arch/parisc/kernel/cache.c

   1 /*
   2  * This file is subject to the terms and conditions of the GNU General Public
   3  * License.  See the file "COPYING" in the main directory of this archive
   4  * for more details.
   5  *
   6  * Copyright (C) 1999-2006 Helge Deller <deller@gmx.de> (07-13-1999)
   7  * Copyright (C) 1999 SuSE GmbH Nuernberg
   8  * Copyright (C) 2000 Philipp Rumpf (prumpf@tux.org)
   9  *
  10  * Cache and TLB management
  11  *
  12  */
  13
  14 #include <linux/init.h>
  15 #include <linux/kernel.h>
  16 #include <linux/mm.h>
  17 #include <linux/module.h>
  18 #include <linux/seq_file.h>
  19 #include <linux/pagemap.h>
  20 #include <linux/sched.h>
  21 #include <asm/pdc.h>
  22 #include <asm/cache.h>
  23 #include <asm/cacheflush.h>
  24 #include <asm/tlbflush.h>
  25 #include <asm/page.h>
  26 #include <asm/pgalloc.h>
  27 #include <asm/processor.h>
  28 #include <asm/sections.h>
  29 #include <asm/shmparam.h>
  30
  31 int split_tlb __read_mostly;
  32 int dcache_stride __read_mostly;
  33 int icache_stride __read_mostly;
  34 EXPORT_SYMBOL(dcache_stride);
  35
  36 void flush_dcache_page_asm(unsigned long phys_addr, unsigned long vaddr);
  37 EXPORT_SYMBOL(flush_dcache_page_asm);
  38 void flush_icache_page_asm(unsigned long phys_addr, unsigned long vaddr);
  39
  40
  41 /* On some machines (e.g. ones with the Merced bus), there can be
  42  * only a single PxTLB broadcast at a time; this must be guaranteed
  43  * by software.  We put a spinlock around all TLB flushes  to
  44  * ensure this.
  45  */
  46 DEFINE_SPINLOCK(pa_tlb_lock);
  47
  48 struct pdc_cache_info cache_info __read_mostly;
  49 #ifndef CONFIG_PA20
  50 static struct pdc_btlb_info btlb_info __read_mostly;
  51 #endif
  52
  53 #ifdef CONFIG_SMP
  54 void
  55 flush_data_cache(void)
  56 {
  57         on_each_cpu(flush_data_cache_local, NULL, 1);
  58 }
  59 void
  60 flush_instruction_cache(void)
  61 {
  62         on_each_cpu(flush_instruction_cache_local, NULL, 1);
  63 }
  64 #endif
  65
  66 void
  67 flush_cache_all_local(void)
  68 {
  69         flush_instruction_cache_local(NULL);
  70         flush_data_cache_local(NULL);
  71 }
  72 EXPORT_SYMBOL(flush_cache_all_local);
  73
  74 /* Virtual address of pfn.  */
  75 #define pfn_va(pfn)     __va(PFN_PHYS(pfn))
  76
  77 void
  78 update_mmu_cache(struct vm_area_struct *vma, unsigned long address, pte_t *ptep)
  79 {
  80         unsigned long pfn = pte_pfn(*ptep);
  81         struct page *page;
  82
  83         /* We don't have pte special.  As a result, we can be called with
  84            an invalid pfn and we don't need to flush the kernel dcache page.
  85            This occurs with FireGL card in C8000.  */
  86         if (!pfn_valid(pfn))
  87                 return;
  88
  89         page = pfn_to_page(pfn);
  90         if (page_mapping(page) && test_bit(PG_dcache_dirty, &page->flags)) {
  91                 flush_kernel_dcache_page_addr(pfn_va(pfn));
  92                 clear_bit(PG_dcache_dirty, &page->flags);
  93         } else if (parisc_requires_coherency())
  94                 flush_kernel_dcache_page_addr(pfn_va(pfn));
  95 }
  96
  97 void
  98 show_cache_info(struct seq_file *m)
  99 {
 100         char buf[32];
 101
 102         seq_printf(m, "I-cache\t\t: %ld KB\n",
 103                 cache_info.ic_size/1024 );
 104         if (cache_info.dc_loop != 1)
 105                 snprintf(buf, 32, "%lu-way associative", cache_info.dc_loop);
 106         seq_printf(m, "D-cache\t\t: %ld KB (%s%s, %s)\n",
 107                 cache_info.dc_size/1024,
 108                 (cache_info.dc_conf.cc_wt ? "WT":"WB"),
 109                 (cache_info.dc_conf.cc_sh ? ", shared I/D":""),
 110                 ((cache_info.dc_loop == 1) ? "direct mapped" : buf));
 111         seq_printf(m, "ITLB entries\t: %ld\n" "DTLB entries\t: %ld%s\n",
 112                 cache_info.it_size,
 113                 cache_info.dt_size,
 114                 cache_info.dt_conf.tc_sh ? " - shared with ITLB":""
 115         );
 116
 117 #ifndef CONFIG_PA20
 118         /* BTLB - Block TLB */
 119         if (btlb_info.max_size==0) {
 120                 seq_printf(m, "BTLB\t\t: not supported\n" );
 121         } else {
 122                 seq_printf(m,
 123                 "BTLB fixed\t: max. %d pages, pagesize=%d (%dMB)\n"
 124                 "BTLB fix-entr.\t: %d instruction, %d data (%d combined)\n"
 125                 "BTLB var-entr.\t: %d instruction, %d data (%d combined)\n",
 126                 btlb_info.max_size, (int)4096,
 127                 btlb_info.max_size>>8,
 128                 btlb_info.fixed_range_info.num_i,
 129                 btlb_info.fixed_range_info.num_d,
 130                 btlb_info.fixed_range_info.num_comb,
 131                 btlb_info.variable_range_info.num_i,
 132                 btlb_info.variable_range_info.num_d,
 133                 btlb_info.variable_range_info.num_comb
 134                 );
 135         }
 136 #endif
 137 }
 138
 139 void __init
 140 parisc_cache_init(void)
 141 {
 142         if (pdc_cache_info(&cache_info) < 0)
 143                 panic("parisc_cache_init: pdc_cache_info failed");
 144
 145 #if 0
 146         printk("ic_size %lx dc_size %lx it_size %lx\n",
 147                 cache_info.ic_size,
 148                 cache_info.dc_size,
 149                 cache_info.it_size);
 150
 151         printk("DC  base 0x%lx stride 0x%lx count 0x%lx loop 0x%lx\n",
 152                 cache_info.dc_base,
 153                 cache_info.dc_stride,
 154                 cache_info.dc_count,
 155                 cache_info.dc_loop);
 156
 157         printk("dc_conf = 0x%lx  alias %d blk %d line %d shift %d\n",
 158                 *(unsigned long *) (&cache_info.dc_conf),
 159                 cache_info.dc_conf.cc_alias,
 160                 cache_info.dc_conf.cc_block,
 161                 cache_info.dc_conf.cc_line,
 162                 cache_info.dc_conf.cc_shift);
 163         printk("        wt %d sh %d cst %d hv %d\n",
 164                 cache_info.dc_conf.cc_wt,
 165                 cache_info.dc_conf.cc_sh,
 166                 cache_info.dc_conf.cc_cst,
 167                 cache_info.dc_conf.cc_hv);
 168
 169         printk("IC  base 0x%lx stride 0x%lx count 0x%lx loop 0x%lx\n",
 170                 cache_info.ic_base,
 171                 cache_info.ic_stride,
 172                 cache_info.ic_count,
 173                 cache_info.ic_loop);
 174
 175         printk("ic_conf = 0x%lx  alias %d blk %d line %d shift %d\n",
 176                 *(unsigned long *) (&cache_info.ic_conf),
 177                 cache_info.ic_conf.cc_alias,
 178                 cache_info.ic_conf.cc_block,
 179                 cache_info.ic_conf.cc_line,
 180                 cache_info.ic_conf.cc_shift);
 181         printk("        wt %d sh %d cst %d hv %d\n",
 182                 cache_info.ic_conf.cc_wt,
 183                 cache_info.ic_conf.cc_sh,
 184                 cache_info.ic_conf.cc_cst,
 185                 cache_info.ic_conf.cc_hv);
 186
 187         printk("D-TLB conf: sh %d page %d cst %d aid %d pad1 %d\n",
 188                 cache_info.dt_conf.tc_sh,
 189                 cache_info.dt_conf.tc_page,
 190                 cache_info.dt_conf.tc_cst,
 191                 cache_info.dt_conf.tc_aid,
 192                 cache_info.dt_conf.tc_pad1);
 193
 194         printk("I-TLB conf: sh %d page %d cst %d aid %d pad1 %d\n",
 195                 cache_info.it_conf.tc_sh,
 196                 cache_info.it_conf.tc_page,
 197                 cache_info.it_conf.tc_cst,
 198                 cache_info.it_conf.tc_aid,
 199                 cache_info.it_conf.tc_pad1);
 200 #endif
 201
 202         split_tlb = 0;
 203         if (cache_info.dt_conf.tc_sh == 0 || cache_info.dt_conf.tc_sh == 2) {
 204                 if (cache_info.dt_conf.tc_sh == 2)
 205                         printk(KERN_WARNING "Unexpected TLB configuration. "
 206                         "Will flush I/D separately (could be optimized).\n");
 207
 208                 split_tlb = 1;
 209         }
 210
 211         /* "New and Improved" version from Jim Hull
 212          *      (1 << (cc_block-1)) * (cc_line << (4 + cnf.cc_shift))
 213          * The following CAFL_STRIDE is an optimized version, see
 214          * http://lists.parisc-linux.org/pipermail/parisc-linux/2004-June/023625.html
 215          * http://lists.parisc-linux.org/pipermail/parisc-linux/2004-June/023671.html
 216          */
 217 #define CAFL_STRIDE(cnf) (cnf.cc_line << (3 + cnf.cc_block + cnf.cc_shift))
 218         dcache_stride = CAFL_STRIDE(cache_info.dc_conf);
 219         icache_stride = CAFL_STRIDE(cache_info.ic_conf);
 220 #undef CAFL_STRIDE
 221
 222 #ifndef CONFIG_PA20
 223         if (pdc_btlb_info(&btlb_info) < 0) {
 224                 memset(&btlb_info, 0, sizeof btlb_info);
 225         }
 226 #endif
 227
 228         if ((boot_cpu_data.pdc.capabilities & PDC_MODEL_NVA_MASK) ==
 229                                                 PDC_MODEL_NVA_UNSUPPORTED) {
 230                 printk(KERN_WARNING "parisc_cache_init: Only equivalent aliasing supported!\n");
 231 #if 0
 232                 panic("SMP kernel required to avoid non-equivalent aliasing");
 233 #endif
 234         }
 235 }
 236
 237 void disable_sr_hashing(void)
 238 {
 239         int srhash_type, retval;
 240         unsigned long space_bits;
 241
 242         switch (boot_cpu_data.cpu_type) {
 243         case pcx: /* We shouldn't get this far.  setup.c should prevent it. */
 244                 BUG();
 245                 return;
 246
 247         case pcxs:
 248         case pcxt:
 249         case pcxt_:
 250                 srhash_type = SRHASH_PCXST;
 251                 break;
 252
 253         case pcxl:
 254                 srhash_type = SRHASH_PCXL;
 255                 break;
 256
 257         case pcxl2: /* pcxl2 doesn't support space register hashing */
 258                 return;
 259
 260         default: /* Currently all PA2.0 machines use the same ins. sequence */
 261                 srhash_type = SRHASH_PA20;
 262                 break;
 263         }
 264
 265         disable_sr_hashing_asm(srhash_type);
 266
 267         retval = pdc_spaceid_bits(&space_bits);
 268         /* If this procedure isn't implemented, don't panic. */
 269         if (retval < 0 && retval != PDC_BAD_OPTION)
 270                 panic("pdc_spaceid_bits call failed.\n");
 271         if (space_bits != 0)
 272                 panic("SpaceID hashing is still on!\n");
 273 }
 274
 275 static inline void
 276 __flush_cache_page(struct vm_area_struct *vma, unsigned long vmaddr,
 277                    unsigned long physaddr)
 278 {
 279         preempt_disable();
 280         flush_dcache_page_asm(physaddr, vmaddr);
 281         if (vma->vm_flags & VM_EXEC)
 282                 flush_icache_page_asm(physaddr, vmaddr);
 283         preempt_enable();
 284 }
 285
 286 void flush_dcache_page(struct page *page)
 287 {
 288         struct address_space *mapping = page_mapping(page);
 289         struct vm_area_struct *mpnt;
 290         unsigned long offset;
 291         unsigned long addr, old_addr = 0;
 292         pgoff_t pgoff;
 293
 294         if (mapping && !mapping_mapped(mapping)) {
 295                 set_bit(PG_dcache_dirty, &page->flags);
 296                 return;
 297         }
 298
 299         flush_kernel_dcache_page(page);
 300
 301         if (!mapping)
 302                 return;
 303
 304         pgoff = page->index << (PAGE_CACHE_SHIFT - PAGE_SHIFT);
 305
 306         /* We have carefully arranged in arch_get_unmapped_area() that
 307          * *any* mappings of a file are always congruently mapped (whether
 308          * declared as MAP_PRIVATE or MAP_SHARED), so we only need
 309          * to flush one address here for them all to become coherent */
 310
 311         flush_dcache_mmap_lock(mapping);
 312         vma_interval_tree_foreach(mpnt, &mapping->i_mmap, pgoff, pgoff) {
 313                 offset = (pgoff - mpnt->vm_pgoff) << PAGE_SHIFT;
 314                 addr = mpnt->vm_start + offset;
 315
 316                 /* The TLB is the engine of coherence on parisc: The
 317                  * CPU is entitled to speculate any page with a TLB
 318                  * mapping, so here we kill the mapping then flush the
 319                  * page along a special flush only alias mapping.
 320                  * This guarantees that the page is no-longer in the
 321                  * cache for any process and nor may it be
 322                  * speculatively read in (until the user or kernel
 323                  * specifically accesses it, of course) */
 324
 325                 flush_tlb_page(mpnt, addr);
 326                 if (old_addr == 0 || (old_addr & (SHM_COLOUR - 1))
 327                                       != (addr & (SHM_COLOUR - 1))) {
 328                         __flush_cache_page(mpnt, addr, page_to_phys(page));
 329                         if (old_addr)
 330                                 printk(KERN_ERR "INEQUIVALENT ALIASES 0x%lx and 0x%lx in file %s\n", old_addr, addr, mpnt->vm_file ? (char *)mpnt->vm_file->f_path.dentry->d_name.name : "(null)");
 331                         old_addr = addr;
 332                 }
 333         }
 334         flush_dcache_mmap_unlock(mapping);
 335 }
 336 EXPORT_SYMBOL(flush_dcache_page);
 337
 338 /* Defined in arch/parisc/kernel/pacache.S */
 339 EXPORT_SYMBOL(flush_kernel_dcache_range_asm);
 340 EXPORT_SYMBOL(flush_kernel_dcache_page_asm);
 341 EXPORT_SYMBOL(flush_data_cache_local);
 342 EXPORT_SYMBOL(flush_kernel_icache_range_asm);
 343
 344 #define FLUSH_THRESHOLD 0x80000 /* 0.5MB */
 345 int parisc_cache_flush_threshold __read_mostly = FLUSH_THRESHOLD;
 346
 347 void __init parisc_setup_cache_timing(void)
 348 {
 349         unsigned long rangetime, alltime;
 350         unsigned long size;
 351
 352         alltime = mfctl(16);
 353         flush_data_cache();
 354         alltime = mfctl(16) - alltime;
 355
 356         size = (unsigned long)(_end - _text);
 357         rangetime = mfctl(16);
 358         flush_kernel_dcache_range((unsigned long)_text, size);
 359         rangetime = mfctl(16) - rangetime;
 360
 361         printk(KERN_DEBUG "Whole cache flush %lu cycles, flushing %lu bytes %lu cycles\n",
 362                 alltime, size, rangetime);
 363
 364         /* Racy, but if we see an intermediate value, it's ok too... */
 365         parisc_cache_flush_threshold = size * alltime / rangetime;
 366
 367         parisc_cache_flush_threshold = (parisc_cache_flush_threshold + L1_CACHE_BYTES - 1) &~ (L1_CACHE_BYTES - 1);
 368         if (!parisc_cache_flush_threshold)
 369                 parisc_cache_flush_threshold = FLUSH_THRESHOLD;
 370
 371         if (parisc_cache_flush_threshold > cache_info.dc_size)
 372                 parisc_cache_flush_threshold = cache_info.dc_size;
 373
 374         printk(KERN_INFO "Setting cache flush threshold to %x (%d CPUs online)\n", parisc_cache_flush_threshold, num_online_cpus());
 375 }
 376
 377 extern void purge_kernel_dcache_page_asm(unsigned long);
 378 extern void clear_user_page_asm(void *, unsigned long);
 379 extern void copy_user_page_asm(void *, void *, unsigned long);
 380
 381 void flush_kernel_dcache_page_addr(void *addr)
 382 {
 383         unsigned long flags;
 384
 385         flush_kernel_dcache_page_asm(addr);
 386         purge_tlb_start(flags);
 387         pdtlb_kernel(addr);
 388         purge_tlb_end(flags);
 389 }
 390 EXPORT_SYMBOL(flush_kernel_dcache_page_addr);
 391
 392 void copy_user_page(void *vto, void *vfrom, unsigned long vaddr,
 393         struct page *pg)
 394 {
 395        /* Copy using kernel mapping.  No coherency is needed (all in
 396           kunmap) for the `to' page.  However, the `from' page needs to
 397           be flushed through a mapping equivalent to the user mapping
 398           before it can be accessed through the kernel mapping. */
 399         preempt_disable();
 400         flush_dcache_page_asm(__pa(vfrom), vaddr);
 401         preempt_enable();
 402         copy_page_asm(vto, vfrom);
 403 }
 404 EXPORT_SYMBOL(copy_user_page);
 405
 406 void purge_tlb_entries(struct mm_struct *mm, unsigned long addr)
 407 {
 408         unsigned long flags;
 409
 410         /* Note: purge_tlb_entries can be called at startup with
 411            no context.  */
 412
 413         purge_tlb_start(flags);
 414         mtsp(mm->context, 1);
 415         pdtlb(addr);
 416         pitlb(addr);
 417         purge_tlb_end(flags);
 418 }
 419 EXPORT_SYMBOL(purge_tlb_entries);
 420
 421 void __flush_tlb_range(unsigned long sid, unsigned long start,
 422                        unsigned long end)
 423 {
 424         unsigned long npages;
 425
 426         npages = ((end - (start & PAGE_MASK)) + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
 427         if (npages >= 512)  /* 2MB of space: arbitrary, should be tuned */
 428                 flush_tlb_all();
 429         else {
 430                 unsigned long flags;
 431
 432                 purge_tlb_start(flags);
 433                 mtsp(sid, 1);
 434                 if (split_tlb) {
 435                         while (npages--) {
 436                                 pdtlb(start);
 437                                 pitlb(start);
 438                                 start += PAGE_SIZE;
 439                         }
 440                 } else {
 441                         while (npages--) {
 442                                 pdtlb(start);
 443                                 start += PAGE_SIZE;
 444                         }
 445                 }
 446                 purge_tlb_end(flags);
 447         }
 448 }
 449
 450 static void cacheflush_h_tmp_function(void *dummy)
 451 {
 452         flush_cache_all_local();
 453 }
 454
 455 void flush_cache_all(void)
 456 {
 457         on_each_cpu(cacheflush_h_tmp_function, NULL, 1);
 458 }
 459
 460 static inline unsigned long mm_total_size(struct mm_struct *mm)
 461 {
 462         struct vm_area_struct *vma;
 463         unsigned long usize = 0;
 464
 465         for (vma = mm->mmap; vma; vma = vma->vm_next)
 466                 usize += vma->vm_end - vma->vm_start;
 467         return usize;
 468 }
 469
 470 static inline pte_t *get_ptep(pgd_t *pgd, unsigned long addr)
 471 {
 472         pte_t *ptep = NULL;
 473
 474         if (!pgd_none(*pgd)) {
 475                 pud_t *pud = pud_offset(pgd, addr);
 476                 if (!pud_none(*pud)) {
 477                         pmd_t *pmd = pmd_offset(pud, addr);
 478                         if (!pmd_none(*pmd))
 479                                 ptep = pte_offset_map(pmd, addr);
 480                 }
 481         }
 482         return ptep;
 483 }
 484
 485 void flush_cache_mm(struct mm_struct *mm)
 486 {
 487         struct vm_area_struct *vma;
 488         pgd_t *pgd;
 489
 490         /* Flushing the whole cache on each cpu takes forever on
 491            rp3440, etc.  So, avoid it if the mm isn't too big.  */
 492         if (mm_total_size(mm) >= parisc_cache_flush_threshold) {
 493                 flush_cache_all();
 494                 return;
 495         }
 496
 497         if (mm->context == mfsp(3)) {
 498                 for (vma = mm->mmap; vma; vma = vma->vm_next) {
 499                         flush_user_dcache_range_asm(vma->vm_start, vma->vm_end);
 500                         if ((vma->vm_flags & VM_EXEC) == 0)
 501                                 continue;
 502                         flush_user_icache_range_asm(vma->vm_start, vma->vm_end);
 503                 }
 504                 return;
 505         }
 506
 507         pgd = mm->pgd;
 508         for (vma = mm->mmap; vma; vma = vma->vm_next) {
 509                 unsigned long addr;
 510
 511                 for (addr = vma->vm_start; addr < vma->vm_end;
 512                      addr += PAGE_SIZE) {
 513                         unsigned long pfn;
 514                         pte_t *ptep = get_ptep(pgd, addr);
 515                         if (!ptep)
 516                                 continue;
 517                         pfn = pte_pfn(*ptep);
 518                         if (!pfn_valid(pfn))
 519                                 continue;
 520                         __flush_cache_page(vma, addr, PFN_PHYS(pfn));
 521                 }
 522         }
 523 }
 524
 525 void
 526 flush_user_dcache_range(unsigned long start, unsigned long end)
 527 {
 528         if ((end - start) < parisc_cache_flush_threshold)
 529                 flush_user_dcache_range_asm(start,end);
 530         else
 531                 flush_data_cache();
 532 }
 533
 534 void
 535 flush_user_icache_range(unsigned long start, unsigned long end)
 536 {
 537         if ((end - start) < parisc_cache_flush_threshold)
 538                 flush_user_icache_range_asm(start,end);
 539         else
 540                 flush_instruction_cache();
 541 }
 542
 543 void flush_cache_range(struct vm_area_struct *vma,
 544                 unsigned long start, unsigned long end)
 545 {
 546         unsigned long addr;
 547         pgd_t *pgd;
 548
 549         BUG_ON(!vma->vm_mm->context);
 550
 551         if ((end - start) >= parisc_cache_flush_threshold) {
 552                 flush_cache_all();
 553                 return;
 554         }
 555
 556         if (vma->vm_mm->context == mfsp(3)) {
 557                 flush_user_dcache_range_asm(start, end);
 558                 if (vma->vm_flags & VM_EXEC)
 559                         flush_user_icache_range_asm(start, end);
 560                 return;
 561         }
 562
 563         pgd = vma->vm_mm->pgd;
 564         for (addr = start & PAGE_MASK; addr < end; addr += PAGE_SIZE) {
 565                 unsigned long pfn;
 566                 pte_t *ptep = get_ptep(pgd, addr);
 567                 if (!ptep)
 568                         continue;
 569                 pfn = pte_pfn(*ptep);
 570                 if (pfn_valid(pfn))
 571                         __flush_cache_page(vma, addr, PFN_PHYS(pfn));
 572         }
 573 }
 574
 575 void
 576 flush_cache_page(struct vm_area_struct *vma, unsigned long vmaddr, unsigned long pfn)
 577 {
 578         BUG_ON(!vma->vm_mm->context);
 579
 580         if (pfn_valid(pfn)) {
 581                 flush_tlb_page(vma, vmaddr);
 582                 __flush_cache_page(vma, vmaddr, PFN_PHYS(pfn));
 583         }
 584 }