1 /* $Id: init.c,v 1.209 2002/02/09 19:49:31 davem Exp $
2 * arch/sparc64/mm/init.c
4 * Copyright (C) 1996-1999 David S. Miller (davem@caip.rutgers.edu)
5 * Copyright (C) 1997-1999 Jakub Jelinek (jj@sunsite.mff.cuni.cz)
8 #include <linux/config.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>
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>
21 #include <linux/seq_file.h>
22 #include <linux/kprobes.h>
23 #include <linux/cache.h>
24 #include <linux/sort.h>
27 #include <asm/system.h>
29 #include <asm/pgalloc.h>
30 #include <asm/pgtable.h>
31 #include <asm/oplib.h>
32 #include <asm/iommu.h>
34 #include <asm/uaccess.h>
35 #include <asm/mmu_context.h>
36 #include <asm/tlbflush.h>
38 #include <asm/starfire.h>
40 #include <asm/spitfire.h>
41 #include <asm/sections.h>
44 extern void device_scan(void);
48 static struct linux_prom64_registers pavail[MAX_BANKS] __initdata;
49 static struct linux_prom64_registers pavail_rescan[MAX_BANKS] __initdata;
50 static int pavail_ents __initdata;
51 static int pavail_rescan_ents __initdata;
53 static int cmp_p64(const void *a, const void *b)
55 const struct linux_prom64_registers *x = a, *y = b;
57 if (x->phys_addr > y->phys_addr)
59 if (x->phys_addr < y->phys_addr)
64 static void __init read_obp_memory(const char *property,
65 struct linux_prom64_registers *regs,
68 int node = prom_finddevice("/memory");
69 int prop_size = prom_getproplen(node, property);
72 ents = prop_size / sizeof(struct linux_prom64_registers);
73 if (ents > MAX_BANKS) {
74 prom_printf("The machine has more %s property entries than "
75 "this kernel can support (%d).\n",
80 ret = prom_getproperty(node, property, (char *) regs, prop_size);
82 prom_printf("Couldn't get %s property from /memory.\n");
88 /* Sanitize what we got from the firmware, by page aligning
91 for (i = 0; i < ents; i++) {
92 unsigned long base, size;
94 base = regs[i].phys_addr;
95 size = regs[i].reg_size;
98 if (base & ~PAGE_MASK) {
99 unsigned long new_base = PAGE_ALIGN(base);
101 size -= new_base - base;
102 if ((long) size < 0L)
106 regs[i].phys_addr = base;
107 regs[i].reg_size = size;
109 sort(regs, ents, sizeof(struct linux_prom64_registers),
113 unsigned long *sparc64_valid_addr_bitmap __read_mostly;
115 /* Ugly, but necessary... -DaveM */
116 unsigned long phys_base __read_mostly;
117 unsigned long kern_base __read_mostly;
118 unsigned long kern_size __read_mostly;
119 unsigned long pfn_base __read_mostly;
121 /* get_new_mmu_context() uses "cache + 1". */
122 DEFINE_SPINLOCK(ctx_alloc_lock);
123 unsigned long tlb_context_cache = CTX_FIRST_VERSION - 1;
124 #define CTX_BMAP_SLOTS (1UL << (CTX_NR_BITS - 6))
125 unsigned long mmu_context_bmap[CTX_BMAP_SLOTS];
127 /* References to special section boundaries */
128 extern char _start[], _end[];
130 /* Initial ramdisk setup */
131 extern unsigned long sparc_ramdisk_image64;
132 extern unsigned int sparc_ramdisk_image;
133 extern unsigned int sparc_ramdisk_size;
135 struct page *mem_map_zero __read_mostly;
137 unsigned int sparc64_highest_unlocked_tlb_ent __read_mostly;
139 unsigned long sparc64_kern_pri_context __read_mostly;
140 unsigned long sparc64_kern_pri_nuc_bits __read_mostly;
141 unsigned long sparc64_kern_sec_context __read_mostly;
145 kmem_cache_t *pgtable_cache __read_mostly;
147 static void zero_ctor(void *addr, kmem_cache_t *cache, unsigned long flags)
152 void pgtable_cache_init(void)
154 pgtable_cache = kmem_cache_create("pgtable_cache",
155 PAGE_SIZE, PAGE_SIZE,
157 SLAB_MUST_HWCACHE_ALIGN,
160 if (!pgtable_cache) {
161 prom_printf("pgtable_cache_init(): Could not create!\n");
166 #ifdef CONFIG_DEBUG_DCFLUSH
167 atomic_t dcpage_flushes = ATOMIC_INIT(0);
169 atomic_t dcpage_flushes_xcall = ATOMIC_INIT(0);
173 __inline__ void flush_dcache_page_impl(struct page *page)
175 #ifdef CONFIG_DEBUG_DCFLUSH
176 atomic_inc(&dcpage_flushes);
179 #ifdef DCACHE_ALIASING_POSSIBLE
180 __flush_dcache_page(page_address(page),
181 ((tlb_type == spitfire) &&
182 page_mapping(page) != NULL));
184 if (page_mapping(page) != NULL &&
185 tlb_type == spitfire)
186 __flush_icache_page(__pa(page_address(page)));
190 #define PG_dcache_dirty PG_arch_1
191 #define PG_dcache_cpu_shift 24
192 #define PG_dcache_cpu_mask (256 - 1)
195 #error D-cache dirty tracking and thread_info->cpu need fixing for > 256 cpus
198 #define dcache_dirty_cpu(page) \
199 (((page)->flags >> PG_dcache_cpu_shift) & PG_dcache_cpu_mask)
201 static __inline__ void set_dcache_dirty(struct page *page, int this_cpu)
203 unsigned long mask = this_cpu;
204 unsigned long non_cpu_bits;
206 non_cpu_bits = ~(PG_dcache_cpu_mask << PG_dcache_cpu_shift);
207 mask = (mask << PG_dcache_cpu_shift) | (1UL << PG_dcache_dirty);
209 __asm__ __volatile__("1:\n\t"
211 "and %%g7, %1, %%g1\n\t"
212 "or %%g1, %0, %%g1\n\t"
213 "casx [%2], %%g7, %%g1\n\t"
215 "membar #StoreLoad | #StoreStore\n\t"
216 "bne,pn %%xcc, 1b\n\t"
219 : "r" (mask), "r" (non_cpu_bits), "r" (&page->flags)
223 static __inline__ void clear_dcache_dirty_cpu(struct page *page, unsigned long cpu)
225 unsigned long mask = (1UL << PG_dcache_dirty);
227 __asm__ __volatile__("! test_and_clear_dcache_dirty\n"
230 "srlx %%g7, %4, %%g1\n\t"
231 "and %%g1, %3, %%g1\n\t"
233 "bne,pn %%icc, 2f\n\t"
234 " andn %%g7, %1, %%g1\n\t"
235 "casx [%2], %%g7, %%g1\n\t"
237 "membar #StoreLoad | #StoreStore\n\t"
238 "bne,pn %%xcc, 1b\n\t"
242 : "r" (cpu), "r" (mask), "r" (&page->flags),
243 "i" (PG_dcache_cpu_mask),
244 "i" (PG_dcache_cpu_shift)
248 static inline void tsb_insert(struct tsb *ent, unsigned long tag, unsigned long pte)
250 unsigned long tsb_addr = (unsigned long) ent;
252 if (tlb_type == cheetah_plus)
253 tsb_addr = __pa(tsb_addr);
255 __tsb_insert(tsb_addr, tag, pte);
258 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address, pte_t pte)
260 struct mm_struct *mm;
263 unsigned long pg_flags;
266 if (pfn_valid(pfn) &&
267 (page = pfn_to_page(pfn), page_mapping(page)) &&
268 ((pg_flags = page->flags) & (1UL << PG_dcache_dirty))) {
269 int cpu = ((pg_flags >> PG_dcache_cpu_shift) &
271 int this_cpu = get_cpu();
273 /* This is just to optimize away some function calls
277 flush_dcache_page_impl(page);
279 smp_flush_dcache_page_impl(page, cpu);
281 clear_dcache_dirty_cpu(page, cpu);
287 if ((pte_val(pte) & _PAGE_ALL_SZ_BITS) == _PAGE_SZBITS) {
291 tsb = &mm->context.tsb[(address >> PAGE_SHIFT) &
292 (mm->context.tsb_nentries - 1UL)];
293 tag = (address >> 22UL) | CTX_HWBITS(mm->context) << 48UL;
294 tsb_insert(tsb, tag, pte_val(pte));
298 void flush_dcache_page(struct page *page)
300 struct address_space *mapping;
303 /* Do not bother with the expensive D-cache flush if it
304 * is merely the zero page. The 'bigcore' testcase in GDB
305 * causes this case to run millions of times.
307 if (page == ZERO_PAGE(0))
310 this_cpu = get_cpu();
312 mapping = page_mapping(page);
313 if (mapping && !mapping_mapped(mapping)) {
314 int dirty = test_bit(PG_dcache_dirty, &page->flags);
316 int dirty_cpu = dcache_dirty_cpu(page);
318 if (dirty_cpu == this_cpu)
320 smp_flush_dcache_page_impl(page, dirty_cpu);
322 set_dcache_dirty(page, this_cpu);
324 /* We could delay the flush for the !page_mapping
325 * case too. But that case is for exec env/arg
326 * pages and those are %99 certainly going to get
327 * faulted into the tlb (and thus flushed) anyways.
329 flush_dcache_page_impl(page);
336 void __kprobes flush_icache_range(unsigned long start, unsigned long end)
338 /* Cheetah and Hypervisor platform cpus have coherent I-cache. */
339 if (tlb_type == spitfire) {
342 for (kaddr = start; kaddr < end; kaddr += PAGE_SIZE)
343 __flush_icache_page(__get_phys(kaddr));
347 unsigned long page_to_pfn(struct page *page)
349 return (unsigned long) ((page - mem_map) + pfn_base);
352 struct page *pfn_to_page(unsigned long pfn)
354 return (mem_map + (pfn - pfn_base));
359 printk("Mem-info:\n");
361 printk("Free swap: %6ldkB\n",
362 nr_swap_pages << (PAGE_SHIFT-10));
363 printk("%ld pages of RAM\n", num_physpages);
364 printk("%d free pages\n", nr_free_pages());
367 void mmu_info(struct seq_file *m)
369 if (tlb_type == cheetah)
370 seq_printf(m, "MMU Type\t: Cheetah\n");
371 else if (tlb_type == cheetah_plus)
372 seq_printf(m, "MMU Type\t: Cheetah+\n");
373 else if (tlb_type == spitfire)
374 seq_printf(m, "MMU Type\t: Spitfire\n");
375 else if (tlb_type == hypervisor)
376 seq_printf(m, "MMU Type\t: Hypervisor (sun4v)\n");
378 seq_printf(m, "MMU Type\t: ???\n");
380 #ifdef CONFIG_DEBUG_DCFLUSH
381 seq_printf(m, "DCPageFlushes\t: %d\n",
382 atomic_read(&dcpage_flushes));
384 seq_printf(m, "DCPageFlushesXC\t: %d\n",
385 atomic_read(&dcpage_flushes_xcall));
386 #endif /* CONFIG_SMP */
387 #endif /* CONFIG_DEBUG_DCFLUSH */
390 struct linux_prom_translation {
396 /* Exported for kernel TLB miss handling in ktlb.S */
397 struct linux_prom_translation prom_trans[512] __read_mostly;
398 unsigned int prom_trans_ents __read_mostly;
400 extern unsigned long prom_boot_page;
401 extern void prom_remap(unsigned long physpage, unsigned long virtpage, int mmu_ihandle);
402 extern int prom_get_mmu_ihandle(void);
403 extern void register_prom_callbacks(void);
405 /* Exported for SMP bootup purposes. */
406 unsigned long kern_locked_tte_data;
409 * Translate PROM's mapping we capture at boot time into physical address.
410 * The second parameter is only set from prom_callback() invocations.
412 unsigned long prom_virt_to_phys(unsigned long promva, int *error)
416 for (i = 0; i < prom_trans_ents; i++) {
417 struct linux_prom_translation *p = &prom_trans[i];
419 if (promva >= p->virt &&
420 promva < (p->virt + p->size)) {
421 unsigned long base = p->data & _PAGE_PADDR;
425 return base + (promva & (8192 - 1));
433 /* The obp translations are saved based on 8k pagesize, since obp can
434 * use a mixture of pagesizes. Misses to the LOW_OBP_ADDRESS ->
435 * HI_OBP_ADDRESS range are handled in ktlb.S.
437 static inline int in_obp_range(unsigned long vaddr)
439 return (vaddr >= LOW_OBP_ADDRESS &&
440 vaddr < HI_OBP_ADDRESS);
443 static int cmp_ptrans(const void *a, const void *b)
445 const struct linux_prom_translation *x = a, *y = b;
447 if (x->virt > y->virt)
449 if (x->virt < y->virt)
454 /* Read OBP translations property into 'prom_trans[]'. */
455 static void __init read_obp_translations(void)
457 int n, node, ents, first, last, i;
459 node = prom_finddevice("/virtual-memory");
460 n = prom_getproplen(node, "translations");
461 if (unlikely(n == 0 || n == -1)) {
462 prom_printf("prom_mappings: Couldn't get size.\n");
465 if (unlikely(n > sizeof(prom_trans))) {
466 prom_printf("prom_mappings: Size %Zd is too big.\n", n);
470 if ((n = prom_getproperty(node, "translations",
471 (char *)&prom_trans[0],
472 sizeof(prom_trans))) == -1) {
473 prom_printf("prom_mappings: Couldn't get property.\n");
477 n = n / sizeof(struct linux_prom_translation);
481 sort(prom_trans, ents, sizeof(struct linux_prom_translation),
484 /* Now kick out all the non-OBP entries. */
485 for (i = 0; i < ents; i++) {
486 if (in_obp_range(prom_trans[i].virt))
490 for (; i < ents; i++) {
491 if (!in_obp_range(prom_trans[i].virt))
496 for (i = 0; i < (last - first); i++) {
497 struct linux_prom_translation *src = &prom_trans[i + first];
498 struct linux_prom_translation *dest = &prom_trans[i];
502 for (; i < ents; i++) {
503 struct linux_prom_translation *dest = &prom_trans[i];
504 dest->virt = dest->size = dest->data = 0x0UL;
507 prom_trans_ents = last - first;
509 if (tlb_type == spitfire) {
510 /* Clear diag TTE bits. */
511 for (i = 0; i < prom_trans_ents; i++)
512 prom_trans[i].data &= ~0x0003fe0000000000UL;
516 static void __init remap_kernel(void)
518 unsigned long phys_page, tte_vaddr, tte_data;
519 int tlb_ent = sparc64_highest_locked_tlbent();
521 tte_vaddr = (unsigned long) KERNBASE;
522 phys_page = (prom_boot_mapping_phys_low >> 22UL) << 22UL;
523 tte_data = (phys_page | (_PAGE_VALID | _PAGE_SZ4MB |
524 _PAGE_CP | _PAGE_CV | _PAGE_P |
527 kern_locked_tte_data = tte_data;
529 /* Now lock us into the TLBs via OBP. */
530 prom_dtlb_load(tlb_ent, tte_data, tte_vaddr);
531 prom_itlb_load(tlb_ent, tte_data, tte_vaddr);
534 prom_dtlb_load(tlb_ent,
536 tte_vaddr + 0x400000);
537 prom_itlb_load(tlb_ent,
539 tte_vaddr + 0x400000);
541 sparc64_highest_unlocked_tlb_ent = tlb_ent - 1;
542 if (tlb_type == cheetah_plus) {
543 sparc64_kern_pri_context = (CTX_CHEETAH_PLUS_CTX0 |
544 CTX_CHEETAH_PLUS_NUC);
545 sparc64_kern_pri_nuc_bits = CTX_CHEETAH_PLUS_NUC;
546 sparc64_kern_sec_context = CTX_CHEETAH_PLUS_CTX0;
551 static void __init inherit_prom_mappings(void)
553 read_obp_translations();
555 /* Now fixup OBP's idea about where we really are mapped. */
556 prom_printf("Remapping the kernel... ");
558 prom_printf("done.\n");
560 prom_printf("Registering callbacks... ");
561 register_prom_callbacks();
562 prom_printf("done.\n");
565 void prom_world(int enter)
568 set_fs((mm_segment_t) { get_thread_current_ds() });
570 __asm__ __volatile__("flushw");
573 #ifdef DCACHE_ALIASING_POSSIBLE
574 void __flush_dcache_range(unsigned long start, unsigned long end)
578 if (tlb_type == spitfire) {
581 for (va = start; va < end; va += 32) {
582 spitfire_put_dcache_tag(va & 0x3fe0, 0x0);
586 } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
589 for (va = start; va < end; va += 32)
590 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
594 "i" (ASI_DCACHE_INVALIDATE));
597 #endif /* DCACHE_ALIASING_POSSIBLE */
599 /* If not locked, zap it. */
600 void __flush_tlb_all(void)
602 unsigned long pstate;
605 __asm__ __volatile__("flushw\n\t"
606 "rdpr %%pstate, %0\n\t"
607 "wrpr %0, %1, %%pstate"
610 if (tlb_type == spitfire) {
611 for (i = 0; i < 64; i++) {
612 /* Spitfire Errata #32 workaround */
613 /* NOTE: Always runs on spitfire, so no
614 * cheetah+ page size encodings.
616 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
620 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
622 if (!(spitfire_get_dtlb_data(i) & _PAGE_L)) {
623 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
626 : "r" (TLB_TAG_ACCESS), "i" (ASI_DMMU));
627 spitfire_put_dtlb_data(i, 0x0UL);
630 /* Spitfire Errata #32 workaround */
631 /* NOTE: Always runs on spitfire, so no
632 * cheetah+ page size encodings.
634 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
638 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
640 if (!(spitfire_get_itlb_data(i) & _PAGE_L)) {
641 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
644 : "r" (TLB_TAG_ACCESS), "i" (ASI_IMMU));
645 spitfire_put_itlb_data(i, 0x0UL);
648 } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
649 cheetah_flush_dtlb_all();
650 cheetah_flush_itlb_all();
652 __asm__ __volatile__("wrpr %0, 0, %%pstate"
656 /* Caller does TLB context flushing on local CPU if necessary.
657 * The caller also ensures that CTX_VALID(mm->context) is false.
659 * We must be careful about boundary cases so that we never
660 * let the user have CTX 0 (nucleus) or we ever use a CTX
661 * version of zero (and thus NO_CONTEXT would not be caught
662 * by version mis-match tests in mmu_context.h).
664 void get_new_mmu_context(struct mm_struct *mm)
666 unsigned long ctx, new_ctx;
667 unsigned long orig_pgsz_bits;
670 spin_lock(&ctx_alloc_lock);
671 orig_pgsz_bits = (mm->context.sparc64_ctx_val & CTX_PGSZ_MASK);
672 ctx = (tlb_context_cache + 1) & CTX_NR_MASK;
673 new_ctx = find_next_zero_bit(mmu_context_bmap, 1 << CTX_NR_BITS, ctx);
674 if (new_ctx >= (1 << CTX_NR_BITS)) {
675 new_ctx = find_next_zero_bit(mmu_context_bmap, ctx, 1);
676 if (new_ctx >= ctx) {
678 new_ctx = (tlb_context_cache & CTX_VERSION_MASK) +
681 new_ctx = CTX_FIRST_VERSION;
683 /* Don't call memset, for 16 entries that's just
686 mmu_context_bmap[0] = 3;
687 mmu_context_bmap[1] = 0;
688 mmu_context_bmap[2] = 0;
689 mmu_context_bmap[3] = 0;
690 for (i = 4; i < CTX_BMAP_SLOTS; i += 4) {
691 mmu_context_bmap[i + 0] = 0;
692 mmu_context_bmap[i + 1] = 0;
693 mmu_context_bmap[i + 2] = 0;
694 mmu_context_bmap[i + 3] = 0;
699 mmu_context_bmap[new_ctx>>6] |= (1UL << (new_ctx & 63));
700 new_ctx |= (tlb_context_cache & CTX_VERSION_MASK);
702 tlb_context_cache = new_ctx;
703 mm->context.sparc64_ctx_val = new_ctx | orig_pgsz_bits;
704 spin_unlock(&ctx_alloc_lock);
707 void sparc_ultra_dump_itlb(void)
711 if (tlb_type == spitfire) {
712 printk ("Contents of itlb: ");
713 for (slot = 0; slot < 14; slot++) printk (" ");
714 printk ("%2x:%016lx,%016lx\n",
716 spitfire_get_itlb_tag(0), spitfire_get_itlb_data(0));
717 for (slot = 1; slot < 64; slot+=3) {
718 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx %2x:%016lx,%016lx\n",
720 spitfire_get_itlb_tag(slot), spitfire_get_itlb_data(slot),
722 spitfire_get_itlb_tag(slot+1), spitfire_get_itlb_data(slot+1),
724 spitfire_get_itlb_tag(slot+2), spitfire_get_itlb_data(slot+2));
726 } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
727 printk ("Contents of itlb0:\n");
728 for (slot = 0; slot < 16; slot+=2) {
729 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
731 cheetah_get_litlb_tag(slot), cheetah_get_litlb_data(slot),
733 cheetah_get_litlb_tag(slot+1), cheetah_get_litlb_data(slot+1));
735 printk ("Contents of itlb2:\n");
736 for (slot = 0; slot < 128; slot+=2) {
737 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
739 cheetah_get_itlb_tag(slot), cheetah_get_itlb_data(slot),
741 cheetah_get_itlb_tag(slot+1), cheetah_get_itlb_data(slot+1));
746 void sparc_ultra_dump_dtlb(void)
750 if (tlb_type == spitfire) {
751 printk ("Contents of dtlb: ");
752 for (slot = 0; slot < 14; slot++) printk (" ");
753 printk ("%2x:%016lx,%016lx\n", 0,
754 spitfire_get_dtlb_tag(0), spitfire_get_dtlb_data(0));
755 for (slot = 1; slot < 64; slot+=3) {
756 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx %2x:%016lx,%016lx\n",
758 spitfire_get_dtlb_tag(slot), spitfire_get_dtlb_data(slot),
760 spitfire_get_dtlb_tag(slot+1), spitfire_get_dtlb_data(slot+1),
762 spitfire_get_dtlb_tag(slot+2), spitfire_get_dtlb_data(slot+2));
764 } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
765 printk ("Contents of dtlb0:\n");
766 for (slot = 0; slot < 16; slot+=2) {
767 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
769 cheetah_get_ldtlb_tag(slot), cheetah_get_ldtlb_data(slot),
771 cheetah_get_ldtlb_tag(slot+1), cheetah_get_ldtlb_data(slot+1));
773 printk ("Contents of dtlb2:\n");
774 for (slot = 0; slot < 512; slot+=2) {
775 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
777 cheetah_get_dtlb_tag(slot, 2), cheetah_get_dtlb_data(slot, 2),
779 cheetah_get_dtlb_tag(slot+1, 2), cheetah_get_dtlb_data(slot+1, 2));
781 if (tlb_type == cheetah_plus) {
782 printk ("Contents of dtlb3:\n");
783 for (slot = 0; slot < 512; slot+=2) {
784 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
786 cheetah_get_dtlb_tag(slot, 3), cheetah_get_dtlb_data(slot, 3),
788 cheetah_get_dtlb_tag(slot+1, 3), cheetah_get_dtlb_data(slot+1, 3));
794 static inline void spitfire_errata32(void)
796 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
800 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
803 extern unsigned long cmdline_memory_size;
805 unsigned long __init bootmem_init(unsigned long *pages_avail)
807 unsigned long bootmap_size, start_pfn, end_pfn;
808 unsigned long end_of_phys_memory = 0UL;
809 unsigned long bootmap_pfn, bytes_avail, size;
812 #ifdef CONFIG_DEBUG_BOOTMEM
813 prom_printf("bootmem_init: Scan pavail, ");
817 for (i = 0; i < pavail_ents; i++) {
818 end_of_phys_memory = pavail[i].phys_addr +
820 bytes_avail += pavail[i].reg_size;
821 if (cmdline_memory_size) {
822 if (bytes_avail > cmdline_memory_size) {
823 unsigned long slack = bytes_avail - cmdline_memory_size;
825 bytes_avail -= slack;
826 end_of_phys_memory -= slack;
828 pavail[i].reg_size -= slack;
829 if ((long)pavail[i].reg_size <= 0L) {
830 pavail[i].phys_addr = 0xdeadbeefUL;
831 pavail[i].reg_size = 0UL;
834 pavail[i+1].reg_size = 0Ul;
835 pavail[i+1].phys_addr = 0xdeadbeefUL;
843 *pages_avail = bytes_avail >> PAGE_SHIFT;
845 /* Start with page aligned address of last symbol in kernel
846 * image. The kernel is hard mapped below PAGE_OFFSET in a
847 * 4MB locked TLB translation.
849 start_pfn = PAGE_ALIGN(kern_base + kern_size) >> PAGE_SHIFT;
851 bootmap_pfn = start_pfn;
853 end_pfn = end_of_phys_memory >> PAGE_SHIFT;
855 #ifdef CONFIG_BLK_DEV_INITRD
856 /* Now have to check initial ramdisk, so that bootmap does not overwrite it */
857 if (sparc_ramdisk_image || sparc_ramdisk_image64) {
858 unsigned long ramdisk_image = sparc_ramdisk_image ?
859 sparc_ramdisk_image : sparc_ramdisk_image64;
860 if (ramdisk_image >= (unsigned long)_end - 2 * PAGE_SIZE)
861 ramdisk_image -= KERNBASE;
862 initrd_start = ramdisk_image + phys_base;
863 initrd_end = initrd_start + sparc_ramdisk_size;
864 if (initrd_end > end_of_phys_memory) {
865 printk(KERN_CRIT "initrd extends beyond end of memory "
866 "(0x%016lx > 0x%016lx)\ndisabling initrd\n",
867 initrd_end, end_of_phys_memory);
871 if (initrd_start >= (start_pfn << PAGE_SHIFT) &&
872 initrd_start < (start_pfn << PAGE_SHIFT) + 2 * PAGE_SIZE)
873 bootmap_pfn = PAGE_ALIGN (initrd_end) >> PAGE_SHIFT;
877 /* Initialize the boot-time allocator. */
878 max_pfn = max_low_pfn = end_pfn;
879 min_low_pfn = pfn_base;
881 #ifdef CONFIG_DEBUG_BOOTMEM
882 prom_printf("init_bootmem(min[%lx], bootmap[%lx], max[%lx])\n",
883 min_low_pfn, bootmap_pfn, max_low_pfn);
885 bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap_pfn, pfn_base, end_pfn);
887 /* Now register the available physical memory with the
890 for (i = 0; i < pavail_ents; i++) {
891 #ifdef CONFIG_DEBUG_BOOTMEM
892 prom_printf("free_bootmem(pavail:%d): base[%lx] size[%lx]\n",
893 i, pavail[i].phys_addr, pavail[i].reg_size);
895 free_bootmem(pavail[i].phys_addr, pavail[i].reg_size);
898 #ifdef CONFIG_BLK_DEV_INITRD
900 size = initrd_end - initrd_start;
902 /* Resert the initrd image area. */
903 #ifdef CONFIG_DEBUG_BOOTMEM
904 prom_printf("reserve_bootmem(initrd): base[%llx] size[%lx]\n",
905 initrd_start, initrd_end);
907 reserve_bootmem(initrd_start, size);
908 *pages_avail -= PAGE_ALIGN(size) >> PAGE_SHIFT;
910 initrd_start += PAGE_OFFSET;
911 initrd_end += PAGE_OFFSET;
914 /* Reserve the kernel text/data/bss. */
915 #ifdef CONFIG_DEBUG_BOOTMEM
916 prom_printf("reserve_bootmem(kernel): base[%lx] size[%lx]\n", kern_base, kern_size);
918 reserve_bootmem(kern_base, kern_size);
919 *pages_avail -= PAGE_ALIGN(kern_size) >> PAGE_SHIFT;
921 /* Reserve the bootmem map. We do not account for it
922 * in pages_avail because we will release that memory
923 * in free_all_bootmem.
926 #ifdef CONFIG_DEBUG_BOOTMEM
927 prom_printf("reserve_bootmem(bootmap): base[%lx] size[%lx]\n",
928 (bootmap_pfn << PAGE_SHIFT), size);
930 reserve_bootmem((bootmap_pfn << PAGE_SHIFT), size);
931 *pages_avail -= PAGE_ALIGN(size) >> PAGE_SHIFT;
936 #ifdef CONFIG_DEBUG_PAGEALLOC
937 static unsigned long kernel_map_range(unsigned long pstart, unsigned long pend, pgprot_t prot)
939 unsigned long vstart = PAGE_OFFSET + pstart;
940 unsigned long vend = PAGE_OFFSET + pend;
941 unsigned long alloc_bytes = 0UL;
943 if ((vstart & ~PAGE_MASK) || (vend & ~PAGE_MASK)) {
944 prom_printf("kernel_map: Unaligned physmem[%lx:%lx]\n",
949 while (vstart < vend) {
950 unsigned long this_end, paddr = __pa(vstart);
951 pgd_t *pgd = pgd_offset_k(vstart);
956 pud = pud_offset(pgd, vstart);
957 if (pud_none(*pud)) {
960 new = __alloc_bootmem(PAGE_SIZE, PAGE_SIZE, PAGE_SIZE);
961 alloc_bytes += PAGE_SIZE;
962 pud_populate(&init_mm, pud, new);
965 pmd = pmd_offset(pud, vstart);
966 if (!pmd_present(*pmd)) {
969 new = __alloc_bootmem(PAGE_SIZE, PAGE_SIZE, PAGE_SIZE);
970 alloc_bytes += PAGE_SIZE;
971 pmd_populate_kernel(&init_mm, pmd, new);
974 pte = pte_offset_kernel(pmd, vstart);
975 this_end = (vstart + PMD_SIZE) & PMD_MASK;
979 while (vstart < this_end) {
980 pte_val(*pte) = (paddr | pgprot_val(prot));
991 static struct linux_prom64_registers pall[MAX_BANKS] __initdata;
992 static int pall_ents __initdata;
994 extern unsigned int kvmap_linear_patch[1];
996 static void __init kernel_physical_mapping_init(void)
998 unsigned long i, mem_alloced = 0UL;
1000 read_obp_memory("reg", &pall[0], &pall_ents);
1002 for (i = 0; i < pall_ents; i++) {
1003 unsigned long phys_start, phys_end;
1005 phys_start = pall[i].phys_addr;
1006 phys_end = phys_start + pall[i].reg_size;
1007 mem_alloced += kernel_map_range(phys_start, phys_end,
1011 printk("Allocated %ld bytes for kernel page tables.\n",
1014 kvmap_linear_patch[0] = 0x01000000; /* nop */
1015 flushi(&kvmap_linear_patch[0]);
1020 void kernel_map_pages(struct page *page, int numpages, int enable)
1022 unsigned long phys_start = page_to_pfn(page) << PAGE_SHIFT;
1023 unsigned long phys_end = phys_start + (numpages * PAGE_SIZE);
1025 kernel_map_range(phys_start, phys_end,
1026 (enable ? PAGE_KERNEL : __pgprot(0)));
1028 flush_tsb_kernel_range(PAGE_OFFSET + phys_start,
1029 PAGE_OFFSET + phys_end);
1031 /* we should perform an IPI and flush all tlbs,
1032 * but that can deadlock->flush only current cpu.
1034 __flush_tlb_kernel_range(PAGE_OFFSET + phys_start,
1035 PAGE_OFFSET + phys_end);
1039 unsigned long __init find_ecache_flush_span(unsigned long size)
1043 for (i = 0; i < pavail_ents; i++) {
1044 if (pavail[i].reg_size >= size)
1045 return pavail[i].phys_addr;
1051 static void __init tsb_phys_patch(void)
1053 struct tsb_phys_patch_entry *p;
1055 p = &__tsb_phys_patch;
1056 while (p < &__tsb_phys_patch_end) {
1057 unsigned long addr = p->addr;
1059 *(unsigned int *) addr = p->insn;
1061 __asm__ __volatile__("flush %0"
1069 /* paging_init() sets up the page tables */
1071 extern void cheetah_ecache_flush_init(void);
1073 static unsigned long last_valid_pfn;
1074 pgd_t swapper_pg_dir[2048];
1076 void __init paging_init(void)
1078 unsigned long end_pfn, pages_avail, shift;
1079 unsigned long real_end, i;
1081 if (tlb_type == cheetah_plus)
1084 /* Find available physical memory... */
1085 read_obp_memory("available", &pavail[0], &pavail_ents);
1087 phys_base = 0xffffffffffffffffUL;
1088 for (i = 0; i < pavail_ents; i++)
1089 phys_base = min(phys_base, pavail[i].phys_addr);
1091 pfn_base = phys_base >> PAGE_SHIFT;
1093 kern_base = (prom_boot_mapping_phys_low >> 22UL) << 22UL;
1094 kern_size = (unsigned long)&_end - (unsigned long)KERNBASE;
1096 set_bit(0, mmu_context_bmap);
1098 shift = kern_base + PAGE_OFFSET - ((unsigned long)KERNBASE);
1100 real_end = (unsigned long)_end;
1101 if ((real_end > ((unsigned long)KERNBASE + 0x400000)))
1103 if ((real_end > ((unsigned long)KERNBASE + 0x800000))) {
1104 prom_printf("paging_init: Kernel > 8MB, too large.\n");
1108 /* Set kernel pgd to upper alias so physical page computations
1111 init_mm.pgd += ((shift) / (sizeof(pgd_t)));
1113 memset(swapper_low_pmd_dir, 0, sizeof(swapper_low_pmd_dir));
1115 /* Now can init the kernel/bad page tables. */
1116 pud_set(pud_offset(&swapper_pg_dir[0], 0),
1117 swapper_low_pmd_dir + (shift / sizeof(pgd_t)));
1119 inherit_prom_mappings();
1121 /* Ok, we can use our TLB miss and window trap handlers safely. */
1126 /* Setup bootmem... */
1128 last_valid_pfn = end_pfn = bootmem_init(&pages_avail);
1130 #ifdef CONFIG_DEBUG_PAGEALLOC
1131 kernel_physical_mapping_init();
1135 unsigned long zones_size[MAX_NR_ZONES];
1136 unsigned long zholes_size[MAX_NR_ZONES];
1137 unsigned long npages;
1140 for (znum = 0; znum < MAX_NR_ZONES; znum++)
1141 zones_size[znum] = zholes_size[znum] = 0;
1143 npages = end_pfn - pfn_base;
1144 zones_size[ZONE_DMA] = npages;
1145 zholes_size[ZONE_DMA] = npages - pages_avail;
1147 free_area_init_node(0, &contig_page_data, zones_size,
1148 phys_base >> PAGE_SHIFT, zholes_size);
1154 static void __init taint_real_pages(void)
1158 read_obp_memory("available", &pavail_rescan[0], &pavail_rescan_ents);
1160 /* Find changes discovered in the physmem available rescan and
1161 * reserve the lost portions in the bootmem maps.
1163 for (i = 0; i < pavail_ents; i++) {
1164 unsigned long old_start, old_end;
1166 old_start = pavail[i].phys_addr;
1167 old_end = old_start +
1169 while (old_start < old_end) {
1172 for (n = 0; pavail_rescan_ents; n++) {
1173 unsigned long new_start, new_end;
1175 new_start = pavail_rescan[n].phys_addr;
1176 new_end = new_start +
1177 pavail_rescan[n].reg_size;
1179 if (new_start <= old_start &&
1180 new_end >= (old_start + PAGE_SIZE)) {
1181 set_bit(old_start >> 22,
1182 sparc64_valid_addr_bitmap);
1186 reserve_bootmem(old_start, PAGE_SIZE);
1189 old_start += PAGE_SIZE;
1194 void __init mem_init(void)
1196 unsigned long codepages, datapages, initpages;
1197 unsigned long addr, last;
1200 i = last_valid_pfn >> ((22 - PAGE_SHIFT) + 6);
1202 sparc64_valid_addr_bitmap = (unsigned long *) alloc_bootmem(i << 3);
1203 if (sparc64_valid_addr_bitmap == NULL) {
1204 prom_printf("mem_init: Cannot alloc valid_addr_bitmap.\n");
1207 memset(sparc64_valid_addr_bitmap, 0, i << 3);
1209 addr = PAGE_OFFSET + kern_base;
1210 last = PAGE_ALIGN(kern_size) + addr;
1211 while (addr < last) {
1212 set_bit(__pa(addr) >> 22, sparc64_valid_addr_bitmap);
1218 max_mapnr = last_valid_pfn - pfn_base;
1219 high_memory = __va(last_valid_pfn << PAGE_SHIFT);
1221 #ifdef CONFIG_DEBUG_BOOTMEM
1222 prom_printf("mem_init: Calling free_all_bootmem().\n");
1224 totalram_pages = num_physpages = free_all_bootmem() - 1;
1227 * Set up the zero page, mark it reserved, so that page count
1228 * is not manipulated when freeing the page from user ptes.
1230 mem_map_zero = alloc_pages(GFP_KERNEL|__GFP_ZERO, 0);
1231 if (mem_map_zero == NULL) {
1232 prom_printf("paging_init: Cannot alloc zero page.\n");
1235 SetPageReserved(mem_map_zero);
1237 codepages = (((unsigned long) _etext) - ((unsigned long) _start));
1238 codepages = PAGE_ALIGN(codepages) >> PAGE_SHIFT;
1239 datapages = (((unsigned long) _edata) - ((unsigned long) _etext));
1240 datapages = PAGE_ALIGN(datapages) >> PAGE_SHIFT;
1241 initpages = (((unsigned long) __init_end) - ((unsigned long) __init_begin));
1242 initpages = PAGE_ALIGN(initpages) >> PAGE_SHIFT;
1244 printk("Memory: %uk available (%ldk kernel code, %ldk data, %ldk init) [%016lx,%016lx]\n",
1245 nr_free_pages() << (PAGE_SHIFT-10),
1246 codepages << (PAGE_SHIFT-10),
1247 datapages << (PAGE_SHIFT-10),
1248 initpages << (PAGE_SHIFT-10),
1249 PAGE_OFFSET, (last_valid_pfn << PAGE_SHIFT));
1251 if (tlb_type == cheetah || tlb_type == cheetah_plus)
1252 cheetah_ecache_flush_init();
1255 void free_initmem(void)
1257 unsigned long addr, initend;
1260 * The init section is aligned to 8k in vmlinux.lds. Page align for >8k pagesizes.
1262 addr = PAGE_ALIGN((unsigned long)(__init_begin));
1263 initend = (unsigned long)(__init_end) & PAGE_MASK;
1264 for (; addr < initend; addr += PAGE_SIZE) {
1269 ((unsigned long) __va(kern_base)) -
1270 ((unsigned long) KERNBASE));
1271 memset((void *)addr, 0xcc, PAGE_SIZE);
1272 p = virt_to_page(page);
1274 ClearPageReserved(p);
1275 set_page_count(p, 1);
1282 #ifdef CONFIG_BLK_DEV_INITRD
1283 void free_initrd_mem(unsigned long start, unsigned long end)
1286 printk ("Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
1287 for (; start < end; start += PAGE_SIZE) {
1288 struct page *p = virt_to_page(start);
1290 ClearPageReserved(p);
1291 set_page_count(p, 1);