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>
43 extern void device_scan(void);
47 static struct linux_prom64_registers pavail[MAX_BANKS] __initdata;
48 static struct linux_prom64_registers pavail_rescan[MAX_BANKS] __initdata;
49 static int pavail_ents __initdata;
50 static int pavail_rescan_ents __initdata;
52 static int cmp_p64(const void *a, const void *b)
54 const struct linux_prom64_registers *x = a, *y = b;
56 if (x->phys_addr > y->phys_addr)
58 if (x->phys_addr < y->phys_addr)
63 static void __init read_obp_memory(const char *property,
64 struct linux_prom64_registers *regs,
67 int node = prom_finddevice("/memory");
68 int prop_size = prom_getproplen(node, property);
71 ents = prop_size / sizeof(struct linux_prom64_registers);
72 if (ents > MAX_BANKS) {
73 prom_printf("The machine has more %s property entries than "
74 "this kernel can support (%d).\n",
79 ret = prom_getproperty(node, property, (char *) regs, prop_size);
81 prom_printf("Couldn't get %s property from /memory.\n");
87 /* Sanitize what we got from the firmware, by page aligning
90 for (i = 0; i < ents; i++) {
91 unsigned long base, size;
93 base = regs[i].phys_addr;
94 size = regs[i].reg_size;
97 if (base & ~PAGE_MASK) {
98 unsigned long new_base = PAGE_ALIGN(base);
100 size -= new_base - base;
101 if ((long) size < 0L)
105 regs[i].phys_addr = base;
106 regs[i].reg_size = size;
108 sort(regs, ents, sizeof(struct linux_prom64_registers),
112 unsigned long *sparc64_valid_addr_bitmap __read_mostly;
114 /* Ugly, but necessary... -DaveM */
115 unsigned long phys_base __read_mostly;
116 unsigned long kern_base __read_mostly;
117 unsigned long kern_size __read_mostly;
118 unsigned long pfn_base __read_mostly;
120 /* get_new_mmu_context() uses "cache + 1". */
121 DEFINE_SPINLOCK(ctx_alloc_lock);
122 unsigned long tlb_context_cache = CTX_FIRST_VERSION - 1;
123 #define CTX_BMAP_SLOTS (1UL << (CTX_NR_BITS - 6))
124 unsigned long mmu_context_bmap[CTX_BMAP_SLOTS];
126 /* References to special section boundaries */
127 extern char _start[], _end[];
129 /* Initial ramdisk setup */
130 extern unsigned long sparc_ramdisk_image64;
131 extern unsigned int sparc_ramdisk_image;
132 extern unsigned int sparc_ramdisk_size;
134 struct page *mem_map_zero __read_mostly;
136 unsigned int sparc64_highest_unlocked_tlb_ent __read_mostly;
138 unsigned long sparc64_kern_pri_context __read_mostly;
139 unsigned long sparc64_kern_pri_nuc_bits __read_mostly;
140 unsigned long sparc64_kern_sec_context __read_mostly;
144 kmem_cache_t *pgtable_cache __read_mostly;
146 static void zero_ctor(void *addr, kmem_cache_t *cache, unsigned long flags)
151 void pgtable_cache_init(void)
153 pgtable_cache = kmem_cache_create("pgtable_cache",
154 PAGE_SIZE, PAGE_SIZE,
156 SLAB_MUST_HWCACHE_ALIGN,
159 if (!pgtable_cache) {
160 prom_printf("pgtable_cache_init(): Could not create!\n");
165 #ifdef CONFIG_DEBUG_DCFLUSH
166 atomic_t dcpage_flushes = ATOMIC_INIT(0);
168 atomic_t dcpage_flushes_xcall = ATOMIC_INIT(0);
172 __inline__ void flush_dcache_page_impl(struct page *page)
174 #ifdef CONFIG_DEBUG_DCFLUSH
175 atomic_inc(&dcpage_flushes);
178 #ifdef DCACHE_ALIASING_POSSIBLE
179 __flush_dcache_page(page_address(page),
180 ((tlb_type == spitfire) &&
181 page_mapping(page) != NULL));
183 if (page_mapping(page) != NULL &&
184 tlb_type == spitfire)
185 __flush_icache_page(__pa(page_address(page)));
189 #define PG_dcache_dirty PG_arch_1
190 #define PG_dcache_cpu_shift 24
191 #define PG_dcache_cpu_mask (256 - 1)
194 #error D-cache dirty tracking and thread_info->cpu need fixing for > 256 cpus
197 #define dcache_dirty_cpu(page) \
198 (((page)->flags >> PG_dcache_cpu_shift) & PG_dcache_cpu_mask)
200 static __inline__ void set_dcache_dirty(struct page *page, int this_cpu)
202 unsigned long mask = this_cpu;
203 unsigned long non_cpu_bits;
205 non_cpu_bits = ~(PG_dcache_cpu_mask << PG_dcache_cpu_shift);
206 mask = (mask << PG_dcache_cpu_shift) | (1UL << PG_dcache_dirty);
208 __asm__ __volatile__("1:\n\t"
210 "and %%g7, %1, %%g1\n\t"
211 "or %%g1, %0, %%g1\n\t"
212 "casx [%2], %%g7, %%g1\n\t"
214 "membar #StoreLoad | #StoreStore\n\t"
215 "bne,pn %%xcc, 1b\n\t"
218 : "r" (mask), "r" (non_cpu_bits), "r" (&page->flags)
222 static __inline__ void clear_dcache_dirty_cpu(struct page *page, unsigned long cpu)
224 unsigned long mask = (1UL << PG_dcache_dirty);
226 __asm__ __volatile__("! test_and_clear_dcache_dirty\n"
229 "srlx %%g7, %4, %%g1\n\t"
230 "and %%g1, %3, %%g1\n\t"
232 "bne,pn %%icc, 2f\n\t"
233 " andn %%g7, %1, %%g1\n\t"
234 "casx [%2], %%g7, %%g1\n\t"
236 "membar #StoreLoad | #StoreStore\n\t"
237 "bne,pn %%xcc, 1b\n\t"
241 : "r" (cpu), "r" (mask), "r" (&page->flags),
242 "i" (PG_dcache_cpu_mask),
243 "i" (PG_dcache_cpu_shift)
247 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address, pte_t pte)
249 struct mm_struct *mm;
252 unsigned long pg_flags;
253 unsigned long mm_rss;
256 if (pfn_valid(pfn) &&
257 (page = pfn_to_page(pfn), page_mapping(page)) &&
258 ((pg_flags = page->flags) & (1UL << PG_dcache_dirty))) {
259 int cpu = ((pg_flags >> PG_dcache_cpu_shift) &
261 int this_cpu = get_cpu();
263 /* This is just to optimize away some function calls
267 flush_dcache_page_impl(page);
269 smp_flush_dcache_page_impl(page, cpu);
271 clear_dcache_dirty_cpu(page, cpu);
277 mm_rss = get_mm_rss(mm);
278 if (mm_rss >= mm->context.tsb_rss_limit)
279 tsb_grow(mm, mm_rss, GFP_ATOMIC);
281 if ((pte_val(pte) & _PAGE_ALL_SZ_BITS) == _PAGE_SZBITS) {
285 tsb = &mm->context.tsb[(address >> PAGE_SHIFT) &
286 (mm->context.tsb_nentries - 1UL)];
287 tag = (address >> 22UL) | CTX_HWBITS(mm->context) << 48UL;
288 tsb_insert(tsb, tag, pte_val(pte));
292 void flush_dcache_page(struct page *page)
294 struct address_space *mapping;
297 /* Do not bother with the expensive D-cache flush if it
298 * is merely the zero page. The 'bigcore' testcase in GDB
299 * causes this case to run millions of times.
301 if (page == ZERO_PAGE(0))
304 this_cpu = get_cpu();
306 mapping = page_mapping(page);
307 if (mapping && !mapping_mapped(mapping)) {
308 int dirty = test_bit(PG_dcache_dirty, &page->flags);
310 int dirty_cpu = dcache_dirty_cpu(page);
312 if (dirty_cpu == this_cpu)
314 smp_flush_dcache_page_impl(page, dirty_cpu);
316 set_dcache_dirty(page, this_cpu);
318 /* We could delay the flush for the !page_mapping
319 * case too. But that case is for exec env/arg
320 * pages and those are %99 certainly going to get
321 * faulted into the tlb (and thus flushed) anyways.
323 flush_dcache_page_impl(page);
330 void __kprobes flush_icache_range(unsigned long start, unsigned long end)
332 /* Cheetah has coherent I-cache. */
333 if (tlb_type == spitfire) {
336 for (kaddr = start; kaddr < end; kaddr += PAGE_SIZE)
337 __flush_icache_page(__get_phys(kaddr));
341 unsigned long page_to_pfn(struct page *page)
343 return (unsigned long) ((page - mem_map) + pfn_base);
346 struct page *pfn_to_page(unsigned long pfn)
348 return (mem_map + (pfn - pfn_base));
353 printk("Mem-info:\n");
355 printk("Free swap: %6ldkB\n",
356 nr_swap_pages << (PAGE_SHIFT-10));
357 printk("%ld pages of RAM\n", num_physpages);
358 printk("%d free pages\n", nr_free_pages());
361 void mmu_info(struct seq_file *m)
363 if (tlb_type == cheetah)
364 seq_printf(m, "MMU Type\t: Cheetah\n");
365 else if (tlb_type == cheetah_plus)
366 seq_printf(m, "MMU Type\t: Cheetah+\n");
367 else if (tlb_type == spitfire)
368 seq_printf(m, "MMU Type\t: Spitfire\n");
370 seq_printf(m, "MMU Type\t: ???\n");
372 #ifdef CONFIG_DEBUG_DCFLUSH
373 seq_printf(m, "DCPageFlushes\t: %d\n",
374 atomic_read(&dcpage_flushes));
376 seq_printf(m, "DCPageFlushesXC\t: %d\n",
377 atomic_read(&dcpage_flushes_xcall));
378 #endif /* CONFIG_SMP */
379 #endif /* CONFIG_DEBUG_DCFLUSH */
382 struct linux_prom_translation {
388 /* Exported for kernel TLB miss handling in ktlb.S */
389 struct linux_prom_translation prom_trans[512] __read_mostly;
390 unsigned int prom_trans_ents __read_mostly;
391 unsigned int swapper_pgd_zero __read_mostly;
393 extern unsigned long prom_boot_page;
394 extern void prom_remap(unsigned long physpage, unsigned long virtpage, int mmu_ihandle);
395 extern int prom_get_mmu_ihandle(void);
396 extern void register_prom_callbacks(void);
398 /* Exported for SMP bootup purposes. */
399 unsigned long kern_locked_tte_data;
402 * Translate PROM's mapping we capture at boot time into physical address.
403 * The second parameter is only set from prom_callback() invocations.
405 unsigned long prom_virt_to_phys(unsigned long promva, int *error)
409 for (i = 0; i < prom_trans_ents; i++) {
410 struct linux_prom_translation *p = &prom_trans[i];
412 if (promva >= p->virt &&
413 promva < (p->virt + p->size)) {
414 unsigned long base = p->data & _PAGE_PADDR;
418 return base + (promva & (8192 - 1));
426 /* The obp translations are saved based on 8k pagesize, since obp can
427 * use a mixture of pagesizes. Misses to the LOW_OBP_ADDRESS ->
428 * HI_OBP_ADDRESS range are handled in ktlb.S.
430 static inline int in_obp_range(unsigned long vaddr)
432 return (vaddr >= LOW_OBP_ADDRESS &&
433 vaddr < HI_OBP_ADDRESS);
436 static int cmp_ptrans(const void *a, const void *b)
438 const struct linux_prom_translation *x = a, *y = b;
440 if (x->virt > y->virt)
442 if (x->virt < y->virt)
447 /* Read OBP translations property into 'prom_trans[]'. */
448 static void __init read_obp_translations(void)
450 int n, node, ents, first, last, i;
452 node = prom_finddevice("/virtual-memory");
453 n = prom_getproplen(node, "translations");
454 if (unlikely(n == 0 || n == -1)) {
455 prom_printf("prom_mappings: Couldn't get size.\n");
458 if (unlikely(n > sizeof(prom_trans))) {
459 prom_printf("prom_mappings: Size %Zd is too big.\n", n);
463 if ((n = prom_getproperty(node, "translations",
464 (char *)&prom_trans[0],
465 sizeof(prom_trans))) == -1) {
466 prom_printf("prom_mappings: Couldn't get property.\n");
470 n = n / sizeof(struct linux_prom_translation);
474 sort(prom_trans, ents, sizeof(struct linux_prom_translation),
477 /* Now kick out all the non-OBP entries. */
478 for (i = 0; i < ents; i++) {
479 if (in_obp_range(prom_trans[i].virt))
483 for (; i < ents; i++) {
484 if (!in_obp_range(prom_trans[i].virt))
489 for (i = 0; i < (last - first); i++) {
490 struct linux_prom_translation *src = &prom_trans[i + first];
491 struct linux_prom_translation *dest = &prom_trans[i];
495 for (; i < ents; i++) {
496 struct linux_prom_translation *dest = &prom_trans[i];
497 dest->virt = dest->size = dest->data = 0x0UL;
500 prom_trans_ents = last - first;
502 if (tlb_type == spitfire) {
503 /* Clear diag TTE bits. */
504 for (i = 0; i < prom_trans_ents; i++)
505 prom_trans[i].data &= ~0x0003fe0000000000UL;
509 static void __init remap_kernel(void)
511 unsigned long phys_page, tte_vaddr, tte_data;
512 int tlb_ent = sparc64_highest_locked_tlbent();
514 tte_vaddr = (unsigned long) KERNBASE;
515 phys_page = (prom_boot_mapping_phys_low >> 22UL) << 22UL;
516 tte_data = (phys_page | (_PAGE_VALID | _PAGE_SZ4MB |
517 _PAGE_CP | _PAGE_CV | _PAGE_P |
520 kern_locked_tte_data = tte_data;
522 /* Now lock us into the TLBs via OBP. */
523 prom_dtlb_load(tlb_ent, tte_data, tte_vaddr);
524 prom_itlb_load(tlb_ent, tte_data, tte_vaddr);
527 prom_dtlb_load(tlb_ent,
529 tte_vaddr + 0x400000);
530 prom_itlb_load(tlb_ent,
532 tte_vaddr + 0x400000);
534 sparc64_highest_unlocked_tlb_ent = tlb_ent - 1;
535 if (tlb_type == cheetah_plus) {
536 sparc64_kern_pri_context = (CTX_CHEETAH_PLUS_CTX0 |
537 CTX_CHEETAH_PLUS_NUC);
538 sparc64_kern_pri_nuc_bits = CTX_CHEETAH_PLUS_NUC;
539 sparc64_kern_sec_context = CTX_CHEETAH_PLUS_CTX0;
544 static void __init inherit_prom_mappings(void)
546 read_obp_translations();
548 /* Now fixup OBP's idea about where we really are mapped. */
549 prom_printf("Remapping the kernel... ");
551 prom_printf("done.\n");
553 prom_printf("Registering callbacks... ");
554 register_prom_callbacks();
555 prom_printf("done.\n");
558 void prom_world(int enter)
561 set_fs((mm_segment_t) { get_thread_current_ds() });
563 __asm__ __volatile__("flushw");
566 #ifdef DCACHE_ALIASING_POSSIBLE
567 void __flush_dcache_range(unsigned long start, unsigned long end)
571 if (tlb_type == spitfire) {
574 for (va = start; va < end; va += 32) {
575 spitfire_put_dcache_tag(va & 0x3fe0, 0x0);
582 for (va = start; va < end; va += 32)
583 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
587 "i" (ASI_DCACHE_INVALIDATE));
590 #endif /* DCACHE_ALIASING_POSSIBLE */
592 /* If not locked, zap it. */
593 void __flush_tlb_all(void)
595 unsigned long pstate;
598 __asm__ __volatile__("flushw\n\t"
599 "rdpr %%pstate, %0\n\t"
600 "wrpr %0, %1, %%pstate"
603 if (tlb_type == spitfire) {
604 for (i = 0; i < 64; i++) {
605 /* Spitfire Errata #32 workaround */
606 /* NOTE: Always runs on spitfire, so no
607 * cheetah+ page size encodings.
609 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
613 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
615 if (!(spitfire_get_dtlb_data(i) & _PAGE_L)) {
616 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
619 : "r" (TLB_TAG_ACCESS), "i" (ASI_DMMU));
620 spitfire_put_dtlb_data(i, 0x0UL);
623 /* Spitfire Errata #32 workaround */
624 /* NOTE: Always runs on spitfire, so no
625 * cheetah+ page size encodings.
627 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
631 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
633 if (!(spitfire_get_itlb_data(i) & _PAGE_L)) {
634 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
637 : "r" (TLB_TAG_ACCESS), "i" (ASI_IMMU));
638 spitfire_put_itlb_data(i, 0x0UL);
641 } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
642 cheetah_flush_dtlb_all();
643 cheetah_flush_itlb_all();
645 __asm__ __volatile__("wrpr %0, 0, %%pstate"
649 /* Caller does TLB context flushing on local CPU if necessary.
650 * The caller also ensures that CTX_VALID(mm->context) is false.
652 * We must be careful about boundary cases so that we never
653 * let the user have CTX 0 (nucleus) or we ever use a CTX
654 * version of zero (and thus NO_CONTEXT would not be caught
655 * by version mis-match tests in mmu_context.h).
657 void get_new_mmu_context(struct mm_struct *mm)
659 unsigned long ctx, new_ctx;
660 unsigned long orig_pgsz_bits;
663 spin_lock(&ctx_alloc_lock);
664 orig_pgsz_bits = (mm->context.sparc64_ctx_val & CTX_PGSZ_MASK);
665 ctx = (tlb_context_cache + 1) & CTX_NR_MASK;
666 new_ctx = find_next_zero_bit(mmu_context_bmap, 1 << CTX_NR_BITS, ctx);
667 if (new_ctx >= (1 << CTX_NR_BITS)) {
668 new_ctx = find_next_zero_bit(mmu_context_bmap, ctx, 1);
669 if (new_ctx >= ctx) {
671 new_ctx = (tlb_context_cache & CTX_VERSION_MASK) +
674 new_ctx = CTX_FIRST_VERSION;
676 /* Don't call memset, for 16 entries that's just
679 mmu_context_bmap[0] = 3;
680 mmu_context_bmap[1] = 0;
681 mmu_context_bmap[2] = 0;
682 mmu_context_bmap[3] = 0;
683 for (i = 4; i < CTX_BMAP_SLOTS; i += 4) {
684 mmu_context_bmap[i + 0] = 0;
685 mmu_context_bmap[i + 1] = 0;
686 mmu_context_bmap[i + 2] = 0;
687 mmu_context_bmap[i + 3] = 0;
692 mmu_context_bmap[new_ctx>>6] |= (1UL << (new_ctx & 63));
693 new_ctx |= (tlb_context_cache & CTX_VERSION_MASK);
695 tlb_context_cache = new_ctx;
696 mm->context.sparc64_ctx_val = new_ctx | orig_pgsz_bits;
697 spin_unlock(&ctx_alloc_lock);
700 void sparc_ultra_dump_itlb(void)
704 if (tlb_type == spitfire) {
705 printk ("Contents of itlb: ");
706 for (slot = 0; slot < 14; slot++) printk (" ");
707 printk ("%2x:%016lx,%016lx\n",
709 spitfire_get_itlb_tag(0), spitfire_get_itlb_data(0));
710 for (slot = 1; slot < 64; slot+=3) {
711 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx %2x:%016lx,%016lx\n",
713 spitfire_get_itlb_tag(slot), spitfire_get_itlb_data(slot),
715 spitfire_get_itlb_tag(slot+1), spitfire_get_itlb_data(slot+1),
717 spitfire_get_itlb_tag(slot+2), spitfire_get_itlb_data(slot+2));
719 } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
720 printk ("Contents of itlb0:\n");
721 for (slot = 0; slot < 16; slot+=2) {
722 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
724 cheetah_get_litlb_tag(slot), cheetah_get_litlb_data(slot),
726 cheetah_get_litlb_tag(slot+1), cheetah_get_litlb_data(slot+1));
728 printk ("Contents of itlb2:\n");
729 for (slot = 0; slot < 128; slot+=2) {
730 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
732 cheetah_get_itlb_tag(slot), cheetah_get_itlb_data(slot),
734 cheetah_get_itlb_tag(slot+1), cheetah_get_itlb_data(slot+1));
739 void sparc_ultra_dump_dtlb(void)
743 if (tlb_type == spitfire) {
744 printk ("Contents of dtlb: ");
745 for (slot = 0; slot < 14; slot++) printk (" ");
746 printk ("%2x:%016lx,%016lx\n", 0,
747 spitfire_get_dtlb_tag(0), spitfire_get_dtlb_data(0));
748 for (slot = 1; slot < 64; slot+=3) {
749 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx %2x:%016lx,%016lx\n",
751 spitfire_get_dtlb_tag(slot), spitfire_get_dtlb_data(slot),
753 spitfire_get_dtlb_tag(slot+1), spitfire_get_dtlb_data(slot+1),
755 spitfire_get_dtlb_tag(slot+2), spitfire_get_dtlb_data(slot+2));
757 } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
758 printk ("Contents of dtlb0:\n");
759 for (slot = 0; slot < 16; slot+=2) {
760 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
762 cheetah_get_ldtlb_tag(slot), cheetah_get_ldtlb_data(slot),
764 cheetah_get_ldtlb_tag(slot+1), cheetah_get_ldtlb_data(slot+1));
766 printk ("Contents of dtlb2:\n");
767 for (slot = 0; slot < 512; slot+=2) {
768 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
770 cheetah_get_dtlb_tag(slot, 2), cheetah_get_dtlb_data(slot, 2),
772 cheetah_get_dtlb_tag(slot+1, 2), cheetah_get_dtlb_data(slot+1, 2));
774 if (tlb_type == cheetah_plus) {
775 printk ("Contents of dtlb3:\n");
776 for (slot = 0; slot < 512; slot+=2) {
777 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
779 cheetah_get_dtlb_tag(slot, 3), cheetah_get_dtlb_data(slot, 3),
781 cheetah_get_dtlb_tag(slot+1, 3), cheetah_get_dtlb_data(slot+1, 3));
787 static inline void spitfire_errata32(void)
789 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
793 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
796 extern unsigned long cmdline_memory_size;
798 unsigned long __init bootmem_init(unsigned long *pages_avail)
800 unsigned long bootmap_size, start_pfn, end_pfn;
801 unsigned long end_of_phys_memory = 0UL;
802 unsigned long bootmap_pfn, bytes_avail, size;
805 #ifdef CONFIG_DEBUG_BOOTMEM
806 prom_printf("bootmem_init: Scan pavail, ");
810 for (i = 0; i < pavail_ents; i++) {
811 end_of_phys_memory = pavail[i].phys_addr +
813 bytes_avail += pavail[i].reg_size;
814 if (cmdline_memory_size) {
815 if (bytes_avail > cmdline_memory_size) {
816 unsigned long slack = bytes_avail - cmdline_memory_size;
818 bytes_avail -= slack;
819 end_of_phys_memory -= slack;
821 pavail[i].reg_size -= slack;
822 if ((long)pavail[i].reg_size <= 0L) {
823 pavail[i].phys_addr = 0xdeadbeefUL;
824 pavail[i].reg_size = 0UL;
827 pavail[i+1].reg_size = 0Ul;
828 pavail[i+1].phys_addr = 0xdeadbeefUL;
836 *pages_avail = bytes_avail >> PAGE_SHIFT;
838 /* Start with page aligned address of last symbol in kernel
839 * image. The kernel is hard mapped below PAGE_OFFSET in a
840 * 4MB locked TLB translation.
842 start_pfn = PAGE_ALIGN(kern_base + kern_size) >> PAGE_SHIFT;
844 bootmap_pfn = start_pfn;
846 end_pfn = end_of_phys_memory >> PAGE_SHIFT;
848 #ifdef CONFIG_BLK_DEV_INITRD
849 /* Now have to check initial ramdisk, so that bootmap does not overwrite it */
850 if (sparc_ramdisk_image || sparc_ramdisk_image64) {
851 unsigned long ramdisk_image = sparc_ramdisk_image ?
852 sparc_ramdisk_image : sparc_ramdisk_image64;
853 if (ramdisk_image >= (unsigned long)_end - 2 * PAGE_SIZE)
854 ramdisk_image -= KERNBASE;
855 initrd_start = ramdisk_image + phys_base;
856 initrd_end = initrd_start + sparc_ramdisk_size;
857 if (initrd_end > end_of_phys_memory) {
858 printk(KERN_CRIT "initrd extends beyond end of memory "
859 "(0x%016lx > 0x%016lx)\ndisabling initrd\n",
860 initrd_end, end_of_phys_memory);
864 if (initrd_start >= (start_pfn << PAGE_SHIFT) &&
865 initrd_start < (start_pfn << PAGE_SHIFT) + 2 * PAGE_SIZE)
866 bootmap_pfn = PAGE_ALIGN (initrd_end) >> PAGE_SHIFT;
870 /* Initialize the boot-time allocator. */
871 max_pfn = max_low_pfn = end_pfn;
872 min_low_pfn = pfn_base;
874 #ifdef CONFIG_DEBUG_BOOTMEM
875 prom_printf("init_bootmem(min[%lx], bootmap[%lx], max[%lx])\n",
876 min_low_pfn, bootmap_pfn, max_low_pfn);
878 bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap_pfn, pfn_base, end_pfn);
880 /* Now register the available physical memory with the
883 for (i = 0; i < pavail_ents; i++) {
884 #ifdef CONFIG_DEBUG_BOOTMEM
885 prom_printf("free_bootmem(pavail:%d): base[%lx] size[%lx]\n",
886 i, pavail[i].phys_addr, pavail[i].reg_size);
888 free_bootmem(pavail[i].phys_addr, pavail[i].reg_size);
891 #ifdef CONFIG_BLK_DEV_INITRD
893 size = initrd_end - initrd_start;
895 /* Resert the initrd image area. */
896 #ifdef CONFIG_DEBUG_BOOTMEM
897 prom_printf("reserve_bootmem(initrd): base[%llx] size[%lx]\n",
898 initrd_start, initrd_end);
900 reserve_bootmem(initrd_start, size);
901 *pages_avail -= PAGE_ALIGN(size) >> PAGE_SHIFT;
903 initrd_start += PAGE_OFFSET;
904 initrd_end += PAGE_OFFSET;
907 /* Reserve the kernel text/data/bss. */
908 #ifdef CONFIG_DEBUG_BOOTMEM
909 prom_printf("reserve_bootmem(kernel): base[%lx] size[%lx]\n", kern_base, kern_size);
911 reserve_bootmem(kern_base, kern_size);
912 *pages_avail -= PAGE_ALIGN(kern_size) >> PAGE_SHIFT;
914 /* Reserve the bootmem map. We do not account for it
915 * in pages_avail because we will release that memory
916 * in free_all_bootmem.
919 #ifdef CONFIG_DEBUG_BOOTMEM
920 prom_printf("reserve_bootmem(bootmap): base[%lx] size[%lx]\n",
921 (bootmap_pfn << PAGE_SHIFT), size);
923 reserve_bootmem((bootmap_pfn << PAGE_SHIFT), size);
924 *pages_avail -= PAGE_ALIGN(size) >> PAGE_SHIFT;
929 #ifdef CONFIG_DEBUG_PAGEALLOC
930 static unsigned long kernel_map_range(unsigned long pstart, unsigned long pend, pgprot_t prot)
932 unsigned long vstart = PAGE_OFFSET + pstart;
933 unsigned long vend = PAGE_OFFSET + pend;
934 unsigned long alloc_bytes = 0UL;
936 if ((vstart & ~PAGE_MASK) || (vend & ~PAGE_MASK)) {
937 prom_printf("kernel_map: Unaligned physmem[%lx:%lx]\n",
942 while (vstart < vend) {
943 unsigned long this_end, paddr = __pa(vstart);
944 pgd_t *pgd = pgd_offset_k(vstart);
949 pud = pud_offset(pgd, vstart);
950 if (pud_none(*pud)) {
953 new = __alloc_bootmem(PAGE_SIZE, PAGE_SIZE, PAGE_SIZE);
954 alloc_bytes += PAGE_SIZE;
955 pud_populate(&init_mm, pud, new);
958 pmd = pmd_offset(pud, vstart);
959 if (!pmd_present(*pmd)) {
962 new = __alloc_bootmem(PAGE_SIZE, PAGE_SIZE, PAGE_SIZE);
963 alloc_bytes += PAGE_SIZE;
964 pmd_populate_kernel(&init_mm, pmd, new);
967 pte = pte_offset_kernel(pmd, vstart);
968 this_end = (vstart + PMD_SIZE) & PMD_MASK;
972 while (vstart < this_end) {
973 pte_val(*pte) = (paddr | pgprot_val(prot));
984 static struct linux_prom64_registers pall[MAX_BANKS] __initdata;
985 static int pall_ents __initdata;
987 extern unsigned int kvmap_linear_patch[1];
989 static void __init kernel_physical_mapping_init(void)
991 unsigned long i, mem_alloced = 0UL;
993 read_obp_memory("reg", &pall[0], &pall_ents);
995 for (i = 0; i < pall_ents; i++) {
996 unsigned long phys_start, phys_end;
998 phys_start = pall[i].phys_addr;
999 phys_end = phys_start + pall[i].reg_size;
1000 mem_alloced += kernel_map_range(phys_start, phys_end,
1004 printk("Allocated %ld bytes for kernel page tables.\n",
1007 kvmap_linear_patch[0] = 0x01000000; /* nop */
1008 flushi(&kvmap_linear_patch[0]);
1013 void kernel_map_pages(struct page *page, int numpages, int enable)
1015 unsigned long phys_start = page_to_pfn(page) << PAGE_SHIFT;
1016 unsigned long phys_end = phys_start + (numpages * PAGE_SIZE);
1018 kernel_map_range(phys_start, phys_end,
1019 (enable ? PAGE_KERNEL : __pgprot(0)));
1021 flush_tsb_kernel_range(PAGE_OFFSET + phys_start,
1022 PAGE_OFFSET + phys_end);
1024 /* we should perform an IPI and flush all tlbs,
1025 * but that can deadlock->flush only current cpu.
1027 __flush_tlb_kernel_range(PAGE_OFFSET + phys_start,
1028 PAGE_OFFSET + phys_end);
1032 unsigned long __init find_ecache_flush_span(unsigned long size)
1036 for (i = 0; i < pavail_ents; i++) {
1037 if (pavail[i].reg_size >= size)
1038 return pavail[i].phys_addr;
1044 /* paging_init() sets up the page tables */
1046 extern void cheetah_ecache_flush_init(void);
1048 static unsigned long last_valid_pfn;
1049 pgd_t swapper_pg_dir[2048];
1051 void __init paging_init(void)
1053 unsigned long end_pfn, pages_avail, shift;
1054 unsigned long real_end, i;
1056 /* Find available physical memory... */
1057 read_obp_memory("available", &pavail[0], &pavail_ents);
1059 phys_base = 0xffffffffffffffffUL;
1060 for (i = 0; i < pavail_ents; i++)
1061 phys_base = min(phys_base, pavail[i].phys_addr);
1063 pfn_base = phys_base >> PAGE_SHIFT;
1065 kern_base = (prom_boot_mapping_phys_low >> 22UL) << 22UL;
1066 kern_size = (unsigned long)&_end - (unsigned long)KERNBASE;
1068 set_bit(0, mmu_context_bmap);
1070 shift = kern_base + PAGE_OFFSET - ((unsigned long)KERNBASE);
1072 real_end = (unsigned long)_end;
1073 if ((real_end > ((unsigned long)KERNBASE + 0x400000)))
1075 if ((real_end > ((unsigned long)KERNBASE + 0x800000))) {
1076 prom_printf("paging_init: Kernel > 8MB, too large.\n");
1080 /* Set kernel pgd to upper alias so physical page computations
1083 init_mm.pgd += ((shift) / (sizeof(pgd_t)));
1085 memset(swapper_low_pmd_dir, 0, sizeof(swapper_low_pmd_dir));
1087 /* Now can init the kernel/bad page tables. */
1088 pud_set(pud_offset(&swapper_pg_dir[0], 0),
1089 swapper_low_pmd_dir + (shift / sizeof(pgd_t)));
1091 swapper_pgd_zero = pgd_val(swapper_pg_dir[0]);
1093 inherit_prom_mappings();
1095 /* Ok, we can use our TLB miss and window trap handlers safely.
1096 * We need to do a quick peek here to see if we are on StarFire
1097 * or not, so setup_tba can setup the IRQ globals correctly (it
1098 * needs to get the hard smp processor id correctly).
1101 extern void setup_tba(int);
1102 setup_tba(this_is_starfire);
1107 /* Setup bootmem... */
1109 last_valid_pfn = end_pfn = bootmem_init(&pages_avail);
1111 #ifdef CONFIG_DEBUG_PAGEALLOC
1112 kernel_physical_mapping_init();
1116 unsigned long zones_size[MAX_NR_ZONES];
1117 unsigned long zholes_size[MAX_NR_ZONES];
1118 unsigned long npages;
1121 for (znum = 0; znum < MAX_NR_ZONES; znum++)
1122 zones_size[znum] = zholes_size[znum] = 0;
1124 npages = end_pfn - pfn_base;
1125 zones_size[ZONE_DMA] = npages;
1126 zholes_size[ZONE_DMA] = npages - pages_avail;
1128 free_area_init_node(0, &contig_page_data, zones_size,
1129 phys_base >> PAGE_SHIFT, zholes_size);
1135 static void __init taint_real_pages(void)
1139 read_obp_memory("available", &pavail_rescan[0], &pavail_rescan_ents);
1141 /* Find changes discovered in the physmem available rescan and
1142 * reserve the lost portions in the bootmem maps.
1144 for (i = 0; i < pavail_ents; i++) {
1145 unsigned long old_start, old_end;
1147 old_start = pavail[i].phys_addr;
1148 old_end = old_start +
1150 while (old_start < old_end) {
1153 for (n = 0; pavail_rescan_ents; n++) {
1154 unsigned long new_start, new_end;
1156 new_start = pavail_rescan[n].phys_addr;
1157 new_end = new_start +
1158 pavail_rescan[n].reg_size;
1160 if (new_start <= old_start &&
1161 new_end >= (old_start + PAGE_SIZE)) {
1162 set_bit(old_start >> 22,
1163 sparc64_valid_addr_bitmap);
1167 reserve_bootmem(old_start, PAGE_SIZE);
1170 old_start += PAGE_SIZE;
1175 void __init mem_init(void)
1177 unsigned long codepages, datapages, initpages;
1178 unsigned long addr, last;
1181 i = last_valid_pfn >> ((22 - PAGE_SHIFT) + 6);
1183 sparc64_valid_addr_bitmap = (unsigned long *) alloc_bootmem(i << 3);
1184 if (sparc64_valid_addr_bitmap == NULL) {
1185 prom_printf("mem_init: Cannot alloc valid_addr_bitmap.\n");
1188 memset(sparc64_valid_addr_bitmap, 0, i << 3);
1190 addr = PAGE_OFFSET + kern_base;
1191 last = PAGE_ALIGN(kern_size) + addr;
1192 while (addr < last) {
1193 set_bit(__pa(addr) >> 22, sparc64_valid_addr_bitmap);
1199 max_mapnr = last_valid_pfn - pfn_base;
1200 high_memory = __va(last_valid_pfn << PAGE_SHIFT);
1202 #ifdef CONFIG_DEBUG_BOOTMEM
1203 prom_printf("mem_init: Calling free_all_bootmem().\n");
1205 totalram_pages = num_physpages = free_all_bootmem() - 1;
1208 * Set up the zero page, mark it reserved, so that page count
1209 * is not manipulated when freeing the page from user ptes.
1211 mem_map_zero = alloc_pages(GFP_KERNEL|__GFP_ZERO, 0);
1212 if (mem_map_zero == NULL) {
1213 prom_printf("paging_init: Cannot alloc zero page.\n");
1216 SetPageReserved(mem_map_zero);
1218 codepages = (((unsigned long) _etext) - ((unsigned long) _start));
1219 codepages = PAGE_ALIGN(codepages) >> PAGE_SHIFT;
1220 datapages = (((unsigned long) _edata) - ((unsigned long) _etext));
1221 datapages = PAGE_ALIGN(datapages) >> PAGE_SHIFT;
1222 initpages = (((unsigned long) __init_end) - ((unsigned long) __init_begin));
1223 initpages = PAGE_ALIGN(initpages) >> PAGE_SHIFT;
1225 printk("Memory: %uk available (%ldk kernel code, %ldk data, %ldk init) [%016lx,%016lx]\n",
1226 nr_free_pages() << (PAGE_SHIFT-10),
1227 codepages << (PAGE_SHIFT-10),
1228 datapages << (PAGE_SHIFT-10),
1229 initpages << (PAGE_SHIFT-10),
1230 PAGE_OFFSET, (last_valid_pfn << PAGE_SHIFT));
1232 if (tlb_type == cheetah || tlb_type == cheetah_plus)
1233 cheetah_ecache_flush_init();
1236 void free_initmem(void)
1238 unsigned long addr, initend;
1241 * The init section is aligned to 8k in vmlinux.lds. Page align for >8k pagesizes.
1243 addr = PAGE_ALIGN((unsigned long)(__init_begin));
1244 initend = (unsigned long)(__init_end) & PAGE_MASK;
1245 for (; addr < initend; addr += PAGE_SIZE) {
1250 ((unsigned long) __va(kern_base)) -
1251 ((unsigned long) KERNBASE));
1252 memset((void *)addr, 0xcc, PAGE_SIZE);
1253 p = virt_to_page(page);
1255 ClearPageReserved(p);
1256 set_page_count(p, 1);
1263 #ifdef CONFIG_BLK_DEV_INITRD
1264 void free_initrd_mem(unsigned long start, unsigned long end)
1267 printk ("Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
1268 for (; start < end; start += PAGE_SIZE) {
1269 struct page *p = virt_to_page(start);
1271 ClearPageReserved(p);
1272 set_page_count(p, 1);