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-noncoherent.h>
12 #include <linux/efi.h>
13 #include <linux/elf.h>
14 #include <linux/memblock.h>
16 #include <linux/sched/signal.h>
17 #include <linux/mmzone.h>
18 #include <linux/module.h>
19 #include <linux/personality.h>
20 #include <linux/reboot.h>
21 #include <linux/slab.h>
22 #include <linux/swap.h>
23 #include <linux/proc_fs.h>
24 #include <linux/bitops.h>
25 #include <linux/kexec.h>
29 #include <asm/machvec.h>
31 #include <asm/patch.h>
32 #include <asm/pgalloc.h>
34 #include <asm/sections.h>
36 #include <linux/uaccess.h>
37 #include <asm/unistd.h>
40 extern void ia64_tlb_init (void);
42 unsigned long MAX_DMA_ADDRESS = PAGE_OFFSET + 0x100000000UL;
44 #ifdef CONFIG_VIRTUAL_MEM_MAP
45 unsigned long VMALLOC_END = VMALLOC_END_INIT;
46 EXPORT_SYMBOL(VMALLOC_END);
47 struct page *vmem_map;
48 EXPORT_SYMBOL(vmem_map);
51 struct page *zero_page_memmap_ptr; /* map entry for zero page */
52 EXPORT_SYMBOL(zero_page_memmap_ptr);
55 __ia64_sync_icache_dcache (pte_t pte)
61 addr = (unsigned long) page_address(page);
63 if (test_bit(PG_arch_1, &page->flags))
64 return; /* i-cache is already coherent with d-cache */
66 flush_icache_range(addr, addr + (PAGE_SIZE << compound_order(page)));
67 set_bit(PG_arch_1, &page->flags); /* mark page as clean */
71 * Since DMA is i-cache coherent, any (complete) pages that were written via
72 * DMA can be marked as "clean" so that lazy_mmu_prot_update() doesn't have to
73 * flush them when they get mapped into an executable vm-area.
75 void arch_sync_dma_for_cpu(struct device *dev, phys_addr_t paddr,
76 size_t size, enum dma_data_direction dir)
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 down_write(¤t->mm->mmap_sem);
121 if (insert_vm_struct(current->mm, vma)) {
122 up_write(¤t->mm->mmap_sem);
126 up_write(¤t->mm->mmap_sem);
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 down_write(¤t->mm->mmap_sem);
139 if (insert_vm_struct(current->mm, vma)) {
140 up_write(¤t->mm->mmap_sem);
144 up_write(¤t->mm->mmap_sem);
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)
214 pgd = pgd_offset_k(address); /* note: this is NOT pgd_offset()! */
217 pud = pud_alloc(&init_mm, pgd, address);
220 pmd = pmd_alloc(&init_mm, pud, address);
223 pte = pte_alloc_kernel(pmd, address);
228 set_pte(pte, mk_pte(page, pgprot));
231 /* no need for flush_tlb */
241 * Map the gate page twice: once read-only to export the ELF
242 * headers etc. and once execute-only page to enable
243 * privilege-promotion via "epc":
245 page = virt_to_page(ia64_imva(__start_gate_section));
246 put_kernel_page(page, GATE_ADDR, PAGE_READONLY);
247 #ifdef HAVE_BUGGY_SEGREL
248 page = virt_to_page(ia64_imva(__start_gate_section + PAGE_SIZE));
249 put_kernel_page(page, GATE_ADDR + PAGE_SIZE, PAGE_GATE);
251 put_kernel_page(page, GATE_ADDR + PERCPU_PAGE_SIZE, PAGE_GATE);
252 /* Fill in the holes (if any) with read-only zero pages: */
256 for (addr = GATE_ADDR + PAGE_SIZE;
257 addr < GATE_ADDR + PERCPU_PAGE_SIZE;
260 put_kernel_page(ZERO_PAGE(0), addr,
262 put_kernel_page(ZERO_PAGE(0), addr + PERCPU_PAGE_SIZE,
270 static struct vm_area_struct gate_vma;
272 static int __init gate_vma_init(void)
274 vma_init(&gate_vma, NULL);
275 gate_vma.vm_start = FIXADDR_USER_START;
276 gate_vma.vm_end = FIXADDR_USER_END;
277 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
278 gate_vma.vm_page_prot = __P101;
282 __initcall(gate_vma_init);
284 struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
289 int in_gate_area_no_mm(unsigned long addr)
291 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
296 int in_gate_area(struct mm_struct *mm, unsigned long addr)
298 return in_gate_area_no_mm(addr);
301 void ia64_mmu_init(void *my_cpu_data)
303 unsigned long pta, impl_va_bits;
304 extern void tlb_init(void);
306 #ifdef CONFIG_DISABLE_VHPT
307 # define VHPT_ENABLE_BIT 0
309 # define VHPT_ENABLE_BIT 1
313 * Check if the virtually mapped linear page table (VMLPT) overlaps with a mapped
314 * address space. The IA-64 architecture guarantees that at least 50 bits of
315 * virtual address space are implemented but if we pick a large enough page size
316 * (e.g., 64KB), the mapped address space is big enough that it will overlap with
317 * VMLPT. I assume that once we run on machines big enough to warrant 64KB pages,
318 * IMPL_VA_MSB will be significantly bigger, so this is unlikely to become a
319 * problem in practice. Alternatively, we could truncate the top of the mapped
320 * address space to not permit mappings that would overlap with the VMLPT.
324 # define mapped_space_bits (3*(PAGE_SHIFT - pte_bits) + PAGE_SHIFT)
326 * The virtual page table has to cover the entire implemented address space within
327 * a region even though not all of this space may be mappable. The reason for
328 * this is that the Access bit and Dirty bit fault handlers perform
329 * non-speculative accesses to the virtual page table, so the address range of the
330 * virtual page table itself needs to be covered by virtual page table.
332 # define vmlpt_bits (impl_va_bits - PAGE_SHIFT + pte_bits)
333 # define POW2(n) (1ULL << (n))
335 impl_va_bits = ffz(~(local_cpu_data->unimpl_va_mask | (7UL << 61)));
337 if (impl_va_bits < 51 || impl_va_bits > 61)
338 panic("CPU has bogus IMPL_VA_MSB value of %lu!\n", impl_va_bits - 1);
340 * mapped_space_bits - PAGE_SHIFT is the total number of ptes we need,
341 * which must fit into "vmlpt_bits - pte_bits" slots. Second half of
342 * the test makes sure that our mapped space doesn't overlap the
343 * unimplemented hole in the middle of the region.
345 if ((mapped_space_bits - PAGE_SHIFT > vmlpt_bits - pte_bits) ||
346 (mapped_space_bits > impl_va_bits - 1))
347 panic("Cannot build a big enough virtual-linear page table"
348 " to cover mapped address space.\n"
349 " Try using a smaller page size.\n");
352 /* place the VMLPT at the end of each page-table mapped region: */
353 pta = POW2(61) - POW2(vmlpt_bits);
356 * Set the (virtually mapped linear) page table address. Bit
357 * 8 selects between the short and long format, bits 2-7 the
358 * size of the table, and bit 0 whether the VHPT walker is
361 ia64_set_pta(pta | (0 << 8) | (vmlpt_bits << 2) | VHPT_ENABLE_BIT);
365 #ifdef CONFIG_HUGETLB_PAGE
366 ia64_set_rr(HPAGE_REGION_BASE, HPAGE_SHIFT << 2);
371 #ifdef CONFIG_VIRTUAL_MEM_MAP
372 int vmemmap_find_next_valid_pfn(int node, int i)
374 unsigned long end_address, hole_next_pfn;
375 unsigned long stop_address;
376 pg_data_t *pgdat = NODE_DATA(node);
378 end_address = (unsigned long) &vmem_map[pgdat->node_start_pfn + i];
379 end_address = PAGE_ALIGN(end_address);
380 stop_address = (unsigned long) &vmem_map[pgdat_end_pfn(pgdat)];
388 pgd = pgd_offset_k(end_address);
389 if (pgd_none(*pgd)) {
390 end_address += PGDIR_SIZE;
394 pud = pud_offset(pgd, end_address);
395 if (pud_none(*pud)) {
396 end_address += PUD_SIZE;
400 pmd = pmd_offset(pud, end_address);
401 if (pmd_none(*pmd)) {
402 end_address += PMD_SIZE;
406 pte = pte_offset_kernel(pmd, end_address);
408 if (pte_none(*pte)) {
409 end_address += PAGE_SIZE;
411 if ((end_address < stop_address) &&
412 (end_address != ALIGN(end_address, 1UL << PMD_SHIFT)))
416 /* Found next valid vmem_map page */
418 } while (end_address < stop_address);
420 end_address = min(end_address, stop_address);
421 end_address = end_address - (unsigned long) vmem_map + sizeof(struct page) - 1;
422 hole_next_pfn = end_address / sizeof(struct page);
423 return hole_next_pfn - pgdat->node_start_pfn;
426 int __init create_mem_map_page_table(u64 start, u64 end, void *arg)
428 unsigned long address, start_page, end_page;
429 struct page *map_start, *map_end;
436 map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
437 map_end = vmem_map + (__pa(end) >> PAGE_SHIFT);
439 start_page = (unsigned long) map_start & PAGE_MASK;
440 end_page = PAGE_ALIGN((unsigned long) map_end);
441 node = paddr_to_nid(__pa(start));
443 for (address = start_page; address < end_page; address += PAGE_SIZE) {
444 pgd = pgd_offset_k(address);
446 pgd_populate(&init_mm, pgd, memblock_alloc_node(PAGE_SIZE, PAGE_SIZE, node));
447 pud = pud_offset(pgd, address);
450 pud_populate(&init_mm, pud, memblock_alloc_node(PAGE_SIZE, PAGE_SIZE, node));
451 pmd = pmd_offset(pud, address);
454 pmd_populate_kernel(&init_mm, pmd, memblock_alloc_node(PAGE_SIZE, PAGE_SIZE, node));
455 pte = pte_offset_kernel(pmd, address);
458 set_pte(pte, pfn_pte(__pa(memblock_alloc_node(PAGE_SIZE, PAGE_SIZE, node)) >> PAGE_SHIFT,
464 struct memmap_init_callback_data {
472 virtual_memmap_init(u64 start, u64 end, void *arg)
474 struct memmap_init_callback_data *args;
475 struct page *map_start, *map_end;
477 args = (struct memmap_init_callback_data *) arg;
478 map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
479 map_end = vmem_map + (__pa(end) >> PAGE_SHIFT);
481 if (map_start < args->start)
482 map_start = args->start;
483 if (map_end > args->end)
487 * We have to initialize "out of bounds" struct page elements that fit completely
488 * on the same pages that were allocated for the "in bounds" elements because they
489 * may be referenced later (and found to be "reserved").
491 map_start -= ((unsigned long) map_start & (PAGE_SIZE - 1)) / sizeof(struct page);
492 map_end += ((PAGE_ALIGN((unsigned long) map_end) - (unsigned long) map_end)
493 / sizeof(struct page));
495 if (map_start < map_end)
496 memmap_init_zone((unsigned long)(map_end - map_start),
497 args->nid, args->zone, page_to_pfn(map_start),
503 memmap_init (unsigned long size, int nid, unsigned long zone,
504 unsigned long start_pfn)
507 memmap_init_zone(size, nid, zone, start_pfn, MEMMAP_EARLY,
511 struct memmap_init_callback_data args;
513 start = pfn_to_page(start_pfn);
515 args.end = start + size;
519 efi_memmap_walk(virtual_memmap_init, &args);
524 ia64_pfn_valid (unsigned long pfn)
527 struct page *pg = pfn_to_page(pfn);
529 return (__get_user(byte, (char __user *) pg) == 0)
530 && ((((u64)pg & PAGE_MASK) == (((u64)(pg + 1) - 1) & PAGE_MASK))
531 || (__get_user(byte, (char __user *) (pg + 1) - 1) == 0));
533 EXPORT_SYMBOL(ia64_pfn_valid);
535 int __init find_largest_hole(u64 start, u64 end, void *arg)
539 static u64 last_end = PAGE_OFFSET;
541 /* NOTE: this algorithm assumes efi memmap table is ordered */
543 if (*max_gap < (start - last_end))
544 *max_gap = start - last_end;
549 #endif /* CONFIG_VIRTUAL_MEM_MAP */
551 int __init register_active_ranges(u64 start, u64 len, int nid)
553 u64 end = start + len;
556 if (start > crashk_res.start && start < crashk_res.end)
557 start = crashk_res.end;
558 if (end > crashk_res.start && end < crashk_res.end)
559 end = crashk_res.start;
563 memblock_add_node(__pa(start), end - start, nid);
568 find_max_min_low_pfn (u64 start, u64 end, void *arg)
570 unsigned long pfn_start, pfn_end;
571 #ifdef CONFIG_FLATMEM
572 pfn_start = (PAGE_ALIGN(__pa(start))) >> PAGE_SHIFT;
573 pfn_end = (PAGE_ALIGN(__pa(end - 1))) >> PAGE_SHIFT;
575 pfn_start = GRANULEROUNDDOWN(__pa(start)) >> PAGE_SHIFT;
576 pfn_end = GRANULEROUNDUP(__pa(end - 1)) >> PAGE_SHIFT;
578 min_low_pfn = min(min_low_pfn, pfn_start);
579 max_low_pfn = max(max_low_pfn, pfn_end);
584 * Boot command-line option "nolwsys" can be used to disable the use of any light-weight
585 * system call handler. When this option is in effect, all fsyscalls will end up bubbling
586 * down into the kernel and calling the normal (heavy-weight) syscall handler. This is
587 * useful for performance testing, but conceivably could also come in handy for debugging
591 static int nolwsys __initdata;
594 nolwsys_setup (char *s)
600 __setup("nolwsys", nolwsys_setup);
607 BUG_ON(PTRS_PER_PGD * sizeof(pgd_t) != PAGE_SIZE);
608 BUG_ON(PTRS_PER_PMD * sizeof(pmd_t) != PAGE_SIZE);
609 BUG_ON(PTRS_PER_PTE * sizeof(pte_t) != PAGE_SIZE);
613 * This needs to be called _after_ the command line has been parsed but _before_
614 * any drivers that may need the PCI DMA interface are initialized or bootmem has
620 #ifdef CONFIG_FLATMEM
624 set_max_mapnr(max_low_pfn);
625 high_memory = __va(max_low_pfn * PAGE_SIZE);
627 mem_init_print_info(NULL);
630 * For fsyscall entrpoints with no light-weight handler, use the ordinary
631 * (heavy-weight) handler, but mark it by setting bit 0, so the fsyscall entry
632 * code can tell them apart.
634 for (i = 0; i < NR_syscalls; ++i) {
635 extern unsigned long fsyscall_table[NR_syscalls];
636 extern unsigned long sys_call_table[NR_syscalls];
638 if (!fsyscall_table[i] || nolwsys)
639 fsyscall_table[i] = sys_call_table[i] | 1;
644 #ifdef CONFIG_MEMORY_HOTPLUG
645 int arch_add_memory(int nid, u64 start, u64 size, struct vmem_altmap *altmap,
648 unsigned long start_pfn = start >> PAGE_SHIFT;
649 unsigned long nr_pages = size >> PAGE_SHIFT;
652 ret = __add_pages(nid, start_pfn, nr_pages, altmap, want_memblock);
654 printk("%s: Problem encountered in __add_pages() as ret=%d\n",
660 #ifdef CONFIG_MEMORY_HOTREMOVE
661 int arch_remove_memory(u64 start, u64 size, struct vmem_altmap *altmap)
663 unsigned long start_pfn = start >> PAGE_SHIFT;
664 unsigned long nr_pages = size >> PAGE_SHIFT;
668 zone = page_zone(pfn_to_page(start_pfn));
669 ret = __remove_pages(zone, start_pfn, nr_pages, altmap);
671 pr_warn("%s: Problem encountered in __remove_pages() as"
672 " ret=%d\n", __func__, ret);