atmel_serial: fix bugs in probe() error path and remove()
[pv_ops_mirror.git] / arch / x86 / mm / init_64.c
blob5fe880fc305d03c90131948bab8c4cd16096e9f9
1 /*
2 * linux/arch/x86_64/mm/init.c
4 * Copyright (C) 1995 Linus Torvalds
5 * Copyright (C) 2000 Pavel Machek <pavel@suse.cz>
6 * Copyright (C) 2002,2003 Andi Kleen <ak@suse.de>
7 */
9 #include <linux/signal.h>
10 #include <linux/sched.h>
11 #include <linux/kernel.h>
12 #include <linux/errno.h>
13 #include <linux/string.h>
14 #include <linux/types.h>
15 #include <linux/ptrace.h>
16 #include <linux/mman.h>
17 #include <linux/mm.h>
18 #include <linux/swap.h>
19 #include <linux/smp.h>
20 #include <linux/init.h>
21 #include <linux/pagemap.h>
22 #include <linux/bootmem.h>
23 #include <linux/proc_fs.h>
24 #include <linux/pci.h>
25 #include <linux/pfn.h>
26 #include <linux/poison.h>
27 #include <linux/dma-mapping.h>
28 #include <linux/module.h>
29 #include <linux/memory_hotplug.h>
30 #include <linux/nmi.h>
32 #include <asm/processor.h>
33 #include <asm/system.h>
34 #include <asm/uaccess.h>
35 #include <asm/pgtable.h>
36 #include <asm/pgalloc.h>
37 #include <asm/dma.h>
38 #include <asm/fixmap.h>
39 #include <asm/e820.h>
40 #include <asm/apic.h>
41 #include <asm/tlb.h>
42 #include <asm/mmu_context.h>
43 #include <asm/proto.h>
44 #include <asm/smp.h>
45 #include <asm/sections.h>
46 #include <asm/kdebug.h>
47 #include <asm/numa.h>
49 const struct dma_mapping_ops *dma_ops;
50 EXPORT_SYMBOL(dma_ops);
52 static unsigned long dma_reserve __initdata;
54 DEFINE_PER_CPU(struct mmu_gather, mmu_gathers);
57 * NOTE: pagetable_init alloc all the fixmap pagetables contiguous on the
58 * physical space so we can cache the place of the first one and move
59 * around without checking the pgd every time.
62 void show_mem(void)
64 long i, total = 0, reserved = 0;
65 long shared = 0, cached = 0;
66 struct page *page;
67 pg_data_t *pgdat;
69 printk(KERN_INFO "Mem-info:\n");
70 show_free_areas();
71 printk(KERN_INFO "Free swap: %6ldkB\n",
72 nr_swap_pages << (PAGE_SHIFT-10));
74 for_each_online_pgdat(pgdat) {
75 for (i = 0; i < pgdat->node_spanned_pages; ++i) {
77 * This loop can take a while with 256 GB and
78 * 4k pages so defer the NMI watchdog:
80 if (unlikely(i % MAX_ORDER_NR_PAGES == 0))
81 touch_nmi_watchdog();
83 if (!pfn_valid(pgdat->node_start_pfn + i))
84 continue;
86 page = pfn_to_page(pgdat->node_start_pfn + i);
87 total++;
88 if (PageReserved(page))
89 reserved++;
90 else if (PageSwapCache(page))
91 cached++;
92 else if (page_count(page))
93 shared += page_count(page) - 1;
96 printk(KERN_INFO "%lu pages of RAM\n", total);
97 printk(KERN_INFO "%lu reserved pages\n", reserved);
98 printk(KERN_INFO "%lu pages shared\n", shared);
99 printk(KERN_INFO "%lu pages swap cached\n", cached);
102 int after_bootmem;
104 static __init void *spp_getpage(void)
106 void *ptr;
108 if (after_bootmem)
109 ptr = (void *) get_zeroed_page(GFP_ATOMIC);
110 else
111 ptr = alloc_bootmem_pages(PAGE_SIZE);
113 if (!ptr || ((unsigned long)ptr & ~PAGE_MASK)) {
114 panic("set_pte_phys: cannot allocate page data %s\n",
115 after_bootmem ? "after bootmem" : "");
118 pr_debug("spp_getpage %p\n", ptr);
120 return ptr;
123 static __init void
124 set_pte_phys(unsigned long vaddr, unsigned long phys, pgprot_t prot)
126 pgd_t *pgd;
127 pud_t *pud;
128 pmd_t *pmd;
129 pte_t *pte, new_pte;
131 pr_debug("set_pte_phys %lx to %lx\n", vaddr, phys);
133 pgd = pgd_offset_k(vaddr);
134 if (pgd_none(*pgd)) {
135 printk(KERN_ERR
136 "PGD FIXMAP MISSING, it should be setup in head.S!\n");
137 return;
139 pud = pud_offset(pgd, vaddr);
140 if (pud_none(*pud)) {
141 pmd = (pmd_t *) spp_getpage();
142 set_pud(pud, __pud(__pa(pmd) | _KERNPG_TABLE | _PAGE_USER));
143 if (pmd != pmd_offset(pud, 0)) {
144 printk(KERN_ERR "PAGETABLE BUG #01! %p <-> %p\n",
145 pmd, pmd_offset(pud, 0));
146 return;
149 pmd = pmd_offset(pud, vaddr);
150 if (pmd_none(*pmd)) {
151 pte = (pte_t *) spp_getpage();
152 set_pmd(pmd, __pmd(__pa(pte) | _KERNPG_TABLE | _PAGE_USER));
153 if (pte != pte_offset_kernel(pmd, 0)) {
154 printk(KERN_ERR "PAGETABLE BUG #02!\n");
155 return;
158 new_pte = pfn_pte(phys >> PAGE_SHIFT, prot);
160 pte = pte_offset_kernel(pmd, vaddr);
161 if (!pte_none(*pte) &&
162 pte_val(*pte) != (pte_val(new_pte) & __supported_pte_mask))
163 pte_ERROR(*pte);
164 set_pte(pte, new_pte);
167 * It's enough to flush this one mapping.
168 * (PGE mappings get flushed as well)
170 __flush_tlb_one(vaddr);
173 /* NOTE: this is meant to be run only at boot */
174 void __init
175 __set_fixmap(enum fixed_addresses idx, unsigned long phys, pgprot_t prot)
177 unsigned long address = __fix_to_virt(idx);
179 if (idx >= __end_of_fixed_addresses) {
180 printk(KERN_ERR "Invalid __set_fixmap\n");
181 return;
183 set_pte_phys(address, phys, prot);
186 static unsigned long __initdata table_start;
187 static unsigned long __meminitdata table_end;
189 static __meminit void *alloc_low_page(unsigned long *phys)
191 unsigned long pfn = table_end++;
192 void *adr;
194 if (after_bootmem) {
195 adr = (void *)get_zeroed_page(GFP_ATOMIC);
196 *phys = __pa(adr);
198 return adr;
201 if (pfn >= end_pfn)
202 panic("alloc_low_page: ran out of memory");
204 adr = early_ioremap(pfn * PAGE_SIZE, PAGE_SIZE);
205 memset(adr, 0, PAGE_SIZE);
206 *phys = pfn * PAGE_SIZE;
207 return adr;
210 static __meminit void unmap_low_page(void *adr)
212 if (after_bootmem)
213 return;
215 early_iounmap(adr, PAGE_SIZE);
218 /* Must run before zap_low_mappings */
219 __meminit void *early_ioremap(unsigned long addr, unsigned long size)
221 pmd_t *pmd, *last_pmd;
222 unsigned long vaddr;
223 int i, pmds;
225 pmds = ((addr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE;
226 vaddr = __START_KERNEL_map;
227 pmd = level2_kernel_pgt;
228 last_pmd = level2_kernel_pgt + PTRS_PER_PMD - 1;
230 for (; pmd <= last_pmd; pmd++, vaddr += PMD_SIZE) {
231 for (i = 0; i < pmds; i++) {
232 if (pmd_present(pmd[i]))
233 goto continue_outer_loop;
235 vaddr += addr & ~PMD_MASK;
236 addr &= PMD_MASK;
238 for (i = 0; i < pmds; i++, addr += PMD_SIZE)
239 set_pmd(pmd+i, __pmd(addr | __PAGE_KERNEL_LARGE_EXEC));
240 __flush_tlb_all();
242 return (void *)vaddr;
243 continue_outer_loop:
246 printk(KERN_ERR "early_ioremap(0x%lx, %lu) failed\n", addr, size);
248 return NULL;
252 * To avoid virtual aliases later:
254 __meminit void early_iounmap(void *addr, unsigned long size)
256 unsigned long vaddr;
257 pmd_t *pmd;
258 int i, pmds;
260 vaddr = (unsigned long)addr;
261 pmds = ((vaddr & ~PMD_MASK) + size + ~PMD_MASK) / PMD_SIZE;
262 pmd = level2_kernel_pgt + pmd_index(vaddr);
264 for (i = 0; i < pmds; i++)
265 pmd_clear(pmd + i);
267 __flush_tlb_all();
270 static void __meminit
271 phys_pmd_init(pmd_t *pmd_page, unsigned long address, unsigned long end)
273 int i = pmd_index(address);
275 for (; i < PTRS_PER_PMD; i++, address += PMD_SIZE) {
276 pmd_t *pmd = pmd_page + pmd_index(address);
278 if (address >= end) {
279 if (!after_bootmem) {
280 for (; i < PTRS_PER_PMD; i++, pmd++)
281 set_pmd(pmd, __pmd(0));
283 break;
286 if (pmd_val(*pmd))
287 continue;
289 set_pte((pte_t *)pmd,
290 pfn_pte(address >> PAGE_SHIFT, PAGE_KERNEL_LARGE));
294 static void __meminit
295 phys_pmd_update(pud_t *pud, unsigned long address, unsigned long end)
297 pmd_t *pmd = pmd_offset(pud, 0);
298 spin_lock(&init_mm.page_table_lock);
299 phys_pmd_init(pmd, address, end);
300 spin_unlock(&init_mm.page_table_lock);
301 __flush_tlb_all();
304 static void __meminit
305 phys_pud_init(pud_t *pud_page, unsigned long addr, unsigned long end)
307 int i = pud_index(addr);
309 for (; i < PTRS_PER_PUD; i++, addr = (addr & PUD_MASK) + PUD_SIZE) {
310 unsigned long pmd_phys;
311 pud_t *pud = pud_page + pud_index(addr);
312 pmd_t *pmd;
314 if (addr >= end)
315 break;
317 if (!after_bootmem &&
318 !e820_any_mapped(addr, addr+PUD_SIZE, 0)) {
319 set_pud(pud, __pud(0));
320 continue;
323 if (pud_val(*pud)) {
324 phys_pmd_update(pud, addr, end);
325 continue;
328 pmd = alloc_low_page(&pmd_phys);
330 spin_lock(&init_mm.page_table_lock);
331 set_pud(pud, __pud(pmd_phys | _KERNPG_TABLE));
332 phys_pmd_init(pmd, addr, end);
333 spin_unlock(&init_mm.page_table_lock);
335 unmap_low_page(pmd);
337 __flush_tlb_all();
340 static void __init find_early_table_space(unsigned long end)
342 unsigned long puds, pmds, tables, start;
344 puds = (end + PUD_SIZE - 1) >> PUD_SHIFT;
345 pmds = (end + PMD_SIZE - 1) >> PMD_SHIFT;
346 tables = round_up(puds * sizeof(pud_t), PAGE_SIZE) +
347 round_up(pmds * sizeof(pmd_t), PAGE_SIZE);
350 * RED-PEN putting page tables only on node 0 could
351 * cause a hotspot and fill up ZONE_DMA. The page tables
352 * need roughly 0.5KB per GB.
354 start = 0x8000;
355 table_start = find_e820_area(start, end, tables, PAGE_SIZE);
356 if (table_start == -1UL)
357 panic("Cannot find space for the kernel page tables");
359 table_start >>= PAGE_SHIFT;
360 table_end = table_start;
362 early_printk("kernel direct mapping tables up to %lx @ %lx-%lx\n",
363 end, table_start << PAGE_SHIFT,
364 (table_start << PAGE_SHIFT) + tables);
368 * Setup the direct mapping of the physical memory at PAGE_OFFSET.
369 * This runs before bootmem is initialized and gets pages directly from
370 * the physical memory. To access them they are temporarily mapped.
372 void __init_refok init_memory_mapping(unsigned long start, unsigned long end)
374 unsigned long next;
376 pr_debug("init_memory_mapping\n");
379 * Find space for the kernel direct mapping tables.
381 * Later we should allocate these tables in the local node of the
382 * memory mapped. Unfortunately this is done currently before the
383 * nodes are discovered.
385 if (!after_bootmem)
386 find_early_table_space(end);
388 start = (unsigned long)__va(start);
389 end = (unsigned long)__va(end);
391 for (; start < end; start = next) {
392 pgd_t *pgd = pgd_offset_k(start);
393 unsigned long pud_phys;
394 pud_t *pud;
396 if (after_bootmem)
397 pud = pud_offset(pgd, start & PGDIR_MASK);
398 else
399 pud = alloc_low_page(&pud_phys);
401 next = start + PGDIR_SIZE;
402 if (next > end)
403 next = end;
404 phys_pud_init(pud, __pa(start), __pa(next));
405 if (!after_bootmem)
406 set_pgd(pgd_offset_k(start), mk_kernel_pgd(pud_phys));
407 unmap_low_page(pud);
410 if (!after_bootmem)
411 mmu_cr4_features = read_cr4();
412 __flush_tlb_all();
414 if (!after_bootmem)
415 reserve_early(table_start << PAGE_SHIFT,
416 table_end << PAGE_SHIFT, "PGTABLE");
419 #ifndef CONFIG_NUMA
420 void __init paging_init(void)
422 unsigned long max_zone_pfns[MAX_NR_ZONES];
424 memset(max_zone_pfns, 0, sizeof(max_zone_pfns));
425 max_zone_pfns[ZONE_DMA] = MAX_DMA_PFN;
426 max_zone_pfns[ZONE_DMA32] = MAX_DMA32_PFN;
427 max_zone_pfns[ZONE_NORMAL] = end_pfn;
429 memory_present(0, 0, end_pfn);
430 sparse_init();
431 free_area_init_nodes(max_zone_pfns);
433 #endif
436 * Memory hotplug specific functions
438 void online_page(struct page *page)
440 ClearPageReserved(page);
441 init_page_count(page);
442 __free_page(page);
443 totalram_pages++;
444 num_physpages++;
447 #ifdef CONFIG_MEMORY_HOTPLUG
449 * Memory is added always to NORMAL zone. This means you will never get
450 * additional DMA/DMA32 memory.
452 int arch_add_memory(int nid, u64 start, u64 size)
454 struct pglist_data *pgdat = NODE_DATA(nid);
455 struct zone *zone = pgdat->node_zones + ZONE_NORMAL;
456 unsigned long start_pfn = start >> PAGE_SHIFT;
457 unsigned long nr_pages = size >> PAGE_SHIFT;
458 int ret;
460 init_memory_mapping(start, start + size-1);
462 ret = __add_pages(zone, start_pfn, nr_pages);
463 WARN_ON(1);
465 return ret;
467 EXPORT_SYMBOL_GPL(arch_add_memory);
469 #if !defined(CONFIG_ACPI_NUMA) && defined(CONFIG_NUMA)
470 int memory_add_physaddr_to_nid(u64 start)
472 return 0;
474 EXPORT_SYMBOL_GPL(memory_add_physaddr_to_nid);
475 #endif
477 #endif /* CONFIG_MEMORY_HOTPLUG */
479 static struct kcore_list kcore_mem, kcore_vmalloc, kcore_kernel,
480 kcore_modules, kcore_vsyscall;
482 void __init mem_init(void)
484 long codesize, reservedpages, datasize, initsize;
486 pci_iommu_alloc();
488 /* clear_bss() already clear the empty_zero_page */
490 /* temporary debugging - double check it's true: */
492 int i;
494 for (i = 0; i < 1024; i++)
495 WARN_ON_ONCE(empty_zero_page[i]);
498 reservedpages = 0;
500 /* this will put all low memory onto the freelists */
501 #ifdef CONFIG_NUMA
502 totalram_pages = numa_free_all_bootmem();
503 #else
504 totalram_pages = free_all_bootmem();
505 #endif
506 reservedpages = end_pfn - totalram_pages -
507 absent_pages_in_range(0, end_pfn);
508 after_bootmem = 1;
510 codesize = (unsigned long) &_etext - (unsigned long) &_text;
511 datasize = (unsigned long) &_edata - (unsigned long) &_etext;
512 initsize = (unsigned long) &__init_end - (unsigned long) &__init_begin;
514 /* Register memory areas for /proc/kcore */
515 kclist_add(&kcore_mem, __va(0), max_low_pfn << PAGE_SHIFT);
516 kclist_add(&kcore_vmalloc, (void *)VMALLOC_START,
517 VMALLOC_END-VMALLOC_START);
518 kclist_add(&kcore_kernel, &_stext, _end - _stext);
519 kclist_add(&kcore_modules, (void *)MODULES_VADDR, MODULES_LEN);
520 kclist_add(&kcore_vsyscall, (void *)VSYSCALL_START,
521 VSYSCALL_END - VSYSCALL_START);
523 printk(KERN_INFO "Memory: %luk/%luk available (%ldk kernel code, "
524 "%ldk reserved, %ldk data, %ldk init)\n",
525 (unsigned long) nr_free_pages() << (PAGE_SHIFT-10),
526 end_pfn << (PAGE_SHIFT-10),
527 codesize >> 10,
528 reservedpages << (PAGE_SHIFT-10),
529 datasize >> 10,
530 initsize >> 10);
533 void free_init_pages(char *what, unsigned long begin, unsigned long end)
535 unsigned long addr;
537 if (begin >= end)
538 return;
541 * If debugging page accesses then do not free this memory but
542 * mark them not present - any buggy init-section access will
543 * create a kernel page fault:
545 #ifdef CONFIG_DEBUG_PAGEALLOC
546 printk(KERN_INFO "debug: unmapping init memory %08lx..%08lx\n",
547 begin, PAGE_ALIGN(end));
548 set_memory_np(begin, (end - begin) >> PAGE_SHIFT);
549 #else
550 printk(KERN_INFO "Freeing %s: %luk freed\n", what, (end - begin) >> 10);
552 for (addr = begin; addr < end; addr += PAGE_SIZE) {
553 ClearPageReserved(virt_to_page(addr));
554 init_page_count(virt_to_page(addr));
555 memset((void *)(addr & ~(PAGE_SIZE-1)),
556 POISON_FREE_INITMEM, PAGE_SIZE);
557 free_page(addr);
558 totalram_pages++;
560 #endif
563 void free_initmem(void)
565 free_init_pages("unused kernel memory",
566 (unsigned long)(&__init_begin),
567 (unsigned long)(&__init_end));
570 #ifdef CONFIG_DEBUG_RODATA
571 const int rodata_test_data = 0xC3;
572 EXPORT_SYMBOL_GPL(rodata_test_data);
574 void mark_rodata_ro(void)
576 unsigned long start = (unsigned long)_stext, end;
578 #ifdef CONFIG_HOTPLUG_CPU
579 /* It must still be possible to apply SMP alternatives. */
580 if (num_possible_cpus() > 1)
581 start = (unsigned long)_etext;
582 #endif
584 #ifdef CONFIG_KPROBES
585 start = (unsigned long)__start_rodata;
586 #endif
588 end = (unsigned long)__end_rodata;
589 start = (start + PAGE_SIZE - 1) & PAGE_MASK;
590 end &= PAGE_MASK;
591 if (end <= start)
592 return;
595 printk(KERN_INFO "Write protecting the kernel read-only data: %luk\n",
596 (end - start) >> 10);
597 set_memory_ro(start, (end - start) >> PAGE_SHIFT);
600 * The rodata section (but not the kernel text!) should also be
601 * not-executable.
603 start = ((unsigned long)__start_rodata + PAGE_SIZE - 1) & PAGE_MASK;
604 set_memory_nx(start, (end - start) >> PAGE_SHIFT);
606 rodata_test();
608 #ifdef CONFIG_CPA_DEBUG
609 printk(KERN_INFO "Testing CPA: undo %lx-%lx\n", start, end);
610 set_memory_rw(start, (end-start) >> PAGE_SHIFT);
612 printk(KERN_INFO "Testing CPA: again\n");
613 set_memory_ro(start, (end-start) >> PAGE_SHIFT);
614 #endif
616 #endif
618 #ifdef CONFIG_BLK_DEV_INITRD
619 void free_initrd_mem(unsigned long start, unsigned long end)
621 free_init_pages("initrd memory", start, end);
623 #endif
625 void __init reserve_bootmem_generic(unsigned long phys, unsigned len)
627 #ifdef CONFIG_NUMA
628 int nid = phys_to_nid(phys);
629 #endif
630 unsigned long pfn = phys >> PAGE_SHIFT;
632 if (pfn >= end_pfn) {
634 * This can happen with kdump kernels when accessing
635 * firmware tables:
637 if (pfn < end_pfn_map)
638 return;
640 printk(KERN_ERR "reserve_bootmem: illegal reserve %lx %u\n",
641 phys, len);
642 return;
645 /* Should check here against the e820 map to avoid double free */
646 #ifdef CONFIG_NUMA
647 reserve_bootmem_node(NODE_DATA(nid), phys, len, BOOTMEM_DEFAULT);
648 #else
649 reserve_bootmem(phys, len, BOOTMEM_DEFAULT);
650 #endif
651 if (phys+len <= MAX_DMA_PFN*PAGE_SIZE) {
652 dma_reserve += len / PAGE_SIZE;
653 set_dma_reserve(dma_reserve);
657 int kern_addr_valid(unsigned long addr)
659 unsigned long above = ((long)addr) >> __VIRTUAL_MASK_SHIFT;
660 pgd_t *pgd;
661 pud_t *pud;
662 pmd_t *pmd;
663 pte_t *pte;
665 if (above != 0 && above != -1UL)
666 return 0;
668 pgd = pgd_offset_k(addr);
669 if (pgd_none(*pgd))
670 return 0;
672 pud = pud_offset(pgd, addr);
673 if (pud_none(*pud))
674 return 0;
676 pmd = pmd_offset(pud, addr);
677 if (pmd_none(*pmd))
678 return 0;
680 if (pmd_large(*pmd))
681 return pfn_valid(pmd_pfn(*pmd));
683 pte = pte_offset_kernel(pmd, addr);
684 if (pte_none(*pte))
685 return 0;
687 return pfn_valid(pte_pfn(*pte));
691 * A pseudo VMA to allow ptrace access for the vsyscall page. This only
692 * covers the 64bit vsyscall page now. 32bit has a real VMA now and does
693 * not need special handling anymore:
695 static struct vm_area_struct gate_vma = {
696 .vm_start = VSYSCALL_START,
697 .vm_end = VSYSCALL_START + (VSYSCALL_MAPPED_PAGES * PAGE_SIZE),
698 .vm_page_prot = PAGE_READONLY_EXEC,
699 .vm_flags = VM_READ | VM_EXEC
702 struct vm_area_struct *get_gate_vma(struct task_struct *tsk)
704 #ifdef CONFIG_IA32_EMULATION
705 if (test_tsk_thread_flag(tsk, TIF_IA32))
706 return NULL;
707 #endif
708 return &gate_vma;
711 int in_gate_area(struct task_struct *task, unsigned long addr)
713 struct vm_area_struct *vma = get_gate_vma(task);
715 if (!vma)
716 return 0;
718 return (addr >= vma->vm_start) && (addr < vma->vm_end);
722 * Use this when you have no reliable task/vma, typically from interrupt
723 * context. It is less reliable than using the task's vma and may give
724 * false positives:
726 int in_gate_area_no_task(unsigned long addr)
728 return (addr >= VSYSCALL_START) && (addr < VSYSCALL_END);
731 const char *arch_vma_name(struct vm_area_struct *vma)
733 if (vma->vm_mm && vma->vm_start == (long)vma->vm_mm->context.vdso)
734 return "[vdso]";
735 if (vma == &gate_vma)
736 return "[vsyscall]";
737 return NULL;
740 #ifdef CONFIG_SPARSEMEM_VMEMMAP
742 * Initialise the sparsemem vmemmap using huge-pages at the PMD level.
744 int __meminit
745 vmemmap_populate(struct page *start_page, unsigned long size, int node)
747 unsigned long addr = (unsigned long)start_page;
748 unsigned long end = (unsigned long)(start_page + size);
749 unsigned long next;
750 pgd_t *pgd;
751 pud_t *pud;
752 pmd_t *pmd;
754 for (; addr < end; addr = next) {
755 next = pmd_addr_end(addr, end);
757 pgd = vmemmap_pgd_populate(addr, node);
758 if (!pgd)
759 return -ENOMEM;
761 pud = vmemmap_pud_populate(pgd, addr, node);
762 if (!pud)
763 return -ENOMEM;
765 pmd = pmd_offset(pud, addr);
766 if (pmd_none(*pmd)) {
767 pte_t entry;
768 void *p;
770 p = vmemmap_alloc_block(PMD_SIZE, node);
771 if (!p)
772 return -ENOMEM;
774 entry = pfn_pte(__pa(p) >> PAGE_SHIFT,
775 PAGE_KERNEL_LARGE);
776 set_pmd(pmd, __pmd(pte_val(entry)));
778 printk(KERN_DEBUG " [%lx-%lx] PMD ->%p on node %d\n",
779 addr, addr + PMD_SIZE - 1, p, node);
780 } else {
781 vmemmap_verify((pte_t *)pmd, node, addr, next);
784 return 0;
786 #endif