Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Based on arch/arm/kernel/process.c
4 *
5 * Original Copyright (C) 1995 Linus Torvalds
6 * Copyright (C) 1996-2000 Russell King - Converted to ARM.
7 * Copyright (C) 2012 ARM Ltd.
8 */
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>
19#include <linux/mm.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>
31#include <linux/pm.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>
44
45#include <asm/alternative.h>
46#include <asm/compat.h>
47#include <asm/cpufeature.h>
48#include <asm/cacheflush.h>
49#include <asm/exec.h>
50#include <asm/fpsimd.h>
51#include <asm/mmu_context.h>
52#include <asm/mte.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>
58
59#if defined(CONFIG_STACKPROTECTOR) && !defined(CONFIG_STACKPROTECTOR_PER_TASK)
60#include <linux/stackprotector.h>
61unsigned long __stack_chk_guard __ro_after_init;
62EXPORT_SYMBOL(__stack_chk_guard);
63#endif
64
65/*
66 * Function pointers to optional machine specific functions
67 */
68void (*pm_power_off)(void);
69EXPORT_SYMBOL_GPL(pm_power_off);
70
71#ifdef CONFIG_HOTPLUG_CPU
72void __noreturn arch_cpu_idle_dead(void)
73{
74 cpu_die();
75}
76#endif
77
78/*
79 * Called by kexec, immediately prior to machine_kexec().
80 *
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.
86 */
87void machine_shutdown(void)
88{
89 smp_shutdown_nonboot_cpus(reboot_cpu);
90}
91
92/*
93 * Halting simply requires that the secondary CPUs stop performing any
94 * activity (executing tasks, handling interrupts). smp_send_stop()
95 * achieves this.
96 */
97void machine_halt(void)
98{
99 local_irq_disable();
100 smp_send_stop();
101 while (1);
102}
103
104/*
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
108 * with it.
109 */
110void machine_power_off(void)
111{
112 local_irq_disable();
113 smp_send_stop();
114 do_kernel_power_off();
115}
116
117/*
118 * Restart requires that the secondary CPUs stop performing any activity
119 * while the primary CPU resets the system. Systems with multiple CPUs must
120 * provide a HW restart implementation, to ensure that all CPUs reset at once.
121 * This is required so that any code running after reset on the primary CPU
122 * doesn't have to co-ordinate with other CPUs to ensure they aren't still
123 * executing pre-reset code, and using RAM that the primary CPU's code wishes
124 * to use. Implementing such co-ordination would be essentially impossible.
125 */
126void machine_restart(char *cmd)
127{
128 /* Disable interrupts first */
129 local_irq_disable();
130 smp_send_stop();
131
132 /*
133 * UpdateCapsule() depends on the system being reset via
134 * ResetSystem().
135 */
136 if (efi_enabled(EFI_RUNTIME_SERVICES))
137 efi_reboot(reboot_mode, NULL);
138
139 /* Now call the architecture specific reboot code. */
140 do_kernel_restart(cmd);
141
142 /*
143 * Whoops - the architecture was unable to reboot.
144 */
145 printk("Reboot failed -- System halted\n");
146 while (1);
147}
148
149#define bstr(suffix, str) [PSR_BTYPE_ ## suffix >> PSR_BTYPE_SHIFT] = str
150static const char *const btypes[] = {
151 bstr(NONE, "--"),
152 bstr( JC, "jc"),
153 bstr( C, "-c"),
154 bstr( J , "j-")
155};
156#undef bstr
157
158static void print_pstate(struct pt_regs *regs)
159{
160 u64 pstate = regs->pstate;
161
162 if (compat_user_mode(regs)) {
163 printk("pstate: %08llx (%c%c%c%c %c %s %s %c%c%c %cDIT %cSSBS)\n",
164 pstate,
165 pstate & PSR_AA32_N_BIT ? 'N' : 'n',
166 pstate & PSR_AA32_Z_BIT ? 'Z' : 'z',
167 pstate & PSR_AA32_C_BIT ? 'C' : 'c',
168 pstate & PSR_AA32_V_BIT ? 'V' : 'v',
169 pstate & PSR_AA32_Q_BIT ? 'Q' : 'q',
170 pstate & PSR_AA32_T_BIT ? "T32" : "A32",
171 pstate & PSR_AA32_E_BIT ? "BE" : "LE",
172 pstate & PSR_AA32_A_BIT ? 'A' : 'a',
173 pstate & PSR_AA32_I_BIT ? 'I' : 'i',
174 pstate & PSR_AA32_F_BIT ? 'F' : 'f',
175 pstate & PSR_AA32_DIT_BIT ? '+' : '-',
176 pstate & PSR_AA32_SSBS_BIT ? '+' : '-');
177 } else {
178 const char *btype_str = btypes[(pstate & PSR_BTYPE_MASK) >>
179 PSR_BTYPE_SHIFT];
180
181 printk("pstate: %08llx (%c%c%c%c %c%c%c%c %cPAN %cUAO %cTCO %cDIT %cSSBS BTYPE=%s)\n",
182 pstate,
183 pstate & PSR_N_BIT ? 'N' : 'n',
184 pstate & PSR_Z_BIT ? 'Z' : 'z',
185 pstate & PSR_C_BIT ? 'C' : 'c',
186 pstate & PSR_V_BIT ? 'V' : 'v',
187 pstate & PSR_D_BIT ? 'D' : 'd',
188 pstate & PSR_A_BIT ? 'A' : 'a',
189 pstate & PSR_I_BIT ? 'I' : 'i',
190 pstate & PSR_F_BIT ? 'F' : 'f',
191 pstate & PSR_PAN_BIT ? '+' : '-',
192 pstate & PSR_UAO_BIT ? '+' : '-',
193 pstate & PSR_TCO_BIT ? '+' : '-',
194 pstate & PSR_DIT_BIT ? '+' : '-',
195 pstate & PSR_SSBS_BIT ? '+' : '-',
196 btype_str);
197 }
198}
199
200void __show_regs(struct pt_regs *regs)
201{
202 int i, top_reg;
203 u64 lr, sp;
204
205 if (compat_user_mode(regs)) {
206 lr = regs->compat_lr;
207 sp = regs->compat_sp;
208 top_reg = 12;
209 } else {
210 lr = regs->regs[30];
211 sp = regs->sp;
212 top_reg = 29;
213 }
214
215 show_regs_print_info(KERN_DEFAULT);
216 print_pstate(regs);
217
218 if (!user_mode(regs)) {
219 printk("pc : %pS\n", (void *)regs->pc);
220 printk("lr : %pS\n", (void *)ptrauth_strip_kernel_insn_pac(lr));
221 } else {
222 printk("pc : %016llx\n", regs->pc);
223 printk("lr : %016llx\n", lr);
224 }
225
226 printk("sp : %016llx\n", sp);
227
228 if (system_uses_irq_prio_masking())
229 printk("pmr_save: %08llx\n", regs->pmr_save);
230
231 i = top_reg;
232
233 while (i >= 0) {
234 printk("x%-2d: %016llx", i, regs->regs[i]);
235
236 while (i-- % 3)
237 pr_cont(" x%-2d: %016llx", i, regs->regs[i]);
238
239 pr_cont("\n");
240 }
241}
242
243void show_regs(struct pt_regs *regs)
244{
245 __show_regs(regs);
246 dump_backtrace(regs, NULL, KERN_DEFAULT);
247}
248
249static void tls_thread_flush(void)
250{
251 write_sysreg(0, tpidr_el0);
252 if (system_supports_tpidr2())
253 write_sysreg_s(0, SYS_TPIDR2_EL0);
254
255 if (is_compat_task()) {
256 current->thread.uw.tp_value = 0;
257
258 /*
259 * We need to ensure ordering between the shadow state and the
260 * hardware state, so that we don't corrupt the hardware state
261 * with a stale shadow state during context switch.
262 */
263 barrier();
264 write_sysreg(0, tpidrro_el0);
265 }
266}
267
268static void flush_tagged_addr_state(void)
269{
270 if (IS_ENABLED(CONFIG_ARM64_TAGGED_ADDR_ABI))
271 clear_thread_flag(TIF_TAGGED_ADDR);
272}
273
274void flush_thread(void)
275{
276 fpsimd_flush_thread();
277 tls_thread_flush();
278 flush_ptrace_hw_breakpoint(current);
279 flush_tagged_addr_state();
280}
281
282void arch_release_task_struct(struct task_struct *tsk)
283{
284 fpsimd_release_task(tsk);
285}
286
287int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
288{
289 if (current->mm)
290 fpsimd_preserve_current_state();
291 *dst = *src;
292
293 /* We rely on the above assignment to initialize dst's thread_flags: */
294 BUILD_BUG_ON(!IS_ENABLED(CONFIG_THREAD_INFO_IN_TASK));
295
296 /*
297 * Detach src's sve_state (if any) from dst so that it does not
298 * get erroneously used or freed prematurely. dst's copies
299 * will be allocated on demand later on if dst uses SVE.
300 * For consistency, also clear TIF_SVE here: this could be done
301 * later in copy_process(), but to avoid tripping up future
302 * maintainers it is best not to leave TIF flags and buffers in
303 * an inconsistent state, even temporarily.
304 */
305 dst->thread.sve_state = NULL;
306 clear_tsk_thread_flag(dst, TIF_SVE);
307
308 /*
309 * In the unlikely event that we create a new thread with ZA
310 * enabled we should retain the ZA and ZT state so duplicate
311 * it here. This may be shortly freed if we exec() or if
312 * CLONE_SETTLS but it's simpler to do it here. To avoid
313 * confusing the rest of the code ensure that we have a
314 * sve_state allocated whenever sme_state is allocated.
315 */
316 if (thread_za_enabled(&src->thread)) {
317 dst->thread.sve_state = kzalloc(sve_state_size(src),
318 GFP_KERNEL);
319 if (!dst->thread.sve_state)
320 return -ENOMEM;
321
322 dst->thread.sme_state = kmemdup(src->thread.sme_state,
323 sme_state_size(src),
324 GFP_KERNEL);
325 if (!dst->thread.sme_state) {
326 kfree(dst->thread.sve_state);
327 dst->thread.sve_state = NULL;
328 return -ENOMEM;
329 }
330 } else {
331 dst->thread.sme_state = NULL;
332 clear_tsk_thread_flag(dst, TIF_SME);
333 }
334
335 dst->thread.fp_type = FP_STATE_FPSIMD;
336
337 /* clear any pending asynchronous tag fault raised by the parent */
338 clear_tsk_thread_flag(dst, TIF_MTE_ASYNC_FAULT);
339
340 return 0;
341}
342
343asmlinkage void ret_from_fork(void) asm("ret_from_fork");
344
345int copy_thread(struct task_struct *p, const struct kernel_clone_args *args)
346{
347 unsigned long clone_flags = args->flags;
348 unsigned long stack_start = args->stack;
349 unsigned long tls = args->tls;
350 struct pt_regs *childregs = task_pt_regs(p);
351
352 memset(&p->thread.cpu_context, 0, sizeof(struct cpu_context));
353
354 /*
355 * In case p was allocated the same task_struct pointer as some
356 * other recently-exited task, make sure p is disassociated from
357 * any cpu that may have run that now-exited task recently.
358 * Otherwise we could erroneously skip reloading the FPSIMD
359 * registers for p.
360 */
361 fpsimd_flush_task_state(p);
362
363 ptrauth_thread_init_kernel(p);
364
365 if (likely(!args->fn)) {
366 *childregs = *current_pt_regs();
367 childregs->regs[0] = 0;
368
369 /*
370 * Read the current TLS pointer from tpidr_el0 as it may be
371 * out-of-sync with the saved value.
372 */
373 *task_user_tls(p) = read_sysreg(tpidr_el0);
374 if (system_supports_tpidr2())
375 p->thread.tpidr2_el0 = read_sysreg_s(SYS_TPIDR2_EL0);
376
377 if (stack_start) {
378 if (is_compat_thread(task_thread_info(p)))
379 childregs->compat_sp = stack_start;
380 else
381 childregs->sp = stack_start;
382 }
383
384 /*
385 * If a TLS pointer was passed to clone, use it for the new
386 * thread. We also reset TPIDR2 if it's in use.
387 */
388 if (clone_flags & CLONE_SETTLS) {
389 p->thread.uw.tp_value = tls;
390 p->thread.tpidr2_el0 = 0;
391 }
392 } else {
393 /*
394 * A kthread has no context to ERET to, so ensure any buggy
395 * ERET is treated as an illegal exception return.
396 *
397 * When a user task is created from a kthread, childregs will
398 * be initialized by start_thread() or start_compat_thread().
399 */
400 memset(childregs, 0, sizeof(struct pt_regs));
401 childregs->pstate = PSR_MODE_EL1h | PSR_IL_BIT;
402
403 p->thread.cpu_context.x19 = (unsigned long)args->fn;
404 p->thread.cpu_context.x20 = (unsigned long)args->fn_arg;
405 }
406 p->thread.cpu_context.pc = (unsigned long)ret_from_fork;
407 p->thread.cpu_context.sp = (unsigned long)childregs;
408 /*
409 * For the benefit of the unwinder, set up childregs->stackframe
410 * as the final frame for the new task.
411 */
412 p->thread.cpu_context.fp = (unsigned long)childregs->stackframe;
413
414 ptrace_hw_copy_thread(p);
415
416 return 0;
417}
418
419void tls_preserve_current_state(void)
420{
421 *task_user_tls(current) = read_sysreg(tpidr_el0);
422 if (system_supports_tpidr2() && !is_compat_task())
423 current->thread.tpidr2_el0 = read_sysreg_s(SYS_TPIDR2_EL0);
424}
425
426static void tls_thread_switch(struct task_struct *next)
427{
428 tls_preserve_current_state();
429
430 if (is_compat_thread(task_thread_info(next)))
431 write_sysreg(next->thread.uw.tp_value, tpidrro_el0);
432 else if (!arm64_kernel_unmapped_at_el0())
433 write_sysreg(0, tpidrro_el0);
434
435 write_sysreg(*task_user_tls(next), tpidr_el0);
436 if (system_supports_tpidr2())
437 write_sysreg_s(next->thread.tpidr2_el0, SYS_TPIDR2_EL0);
438}
439
440/*
441 * Force SSBS state on context-switch, since it may be lost after migrating
442 * from a CPU which treats the bit as RES0 in a heterogeneous system.
443 */
444static void ssbs_thread_switch(struct task_struct *next)
445{
446 /*
447 * Nothing to do for kernel threads, but 'regs' may be junk
448 * (e.g. idle task) so check the flags and bail early.
449 */
450 if (unlikely(next->flags & PF_KTHREAD))
451 return;
452
453 /*
454 * If all CPUs implement the SSBS extension, then we just need to
455 * context-switch the PSTATE field.
456 */
457 if (alternative_has_cap_unlikely(ARM64_SSBS))
458 return;
459
460 spectre_v4_enable_task_mitigation(next);
461}
462
463/*
464 * We store our current task in sp_el0, which is clobbered by userspace. Keep a
465 * shadow copy so that we can restore this upon entry from userspace.
466 *
467 * This is *only* for exception entry from EL0, and is not valid until we
468 * __switch_to() a user task.
469 */
470DEFINE_PER_CPU(struct task_struct *, __entry_task);
471
472static void entry_task_switch(struct task_struct *next)
473{
474 __this_cpu_write(__entry_task, next);
475}
476
477/*
478 * ARM erratum 1418040 handling, affecting the 32bit view of CNTVCT.
479 * Ensure access is disabled when switching to a 32bit task, ensure
480 * access is enabled when switching to a 64bit task.
481 */
482static void erratum_1418040_thread_switch(struct task_struct *next)
483{
484 if (!IS_ENABLED(CONFIG_ARM64_ERRATUM_1418040) ||
485 !this_cpu_has_cap(ARM64_WORKAROUND_1418040))
486 return;
487
488 if (is_compat_thread(task_thread_info(next)))
489 sysreg_clear_set(cntkctl_el1, ARCH_TIMER_USR_VCT_ACCESS_EN, 0);
490 else
491 sysreg_clear_set(cntkctl_el1, 0, ARCH_TIMER_USR_VCT_ACCESS_EN);
492}
493
494static void erratum_1418040_new_exec(void)
495{
496 preempt_disable();
497 erratum_1418040_thread_switch(current);
498 preempt_enable();
499}
500
501/*
502 * __switch_to() checks current->thread.sctlr_user as an optimisation. Therefore
503 * this function must be called with preemption disabled and the update to
504 * sctlr_user must be made in the same preemption disabled block so that
505 * __switch_to() does not see the variable update before the SCTLR_EL1 one.
506 */
507void update_sctlr_el1(u64 sctlr)
508{
509 /*
510 * EnIA must not be cleared while in the kernel as this is necessary for
511 * in-kernel PAC. It will be cleared on kernel exit if needed.
512 */
513 sysreg_clear_set(sctlr_el1, SCTLR_USER_MASK & ~SCTLR_ELx_ENIA, sctlr);
514
515 /* ISB required for the kernel uaccess routines when setting TCF0. */
516 isb();
517}
518
519/*
520 * Thread switching.
521 */
522__notrace_funcgraph __sched
523struct task_struct *__switch_to(struct task_struct *prev,
524 struct task_struct *next)
525{
526 struct task_struct *last;
527
528 fpsimd_thread_switch(next);
529 tls_thread_switch(next);
530 hw_breakpoint_thread_switch(next);
531 contextidr_thread_switch(next);
532 entry_task_switch(next);
533 ssbs_thread_switch(next);
534 erratum_1418040_thread_switch(next);
535 ptrauth_thread_switch_user(next);
536
537 /*
538 * Complete any pending TLB or cache maintenance on this CPU in case
539 * the thread migrates to a different CPU.
540 * This full barrier is also required by the membarrier system
541 * call.
542 */
543 dsb(ish);
544
545 /*
546 * MTE thread switching must happen after the DSB above to ensure that
547 * any asynchronous tag check faults have been logged in the TFSR*_EL1
548 * registers.
549 */
550 mte_thread_switch(next);
551 /* avoid expensive SCTLR_EL1 accesses if no change */
552 if (prev->thread.sctlr_user != next->thread.sctlr_user)
553 update_sctlr_el1(next->thread.sctlr_user);
554
555 /* the actual thread switch */
556 last = cpu_switch_to(prev, next);
557
558 return last;
559}
560
561struct wchan_info {
562 unsigned long pc;
563 int count;
564};
565
566static bool get_wchan_cb(void *arg, unsigned long pc)
567{
568 struct wchan_info *wchan_info = arg;
569
570 if (!in_sched_functions(pc)) {
571 wchan_info->pc = pc;
572 return false;
573 }
574 return wchan_info->count++ < 16;
575}
576
577unsigned long __get_wchan(struct task_struct *p)
578{
579 struct wchan_info wchan_info = {
580 .pc = 0,
581 .count = 0,
582 };
583
584 if (!try_get_task_stack(p))
585 return 0;
586
587 arch_stack_walk(get_wchan_cb, &wchan_info, p, NULL);
588
589 put_task_stack(p);
590
591 return wchan_info.pc;
592}
593
594unsigned long arch_align_stack(unsigned long sp)
595{
596 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
597 sp -= get_random_u32_below(PAGE_SIZE);
598 return sp & ~0xf;
599}
600
601#ifdef CONFIG_COMPAT
602int compat_elf_check_arch(const struct elf32_hdr *hdr)
603{
604 if (!system_supports_32bit_el0())
605 return false;
606
607 if ((hdr)->e_machine != EM_ARM)
608 return false;
609
610 if (!((hdr)->e_flags & EF_ARM_EABI_MASK))
611 return false;
612
613 /*
614 * Prevent execve() of a 32-bit program from a deadline task
615 * if the restricted affinity mask would be inadmissible on an
616 * asymmetric system.
617 */
618 return !static_branch_unlikely(&arm64_mismatched_32bit_el0) ||
619 !dl_task_check_affinity(current, system_32bit_el0_cpumask());
620}
621#endif
622
623/*
624 * Called from setup_new_exec() after (COMPAT_)SET_PERSONALITY.
625 */
626void arch_setup_new_exec(void)
627{
628 unsigned long mmflags = 0;
629
630 if (is_compat_task()) {
631 mmflags = MMCF_AARCH32;
632
633 /*
634 * Restrict the CPU affinity mask for a 32-bit task so that
635 * it contains only 32-bit-capable CPUs.
636 *
637 * From the perspective of the task, this looks similar to
638 * what would happen if the 64-bit-only CPUs were hot-unplugged
639 * at the point of execve(), although we try a bit harder to
640 * honour the cpuset hierarchy.
641 */
642 if (static_branch_unlikely(&arm64_mismatched_32bit_el0))
643 force_compatible_cpus_allowed_ptr(current);
644 } else if (static_branch_unlikely(&arm64_mismatched_32bit_el0)) {
645 relax_compatible_cpus_allowed_ptr(current);
646 }
647
648 current->mm->context.flags = mmflags;
649 ptrauth_thread_init_user();
650 mte_thread_init_user();
651 erratum_1418040_new_exec();
652
653 if (task_spec_ssb_noexec(current)) {
654 arch_prctl_spec_ctrl_set(current, PR_SPEC_STORE_BYPASS,
655 PR_SPEC_ENABLE);
656 }
657}
658
659#ifdef CONFIG_ARM64_TAGGED_ADDR_ABI
660/*
661 * Control the relaxed ABI allowing tagged user addresses into the kernel.
662 */
663static unsigned int tagged_addr_disabled;
664
665long set_tagged_addr_ctrl(struct task_struct *task, unsigned long arg)
666{
667 unsigned long valid_mask = PR_TAGGED_ADDR_ENABLE;
668 struct thread_info *ti = task_thread_info(task);
669
670 if (is_compat_thread(ti))
671 return -EINVAL;
672
673 if (system_supports_mte())
674 valid_mask |= PR_MTE_TCF_SYNC | PR_MTE_TCF_ASYNC \
675 | PR_MTE_TAG_MASK;
676
677 if (arg & ~valid_mask)
678 return -EINVAL;
679
680 /*
681 * Do not allow the enabling of the tagged address ABI if globally
682 * disabled via sysctl abi.tagged_addr_disabled.
683 */
684 if (arg & PR_TAGGED_ADDR_ENABLE && tagged_addr_disabled)
685 return -EINVAL;
686
687 if (set_mte_ctrl(task, arg) != 0)
688 return -EINVAL;
689
690 update_ti_thread_flag(ti, TIF_TAGGED_ADDR, arg & PR_TAGGED_ADDR_ENABLE);
691
692 return 0;
693}
694
695long get_tagged_addr_ctrl(struct task_struct *task)
696{
697 long ret = 0;
698 struct thread_info *ti = task_thread_info(task);
699
700 if (is_compat_thread(ti))
701 return -EINVAL;
702
703 if (test_ti_thread_flag(ti, TIF_TAGGED_ADDR))
704 ret = PR_TAGGED_ADDR_ENABLE;
705
706 ret |= get_mte_ctrl(task);
707
708 return ret;
709}
710
711/*
712 * Global sysctl to disable the tagged user addresses support. This control
713 * only prevents the tagged address ABI enabling via prctl() and does not
714 * disable it for tasks that already opted in to the relaxed ABI.
715 */
716
717static struct ctl_table tagged_addr_sysctl_table[] = {
718 {
719 .procname = "tagged_addr_disabled",
720 .mode = 0644,
721 .data = &tagged_addr_disabled,
722 .maxlen = sizeof(int),
723 .proc_handler = proc_dointvec_minmax,
724 .extra1 = SYSCTL_ZERO,
725 .extra2 = SYSCTL_ONE,
726 },
727};
728
729static int __init tagged_addr_init(void)
730{
731 if (!register_sysctl("abi", tagged_addr_sysctl_table))
732 return -EINVAL;
733 return 0;
734}
735
736core_initcall(tagged_addr_init);
737#endif /* CONFIG_ARM64_TAGGED_ADDR_ABI */
738
739#ifdef CONFIG_BINFMT_ELF
740int arch_elf_adjust_prot(int prot, const struct arch_elf_state *state,
741 bool has_interp, bool is_interp)
742{
743 /*
744 * For dynamically linked executables the interpreter is
745 * responsible for setting PROT_BTI on everything except
746 * itself.
747 */
748 if (is_interp != has_interp)
749 return prot;
750
751 if (!(state->flags & ARM64_ELF_BTI))
752 return prot;
753
754 if (prot & PROT_EXEC)
755 prot |= PROT_BTI;
756
757 return prot;
758}
759#endif
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Based on arch/arm/kernel/process.c
4 *
5 * Original Copyright (C) 1995 Linus Torvalds
6 * Copyright (C) 1996-2000 Russell King - Converted to ARM.
7 * Copyright (C) 2012 ARM Ltd.
8 */
9
10#include <stdarg.h>
11
12#include <linux/compat.h>
13#include <linux/efi.h>
14#include <linux/export.h>
15#include <linux/sched.h>
16#include <linux/sched/debug.h>
17#include <linux/sched/task.h>
18#include <linux/sched/task_stack.h>
19#include <linux/kernel.h>
20#include <linux/lockdep.h>
21#include <linux/mm.h>
22#include <linux/stddef.h>
23#include <linux/sysctl.h>
24#include <linux/unistd.h>
25#include <linux/user.h>
26#include <linux/delay.h>
27#include <linux/reboot.h>
28#include <linux/interrupt.h>
29#include <linux/init.h>
30#include <linux/cpu.h>
31#include <linux/elfcore.h>
32#include <linux/pm.h>
33#include <linux/tick.h>
34#include <linux/utsname.h>
35#include <linux/uaccess.h>
36#include <linux/random.h>
37#include <linux/hw_breakpoint.h>
38#include <linux/personality.h>
39#include <linux/notifier.h>
40#include <trace/events/power.h>
41#include <linux/percpu.h>
42#include <linux/thread_info.h>
43#include <linux/prctl.h>
44
45#include <asm/alternative.h>
46#include <asm/arch_gicv3.h>
47#include <asm/compat.h>
48#include <asm/cpufeature.h>
49#include <asm/cacheflush.h>
50#include <asm/exec.h>
51#include <asm/fpsimd.h>
52#include <asm/mmu_context.h>
53#include <asm/processor.h>
54#include <asm/pointer_auth.h>
55#include <asm/stacktrace.h>
56
57#if defined(CONFIG_STACKPROTECTOR) && !defined(CONFIG_STACKPROTECTOR_PER_TASK)
58#include <linux/stackprotector.h>
59unsigned long __stack_chk_guard __read_mostly;
60EXPORT_SYMBOL(__stack_chk_guard);
61#endif
62
63/*
64 * Function pointers to optional machine specific functions
65 */
66void (*pm_power_off)(void);
67EXPORT_SYMBOL_GPL(pm_power_off);
68
69void (*arm_pm_restart)(enum reboot_mode reboot_mode, const char *cmd);
70
71static void __cpu_do_idle(void)
72{
73 dsb(sy);
74 wfi();
75}
76
77static void __cpu_do_idle_irqprio(void)
78{
79 unsigned long pmr;
80 unsigned long daif_bits;
81
82 daif_bits = read_sysreg(daif);
83 write_sysreg(daif_bits | PSR_I_BIT, daif);
84
85 /*
86 * Unmask PMR before going idle to make sure interrupts can
87 * be raised.
88 */
89 pmr = gic_read_pmr();
90 gic_write_pmr(GIC_PRIO_IRQON | GIC_PRIO_PSR_I_SET);
91
92 __cpu_do_idle();
93
94 gic_write_pmr(pmr);
95 write_sysreg(daif_bits, daif);
96}
97
98/*
99 * cpu_do_idle()
100 *
101 * Idle the processor (wait for interrupt).
102 *
103 * If the CPU supports priority masking we must do additional work to
104 * ensure that interrupts are not masked at the PMR (because the core will
105 * not wake up if we block the wake up signal in the interrupt controller).
106 */
107void cpu_do_idle(void)
108{
109 if (system_uses_irq_prio_masking())
110 __cpu_do_idle_irqprio();
111 else
112 __cpu_do_idle();
113}
114
115/*
116 * This is our default idle handler.
117 */
118void arch_cpu_idle(void)
119{
120 /*
121 * This should do all the clock switching and wait for interrupt
122 * tricks
123 */
124 trace_cpu_idle_rcuidle(1, smp_processor_id());
125 cpu_do_idle();
126 local_irq_enable();
127 trace_cpu_idle_rcuidle(PWR_EVENT_EXIT, smp_processor_id());
128}
129
130#ifdef CONFIG_HOTPLUG_CPU
131void arch_cpu_idle_dead(void)
132{
133 cpu_die();
134}
135#endif
136
137/*
138 * Called by kexec, immediately prior to machine_kexec().
139 *
140 * This must completely disable all secondary CPUs; simply causing those CPUs
141 * to execute e.g. a RAM-based pin loop is not sufficient. This allows the
142 * kexec'd kernel to use any and all RAM as it sees fit, without having to
143 * avoid any code or data used by any SW CPU pin loop. The CPU hotplug
144 * functionality embodied in disable_nonboot_cpus() to achieve this.
145 */
146void machine_shutdown(void)
147{
148 disable_nonboot_cpus();
149}
150
151/*
152 * Halting simply requires that the secondary CPUs stop performing any
153 * activity (executing tasks, handling interrupts). smp_send_stop()
154 * achieves this.
155 */
156void machine_halt(void)
157{
158 local_irq_disable();
159 smp_send_stop();
160 while (1);
161}
162
163/*
164 * Power-off simply requires that the secondary CPUs stop performing any
165 * activity (executing tasks, handling interrupts). smp_send_stop()
166 * achieves this. When the system power is turned off, it will take all CPUs
167 * with it.
168 */
169void machine_power_off(void)
170{
171 local_irq_disable();
172 smp_send_stop();
173 if (pm_power_off)
174 pm_power_off();
175}
176
177/*
178 * Restart requires that the secondary CPUs stop performing any activity
179 * while the primary CPU resets the system. Systems with multiple CPUs must
180 * provide a HW restart implementation, to ensure that all CPUs reset at once.
181 * This is required so that any code running after reset on the primary CPU
182 * doesn't have to co-ordinate with other CPUs to ensure they aren't still
183 * executing pre-reset code, and using RAM that the primary CPU's code wishes
184 * to use. Implementing such co-ordination would be essentially impossible.
185 */
186void machine_restart(char *cmd)
187{
188 /* Disable interrupts first */
189 local_irq_disable();
190 smp_send_stop();
191
192 /*
193 * UpdateCapsule() depends on the system being reset via
194 * ResetSystem().
195 */
196 if (efi_enabled(EFI_RUNTIME_SERVICES))
197 efi_reboot(reboot_mode, NULL);
198
199 /* Now call the architecture specific reboot code. */
200 if (arm_pm_restart)
201 arm_pm_restart(reboot_mode, cmd);
202 else
203 do_kernel_restart(cmd);
204
205 /*
206 * Whoops - the architecture was unable to reboot.
207 */
208 printk("Reboot failed -- System halted\n");
209 while (1);
210}
211
212static void print_pstate(struct pt_regs *regs)
213{
214 u64 pstate = regs->pstate;
215
216 if (compat_user_mode(regs)) {
217 printk("pstate: %08llx (%c%c%c%c %c %s %s %c%c%c)\n",
218 pstate,
219 pstate & PSR_AA32_N_BIT ? 'N' : 'n',
220 pstate & PSR_AA32_Z_BIT ? 'Z' : 'z',
221 pstate & PSR_AA32_C_BIT ? 'C' : 'c',
222 pstate & PSR_AA32_V_BIT ? 'V' : 'v',
223 pstate & PSR_AA32_Q_BIT ? 'Q' : 'q',
224 pstate & PSR_AA32_T_BIT ? "T32" : "A32",
225 pstate & PSR_AA32_E_BIT ? "BE" : "LE",
226 pstate & PSR_AA32_A_BIT ? 'A' : 'a',
227 pstate & PSR_AA32_I_BIT ? 'I' : 'i',
228 pstate & PSR_AA32_F_BIT ? 'F' : 'f');
229 } else {
230 printk("pstate: %08llx (%c%c%c%c %c%c%c%c %cPAN %cUAO)\n",
231 pstate,
232 pstate & PSR_N_BIT ? 'N' : 'n',
233 pstate & PSR_Z_BIT ? 'Z' : 'z',
234 pstate & PSR_C_BIT ? 'C' : 'c',
235 pstate & PSR_V_BIT ? 'V' : 'v',
236 pstate & PSR_D_BIT ? 'D' : 'd',
237 pstate & PSR_A_BIT ? 'A' : 'a',
238 pstate & PSR_I_BIT ? 'I' : 'i',
239 pstate & PSR_F_BIT ? 'F' : 'f',
240 pstate & PSR_PAN_BIT ? '+' : '-',
241 pstate & PSR_UAO_BIT ? '+' : '-');
242 }
243}
244
245void __show_regs(struct pt_regs *regs)
246{
247 int i, top_reg;
248 u64 lr, sp;
249
250 if (compat_user_mode(regs)) {
251 lr = regs->compat_lr;
252 sp = regs->compat_sp;
253 top_reg = 12;
254 } else {
255 lr = regs->regs[30];
256 sp = regs->sp;
257 top_reg = 29;
258 }
259
260 show_regs_print_info(KERN_DEFAULT);
261 print_pstate(regs);
262
263 if (!user_mode(regs)) {
264 printk("pc : %pS\n", (void *)regs->pc);
265 printk("lr : %pS\n", (void *)lr);
266 } else {
267 printk("pc : %016llx\n", regs->pc);
268 printk("lr : %016llx\n", lr);
269 }
270
271 printk("sp : %016llx\n", sp);
272
273 if (system_uses_irq_prio_masking())
274 printk("pmr_save: %08llx\n", regs->pmr_save);
275
276 i = top_reg;
277
278 while (i >= 0) {
279 printk("x%-2d: %016llx ", i, regs->regs[i]);
280 i--;
281
282 if (i % 2 == 0) {
283 pr_cont("x%-2d: %016llx ", i, regs->regs[i]);
284 i--;
285 }
286
287 pr_cont("\n");
288 }
289}
290
291void show_regs(struct pt_regs * regs)
292{
293 __show_regs(regs);
294 dump_backtrace(regs, NULL);
295}
296
297static void tls_thread_flush(void)
298{
299 write_sysreg(0, tpidr_el0);
300
301 if (is_compat_task()) {
302 current->thread.uw.tp_value = 0;
303
304 /*
305 * We need to ensure ordering between the shadow state and the
306 * hardware state, so that we don't corrupt the hardware state
307 * with a stale shadow state during context switch.
308 */
309 barrier();
310 write_sysreg(0, tpidrro_el0);
311 }
312}
313
314static void flush_tagged_addr_state(void)
315{
316 if (IS_ENABLED(CONFIG_ARM64_TAGGED_ADDR_ABI))
317 clear_thread_flag(TIF_TAGGED_ADDR);
318}
319
320void flush_thread(void)
321{
322 fpsimd_flush_thread();
323 tls_thread_flush();
324 flush_ptrace_hw_breakpoint(current);
325 flush_tagged_addr_state();
326}
327
328void release_thread(struct task_struct *dead_task)
329{
330}
331
332void arch_release_task_struct(struct task_struct *tsk)
333{
334 fpsimd_release_task(tsk);
335}
336
337int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
338{
339 if (current->mm)
340 fpsimd_preserve_current_state();
341 *dst = *src;
342
343 /* We rely on the above assignment to initialize dst's thread_flags: */
344 BUILD_BUG_ON(!IS_ENABLED(CONFIG_THREAD_INFO_IN_TASK));
345
346 /*
347 * Detach src's sve_state (if any) from dst so that it does not
348 * get erroneously used or freed prematurely. dst's sve_state
349 * will be allocated on demand later on if dst uses SVE.
350 * For consistency, also clear TIF_SVE here: this could be done
351 * later in copy_process(), but to avoid tripping up future
352 * maintainers it is best not to leave TIF_SVE and sve_state in
353 * an inconsistent state, even temporarily.
354 */
355 dst->thread.sve_state = NULL;
356 clear_tsk_thread_flag(dst, TIF_SVE);
357
358 return 0;
359}
360
361asmlinkage void ret_from_fork(void) asm("ret_from_fork");
362
363int copy_thread(unsigned long clone_flags, unsigned long stack_start,
364 unsigned long stk_sz, struct task_struct *p)
365{
366 struct pt_regs *childregs = task_pt_regs(p);
367
368 memset(&p->thread.cpu_context, 0, sizeof(struct cpu_context));
369
370 /*
371 * In case p was allocated the same task_struct pointer as some
372 * other recently-exited task, make sure p is disassociated from
373 * any cpu that may have run that now-exited task recently.
374 * Otherwise we could erroneously skip reloading the FPSIMD
375 * registers for p.
376 */
377 fpsimd_flush_task_state(p);
378
379 if (likely(!(p->flags & PF_KTHREAD))) {
380 *childregs = *current_pt_regs();
381 childregs->regs[0] = 0;
382
383 /*
384 * Read the current TLS pointer from tpidr_el0 as it may be
385 * out-of-sync with the saved value.
386 */
387 *task_user_tls(p) = read_sysreg(tpidr_el0);
388
389 if (stack_start) {
390 if (is_compat_thread(task_thread_info(p)))
391 childregs->compat_sp = stack_start;
392 else
393 childregs->sp = stack_start;
394 }
395
396 /*
397 * If a TLS pointer was passed to clone (4th argument), use it
398 * for the new thread.
399 */
400 if (clone_flags & CLONE_SETTLS)
401 p->thread.uw.tp_value = childregs->regs[3];
402 } else {
403 memset(childregs, 0, sizeof(struct pt_regs));
404 childregs->pstate = PSR_MODE_EL1h;
405 if (IS_ENABLED(CONFIG_ARM64_UAO) &&
406 cpus_have_const_cap(ARM64_HAS_UAO))
407 childregs->pstate |= PSR_UAO_BIT;
408
409 if (arm64_get_ssbd_state() == ARM64_SSBD_FORCE_DISABLE)
410 set_ssbs_bit(childregs);
411
412 if (system_uses_irq_prio_masking())
413 childregs->pmr_save = GIC_PRIO_IRQON;
414
415 p->thread.cpu_context.x19 = stack_start;
416 p->thread.cpu_context.x20 = stk_sz;
417 }
418 p->thread.cpu_context.pc = (unsigned long)ret_from_fork;
419 p->thread.cpu_context.sp = (unsigned long)childregs;
420
421 ptrace_hw_copy_thread(p);
422
423 return 0;
424}
425
426void tls_preserve_current_state(void)
427{
428 *task_user_tls(current) = read_sysreg(tpidr_el0);
429}
430
431static void tls_thread_switch(struct task_struct *next)
432{
433 tls_preserve_current_state();
434
435 if (is_compat_thread(task_thread_info(next)))
436 write_sysreg(next->thread.uw.tp_value, tpidrro_el0);
437 else if (!arm64_kernel_unmapped_at_el0())
438 write_sysreg(0, tpidrro_el0);
439
440 write_sysreg(*task_user_tls(next), tpidr_el0);
441}
442
443/* Restore the UAO state depending on next's addr_limit */
444void uao_thread_switch(struct task_struct *next)
445{
446 if (IS_ENABLED(CONFIG_ARM64_UAO)) {
447 if (task_thread_info(next)->addr_limit == KERNEL_DS)
448 asm(ALTERNATIVE("nop", SET_PSTATE_UAO(1), ARM64_HAS_UAO));
449 else
450 asm(ALTERNATIVE("nop", SET_PSTATE_UAO(0), ARM64_HAS_UAO));
451 }
452}
453
454/*
455 * Force SSBS state on context-switch, since it may be lost after migrating
456 * from a CPU which treats the bit as RES0 in a heterogeneous system.
457 */
458static void ssbs_thread_switch(struct task_struct *next)
459{
460 struct pt_regs *regs = task_pt_regs(next);
461
462 /*
463 * Nothing to do for kernel threads, but 'regs' may be junk
464 * (e.g. idle task) so check the flags and bail early.
465 */
466 if (unlikely(next->flags & PF_KTHREAD))
467 return;
468
469 /* If the mitigation is enabled, then we leave SSBS clear. */
470 if ((arm64_get_ssbd_state() == ARM64_SSBD_FORCE_ENABLE) ||
471 test_tsk_thread_flag(next, TIF_SSBD))
472 return;
473
474 if (compat_user_mode(regs))
475 set_compat_ssbs_bit(regs);
476 else if (user_mode(regs))
477 set_ssbs_bit(regs);
478}
479
480/*
481 * We store our current task in sp_el0, which is clobbered by userspace. Keep a
482 * shadow copy so that we can restore this upon entry from userspace.
483 *
484 * This is *only* for exception entry from EL0, and is not valid until we
485 * __switch_to() a user task.
486 */
487DEFINE_PER_CPU(struct task_struct *, __entry_task);
488
489static void entry_task_switch(struct task_struct *next)
490{
491 __this_cpu_write(__entry_task, next);
492}
493
494/*
495 * Thread switching.
496 */
497__notrace_funcgraph struct task_struct *__switch_to(struct task_struct *prev,
498 struct task_struct *next)
499{
500 struct task_struct *last;
501
502 fpsimd_thread_switch(next);
503 tls_thread_switch(next);
504 hw_breakpoint_thread_switch(next);
505 contextidr_thread_switch(next);
506 entry_task_switch(next);
507 uao_thread_switch(next);
508 ptrauth_thread_switch(next);
509 ssbs_thread_switch(next);
510
511 /*
512 * Complete any pending TLB or cache maintenance on this CPU in case
513 * the thread migrates to a different CPU.
514 * This full barrier is also required by the membarrier system
515 * call.
516 */
517 dsb(ish);
518
519 /* the actual thread switch */
520 last = cpu_switch_to(prev, next);
521
522 return last;
523}
524
525unsigned long get_wchan(struct task_struct *p)
526{
527 struct stackframe frame;
528 unsigned long stack_page, ret = 0;
529 int count = 0;
530 if (!p || p == current || p->state == TASK_RUNNING)
531 return 0;
532
533 stack_page = (unsigned long)try_get_task_stack(p);
534 if (!stack_page)
535 return 0;
536
537 start_backtrace(&frame, thread_saved_fp(p), thread_saved_pc(p));
538
539 do {
540 if (unwind_frame(p, &frame))
541 goto out;
542 if (!in_sched_functions(frame.pc)) {
543 ret = frame.pc;
544 goto out;
545 }
546 } while (count ++ < 16);
547
548out:
549 put_task_stack(p);
550 return ret;
551}
552
553unsigned long arch_align_stack(unsigned long sp)
554{
555 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
556 sp -= get_random_int() & ~PAGE_MASK;
557 return sp & ~0xf;
558}
559
560/*
561 * Called from setup_new_exec() after (COMPAT_)SET_PERSONALITY.
562 */
563void arch_setup_new_exec(void)
564{
565 current->mm->context.flags = is_compat_task() ? MMCF_AARCH32 : 0;
566
567 ptrauth_thread_init_user(current);
568}
569
570#ifdef CONFIG_ARM64_TAGGED_ADDR_ABI
571/*
572 * Control the relaxed ABI allowing tagged user addresses into the kernel.
573 */
574static unsigned int tagged_addr_disabled;
575
576long set_tagged_addr_ctrl(unsigned long arg)
577{
578 if (is_compat_task())
579 return -EINVAL;
580 if (arg & ~PR_TAGGED_ADDR_ENABLE)
581 return -EINVAL;
582
583 /*
584 * Do not allow the enabling of the tagged address ABI if globally
585 * disabled via sysctl abi.tagged_addr_disabled.
586 */
587 if (arg & PR_TAGGED_ADDR_ENABLE && tagged_addr_disabled)
588 return -EINVAL;
589
590 update_thread_flag(TIF_TAGGED_ADDR, arg & PR_TAGGED_ADDR_ENABLE);
591
592 return 0;
593}
594
595long get_tagged_addr_ctrl(void)
596{
597 if (is_compat_task())
598 return -EINVAL;
599
600 if (test_thread_flag(TIF_TAGGED_ADDR))
601 return PR_TAGGED_ADDR_ENABLE;
602
603 return 0;
604}
605
606/*
607 * Global sysctl to disable the tagged user addresses support. This control
608 * only prevents the tagged address ABI enabling via prctl() and does not
609 * disable it for tasks that already opted in to the relaxed ABI.
610 */
611static int zero;
612static int one = 1;
613
614static struct ctl_table tagged_addr_sysctl_table[] = {
615 {
616 .procname = "tagged_addr_disabled",
617 .mode = 0644,
618 .data = &tagged_addr_disabled,
619 .maxlen = sizeof(int),
620 .proc_handler = proc_dointvec_minmax,
621 .extra1 = &zero,
622 .extra2 = &one,
623 },
624 { }
625};
626
627static int __init tagged_addr_init(void)
628{
629 if (!register_sysctl("abi", tagged_addr_sysctl_table))
630 return -EINVAL;
631 return 0;
632}
633
634core_initcall(tagged_addr_init);
635#endif /* CONFIG_ARM64_TAGGED_ADDR_ABI */
636
637asmlinkage void __sched arm64_preempt_schedule_irq(void)
638{
639 lockdep_assert_irqs_disabled();
640
641 /*
642 * Preempting a task from an IRQ means we leave copies of PSTATE
643 * on the stack. cpufeature's enable calls may modify PSTATE, but
644 * resuming one of these preempted tasks would undo those changes.
645 *
646 * Only allow a task to be preempted once cpufeatures have been
647 * enabled.
648 */
649 if (static_branch_likely(&arm64_const_caps_ready))
650 preempt_schedule_irq();
651}