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/module.h>
10 #include <linux/kernel.h>
11 #include <linux/sched.h>
12 #include <linux/string.h>
13 #include <linux/init.h>
14 #include <linux/bootmem.h>
16 #include <linux/hugetlb.h>
17 #include <linux/slab.h>
18 #include <linux/initrd.h>
19 #include <linux/swap.h>
20 #include <linux/pagemap.h>
21 #include <linux/poison.h>
23 #include <linux/seq_file.h>
24 #include <linux/kprobes.h>
25 #include <linux/cache.h>
26 #include <linux/sort.h>
29 #include <asm/system.h>
31 #include <asm/pgalloc.h>
32 #include <asm/pgtable.h>
33 #include <asm/oplib.h>
34 #include <asm/iommu.h>
36 #include <asm/uaccess.h>
37 #include <asm/mmu_context.h>
38 #include <asm/tlbflush.h>
40 #include <asm/starfire.h>
42 #include <asm/spitfire.h>
43 #include <asm/sections.h>
45 #include <asm/hypervisor.h>
48 extern void device_scan(void);
50 #define MAX_PHYS_ADDRESS (1UL << 42UL)
51 #define KPTE_BITMAP_CHUNK_SZ (256UL * 1024UL * 1024UL)
52 #define KPTE_BITMAP_BYTES \
53 ((MAX_PHYS_ADDRESS / KPTE_BITMAP_CHUNK_SZ) / 8)
55 unsigned long kern_linear_pte_xor
[2] __read_mostly
;
57 /* A bitmap, one bit for every 256MB of physical memory. If the bit
58 * is clear, we should use a 4MB page (via kern_linear_pte_xor[0]) else
59 * if set we should use a 256MB page (via kern_linear_pte_xor[1]).
61 unsigned long kpte_linear_bitmap
[KPTE_BITMAP_BYTES
/ sizeof(unsigned long)];
63 /* A special kernel TSB for 4MB and 256MB linear mappings. */
64 struct tsb swapper_4m_tsb
[KERNEL_TSB4M_NENTRIES
];
68 static struct linux_prom64_registers pavail
[MAX_BANKS
] __initdata
;
69 static struct linux_prom64_registers pavail_rescan
[MAX_BANKS
] __initdata
;
70 static int pavail_ents __initdata
;
71 static int pavail_rescan_ents __initdata
;
73 static int cmp_p64(const void *a
, const void *b
)
75 const struct linux_prom64_registers
*x
= a
, *y
= b
;
77 if (x
->phys_addr
> y
->phys_addr
)
79 if (x
->phys_addr
< y
->phys_addr
)
84 static void __init
read_obp_memory(const char *property
,
85 struct linux_prom64_registers
*regs
,
88 int node
= prom_finddevice("/memory");
89 int prop_size
= prom_getproplen(node
, property
);
92 ents
= prop_size
/ sizeof(struct linux_prom64_registers
);
93 if (ents
> MAX_BANKS
) {
94 prom_printf("The machine has more %s property entries than "
95 "this kernel can support (%d).\n",
100 ret
= prom_getproperty(node
, property
, (char *) regs
, prop_size
);
102 prom_printf("Couldn't get %s property from /memory.\n");
106 /* Sanitize what we got from the firmware, by page aligning
109 for (i
= 0; i
< ents
; i
++) {
110 unsigned long base
, size
;
112 base
= regs
[i
].phys_addr
;
113 size
= regs
[i
].reg_size
;
116 if (base
& ~PAGE_MASK
) {
117 unsigned long new_base
= PAGE_ALIGN(base
);
119 size
-= new_base
- base
;
120 if ((long) size
< 0L)
124 regs
[i
].phys_addr
= base
;
125 regs
[i
].reg_size
= size
;
128 for (i
= 0; i
< ents
; i
++) {
129 if (regs
[i
].reg_size
== 0UL) {
132 for (j
= i
; j
< ents
- 1; j
++) {
146 sort(regs
, ents
, sizeof(struct linux_prom64_registers
),
150 unsigned long *sparc64_valid_addr_bitmap __read_mostly
;
152 /* Kernel physical address base and size in bytes. */
153 unsigned long kern_base __read_mostly
;
154 unsigned long kern_size __read_mostly
;
156 /* get_new_mmu_context() uses "cache + 1". */
157 DEFINE_SPINLOCK(ctx_alloc_lock
);
158 unsigned long tlb_context_cache
= CTX_FIRST_VERSION
- 1;
159 #define CTX_BMAP_SLOTS (1UL << (CTX_NR_BITS - 6))
160 unsigned long mmu_context_bmap
[CTX_BMAP_SLOTS
];
162 /* References to special section boundaries */
163 extern char _start
[], _end
[];
165 /* Initial ramdisk setup */
166 extern unsigned long sparc_ramdisk_image64
;
167 extern unsigned int sparc_ramdisk_image
;
168 extern unsigned int sparc_ramdisk_size
;
170 struct page
*mem_map_zero __read_mostly
;
172 unsigned int sparc64_highest_unlocked_tlb_ent __read_mostly
;
174 unsigned long sparc64_kern_pri_context __read_mostly
;
175 unsigned long sparc64_kern_pri_nuc_bits __read_mostly
;
176 unsigned long sparc64_kern_sec_context __read_mostly
;
180 kmem_cache_t
*pgtable_cache __read_mostly
;
182 static void zero_ctor(void *addr
, kmem_cache_t
*cache
, unsigned long flags
)
187 extern void tsb_cache_init(void);
189 void pgtable_cache_init(void)
191 pgtable_cache
= kmem_cache_create("pgtable_cache",
192 PAGE_SIZE
, PAGE_SIZE
,
194 SLAB_MUST_HWCACHE_ALIGN
,
197 if (!pgtable_cache
) {
198 prom_printf("Could not create pgtable_cache\n");
204 #ifdef CONFIG_DEBUG_DCFLUSH
205 atomic_t dcpage_flushes
= ATOMIC_INIT(0);
207 atomic_t dcpage_flushes_xcall
= ATOMIC_INIT(0);
211 inline void flush_dcache_page_impl(struct page
*page
)
213 BUG_ON(tlb_type
== hypervisor
);
214 #ifdef CONFIG_DEBUG_DCFLUSH
215 atomic_inc(&dcpage_flushes
);
218 #ifdef DCACHE_ALIASING_POSSIBLE
219 __flush_dcache_page(page_address(page
),
220 ((tlb_type
== spitfire
) &&
221 page_mapping(page
) != NULL
));
223 if (page_mapping(page
) != NULL
&&
224 tlb_type
== spitfire
)
225 __flush_icache_page(__pa(page_address(page
)));
229 #define PG_dcache_dirty PG_arch_1
230 #define PG_dcache_cpu_shift 24UL
231 #define PG_dcache_cpu_mask (256UL - 1UL)
234 #error D-cache dirty tracking and thread_info->cpu need fixing for > 256 cpus
237 #define dcache_dirty_cpu(page) \
238 (((page)->flags >> PG_dcache_cpu_shift) & PG_dcache_cpu_mask)
240 static __inline__
void set_dcache_dirty(struct page
*page
, int this_cpu
)
242 unsigned long mask
= this_cpu
;
243 unsigned long non_cpu_bits
;
245 non_cpu_bits
= ~(PG_dcache_cpu_mask
<< PG_dcache_cpu_shift
);
246 mask
= (mask
<< PG_dcache_cpu_shift
) | (1UL << PG_dcache_dirty
);
248 __asm__
__volatile__("1:\n\t"
250 "and %%g7, %1, %%g1\n\t"
251 "or %%g1, %0, %%g1\n\t"
252 "casx [%2], %%g7, %%g1\n\t"
254 "membar #StoreLoad | #StoreStore\n\t"
255 "bne,pn %%xcc, 1b\n\t"
258 : "r" (mask
), "r" (non_cpu_bits
), "r" (&page
->flags
)
262 static __inline__
void clear_dcache_dirty_cpu(struct page
*page
, unsigned long cpu
)
264 unsigned long mask
= (1UL << PG_dcache_dirty
);
266 __asm__
__volatile__("! test_and_clear_dcache_dirty\n"
269 "srlx %%g7, %4, %%g1\n\t"
270 "and %%g1, %3, %%g1\n\t"
272 "bne,pn %%icc, 2f\n\t"
273 " andn %%g7, %1, %%g1\n\t"
274 "casx [%2], %%g7, %%g1\n\t"
276 "membar #StoreLoad | #StoreStore\n\t"
277 "bne,pn %%xcc, 1b\n\t"
281 : "r" (cpu
), "r" (mask
), "r" (&page
->flags
),
282 "i" (PG_dcache_cpu_mask
),
283 "i" (PG_dcache_cpu_shift
)
287 static inline void tsb_insert(struct tsb
*ent
, unsigned long tag
, unsigned long pte
)
289 unsigned long tsb_addr
= (unsigned long) ent
;
291 if (tlb_type
== cheetah_plus
|| tlb_type
== hypervisor
)
292 tsb_addr
= __pa(tsb_addr
);
294 __tsb_insert(tsb_addr
, tag
, pte
);
297 unsigned long _PAGE_ALL_SZ_BITS __read_mostly
;
298 unsigned long _PAGE_SZBITS __read_mostly
;
300 void update_mmu_cache(struct vm_area_struct
*vma
, unsigned long address
, pte_t pte
)
302 struct mm_struct
*mm
;
304 unsigned long tag
, flags
;
305 unsigned long tsb_index
, tsb_hash_shift
;
307 if (tlb_type
!= hypervisor
) {
308 unsigned long pfn
= pte_pfn(pte
);
309 unsigned long pg_flags
;
312 if (pfn_valid(pfn
) &&
313 (page
= pfn_to_page(pfn
), page_mapping(page
)) &&
314 ((pg_flags
= page
->flags
) & (1UL << PG_dcache_dirty
))) {
315 int cpu
= ((pg_flags
>> PG_dcache_cpu_shift
) &
317 int this_cpu
= get_cpu();
319 /* This is just to optimize away some function calls
323 flush_dcache_page_impl(page
);
325 smp_flush_dcache_page_impl(page
, cpu
);
327 clear_dcache_dirty_cpu(page
, cpu
);
335 tsb_index
= MM_TSB_BASE
;
336 tsb_hash_shift
= PAGE_SHIFT
;
338 spin_lock_irqsave(&mm
->context
.lock
, flags
);
340 #ifdef CONFIG_HUGETLB_PAGE
341 if (mm
->context
.tsb_block
[MM_TSB_HUGE
].tsb
!= NULL
) {
342 if ((tlb_type
== hypervisor
&&
343 (pte_val(pte
) & _PAGE_SZALL_4V
) == _PAGE_SZHUGE_4V
) ||
344 (tlb_type
!= hypervisor
&&
345 (pte_val(pte
) & _PAGE_SZALL_4U
) == _PAGE_SZHUGE_4U
)) {
346 tsb_index
= MM_TSB_HUGE
;
347 tsb_hash_shift
= HPAGE_SHIFT
;
352 tsb
= mm
->context
.tsb_block
[tsb_index
].tsb
;
353 tsb
+= ((address
>> tsb_hash_shift
) &
354 (mm
->context
.tsb_block
[tsb_index
].tsb_nentries
- 1UL));
355 tag
= (address
>> 22UL);
356 tsb_insert(tsb
, tag
, pte_val(pte
));
358 spin_unlock_irqrestore(&mm
->context
.lock
, flags
);
361 void flush_dcache_page(struct page
*page
)
363 struct address_space
*mapping
;
366 if (tlb_type
== hypervisor
)
369 /* Do not bother with the expensive D-cache flush if it
370 * is merely the zero page. The 'bigcore' testcase in GDB
371 * causes this case to run millions of times.
373 if (page
== ZERO_PAGE(0))
376 this_cpu
= get_cpu();
378 mapping
= page_mapping(page
);
379 if (mapping
&& !mapping_mapped(mapping
)) {
380 int dirty
= test_bit(PG_dcache_dirty
, &page
->flags
);
382 int dirty_cpu
= dcache_dirty_cpu(page
);
384 if (dirty_cpu
== this_cpu
)
386 smp_flush_dcache_page_impl(page
, dirty_cpu
);
388 set_dcache_dirty(page
, this_cpu
);
390 /* We could delay the flush for the !page_mapping
391 * case too. But that case is for exec env/arg
392 * pages and those are %99 certainly going to get
393 * faulted into the tlb (and thus flushed) anyways.
395 flush_dcache_page_impl(page
);
402 void __kprobes
flush_icache_range(unsigned long start
, unsigned long end
)
404 /* Cheetah and Hypervisor platform cpus have coherent I-cache. */
405 if (tlb_type
== spitfire
) {
408 for (kaddr
= start
; kaddr
< end
; kaddr
+= PAGE_SIZE
)
409 __flush_icache_page(__get_phys(kaddr
));
415 printk("Mem-info:\n");
417 printk("Free swap: %6ldkB\n",
418 nr_swap_pages
<< (PAGE_SHIFT
-10));
419 printk("%ld pages of RAM\n", num_physpages
);
420 printk("%d free pages\n", nr_free_pages());
423 void mmu_info(struct seq_file
*m
)
425 if (tlb_type
== cheetah
)
426 seq_printf(m
, "MMU Type\t: Cheetah\n");
427 else if (tlb_type
== cheetah_plus
)
428 seq_printf(m
, "MMU Type\t: Cheetah+\n");
429 else if (tlb_type
== spitfire
)
430 seq_printf(m
, "MMU Type\t: Spitfire\n");
431 else if (tlb_type
== hypervisor
)
432 seq_printf(m
, "MMU Type\t: Hypervisor (sun4v)\n");
434 seq_printf(m
, "MMU Type\t: ???\n");
436 #ifdef CONFIG_DEBUG_DCFLUSH
437 seq_printf(m
, "DCPageFlushes\t: %d\n",
438 atomic_read(&dcpage_flushes
));
440 seq_printf(m
, "DCPageFlushesXC\t: %d\n",
441 atomic_read(&dcpage_flushes_xcall
));
442 #endif /* CONFIG_SMP */
443 #endif /* CONFIG_DEBUG_DCFLUSH */
446 struct linux_prom_translation
{
452 /* Exported for kernel TLB miss handling in ktlb.S */
453 struct linux_prom_translation prom_trans
[512] __read_mostly
;
454 unsigned int prom_trans_ents __read_mostly
;
456 /* Exported for SMP bootup purposes. */
457 unsigned long kern_locked_tte_data
;
459 /* The obp translations are saved based on 8k pagesize, since obp can
460 * use a mixture of pagesizes. Misses to the LOW_OBP_ADDRESS ->
461 * HI_OBP_ADDRESS range are handled in ktlb.S.
463 static inline int in_obp_range(unsigned long vaddr
)
465 return (vaddr
>= LOW_OBP_ADDRESS
&&
466 vaddr
< HI_OBP_ADDRESS
);
469 static int cmp_ptrans(const void *a
, const void *b
)
471 const struct linux_prom_translation
*x
= a
, *y
= b
;
473 if (x
->virt
> y
->virt
)
475 if (x
->virt
< y
->virt
)
480 /* Read OBP translations property into 'prom_trans[]'. */
481 static void __init
read_obp_translations(void)
483 int n
, node
, ents
, first
, last
, i
;
485 node
= prom_finddevice("/virtual-memory");
486 n
= prom_getproplen(node
, "translations");
487 if (unlikely(n
== 0 || n
== -1)) {
488 prom_printf("prom_mappings: Couldn't get size.\n");
491 if (unlikely(n
> sizeof(prom_trans
))) {
492 prom_printf("prom_mappings: Size %Zd is too big.\n", n
);
496 if ((n
= prom_getproperty(node
, "translations",
497 (char *)&prom_trans
[0],
498 sizeof(prom_trans
))) == -1) {
499 prom_printf("prom_mappings: Couldn't get property.\n");
503 n
= n
/ sizeof(struct linux_prom_translation
);
507 sort(prom_trans
, ents
, sizeof(struct linux_prom_translation
),
510 /* Now kick out all the non-OBP entries. */
511 for (i
= 0; i
< ents
; i
++) {
512 if (in_obp_range(prom_trans
[i
].virt
))
516 for (; i
< ents
; i
++) {
517 if (!in_obp_range(prom_trans
[i
].virt
))
522 for (i
= 0; i
< (last
- first
); i
++) {
523 struct linux_prom_translation
*src
= &prom_trans
[i
+ first
];
524 struct linux_prom_translation
*dest
= &prom_trans
[i
];
528 for (; i
< ents
; i
++) {
529 struct linux_prom_translation
*dest
= &prom_trans
[i
];
530 dest
->virt
= dest
->size
= dest
->data
= 0x0UL
;
533 prom_trans_ents
= last
- first
;
535 if (tlb_type
== spitfire
) {
536 /* Clear diag TTE bits. */
537 for (i
= 0; i
< prom_trans_ents
; i
++)
538 prom_trans
[i
].data
&= ~0x0003fe0000000000UL
;
542 static void __init
hypervisor_tlb_lock(unsigned long vaddr
,
546 register unsigned long func
asm("%o5");
547 register unsigned long arg0
asm("%o0");
548 register unsigned long arg1
asm("%o1");
549 register unsigned long arg2
asm("%o2");
550 register unsigned long arg3
asm("%o3");
552 func
= HV_FAST_MMU_MAP_PERM_ADDR
;
557 __asm__
__volatile__("ta 0x80"
558 : "=&r" (func
), "=&r" (arg0
),
559 "=&r" (arg1
), "=&r" (arg2
),
561 : "0" (func
), "1" (arg0
), "2" (arg1
),
562 "3" (arg2
), "4" (arg3
));
564 prom_printf("hypervisor_tlb_lock[%lx:%lx:%lx:%lx]: "
565 "errors with %lx\n", vaddr
, 0, pte
, mmu
, arg0
);
570 static unsigned long kern_large_tte(unsigned long paddr
);
572 static void __init
remap_kernel(void)
574 unsigned long phys_page
, tte_vaddr
, tte_data
;
575 int tlb_ent
= sparc64_highest_locked_tlbent();
577 tte_vaddr
= (unsigned long) KERNBASE
;
578 phys_page
= (prom_boot_mapping_phys_low
>> 22UL) << 22UL;
579 tte_data
= kern_large_tte(phys_page
);
581 kern_locked_tte_data
= tte_data
;
583 /* Now lock us into the TLBs via Hypervisor or OBP. */
584 if (tlb_type
== hypervisor
) {
585 hypervisor_tlb_lock(tte_vaddr
, tte_data
, HV_MMU_DMMU
);
586 hypervisor_tlb_lock(tte_vaddr
, tte_data
, HV_MMU_IMMU
);
588 tte_vaddr
+= 0x400000;
589 tte_data
+= 0x400000;
590 hypervisor_tlb_lock(tte_vaddr
, tte_data
, HV_MMU_DMMU
);
591 hypervisor_tlb_lock(tte_vaddr
, tte_data
, HV_MMU_IMMU
);
594 prom_dtlb_load(tlb_ent
, tte_data
, tte_vaddr
);
595 prom_itlb_load(tlb_ent
, tte_data
, tte_vaddr
);
598 prom_dtlb_load(tlb_ent
,
600 tte_vaddr
+ 0x400000);
601 prom_itlb_load(tlb_ent
,
603 tte_vaddr
+ 0x400000);
605 sparc64_highest_unlocked_tlb_ent
= tlb_ent
- 1;
607 if (tlb_type
== cheetah_plus
) {
608 sparc64_kern_pri_context
= (CTX_CHEETAH_PLUS_CTX0
|
609 CTX_CHEETAH_PLUS_NUC
);
610 sparc64_kern_pri_nuc_bits
= CTX_CHEETAH_PLUS_NUC
;
611 sparc64_kern_sec_context
= CTX_CHEETAH_PLUS_CTX0
;
616 static void __init
inherit_prom_mappings(void)
618 read_obp_translations();
620 /* Now fixup OBP's idea about where we really are mapped. */
621 prom_printf("Remapping the kernel... ");
623 prom_printf("done.\n");
626 void prom_world(int enter
)
629 set_fs((mm_segment_t
) { get_thread_current_ds() });
631 __asm__
__volatile__("flushw");
634 #ifdef DCACHE_ALIASING_POSSIBLE
635 void __flush_dcache_range(unsigned long start
, unsigned long end
)
639 if (tlb_type
== spitfire
) {
642 for (va
= start
; va
< end
; va
+= 32) {
643 spitfire_put_dcache_tag(va
& 0x3fe0, 0x0);
647 } else if (tlb_type
== cheetah
|| tlb_type
== cheetah_plus
) {
650 for (va
= start
; va
< end
; va
+= 32)
651 __asm__
__volatile__("stxa %%g0, [%0] %1\n\t"
655 "i" (ASI_DCACHE_INVALIDATE
));
658 #endif /* DCACHE_ALIASING_POSSIBLE */
660 /* Caller does TLB context flushing on local CPU if necessary.
661 * The caller also ensures that CTX_VALID(mm->context) is false.
663 * We must be careful about boundary cases so that we never
664 * let the user have CTX 0 (nucleus) or we ever use a CTX
665 * version of zero (and thus NO_CONTEXT would not be caught
666 * by version mis-match tests in mmu_context.h).
668 * Always invoked with interrupts disabled.
670 void get_new_mmu_context(struct mm_struct
*mm
)
672 unsigned long ctx
, new_ctx
;
673 unsigned long orig_pgsz_bits
;
677 spin_lock_irqsave(&ctx_alloc_lock
, flags
);
678 orig_pgsz_bits
= (mm
->context
.sparc64_ctx_val
& CTX_PGSZ_MASK
);
679 ctx
= (tlb_context_cache
+ 1) & CTX_NR_MASK
;
680 new_ctx
= find_next_zero_bit(mmu_context_bmap
, 1 << CTX_NR_BITS
, ctx
);
682 if (new_ctx
>= (1 << CTX_NR_BITS
)) {
683 new_ctx
= find_next_zero_bit(mmu_context_bmap
, ctx
, 1);
684 if (new_ctx
>= ctx
) {
686 new_ctx
= (tlb_context_cache
& CTX_VERSION_MASK
) +
689 new_ctx
= CTX_FIRST_VERSION
;
691 /* Don't call memset, for 16 entries that's just
694 mmu_context_bmap
[0] = 3;
695 mmu_context_bmap
[1] = 0;
696 mmu_context_bmap
[2] = 0;
697 mmu_context_bmap
[3] = 0;
698 for (i
= 4; i
< CTX_BMAP_SLOTS
; i
+= 4) {
699 mmu_context_bmap
[i
+ 0] = 0;
700 mmu_context_bmap
[i
+ 1] = 0;
701 mmu_context_bmap
[i
+ 2] = 0;
702 mmu_context_bmap
[i
+ 3] = 0;
708 mmu_context_bmap
[new_ctx
>>6] |= (1UL << (new_ctx
& 63));
709 new_ctx
|= (tlb_context_cache
& CTX_VERSION_MASK
);
711 tlb_context_cache
= new_ctx
;
712 mm
->context
.sparc64_ctx_val
= new_ctx
| orig_pgsz_bits
;
713 spin_unlock_irqrestore(&ctx_alloc_lock
, flags
);
715 if (unlikely(new_version
))
716 smp_new_mmu_context_version();
719 void sparc_ultra_dump_itlb(void)
723 if (tlb_type
== spitfire
) {
724 printk ("Contents of itlb: ");
725 for (slot
= 0; slot
< 14; slot
++) printk (" ");
726 printk ("%2x:%016lx,%016lx\n",
728 spitfire_get_itlb_tag(0), spitfire_get_itlb_data(0));
729 for (slot
= 1; slot
< 64; slot
+=3) {
730 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx %2x:%016lx,%016lx\n",
732 spitfire_get_itlb_tag(slot
), spitfire_get_itlb_data(slot
),
734 spitfire_get_itlb_tag(slot
+1), spitfire_get_itlb_data(slot
+1),
736 spitfire_get_itlb_tag(slot
+2), spitfire_get_itlb_data(slot
+2));
738 } else if (tlb_type
== cheetah
|| tlb_type
== cheetah_plus
) {
739 printk ("Contents of itlb0:\n");
740 for (slot
= 0; slot
< 16; slot
+=2) {
741 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
743 cheetah_get_litlb_tag(slot
), cheetah_get_litlb_data(slot
),
745 cheetah_get_litlb_tag(slot
+1), cheetah_get_litlb_data(slot
+1));
747 printk ("Contents of itlb2:\n");
748 for (slot
= 0; slot
< 128; slot
+=2) {
749 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
751 cheetah_get_itlb_tag(slot
), cheetah_get_itlb_data(slot
),
753 cheetah_get_itlb_tag(slot
+1), cheetah_get_itlb_data(slot
+1));
758 void sparc_ultra_dump_dtlb(void)
762 if (tlb_type
== spitfire
) {
763 printk ("Contents of dtlb: ");
764 for (slot
= 0; slot
< 14; slot
++) printk (" ");
765 printk ("%2x:%016lx,%016lx\n", 0,
766 spitfire_get_dtlb_tag(0), spitfire_get_dtlb_data(0));
767 for (slot
= 1; slot
< 64; slot
+=3) {
768 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx %2x:%016lx,%016lx\n",
770 spitfire_get_dtlb_tag(slot
), spitfire_get_dtlb_data(slot
),
772 spitfire_get_dtlb_tag(slot
+1), spitfire_get_dtlb_data(slot
+1),
774 spitfire_get_dtlb_tag(slot
+2), spitfire_get_dtlb_data(slot
+2));
776 } else if (tlb_type
== cheetah
|| tlb_type
== cheetah_plus
) {
777 printk ("Contents of dtlb0:\n");
778 for (slot
= 0; slot
< 16; slot
+=2) {
779 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
781 cheetah_get_ldtlb_tag(slot
), cheetah_get_ldtlb_data(slot
),
783 cheetah_get_ldtlb_tag(slot
+1), cheetah_get_ldtlb_data(slot
+1));
785 printk ("Contents of dtlb2:\n");
786 for (slot
= 0; slot
< 512; slot
+=2) {
787 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
789 cheetah_get_dtlb_tag(slot
, 2), cheetah_get_dtlb_data(slot
, 2),
791 cheetah_get_dtlb_tag(slot
+1, 2), cheetah_get_dtlb_data(slot
+1, 2));
793 if (tlb_type
== cheetah_plus
) {
794 printk ("Contents of dtlb3:\n");
795 for (slot
= 0; slot
< 512; slot
+=2) {
796 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
798 cheetah_get_dtlb_tag(slot
, 3), cheetah_get_dtlb_data(slot
, 3),
800 cheetah_get_dtlb_tag(slot
+1, 3), cheetah_get_dtlb_data(slot
+1, 3));
806 extern unsigned long cmdline_memory_size
;
808 /* Find a free area for the bootmem map, avoiding the kernel image
809 * and the initial ramdisk.
811 static unsigned long __init
choose_bootmap_pfn(unsigned long start_pfn
,
812 unsigned long end_pfn
)
814 unsigned long avoid_start
, avoid_end
, bootmap_size
;
817 bootmap_size
= ((end_pfn
- start_pfn
) + 7) / 8;
818 bootmap_size
= ALIGN(bootmap_size
, sizeof(long));
820 avoid_start
= avoid_end
= 0;
821 #ifdef CONFIG_BLK_DEV_INITRD
822 avoid_start
= initrd_start
;
823 avoid_end
= PAGE_ALIGN(initrd_end
);
826 #ifdef CONFIG_DEBUG_BOOTMEM
827 prom_printf("choose_bootmap_pfn: kern[%lx:%lx] avoid[%lx:%lx]\n",
828 kern_base
, PAGE_ALIGN(kern_base
+ kern_size
),
829 avoid_start
, avoid_end
);
831 for (i
= 0; i
< pavail_ents
; i
++) {
832 unsigned long start
, end
;
834 start
= pavail
[i
].phys_addr
;
835 end
= start
+ pavail
[i
].reg_size
;
837 while (start
< end
) {
838 if (start
>= kern_base
&&
839 start
< PAGE_ALIGN(kern_base
+ kern_size
)) {
840 start
= PAGE_ALIGN(kern_base
+ kern_size
);
843 if (start
>= avoid_start
&& start
< avoid_end
) {
848 if ((end
- start
) < bootmap_size
)
851 if (start
< kern_base
&&
852 (start
+ bootmap_size
) > kern_base
) {
853 start
= PAGE_ALIGN(kern_base
+ kern_size
);
857 if (start
< avoid_start
&&
858 (start
+ bootmap_size
) > avoid_start
) {
863 /* OK, it doesn't overlap anything, use it. */
864 #ifdef CONFIG_DEBUG_BOOTMEM
865 prom_printf("choose_bootmap_pfn: Using %lx [%lx]\n",
866 start
>> PAGE_SHIFT
, start
);
868 return start
>> PAGE_SHIFT
;
872 prom_printf("Cannot find free area for bootmap, aborting.\n");
876 static unsigned long __init
bootmem_init(unsigned long *pages_avail
,
877 unsigned long phys_base
)
879 unsigned long bootmap_size
, end_pfn
;
880 unsigned long end_of_phys_memory
= 0UL;
881 unsigned long bootmap_pfn
, bytes_avail
, size
;
884 #ifdef CONFIG_DEBUG_BOOTMEM
885 prom_printf("bootmem_init: Scan pavail, ");
889 for (i
= 0; i
< pavail_ents
; i
++) {
890 end_of_phys_memory
= pavail
[i
].phys_addr
+
892 bytes_avail
+= pavail
[i
].reg_size
;
893 if (cmdline_memory_size
) {
894 if (bytes_avail
> cmdline_memory_size
) {
895 unsigned long slack
= bytes_avail
- cmdline_memory_size
;
897 bytes_avail
-= slack
;
898 end_of_phys_memory
-= slack
;
900 pavail
[i
].reg_size
-= slack
;
901 if ((long)pavail
[i
].reg_size
<= 0L) {
902 pavail
[i
].phys_addr
= 0xdeadbeefUL
;
903 pavail
[i
].reg_size
= 0UL;
906 pavail
[i
+1].reg_size
= 0Ul;
907 pavail
[i
+1].phys_addr
= 0xdeadbeefUL
;
915 *pages_avail
= bytes_avail
>> PAGE_SHIFT
;
917 end_pfn
= end_of_phys_memory
>> PAGE_SHIFT
;
919 #ifdef CONFIG_BLK_DEV_INITRD
920 /* Now have to check initial ramdisk, so that bootmap does not overwrite it */
921 if (sparc_ramdisk_image
|| sparc_ramdisk_image64
) {
922 unsigned long ramdisk_image
= sparc_ramdisk_image
?
923 sparc_ramdisk_image
: sparc_ramdisk_image64
;
924 if (ramdisk_image
>= (unsigned long)_end
- 2 * PAGE_SIZE
)
925 ramdisk_image
-= KERNBASE
;
926 initrd_start
= ramdisk_image
+ phys_base
;
927 initrd_end
= initrd_start
+ sparc_ramdisk_size
;
928 if (initrd_end
> end_of_phys_memory
) {
929 printk(KERN_CRIT
"initrd extends beyond end of memory "
930 "(0x%016lx > 0x%016lx)\ndisabling initrd\n",
931 initrd_end
, end_of_phys_memory
);
937 /* Initialize the boot-time allocator. */
938 max_pfn
= max_low_pfn
= end_pfn
;
939 min_low_pfn
= (phys_base
>> PAGE_SHIFT
);
941 bootmap_pfn
= choose_bootmap_pfn(min_low_pfn
, end_pfn
);
943 #ifdef CONFIG_DEBUG_BOOTMEM
944 prom_printf("init_bootmem(min[%lx], bootmap[%lx], max[%lx])\n",
945 min_low_pfn
, bootmap_pfn
, max_low_pfn
);
947 bootmap_size
= init_bootmem_node(NODE_DATA(0), bootmap_pfn
,
948 min_low_pfn
, end_pfn
);
950 /* Now register the available physical memory with the
953 for (i
= 0; i
< pavail_ents
; i
++) {
954 #ifdef CONFIG_DEBUG_BOOTMEM
955 prom_printf("free_bootmem(pavail:%d): base[%lx] size[%lx]\n",
956 i
, pavail
[i
].phys_addr
, pavail
[i
].reg_size
);
958 free_bootmem(pavail
[i
].phys_addr
, pavail
[i
].reg_size
);
961 #ifdef CONFIG_BLK_DEV_INITRD
963 size
= initrd_end
- initrd_start
;
965 /* Resert the initrd image area. */
966 #ifdef CONFIG_DEBUG_BOOTMEM
967 prom_printf("reserve_bootmem(initrd): base[%llx] size[%lx]\n",
968 initrd_start
, initrd_end
);
970 reserve_bootmem(initrd_start
, size
);
971 *pages_avail
-= PAGE_ALIGN(size
) >> PAGE_SHIFT
;
973 initrd_start
+= PAGE_OFFSET
;
974 initrd_end
+= PAGE_OFFSET
;
977 /* Reserve the kernel text/data/bss. */
978 #ifdef CONFIG_DEBUG_BOOTMEM
979 prom_printf("reserve_bootmem(kernel): base[%lx] size[%lx]\n", kern_base
, kern_size
);
981 reserve_bootmem(kern_base
, kern_size
);
982 *pages_avail
-= PAGE_ALIGN(kern_size
) >> PAGE_SHIFT
;
984 /* Reserve the bootmem map. We do not account for it
985 * in pages_avail because we will release that memory
986 * in free_all_bootmem.
989 #ifdef CONFIG_DEBUG_BOOTMEM
990 prom_printf("reserve_bootmem(bootmap): base[%lx] size[%lx]\n",
991 (bootmap_pfn
<< PAGE_SHIFT
), size
);
993 reserve_bootmem((bootmap_pfn
<< PAGE_SHIFT
), size
);
994 *pages_avail
-= PAGE_ALIGN(size
) >> PAGE_SHIFT
;
996 for (i
= 0; i
< pavail_ents
; i
++) {
997 unsigned long start_pfn
, end_pfn
;
999 start_pfn
= pavail
[i
].phys_addr
>> PAGE_SHIFT
;
1000 end_pfn
= (start_pfn
+ (pavail
[i
].reg_size
>> PAGE_SHIFT
));
1001 #ifdef CONFIG_DEBUG_BOOTMEM
1002 prom_printf("memory_present(0, %lx, %lx)\n",
1003 start_pfn
, end_pfn
);
1005 memory_present(0, start_pfn
, end_pfn
);
1013 static struct linux_prom64_registers pall
[MAX_BANKS
] __initdata
;
1014 static int pall_ents __initdata
;
1016 #ifdef CONFIG_DEBUG_PAGEALLOC
1017 static unsigned long kernel_map_range(unsigned long pstart
, unsigned long pend
, pgprot_t prot
)
1019 unsigned long vstart
= PAGE_OFFSET
+ pstart
;
1020 unsigned long vend
= PAGE_OFFSET
+ pend
;
1021 unsigned long alloc_bytes
= 0UL;
1023 if ((vstart
& ~PAGE_MASK
) || (vend
& ~PAGE_MASK
)) {
1024 prom_printf("kernel_map: Unaligned physmem[%lx:%lx]\n",
1029 while (vstart
< vend
) {
1030 unsigned long this_end
, paddr
= __pa(vstart
);
1031 pgd_t
*pgd
= pgd_offset_k(vstart
);
1036 pud
= pud_offset(pgd
, vstart
);
1037 if (pud_none(*pud
)) {
1040 new = __alloc_bootmem(PAGE_SIZE
, PAGE_SIZE
, PAGE_SIZE
);
1041 alloc_bytes
+= PAGE_SIZE
;
1042 pud_populate(&init_mm
, pud
, new);
1045 pmd
= pmd_offset(pud
, vstart
);
1046 if (!pmd_present(*pmd
)) {
1049 new = __alloc_bootmem(PAGE_SIZE
, PAGE_SIZE
, PAGE_SIZE
);
1050 alloc_bytes
+= PAGE_SIZE
;
1051 pmd_populate_kernel(&init_mm
, pmd
, new);
1054 pte
= pte_offset_kernel(pmd
, vstart
);
1055 this_end
= (vstart
+ PMD_SIZE
) & PMD_MASK
;
1056 if (this_end
> vend
)
1059 while (vstart
< this_end
) {
1060 pte_val(*pte
) = (paddr
| pgprot_val(prot
));
1062 vstart
+= PAGE_SIZE
;
1071 extern unsigned int kvmap_linear_patch
[1];
1072 #endif /* CONFIG_DEBUG_PAGEALLOC */
1074 static void __init
mark_kpte_bitmap(unsigned long start
, unsigned long end
)
1076 const unsigned long shift_256MB
= 28;
1077 const unsigned long mask_256MB
= ((1UL << shift_256MB
) - 1UL);
1078 const unsigned long size_256MB
= (1UL << shift_256MB
);
1080 while (start
< end
) {
1083 remains
= end
- start
;
1084 if (remains
< size_256MB
)
1087 if (start
& mask_256MB
) {
1088 start
= (start
+ size_256MB
) & ~mask_256MB
;
1092 while (remains
>= size_256MB
) {
1093 unsigned long index
= start
>> shift_256MB
;
1095 __set_bit(index
, kpte_linear_bitmap
);
1097 start
+= size_256MB
;
1098 remains
-= size_256MB
;
1103 static void __init
kernel_physical_mapping_init(void)
1106 #ifdef CONFIG_DEBUG_PAGEALLOC
1107 unsigned long mem_alloced
= 0UL;
1110 read_obp_memory("reg", &pall
[0], &pall_ents
);
1112 for (i
= 0; i
< pall_ents
; i
++) {
1113 unsigned long phys_start
, phys_end
;
1115 phys_start
= pall
[i
].phys_addr
;
1116 phys_end
= phys_start
+ pall
[i
].reg_size
;
1118 mark_kpte_bitmap(phys_start
, phys_end
);
1120 #ifdef CONFIG_DEBUG_PAGEALLOC
1121 mem_alloced
+= kernel_map_range(phys_start
, phys_end
,
1126 #ifdef CONFIG_DEBUG_PAGEALLOC
1127 printk("Allocated %ld bytes for kernel page tables.\n",
1130 kvmap_linear_patch
[0] = 0x01000000; /* nop */
1131 flushi(&kvmap_linear_patch
[0]);
1137 #ifdef CONFIG_DEBUG_PAGEALLOC
1138 void kernel_map_pages(struct page
*page
, int numpages
, int enable
)
1140 unsigned long phys_start
= page_to_pfn(page
) << PAGE_SHIFT
;
1141 unsigned long phys_end
= phys_start
+ (numpages
* PAGE_SIZE
);
1143 kernel_map_range(phys_start
, phys_end
,
1144 (enable
? PAGE_KERNEL
: __pgprot(0)));
1146 flush_tsb_kernel_range(PAGE_OFFSET
+ phys_start
,
1147 PAGE_OFFSET
+ phys_end
);
1149 /* we should perform an IPI and flush all tlbs,
1150 * but that can deadlock->flush only current cpu.
1152 __flush_tlb_kernel_range(PAGE_OFFSET
+ phys_start
,
1153 PAGE_OFFSET
+ phys_end
);
1157 unsigned long __init
find_ecache_flush_span(unsigned long size
)
1161 for (i
= 0; i
< pavail_ents
; i
++) {
1162 if (pavail
[i
].reg_size
>= size
)
1163 return pavail
[i
].phys_addr
;
1169 static void __init
tsb_phys_patch(void)
1171 struct tsb_ldquad_phys_patch_entry
*pquad
;
1172 struct tsb_phys_patch_entry
*p
;
1174 pquad
= &__tsb_ldquad_phys_patch
;
1175 while (pquad
< &__tsb_ldquad_phys_patch_end
) {
1176 unsigned long addr
= pquad
->addr
;
1178 if (tlb_type
== hypervisor
)
1179 *(unsigned int *) addr
= pquad
->sun4v_insn
;
1181 *(unsigned int *) addr
= pquad
->sun4u_insn
;
1183 __asm__
__volatile__("flush %0"
1190 p
= &__tsb_phys_patch
;
1191 while (p
< &__tsb_phys_patch_end
) {
1192 unsigned long addr
= p
->addr
;
1194 *(unsigned int *) addr
= p
->insn
;
1196 __asm__
__volatile__("flush %0"
1204 /* Don't mark as init, we give this to the Hypervisor. */
1205 static struct hv_tsb_descr ktsb_descr
[2];
1206 extern struct tsb swapper_tsb
[KERNEL_TSB_NENTRIES
];
1208 static void __init
sun4v_ktsb_init(void)
1210 unsigned long ktsb_pa
;
1212 /* First KTSB for PAGE_SIZE mappings. */
1213 ktsb_pa
= kern_base
+ ((unsigned long)&swapper_tsb
[0] - KERNBASE
);
1215 switch (PAGE_SIZE
) {
1218 ktsb_descr
[0].pgsz_idx
= HV_PGSZ_IDX_8K
;
1219 ktsb_descr
[0].pgsz_mask
= HV_PGSZ_MASK_8K
;
1223 ktsb_descr
[0].pgsz_idx
= HV_PGSZ_IDX_64K
;
1224 ktsb_descr
[0].pgsz_mask
= HV_PGSZ_MASK_64K
;
1228 ktsb_descr
[0].pgsz_idx
= HV_PGSZ_IDX_512K
;
1229 ktsb_descr
[0].pgsz_mask
= HV_PGSZ_MASK_512K
;
1232 case 4 * 1024 * 1024:
1233 ktsb_descr
[0].pgsz_idx
= HV_PGSZ_IDX_4MB
;
1234 ktsb_descr
[0].pgsz_mask
= HV_PGSZ_MASK_4MB
;
1238 ktsb_descr
[0].assoc
= 1;
1239 ktsb_descr
[0].num_ttes
= KERNEL_TSB_NENTRIES
;
1240 ktsb_descr
[0].ctx_idx
= 0;
1241 ktsb_descr
[0].tsb_base
= ktsb_pa
;
1242 ktsb_descr
[0].resv
= 0;
1244 /* Second KTSB for 4MB/256MB mappings. */
1245 ktsb_pa
= (kern_base
+
1246 ((unsigned long)&swapper_4m_tsb
[0] - KERNBASE
));
1248 ktsb_descr
[1].pgsz_idx
= HV_PGSZ_IDX_4MB
;
1249 ktsb_descr
[1].pgsz_mask
= (HV_PGSZ_MASK_4MB
|
1250 HV_PGSZ_MASK_256MB
);
1251 ktsb_descr
[1].assoc
= 1;
1252 ktsb_descr
[1].num_ttes
= KERNEL_TSB4M_NENTRIES
;
1253 ktsb_descr
[1].ctx_idx
= 0;
1254 ktsb_descr
[1].tsb_base
= ktsb_pa
;
1255 ktsb_descr
[1].resv
= 0;
1258 void __cpuinit
sun4v_ktsb_register(void)
1260 register unsigned long func
asm("%o5");
1261 register unsigned long arg0
asm("%o0");
1262 register unsigned long arg1
asm("%o1");
1265 pa
= kern_base
+ ((unsigned long)&ktsb_descr
[0] - KERNBASE
);
1267 func
= HV_FAST_MMU_TSB_CTX0
;
1270 __asm__
__volatile__("ta %6"
1271 : "=&r" (func
), "=&r" (arg0
), "=&r" (arg1
)
1272 : "0" (func
), "1" (arg0
), "2" (arg1
),
1273 "i" (HV_FAST_TRAP
));
1276 /* paging_init() sets up the page tables */
1278 extern void cheetah_ecache_flush_init(void);
1279 extern void sun4v_patch_tlb_handlers(void);
1281 static unsigned long last_valid_pfn
;
1282 pgd_t swapper_pg_dir
[2048];
1284 static void sun4u_pgprot_init(void);
1285 static void sun4v_pgprot_init(void);
1287 void __init
paging_init(void)
1289 unsigned long end_pfn
, pages_avail
, shift
, phys_base
;
1290 unsigned long real_end
, i
;
1292 kern_base
= (prom_boot_mapping_phys_low
>> 22UL) << 22UL;
1293 kern_size
= (unsigned long)&_end
- (unsigned long)KERNBASE
;
1295 /* Invalidate both kernel TSBs. */
1296 memset(swapper_tsb
, 0x40, sizeof(swapper_tsb
));
1297 memset(swapper_4m_tsb
, 0x40, sizeof(swapper_4m_tsb
));
1299 if (tlb_type
== hypervisor
)
1300 sun4v_pgprot_init();
1302 sun4u_pgprot_init();
1304 if (tlb_type
== cheetah_plus
||
1305 tlb_type
== hypervisor
)
1308 if (tlb_type
== hypervisor
) {
1309 sun4v_patch_tlb_handlers();
1313 /* Find available physical memory... */
1314 read_obp_memory("available", &pavail
[0], &pavail_ents
);
1316 phys_base
= 0xffffffffffffffffUL
;
1317 for (i
= 0; i
< pavail_ents
; i
++)
1318 phys_base
= min(phys_base
, pavail
[i
].phys_addr
);
1320 set_bit(0, mmu_context_bmap
);
1322 shift
= kern_base
+ PAGE_OFFSET
- ((unsigned long)KERNBASE
);
1324 real_end
= (unsigned long)_end
;
1325 if ((real_end
> ((unsigned long)KERNBASE
+ 0x400000)))
1327 if ((real_end
> ((unsigned long)KERNBASE
+ 0x800000))) {
1328 prom_printf("paging_init: Kernel > 8MB, too large.\n");
1332 /* Set kernel pgd to upper alias so physical page computations
1335 init_mm
.pgd
+= ((shift
) / (sizeof(pgd_t
)));
1337 memset(swapper_low_pmd_dir
, 0, sizeof(swapper_low_pmd_dir
));
1339 /* Now can init the kernel/bad page tables. */
1340 pud_set(pud_offset(&swapper_pg_dir
[0], 0),
1341 swapper_low_pmd_dir
+ (shift
/ sizeof(pgd_t
)));
1343 inherit_prom_mappings();
1345 /* Ok, we can use our TLB miss and window trap handlers safely. */
1350 if (tlb_type
== hypervisor
)
1351 sun4v_ktsb_register();
1353 /* Setup bootmem... */
1355 last_valid_pfn
= end_pfn
= bootmem_init(&pages_avail
, phys_base
);
1357 max_mapnr
= last_valid_pfn
;
1359 kernel_physical_mapping_init();
1361 prom_build_devicetree();
1364 unsigned long zones_size
[MAX_NR_ZONES
];
1365 unsigned long zholes_size
[MAX_NR_ZONES
];
1368 for (znum
= 0; znum
< MAX_NR_ZONES
; znum
++)
1369 zones_size
[znum
] = zholes_size
[znum
] = 0;
1371 zones_size
[ZONE_DMA
] = end_pfn
;
1372 zholes_size
[ZONE_DMA
] = end_pfn
- pages_avail
;
1374 free_area_init_node(0, &contig_page_data
, zones_size
,
1375 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
,
1382 static void __init
taint_real_pages(void)
1386 read_obp_memory("available", &pavail_rescan
[0], &pavail_rescan_ents
);
1388 /* Find changes discovered in the physmem available rescan and
1389 * reserve the lost portions in the bootmem maps.
1391 for (i
= 0; i
< pavail_ents
; i
++) {
1392 unsigned long old_start
, old_end
;
1394 old_start
= pavail
[i
].phys_addr
;
1395 old_end
= old_start
+
1397 while (old_start
< old_end
) {
1400 for (n
= 0; n
< pavail_rescan_ents
; n
++) {
1401 unsigned long new_start
, new_end
;
1403 new_start
= pavail_rescan
[n
].phys_addr
;
1404 new_end
= new_start
+
1405 pavail_rescan
[n
].reg_size
;
1407 if (new_start
<= old_start
&&
1408 new_end
>= (old_start
+ PAGE_SIZE
)) {
1409 set_bit(old_start
>> 22,
1410 sparc64_valid_addr_bitmap
);
1414 reserve_bootmem(old_start
, PAGE_SIZE
);
1417 old_start
+= PAGE_SIZE
;
1422 int __init
page_in_phys_avail(unsigned long paddr
)
1428 for (i
= 0; i
< pavail_rescan_ents
; i
++) {
1429 unsigned long start
, end
;
1431 start
= pavail_rescan
[i
].phys_addr
;
1432 end
= start
+ pavail_rescan
[i
].reg_size
;
1434 if (paddr
>= start
&& paddr
< end
)
1437 if (paddr
>= kern_base
&& paddr
< (kern_base
+ kern_size
))
1439 #ifdef CONFIG_BLK_DEV_INITRD
1440 if (paddr
>= __pa(initrd_start
) &&
1441 paddr
< __pa(PAGE_ALIGN(initrd_end
)))
1448 void __init
mem_init(void)
1450 unsigned long codepages
, datapages
, initpages
;
1451 unsigned long addr
, last
;
1454 i
= last_valid_pfn
>> ((22 - PAGE_SHIFT
) + 6);
1456 sparc64_valid_addr_bitmap
= (unsigned long *) alloc_bootmem(i
<< 3);
1457 if (sparc64_valid_addr_bitmap
== NULL
) {
1458 prom_printf("mem_init: Cannot alloc valid_addr_bitmap.\n");
1461 memset(sparc64_valid_addr_bitmap
, 0, i
<< 3);
1463 addr
= PAGE_OFFSET
+ kern_base
;
1464 last
= PAGE_ALIGN(kern_size
) + addr
;
1465 while (addr
< last
) {
1466 set_bit(__pa(addr
) >> 22, sparc64_valid_addr_bitmap
);
1472 high_memory
= __va(last_valid_pfn
<< PAGE_SHIFT
);
1474 #ifdef CONFIG_DEBUG_BOOTMEM
1475 prom_printf("mem_init: Calling free_all_bootmem().\n");
1477 totalram_pages
= num_physpages
= free_all_bootmem() - 1;
1480 * Set up the zero page, mark it reserved, so that page count
1481 * is not manipulated when freeing the page from user ptes.
1483 mem_map_zero
= alloc_pages(GFP_KERNEL
|__GFP_ZERO
, 0);
1484 if (mem_map_zero
== NULL
) {
1485 prom_printf("paging_init: Cannot alloc zero page.\n");
1488 SetPageReserved(mem_map_zero
);
1490 codepages
= (((unsigned long) _etext
) - ((unsigned long) _start
));
1491 codepages
= PAGE_ALIGN(codepages
) >> PAGE_SHIFT
;
1492 datapages
= (((unsigned long) _edata
) - ((unsigned long) _etext
));
1493 datapages
= PAGE_ALIGN(datapages
) >> PAGE_SHIFT
;
1494 initpages
= (((unsigned long) __init_end
) - ((unsigned long) __init_begin
));
1495 initpages
= PAGE_ALIGN(initpages
) >> PAGE_SHIFT
;
1497 printk("Memory: %uk available (%ldk kernel code, %ldk data, %ldk init) [%016lx,%016lx]\n",
1498 nr_free_pages() << (PAGE_SHIFT
-10),
1499 codepages
<< (PAGE_SHIFT
-10),
1500 datapages
<< (PAGE_SHIFT
-10),
1501 initpages
<< (PAGE_SHIFT
-10),
1502 PAGE_OFFSET
, (last_valid_pfn
<< PAGE_SHIFT
));
1504 if (tlb_type
== cheetah
|| tlb_type
== cheetah_plus
)
1505 cheetah_ecache_flush_init();
1508 void free_initmem(void)
1510 unsigned long addr
, initend
;
1513 * The init section is aligned to 8k in vmlinux.lds. Page align for >8k pagesizes.
1515 addr
= PAGE_ALIGN((unsigned long)(__init_begin
));
1516 initend
= (unsigned long)(__init_end
) & PAGE_MASK
;
1517 for (; addr
< initend
; addr
+= PAGE_SIZE
) {
1522 ((unsigned long) __va(kern_base
)) -
1523 ((unsigned long) KERNBASE
));
1524 memset((void *)addr
, POISON_FREE_INITMEM
, PAGE_SIZE
);
1525 p
= virt_to_page(page
);
1527 ClearPageReserved(p
);
1535 #ifdef CONFIG_BLK_DEV_INITRD
1536 void free_initrd_mem(unsigned long start
, unsigned long end
)
1539 printk ("Freeing initrd memory: %ldk freed\n", (end
- start
) >> 10);
1540 for (; start
< end
; start
+= PAGE_SIZE
) {
1541 struct page
*p
= virt_to_page(start
);
1543 ClearPageReserved(p
);
1552 #define _PAGE_CACHE_4U (_PAGE_CP_4U | _PAGE_CV_4U)
1553 #define _PAGE_CACHE_4V (_PAGE_CP_4V | _PAGE_CV_4V)
1554 #define __DIRTY_BITS_4U (_PAGE_MODIFIED_4U | _PAGE_WRITE_4U | _PAGE_W_4U)
1555 #define __DIRTY_BITS_4V (_PAGE_MODIFIED_4V | _PAGE_WRITE_4V | _PAGE_W_4V)
1556 #define __ACCESS_BITS_4U (_PAGE_ACCESSED_4U | _PAGE_READ_4U | _PAGE_R)
1557 #define __ACCESS_BITS_4V (_PAGE_ACCESSED_4V | _PAGE_READ_4V | _PAGE_R)
1559 pgprot_t PAGE_KERNEL __read_mostly
;
1560 EXPORT_SYMBOL(PAGE_KERNEL
);
1562 pgprot_t PAGE_KERNEL_LOCKED __read_mostly
;
1563 pgprot_t PAGE_COPY __read_mostly
;
1565 pgprot_t PAGE_SHARED __read_mostly
;
1566 EXPORT_SYMBOL(PAGE_SHARED
);
1568 pgprot_t PAGE_EXEC __read_mostly
;
1569 unsigned long pg_iobits __read_mostly
;
1571 unsigned long _PAGE_IE __read_mostly
;
1572 EXPORT_SYMBOL(_PAGE_IE
);
1574 unsigned long _PAGE_E __read_mostly
;
1575 EXPORT_SYMBOL(_PAGE_E
);
1577 unsigned long _PAGE_CACHE __read_mostly
;
1578 EXPORT_SYMBOL(_PAGE_CACHE
);
1580 static void prot_init_common(unsigned long page_none
,
1581 unsigned long page_shared
,
1582 unsigned long page_copy
,
1583 unsigned long page_readonly
,
1584 unsigned long page_exec_bit
)
1586 PAGE_COPY
= __pgprot(page_copy
);
1587 PAGE_SHARED
= __pgprot(page_shared
);
1589 protection_map
[0x0] = __pgprot(page_none
);
1590 protection_map
[0x1] = __pgprot(page_readonly
& ~page_exec_bit
);
1591 protection_map
[0x2] = __pgprot(page_copy
& ~page_exec_bit
);
1592 protection_map
[0x3] = __pgprot(page_copy
& ~page_exec_bit
);
1593 protection_map
[0x4] = __pgprot(page_readonly
);
1594 protection_map
[0x5] = __pgprot(page_readonly
);
1595 protection_map
[0x6] = __pgprot(page_copy
);
1596 protection_map
[0x7] = __pgprot(page_copy
);
1597 protection_map
[0x8] = __pgprot(page_none
);
1598 protection_map
[0x9] = __pgprot(page_readonly
& ~page_exec_bit
);
1599 protection_map
[0xa] = __pgprot(page_shared
& ~page_exec_bit
);
1600 protection_map
[0xb] = __pgprot(page_shared
& ~page_exec_bit
);
1601 protection_map
[0xc] = __pgprot(page_readonly
);
1602 protection_map
[0xd] = __pgprot(page_readonly
);
1603 protection_map
[0xe] = __pgprot(page_shared
);
1604 protection_map
[0xf] = __pgprot(page_shared
);
1607 static void __init
sun4u_pgprot_init(void)
1609 unsigned long page_none
, page_shared
, page_copy
, page_readonly
;
1610 unsigned long page_exec_bit
;
1612 PAGE_KERNEL
= __pgprot (_PAGE_PRESENT_4U
| _PAGE_VALID
|
1613 _PAGE_CACHE_4U
| _PAGE_P_4U
|
1614 __ACCESS_BITS_4U
| __DIRTY_BITS_4U
|
1616 PAGE_KERNEL_LOCKED
= __pgprot (_PAGE_PRESENT_4U
| _PAGE_VALID
|
1617 _PAGE_CACHE_4U
| _PAGE_P_4U
|
1618 __ACCESS_BITS_4U
| __DIRTY_BITS_4U
|
1619 _PAGE_EXEC_4U
| _PAGE_L_4U
);
1620 PAGE_EXEC
= __pgprot(_PAGE_EXEC_4U
);
1622 _PAGE_IE
= _PAGE_IE_4U
;
1623 _PAGE_E
= _PAGE_E_4U
;
1624 _PAGE_CACHE
= _PAGE_CACHE_4U
;
1626 pg_iobits
= (_PAGE_VALID
| _PAGE_PRESENT_4U
| __DIRTY_BITS_4U
|
1627 __ACCESS_BITS_4U
| _PAGE_E_4U
);
1629 kern_linear_pte_xor
[0] = (_PAGE_VALID
| _PAGE_SZ4MB_4U
) ^
1631 kern_linear_pte_xor
[0] |= (_PAGE_CP_4U
| _PAGE_CV_4U
|
1632 _PAGE_P_4U
| _PAGE_W_4U
);
1634 /* XXX Should use 256MB on Panther. XXX */
1635 kern_linear_pte_xor
[1] = kern_linear_pte_xor
[0];
1637 _PAGE_SZBITS
= _PAGE_SZBITS_4U
;
1638 _PAGE_ALL_SZ_BITS
= (_PAGE_SZ4MB_4U
| _PAGE_SZ512K_4U
|
1639 _PAGE_SZ64K_4U
| _PAGE_SZ8K_4U
|
1640 _PAGE_SZ32MB_4U
| _PAGE_SZ256MB_4U
);
1643 page_none
= _PAGE_PRESENT_4U
| _PAGE_ACCESSED_4U
| _PAGE_CACHE_4U
;
1644 page_shared
= (_PAGE_VALID
| _PAGE_PRESENT_4U
| _PAGE_CACHE_4U
|
1645 __ACCESS_BITS_4U
| _PAGE_WRITE_4U
| _PAGE_EXEC_4U
);
1646 page_copy
= (_PAGE_VALID
| _PAGE_PRESENT_4U
| _PAGE_CACHE_4U
|
1647 __ACCESS_BITS_4U
| _PAGE_EXEC_4U
);
1648 page_readonly
= (_PAGE_VALID
| _PAGE_PRESENT_4U
| _PAGE_CACHE_4U
|
1649 __ACCESS_BITS_4U
| _PAGE_EXEC_4U
);
1651 page_exec_bit
= _PAGE_EXEC_4U
;
1653 prot_init_common(page_none
, page_shared
, page_copy
, page_readonly
,
1657 static void __init
sun4v_pgprot_init(void)
1659 unsigned long page_none
, page_shared
, page_copy
, page_readonly
;
1660 unsigned long page_exec_bit
;
1662 PAGE_KERNEL
= __pgprot (_PAGE_PRESENT_4V
| _PAGE_VALID
|
1663 _PAGE_CACHE_4V
| _PAGE_P_4V
|
1664 __ACCESS_BITS_4V
| __DIRTY_BITS_4V
|
1666 PAGE_KERNEL_LOCKED
= PAGE_KERNEL
;
1667 PAGE_EXEC
= __pgprot(_PAGE_EXEC_4V
);
1669 _PAGE_IE
= _PAGE_IE_4V
;
1670 _PAGE_E
= _PAGE_E_4V
;
1671 _PAGE_CACHE
= _PAGE_CACHE_4V
;
1673 kern_linear_pte_xor
[0] = (_PAGE_VALID
| _PAGE_SZ4MB_4V
) ^
1675 kern_linear_pte_xor
[0] |= (_PAGE_CP_4V
| _PAGE_CV_4V
|
1676 _PAGE_P_4V
| _PAGE_W_4V
);
1678 kern_linear_pte_xor
[1] = (_PAGE_VALID
| _PAGE_SZ256MB_4V
) ^
1680 kern_linear_pte_xor
[1] |= (_PAGE_CP_4V
| _PAGE_CV_4V
|
1681 _PAGE_P_4V
| _PAGE_W_4V
);
1683 pg_iobits
= (_PAGE_VALID
| _PAGE_PRESENT_4V
| __DIRTY_BITS_4V
|
1684 __ACCESS_BITS_4V
| _PAGE_E_4V
);
1686 _PAGE_SZBITS
= _PAGE_SZBITS_4V
;
1687 _PAGE_ALL_SZ_BITS
= (_PAGE_SZ16GB_4V
| _PAGE_SZ2GB_4V
|
1688 _PAGE_SZ256MB_4V
| _PAGE_SZ32MB_4V
|
1689 _PAGE_SZ4MB_4V
| _PAGE_SZ512K_4V
|
1690 _PAGE_SZ64K_4V
| _PAGE_SZ8K_4V
);
1692 page_none
= _PAGE_PRESENT_4V
| _PAGE_ACCESSED_4V
| _PAGE_CACHE_4V
;
1693 page_shared
= (_PAGE_VALID
| _PAGE_PRESENT_4V
| _PAGE_CACHE_4V
|
1694 __ACCESS_BITS_4V
| _PAGE_WRITE_4V
| _PAGE_EXEC_4V
);
1695 page_copy
= (_PAGE_VALID
| _PAGE_PRESENT_4V
| _PAGE_CACHE_4V
|
1696 __ACCESS_BITS_4V
| _PAGE_EXEC_4V
);
1697 page_readonly
= (_PAGE_VALID
| _PAGE_PRESENT_4V
| _PAGE_CACHE_4V
|
1698 __ACCESS_BITS_4V
| _PAGE_EXEC_4V
);
1700 page_exec_bit
= _PAGE_EXEC_4V
;
1702 prot_init_common(page_none
, page_shared
, page_copy
, page_readonly
,
1706 unsigned long pte_sz_bits(unsigned long sz
)
1708 if (tlb_type
== hypervisor
) {
1712 return _PAGE_SZ8K_4V
;
1714 return _PAGE_SZ64K_4V
;
1716 return _PAGE_SZ512K_4V
;
1717 case 4 * 1024 * 1024:
1718 return _PAGE_SZ4MB_4V
;
1724 return _PAGE_SZ8K_4U
;
1726 return _PAGE_SZ64K_4U
;
1728 return _PAGE_SZ512K_4U
;
1729 case 4 * 1024 * 1024:
1730 return _PAGE_SZ4MB_4U
;
1735 pte_t
mk_pte_io(unsigned long page
, pgprot_t prot
, int space
, unsigned long page_size
)
1739 pte_val(pte
) = page
| pgprot_val(pgprot_noncached(prot
));
1740 pte_val(pte
) |= (((unsigned long)space
) << 32);
1741 pte_val(pte
) |= pte_sz_bits(page_size
);
1746 static unsigned long kern_large_tte(unsigned long paddr
)
1750 val
= (_PAGE_VALID
| _PAGE_SZ4MB_4U
|
1751 _PAGE_CP_4U
| _PAGE_CV_4U
| _PAGE_P_4U
|
1752 _PAGE_EXEC_4U
| _PAGE_L_4U
| _PAGE_W_4U
);
1753 if (tlb_type
== hypervisor
)
1754 val
= (_PAGE_VALID
| _PAGE_SZ4MB_4V
|
1755 _PAGE_CP_4V
| _PAGE_CV_4V
| _PAGE_P_4V
|
1756 _PAGE_EXEC_4V
| _PAGE_W_4V
);
1762 * Translate PROM's mapping we capture at boot time into physical address.
1763 * The second parameter is only set from prom_callback() invocations.
1765 unsigned long prom_virt_to_phys(unsigned long promva
, int *error
)
1770 mask
= _PAGE_PADDR_4U
;
1771 if (tlb_type
== hypervisor
)
1772 mask
= _PAGE_PADDR_4V
;
1774 for (i
= 0; i
< prom_trans_ents
; i
++) {
1775 struct linux_prom_translation
*p
= &prom_trans
[i
];
1777 if (promva
>= p
->virt
&&
1778 promva
< (p
->virt
+ p
->size
)) {
1779 unsigned long base
= p
->data
& mask
;
1783 return base
+ (promva
& (8192 - 1));
1791 /* XXX We should kill off this ugly thing at so me point. XXX */
1792 unsigned long sun4u_get_pte(unsigned long addr
)
1798 unsigned long mask
= _PAGE_PADDR_4U
;
1800 if (tlb_type
== hypervisor
)
1801 mask
= _PAGE_PADDR_4V
;
1803 if (addr
>= PAGE_OFFSET
)
1806 if ((addr
>= LOW_OBP_ADDRESS
) && (addr
< HI_OBP_ADDRESS
))
1807 return prom_virt_to_phys(addr
, NULL
);
1809 pgdp
= pgd_offset_k(addr
);
1810 pudp
= pud_offset(pgdp
, addr
);
1811 pmdp
= pmd_offset(pudp
, addr
);
1812 ptep
= pte_offset_kernel(pmdp
, addr
);
1814 return pte_val(*ptep
) & mask
;
1817 /* If not locked, zap it. */
1818 void __flush_tlb_all(void)
1820 unsigned long pstate
;
1823 __asm__
__volatile__("flushw\n\t"
1824 "rdpr %%pstate, %0\n\t"
1825 "wrpr %0, %1, %%pstate"
1828 if (tlb_type
== spitfire
) {
1829 for (i
= 0; i
< 64; i
++) {
1830 /* Spitfire Errata #32 workaround */
1831 /* NOTE: Always runs on spitfire, so no
1832 * cheetah+ page size encodings.
1834 __asm__
__volatile__("stxa %0, [%1] %2\n\t"
1838 "r" (PRIMARY_CONTEXT
), "i" (ASI_DMMU
));
1840 if (!(spitfire_get_dtlb_data(i
) & _PAGE_L_4U
)) {
1841 __asm__
__volatile__("stxa %%g0, [%0] %1\n\t"
1844 : "r" (TLB_TAG_ACCESS
), "i" (ASI_DMMU
));
1845 spitfire_put_dtlb_data(i
, 0x0UL
);
1848 /* Spitfire Errata #32 workaround */
1849 /* NOTE: Always runs on spitfire, so no
1850 * cheetah+ page size encodings.
1852 __asm__
__volatile__("stxa %0, [%1] %2\n\t"
1856 "r" (PRIMARY_CONTEXT
), "i" (ASI_DMMU
));
1858 if (!(spitfire_get_itlb_data(i
) & _PAGE_L_4U
)) {
1859 __asm__
__volatile__("stxa %%g0, [%0] %1\n\t"
1862 : "r" (TLB_TAG_ACCESS
), "i" (ASI_IMMU
));
1863 spitfire_put_itlb_data(i
, 0x0UL
);
1866 } else if (tlb_type
== cheetah
|| tlb_type
== cheetah_plus
) {
1867 cheetah_flush_dtlb_all();
1868 cheetah_flush_itlb_all();
1870 __asm__
__volatile__("wrpr %0, 0, %%pstate"
1874 #ifdef CONFIG_MEMORY_HOTPLUG
1876 void online_page(struct page
*page
)
1878 ClearPageReserved(page
);
1879 init_page_count(page
);
1885 int remove_memory(u64 start
, u64 size
)
1890 #endif /* CONFIG_MEMORY_HOTPLUG */