[TG3]: Set minimal hw interrupt mitigation.
[linux-2.6/verdex.git] / arch / sparc64 / mm / init.c
blob9c5222075da93ac61018cb996f32008178dffcbd
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)
6 */
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>
14 #include <linux/mm.h>
15 #include <linux/hugetlb.h>
16 #include <linux/slab.h>
17 #include <linux/initrd.h>
18 #include <linux/swap.h>
19 #include <linux/pagemap.h>
20 #include <linux/fs.h>
21 #include <linux/seq_file.h>
23 #include <asm/head.h>
24 #include <asm/system.h>
25 #include <asm/page.h>
26 #include <asm/pgalloc.h>
27 #include <asm/pgtable.h>
28 #include <asm/oplib.h>
29 #include <asm/iommu.h>
30 #include <asm/io.h>
31 #include <asm/uaccess.h>
32 #include <asm/mmu_context.h>
33 #include <asm/tlbflush.h>
34 #include <asm/dma.h>
35 #include <asm/starfire.h>
36 #include <asm/tlb.h>
37 #include <asm/spitfire.h>
38 #include <asm/sections.h>
40 extern void device_scan(void);
42 struct sparc_phys_banks sp_banks[SPARC_PHYS_BANKS];
44 unsigned long *sparc64_valid_addr_bitmap;
46 /* Ugly, but necessary... -DaveM */
47 unsigned long phys_base;
48 unsigned long kern_base;
49 unsigned long kern_size;
50 unsigned long pfn_base;
52 /* This is even uglier. We have a problem where the kernel may not be
53 * located at phys_base. However, initial __alloc_bootmem() calls need to
54 * be adjusted to be within the 4-8Megs that the kernel is mapped to, else
55 * those page mappings wont work. Things are ok after inherit_prom_mappings
56 * is called though. Dave says he'll clean this up some other time.
57 * -- BenC
59 static unsigned long bootmap_base;
61 /* get_new_mmu_context() uses "cache + 1". */
62 DEFINE_SPINLOCK(ctx_alloc_lock);
63 unsigned long tlb_context_cache = CTX_FIRST_VERSION - 1;
64 #define CTX_BMAP_SLOTS (1UL << (CTX_NR_BITS - 6))
65 unsigned long mmu_context_bmap[CTX_BMAP_SLOTS];
67 /* References to special section boundaries */
68 extern char _start[], _end[];
70 /* Initial ramdisk setup */
71 extern unsigned long sparc_ramdisk_image64;
72 extern unsigned int sparc_ramdisk_image;
73 extern unsigned int sparc_ramdisk_size;
75 struct page *mem_map_zero;
77 int bigkernel = 0;
79 /* XXX Tune this... */
80 #define PGT_CACHE_LOW 25
81 #define PGT_CACHE_HIGH 50
83 void check_pgt_cache(void)
85 preempt_disable();
86 if (pgtable_cache_size > PGT_CACHE_HIGH) {
87 do {
88 if (pgd_quicklist)
89 free_pgd_slow(get_pgd_fast());
90 if (pte_quicklist[0])
91 free_pte_slow(pte_alloc_one_fast(NULL, 0));
92 if (pte_quicklist[1])
93 free_pte_slow(pte_alloc_one_fast(NULL, 1 << (PAGE_SHIFT + 10)));
94 } while (pgtable_cache_size > PGT_CACHE_LOW);
96 preempt_enable();
99 #ifdef CONFIG_DEBUG_DCFLUSH
100 atomic_t dcpage_flushes = ATOMIC_INIT(0);
101 #ifdef CONFIG_SMP
102 atomic_t dcpage_flushes_xcall = ATOMIC_INIT(0);
103 #endif
104 #endif
106 __inline__ void flush_dcache_page_impl(struct page *page)
108 #ifdef CONFIG_DEBUG_DCFLUSH
109 atomic_inc(&dcpage_flushes);
110 #endif
112 #ifdef DCACHE_ALIASING_POSSIBLE
113 __flush_dcache_page(page_address(page),
114 ((tlb_type == spitfire) &&
115 page_mapping(page) != NULL));
116 #else
117 if (page_mapping(page) != NULL &&
118 tlb_type == spitfire)
119 __flush_icache_page(__pa(page_address(page)));
120 #endif
123 #define PG_dcache_dirty PG_arch_1
125 #define dcache_dirty_cpu(page) \
126 (((page)->flags >> 24) & (NR_CPUS - 1UL))
128 static __inline__ void set_dcache_dirty(struct page *page, int this_cpu)
130 unsigned long mask = this_cpu;
131 unsigned long non_cpu_bits = ~((NR_CPUS - 1UL) << 24UL);
132 mask = (mask << 24) | (1UL << PG_dcache_dirty);
133 __asm__ __volatile__("1:\n\t"
134 "ldx [%2], %%g7\n\t"
135 "and %%g7, %1, %%g1\n\t"
136 "or %%g1, %0, %%g1\n\t"
137 "casx [%2], %%g7, %%g1\n\t"
138 "cmp %%g7, %%g1\n\t"
139 "bne,pn %%xcc, 1b\n\t"
140 " membar #StoreLoad | #StoreStore"
141 : /* no outputs */
142 : "r" (mask), "r" (non_cpu_bits), "r" (&page->flags)
143 : "g1", "g7");
146 static __inline__ void clear_dcache_dirty_cpu(struct page *page, unsigned long cpu)
148 unsigned long mask = (1UL << PG_dcache_dirty);
150 __asm__ __volatile__("! test_and_clear_dcache_dirty\n"
151 "1:\n\t"
152 "ldx [%2], %%g7\n\t"
153 "srlx %%g7, 24, %%g1\n\t"
154 "and %%g1, %3, %%g1\n\t"
155 "cmp %%g1, %0\n\t"
156 "bne,pn %%icc, 2f\n\t"
157 " andn %%g7, %1, %%g1\n\t"
158 "casx [%2], %%g7, %%g1\n\t"
159 "cmp %%g7, %%g1\n\t"
160 "bne,pn %%xcc, 1b\n\t"
161 " membar #StoreLoad | #StoreStore\n"
162 "2:"
163 : /* no outputs */
164 : "r" (cpu), "r" (mask), "r" (&page->flags),
165 "i" (NR_CPUS - 1UL)
166 : "g1", "g7");
169 extern void __update_mmu_cache(unsigned long mmu_context_hw, unsigned long address, pte_t pte, int code);
171 void update_mmu_cache(struct vm_area_struct *vma, unsigned long address, pte_t pte)
173 struct page *page;
174 unsigned long pfn;
175 unsigned long pg_flags;
177 pfn = pte_pfn(pte);
178 if (pfn_valid(pfn) &&
179 (page = pfn_to_page(pfn), page_mapping(page)) &&
180 ((pg_flags = page->flags) & (1UL << PG_dcache_dirty))) {
181 int cpu = ((pg_flags >> 24) & (NR_CPUS - 1UL));
182 int this_cpu = get_cpu();
184 /* This is just to optimize away some function calls
185 * in the SMP case.
187 if (cpu == this_cpu)
188 flush_dcache_page_impl(page);
189 else
190 smp_flush_dcache_page_impl(page, cpu);
192 clear_dcache_dirty_cpu(page, cpu);
194 put_cpu();
197 if (get_thread_fault_code())
198 __update_mmu_cache(CTX_NRBITS(vma->vm_mm->context),
199 address, pte, get_thread_fault_code());
202 void flush_dcache_page(struct page *page)
204 struct address_space *mapping;
205 int this_cpu;
207 /* Do not bother with the expensive D-cache flush if it
208 * is merely the zero page. The 'bigcore' testcase in GDB
209 * causes this case to run millions of times.
211 if (page == ZERO_PAGE(0))
212 return;
214 this_cpu = get_cpu();
216 mapping = page_mapping(page);
217 if (mapping && !mapping_mapped(mapping)) {
218 int dirty = test_bit(PG_dcache_dirty, &page->flags);
219 if (dirty) {
220 int dirty_cpu = dcache_dirty_cpu(page);
222 if (dirty_cpu == this_cpu)
223 goto out;
224 smp_flush_dcache_page_impl(page, dirty_cpu);
226 set_dcache_dirty(page, this_cpu);
227 } else {
228 /* We could delay the flush for the !page_mapping
229 * case too. But that case is for exec env/arg
230 * pages and those are %99 certainly going to get
231 * faulted into the tlb (and thus flushed) anyways.
233 flush_dcache_page_impl(page);
236 out:
237 put_cpu();
240 void flush_icache_range(unsigned long start, unsigned long end)
242 /* Cheetah has coherent I-cache. */
243 if (tlb_type == spitfire) {
244 unsigned long kaddr;
246 for (kaddr = start; kaddr < end; kaddr += PAGE_SIZE)
247 __flush_icache_page(__get_phys(kaddr));
251 unsigned long page_to_pfn(struct page *page)
253 return (unsigned long) ((page - mem_map) + pfn_base);
256 struct page *pfn_to_page(unsigned long pfn)
258 return (mem_map + (pfn - pfn_base));
261 void show_mem(void)
263 printk("Mem-info:\n");
264 show_free_areas();
265 printk("Free swap: %6ldkB\n",
266 nr_swap_pages << (PAGE_SHIFT-10));
267 printk("%ld pages of RAM\n", num_physpages);
268 printk("%d free pages\n", nr_free_pages());
269 printk("%d pages in page table cache\n",pgtable_cache_size);
272 void mmu_info(struct seq_file *m)
274 if (tlb_type == cheetah)
275 seq_printf(m, "MMU Type\t: Cheetah\n");
276 else if (tlb_type == cheetah_plus)
277 seq_printf(m, "MMU Type\t: Cheetah+\n");
278 else if (tlb_type == spitfire)
279 seq_printf(m, "MMU Type\t: Spitfire\n");
280 else
281 seq_printf(m, "MMU Type\t: ???\n");
283 #ifdef CONFIG_DEBUG_DCFLUSH
284 seq_printf(m, "DCPageFlushes\t: %d\n",
285 atomic_read(&dcpage_flushes));
286 #ifdef CONFIG_SMP
287 seq_printf(m, "DCPageFlushesXC\t: %d\n",
288 atomic_read(&dcpage_flushes_xcall));
289 #endif /* CONFIG_SMP */
290 #endif /* CONFIG_DEBUG_DCFLUSH */
293 struct linux_prom_translation {
294 unsigned long virt;
295 unsigned long size;
296 unsigned long data;
299 extern unsigned long prom_boot_page;
300 extern void prom_remap(unsigned long physpage, unsigned long virtpage, int mmu_ihandle);
301 extern int prom_get_mmu_ihandle(void);
302 extern void register_prom_callbacks(void);
304 /* Exported for SMP bootup purposes. */
305 unsigned long kern_locked_tte_data;
307 void __init early_pgtable_allocfail(char *type)
309 prom_printf("inherit_prom_mappings: Cannot alloc kernel %s.\n", type);
310 prom_halt();
313 #define BASE_PAGE_SIZE 8192
314 static pmd_t *prompmd;
317 * Translate PROM's mapping we capture at boot time into physical address.
318 * The second parameter is only set from prom_callback() invocations.
320 unsigned long prom_virt_to_phys(unsigned long promva, int *error)
322 pmd_t *pmdp = prompmd + ((promva >> 23) & 0x7ff);
323 pte_t *ptep;
324 unsigned long base;
326 if (pmd_none(*pmdp)) {
327 if (error)
328 *error = 1;
329 return(0);
331 ptep = (pte_t *)__pmd_page(*pmdp) + ((promva >> 13) & 0x3ff);
332 if (!pte_present(*ptep)) {
333 if (error)
334 *error = 1;
335 return(0);
337 if (error) {
338 *error = 0;
339 return(pte_val(*ptep));
341 base = pte_val(*ptep) & _PAGE_PADDR;
342 return(base + (promva & (BASE_PAGE_SIZE - 1)));
345 static void inherit_prom_mappings(void)
347 struct linux_prom_translation *trans;
348 unsigned long phys_page, tte_vaddr, tte_data;
349 void (*remap_func)(unsigned long, unsigned long, int);
350 pmd_t *pmdp;
351 pte_t *ptep;
352 int node, n, i, tsz;
353 extern unsigned int obp_iaddr_patch[2], obp_daddr_patch[2];
355 node = prom_finddevice("/virtual-memory");
356 n = prom_getproplen(node, "translations");
357 if (n == 0 || n == -1) {
358 prom_printf("Couldn't get translation property\n");
359 prom_halt();
361 n += 5 * sizeof(struct linux_prom_translation);
362 for (tsz = 1; tsz < n; tsz <<= 1)
363 /* empty */;
364 trans = __alloc_bootmem(tsz, SMP_CACHE_BYTES, bootmap_base);
365 if (trans == NULL) {
366 prom_printf("inherit_prom_mappings: Cannot alloc translations.\n");
367 prom_halt();
369 memset(trans, 0, tsz);
371 if ((n = prom_getproperty(node, "translations", (char *)trans, tsz)) == -1) {
372 prom_printf("Couldn't get translation property\n");
373 prom_halt();
375 n = n / sizeof(*trans);
378 * The obp translations are saved based on 8k pagesize, since obp can
379 * use a mixture of pagesizes. Misses to the 0xf0000000 - 0x100000000,
380 * ie obp range, are handled in entry.S and do not use the vpte scheme
381 * (see rant in inherit_locked_prom_mappings()).
383 #define OBP_PMD_SIZE 2048
384 prompmd = __alloc_bootmem(OBP_PMD_SIZE, OBP_PMD_SIZE, bootmap_base);
385 if (prompmd == NULL)
386 early_pgtable_allocfail("pmd");
387 memset(prompmd, 0, OBP_PMD_SIZE);
388 for (i = 0; i < n; i++) {
389 unsigned long vaddr;
391 if (trans[i].virt >= LOW_OBP_ADDRESS && trans[i].virt < HI_OBP_ADDRESS) {
392 for (vaddr = trans[i].virt;
393 ((vaddr < trans[i].virt + trans[i].size) &&
394 (vaddr < HI_OBP_ADDRESS));
395 vaddr += BASE_PAGE_SIZE) {
396 unsigned long val;
398 pmdp = prompmd + ((vaddr >> 23) & 0x7ff);
399 if (pmd_none(*pmdp)) {
400 ptep = __alloc_bootmem(BASE_PAGE_SIZE,
401 BASE_PAGE_SIZE,
402 bootmap_base);
403 if (ptep == NULL)
404 early_pgtable_allocfail("pte");
405 memset(ptep, 0, BASE_PAGE_SIZE);
406 pmd_set(pmdp, ptep);
408 ptep = (pte_t *)__pmd_page(*pmdp) +
409 ((vaddr >> 13) & 0x3ff);
411 val = trans[i].data;
413 /* Clear diag TTE bits. */
414 if (tlb_type == spitfire)
415 val &= ~0x0003fe0000000000UL;
417 set_pte_at(&init_mm, vaddr,
418 ptep, __pte(val | _PAGE_MODIFIED));
419 trans[i].data += BASE_PAGE_SIZE;
423 phys_page = __pa(prompmd);
424 obp_iaddr_patch[0] |= (phys_page >> 10);
425 obp_iaddr_patch[1] |= (phys_page & 0x3ff);
426 flushi((long)&obp_iaddr_patch[0]);
427 obp_daddr_patch[0] |= (phys_page >> 10);
428 obp_daddr_patch[1] |= (phys_page & 0x3ff);
429 flushi((long)&obp_daddr_patch[0]);
431 /* Now fixup OBP's idea about where we really are mapped. */
432 prom_printf("Remapping the kernel... ");
434 /* Spitfire Errata #32 workaround */
435 /* NOTE: Using plain zero for the context value is
436 * correct here, we are not using the Linux trap
437 * tables yet so we should not use the special
438 * UltraSPARC-III+ page size encodings yet.
440 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
441 "flush %%g6"
442 : /* No outputs */
443 : "r" (0), "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
445 switch (tlb_type) {
446 default:
447 case spitfire:
448 phys_page = spitfire_get_dtlb_data(sparc64_highest_locked_tlbent());
449 break;
451 case cheetah:
452 case cheetah_plus:
453 phys_page = cheetah_get_litlb_data(sparc64_highest_locked_tlbent());
454 break;
457 phys_page &= _PAGE_PADDR;
458 phys_page += ((unsigned long)&prom_boot_page -
459 (unsigned long)KERNBASE);
461 if (tlb_type == spitfire) {
462 /* Lock this into i/d tlb entry 59 */
463 __asm__ __volatile__(
464 "stxa %%g0, [%2] %3\n\t"
465 "stxa %0, [%1] %4\n\t"
466 "membar #Sync\n\t"
467 "flush %%g6\n\t"
468 "stxa %%g0, [%2] %5\n\t"
469 "stxa %0, [%1] %6\n\t"
470 "membar #Sync\n\t"
471 "flush %%g6"
472 : : "r" (phys_page | _PAGE_VALID | _PAGE_SZ8K | _PAGE_CP |
473 _PAGE_CV | _PAGE_P | _PAGE_L | _PAGE_W),
474 "r" (59 << 3), "r" (TLB_TAG_ACCESS),
475 "i" (ASI_DMMU), "i" (ASI_DTLB_DATA_ACCESS),
476 "i" (ASI_IMMU), "i" (ASI_ITLB_DATA_ACCESS)
477 : "memory");
478 } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
479 /* Lock this into i/d tlb-0 entry 11 */
480 __asm__ __volatile__(
481 "stxa %%g0, [%2] %3\n\t"
482 "stxa %0, [%1] %4\n\t"
483 "membar #Sync\n\t"
484 "flush %%g6\n\t"
485 "stxa %%g0, [%2] %5\n\t"
486 "stxa %0, [%1] %6\n\t"
487 "membar #Sync\n\t"
488 "flush %%g6"
489 : : "r" (phys_page | _PAGE_VALID | _PAGE_SZ8K | _PAGE_CP |
490 _PAGE_CV | _PAGE_P | _PAGE_L | _PAGE_W),
491 "r" ((0 << 16) | (11 << 3)), "r" (TLB_TAG_ACCESS),
492 "i" (ASI_DMMU), "i" (ASI_DTLB_DATA_ACCESS),
493 "i" (ASI_IMMU), "i" (ASI_ITLB_DATA_ACCESS)
494 : "memory");
495 } else {
496 /* Implement me :-) */
497 BUG();
500 tte_vaddr = (unsigned long) KERNBASE;
502 /* Spitfire Errata #32 workaround */
503 /* NOTE: Using plain zero for the context value is
504 * correct here, we are not using the Linux trap
505 * tables yet so we should not use the special
506 * UltraSPARC-III+ page size encodings yet.
508 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
509 "flush %%g6"
510 : /* No outputs */
511 : "r" (0),
512 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
514 if (tlb_type == spitfire)
515 tte_data = spitfire_get_dtlb_data(sparc64_highest_locked_tlbent());
516 else
517 tte_data = cheetah_get_ldtlb_data(sparc64_highest_locked_tlbent());
519 kern_locked_tte_data = tte_data;
521 remap_func = (void *) ((unsigned long) &prom_remap -
522 (unsigned long) &prom_boot_page);
525 /* Spitfire Errata #32 workaround */
526 /* NOTE: Using plain zero for the context value is
527 * correct here, we are not using the Linux trap
528 * tables yet so we should not use the special
529 * UltraSPARC-III+ page size encodings yet.
531 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
532 "flush %%g6"
533 : /* No outputs */
534 : "r" (0),
535 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
537 remap_func((tlb_type == spitfire ?
538 (spitfire_get_dtlb_data(sparc64_highest_locked_tlbent()) & _PAGE_PADDR) :
539 (cheetah_get_litlb_data(sparc64_highest_locked_tlbent()) & _PAGE_PADDR)),
540 (unsigned long) KERNBASE,
541 prom_get_mmu_ihandle());
543 if (bigkernel)
544 remap_func(((tte_data + 0x400000) & _PAGE_PADDR),
545 (unsigned long) KERNBASE + 0x400000, prom_get_mmu_ihandle());
547 /* Flush out that temporary mapping. */
548 spitfire_flush_dtlb_nucleus_page(0x0);
549 spitfire_flush_itlb_nucleus_page(0x0);
551 /* Now lock us back into the TLBs via OBP. */
552 prom_dtlb_load(sparc64_highest_locked_tlbent(), tte_data, tte_vaddr);
553 prom_itlb_load(sparc64_highest_locked_tlbent(), tte_data, tte_vaddr);
554 if (bigkernel) {
555 prom_dtlb_load(sparc64_highest_locked_tlbent()-1, tte_data + 0x400000,
556 tte_vaddr + 0x400000);
557 prom_itlb_load(sparc64_highest_locked_tlbent()-1, tte_data + 0x400000,
558 tte_vaddr + 0x400000);
561 /* Re-read translations property. */
562 if ((n = prom_getproperty(node, "translations", (char *)trans, tsz)) == -1) {
563 prom_printf("Couldn't get translation property\n");
564 prom_halt();
566 n = n / sizeof(*trans);
568 for (i = 0; i < n; i++) {
569 unsigned long vaddr = trans[i].virt;
570 unsigned long size = trans[i].size;
572 if (vaddr < 0xf0000000UL) {
573 unsigned long avoid_start = (unsigned long) KERNBASE;
574 unsigned long avoid_end = avoid_start + (4 * 1024 * 1024);
576 if (bigkernel)
577 avoid_end += (4 * 1024 * 1024);
578 if (vaddr < avoid_start) {
579 unsigned long top = vaddr + size;
581 if (top > avoid_start)
582 top = avoid_start;
583 prom_unmap(top - vaddr, vaddr);
585 if ((vaddr + size) > avoid_end) {
586 unsigned long bottom = vaddr;
588 if (bottom < avoid_end)
589 bottom = avoid_end;
590 prom_unmap((vaddr + size) - bottom, bottom);
595 prom_printf("done.\n");
597 register_prom_callbacks();
600 /* The OBP specifications for sun4u mark 0xfffffffc00000000 and
601 * upwards as reserved for use by the firmware (I wonder if this
602 * will be the same on Cheetah...). We use this virtual address
603 * range for the VPTE table mappings of the nucleus so we need
604 * to zap them when we enter the PROM. -DaveM
606 static void __flush_nucleus_vptes(void)
608 unsigned long prom_reserved_base = 0xfffffffc00000000UL;
609 int i;
611 /* Only DTLB must be checked for VPTE entries. */
612 if (tlb_type == spitfire) {
613 for (i = 0; i < 63; i++) {
614 unsigned long tag;
616 /* Spitfire Errata #32 workaround */
617 /* NOTE: Always runs on spitfire, so no cheetah+
618 * page size encodings.
620 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
621 "flush %%g6"
622 : /* No outputs */
623 : "r" (0),
624 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
626 tag = spitfire_get_dtlb_tag(i);
627 if (((tag & ~(PAGE_MASK)) == 0) &&
628 ((tag & (PAGE_MASK)) >= prom_reserved_base)) {
629 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
630 "membar #Sync"
631 : /* no outputs */
632 : "r" (TLB_TAG_ACCESS), "i" (ASI_DMMU));
633 spitfire_put_dtlb_data(i, 0x0UL);
636 } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
637 for (i = 0; i < 512; i++) {
638 unsigned long tag = cheetah_get_dtlb_tag(i, 2);
640 if ((tag & ~PAGE_MASK) == 0 &&
641 (tag & PAGE_MASK) >= prom_reserved_base) {
642 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
643 "membar #Sync"
644 : /* no outputs */
645 : "r" (TLB_TAG_ACCESS), "i" (ASI_DMMU));
646 cheetah_put_dtlb_data(i, 0x0UL, 2);
649 if (tlb_type != cheetah_plus)
650 continue;
652 tag = cheetah_get_dtlb_tag(i, 3);
654 if ((tag & ~PAGE_MASK) == 0 &&
655 (tag & PAGE_MASK) >= prom_reserved_base) {
656 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
657 "membar #Sync"
658 : /* no outputs */
659 : "r" (TLB_TAG_ACCESS), "i" (ASI_DMMU));
660 cheetah_put_dtlb_data(i, 0x0UL, 3);
663 } else {
664 /* Implement me :-) */
665 BUG();
669 static int prom_ditlb_set;
670 struct prom_tlb_entry {
671 int tlb_ent;
672 unsigned long tlb_tag;
673 unsigned long tlb_data;
675 struct prom_tlb_entry prom_itlb[16], prom_dtlb[16];
677 void prom_world(int enter)
679 unsigned long pstate;
680 int i;
682 if (!enter)
683 set_fs((mm_segment_t) { get_thread_current_ds() });
685 if (!prom_ditlb_set)
686 return;
688 /* Make sure the following runs atomically. */
689 __asm__ __volatile__("flushw\n\t"
690 "rdpr %%pstate, %0\n\t"
691 "wrpr %0, %1, %%pstate"
692 : "=r" (pstate)
693 : "i" (PSTATE_IE));
695 if (enter) {
696 /* Kick out nucleus VPTEs. */
697 __flush_nucleus_vptes();
699 /* Install PROM world. */
700 for (i = 0; i < 16; i++) {
701 if (prom_dtlb[i].tlb_ent != -1) {
702 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
703 "membar #Sync"
704 : : "r" (prom_dtlb[i].tlb_tag), "r" (TLB_TAG_ACCESS),
705 "i" (ASI_DMMU));
706 if (tlb_type == spitfire)
707 spitfire_put_dtlb_data(prom_dtlb[i].tlb_ent,
708 prom_dtlb[i].tlb_data);
709 else if (tlb_type == cheetah || tlb_type == cheetah_plus)
710 cheetah_put_ldtlb_data(prom_dtlb[i].tlb_ent,
711 prom_dtlb[i].tlb_data);
713 if (prom_itlb[i].tlb_ent != -1) {
714 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
715 "membar #Sync"
716 : : "r" (prom_itlb[i].tlb_tag),
717 "r" (TLB_TAG_ACCESS),
718 "i" (ASI_IMMU));
719 if (tlb_type == spitfire)
720 spitfire_put_itlb_data(prom_itlb[i].tlb_ent,
721 prom_itlb[i].tlb_data);
722 else if (tlb_type == cheetah || tlb_type == cheetah_plus)
723 cheetah_put_litlb_data(prom_itlb[i].tlb_ent,
724 prom_itlb[i].tlb_data);
727 } else {
728 for (i = 0; i < 16; i++) {
729 if (prom_dtlb[i].tlb_ent != -1) {
730 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
731 "membar #Sync"
732 : : "r" (TLB_TAG_ACCESS), "i" (ASI_DMMU));
733 if (tlb_type == spitfire)
734 spitfire_put_dtlb_data(prom_dtlb[i].tlb_ent, 0x0UL);
735 else
736 cheetah_put_ldtlb_data(prom_dtlb[i].tlb_ent, 0x0UL);
738 if (prom_itlb[i].tlb_ent != -1) {
739 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
740 "membar #Sync"
741 : : "r" (TLB_TAG_ACCESS),
742 "i" (ASI_IMMU));
743 if (tlb_type == spitfire)
744 spitfire_put_itlb_data(prom_itlb[i].tlb_ent, 0x0UL);
745 else
746 cheetah_put_litlb_data(prom_itlb[i].tlb_ent, 0x0UL);
750 __asm__ __volatile__("wrpr %0, 0, %%pstate"
751 : : "r" (pstate));
754 void inherit_locked_prom_mappings(int save_p)
756 int i;
757 int dtlb_seen = 0;
758 int itlb_seen = 0;
760 /* Fucking losing PROM has more mappings in the TLB, but
761 * it (conveniently) fails to mention any of these in the
762 * translations property. The only ones that matter are
763 * the locked PROM tlb entries, so we impose the following
764 * irrecovable rule on the PROM, it is allowed 8 locked
765 * entries in the ITLB and 8 in the DTLB.
767 * Supposedly the upper 16GB of the address space is
768 * reserved for OBP, BUT I WISH THIS WAS DOCUMENTED
769 * SOMEWHERE!!!!!!!!!!!!!!!!! Furthermore the entire interface
770 * used between the client program and the firmware on sun5
771 * systems to coordinate mmu mappings is also COMPLETELY
772 * UNDOCUMENTED!!!!!! Thanks S(t)un!
774 if (save_p) {
775 for (i = 0; i < 16; i++) {
776 prom_itlb[i].tlb_ent = -1;
777 prom_dtlb[i].tlb_ent = -1;
780 if (tlb_type == spitfire) {
781 int high = SPITFIRE_HIGHEST_LOCKED_TLBENT - bigkernel;
782 for (i = 0; i < high; i++) {
783 unsigned long data;
785 /* Spitfire Errata #32 workaround */
786 /* NOTE: Always runs on spitfire, so no cheetah+
787 * page size encodings.
789 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
790 "flush %%g6"
791 : /* No outputs */
792 : "r" (0),
793 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
795 data = spitfire_get_dtlb_data(i);
796 if ((data & (_PAGE_L|_PAGE_VALID)) == (_PAGE_L|_PAGE_VALID)) {
797 unsigned long tag;
799 /* Spitfire Errata #32 workaround */
800 /* NOTE: Always runs on spitfire, so no
801 * cheetah+ page size encodings.
803 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
804 "flush %%g6"
805 : /* No outputs */
806 : "r" (0),
807 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
809 tag = spitfire_get_dtlb_tag(i);
810 if (save_p) {
811 prom_dtlb[dtlb_seen].tlb_ent = i;
812 prom_dtlb[dtlb_seen].tlb_tag = tag;
813 prom_dtlb[dtlb_seen].tlb_data = data;
815 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
816 "membar #Sync"
817 : : "r" (TLB_TAG_ACCESS), "i" (ASI_DMMU));
818 spitfire_put_dtlb_data(i, 0x0UL);
820 dtlb_seen++;
821 if (dtlb_seen > 15)
822 break;
826 for (i = 0; i < high; i++) {
827 unsigned long data;
829 /* Spitfire Errata #32 workaround */
830 /* NOTE: Always runs on spitfire, so no
831 * cheetah+ page size encodings.
833 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
834 "flush %%g6"
835 : /* No outputs */
836 : "r" (0),
837 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
839 data = spitfire_get_itlb_data(i);
840 if ((data & (_PAGE_L|_PAGE_VALID)) == (_PAGE_L|_PAGE_VALID)) {
841 unsigned long tag;
843 /* Spitfire Errata #32 workaround */
844 /* NOTE: Always runs on spitfire, so no
845 * cheetah+ page size encodings.
847 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
848 "flush %%g6"
849 : /* No outputs */
850 : "r" (0),
851 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
853 tag = spitfire_get_itlb_tag(i);
854 if (save_p) {
855 prom_itlb[itlb_seen].tlb_ent = i;
856 prom_itlb[itlb_seen].tlb_tag = tag;
857 prom_itlb[itlb_seen].tlb_data = data;
859 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
860 "membar #Sync"
861 : : "r" (TLB_TAG_ACCESS), "i" (ASI_IMMU));
862 spitfire_put_itlb_data(i, 0x0UL);
864 itlb_seen++;
865 if (itlb_seen > 15)
866 break;
869 } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
870 int high = CHEETAH_HIGHEST_LOCKED_TLBENT - bigkernel;
872 for (i = 0; i < high; i++) {
873 unsigned long data;
875 data = cheetah_get_ldtlb_data(i);
876 if ((data & (_PAGE_L|_PAGE_VALID)) == (_PAGE_L|_PAGE_VALID)) {
877 unsigned long tag;
879 tag = cheetah_get_ldtlb_tag(i);
880 if (save_p) {
881 prom_dtlb[dtlb_seen].tlb_ent = i;
882 prom_dtlb[dtlb_seen].tlb_tag = tag;
883 prom_dtlb[dtlb_seen].tlb_data = data;
885 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
886 "membar #Sync"
887 : : "r" (TLB_TAG_ACCESS), "i" (ASI_DMMU));
888 cheetah_put_ldtlb_data(i, 0x0UL);
890 dtlb_seen++;
891 if (dtlb_seen > 15)
892 break;
896 for (i = 0; i < high; i++) {
897 unsigned long data;
899 data = cheetah_get_litlb_data(i);
900 if ((data & (_PAGE_L|_PAGE_VALID)) == (_PAGE_L|_PAGE_VALID)) {
901 unsigned long tag;
903 tag = cheetah_get_litlb_tag(i);
904 if (save_p) {
905 prom_itlb[itlb_seen].tlb_ent = i;
906 prom_itlb[itlb_seen].tlb_tag = tag;
907 prom_itlb[itlb_seen].tlb_data = data;
909 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
910 "membar #Sync"
911 : : "r" (TLB_TAG_ACCESS), "i" (ASI_IMMU));
912 cheetah_put_litlb_data(i, 0x0UL);
914 itlb_seen++;
915 if (itlb_seen > 15)
916 break;
919 } else {
920 /* Implement me :-) */
921 BUG();
923 if (save_p)
924 prom_ditlb_set = 1;
927 /* Give PROM back his world, done during reboots... */
928 void prom_reload_locked(void)
930 int i;
932 for (i = 0; i < 16; i++) {
933 if (prom_dtlb[i].tlb_ent != -1) {
934 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
935 "membar #Sync"
936 : : "r" (prom_dtlb[i].tlb_tag), "r" (TLB_TAG_ACCESS),
937 "i" (ASI_DMMU));
938 if (tlb_type == spitfire)
939 spitfire_put_dtlb_data(prom_dtlb[i].tlb_ent,
940 prom_dtlb[i].tlb_data);
941 else if (tlb_type == cheetah || tlb_type == cheetah_plus)
942 cheetah_put_ldtlb_data(prom_dtlb[i].tlb_ent,
943 prom_dtlb[i].tlb_data);
946 if (prom_itlb[i].tlb_ent != -1) {
947 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
948 "membar #Sync"
949 : : "r" (prom_itlb[i].tlb_tag),
950 "r" (TLB_TAG_ACCESS),
951 "i" (ASI_IMMU));
952 if (tlb_type == spitfire)
953 spitfire_put_itlb_data(prom_itlb[i].tlb_ent,
954 prom_itlb[i].tlb_data);
955 else
956 cheetah_put_litlb_data(prom_itlb[i].tlb_ent,
957 prom_itlb[i].tlb_data);
962 #ifdef DCACHE_ALIASING_POSSIBLE
963 void __flush_dcache_range(unsigned long start, unsigned long end)
965 unsigned long va;
967 if (tlb_type == spitfire) {
968 int n = 0;
970 for (va = start; va < end; va += 32) {
971 spitfire_put_dcache_tag(va & 0x3fe0, 0x0);
972 if (++n >= 512)
973 break;
975 } else {
976 start = __pa(start);
977 end = __pa(end);
978 for (va = start; va < end; va += 32)
979 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
980 "membar #Sync"
981 : /* no outputs */
982 : "r" (va),
983 "i" (ASI_DCACHE_INVALIDATE));
986 #endif /* DCACHE_ALIASING_POSSIBLE */
988 /* If not locked, zap it. */
989 void __flush_tlb_all(void)
991 unsigned long pstate;
992 int i;
994 __asm__ __volatile__("flushw\n\t"
995 "rdpr %%pstate, %0\n\t"
996 "wrpr %0, %1, %%pstate"
997 : "=r" (pstate)
998 : "i" (PSTATE_IE));
999 if (tlb_type == spitfire) {
1000 for (i = 0; i < 64; i++) {
1001 /* Spitfire Errata #32 workaround */
1002 /* NOTE: Always runs on spitfire, so no
1003 * cheetah+ page size encodings.
1005 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
1006 "flush %%g6"
1007 : /* No outputs */
1008 : "r" (0),
1009 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
1011 if (!(spitfire_get_dtlb_data(i) & _PAGE_L)) {
1012 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
1013 "membar #Sync"
1014 : /* no outputs */
1015 : "r" (TLB_TAG_ACCESS), "i" (ASI_DMMU));
1016 spitfire_put_dtlb_data(i, 0x0UL);
1019 /* Spitfire Errata #32 workaround */
1020 /* NOTE: Always runs on spitfire, so no
1021 * cheetah+ page size encodings.
1023 __asm__ __volatile__("stxa %0, [%1] %2\n\t"
1024 "flush %%g6"
1025 : /* No outputs */
1026 : "r" (0),
1027 "r" (PRIMARY_CONTEXT), "i" (ASI_DMMU));
1029 if (!(spitfire_get_itlb_data(i) & _PAGE_L)) {
1030 __asm__ __volatile__("stxa %%g0, [%0] %1\n\t"
1031 "membar #Sync"
1032 : /* no outputs */
1033 : "r" (TLB_TAG_ACCESS), "i" (ASI_IMMU));
1034 spitfire_put_itlb_data(i, 0x0UL);
1037 } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
1038 cheetah_flush_dtlb_all();
1039 cheetah_flush_itlb_all();
1041 __asm__ __volatile__("wrpr %0, 0, %%pstate"
1042 : : "r" (pstate));
1045 /* Caller does TLB context flushing on local CPU if necessary.
1046 * The caller also ensures that CTX_VALID(mm->context) is false.
1048 * We must be careful about boundary cases so that we never
1049 * let the user have CTX 0 (nucleus) or we ever use a CTX
1050 * version of zero (and thus NO_CONTEXT would not be caught
1051 * by version mis-match tests in mmu_context.h).
1053 void get_new_mmu_context(struct mm_struct *mm)
1055 unsigned long ctx, new_ctx;
1056 unsigned long orig_pgsz_bits;
1059 spin_lock(&ctx_alloc_lock);
1060 orig_pgsz_bits = (mm->context.sparc64_ctx_val & CTX_PGSZ_MASK);
1061 ctx = (tlb_context_cache + 1) & CTX_NR_MASK;
1062 new_ctx = find_next_zero_bit(mmu_context_bmap, 1 << CTX_NR_BITS, ctx);
1063 if (new_ctx >= (1 << CTX_NR_BITS)) {
1064 new_ctx = find_next_zero_bit(mmu_context_bmap, ctx, 1);
1065 if (new_ctx >= ctx) {
1066 int i;
1067 new_ctx = (tlb_context_cache & CTX_VERSION_MASK) +
1068 CTX_FIRST_VERSION;
1069 if (new_ctx == 1)
1070 new_ctx = CTX_FIRST_VERSION;
1072 /* Don't call memset, for 16 entries that's just
1073 * plain silly...
1075 mmu_context_bmap[0] = 3;
1076 mmu_context_bmap[1] = 0;
1077 mmu_context_bmap[2] = 0;
1078 mmu_context_bmap[3] = 0;
1079 for (i = 4; i < CTX_BMAP_SLOTS; i += 4) {
1080 mmu_context_bmap[i + 0] = 0;
1081 mmu_context_bmap[i + 1] = 0;
1082 mmu_context_bmap[i + 2] = 0;
1083 mmu_context_bmap[i + 3] = 0;
1085 goto out;
1088 mmu_context_bmap[new_ctx>>6] |= (1UL << (new_ctx & 63));
1089 new_ctx |= (tlb_context_cache & CTX_VERSION_MASK);
1090 out:
1091 tlb_context_cache = new_ctx;
1092 mm->context.sparc64_ctx_val = new_ctx | orig_pgsz_bits;
1093 spin_unlock(&ctx_alloc_lock);
1096 #ifndef CONFIG_SMP
1097 struct pgtable_cache_struct pgt_quicklists;
1098 #endif
1100 /* OK, we have to color these pages. The page tables are accessed
1101 * by non-Dcache enabled mapping in the VPTE area by the dtlb_backend.S
1102 * code, as well as by PAGE_OFFSET range direct-mapped addresses by
1103 * other parts of the kernel. By coloring, we make sure that the tlbmiss
1104 * fast handlers do not get data from old/garbage dcache lines that
1105 * correspond to an old/stale virtual address (user/kernel) that
1106 * previously mapped the pagetable page while accessing vpte range
1107 * addresses. The idea is that if the vpte color and PAGE_OFFSET range
1108 * color is the same, then when the kernel initializes the pagetable
1109 * using the later address range, accesses with the first address
1110 * range will see the newly initialized data rather than the garbage.
1112 #ifdef DCACHE_ALIASING_POSSIBLE
1113 #define DC_ALIAS_SHIFT 1
1114 #else
1115 #define DC_ALIAS_SHIFT 0
1116 #endif
1117 pte_t *pte_alloc_one_kernel(struct mm_struct *mm, unsigned long address)
1119 struct page *page;
1120 unsigned long color;
1123 pte_t *ptep = pte_alloc_one_fast(mm, address);
1125 if (ptep)
1126 return ptep;
1129 color = VPTE_COLOR(address);
1130 page = alloc_pages(GFP_KERNEL|__GFP_REPEAT, DC_ALIAS_SHIFT);
1131 if (page) {
1132 unsigned long *to_free;
1133 unsigned long paddr;
1134 pte_t *pte;
1136 #ifdef DCACHE_ALIASING_POSSIBLE
1137 set_page_count(page, 1);
1138 ClearPageCompound(page);
1140 set_page_count((page + 1), 1);
1141 ClearPageCompound(page + 1);
1142 #endif
1143 paddr = (unsigned long) page_address(page);
1144 memset((char *)paddr, 0, (PAGE_SIZE << DC_ALIAS_SHIFT));
1146 if (!color) {
1147 pte = (pte_t *) paddr;
1148 to_free = (unsigned long *) (paddr + PAGE_SIZE);
1149 } else {
1150 pte = (pte_t *) (paddr + PAGE_SIZE);
1151 to_free = (unsigned long *) paddr;
1154 #ifdef DCACHE_ALIASING_POSSIBLE
1155 /* Now free the other one up, adjust cache size. */
1156 preempt_disable();
1157 *to_free = (unsigned long) pte_quicklist[color ^ 0x1];
1158 pte_quicklist[color ^ 0x1] = to_free;
1159 pgtable_cache_size++;
1160 preempt_enable();
1161 #endif
1163 return pte;
1165 return NULL;
1168 void sparc_ultra_dump_itlb(void)
1170 int slot;
1172 if (tlb_type == spitfire) {
1173 printk ("Contents of itlb: ");
1174 for (slot = 0; slot < 14; slot++) printk (" ");
1175 printk ("%2x:%016lx,%016lx\n",
1177 spitfire_get_itlb_tag(0), spitfire_get_itlb_data(0));
1178 for (slot = 1; slot < 64; slot+=3) {
1179 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx %2x:%016lx,%016lx\n",
1180 slot,
1181 spitfire_get_itlb_tag(slot), spitfire_get_itlb_data(slot),
1182 slot+1,
1183 spitfire_get_itlb_tag(slot+1), spitfire_get_itlb_data(slot+1),
1184 slot+2,
1185 spitfire_get_itlb_tag(slot+2), spitfire_get_itlb_data(slot+2));
1187 } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
1188 printk ("Contents of itlb0:\n");
1189 for (slot = 0; slot < 16; slot+=2) {
1190 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
1191 slot,
1192 cheetah_get_litlb_tag(slot), cheetah_get_litlb_data(slot),
1193 slot+1,
1194 cheetah_get_litlb_tag(slot+1), cheetah_get_litlb_data(slot+1));
1196 printk ("Contents of itlb2:\n");
1197 for (slot = 0; slot < 128; slot+=2) {
1198 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
1199 slot,
1200 cheetah_get_itlb_tag(slot), cheetah_get_itlb_data(slot),
1201 slot+1,
1202 cheetah_get_itlb_tag(slot+1), cheetah_get_itlb_data(slot+1));
1207 void sparc_ultra_dump_dtlb(void)
1209 int slot;
1211 if (tlb_type == spitfire) {
1212 printk ("Contents of dtlb: ");
1213 for (slot = 0; slot < 14; slot++) printk (" ");
1214 printk ("%2x:%016lx,%016lx\n", 0,
1215 spitfire_get_dtlb_tag(0), spitfire_get_dtlb_data(0));
1216 for (slot = 1; slot < 64; slot+=3) {
1217 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx %2x:%016lx,%016lx\n",
1218 slot,
1219 spitfire_get_dtlb_tag(slot), spitfire_get_dtlb_data(slot),
1220 slot+1,
1221 spitfire_get_dtlb_tag(slot+1), spitfire_get_dtlb_data(slot+1),
1222 slot+2,
1223 spitfire_get_dtlb_tag(slot+2), spitfire_get_dtlb_data(slot+2));
1225 } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
1226 printk ("Contents of dtlb0:\n");
1227 for (slot = 0; slot < 16; slot+=2) {
1228 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
1229 slot,
1230 cheetah_get_ldtlb_tag(slot), cheetah_get_ldtlb_data(slot),
1231 slot+1,
1232 cheetah_get_ldtlb_tag(slot+1), cheetah_get_ldtlb_data(slot+1));
1234 printk ("Contents of dtlb2:\n");
1235 for (slot = 0; slot < 512; slot+=2) {
1236 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
1237 slot,
1238 cheetah_get_dtlb_tag(slot, 2), cheetah_get_dtlb_data(slot, 2),
1239 slot+1,
1240 cheetah_get_dtlb_tag(slot+1, 2), cheetah_get_dtlb_data(slot+1, 2));
1242 if (tlb_type == cheetah_plus) {
1243 printk ("Contents of dtlb3:\n");
1244 for (slot = 0; slot < 512; slot+=2) {
1245 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
1246 slot,
1247 cheetah_get_dtlb_tag(slot, 3), cheetah_get_dtlb_data(slot, 3),
1248 slot+1,
1249 cheetah_get_dtlb_tag(slot+1, 3), cheetah_get_dtlb_data(slot+1, 3));
1255 extern unsigned long cmdline_memory_size;
1257 unsigned long __init bootmem_init(unsigned long *pages_avail)
1259 unsigned long bootmap_size, start_pfn, end_pfn;
1260 unsigned long end_of_phys_memory = 0UL;
1261 unsigned long bootmap_pfn, bytes_avail, size;
1262 int i;
1264 #ifdef CONFIG_DEBUG_BOOTMEM
1265 prom_printf("bootmem_init: Scan sp_banks, ");
1266 #endif
1268 bytes_avail = 0UL;
1269 for (i = 0; sp_banks[i].num_bytes != 0; i++) {
1270 end_of_phys_memory = sp_banks[i].base_addr +
1271 sp_banks[i].num_bytes;
1272 bytes_avail += sp_banks[i].num_bytes;
1273 if (cmdline_memory_size) {
1274 if (bytes_avail > cmdline_memory_size) {
1275 unsigned long slack = bytes_avail - cmdline_memory_size;
1277 bytes_avail -= slack;
1278 end_of_phys_memory -= slack;
1280 sp_banks[i].num_bytes -= slack;
1281 if (sp_banks[i].num_bytes == 0) {
1282 sp_banks[i].base_addr = 0xdeadbeef;
1283 } else {
1284 sp_banks[i+1].num_bytes = 0;
1285 sp_banks[i+1].base_addr = 0xdeadbeef;
1287 break;
1292 *pages_avail = bytes_avail >> PAGE_SHIFT;
1294 /* Start with page aligned address of last symbol in kernel
1295 * image. The kernel is hard mapped below PAGE_OFFSET in a
1296 * 4MB locked TLB translation.
1298 start_pfn = PAGE_ALIGN(kern_base + kern_size) >> PAGE_SHIFT;
1300 bootmap_pfn = start_pfn;
1302 end_pfn = end_of_phys_memory >> PAGE_SHIFT;
1304 #ifdef CONFIG_BLK_DEV_INITRD
1305 /* Now have to check initial ramdisk, so that bootmap does not overwrite it */
1306 if (sparc_ramdisk_image || sparc_ramdisk_image64) {
1307 unsigned long ramdisk_image = sparc_ramdisk_image ?
1308 sparc_ramdisk_image : sparc_ramdisk_image64;
1309 if (ramdisk_image >= (unsigned long)_end - 2 * PAGE_SIZE)
1310 ramdisk_image -= KERNBASE;
1311 initrd_start = ramdisk_image + phys_base;
1312 initrd_end = initrd_start + sparc_ramdisk_size;
1313 if (initrd_end > end_of_phys_memory) {
1314 printk(KERN_CRIT "initrd extends beyond end of memory "
1315 "(0x%016lx > 0x%016lx)\ndisabling initrd\n",
1316 initrd_end, end_of_phys_memory);
1317 initrd_start = 0;
1319 if (initrd_start) {
1320 if (initrd_start >= (start_pfn << PAGE_SHIFT) &&
1321 initrd_start < (start_pfn << PAGE_SHIFT) + 2 * PAGE_SIZE)
1322 bootmap_pfn = PAGE_ALIGN (initrd_end) >> PAGE_SHIFT;
1325 #endif
1326 /* Initialize the boot-time allocator. */
1327 max_pfn = max_low_pfn = end_pfn;
1328 min_low_pfn = pfn_base;
1330 #ifdef CONFIG_DEBUG_BOOTMEM
1331 prom_printf("init_bootmem(min[%lx], bootmap[%lx], max[%lx])\n",
1332 min_low_pfn, bootmap_pfn, max_low_pfn);
1333 #endif
1334 bootmap_size = init_bootmem_node(NODE_DATA(0), bootmap_pfn, pfn_base, end_pfn);
1336 bootmap_base = bootmap_pfn << PAGE_SHIFT;
1338 /* Now register the available physical memory with the
1339 * allocator.
1341 for (i = 0; sp_banks[i].num_bytes != 0; i++) {
1342 #ifdef CONFIG_DEBUG_BOOTMEM
1343 prom_printf("free_bootmem(sp_banks:%d): base[%lx] size[%lx]\n",
1344 i, sp_banks[i].base_addr, sp_banks[i].num_bytes);
1345 #endif
1346 free_bootmem(sp_banks[i].base_addr, sp_banks[i].num_bytes);
1349 #ifdef CONFIG_BLK_DEV_INITRD
1350 if (initrd_start) {
1351 size = initrd_end - initrd_start;
1353 /* Resert the initrd image area. */
1354 #ifdef CONFIG_DEBUG_BOOTMEM
1355 prom_printf("reserve_bootmem(initrd): base[%llx] size[%lx]\n",
1356 initrd_start, initrd_end);
1357 #endif
1358 reserve_bootmem(initrd_start, size);
1359 *pages_avail -= PAGE_ALIGN(size) >> PAGE_SHIFT;
1361 initrd_start += PAGE_OFFSET;
1362 initrd_end += PAGE_OFFSET;
1364 #endif
1365 /* Reserve the kernel text/data/bss. */
1366 #ifdef CONFIG_DEBUG_BOOTMEM
1367 prom_printf("reserve_bootmem(kernel): base[%lx] size[%lx]\n", kern_base, kern_size);
1368 #endif
1369 reserve_bootmem(kern_base, kern_size);
1370 *pages_avail -= PAGE_ALIGN(kern_size) >> PAGE_SHIFT;
1372 /* Reserve the bootmem map. We do not account for it
1373 * in pages_avail because we will release that memory
1374 * in free_all_bootmem.
1376 size = bootmap_size;
1377 #ifdef CONFIG_DEBUG_BOOTMEM
1378 prom_printf("reserve_bootmem(bootmap): base[%lx] size[%lx]\n",
1379 (bootmap_pfn << PAGE_SHIFT), size);
1380 #endif
1381 reserve_bootmem((bootmap_pfn << PAGE_SHIFT), size);
1382 *pages_avail -= PAGE_ALIGN(size) >> PAGE_SHIFT;
1384 return end_pfn;
1387 /* paging_init() sets up the page tables */
1389 extern void cheetah_ecache_flush_init(void);
1391 static unsigned long last_valid_pfn;
1393 void __init paging_init(void)
1395 extern pmd_t swapper_pmd_dir[1024];
1396 extern unsigned int sparc64_vpte_patchme1[1];
1397 extern unsigned int sparc64_vpte_patchme2[1];
1398 unsigned long alias_base = kern_base + PAGE_OFFSET;
1399 unsigned long second_alias_page = 0;
1400 unsigned long pt, flags, end_pfn, pages_avail;
1401 unsigned long shift = alias_base - ((unsigned long)KERNBASE);
1402 unsigned long real_end;
1404 set_bit(0, mmu_context_bmap);
1406 real_end = (unsigned long)_end;
1407 if ((real_end > ((unsigned long)KERNBASE + 0x400000)))
1408 bigkernel = 1;
1409 #ifdef CONFIG_BLK_DEV_INITRD
1410 if (sparc_ramdisk_image || sparc_ramdisk_image64)
1411 real_end = (PAGE_ALIGN(real_end) + PAGE_ALIGN(sparc_ramdisk_size));
1412 #endif
1414 /* We assume physical memory starts at some 4mb multiple,
1415 * if this were not true we wouldn't boot up to this point
1416 * anyways.
1418 pt = kern_base | _PAGE_VALID | _PAGE_SZ4MB;
1419 pt |= _PAGE_CP | _PAGE_CV | _PAGE_P | _PAGE_L | _PAGE_W;
1420 local_irq_save(flags);
1421 if (tlb_type == spitfire) {
1422 __asm__ __volatile__(
1423 " stxa %1, [%0] %3\n"
1424 " stxa %2, [%5] %4\n"
1425 " membar #Sync\n"
1426 " flush %%g6\n"
1427 " nop\n"
1428 " nop\n"
1429 " nop\n"
1430 : /* No outputs */
1431 : "r" (TLB_TAG_ACCESS), "r" (alias_base), "r" (pt),
1432 "i" (ASI_DMMU), "i" (ASI_DTLB_DATA_ACCESS), "r" (61 << 3)
1433 : "memory");
1434 if (real_end >= KERNBASE + 0x340000) {
1435 second_alias_page = alias_base + 0x400000;
1436 __asm__ __volatile__(
1437 " stxa %1, [%0] %3\n"
1438 " stxa %2, [%5] %4\n"
1439 " membar #Sync\n"
1440 " flush %%g6\n"
1441 " nop\n"
1442 " nop\n"
1443 " nop\n"
1444 : /* No outputs */
1445 : "r" (TLB_TAG_ACCESS), "r" (second_alias_page), "r" (pt + 0x400000),
1446 "i" (ASI_DMMU), "i" (ASI_DTLB_DATA_ACCESS), "r" (60 << 3)
1447 : "memory");
1449 } else if (tlb_type == cheetah || tlb_type == cheetah_plus) {
1450 __asm__ __volatile__(
1451 " stxa %1, [%0] %3\n"
1452 " stxa %2, [%5] %4\n"
1453 " membar #Sync\n"
1454 " flush %%g6\n"
1455 " nop\n"
1456 " nop\n"
1457 " nop\n"
1458 : /* No outputs */
1459 : "r" (TLB_TAG_ACCESS), "r" (alias_base), "r" (pt),
1460 "i" (ASI_DMMU), "i" (ASI_DTLB_DATA_ACCESS), "r" ((0<<16) | (13<<3))
1461 : "memory");
1462 if (real_end >= KERNBASE + 0x340000) {
1463 second_alias_page = alias_base + 0x400000;
1464 __asm__ __volatile__(
1465 " stxa %1, [%0] %3\n"
1466 " stxa %2, [%5] %4\n"
1467 " membar #Sync\n"
1468 " flush %%g6\n"
1469 " nop\n"
1470 " nop\n"
1471 " nop\n"
1472 : /* No outputs */
1473 : "r" (TLB_TAG_ACCESS), "r" (second_alias_page), "r" (pt + 0x400000),
1474 "i" (ASI_DMMU), "i" (ASI_DTLB_DATA_ACCESS), "r" ((0<<16) | (12<<3))
1475 : "memory");
1478 local_irq_restore(flags);
1480 /* Now set kernel pgd to upper alias so physical page computations
1481 * work.
1483 init_mm.pgd += ((shift) / (sizeof(pgd_t)));
1485 memset(swapper_pmd_dir, 0, sizeof(swapper_pmd_dir));
1487 /* Now can init the kernel/bad page tables. */
1488 pud_set(pud_offset(&swapper_pg_dir[0], 0),
1489 swapper_pmd_dir + (shift / sizeof(pgd_t)));
1491 sparc64_vpte_patchme1[0] |=
1492 (((unsigned long)pgd_val(init_mm.pgd[0])) >> 10);
1493 sparc64_vpte_patchme2[0] |=
1494 (((unsigned long)pgd_val(init_mm.pgd[0])) & 0x3ff);
1495 flushi((long)&sparc64_vpte_patchme1[0]);
1497 /* Setup bootmem... */
1498 pages_avail = 0;
1499 last_valid_pfn = end_pfn = bootmem_init(&pages_avail);
1501 /* Inherit non-locked OBP mappings. */
1502 inherit_prom_mappings();
1504 /* Ok, we can use our TLB miss and window trap handlers safely.
1505 * We need to do a quick peek here to see if we are on StarFire
1506 * or not, so setup_tba can setup the IRQ globals correctly (it
1507 * needs to get the hard smp processor id correctly).
1510 extern void setup_tba(int);
1511 setup_tba(this_is_starfire);
1514 inherit_locked_prom_mappings(1);
1516 /* We only created DTLB mapping of this stuff. */
1517 spitfire_flush_dtlb_nucleus_page(alias_base);
1518 if (second_alias_page)
1519 spitfire_flush_dtlb_nucleus_page(second_alias_page);
1521 __flush_tlb_all();
1524 unsigned long zones_size[MAX_NR_ZONES];
1525 unsigned long zholes_size[MAX_NR_ZONES];
1526 unsigned long npages;
1527 int znum;
1529 for (znum = 0; znum < MAX_NR_ZONES; znum++)
1530 zones_size[znum] = zholes_size[znum] = 0;
1532 npages = end_pfn - pfn_base;
1533 zones_size[ZONE_DMA] = npages;
1534 zholes_size[ZONE_DMA] = npages - pages_avail;
1536 free_area_init_node(0, &contig_page_data, zones_size,
1537 phys_base >> PAGE_SHIFT, zholes_size);
1540 device_scan();
1543 /* Ok, it seems that the prom can allocate some more memory chunks
1544 * as a side effect of some prom calls we perform during the
1545 * boot sequence. My most likely theory is that it is from the
1546 * prom_set_traptable() call, and OBP is allocating a scratchpad
1547 * for saving client program register state etc.
1549 static void __init sort_memlist(struct linux_mlist_p1275 *thislist)
1551 int swapi = 0;
1552 int i, mitr;
1553 unsigned long tmpaddr, tmpsize;
1554 unsigned long lowest;
1556 for (i = 0; thislist[i].theres_more != 0; i++) {
1557 lowest = thislist[i].start_adr;
1558 for (mitr = i+1; thislist[mitr-1].theres_more != 0; mitr++)
1559 if (thislist[mitr].start_adr < lowest) {
1560 lowest = thislist[mitr].start_adr;
1561 swapi = mitr;
1563 if (lowest == thislist[i].start_adr)
1564 continue;
1565 tmpaddr = thislist[swapi].start_adr;
1566 tmpsize = thislist[swapi].num_bytes;
1567 for (mitr = swapi; mitr > i; mitr--) {
1568 thislist[mitr].start_adr = thislist[mitr-1].start_adr;
1569 thislist[mitr].num_bytes = thislist[mitr-1].num_bytes;
1571 thislist[i].start_adr = tmpaddr;
1572 thislist[i].num_bytes = tmpsize;
1576 void __init rescan_sp_banks(void)
1578 struct linux_prom64_registers memlist[64];
1579 struct linux_mlist_p1275 avail[64], *mlist;
1580 unsigned long bytes, base_paddr;
1581 int num_regs, node = prom_finddevice("/memory");
1582 int i;
1584 num_regs = prom_getproperty(node, "available",
1585 (char *) memlist, sizeof(memlist));
1586 num_regs = (num_regs / sizeof(struct linux_prom64_registers));
1587 for (i = 0; i < num_regs; i++) {
1588 avail[i].start_adr = memlist[i].phys_addr;
1589 avail[i].num_bytes = memlist[i].reg_size;
1590 avail[i].theres_more = &avail[i + 1];
1592 avail[i - 1].theres_more = NULL;
1593 sort_memlist(avail);
1595 mlist = &avail[0];
1596 i = 0;
1597 bytes = mlist->num_bytes;
1598 base_paddr = mlist->start_adr;
1600 sp_banks[0].base_addr = base_paddr;
1601 sp_banks[0].num_bytes = bytes;
1603 while (mlist->theres_more != NULL){
1604 i++;
1605 mlist = mlist->theres_more;
1606 bytes = mlist->num_bytes;
1607 if (i >= SPARC_PHYS_BANKS-1) {
1608 printk ("The machine has more banks than "
1609 "this kernel can support\n"
1610 "Increase the SPARC_PHYS_BANKS "
1611 "setting (currently %d)\n",
1612 SPARC_PHYS_BANKS);
1613 i = SPARC_PHYS_BANKS-1;
1614 break;
1617 sp_banks[i].base_addr = mlist->start_adr;
1618 sp_banks[i].num_bytes = mlist->num_bytes;
1621 i++;
1622 sp_banks[i].base_addr = 0xdeadbeefbeefdeadUL;
1623 sp_banks[i].num_bytes = 0;
1625 for (i = 0; sp_banks[i].num_bytes != 0; i++)
1626 sp_banks[i].num_bytes &= PAGE_MASK;
1629 static void __init taint_real_pages(void)
1631 struct sparc_phys_banks saved_sp_banks[SPARC_PHYS_BANKS];
1632 int i;
1634 for (i = 0; i < SPARC_PHYS_BANKS; i++) {
1635 saved_sp_banks[i].base_addr =
1636 sp_banks[i].base_addr;
1637 saved_sp_banks[i].num_bytes =
1638 sp_banks[i].num_bytes;
1641 rescan_sp_banks();
1643 /* Find changes discovered in the sp_bank rescan and
1644 * reserve the lost portions in the bootmem maps.
1646 for (i = 0; saved_sp_banks[i].num_bytes; i++) {
1647 unsigned long old_start, old_end;
1649 old_start = saved_sp_banks[i].base_addr;
1650 old_end = old_start +
1651 saved_sp_banks[i].num_bytes;
1652 while (old_start < old_end) {
1653 int n;
1655 for (n = 0; sp_banks[n].num_bytes; n++) {
1656 unsigned long new_start, new_end;
1658 new_start = sp_banks[n].base_addr;
1659 new_end = new_start + sp_banks[n].num_bytes;
1661 if (new_start <= old_start &&
1662 new_end >= (old_start + PAGE_SIZE)) {
1663 set_bit (old_start >> 22,
1664 sparc64_valid_addr_bitmap);
1665 goto do_next_page;
1668 reserve_bootmem(old_start, PAGE_SIZE);
1670 do_next_page:
1671 old_start += PAGE_SIZE;
1676 void __init mem_init(void)
1678 unsigned long codepages, datapages, initpages;
1679 unsigned long addr, last;
1680 int i;
1682 i = last_valid_pfn >> ((22 - PAGE_SHIFT) + 6);
1683 i += 1;
1684 sparc64_valid_addr_bitmap = (unsigned long *)
1685 __alloc_bootmem(i << 3, SMP_CACHE_BYTES, bootmap_base);
1686 if (sparc64_valid_addr_bitmap == NULL) {
1687 prom_printf("mem_init: Cannot alloc valid_addr_bitmap.\n");
1688 prom_halt();
1690 memset(sparc64_valid_addr_bitmap, 0, i << 3);
1692 addr = PAGE_OFFSET + kern_base;
1693 last = PAGE_ALIGN(kern_size) + addr;
1694 while (addr < last) {
1695 set_bit(__pa(addr) >> 22, sparc64_valid_addr_bitmap);
1696 addr += PAGE_SIZE;
1699 taint_real_pages();
1701 max_mapnr = last_valid_pfn - pfn_base;
1702 high_memory = __va(last_valid_pfn << PAGE_SHIFT);
1704 #ifdef CONFIG_DEBUG_BOOTMEM
1705 prom_printf("mem_init: Calling free_all_bootmem().\n");
1706 #endif
1707 totalram_pages = num_physpages = free_all_bootmem() - 1;
1710 * Set up the zero page, mark it reserved, so that page count
1711 * is not manipulated when freeing the page from user ptes.
1713 mem_map_zero = alloc_pages(GFP_KERNEL|__GFP_ZERO, 0);
1714 if (mem_map_zero == NULL) {
1715 prom_printf("paging_init: Cannot alloc zero page.\n");
1716 prom_halt();
1718 SetPageReserved(mem_map_zero);
1720 codepages = (((unsigned long) _etext) - ((unsigned long) _start));
1721 codepages = PAGE_ALIGN(codepages) >> PAGE_SHIFT;
1722 datapages = (((unsigned long) _edata) - ((unsigned long) _etext));
1723 datapages = PAGE_ALIGN(datapages) >> PAGE_SHIFT;
1724 initpages = (((unsigned long) __init_end) - ((unsigned long) __init_begin));
1725 initpages = PAGE_ALIGN(initpages) >> PAGE_SHIFT;
1727 printk("Memory: %uk available (%ldk kernel code, %ldk data, %ldk init) [%016lx,%016lx]\n",
1728 nr_free_pages() << (PAGE_SHIFT-10),
1729 codepages << (PAGE_SHIFT-10),
1730 datapages << (PAGE_SHIFT-10),
1731 initpages << (PAGE_SHIFT-10),
1732 PAGE_OFFSET, (last_valid_pfn << PAGE_SHIFT));
1734 if (tlb_type == cheetah || tlb_type == cheetah_plus)
1735 cheetah_ecache_flush_init();
1738 void free_initmem (void)
1740 unsigned long addr, initend;
1743 * The init section is aligned to 8k in vmlinux.lds. Page align for >8k pagesizes.
1745 addr = PAGE_ALIGN((unsigned long)(__init_begin));
1746 initend = (unsigned long)(__init_end) & PAGE_MASK;
1747 for (; addr < initend; addr += PAGE_SIZE) {
1748 unsigned long page;
1749 struct page *p;
1751 page = (addr +
1752 ((unsigned long) __va(kern_base)) -
1753 ((unsigned long) KERNBASE));
1754 memset((void *)addr, 0xcc, PAGE_SIZE);
1755 p = virt_to_page(page);
1757 ClearPageReserved(p);
1758 set_page_count(p, 1);
1759 __free_page(p);
1760 num_physpages++;
1761 totalram_pages++;
1765 #ifdef CONFIG_BLK_DEV_INITRD
1766 void free_initrd_mem(unsigned long start, unsigned long end)
1768 if (start < end)
1769 printk ("Freeing initrd memory: %ldk freed\n", (end - start) >> 10);
1770 for (; start < end; start += PAGE_SIZE) {
1771 struct page *p = virt_to_page(start);
1773 ClearPageReserved(p);
1774 set_page_count(p, 1);
1775 __free_page(p);
1776 num_physpages++;
1777 totalram_pages++;
1780 #endif