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;
264 unsigned long mm_rss;
267 if (pfn_valid(pfn) &&
268 (page = pfn_to_page(pfn), page_mapping(page)) &&
269 ((pg_flags = page->flags) & (1UL << PG_dcache_dirty))) {
270 int cpu = ((pg_flags >> PG_dcache_cpu_shift) &
272 int this_cpu = get_cpu();
274 /* This is just to optimize away some function calls
278 flush_dcache_page_impl(page);
280 smp_flush_dcache_page_impl(page, cpu);
282 clear_dcache_dirty_cpu(page, cpu);
288 mm_rss = get_mm_rss(mm);
289 if (mm_rss >= mm->context.tsb_rss_limit)
290 tsb_grow(mm, mm_rss, GFP_ATOMIC);
292 if ((pte_val(pte) & _PAGE_ALL_SZ_BITS) == _PAGE_SZBITS) {
296 tsb = &mm->context.tsb[(address >> PAGE_SHIFT) &
297 (mm->context.tsb_nentries - 1UL)];
298 tag = (address >> 22UL) | CTX_HWBITS(mm->context) << 48UL;
299 tsb_insert(tsb, tag, pte_val(pte));
303 void flush_dcache_page(struct page *page)
305 struct address_space *mapping;
308 /* Do not bother with the expensive D-cache flush if it
309 * is merely the zero page. The 'bigcore' testcase in GDB
310 * causes this case to run millions of times.
312 if (page == ZERO_PAGE(0))
315 this_cpu = get_cpu();
317 mapping = page_mapping(page);
318 if (mapping && !mapping_mapped(mapping)) {
319 int dirty = test_bit(PG_dcache_dirty, &page->flags);
321 int dirty_cpu = dcache_dirty_cpu(page);
323 if (dirty_cpu == this_cpu)
325 smp_flush_dcache_page_impl(page, dirty_cpu);
327 set_dcache_dirty(page, this_cpu);
329 /* We could delay the flush for the !page_mapping
330 * case too. But that case is for exec env/arg
331 * pages and those are %99 certainly going to get
332 * faulted into the tlb (and thus flushed) anyways.
334 flush_dcache_page_impl(page);
341 void __kprobes flush_icache_range(unsigned long start, unsigned long end)
343 /* Cheetah has coherent I-cache. */
344 if (tlb_type == spitfire) {
347 for (kaddr = start; kaddr < end; kaddr += PAGE_SIZE)
348 __flush_icache_page(__get_phys(kaddr));
352 unsigned long page_to_pfn(struct page *page)
354 return (unsigned long) ((page - mem_map) + pfn_base);
357 struct page *pfn_to_page(unsigned long pfn)
359 return (mem_map + (pfn - pfn_base));
364 printk("Mem-info:\n");
366 printk("Free swap: %6ldkB\n",
367 nr_swap_pages << (PAGE_SHIFT-10));
368 printk("%ld pages of RAM\n", num_physpages);
369 printk("%d free pages\n", nr_free_pages());
372 void mmu_info(struct seq_file *m)
374 if (tlb_type == cheetah)
375 seq_printf(m, "MMU Type\t: Cheetah\n");
376 else if (tlb_type == cheetah_plus)
377 seq_printf(m, "MMU Type\t: Cheetah+\n");
378 else if (tlb_type == spitfire)
379 seq_printf(m, "MMU Type\t: Spitfire\n");
381 seq_printf(m, "MMU Type\t: ???\n");
383 #ifdef CONFIG_DEBUG_DCFLUSH
384 seq_printf(m, "DCPageFlushes\t: %d\n",
385 atomic_read(&dcpage_flushes));
387 seq_printf(m, "DCPageFlushesXC\t: %d\n",
388 atomic_read(&dcpage_flushes_xcall));
389 #endif /* CONFIG_SMP */
390 #endif /* CONFIG_DEBUG_DCFLUSH */
393 struct linux_prom_translation {
399 /* Exported for kernel TLB miss handling in ktlb.S */
400 struct linux_prom_translation prom_trans[512] __read_mostly;
401 unsigned int prom_trans_ents __read_mostly;
403 extern unsigned long prom_boot_page;
404 extern void prom_remap(unsigned long physpage, unsigned long virtpage, int mmu_ihandle);
405 extern int prom_get_mmu_ihandle(void);
406 extern void register_prom_callbacks(void);
408 /* Exported for SMP bootup purposes. */
409 unsigned long kern_locked_tte_data;
412 * Translate PROM's mapping we capture at boot time into physical address.
413 * The second parameter is only set from prom_callback() invocations.
415 unsigned long prom_virt_to_phys(unsigned long promva, int *error)
419 for (i = 0; i < prom_trans_ents; i++) {
420 struct linux_prom_translation *p = &prom_trans[i];
422 if (promva >= p->virt &&
423 promva < (p->virt + p->size)) {
424 unsigned long base = p->data & _PAGE_PADDR;
428 return base + (promva & (8192 - 1));
436 /* The obp translations are saved based on 8k pagesize, since obp can
437 * use a mixture of pagesizes. Misses to the LOW_OBP_ADDRESS ->
438 * HI_OBP_ADDRESS range are handled in ktlb.S.
440 static inline int in_obp_range(unsigned long vaddr)
442 return (vaddr >= LOW_OBP_ADDRESS &&
443 vaddr < HI_OBP_ADDRESS);
446 static int cmp_ptrans(const void *a, const void *b)
448 const struct linux_prom_translation *x = a, *y = b;
450 if (x->virt > y->virt)
452 if (x->virt < y->virt)
457 /* Read OBP translations property into 'prom_trans[]'. */
458 static void __init read_obp_translations(void)
460 int n, node, ents, first, last, i;
462 node = prom_finddevice("/virtual-memory");
463 n = prom_getproplen(node, "translations");
464 if (unlikely(n == 0 || n == -1)) {
465 prom_printf("prom_mappings: Couldn't get size.\n");
468 if (unlikely(n > sizeof(prom_trans))) {
469 prom_printf("prom_mappings: Size %Zd is too big.\n", n);
473 if ((n = prom_getproperty(node, "translations",
474 (char *)&prom_trans[0],
475 sizeof(prom_trans))) == -1) {
476 prom_printf("prom_mappings: Couldn't get property.\n");
480 n = n / sizeof(struct linux_prom_translation);
484 sort(prom_trans, ents, sizeof(struct linux_prom_translation),
487 /* Now kick out all the non-OBP entries. */
488 for (i = 0; i < ents; i++) {
489 if (in_obp_range(prom_trans[i].virt))
493 for (; i < ents; i++) {
494 if (!in_obp_range(prom_trans[i].virt))
499 for (i = 0; i < (last - first); i++) {
500 struct linux_prom_translation *src = &prom_trans[i + first];
501 struct linux_prom_translation *dest = &prom_trans[i];
505 for (; i < ents; i++) {
506 struct linux_prom_translation *dest = &prom_trans[i];
507 dest->virt = dest->size = dest->data = 0x0UL;
510 prom_trans_ents = last - first;
512 if (tlb_type == spitfire) {
513 /* Clear diag TTE bits. */
514 for (i = 0; i < prom_trans_ents; i++)
515 prom_trans[i].data &= ~0x0003fe0000000000UL;
519 static void __init remap_kernel(void)
521 unsigned long phys_page, tte_vaddr, tte_data;
522 int tlb_ent = sparc64_highest_locked_tlbent();
524 tte_vaddr = (unsigned long) KERNBASE;
525 phys_page = (prom_boot_mapping_phys_low >> 22UL) << 22UL;
526 tte_data = (phys_page | (_PAGE_VALID | _PAGE_SZ4MB |
527 _PAGE_CP | _PAGE_CV | _PAGE_P |
530 kern_locked_tte_data = tte_data;
532 /* Now lock us into the TLBs via OBP. */
533 prom_dtlb_load(tlb_ent, tte_data, tte_vaddr);
534 prom_itlb_load(tlb_ent, tte_data, tte_vaddr);
537 prom_dtlb_load(tlb_ent,
539 tte_vaddr + 0x400000);
540 prom_itlb_load(tlb_ent,
542 tte_vaddr + 0x400000);
544 sparc64_highest_unlocked_tlb_ent = tlb_ent - 1;
545 if (tlb_type == cheetah_plus) {
546 sparc64_kern_pri_context = (CTX_CHEETAH_PLUS_CTX0 |
547 CTX_CHEETAH_PLUS_NUC);
548 sparc64_kern_pri_nuc_bits = CTX_CHEETAH_PLUS_NUC;
549 sparc64_kern_sec_context = CTX_CHEETAH_PLUS_CTX0;
554 static void __init inherit_prom_mappings(void)
556 read_obp_translations();
558 /* Now fixup OBP's idea about where we really are mapped. */
559 prom_printf("Remapping the kernel... ");
561 prom_printf("done.\n");
563 prom_printf("Registering callbacks... ");
564 register_prom_callbacks();
565 prom_printf("done.\n");
568 void prom_world(int enter)
571 set_fs((mm_segment_t) { get_thread_current_ds() });
573 __asm__ __volatile__("flushw");
576 #ifdef DCACHE_ALIASING_POSSIBLE
577 void __flush_dcache_range(unsigned long start, unsigned long end)
581 if (tlb_type == spitfire) {
584 for (va = start; va < end; va += 32) {
585 spitfire_put_dcache_tag(va & 0x3fe0, 0x0);
592 for (va = start; va < end; va += 32)
593 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
597 "i" (ASI_DCACHE_INVALIDATE));
600 #endif /* DCACHE_ALIASING_POSSIBLE */
602 /* If not locked, zap it. */
603 void __flush_tlb_all(void)
605 unsigned long pstate;
608 __asm__ __volatile__("flushw\n\t"
609 "rdpr %%pstate, %0\n\t"
610 "wrpr %0, %1, %%pstate"
613 if (tlb_type == spitfire) {
614 for (i = 0; i < 64; i++) {
615 /* Spitfire Errata #32 workaround */
616 /* NOTE: Always runs on spitfire, so no
617 * cheetah+ page size encodings.
619 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
623 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
625 if (!(spitfire_get_dtlb_data(i) & _PAGE_L)) {
626 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
629 : "r" (TLB_TAG_ACCESS), "i" (ASI_DMMU));
630 spitfire_put_dtlb_data(i, 0x0UL);
633 /* Spitfire Errata #32 workaround */
634 /* NOTE: Always runs on spitfire, so no
635 * cheetah+ page size encodings.
637 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
641 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
643 if (!(spitfire_get_itlb_data(i) & _PAGE_L)) {
644 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
647 : "r" (TLB_TAG_ACCESS), "i" (ASI_IMMU));
648 spitfire_put_itlb_data(i, 0x0UL);
651 } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
652 cheetah_flush_dtlb_all();
653 cheetah_flush_itlb_all();
655 __asm__ __volatile__("wrpr %0, 0, %%pstate"
659 /* Caller does TLB context flushing on local CPU if necessary.
660 * The caller also ensures that CTX_VALID(mm->context) is false.
662 * We must be careful about boundary cases so that we never
663 * let the user have CTX 0 (nucleus) or we ever use a CTX
664 * version of zero (and thus NO_CONTEXT would not be caught
665 * by version mis-match tests in mmu_context.h).
667 void get_new_mmu_context(struct mm_struct *mm)
669 unsigned long ctx, new_ctx;
670 unsigned long orig_pgsz_bits;
673 spin_lock(&ctx_alloc_lock);
674 orig_pgsz_bits = (mm->context.sparc64_ctx_val & CTX_PGSZ_MASK);
675 ctx = (tlb_context_cache + 1) & CTX_NR_MASK;
676 new_ctx = find_next_zero_bit(mmu_context_bmap, 1 << CTX_NR_BITS, ctx);
677 if (new_ctx >= (1 << CTX_NR_BITS)) {
678 new_ctx = find_next_zero_bit(mmu_context_bmap, ctx, 1);
679 if (new_ctx >= ctx) {
681 new_ctx = (tlb_context_cache & CTX_VERSION_MASK) +
684 new_ctx = CTX_FIRST_VERSION;
686 /* Don't call memset, for 16 entries that's just
689 mmu_context_bmap[0] = 3;
690 mmu_context_bmap[1] = 0;
691 mmu_context_bmap[2] = 0;
692 mmu_context_bmap[3] = 0;
693 for (i = 4; i < CTX_BMAP_SLOTS; i += 4) {
694 mmu_context_bmap[i + 0] = 0;
695 mmu_context_bmap[i + 1] = 0;
696 mmu_context_bmap[i + 2] = 0;
697 mmu_context_bmap[i + 3] = 0;
702 mmu_context_bmap[new_ctx>>6] |= (1UL << (new_ctx & 63));
703 new_ctx |= (tlb_context_cache & CTX_VERSION_MASK);
705 tlb_context_cache = new_ctx;
706 mm->context.sparc64_ctx_val = new_ctx | orig_pgsz_bits;
707 spin_unlock(&ctx_alloc_lock);
710 void sparc_ultra_dump_itlb(void)
714 if (tlb_type == spitfire) {
715 printk ("Contents of itlb: ");
716 for (slot = 0; slot < 14; slot++) printk (" ");
717 printk ("%2x:%016lx,%016lx\n",
719 spitfire_get_itlb_tag(0), spitfire_get_itlb_data(0));
720 for (slot = 1; slot < 64; slot+=3) {
721 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx %2x:%016lx,%016lx\n",
723 spitfire_get_itlb_tag(slot), spitfire_get_itlb_data(slot),
725 spitfire_get_itlb_tag(slot+1), spitfire_get_itlb_data(slot+1),
727 spitfire_get_itlb_tag(slot+2), spitfire_get_itlb_data(slot+2));
729 } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
730 printk ("Contents of itlb0:\n");
731 for (slot = 0; slot < 16; slot+=2) {
732 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
734 cheetah_get_litlb_tag(slot), cheetah_get_litlb_data(slot),
736 cheetah_get_litlb_tag(slot+1), cheetah_get_litlb_data(slot+1));
738 printk ("Contents of itlb2:\n");
739 for (slot = 0; slot < 128; slot+=2) {
740 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
742 cheetah_get_itlb_tag(slot), cheetah_get_itlb_data(slot),
744 cheetah_get_itlb_tag(slot+1), cheetah_get_itlb_data(slot+1));
749 void sparc_ultra_dump_dtlb(void)
753 if (tlb_type == spitfire) {
754 printk ("Contents of dtlb: ");
755 for (slot = 0; slot < 14; slot++) printk (" ");
756 printk ("%2x:%016lx,%016lx\n", 0,
757 spitfire_get_dtlb_tag(0), spitfire_get_dtlb_data(0));
758 for (slot = 1; slot < 64; slot+=3) {
759 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx %2x:%016lx,%016lx\n",
761 spitfire_get_dtlb_tag(slot), spitfire_get_dtlb_data(slot),
763 spitfire_get_dtlb_tag(slot+1), spitfire_get_dtlb_data(slot+1),
765 spitfire_get_dtlb_tag(slot+2), spitfire_get_dtlb_data(slot+2));
767 } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
768 printk ("Contents of dtlb0:\n");
769 for (slot = 0; slot < 16; slot+=2) {
770 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
772 cheetah_get_ldtlb_tag(slot), cheetah_get_ldtlb_data(slot),
774 cheetah_get_ldtlb_tag(slot+1), cheetah_get_ldtlb_data(slot+1));
776 printk ("Contents of dtlb2:\n");
777 for (slot = 0; slot < 512; slot+=2) {
778 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
780 cheetah_get_dtlb_tag(slot, 2), cheetah_get_dtlb_data(slot, 2),
782 cheetah_get_dtlb_tag(slot+1, 2), cheetah_get_dtlb_data(slot+1, 2));
784 if (tlb_type == cheetah_plus) {
785 printk ("Contents of dtlb3:\n");
786 for (slot = 0; slot < 512; slot+=2) {
787 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
789 cheetah_get_dtlb_tag(slot, 3), cheetah_get_dtlb_data(slot, 3),
791 cheetah_get_dtlb_tag(slot+1, 3), cheetah_get_dtlb_data(slot+1, 3));
797 static inline void spitfire_errata32(void)
799 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
803 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
806 extern unsigned long cmdline_memory_size;
808 unsigned long __init bootmem_init(unsigned long *pages_avail)
810 unsigned long bootmap_size, start_pfn, end_pfn;
811 unsigned long end_of_phys_memory = 0UL;
812 unsigned long bootmap_pfn, bytes_avail, size;
815 #ifdef CONFIG_DEBUG_BOOTMEM
816 prom_printf("bootmem_init: Scan pavail, ");
820 for (i = 0; i < pavail_ents; i++) {
821 end_of_phys_memory = pavail[i].phys_addr +
823 bytes_avail += pavail[i].reg_size;
824 if (cmdline_memory_size) {
825 if (bytes_avail > cmdline_memory_size) {
826 unsigned long slack = bytes_avail - cmdline_memory_size;
828 bytes_avail -= slack;
829 end_of_phys_memory -= slack;
831 pavail[i].reg_size -= slack;
832 if ((long)pavail[i].reg_size <= 0L) {
833 pavail[i].phys_addr = 0xdeadbeefUL;
834 pavail[i].reg_size = 0UL;
837 pavail[i+1].reg_size = 0Ul;
838 pavail[i+1].phys_addr = 0xdeadbeefUL;
846 *pages_avail = bytes_avail >> PAGE_SHIFT;
848 /* Start with page aligned address of last symbol in kernel
849 * image. The kernel is hard mapped below PAGE_OFFSET in a
850 * 4MB locked TLB translation.
852 start_pfn = PAGE_ALIGN(kern_base + kern_size) >> PAGE_SHIFT;
854 bootmap_pfn = start_pfn;
856 end_pfn = end_of_phys_memory >> PAGE_SHIFT;
858 #ifdef CONFIG_BLK_DEV_INITRD
859 /* Now have to check initial ramdisk, so that bootmap does not overwrite it */
860 if (sparc_ramdisk_image || sparc_ramdisk_image64) {
861 unsigned long ramdisk_image = sparc_ramdisk_image ?
862 sparc_ramdisk_image : sparc_ramdisk_image64;
863 if (ramdisk_image >= (unsigned long)_end - 2 * PAGE_SIZE)
864 ramdisk_image -= KERNBASE;
865 initrd_start = ramdisk_image + phys_base;
866 initrd_end = initrd_start + sparc_ramdisk_size;
867 if (initrd_end > end_of_phys_memory) {
868 printk(KERN_CRIT "initrd extends beyond end of memory "
869 "(0x%016lx > 0x%016lx)\ndisabling initrd\n",
870 initrd_end, end_of_phys_memory);
874 if (initrd_start >= (start_pfn << PAGE_SHIFT) &&
875 initrd_start < (start_pfn << PAGE_SHIFT) + 2 * PAGE_SIZE)
876 bootmap_pfn = PAGE_ALIGN (initrd_end) >> PAGE_SHIFT;
880 /* Initialize the boot-time allocator. */
881 max_pfn = max_low_pfn = end_pfn;
882 min_low_pfn = pfn_base;
884 #ifdef CONFIG_DEBUG_BOOTMEM
885 prom_printf("init_bootmem(min[%lx], bootmap[%lx], max[%lx])\n",
886 min_low_pfn, bootmap_pfn, max_low_pfn);
888 bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap_pfn, pfn_base, end_pfn);
890 /* Now register the available physical memory with the
893 for (i = 0; i < pavail_ents; i++) {
894 #ifdef CONFIG_DEBUG_BOOTMEM
895 prom_printf("free_bootmem(pavail:%d): base[%lx] size[%lx]\n",
896 i, pavail[i].phys_addr, pavail[i].reg_size);
898 free_bootmem(pavail[i].phys_addr, pavail[i].reg_size);
901 #ifdef CONFIG_BLK_DEV_INITRD
903 size = initrd_end - initrd_start;
905 /* Resert the initrd image area. */
906 #ifdef CONFIG_DEBUG_BOOTMEM
907 prom_printf("reserve_bootmem(initrd): base[%llx] size[%lx]\n",
908 initrd_start, initrd_end);
910 reserve_bootmem(initrd_start, size);
911 *pages_avail -= PAGE_ALIGN(size) >> PAGE_SHIFT;
913 initrd_start += PAGE_OFFSET;
914 initrd_end += PAGE_OFFSET;
917 /* Reserve the kernel text/data/bss. */
918 #ifdef CONFIG_DEBUG_BOOTMEM
919 prom_printf("reserve_bootmem(kernel): base[%lx] size[%lx]\n", kern_base, kern_size);
921 reserve_bootmem(kern_base, kern_size);
922 *pages_avail -= PAGE_ALIGN(kern_size) >> PAGE_SHIFT;
924 /* Reserve the bootmem map. We do not account for it
925 * in pages_avail because we will release that memory
926 * in free_all_bootmem.
929 #ifdef CONFIG_DEBUG_BOOTMEM
930 prom_printf("reserve_bootmem(bootmap): base[%lx] size[%lx]\n",
931 (bootmap_pfn << PAGE_SHIFT), size);
933 reserve_bootmem((bootmap_pfn << PAGE_SHIFT), size);
934 *pages_avail -= PAGE_ALIGN(size) >> PAGE_SHIFT;
939 #ifdef CONFIG_DEBUG_PAGEALLOC
940 static unsigned long kernel_map_range(unsigned long pstart, unsigned long pend, pgprot_t prot)
942 unsigned long vstart = PAGE_OFFSET + pstart;
943 unsigned long vend = PAGE_OFFSET + pend;
944 unsigned long alloc_bytes = 0UL;
946 if ((vstart & ~PAGE_MASK) || (vend & ~PAGE_MASK)) {
947 prom_printf("kernel_map: Unaligned physmem[%lx:%lx]\n",
952 while (vstart < vend) {
953 unsigned long this_end, paddr = __pa(vstart);
954 pgd_t *pgd = pgd_offset_k(vstart);
959 pud = pud_offset(pgd, vstart);
960 if (pud_none(*pud)) {
963 new = __alloc_bootmem(PAGE_SIZE, PAGE_SIZE, PAGE_SIZE);
964 alloc_bytes += PAGE_SIZE;
965 pud_populate(&init_mm, pud, new);
968 pmd = pmd_offset(pud, vstart);
969 if (!pmd_present(*pmd)) {
972 new = __alloc_bootmem(PAGE_SIZE, PAGE_SIZE, PAGE_SIZE);
973 alloc_bytes += PAGE_SIZE;
974 pmd_populate_kernel(&init_mm, pmd, new);
977 pte = pte_offset_kernel(pmd, vstart);
978 this_end = (vstart + PMD_SIZE) & PMD_MASK;
982 while (vstart < this_end) {
983 pte_val(*pte) = (paddr | pgprot_val(prot));
994 static struct linux_prom64_registers pall[MAX_BANKS] __initdata;
995 static int pall_ents __initdata;
997 extern unsigned int kvmap_linear_patch[1];
999 static void __init kernel_physical_mapping_init(void)
1001 unsigned long i, mem_alloced = 0UL;
1003 read_obp_memory("reg", &pall[0], &pall_ents);
1005 for (i = 0; i < pall_ents; i++) {
1006 unsigned long phys_start, phys_end;
1008 phys_start = pall[i].phys_addr;
1009 phys_end = phys_start + pall[i].reg_size;
1010 mem_alloced += kernel_map_range(phys_start, phys_end,
1014 printk("Allocated %ld bytes for kernel page tables.\n",
1017 kvmap_linear_patch[0] = 0x01000000; /* nop */
1018 flushi(&kvmap_linear_patch[0]);
1023 void kernel_map_pages(struct page *page, int numpages, int enable)
1025 unsigned long phys_start = page_to_pfn(page) << PAGE_SHIFT;
1026 unsigned long phys_end = phys_start + (numpages * PAGE_SIZE);
1028 kernel_map_range(phys_start, phys_end,
1029 (enable ? PAGE_KERNEL : __pgprot(0)));
1031 flush_tsb_kernel_range(PAGE_OFFSET + phys_start,
1032 PAGE_OFFSET + phys_end);
1034 /* we should perform an IPI and flush all tlbs,
1035 * but that can deadlock->flush only current cpu.
1037 __flush_tlb_kernel_range(PAGE_OFFSET + phys_start,
1038 PAGE_OFFSET + phys_end);
1042 unsigned long __init find_ecache_flush_span(unsigned long size)
1046 for (i = 0; i < pavail_ents; i++) {
1047 if (pavail[i].reg_size >= size)
1048 return pavail[i].phys_addr;
1054 static void __init tsb_phys_patch(void)
1056 struct tsb_phys_patch_entry *p;
1058 p = &__tsb_phys_patch;
1059 while (p < &__tsb_phys_patch_end) {
1060 unsigned long addr = p->addr;
1062 *(unsigned int *) addr = p->insn;
1064 __asm__ __volatile__("flush %0"
1072 /* paging_init() sets up the page tables */
1074 extern void cheetah_ecache_flush_init(void);
1076 static unsigned long last_valid_pfn;
1077 pgd_t swapper_pg_dir[2048];
1079 void __init paging_init(void)
1081 unsigned long end_pfn, pages_avail, shift;
1082 unsigned long real_end, i;
1084 if (tlb_type == cheetah_plus)
1087 /* Find available physical memory... */
1088 read_obp_memory("available", &pavail[0], &pavail_ents);
1090 phys_base = 0xffffffffffffffffUL;
1091 for (i = 0; i < pavail_ents; i++)
1092 phys_base = min(phys_base, pavail[i].phys_addr);
1094 pfn_base = phys_base >> PAGE_SHIFT;
1096 kern_base = (prom_boot_mapping_phys_low >> 22UL) << 22UL;
1097 kern_size = (unsigned long)&_end - (unsigned long)KERNBASE;
1099 set_bit(0, mmu_context_bmap);
1101 shift = kern_base + PAGE_OFFSET - ((unsigned long)KERNBASE);
1103 real_end = (unsigned long)_end;
1104 if ((real_end > ((unsigned long)KERNBASE + 0x400000)))
1106 if ((real_end > ((unsigned long)KERNBASE + 0x800000))) {
1107 prom_printf("paging_init: Kernel > 8MB, too large.\n");
1111 /* Set kernel pgd to upper alias so physical page computations
1114 init_mm.pgd += ((shift) / (sizeof(pgd_t)));
1116 memset(swapper_low_pmd_dir, 0, sizeof(swapper_low_pmd_dir));
1118 /* Now can init the kernel/bad page tables. */
1119 pud_set(pud_offset(&swapper_pg_dir[0], 0),
1120 swapper_low_pmd_dir + (shift / sizeof(pgd_t)));
1122 inherit_prom_mappings();
1124 /* Ok, we can use our TLB miss and window trap handlers safely. */
1129 /* Setup bootmem... */
1131 last_valid_pfn = end_pfn = bootmem_init(&pages_avail);
1133 #ifdef CONFIG_DEBUG_PAGEALLOC
1134 kernel_physical_mapping_init();
1138 unsigned long zones_size[MAX_NR_ZONES];
1139 unsigned long zholes_size[MAX_NR_ZONES];
1140 unsigned long npages;
1143 for (znum = 0; znum < MAX_NR_ZONES; znum++)
1144 zones_size[znum] = zholes_size[znum] = 0;
1146 npages = end_pfn - pfn_base;
1147 zones_size[ZONE_DMA] = npages;
1148 zholes_size[ZONE_DMA] = npages - pages_avail;
1150 free_area_init_node(0, &contig_page_data, zones_size,
1151 phys_base >> PAGE_SHIFT, zholes_size);
1157 static void __init taint_real_pages(void)
1161 read_obp_memory("available", &pavail_rescan[0], &pavail_rescan_ents);
1163 /* Find changes discovered in the physmem available rescan and
1164 * reserve the lost portions in the bootmem maps.
1166 for (i = 0; i < pavail_ents; i++) {
1167 unsigned long old_start, old_end;
1169 old_start = pavail[i].phys_addr;
1170 old_end = old_start +
1172 while (old_start < old_end) {
1175 for (n = 0; pavail_rescan_ents; n++) {
1176 unsigned long new_start, new_end;
1178 new_start = pavail_rescan[n].phys_addr;
1179 new_end = new_start +
1180 pavail_rescan[n].reg_size;
1182 if (new_start <= old_start &&
1183 new_end >= (old_start + PAGE_SIZE)) {
1184 set_bit(old_start >> 22,
1185 sparc64_valid_addr_bitmap);
1189 reserve_bootmem(old_start, PAGE_SIZE);
1192 old_start += PAGE_SIZE;
1197 void __init mem_init(void)
1199 unsigned long codepages, datapages, initpages;
1200 unsigned long addr, last;
1203 i = last_valid_pfn >> ((22 - PAGE_SHIFT) + 6);
1205 sparc64_valid_addr_bitmap = (unsigned long *) alloc_bootmem(i << 3);
1206 if (sparc64_valid_addr_bitmap == NULL) {
1207 prom_printf("mem_init: Cannot alloc valid_addr_bitmap.\n");
1210 memset(sparc64_valid_addr_bitmap, 0, i << 3);
1212 addr = PAGE_OFFSET + kern_base;
1213 last = PAGE_ALIGN(kern_size) + addr;
1214 while (addr < last) {
1215 set_bit(__pa(addr) >> 22, sparc64_valid_addr_bitmap);
1221 max_mapnr = last_valid_pfn - pfn_base;
1222 high_memory = __va(last_valid_pfn << PAGE_SHIFT);
1224 #ifdef CONFIG_DEBUG_BOOTMEM
1225 prom_printf("mem_init: Calling free_all_bootmem().\n");
1227 totalram_pages = num_physpages = free_all_bootmem() - 1;
1230 * Set up the zero page, mark it reserved, so that page count
1231 * is not manipulated when freeing the page from user ptes.
1233 mem_map_zero = alloc_pages(GFP_KERNEL|__GFP_ZERO, 0);
1234 if (mem_map_zero == NULL) {
1235 prom_printf("paging_init: Cannot alloc zero page.\n");
1238 SetPageReserved(mem_map_zero);
1240 codepages = (((unsigned long) _etext) - ((unsigned long) _start));
1241 codepages = PAGE_ALIGN(codepages) >> PAGE_SHIFT;
1242 datapages = (((unsigned long) _edata) - ((unsigned long) _etext));
1243 datapages = PAGE_ALIGN(datapages) >> PAGE_SHIFT;
1244 initpages = (((unsigned long) __init_end) - ((unsigned long) __init_begin));
1245 initpages = PAGE_ALIGN(initpages) >> PAGE_SHIFT;
1247 printk("Memory: %uk available (%ldk kernel code, %ldk data, %ldk init) [%016lx,%016lx]\n",
1248 nr_free_pages() << (PAGE_SHIFT-10),
1249 codepages << (PAGE_SHIFT-10),
1250 datapages << (PAGE_SHIFT-10),
1251 initpages << (PAGE_SHIFT-10),
1252 PAGE_OFFSET, (last_valid_pfn << PAGE_SHIFT));
1254 if (tlb_type == cheetah || tlb_type == cheetah_plus)
1255 cheetah_ecache_flush_init();
1258 void free_initmem(void)
1260 unsigned long addr, initend;
1263 * The init section is aligned to 8k in vmlinux.lds. Page align for >8k pagesizes.
1265 addr = PAGE_ALIGN((unsigned long)(__init_begin));
1266 initend = (unsigned long)(__init_end) & PAGE_MASK;
1267 for (; addr < initend; addr += PAGE_SIZE) {
1272 ((unsigned long) __va(kern_base)) -
1273 ((unsigned long) KERNBASE));
1274 memset((void *)addr, 0xcc, PAGE_SIZE);
1275 p = virt_to_page(page);
1277 ClearPageReserved(p);
1278 set_page_count(p, 1);
1285 #ifdef CONFIG_BLK_DEV_INITRD
1286 void free_initrd_mem(unsigned long start, unsigned long end)
1289 printk ("Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
1290 for (; start < end; start += PAGE_SIZE) {
1291 struct page *p = virt_to_page(start);
1293 ClearPageReserved(p);
1294 set_page_count(p, 1);