2 * Initialize MMU support.
4 * Copyright (C) 1998-2003 Hewlett-Packard Co
5 * David Mosberger-Tang <davidm@hpl.hp.com>
7 #include <linux/kernel.h>
8 #include <linux/init.h>
10 #include <linux/bootmem.h>
11 #include <linux/efi.h>
12 #include <linux/elf.h>
13 #include <linux/memblock.h>
15 #include <linux/mmzone.h>
16 #include <linux/module.h>
17 #include <linux/personality.h>
18 #include <linux/reboot.h>
19 #include <linux/slab.h>
20 #include <linux/swap.h>
21 #include <linux/proc_fs.h>
22 #include <linux/bitops.h>
23 #include <linux/kexec.h>
27 #include <asm/machvec.h>
29 #include <asm/patch.h>
30 #include <asm/pgalloc.h>
32 #include <asm/sections.h>
34 #include <asm/uaccess.h>
35 #include <asm/unistd.h>
38 extern void ia64_tlb_init (void);
40 unsigned long MAX_DMA_ADDRESS
= PAGE_OFFSET
+ 0x100000000UL
;
42 #ifdef CONFIG_VIRTUAL_MEM_MAP
43 unsigned long VMALLOC_END
= VMALLOC_END_INIT
;
44 EXPORT_SYMBOL(VMALLOC_END
);
45 struct page
*vmem_map
;
46 EXPORT_SYMBOL(vmem_map
);
49 struct page
*zero_page_memmap_ptr
; /* map entry for zero page */
50 EXPORT_SYMBOL(zero_page_memmap_ptr
);
53 __ia64_sync_icache_dcache (pte_t pte
)
59 addr
= (unsigned long) page_address(page
);
61 if (test_bit(PG_arch_1
, &page
->flags
))
62 return; /* i-cache is already coherent with d-cache */
64 flush_icache_range(addr
, addr
+ (PAGE_SIZE
<< compound_order(page
)));
65 set_bit(PG_arch_1
, &page
->flags
); /* mark page as clean */
69 * Since DMA is i-cache coherent, any (complete) pages that were written via
70 * DMA can be marked as "clean" so that lazy_mmu_prot_update() doesn't have to
71 * flush them when they get mapped into an executable vm-area.
74 dma_mark_clean(void *addr
, size_t size
)
76 unsigned long pg_addr
, end
;
78 pg_addr
= PAGE_ALIGN((unsigned long) addr
);
79 end
= (unsigned long) addr
+ size
;
80 while (pg_addr
+ PAGE_SIZE
<= end
) {
81 struct page
*page
= virt_to_page(pg_addr
);
82 set_bit(PG_arch_1
, &page
->flags
);
88 ia64_set_rbs_bot (void)
90 unsigned long stack_size
= rlimit_max(RLIMIT_STACK
) & -16;
92 if (stack_size
> MAX_USER_STACK_SIZE
)
93 stack_size
= MAX_USER_STACK_SIZE
;
94 current
->thread
.rbs_bot
= PAGE_ALIGN(current
->mm
->start_stack
- stack_size
);
98 * This performs some platform-dependent address space initialization.
99 * On IA-64, we want to setup the VM area for the register backing
100 * store (which grows upwards) and install the gateway page which is
101 * used for signal trampolines, etc.
104 ia64_init_addr_space (void)
106 struct vm_area_struct
*vma
;
111 * If we're out of memory and kmem_cache_alloc() returns NULL, we simply ignore
112 * the problem. When the process attempts to write to the register backing store
113 * for the first time, it will get a SEGFAULT in this case.
115 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
117 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
118 vma
->vm_mm
= current
->mm
;
119 vma
->vm_start
= current
->thread
.rbs_bot
& PAGE_MASK
;
120 vma
->vm_end
= vma
->vm_start
+ PAGE_SIZE
;
121 vma
->vm_flags
= VM_DATA_DEFAULT_FLAGS
|VM_GROWSUP
|VM_ACCOUNT
;
122 vma
->vm_page_prot
= vm_get_page_prot(vma
->vm_flags
);
123 down_write(¤t
->mm
->mmap_sem
);
124 if (insert_vm_struct(current
->mm
, vma
)) {
125 up_write(¤t
->mm
->mmap_sem
);
126 kmem_cache_free(vm_area_cachep
, vma
);
129 up_write(¤t
->mm
->mmap_sem
);
132 /* map NaT-page at address zero to speed up speculative dereferencing of NULL: */
133 if (!(current
->personality
& MMAP_PAGE_ZERO
)) {
134 vma
= kmem_cache_zalloc(vm_area_cachep
, GFP_KERNEL
);
136 INIT_LIST_HEAD(&vma
->anon_vma_chain
);
137 vma
->vm_mm
= current
->mm
;
138 vma
->vm_end
= PAGE_SIZE
;
139 vma
->vm_page_prot
= __pgprot(pgprot_val(PAGE_READONLY
) | _PAGE_MA_NAT
);
140 vma
->vm_flags
= VM_READ
| VM_MAYREAD
| VM_IO
|
141 VM_DONTEXPAND
| VM_DONTDUMP
;
142 down_write(¤t
->mm
->mmap_sem
);
143 if (insert_vm_struct(current
->mm
, vma
)) {
144 up_write(¤t
->mm
->mmap_sem
);
145 kmem_cache_free(vm_area_cachep
, vma
);
148 up_write(¤t
->mm
->mmap_sem
);
156 free_reserved_area(ia64_imva(__init_begin
), ia64_imva(__init_end
),
157 -1, "unused kernel");
161 free_initrd_mem (unsigned long start
, unsigned long end
)
164 * EFI uses 4KB pages while the kernel can use 4KB or bigger.
165 * Thus EFI and the kernel may have different page sizes. It is
166 * therefore possible to have the initrd share the same page as
167 * the end of the kernel (given current setup).
169 * To avoid freeing/using the wrong page (kernel sized) we:
170 * - align up the beginning of initrd
171 * - align down the end of initrd
174 * |=============| a000
180 * |=============| 8000
183 * |/////////////| 7000
186 * |=============| 6000
189 * K=kernel using 8KB pages
191 * In this example, we must free page 8000 ONLY. So we must align up
192 * initrd_start and keep initrd_end as is.
194 start
= PAGE_ALIGN(start
);
195 end
= end
& PAGE_MASK
;
198 printk(KERN_INFO
"Freeing initrd memory: %ldkB freed\n", (end
- start
) >> 10);
200 for (; start
< end
; start
+= PAGE_SIZE
) {
201 if (!virt_addr_valid(start
))
203 free_reserved_page(virt_to_page(start
));
208 * This installs a clean page in the kernel's page table.
210 static struct page
* __init
211 put_kernel_page (struct page
*page
, unsigned long address
, pgprot_t pgprot
)
218 pgd
= pgd_offset_k(address
); /* note: this is NOT pgd_offset()! */
221 pud
= pud_alloc(&init_mm
, pgd
, address
);
224 pmd
= pmd_alloc(&init_mm
, pud
, address
);
227 pte
= pte_alloc_kernel(pmd
, address
);
232 set_pte(pte
, mk_pte(page
, pgprot
));
235 /* no need for flush_tlb */
245 * Map the gate page twice: once read-only to export the ELF
246 * headers etc. and once execute-only page to enable
247 * privilege-promotion via "epc":
249 page
= virt_to_page(ia64_imva(__start_gate_section
));
250 put_kernel_page(page
, GATE_ADDR
, PAGE_READONLY
);
251 #ifdef HAVE_BUGGY_SEGREL
252 page
= virt_to_page(ia64_imva(__start_gate_section
+ PAGE_SIZE
));
253 put_kernel_page(page
, GATE_ADDR
+ PAGE_SIZE
, PAGE_GATE
);
255 put_kernel_page(page
, GATE_ADDR
+ PERCPU_PAGE_SIZE
, PAGE_GATE
);
256 /* Fill in the holes (if any) with read-only zero pages: */
260 for (addr
= GATE_ADDR
+ PAGE_SIZE
;
261 addr
< GATE_ADDR
+ PERCPU_PAGE_SIZE
;
264 put_kernel_page(ZERO_PAGE(0), addr
,
266 put_kernel_page(ZERO_PAGE(0), addr
+ PERCPU_PAGE_SIZE
,
274 static struct vm_area_struct gate_vma
;
276 static int __init
gate_vma_init(void)
278 gate_vma
.vm_mm
= NULL
;
279 gate_vma
.vm_start
= FIXADDR_USER_START
;
280 gate_vma
.vm_end
= FIXADDR_USER_END
;
281 gate_vma
.vm_flags
= VM_READ
| VM_MAYREAD
| VM_EXEC
| VM_MAYEXEC
;
282 gate_vma
.vm_page_prot
= __P101
;
286 __initcall(gate_vma_init
);
288 struct vm_area_struct
*get_gate_vma(struct mm_struct
*mm
)
293 int in_gate_area_no_mm(unsigned long addr
)
295 if ((addr
>= FIXADDR_USER_START
) && (addr
< FIXADDR_USER_END
))
300 int in_gate_area(struct mm_struct
*mm
, unsigned long addr
)
302 return in_gate_area_no_mm(addr
);
305 void ia64_mmu_init(void *my_cpu_data
)
307 unsigned long pta
, impl_va_bits
;
308 extern void tlb_init(void);
310 #ifdef CONFIG_DISABLE_VHPT
311 # define VHPT_ENABLE_BIT 0
313 # define VHPT_ENABLE_BIT 1
317 * Check if the virtually mapped linear page table (VMLPT) overlaps with a mapped
318 * address space. The IA-64 architecture guarantees that at least 50 bits of
319 * virtual address space are implemented but if we pick a large enough page size
320 * (e.g., 64KB), the mapped address space is big enough that it will overlap with
321 * VMLPT. I assume that once we run on machines big enough to warrant 64KB pages,
322 * IMPL_VA_MSB will be significantly bigger, so this is unlikely to become a
323 * problem in practice. Alternatively, we could truncate the top of the mapped
324 * address space to not permit mappings that would overlap with the VMLPT.
328 # define mapped_space_bits (3*(PAGE_SHIFT - pte_bits) + PAGE_SHIFT)
330 * The virtual page table has to cover the entire implemented address space within
331 * a region even though not all of this space may be mappable. The reason for
332 * this is that the Access bit and Dirty bit fault handlers perform
333 * non-speculative accesses to the virtual page table, so the address range of the
334 * virtual page table itself needs to be covered by virtual page table.
336 # define vmlpt_bits (impl_va_bits - PAGE_SHIFT + pte_bits)
337 # define POW2(n) (1ULL << (n))
339 impl_va_bits
= ffz(~(local_cpu_data
->unimpl_va_mask
| (7UL << 61)));
341 if (impl_va_bits
< 51 || impl_va_bits
> 61)
342 panic("CPU has bogus IMPL_VA_MSB value of %lu!\n", impl_va_bits
- 1);
344 * mapped_space_bits - PAGE_SHIFT is the total number of ptes we need,
345 * which must fit into "vmlpt_bits - pte_bits" slots. Second half of
346 * the test makes sure that our mapped space doesn't overlap the
347 * unimplemented hole in the middle of the region.
349 if ((mapped_space_bits
- PAGE_SHIFT
> vmlpt_bits
- pte_bits
) ||
350 (mapped_space_bits
> impl_va_bits
- 1))
351 panic("Cannot build a big enough virtual-linear page table"
352 " to cover mapped address space.\n"
353 " Try using a smaller page size.\n");
356 /* place the VMLPT at the end of each page-table mapped region: */
357 pta
= POW2(61) - POW2(vmlpt_bits
);
360 * Set the (virtually mapped linear) page table address. Bit
361 * 8 selects between the short and long format, bits 2-7 the
362 * size of the table, and bit 0 whether the VHPT walker is
365 ia64_set_pta(pta
| (0 << 8) | (vmlpt_bits
<< 2) | VHPT_ENABLE_BIT
);
369 #ifdef CONFIG_HUGETLB_PAGE
370 ia64_set_rr(HPAGE_REGION_BASE
, HPAGE_SHIFT
<< 2);
375 #ifdef CONFIG_VIRTUAL_MEM_MAP
376 int vmemmap_find_next_valid_pfn(int node
, int i
)
378 unsigned long end_address
, hole_next_pfn
;
379 unsigned long stop_address
;
380 pg_data_t
*pgdat
= NODE_DATA(node
);
382 end_address
= (unsigned long) &vmem_map
[pgdat
->node_start_pfn
+ i
];
383 end_address
= PAGE_ALIGN(end_address
);
384 stop_address
= (unsigned long) &vmem_map
[pgdat_end_pfn(pgdat
)];
392 pgd
= pgd_offset_k(end_address
);
393 if (pgd_none(*pgd
)) {
394 end_address
+= PGDIR_SIZE
;
398 pud
= pud_offset(pgd
, end_address
);
399 if (pud_none(*pud
)) {
400 end_address
+= PUD_SIZE
;
404 pmd
= pmd_offset(pud
, end_address
);
405 if (pmd_none(*pmd
)) {
406 end_address
+= PMD_SIZE
;
410 pte
= pte_offset_kernel(pmd
, end_address
);
412 if (pte_none(*pte
)) {
413 end_address
+= PAGE_SIZE
;
415 if ((end_address
< stop_address
) &&
416 (end_address
!= ALIGN(end_address
, 1UL << PMD_SHIFT
)))
420 /* Found next valid vmem_map page */
422 } while (end_address
< stop_address
);
424 end_address
= min(end_address
, stop_address
);
425 end_address
= end_address
- (unsigned long) vmem_map
+ sizeof(struct page
) - 1;
426 hole_next_pfn
= end_address
/ sizeof(struct page
);
427 return hole_next_pfn
- pgdat
->node_start_pfn
;
430 int __init
create_mem_map_page_table(u64 start
, u64 end
, void *arg
)
432 unsigned long address
, start_page
, end_page
;
433 struct page
*map_start
, *map_end
;
440 map_start
= vmem_map
+ (__pa(start
) >> PAGE_SHIFT
);
441 map_end
= vmem_map
+ (__pa(end
) >> PAGE_SHIFT
);
443 start_page
= (unsigned long) map_start
& PAGE_MASK
;
444 end_page
= PAGE_ALIGN((unsigned long) map_end
);
445 node
= paddr_to_nid(__pa(start
));
447 for (address
= start_page
; address
< end_page
; address
+= PAGE_SIZE
) {
448 pgd
= pgd_offset_k(address
);
450 pgd_populate(&init_mm
, pgd
, alloc_bootmem_pages_node(NODE_DATA(node
), PAGE_SIZE
));
451 pud
= pud_offset(pgd
, address
);
454 pud_populate(&init_mm
, pud
, alloc_bootmem_pages_node(NODE_DATA(node
), PAGE_SIZE
));
455 pmd
= pmd_offset(pud
, address
);
458 pmd_populate_kernel(&init_mm
, pmd
, alloc_bootmem_pages_node(NODE_DATA(node
), PAGE_SIZE
));
459 pte
= pte_offset_kernel(pmd
, address
);
462 set_pte(pte
, pfn_pte(__pa(alloc_bootmem_pages_node(NODE_DATA(node
), PAGE_SIZE
)) >> PAGE_SHIFT
,
468 struct memmap_init_callback_data
{
476 virtual_memmap_init(u64 start
, u64 end
, void *arg
)
478 struct memmap_init_callback_data
*args
;
479 struct page
*map_start
, *map_end
;
481 args
= (struct memmap_init_callback_data
*) arg
;
482 map_start
= vmem_map
+ (__pa(start
) >> PAGE_SHIFT
);
483 map_end
= vmem_map
+ (__pa(end
) >> PAGE_SHIFT
);
485 if (map_start
< args
->start
)
486 map_start
= args
->start
;
487 if (map_end
> args
->end
)
491 * We have to initialize "out of bounds" struct page elements that fit completely
492 * on the same pages that were allocated for the "in bounds" elements because they
493 * may be referenced later (and found to be "reserved").
495 map_start
-= ((unsigned long) map_start
& (PAGE_SIZE
- 1)) / sizeof(struct page
);
496 map_end
+= ((PAGE_ALIGN((unsigned long) map_end
) - (unsigned long) map_end
)
497 / sizeof(struct page
));
499 if (map_start
< map_end
)
500 memmap_init_zone((unsigned long)(map_end
- map_start
),
501 args
->nid
, args
->zone
, page_to_pfn(map_start
),
507 memmap_init (unsigned long size
, int nid
, unsigned long zone
,
508 unsigned long start_pfn
)
511 memmap_init_zone(size
, nid
, zone
, start_pfn
, MEMMAP_EARLY
);
514 struct memmap_init_callback_data args
;
516 start
= pfn_to_page(start_pfn
);
518 args
.end
= start
+ size
;
522 efi_memmap_walk(virtual_memmap_init
, &args
);
527 ia64_pfn_valid (unsigned long pfn
)
530 struct page
*pg
= pfn_to_page(pfn
);
532 return (__get_user(byte
, (char __user
*) pg
) == 0)
533 && ((((u64
)pg
& PAGE_MASK
) == (((u64
)(pg
+ 1) - 1) & PAGE_MASK
))
534 || (__get_user(byte
, (char __user
*) (pg
+ 1) - 1) == 0));
536 EXPORT_SYMBOL(ia64_pfn_valid
);
538 int __init
find_largest_hole(u64 start
, u64 end
, void *arg
)
542 static u64 last_end
= PAGE_OFFSET
;
544 /* NOTE: this algorithm assumes efi memmap table is ordered */
546 if (*max_gap
< (start
- last_end
))
547 *max_gap
= start
- last_end
;
552 #endif /* CONFIG_VIRTUAL_MEM_MAP */
554 int __init
register_active_ranges(u64 start
, u64 len
, int nid
)
556 u64 end
= start
+ len
;
559 if (start
> crashk_res
.start
&& start
< crashk_res
.end
)
560 start
= crashk_res
.end
;
561 if (end
> crashk_res
.start
&& end
< crashk_res
.end
)
562 end
= crashk_res
.start
;
566 memblock_add_node(__pa(start
), end
- start
, nid
);
571 find_max_min_low_pfn (u64 start
, u64 end
, void *arg
)
573 unsigned long pfn_start
, pfn_end
;
574 #ifdef CONFIG_FLATMEM
575 pfn_start
= (PAGE_ALIGN(__pa(start
))) >> PAGE_SHIFT
;
576 pfn_end
= (PAGE_ALIGN(__pa(end
- 1))) >> PAGE_SHIFT
;
578 pfn_start
= GRANULEROUNDDOWN(__pa(start
)) >> PAGE_SHIFT
;
579 pfn_end
= GRANULEROUNDUP(__pa(end
- 1)) >> PAGE_SHIFT
;
581 min_low_pfn
= min(min_low_pfn
, pfn_start
);
582 max_low_pfn
= max(max_low_pfn
, pfn_end
);
587 * Boot command-line option "nolwsys" can be used to disable the use of any light-weight
588 * system call handler. When this option is in effect, all fsyscalls will end up bubbling
589 * down into the kernel and calling the normal (heavy-weight) syscall handler. This is
590 * useful for performance testing, but conceivably could also come in handy for debugging
594 static int nolwsys __initdata
;
597 nolwsys_setup (char *s
)
603 __setup("nolwsys", nolwsys_setup
);
610 BUG_ON(PTRS_PER_PGD
* sizeof(pgd_t
) != PAGE_SIZE
);
611 BUG_ON(PTRS_PER_PMD
* sizeof(pmd_t
) != PAGE_SIZE
);
612 BUG_ON(PTRS_PER_PTE
* sizeof(pte_t
) != PAGE_SIZE
);
616 * This needs to be called _after_ the command line has been parsed but _before_
617 * any drivers that may need the PCI DMA interface are initialized or bootmem has
623 #ifdef CONFIG_FLATMEM
627 set_max_mapnr(max_low_pfn
);
628 high_memory
= __va(max_low_pfn
* PAGE_SIZE
);
630 mem_init_print_info(NULL
);
633 * For fsyscall entrpoints with no light-weight handler, use the ordinary
634 * (heavy-weight) handler, but mark it by setting bit 0, so the fsyscall entry
635 * code can tell them apart.
637 for (i
= 0; i
< NR_syscalls
; ++i
) {
638 extern unsigned long fsyscall_table
[NR_syscalls
];
639 extern unsigned long sys_call_table
[NR_syscalls
];
641 if (!fsyscall_table
[i
] || nolwsys
)
642 fsyscall_table
[i
] = sys_call_table
[i
] | 1;
647 #ifdef CONFIG_MEMORY_HOTPLUG
648 int arch_add_memory(int nid
, u64 start
, u64 size
, bool for_device
)
652 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
653 unsigned long nr_pages
= size
>> PAGE_SHIFT
;
656 pgdat
= NODE_DATA(nid
);
658 zone
= pgdat
->node_zones
+
659 zone_for_memory(nid
, start
, size
, ZONE_NORMAL
, for_device
);
660 ret
= __add_pages(nid
, zone
, start_pfn
, nr_pages
);
663 printk("%s: Problem encountered in __add_pages() as ret=%d\n",
669 #ifdef CONFIG_MEMORY_HOTREMOVE
670 int arch_remove_memory(u64 start
, u64 size
)
672 unsigned long start_pfn
= start
>> PAGE_SHIFT
;
673 unsigned long nr_pages
= size
>> PAGE_SHIFT
;
677 zone
= page_zone(pfn_to_page(start_pfn
));
678 ret
= __remove_pages(zone
, start_pfn
, nr_pages
);
680 pr_warn("%s: Problem encountered in __remove_pages() as"
681 " ret=%d\n", __func__
, ret
);
689 * show_mem - give short summary of memory stats
691 * Shows a simple page count of reserved and used pages in the system.
692 * For discontig machines, it does this on a per-pgdat basis.
694 void show_mem(unsigned int filter
)
696 int total_reserved
= 0;
697 unsigned long total_present
= 0;
700 printk(KERN_INFO
"Mem-info:\n");
701 show_free_areas(filter
);
702 printk(KERN_INFO
"Node memory in pages:\n");
703 for_each_online_pgdat(pgdat
) {
704 unsigned long present
;
707 int nid
= pgdat
->node_id
;
710 if (skip_free_areas_node(filter
, nid
))
712 pgdat_resize_lock(pgdat
, &flags
);
714 for (zoneid
= 0; zoneid
< MAX_NR_ZONES
; zoneid
++) {
715 struct zone
*zone
= &pgdat
->node_zones
[zoneid
];
716 if (!populated_zone(zone
))
719 reserved
+= zone
->present_pages
- zone
->managed_pages
;
721 present
= pgdat
->node_present_pages
;
723 pgdat_resize_unlock(pgdat
, &flags
);
724 total_present
+= present
;
725 total_reserved
+= reserved
;
726 printk(KERN_INFO
"Node %4d: RAM: %11ld, rsvd: %8d, ",
727 nid
, present
, reserved
);
729 printk(KERN_INFO
"%ld pages of RAM\n", total_present
);
730 printk(KERN_INFO
"%d reserved pages\n", total_reserved
);
731 printk(KERN_INFO
"Total of %ld pages in page table cache\n",
732 quicklist_total_size());
733 printk(KERN_INFO
"%ld free buffer pages\n", nr_free_buffer_pages());