Merge branch 'avr32-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/hskinnemo...
[linux-2.6] / arch / sparc64 / mm / init.c
1 /*  $Id: init.c,v 1.209 2002/02/09 19:49:31 davem Exp $
2  *  arch/sparc64/mm/init.c
3  *
4  *  Copyright (C) 1996-1999 David S. Miller (davem@caip.rutgers.edu)
5  *  Copyright (C) 1997-1999 Jakub Jelinek (jj@sunsite.mff.cuni.cz)
6  */
7  
8 #include <linux/module.h>
9 #include <linux/kernel.h>
10 #include <linux/sched.h>
11 #include <linux/string.h>
12 #include <linux/init.h>
13 #include <linux/bootmem.h>
14 #include <linux/mm.h>
15 #include <linux/hugetlb.h>
16 #include <linux/slab.h>
17 #include <linux/initrd.h>
18 #include <linux/swap.h>
19 #include <linux/pagemap.h>
20 #include <linux/poison.h>
21 #include <linux/fs.h>
22 #include <linux/seq_file.h>
23 #include <linux/kprobes.h>
24 #include <linux/cache.h>
25 #include <linux/sort.h>
26 #include <linux/percpu.h>
27
28 #include <asm/head.h>
29 #include <asm/system.h>
30 #include <asm/page.h>
31 #include <asm/pgalloc.h>
32 #include <asm/pgtable.h>
33 #include <asm/oplib.h>
34 #include <asm/iommu.h>
35 #include <asm/io.h>
36 #include <asm/uaccess.h>
37 #include <asm/mmu_context.h>
38 #include <asm/tlbflush.h>
39 #include <asm/dma.h>
40 #include <asm/starfire.h>
41 #include <asm/tlb.h>
42 #include <asm/spitfire.h>
43 #include <asm/sections.h>
44 #include <asm/tsb.h>
45 #include <asm/hypervisor.h>
46 #include <asm/prom.h>
47 #include <asm/sstate.h>
48 #include <asm/mdesc.h>
49
50 #define MAX_PHYS_ADDRESS        (1UL << 42UL)
51 #define KPTE_BITMAP_CHUNK_SZ    (256UL * 1024UL * 1024UL)
52 #define KPTE_BITMAP_BYTES       \
53         ((MAX_PHYS_ADDRESS / KPTE_BITMAP_CHUNK_SZ) / 8)
54
55 unsigned long kern_linear_pte_xor[2] __read_mostly;
56
57 /* A bitmap, one bit for every 256MB of physical memory.  If the bit
58  * is clear, we should use a 4MB page (via kern_linear_pte_xor[0]) else
59  * if set we should use a 256MB page (via kern_linear_pte_xor[1]).
60  */
61 unsigned long kpte_linear_bitmap[KPTE_BITMAP_BYTES / sizeof(unsigned long)];
62
63 #ifndef CONFIG_DEBUG_PAGEALLOC
64 /* A special kernel TSB for 4MB and 256MB linear mappings.
65  * Space is allocated for this right after the trap table
66  * in arch/sparc64/kernel/head.S
67  */
68 extern struct tsb swapper_4m_tsb[KERNEL_TSB4M_NENTRIES];
69 #endif
70
71 #define MAX_BANKS       32
72
73 static struct linux_prom64_registers pavail[MAX_BANKS] __initdata;
74 static struct linux_prom64_registers pavail_rescan[MAX_BANKS] __initdata;
75 static int pavail_ents __initdata;
76 static int pavail_rescan_ents __initdata;
77
78 static int cmp_p64(const void *a, const void *b)
79 {
80         const struct linux_prom64_registers *x = a, *y = b;
81
82         if (x->phys_addr > y->phys_addr)
83                 return 1;
84         if (x->phys_addr < y->phys_addr)
85                 return -1;
86         return 0;
87 }
88
89 static void __init read_obp_memory(const char *property,
90                                    struct linux_prom64_registers *regs,
91                                    int *num_ents)
92 {
93         int node = prom_finddevice("/memory");
94         int prop_size = prom_getproplen(node, property);
95         int ents, ret, i;
96
97         ents = prop_size / sizeof(struct linux_prom64_registers);
98         if (ents > MAX_BANKS) {
99                 prom_printf("The machine has more %s property entries than "
100                             "this kernel can support (%d).\n",
101                             property, MAX_BANKS);
102                 prom_halt();
103         }
104
105         ret = prom_getproperty(node, property, (char *) regs, prop_size);
106         if (ret == -1) {
107                 prom_printf("Couldn't get %s property from /memory.\n");
108                 prom_halt();
109         }
110
111         /* Sanitize what we got from the firmware, by page aligning
112          * everything.
113          */
114         for (i = 0; i < ents; i++) {
115                 unsigned long base, size;
116
117                 base = regs[i].phys_addr;
118                 size = regs[i].reg_size;
119
120                 size &= PAGE_MASK;
121                 if (base & ~PAGE_MASK) {
122                         unsigned long new_base = PAGE_ALIGN(base);
123
124                         size -= new_base - base;
125                         if ((long) size < 0L)
126                                 size = 0UL;
127                         base = new_base;
128                 }
129                 if (size == 0UL) {
130                         /* If it is empty, simply get rid of it.
131                          * This simplifies the logic of the other
132                          * functions that process these arrays.
133                          */
134                         memmove(&regs[i], &regs[i + 1],
135                                 (ents - i - 1) * sizeof(regs[0]));
136                         i--;
137                         ents--;
138                         continue;
139                 }
140                 regs[i].phys_addr = base;
141                 regs[i].reg_size = size;
142         }
143
144         *num_ents = ents;
145
146         sort(regs, ents, sizeof(struct linux_prom64_registers),
147              cmp_p64, NULL);
148 }
149
150 unsigned long *sparc64_valid_addr_bitmap __read_mostly;
151
152 /* Kernel physical address base and size in bytes.  */
153 unsigned long kern_base __read_mostly;
154 unsigned long kern_size __read_mostly;
155
156 /* Initial ramdisk setup */
157 extern unsigned long sparc_ramdisk_image64;
158 extern unsigned int sparc_ramdisk_image;
159 extern unsigned int sparc_ramdisk_size;
160
161 struct page *mem_map_zero __read_mostly;
162
163 unsigned int sparc64_highest_unlocked_tlb_ent __read_mostly;
164
165 unsigned long sparc64_kern_pri_context __read_mostly;
166 unsigned long sparc64_kern_pri_nuc_bits __read_mostly;
167 unsigned long sparc64_kern_sec_context __read_mostly;
168
169 int bigkernel = 0;
170
171 #ifdef CONFIG_DEBUG_DCFLUSH
172 atomic_t dcpage_flushes = ATOMIC_INIT(0);
173 #ifdef CONFIG_SMP
174 atomic_t dcpage_flushes_xcall = ATOMIC_INIT(0);
175 #endif
176 #endif
177
178 inline void flush_dcache_page_impl(struct page *page)
179 {
180         BUG_ON(tlb_type == hypervisor);
181 #ifdef CONFIG_DEBUG_DCFLUSH
182         atomic_inc(&dcpage_flushes);
183 #endif
184
185 #ifdef DCACHE_ALIASING_POSSIBLE
186         __flush_dcache_page(page_address(page),
187                             ((tlb_type == spitfire) &&
188                              page_mapping(page) != NULL));
189 #else
190         if (page_mapping(page) != NULL &&
191             tlb_type == spitfire)
192                 __flush_icache_page(__pa(page_address(page)));
193 #endif
194 }
195
196 #define PG_dcache_dirty         PG_arch_1
197 #define PG_dcache_cpu_shift     32UL
198 #define PG_dcache_cpu_mask      \
199         ((1UL<<ilog2(roundup_pow_of_two(NR_CPUS)))-1UL)
200
201 #define dcache_dirty_cpu(page) \
202         (((page)->flags >> PG_dcache_cpu_shift) & PG_dcache_cpu_mask)
203
204 static inline void set_dcache_dirty(struct page *page, int this_cpu)
205 {
206         unsigned long mask = this_cpu;
207         unsigned long non_cpu_bits;
208
209         non_cpu_bits = ~(PG_dcache_cpu_mask << PG_dcache_cpu_shift);
210         mask = (mask << PG_dcache_cpu_shift) | (1UL << PG_dcache_dirty);
211
212         __asm__ __volatile__("1:\n\t"
213                              "ldx       [%2], %%g7\n\t"
214                              "and       %%g7, %1, %%g1\n\t"
215                              "or        %%g1, %0, %%g1\n\t"
216                              "casx      [%2], %%g7, %%g1\n\t"
217                              "cmp       %%g7, %%g1\n\t"
218                              "membar    #StoreLoad | #StoreStore\n\t"
219                              "bne,pn    %%xcc, 1b\n\t"
220                              " nop"
221                              : /* no outputs */
222                              : "r" (mask), "r" (non_cpu_bits), "r" (&page->flags)
223                              : "g1", "g7");
224 }
225
226 static inline void clear_dcache_dirty_cpu(struct page *page, unsigned long cpu)
227 {
228         unsigned long mask = (1UL << PG_dcache_dirty);
229
230         __asm__ __volatile__("! test_and_clear_dcache_dirty\n"
231                              "1:\n\t"
232                              "ldx       [%2], %%g7\n\t"
233                              "srlx      %%g7, %4, %%g1\n\t"
234                              "and       %%g1, %3, %%g1\n\t"
235                              "cmp       %%g1, %0\n\t"
236                              "bne,pn    %%icc, 2f\n\t"
237                              " andn     %%g7, %1, %%g1\n\t"
238                              "casx      [%2], %%g7, %%g1\n\t"
239                              "cmp       %%g7, %%g1\n\t"
240                              "membar    #StoreLoad | #StoreStore\n\t"
241                              "bne,pn    %%xcc, 1b\n\t"
242                              " nop\n"
243                              "2:"
244                              : /* no outputs */
245                              : "r" (cpu), "r" (mask), "r" (&page->flags),
246                                "i" (PG_dcache_cpu_mask),
247                                "i" (PG_dcache_cpu_shift)
248                              : "g1", "g7");
249 }
250
251 static inline void tsb_insert(struct tsb *ent, unsigned long tag, unsigned long pte)
252 {
253         unsigned long tsb_addr = (unsigned long) ent;
254
255         if (tlb_type == cheetah_plus || tlb_type == hypervisor)
256                 tsb_addr = __pa(tsb_addr);
257
258         __tsb_insert(tsb_addr, tag, pte);
259 }
260
261 unsigned long _PAGE_ALL_SZ_BITS __read_mostly;
262 unsigned long _PAGE_SZBITS __read_mostly;
263
264 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address, pte_t pte)
265 {
266         struct mm_struct *mm;
267         struct tsb *tsb;
268         unsigned long tag, flags;
269         unsigned long tsb_index, tsb_hash_shift;
270
271         if (tlb_type != hypervisor) {
272                 unsigned long pfn = pte_pfn(pte);
273                 unsigned long pg_flags;
274                 struct page *page;
275
276                 if (pfn_valid(pfn) &&
277                     (page = pfn_to_page(pfn), page_mapping(page)) &&
278                     ((pg_flags = page->flags) & (1UL << PG_dcache_dirty))) {
279                         int cpu = ((pg_flags >> PG_dcache_cpu_shift) &
280                                    PG_dcache_cpu_mask);
281                         int this_cpu = get_cpu();
282
283                         /* This is just to optimize away some function calls
284                          * in the SMP case.
285                          */
286                         if (cpu == this_cpu)
287                                 flush_dcache_page_impl(page);
288                         else
289                                 smp_flush_dcache_page_impl(page, cpu);
290
291                         clear_dcache_dirty_cpu(page, cpu);
292
293                         put_cpu();
294                 }
295         }
296
297         mm = vma->vm_mm;
298
299         tsb_index = MM_TSB_BASE;
300         tsb_hash_shift = PAGE_SHIFT;
301
302         spin_lock_irqsave(&mm->context.lock, flags);
303
304 #ifdef CONFIG_HUGETLB_PAGE
305         if (mm->context.tsb_block[MM_TSB_HUGE].tsb != NULL) {
306                 if ((tlb_type == hypervisor &&
307                      (pte_val(pte) & _PAGE_SZALL_4V) == _PAGE_SZHUGE_4V) ||
308                     (tlb_type != hypervisor &&
309                      (pte_val(pte) & _PAGE_SZALL_4U) == _PAGE_SZHUGE_4U)) {
310                         tsb_index = MM_TSB_HUGE;
311                         tsb_hash_shift = HPAGE_SHIFT;
312                 }
313         }
314 #endif
315
316         tsb = mm->context.tsb_block[tsb_index].tsb;
317         tsb += ((address >> tsb_hash_shift) &
318                 (mm->context.tsb_block[tsb_index].tsb_nentries - 1UL));
319         tag = (address >> 22UL);
320         tsb_insert(tsb, tag, pte_val(pte));
321
322         spin_unlock_irqrestore(&mm->context.lock, flags);
323 }
324
325 void flush_dcache_page(struct page *page)
326 {
327         struct address_space *mapping;
328         int this_cpu;
329
330         if (tlb_type == hypervisor)
331                 return;
332
333         /* Do not bother with the expensive D-cache flush if it
334          * is merely the zero page.  The 'bigcore' testcase in GDB
335          * causes this case to run millions of times.
336          */
337         if (page == ZERO_PAGE(0))
338                 return;
339
340         this_cpu = get_cpu();
341
342         mapping = page_mapping(page);
343         if (mapping && !mapping_mapped(mapping)) {
344                 int dirty = test_bit(PG_dcache_dirty, &page->flags);
345                 if (dirty) {
346                         int dirty_cpu = dcache_dirty_cpu(page);
347
348                         if (dirty_cpu == this_cpu)
349                                 goto out;
350                         smp_flush_dcache_page_impl(page, dirty_cpu);
351                 }
352                 set_dcache_dirty(page, this_cpu);
353         } else {
354                 /* We could delay the flush for the !page_mapping
355                  * case too.  But that case is for exec env/arg
356                  * pages and those are %99 certainly going to get
357                  * faulted into the tlb (and thus flushed) anyways.
358                  */
359                 flush_dcache_page_impl(page);
360         }
361
362 out:
363         put_cpu();
364 }
365
366 void __kprobes flush_icache_range(unsigned long start, unsigned long end)
367 {
368         /* Cheetah and Hypervisor platform cpus have coherent I-cache. */
369         if (tlb_type == spitfire) {
370                 unsigned long kaddr;
371
372                 /* This code only runs on Spitfire cpus so this is
373                  * why we can assume _PAGE_PADDR_4U.
374                  */
375                 for (kaddr = start; kaddr < end; kaddr += PAGE_SIZE) {
376                         unsigned long paddr, mask = _PAGE_PADDR_4U;
377
378                         if (kaddr >= PAGE_OFFSET)
379                                 paddr = kaddr & mask;
380                         else {
381                                 pgd_t *pgdp = pgd_offset_k(kaddr);
382                                 pud_t *pudp = pud_offset(pgdp, kaddr);
383                                 pmd_t *pmdp = pmd_offset(pudp, kaddr);
384                                 pte_t *ptep = pte_offset_kernel(pmdp, kaddr);
385
386                                 paddr = pte_val(*ptep) & mask;
387                         }
388                         __flush_icache_page(paddr);
389                 }
390         }
391 }
392
393 void show_mem(void)
394 {
395         unsigned long total = 0, reserved = 0;
396         unsigned long shared = 0, cached = 0;
397         pg_data_t *pgdat;
398
399         printk(KERN_INFO "Mem-info:\n");
400         show_free_areas();
401         printk(KERN_INFO "Free swap:       %6ldkB\n",
402                nr_swap_pages << (PAGE_SHIFT-10));
403         for_each_online_pgdat(pgdat) {
404                 unsigned long i, flags;
405
406                 pgdat_resize_lock(pgdat, &flags);
407                 for (i = 0; i < pgdat->node_spanned_pages; i++) {
408                         struct page *page = pgdat_page_nr(pgdat, i);
409                         total++;
410                         if (PageReserved(page))
411                                 reserved++;
412                         else if (PageSwapCache(page))
413                                 cached++;
414                         else if (page_count(page))
415                                 shared += page_count(page) - 1;
416                 }
417                 pgdat_resize_unlock(pgdat, &flags);
418         }
419
420         printk(KERN_INFO "%lu pages of RAM\n", total);
421         printk(KERN_INFO "%lu reserved pages\n", reserved);
422         printk(KERN_INFO "%lu pages shared\n", shared);
423         printk(KERN_INFO "%lu pages swap cached\n", cached);
424
425         printk(KERN_INFO "%lu pages dirty\n",
426                global_page_state(NR_FILE_DIRTY));
427         printk(KERN_INFO "%lu pages writeback\n",
428                global_page_state(NR_WRITEBACK));
429         printk(KERN_INFO "%lu pages mapped\n",
430                global_page_state(NR_FILE_MAPPED));
431         printk(KERN_INFO "%lu pages slab\n",
432                 global_page_state(NR_SLAB_RECLAIMABLE) +
433                 global_page_state(NR_SLAB_UNRECLAIMABLE));
434         printk(KERN_INFO "%lu pages pagetables\n",
435                global_page_state(NR_PAGETABLE));
436 }
437
438 void mmu_info(struct seq_file *m)
439 {
440         if (tlb_type == cheetah)
441                 seq_printf(m, "MMU Type\t: Cheetah\n");
442         else if (tlb_type == cheetah_plus)
443                 seq_printf(m, "MMU Type\t: Cheetah+\n");
444         else if (tlb_type == spitfire)
445                 seq_printf(m, "MMU Type\t: Spitfire\n");
446         else if (tlb_type == hypervisor)
447                 seq_printf(m, "MMU Type\t: Hypervisor (sun4v)\n");
448         else
449                 seq_printf(m, "MMU Type\t: ???\n");
450
451 #ifdef CONFIG_DEBUG_DCFLUSH
452         seq_printf(m, "DCPageFlushes\t: %d\n",
453                    atomic_read(&dcpage_flushes));
454 #ifdef CONFIG_SMP
455         seq_printf(m, "DCPageFlushesXC\t: %d\n",
456                    atomic_read(&dcpage_flushes_xcall));
457 #endif /* CONFIG_SMP */
458 #endif /* CONFIG_DEBUG_DCFLUSH */
459 }
460
461 struct linux_prom_translation {
462         unsigned long virt;
463         unsigned long size;
464         unsigned long data;
465 };
466
467 /* Exported for kernel TLB miss handling in ktlb.S */
468 struct linux_prom_translation prom_trans[512] __read_mostly;
469 unsigned int prom_trans_ents __read_mostly;
470
471 /* Exported for SMP bootup purposes. */
472 unsigned long kern_locked_tte_data;
473
474 /* The obp translations are saved based on 8k pagesize, since obp can
475  * use a mixture of pagesizes. Misses to the LOW_OBP_ADDRESS ->
476  * HI_OBP_ADDRESS range are handled in ktlb.S.
477  */
478 static inline int in_obp_range(unsigned long vaddr)
479 {
480         return (vaddr >= LOW_OBP_ADDRESS &&
481                 vaddr < HI_OBP_ADDRESS);
482 }
483
484 static int cmp_ptrans(const void *a, const void *b)
485 {
486         const struct linux_prom_translation *x = a, *y = b;
487
488         if (x->virt > y->virt)
489                 return 1;
490         if (x->virt < y->virt)
491                 return -1;
492         return 0;
493 }
494
495 /* Read OBP translations property into 'prom_trans[]'.  */
496 static void __init read_obp_translations(void)
497 {
498         int n, node, ents, first, last, i;
499
500         node = prom_finddevice("/virtual-memory");
501         n = prom_getproplen(node, "translations");
502         if (unlikely(n == 0 || n == -1)) {
503                 prom_printf("prom_mappings: Couldn't get size.\n");
504                 prom_halt();
505         }
506         if (unlikely(n > sizeof(prom_trans))) {
507                 prom_printf("prom_mappings: Size %Zd is too big.\n", n);
508                 prom_halt();
509         }
510
511         if ((n = prom_getproperty(node, "translations",
512                                   (char *)&prom_trans[0],
513                                   sizeof(prom_trans))) == -1) {
514                 prom_printf("prom_mappings: Couldn't get property.\n");
515                 prom_halt();
516         }
517
518         n = n / sizeof(struct linux_prom_translation);
519
520         ents = n;
521
522         sort(prom_trans, ents, sizeof(struct linux_prom_translation),
523              cmp_ptrans, NULL);
524
525         /* Now kick out all the non-OBP entries.  */
526         for (i = 0; i < ents; i++) {
527                 if (in_obp_range(prom_trans[i].virt))
528                         break;
529         }
530         first = i;
531         for (; i < ents; i++) {
532                 if (!in_obp_range(prom_trans[i].virt))
533                         break;
534         }
535         last = i;
536
537         for (i = 0; i < (last - first); i++) {
538                 struct linux_prom_translation *src = &prom_trans[i + first];
539                 struct linux_prom_translation *dest = &prom_trans[i];
540
541                 *dest = *src;
542         }
543         for (; i < ents; i++) {
544                 struct linux_prom_translation *dest = &prom_trans[i];
545                 dest->virt = dest->size = dest->data = 0x0UL;
546         }
547
548         prom_trans_ents = last - first;
549
550         if (tlb_type == spitfire) {
551                 /* Clear diag TTE bits. */
552                 for (i = 0; i < prom_trans_ents; i++)
553                         prom_trans[i].data &= ~0x0003fe0000000000UL;
554         }
555 }
556
557 static void __init hypervisor_tlb_lock(unsigned long vaddr,
558                                        unsigned long pte,
559                                        unsigned long mmu)
560 {
561         unsigned long ret = sun4v_mmu_map_perm_addr(vaddr, 0, pte, mmu);
562
563         if (ret != 0) {
564                 prom_printf("hypervisor_tlb_lock[%lx:%lx:%lx:%lx]: "
565                             "errors with %lx\n", vaddr, 0, pte, mmu, ret);
566                 prom_halt();
567         }
568 }
569
570 static unsigned long kern_large_tte(unsigned long paddr);
571
572 static void __init remap_kernel(void)
573 {
574         unsigned long phys_page, tte_vaddr, tte_data;
575         int tlb_ent = sparc64_highest_locked_tlbent();
576
577         tte_vaddr = (unsigned long) KERNBASE;
578         phys_page = (prom_boot_mapping_phys_low >> 22UL) << 22UL;
579         tte_data = kern_large_tte(phys_page);
580
581         kern_locked_tte_data = tte_data;
582
583         /* Now lock us into the TLBs via Hypervisor or OBP. */
584         if (tlb_type == hypervisor) {
585                 hypervisor_tlb_lock(tte_vaddr, tte_data, HV_MMU_DMMU);
586                 hypervisor_tlb_lock(tte_vaddr, tte_data, HV_MMU_IMMU);
587                 if (bigkernel) {
588                         tte_vaddr += 0x400000;
589                         tte_data += 0x400000;
590                         hypervisor_tlb_lock(tte_vaddr, tte_data, HV_MMU_DMMU);
591                         hypervisor_tlb_lock(tte_vaddr, tte_data, HV_MMU_IMMU);
592                 }
593         } else {
594                 prom_dtlb_load(tlb_ent, tte_data, tte_vaddr);
595                 prom_itlb_load(tlb_ent, tte_data, tte_vaddr);
596                 if (bigkernel) {
597                         tlb_ent -= 1;
598                         prom_dtlb_load(tlb_ent,
599                                        tte_data + 0x400000, 
600                                        tte_vaddr + 0x400000);
601                         prom_itlb_load(tlb_ent,
602                                        tte_data + 0x400000, 
603                                        tte_vaddr + 0x400000);
604                 }
605                 sparc64_highest_unlocked_tlb_ent = tlb_ent - 1;
606         }
607         if (tlb_type == cheetah_plus) {
608                 sparc64_kern_pri_context = (CTX_CHEETAH_PLUS_CTX0 |
609                                             CTX_CHEETAH_PLUS_NUC);
610                 sparc64_kern_pri_nuc_bits = CTX_CHEETAH_PLUS_NUC;
611                 sparc64_kern_sec_context = CTX_CHEETAH_PLUS_CTX0;
612         }
613 }
614
615
616 static void __init inherit_prom_mappings(void)
617 {
618         read_obp_translations();
619
620         /* Now fixup OBP's idea about where we really are mapped. */
621         printk("Remapping the kernel... ");
622         remap_kernel();
623         printk("done.\n");
624 }
625
626 void prom_world(int enter)
627 {
628         if (!enter)
629                 set_fs((mm_segment_t) { get_thread_current_ds() });
630
631         __asm__ __volatile__("flushw");
632 }
633
634 void __flush_dcache_range(unsigned long start, unsigned long end)
635 {
636         unsigned long va;
637
638         if (tlb_type == spitfire) {
639                 int n = 0;
640
641                 for (va = start; va < end; va += 32) {
642                         spitfire_put_dcache_tag(va & 0x3fe0, 0x0);
643                         if (++n >= 512)
644                                 break;
645                 }
646         } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
647                 start = __pa(start);
648                 end = __pa(end);
649                 for (va = start; va < end; va += 32)
650                         __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
651                                              "membar #Sync"
652                                              : /* no outputs */
653                                              : "r" (va),
654                                                "i" (ASI_DCACHE_INVALIDATE));
655         }
656 }
657
658 /* get_new_mmu_context() uses "cache + 1".  */
659 DEFINE_SPINLOCK(ctx_alloc_lock);
660 unsigned long tlb_context_cache = CTX_FIRST_VERSION - 1;
661 #define MAX_CTX_NR      (1UL << CTX_NR_BITS)
662 #define CTX_BMAP_SLOTS  BITS_TO_LONGS(MAX_CTX_NR)
663 DECLARE_BITMAP(mmu_context_bmap, MAX_CTX_NR);
664
665 /* Caller does TLB context flushing on local CPU if necessary.
666  * The caller also ensures that CTX_VALID(mm->context) is false.
667  *
668  * We must be careful about boundary cases so that we never
669  * let the user have CTX 0 (nucleus) or we ever use a CTX
670  * version of zero (and thus NO_CONTEXT would not be caught
671  * by version mis-match tests in mmu_context.h).
672  *
673  * Always invoked with interrupts disabled.
674  */
675 void get_new_mmu_context(struct mm_struct *mm)
676 {
677         unsigned long ctx, new_ctx;
678         unsigned long orig_pgsz_bits;
679         unsigned long flags;
680         int new_version;
681
682         spin_lock_irqsave(&ctx_alloc_lock, flags);
683         orig_pgsz_bits = (mm->context.sparc64_ctx_val & CTX_PGSZ_MASK);
684         ctx = (tlb_context_cache + 1) & CTX_NR_MASK;
685         new_ctx = find_next_zero_bit(mmu_context_bmap, 1 << CTX_NR_BITS, ctx);
686         new_version = 0;
687         if (new_ctx >= (1 << CTX_NR_BITS)) {
688                 new_ctx = find_next_zero_bit(mmu_context_bmap, ctx, 1);
689                 if (new_ctx >= ctx) {
690                         int i;
691                         new_ctx = (tlb_context_cache & CTX_VERSION_MASK) +
692                                 CTX_FIRST_VERSION;
693                         if (new_ctx == 1)
694                                 new_ctx = CTX_FIRST_VERSION;
695
696                         /* Don't call memset, for 16 entries that's just
697                          * plain silly...
698                          */
699                         mmu_context_bmap[0] = 3;
700                         mmu_context_bmap[1] = 0;
701                         mmu_context_bmap[2] = 0;
702                         mmu_context_bmap[3] = 0;
703                         for (i = 4; i < CTX_BMAP_SLOTS; i += 4) {
704                                 mmu_context_bmap[i + 0] = 0;
705                                 mmu_context_bmap[i + 1] = 0;
706                                 mmu_context_bmap[i + 2] = 0;
707                                 mmu_context_bmap[i + 3] = 0;
708                         }
709                         new_version = 1;
710                         goto out;
711                 }
712         }
713         mmu_context_bmap[new_ctx>>6] |= (1UL << (new_ctx & 63));
714         new_ctx |= (tlb_context_cache & CTX_VERSION_MASK);
715 out:
716         tlb_context_cache = new_ctx;
717         mm->context.sparc64_ctx_val = new_ctx | orig_pgsz_bits;
718         spin_unlock_irqrestore(&ctx_alloc_lock, flags);
719
720         if (unlikely(new_version))
721                 smp_new_mmu_context_version();
722 }
723
724 /* Find a free area for the bootmem map, avoiding the kernel image
725  * and the initial ramdisk.
726  */
727 static unsigned long __init choose_bootmap_pfn(unsigned long start_pfn,
728                                                unsigned long end_pfn)
729 {
730         unsigned long avoid_start, avoid_end, bootmap_size;
731         int i;
732
733         bootmap_size = bootmem_bootmap_pages(end_pfn - start_pfn);
734         bootmap_size <<= PAGE_SHIFT;
735
736         avoid_start = avoid_end = 0;
737 #ifdef CONFIG_BLK_DEV_INITRD
738         avoid_start = initrd_start;
739         avoid_end = PAGE_ALIGN(initrd_end);
740 #endif
741
742         for (i = 0; i < pavail_ents; i++) {
743                 unsigned long start, end;
744
745                 start = pavail[i].phys_addr;
746                 end = start + pavail[i].reg_size;
747
748                 while (start < end) {
749                         if (start >= kern_base &&
750                             start < PAGE_ALIGN(kern_base + kern_size)) {
751                                 start = PAGE_ALIGN(kern_base + kern_size);
752                                 continue;
753                         }
754                         if (start >= avoid_start && start < avoid_end) {
755                                 start = avoid_end;
756                                 continue;
757                         }
758
759                         if ((end - start) < bootmap_size)
760                                 break;
761
762                         if (start < kern_base &&
763                             (start + bootmap_size) > kern_base) {
764                                 start = PAGE_ALIGN(kern_base + kern_size);
765                                 continue;
766                         }
767
768                         if (start < avoid_start &&
769                             (start + bootmap_size) > avoid_start) {
770                                 start = avoid_end;
771                                 continue;
772                         }
773
774                         /* OK, it doesn't overlap anything, use it.  */
775                         return start >> PAGE_SHIFT;
776                 }
777         }
778
779         prom_printf("Cannot find free area for bootmap, aborting.\n");
780         prom_halt();
781 }
782
783 static void __init trim_pavail(unsigned long *cur_size_p,
784                                unsigned long *end_of_phys_p)
785 {
786         unsigned long to_trim = *cur_size_p - cmdline_memory_size;
787         unsigned long avoid_start, avoid_end;
788         int i;
789
790         to_trim = PAGE_ALIGN(to_trim);
791
792         avoid_start = avoid_end = 0;
793 #ifdef CONFIG_BLK_DEV_INITRD
794         avoid_start = initrd_start;
795         avoid_end = PAGE_ALIGN(initrd_end);
796 #endif
797
798         /* Trim some pavail[] entries in order to satisfy the
799          * requested "mem=xxx" kernel command line specification.
800          *
801          * We must not trim off the kernel image area nor the
802          * initial ramdisk range (if any).  Also, we must not trim
803          * any pavail[] entry down to zero in order to preserve
804          * the invariant that all pavail[] entries have a non-zero
805          * size which is assumed by all of the code in here.
806          */
807         for (i = 0; i < pavail_ents; i++) {
808                 unsigned long start, end, kern_end;
809                 unsigned long trim_low, trim_high, n;
810
811                 kern_end = PAGE_ALIGN(kern_base + kern_size);
812
813                 trim_low = start = pavail[i].phys_addr;
814                 trim_high = end = start + pavail[i].reg_size;
815
816                 if (kern_base >= start &&
817                     kern_base < end) {
818                         trim_low = kern_base;
819                         if (kern_end >= end)
820                                 continue;
821                 }
822                 if (kern_end >= start &&
823                     kern_end < end) {
824                         trim_high = kern_end;
825                 }
826                 if (avoid_start &&
827                     avoid_start >= start &&
828                     avoid_start < end) {
829                         if (trim_low > avoid_start)
830                                 trim_low = avoid_start;
831                         if (avoid_end >= end)
832                                 continue;
833                 }
834                 if (avoid_end &&
835                     avoid_end >= start &&
836                     avoid_end < end) {
837                         if (trim_high < avoid_end)
838                                 trim_high = avoid_end;
839                 }
840
841                 if (trim_high <= trim_low)
842                         continue;
843
844                 if (trim_low == start && trim_high == end) {
845                         /* Whole chunk is available for trimming.
846                          * Trim all except one page, in order to keep
847                          * entry non-empty.
848                          */
849                         n = (end - start) - PAGE_SIZE;
850                         if (n > to_trim)
851                                 n = to_trim;
852
853                         if (n) {
854                                 pavail[i].phys_addr += n;
855                                 pavail[i].reg_size -= n;
856                                 to_trim -= n;
857                         }
858                 } else {
859                         n = (trim_low - start);
860                         if (n > to_trim)
861                                 n = to_trim;
862
863                         if (n) {
864                                 pavail[i].phys_addr += n;
865                                 pavail[i].reg_size -= n;
866                                 to_trim -= n;
867                         }
868                         if (to_trim) {
869                                 n = end - trim_high;
870                                 if (n > to_trim)
871                                         n = to_trim;
872                                 if (n) {
873                                         pavail[i].reg_size -= n;
874                                         to_trim -= n;
875                                 }
876                         }
877                 }
878
879                 if (!to_trim)
880                         break;
881         }
882
883         /* Recalculate.  */
884         *cur_size_p = 0UL;
885         for (i = 0; i < pavail_ents; i++) {
886                 *end_of_phys_p = pavail[i].phys_addr +
887                         pavail[i].reg_size;
888                 *cur_size_p += pavail[i].reg_size;
889         }
890 }
891
892 /* About pages_avail, this is the value we will use to calculate
893  * the zholes_size[] argument given to free_area_init_node().  The
894  * page allocator uses this to calculate nr_kernel_pages,
895  * nr_all_pages and zone->present_pages.  On NUMA it is used
896  * to calculate zone->min_unmapped_pages and zone->min_slab_pages.
897  *
898  * So this number should really be set to what the page allocator
899  * actually ends up with.  This means:
900  * 1) It should include bootmem map pages, we'll release those.
901  * 2) It should not include the kernel image, except for the
902  *    __init sections which we will also release.
903  * 3) It should include the initrd image, since we'll release
904  *    that too.
905  */
906 static unsigned long __init bootmem_init(unsigned long *pages_avail,
907                                          unsigned long phys_base)
908 {
909         unsigned long bootmap_size, end_pfn;
910         unsigned long end_of_phys_memory = 0UL;
911         unsigned long bootmap_pfn, bytes_avail, size;
912         int i;
913
914         bytes_avail = 0UL;
915         for (i = 0; i < pavail_ents; i++) {
916                 end_of_phys_memory = pavail[i].phys_addr +
917                         pavail[i].reg_size;
918                 bytes_avail += pavail[i].reg_size;
919         }
920
921         /* Determine the location of the initial ramdisk before trying
922          * to honor the "mem=xxx" command line argument.  We must know
923          * where the kernel image and the ramdisk image are so that we
924          * do not trim those two areas from the physical memory map.
925          */
926
927 #ifdef CONFIG_BLK_DEV_INITRD
928         /* Now have to check initial ramdisk, so that bootmap does not overwrite it */
929         if (sparc_ramdisk_image || sparc_ramdisk_image64) {
930                 unsigned long ramdisk_image = sparc_ramdisk_image ?
931                         sparc_ramdisk_image : sparc_ramdisk_image64;
932                 ramdisk_image -= KERNBASE;
933                 initrd_start = ramdisk_image + phys_base;
934                 initrd_end = initrd_start + sparc_ramdisk_size;
935                 if (initrd_end > end_of_phys_memory) {
936                         printk(KERN_CRIT "initrd extends beyond end of memory "
937                                          "(0x%016lx > 0x%016lx)\ndisabling initrd\n",
938                                initrd_end, end_of_phys_memory);
939                         initrd_start = 0;
940                         initrd_end = 0;
941                 }
942         }
943 #endif  
944
945         if (cmdline_memory_size &&
946             bytes_avail > cmdline_memory_size)
947                 trim_pavail(&bytes_avail,
948                             &end_of_phys_memory);
949
950         *pages_avail = bytes_avail >> PAGE_SHIFT;
951
952         end_pfn = end_of_phys_memory >> PAGE_SHIFT;
953
954         /* Initialize the boot-time allocator. */
955         max_pfn = max_low_pfn = end_pfn;
956         min_low_pfn = (phys_base >> PAGE_SHIFT);
957
958         bootmap_pfn = choose_bootmap_pfn(min_low_pfn, end_pfn);
959
960         bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap_pfn,
961                                          min_low_pfn, end_pfn);
962
963         /* Now register the available physical memory with the
964          * allocator.
965          */
966         for (i = 0; i < pavail_ents; i++)
967                 free_bootmem(pavail[i].phys_addr, pavail[i].reg_size);
968
969 #ifdef CONFIG_BLK_DEV_INITRD
970         if (initrd_start) {
971                 size = initrd_end - initrd_start;
972
973                 /* Reserve the initrd image area. */
974                 reserve_bootmem(initrd_start, size, BOOTMEM_DEFAULT);
975
976                 initrd_start += PAGE_OFFSET;
977                 initrd_end += PAGE_OFFSET;
978         }
979 #endif
980         /* Reserve the kernel text/data/bss. */
981         reserve_bootmem(kern_base, kern_size, BOOTMEM_DEFAULT);
982         *pages_avail -= PAGE_ALIGN(kern_size) >> PAGE_SHIFT;
983
984         /* Add back in the initmem pages. */
985         size = ((unsigned long)(__init_end) & PAGE_MASK) -
986                 PAGE_ALIGN((unsigned long)__init_begin);
987         *pages_avail += size >> PAGE_SHIFT;
988
989         /* Reserve the bootmem map.   We do not account for it
990          * in pages_avail because we will release that memory
991          * in free_all_bootmem.
992          */
993         size = bootmap_size;
994         reserve_bootmem((bootmap_pfn << PAGE_SHIFT), size, BOOTMEM_DEFAULT);
995
996         for (i = 0; i < pavail_ents; i++) {
997                 unsigned long start_pfn, end_pfn;
998
999                 start_pfn = pavail[i].phys_addr >> PAGE_SHIFT;
1000                 end_pfn = (start_pfn + (pavail[i].reg_size >> PAGE_SHIFT));
1001                 memory_present(0, start_pfn, end_pfn);
1002         }
1003
1004         sparse_init();
1005
1006         return end_pfn;
1007 }
1008
1009 static struct linux_prom64_registers pall[MAX_BANKS] __initdata;
1010 static int pall_ents __initdata;
1011
1012 #ifdef CONFIG_DEBUG_PAGEALLOC
1013 static unsigned long __ref kernel_map_range(unsigned long pstart,
1014                                             unsigned long pend, pgprot_t prot)
1015 {
1016         unsigned long vstart = PAGE_OFFSET + pstart;
1017         unsigned long vend = PAGE_OFFSET + pend;
1018         unsigned long alloc_bytes = 0UL;
1019
1020         if ((vstart & ~PAGE_MASK) || (vend & ~PAGE_MASK)) {
1021                 prom_printf("kernel_map: Unaligned physmem[%lx:%lx]\n",
1022                             vstart, vend);
1023                 prom_halt();
1024         }
1025
1026         while (vstart < vend) {
1027                 unsigned long this_end, paddr = __pa(vstart);
1028                 pgd_t *pgd = pgd_offset_k(vstart);
1029                 pud_t *pud;
1030                 pmd_t *pmd;
1031                 pte_t *pte;
1032
1033                 pud = pud_offset(pgd, vstart);
1034                 if (pud_none(*pud)) {
1035                         pmd_t *new;
1036
1037                         new = __alloc_bootmem(PAGE_SIZE, PAGE_SIZE, PAGE_SIZE);
1038                         alloc_bytes += PAGE_SIZE;
1039                         pud_populate(&init_mm, pud, new);
1040                 }
1041
1042                 pmd = pmd_offset(pud, vstart);
1043                 if (!pmd_present(*pmd)) {
1044                         pte_t *new;
1045
1046                         new = __alloc_bootmem(PAGE_SIZE, PAGE_SIZE, PAGE_SIZE);
1047                         alloc_bytes += PAGE_SIZE;
1048                         pmd_populate_kernel(&init_mm, pmd, new);
1049                 }
1050
1051                 pte = pte_offset_kernel(pmd, vstart);
1052                 this_end = (vstart + PMD_SIZE) & PMD_MASK;
1053                 if (this_end > vend)
1054                         this_end = vend;
1055
1056                 while (vstart < this_end) {
1057                         pte_val(*pte) = (paddr | pgprot_val(prot));
1058
1059                         vstart += PAGE_SIZE;
1060                         paddr += PAGE_SIZE;
1061                         pte++;
1062                 }
1063         }
1064
1065         return alloc_bytes;
1066 }
1067
1068 extern unsigned int kvmap_linear_patch[1];
1069 #endif /* CONFIG_DEBUG_PAGEALLOC */
1070
1071 static void __init mark_kpte_bitmap(unsigned long start, unsigned long end)
1072 {
1073         const unsigned long shift_256MB = 28;
1074         const unsigned long mask_256MB = ((1UL << shift_256MB) - 1UL);
1075         const unsigned long size_256MB = (1UL << shift_256MB);
1076
1077         while (start < end) {
1078                 long remains;
1079
1080                 remains = end - start;
1081                 if (remains < size_256MB)
1082                         break;
1083
1084                 if (start & mask_256MB) {
1085                         start = (start + size_256MB) & ~mask_256MB;
1086                         continue;
1087                 }
1088
1089                 while (remains >= size_256MB) {
1090                         unsigned long index = start >> shift_256MB;
1091
1092                         __set_bit(index, kpte_linear_bitmap);
1093
1094                         start += size_256MB;
1095                         remains -= size_256MB;
1096                 }
1097         }
1098 }
1099
1100 static void __init init_kpte_bitmap(void)
1101 {
1102         unsigned long i;
1103
1104         for (i = 0; i < pall_ents; i++) {
1105                 unsigned long phys_start, phys_end;
1106
1107                 phys_start = pall[i].phys_addr;
1108                 phys_end = phys_start + pall[i].reg_size;
1109
1110                 mark_kpte_bitmap(phys_start, phys_end);
1111         }
1112 }
1113
1114 static void __init kernel_physical_mapping_init(void)
1115 {
1116 #ifdef CONFIG_DEBUG_PAGEALLOC
1117         unsigned long i, mem_alloced = 0UL;
1118
1119         for (i = 0; i < pall_ents; i++) {
1120                 unsigned long phys_start, phys_end;
1121
1122                 phys_start = pall[i].phys_addr;
1123                 phys_end = phys_start + pall[i].reg_size;
1124
1125                 mem_alloced += kernel_map_range(phys_start, phys_end,
1126                                                 PAGE_KERNEL);
1127         }
1128
1129         printk("Allocated %ld bytes for kernel page tables.\n",
1130                mem_alloced);
1131
1132         kvmap_linear_patch[0] = 0x01000000; /* nop */
1133         flushi(&kvmap_linear_patch[0]);
1134
1135         __flush_tlb_all();
1136 #endif
1137 }
1138
1139 #ifdef CONFIG_DEBUG_PAGEALLOC
1140 void kernel_map_pages(struct page *page, int numpages, int enable)
1141 {
1142         unsigned long phys_start = page_to_pfn(page) << PAGE_SHIFT;
1143         unsigned long phys_end = phys_start + (numpages * PAGE_SIZE);
1144
1145         kernel_map_range(phys_start, phys_end,
1146                          (enable ? PAGE_KERNEL : __pgprot(0)));
1147
1148         flush_tsb_kernel_range(PAGE_OFFSET + phys_start,
1149                                PAGE_OFFSET + phys_end);
1150
1151         /* we should perform an IPI and flush all tlbs,
1152          * but that can deadlock->flush only current cpu.
1153          */
1154         __flush_tlb_kernel_range(PAGE_OFFSET + phys_start,
1155                                  PAGE_OFFSET + phys_end);
1156 }
1157 #endif
1158
1159 unsigned long __init find_ecache_flush_span(unsigned long size)
1160 {
1161         int i;
1162
1163         for (i = 0; i < pavail_ents; i++) {
1164                 if (pavail[i].reg_size >= size)
1165                         return pavail[i].phys_addr;
1166         }
1167
1168         return ~0UL;
1169 }
1170
1171 static void __init tsb_phys_patch(void)
1172 {
1173         struct tsb_ldquad_phys_patch_entry *pquad;
1174         struct tsb_phys_patch_entry *p;
1175
1176         pquad = &__tsb_ldquad_phys_patch;
1177         while (pquad < &__tsb_ldquad_phys_patch_end) {
1178                 unsigned long addr = pquad->addr;
1179
1180                 if (tlb_type == hypervisor)
1181                         *(unsigned int *) addr = pquad->sun4v_insn;
1182                 else
1183                         *(unsigned int *) addr = pquad->sun4u_insn;
1184                 wmb();
1185                 __asm__ __volatile__("flush     %0"
1186                                      : /* no outputs */
1187                                      : "r" (addr));
1188
1189                 pquad++;
1190         }
1191
1192         p = &__tsb_phys_patch;
1193         while (p < &__tsb_phys_patch_end) {
1194                 unsigned long addr = p->addr;
1195
1196                 *(unsigned int *) addr = p->insn;
1197                 wmb();
1198                 __asm__ __volatile__("flush     %0"
1199                                      : /* no outputs */
1200                                      : "r" (addr));
1201
1202                 p++;
1203         }
1204 }
1205
1206 /* Don't mark as init, we give this to the Hypervisor.  */
1207 #ifndef CONFIG_DEBUG_PAGEALLOC
1208 #define NUM_KTSB_DESCR  2
1209 #else
1210 #define NUM_KTSB_DESCR  1
1211 #endif
1212 static struct hv_tsb_descr ktsb_descr[NUM_KTSB_DESCR];
1213 extern struct tsb swapper_tsb[KERNEL_TSB_NENTRIES];
1214
1215 static void __init sun4v_ktsb_init(void)
1216 {
1217         unsigned long ktsb_pa;
1218
1219         /* First KTSB for PAGE_SIZE mappings.  */
1220         ktsb_pa = kern_base + ((unsigned long)&swapper_tsb[0] - KERNBASE);
1221
1222         switch (PAGE_SIZE) {
1223         case 8 * 1024:
1224         default:
1225                 ktsb_descr[0].pgsz_idx = HV_PGSZ_IDX_8K;
1226                 ktsb_descr[0].pgsz_mask = HV_PGSZ_MASK_8K;
1227                 break;
1228
1229         case 64 * 1024:
1230                 ktsb_descr[0].pgsz_idx = HV_PGSZ_IDX_64K;
1231                 ktsb_descr[0].pgsz_mask = HV_PGSZ_MASK_64K;
1232                 break;
1233
1234         case 512 * 1024:
1235                 ktsb_descr[0].pgsz_idx = HV_PGSZ_IDX_512K;
1236                 ktsb_descr[0].pgsz_mask = HV_PGSZ_MASK_512K;
1237                 break;
1238
1239         case 4 * 1024 * 1024:
1240                 ktsb_descr[0].pgsz_idx = HV_PGSZ_IDX_4MB;
1241                 ktsb_descr[0].pgsz_mask = HV_PGSZ_MASK_4MB;
1242                 break;
1243         };
1244
1245         ktsb_descr[0].assoc = 1;
1246         ktsb_descr[0].num_ttes = KERNEL_TSB_NENTRIES;
1247         ktsb_descr[0].ctx_idx = 0;
1248         ktsb_descr[0].tsb_base = ktsb_pa;
1249         ktsb_descr[0].resv = 0;
1250
1251 #ifndef CONFIG_DEBUG_PAGEALLOC
1252         /* Second KTSB for 4MB/256MB mappings.  */
1253         ktsb_pa = (kern_base +
1254                    ((unsigned long)&swapper_4m_tsb[0] - KERNBASE));
1255
1256         ktsb_descr[1].pgsz_idx = HV_PGSZ_IDX_4MB;
1257         ktsb_descr[1].pgsz_mask = (HV_PGSZ_MASK_4MB |
1258                                    HV_PGSZ_MASK_256MB);
1259         ktsb_descr[1].assoc = 1;
1260         ktsb_descr[1].num_ttes = KERNEL_TSB4M_NENTRIES;
1261         ktsb_descr[1].ctx_idx = 0;
1262         ktsb_descr[1].tsb_base = ktsb_pa;
1263         ktsb_descr[1].resv = 0;
1264 #endif
1265 }
1266
1267 void __cpuinit sun4v_ktsb_register(void)
1268 {
1269         unsigned long pa, ret;
1270
1271         pa = kern_base + ((unsigned long)&ktsb_descr[0] - KERNBASE);
1272
1273         ret = sun4v_mmu_tsb_ctx0(NUM_KTSB_DESCR, pa);
1274         if (ret != 0) {
1275                 prom_printf("hypervisor_mmu_tsb_ctx0[%lx]: "
1276                             "errors with %lx\n", pa, ret);
1277                 prom_halt();
1278         }
1279 }
1280
1281 /* paging_init() sets up the page tables */
1282
1283 extern void cheetah_ecache_flush_init(void);
1284 extern void sun4v_patch_tlb_handlers(void);
1285
1286 extern void cpu_probe(void);
1287 extern void central_probe(void);
1288
1289 static unsigned long last_valid_pfn;
1290 pgd_t swapper_pg_dir[2048];
1291
1292 static void sun4u_pgprot_init(void);
1293 static void sun4v_pgprot_init(void);
1294
1295 /* Dummy function */
1296 void __init setup_per_cpu_areas(void)
1297 {
1298 }
1299
1300 void __init paging_init(void)
1301 {
1302         unsigned long end_pfn, pages_avail, shift, phys_base;
1303         unsigned long real_end, i;
1304
1305         /* These build time checkes make sure that the dcache_dirty_cpu()
1306          * page->flags usage will work.
1307          *
1308          * When a page gets marked as dcache-dirty, we store the
1309          * cpu number starting at bit 32 in the page->flags.  Also,
1310          * functions like clear_dcache_dirty_cpu use the cpu mask
1311          * in 13-bit signed-immediate instruction fields.
1312          */
1313         BUILD_BUG_ON(FLAGS_RESERVED != 32);
1314         BUILD_BUG_ON(SECTIONS_WIDTH + NODES_WIDTH + ZONES_WIDTH +
1315                      ilog2(roundup_pow_of_two(NR_CPUS)) > FLAGS_RESERVED);
1316         BUILD_BUG_ON(NR_CPUS > 4096);
1317
1318         kern_base = (prom_boot_mapping_phys_low >> 22UL) << 22UL;
1319         kern_size = (unsigned long)&_end - (unsigned long)KERNBASE;
1320
1321         sstate_booting();
1322
1323         /* Invalidate both kernel TSBs.  */
1324         memset(swapper_tsb, 0x40, sizeof(swapper_tsb));
1325 #ifndef CONFIG_DEBUG_PAGEALLOC
1326         memset(swapper_4m_tsb, 0x40, sizeof(swapper_4m_tsb));
1327 #endif
1328
1329         if (tlb_type == hypervisor)
1330                 sun4v_pgprot_init();
1331         else
1332                 sun4u_pgprot_init();
1333
1334         if (tlb_type == cheetah_plus ||
1335             tlb_type == hypervisor)
1336                 tsb_phys_patch();
1337
1338         if (tlb_type == hypervisor) {
1339                 sun4v_patch_tlb_handlers();
1340                 sun4v_ktsb_init();
1341         }
1342
1343         /* Find available physical memory... */
1344         read_obp_memory("available", &pavail[0], &pavail_ents);
1345
1346         phys_base = 0xffffffffffffffffUL;
1347         for (i = 0; i < pavail_ents; i++)
1348                 phys_base = min(phys_base, pavail[i].phys_addr);
1349
1350         set_bit(0, mmu_context_bmap);
1351
1352         shift = kern_base + PAGE_OFFSET - ((unsigned long)KERNBASE);
1353
1354         real_end = (unsigned long)_end;
1355         if ((real_end > ((unsigned long)KERNBASE + 0x400000)))
1356                 bigkernel = 1;
1357         if ((real_end > ((unsigned long)KERNBASE + 0x800000))) {
1358                 prom_printf("paging_init: Kernel > 8MB, too large.\n");
1359                 prom_halt();
1360         }
1361
1362         /* Set kernel pgd to upper alias so physical page computations
1363          * work.
1364          */
1365         init_mm.pgd += ((shift) / (sizeof(pgd_t)));
1366         
1367         memset(swapper_low_pmd_dir, 0, sizeof(swapper_low_pmd_dir));
1368
1369         /* Now can init the kernel/bad page tables. */
1370         pud_set(pud_offset(&swapper_pg_dir[0], 0),
1371                 swapper_low_pmd_dir + (shift / sizeof(pgd_t)));
1372         
1373         inherit_prom_mappings();
1374         
1375         read_obp_memory("reg", &pall[0], &pall_ents);
1376
1377         init_kpte_bitmap();
1378
1379         /* Ok, we can use our TLB miss and window trap handlers safely.  */
1380         setup_tba();
1381
1382         __flush_tlb_all();
1383
1384         if (tlb_type == hypervisor)
1385                 sun4v_ktsb_register();
1386
1387         /* Setup bootmem... */
1388         pages_avail = 0;
1389         last_valid_pfn = end_pfn = bootmem_init(&pages_avail, phys_base);
1390
1391         max_mapnr = last_valid_pfn;
1392
1393         kernel_physical_mapping_init();
1394
1395         real_setup_per_cpu_areas();
1396
1397         prom_build_devicetree();
1398
1399         if (tlb_type == hypervisor)
1400                 sun4v_mdesc_init();
1401
1402         {
1403                 unsigned long zones_size[MAX_NR_ZONES];
1404                 unsigned long zholes_size[MAX_NR_ZONES];
1405                 int znum;
1406
1407                 for (znum = 0; znum < MAX_NR_ZONES; znum++)
1408                         zones_size[znum] = zholes_size[znum] = 0;
1409
1410                 zones_size[ZONE_NORMAL] = end_pfn;
1411                 zholes_size[ZONE_NORMAL] = end_pfn - pages_avail;
1412
1413                 free_area_init_node(0, &contig_page_data, zones_size,
1414                                     __pa(PAGE_OFFSET) >> PAGE_SHIFT,
1415                                     zholes_size);
1416         }
1417
1418         printk("Booting Linux...\n");
1419
1420         central_probe();
1421         cpu_probe();
1422 }
1423
1424 static void __init taint_real_pages(void)
1425 {
1426         int i;
1427
1428         read_obp_memory("available", &pavail_rescan[0], &pavail_rescan_ents);
1429
1430         /* Find changes discovered in the physmem available rescan and
1431          * reserve the lost portions in the bootmem maps.
1432          */
1433         for (i = 0; i < pavail_ents; i++) {
1434                 unsigned long old_start, old_end;
1435
1436                 old_start = pavail[i].phys_addr;
1437                 old_end = old_start +
1438                         pavail[i].reg_size;
1439                 while (old_start < old_end) {
1440                         int n;
1441
1442                         for (n = 0; n < pavail_rescan_ents; n++) {
1443                                 unsigned long new_start, new_end;
1444
1445                                 new_start = pavail_rescan[n].phys_addr;
1446                                 new_end = new_start +
1447                                         pavail_rescan[n].reg_size;
1448
1449                                 if (new_start <= old_start &&
1450                                     new_end >= (old_start + PAGE_SIZE)) {
1451                                         set_bit(old_start >> 22,
1452                                                 sparc64_valid_addr_bitmap);
1453                                         goto do_next_page;
1454                                 }
1455                         }
1456                         reserve_bootmem(old_start, PAGE_SIZE, BOOTMEM_DEFAULT);
1457
1458                 do_next_page:
1459                         old_start += PAGE_SIZE;
1460                 }
1461         }
1462 }
1463
1464 int __init page_in_phys_avail(unsigned long paddr)
1465 {
1466         int i;
1467
1468         paddr &= PAGE_MASK;
1469
1470         for (i = 0; i < pavail_rescan_ents; i++) {
1471                 unsigned long start, end;
1472
1473                 start = pavail_rescan[i].phys_addr;
1474                 end = start + pavail_rescan[i].reg_size;
1475
1476                 if (paddr >= start && paddr < end)
1477                         return 1;
1478         }
1479         if (paddr >= kern_base && paddr < (kern_base + kern_size))
1480                 return 1;
1481 #ifdef CONFIG_BLK_DEV_INITRD
1482         if (paddr >= __pa(initrd_start) &&
1483             paddr < __pa(PAGE_ALIGN(initrd_end)))
1484                 return 1;
1485 #endif
1486
1487         return 0;
1488 }
1489
1490 void __init mem_init(void)
1491 {
1492         unsigned long codepages, datapages, initpages;
1493         unsigned long addr, last;
1494         int i;
1495
1496         i = last_valid_pfn >> ((22 - PAGE_SHIFT) + 6);
1497         i += 1;
1498         sparc64_valid_addr_bitmap = (unsigned long *) alloc_bootmem(i << 3);
1499         if (sparc64_valid_addr_bitmap == NULL) {
1500                 prom_printf("mem_init: Cannot alloc valid_addr_bitmap.\n");
1501                 prom_halt();
1502         }
1503         memset(sparc64_valid_addr_bitmap, 0, i << 3);
1504
1505         addr = PAGE_OFFSET + kern_base;
1506         last = PAGE_ALIGN(kern_size) + addr;
1507         while (addr < last) {
1508                 set_bit(__pa(addr) >> 22, sparc64_valid_addr_bitmap);
1509                 addr += PAGE_SIZE;
1510         }
1511
1512         taint_real_pages();
1513
1514         high_memory = __va(last_valid_pfn << PAGE_SHIFT);
1515
1516         /* We subtract one to account for the mem_map_zero page
1517          * allocated below.
1518          */
1519         totalram_pages = num_physpages = free_all_bootmem() - 1;
1520
1521         /*
1522          * Set up the zero page, mark it reserved, so that page count
1523          * is not manipulated when freeing the page from user ptes.
1524          */
1525         mem_map_zero = alloc_pages(GFP_KERNEL|__GFP_ZERO, 0);
1526         if (mem_map_zero == NULL) {
1527                 prom_printf("paging_init: Cannot alloc zero page.\n");
1528                 prom_halt();
1529         }
1530         SetPageReserved(mem_map_zero);
1531
1532         codepages = (((unsigned long) _etext) - ((unsigned long) _start));
1533         codepages = PAGE_ALIGN(codepages) >> PAGE_SHIFT;
1534         datapages = (((unsigned long) _edata) - ((unsigned long) _etext));
1535         datapages = PAGE_ALIGN(datapages) >> PAGE_SHIFT;
1536         initpages = (((unsigned long) __init_end) - ((unsigned long) __init_begin));
1537         initpages = PAGE_ALIGN(initpages) >> PAGE_SHIFT;
1538
1539         printk("Memory: %luk available (%ldk kernel code, %ldk data, %ldk init) [%016lx,%016lx]\n",
1540                nr_free_pages() << (PAGE_SHIFT-10),
1541                codepages << (PAGE_SHIFT-10),
1542                datapages << (PAGE_SHIFT-10), 
1543                initpages << (PAGE_SHIFT-10), 
1544                PAGE_OFFSET, (last_valid_pfn << PAGE_SHIFT));
1545
1546         if (tlb_type == cheetah || tlb_type == cheetah_plus)
1547                 cheetah_ecache_flush_init();
1548 }
1549
1550 void free_initmem(void)
1551 {
1552         unsigned long addr, initend;
1553
1554         /*
1555          * The init section is aligned to 8k in vmlinux.lds. Page align for >8k pagesizes.
1556          */
1557         addr = PAGE_ALIGN((unsigned long)(__init_begin));
1558         initend = (unsigned long)(__init_end) & PAGE_MASK;
1559         for (; addr < initend; addr += PAGE_SIZE) {
1560                 unsigned long page;
1561                 struct page *p;
1562
1563                 page = (addr +
1564                         ((unsigned long) __va(kern_base)) -
1565                         ((unsigned long) KERNBASE));
1566                 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
1567                 p = virt_to_page(page);
1568
1569                 ClearPageReserved(p);
1570                 init_page_count(p);
1571                 __free_page(p);
1572                 num_physpages++;
1573                 totalram_pages++;
1574         }
1575 }
1576
1577 #ifdef CONFIG_BLK_DEV_INITRD
1578 void free_initrd_mem(unsigned long start, unsigned long end)
1579 {
1580         if (start < end)
1581                 printk ("Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
1582         for (; start < end; start += PAGE_SIZE) {
1583                 struct page *p = virt_to_page(start);
1584
1585                 ClearPageReserved(p);
1586                 init_page_count(p);
1587                 __free_page(p);
1588                 num_physpages++;
1589                 totalram_pages++;
1590         }
1591 }
1592 #endif
1593
1594 #define _PAGE_CACHE_4U  (_PAGE_CP_4U | _PAGE_CV_4U)
1595 #define _PAGE_CACHE_4V  (_PAGE_CP_4V | _PAGE_CV_4V)
1596 #define __DIRTY_BITS_4U  (_PAGE_MODIFIED_4U | _PAGE_WRITE_4U | _PAGE_W_4U)
1597 #define __DIRTY_BITS_4V  (_PAGE_MODIFIED_4V | _PAGE_WRITE_4V | _PAGE_W_4V)
1598 #define __ACCESS_BITS_4U (_PAGE_ACCESSED_4U | _PAGE_READ_4U | _PAGE_R)
1599 #define __ACCESS_BITS_4V (_PAGE_ACCESSED_4V | _PAGE_READ_4V | _PAGE_R)
1600
1601 pgprot_t PAGE_KERNEL __read_mostly;
1602 EXPORT_SYMBOL(PAGE_KERNEL);
1603
1604 pgprot_t PAGE_KERNEL_LOCKED __read_mostly;
1605 pgprot_t PAGE_COPY __read_mostly;
1606
1607 pgprot_t PAGE_SHARED __read_mostly;
1608 EXPORT_SYMBOL(PAGE_SHARED);
1609
1610 pgprot_t PAGE_EXEC __read_mostly;
1611 unsigned long pg_iobits __read_mostly;
1612
1613 unsigned long _PAGE_IE __read_mostly;
1614 EXPORT_SYMBOL(_PAGE_IE);
1615
1616 unsigned long _PAGE_E __read_mostly;
1617 EXPORT_SYMBOL(_PAGE_E);
1618
1619 unsigned long _PAGE_CACHE __read_mostly;
1620 EXPORT_SYMBOL(_PAGE_CACHE);
1621
1622 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1623
1624 #define VMEMMAP_CHUNK_SHIFT     22
1625 #define VMEMMAP_CHUNK           (1UL << VMEMMAP_CHUNK_SHIFT)
1626 #define VMEMMAP_CHUNK_MASK      ~(VMEMMAP_CHUNK - 1UL)
1627 #define VMEMMAP_ALIGN(x)        (((x)+VMEMMAP_CHUNK-1UL)&VMEMMAP_CHUNK_MASK)
1628
1629 #define VMEMMAP_SIZE    ((((1UL << MAX_PHYSADDR_BITS) >> PAGE_SHIFT) * \
1630                           sizeof(struct page *)) >> VMEMMAP_CHUNK_SHIFT)
1631 unsigned long vmemmap_table[VMEMMAP_SIZE];
1632
1633 int __meminit vmemmap_populate(struct page *start, unsigned long nr, int node)
1634 {
1635         unsigned long vstart = (unsigned long) start;
1636         unsigned long vend = (unsigned long) (start + nr);
1637         unsigned long phys_start = (vstart - VMEMMAP_BASE);
1638         unsigned long phys_end = (vend - VMEMMAP_BASE);
1639         unsigned long addr = phys_start & VMEMMAP_CHUNK_MASK;
1640         unsigned long end = VMEMMAP_ALIGN(phys_end);
1641         unsigned long pte_base;
1642
1643         pte_base = (_PAGE_VALID | _PAGE_SZ4MB_4U |
1644                     _PAGE_CP_4U | _PAGE_CV_4U |
1645                     _PAGE_P_4U | _PAGE_W_4U);
1646         if (tlb_type == hypervisor)
1647                 pte_base = (_PAGE_VALID | _PAGE_SZ4MB_4V |
1648                             _PAGE_CP_4V | _PAGE_CV_4V |
1649                             _PAGE_P_4V | _PAGE_W_4V);
1650
1651         for (; addr < end; addr += VMEMMAP_CHUNK) {
1652                 unsigned long *vmem_pp =
1653                         vmemmap_table + (addr >> VMEMMAP_CHUNK_SHIFT);
1654                 void *block;
1655
1656                 if (!(*vmem_pp & _PAGE_VALID)) {
1657                         block = vmemmap_alloc_block(1UL << 22, node);
1658                         if (!block)
1659                                 return -ENOMEM;
1660
1661                         *vmem_pp = pte_base | __pa(block);
1662
1663                         printk(KERN_INFO "[%p-%p] page_structs=%lu "
1664                                "node=%d entry=%lu/%lu\n", start, block, nr,
1665                                node,
1666                                addr >> VMEMMAP_CHUNK_SHIFT,
1667                                VMEMMAP_SIZE >> VMEMMAP_CHUNK_SHIFT);
1668                 }
1669         }
1670         return 0;
1671 }
1672 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
1673
1674 static void prot_init_common(unsigned long page_none,
1675                              unsigned long page_shared,
1676                              unsigned long page_copy,
1677                              unsigned long page_readonly,
1678                              unsigned long page_exec_bit)
1679 {
1680         PAGE_COPY = __pgprot(page_copy);
1681         PAGE_SHARED = __pgprot(page_shared);
1682
1683         protection_map[0x0] = __pgprot(page_none);
1684         protection_map[0x1] = __pgprot(page_readonly & ~page_exec_bit);
1685         protection_map[0x2] = __pgprot(page_copy & ~page_exec_bit);
1686         protection_map[0x3] = __pgprot(page_copy & ~page_exec_bit);
1687         protection_map[0x4] = __pgprot(page_readonly);
1688         protection_map[0x5] = __pgprot(page_readonly);
1689         protection_map[0x6] = __pgprot(page_copy);
1690         protection_map[0x7] = __pgprot(page_copy);
1691         protection_map[0x8] = __pgprot(page_none);
1692         protection_map[0x9] = __pgprot(page_readonly & ~page_exec_bit);
1693         protection_map[0xa] = __pgprot(page_shared & ~page_exec_bit);
1694         protection_map[0xb] = __pgprot(page_shared & ~page_exec_bit);
1695         protection_map[0xc] = __pgprot(page_readonly);
1696         protection_map[0xd] = __pgprot(page_readonly);
1697         protection_map[0xe] = __pgprot(page_shared);
1698         protection_map[0xf] = __pgprot(page_shared);
1699 }
1700
1701 static void __init sun4u_pgprot_init(void)
1702 {
1703         unsigned long page_none, page_shared, page_copy, page_readonly;
1704         unsigned long page_exec_bit;
1705
1706         PAGE_KERNEL = __pgprot (_PAGE_PRESENT_4U | _PAGE_VALID |
1707                                 _PAGE_CACHE_4U | _PAGE_P_4U |
1708                                 __ACCESS_BITS_4U | __DIRTY_BITS_4U |
1709                                 _PAGE_EXEC_4U);
1710         PAGE_KERNEL_LOCKED = __pgprot (_PAGE_PRESENT_4U | _PAGE_VALID |
1711                                        _PAGE_CACHE_4U | _PAGE_P_4U |
1712                                        __ACCESS_BITS_4U | __DIRTY_BITS_4U |
1713                                        _PAGE_EXEC_4U | _PAGE_L_4U);
1714         PAGE_EXEC = __pgprot(_PAGE_EXEC_4U);
1715
1716         _PAGE_IE = _PAGE_IE_4U;
1717         _PAGE_E = _PAGE_E_4U;
1718         _PAGE_CACHE = _PAGE_CACHE_4U;
1719
1720         pg_iobits = (_PAGE_VALID | _PAGE_PRESENT_4U | __DIRTY_BITS_4U |
1721                      __ACCESS_BITS_4U | _PAGE_E_4U);
1722
1723 #ifdef CONFIG_DEBUG_PAGEALLOC
1724         kern_linear_pte_xor[0] = (_PAGE_VALID | _PAGE_SZBITS_4U) ^
1725                 0xfffff80000000000;
1726 #else
1727         kern_linear_pte_xor[0] = (_PAGE_VALID | _PAGE_SZ4MB_4U) ^
1728                 0xfffff80000000000;
1729 #endif
1730         kern_linear_pte_xor[0] |= (_PAGE_CP_4U | _PAGE_CV_4U |
1731                                    _PAGE_P_4U | _PAGE_W_4U);
1732
1733         /* XXX Should use 256MB on Panther. XXX */
1734         kern_linear_pte_xor[1] = kern_linear_pte_xor[0];
1735
1736         _PAGE_SZBITS = _PAGE_SZBITS_4U;
1737         _PAGE_ALL_SZ_BITS =  (_PAGE_SZ4MB_4U | _PAGE_SZ512K_4U |
1738                               _PAGE_SZ64K_4U | _PAGE_SZ8K_4U |
1739                               _PAGE_SZ32MB_4U | _PAGE_SZ256MB_4U);
1740
1741
1742         page_none = _PAGE_PRESENT_4U | _PAGE_ACCESSED_4U | _PAGE_CACHE_4U;
1743         page_shared = (_PAGE_VALID | _PAGE_PRESENT_4U | _PAGE_CACHE_4U |
1744                        __ACCESS_BITS_4U | _PAGE_WRITE_4U | _PAGE_EXEC_4U);
1745         page_copy   = (_PAGE_VALID | _PAGE_PRESENT_4U | _PAGE_CACHE_4U |
1746                        __ACCESS_BITS_4U | _PAGE_EXEC_4U);
1747         page_readonly   = (_PAGE_VALID | _PAGE_PRESENT_4U | _PAGE_CACHE_4U |
1748                            __ACCESS_BITS_4U | _PAGE_EXEC_4U);
1749
1750         page_exec_bit = _PAGE_EXEC_4U;
1751
1752         prot_init_common(page_none, page_shared, page_copy, page_readonly,
1753                          page_exec_bit);
1754 }
1755
1756 static void __init sun4v_pgprot_init(void)
1757 {
1758         unsigned long page_none, page_shared, page_copy, page_readonly;
1759         unsigned long page_exec_bit;
1760
1761         PAGE_KERNEL = __pgprot (_PAGE_PRESENT_4V | _PAGE_VALID |
1762                                 _PAGE_CACHE_4V | _PAGE_P_4V |
1763                                 __ACCESS_BITS_4V | __DIRTY_BITS_4V |
1764                                 _PAGE_EXEC_4V);
1765         PAGE_KERNEL_LOCKED = PAGE_KERNEL;
1766         PAGE_EXEC = __pgprot(_PAGE_EXEC_4V);
1767
1768         _PAGE_IE = _PAGE_IE_4V;
1769         _PAGE_E = _PAGE_E_4V;
1770         _PAGE_CACHE = _PAGE_CACHE_4V;
1771
1772 #ifdef CONFIG_DEBUG_PAGEALLOC
1773         kern_linear_pte_xor[0] = (_PAGE_VALID | _PAGE_SZBITS_4V) ^
1774                 0xfffff80000000000;
1775 #else
1776         kern_linear_pte_xor[0] = (_PAGE_VALID | _PAGE_SZ4MB_4V) ^
1777                 0xfffff80000000000;
1778 #endif
1779         kern_linear_pte_xor[0] |= (_PAGE_CP_4V | _PAGE_CV_4V |
1780                                    _PAGE_P_4V | _PAGE_W_4V);
1781
1782 #ifdef CONFIG_DEBUG_PAGEALLOC
1783         kern_linear_pte_xor[1] = (_PAGE_VALID | _PAGE_SZBITS_4V) ^
1784                 0xfffff80000000000;
1785 #else
1786         kern_linear_pte_xor[1] = (_PAGE_VALID | _PAGE_SZ256MB_4V) ^
1787                 0xfffff80000000000;
1788 #endif
1789         kern_linear_pte_xor[1] |= (_PAGE_CP_4V | _PAGE_CV_4V |
1790                                    _PAGE_P_4V | _PAGE_W_4V);
1791
1792         pg_iobits = (_PAGE_VALID | _PAGE_PRESENT_4V | __DIRTY_BITS_4V |
1793                      __ACCESS_BITS_4V | _PAGE_E_4V);
1794
1795         _PAGE_SZBITS = _PAGE_SZBITS_4V;
1796         _PAGE_ALL_SZ_BITS = (_PAGE_SZ16GB_4V | _PAGE_SZ2GB_4V |
1797                              _PAGE_SZ256MB_4V | _PAGE_SZ32MB_4V |
1798                              _PAGE_SZ4MB_4V | _PAGE_SZ512K_4V |
1799                              _PAGE_SZ64K_4V | _PAGE_SZ8K_4V);
1800
1801         page_none = _PAGE_PRESENT_4V | _PAGE_ACCESSED_4V | _PAGE_CACHE_4V;
1802         page_shared = (_PAGE_VALID | _PAGE_PRESENT_4V | _PAGE_CACHE_4V |
1803                        __ACCESS_BITS_4V | _PAGE_WRITE_4V | _PAGE_EXEC_4V);
1804         page_copy   = (_PAGE_VALID | _PAGE_PRESENT_4V | _PAGE_CACHE_4V |
1805                        __ACCESS_BITS_4V | _PAGE_EXEC_4V);
1806         page_readonly = (_PAGE_VALID | _PAGE_PRESENT_4V | _PAGE_CACHE_4V |
1807                          __ACCESS_BITS_4V | _PAGE_EXEC_4V);
1808
1809         page_exec_bit = _PAGE_EXEC_4V;
1810
1811         prot_init_common(page_none, page_shared, page_copy, page_readonly,
1812                          page_exec_bit);
1813 }
1814
1815 unsigned long pte_sz_bits(unsigned long sz)
1816 {
1817         if (tlb_type == hypervisor) {
1818                 switch (sz) {
1819                 case 8 * 1024:
1820                 default:
1821                         return _PAGE_SZ8K_4V;
1822                 case 64 * 1024:
1823                         return _PAGE_SZ64K_4V;
1824                 case 512 * 1024:
1825                         return _PAGE_SZ512K_4V;
1826                 case 4 * 1024 * 1024:
1827                         return _PAGE_SZ4MB_4V;
1828                 };
1829         } else {
1830                 switch (sz) {
1831                 case 8 * 1024:
1832                 default:
1833                         return _PAGE_SZ8K_4U;
1834                 case 64 * 1024:
1835                         return _PAGE_SZ64K_4U;
1836                 case 512 * 1024:
1837                         return _PAGE_SZ512K_4U;
1838                 case 4 * 1024 * 1024:
1839                         return _PAGE_SZ4MB_4U;
1840                 };
1841         }
1842 }
1843
1844 pte_t mk_pte_io(unsigned long page, pgprot_t prot, int space, unsigned long page_size)
1845 {
1846         pte_t pte;
1847
1848         pte_val(pte)  = page | pgprot_val(pgprot_noncached(prot));
1849         pte_val(pte) |= (((unsigned long)space) << 32);
1850         pte_val(pte) |= pte_sz_bits(page_size);
1851
1852         return pte;
1853 }
1854
1855 static unsigned long kern_large_tte(unsigned long paddr)
1856 {
1857         unsigned long val;
1858
1859         val = (_PAGE_VALID | _PAGE_SZ4MB_4U |
1860                _PAGE_CP_4U | _PAGE_CV_4U | _PAGE_P_4U |
1861                _PAGE_EXEC_4U | _PAGE_L_4U | _PAGE_W_4U);
1862         if (tlb_type == hypervisor)
1863                 val = (_PAGE_VALID | _PAGE_SZ4MB_4V |
1864                        _PAGE_CP_4V | _PAGE_CV_4V | _PAGE_P_4V |
1865                        _PAGE_EXEC_4V | _PAGE_W_4V);
1866
1867         return val | paddr;
1868 }
1869
1870 /* If not locked, zap it. */
1871 void __flush_tlb_all(void)
1872 {
1873         unsigned long pstate;
1874         int i;
1875
1876         __asm__ __volatile__("flushw\n\t"
1877                              "rdpr      %%pstate, %0\n\t"
1878                              "wrpr      %0, %1, %%pstate"
1879                              : "=r" (pstate)
1880                              : "i" (PSTATE_IE));
1881         if (tlb_type == hypervisor) {
1882                 sun4v_mmu_demap_all();
1883         } else if (tlb_type == spitfire) {
1884                 for (i = 0; i < 64; i++) {
1885                         /* Spitfire Errata #32 workaround */
1886                         /* NOTE: Always runs on spitfire, so no
1887                          *       cheetah+ page size encodings.
1888                          */
1889                         __asm__ __volatile__("stxa      %0, [%1] %2\n\t"
1890                                              "flush     %%g6"
1891                                              : /* No outputs */
1892                                              : "r" (0),
1893                                              "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
1894
1895                         if (!(spitfire_get_dtlb_data(i) & _PAGE_L_4U)) {
1896                                 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
1897                                                      "membar #Sync"
1898                                                      : /* no outputs */
1899                                                      : "r" (TLB_TAG_ACCESS), "i" (ASI_DMMU));
1900                                 spitfire_put_dtlb_data(i, 0x0UL);
1901                         }
1902
1903                         /* Spitfire Errata #32 workaround */
1904                         /* NOTE: Always runs on spitfire, so no
1905                          *       cheetah+ page size encodings.
1906                          */
1907                         __asm__ __volatile__("stxa      %0, [%1] %2\n\t"
1908                                              "flush     %%g6"
1909                                              : /* No outputs */
1910                                              : "r" (0),
1911                                              "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
1912
1913                         if (!(spitfire_get_itlb_data(i) & _PAGE_L_4U)) {
1914                                 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
1915                                                      "membar #Sync"
1916                                                      : /* no outputs */
1917                                                      : "r" (TLB_TAG_ACCESS), "i" (ASI_IMMU));
1918                                 spitfire_put_itlb_data(i, 0x0UL);
1919                         }
1920                 }
1921         } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
1922                 cheetah_flush_dtlb_all();
1923                 cheetah_flush_itlb_all();
1924         }
1925         __asm__ __volatile__("wrpr      %0, 0, %%pstate"
1926                              : : "r" (pstate));
1927 }
1928
1929 #ifdef CONFIG_MEMORY_HOTPLUG
1930
1931 void online_page(struct page *page)
1932 {
1933         ClearPageReserved(page);
1934         init_page_count(page);
1935         __free_page(page);
1936         totalram_pages++;
1937         num_physpages++;
1938 }
1939
1940 #endif /* CONFIG_MEMORY_HOTPLUG */