| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | |
| 3 | #include <linux/ptrace.h> |
| 4 | #include <asm/bugs.h> |
| 5 | #include <asm/msr.h> |
| 6 | #include <asm/traps.h> |
| 7 | |
| 8 | enum cp_error_code { |
| 9 | CP_EC = (1 << 15) - 1, |
| 10 | |
| 11 | CP_RET = 1, |
| 12 | CP_IRET = 2, |
| 13 | CP_ENDBR = 3, |
| 14 | = 4, |
| 15 | CP_SETSSBSY = 5, |
| 16 | |
| 17 | CP_ENCL = 1 << 15, |
| 18 | }; |
| 19 | |
| 20 | static const char cp_err[][10] = { |
| 21 | [0] = "unknown" , |
| 22 | [1] = "near ret" , |
| 23 | [2] = "far/iret" , |
| 24 | [3] = "endbranch" , |
| 25 | [4] = "rstorssp" , |
| 26 | [5] = "setssbsy" , |
| 27 | }; |
| 28 | |
| 29 | static const char *cp_err_string(unsigned long error_code) |
| 30 | { |
| 31 | unsigned int cpec = error_code & CP_EC; |
| 32 | |
| 33 | if (cpec >= ARRAY_SIZE(cp_err)) |
| 34 | cpec = 0; |
| 35 | return cp_err[cpec]; |
| 36 | } |
| 37 | |
| 38 | static void do_unexpected_cp(struct pt_regs *regs, unsigned long error_code) |
| 39 | { |
| 40 | WARN_ONCE(1, "Unexpected %s #CP, error_code: %s\n" , |
| 41 | user_mode(regs) ? "user mode" : "kernel mode" , |
| 42 | cp_err_string(error_code)); |
| 43 | } |
| 44 | |
| 45 | static DEFINE_RATELIMIT_STATE(cpf_rate, DEFAULT_RATELIMIT_INTERVAL, |
| 46 | DEFAULT_RATELIMIT_BURST); |
| 47 | |
| 48 | static void do_user_cp_fault(struct pt_regs *regs, unsigned long error_code) |
| 49 | { |
| 50 | struct task_struct *tsk; |
| 51 | unsigned long ssp; |
| 52 | |
| 53 | /* |
| 54 | * An exception was just taken from userspace. Since interrupts are disabled |
| 55 | * here, no scheduling should have messed with the registers yet and they |
| 56 | * will be whatever is live in userspace. So read the SSP before enabling |
| 57 | * interrupts so locking the fpregs to do it later is not required. |
| 58 | */ |
| 59 | rdmsrq(MSR_IA32_PL3_SSP, ssp); |
| 60 | |
| 61 | cond_local_irq_enable(regs); |
| 62 | |
| 63 | tsk = current; |
| 64 | tsk->thread.error_code = error_code; |
| 65 | tsk->thread.trap_nr = X86_TRAP_CP; |
| 66 | |
| 67 | /* Ratelimit to prevent log spamming. */ |
| 68 | if (show_unhandled_signals && unhandled_signal(tsk, SIGSEGV) && |
| 69 | __ratelimit(&cpf_rate)) { |
| 70 | pr_emerg("%s[%d] control protection ip:%lx sp:%lx ssp:%lx error:%lx(%s)%s" , |
| 71 | tsk->comm, task_pid_nr(tsk), |
| 72 | regs->ip, regs->sp, ssp, error_code, |
| 73 | cp_err_string(error_code), |
| 74 | error_code & CP_ENCL ? " in enclave" : "" ); |
| 75 | print_vma_addr(KERN_CONT " in " , rip: regs->ip); |
| 76 | pr_cont("\n" ); |
| 77 | } |
| 78 | |
| 79 | force_sig_fault(SIGSEGV, SEGV_CPERR, addr: (void __user *)0); |
| 80 | cond_local_irq_disable(regs); |
| 81 | } |
| 82 | |
| 83 | static __ro_after_init bool ibt_fatal = true; |
| 84 | |
| 85 | /* |
| 86 | * By definition, all missing-ENDBRANCH #CPs are a result of WFE && !ENDBR. |
| 87 | * |
| 88 | * For the kernel IBT no ENDBR selftest where #CPs are deliberately triggered, |
| 89 | * the WFE state of the interrupted context needs to be cleared to let execution |
| 90 | * continue. Otherwise when the CPU resumes from the instruction that just |
| 91 | * caused the previous #CP, another missing-ENDBRANCH #CP is raised and the CPU |
| 92 | * enters a dead loop. |
| 93 | * |
| 94 | * This is not a problem with IDT because it doesn't preserve WFE and IRET doesn't |
| 95 | * set WFE. But FRED provides space on the entry stack (in an expanded CS area) |
| 96 | * to save and restore the WFE state, thus the WFE state is no longer clobbered, |
| 97 | * so software must clear it. |
| 98 | */ |
| 99 | static void ibt_clear_fred_wfe(struct pt_regs *regs) |
| 100 | { |
| 101 | /* |
| 102 | * No need to do any FRED checks. |
| 103 | * |
| 104 | * For IDT event delivery, the high-order 48 bits of CS are pushed |
| 105 | * as 0s into the stack, and later IRET ignores these bits. |
| 106 | * |
| 107 | * For FRED, a test to check if fred_cs.wfe is set would be dropped |
| 108 | * by compilers. |
| 109 | */ |
| 110 | regs->fred_cs.wfe = 0; |
| 111 | } |
| 112 | |
| 113 | static void do_kernel_cp_fault(struct pt_regs *regs, unsigned long error_code) |
| 114 | { |
| 115 | if ((error_code & CP_EC) != CP_ENDBR) { |
| 116 | do_unexpected_cp(regs, error_code); |
| 117 | return; |
| 118 | } |
| 119 | |
| 120 | if (unlikely(regs->ip == (unsigned long)&ibt_selftest_noendbr)) { |
| 121 | regs->ax = 0; |
| 122 | ibt_clear_fred_wfe(regs); |
| 123 | return; |
| 124 | } |
| 125 | |
| 126 | pr_err("Missing ENDBR: %pS\n" , (void *)instruction_pointer(regs)); |
| 127 | if (!ibt_fatal) { |
| 128 | printk(KERN_DEFAULT CUT_HERE); |
| 129 | __warn(__FILE__, __LINE__, caller: (void *)regs->ip, TAINT_WARN, regs, NULL); |
| 130 | ibt_clear_fred_wfe(regs); |
| 131 | return; |
| 132 | } |
| 133 | BUG(); |
| 134 | } |
| 135 | |
| 136 | static int __init ibt_setup(char *str) |
| 137 | { |
| 138 | if (!strcmp(str, "off" )) |
| 139 | setup_clear_cpu_cap(X86_FEATURE_IBT); |
| 140 | |
| 141 | if (!strcmp(str, "warn" )) |
| 142 | ibt_fatal = false; |
| 143 | |
| 144 | return 1; |
| 145 | } |
| 146 | |
| 147 | __setup("ibt=" , ibt_setup); |
| 148 | |
| 149 | DEFINE_IDTENTRY_ERRORCODE(exc_control_protection) |
| 150 | { |
| 151 | if (user_mode(regs)) { |
| 152 | if (cpu_feature_enabled(X86_FEATURE_USER_SHSTK)) |
| 153 | do_user_cp_fault(regs, error_code); |
| 154 | else |
| 155 | do_unexpected_cp(regs, error_code); |
| 156 | } else { |
| 157 | if (cpu_feature_enabled(X86_FEATURE_IBT)) |
| 158 | do_kernel_cp_fault(regs, error_code); |
| 159 | else |
| 160 | do_unexpected_cp(regs, error_code); |
| 161 | } |
| 162 | } |
| 163 | |