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>
26 #include <linux/percpu.h>
27 #include <linux/lmb.h>
28 #include <linux/mmzone.h>
31 #include <asm/system.h>
33 #include <asm/pgalloc.h>
34 #include <asm/pgtable.h>
35 #include <asm/oplib.h>
36 #include <asm/iommu.h>
38 #include <asm/uaccess.h>
39 #include <asm/mmu_context.h>
40 #include <asm/tlbflush.h>
42 #include <asm/starfire.h>
44 #include <asm/spitfire.h>
45 #include <asm/sections.h>
47 #include <asm/hypervisor.h>
49 #include <asm/sstate.h>
50 #include <asm/mdesc.h>
51 #include <asm/cpudata.h>
53 #define MAX_PHYS_ADDRESS (1UL << 42UL)
54 #define KPTE_BITMAP_CHUNK_SZ (256UL * 1024UL * 1024UL)
55 #define KPTE_BITMAP_BYTES \
56 ((MAX_PHYS_ADDRESS / KPTE_BITMAP_CHUNK_SZ) / 8)
58 unsigned long kern_linear_pte_xor
[2] __read_mostly
;
60 /* A bitmap, one bit for every 256MB of physical memory. If the bit
61 * is clear, we should use a 4MB page (via kern_linear_pte_xor[0]) else
62 * if set we should use a 256MB page (via kern_linear_pte_xor[1]).
64 unsigned long kpte_linear_bitmap
[KPTE_BITMAP_BYTES
/ sizeof(unsigned long)];
66 #ifndef CONFIG_DEBUG_PAGEALLOC
67 /* A special kernel TSB for 4MB and 256MB linear mappings.
68 * Space is allocated for this right after the trap table
69 * in arch/sparc64/kernel/head.S
71 extern struct tsb swapper_4m_tsb
[KERNEL_TSB4M_NENTRIES
];
76 static struct linux_prom64_registers pavail
[MAX_BANKS
] __initdata
;
77 static int pavail_ents __initdata
;
79 static int cmp_p64(const void *a
, const void *b
)
81 const struct linux_prom64_registers
*x
= a
, *y
= b
;
83 if (x
->phys_addr
> y
->phys_addr
)
85 if (x
->phys_addr
< y
->phys_addr
)
90 static void __init
read_obp_memory(const char *property
,
91 struct linux_prom64_registers
*regs
,
94 int node
= prom_finddevice("/memory");
95 int prop_size
= prom_getproplen(node
, property
);
98 ents
= prop_size
/ sizeof(struct linux_prom64_registers
);
99 if (ents
> MAX_BANKS
) {
100 prom_printf("The machine has more %s property entries than "
101 "this kernel can support (%d).\n",
102 property
, MAX_BANKS
);
106 ret
= prom_getproperty(node
, property
, (char *) regs
, prop_size
);
108 prom_printf("Couldn't get %s property from /memory.\n");
112 /* Sanitize what we got from the firmware, by page aligning
115 for (i
= 0; i
< ents
; i
++) {
116 unsigned long base
, size
;
118 base
= regs
[i
].phys_addr
;
119 size
= regs
[i
].reg_size
;
122 if (base
& ~PAGE_MASK
) {
123 unsigned long new_base
= PAGE_ALIGN(base
);
125 size
-= new_base
- base
;
126 if ((long) size
< 0L)
131 /* If it is empty, simply get rid of it.
132 * This simplifies the logic of the other
133 * functions that process these arrays.
135 memmove(®s
[i
], ®s
[i
+ 1],
136 (ents
- i
- 1) * sizeof(regs
[0]));
141 regs
[i
].phys_addr
= base
;
142 regs
[i
].reg_size
= size
;
147 sort(regs
, ents
, sizeof(struct linux_prom64_registers
),
151 unsigned long *sparc64_valid_addr_bitmap __read_mostly
;
153 /* Kernel physical address base and size in bytes. */
154 unsigned long kern_base __read_mostly
;
155 unsigned long kern_size __read_mostly
;
157 /* Initial ramdisk setup */
158 extern unsigned long sparc_ramdisk_image64
;
159 extern unsigned int sparc_ramdisk_image
;
160 extern unsigned int sparc_ramdisk_size
;
162 struct page
*mem_map_zero __read_mostly
;
163 EXPORT_SYMBOL(mem_map_zero
);
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
;
171 int num_kernel_image_mappings
;
173 #ifdef CONFIG_DEBUG_DCFLUSH
174 atomic_t dcpage_flushes
= ATOMIC_INIT(0);
176 atomic_t dcpage_flushes_xcall
= ATOMIC_INIT(0);
180 inline void flush_dcache_page_impl(struct page
*page
)
182 BUG_ON(tlb_type
== hypervisor
);
183 #ifdef CONFIG_DEBUG_DCFLUSH
184 atomic_inc(&dcpage_flushes
);
187 #ifdef DCACHE_ALIASING_POSSIBLE
188 __flush_dcache_page(page_address(page
),
189 ((tlb_type
== spitfire
) &&
190 page_mapping(page
) != NULL
));
192 if (page_mapping(page
) != NULL
&&
193 tlb_type
== spitfire
)
194 __flush_icache_page(__pa(page_address(page
)));
198 #define PG_dcache_dirty PG_arch_1
199 #define PG_dcache_cpu_shift 32UL
200 #define PG_dcache_cpu_mask \
201 ((1UL<<ilog2(roundup_pow_of_two(NR_CPUS)))-1UL)
203 #define dcache_dirty_cpu(page) \
204 (((page)->flags >> PG_dcache_cpu_shift) & PG_dcache_cpu_mask)
206 static inline void set_dcache_dirty(struct page
*page
, int this_cpu
)
208 unsigned long mask
= this_cpu
;
209 unsigned long non_cpu_bits
;
211 non_cpu_bits
= ~(PG_dcache_cpu_mask
<< PG_dcache_cpu_shift
);
212 mask
= (mask
<< PG_dcache_cpu_shift
) | (1UL << PG_dcache_dirty
);
214 __asm__
__volatile__("1:\n\t"
216 "and %%g7, %1, %%g1\n\t"
217 "or %%g1, %0, %%g1\n\t"
218 "casx [%2], %%g7, %%g1\n\t"
220 "membar #StoreLoad | #StoreStore\n\t"
221 "bne,pn %%xcc, 1b\n\t"
224 : "r" (mask
), "r" (non_cpu_bits
), "r" (&page
->flags
)
228 static inline void clear_dcache_dirty_cpu(struct page
*page
, unsigned long cpu
)
230 unsigned long mask
= (1UL << PG_dcache_dirty
);
232 __asm__
__volatile__("! test_and_clear_dcache_dirty\n"
235 "srlx %%g7, %4, %%g1\n\t"
236 "and %%g1, %3, %%g1\n\t"
238 "bne,pn %%icc, 2f\n\t"
239 " andn %%g7, %1, %%g1\n\t"
240 "casx [%2], %%g7, %%g1\n\t"
242 "membar #StoreLoad | #StoreStore\n\t"
243 "bne,pn %%xcc, 1b\n\t"
247 : "r" (cpu
), "r" (mask
), "r" (&page
->flags
),
248 "i" (PG_dcache_cpu_mask
),
249 "i" (PG_dcache_cpu_shift
)
253 static inline void tsb_insert(struct tsb
*ent
, unsigned long tag
, unsigned long pte
)
255 unsigned long tsb_addr
= (unsigned long) ent
;
257 if (tlb_type
== cheetah_plus
|| tlb_type
== hypervisor
)
258 tsb_addr
= __pa(tsb_addr
);
260 __tsb_insert(tsb_addr
, tag
, pte
);
263 unsigned long _PAGE_ALL_SZ_BITS __read_mostly
;
264 unsigned long _PAGE_SZBITS __read_mostly
;
266 void update_mmu_cache(struct vm_area_struct
*vma
, unsigned long address
, pte_t pte
)
268 struct mm_struct
*mm
;
270 unsigned long tag
, flags
;
271 unsigned long tsb_index
, tsb_hash_shift
;
273 if (tlb_type
!= hypervisor
) {
274 unsigned long pfn
= pte_pfn(pte
);
275 unsigned long pg_flags
;
278 if (pfn_valid(pfn
) &&
279 (page
= pfn_to_page(pfn
), page_mapping(page
)) &&
280 ((pg_flags
= page
->flags
) & (1UL << PG_dcache_dirty
))) {
281 int cpu
= ((pg_flags
>> PG_dcache_cpu_shift
) &
283 int this_cpu
= get_cpu();
285 /* This is just to optimize away some function calls
289 flush_dcache_page_impl(page
);
291 smp_flush_dcache_page_impl(page
, cpu
);
293 clear_dcache_dirty_cpu(page
, cpu
);
301 tsb_index
= MM_TSB_BASE
;
302 tsb_hash_shift
= PAGE_SHIFT
;
304 spin_lock_irqsave(&mm
->context
.lock
, flags
);
306 #ifdef CONFIG_HUGETLB_PAGE
307 if (mm
->context
.tsb_block
[MM_TSB_HUGE
].tsb
!= NULL
) {
308 if ((tlb_type
== hypervisor
&&
309 (pte_val(pte
) & _PAGE_SZALL_4V
) == _PAGE_SZHUGE_4V
) ||
310 (tlb_type
!= hypervisor
&&
311 (pte_val(pte
) & _PAGE_SZALL_4U
) == _PAGE_SZHUGE_4U
)) {
312 tsb_index
= MM_TSB_HUGE
;
313 tsb_hash_shift
= HPAGE_SHIFT
;
318 tsb
= mm
->context
.tsb_block
[tsb_index
].tsb
;
319 tsb
+= ((address
>> tsb_hash_shift
) &
320 (mm
->context
.tsb_block
[tsb_index
].tsb_nentries
- 1UL));
321 tag
= (address
>> 22UL);
322 tsb_insert(tsb
, tag
, pte_val(pte
));
324 spin_unlock_irqrestore(&mm
->context
.lock
, flags
);
327 void flush_dcache_page(struct page
*page
)
329 struct address_space
*mapping
;
332 if (tlb_type
== hypervisor
)
335 /* Do not bother with the expensive D-cache flush if it
336 * is merely the zero page. The 'bigcore' testcase in GDB
337 * causes this case to run millions of times.
339 if (page
== ZERO_PAGE(0))
342 this_cpu
= get_cpu();
344 mapping
= page_mapping(page
);
345 if (mapping
&& !mapping_mapped(mapping
)) {
346 int dirty
= test_bit(PG_dcache_dirty
, &page
->flags
);
348 int dirty_cpu
= dcache_dirty_cpu(page
);
350 if (dirty_cpu
== this_cpu
)
352 smp_flush_dcache_page_impl(page
, dirty_cpu
);
354 set_dcache_dirty(page
, this_cpu
);
356 /* We could delay the flush for the !page_mapping
357 * case too. But that case is for exec env/arg
358 * pages and those are %99 certainly going to get
359 * faulted into the tlb (and thus flushed) anyways.
361 flush_dcache_page_impl(page
);
368 void __kprobes
flush_icache_range(unsigned long start
, unsigned long end
)
370 /* Cheetah and Hypervisor platform cpus have coherent I-cache. */
371 if (tlb_type
== spitfire
) {
374 /* This code only runs on Spitfire cpus so this is
375 * why we can assume _PAGE_PADDR_4U.
377 for (kaddr
= start
; kaddr
< end
; kaddr
+= PAGE_SIZE
) {
378 unsigned long paddr
, mask
= _PAGE_PADDR_4U
;
380 if (kaddr
>= PAGE_OFFSET
)
381 paddr
= kaddr
& mask
;
383 pgd_t
*pgdp
= pgd_offset_k(kaddr
);
384 pud_t
*pudp
= pud_offset(pgdp
, kaddr
);
385 pmd_t
*pmdp
= pmd_offset(pudp
, kaddr
);
386 pte_t
*ptep
= pte_offset_kernel(pmdp
, kaddr
);
388 paddr
= pte_val(*ptep
) & mask
;
390 __flush_icache_page(paddr
);
397 unsigned long total
= 0, reserved
= 0;
398 unsigned long shared
= 0, cached
= 0;
401 printk(KERN_INFO
"Mem-info:\n");
403 printk(KERN_INFO
"Free swap: %6ldkB\n",
404 nr_swap_pages
<< (PAGE_SHIFT
-10));
405 for_each_online_pgdat(pgdat
) {
406 unsigned long i
, flags
;
408 pgdat_resize_lock(pgdat
, &flags
);
409 for (i
= 0; i
< pgdat
->node_spanned_pages
; i
++) {
410 struct page
*page
= pgdat_page_nr(pgdat
, i
);
412 if (PageReserved(page
))
414 else if (PageSwapCache(page
))
416 else if (page_count(page
))
417 shared
+= page_count(page
) - 1;
419 pgdat_resize_unlock(pgdat
, &flags
);
422 printk(KERN_INFO
"%lu pages of RAM\n", total
);
423 printk(KERN_INFO
"%lu reserved pages\n", reserved
);
424 printk(KERN_INFO
"%lu pages shared\n", shared
);
425 printk(KERN_INFO
"%lu pages swap cached\n", cached
);
427 printk(KERN_INFO
"%lu pages dirty\n",
428 global_page_state(NR_FILE_DIRTY
));
429 printk(KERN_INFO
"%lu pages writeback\n",
430 global_page_state(NR_WRITEBACK
));
431 printk(KERN_INFO
"%lu pages mapped\n",
432 global_page_state(NR_FILE_MAPPED
));
433 printk(KERN_INFO
"%lu pages slab\n",
434 global_page_state(NR_SLAB_RECLAIMABLE
) +
435 global_page_state(NR_SLAB_UNRECLAIMABLE
));
436 printk(KERN_INFO
"%lu pages pagetables\n",
437 global_page_state(NR_PAGETABLE
));
440 void mmu_info(struct seq_file
*m
)
442 if (tlb_type
== cheetah
)
443 seq_printf(m
, "MMU Type\t: Cheetah\n");
444 else if (tlb_type
== cheetah_plus
)
445 seq_printf(m
, "MMU Type\t: Cheetah+\n");
446 else if (tlb_type
== spitfire
)
447 seq_printf(m
, "MMU Type\t: Spitfire\n");
448 else if (tlb_type
== hypervisor
)
449 seq_printf(m
, "MMU Type\t: Hypervisor (sun4v)\n");
451 seq_printf(m
, "MMU Type\t: ???\n");
453 #ifdef CONFIG_DEBUG_DCFLUSH
454 seq_printf(m
, "DCPageFlushes\t: %d\n",
455 atomic_read(&dcpage_flushes
));
457 seq_printf(m
, "DCPageFlushesXC\t: %d\n",
458 atomic_read(&dcpage_flushes_xcall
));
459 #endif /* CONFIG_SMP */
460 #endif /* CONFIG_DEBUG_DCFLUSH */
463 struct linux_prom_translation
{
469 /* Exported for kernel TLB miss handling in ktlb.S */
470 struct linux_prom_translation prom_trans
[512] __read_mostly
;
471 unsigned int prom_trans_ents __read_mostly
;
473 /* Exported for SMP bootup purposes. */
474 unsigned long kern_locked_tte_data
;
476 /* The obp translations are saved based on 8k pagesize, since obp can
477 * use a mixture of pagesizes. Misses to the LOW_OBP_ADDRESS ->
478 * HI_OBP_ADDRESS range are handled in ktlb.S.
480 static inline int in_obp_range(unsigned long vaddr
)
482 return (vaddr
>= LOW_OBP_ADDRESS
&&
483 vaddr
< HI_OBP_ADDRESS
);
486 static int cmp_ptrans(const void *a
, const void *b
)
488 const struct linux_prom_translation
*x
= a
, *y
= b
;
490 if (x
->virt
> y
->virt
)
492 if (x
->virt
< y
->virt
)
497 /* Read OBP translations property into 'prom_trans[]'. */
498 static void __init
read_obp_translations(void)
500 int n
, node
, ents
, first
, last
, i
;
502 node
= prom_finddevice("/virtual-memory");
503 n
= prom_getproplen(node
, "translations");
504 if (unlikely(n
== 0 || n
== -1)) {
505 prom_printf("prom_mappings: Couldn't get size.\n");
508 if (unlikely(n
> sizeof(prom_trans
))) {
509 prom_printf("prom_mappings: Size %Zd is too big.\n", n
);
513 if ((n
= prom_getproperty(node
, "translations",
514 (char *)&prom_trans
[0],
515 sizeof(prom_trans
))) == -1) {
516 prom_printf("prom_mappings: Couldn't get property.\n");
520 n
= n
/ sizeof(struct linux_prom_translation
);
524 sort(prom_trans
, ents
, sizeof(struct linux_prom_translation
),
527 /* Now kick out all the non-OBP entries. */
528 for (i
= 0; i
< ents
; i
++) {
529 if (in_obp_range(prom_trans
[i
].virt
))
533 for (; i
< ents
; i
++) {
534 if (!in_obp_range(prom_trans
[i
].virt
))
539 for (i
= 0; i
< (last
- first
); i
++) {
540 struct linux_prom_translation
*src
= &prom_trans
[i
+ first
];
541 struct linux_prom_translation
*dest
= &prom_trans
[i
];
545 for (; i
< ents
; i
++) {
546 struct linux_prom_translation
*dest
= &prom_trans
[i
];
547 dest
->virt
= dest
->size
= dest
->data
= 0x0UL
;
550 prom_trans_ents
= last
- first
;
552 if (tlb_type
== spitfire
) {
553 /* Clear diag TTE bits. */
554 for (i
= 0; i
< prom_trans_ents
; i
++)
555 prom_trans
[i
].data
&= ~0x0003fe0000000000UL
;
559 static void __init
hypervisor_tlb_lock(unsigned long vaddr
,
563 unsigned long ret
= sun4v_mmu_map_perm_addr(vaddr
, 0, pte
, mmu
);
566 prom_printf("hypervisor_tlb_lock[%lx:%lx:%lx:%lx]: "
567 "errors with %lx\n", vaddr
, 0, pte
, mmu
, ret
);
572 static unsigned long kern_large_tte(unsigned long paddr
);
574 static void __init
remap_kernel(void)
576 unsigned long phys_page
, tte_vaddr
, tte_data
;
577 int i
, tlb_ent
= sparc64_highest_locked_tlbent();
579 tte_vaddr
= (unsigned long) KERNBASE
;
580 phys_page
= (prom_boot_mapping_phys_low
>> 22UL) << 22UL;
581 tte_data
= kern_large_tte(phys_page
);
583 kern_locked_tte_data
= tte_data
;
585 /* Now lock us into the TLBs via Hypervisor or OBP. */
586 if (tlb_type
== hypervisor
) {
587 for (i
= 0; i
< num_kernel_image_mappings
; i
++) {
588 hypervisor_tlb_lock(tte_vaddr
, tte_data
, HV_MMU_DMMU
);
589 hypervisor_tlb_lock(tte_vaddr
, tte_data
, HV_MMU_IMMU
);
590 tte_vaddr
+= 0x400000;
591 tte_data
+= 0x400000;
594 for (i
= 0; i
< num_kernel_image_mappings
; i
++) {
595 prom_dtlb_load(tlb_ent
- i
, tte_data
, tte_vaddr
);
596 prom_itlb_load(tlb_ent
- i
, tte_data
, tte_vaddr
);
597 tte_vaddr
+= 0x400000;
598 tte_data
+= 0x400000;
600 sparc64_highest_unlocked_tlb_ent
= tlb_ent
- i
;
602 if (tlb_type
== cheetah_plus
) {
603 sparc64_kern_pri_context
= (CTX_CHEETAH_PLUS_CTX0
|
604 CTX_CHEETAH_PLUS_NUC
);
605 sparc64_kern_pri_nuc_bits
= CTX_CHEETAH_PLUS_NUC
;
606 sparc64_kern_sec_context
= CTX_CHEETAH_PLUS_CTX0
;
611 static void __init
inherit_prom_mappings(void)
613 read_obp_translations();
615 /* Now fixup OBP's idea about where we really are mapped. */
616 printk("Remapping the kernel... ");
621 void prom_world(int enter
)
624 set_fs((mm_segment_t
) { get_thread_current_ds() });
626 __asm__
__volatile__("flushw");
629 void __flush_dcache_range(unsigned long start
, unsigned long end
)
633 if (tlb_type
== spitfire
) {
636 for (va
= start
; va
< end
; va
+= 32) {
637 spitfire_put_dcache_tag(va
& 0x3fe0, 0x0);
641 } else if (tlb_type
== cheetah
|| tlb_type
== cheetah_plus
) {
644 for (va
= start
; va
< end
; va
+= 32)
645 __asm__
__volatile__("stxa %%g0, [%0] %1\n\t"
649 "i" (ASI_DCACHE_INVALIDATE
));
653 /* get_new_mmu_context() uses "cache + 1". */
654 DEFINE_SPINLOCK(ctx_alloc_lock
);
655 unsigned long tlb_context_cache
= CTX_FIRST_VERSION
- 1;
656 #define MAX_CTX_NR (1UL << CTX_NR_BITS)
657 #define CTX_BMAP_SLOTS BITS_TO_LONGS(MAX_CTX_NR)
658 DECLARE_BITMAP(mmu_context_bmap
, MAX_CTX_NR
);
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 static int numa_enabled
= 1;
720 static int numa_debug
;
722 static int __init
early_numa(char *p
)
727 if (strstr(p
, "off"))
730 if (strstr(p
, "debug"))
735 early_param("numa", early_numa
);
737 #define numadbg(f, a...) \
738 do { if (numa_debug) \
739 printk(KERN_INFO f, ## a); \
742 static void __init
find_ramdisk(unsigned long phys_base
)
744 #ifdef CONFIG_BLK_DEV_INITRD
745 if (sparc_ramdisk_image
|| sparc_ramdisk_image64
) {
746 unsigned long ramdisk_image
;
748 /* Older versions of the bootloader only supported a
749 * 32-bit physical address for the ramdisk image
750 * location, stored at sparc_ramdisk_image. Newer
751 * SILO versions set sparc_ramdisk_image to zero and
752 * provide a full 64-bit physical address at
753 * sparc_ramdisk_image64.
755 ramdisk_image
= sparc_ramdisk_image
;
757 ramdisk_image
= sparc_ramdisk_image64
;
759 /* Another bootloader quirk. The bootloader normalizes
760 * the physical address to KERNBASE, so we have to
761 * factor that back out and add in the lowest valid
762 * physical page address to get the true physical address.
764 ramdisk_image
-= KERNBASE
;
765 ramdisk_image
+= phys_base
;
767 numadbg("Found ramdisk at physical address 0x%lx, size %u\n",
768 ramdisk_image
, sparc_ramdisk_size
);
770 initrd_start
= ramdisk_image
;
771 initrd_end
= ramdisk_image
+ sparc_ramdisk_size
;
773 lmb_reserve(initrd_start
, initrd_end
);
775 initrd_start
+= PAGE_OFFSET
;
776 initrd_end
+= PAGE_OFFSET
;
781 struct node_mem_mask
{
784 unsigned long bootmem_paddr
;
786 static struct node_mem_mask node_masks
[MAX_NUMNODES
];
787 static int num_node_masks
;
789 int numa_cpu_lookup_table
[NR_CPUS
];
790 cpumask_t numa_cpumask_lookup_table
[MAX_NUMNODES
];
792 #ifdef CONFIG_NEED_MULTIPLE_NODES
793 static bootmem_data_t plat_node_bdata
[MAX_NUMNODES
];
795 struct mdesc_mblock
{
798 u64 offset
; /* RA-to-PA */
800 static struct mdesc_mblock
*mblocks
;
801 static int num_mblocks
;
803 static unsigned long ra_to_pa(unsigned long addr
)
807 for (i
= 0; i
< num_mblocks
; i
++) {
808 struct mdesc_mblock
*m
= &mblocks
[i
];
810 if (addr
>= m
->base
&&
811 addr
< (m
->base
+ m
->size
)) {
819 static int find_node(unsigned long addr
)
823 addr
= ra_to_pa(addr
);
824 for (i
= 0; i
< num_node_masks
; i
++) {
825 struct node_mem_mask
*p
= &node_masks
[i
];
827 if ((addr
& p
->mask
) == p
->val
)
833 static unsigned long nid_range(unsigned long start
, unsigned long end
,
836 *nid
= find_node(start
);
838 while (start
< end
) {
839 int n
= find_node(start
);
849 static unsigned long nid_range(unsigned long start
, unsigned long end
,
857 /* This must be invoked after performing all of the necessary
858 * add_active_range() calls for 'nid'. We need to be able to get
859 * correct data from get_pfn_range_for_nid().
861 static void __init
allocate_node_data(int nid
)
863 unsigned long paddr
, num_pages
, start_pfn
, end_pfn
;
864 struct pglist_data
*p
;
866 #ifdef CONFIG_NEED_MULTIPLE_NODES
867 paddr
= lmb_alloc_nid(sizeof(struct pglist_data
),
868 SMP_CACHE_BYTES
, nid
, nid_range
);
870 prom_printf("Cannot allocate pglist_data for nid[%d]\n", nid
);
873 NODE_DATA(nid
) = __va(paddr
);
874 memset(NODE_DATA(nid
), 0, sizeof(struct pglist_data
));
876 NODE_DATA(nid
)->bdata
= &plat_node_bdata
[nid
];
881 get_pfn_range_for_nid(nid
, &start_pfn
, &end_pfn
);
882 p
->node_start_pfn
= start_pfn
;
883 p
->node_spanned_pages
= end_pfn
- start_pfn
;
885 if (p
->node_spanned_pages
) {
886 num_pages
= bootmem_bootmap_pages(p
->node_spanned_pages
);
888 paddr
= lmb_alloc_nid(num_pages
<< PAGE_SHIFT
, PAGE_SIZE
, nid
,
891 prom_printf("Cannot allocate bootmap for nid[%d]\n",
895 node_masks
[nid
].bootmem_paddr
= paddr
;
899 static void init_node_masks_nonnuma(void)
903 numadbg("Initializing tables for non-numa.\n");
905 node_masks
[0].mask
= node_masks
[0].val
= 0;
908 for (i
= 0; i
< NR_CPUS
; i
++)
909 numa_cpu_lookup_table
[i
] = 0;
911 numa_cpumask_lookup_table
[0] = CPU_MASK_ALL
;
914 #ifdef CONFIG_NEED_MULTIPLE_NODES
915 struct pglist_data
*node_data
[MAX_NUMNODES
];
917 EXPORT_SYMBOL(numa_cpu_lookup_table
);
918 EXPORT_SYMBOL(numa_cpumask_lookup_table
);
919 EXPORT_SYMBOL(node_data
);
921 struct mdesc_mlgroup
{
927 static struct mdesc_mlgroup
*mlgroups
;
928 static int num_mlgroups
;
930 static int scan_pio_for_cfg_handle(struct mdesc_handle
*md
, u64 pio
,
935 mdesc_for_each_arc(arc
, md
, pio
, MDESC_ARC_TYPE_FWD
) {
936 u64 target
= mdesc_arc_target(md
, arc
);
939 val
= mdesc_get_property(md
, target
,
941 if (val
&& *val
== cfg_handle
)
947 static int scan_arcs_for_cfg_handle(struct mdesc_handle
*md
, u64 grp
,
950 u64 arc
, candidate
, best_latency
= ~(u64
)0;
952 candidate
= MDESC_NODE_NULL
;
953 mdesc_for_each_arc(arc
, md
, grp
, MDESC_ARC_TYPE_FWD
) {
954 u64 target
= mdesc_arc_target(md
, arc
);
955 const char *name
= mdesc_node_name(md
, target
);
958 if (strcmp(name
, "pio-latency-group"))
961 val
= mdesc_get_property(md
, target
, "latency", NULL
);
965 if (*val
< best_latency
) {
971 if (candidate
== MDESC_NODE_NULL
)
974 return scan_pio_for_cfg_handle(md
, candidate
, cfg_handle
);
977 int of_node_to_nid(struct device_node
*dp
)
979 const struct linux_prom64_registers
*regs
;
980 struct mdesc_handle
*md
;
988 regs
= of_get_property(dp
, "reg", NULL
);
992 cfg_handle
= (regs
->phys_addr
>> 32UL) & 0x0fffffff;
998 mdesc_for_each_node_by_name(md
, grp
, "group") {
999 if (!scan_arcs_for_cfg_handle(md
, grp
, cfg_handle
)) {
1011 static void add_node_ranges(void)
1015 for (i
= 0; i
< lmb
.memory
.cnt
; i
++) {
1016 unsigned long size
= lmb_size_bytes(&lmb
.memory
, i
);
1017 unsigned long start
, end
;
1019 start
= lmb
.memory
.region
[i
].base
;
1021 while (start
< end
) {
1022 unsigned long this_end
;
1025 this_end
= nid_range(start
, end
, &nid
);
1027 numadbg("Adding active range nid[%d] "
1028 "start[%lx] end[%lx]\n",
1029 nid
, start
, this_end
);
1031 add_active_range(nid
,
1032 start
>> PAGE_SHIFT
,
1033 this_end
>> PAGE_SHIFT
);
1040 static int __init
grab_mlgroups(struct mdesc_handle
*md
)
1042 unsigned long paddr
;
1046 mdesc_for_each_node_by_name(md
, node
, "memory-latency-group")
1051 paddr
= lmb_alloc(count
* sizeof(struct mdesc_mlgroup
),
1056 mlgroups
= __va(paddr
);
1057 num_mlgroups
= count
;
1060 mdesc_for_each_node_by_name(md
, node
, "memory-latency-group") {
1061 struct mdesc_mlgroup
*m
= &mlgroups
[count
++];
1066 val
= mdesc_get_property(md
, node
, "latency", NULL
);
1068 val
= mdesc_get_property(md
, node
, "address-match", NULL
);
1070 val
= mdesc_get_property(md
, node
, "address-mask", NULL
);
1073 numadbg("MLGROUP[%d]: node[%lx] latency[%lx] "
1074 "match[%lx] mask[%lx]\n",
1075 count
- 1, m
->node
, m
->latency
, m
->match
, m
->mask
);
1081 static int __init
grab_mblocks(struct mdesc_handle
*md
)
1083 unsigned long paddr
;
1087 mdesc_for_each_node_by_name(md
, node
, "mblock")
1092 paddr
= lmb_alloc(count
* sizeof(struct mdesc_mblock
),
1097 mblocks
= __va(paddr
);
1098 num_mblocks
= count
;
1101 mdesc_for_each_node_by_name(md
, node
, "mblock") {
1102 struct mdesc_mblock
*m
= &mblocks
[count
++];
1105 val
= mdesc_get_property(md
, node
, "base", NULL
);
1107 val
= mdesc_get_property(md
, node
, "size", NULL
);
1109 val
= mdesc_get_property(md
, node
,
1110 "address-congruence-offset", NULL
);
1113 numadbg("MBLOCK[%d]: base[%lx] size[%lx] offset[%lx]\n",
1114 count
- 1, m
->base
, m
->size
, m
->offset
);
1120 static void __init
numa_parse_mdesc_group_cpus(struct mdesc_handle
*md
,
1121 u64 grp
, cpumask_t
*mask
)
1127 mdesc_for_each_arc(arc
, md
, grp
, MDESC_ARC_TYPE_BACK
) {
1128 u64 target
= mdesc_arc_target(md
, arc
);
1129 const char *name
= mdesc_node_name(md
, target
);
1132 if (strcmp(name
, "cpu"))
1134 id
= mdesc_get_property(md
, target
, "id", NULL
);
1136 cpu_set(*id
, *mask
);
1140 static struct mdesc_mlgroup
* __init
find_mlgroup(u64 node
)
1144 for (i
= 0; i
< num_mlgroups
; i
++) {
1145 struct mdesc_mlgroup
*m
= &mlgroups
[i
];
1146 if (m
->node
== node
)
1152 static int __init
numa_attach_mlgroup(struct mdesc_handle
*md
, u64 grp
,
1155 struct mdesc_mlgroup
*candidate
= NULL
;
1156 u64 arc
, best_latency
= ~(u64
)0;
1157 struct node_mem_mask
*n
;
1159 mdesc_for_each_arc(arc
, md
, grp
, MDESC_ARC_TYPE_FWD
) {
1160 u64 target
= mdesc_arc_target(md
, arc
);
1161 struct mdesc_mlgroup
*m
= find_mlgroup(target
);
1164 if (m
->latency
< best_latency
) {
1166 best_latency
= m
->latency
;
1172 if (num_node_masks
!= index
) {
1173 printk(KERN_ERR
"Inconsistent NUMA state, "
1174 "index[%d] != num_node_masks[%d]\n",
1175 index
, num_node_masks
);
1179 n
= &node_masks
[num_node_masks
++];
1181 n
->mask
= candidate
->mask
;
1182 n
->val
= candidate
->match
;
1184 numadbg("NUMA NODE[%d]: mask[%lx] val[%lx] (latency[%lx])\n",
1185 index
, n
->mask
, n
->val
, candidate
->latency
);
1190 static int __init
numa_parse_mdesc_group(struct mdesc_handle
*md
, u64 grp
,
1196 numa_parse_mdesc_group_cpus(md
, grp
, &mask
);
1198 for_each_cpu_mask(cpu
, mask
)
1199 numa_cpu_lookup_table
[cpu
] = index
;
1200 numa_cpumask_lookup_table
[index
] = mask
;
1203 printk(KERN_INFO
"NUMA GROUP[%d]: cpus [ ", index
);
1204 for_each_cpu_mask(cpu
, mask
)
1209 return numa_attach_mlgroup(md
, grp
, index
);
1212 static int __init
numa_parse_mdesc(void)
1214 struct mdesc_handle
*md
= mdesc_grab();
1218 node
= mdesc_node_by_name(md
, MDESC_NODE_NULL
, "latency-groups");
1219 if (node
== MDESC_NODE_NULL
) {
1224 err
= grab_mblocks(md
);
1228 err
= grab_mlgroups(md
);
1233 mdesc_for_each_node_by_name(md
, node
, "group") {
1234 err
= numa_parse_mdesc_group(md
, node
, count
);
1242 for (i
= 0; i
< num_node_masks
; i
++) {
1243 allocate_node_data(i
);
1253 static int __init
numa_parse_sun4u(void)
1258 static int __init
bootmem_init_numa(void)
1262 numadbg("bootmem_init_numa()\n");
1265 if (tlb_type
== hypervisor
)
1266 err
= numa_parse_mdesc();
1268 err
= numa_parse_sun4u();
1275 static int bootmem_init_numa(void)
1282 static void __init
bootmem_init_nonnuma(void)
1284 unsigned long top_of_ram
= lmb_end_of_DRAM();
1285 unsigned long total_ram
= lmb_phys_mem_size();
1288 numadbg("bootmem_init_nonnuma()\n");
1290 printk(KERN_INFO
"Top of RAM: 0x%lx, Total RAM: 0x%lx\n",
1291 top_of_ram
, total_ram
);
1292 printk(KERN_INFO
"Memory hole size: %ldMB\n",
1293 (top_of_ram
- total_ram
) >> 20);
1295 init_node_masks_nonnuma();
1297 for (i
= 0; i
< lmb
.memory
.cnt
; i
++) {
1298 unsigned long size
= lmb_size_bytes(&lmb
.memory
, i
);
1299 unsigned long start_pfn
, end_pfn
;
1304 start_pfn
= lmb
.memory
.region
[i
].base
>> PAGE_SHIFT
;
1305 end_pfn
= start_pfn
+ lmb_size_pages(&lmb
.memory
, i
);
1306 add_active_range(0, start_pfn
, end_pfn
);
1309 allocate_node_data(0);
1314 static void __init
reserve_range_in_node(int nid
, unsigned long start
,
1317 numadbg(" reserve_range_in_node(nid[%d],start[%lx],end[%lx]\n",
1319 while (start
< end
) {
1320 unsigned long this_end
;
1323 this_end
= nid_range(start
, end
, &n
);
1325 numadbg(" MATCH reserving range [%lx:%lx]\n",
1327 reserve_bootmem_node(NODE_DATA(nid
), start
,
1328 (this_end
- start
), BOOTMEM_DEFAULT
);
1330 numadbg(" NO MATCH, advancing start to %lx\n",
1337 static void __init
trim_reserved_in_node(int nid
)
1341 numadbg(" trim_reserved_in_node(%d)\n", nid
);
1343 for (i
= 0; i
< lmb
.reserved
.cnt
; i
++) {
1344 unsigned long start
= lmb
.reserved
.region
[i
].base
;
1345 unsigned long size
= lmb_size_bytes(&lmb
.reserved
, i
);
1346 unsigned long end
= start
+ size
;
1348 reserve_range_in_node(nid
, start
, end
);
1352 static void __init
bootmem_init_one_node(int nid
)
1354 struct pglist_data
*p
;
1356 numadbg("bootmem_init_one_node(%d)\n", nid
);
1360 if (p
->node_spanned_pages
) {
1361 unsigned long paddr
= node_masks
[nid
].bootmem_paddr
;
1362 unsigned long end_pfn
;
1364 end_pfn
= p
->node_start_pfn
+ p
->node_spanned_pages
;
1366 numadbg(" init_bootmem_node(%d, %lx, %lx, %lx)\n",
1367 nid
, paddr
>> PAGE_SHIFT
, p
->node_start_pfn
, end_pfn
);
1369 init_bootmem_node(p
, paddr
>> PAGE_SHIFT
,
1370 p
->node_start_pfn
, end_pfn
);
1372 numadbg(" free_bootmem_with_active_regions(%d, %lx)\n",
1374 free_bootmem_with_active_regions(nid
, end_pfn
);
1376 trim_reserved_in_node(nid
);
1378 numadbg(" sparse_memory_present_with_active_regions(%d)\n",
1380 sparse_memory_present_with_active_regions(nid
);
1384 static unsigned long __init
bootmem_init(unsigned long phys_base
)
1386 unsigned long end_pfn
;
1389 end_pfn
= lmb_end_of_DRAM() >> PAGE_SHIFT
;
1390 max_pfn
= max_low_pfn
= end_pfn
;
1391 min_low_pfn
= (phys_base
>> PAGE_SHIFT
);
1393 if (bootmem_init_numa() < 0)
1394 bootmem_init_nonnuma();
1396 /* XXX cpu notifier XXX */
1398 for_each_online_node(nid
)
1399 bootmem_init_one_node(nid
);
1406 static struct linux_prom64_registers pall
[MAX_BANKS
] __initdata
;
1407 static int pall_ents __initdata
;
1409 #ifdef CONFIG_DEBUG_PAGEALLOC
1410 static unsigned long __ref
kernel_map_range(unsigned long pstart
,
1411 unsigned long pend
, pgprot_t prot
)
1413 unsigned long vstart
= PAGE_OFFSET
+ pstart
;
1414 unsigned long vend
= PAGE_OFFSET
+ pend
;
1415 unsigned long alloc_bytes
= 0UL;
1417 if ((vstart
& ~PAGE_MASK
) || (vend
& ~PAGE_MASK
)) {
1418 prom_printf("kernel_map: Unaligned physmem[%lx:%lx]\n",
1423 while (vstart
< vend
) {
1424 unsigned long this_end
, paddr
= __pa(vstart
);
1425 pgd_t
*pgd
= pgd_offset_k(vstart
);
1430 pud
= pud_offset(pgd
, vstart
);
1431 if (pud_none(*pud
)) {
1434 new = __alloc_bootmem(PAGE_SIZE
, PAGE_SIZE
, PAGE_SIZE
);
1435 alloc_bytes
+= PAGE_SIZE
;
1436 pud_populate(&init_mm
, pud
, new);
1439 pmd
= pmd_offset(pud
, vstart
);
1440 if (!pmd_present(*pmd
)) {
1443 new = __alloc_bootmem(PAGE_SIZE
, PAGE_SIZE
, PAGE_SIZE
);
1444 alloc_bytes
+= PAGE_SIZE
;
1445 pmd_populate_kernel(&init_mm
, pmd
, new);
1448 pte
= pte_offset_kernel(pmd
, vstart
);
1449 this_end
= (vstart
+ PMD_SIZE
) & PMD_MASK
;
1450 if (this_end
> vend
)
1453 while (vstart
< this_end
) {
1454 pte_val(*pte
) = (paddr
| pgprot_val(prot
));
1456 vstart
+= PAGE_SIZE
;
1465 extern unsigned int kvmap_linear_patch
[1];
1466 #endif /* CONFIG_DEBUG_PAGEALLOC */
1468 static void __init
mark_kpte_bitmap(unsigned long start
, unsigned long end
)
1470 const unsigned long shift_256MB
= 28;
1471 const unsigned long mask_256MB
= ((1UL << shift_256MB
) - 1UL);
1472 const unsigned long size_256MB
= (1UL << shift_256MB
);
1474 while (start
< end
) {
1477 remains
= end
- start
;
1478 if (remains
< size_256MB
)
1481 if (start
& mask_256MB
) {
1482 start
= (start
+ size_256MB
) & ~mask_256MB
;
1486 while (remains
>= size_256MB
) {
1487 unsigned long index
= start
>> shift_256MB
;
1489 __set_bit(index
, kpte_linear_bitmap
);
1491 start
+= size_256MB
;
1492 remains
-= size_256MB
;
1497 static void __init
init_kpte_bitmap(void)
1501 for (i
= 0; i
< pall_ents
; i
++) {
1502 unsigned long phys_start
, phys_end
;
1504 phys_start
= pall
[i
].phys_addr
;
1505 phys_end
= phys_start
+ pall
[i
].reg_size
;
1507 mark_kpte_bitmap(phys_start
, phys_end
);
1511 static void __init
kernel_physical_mapping_init(void)
1513 #ifdef CONFIG_DEBUG_PAGEALLOC
1514 unsigned long i
, mem_alloced
= 0UL;
1516 for (i
= 0; i
< pall_ents
; i
++) {
1517 unsigned long phys_start
, phys_end
;
1519 phys_start
= pall
[i
].phys_addr
;
1520 phys_end
= phys_start
+ pall
[i
].reg_size
;
1522 mem_alloced
+= kernel_map_range(phys_start
, phys_end
,
1526 printk("Allocated %ld bytes for kernel page tables.\n",
1529 kvmap_linear_patch
[0] = 0x01000000; /* nop */
1530 flushi(&kvmap_linear_patch
[0]);
1536 #ifdef CONFIG_DEBUG_PAGEALLOC
1537 void kernel_map_pages(struct page
*page
, int numpages
, int enable
)
1539 unsigned long phys_start
= page_to_pfn(page
) << PAGE_SHIFT
;
1540 unsigned long phys_end
= phys_start
+ (numpages
* PAGE_SIZE
);
1542 kernel_map_range(phys_start
, phys_end
,
1543 (enable
? PAGE_KERNEL
: __pgprot(0)));
1545 flush_tsb_kernel_range(PAGE_OFFSET
+ phys_start
,
1546 PAGE_OFFSET
+ phys_end
);
1548 /* we should perform an IPI and flush all tlbs,
1549 * but that can deadlock->flush only current cpu.
1551 __flush_tlb_kernel_range(PAGE_OFFSET
+ phys_start
,
1552 PAGE_OFFSET
+ phys_end
);
1556 unsigned long __init
find_ecache_flush_span(unsigned long size
)
1560 for (i
= 0; i
< pavail_ents
; i
++) {
1561 if (pavail
[i
].reg_size
>= size
)
1562 return pavail
[i
].phys_addr
;
1568 static void __init
tsb_phys_patch(void)
1570 struct tsb_ldquad_phys_patch_entry
*pquad
;
1571 struct tsb_phys_patch_entry
*p
;
1573 pquad
= &__tsb_ldquad_phys_patch
;
1574 while (pquad
< &__tsb_ldquad_phys_patch_end
) {
1575 unsigned long addr
= pquad
->addr
;
1577 if (tlb_type
== hypervisor
)
1578 *(unsigned int *) addr
= pquad
->sun4v_insn
;
1580 *(unsigned int *) addr
= pquad
->sun4u_insn
;
1582 __asm__
__volatile__("flush %0"
1589 p
= &__tsb_phys_patch
;
1590 while (p
< &__tsb_phys_patch_end
) {
1591 unsigned long addr
= p
->addr
;
1593 *(unsigned int *) addr
= p
->insn
;
1595 __asm__
__volatile__("flush %0"
1603 /* Don't mark as init, we give this to the Hypervisor. */
1604 #ifndef CONFIG_DEBUG_PAGEALLOC
1605 #define NUM_KTSB_DESCR 2
1607 #define NUM_KTSB_DESCR 1
1609 static struct hv_tsb_descr ktsb_descr
[NUM_KTSB_DESCR
];
1610 extern struct tsb swapper_tsb
[KERNEL_TSB_NENTRIES
];
1612 static void __init
sun4v_ktsb_init(void)
1614 unsigned long ktsb_pa
;
1616 /* First KTSB for PAGE_SIZE mappings. */
1617 ktsb_pa
= kern_base
+ ((unsigned long)&swapper_tsb
[0] - KERNBASE
);
1619 switch (PAGE_SIZE
) {
1622 ktsb_descr
[0].pgsz_idx
= HV_PGSZ_IDX_8K
;
1623 ktsb_descr
[0].pgsz_mask
= HV_PGSZ_MASK_8K
;
1627 ktsb_descr
[0].pgsz_idx
= HV_PGSZ_IDX_64K
;
1628 ktsb_descr
[0].pgsz_mask
= HV_PGSZ_MASK_64K
;
1632 ktsb_descr
[0].pgsz_idx
= HV_PGSZ_IDX_512K
;
1633 ktsb_descr
[0].pgsz_mask
= HV_PGSZ_MASK_512K
;
1636 case 4 * 1024 * 1024:
1637 ktsb_descr
[0].pgsz_idx
= HV_PGSZ_IDX_4MB
;
1638 ktsb_descr
[0].pgsz_mask
= HV_PGSZ_MASK_4MB
;
1642 ktsb_descr
[0].assoc
= 1;
1643 ktsb_descr
[0].num_ttes
= KERNEL_TSB_NENTRIES
;
1644 ktsb_descr
[0].ctx_idx
= 0;
1645 ktsb_descr
[0].tsb_base
= ktsb_pa
;
1646 ktsb_descr
[0].resv
= 0;
1648 #ifndef CONFIG_DEBUG_PAGEALLOC
1649 /* Second KTSB for 4MB/256MB mappings. */
1650 ktsb_pa
= (kern_base
+
1651 ((unsigned long)&swapper_4m_tsb
[0] - KERNBASE
));
1653 ktsb_descr
[1].pgsz_idx
= HV_PGSZ_IDX_4MB
;
1654 ktsb_descr
[1].pgsz_mask
= (HV_PGSZ_MASK_4MB
|
1655 HV_PGSZ_MASK_256MB
);
1656 ktsb_descr
[1].assoc
= 1;
1657 ktsb_descr
[1].num_ttes
= KERNEL_TSB4M_NENTRIES
;
1658 ktsb_descr
[1].ctx_idx
= 0;
1659 ktsb_descr
[1].tsb_base
= ktsb_pa
;
1660 ktsb_descr
[1].resv
= 0;
1664 void __cpuinit
sun4v_ktsb_register(void)
1666 unsigned long pa
, ret
;
1668 pa
= kern_base
+ ((unsigned long)&ktsb_descr
[0] - KERNBASE
);
1670 ret
= sun4v_mmu_tsb_ctx0(NUM_KTSB_DESCR
, pa
);
1672 prom_printf("hypervisor_mmu_tsb_ctx0[%lx]: "
1673 "errors with %lx\n", pa
, ret
);
1678 /* paging_init() sets up the page tables */
1680 extern void central_probe(void);
1682 static unsigned long last_valid_pfn
;
1683 pgd_t swapper_pg_dir
[2048];
1685 static void sun4u_pgprot_init(void);
1686 static void sun4v_pgprot_init(void);
1688 /* Dummy function */
1689 void __init
setup_per_cpu_areas(void)
1693 void __init
paging_init(void)
1695 unsigned long end_pfn
, shift
, phys_base
;
1696 unsigned long real_end
, i
;
1698 /* These build time checkes make sure that the dcache_dirty_cpu()
1699 * page->flags usage will work.
1701 * When a page gets marked as dcache-dirty, we store the
1702 * cpu number starting at bit 32 in the page->flags. Also,
1703 * functions like clear_dcache_dirty_cpu use the cpu mask
1704 * in 13-bit signed-immediate instruction fields.
1708 * Page flags must not reach into upper 32 bits that are used
1709 * for the cpu number
1711 BUILD_BUG_ON(NR_PAGEFLAGS
> 32);
1714 * The bit fields placed in the high range must not reach below
1715 * the 32 bit boundary. Otherwise we cannot place the cpu field
1716 * at the 32 bit boundary.
1718 BUILD_BUG_ON(SECTIONS_WIDTH
+ NODES_WIDTH
+ ZONES_WIDTH
+
1719 ilog2(roundup_pow_of_two(NR_CPUS
)) > 32);
1721 BUILD_BUG_ON(NR_CPUS
> 4096);
1723 kern_base
= (prom_boot_mapping_phys_low
>> 22UL) << 22UL;
1724 kern_size
= (unsigned long)&_end
- (unsigned long)KERNBASE
;
1728 /* Invalidate both kernel TSBs. */
1729 memset(swapper_tsb
, 0x40, sizeof(swapper_tsb
));
1730 #ifndef CONFIG_DEBUG_PAGEALLOC
1731 memset(swapper_4m_tsb
, 0x40, sizeof(swapper_4m_tsb
));
1734 if (tlb_type
== hypervisor
)
1735 sun4v_pgprot_init();
1737 sun4u_pgprot_init();
1739 if (tlb_type
== cheetah_plus
||
1740 tlb_type
== hypervisor
)
1743 if (tlb_type
== hypervisor
) {
1744 sun4v_patch_tlb_handlers();
1750 /* Find available physical memory... */
1751 read_obp_memory("available", &pavail
[0], &pavail_ents
);
1753 phys_base
= 0xffffffffffffffffUL
;
1754 for (i
= 0; i
< pavail_ents
; i
++) {
1755 phys_base
= min(phys_base
, pavail
[i
].phys_addr
);
1756 lmb_add(pavail
[i
].phys_addr
, pavail
[i
].reg_size
);
1759 lmb_reserve(kern_base
, kern_size
);
1761 find_ramdisk(phys_base
);
1763 if (cmdline_memory_size
)
1764 lmb_enforce_memory_limit(phys_base
+ cmdline_memory_size
);
1769 set_bit(0, mmu_context_bmap
);
1771 shift
= kern_base
+ PAGE_OFFSET
- ((unsigned long)KERNBASE
);
1773 real_end
= (unsigned long)_end
;
1774 num_kernel_image_mappings
= DIV_ROUND_UP(real_end
- KERNBASE
, 1 << 22);
1775 printk("Kernel: Using %d locked TLB entries for main kernel image.\n",
1776 num_kernel_image_mappings
);
1778 /* Set kernel pgd to upper alias so physical page computations
1781 init_mm
.pgd
+= ((shift
) / (sizeof(pgd_t
)));
1783 memset(swapper_low_pmd_dir
, 0, sizeof(swapper_low_pmd_dir
));
1785 /* Now can init the kernel/bad page tables. */
1786 pud_set(pud_offset(&swapper_pg_dir
[0], 0),
1787 swapper_low_pmd_dir
+ (shift
/ sizeof(pgd_t
)));
1789 inherit_prom_mappings();
1791 read_obp_memory("reg", &pall
[0], &pall_ents
);
1795 /* Ok, we can use our TLB miss and window trap handlers safely. */
1800 if (tlb_type
== hypervisor
)
1801 sun4v_ktsb_register();
1803 /* We must setup the per-cpu areas before we pull in the
1804 * PROM and the MDESC. The code there fills in cpu and
1805 * other information into per-cpu data structures.
1807 real_setup_per_cpu_areas();
1809 prom_build_devicetree();
1811 if (tlb_type
== hypervisor
)
1814 /* Setup bootmem... */
1815 last_valid_pfn
= end_pfn
= bootmem_init(phys_base
);
1817 #ifndef CONFIG_NEED_MULTIPLE_NODES
1818 max_mapnr
= last_valid_pfn
;
1820 kernel_physical_mapping_init();
1823 unsigned long max_zone_pfns
[MAX_NR_ZONES
];
1825 memset(max_zone_pfns
, 0, sizeof(max_zone_pfns
));
1827 max_zone_pfns
[ZONE_NORMAL
] = end_pfn
;
1829 free_area_init_nodes(max_zone_pfns
);
1832 printk("Booting Linux...\n");
1838 int __init
page_in_phys_avail(unsigned long paddr
)
1844 for (i
= 0; i
< pavail_ents
; i
++) {
1845 unsigned long start
, end
;
1847 start
= pavail
[i
].phys_addr
;
1848 end
= start
+ pavail
[i
].reg_size
;
1850 if (paddr
>= start
&& paddr
< end
)
1853 if (paddr
>= kern_base
&& paddr
< (kern_base
+ kern_size
))
1855 #ifdef CONFIG_BLK_DEV_INITRD
1856 if (paddr
>= __pa(initrd_start
) &&
1857 paddr
< __pa(PAGE_ALIGN(initrd_end
)))
1864 static struct linux_prom64_registers pavail_rescan
[MAX_BANKS
] __initdata
;
1865 static int pavail_rescan_ents __initdata
;
1867 /* Certain OBP calls, such as fetching "available" properties, can
1868 * claim physical memory. So, along with initializing the valid
1869 * address bitmap, what we do here is refetch the physical available
1870 * memory list again, and make sure it provides at least as much
1871 * memory as 'pavail' does.
1873 static void setup_valid_addr_bitmap_from_pavail(void)
1877 read_obp_memory("available", &pavail_rescan
[0], &pavail_rescan_ents
);
1879 for (i
= 0; i
< pavail_ents
; i
++) {
1880 unsigned long old_start
, old_end
;
1882 old_start
= pavail
[i
].phys_addr
;
1883 old_end
= old_start
+ pavail
[i
].reg_size
;
1884 while (old_start
< old_end
) {
1887 for (n
= 0; n
< pavail_rescan_ents
; n
++) {
1888 unsigned long new_start
, new_end
;
1890 new_start
= pavail_rescan
[n
].phys_addr
;
1891 new_end
= new_start
+
1892 pavail_rescan
[n
].reg_size
;
1894 if (new_start
<= old_start
&&
1895 new_end
>= (old_start
+ PAGE_SIZE
)) {
1896 set_bit(old_start
>> 22,
1897 sparc64_valid_addr_bitmap
);
1902 prom_printf("mem_init: Lost memory in pavail\n");
1903 prom_printf("mem_init: OLD start[%lx] size[%lx]\n",
1904 pavail
[i
].phys_addr
,
1905 pavail
[i
].reg_size
);
1906 prom_printf("mem_init: NEW start[%lx] size[%lx]\n",
1907 pavail_rescan
[i
].phys_addr
,
1908 pavail_rescan
[i
].reg_size
);
1909 prom_printf("mem_init: Cannot continue, aborting.\n");
1913 old_start
+= PAGE_SIZE
;
1918 void __init
mem_init(void)
1920 unsigned long codepages
, datapages
, initpages
;
1921 unsigned long addr
, last
;
1924 i
= last_valid_pfn
>> ((22 - PAGE_SHIFT
) + 6);
1926 sparc64_valid_addr_bitmap
= (unsigned long *) alloc_bootmem(i
<< 3);
1927 if (sparc64_valid_addr_bitmap
== NULL
) {
1928 prom_printf("mem_init: Cannot alloc valid_addr_bitmap.\n");
1931 memset(sparc64_valid_addr_bitmap
, 0, i
<< 3);
1933 addr
= PAGE_OFFSET
+ kern_base
;
1934 last
= PAGE_ALIGN(kern_size
) + addr
;
1935 while (addr
< last
) {
1936 set_bit(__pa(addr
) >> 22, sparc64_valid_addr_bitmap
);
1940 setup_valid_addr_bitmap_from_pavail();
1942 high_memory
= __va(last_valid_pfn
<< PAGE_SHIFT
);
1944 #ifdef CONFIG_NEED_MULTIPLE_NODES
1945 for_each_online_node(i
) {
1946 if (NODE_DATA(i
)->node_spanned_pages
!= 0) {
1948 free_all_bootmem_node(NODE_DATA(i
));
1952 totalram_pages
= free_all_bootmem();
1955 /* We subtract one to account for the mem_map_zero page
1958 totalram_pages
-= 1;
1959 num_physpages
= totalram_pages
;
1962 * Set up the zero page, mark it reserved, so that page count
1963 * is not manipulated when freeing the page from user ptes.
1965 mem_map_zero
= alloc_pages(GFP_KERNEL
|__GFP_ZERO
, 0);
1966 if (mem_map_zero
== NULL
) {
1967 prom_printf("paging_init: Cannot alloc zero page.\n");
1970 SetPageReserved(mem_map_zero
);
1972 codepages
= (((unsigned long) _etext
) - ((unsigned long) _start
));
1973 codepages
= PAGE_ALIGN(codepages
) >> PAGE_SHIFT
;
1974 datapages
= (((unsigned long) _edata
) - ((unsigned long) _etext
));
1975 datapages
= PAGE_ALIGN(datapages
) >> PAGE_SHIFT
;
1976 initpages
= (((unsigned long) __init_end
) - ((unsigned long) __init_begin
));
1977 initpages
= PAGE_ALIGN(initpages
) >> PAGE_SHIFT
;
1979 printk("Memory: %luk available (%ldk kernel code, %ldk data, %ldk init) [%016lx,%016lx]\n",
1980 nr_free_pages() << (PAGE_SHIFT
-10),
1981 codepages
<< (PAGE_SHIFT
-10),
1982 datapages
<< (PAGE_SHIFT
-10),
1983 initpages
<< (PAGE_SHIFT
-10),
1984 PAGE_OFFSET
, (last_valid_pfn
<< PAGE_SHIFT
));
1986 if (tlb_type
== cheetah
|| tlb_type
== cheetah_plus
)
1987 cheetah_ecache_flush_init();
1990 void free_initmem(void)
1992 unsigned long addr
, initend
;
1995 * The init section is aligned to 8k in vmlinux.lds. Page align for >8k pagesizes.
1997 addr
= PAGE_ALIGN((unsigned long)(__init_begin
));
1998 initend
= (unsigned long)(__init_end
) & PAGE_MASK
;
1999 for (; addr
< initend
; addr
+= PAGE_SIZE
) {
2004 ((unsigned long) __va(kern_base
)) -
2005 ((unsigned long) KERNBASE
));
2006 memset((void *)addr
, POISON_FREE_INITMEM
, PAGE_SIZE
);
2007 p
= virt_to_page(page
);
2009 ClearPageReserved(p
);
2017 #ifdef CONFIG_BLK_DEV_INITRD
2018 void free_initrd_mem(unsigned long start
, unsigned long end
)
2021 printk ("Freeing initrd memory: %ldk freed\n", (end
- start
) >> 10);
2022 for (; start
< end
; start
+= PAGE_SIZE
) {
2023 struct page
*p
= virt_to_page(start
);
2025 ClearPageReserved(p
);
2034 #define _PAGE_CACHE_4U (_PAGE_CP_4U | _PAGE_CV_4U)
2035 #define _PAGE_CACHE_4V (_PAGE_CP_4V | _PAGE_CV_4V)
2036 #define __DIRTY_BITS_4U (_PAGE_MODIFIED_4U | _PAGE_WRITE_4U | _PAGE_W_4U)
2037 #define __DIRTY_BITS_4V (_PAGE_MODIFIED_4V | _PAGE_WRITE_4V | _PAGE_W_4V)
2038 #define __ACCESS_BITS_4U (_PAGE_ACCESSED_4U | _PAGE_READ_4U | _PAGE_R)
2039 #define __ACCESS_BITS_4V (_PAGE_ACCESSED_4V | _PAGE_READ_4V | _PAGE_R)
2041 pgprot_t PAGE_KERNEL __read_mostly
;
2042 EXPORT_SYMBOL(PAGE_KERNEL
);
2044 pgprot_t PAGE_KERNEL_LOCKED __read_mostly
;
2045 pgprot_t PAGE_COPY __read_mostly
;
2047 pgprot_t PAGE_SHARED __read_mostly
;
2048 EXPORT_SYMBOL(PAGE_SHARED
);
2050 pgprot_t PAGE_EXEC __read_mostly
;
2051 unsigned long pg_iobits __read_mostly
;
2053 unsigned long _PAGE_IE __read_mostly
;
2054 EXPORT_SYMBOL(_PAGE_IE
);
2056 unsigned long _PAGE_E __read_mostly
;
2057 EXPORT_SYMBOL(_PAGE_E
);
2059 unsigned long _PAGE_CACHE __read_mostly
;
2060 EXPORT_SYMBOL(_PAGE_CACHE
);
2062 #ifdef CONFIG_SPARSEMEM_VMEMMAP
2064 #define VMEMMAP_CHUNK_SHIFT 22
2065 #define VMEMMAP_CHUNK (1UL << VMEMMAP_CHUNK_SHIFT)
2066 #define VMEMMAP_CHUNK_MASK ~(VMEMMAP_CHUNK - 1UL)
2067 #define VMEMMAP_ALIGN(x) (((x)+VMEMMAP_CHUNK-1UL)&VMEMMAP_CHUNK_MASK)
2069 #define VMEMMAP_SIZE ((((1UL << MAX_PHYSADDR_BITS) >> PAGE_SHIFT) * \
2070 sizeof(struct page *)) >> VMEMMAP_CHUNK_SHIFT)
2071 unsigned long vmemmap_table
[VMEMMAP_SIZE
];
2073 int __meminit
vmemmap_populate(struct page
*start
, unsigned long nr
, int node
)
2075 unsigned long vstart
= (unsigned long) start
;
2076 unsigned long vend
= (unsigned long) (start
+ nr
);
2077 unsigned long phys_start
= (vstart
- VMEMMAP_BASE
);
2078 unsigned long phys_end
= (vend
- VMEMMAP_BASE
);
2079 unsigned long addr
= phys_start
& VMEMMAP_CHUNK_MASK
;
2080 unsigned long end
= VMEMMAP_ALIGN(phys_end
);
2081 unsigned long pte_base
;
2083 pte_base
= (_PAGE_VALID
| _PAGE_SZ4MB_4U
|
2084 _PAGE_CP_4U
| _PAGE_CV_4U
|
2085 _PAGE_P_4U
| _PAGE_W_4U
);
2086 if (tlb_type
== hypervisor
)
2087 pte_base
= (_PAGE_VALID
| _PAGE_SZ4MB_4V
|
2088 _PAGE_CP_4V
| _PAGE_CV_4V
|
2089 _PAGE_P_4V
| _PAGE_W_4V
);
2091 for (; addr
< end
; addr
+= VMEMMAP_CHUNK
) {
2092 unsigned long *vmem_pp
=
2093 vmemmap_table
+ (addr
>> VMEMMAP_CHUNK_SHIFT
);
2096 if (!(*vmem_pp
& _PAGE_VALID
)) {
2097 block
= vmemmap_alloc_block(1UL << 22, node
);
2101 *vmem_pp
= pte_base
| __pa(block
);
2103 printk(KERN_INFO
"[%p-%p] page_structs=%lu "
2104 "node=%d entry=%lu/%lu\n", start
, block
, nr
,
2106 addr
>> VMEMMAP_CHUNK_SHIFT
,
2107 VMEMMAP_SIZE
>> VMEMMAP_CHUNK_SHIFT
);
2112 #endif /* CONFIG_SPARSEMEM_VMEMMAP */
2114 static void prot_init_common(unsigned long page_none
,
2115 unsigned long page_shared
,
2116 unsigned long page_copy
,
2117 unsigned long page_readonly
,
2118 unsigned long page_exec_bit
)
2120 PAGE_COPY
= __pgprot(page_copy
);
2121 PAGE_SHARED
= __pgprot(page_shared
);
2123 protection_map
[0x0] = __pgprot(page_none
);
2124 protection_map
[0x1] = __pgprot(page_readonly
& ~page_exec_bit
);
2125 protection_map
[0x2] = __pgprot(page_copy
& ~page_exec_bit
);
2126 protection_map
[0x3] = __pgprot(page_copy
& ~page_exec_bit
);
2127 protection_map
[0x4] = __pgprot(page_readonly
);
2128 protection_map
[0x5] = __pgprot(page_readonly
);
2129 protection_map
[0x6] = __pgprot(page_copy
);
2130 protection_map
[0x7] = __pgprot(page_copy
);
2131 protection_map
[0x8] = __pgprot(page_none
);
2132 protection_map
[0x9] = __pgprot(page_readonly
& ~page_exec_bit
);
2133 protection_map
[0xa] = __pgprot(page_shared
& ~page_exec_bit
);
2134 protection_map
[0xb] = __pgprot(page_shared
& ~page_exec_bit
);
2135 protection_map
[0xc] = __pgprot(page_readonly
);
2136 protection_map
[0xd] = __pgprot(page_readonly
);
2137 protection_map
[0xe] = __pgprot(page_shared
);
2138 protection_map
[0xf] = __pgprot(page_shared
);
2141 static void __init
sun4u_pgprot_init(void)
2143 unsigned long page_none
, page_shared
, page_copy
, page_readonly
;
2144 unsigned long page_exec_bit
;
2146 PAGE_KERNEL
= __pgprot (_PAGE_PRESENT_4U
| _PAGE_VALID
|
2147 _PAGE_CACHE_4U
| _PAGE_P_4U
|
2148 __ACCESS_BITS_4U
| __DIRTY_BITS_4U
|
2150 PAGE_KERNEL_LOCKED
= __pgprot (_PAGE_PRESENT_4U
| _PAGE_VALID
|
2151 _PAGE_CACHE_4U
| _PAGE_P_4U
|
2152 __ACCESS_BITS_4U
| __DIRTY_BITS_4U
|
2153 _PAGE_EXEC_4U
| _PAGE_L_4U
);
2154 PAGE_EXEC
= __pgprot(_PAGE_EXEC_4U
);
2156 _PAGE_IE
= _PAGE_IE_4U
;
2157 _PAGE_E
= _PAGE_E_4U
;
2158 _PAGE_CACHE
= _PAGE_CACHE_4U
;
2160 pg_iobits
= (_PAGE_VALID
| _PAGE_PRESENT_4U
| __DIRTY_BITS_4U
|
2161 __ACCESS_BITS_4U
| _PAGE_E_4U
);
2163 #ifdef CONFIG_DEBUG_PAGEALLOC
2164 kern_linear_pte_xor
[0] = (_PAGE_VALID
| _PAGE_SZBITS_4U
) ^
2167 kern_linear_pte_xor
[0] = (_PAGE_VALID
| _PAGE_SZ4MB_4U
) ^
2170 kern_linear_pte_xor
[0] |= (_PAGE_CP_4U
| _PAGE_CV_4U
|
2171 _PAGE_P_4U
| _PAGE_W_4U
);
2173 /* XXX Should use 256MB on Panther. XXX */
2174 kern_linear_pte_xor
[1] = kern_linear_pte_xor
[0];
2176 _PAGE_SZBITS
= _PAGE_SZBITS_4U
;
2177 _PAGE_ALL_SZ_BITS
= (_PAGE_SZ4MB_4U
| _PAGE_SZ512K_4U
|
2178 _PAGE_SZ64K_4U
| _PAGE_SZ8K_4U
|
2179 _PAGE_SZ32MB_4U
| _PAGE_SZ256MB_4U
);
2182 page_none
= _PAGE_PRESENT_4U
| _PAGE_ACCESSED_4U
| _PAGE_CACHE_4U
;
2183 page_shared
= (_PAGE_VALID
| _PAGE_PRESENT_4U
| _PAGE_CACHE_4U
|
2184 __ACCESS_BITS_4U
| _PAGE_WRITE_4U
| _PAGE_EXEC_4U
);
2185 page_copy
= (_PAGE_VALID
| _PAGE_PRESENT_4U
| _PAGE_CACHE_4U
|
2186 __ACCESS_BITS_4U
| _PAGE_EXEC_4U
);
2187 page_readonly
= (_PAGE_VALID
| _PAGE_PRESENT_4U
| _PAGE_CACHE_4U
|
2188 __ACCESS_BITS_4U
| _PAGE_EXEC_4U
);
2190 page_exec_bit
= _PAGE_EXEC_4U
;
2192 prot_init_common(page_none
, page_shared
, page_copy
, page_readonly
,
2196 static void __init
sun4v_pgprot_init(void)
2198 unsigned long page_none
, page_shared
, page_copy
, page_readonly
;
2199 unsigned long page_exec_bit
;
2201 PAGE_KERNEL
= __pgprot (_PAGE_PRESENT_4V
| _PAGE_VALID
|
2202 _PAGE_CACHE_4V
| _PAGE_P_4V
|
2203 __ACCESS_BITS_4V
| __DIRTY_BITS_4V
|
2205 PAGE_KERNEL_LOCKED
= PAGE_KERNEL
;
2206 PAGE_EXEC
= __pgprot(_PAGE_EXEC_4V
);
2208 _PAGE_IE
= _PAGE_IE_4V
;
2209 _PAGE_E
= _PAGE_E_4V
;
2210 _PAGE_CACHE
= _PAGE_CACHE_4V
;
2212 #ifdef CONFIG_DEBUG_PAGEALLOC
2213 kern_linear_pte_xor
[0] = (_PAGE_VALID
| _PAGE_SZBITS_4V
) ^
2216 kern_linear_pte_xor
[0] = (_PAGE_VALID
| _PAGE_SZ4MB_4V
) ^
2219 kern_linear_pte_xor
[0] |= (_PAGE_CP_4V
| _PAGE_CV_4V
|
2220 _PAGE_P_4V
| _PAGE_W_4V
);
2222 #ifdef CONFIG_DEBUG_PAGEALLOC
2223 kern_linear_pte_xor
[1] = (_PAGE_VALID
| _PAGE_SZBITS_4V
) ^
2226 kern_linear_pte_xor
[1] = (_PAGE_VALID
| _PAGE_SZ256MB_4V
) ^
2229 kern_linear_pte_xor
[1] |= (_PAGE_CP_4V
| _PAGE_CV_4V
|
2230 _PAGE_P_4V
| _PAGE_W_4V
);
2232 pg_iobits
= (_PAGE_VALID
| _PAGE_PRESENT_4V
| __DIRTY_BITS_4V
|
2233 __ACCESS_BITS_4V
| _PAGE_E_4V
);
2235 _PAGE_SZBITS
= _PAGE_SZBITS_4V
;
2236 _PAGE_ALL_SZ_BITS
= (_PAGE_SZ16GB_4V
| _PAGE_SZ2GB_4V
|
2237 _PAGE_SZ256MB_4V
| _PAGE_SZ32MB_4V
|
2238 _PAGE_SZ4MB_4V
| _PAGE_SZ512K_4V
|
2239 _PAGE_SZ64K_4V
| _PAGE_SZ8K_4V
);
2241 page_none
= _PAGE_PRESENT_4V
| _PAGE_ACCESSED_4V
| _PAGE_CACHE_4V
;
2242 page_shared
= (_PAGE_VALID
| _PAGE_PRESENT_4V
| _PAGE_CACHE_4V
|
2243 __ACCESS_BITS_4V
| _PAGE_WRITE_4V
| _PAGE_EXEC_4V
);
2244 page_copy
= (_PAGE_VALID
| _PAGE_PRESENT_4V
| _PAGE_CACHE_4V
|
2245 __ACCESS_BITS_4V
| _PAGE_EXEC_4V
);
2246 page_readonly
= (_PAGE_VALID
| _PAGE_PRESENT_4V
| _PAGE_CACHE_4V
|
2247 __ACCESS_BITS_4V
| _PAGE_EXEC_4V
);
2249 page_exec_bit
= _PAGE_EXEC_4V
;
2251 prot_init_common(page_none
, page_shared
, page_copy
, page_readonly
,
2255 unsigned long pte_sz_bits(unsigned long sz
)
2257 if (tlb_type
== hypervisor
) {
2261 return _PAGE_SZ8K_4V
;
2263 return _PAGE_SZ64K_4V
;
2265 return _PAGE_SZ512K_4V
;
2266 case 4 * 1024 * 1024:
2267 return _PAGE_SZ4MB_4V
;
2273 return _PAGE_SZ8K_4U
;
2275 return _PAGE_SZ64K_4U
;
2277 return _PAGE_SZ512K_4U
;
2278 case 4 * 1024 * 1024:
2279 return _PAGE_SZ4MB_4U
;
2284 pte_t
mk_pte_io(unsigned long page
, pgprot_t prot
, int space
, unsigned long page_size
)
2288 pte_val(pte
) = page
| pgprot_val(pgprot_noncached(prot
));
2289 pte_val(pte
) |= (((unsigned long)space
) << 32);
2290 pte_val(pte
) |= pte_sz_bits(page_size
);
2295 static unsigned long kern_large_tte(unsigned long paddr
)
2299 val
= (_PAGE_VALID
| _PAGE_SZ4MB_4U
|
2300 _PAGE_CP_4U
| _PAGE_CV_4U
| _PAGE_P_4U
|
2301 _PAGE_EXEC_4U
| _PAGE_L_4U
| _PAGE_W_4U
);
2302 if (tlb_type
== hypervisor
)
2303 val
= (_PAGE_VALID
| _PAGE_SZ4MB_4V
|
2304 _PAGE_CP_4V
| _PAGE_CV_4V
| _PAGE_P_4V
|
2305 _PAGE_EXEC_4V
| _PAGE_W_4V
);
2310 /* If not locked, zap it. */
2311 void __flush_tlb_all(void)
2313 unsigned long pstate
;
2316 __asm__
__volatile__("flushw\n\t"
2317 "rdpr %%pstate, %0\n\t"
2318 "wrpr %0, %1, %%pstate"
2321 if (tlb_type
== hypervisor
) {
2322 sun4v_mmu_demap_all();
2323 } else if (tlb_type
== spitfire
) {
2324 for (i
= 0; i
< 64; i
++) {
2325 /* Spitfire Errata #32 workaround */
2326 /* NOTE: Always runs on spitfire, so no
2327 * cheetah+ page size encodings.
2329 __asm__
__volatile__("stxa %0, [%1] %2\n\t"
2333 "r" (PRIMARY_CONTEXT
), "i" (ASI_DMMU
));
2335 if (!(spitfire_get_dtlb_data(i
) & _PAGE_L_4U
)) {
2336 __asm__
__volatile__("stxa %%g0, [%0] %1\n\t"
2339 : "r" (TLB_TAG_ACCESS
), "i" (ASI_DMMU
));
2340 spitfire_put_dtlb_data(i
, 0x0UL
);
2343 /* Spitfire Errata #32 workaround */
2344 /* NOTE: Always runs on spitfire, so no
2345 * cheetah+ page size encodings.
2347 __asm__
__volatile__("stxa %0, [%1] %2\n\t"
2351 "r" (PRIMARY_CONTEXT
), "i" (ASI_DMMU
));
2353 if (!(spitfire_get_itlb_data(i
) & _PAGE_L_4U
)) {
2354 __asm__
__volatile__("stxa %%g0, [%0] %1\n\t"
2357 : "r" (TLB_TAG_ACCESS
), "i" (ASI_IMMU
));
2358 spitfire_put_itlb_data(i
, 0x0UL
);
2361 } else if (tlb_type
== cheetah
|| tlb_type
== cheetah_plus
) {
2362 cheetah_flush_dtlb_all();
2363 cheetah_flush_itlb_all();
2365 __asm__
__volatile__("wrpr %0, 0, %%pstate"