Merge branch 'for-linus' of git://git.o-hand.com/linux-rpurdie-backlight
[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         prom_printf("Remapping the kernel... ");
622         remap_kernel();
623         prom_printf("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 #ifdef CONFIG_DEBUG_BOOTMEM
743         prom_printf("choose_bootmap_pfn: kern[%lx:%lx] avoid[%lx:%lx]\n",
744                     kern_base, PAGE_ALIGN(kern_base + kern_size),
745                     avoid_start, avoid_end);
746 #endif
747         for (i = 0; i < pavail_ents; i++) {
748                 unsigned long start, end;
749
750                 start = pavail[i].phys_addr;
751                 end = start + pavail[i].reg_size;
752
753                 while (start < end) {
754                         if (start >= kern_base &&
755                             start < PAGE_ALIGN(kern_base + kern_size)) {
756                                 start = PAGE_ALIGN(kern_base + kern_size);
757                                 continue;
758                         }
759                         if (start >= avoid_start && start < avoid_end) {
760                                 start = avoid_end;
761                                 continue;
762                         }
763
764                         if ((end - start) < bootmap_size)
765                                 break;
766
767                         if (start < kern_base &&
768                             (start + bootmap_size) > kern_base) {
769                                 start = PAGE_ALIGN(kern_base + kern_size);
770                                 continue;
771                         }
772
773                         if (start < avoid_start &&
774                             (start + bootmap_size) > avoid_start) {
775                                 start = avoid_end;
776                                 continue;
777                         }
778
779                         /* OK, it doesn't overlap anything, use it.  */
780 #ifdef CONFIG_DEBUG_BOOTMEM
781                         prom_printf("choose_bootmap_pfn: Using %lx [%lx]\n",
782                                     start >> PAGE_SHIFT, start);
783 #endif
784                         return start >> PAGE_SHIFT;
785                 }
786         }
787
788         prom_printf("Cannot find free area for bootmap, aborting.\n");
789         prom_halt();
790 }
791
792 static void __init trim_pavail(unsigned long *cur_size_p,
793                                unsigned long *end_of_phys_p)
794 {
795         unsigned long to_trim = *cur_size_p - cmdline_memory_size;
796         unsigned long avoid_start, avoid_end;
797         int i;
798
799         to_trim = PAGE_ALIGN(to_trim);
800
801         avoid_start = avoid_end = 0;
802 #ifdef CONFIG_BLK_DEV_INITRD
803         avoid_start = initrd_start;
804         avoid_end = PAGE_ALIGN(initrd_end);
805 #endif
806
807         /* Trim some pavail[] entries in order to satisfy the
808          * requested "mem=xxx" kernel command line specification.
809          *
810          * We must not trim off the kernel image area nor the
811          * initial ramdisk range (if any).  Also, we must not trim
812          * any pavail[] entry down to zero in order to preserve
813          * the invariant that all pavail[] entries have a non-zero
814          * size which is assumed by all of the code in here.
815          */
816         for (i = 0; i < pavail_ents; i++) {
817                 unsigned long start, end, kern_end;
818                 unsigned long trim_low, trim_high, n;
819
820                 kern_end = PAGE_ALIGN(kern_base + kern_size);
821
822                 trim_low = start = pavail[i].phys_addr;
823                 trim_high = end = start + pavail[i].reg_size;
824
825                 if (kern_base >= start &&
826                     kern_base < end) {
827                         trim_low = kern_base;
828                         if (kern_end >= end)
829                                 continue;
830                 }
831                 if (kern_end >= start &&
832                     kern_end < end) {
833                         trim_high = kern_end;
834                 }
835                 if (avoid_start &&
836                     avoid_start >= start &&
837                     avoid_start < end) {
838                         if (trim_low > avoid_start)
839                                 trim_low = avoid_start;
840                         if (avoid_end >= end)
841                                 continue;
842                 }
843                 if (avoid_end &&
844                     avoid_end >= start &&
845                     avoid_end < end) {
846                         if (trim_high < avoid_end)
847                                 trim_high = avoid_end;
848                 }
849
850                 if (trim_high <= trim_low)
851                         continue;
852
853                 if (trim_low == start && trim_high == end) {
854                         /* Whole chunk is available for trimming.
855                          * Trim all except one page, in order to keep
856                          * entry non-empty.
857                          */
858                         n = (end - start) - PAGE_SIZE;
859                         if (n > to_trim)
860                                 n = to_trim;
861
862                         if (n) {
863                                 pavail[i].phys_addr += n;
864                                 pavail[i].reg_size -= n;
865                                 to_trim -= n;
866                         }
867                 } else {
868                         n = (trim_low - start);
869                         if (n > to_trim)
870                                 n = to_trim;
871
872                         if (n) {
873                                 pavail[i].phys_addr += n;
874                                 pavail[i].reg_size -= n;
875                                 to_trim -= n;
876                         }
877                         if (to_trim) {
878                                 n = end - trim_high;
879                                 if (n > to_trim)
880                                         n = to_trim;
881                                 if (n) {
882                                         pavail[i].reg_size -= n;
883                                         to_trim -= n;
884                                 }
885                         }
886                 }
887
888                 if (!to_trim)
889                         break;
890         }
891
892         /* Recalculate.  */
893         *cur_size_p = 0UL;
894         for (i = 0; i < pavail_ents; i++) {
895                 *end_of_phys_p = pavail[i].phys_addr +
896                         pavail[i].reg_size;
897                 *cur_size_p += pavail[i].reg_size;
898         }
899 }
900
901 /* About pages_avail, this is the value we will use to calculate
902  * the zholes_size[] argument given to free_area_init_node().  The
903  * page allocator uses this to calculate nr_kernel_pages,
904  * nr_all_pages and zone->present_pages.  On NUMA it is used
905  * to calculate zone->min_unmapped_pages and zone->min_slab_pages.
906  *
907  * So this number should really be set to what the page allocator
908  * actually ends up with.  This means:
909  * 1) It should include bootmem map pages, we'll release those.
910  * 2) It should not include the kernel image, except for the
911  *    __init sections which we will also release.
912  * 3) It should include the initrd image, since we'll release
913  *    that too.
914  */
915 static unsigned long __init bootmem_init(unsigned long *pages_avail,
916                                          unsigned long phys_base)
917 {
918         unsigned long bootmap_size, end_pfn;
919         unsigned long end_of_phys_memory = 0UL;
920         unsigned long bootmap_pfn, bytes_avail, size;
921         int i;
922
923 #ifdef CONFIG_DEBUG_BOOTMEM
924         prom_printf("bootmem_init: Scan pavail, ");
925 #endif
926
927         bytes_avail = 0UL;
928         for (i = 0; i < pavail_ents; i++) {
929                 end_of_phys_memory = pavail[i].phys_addr +
930                         pavail[i].reg_size;
931                 bytes_avail += pavail[i].reg_size;
932         }
933
934         /* Determine the location of the initial ramdisk before trying
935          * to honor the "mem=xxx" command line argument.  We must know
936          * where the kernel image and the ramdisk image are so that we
937          * do not trim those two areas from the physical memory map.
938          */
939
940 #ifdef CONFIG_BLK_DEV_INITRD
941         /* Now have to check initial ramdisk, so that bootmap does not overwrite it */
942         if (sparc_ramdisk_image || sparc_ramdisk_image64) {
943                 unsigned long ramdisk_image = sparc_ramdisk_image ?
944                         sparc_ramdisk_image : sparc_ramdisk_image64;
945                 ramdisk_image -= KERNBASE;
946                 initrd_start = ramdisk_image + phys_base;
947                 initrd_end = initrd_start + sparc_ramdisk_size;
948                 if (initrd_end > end_of_phys_memory) {
949                         printk(KERN_CRIT "initrd extends beyond end of memory "
950                                          "(0x%016lx > 0x%016lx)\ndisabling initrd\n",
951                                initrd_end, end_of_phys_memory);
952                         initrd_start = 0;
953                         initrd_end = 0;
954                 }
955         }
956 #endif  
957
958         if (cmdline_memory_size &&
959             bytes_avail > cmdline_memory_size)
960                 trim_pavail(&bytes_avail,
961                             &end_of_phys_memory);
962
963         *pages_avail = bytes_avail >> PAGE_SHIFT;
964
965         end_pfn = end_of_phys_memory >> PAGE_SHIFT;
966
967         /* Initialize the boot-time allocator. */
968         max_pfn = max_low_pfn = end_pfn;
969         min_low_pfn = (phys_base >> PAGE_SHIFT);
970
971         bootmap_pfn = choose_bootmap_pfn(min_low_pfn, end_pfn);
972
973 #ifdef CONFIG_DEBUG_BOOTMEM
974         prom_printf("init_bootmem(min[%lx], bootmap[%lx], max[%lx])\n",
975                     min_low_pfn, bootmap_pfn, max_low_pfn);
976 #endif
977         bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap_pfn,
978                                          min_low_pfn, end_pfn);
979
980         /* Now register the available physical memory with the
981          * allocator.
982          */
983         for (i = 0; i < pavail_ents; i++) {
984 #ifdef CONFIG_DEBUG_BOOTMEM
985                 prom_printf("free_bootmem(pavail:%d): base[%lx] size[%lx]\n",
986                             i, pavail[i].phys_addr, pavail[i].reg_size);
987 #endif
988                 free_bootmem(pavail[i].phys_addr, pavail[i].reg_size);
989         }
990
991 #ifdef CONFIG_BLK_DEV_INITRD
992         if (initrd_start) {
993                 size = initrd_end - initrd_start;
994
995                 /* Reserve the initrd image area. */
996 #ifdef CONFIG_DEBUG_BOOTMEM
997                 prom_printf("reserve_bootmem(initrd): base[%llx] size[%lx]\n",
998                         initrd_start, initrd_end);
999 #endif
1000                 reserve_bootmem(initrd_start, size, BOOTMEM_DEFAULT);
1001
1002                 initrd_start += PAGE_OFFSET;
1003                 initrd_end += PAGE_OFFSET;
1004         }
1005 #endif
1006         /* Reserve the kernel text/data/bss. */
1007 #ifdef CONFIG_DEBUG_BOOTMEM
1008         prom_printf("reserve_bootmem(kernel): base[%lx] size[%lx]\n", kern_base, kern_size);
1009 #endif
1010         reserve_bootmem(kern_base, kern_size, BOOTMEM_DEFAULT);
1011         *pages_avail -= PAGE_ALIGN(kern_size) >> PAGE_SHIFT;
1012
1013         /* Add back in the initmem pages. */
1014         size = ((unsigned long)(__init_end) & PAGE_MASK) -
1015                 PAGE_ALIGN((unsigned long)__init_begin);
1016         *pages_avail += size >> PAGE_SHIFT;
1017
1018         /* Reserve the bootmem map.   We do not account for it
1019          * in pages_avail because we will release that memory
1020          * in free_all_bootmem.
1021          */
1022         size = bootmap_size;
1023 #ifdef CONFIG_DEBUG_BOOTMEM
1024         prom_printf("reserve_bootmem(bootmap): base[%lx] size[%lx]\n",
1025                     (bootmap_pfn << PAGE_SHIFT), size);
1026 #endif
1027         reserve_bootmem((bootmap_pfn << PAGE_SHIFT), size, BOOTMEM_DEFAULT);
1028
1029         for (i = 0; i < pavail_ents; i++) {
1030                 unsigned long start_pfn, end_pfn;
1031
1032                 start_pfn = pavail[i].phys_addr >> PAGE_SHIFT;
1033                 end_pfn = (start_pfn + (pavail[i].reg_size >> PAGE_SHIFT));
1034 #ifdef CONFIG_DEBUG_BOOTMEM
1035                 prom_printf("memory_present(0, %lx, %lx)\n",
1036                             start_pfn, end_pfn);
1037 #endif
1038                 memory_present(0, start_pfn, end_pfn);
1039         }
1040
1041         sparse_init();
1042
1043         return end_pfn;
1044 }
1045
1046 static struct linux_prom64_registers pall[MAX_BANKS] __initdata;
1047 static int pall_ents __initdata;
1048
1049 #ifdef CONFIG_DEBUG_PAGEALLOC
1050 static unsigned long kernel_map_range(unsigned long pstart, unsigned long pend, pgprot_t prot)
1051 {
1052         unsigned long vstart = PAGE_OFFSET + pstart;
1053         unsigned long vend = PAGE_OFFSET + pend;
1054         unsigned long alloc_bytes = 0UL;
1055
1056         if ((vstart & ~PAGE_MASK) || (vend & ~PAGE_MASK)) {
1057                 prom_printf("kernel_map: Unaligned physmem[%lx:%lx]\n",
1058                             vstart, vend);
1059                 prom_halt();
1060         }
1061
1062         while (vstart < vend) {
1063                 unsigned long this_end, paddr = __pa(vstart);
1064                 pgd_t *pgd = pgd_offset_k(vstart);
1065                 pud_t *pud;
1066                 pmd_t *pmd;
1067                 pte_t *pte;
1068
1069                 pud = pud_offset(pgd, vstart);
1070                 if (pud_none(*pud)) {
1071                         pmd_t *new;
1072
1073                         new = __alloc_bootmem(PAGE_SIZE, PAGE_SIZE, PAGE_SIZE);
1074                         alloc_bytes += PAGE_SIZE;
1075                         pud_populate(&init_mm, pud, new);
1076                 }
1077
1078                 pmd = pmd_offset(pud, vstart);
1079                 if (!pmd_present(*pmd)) {
1080                         pte_t *new;
1081
1082                         new = __alloc_bootmem(PAGE_SIZE, PAGE_SIZE, PAGE_SIZE);
1083                         alloc_bytes += PAGE_SIZE;
1084                         pmd_populate_kernel(&init_mm, pmd, new);
1085                 }
1086
1087                 pte = pte_offset_kernel(pmd, vstart);
1088                 this_end = (vstart + PMD_SIZE) & PMD_MASK;
1089                 if (this_end > vend)
1090                         this_end = vend;
1091
1092                 while (vstart < this_end) {
1093                         pte_val(*pte) = (paddr | pgprot_val(prot));
1094
1095                         vstart += PAGE_SIZE;
1096                         paddr += PAGE_SIZE;
1097                         pte++;
1098                 }
1099         }
1100
1101         return alloc_bytes;
1102 }
1103
1104 extern unsigned int kvmap_linear_patch[1];
1105 #endif /* CONFIG_DEBUG_PAGEALLOC */
1106
1107 static void __init mark_kpte_bitmap(unsigned long start, unsigned long end)
1108 {
1109         const unsigned long shift_256MB = 28;
1110         const unsigned long mask_256MB = ((1UL << shift_256MB) - 1UL);
1111         const unsigned long size_256MB = (1UL << shift_256MB);
1112
1113         while (start < end) {
1114                 long remains;
1115
1116                 remains = end - start;
1117                 if (remains < size_256MB)
1118                         break;
1119
1120                 if (start & mask_256MB) {
1121                         start = (start + size_256MB) & ~mask_256MB;
1122                         continue;
1123                 }
1124
1125                 while (remains >= size_256MB) {
1126                         unsigned long index = start >> shift_256MB;
1127
1128                         __set_bit(index, kpte_linear_bitmap);
1129
1130                         start += size_256MB;
1131                         remains -= size_256MB;
1132                 }
1133         }
1134 }
1135
1136 static void __init init_kpte_bitmap(void)
1137 {
1138         unsigned long i;
1139
1140         for (i = 0; i < pall_ents; i++) {
1141                 unsigned long phys_start, phys_end;
1142
1143                 phys_start = pall[i].phys_addr;
1144                 phys_end = phys_start + pall[i].reg_size;
1145
1146                 mark_kpte_bitmap(phys_start, phys_end);
1147         }
1148 }
1149
1150 static void __init kernel_physical_mapping_init(void)
1151 {
1152 #ifdef CONFIG_DEBUG_PAGEALLOC
1153         unsigned long i, mem_alloced = 0UL;
1154
1155         for (i = 0; i < pall_ents; i++) {
1156                 unsigned long phys_start, phys_end;
1157
1158                 phys_start = pall[i].phys_addr;
1159                 phys_end = phys_start + pall[i].reg_size;
1160
1161                 mem_alloced += kernel_map_range(phys_start, phys_end,
1162                                                 PAGE_KERNEL);
1163         }
1164
1165         printk("Allocated %ld bytes for kernel page tables.\n",
1166                mem_alloced);
1167
1168         kvmap_linear_patch[0] = 0x01000000; /* nop */
1169         flushi(&kvmap_linear_patch[0]);
1170
1171         __flush_tlb_all();
1172 #endif
1173 }
1174
1175 #ifdef CONFIG_DEBUG_PAGEALLOC
1176 void kernel_map_pages(struct page *page, int numpages, int enable)
1177 {
1178         unsigned long phys_start = page_to_pfn(page) << PAGE_SHIFT;
1179         unsigned long phys_end = phys_start + (numpages * PAGE_SIZE);
1180
1181         kernel_map_range(phys_start, phys_end,
1182                          (enable ? PAGE_KERNEL : __pgprot(0)));
1183
1184         flush_tsb_kernel_range(PAGE_OFFSET + phys_start,
1185                                PAGE_OFFSET + phys_end);
1186
1187         /* we should perform an IPI and flush all tlbs,
1188          * but that can deadlock->flush only current cpu.
1189          */
1190         __flush_tlb_kernel_range(PAGE_OFFSET + phys_start,
1191                                  PAGE_OFFSET + phys_end);
1192 }
1193 #endif
1194
1195 unsigned long __init find_ecache_flush_span(unsigned long size)
1196 {
1197         int i;
1198
1199         for (i = 0; i < pavail_ents; i++) {
1200                 if (pavail[i].reg_size >= size)
1201                         return pavail[i].phys_addr;
1202         }
1203
1204         return ~0UL;
1205 }
1206
1207 static void __init tsb_phys_patch(void)
1208 {
1209         struct tsb_ldquad_phys_patch_entry *pquad;
1210         struct tsb_phys_patch_entry *p;
1211
1212         pquad = &__tsb_ldquad_phys_patch;
1213         while (pquad < &__tsb_ldquad_phys_patch_end) {
1214                 unsigned long addr = pquad->addr;
1215
1216                 if (tlb_type == hypervisor)
1217                         *(unsigned int *) addr = pquad->sun4v_insn;
1218                 else
1219                         *(unsigned int *) addr = pquad->sun4u_insn;
1220                 wmb();
1221                 __asm__ __volatile__("flush     %0"
1222                                      : /* no outputs */
1223                                      : "r" (addr));
1224
1225                 pquad++;
1226         }
1227
1228         p = &__tsb_phys_patch;
1229         while (p < &__tsb_phys_patch_end) {
1230                 unsigned long addr = p->addr;
1231
1232                 *(unsigned int *) addr = p->insn;
1233                 wmb();
1234                 __asm__ __volatile__("flush     %0"
1235                                      : /* no outputs */
1236                                      : "r" (addr));
1237
1238                 p++;
1239         }
1240 }
1241
1242 /* Don't mark as init, we give this to the Hypervisor.  */
1243 #ifndef CONFIG_DEBUG_PAGEALLOC
1244 #define NUM_KTSB_DESCR  2
1245 #else
1246 #define NUM_KTSB_DESCR  1
1247 #endif
1248 static struct hv_tsb_descr ktsb_descr[NUM_KTSB_DESCR];
1249 extern struct tsb swapper_tsb[KERNEL_TSB_NENTRIES];
1250
1251 static void __init sun4v_ktsb_init(void)
1252 {
1253         unsigned long ktsb_pa;
1254
1255         /* First KTSB for PAGE_SIZE mappings.  */
1256         ktsb_pa = kern_base + ((unsigned long)&swapper_tsb[0] - KERNBASE);
1257
1258         switch (PAGE_SIZE) {
1259         case 8 * 1024:
1260         default:
1261                 ktsb_descr[0].pgsz_idx = HV_PGSZ_IDX_8K;
1262                 ktsb_descr[0].pgsz_mask = HV_PGSZ_MASK_8K;
1263                 break;
1264
1265         case 64 * 1024:
1266                 ktsb_descr[0].pgsz_idx = HV_PGSZ_IDX_64K;
1267                 ktsb_descr[0].pgsz_mask = HV_PGSZ_MASK_64K;
1268                 break;
1269
1270         case 512 * 1024:
1271                 ktsb_descr[0].pgsz_idx = HV_PGSZ_IDX_512K;
1272                 ktsb_descr[0].pgsz_mask = HV_PGSZ_MASK_512K;
1273                 break;
1274
1275         case 4 * 1024 * 1024:
1276                 ktsb_descr[0].pgsz_idx = HV_PGSZ_IDX_4MB;
1277                 ktsb_descr[0].pgsz_mask = HV_PGSZ_MASK_4MB;
1278                 break;
1279         };
1280
1281         ktsb_descr[0].assoc = 1;
1282         ktsb_descr[0].num_ttes = KERNEL_TSB_NENTRIES;
1283         ktsb_descr[0].ctx_idx = 0;
1284         ktsb_descr[0].tsb_base = ktsb_pa;
1285         ktsb_descr[0].resv = 0;
1286
1287 #ifndef CONFIG_DEBUG_PAGEALLOC
1288         /* Second KTSB for 4MB/256MB mappings.  */
1289         ktsb_pa = (kern_base +
1290                    ((unsigned long)&swapper_4m_tsb[0] - KERNBASE));
1291
1292         ktsb_descr[1].pgsz_idx = HV_PGSZ_IDX_4MB;
1293         ktsb_descr[1].pgsz_mask = (HV_PGSZ_MASK_4MB |
1294                                    HV_PGSZ_MASK_256MB);
1295         ktsb_descr[1].assoc = 1;
1296         ktsb_descr[1].num_ttes = KERNEL_TSB4M_NENTRIES;
1297         ktsb_descr[1].ctx_idx = 0;
1298         ktsb_descr[1].tsb_base = ktsb_pa;
1299         ktsb_descr[1].resv = 0;
1300 #endif
1301 }
1302
1303 void __cpuinit sun4v_ktsb_register(void)
1304 {
1305         unsigned long pa, ret;
1306
1307         pa = kern_base + ((unsigned long)&ktsb_descr[0] - KERNBASE);
1308
1309         ret = sun4v_mmu_tsb_ctx0(NUM_KTSB_DESCR, pa);
1310         if (ret != 0) {
1311                 prom_printf("hypervisor_mmu_tsb_ctx0[%lx]: "
1312                             "errors with %lx\n", pa, ret);
1313                 prom_halt();
1314         }
1315 }
1316
1317 /* paging_init() sets up the page tables */
1318
1319 extern void cheetah_ecache_flush_init(void);
1320 extern void sun4v_patch_tlb_handlers(void);
1321
1322 extern void cpu_probe(void);
1323 extern void central_probe(void);
1324
1325 static unsigned long last_valid_pfn;
1326 pgd_t swapper_pg_dir[2048];
1327
1328 static void sun4u_pgprot_init(void);
1329 static void sun4v_pgprot_init(void);
1330
1331 /* Dummy function */
1332 void __init setup_per_cpu_areas(void)
1333 {
1334 }
1335
1336 void __init paging_init(void)
1337 {
1338         unsigned long end_pfn, pages_avail, shift, phys_base;
1339         unsigned long real_end, i;
1340
1341         /* These build time checkes make sure that the dcache_dirty_cpu()
1342          * page->flags usage will work.
1343          *
1344          * When a page gets marked as dcache-dirty, we store the
1345          * cpu number starting at bit 32 in the page->flags.  Also,
1346          * functions like clear_dcache_dirty_cpu use the cpu mask
1347          * in 13-bit signed-immediate instruction fields.
1348          */
1349         BUILD_BUG_ON(FLAGS_RESERVED != 32);
1350         BUILD_BUG_ON(SECTIONS_WIDTH + NODES_WIDTH + ZONES_WIDTH +
1351                      ilog2(roundup_pow_of_two(NR_CPUS)) > FLAGS_RESERVED);
1352         BUILD_BUG_ON(NR_CPUS > 4096);
1353
1354         kern_base = (prom_boot_mapping_phys_low >> 22UL) << 22UL;
1355         kern_size = (unsigned long)&_end - (unsigned long)KERNBASE;
1356
1357         sstate_booting();
1358
1359         /* Invalidate both kernel TSBs.  */
1360         memset(swapper_tsb, 0x40, sizeof(swapper_tsb));
1361 #ifndef CONFIG_DEBUG_PAGEALLOC
1362         memset(swapper_4m_tsb, 0x40, sizeof(swapper_4m_tsb));
1363 #endif
1364
1365         if (tlb_type == hypervisor)
1366                 sun4v_pgprot_init();
1367         else
1368                 sun4u_pgprot_init();
1369
1370         if (tlb_type == cheetah_plus ||
1371             tlb_type == hypervisor)
1372                 tsb_phys_patch();
1373
1374         if (tlb_type == hypervisor) {
1375                 sun4v_patch_tlb_handlers();
1376                 sun4v_ktsb_init();
1377         }
1378
1379         /* Find available physical memory... */
1380         read_obp_memory("available", &pavail[0], &pavail_ents);
1381
1382         phys_base = 0xffffffffffffffffUL;
1383         for (i = 0; i < pavail_ents; i++)
1384                 phys_base = min(phys_base, pavail[i].phys_addr);
1385
1386         set_bit(0, mmu_context_bmap);
1387
1388         shift = kern_base + PAGE_OFFSET - ((unsigned long)KERNBASE);
1389
1390         real_end = (unsigned long)_end;
1391         if ((real_end > ((unsigned long)KERNBASE + 0x400000)))
1392                 bigkernel = 1;
1393         if ((real_end > ((unsigned long)KERNBASE + 0x800000))) {
1394                 prom_printf("paging_init: Kernel > 8MB, too large.\n");
1395                 prom_halt();
1396         }
1397
1398         /* Set kernel pgd to upper alias so physical page computations
1399          * work.
1400          */
1401         init_mm.pgd += ((shift) / (sizeof(pgd_t)));
1402         
1403         memset(swapper_low_pmd_dir, 0, sizeof(swapper_low_pmd_dir));
1404
1405         /* Now can init the kernel/bad page tables. */
1406         pud_set(pud_offset(&swapper_pg_dir[0], 0),
1407                 swapper_low_pmd_dir + (shift / sizeof(pgd_t)));
1408         
1409         inherit_prom_mappings();
1410         
1411         read_obp_memory("reg", &pall[0], &pall_ents);
1412
1413         init_kpte_bitmap();
1414
1415         /* Ok, we can use our TLB miss and window trap handlers safely.  */
1416         setup_tba();
1417
1418         __flush_tlb_all();
1419
1420         if (tlb_type == hypervisor)
1421                 sun4v_ktsb_register();
1422
1423         /* Setup bootmem... */
1424         pages_avail = 0;
1425         last_valid_pfn = end_pfn = bootmem_init(&pages_avail, phys_base);
1426
1427         max_mapnr = last_valid_pfn;
1428
1429         kernel_physical_mapping_init();
1430
1431         real_setup_per_cpu_areas();
1432
1433         prom_build_devicetree();
1434
1435         if (tlb_type == hypervisor)
1436                 sun4v_mdesc_init();
1437
1438         {
1439                 unsigned long zones_size[MAX_NR_ZONES];
1440                 unsigned long zholes_size[MAX_NR_ZONES];
1441                 int znum;
1442
1443                 for (znum = 0; znum < MAX_NR_ZONES; znum++)
1444                         zones_size[znum] = zholes_size[znum] = 0;
1445
1446                 zones_size[ZONE_NORMAL] = end_pfn;
1447                 zholes_size[ZONE_NORMAL] = end_pfn - pages_avail;
1448
1449                 free_area_init_node(0, &contig_page_data, zones_size,
1450                                     __pa(PAGE_OFFSET) >> PAGE_SHIFT,
1451                                     zholes_size);
1452         }
1453
1454         prom_printf("Booting Linux...\n");
1455
1456         central_probe();
1457         cpu_probe();
1458 }
1459
1460 static void __init taint_real_pages(void)
1461 {
1462         int i;
1463
1464         read_obp_memory("available", &pavail_rescan[0], &pavail_rescan_ents);
1465
1466         /* Find changes discovered in the physmem available rescan and
1467          * reserve the lost portions in the bootmem maps.
1468          */
1469         for (i = 0; i < pavail_ents; i++) {
1470                 unsigned long old_start, old_end;
1471
1472                 old_start = pavail[i].phys_addr;
1473                 old_end = old_start +
1474                         pavail[i].reg_size;
1475                 while (old_start < old_end) {
1476                         int n;
1477
1478                         for (n = 0; n < pavail_rescan_ents; n++) {
1479                                 unsigned long new_start, new_end;
1480
1481                                 new_start = pavail_rescan[n].phys_addr;
1482                                 new_end = new_start +
1483                                         pavail_rescan[n].reg_size;
1484
1485                                 if (new_start <= old_start &&
1486                                     new_end >= (old_start + PAGE_SIZE)) {
1487                                         set_bit(old_start >> 22,
1488                                                 sparc64_valid_addr_bitmap);
1489                                         goto do_next_page;
1490                                 }
1491                         }
1492                         reserve_bootmem(old_start, PAGE_SIZE, BOOTMEM_DEFAULT);
1493
1494                 do_next_page:
1495                         old_start += PAGE_SIZE;
1496                 }
1497         }
1498 }
1499
1500 int __init page_in_phys_avail(unsigned long paddr)
1501 {
1502         int i;
1503
1504         paddr &= PAGE_MASK;
1505
1506         for (i = 0; i < pavail_rescan_ents; i++) {
1507                 unsigned long start, end;
1508
1509                 start = pavail_rescan[i].phys_addr;
1510                 end = start + pavail_rescan[i].reg_size;
1511
1512                 if (paddr >= start && paddr < end)
1513                         return 1;
1514         }
1515         if (paddr >= kern_base && paddr < (kern_base + kern_size))
1516                 return 1;
1517 #ifdef CONFIG_BLK_DEV_INITRD
1518         if (paddr >= __pa(initrd_start) &&
1519             paddr < __pa(PAGE_ALIGN(initrd_end)))
1520                 return 1;
1521 #endif
1522
1523         return 0;
1524 }
1525
1526 void __init mem_init(void)
1527 {
1528         unsigned long codepages, datapages, initpages;
1529         unsigned long addr, last;
1530         int i;
1531
1532         i = last_valid_pfn >> ((22 - PAGE_SHIFT) + 6);
1533         i += 1;
1534         sparc64_valid_addr_bitmap = (unsigned long *) alloc_bootmem(i << 3);
1535         if (sparc64_valid_addr_bitmap == NULL) {
1536                 prom_printf("mem_init: Cannot alloc valid_addr_bitmap.\n");
1537                 prom_halt();
1538         }
1539         memset(sparc64_valid_addr_bitmap, 0, i << 3);
1540
1541         addr = PAGE_OFFSET + kern_base;
1542         last = PAGE_ALIGN(kern_size) + addr;
1543         while (addr < last) {
1544                 set_bit(__pa(addr) >> 22, sparc64_valid_addr_bitmap);
1545                 addr += PAGE_SIZE;
1546         }
1547
1548         taint_real_pages();
1549
1550         high_memory = __va(last_valid_pfn << PAGE_SHIFT);
1551
1552 #ifdef CONFIG_DEBUG_BOOTMEM
1553         prom_printf("mem_init: Calling free_all_bootmem().\n");
1554 #endif
1555
1556         /* We subtract one to account for the mem_map_zero page
1557          * allocated below.
1558          */
1559         totalram_pages = num_physpages = free_all_bootmem() - 1;
1560
1561         /*
1562          * Set up the zero page, mark it reserved, so that page count
1563          * is not manipulated when freeing the page from user ptes.
1564          */
1565         mem_map_zero = alloc_pages(GFP_KERNEL|__GFP_ZERO, 0);
1566         if (mem_map_zero == NULL) {
1567                 prom_printf("paging_init: Cannot alloc zero page.\n");
1568                 prom_halt();
1569         }
1570         SetPageReserved(mem_map_zero);
1571
1572         codepages = (((unsigned long) _etext) - ((unsigned long) _start));
1573         codepages = PAGE_ALIGN(codepages) >> PAGE_SHIFT;
1574         datapages = (((unsigned long) _edata) - ((unsigned long) _etext));
1575         datapages = PAGE_ALIGN(datapages) >> PAGE_SHIFT;
1576         initpages = (((unsigned long) __init_end) - ((unsigned long) __init_begin));
1577         initpages = PAGE_ALIGN(initpages) >> PAGE_SHIFT;
1578
1579         printk("Memory: %luk available (%ldk kernel code, %ldk data, %ldk init) [%016lx,%016lx]\n",
1580                nr_free_pages() << (PAGE_SHIFT-10),
1581                codepages << (PAGE_SHIFT-10),
1582                datapages << (PAGE_SHIFT-10), 
1583                initpages << (PAGE_SHIFT-10), 
1584                PAGE_OFFSET, (last_valid_pfn << PAGE_SHIFT));
1585
1586         if (tlb_type == cheetah || tlb_type == cheetah_plus)
1587                 cheetah_ecache_flush_init();
1588 }
1589
1590 void free_initmem(void)
1591 {
1592         unsigned long addr, initend;
1593
1594         /*
1595          * The init section is aligned to 8k in vmlinux.lds. Page align for >8k pagesizes.
1596          */
1597         addr = PAGE_ALIGN((unsigned long)(__init_begin));
1598         initend = (unsigned long)(__init_end) & PAGE_MASK;
1599         for (; addr < initend; addr += PAGE_SIZE) {
1600                 unsigned long page;
1601                 struct page *p;
1602
1603                 page = (addr +
1604                         ((unsigned long) __va(kern_base)) -
1605                         ((unsigned long) KERNBASE));
1606                 memset((void *)addr, POISON_FREE_INITMEM, PAGE_SIZE);
1607                 p = virt_to_page(page);
1608
1609                 ClearPageReserved(p);
1610                 init_page_count(p);
1611                 __free_page(p);
1612                 num_physpages++;
1613                 totalram_pages++;
1614         }
1615 }
1616
1617 #ifdef CONFIG_BLK_DEV_INITRD
1618 void free_initrd_mem(unsigned long start, unsigned long end)
1619 {
1620         if (start < end)
1621                 printk ("Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
1622         for (; start < end; start += PAGE_SIZE) {
1623                 struct page *p = virt_to_page(start);
1624
1625                 ClearPageReserved(p);
1626                 init_page_count(p);
1627                 __free_page(p);
1628                 num_physpages++;
1629                 totalram_pages++;
1630         }
1631 }
1632 #endif
1633
1634 #define _PAGE_CACHE_4U  (_PAGE_CP_4U | _PAGE_CV_4U)
1635 #define _PAGE_CACHE_4V  (_PAGE_CP_4V | _PAGE_CV_4V)
1636 #define __DIRTY_BITS_4U  (_PAGE_MODIFIED_4U | _PAGE_WRITE_4U | _PAGE_W_4U)
1637 #define __DIRTY_BITS_4V  (_PAGE_MODIFIED_4V | _PAGE_WRITE_4V | _PAGE_W_4V)
1638 #define __ACCESS_BITS_4U (_PAGE_ACCESSED_4U | _PAGE_READ_4U | _PAGE_R)
1639 #define __ACCESS_BITS_4V (_PAGE_ACCESSED_4V | _PAGE_READ_4V | _PAGE_R)
1640
1641 pgprot_t PAGE_KERNEL __read_mostly;
1642 EXPORT_SYMBOL(PAGE_KERNEL);
1643
1644 pgprot_t PAGE_KERNEL_LOCKED __read_mostly;
1645 pgprot_t PAGE_COPY __read_mostly;
1646
1647 pgprot_t PAGE_SHARED __read_mostly;
1648 EXPORT_SYMBOL(PAGE_SHARED);
1649
1650 pgprot_t PAGE_EXEC __read_mostly;
1651 unsigned long pg_iobits __read_mostly;
1652
1653 unsigned long _PAGE_IE __read_mostly;
1654 EXPORT_SYMBOL(_PAGE_IE);
1655
1656 unsigned long _PAGE_E __read_mostly;
1657 EXPORT_SYMBOL(_PAGE_E);
1658
1659 unsigned long _PAGE_CACHE __read_mostly;
1660 EXPORT_SYMBOL(_PAGE_CACHE);
1661
1662 #ifdef CONFIG_SPARSEMEM_VMEMMAP
1663
1664 #define VMEMMAP_CHUNK_SHIFT     22
1665 #define VMEMMAP_CHUNK           (1UL << VMEMMAP_CHUNK_SHIFT)
1666 #define VMEMMAP_CHUNK_MASK      ~(VMEMMAP_CHUNK - 1UL)
1667 #define VMEMMAP_ALIGN(x)        (((x)+VMEMMAP_CHUNK-1UL)&VMEMMAP_CHUNK_MASK)
1668
1669 #define VMEMMAP_SIZE    ((((1UL << MAX_PHYSADDR_BITS) >> PAGE_SHIFT) * \
1670                           sizeof(struct page *)) >> VMEMMAP_CHUNK_SHIFT)
1671 unsigned long vmemmap_table[VMEMMAP_SIZE];
1672
1673 int __meminit vmemmap_populate(struct page *start, unsigned long nr, int node)
1674 {
1675         unsigned long vstart = (unsigned long) start;
1676         unsigned long vend = (unsigned long) (start + nr);
1677         unsigned long phys_start = (vstart - VMEMMAP_BASE);
1678         unsigned long phys_end = (vend - VMEMMAP_BASE);
1679         unsigned long addr = phys_start & VMEMMAP_CHUNK_MASK;
1680         unsigned long end = VMEMMAP_ALIGN(phys_end);
1681         unsigned long pte_base;
1682
1683         pte_base = (_PAGE_VALID | _PAGE_SZ4MB_4U |
1684                     _PAGE_CP_4U | _PAGE_CV_4U |
1685                     _PAGE_P_4U | _PAGE_W_4U);
1686         if (tlb_type == hypervisor)
1687                 pte_base = (_PAGE_VALID | _PAGE_SZ4MB_4V |
1688                             _PAGE_CP_4V | _PAGE_CV_4V |
1689                             _PAGE_P_4V | _PAGE_W_4V);
1690
1691         for (; addr < end; addr += VMEMMAP_CHUNK) {
1692                 unsigned long *vmem_pp =
1693                         vmemmap_table + (addr >> VMEMMAP_CHUNK_SHIFT);
1694                 void *block;
1695
1696                 if (!(*vmem_pp & _PAGE_VALID)) {
1697                         block = vmemmap_alloc_block(1UL << 22, node);
1698                         if (!block)
1699                                 return -ENOMEM;
1700
1701                         *vmem_pp = pte_base | __pa(block);
1702
1703                         printk(KERN_INFO "[%p-%p] page_structs=%lu "
1704                                "node=%d entry=%lu/%lu\n", start, block, nr,
1705                                node,
1706                                addr >> VMEMMAP_CHUNK_SHIFT,
1707                                VMEMMAP_SIZE >> VMEMMAP_CHUNK_SHIFT);
1708                 }
1709         }
1710         return 0;
1711 }
1712 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
1713
1714 static void prot_init_common(unsigned long page_none,
1715                              unsigned long page_shared,
1716                              unsigned long page_copy,
1717                              unsigned long page_readonly,
1718                              unsigned long page_exec_bit)
1719 {
1720         PAGE_COPY = __pgprot(page_copy);
1721         PAGE_SHARED = __pgprot(page_shared);
1722
1723         protection_map[0x0] = __pgprot(page_none);
1724         protection_map[0x1] = __pgprot(page_readonly & ~page_exec_bit);
1725         protection_map[0x2] = __pgprot(page_copy & ~page_exec_bit);
1726         protection_map[0x3] = __pgprot(page_copy & ~page_exec_bit);
1727         protection_map[0x4] = __pgprot(page_readonly);
1728         protection_map[0x5] = __pgprot(page_readonly);
1729         protection_map[0x6] = __pgprot(page_copy);
1730         protection_map[0x7] = __pgprot(page_copy);
1731         protection_map[0x8] = __pgprot(page_none);
1732         protection_map[0x9] = __pgprot(page_readonly & ~page_exec_bit);
1733         protection_map[0xa] = __pgprot(page_shared & ~page_exec_bit);
1734         protection_map[0xb] = __pgprot(page_shared & ~page_exec_bit);
1735         protection_map[0xc] = __pgprot(page_readonly);
1736         protection_map[0xd] = __pgprot(page_readonly);
1737         protection_map[0xe] = __pgprot(page_shared);
1738         protection_map[0xf] = __pgprot(page_shared);
1739 }
1740
1741 static void __init sun4u_pgprot_init(void)
1742 {
1743         unsigned long page_none, page_shared, page_copy, page_readonly;
1744         unsigned long page_exec_bit;
1745
1746         PAGE_KERNEL = __pgprot (_PAGE_PRESENT_4U | _PAGE_VALID |
1747                                 _PAGE_CACHE_4U | _PAGE_P_4U |
1748                                 __ACCESS_BITS_4U | __DIRTY_BITS_4U |
1749                                 _PAGE_EXEC_4U);
1750         PAGE_KERNEL_LOCKED = __pgprot (_PAGE_PRESENT_4U | _PAGE_VALID |
1751                                        _PAGE_CACHE_4U | _PAGE_P_4U |
1752                                        __ACCESS_BITS_4U | __DIRTY_BITS_4U |
1753                                        _PAGE_EXEC_4U | _PAGE_L_4U);
1754         PAGE_EXEC = __pgprot(_PAGE_EXEC_4U);
1755
1756         _PAGE_IE = _PAGE_IE_4U;
1757         _PAGE_E = _PAGE_E_4U;
1758         _PAGE_CACHE = _PAGE_CACHE_4U;
1759
1760         pg_iobits = (_PAGE_VALID | _PAGE_PRESENT_4U | __DIRTY_BITS_4U |
1761                      __ACCESS_BITS_4U | _PAGE_E_4U);
1762
1763 #ifdef CONFIG_DEBUG_PAGEALLOC
1764         kern_linear_pte_xor[0] = (_PAGE_VALID | _PAGE_SZBITS_4U) ^
1765                 0xfffff80000000000;
1766 #else
1767         kern_linear_pte_xor[0] = (_PAGE_VALID | _PAGE_SZ4MB_4U) ^
1768                 0xfffff80000000000;
1769 #endif
1770         kern_linear_pte_xor[0] |= (_PAGE_CP_4U | _PAGE_CV_4U |
1771                                    _PAGE_P_4U | _PAGE_W_4U);
1772
1773         /* XXX Should use 256MB on Panther. XXX */
1774         kern_linear_pte_xor[1] = kern_linear_pte_xor[0];
1775
1776         _PAGE_SZBITS = _PAGE_SZBITS_4U;
1777         _PAGE_ALL_SZ_BITS =  (_PAGE_SZ4MB_4U | _PAGE_SZ512K_4U |
1778                               _PAGE_SZ64K_4U | _PAGE_SZ8K_4U |
1779                               _PAGE_SZ32MB_4U | _PAGE_SZ256MB_4U);
1780
1781
1782         page_none = _PAGE_PRESENT_4U | _PAGE_ACCESSED_4U | _PAGE_CACHE_4U;
1783         page_shared = (_PAGE_VALID | _PAGE_PRESENT_4U | _PAGE_CACHE_4U |
1784                        __ACCESS_BITS_4U | _PAGE_WRITE_4U | _PAGE_EXEC_4U);
1785         page_copy   = (_PAGE_VALID | _PAGE_PRESENT_4U | _PAGE_CACHE_4U |
1786                        __ACCESS_BITS_4U | _PAGE_EXEC_4U);
1787         page_readonly   = (_PAGE_VALID | _PAGE_PRESENT_4U | _PAGE_CACHE_4U |
1788                            __ACCESS_BITS_4U | _PAGE_EXEC_4U);
1789
1790         page_exec_bit = _PAGE_EXEC_4U;
1791
1792         prot_init_common(page_none, page_shared, page_copy, page_readonly,
1793                          page_exec_bit);
1794 }
1795
1796 static void __init sun4v_pgprot_init(void)
1797 {
1798         unsigned long page_none, page_shared, page_copy, page_readonly;
1799         unsigned long page_exec_bit;
1800
1801         PAGE_KERNEL = __pgprot (_PAGE_PRESENT_4V | _PAGE_VALID |
1802                                 _PAGE_CACHE_4V | _PAGE_P_4V |
1803                                 __ACCESS_BITS_4V | __DIRTY_BITS_4V |
1804                                 _PAGE_EXEC_4V);
1805         PAGE_KERNEL_LOCKED = PAGE_KERNEL;
1806         PAGE_EXEC = __pgprot(_PAGE_EXEC_4V);
1807
1808         _PAGE_IE = _PAGE_IE_4V;
1809         _PAGE_E = _PAGE_E_4V;
1810         _PAGE_CACHE = _PAGE_CACHE_4V;
1811
1812 #ifdef CONFIG_DEBUG_PAGEALLOC
1813         kern_linear_pte_xor[0] = (_PAGE_VALID | _PAGE_SZBITS_4V) ^
1814                 0xfffff80000000000;
1815 #else
1816         kern_linear_pte_xor[0] = (_PAGE_VALID | _PAGE_SZ4MB_4V) ^
1817                 0xfffff80000000000;
1818 #endif
1819         kern_linear_pte_xor[0] |= (_PAGE_CP_4V | _PAGE_CV_4V |
1820                                    _PAGE_P_4V | _PAGE_W_4V);
1821
1822 #ifdef CONFIG_DEBUG_PAGEALLOC
1823         kern_linear_pte_xor[1] = (_PAGE_VALID | _PAGE_SZBITS_4V) ^
1824                 0xfffff80000000000;
1825 #else
1826         kern_linear_pte_xor[1] = (_PAGE_VALID | _PAGE_SZ256MB_4V) ^
1827                 0xfffff80000000000;
1828 #endif
1829         kern_linear_pte_xor[1] |= (_PAGE_CP_4V | _PAGE_CV_4V |
1830                                    _PAGE_P_4V | _PAGE_W_4V);
1831
1832         pg_iobits = (_PAGE_VALID | _PAGE_PRESENT_4V | __DIRTY_BITS_4V |
1833                      __ACCESS_BITS_4V | _PAGE_E_4V);
1834
1835         _PAGE_SZBITS = _PAGE_SZBITS_4V;
1836         _PAGE_ALL_SZ_BITS = (_PAGE_SZ16GB_4V | _PAGE_SZ2GB_4V |
1837                              _PAGE_SZ256MB_4V | _PAGE_SZ32MB_4V |
1838                              _PAGE_SZ4MB_4V | _PAGE_SZ512K_4V |
1839                              _PAGE_SZ64K_4V | _PAGE_SZ8K_4V);
1840
1841         page_none = _PAGE_PRESENT_4V | _PAGE_ACCESSED_4V | _PAGE_CACHE_4V;
1842         page_shared = (_PAGE_VALID | _PAGE_PRESENT_4V | _PAGE_CACHE_4V |
1843                        __ACCESS_BITS_4V | _PAGE_WRITE_4V | _PAGE_EXEC_4V);
1844         page_copy   = (_PAGE_VALID | _PAGE_PRESENT_4V | _PAGE_CACHE_4V |
1845                        __ACCESS_BITS_4V | _PAGE_EXEC_4V);
1846         page_readonly = (_PAGE_VALID | _PAGE_PRESENT_4V | _PAGE_CACHE_4V |
1847                          __ACCESS_BITS_4V | _PAGE_EXEC_4V);
1848
1849         page_exec_bit = _PAGE_EXEC_4V;
1850
1851         prot_init_common(page_none, page_shared, page_copy, page_readonly,
1852                          page_exec_bit);
1853 }
1854
1855 unsigned long pte_sz_bits(unsigned long sz)
1856 {
1857         if (tlb_type == hypervisor) {
1858                 switch (sz) {
1859                 case 8 * 1024:
1860                 default:
1861                         return _PAGE_SZ8K_4V;
1862                 case 64 * 1024:
1863                         return _PAGE_SZ64K_4V;
1864                 case 512 * 1024:
1865                         return _PAGE_SZ512K_4V;
1866                 case 4 * 1024 * 1024:
1867                         return _PAGE_SZ4MB_4V;
1868                 };
1869         } else {
1870                 switch (sz) {
1871                 case 8 * 1024:
1872                 default:
1873                         return _PAGE_SZ8K_4U;
1874                 case 64 * 1024:
1875                         return _PAGE_SZ64K_4U;
1876                 case 512 * 1024:
1877                         return _PAGE_SZ512K_4U;
1878                 case 4 * 1024 * 1024:
1879                         return _PAGE_SZ4MB_4U;
1880                 };
1881         }
1882 }
1883
1884 pte_t mk_pte_io(unsigned long page, pgprot_t prot, int space, unsigned long page_size)
1885 {
1886         pte_t pte;
1887
1888         pte_val(pte)  = page | pgprot_val(pgprot_noncached(prot));
1889         pte_val(pte) |= (((unsigned long)space) << 32);
1890         pte_val(pte) |= pte_sz_bits(page_size);
1891
1892         return pte;
1893 }
1894
1895 static unsigned long kern_large_tte(unsigned long paddr)
1896 {
1897         unsigned long val;
1898
1899         val = (_PAGE_VALID | _PAGE_SZ4MB_4U |
1900                _PAGE_CP_4U | _PAGE_CV_4U | _PAGE_P_4U |
1901                _PAGE_EXEC_4U | _PAGE_L_4U | _PAGE_W_4U);
1902         if (tlb_type == hypervisor)
1903                 val = (_PAGE_VALID | _PAGE_SZ4MB_4V |
1904                        _PAGE_CP_4V | _PAGE_CV_4V | _PAGE_P_4V |
1905                        _PAGE_EXEC_4V | _PAGE_W_4V);
1906
1907         return val | paddr;
1908 }
1909
1910 /* If not locked, zap it. */
1911 void __flush_tlb_all(void)
1912 {
1913         unsigned long pstate;
1914         int i;
1915
1916         __asm__ __volatile__("flushw\n\t"
1917                              "rdpr      %%pstate, %0\n\t"
1918                              "wrpr      %0, %1, %%pstate"
1919                              : "=r" (pstate)
1920                              : "i" (PSTATE_IE));
1921         if (tlb_type == hypervisor) {
1922                 sun4v_mmu_demap_all();
1923         } else if (tlb_type == spitfire) {
1924                 for (i = 0; i < 64; i++) {
1925                         /* Spitfire Errata #32 workaround */
1926                         /* NOTE: Always runs on spitfire, so no
1927                          *       cheetah+ page size encodings.
1928                          */
1929                         __asm__ __volatile__("stxa      %0, [%1] %2\n\t"
1930                                              "flush     %%g6"
1931                                              : /* No outputs */
1932                                              : "r" (0),
1933                                              "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
1934
1935                         if (!(spitfire_get_dtlb_data(i) & _PAGE_L_4U)) {
1936                                 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
1937                                                      "membar #Sync"
1938                                                      : /* no outputs */
1939                                                      : "r" (TLB_TAG_ACCESS), "i" (ASI_DMMU));
1940                                 spitfire_put_dtlb_data(i, 0x0UL);
1941                         }
1942
1943                         /* Spitfire Errata #32 workaround */
1944                         /* NOTE: Always runs on spitfire, so no
1945                          *       cheetah+ page size encodings.
1946                          */
1947                         __asm__ __volatile__("stxa      %0, [%1] %2\n\t"
1948                                              "flush     %%g6"
1949                                              : /* No outputs */
1950                                              : "r" (0),
1951                                              "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
1952
1953                         if (!(spitfire_get_itlb_data(i) & _PAGE_L_4U)) {
1954                                 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
1955                                                      "membar #Sync"
1956                                                      : /* no outputs */
1957                                                      : "r" (TLB_TAG_ACCESS), "i" (ASI_IMMU));
1958                                 spitfire_put_itlb_data(i, 0x0UL);
1959                         }
1960                 }
1961         } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
1962                 cheetah_flush_dtlb_all();
1963                 cheetah_flush_itlb_all();
1964         }
1965         __asm__ __volatile__("wrpr      %0, 0, %%pstate"
1966                              : : "r" (pstate));
1967 }
1968
1969 #ifdef CONFIG_MEMORY_HOTPLUG
1970
1971 void online_page(struct page *page)
1972 {
1973         ClearPageReserved(page);
1974         init_page_count(page);
1975         __free_page(page);
1976         totalram_pages++;
1977         num_physpages++;
1978 }
1979
1980 #endif /* CONFIG_MEMORY_HOTPLUG */