mm: fix exec activate_mm vs TLB shootdown and lazy tlb switching race
[linux/fpc-iii.git] / arch / ia64 / mm / init.c
blob7af4e05bb61e785989dac50c236b2cef64713f8f
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3 * Initialize MMU support.
5 * Copyright (C) 1998-2003 Hewlett-Packard Co
6 * David Mosberger-Tang <davidm@hpl.hp.com>
7 */
8 #include <linux/kernel.h>
9 #include <linux/init.h>
11 #include <linux/bootmem.h>
12 #include <linux/efi.h>
13 #include <linux/elf.h>
14 #include <linux/memblock.h>
15 #include <linux/mm.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>
27 #include <asm/dma.h>
28 #include <asm/io.h>
29 #include <asm/machvec.h>
30 #include <asm/numa.h>
31 #include <asm/patch.h>
32 #include <asm/pgalloc.h>
33 #include <asm/sal.h>
34 #include <asm/sections.h>
35 #include <asm/tlb.h>
36 #include <linux/uaccess.h>
37 #include <asm/unistd.h>
38 #include <asm/mca.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);
49 #endif
51 struct page *zero_page_memmap_ptr; /* map entry for zero page */
52 EXPORT_SYMBOL(zero_page_memmap_ptr);
54 void
55 __ia64_sync_icache_dcache (pte_t pte)
57 unsigned long addr;
58 struct page *page;
60 page = pte_page(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
76 dma_mark_clean(void *addr, size_t size)
78 unsigned long pg_addr, end;
80 pg_addr = PAGE_ALIGN((unsigned long) addr);
81 end = (unsigned long) addr + size;
82 while (pg_addr + PAGE_SIZE <= end) {
83 struct page *page = virt_to_page(pg_addr);
84 set_bit(PG_arch_1, &page->flags);
85 pg_addr += PAGE_SIZE;
89 inline void
90 ia64_set_rbs_bot (void)
92 unsigned long stack_size = rlimit_max(RLIMIT_STACK) & -16;
94 if (stack_size > MAX_USER_STACK_SIZE)
95 stack_size = MAX_USER_STACK_SIZE;
96 current->thread.rbs_bot = PAGE_ALIGN(current->mm->start_stack - stack_size);
100 * This performs some platform-dependent address space initialization.
101 * On IA-64, we want to setup the VM area for the register backing
102 * store (which grows upwards) and install the gateway page which is
103 * used for signal trampolines, etc.
105 void
106 ia64_init_addr_space (void)
108 struct vm_area_struct *vma;
110 ia64_set_rbs_bot();
113 * If we're out of memory and kmem_cache_alloc() returns NULL, we simply ignore
114 * the problem. When the process attempts to write to the register backing store
115 * for the first time, it will get a SEGFAULT in this case.
117 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
118 if (vma) {
119 INIT_LIST_HEAD(&vma->anon_vma_chain);
120 vma->vm_mm = current->mm;
121 vma->vm_start = current->thread.rbs_bot & PAGE_MASK;
122 vma->vm_end = vma->vm_start + PAGE_SIZE;
123 vma->vm_flags = VM_DATA_DEFAULT_FLAGS|VM_GROWSUP|VM_ACCOUNT;
124 vma->vm_page_prot = vm_get_page_prot(vma->vm_flags);
125 down_write(&current->mm->mmap_sem);
126 if (insert_vm_struct(current->mm, vma)) {
127 up_write(&current->mm->mmap_sem);
128 kmem_cache_free(vm_area_cachep, vma);
129 return;
131 up_write(&current->mm->mmap_sem);
134 /* map NaT-page at address zero to speed up speculative dereferencing of NULL: */
135 if (!(current->personality & MMAP_PAGE_ZERO)) {
136 vma = kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL);
137 if (vma) {
138 INIT_LIST_HEAD(&vma->anon_vma_chain);
139 vma->vm_mm = current->mm;
140 vma->vm_end = PAGE_SIZE;
141 vma->vm_page_prot = __pgprot(pgprot_val(PAGE_READONLY) | _PAGE_MA_NAT);
142 vma->vm_flags = VM_READ | VM_MAYREAD | VM_IO |
143 VM_DONTEXPAND | VM_DONTDUMP;
144 down_write(&current->mm->mmap_sem);
145 if (insert_vm_struct(current->mm, vma)) {
146 up_write(&current->mm->mmap_sem);
147 kmem_cache_free(vm_area_cachep, vma);
148 return;
150 up_write(&current->mm->mmap_sem);
155 void
156 free_initmem (void)
158 free_reserved_area(ia64_imva(__init_begin), ia64_imva(__init_end),
159 -1, "unused kernel");
162 void __init
163 free_initrd_mem (unsigned long start, unsigned long end)
166 * EFI uses 4KB pages while the kernel can use 4KB or bigger.
167 * Thus EFI and the kernel may have different page sizes. It is
168 * therefore possible to have the initrd share the same page as
169 * the end of the kernel (given current setup).
171 * To avoid freeing/using the wrong page (kernel sized) we:
172 * - align up the beginning of initrd
173 * - align down the end of initrd
175 * | |
176 * |=============| a000
177 * | |
178 * | |
179 * | | 9000
180 * |/////////////|
181 * |/////////////|
182 * |=============| 8000
183 * |///INITRD////|
184 * |/////////////|
185 * |/////////////| 7000
186 * | |
187 * |KKKKKKKKKKKKK|
188 * |=============| 6000
189 * |KKKKKKKKKKKKK|
190 * |KKKKKKKKKKKKK|
191 * K=kernel using 8KB pages
193 * In this example, we must free page 8000 ONLY. So we must align up
194 * initrd_start and keep initrd_end as is.
196 start = PAGE_ALIGN(start);
197 end = end & PAGE_MASK;
199 if (start < end)
200 printk(KERN_INFO "Freeing initrd memory: %ldkB freed\n", (end - start) >> 10);
202 for (; start < end; start += PAGE_SIZE) {
203 if (!virt_addr_valid(start))
204 continue;
205 free_reserved_page(virt_to_page(start));
210 * This installs a clean page in the kernel's page table.
212 static struct page * __init
213 put_kernel_page (struct page *page, unsigned long address, pgprot_t pgprot)
215 pgd_t *pgd;
216 pud_t *pud;
217 pmd_t *pmd;
218 pte_t *pte;
220 pgd = pgd_offset_k(address); /* note: this is NOT pgd_offset()! */
223 pud = pud_alloc(&init_mm, pgd, address);
224 if (!pud)
225 goto out;
226 pmd = pmd_alloc(&init_mm, pud, address);
227 if (!pmd)
228 goto out;
229 pte = pte_alloc_kernel(pmd, address);
230 if (!pte)
231 goto out;
232 if (!pte_none(*pte))
233 goto out;
234 set_pte(pte, mk_pte(page, pgprot));
236 out:
237 /* no need for flush_tlb */
238 return page;
241 static void __init
242 setup_gate (void)
244 struct page *page;
247 * Map the gate page twice: once read-only to export the ELF
248 * headers etc. and once execute-only page to enable
249 * privilege-promotion via "epc":
251 page = virt_to_page(ia64_imva(__start_gate_section));
252 put_kernel_page(page, GATE_ADDR, PAGE_READONLY);
253 #ifdef HAVE_BUGGY_SEGREL
254 page = virt_to_page(ia64_imva(__start_gate_section + PAGE_SIZE));
255 put_kernel_page(page, GATE_ADDR + PAGE_SIZE, PAGE_GATE);
256 #else
257 put_kernel_page(page, GATE_ADDR + PERCPU_PAGE_SIZE, PAGE_GATE);
258 /* Fill in the holes (if any) with read-only zero pages: */
260 unsigned long addr;
262 for (addr = GATE_ADDR + PAGE_SIZE;
263 addr < GATE_ADDR + PERCPU_PAGE_SIZE;
264 addr += PAGE_SIZE)
266 put_kernel_page(ZERO_PAGE(0), addr,
267 PAGE_READONLY);
268 put_kernel_page(ZERO_PAGE(0), addr + PERCPU_PAGE_SIZE,
269 PAGE_READONLY);
272 #endif
273 ia64_patch_gate();
276 static struct vm_area_struct gate_vma;
278 static int __init gate_vma_init(void)
280 gate_vma.vm_mm = NULL;
281 gate_vma.vm_start = FIXADDR_USER_START;
282 gate_vma.vm_end = FIXADDR_USER_END;
283 gate_vma.vm_flags = VM_READ | VM_MAYREAD | VM_EXEC | VM_MAYEXEC;
284 gate_vma.vm_page_prot = __P101;
286 return 0;
288 __initcall(gate_vma_init);
290 struct vm_area_struct *get_gate_vma(struct mm_struct *mm)
292 return &gate_vma;
295 int in_gate_area_no_mm(unsigned long addr)
297 if ((addr >= FIXADDR_USER_START) && (addr < FIXADDR_USER_END))
298 return 1;
299 return 0;
302 int in_gate_area(struct mm_struct *mm, unsigned long addr)
304 return in_gate_area_no_mm(addr);
307 void ia64_mmu_init(void *my_cpu_data)
309 unsigned long pta, impl_va_bits;
310 extern void tlb_init(void);
312 #ifdef CONFIG_DISABLE_VHPT
313 # define VHPT_ENABLE_BIT 0
314 #else
315 # define VHPT_ENABLE_BIT 1
316 #endif
319 * Check if the virtually mapped linear page table (VMLPT) overlaps with a mapped
320 * address space. The IA-64 architecture guarantees that at least 50 bits of
321 * virtual address space are implemented but if we pick a large enough page size
322 * (e.g., 64KB), the mapped address space is big enough that it will overlap with
323 * VMLPT. I assume that once we run on machines big enough to warrant 64KB pages,
324 * IMPL_VA_MSB will be significantly bigger, so this is unlikely to become a
325 * problem in practice. Alternatively, we could truncate the top of the mapped
326 * address space to not permit mappings that would overlap with the VMLPT.
327 * --davidm 00/12/06
329 # define pte_bits 3
330 # define mapped_space_bits (3*(PAGE_SHIFT - pte_bits) + PAGE_SHIFT)
332 * The virtual page table has to cover the entire implemented address space within
333 * a region even though not all of this space may be mappable. The reason for
334 * this is that the Access bit and Dirty bit fault handlers perform
335 * non-speculative accesses to the virtual page table, so the address range of the
336 * virtual page table itself needs to be covered by virtual page table.
338 # define vmlpt_bits (impl_va_bits - PAGE_SHIFT + pte_bits)
339 # define POW2(n) (1ULL << (n))
341 impl_va_bits = ffz(~(local_cpu_data->unimpl_va_mask | (7UL << 61)));
343 if (impl_va_bits < 51 || impl_va_bits > 61)
344 panic("CPU has bogus IMPL_VA_MSB value of %lu!\n", impl_va_bits - 1);
346 * mapped_space_bits - PAGE_SHIFT is the total number of ptes we need,
347 * which must fit into "vmlpt_bits - pte_bits" slots. Second half of
348 * the test makes sure that our mapped space doesn't overlap the
349 * unimplemented hole in the middle of the region.
351 if ((mapped_space_bits - PAGE_SHIFT > vmlpt_bits - pte_bits) ||
352 (mapped_space_bits > impl_va_bits - 1))
353 panic("Cannot build a big enough virtual-linear page table"
354 " to cover mapped address space.\n"
355 " Try using a smaller page size.\n");
358 /* place the VMLPT at the end of each page-table mapped region: */
359 pta = POW2(61) - POW2(vmlpt_bits);
362 * Set the (virtually mapped linear) page table address. Bit
363 * 8 selects between the short and long format, bits 2-7 the
364 * size of the table, and bit 0 whether the VHPT walker is
365 * enabled.
367 ia64_set_pta(pta | (0 << 8) | (vmlpt_bits << 2) | VHPT_ENABLE_BIT);
369 ia64_tlb_init();
371 #ifdef CONFIG_HUGETLB_PAGE
372 ia64_set_rr(HPAGE_REGION_BASE, HPAGE_SHIFT << 2);
373 ia64_srlz_d();
374 #endif
377 #ifdef CONFIG_VIRTUAL_MEM_MAP
378 int vmemmap_find_next_valid_pfn(int node, int i)
380 unsigned long end_address, hole_next_pfn;
381 unsigned long stop_address;
382 pg_data_t *pgdat = NODE_DATA(node);
384 end_address = (unsigned long) &vmem_map[pgdat->node_start_pfn + i];
385 end_address = PAGE_ALIGN(end_address);
386 stop_address = (unsigned long) &vmem_map[pgdat_end_pfn(pgdat)];
388 do {
389 pgd_t *pgd;
390 pud_t *pud;
391 pmd_t *pmd;
392 pte_t *pte;
394 pgd = pgd_offset_k(end_address);
395 if (pgd_none(*pgd)) {
396 end_address += PGDIR_SIZE;
397 continue;
400 pud = pud_offset(pgd, end_address);
401 if (pud_none(*pud)) {
402 end_address += PUD_SIZE;
403 continue;
406 pmd = pmd_offset(pud, end_address);
407 if (pmd_none(*pmd)) {
408 end_address += PMD_SIZE;
409 continue;
412 pte = pte_offset_kernel(pmd, end_address);
413 retry_pte:
414 if (pte_none(*pte)) {
415 end_address += PAGE_SIZE;
416 pte++;
417 if ((end_address < stop_address) &&
418 (end_address != ALIGN(end_address, 1UL << PMD_SHIFT)))
419 goto retry_pte;
420 continue;
422 /* Found next valid vmem_map page */
423 break;
424 } while (end_address < stop_address);
426 end_address = min(end_address, stop_address);
427 end_address = end_address - (unsigned long) vmem_map + sizeof(struct page) - 1;
428 hole_next_pfn = end_address / sizeof(struct page);
429 return hole_next_pfn - pgdat->node_start_pfn;
432 int __init create_mem_map_page_table(u64 start, u64 end, void *arg)
434 unsigned long address, start_page, end_page;
435 struct page *map_start, *map_end;
436 int node;
437 pgd_t *pgd;
438 pud_t *pud;
439 pmd_t *pmd;
440 pte_t *pte;
442 map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
443 map_end = vmem_map + (__pa(end) >> PAGE_SHIFT);
445 start_page = (unsigned long) map_start & PAGE_MASK;
446 end_page = PAGE_ALIGN((unsigned long) map_end);
447 node = paddr_to_nid(__pa(start));
449 for (address = start_page; address < end_page; address += PAGE_SIZE) {
450 pgd = pgd_offset_k(address);
451 if (pgd_none(*pgd))
452 pgd_populate(&init_mm, pgd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
453 pud = pud_offset(pgd, address);
455 if (pud_none(*pud))
456 pud_populate(&init_mm, pud, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
457 pmd = pmd_offset(pud, address);
459 if (pmd_none(*pmd))
460 pmd_populate_kernel(&init_mm, pmd, alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE));
461 pte = pte_offset_kernel(pmd, address);
463 if (pte_none(*pte))
464 set_pte(pte, pfn_pte(__pa(alloc_bootmem_pages_node(NODE_DATA(node), PAGE_SIZE)) >> PAGE_SHIFT,
465 PAGE_KERNEL));
467 return 0;
470 struct memmap_init_callback_data {
471 struct page *start;
472 struct page *end;
473 int nid;
474 unsigned long zone;
477 static int __meminit
478 virtual_memmap_init(u64 start, u64 end, void *arg)
480 struct memmap_init_callback_data *args;
481 struct page *map_start, *map_end;
483 args = (struct memmap_init_callback_data *) arg;
484 map_start = vmem_map + (__pa(start) >> PAGE_SHIFT);
485 map_end = vmem_map + (__pa(end) >> PAGE_SHIFT);
487 if (map_start < args->start)
488 map_start = args->start;
489 if (map_end > args->end)
490 map_end = args->end;
493 * We have to initialize "out of bounds" struct page elements that fit completely
494 * on the same pages that were allocated for the "in bounds" elements because they
495 * may be referenced later (and found to be "reserved").
497 map_start -= ((unsigned long) map_start & (PAGE_SIZE - 1)) / sizeof(struct page);
498 map_end += ((PAGE_ALIGN((unsigned long) map_end) - (unsigned long) map_end)
499 / sizeof(struct page));
501 if (map_start < map_end)
502 memmap_init_zone((unsigned long)(map_end - map_start),
503 args->nid, args->zone, page_to_pfn(map_start),
504 MEMMAP_EARLY);
505 return 0;
508 void __meminit
509 memmap_init (unsigned long size, int nid, unsigned long zone,
510 unsigned long start_pfn)
512 if (!vmem_map)
513 memmap_init_zone(size, nid, zone, start_pfn, MEMMAP_EARLY);
514 else {
515 struct page *start;
516 struct memmap_init_callback_data args;
518 start = pfn_to_page(start_pfn);
519 args.start = start;
520 args.end = start + size;
521 args.nid = nid;
522 args.zone = zone;
524 efi_memmap_walk(virtual_memmap_init, &args);
529 ia64_pfn_valid (unsigned long pfn)
531 char byte;
532 struct page *pg = pfn_to_page(pfn);
534 return (__get_user(byte, (char __user *) pg) == 0)
535 && ((((u64)pg & PAGE_MASK) == (((u64)(pg + 1) - 1) & PAGE_MASK))
536 || (__get_user(byte, (char __user *) (pg + 1) - 1) == 0));
538 EXPORT_SYMBOL(ia64_pfn_valid);
540 int __init find_largest_hole(u64 start, u64 end, void *arg)
542 u64 *max_gap = arg;
544 static u64 last_end = PAGE_OFFSET;
546 /* NOTE: this algorithm assumes efi memmap table is ordered */
548 if (*max_gap < (start - last_end))
549 *max_gap = start - last_end;
550 last_end = end;
551 return 0;
554 #endif /* CONFIG_VIRTUAL_MEM_MAP */
556 int __init register_active_ranges(u64 start, u64 len, int nid)
558 u64 end = start + len;
560 #ifdef CONFIG_KEXEC
561 if (start > crashk_res.start && start < crashk_res.end)
562 start = crashk_res.end;
563 if (end > crashk_res.start && end < crashk_res.end)
564 end = crashk_res.start;
565 #endif
567 if (start < end)
568 memblock_add_node(__pa(start), end - start, nid);
569 return 0;
573 find_max_min_low_pfn (u64 start, u64 end, void *arg)
575 unsigned long pfn_start, pfn_end;
576 #ifdef CONFIG_FLATMEM
577 pfn_start = (PAGE_ALIGN(__pa(start))) >> PAGE_SHIFT;
578 pfn_end = (PAGE_ALIGN(__pa(end - 1))) >> PAGE_SHIFT;
579 #else
580 pfn_start = GRANULEROUNDDOWN(__pa(start)) >> PAGE_SHIFT;
581 pfn_end = GRANULEROUNDUP(__pa(end - 1)) >> PAGE_SHIFT;
582 #endif
583 min_low_pfn = min(min_low_pfn, pfn_start);
584 max_low_pfn = max(max_low_pfn, pfn_end);
585 return 0;
589 * Boot command-line option "nolwsys" can be used to disable the use of any light-weight
590 * system call handler. When this option is in effect, all fsyscalls will end up bubbling
591 * down into the kernel and calling the normal (heavy-weight) syscall handler. This is
592 * useful for performance testing, but conceivably could also come in handy for debugging
593 * purposes.
596 static int nolwsys __initdata;
598 static int __init
599 nolwsys_setup (char *s)
601 nolwsys = 1;
602 return 1;
605 __setup("nolwsys", nolwsys_setup);
607 void __init
608 mem_init (void)
610 int i;
612 BUG_ON(PTRS_PER_PGD * sizeof(pgd_t) != PAGE_SIZE);
613 BUG_ON(PTRS_PER_PMD * sizeof(pmd_t) != PAGE_SIZE);
614 BUG_ON(PTRS_PER_PTE * sizeof(pte_t) != PAGE_SIZE);
616 #ifdef CONFIG_PCI
618 * This needs to be called _after_ the command line has been parsed but _before_
619 * any drivers that may need the PCI DMA interface are initialized or bootmem has
620 * been freed.
622 platform_dma_init();
623 #endif
625 #ifdef CONFIG_FLATMEM
626 BUG_ON(!mem_map);
627 #endif
629 set_max_mapnr(max_low_pfn);
630 high_memory = __va(max_low_pfn * PAGE_SIZE);
631 free_all_bootmem();
632 mem_init_print_info(NULL);
635 * For fsyscall entrpoints with no light-weight handler, use the ordinary
636 * (heavy-weight) handler, but mark it by setting bit 0, so the fsyscall entry
637 * code can tell them apart.
639 for (i = 0; i < NR_syscalls; ++i) {
640 extern unsigned long fsyscall_table[NR_syscalls];
641 extern unsigned long sys_call_table[NR_syscalls];
643 if (!fsyscall_table[i] || nolwsys)
644 fsyscall_table[i] = sys_call_table[i] | 1;
646 setup_gate();
649 #ifdef CONFIG_MEMORY_HOTPLUG
650 int arch_add_memory(int nid, u64 start, u64 size, bool want_memblock)
652 unsigned long start_pfn = start >> PAGE_SHIFT;
653 unsigned long nr_pages = size >> PAGE_SHIFT;
654 int ret;
656 ret = __add_pages(nid, start_pfn, nr_pages, want_memblock);
657 if (ret)
658 printk("%s: Problem encountered in __add_pages() as ret=%d\n",
659 __func__, ret);
661 return ret;
664 #ifdef CONFIG_MEMORY_HOTREMOVE
665 int arch_remove_memory(u64 start, u64 size)
667 unsigned long start_pfn = start >> PAGE_SHIFT;
668 unsigned long nr_pages = size >> PAGE_SHIFT;
669 struct zone *zone;
670 int ret;
672 zone = page_zone(pfn_to_page(start_pfn));
673 ret = __remove_pages(zone, start_pfn, nr_pages);
674 if (ret)
675 pr_warn("%s: Problem encountered in __remove_pages() as"
676 " ret=%d\n", __func__, ret);
678 return ret;
680 #endif
681 #endif