1 // SPDX-License-Identifier: GPL-2.0
3 * Initialize MMU support.
5 * Copyright (C) 1998-2003 Hewlett-Packard Co
6 * David Mosberger-Tang <davidm@hpl.hp.com>
8 #include <linux/kernel.h>
9 #include <linux/init.h>
11 #include <linux/dma-map-ops.h>
12 #include <linux/dmar.h>
13 #include <linux/efi.h>
14 #include <linux/elf.h>
15 #include <linux/memblock.h>
17 #include <linux/sched/signal.h>
18 #include <linux/mmzone.h>
19 #include <linux/module.h>
20 #include <linux/personality.h>
21 #include <linux/reboot.h>
22 #include <linux/slab.h>
23 #include <linux/swap.h>
24 #include <linux/proc_fs.h>
25 #include <linux/bitops.h>
26 #include <linux/kexec.h>
27 #include <linux/swiotlb.h>
32 #include <asm/patch.h>
33 #include <asm/pgalloc.h>
35 #include <asm/sections.h>
37 #include <linux/uaccess.h>
38 #include <asm/unistd.h>
41 extern void ia64_tlb_init (void);
43 unsigned long MAX_DMA_ADDRESS = PAGE_OFFSET + 0x100000000UL;
45 #ifdef CONFIG_VIRTUAL_MEM_MAP
46 unsigned long VMALLOC_END = VMALLOC_END_INIT;
47 EXPORT_SYMBOL(VMALLOC_END);
48 struct page *vmem_map;
49 EXPORT_SYMBOL(vmem_map);
52 struct page *zero_page_memmap_ptr; /* map entry for zero page */
53 EXPORT_SYMBOL(zero_page_memmap_ptr);
56 __ia64_sync_icache_dcache (pte_t pte)
62 addr = (unsigned long) page_address(page);
64 if (test_bit(PG_arch_1, &page->flags))
65 return; /* i-cache is already coherent with d-cache */
67 flush_icache_range(addr, addr + page_size(page));
68 set_bit(PG_arch_1, &page->flags); /* mark page as clean */
72 * Since DMA is i-cache coherent, any (complete) pages that were written via
73 * DMA can be marked as "clean" so that lazy_mmu_prot_update() doesn't have to
74 * flush them when they get mapped into an executable vm-area.
76 void arch_dma_mark_clean(phys_addr_t paddr, size_t size)
78 unsigned long pfn = PHYS_PFN(paddr);
81 set_bit(PG_arch_1, &pfn_to_page(pfn)->flags);
82 } while (++pfn <= PHYS_PFN(paddr + size - 1));
86 ia64_set_rbs_bot (void)
88 unsigned long stack_size = rlimit_max(RLIMIT_STACK) & -16;
90 if (stack_size > MAX_USER_STACK_SIZE)
91 stack_size = MAX_USER_STACK_SIZE;
92 current->thread.rbs_bot = PAGE_ALIGN(current->mm->start_stack - stack_size);
96 * This performs some platform-dependent address space initialization.
97 * On IA-64, we want to setup the VM area for the register backing
98 * store (which grows upwards) and install the gateway page which is
99 * used for signal trampolines, etc.
102 ia64_init_addr_space (void)
104 struct vm_area_struct *vma;
109 * If we're out of memory and kmem_cache_alloc() returns NULL, we simply ignore
110 * the problem. When the process attempts to write to the register backing store
111 * for the first time, it will get a SEGFAULT in this case.
113 vma = vm_area_alloc(current->mm);
115 vma_set_anonymous(vma);
116 vma->vm_start = current->thread.rbs_bot & PAGE_MASK;
117 vma->vm_end = vma->vm_start + PAGE_SIZE;
118 vma->vm_flags = VM_DATA_DEFAULT_FLAGS|VM_GROWSUP|VM_ACCOUNT;
119 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
120 mmap_write_lock(current->mm);
121 if (insert_vm_struct(current->mm, vma)) {
122 mmap_write_unlock(current->mm);
126 mmap_write_unlock(current->mm);
129 /* map NaT-page at address zero to speed up speculative dereferencing of NULL: */
130 if (!(current->personality & MMAP_PAGE_ZERO)) {
131 vma = vm_area_alloc(current->mm);
133 vma_set_anonymous(vma);
134 vma->vm_end = PAGE_SIZE;
135 vma->vm_page_prot = __pgprot(pgprot_val(PAGE_READONLY) | _PAGE_MA_NAT);
136 vma->vm_flags = VM_READ | VM_MAYREAD | VM_IO |
137 VM_DONTEXPAND | VM_DONTDUMP;
138 mmap_write_lock(current->mm);
139 if (insert_vm_struct(current->mm, vma)) {
140 mmap_write_unlock(current->mm);
144 mmap_write_unlock(current->mm);
152 free_reserved_area(ia64_imva(__init_begin), ia64_imva(__init_end),
153 -1, "unused kernel");
157 free_initrd_mem (unsigned long start, unsigned long end)
160 * EFI uses 4KB pages while the kernel can use 4KB or bigger.
161 * Thus EFI and the kernel may have different page sizes. It is
162 * therefore possible to have the initrd share the same page as
163 * the end of the kernel (given current setup).
165 * To avoid freeing/using the wrong page (kernel sized) we:
166 * - align up the beginning of initrd
167 * - align down the end of initrd
170 * |=============| a000
176 * |=============| 8000
179 * |/////////////| 7000
182 * |=============| 6000
185 * K=kernel using 8KB pages
187 * In this example, we must free page 8000 ONLY. So we must align up
188 * initrd_start and keep initrd_end as is.
190 start = PAGE_ALIGN(start);
191 end = end & PAGE_MASK;
194 printk(KERN_INFO "Freeing initrd memory: %ldkB freed\n", (end - start) >> 10);
196 for (; start < end; start += PAGE_SIZE) {
197 if (!virt_addr_valid(start))
199 free_reserved_page(virt_to_page(start));
204 * This installs a clean page in the kernel's page table.
206 static struct page * __init
207 put_kernel_page (struct page *page, unsigned long address, pgprot_t pgprot)
215 pgd = pgd_offset_k(address); /* note: this is NOT pgd_offset()! */
218 p4d = p4d_alloc(&init_mm, pgd, address);
221 pud = pud_alloc(&init_mm, p4d, 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 vma_init(&gate_vma, 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)];
393 pgd = pgd_offset_k(end_address);
394 if (pgd_none(*pgd)) {
395 end_address += PGDIR_SIZE;
399 p4d = p4d_offset(pgd, end_address);
400 if (p4d_none(*p4d)) {
401 end_address += P4D_SIZE;
405 pud = pud_offset(p4d, end_address);
406 if (pud_none(*pud)) {
407 end_address += PUD_SIZE;
411 pmd = pmd_offset(pud, end_address);
412 if (pmd_none(*pmd)) {
413 end_address += PMD_SIZE;
417 pte = pte_offset_kernel(pmd, end_address);
419 if (pte_none(*pte)) {
420 end_address += PAGE_SIZE;
422 if ((end_address < stop_address) &&
423 (end_address != ALIGN(end_address, 1UL << PMD_SHIFT)))
427 /* Found next valid vmem_map page */
429 } while (end_address < stop_address);
431 end_address = min(end_address, stop_address);
432 end_address = end_address - (unsigned long) vmem_map + sizeof(struct page) - 1;
433 hole_next_pfn = end_address / sizeof(struct page);
434 return hole_next_pfn - pgdat->node_start_pfn;
437 int __init create_mem_map_page_table(u64 start, u64 end, void *arg)
439 unsigned long address, start_page, end_page;
440 struct page *map_start, *map_end;
448 map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
449 map_end = vmem_map + (__pa(end) >> PAGE_SHIFT);
451 start_page = (unsigned long) map_start & PAGE_MASK;
452 end_page = PAGE_ALIGN((unsigned long) map_end);
453 node = paddr_to_nid(__pa(start));
455 for (address = start_page; address < end_page; address += PAGE_SIZE) {
456 pgd = pgd_offset_k(address);
457 if (pgd_none(*pgd)) {
458 p4d = memblock_alloc_node(PAGE_SIZE, PAGE_SIZE, node);
461 pgd_populate(&init_mm, pgd, p4d);
463 p4d = p4d_offset(pgd, address);
465 if (p4d_none(*p4d)) {
466 pud = memblock_alloc_node(PAGE_SIZE, PAGE_SIZE, node);
469 p4d_populate(&init_mm, p4d, pud);
471 pud = pud_offset(p4d, address);
473 if (pud_none(*pud)) {
474 pmd = memblock_alloc_node(PAGE_SIZE, PAGE_SIZE, node);
477 pud_populate(&init_mm, pud, pmd);
479 pmd = pmd_offset(pud, address);
481 if (pmd_none(*pmd)) {
482 pte = memblock_alloc_node(PAGE_SIZE, PAGE_SIZE, node);
485 pmd_populate_kernel(&init_mm, pmd, pte);
487 pte = pte_offset_kernel(pmd, address);
489 if (pte_none(*pte)) {
490 void *page = memblock_alloc_node(PAGE_SIZE, PAGE_SIZE,
494 set_pte(pte, pfn_pte(__pa(page) >> PAGE_SHIFT,
501 panic("%s: Failed to allocate %lu bytes align=0x%lx nid=%d\n",
502 __func__, PAGE_SIZE, PAGE_SIZE, node);
506 struct memmap_init_callback_data {
514 virtual_memmap_init(u64 start, u64 end, void *arg)
516 struct memmap_init_callback_data *args;
517 struct page *map_start, *map_end;
519 args = (struct memmap_init_callback_data *) arg;
520 map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
521 map_end = vmem_map + (__pa(end) >> PAGE_SHIFT);
523 if (map_start < args->start)
524 map_start = args->start;
525 if (map_end > args->end)
529 * We have to initialize "out of bounds" struct page elements that fit completely
530 * on the same pages that were allocated for the "in bounds" elements because they
531 * may be referenced later (and found to be "reserved").
533 map_start -= ((unsigned long) map_start & (PAGE_SIZE - 1)) / sizeof(struct page);
534 map_end += ((PAGE_ALIGN((unsigned long) map_end) - (unsigned long) map_end)
535 / sizeof(struct page));
537 if (map_start < map_end)
538 memmap_init_zone((unsigned long)(map_end - map_start),
539 args->nid, args->zone, page_to_pfn(map_start), page_to_pfn(map_end),
540 MEMINIT_EARLY, NULL, MIGRATE_MOVABLE);
545 arch_memmap_init (unsigned long size, int nid, unsigned long zone,
546 unsigned long start_pfn)
549 memmap_init_zone(size, nid, zone, start_pfn, start_pfn + size,
550 MEMINIT_EARLY, NULL, MIGRATE_MOVABLE);
553 struct memmap_init_callback_data args;
555 start = pfn_to_page(start_pfn);
557 args.end = start + size;
561 efi_memmap_walk(virtual_memmap_init, &args);
565 void __init memmap_init(void)
570 ia64_pfn_valid (unsigned long pfn)
573 struct page *pg = pfn_to_page(pfn);
575 return (__get_user(byte, (char __user *) pg) == 0)
576 && ((((u64)pg & PAGE_MASK) == (((u64)(pg + 1) - 1) & PAGE_MASK))
577 || (__get_user(byte, (char __user *) (pg + 1) - 1) == 0));
579 EXPORT_SYMBOL(ia64_pfn_valid);
581 int __init find_largest_hole(u64 start, u64 end, void *arg)
585 static u64 last_end = PAGE_OFFSET;
587 /* NOTE: this algorithm assumes efi memmap table is ordered */
589 if (*max_gap < (start - last_end))
590 *max_gap = start - last_end;
595 #endif /* CONFIG_VIRTUAL_MEM_MAP */
597 int __init register_active_ranges(u64 start, u64 len, int nid)
599 u64 end = start + len;
602 if (start > crashk_res.start && start < crashk_res.end)
603 start = crashk_res.end;
604 if (end > crashk_res.start && end < crashk_res.end)
605 end = crashk_res.start;
609 memblock_add_node(__pa(start), end - start, nid);
614 find_max_min_low_pfn (u64 start, u64 end, void *arg)
616 unsigned long pfn_start, pfn_end;
617 #ifdef CONFIG_FLATMEM
618 pfn_start = (PAGE_ALIGN(__pa(start))) >> PAGE_SHIFT;
619 pfn_end = (PAGE_ALIGN(__pa(end - 1))) >> PAGE_SHIFT;
621 pfn_start = GRANULEROUNDDOWN(__pa(start)) >> PAGE_SHIFT;
622 pfn_end = GRANULEROUNDUP(__pa(end - 1)) >> PAGE_SHIFT;
624 min_low_pfn = min(min_low_pfn, pfn_start);
625 max_low_pfn = max(max_low_pfn, pfn_end);
630 * Boot command-line option "nolwsys" can be used to disable the use of any light-weight
631 * system call handler. When this option is in effect, all fsyscalls will end up bubbling
632 * down into the kernel and calling the normal (heavy-weight) syscall handler. This is
633 * useful for performance testing, but conceivably could also come in handy for debugging
637 static int nolwsys __initdata;
640 nolwsys_setup (char *s)
646 __setup("nolwsys", nolwsys_setup);
653 BUG_ON(PTRS_PER_PGD * sizeof(pgd_t) != PAGE_SIZE);
654 BUG_ON(PTRS_PER_PMD * sizeof(pmd_t) != PAGE_SIZE);
655 BUG_ON(PTRS_PER_PTE * sizeof(pte_t) != PAGE_SIZE);
658 * This needs to be called _after_ the command line has been parsed but
659 * _before_ any drivers that may need the PCI DMA interface are
660 * initialized or bootmem has been freed.
662 #ifdef CONFIG_INTEL_IOMMU
663 detect_intel_iommu();
666 #ifdef CONFIG_SWIOTLB
670 #ifdef CONFIG_FLATMEM
674 set_max_mapnr(max_low_pfn);
675 high_memory = __va(max_low_pfn * PAGE_SIZE);
677 mem_init_print_info(NULL);
680 * For fsyscall entrpoints with no light-weight handler, use the ordinary
681 * (heavy-weight) handler, but mark it by setting bit 0, so the fsyscall entry
682 * code can tell them apart.
684 for (i = 0; i < NR_syscalls; ++i) {
685 extern unsigned long fsyscall_table[NR_syscalls];
686 extern unsigned long sys_call_table[NR_syscalls];
688 if (!fsyscall_table[i] || nolwsys)
689 fsyscall_table[i] = sys_call_table[i] | 1;
694 #ifdef CONFIG_MEMORY_HOTPLUG
695 int arch_add_memory(int nid, u64 start, u64 size,
696 struct mhp_params *params)
698 unsigned long start_pfn = start >> PAGE_SHIFT;
699 unsigned long nr_pages = size >> PAGE_SHIFT;
702 if (WARN_ON_ONCE(params->pgprot.pgprot != PAGE_KERNEL.pgprot))
705 ret = __add_pages(nid, start_pfn, nr_pages, params);
707 printk("%s: Problem encountered in __add_pages() as ret=%d\n",
713 void arch_remove_memory(int nid, u64 start, u64 size,
714 struct vmem_altmap *altmap)
716 unsigned long start_pfn = start >> PAGE_SHIFT;
717 unsigned long nr_pages = size >> PAGE_SHIFT;
719 __remove_pages(start_pfn, nr_pages, altmap);