Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame^] | 1 | /* Copyright 2002,2003 Andi Kleen, SuSE Labs */ |
| 2 | |
| 3 | /* vsyscall handling for 32bit processes. Map a stub page into it |
| 4 | on demand because 32bit cannot reach the kernel's fixmaps */ |
| 5 | |
| 6 | #include <linux/mm.h> |
| 7 | #include <linux/string.h> |
| 8 | #include <linux/kernel.h> |
| 9 | #include <linux/gfp.h> |
| 10 | #include <linux/init.h> |
| 11 | #include <linux/stringify.h> |
| 12 | #include <asm/proto.h> |
| 13 | #include <asm/tlbflush.h> |
| 14 | #include <asm/ia32_unistd.h> |
| 15 | |
| 16 | /* 32bit VDSOs mapped into user space. */ |
| 17 | asm(".section \".init.data\",\"aw\"\n" |
| 18 | "syscall32_syscall:\n" |
| 19 | ".incbin \"arch/x86_64/ia32/vsyscall-syscall.so\"\n" |
| 20 | "syscall32_syscall_end:\n" |
| 21 | "syscall32_sysenter:\n" |
| 22 | ".incbin \"arch/x86_64/ia32/vsyscall-sysenter.so\"\n" |
| 23 | "syscall32_sysenter_end:\n" |
| 24 | ".previous"); |
| 25 | |
| 26 | extern unsigned char syscall32_syscall[], syscall32_syscall_end[]; |
| 27 | extern unsigned char syscall32_sysenter[], syscall32_sysenter_end[]; |
| 28 | extern int sysctl_vsyscall32; |
| 29 | |
| 30 | char *syscall32_page; |
| 31 | static int use_sysenter = -1; |
| 32 | |
| 33 | /* |
| 34 | * Map the 32bit vsyscall page on demand. |
| 35 | * |
| 36 | * RED-PEN: This knows too much about high level VM. |
| 37 | * |
| 38 | * Alternative would be to generate a vma with appropriate backing options |
| 39 | * and let it be handled by generic VM. |
| 40 | */ |
| 41 | int __map_syscall32(struct mm_struct *mm, unsigned long address) |
| 42 | { |
| 43 | pgd_t *pgd; |
| 44 | pud_t *pud; |
| 45 | pte_t *pte; |
| 46 | pmd_t *pmd; |
| 47 | int err = -ENOMEM; |
| 48 | |
| 49 | spin_lock(&mm->page_table_lock); |
| 50 | pgd = pgd_offset(mm, address); |
| 51 | pud = pud_alloc(mm, pgd, address); |
| 52 | if (pud) { |
| 53 | pmd = pmd_alloc(mm, pud, address); |
| 54 | if (pmd && (pte = pte_alloc_map(mm, pmd, address)) != NULL) { |
| 55 | if (pte_none(*pte)) { |
| 56 | set_pte(pte, |
| 57 | mk_pte(virt_to_page(syscall32_page), |
| 58 | PAGE_KERNEL_VSYSCALL32)); |
| 59 | } |
| 60 | /* Flush only the local CPU. Other CPUs taking a fault |
| 61 | will just end up here again |
| 62 | This probably not needed and just paranoia. */ |
| 63 | __flush_tlb_one(address); |
| 64 | err = 0; |
| 65 | } |
| 66 | } |
| 67 | spin_unlock(&mm->page_table_lock); |
| 68 | return err; |
| 69 | } |
| 70 | |
| 71 | int map_syscall32(struct mm_struct *mm, unsigned long address) |
| 72 | { |
| 73 | int err; |
| 74 | down_read(&mm->mmap_sem); |
| 75 | err = __map_syscall32(mm, address); |
| 76 | up_read(&mm->mmap_sem); |
| 77 | return err; |
| 78 | } |
| 79 | |
| 80 | static int __init init_syscall32(void) |
| 81 | { |
| 82 | syscall32_page = (void *)get_zeroed_page(GFP_KERNEL); |
| 83 | if (!syscall32_page) |
| 84 | panic("Cannot allocate syscall32 page"); |
| 85 | SetPageReserved(virt_to_page(syscall32_page)); |
| 86 | if (use_sysenter > 0) { |
| 87 | memcpy(syscall32_page, syscall32_sysenter, |
| 88 | syscall32_sysenter_end - syscall32_sysenter); |
| 89 | } else { |
| 90 | memcpy(syscall32_page, syscall32_syscall, |
| 91 | syscall32_syscall_end - syscall32_syscall); |
| 92 | } |
| 93 | return 0; |
| 94 | } |
| 95 | |
| 96 | __initcall(init_syscall32); |
| 97 | |
| 98 | /* May not be __init: called during resume */ |
| 99 | void syscall32_cpu_init(void) |
| 100 | { |
| 101 | if (use_sysenter < 0) |
| 102 | use_sysenter = (boot_cpu_data.x86_vendor == X86_VENDOR_INTEL); |
| 103 | |
| 104 | /* Load these always in case some future AMD CPU supports |
| 105 | SYSENTER from compat mode too. */ |
| 106 | checking_wrmsrl(MSR_IA32_SYSENTER_CS, (u64)__KERNEL_CS); |
| 107 | checking_wrmsrl(MSR_IA32_SYSENTER_ESP, 0ULL); |
| 108 | checking_wrmsrl(MSR_IA32_SYSENTER_EIP, (u64)ia32_sysenter_target); |
| 109 | |
| 110 | wrmsrl(MSR_CSTAR, ia32_cstar_target); |
| 111 | } |