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>
33 #include <asm/patch.h>
34 #include <asm/pgalloc.h>
36 #include <asm/sections.h>
38 #include <linux/uaccess.h>
39 #include <asm/unistd.h>
42 extern void ia64_tlb_init (void);
44 unsigned long MAX_DMA_ADDRESS = PAGE_OFFSET + 0x100000000UL;
46 #ifdef CONFIG_VIRTUAL_MEM_MAP
47 unsigned long VMALLOC_END = VMALLOC_END_INIT;
48 EXPORT_SYMBOL(VMALLOC_END);
49 struct page *vmem_map;
50 EXPORT_SYMBOL(vmem_map);
53 struct page *zero_page_memmap_ptr; /* map entry for zero page */
54 EXPORT_SYMBOL(zero_page_memmap_ptr);
57 __ia64_sync_icache_dcache (pte_t pte)
63 addr = (unsigned long) page_address(page);
65 if (test_bit(PG_arch_1, &page->flags))
66 return; /* i-cache is already coherent with d-cache */
68 flush_icache_range(addr, addr + page_size(page));
69 set_bit(PG_arch_1, &page->flags); /* mark page as clean */
73 * Since DMA is i-cache coherent, any (complete) pages that were written via
74 * DMA can be marked as "clean" so that lazy_mmu_prot_update() doesn't have to
75 * flush them when they get mapped into an executable vm-area.
77 void arch_dma_mark_clean(phys_addr_t paddr, size_t size)
79 unsigned long pfn = PHYS_PFN(paddr);
82 set_bit(PG_arch_1, &pfn_to_page(pfn)->flags);
83 } while (++pfn <= PHYS_PFN(paddr + size - 1));
87 ia64_set_rbs_bot (void)
89 unsigned long stack_size = rlimit_max(RLIMIT_STACK) & -16;
91 if (stack_size > MAX_USER_STACK_SIZE)
92 stack_size = MAX_USER_STACK_SIZE;
93 current->thread.rbs_bot = PAGE_ALIGN(current->mm->start_stack - stack_size);
97 * This performs some platform-dependent address space initialization.
98 * On IA-64, we want to setup the VM area for the register backing
99 * store (which grows upwards) and install the gateway page which is
100 * used for signal trampolines, etc.
103 ia64_init_addr_space (void)
105 struct vm_area_struct *vma;
110 * If we're out of memory and kmem_cache_alloc() returns NULL, we simply ignore
111 * the problem. When the process attempts to write to the register backing store
112 * for the first time, it will get a SEGFAULT in this case.
114 vma = vm_area_alloc(current->mm);
116 vma_set_anonymous(vma);
117 vma->vm_start = current->thread.rbs_bot & PAGE_MASK;
118 vma->vm_end = vma->vm_start + PAGE_SIZE;
119 vma->vm_flags = VM_DATA_DEFAULT_FLAGS|VM_GROWSUP|VM_ACCOUNT;
120 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
121 mmap_write_lock(current->mm);
122 if (insert_vm_struct(current->mm, vma)) {
123 mmap_write_unlock(current->mm);
127 mmap_write_unlock(current->mm);
130 /* map NaT-page at address zero to speed up speculative dereferencing of NULL: */
131 if (!(current->personality & MMAP_PAGE_ZERO)) {
132 vma = vm_area_alloc(current->mm);
134 vma_set_anonymous(vma);
135 vma->vm_end = PAGE_SIZE;
136 vma->vm_page_prot = __pgprot(pgprot_val(PAGE_READONLY) | _PAGE_MA_NAT);
137 vma->vm_flags = VM_READ | VM_MAYREAD | VM_IO |
138 VM_DONTEXPAND | VM_DONTDUMP;
139 mmap_write_lock(current->mm);
140 if (insert_vm_struct(current->mm, vma)) {
141 mmap_write_unlock(current->mm);
145 mmap_write_unlock(current->mm);
153 free_reserved_area(ia64_imva(__init_begin), ia64_imva(__init_end),
154 -1, "unused kernel");
158 free_initrd_mem (unsigned long start, unsigned long end)
161 * EFI uses 4KB pages while the kernel can use 4KB or bigger.
162 * Thus EFI and the kernel may have different page sizes. It is
163 * therefore possible to have the initrd share the same page as
164 * the end of the kernel (given current setup).
166 * To avoid freeing/using the wrong page (kernel sized) we:
167 * - align up the beginning of initrd
168 * - align down the end of initrd
171 * |=============| a000
177 * |=============| 8000
180 * |/////////////| 7000
183 * |=============| 6000
186 * K=kernel using 8KB pages
188 * In this example, we must free page 8000 ONLY. So we must align up
189 * initrd_start and keep initrd_end as is.
191 start = PAGE_ALIGN(start);
192 end = end & PAGE_MASK;
195 printk(KERN_INFO "Freeing initrd memory: %ldkB freed\n", (end - start) >> 10);
197 for (; start < end; start += PAGE_SIZE) {
198 if (!virt_addr_valid(start))
200 free_reserved_page(virt_to_page(start));
205 * This installs a clean page in the kernel's page table.
207 static struct page * __init
208 put_kernel_page (struct page *page, unsigned long address, pgprot_t pgprot)
216 pgd = pgd_offset_k(address); /* note: this is NOT pgd_offset()! */
219 p4d = p4d_alloc(&init_mm, pgd, address);
222 pud = pud_alloc(&init_mm, p4d, address);
225 pmd = pmd_alloc(&init_mm, pud, address);
228 pte = pte_alloc_kernel(pmd, address);
233 set_pte(pte, mk_pte(page, pgprot));
236 /* no need for flush_tlb */
246 * Map the gate page twice: once read-only to export the ELF
247 * headers etc. and once execute-only page to enable
248 * privilege-promotion via "epc":
250 page = virt_to_page(ia64_imva(__start_gate_section));
251 put_kernel_page(page, GATE_ADDR, PAGE_READONLY);
252 #ifdef HAVE_BUGGY_SEGREL
253 page = virt_to_page(ia64_imva(__start_gate_section + PAGE_SIZE));
254 put_kernel_page(page, GATE_ADDR + PAGE_SIZE, PAGE_GATE);
256 put_kernel_page(page, GATE_ADDR + PERCPU_PAGE_SIZE, PAGE_GATE);
257 /* Fill in the holes (if any) with read-only zero pages: */
261 for (addr = GATE_ADDR + PAGE_SIZE;
262 addr < GATE_ADDR + PERCPU_PAGE_SIZE;
265 put_kernel_page(ZERO_PAGE(0), addr,
267 put_kernel_page(ZERO_PAGE(0), addr + PERCPU_PAGE_SIZE,
275 static struct vm_area_struct gate_vma;
277 static int __init gate_vma_init(void)
279 vma_init(&gate_vma, NULL);
280 gate_vma.vm_start = FIXADDR_USER_START;
281 gate_vma.vm_end = FIXADDR_USER_END;
282 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
283 gate_vma.vm_page_prot = __P101;
287 __initcall(gate_vma_init);
289 struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
294 int in_gate_area_no_mm(unsigned long addr)
296 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
301 int in_gate_area(struct mm_struct *mm, unsigned long addr)
303 return in_gate_area_no_mm(addr);
306 void ia64_mmu_init(void *my_cpu_data)
308 unsigned long pta, impl_va_bits;
309 extern void tlb_init(void);
311 #ifdef CONFIG_DISABLE_VHPT
312 # define VHPT_ENABLE_BIT 0
314 # define VHPT_ENABLE_BIT 1
318 * Check if the virtually mapped linear page table (VMLPT) overlaps with a mapped
319 * address space. The IA-64 architecture guarantees that at least 50 bits of
320 * virtual address space are implemented but if we pick a large enough page size
321 * (e.g., 64KB), the mapped address space is big enough that it will overlap with
322 * VMLPT. I assume that once we run on machines big enough to warrant 64KB pages,
323 * IMPL_VA_MSB will be significantly bigger, so this is unlikely to become a
324 * problem in practice. Alternatively, we could truncate the top of the mapped
325 * address space to not permit mappings that would overlap with the VMLPT.
329 # define mapped_space_bits (3*(PAGE_SHIFT - pte_bits) + PAGE_SHIFT)
331 * The virtual page table has to cover the entire implemented address space within
332 * a region even though not all of this space may be mappable. The reason for
333 * this is that the Access bit and Dirty bit fault handlers perform
334 * non-speculative accesses to the virtual page table, so the address range of the
335 * virtual page table itself needs to be covered by virtual page table.
337 # define vmlpt_bits (impl_va_bits - PAGE_SHIFT + pte_bits)
338 # define POW2(n) (1ULL << (n))
340 impl_va_bits = ffz(~(local_cpu_data->unimpl_va_mask | (7UL << 61)));
342 if (impl_va_bits < 51 || impl_va_bits > 61)
343 panic("CPU has bogus IMPL_VA_MSB value of %lu!\n", impl_va_bits - 1);
345 * mapped_space_bits - PAGE_SHIFT is the total number of ptes we need,
346 * which must fit into "vmlpt_bits - pte_bits" slots. Second half of
347 * the test makes sure that our mapped space doesn't overlap the
348 * unimplemented hole in the middle of the region.
350 if ((mapped_space_bits - PAGE_SHIFT > vmlpt_bits - pte_bits) ||
351 (mapped_space_bits > impl_va_bits - 1))
352 panic("Cannot build a big enough virtual-linear page table"
353 " to cover mapped address space.\n"
354 " Try using a smaller page size.\n");
357 /* place the VMLPT at the end of each page-table mapped region: */
358 pta = POW2(61) - POW2(vmlpt_bits);
361 * Set the (virtually mapped linear) page table address. Bit
362 * 8 selects between the short and long format, bits 2-7 the
363 * size of the table, and bit 0 whether the VHPT walker is
366 ia64_set_pta(pta | (0 << 8) | (vmlpt_bits << 2) | VHPT_ENABLE_BIT);
370 #ifdef CONFIG_HUGETLB_PAGE
371 ia64_set_rr(HPAGE_REGION_BASE, HPAGE_SHIFT << 2);
376 #ifdef CONFIG_VIRTUAL_MEM_MAP
377 int vmemmap_find_next_valid_pfn(int node, int i)
379 unsigned long end_address, hole_next_pfn;
380 unsigned long stop_address;
381 pg_data_t *pgdat = NODE_DATA(node);
383 end_address = (unsigned long) &vmem_map[pgdat->node_start_pfn + i];
384 end_address = PAGE_ALIGN(end_address);
385 stop_address = (unsigned long) &vmem_map[pgdat_end_pfn(pgdat)];
394 pgd = pgd_offset_k(end_address);
395 if (pgd_none(*pgd)) {
396 end_address += PGDIR_SIZE;
400 p4d = p4d_offset(pgd, end_address);
401 if (p4d_none(*p4d)) {
402 end_address += P4D_SIZE;
406 pud = pud_offset(p4d, end_address);
407 if (pud_none(*pud)) {
408 end_address += PUD_SIZE;
412 pmd = pmd_offset(pud, end_address);
413 if (pmd_none(*pmd)) {
414 end_address += PMD_SIZE;
418 pte = pte_offset_kernel(pmd, end_address);
420 if (pte_none(*pte)) {
421 end_address += PAGE_SIZE;
423 if ((end_address < stop_address) &&
424 (end_address != ALIGN(end_address, 1UL << PMD_SHIFT)))
428 /* Found next valid vmem_map page */
430 } while (end_address < stop_address);
432 end_address = min(end_address, stop_address);
433 end_address = end_address - (unsigned long) vmem_map + sizeof(struct page) - 1;
434 hole_next_pfn = end_address / sizeof(struct page);
435 return hole_next_pfn - pgdat->node_start_pfn;
438 int __init create_mem_map_page_table(u64 start, u64 end, void *arg)
440 unsigned long address, start_page, end_page;
441 struct page *map_start, *map_end;
449 map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
450 map_end = vmem_map + (__pa(end) >> PAGE_SHIFT);
452 start_page = (unsigned long) map_start & PAGE_MASK;
453 end_page = PAGE_ALIGN((unsigned long) map_end);
454 node = paddr_to_nid(__pa(start));
456 for (address = start_page; address < end_page; address += PAGE_SIZE) {
457 pgd = pgd_offset_k(address);
458 if (pgd_none(*pgd)) {
459 p4d = memblock_alloc_node(PAGE_SIZE, PAGE_SIZE, node);
462 pgd_populate(&init_mm, pgd, p4d);
464 p4d = p4d_offset(pgd, address);
466 if (p4d_none(*p4d)) {
467 pud = memblock_alloc_node(PAGE_SIZE, PAGE_SIZE, node);
470 p4d_populate(&init_mm, p4d, pud);
472 pud = pud_offset(p4d, address);
474 if (pud_none(*pud)) {
475 pmd = memblock_alloc_node(PAGE_SIZE, PAGE_SIZE, node);
478 pud_populate(&init_mm, pud, pmd);
480 pmd = pmd_offset(pud, address);
482 if (pmd_none(*pmd)) {
483 pte = memblock_alloc_node(PAGE_SIZE, PAGE_SIZE, node);
486 pmd_populate_kernel(&init_mm, pmd, pte);
488 pte = pte_offset_kernel(pmd, address);
490 if (pte_none(*pte)) {
491 void *page = memblock_alloc_node(PAGE_SIZE, PAGE_SIZE,
495 set_pte(pte, pfn_pte(__pa(page) >> PAGE_SHIFT,
502 panic("%s: Failed to allocate %lu bytes align=0x%lx nid=%d\n",
503 __func__, PAGE_SIZE, PAGE_SIZE, node);
507 struct memmap_init_callback_data {
515 virtual_memmap_init(u64 start, u64 end, void *arg)
517 struct memmap_init_callback_data *args;
518 struct page *map_start, *map_end;
520 args = (struct memmap_init_callback_data *) arg;
521 map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
522 map_end = vmem_map + (__pa(end) >> PAGE_SHIFT);
524 if (map_start < args->start)
525 map_start = args->start;
526 if (map_end > args->end)
530 * We have to initialize "out of bounds" struct page elements that fit completely
531 * on the same pages that were allocated for the "in bounds" elements because they
532 * may be referenced later (and found to be "reserved").
534 map_start -= ((unsigned long) map_start & (PAGE_SIZE - 1)) / sizeof(struct page);
535 map_end += ((PAGE_ALIGN((unsigned long) map_end) - (unsigned long) map_end)
536 / sizeof(struct page));
538 if (map_start < map_end)
539 memmap_init_zone((unsigned long)(map_end - map_start),
540 args->nid, args->zone, page_to_pfn(map_start),
541 MEMINIT_EARLY, NULL, MIGRATE_MOVABLE);
546 memmap_init (unsigned long size, int nid, unsigned long zone,
547 unsigned long start_pfn)
550 memmap_init_zone(size, nid, zone, start_pfn,
551 MEMINIT_EARLY, NULL, MIGRATE_MOVABLE);
554 struct memmap_init_callback_data args;
556 start = pfn_to_page(start_pfn);
558 args.end = start + size;
562 efi_memmap_walk(virtual_memmap_init, &args);
567 ia64_pfn_valid (unsigned long pfn)
570 struct page *pg = pfn_to_page(pfn);
572 return (__get_user(byte, (char __user *) pg) == 0)
573 && ((((u64)pg & PAGE_MASK) == (((u64)(pg + 1) - 1) & PAGE_MASK))
574 || (__get_user(byte, (char __user *) (pg + 1) - 1) == 0));
576 EXPORT_SYMBOL(ia64_pfn_valid);
578 #endif /* CONFIG_VIRTUAL_MEM_MAP */
580 int __init register_active_ranges(u64 start, u64 len, int nid)
582 u64 end = start + len;
585 if (start > crashk_res.start && start < crashk_res.end)
586 start = crashk_res.end;
587 if (end > crashk_res.start && end < crashk_res.end)
588 end = crashk_res.start;
592 memblock_add_node(__pa(start), end - start, nid);
597 find_max_min_low_pfn (u64 start, u64 end, void *arg)
599 unsigned long pfn_start, pfn_end;
600 #ifdef CONFIG_FLATMEM
601 pfn_start = (PAGE_ALIGN(__pa(start))) >> PAGE_SHIFT;
602 pfn_end = (PAGE_ALIGN(__pa(end - 1))) >> PAGE_SHIFT;
604 pfn_start = GRANULEROUNDDOWN(__pa(start)) >> PAGE_SHIFT;
605 pfn_end = GRANULEROUNDUP(__pa(end - 1)) >> PAGE_SHIFT;
607 min_low_pfn = min(min_low_pfn, pfn_start);
608 max_low_pfn = max(max_low_pfn, pfn_end);
613 * Boot command-line option "nolwsys" can be used to disable the use of any light-weight
614 * system call handler. When this option is in effect, all fsyscalls will end up bubbling
615 * down into the kernel and calling the normal (heavy-weight) syscall handler. This is
616 * useful for performance testing, but conceivably could also come in handy for debugging
620 static int nolwsys __initdata;
623 nolwsys_setup (char *s)
629 __setup("nolwsys", nolwsys_setup);
636 BUG_ON(PTRS_PER_PGD * sizeof(pgd_t) != PAGE_SIZE);
637 BUG_ON(PTRS_PER_PMD * sizeof(pmd_t) != PAGE_SIZE);
638 BUG_ON(PTRS_PER_PTE * sizeof(pte_t) != PAGE_SIZE);
641 * This needs to be called _after_ the command line has been parsed but
642 * _before_ any drivers that may need the PCI DMA interface are
643 * initialized or bootmem has been freed.
645 #ifdef CONFIG_INTEL_IOMMU
646 detect_intel_iommu();
649 #ifdef CONFIG_SWIOTLB
653 #ifdef CONFIG_FLATMEM
657 set_max_mapnr(max_low_pfn);
658 high_memory = __va(max_low_pfn * PAGE_SIZE);
660 mem_init_print_info(NULL);
663 * For fsyscall entrpoints with no light-weight handler, use the ordinary
664 * (heavy-weight) handler, but mark it by setting bit 0, so the fsyscall entry
665 * code can tell them apart.
667 for (i = 0; i < NR_syscalls; ++i) {
668 extern unsigned long fsyscall_table[NR_syscalls];
669 extern unsigned long sys_call_table[NR_syscalls];
671 if (!fsyscall_table[i] || nolwsys)
672 fsyscall_table[i] = sys_call_table[i] | 1;
677 #ifdef CONFIG_MEMORY_HOTPLUG
678 int arch_add_memory(int nid, u64 start, u64 size,
679 struct mhp_params *params)
681 unsigned long start_pfn = start >> PAGE_SHIFT;
682 unsigned long nr_pages = size >> PAGE_SHIFT;
685 if (WARN_ON_ONCE(params->pgprot.pgprot != PAGE_KERNEL.pgprot))
688 ret = __add_pages(nid, start_pfn, nr_pages, params);
690 printk("%s: Problem encountered in __add_pages() as ret=%d\n",
696 void arch_remove_memory(int nid, u64 start, u64 size,
697 struct vmem_altmap *altmap)
699 unsigned long start_pfn = start >> PAGE_SHIFT;
700 unsigned long nr_pages = size >> PAGE_SHIFT;
702 __remove_pages(start_pfn, nr_pages, altmap);