & slb_esid_mask(ssize)) | SLB_ESID_V | slot;
  47 }
  48
  49 #define slb_vsid_shift(ssize)   \
  50         ((ssize) == MMU_SEGSIZE_256M? SLB_VSID_SHIFT: SLB_VSID_SHIFT_1T)
  51
  52 static inline unsigned long mk_vsid_data(unsigned long ea, int ssize,
  53                                          unsigned long flags)
  54 {
  55         return (get_kernel_vsid(ea, ssize) << slb_vsid_shift(ssize)) | flags |
  56                 ((unsigned long) ssize << SLB_VSID_SSIZE_SHIFT);
  57 }
  58
  59 static inline void slb_shadow_update(unsigned long ea, int ssize,
  60                                      unsigned long flags,
  61                                      unsigned long entry)
  62 {
  63         /*
  64          * Clear the ESID first so the entry is not valid while we are
  65          * updating it.  No write barriers are needed here, provided
  66          * we only update the current CPU's SLB shadow buffer.
  67          */
  68         get_slb_shadow()->save_area[entry].esid = 0;
  69         get_slb_shadow()->save_area[entry].vsid = mk_vsid_data(ea, ssize, flags);
  70         get_slb_shadow()->save_area[entry].esid = mk_esid_data(ea, ssize, entry);
  71 }
  72
  73 static inline void slb_shadow_clear(unsigned long entry)
  74 {
  75         get_slb_shadow()->save_area[entry].esid = 0;
  76 }
  77
  78 static inline void create_shadowed_slbe(unsigned long ea, int ssize,
  79                                         unsigned long flags,
  80                                         unsigned long entry)
  81 {
  82         /*
  83          * Updating the shadow buffer before writing the SLB ensures
  84          * we don't get a stale entry here if we get preempted by PHYP
  85          * between these two statements.
  86          */
  87         slb_shadow_update(ea, ssize, flags, entry);
  88
  89         asm volatile("slbmte  %0,%1" :
  90                      : "r" (mk_vsid_data(ea, ssize, flags)),
  91                        "r" (mk_esid_data(ea, ssize, entry))
  92                      : "memory" );
  93 }
  94
  95 void slb_flush_and_rebolt(void)
  96 {
  97         /* If you change this make sure you change SLB_NUM_BOLTED
  98          * appropriately too. */
  99         unsigned long linear_llp, vmalloc_llp, lflags, vflags;
 100         unsigned long ksp_esid_data, ksp_vsid_data;
 101
 102         WARN_ON(!irqs_disabled());
 103
 104         linear_llp = mmu_psize_defs[mmu_linear_psize].sllp;
 105         vmalloc_llp = mmu_psize_defs[mmu_vmalloc_psize].sllp;
 106         lflags = SLB_VSID_KERNEL | linear_llp;
 107         vflags = SLB_VSID_KERNEL | vmalloc_llp;
 108
 109         ksp_esid_data = mk_esid_data(get_paca()->kstack, mmu_kernel_ssize, 2);
 110         if ((ksp_esid_data & ~0xfffffffUL) <= PAGE_OFFSET) {
 111                 ksp_esid_data &= ~SLB_ESID_V;
 112                 ksp_vsid_data = 0;
 113                 slb_shadow_clear(2);
 114         } else {
 115                 /* Update stack entry; others don't change */
 116                 slb_shadow_update(get_paca()->kstack, mmu_kernel_ssize, lflags, 2);
 117                 ksp_vsid_data = get_slb_shadow()->save_area[2].vsid;
 118         }
 119
 120         /*
 121          * We can't take a PMU exception in the following code, so hard
 122          * disable interrupts.
 123          */
 124         hard_irq_disable();
 125
 126         /* We need to do this all in asm, so we're sure we don't touch
 127          * the stack between the slbia and rebolting it. */
 128         asm volatile("isync\n"
 129                      "slbia\n"
 130                      /* Slot 1 - first VMALLOC segment */
 131                      "slbmte    %0,%1\n"
 132                      /* Slot 2 - kernel stack */
 133                      "slbmte    %2,%3\n"
 134                      "isync"
 135                      :: "r"(mk_vsid_data(VMALLOC_START, mmu_kernel_ssize, vflags)),
 136                         "r"(mk_esid_data(VMALLOC_START, mmu_kernel_ssize, 1)),
 137                         "r"(ksp_vsid_data),
 138                         "r"(ksp_esid_data)
 139                      : "memory");
 140 }
 141
 142 void slb_vmalloc_update(void)
 143 {
 144         unsigned long vflags;
 145
 146         vflags = SLB_VSID_KERNEL | mmu_psize_defs[mmu_vmalloc_psize].sllp;
 147         slb_shadow_update(VMALLOC_START, mmu_kernel_ssize, vflags, 1);
 148         slb_flush_and_rebolt();
 149 }
 150
 151 /* Helper function to compare esids.  There are four cases to handle.
 152  * 1. The system is not 1T segment size capable.  Use the GET_ESID compare.
 153  * 2. The system is 1T capable, both addresses are < 1T, use the GET_ESID compare.
 154  * 3. The system is 1T capable, only one of the two addresses is > 1T.  This is not a match.
 155  * 4. The system is 1T capable, both addresses are > 1T, use the GET_ESID_1T macro to compare.
 156  */
 157 static inline int esids_match(unsigned long addr1, unsigned long addr2)
 158 {
 159         int esid_1t_count;
 160
 161         /* System is not 1T segment size capable. */
 162         if (!cpu_has_feature(CPU_FTR_1T_SEGMENT))
 163                 return (GET_ESID(addr1) == GET_ESID(addr2));
 164
 165         esid_1t_count = (((addr1 >> SID_SHIFT_1T) != 0) +
 166                                 ((addr2 >> SID_SHIFT_1T) != 0));
 167
 168         /* both addresses are < 1T */
 169         if (esid_1t_count == 0)
 170                 return (GET_ESID(addr1) == GET_ESID(addr2));
 171
 172         /* One address < 1T, the other > 1T.  Not a match */
 173         if (esid_1t_count == 1)
 174                 return 0;
 175
 176         /* Both addresses are > 1T. */
 177         return (GET_ESID_1T(addr1) == GET_ESID_1T(addr2));
 178 }
 179
 180 /* Flush all user entries from the segment table of the current processor. */
 181 void switch_slb(struct task_struct *tsk, struct mm_struct *mm)
 182 {
 183         unsigned long offset = get_paca()->slb_cache_ptr;
 184         unsigned long slbie_data = 0;
 185         unsigned long pc = KSTK_EIP(tsk);
 186         unsigned long stack = KSTK_ESP(tsk);
 187         unsigned long unmapped_base;
 188
 189         if (!cpu_has_feature(CPU_FTR_NO_SLBIE_B) &&
 190             offset <= SLB_CACHE_ENTRIES) {
 191                 int i;
 192                 asm volatile("isync" : : : "memory");
 193                 for (i = 0; i < offset; i++) {
 194                         slbie_data = (unsigned long)get_paca()->slb_cache[i]
 195                                 << SID_SHIFT; /* EA */
 196                         slbie_data |= user_segment_size(slbie_data)
 197                                 << SLBIE_SSIZE_SHIFT;
 198                         slbie_data |= SLBIE_C; /* C set for user addresses */
 199                         asm volatile("slbie %0" : : "r" (slbie_data));
 200                 }
 201                 asm volatile("isync" : : : "memory");
 202         } else {
 203                 slb_flush_and_rebolt();
 204         }
 205
 206         /* Workaround POWER5 < DD2.1 issue */
 207         if (offset == 1 || offset > SLB_CACHE_ENTRIES)
 208                 asm volatile("slbie %0" : : "r" (slbie_data));
 209
 210         get_paca()->slb_cache_ptr = 0;
 211         get_paca()->context = mm->context;
 212
 213         /*
 214          * preload some userspace segments into the SLB.
 215          */
 216         if (test_tsk_thread_flag(tsk, TIF_32BIT))
 217                 unmapped_base = TASK_UNMAPPED_BASE_USER32;
 218         else
 219                 unmapped_base = TASK_UNMAPPED_BASE_USER64;
 220
 221         if (is_kernel_addr(pc))
 222                 return;
 223         slb_allocate(pc);
 224
 225         if (esids_match(pc,stack))
 226                 return;
 227
 228         if (is_kernel_addr(stack))
 229                 return;
 230         slb_allocate(stack);
 231
 232         if (esids_match(pc,unmapped_base) || esids_match(stack,unmapped_base))
 233                 return;
 234
 235         if (is_kernel_addr(unmapped_base))
 236                 return;
 237         slb_allocate(unmapped_base);
 238 }
 239
 240 static inline void patch_slb_encoding(unsigned int *insn_addr,
 241                                       unsigned int immed)
 242 {
 243         *insn_addr = (*insn_addr & 0xffff0000) | immed;
 244         flush_icache_range((unsigned long)insn_addr, 4+
 245                            (unsigned long)insn_addr);
 246 }
 247
 248 void slb_set_size(u16 size)
 249 {
 250         extern unsigned int *slb_compare_rr_to_size;
 251
 252         if (mmu_slb_size == size)
 253                 return;
 254
 255         mmu_slb_size = size;
 256         patch_slb_encoding(slb_compare_rr_to_size, mmu_slb_size);
 257 }
 258
 259 void slb_initialize(void)
 260 {
 261         unsigned long linear_llp, vmalloc_llp, io_llp;
 262         unsigned long lflags, vflags;
 263         static int slb_encoding_inited;
 264         extern unsigned int *slb_miss_kernel_load_linear;
 265         extern unsigned int *slb_miss_kernel_load_io;
 266         extern unsigned int *slb_compare_rr_to_size;
 267 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 268         extern unsigned int *slb_miss_kernel_load_vmemmap;
 269         unsigned long vmemmap_llp;
 270 #endif
 271
 272         /* Prepare our SLB miss handler based on our page size */
 273         linear_llp = mmu_psize_defs[mmu_linear_psize].sllp;
 274         io_llp = mmu_psize_defs[mmu_io_psize].sllp;
 275         vmalloc_llp = mmu_psize_defs[mmu_vmalloc_psize].sllp;
 276         get_paca()->vmalloc_sllp = SLB_VSID_KERNEL | vmalloc_llp;
 277 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 278         vmemmap_llp = mmu_psize_defs[mmu_vmemmap_psize].sllp;
 279 #endif
 280         if (!slb_encoding_inited) {
 281                 slb_encoding_inited = 1;
 282                 patch_slb_encoding(slb_miss_kernel_load_linear,
 283                                    SLB_VSID_KERNEL | linear_llp);
 284                 patch_slb_encoding(slb_miss_kernel_load_io,
 285                                    SLB_VSID_KERNEL | io_llp);
 286                 patch_slb_encoding(slb_compare_rr_to_size,
 287                                    mmu_slb_size);
 288
 289                 pr_devel("SLB: linear  LLP = %04lx\n", linear_llp);
 290                 pr_devel("SLB: io      LLP = %04lx\n", io_llp);
 291
 292 #ifdef CONFIG_SPARSEMEM_VMEMMAP
 293                 patch_slb_encoding(slb_miss_kernel_load_vmemmap,
 294                                    SLB_VSID_KERNEL | vmemmap_llp);
 295                 pr_devel("SLB: vmemmap LLP = %04lx\n", vmemmap_llp);
 296 #endif
 297         }
 298
 299         get_paca()->stab_rr = SLB_NUM_BOLTED;
 300
 301         /* On iSeries the bolted entries have already been set up by
 302          * the hypervisor from the lparMap data in head.S */
 303         if (firmware_has_feature(FW_FEATURE_ISERIES))
 304                 return;
 305
 306         lflags = SLB_VSID_KERNEL | linear_llp;
 307         vflags = SLB_VSID_KERNEL | vmalloc_llp;
 308
 309         /* Invalidate the entire SLB (even slot 0) & all the ERATS */
 310         asm volatile("isync":::"memory");
 311         asm volatile("slbmte  %0,%0"::"r" (0) : "memory");
 312         asm volatile("isync; slbia; isync":::"memory");
 313         create_shadowed_slbe(PAGE_OFFSET, mmu_kernel_ssize, lflags, 0);
 314
 315         create_shadowed_slbe(VMALLOC_START, mmu_kernel_ssize, vflags, 1);
 316
 317         /* For the boot cpu, we're running on the stack in init_thread_union,
 318          * which is in the first segment of the linear mapping, and also
 319          * get_paca()->kstack hasn't been initialized yet.
 320          * For secondary cpus, we need to bolt the kernel stack entry now.
 321          */
 322         slb_shadow_clear(2);
 323         if (raw_smp_processor_id() != boot_cpuid &&
 324             (get_paca()->kstack & slb_esid_mask(mmu_kernel_ssize)) > PAGE_OFFSET)
 325                 create_shadowed_slbe(get_paca()->kstack,
 326                                      mmu_kernel_ssize, lflags, 2);
 327
 328         asm volatile("isync":::"memory");
 329 }