1 // SPDX-License-Identifier: GPL-2.0-only
3 * Based on arch/arm/kernel/process.c
5 * Original Copyright (C) 1995 Linus Torvalds
6 * Copyright (C) 1996-2000 Russell King - Converted to ARM.
7 * Copyright (C) 2012 ARM Ltd.
9 #include <linux/compat.h>
10 #include <linux/efi.h>
11 #include <linux/elf.h>
12 #include <linux/export.h>
13 #include <linux/sched.h>
14 #include <linux/sched/debug.h>
15 #include <linux/sched/task.h>
16 #include <linux/sched/task_stack.h>
17 #include <linux/kernel.h>
18 #include <linux/mman.h>
20 #include <linux/nospec.h>
21 #include <linux/stddef.h>
22 #include <linux/sysctl.h>
23 #include <linux/unistd.h>
24 #include <linux/user.h>
25 #include <linux/delay.h>
26 #include <linux/reboot.h>
27 #include <linux/interrupt.h>
28 #include <linux/init.h>
29 #include <linux/cpu.h>
30 #include <linux/elfcore.h>
32 #include <linux/tick.h>
33 #include <linux/utsname.h>
34 #include <linux/uaccess.h>
35 #include <linux/random.h>
36 #include <linux/hw_breakpoint.h>
37 #include <linux/personality.h>
38 #include <linux/notifier.h>
39 #include <trace/events/power.h>
40 #include <linux/percpu.h>
41 #include <linux/thread_info.h>
42 #include <linux/prctl.h>
43 #include <linux/stacktrace.h>
45 #include <asm/alternative.h>
46 #include <asm/compat.h>
47 #include <asm/cpufeature.h>
48 #include <asm/cacheflush.h>
50 #include <asm/fpsimd.h>
51 #include <asm/mmu_context.h>
53 #include <asm/processor.h>
54 #include <asm/pointer_auth.h>
55 #include <asm/stacktrace.h>
56 #include <asm/switch_to.h>
57 #include <asm/system_misc.h>
59 #if defined(CONFIG_STACKPROTECTOR) && !defined(CONFIG_STACKPROTECTOR_PER_TASK)
60 #include <linux/stackprotector.h>
61 unsigned long __stack_chk_guard __ro_after_init;
62 EXPORT_SYMBOL(__stack_chk_guard);
66 * Function pointers to optional machine specific functions
68 void (*pm_power_off)(void);
69 EXPORT_SYMBOL_GPL(pm_power_off);
71 #ifdef CONFIG_HOTPLUG_CPU
72 void arch_cpu_idle_dead(void)
79 * Called by kexec, immediately prior to machine_kexec().
81 * This must completely disable all secondary CPUs; simply causing those CPUs
82 * to execute e.g. a RAM-based pin loop is not sufficient. This allows the
83 * kexec'd kernel to use any and all RAM as it sees fit, without having to
84 * avoid any code or data used by any SW CPU pin loop. The CPU hotplug
85 * functionality embodied in smpt_shutdown_nonboot_cpus() to achieve this.
87 void machine_shutdown(void)
89 smp_shutdown_nonboot_cpus(reboot_cpu);
93 * Halting simply requires that the secondary CPUs stop performing any
94 * activity (executing tasks, handling interrupts). smp_send_stop()
97 void machine_halt(void)
105 * Power-off simply requires that the secondary CPUs stop performing any
106 * activity (executing tasks, handling interrupts). smp_send_stop()
107 * achieves this. When the system power is turned off, it will take all CPUs
110 void machine_power_off(void)
119 * Restart requires that the secondary CPUs stop performing any activity
120 * while the primary CPU resets the system. Systems with multiple CPUs must
121 * provide a HW restart implementation, to ensure that all CPUs reset at once.
122 * This is required so that any code running after reset on the primary CPU
123 * doesn't have to co-ordinate with other CPUs to ensure they aren't still
124 * executing pre-reset code, and using RAM that the primary CPU's code wishes
125 * to use. Implementing such co-ordination would be essentially impossible.
127 void machine_restart(char *cmd)
129 /* Disable interrupts first */
134 * UpdateCapsule() depends on the system being reset via
137 if (efi_enabled(EFI_RUNTIME_SERVICES))
138 efi_reboot(reboot_mode, NULL);
140 /* Now call the architecture specific reboot code. */
141 do_kernel_restart(cmd);
144 * Whoops - the architecture was unable to reboot.
146 printk("Reboot failed -- System halted\n");
150 #define bstr(suffix, str) [PSR_BTYPE_ ## suffix >> PSR_BTYPE_SHIFT] = str
151 static const char *const btypes[] = {
159 static void print_pstate(struct pt_regs *regs)
161 u64 pstate = regs->pstate;
163 if (compat_user_mode(regs)) {
164 printk("pstate: %08llx (%c%c%c%c %c %s %s %c%c%c %cDIT %cSSBS)\n",
166 pstate & PSR_AA32_N_BIT ? 'N' : 'n',
167 pstate & PSR_AA32_Z_BIT ? 'Z' : 'z',
168 pstate & PSR_AA32_C_BIT ? 'C' : 'c',
169 pstate & PSR_AA32_V_BIT ? 'V' : 'v',
170 pstate & PSR_AA32_Q_BIT ? 'Q' : 'q',
171 pstate & PSR_AA32_T_BIT ? "T32" : "A32",
172 pstate & PSR_AA32_E_BIT ? "BE" : "LE",
173 pstate & PSR_AA32_A_BIT ? 'A' : 'a',
174 pstate & PSR_AA32_I_BIT ? 'I' : 'i',
175 pstate & PSR_AA32_F_BIT ? 'F' : 'f',
176 pstate & PSR_AA32_DIT_BIT ? '+' : '-',
177 pstate & PSR_AA32_SSBS_BIT ? '+' : '-');
179 const char *btype_str = btypes[(pstate & PSR_BTYPE_MASK) >>
182 printk("pstate: %08llx (%c%c%c%c %c%c%c%c %cPAN %cUAO %cTCO %cDIT %cSSBS BTYPE=%s)\n",
184 pstate & PSR_N_BIT ? 'N' : 'n',
185 pstate & PSR_Z_BIT ? 'Z' : 'z',
186 pstate & PSR_C_BIT ? 'C' : 'c',
187 pstate & PSR_V_BIT ? 'V' : 'v',
188 pstate & PSR_D_BIT ? 'D' : 'd',
189 pstate & PSR_A_BIT ? 'A' : 'a',
190 pstate & PSR_I_BIT ? 'I' : 'i',
191 pstate & PSR_F_BIT ? 'F' : 'f',
192 pstate & PSR_PAN_BIT ? '+' : '-',
193 pstate & PSR_UAO_BIT ? '+' : '-',
194 pstate & PSR_TCO_BIT ? '+' : '-',
195 pstate & PSR_DIT_BIT ? '+' : '-',
196 pstate & PSR_SSBS_BIT ? '+' : '-',
201 void __show_regs(struct pt_regs *regs)
206 if (compat_user_mode(regs)) {
207 lr = regs->compat_lr;
208 sp = regs->compat_sp;
216 show_regs_print_info(KERN_DEFAULT);
219 if (!user_mode(regs)) {
220 printk("pc : %pS\n", (void *)regs->pc);
221 printk("lr : %pS\n", (void *)ptrauth_strip_insn_pac(lr));
223 printk("pc : %016llx\n", regs->pc);
224 printk("lr : %016llx\n", lr);
227 printk("sp : %016llx\n", sp);
229 if (system_uses_irq_prio_masking())
230 printk("pmr_save: %08llx\n", regs->pmr_save);
235 printk("x%-2d: %016llx", i, regs->regs[i]);
238 pr_cont(" x%-2d: %016llx", i, regs->regs[i]);
244 void show_regs(struct pt_regs *regs)
247 dump_backtrace(regs, NULL, KERN_DEFAULT);
250 static void tls_thread_flush(void)
252 write_sysreg(0, tpidr_el0);
254 if (is_compat_task()) {
255 current->thread.uw.tp_value = 0;
258 * We need to ensure ordering between the shadow state and the
259 * hardware state, so that we don't corrupt the hardware state
260 * with a stale shadow state during context switch.
263 write_sysreg(0, tpidrro_el0);
267 static void flush_tagged_addr_state(void)
269 if (IS_ENABLED(CONFIG_ARM64_TAGGED_ADDR_ABI))
270 clear_thread_flag(TIF_TAGGED_ADDR);
273 void flush_thread(void)
275 fpsimd_flush_thread();
277 flush_ptrace_hw_breakpoint(current);
278 flush_tagged_addr_state();
281 void release_thread(struct task_struct *dead_task)
285 void arch_release_task_struct(struct task_struct *tsk)
287 fpsimd_release_task(tsk);
290 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
293 fpsimd_preserve_current_state();
296 /* We rely on the above assignment to initialize dst's thread_flags: */
297 BUILD_BUG_ON(!IS_ENABLED(CONFIG_THREAD_INFO_IN_TASK));
300 * Detach src's sve_state (if any) from dst so that it does not
301 * get erroneously used or freed prematurely. dst's sve_state
302 * will be allocated on demand later on if dst uses SVE.
303 * For consistency, also clear TIF_SVE here: this could be done
304 * later in copy_process(), but to avoid tripping up future
305 * maintainers it is best not to leave TIF_SVE and sve_state in
306 * an inconsistent state, even temporarily.
308 dst->thread.sve_state = NULL;
309 clear_tsk_thread_flag(dst, TIF_SVE);
311 /* clear any pending asynchronous tag fault raised by the parent */
312 clear_tsk_thread_flag(dst, TIF_MTE_ASYNC_FAULT);
317 asmlinkage void ret_from_fork(void) asm("ret_from_fork");
319 int copy_thread(unsigned long clone_flags, unsigned long stack_start,
320 unsigned long stk_sz, struct task_struct *p, unsigned long tls)
322 struct pt_regs *childregs = task_pt_regs(p);
324 memset(&p->thread.cpu_context, 0, sizeof(struct cpu_context));
327 * In case p was allocated the same task_struct pointer as some
328 * other recently-exited task, make sure p is disassociated from
329 * any cpu that may have run that now-exited task recently.
330 * Otherwise we could erroneously skip reloading the FPSIMD
333 fpsimd_flush_task_state(p);
335 ptrauth_thread_init_kernel(p);
337 if (likely(!(p->flags & (PF_KTHREAD | PF_IO_WORKER)))) {
338 *childregs = *current_pt_regs();
339 childregs->regs[0] = 0;
342 * Read the current TLS pointer from tpidr_el0 as it may be
343 * out-of-sync with the saved value.
345 *task_user_tls(p) = read_sysreg(tpidr_el0);
348 if (is_compat_thread(task_thread_info(p)))
349 childregs->compat_sp = stack_start;
351 childregs->sp = stack_start;
355 * If a TLS pointer was passed to clone, use it for the new
358 if (clone_flags & CLONE_SETTLS)
359 p->thread.uw.tp_value = tls;
362 * A kthread has no context to ERET to, so ensure any buggy
363 * ERET is treated as an illegal exception return.
365 * When a user task is created from a kthread, childregs will
366 * be initialized by start_thread() or start_compat_thread().
368 memset(childregs, 0, sizeof(struct pt_regs));
369 childregs->pstate = PSR_MODE_EL1h | PSR_IL_BIT;
371 p->thread.cpu_context.x19 = stack_start;
372 p->thread.cpu_context.x20 = stk_sz;
374 p->thread.cpu_context.pc = (unsigned long)ret_from_fork;
375 p->thread.cpu_context.sp = (unsigned long)childregs;
377 * For the benefit of the unwinder, set up childregs->stackframe
378 * as the final frame for the new task.
380 p->thread.cpu_context.fp = (unsigned long)childregs->stackframe;
382 ptrace_hw_copy_thread(p);
387 void tls_preserve_current_state(void)
389 *task_user_tls(current) = read_sysreg(tpidr_el0);
392 static void tls_thread_switch(struct task_struct *next)
394 tls_preserve_current_state();
396 if (is_compat_thread(task_thread_info(next)))
397 write_sysreg(next->thread.uw.tp_value, tpidrro_el0);
398 else if (!arm64_kernel_unmapped_at_el0())
399 write_sysreg(0, tpidrro_el0);
401 write_sysreg(*task_user_tls(next), tpidr_el0);
405 * Force SSBS state on context-switch, since it may be lost after migrating
406 * from a CPU which treats the bit as RES0 in a heterogeneous system.
408 static void ssbs_thread_switch(struct task_struct *next)
411 * Nothing to do for kernel threads, but 'regs' may be junk
412 * (e.g. idle task) so check the flags and bail early.
414 if (unlikely(next->flags & PF_KTHREAD))
418 * If all CPUs implement the SSBS extension, then we just need to
419 * context-switch the PSTATE field.
421 if (cpus_have_const_cap(ARM64_SSBS))
424 spectre_v4_enable_task_mitigation(next);
428 * We store our current task in sp_el0, which is clobbered by userspace. Keep a
429 * shadow copy so that we can restore this upon entry from userspace.
431 * This is *only* for exception entry from EL0, and is not valid until we
432 * __switch_to() a user task.
434 DEFINE_PER_CPU(struct task_struct *, __entry_task);
436 static void entry_task_switch(struct task_struct *next)
438 __this_cpu_write(__entry_task, next);
442 * ARM erratum 1418040 handling, affecting the 32bit view of CNTVCT.
443 * Ensure access is disabled when switching to a 32bit task, ensure
444 * access is enabled when switching to a 64bit task.
446 static void erratum_1418040_thread_switch(struct task_struct *next)
448 if (!IS_ENABLED(CONFIG_ARM64_ERRATUM_1418040) ||
449 !this_cpu_has_cap(ARM64_WORKAROUND_1418040))
452 if (is_compat_thread(task_thread_info(next)))
453 sysreg_clear_set(cntkctl_el1, ARCH_TIMER_USR_VCT_ACCESS_EN, 0);
455 sysreg_clear_set(cntkctl_el1, 0, ARCH_TIMER_USR_VCT_ACCESS_EN);
458 static void erratum_1418040_new_exec(void)
461 erratum_1418040_thread_switch(current);
466 * __switch_to() checks current->thread.sctlr_user as an optimisation. Therefore
467 * this function must be called with preemption disabled and the update to
468 * sctlr_user must be made in the same preemption disabled block so that
469 * __switch_to() does not see the variable update before the SCTLR_EL1 one.
471 void update_sctlr_el1(u64 sctlr)
474 * EnIA must not be cleared while in the kernel as this is necessary for
475 * in-kernel PAC. It will be cleared on kernel exit if needed.
477 sysreg_clear_set(sctlr_el1, SCTLR_USER_MASK & ~SCTLR_ELx_ENIA, sctlr);
479 /* ISB required for the kernel uaccess routines when setting TCF0. */
486 __notrace_funcgraph __sched
487 struct task_struct *__switch_to(struct task_struct *prev,
488 struct task_struct *next)
490 struct task_struct *last;
492 fpsimd_thread_switch(next);
493 tls_thread_switch(next);
494 hw_breakpoint_thread_switch(next);
495 contextidr_thread_switch(next);
496 entry_task_switch(next);
497 ssbs_thread_switch(next);
498 erratum_1418040_thread_switch(next);
499 ptrauth_thread_switch_user(next);
502 * Complete any pending TLB or cache maintenance on this CPU in case
503 * the thread migrates to a different CPU.
504 * This full barrier is also required by the membarrier system
510 * MTE thread switching must happen after the DSB above to ensure that
511 * any asynchronous tag check faults have been logged in the TFSR*_EL1
514 mte_thread_switch(next);
515 /* avoid expensive SCTLR_EL1 accesses if no change */
516 if (prev->thread.sctlr_user != next->thread.sctlr_user)
517 update_sctlr_el1(next->thread.sctlr_user);
519 /* the actual thread switch */
520 last = cpu_switch_to(prev, next);
530 static bool get_wchan_cb(void *arg, unsigned long pc)
532 struct wchan_info *wchan_info = arg;
534 if (!in_sched_functions(pc)) {
538 return wchan_info->count++ < 16;
541 unsigned long __get_wchan(struct task_struct *p)
543 struct wchan_info wchan_info = {
548 if (!try_get_task_stack(p))
551 arch_stack_walk(get_wchan_cb, &wchan_info, p, NULL);
555 return wchan_info.pc;
558 unsigned long arch_align_stack(unsigned long sp)
560 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
561 sp -= get_random_int() & ~PAGE_MASK;
566 int compat_elf_check_arch(const struct elf32_hdr *hdr)
568 if (!system_supports_32bit_el0())
571 if ((hdr)->e_machine != EM_ARM)
574 if (!((hdr)->e_flags & EF_ARM_EABI_MASK))
578 * Prevent execve() of a 32-bit program from a deadline task
579 * if the restricted affinity mask would be inadmissible on an
582 return !static_branch_unlikely(&arm64_mismatched_32bit_el0) ||
583 !dl_task_check_affinity(current, system_32bit_el0_cpumask());
588 * Called from setup_new_exec() after (COMPAT_)SET_PERSONALITY.
590 void arch_setup_new_exec(void)
592 unsigned long mmflags = 0;
594 if (is_compat_task()) {
595 mmflags = MMCF_AARCH32;
598 * Restrict the CPU affinity mask for a 32-bit task so that
599 * it contains only 32-bit-capable CPUs.
601 * From the perspective of the task, this looks similar to
602 * what would happen if the 64-bit-only CPUs were hot-unplugged
603 * at the point of execve(), although we try a bit harder to
604 * honour the cpuset hierarchy.
606 if (static_branch_unlikely(&arm64_mismatched_32bit_el0))
607 force_compatible_cpus_allowed_ptr(current);
608 } else if (static_branch_unlikely(&arm64_mismatched_32bit_el0)) {
609 relax_compatible_cpus_allowed_ptr(current);
612 current->mm->context.flags = mmflags;
613 ptrauth_thread_init_user();
614 mte_thread_init_user();
615 erratum_1418040_new_exec();
617 if (task_spec_ssb_noexec(current)) {
618 arch_prctl_spec_ctrl_set(current, PR_SPEC_STORE_BYPASS,
623 #ifdef CONFIG_ARM64_TAGGED_ADDR_ABI
625 * Control the relaxed ABI allowing tagged user addresses into the kernel.
627 static unsigned int tagged_addr_disabled;
629 long set_tagged_addr_ctrl(struct task_struct *task, unsigned long arg)
631 unsigned long valid_mask = PR_TAGGED_ADDR_ENABLE;
632 struct thread_info *ti = task_thread_info(task);
634 if (is_compat_thread(ti))
637 if (system_supports_mte())
638 valid_mask |= PR_MTE_TCF_SYNC | PR_MTE_TCF_ASYNC \
641 if (arg & ~valid_mask)
645 * Do not allow the enabling of the tagged address ABI if globally
646 * disabled via sysctl abi.tagged_addr_disabled.
648 if (arg & PR_TAGGED_ADDR_ENABLE && tagged_addr_disabled)
651 if (set_mte_ctrl(task, arg) != 0)
654 update_ti_thread_flag(ti, TIF_TAGGED_ADDR, arg & PR_TAGGED_ADDR_ENABLE);
659 long get_tagged_addr_ctrl(struct task_struct *task)
662 struct thread_info *ti = task_thread_info(task);
664 if (is_compat_thread(ti))
667 if (test_ti_thread_flag(ti, TIF_TAGGED_ADDR))
668 ret = PR_TAGGED_ADDR_ENABLE;
670 ret |= get_mte_ctrl(task);
676 * Global sysctl to disable the tagged user addresses support. This control
677 * only prevents the tagged address ABI enabling via prctl() and does not
678 * disable it for tasks that already opted in to the relaxed ABI.
681 static struct ctl_table tagged_addr_sysctl_table[] = {
683 .procname = "tagged_addr_disabled",
685 .data = &tagged_addr_disabled,
686 .maxlen = sizeof(int),
687 .proc_handler = proc_dointvec_minmax,
688 .extra1 = SYSCTL_ZERO,
689 .extra2 = SYSCTL_ONE,
694 static int __init tagged_addr_init(void)
696 if (!register_sysctl("abi", tagged_addr_sysctl_table))
701 core_initcall(tagged_addr_init);
702 #endif /* CONFIG_ARM64_TAGGED_ADDR_ABI */
704 #ifdef CONFIG_BINFMT_ELF
705 int arch_elf_adjust_prot(int prot, const struct arch_elf_state *state,
706 bool has_interp, bool is_interp)
709 * For dynamically linked executables the interpreter is
710 * responsible for setting PROT_BTI on everything except
713 if (is_interp != has_interp)
716 if (!(state->flags & ARM64_ELF_BTI))
719 if (prot & PROT_EXEC)