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/module.h>
9 #include <linux/kernel.h>
10 #include <linux/sched.h>
11 #include <linux/string.h>
12 #include <linux/init.h>
13 #include <linux/bootmem.h>
15 #include <linux/hugetlb.h>
16 #include <linux/slab.h>
17 #include <linux/initrd.h>
18 #include <linux/swap.h>
19 #include <linux/pagemap.h>
20 #include <linux/poison.h>
22 #include <linux/seq_file.h>
23 #include <linux/kprobes.h>
24 #include <linux/cache.h>
25 #include <linux/sort.h>
28 #include <asm/system.h>
30 #include <asm/pgalloc.h>
31 #include <asm/pgtable.h>
32 #include <asm/oplib.h>
33 #include <asm/iommu.h>
35 #include <asm/uaccess.h>
36 #include <asm/mmu_context.h>
37 #include <asm/tlbflush.h>
39 #include <asm/starfire.h>
41 #include <asm/spitfire.h>
42 #include <asm/sections.h>
44 #include <asm/hypervisor.h>
47 extern void device_scan(void);
49 #define MAX_PHYS_ADDRESS (1UL << 42UL)
50 #define KPTE_BITMAP_CHUNK_SZ (256UL * 1024UL * 1024UL)
51 #define KPTE_BITMAP_BYTES \
52 ((MAX_PHYS_ADDRESS / KPTE_BITMAP_CHUNK_SZ) / 8)
54 unsigned long kern_linear_pte_xor
[2] __read_mostly
;
56 /* A bitmap, one bit for every 256MB of physical memory. If the bit
57 * is clear, we should use a 4MB page (via kern_linear_pte_xor[0]) else
58 * if set we should use a 256MB page (via kern_linear_pte_xor[1]).
60 unsigned long kpte_linear_bitmap
[KPTE_BITMAP_BYTES
/ sizeof(unsigned long)];
62 #ifndef CONFIG_DEBUG_PAGEALLOC
63 /* A special kernel TSB for 4MB and 256MB linear mappings. */
64 struct tsb swapper_4m_tsb
[KERNEL_TSB4M_NENTRIES
];
69 static struct linux_prom64_registers pavail
[MAX_BANKS
] __initdata
;
70 static struct linux_prom64_registers pavail_rescan
[MAX_BANKS
] __initdata
;
71 static int pavail_ents __initdata
;
72 static int pavail_rescan_ents __initdata
;
74 static int cmp_p64(const void *a
, const void *b
)
76 const struct linux_prom64_registers
*x
= a
, *y
= b
;
78 if (x
->phys_addr
> y
->phys_addr
)
80 if (x
->phys_addr
< y
->phys_addr
)
85 static void __init
read_obp_memory(const char *property
,
86 struct linux_prom64_registers
*regs
,
89 int node
= prom_finddevice("/memory");
90 int prop_size
= prom_getproplen(node
, property
);
93 ents
= prop_size
/ sizeof(struct linux_prom64_registers
);
94 if (ents
> MAX_BANKS
) {
95 prom_printf("The machine has more %s property entries than "
96 "this kernel can support (%d).\n",
101 ret
= prom_getproperty(node
, property
, (char *) regs
, prop_size
);
103 prom_printf("Couldn't get %s property from /memory.\n");
107 /* Sanitize what we got from the firmware, by page aligning
110 for (i
= 0; i
< ents
; i
++) {
111 unsigned long base
, size
;
113 base
= regs
[i
].phys_addr
;
114 size
= regs
[i
].reg_size
;
117 if (base
& ~PAGE_MASK
) {
118 unsigned long new_base
= PAGE_ALIGN(base
);
120 size
-= new_base
- base
;
121 if ((long) size
< 0L)
126 /* If it is empty, simply get rid of it.
127 * This simplifies the logic of the other
128 * functions that process these arrays.
130 memmove(®s
[i
], ®s
[i
+ 1],
131 (ents
- i
- 1) * sizeof(regs
[0]));
136 regs
[i
].phys_addr
= base
;
137 regs
[i
].reg_size
= size
;
142 sort(regs
, ents
, sizeof(struct linux_prom64_registers
),
146 unsigned long *sparc64_valid_addr_bitmap __read_mostly
;
148 /* Kernel physical address base and size in bytes. */
149 unsigned long kern_base __read_mostly
;
150 unsigned long kern_size __read_mostly
;
152 /* get_new_mmu_context() uses "cache + 1". */
153 DEFINE_SPINLOCK(ctx_alloc_lock
);
154 unsigned long tlb_context_cache
= CTX_FIRST_VERSION
- 1;
155 #define CTX_BMAP_SLOTS (1UL << (CTX_NR_BITS - 6))
156 unsigned long mmu_context_bmap
[CTX_BMAP_SLOTS
];
158 /* Initial ramdisk setup */
159 extern unsigned long sparc_ramdisk_image64
;
160 extern unsigned int sparc_ramdisk_image
;
161 extern unsigned int sparc_ramdisk_size
;
163 struct page
*mem_map_zero __read_mostly
;
165 unsigned int sparc64_highest_unlocked_tlb_ent __read_mostly
;
167 unsigned long sparc64_kern_pri_context __read_mostly
;
168 unsigned long sparc64_kern_pri_nuc_bits __read_mostly
;
169 unsigned long sparc64_kern_sec_context __read_mostly
;
173 struct kmem_cache
*pgtable_cache __read_mostly
;
175 static void zero_ctor(void *addr
, struct kmem_cache
*cache
, unsigned long flags
)
180 extern void tsb_cache_init(void);
182 void pgtable_cache_init(void)
184 pgtable_cache
= kmem_cache_create("pgtable_cache",
185 PAGE_SIZE
, PAGE_SIZE
,
187 SLAB_MUST_HWCACHE_ALIGN
,
190 if (!pgtable_cache
) {
191 prom_printf("Could not create pgtable_cache\n");
197 #ifdef CONFIG_DEBUG_DCFLUSH
198 atomic_t dcpage_flushes
= ATOMIC_INIT(0);
200 atomic_t dcpage_flushes_xcall
= ATOMIC_INIT(0);
204 inline void flush_dcache_page_impl(struct page
*page
)
206 BUG_ON(tlb_type
== hypervisor
);
207 #ifdef CONFIG_DEBUG_DCFLUSH
208 atomic_inc(&dcpage_flushes
);
211 #ifdef DCACHE_ALIASING_POSSIBLE
212 __flush_dcache_page(page_address(page
),
213 ((tlb_type
== spitfire
) &&
214 page_mapping(page
) != NULL
));
216 if (page_mapping(page
) != NULL
&&
217 tlb_type
== spitfire
)
218 __flush_icache_page(__pa(page_address(page
)));
222 #define PG_dcache_dirty PG_arch_1
223 #define PG_dcache_cpu_shift 24UL
224 #define PG_dcache_cpu_mask (256UL - 1UL)
227 #error D-cache dirty tracking and thread_info->cpu need fixing for > 256 cpus
230 #define dcache_dirty_cpu(page) \
231 (((page)->flags >> PG_dcache_cpu_shift) & PG_dcache_cpu_mask)
233 static __inline__
void set_dcache_dirty(struct page
*page
, int this_cpu
)
235 unsigned long mask
= this_cpu
;
236 unsigned long non_cpu_bits
;
238 non_cpu_bits
= ~(PG_dcache_cpu_mask
<< PG_dcache_cpu_shift
);
239 mask
= (mask
<< PG_dcache_cpu_shift
) | (1UL << PG_dcache_dirty
);
241 __asm__
__volatile__("1:\n\t"
243 "and %%g7, %1, %%g1\n\t"
244 "or %%g1, %0, %%g1\n\t"
245 "casx [%2], %%g7, %%g1\n\t"
247 "membar #StoreLoad | #StoreStore\n\t"
248 "bne,pn %%xcc, 1b\n\t"
251 : "r" (mask
), "r" (non_cpu_bits
), "r" (&page
->flags
)
255 static __inline__
void clear_dcache_dirty_cpu(struct page
*page
, unsigned long cpu
)
257 unsigned long mask
= (1UL << PG_dcache_dirty
);
259 __asm__
__volatile__("! test_and_clear_dcache_dirty\n"
262 "srlx %%g7, %4, %%g1\n\t"
263 "and %%g1, %3, %%g1\n\t"
265 "bne,pn %%icc, 2f\n\t"
266 " andn %%g7, %1, %%g1\n\t"
267 "casx [%2], %%g7, %%g1\n\t"
269 "membar #StoreLoad | #StoreStore\n\t"
270 "bne,pn %%xcc, 1b\n\t"
274 : "r" (cpu
), "r" (mask
), "r" (&page
->flags
),
275 "i" (PG_dcache_cpu_mask
),
276 "i" (PG_dcache_cpu_shift
)
280 static inline void tsb_insert(struct tsb
*ent
, unsigned long tag
, unsigned long pte
)
282 unsigned long tsb_addr
= (unsigned long) ent
;
284 if (tlb_type
== cheetah_plus
|| tlb_type
== hypervisor
)
285 tsb_addr
= __pa(tsb_addr
);
287 __tsb_insert(tsb_addr
, tag
, pte
);
290 unsigned long _PAGE_ALL_SZ_BITS __read_mostly
;
291 unsigned long _PAGE_SZBITS __read_mostly
;
293 void update_mmu_cache(struct vm_area_struct
*vma
, unsigned long address
, pte_t pte
)
295 struct mm_struct
*mm
;
297 unsigned long tag
, flags
;
298 unsigned long tsb_index
, tsb_hash_shift
;
300 if (tlb_type
!= hypervisor
) {
301 unsigned long pfn
= pte_pfn(pte
);
302 unsigned long pg_flags
;
305 if (pfn_valid(pfn
) &&
306 (page
= pfn_to_page(pfn
), page_mapping(page
)) &&
307 ((pg_flags
= page
->flags
) & (1UL << PG_dcache_dirty
))) {
308 int cpu
= ((pg_flags
>> PG_dcache_cpu_shift
) &
310 int this_cpu
= get_cpu();
312 /* This is just to optimize away some function calls
316 flush_dcache_page_impl(page
);
318 smp_flush_dcache_page_impl(page
, cpu
);
320 clear_dcache_dirty_cpu(page
, cpu
);
328 tsb_index
= MM_TSB_BASE
;
329 tsb_hash_shift
= PAGE_SHIFT
;
331 spin_lock_irqsave(&mm
->context
.lock
, flags
);
333 #ifdef CONFIG_HUGETLB_PAGE
334 if (mm
->context
.tsb_block
[MM_TSB_HUGE
].tsb
!= NULL
) {
335 if ((tlb_type
== hypervisor
&&
336 (pte_val(pte
) & _PAGE_SZALL_4V
) == _PAGE_SZHUGE_4V
) ||
337 (tlb_type
!= hypervisor
&&
338 (pte_val(pte
) & _PAGE_SZALL_4U
) == _PAGE_SZHUGE_4U
)) {
339 tsb_index
= MM_TSB_HUGE
;
340 tsb_hash_shift
= HPAGE_SHIFT
;
345 tsb
= mm
->context
.tsb_block
[tsb_index
].tsb
;
346 tsb
+= ((address
>> tsb_hash_shift
) &
347 (mm
->context
.tsb_block
[tsb_index
].tsb_nentries
- 1UL));
348 tag
= (address
>> 22UL);
349 tsb_insert(tsb
, tag
, pte_val(pte
));
351 spin_unlock_irqrestore(&mm
->context
.lock
, flags
);
354 void flush_dcache_page(struct page
*page
)
356 struct address_space
*mapping
;
359 if (tlb_type
== hypervisor
)
362 /* Do not bother with the expensive D-cache flush if it
363 * is merely the zero page. The 'bigcore' testcase in GDB
364 * causes this case to run millions of times.
366 if (page
== ZERO_PAGE(0))
369 this_cpu
= get_cpu();
371 mapping
= page_mapping(page
);
372 if (mapping
&& !mapping_mapped(mapping
)) {
373 int dirty
= test_bit(PG_dcache_dirty
, &page
->flags
);
375 int dirty_cpu
= dcache_dirty_cpu(page
);
377 if (dirty_cpu
== this_cpu
)
379 smp_flush_dcache_page_impl(page
, dirty_cpu
);
381 set_dcache_dirty(page
, this_cpu
);
383 /* We could delay the flush for the !page_mapping
384 * case too. But that case is for exec env/arg
385 * pages and those are %99 certainly going to get
386 * faulted into the tlb (and thus flushed) anyways.
388 flush_dcache_page_impl(page
);
395 struct linux_prom_translation
{
401 /* Exported for kernel TLB miss handling in ktlb.S */
402 struct linux_prom_translation prom_trans
[512] __read_mostly
;
403 unsigned int prom_trans_ents __read_mostly
;
406 * Translate PROM's mapping we capture at boot time into physical address.
407 * The second parameter is only set from prom_callback() invocations.
409 static unsigned long prom_virt_to_phys(unsigned long promva
)
414 mask
= _PAGE_PADDR_4U
;
415 if (tlb_type
== hypervisor
)
416 mask
= _PAGE_PADDR_4V
;
418 for (i
= 0; i
< prom_trans_ents
; i
++) {
419 struct linux_prom_translation
*p
= &prom_trans
[i
];
421 if (promva
>= p
->virt
&&
422 promva
< (p
->virt
+ p
->size
)) {
423 unsigned long base
= p
->data
& mask
;
425 return base
+ (promva
& (8192 - 1));
431 static unsigned long kvaddr_to_phys(unsigned long addr
)
437 unsigned long mask
= _PAGE_PADDR_4U
;
439 if (tlb_type
== hypervisor
)
440 mask
= _PAGE_PADDR_4V
;
442 if (addr
>= PAGE_OFFSET
)
445 if ((addr
>= LOW_OBP_ADDRESS
) && (addr
< HI_OBP_ADDRESS
))
446 return prom_virt_to_phys(addr
);
448 pgdp
= pgd_offset_k(addr
);
449 pudp
= pud_offset(pgdp
, addr
);
450 pmdp
= pmd_offset(pudp
, addr
);
451 ptep
= pte_offset_kernel(pmdp
, addr
);
453 return pte_val(*ptep
) & mask
;
456 void __kprobes
flush_icache_range(unsigned long start
, unsigned long end
)
458 /* Cheetah and Hypervisor platform cpus have coherent I-cache. */
459 if (tlb_type
== spitfire
) {
462 for (kaddr
= start
; kaddr
< end
; kaddr
+= PAGE_SIZE
)
463 __flush_icache_page(kvaddr_to_phys(kaddr
));
469 printk("Mem-info:\n");
471 printk("Free swap: %6ldkB\n",
472 nr_swap_pages
<< (PAGE_SHIFT
-10));
473 printk("%ld pages of RAM\n", num_physpages
);
474 printk("%lu free pages\n", nr_free_pages());
477 void mmu_info(struct seq_file
*m
)
479 if (tlb_type
== cheetah
)
480 seq_printf(m
, "MMU Type\t: Cheetah\n");
481 else if (tlb_type
== cheetah_plus
)
482 seq_printf(m
, "MMU Type\t: Cheetah+\n");
483 else if (tlb_type
== spitfire
)
484 seq_printf(m
, "MMU Type\t: Spitfire\n");
485 else if (tlb_type
== hypervisor
)
486 seq_printf(m
, "MMU Type\t: Hypervisor (sun4v)\n");
488 seq_printf(m
, "MMU Type\t: ???\n");
490 #ifdef CONFIG_DEBUG_DCFLUSH
491 seq_printf(m
, "DCPageFlushes\t: %d\n",
492 atomic_read(&dcpage_flushes
));
494 seq_printf(m
, "DCPageFlushesXC\t: %d\n",
495 atomic_read(&dcpage_flushes_xcall
));
496 #endif /* CONFIG_SMP */
497 #endif /* CONFIG_DEBUG_DCFLUSH */
500 /* Exported for SMP bootup purposes. */
501 unsigned long kern_locked_tte_data
;
503 /* The obp translations are saved based on 8k pagesize, since obp can
504 * use a mixture of pagesizes. Misses to the LOW_OBP_ADDRESS ->
505 * HI_OBP_ADDRESS range are handled in ktlb.S.
507 static inline int in_obp_range(unsigned long vaddr
)
509 return (vaddr
>= LOW_OBP_ADDRESS
&&
510 vaddr
< HI_OBP_ADDRESS
);
513 static int cmp_ptrans(const void *a
, const void *b
)
515 const struct linux_prom_translation
*x
= a
, *y
= b
;
517 if (x
->virt
> y
->virt
)
519 if (x
->virt
< y
->virt
)
524 /* Read OBP translations property into 'prom_trans[]'. */
525 static void __init
read_obp_translations(void)
527 int n
, node
, ents
, first
, last
, i
;
529 node
= prom_finddevice("/virtual-memory");
530 n
= prom_getproplen(node
, "translations");
531 if (unlikely(n
== 0 || n
== -1)) {
532 prom_printf("prom_mappings: Couldn't get size.\n");
535 if (unlikely(n
> sizeof(prom_trans
))) {
536 prom_printf("prom_mappings: Size %Zd is too big.\n", n
);
540 if ((n
= prom_getproperty(node
, "translations",
541 (char *)&prom_trans
[0],
542 sizeof(prom_trans
))) == -1) {
543 prom_printf("prom_mappings: Couldn't get property.\n");
547 n
= n
/ sizeof(struct linux_prom_translation
);
551 sort(prom_trans
, ents
, sizeof(struct linux_prom_translation
),
554 /* Now kick out all the non-OBP entries. */
555 for (i
= 0; i
< ents
; i
++) {
556 if (in_obp_range(prom_trans
[i
].virt
))
560 for (; i
< ents
; i
++) {
561 if (!in_obp_range(prom_trans
[i
].virt
))
566 for (i
= 0; i
< (last
- first
); i
++) {
567 struct linux_prom_translation
*src
= &prom_trans
[i
+ first
];
568 struct linux_prom_translation
*dest
= &prom_trans
[i
];
572 for (; i
< ents
; i
++) {
573 struct linux_prom_translation
*dest
= &prom_trans
[i
];
574 dest
->virt
= dest
->size
= dest
->data
= 0x0UL
;
577 prom_trans_ents
= last
- first
;
579 if (tlb_type
== spitfire
) {
580 /* Clear diag TTE bits. */
581 for (i
= 0; i
< prom_trans_ents
; i
++)
582 prom_trans
[i
].data
&= ~0x0003fe0000000000UL
;
586 static void __init
hypervisor_tlb_lock(unsigned long vaddr
,
590 register unsigned long func
asm("%o5");
591 register unsigned long arg0
asm("%o0");
592 register unsigned long arg1
asm("%o1");
593 register unsigned long arg2
asm("%o2");
594 register unsigned long arg3
asm("%o3");
596 func
= HV_FAST_MMU_MAP_PERM_ADDR
;
601 __asm__
__volatile__("ta 0x80"
602 : "=&r" (func
), "=&r" (arg0
),
603 "=&r" (arg1
), "=&r" (arg2
),
605 : "0" (func
), "1" (arg0
), "2" (arg1
),
606 "3" (arg2
), "4" (arg3
));
608 prom_printf("hypervisor_tlb_lock[%lx:%lx:%lx:%lx]: "
609 "errors with %lx\n", vaddr
, 0, pte
, mmu
, arg0
);
614 static unsigned long kern_large_tte(unsigned long paddr
);
616 static void __init
remap_kernel(void)
618 unsigned long phys_page
, tte_vaddr
, tte_data
;
619 int tlb_ent
= sparc64_highest_locked_tlbent();
621 tte_vaddr
= (unsigned long) KERNBASE
;
622 phys_page
= (prom_boot_mapping_phys_low
>> 22UL) << 22UL;
623 tte_data
= kern_large_tte(phys_page
);
625 kern_locked_tte_data
= tte_data
;
627 /* Now lock us into the TLBs via Hypervisor or OBP. */
628 if (tlb_type
== hypervisor
) {
629 hypervisor_tlb_lock(tte_vaddr
, tte_data
, HV_MMU_DMMU
);
630 hypervisor_tlb_lock(tte_vaddr
, tte_data
, HV_MMU_IMMU
);
632 tte_vaddr
+= 0x400000;
633 tte_data
+= 0x400000;
634 hypervisor_tlb_lock(tte_vaddr
, tte_data
, HV_MMU_DMMU
);
635 hypervisor_tlb_lock(tte_vaddr
, tte_data
, HV_MMU_IMMU
);
638 prom_dtlb_load(tlb_ent
, tte_data
, tte_vaddr
);
639 prom_itlb_load(tlb_ent
, tte_data
, tte_vaddr
);
642 prom_dtlb_load(tlb_ent
,
644 tte_vaddr
+ 0x400000);
645 prom_itlb_load(tlb_ent
,
647 tte_vaddr
+ 0x400000);
649 sparc64_highest_unlocked_tlb_ent
= tlb_ent
- 1;
651 if (tlb_type
== cheetah_plus
) {
652 sparc64_kern_pri_context
= (CTX_CHEETAH_PLUS_CTX0
|
653 CTX_CHEETAH_PLUS_NUC
);
654 sparc64_kern_pri_nuc_bits
= CTX_CHEETAH_PLUS_NUC
;
655 sparc64_kern_sec_context
= CTX_CHEETAH_PLUS_CTX0
;
660 static void __init
inherit_prom_mappings(void)
662 read_obp_translations();
664 /* Now fixup OBP's idea about where we really are mapped. */
665 prom_printf("Remapping the kernel... ");
667 prom_printf("done.\n");
670 void prom_world(int enter
)
673 set_fs((mm_segment_t
) { get_thread_current_ds() });
675 __asm__
__volatile__("flushw");
678 #ifdef DCACHE_ALIASING_POSSIBLE
679 void __flush_dcache_range(unsigned long start
, unsigned long end
)
683 if (tlb_type
== spitfire
) {
686 for (va
= start
; va
< end
; va
+= 32) {
687 spitfire_put_dcache_tag(va
& 0x3fe0, 0x0);
691 } else if (tlb_type
== cheetah
|| tlb_type
== cheetah_plus
) {
694 for (va
= start
; va
< end
; va
+= 32)
695 __asm__
__volatile__("stxa %%g0, [%0] %1\n\t"
699 "i" (ASI_DCACHE_INVALIDATE
));
702 #endif /* DCACHE_ALIASING_POSSIBLE */
704 /* Caller does TLB context flushing on local CPU if necessary.
705 * The caller also ensures that CTX_VALID(mm->context) is false.
707 * We must be careful about boundary cases so that we never
708 * let the user have CTX 0 (nucleus) or we ever use a CTX
709 * version of zero (and thus NO_CONTEXT would not be caught
710 * by version mis-match tests in mmu_context.h).
712 * Always invoked with interrupts disabled.
714 void get_new_mmu_context(struct mm_struct
*mm
)
716 unsigned long ctx
, new_ctx
;
717 unsigned long orig_pgsz_bits
;
721 spin_lock_irqsave(&ctx_alloc_lock
, flags
);
722 orig_pgsz_bits
= (mm
->context
.sparc64_ctx_val
& CTX_PGSZ_MASK
);
723 ctx
= (tlb_context_cache
+ 1) & CTX_NR_MASK
;
724 new_ctx
= find_next_zero_bit(mmu_context_bmap
, 1 << CTX_NR_BITS
, ctx
);
726 if (new_ctx
>= (1 << CTX_NR_BITS
)) {
727 new_ctx
= find_next_zero_bit(mmu_context_bmap
, ctx
, 1);
728 if (new_ctx
>= ctx
) {
730 new_ctx
= (tlb_context_cache
& CTX_VERSION_MASK
) +
733 new_ctx
= CTX_FIRST_VERSION
;
735 /* Don't call memset, for 16 entries that's just
738 mmu_context_bmap
[0] = 3;
739 mmu_context_bmap
[1] = 0;
740 mmu_context_bmap
[2] = 0;
741 mmu_context_bmap
[3] = 0;
742 for (i
= 4; i
< CTX_BMAP_SLOTS
; i
+= 4) {
743 mmu_context_bmap
[i
+ 0] = 0;
744 mmu_context_bmap
[i
+ 1] = 0;
745 mmu_context_bmap
[i
+ 2] = 0;
746 mmu_context_bmap
[i
+ 3] = 0;
752 mmu_context_bmap
[new_ctx
>>6] |= (1UL << (new_ctx
& 63));
753 new_ctx
|= (tlb_context_cache
& CTX_VERSION_MASK
);
755 tlb_context_cache
= new_ctx
;
756 mm
->context
.sparc64_ctx_val
= new_ctx
| orig_pgsz_bits
;
757 spin_unlock_irqrestore(&ctx_alloc_lock
, flags
);
759 if (unlikely(new_version
))
760 smp_new_mmu_context_version();
763 void sparc_ultra_dump_itlb(void)
767 if (tlb_type
== spitfire
) {
768 printk ("Contents of itlb: ");
769 for (slot
= 0; slot
< 14; slot
++) printk (" ");
770 printk ("%2x:%016lx,%016lx\n",
772 spitfire_get_itlb_tag(0), spitfire_get_itlb_data(0));
773 for (slot
= 1; slot
< 64; slot
+=3) {
774 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx %2x:%016lx,%016lx\n",
776 spitfire_get_itlb_tag(slot
), spitfire_get_itlb_data(slot
),
778 spitfire_get_itlb_tag(slot
+1), spitfire_get_itlb_data(slot
+1),
780 spitfire_get_itlb_tag(slot
+2), spitfire_get_itlb_data(slot
+2));
782 } else if (tlb_type
== cheetah
|| tlb_type
== cheetah_plus
) {
783 printk ("Contents of itlb0:\n");
784 for (slot
= 0; slot
< 16; slot
+=2) {
785 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
787 cheetah_get_litlb_tag(slot
), cheetah_get_litlb_data(slot
),
789 cheetah_get_litlb_tag(slot
+1), cheetah_get_litlb_data(slot
+1));
791 printk ("Contents of itlb2:\n");
792 for (slot
= 0; slot
< 128; slot
+=2) {
793 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
795 cheetah_get_itlb_tag(slot
), cheetah_get_itlb_data(slot
),
797 cheetah_get_itlb_tag(slot
+1), cheetah_get_itlb_data(slot
+1));
802 void sparc_ultra_dump_dtlb(void)
806 if (tlb_type
== spitfire
) {
807 printk ("Contents of dtlb: ");
808 for (slot
= 0; slot
< 14; slot
++) printk (" ");
809 printk ("%2x:%016lx,%016lx\n", 0,
810 spitfire_get_dtlb_tag(0), spitfire_get_dtlb_data(0));
811 for (slot
= 1; slot
< 64; slot
+=3) {
812 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx %2x:%016lx,%016lx\n",
814 spitfire_get_dtlb_tag(slot
), spitfire_get_dtlb_data(slot
),
816 spitfire_get_dtlb_tag(slot
+1), spitfire_get_dtlb_data(slot
+1),
818 spitfire_get_dtlb_tag(slot
+2), spitfire_get_dtlb_data(slot
+2));
820 } else if (tlb_type
== cheetah
|| tlb_type
== cheetah_plus
) {
821 printk ("Contents of dtlb0:\n");
822 for (slot
= 0; slot
< 16; slot
+=2) {
823 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
825 cheetah_get_ldtlb_tag(slot
), cheetah_get_ldtlb_data(slot
),
827 cheetah_get_ldtlb_tag(slot
+1), cheetah_get_ldtlb_data(slot
+1));
829 printk ("Contents of dtlb2:\n");
830 for (slot
= 0; slot
< 512; slot
+=2) {
831 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
833 cheetah_get_dtlb_tag(slot
, 2), cheetah_get_dtlb_data(slot
, 2),
835 cheetah_get_dtlb_tag(slot
+1, 2), cheetah_get_dtlb_data(slot
+1, 2));
837 if (tlb_type
== cheetah_plus
) {
838 printk ("Contents of dtlb3:\n");
839 for (slot
= 0; slot
< 512; slot
+=2) {
840 printk ("%2x:%016lx,%016lx %2x:%016lx,%016lx\n",
842 cheetah_get_dtlb_tag(slot
, 3), cheetah_get_dtlb_data(slot
, 3),
844 cheetah_get_dtlb_tag(slot
+1, 3), cheetah_get_dtlb_data(slot
+1, 3));
850 extern unsigned long cmdline_memory_size
;
852 /* Find a free area for the bootmem map, avoiding the kernel image
853 * and the initial ramdisk.
855 static unsigned long __init
choose_bootmap_pfn(unsigned long start_pfn
,
856 unsigned long end_pfn
)
858 unsigned long avoid_start
, avoid_end
, bootmap_size
;
861 bootmap_size
= ((end_pfn
- start_pfn
) + 7) / 8;
862 bootmap_size
= ALIGN(bootmap_size
, sizeof(long));
864 avoid_start
= avoid_end
= 0;
865 #ifdef CONFIG_BLK_DEV_INITRD
866 avoid_start
= initrd_start
;
867 avoid_end
= PAGE_ALIGN(initrd_end
);
870 #ifdef CONFIG_DEBUG_BOOTMEM
871 prom_printf("choose_bootmap_pfn: kern[%lx:%lx] avoid[%lx:%lx]\n",
872 kern_base
, PAGE_ALIGN(kern_base
+ kern_size
),
873 avoid_start
, avoid_end
);
875 for (i
= 0; i
< pavail_ents
; i
++) {
876 unsigned long start
, end
;
878 start
= pavail
[i
].phys_addr
;
879 end
= start
+ pavail
[i
].reg_size
;
881 while (start
< end
) {
882 if (start
>= kern_base
&&
883 start
< PAGE_ALIGN(kern_base
+ kern_size
)) {
884 start
= PAGE_ALIGN(kern_base
+ kern_size
);
887 if (start
>= avoid_start
&& start
< avoid_end
) {
892 if ((end
- start
) < bootmap_size
)
895 if (start
< kern_base
&&
896 (start
+ bootmap_size
) > kern_base
) {
897 start
= PAGE_ALIGN(kern_base
+ kern_size
);
901 if (start
< avoid_start
&&
902 (start
+ bootmap_size
) > avoid_start
) {
907 /* OK, it doesn't overlap anything, use it. */
908 #ifdef CONFIG_DEBUG_BOOTMEM
909 prom_printf("choose_bootmap_pfn: Using %lx [%lx]\n",
910 start
>> PAGE_SHIFT
, start
);
912 return start
>> PAGE_SHIFT
;
916 prom_printf("Cannot find free area for bootmap, aborting.\n");
920 static void __init
trim_pavail(unsigned long *cur_size_p
,
921 unsigned long *end_of_phys_p
)
923 unsigned long to_trim
= *cur_size_p
- cmdline_memory_size
;
924 unsigned long avoid_start
, avoid_end
;
927 to_trim
= PAGE_ALIGN(to_trim
);
929 avoid_start
= avoid_end
= 0;
930 #ifdef CONFIG_BLK_DEV_INITRD
931 avoid_start
= initrd_start
;
932 avoid_end
= PAGE_ALIGN(initrd_end
);
935 /* Trim some pavail[] entries in order to satisfy the
936 * requested "mem=xxx" kernel command line specification.
938 * We must not trim off the kernel image area nor the
939 * initial ramdisk range (if any). Also, we must not trim
940 * any pavail[] entry down to zero in order to preserve
941 * the invariant that all pavail[] entries have a non-zero
942 * size which is assumed by all of the code in here.
944 for (i
= 0; i
< pavail_ents
; i
++) {
945 unsigned long start
, end
, kern_end
;
946 unsigned long trim_low
, trim_high
, n
;
948 kern_end
= PAGE_ALIGN(kern_base
+ kern_size
);
950 trim_low
= start
= pavail
[i
].phys_addr
;
951 trim_high
= end
= start
+ pavail
[i
].reg_size
;
953 if (kern_base
>= start
&&
955 trim_low
= kern_base
;
959 if (kern_end
>= start
&&
961 trim_high
= kern_end
;
964 avoid_start
>= start
&&
966 if (trim_low
> avoid_start
)
967 trim_low
= avoid_start
;
968 if (avoid_end
>= end
)
972 avoid_end
>= start
&&
974 if (trim_high
< avoid_end
)
975 trim_high
= avoid_end
;
978 if (trim_high
<= trim_low
)
981 if (trim_low
== start
&& trim_high
== end
) {
982 /* Whole chunk is available for trimming.
983 * Trim all except one page, in order to keep
986 n
= (end
- start
) - PAGE_SIZE
;
991 pavail
[i
].phys_addr
+= n
;
992 pavail
[i
].reg_size
-= n
;
996 n
= (trim_low
- start
);
1001 pavail
[i
].phys_addr
+= n
;
1002 pavail
[i
].reg_size
-= n
;
1006 n
= end
- trim_high
;
1010 pavail
[i
].reg_size
-= n
;
1022 for (i
= 0; i
< pavail_ents
; i
++) {
1023 *end_of_phys_p
= pavail
[i
].phys_addr
+
1025 *cur_size_p
+= pavail
[i
].reg_size
;
1029 static unsigned long __init
bootmem_init(unsigned long *pages_avail
,
1030 unsigned long phys_base
)
1032 unsigned long bootmap_size
, end_pfn
;
1033 unsigned long end_of_phys_memory
= 0UL;
1034 unsigned long bootmap_pfn
, bytes_avail
, size
;
1037 #ifdef CONFIG_DEBUG_BOOTMEM
1038 prom_printf("bootmem_init: Scan pavail, ");
1042 for (i
= 0; i
< pavail_ents
; i
++) {
1043 end_of_phys_memory
= pavail
[i
].phys_addr
+
1045 bytes_avail
+= pavail
[i
].reg_size
;
1048 /* Determine the location of the initial ramdisk before trying
1049 * to honor the "mem=xxx" command line argument. We must know
1050 * where the kernel image and the ramdisk image are so that we
1051 * do not trim those two areas from the physical memory map.
1054 #ifdef CONFIG_BLK_DEV_INITRD
1055 /* Now have to check initial ramdisk, so that bootmap does not overwrite it */
1056 if (sparc_ramdisk_image
|| sparc_ramdisk_image64
) {
1057 unsigned long ramdisk_image
= sparc_ramdisk_image
?
1058 sparc_ramdisk_image
: sparc_ramdisk_image64
;
1059 ramdisk_image
-= KERNBASE
;
1060 initrd_start
= ramdisk_image
+ phys_base
;
1061 initrd_end
= initrd_start
+ sparc_ramdisk_size
;
1062 if (initrd_end
> end_of_phys_memory
) {
1063 printk(KERN_CRIT
"initrd extends beyond end of memory "
1064 "(0x%016lx > 0x%016lx)\ndisabling initrd\n",
1065 initrd_end
, end_of_phys_memory
);
1072 if (cmdline_memory_size
&&
1073 bytes_avail
> cmdline_memory_size
)
1074 trim_pavail(&bytes_avail
,
1075 &end_of_phys_memory
);
1077 *pages_avail
= bytes_avail
>> PAGE_SHIFT
;
1079 end_pfn
= end_of_phys_memory
>> PAGE_SHIFT
;
1081 /* Initialize the boot-time allocator. */
1082 max_pfn
= max_low_pfn
= end_pfn
;
1083 min_low_pfn
= (phys_base
>> PAGE_SHIFT
);
1085 bootmap_pfn
= choose_bootmap_pfn(min_low_pfn
, end_pfn
);
1087 #ifdef CONFIG_DEBUG_BOOTMEM
1088 prom_printf("init_bootmem(min[%lx], bootmap[%lx], max[%lx])\n",
1089 min_low_pfn
, bootmap_pfn
, max_low_pfn
);
1091 bootmap_size
= init_bootmem_node(NODE_DATA(0), bootmap_pfn
,
1092 min_low_pfn
, end_pfn
);
1094 /* Now register the available physical memory with the
1097 for (i
= 0; i
< pavail_ents
; i
++) {
1098 #ifdef CONFIG_DEBUG_BOOTMEM
1099 prom_printf("free_bootmem(pavail:%d): base[%lx] size[%lx]\n",
1100 i
, pavail
[i
].phys_addr
, pavail
[i
].reg_size
);
1102 free_bootmem(pavail
[i
].phys_addr
, pavail
[i
].reg_size
);
1105 #ifdef CONFIG_BLK_DEV_INITRD
1107 size
= initrd_end
- initrd_start
;
1109 /* Resert the initrd image area. */
1110 #ifdef CONFIG_DEBUG_BOOTMEM
1111 prom_printf("reserve_bootmem(initrd): base[%llx] size[%lx]\n",
1112 initrd_start
, initrd_end
);
1114 reserve_bootmem(initrd_start
, size
);
1115 *pages_avail
-= PAGE_ALIGN(size
) >> PAGE_SHIFT
;
1117 initrd_start
+= PAGE_OFFSET
;
1118 initrd_end
+= PAGE_OFFSET
;
1121 /* Reserve the kernel text/data/bss. */
1122 #ifdef CONFIG_DEBUG_BOOTMEM
1123 prom_printf("reserve_bootmem(kernel): base[%lx] size[%lx]\n", kern_base
, kern_size
);
1125 reserve_bootmem(kern_base
, kern_size
);
1126 *pages_avail
-= PAGE_ALIGN(kern_size
) >> PAGE_SHIFT
;
1128 /* Reserve the bootmem map. We do not account for it
1129 * in pages_avail because we will release that memory
1130 * in free_all_bootmem.
1132 size
= bootmap_size
;
1133 #ifdef CONFIG_DEBUG_BOOTMEM
1134 prom_printf("reserve_bootmem(bootmap): base[%lx] size[%lx]\n",
1135 (bootmap_pfn
<< PAGE_SHIFT
), size
);
1137 reserve_bootmem((bootmap_pfn
<< PAGE_SHIFT
), size
);
1138 *pages_avail
-= PAGE_ALIGN(size
) >> PAGE_SHIFT
;
1140 for (i
= 0; i
< pavail_ents
; i
++) {
1141 unsigned long start_pfn
, end_pfn
;
1143 start_pfn
= pavail
[i
].phys_addr
>> PAGE_SHIFT
;
1144 end_pfn
= (start_pfn
+ (pavail
[i
].reg_size
>> PAGE_SHIFT
));
1145 #ifdef CONFIG_DEBUG_BOOTMEM
1146 prom_printf("memory_present(0, %lx, %lx)\n",
1147 start_pfn
, end_pfn
);
1149 memory_present(0, start_pfn
, end_pfn
);
1157 static struct linux_prom64_registers pall
[MAX_BANKS
] __initdata
;
1158 static int pall_ents __initdata
;
1160 #ifdef CONFIG_DEBUG_PAGEALLOC
1161 static unsigned long kernel_map_range(unsigned long pstart
, unsigned long pend
, pgprot_t prot
)
1163 unsigned long vstart
= PAGE_OFFSET
+ pstart
;
1164 unsigned long vend
= PAGE_OFFSET
+ pend
;
1165 unsigned long alloc_bytes
= 0UL;
1167 if ((vstart
& ~PAGE_MASK
) || (vend
& ~PAGE_MASK
)) {
1168 prom_printf("kernel_map: Unaligned physmem[%lx:%lx]\n",
1173 while (vstart
< vend
) {
1174 unsigned long this_end
, paddr
= __pa(vstart
);
1175 pgd_t
*pgd
= pgd_offset_k(vstart
);
1180 pud
= pud_offset(pgd
, vstart
);
1181 if (pud_none(*pud
)) {
1184 new = __alloc_bootmem(PAGE_SIZE
, PAGE_SIZE
, PAGE_SIZE
);
1185 alloc_bytes
+= PAGE_SIZE
;
1186 pud_populate(&init_mm
, pud
, new);
1189 pmd
= pmd_offset(pud
, vstart
);
1190 if (!pmd_present(*pmd
)) {
1193 new = __alloc_bootmem(PAGE_SIZE
, PAGE_SIZE
, PAGE_SIZE
);
1194 alloc_bytes
+= PAGE_SIZE
;
1195 pmd_populate_kernel(&init_mm
, pmd
, new);
1198 pte
= pte_offset_kernel(pmd
, vstart
);
1199 this_end
= (vstart
+ PMD_SIZE
) & PMD_MASK
;
1200 if (this_end
> vend
)
1203 while (vstart
< this_end
) {
1204 pte_val(*pte
) = (paddr
| pgprot_val(prot
));
1206 vstart
+= PAGE_SIZE
;
1215 extern unsigned int kvmap_linear_patch
[1];
1216 #endif /* CONFIG_DEBUG_PAGEALLOC */
1218 static void __init
mark_kpte_bitmap(unsigned long start
, unsigned long end
)
1220 const unsigned long shift_256MB
= 28;
1221 const unsigned long mask_256MB
= ((1UL << shift_256MB
) - 1UL);
1222 const unsigned long size_256MB
= (1UL << shift_256MB
);
1224 while (start
< end
) {
1227 remains
= end
- start
;
1228 if (remains
< size_256MB
)
1231 if (start
& mask_256MB
) {
1232 start
= (start
+ size_256MB
) & ~mask_256MB
;
1236 while (remains
>= size_256MB
) {
1237 unsigned long index
= start
>> shift_256MB
;
1239 __set_bit(index
, kpte_linear_bitmap
);
1241 start
+= size_256MB
;
1242 remains
-= size_256MB
;
1247 static void __init
kernel_physical_mapping_init(void)
1250 #ifdef CONFIG_DEBUG_PAGEALLOC
1251 unsigned long mem_alloced
= 0UL;
1254 read_obp_memory("reg", &pall
[0], &pall_ents
);
1256 for (i
= 0; i
< pall_ents
; i
++) {
1257 unsigned long phys_start
, phys_end
;
1259 phys_start
= pall
[i
].phys_addr
;
1260 phys_end
= phys_start
+ pall
[i
].reg_size
;
1262 mark_kpte_bitmap(phys_start
, phys_end
);
1264 #ifdef CONFIG_DEBUG_PAGEALLOC
1265 mem_alloced
+= kernel_map_range(phys_start
, phys_end
,
1270 #ifdef CONFIG_DEBUG_PAGEALLOC
1271 printk("Allocated %ld bytes for kernel page tables.\n",
1274 kvmap_linear_patch
[0] = 0x01000000; /* nop */
1275 flushi(&kvmap_linear_patch
[0]);
1281 #ifdef CONFIG_DEBUG_PAGEALLOC
1282 void kernel_map_pages(struct page
*page
, int numpages
, int enable
)
1284 unsigned long phys_start
= page_to_pfn(page
) << PAGE_SHIFT
;
1285 unsigned long phys_end
= phys_start
+ (numpages
* PAGE_SIZE
);
1287 kernel_map_range(phys_start
, phys_end
,
1288 (enable
? PAGE_KERNEL
: __pgprot(0)));
1290 flush_tsb_kernel_range(PAGE_OFFSET
+ phys_start
,
1291 PAGE_OFFSET
+ phys_end
);
1293 /* we should perform an IPI and flush all tlbs,
1294 * but that can deadlock->flush only current cpu.
1296 __flush_tlb_kernel_range(PAGE_OFFSET
+ phys_start
,
1297 PAGE_OFFSET
+ phys_end
);
1301 unsigned long __init
find_ecache_flush_span(unsigned long size
)
1305 for (i
= 0; i
< pavail_ents
; i
++) {
1306 if (pavail
[i
].reg_size
>= size
)
1307 return pavail
[i
].phys_addr
;
1313 static void __init
tsb_phys_patch(void)
1315 struct tsb_ldquad_phys_patch_entry
*pquad
;
1316 struct tsb_phys_patch_entry
*p
;
1318 pquad
= &__tsb_ldquad_phys_patch
;
1319 while (pquad
< &__tsb_ldquad_phys_patch_end
) {
1320 unsigned long addr
= pquad
->addr
;
1322 if (tlb_type
== hypervisor
)
1323 *(unsigned int *) addr
= pquad
->sun4v_insn
;
1325 *(unsigned int *) addr
= pquad
->sun4u_insn
;
1327 __asm__
__volatile__("flush %0"
1334 p
= &__tsb_phys_patch
;
1335 while (p
< &__tsb_phys_patch_end
) {
1336 unsigned long addr
= p
->addr
;
1338 *(unsigned int *) addr
= p
->insn
;
1340 __asm__
__volatile__("flush %0"
1348 /* Don't mark as init, we give this to the Hypervisor. */
1349 #ifndef CONFIG_DEBUG_PAGEALLOC
1350 #define NUM_KTSB_DESCR 2
1352 #define NUM_KTSB_DESCR 1
1354 static struct hv_tsb_descr ktsb_descr
[NUM_KTSB_DESCR
];
1355 extern struct tsb swapper_tsb
[KERNEL_TSB_NENTRIES
];
1357 static void __init
sun4v_ktsb_init(void)
1359 unsigned long ktsb_pa
;
1361 /* First KTSB for PAGE_SIZE mappings. */
1362 ktsb_pa
= kern_base
+ ((unsigned long)&swapper_tsb
[0] - KERNBASE
);
1364 switch (PAGE_SIZE
) {
1367 ktsb_descr
[0].pgsz_idx
= HV_PGSZ_IDX_8K
;
1368 ktsb_descr
[0].pgsz_mask
= HV_PGSZ_MASK_8K
;
1372 ktsb_descr
[0].pgsz_idx
= HV_PGSZ_IDX_64K
;
1373 ktsb_descr
[0].pgsz_mask
= HV_PGSZ_MASK_64K
;
1377 ktsb_descr
[0].pgsz_idx
= HV_PGSZ_IDX_512K
;
1378 ktsb_descr
[0].pgsz_mask
= HV_PGSZ_MASK_512K
;
1381 case 4 * 1024 * 1024:
1382 ktsb_descr
[0].pgsz_idx
= HV_PGSZ_IDX_4MB
;
1383 ktsb_descr
[0].pgsz_mask
= HV_PGSZ_MASK_4MB
;
1387 ktsb_descr
[0].assoc
= 1;
1388 ktsb_descr
[0].num_ttes
= KERNEL_TSB_NENTRIES
;
1389 ktsb_descr
[0].ctx_idx
= 0;
1390 ktsb_descr
[0].tsb_base
= ktsb_pa
;
1391 ktsb_descr
[0].resv
= 0;
1393 #ifndef CONFIG_DEBUG_PAGEALLOC
1394 /* Second KTSB for 4MB/256MB mappings. */
1395 ktsb_pa
= (kern_base
+
1396 ((unsigned long)&swapper_4m_tsb
[0] - KERNBASE
));
1398 ktsb_descr
[1].pgsz_idx
= HV_PGSZ_IDX_4MB
;
1399 ktsb_descr
[1].pgsz_mask
= (HV_PGSZ_MASK_4MB
|
1400 HV_PGSZ_MASK_256MB
);
1401 ktsb_descr
[1].assoc
= 1;
1402 ktsb_descr
[1].num_ttes
= KERNEL_TSB4M_NENTRIES
;
1403 ktsb_descr
[1].ctx_idx
= 0;
1404 ktsb_descr
[1].tsb_base
= ktsb_pa
;
1405 ktsb_descr
[1].resv
= 0;
1409 void __cpuinit
sun4v_ktsb_register(void)
1411 register unsigned long func
asm("%o5");
1412 register unsigned long arg0
asm("%o0");
1413 register unsigned long arg1
asm("%o1");
1416 pa
= kern_base
+ ((unsigned long)&ktsb_descr
[0] - KERNBASE
);
1418 func
= HV_FAST_MMU_TSB_CTX0
;
1419 arg0
= NUM_KTSB_DESCR
;
1421 __asm__
__volatile__("ta %6"
1422 : "=&r" (func
), "=&r" (arg0
), "=&r" (arg1
)
1423 : "0" (func
), "1" (arg0
), "2" (arg1
),
1424 "i" (HV_FAST_TRAP
));
1427 /* paging_init() sets up the page tables */
1429 extern void cheetah_ecache_flush_init(void);
1430 extern void sun4v_patch_tlb_handlers(void);
1432 static unsigned long last_valid_pfn
;
1433 pgd_t swapper_pg_dir
[2048];
1435 static void sun4u_pgprot_init(void);
1436 static void sun4v_pgprot_init(void);
1438 void __init
paging_init(void)
1440 unsigned long end_pfn
, pages_avail
, shift
, phys_base
;
1441 unsigned long real_end
, i
;
1443 kern_base
= (prom_boot_mapping_phys_low
>> 22UL) << 22UL;
1444 kern_size
= (unsigned long)&_end
- (unsigned long)KERNBASE
;
1446 /* Invalidate both kernel TSBs. */
1447 memset(swapper_tsb
, 0x40, sizeof(swapper_tsb
));
1448 #ifndef CONFIG_DEBUG_PAGEALLOC
1449 memset(swapper_4m_tsb
, 0x40, sizeof(swapper_4m_tsb
));
1452 if (tlb_type
== hypervisor
)
1453 sun4v_pgprot_init();
1455 sun4u_pgprot_init();
1457 if (tlb_type
== cheetah_plus
||
1458 tlb_type
== hypervisor
)
1461 if (tlb_type
== hypervisor
) {
1462 sun4v_patch_tlb_handlers();
1466 /* Find available physical memory... */
1467 read_obp_memory("available", &pavail
[0], &pavail_ents
);
1469 phys_base
= 0xffffffffffffffffUL
;
1470 for (i
= 0; i
< pavail_ents
; i
++)
1471 phys_base
= min(phys_base
, pavail
[i
].phys_addr
);
1473 set_bit(0, mmu_context_bmap
);
1475 shift
= kern_base
+ PAGE_OFFSET
- ((unsigned long)KERNBASE
);
1477 real_end
= (unsigned long)_end
;
1478 if ((real_end
> ((unsigned long)KERNBASE
+ 0x400000)))
1480 if ((real_end
> ((unsigned long)KERNBASE
+ 0x800000))) {
1481 prom_printf("paging_init: Kernel > 8MB, too large.\n");
1485 /* Set kernel pgd to upper alias so physical page computations
1488 init_mm
.pgd
+= ((shift
) / (sizeof(pgd_t
)));
1490 memset(swapper_low_pmd_dir
, 0, sizeof(swapper_low_pmd_dir
));
1492 /* Now can init the kernel/bad page tables. */
1493 pud_set(pud_offset(&swapper_pg_dir
[0], 0),
1494 swapper_low_pmd_dir
+ (shift
/ sizeof(pgd_t
)));
1496 inherit_prom_mappings();
1498 /* Ok, we can use our TLB miss and window trap handlers safely. */
1503 if (tlb_type
== hypervisor
)
1504 sun4v_ktsb_register();
1506 /* Setup bootmem... */
1508 last_valid_pfn
= end_pfn
= bootmem_init(&pages_avail
, phys_base
);
1510 max_mapnr
= last_valid_pfn
;
1512 kernel_physical_mapping_init();
1514 prom_build_devicetree();
1517 unsigned long zones_size
[MAX_NR_ZONES
];
1518 unsigned long zholes_size
[MAX_NR_ZONES
];
1521 for (znum
= 0; znum
< MAX_NR_ZONES
; znum
++)
1522 zones_size
[znum
] = zholes_size
[znum
] = 0;
1524 zones_size
[ZONE_NORMAL
] = end_pfn
;
1525 zholes_size
[ZONE_NORMAL
] = end_pfn
- pages_avail
;
1527 free_area_init_node(0, &contig_page_data
, zones_size
,
1528 __pa(PAGE_OFFSET
) >> PAGE_SHIFT
,
1535 static void __init
taint_real_pages(void)
1539 read_obp_memory("available", &pavail_rescan
[0], &pavail_rescan_ents
);
1541 /* Find changes discovered in the physmem available rescan and
1542 * reserve the lost portions in the bootmem maps.
1544 for (i
= 0; i
< pavail_ents
; i
++) {
1545 unsigned long old_start
, old_end
;
1547 old_start
= pavail
[i
].phys_addr
;
1548 old_end
= old_start
+
1550 while (old_start
< old_end
) {
1553 for (n
= 0; n
< pavail_rescan_ents
; n
++) {
1554 unsigned long new_start
, new_end
;
1556 new_start
= pavail_rescan
[n
].phys_addr
;
1557 new_end
= new_start
+
1558 pavail_rescan
[n
].reg_size
;
1560 if (new_start
<= old_start
&&
1561 new_end
>= (old_start
+ PAGE_SIZE
)) {
1562 set_bit(old_start
>> 22,
1563 sparc64_valid_addr_bitmap
);
1567 reserve_bootmem(old_start
, PAGE_SIZE
);
1570 old_start
+= PAGE_SIZE
;
1575 int __init
page_in_phys_avail(unsigned long paddr
)
1581 for (i
= 0; i
< pavail_rescan_ents
; i
++) {
1582 unsigned long start
, end
;
1584 start
= pavail_rescan
[i
].phys_addr
;
1585 end
= start
+ pavail_rescan
[i
].reg_size
;
1587 if (paddr
>= start
&& paddr
< end
)
1590 if (paddr
>= kern_base
&& paddr
< (kern_base
+ kern_size
))
1592 #ifdef CONFIG_BLK_DEV_INITRD
1593 if (paddr
>= __pa(initrd_start
) &&
1594 paddr
< __pa(PAGE_ALIGN(initrd_end
)))
1601 void __init
mem_init(void)
1603 unsigned long codepages
, datapages
, initpages
;
1604 unsigned long addr
, last
;
1607 i
= last_valid_pfn
>> ((22 - PAGE_SHIFT
) + 6);
1609 sparc64_valid_addr_bitmap
= (unsigned long *) alloc_bootmem(i
<< 3);
1610 if (sparc64_valid_addr_bitmap
== NULL
) {
1611 prom_printf("mem_init: Cannot alloc valid_addr_bitmap.\n");
1614 memset(sparc64_valid_addr_bitmap
, 0, i
<< 3);
1616 addr
= PAGE_OFFSET
+ kern_base
;
1617 last
= PAGE_ALIGN(kern_size
) + addr
;
1618 while (addr
< last
) {
1619 set_bit(__pa(addr
) >> 22, sparc64_valid_addr_bitmap
);
1625 high_memory
= __va(last_valid_pfn
<< PAGE_SHIFT
);
1627 #ifdef CONFIG_DEBUG_BOOTMEM
1628 prom_printf("mem_init: Calling free_all_bootmem().\n");
1630 totalram_pages
= num_physpages
= free_all_bootmem() - 1;
1633 * Set up the zero page, mark it reserved, so that page count
1634 * is not manipulated when freeing the page from user ptes.
1636 mem_map_zero
= alloc_pages(GFP_KERNEL
|__GFP_ZERO
, 0);
1637 if (mem_map_zero
== NULL
) {
1638 prom_printf("paging_init: Cannot alloc zero page.\n");
1641 SetPageReserved(mem_map_zero
);
1643 codepages
= (((unsigned long) _etext
) - ((unsigned long) _start
));
1644 codepages
= PAGE_ALIGN(codepages
) >> PAGE_SHIFT
;
1645 datapages
= (((unsigned long) _edata
) - ((unsigned long) _etext
));
1646 datapages
= PAGE_ALIGN(datapages
) >> PAGE_SHIFT
;
1647 initpages
= (((unsigned long) __init_end
) - ((unsigned long) __init_begin
));
1648 initpages
= PAGE_ALIGN(initpages
) >> PAGE_SHIFT
;
1650 printk("Memory: %luk available (%ldk kernel code, %ldk data, %ldk init) [%016lx,%016lx]\n",
1651 nr_free_pages() << (PAGE_SHIFT
-10),
1652 codepages
<< (PAGE_SHIFT
-10),
1653 datapages
<< (PAGE_SHIFT
-10),
1654 initpages
<< (PAGE_SHIFT
-10),
1655 PAGE_OFFSET
, (last_valid_pfn
<< PAGE_SHIFT
));
1657 if (tlb_type
== cheetah
|| tlb_type
== cheetah_plus
)
1658 cheetah_ecache_flush_init();
1661 void free_initmem(void)
1663 unsigned long addr
, initend
;
1666 * The init section is aligned to 8k in vmlinux.lds. Page align for >8k pagesizes.
1668 addr
= PAGE_ALIGN((unsigned long)(__init_begin
));
1669 initend
= (unsigned long)(__init_end
) & PAGE_MASK
;
1670 for (; addr
< initend
; addr
+= PAGE_SIZE
) {
1675 ((unsigned long) __va(kern_base
)) -
1676 ((unsigned long) KERNBASE
));
1677 memset((void *)addr
, POISON_FREE_INITMEM
, PAGE_SIZE
);
1678 p
= virt_to_page(page
);
1680 ClearPageReserved(p
);
1688 #ifdef CONFIG_BLK_DEV_INITRD
1689 void free_initrd_mem(unsigned long start
, unsigned long end
)
1692 printk ("Freeing initrd memory: %ldk freed\n", (end
- start
) >> 10);
1693 for (; start
< end
; start
+= PAGE_SIZE
) {
1694 struct page
*p
= virt_to_page(start
);
1696 ClearPageReserved(p
);
1705 #define _PAGE_CACHE_4U (_PAGE_CP_4U | _PAGE_CV_4U)
1706 #define _PAGE_CACHE_4V (_PAGE_CP_4V | _PAGE_CV_4V)
1707 #define __DIRTY_BITS_4U (_PAGE_MODIFIED_4U | _PAGE_WRITE_4U | _PAGE_W_4U)
1708 #define __DIRTY_BITS_4V (_PAGE_MODIFIED_4V | _PAGE_WRITE_4V | _PAGE_W_4V)
1709 #define __ACCESS_BITS_4U (_PAGE_ACCESSED_4U | _PAGE_READ_4U | _PAGE_R)
1710 #define __ACCESS_BITS_4V (_PAGE_ACCESSED_4V | _PAGE_READ_4V | _PAGE_R)
1712 pgprot_t PAGE_KERNEL __read_mostly
;
1713 EXPORT_SYMBOL(PAGE_KERNEL
);
1715 pgprot_t PAGE_KERNEL_LOCKED __read_mostly
;
1716 pgprot_t PAGE_COPY __read_mostly
;
1718 pgprot_t PAGE_SHARED __read_mostly
;
1719 EXPORT_SYMBOL(PAGE_SHARED
);
1721 pgprot_t PAGE_EXEC __read_mostly
;
1722 unsigned long pg_iobits __read_mostly
;
1724 unsigned long _PAGE_IE __read_mostly
;
1725 EXPORT_SYMBOL(_PAGE_IE
);
1727 unsigned long _PAGE_E __read_mostly
;
1728 EXPORT_SYMBOL(_PAGE_E
);
1730 unsigned long _PAGE_CACHE __read_mostly
;
1731 EXPORT_SYMBOL(_PAGE_CACHE
);
1733 static void prot_init_common(unsigned long page_none
,
1734 unsigned long page_shared
,
1735 unsigned long page_copy
,
1736 unsigned long page_readonly
,
1737 unsigned long page_exec_bit
)
1739 PAGE_COPY
= __pgprot(page_copy
);
1740 PAGE_SHARED
= __pgprot(page_shared
);
1742 protection_map
[0x0] = __pgprot(page_none
);
1743 protection_map
[0x1] = __pgprot(page_readonly
& ~page_exec_bit
);
1744 protection_map
[0x2] = __pgprot(page_copy
& ~page_exec_bit
);
1745 protection_map
[0x3] = __pgprot(page_copy
& ~page_exec_bit
);
1746 protection_map
[0x4] = __pgprot(page_readonly
);
1747 protection_map
[0x5] = __pgprot(page_readonly
);
1748 protection_map
[0x6] = __pgprot(page_copy
);
1749 protection_map
[0x7] = __pgprot(page_copy
);
1750 protection_map
[0x8] = __pgprot(page_none
);
1751 protection_map
[0x9] = __pgprot(page_readonly
& ~page_exec_bit
);
1752 protection_map
[0xa] = __pgprot(page_shared
& ~page_exec_bit
);
1753 protection_map
[0xb] = __pgprot(page_shared
& ~page_exec_bit
);
1754 protection_map
[0xc] = __pgprot(page_readonly
);
1755 protection_map
[0xd] = __pgprot(page_readonly
);
1756 protection_map
[0xe] = __pgprot(page_shared
);
1757 protection_map
[0xf] = __pgprot(page_shared
);
1760 static void __init
sun4u_pgprot_init(void)
1762 unsigned long page_none
, page_shared
, page_copy
, page_readonly
;
1763 unsigned long page_exec_bit
;
1765 PAGE_KERNEL
= __pgprot (_PAGE_PRESENT_4U
| _PAGE_VALID
|
1766 _PAGE_CACHE_4U
| _PAGE_P_4U
|
1767 __ACCESS_BITS_4U
| __DIRTY_BITS_4U
|
1769 PAGE_KERNEL_LOCKED
= __pgprot (_PAGE_PRESENT_4U
| _PAGE_VALID
|
1770 _PAGE_CACHE_4U
| _PAGE_P_4U
|
1771 __ACCESS_BITS_4U
| __DIRTY_BITS_4U
|
1772 _PAGE_EXEC_4U
| _PAGE_L_4U
);
1773 PAGE_EXEC
= __pgprot(_PAGE_EXEC_4U
);
1775 _PAGE_IE
= _PAGE_IE_4U
;
1776 _PAGE_E
= _PAGE_E_4U
;
1777 _PAGE_CACHE
= _PAGE_CACHE_4U
;
1779 pg_iobits
= (_PAGE_VALID
| _PAGE_PRESENT_4U
| __DIRTY_BITS_4U
|
1780 __ACCESS_BITS_4U
| _PAGE_E_4U
);
1782 #ifdef CONFIG_DEBUG_PAGEALLOC
1783 kern_linear_pte_xor
[0] = (_PAGE_VALID
| _PAGE_SZBITS_4U
) ^
1786 kern_linear_pte_xor
[0] = (_PAGE_VALID
| _PAGE_SZ4MB_4U
) ^
1789 kern_linear_pte_xor
[0] |= (_PAGE_CP_4U
| _PAGE_CV_4U
|
1790 _PAGE_P_4U
| _PAGE_W_4U
);
1792 /* XXX Should use 256MB on Panther. XXX */
1793 kern_linear_pte_xor
[1] = kern_linear_pte_xor
[0];
1795 _PAGE_SZBITS
= _PAGE_SZBITS_4U
;
1796 _PAGE_ALL_SZ_BITS
= (_PAGE_SZ4MB_4U
| _PAGE_SZ512K_4U
|
1797 _PAGE_SZ64K_4U
| _PAGE_SZ8K_4U
|
1798 _PAGE_SZ32MB_4U
| _PAGE_SZ256MB_4U
);
1801 page_none
= _PAGE_PRESENT_4U
| _PAGE_ACCESSED_4U
| _PAGE_CACHE_4U
;
1802 page_shared
= (_PAGE_VALID
| _PAGE_PRESENT_4U
| _PAGE_CACHE_4U
|
1803 __ACCESS_BITS_4U
| _PAGE_WRITE_4U
| _PAGE_EXEC_4U
);
1804 page_copy
= (_PAGE_VALID
| _PAGE_PRESENT_4U
| _PAGE_CACHE_4U
|
1805 __ACCESS_BITS_4U
| _PAGE_EXEC_4U
);
1806 page_readonly
= (_PAGE_VALID
| _PAGE_PRESENT_4U
| _PAGE_CACHE_4U
|
1807 __ACCESS_BITS_4U
| _PAGE_EXEC_4U
);
1809 page_exec_bit
= _PAGE_EXEC_4U
;
1811 prot_init_common(page_none
, page_shared
, page_copy
, page_readonly
,
1815 static void __init
sun4v_pgprot_init(void)
1817 unsigned long page_none
, page_shared
, page_copy
, page_readonly
;
1818 unsigned long page_exec_bit
;
1820 PAGE_KERNEL
= __pgprot (_PAGE_PRESENT_4V
| _PAGE_VALID
|
1821 _PAGE_CACHE_4V
| _PAGE_P_4V
|
1822 __ACCESS_BITS_4V
| __DIRTY_BITS_4V
|
1824 PAGE_KERNEL_LOCKED
= PAGE_KERNEL
;
1825 PAGE_EXEC
= __pgprot(_PAGE_EXEC_4V
);
1827 _PAGE_IE
= _PAGE_IE_4V
;
1828 _PAGE_E
= _PAGE_E_4V
;
1829 _PAGE_CACHE
= _PAGE_CACHE_4V
;
1831 #ifdef CONFIG_DEBUG_PAGEALLOC
1832 kern_linear_pte_xor
[0] = (_PAGE_VALID
| _PAGE_SZBITS_4V
) ^
1835 kern_linear_pte_xor
[0] = (_PAGE_VALID
| _PAGE_SZ4MB_4V
) ^
1838 kern_linear_pte_xor
[0] |= (_PAGE_CP_4V
| _PAGE_CV_4V
|
1839 _PAGE_P_4V
| _PAGE_W_4V
);
1841 #ifdef CONFIG_DEBUG_PAGEALLOC
1842 kern_linear_pte_xor
[1] = (_PAGE_VALID
| _PAGE_SZBITS_4V
) ^
1845 kern_linear_pte_xor
[1] = (_PAGE_VALID
| _PAGE_SZ256MB_4V
) ^
1848 kern_linear_pte_xor
[1] |= (_PAGE_CP_4V
| _PAGE_CV_4V
|
1849 _PAGE_P_4V
| _PAGE_W_4V
);
1851 pg_iobits
= (_PAGE_VALID
| _PAGE_PRESENT_4V
| __DIRTY_BITS_4V
|
1852 __ACCESS_BITS_4V
| _PAGE_E_4V
);
1854 _PAGE_SZBITS
= _PAGE_SZBITS_4V
;
1855 _PAGE_ALL_SZ_BITS
= (_PAGE_SZ16GB_4V
| _PAGE_SZ2GB_4V
|
1856 _PAGE_SZ256MB_4V
| _PAGE_SZ32MB_4V
|
1857 _PAGE_SZ4MB_4V
| _PAGE_SZ512K_4V
|
1858 _PAGE_SZ64K_4V
| _PAGE_SZ8K_4V
);
1860 page_none
= _PAGE_PRESENT_4V
| _PAGE_ACCESSED_4V
| _PAGE_CACHE_4V
;
1861 page_shared
= (_PAGE_VALID
| _PAGE_PRESENT_4V
| _PAGE_CACHE_4V
|
1862 __ACCESS_BITS_4V
| _PAGE_WRITE_4V
| _PAGE_EXEC_4V
);
1863 page_copy
= (_PAGE_VALID
| _PAGE_PRESENT_4V
| _PAGE_CACHE_4V
|
1864 __ACCESS_BITS_4V
| _PAGE_EXEC_4V
);
1865 page_readonly
= (_PAGE_VALID
| _PAGE_PRESENT_4V
| _PAGE_CACHE_4V
|
1866 __ACCESS_BITS_4V
| _PAGE_EXEC_4V
);
1868 page_exec_bit
= _PAGE_EXEC_4V
;
1870 prot_init_common(page_none
, page_shared
, page_copy
, page_readonly
,
1874 unsigned long pte_sz_bits(unsigned long sz
)
1876 if (tlb_type
== hypervisor
) {
1880 return _PAGE_SZ8K_4V
;
1882 return _PAGE_SZ64K_4V
;
1884 return _PAGE_SZ512K_4V
;
1885 case 4 * 1024 * 1024:
1886 return _PAGE_SZ4MB_4V
;
1892 return _PAGE_SZ8K_4U
;
1894 return _PAGE_SZ64K_4U
;
1896 return _PAGE_SZ512K_4U
;
1897 case 4 * 1024 * 1024:
1898 return _PAGE_SZ4MB_4U
;
1903 pte_t
mk_pte_io(unsigned long page
, pgprot_t prot
, int space
, unsigned long page_size
)
1907 pte_val(pte
) = page
| pgprot_val(pgprot_noncached(prot
));
1908 pte_val(pte
) |= (((unsigned long)space
) << 32);
1909 pte_val(pte
) |= pte_sz_bits(page_size
);
1914 static unsigned long kern_large_tte(unsigned long paddr
)
1918 val
= (_PAGE_VALID
| _PAGE_SZ4MB_4U
|
1919 _PAGE_CP_4U
| _PAGE_CV_4U
| _PAGE_P_4U
|
1920 _PAGE_EXEC_4U
| _PAGE_L_4U
| _PAGE_W_4U
);
1921 if (tlb_type
== hypervisor
)
1922 val
= (_PAGE_VALID
| _PAGE_SZ4MB_4V
|
1923 _PAGE_CP_4V
| _PAGE_CV_4V
| _PAGE_P_4V
|
1924 _PAGE_EXEC_4V
| _PAGE_W_4V
);
1929 /* If not locked, zap it. */
1930 void __flush_tlb_all(void)
1932 unsigned long pstate
;
1935 __asm__
__volatile__("flushw\n\t"
1936 "rdpr %%pstate, %0\n\t"
1937 "wrpr %0, %1, %%pstate"
1940 if (tlb_type
== spitfire
) {
1941 for (i
= 0; i
< 64; i
++) {
1942 /* Spitfire Errata #32 workaround */
1943 /* NOTE: Always runs on spitfire, so no
1944 * cheetah+ page size encodings.
1946 __asm__
__volatile__("stxa %0, [%1] %2\n\t"
1950 "r" (PRIMARY_CONTEXT
), "i" (ASI_DMMU
));
1952 if (!(spitfire_get_dtlb_data(i
) & _PAGE_L_4U
)) {
1953 __asm__
__volatile__("stxa %%g0, [%0] %1\n\t"
1956 : "r" (TLB_TAG_ACCESS
), "i" (ASI_DMMU
));
1957 spitfire_put_dtlb_data(i
, 0x0UL
);
1960 /* Spitfire Errata #32 workaround */
1961 /* NOTE: Always runs on spitfire, so no
1962 * cheetah+ page size encodings.
1964 __asm__
__volatile__("stxa %0, [%1] %2\n\t"
1968 "r" (PRIMARY_CONTEXT
), "i" (ASI_DMMU
));
1970 if (!(spitfire_get_itlb_data(i
) & _PAGE_L_4U
)) {
1971 __asm__
__volatile__("stxa %%g0, [%0] %1\n\t"
1974 : "r" (TLB_TAG_ACCESS
), "i" (ASI_IMMU
));
1975 spitfire_put_itlb_data(i
, 0x0UL
);
1978 } else if (tlb_type
== cheetah
|| tlb_type
== cheetah_plus
) {
1979 cheetah_flush_dtlb_all();
1980 cheetah_flush_itlb_all();
1982 __asm__
__volatile__("wrpr %0, 0, %%pstate"
1986 #ifdef CONFIG_MEMORY_HOTPLUG
1988 void online_page(struct page
*page
)
1990 ClearPageReserved(page
);
1991 init_page_count(page
);
1997 int remove_memory(u64 start
, u64 size
)
2002 #endif /* CONFIG_MEMORY_HOTPLUG */