• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 #include <trace/hooks/fpsimd.h>
45 #include <trace/hooks/mpam.h>
46 
47 #include <asm/alternative.h>
48 #include <asm/compat.h>
49 #include <asm/cpufeature.h>
50 #include <asm/cacheflush.h>
51 #include <asm/exec.h>
52 #include <asm/fpsimd.h>
53 #include <asm/mmu_context.h>
54 #include <asm/mte.h>
55 #include <asm/processor.h>
56 #include <asm/pointer_auth.h>
57 #include <asm/stacktrace.h>
58 #include <asm/switch_to.h>
59 #include <asm/system_misc.h>
60 
61 #if defined(CONFIG_STACKPROTECTOR) && !defined(CONFIG_STACKPROTECTOR_PER_TASK)
62 #include <linux/stackprotector.h>
63 unsigned long __stack_chk_guard __ro_after_init;
64 EXPORT_SYMBOL(__stack_chk_guard);
65 #endif
66 
67 /*
68  * Function pointers to optional machine specific functions
69  */
70 void (*pm_power_off)(void);
71 EXPORT_SYMBOL_GPL(pm_power_off);
72 
73 #ifdef CONFIG_HOTPLUG_CPU
arch_cpu_idle_dead(void)74 void arch_cpu_idle_dead(void)
75 {
76        cpu_die();
77 }
78 #endif
79 
80 /*
81  * Called by kexec, immediately prior to machine_kexec().
82  *
83  * This must completely disable all secondary CPUs; simply causing those CPUs
84  * to execute e.g. a RAM-based pin loop is not sufficient. This allows the
85  * kexec'd kernel to use any and all RAM as it sees fit, without having to
86  * avoid any code or data used by any SW CPU pin loop. The CPU hotplug
87  * functionality embodied in smpt_shutdown_nonboot_cpus() to achieve this.
88  */
machine_shutdown(void)89 void machine_shutdown(void)
90 {
91 	smp_shutdown_nonboot_cpus(reboot_cpu);
92 }
93 
94 /*
95  * Halting simply requires that the secondary CPUs stop performing any
96  * activity (executing tasks, handling interrupts). smp_send_stop()
97  * achieves this.
98  */
machine_halt(void)99 void machine_halt(void)
100 {
101 	local_irq_disable();
102 	smp_send_stop();
103 	while (1);
104 }
105 
106 /*
107  * Power-off simply requires that the secondary CPUs stop performing any
108  * activity (executing tasks, handling interrupts). smp_send_stop()
109  * achieves this. When the system power is turned off, it will take all CPUs
110  * with it.
111  */
machine_power_off(void)112 void machine_power_off(void)
113 {
114 	local_irq_disable();
115 	smp_send_stop();
116 	if (pm_power_off)
117 		pm_power_off();
118 }
119 
120 /*
121  * Restart requires that the secondary CPUs stop performing any activity
122  * while the primary CPU resets the system. Systems with multiple CPUs must
123  * provide a HW restart implementation, to ensure that all CPUs reset at once.
124  * This is required so that any code running after reset on the primary CPU
125  * doesn't have to co-ordinate with other CPUs to ensure they aren't still
126  * executing pre-reset code, and using RAM that the primary CPU's code wishes
127  * to use. Implementing such co-ordination would be essentially impossible.
128  */
machine_restart(char * cmd)129 void machine_restart(char *cmd)
130 {
131 	/* Disable interrupts first */
132 	local_irq_disable();
133 	smp_send_stop();
134 
135 	/*
136 	 * UpdateCapsule() depends on the system being reset via
137 	 * ResetSystem().
138 	 */
139 	if (efi_enabled(EFI_RUNTIME_SERVICES))
140 		efi_reboot(reboot_mode, NULL);
141 
142 	/* Now call the architecture specific reboot code. */
143 	do_kernel_restart(cmd);
144 
145 	/*
146 	 * Whoops - the architecture was unable to reboot.
147 	 */
148 	printk("Reboot failed -- System halted\n");
149 	while (1);
150 }
151 
152 #define bstr(suffix, str) [PSR_BTYPE_ ## suffix >> PSR_BTYPE_SHIFT] = str
153 static const char *const btypes[] = {
154 	bstr(NONE, "--"),
155 	bstr(  JC, "jc"),
156 	bstr(   C, "-c"),
157 	bstr(  J , "j-")
158 };
159 #undef bstr
160 
print_pstate(struct pt_regs * regs)161 static void print_pstate(struct pt_regs *regs)
162 {
163 	u64 pstate = regs->pstate;
164 
165 	if (compat_user_mode(regs)) {
166 		printk("pstate: %08llx (%c%c%c%c %c %s %s %c%c%c %cDIT %cSSBS)\n",
167 			pstate,
168 			pstate & PSR_AA32_N_BIT ? 'N' : 'n',
169 			pstate & PSR_AA32_Z_BIT ? 'Z' : 'z',
170 			pstate & PSR_AA32_C_BIT ? 'C' : 'c',
171 			pstate & PSR_AA32_V_BIT ? 'V' : 'v',
172 			pstate & PSR_AA32_Q_BIT ? 'Q' : 'q',
173 			pstate & PSR_AA32_T_BIT ? "T32" : "A32",
174 			pstate & PSR_AA32_E_BIT ? "BE" : "LE",
175 			pstate & PSR_AA32_A_BIT ? 'A' : 'a',
176 			pstate & PSR_AA32_I_BIT ? 'I' : 'i',
177 			pstate & PSR_AA32_F_BIT ? 'F' : 'f',
178 			pstate & PSR_AA32_DIT_BIT ? '+' : '-',
179 			pstate & PSR_AA32_SSBS_BIT ? '+' : '-');
180 	} else {
181 		const char *btype_str = btypes[(pstate & PSR_BTYPE_MASK) >>
182 					       PSR_BTYPE_SHIFT];
183 
184 		printk("pstate: %08llx (%c%c%c%c %c%c%c%c %cPAN %cUAO %cTCO %cDIT %cSSBS BTYPE=%s)\n",
185 			pstate,
186 			pstate & PSR_N_BIT ? 'N' : 'n',
187 			pstate & PSR_Z_BIT ? 'Z' : 'z',
188 			pstate & PSR_C_BIT ? 'C' : 'c',
189 			pstate & PSR_V_BIT ? 'V' : 'v',
190 			pstate & PSR_D_BIT ? 'D' : 'd',
191 			pstate & PSR_A_BIT ? 'A' : 'a',
192 			pstate & PSR_I_BIT ? 'I' : 'i',
193 			pstate & PSR_F_BIT ? 'F' : 'f',
194 			pstate & PSR_PAN_BIT ? '+' : '-',
195 			pstate & PSR_UAO_BIT ? '+' : '-',
196 			pstate & PSR_TCO_BIT ? '+' : '-',
197 			pstate & PSR_DIT_BIT ? '+' : '-',
198 			pstate & PSR_SSBS_BIT ? '+' : '-',
199 			btype_str);
200 	}
201 }
202 
__show_regs(struct pt_regs * regs)203 void __show_regs(struct pt_regs *regs)
204 {
205 	int i, top_reg;
206 	u64 lr, sp;
207 
208 	if (compat_user_mode(regs)) {
209 		lr = regs->compat_lr;
210 		sp = regs->compat_sp;
211 		top_reg = 12;
212 	} else {
213 		lr = regs->regs[30];
214 		sp = regs->sp;
215 		top_reg = 29;
216 	}
217 
218 	show_regs_print_info(KERN_DEFAULT);
219 	print_pstate(regs);
220 
221 	if (!user_mode(regs)) {
222 		printk("pc : %pS\n", (void *)regs->pc);
223 		printk("lr : %pS\n", (void *)ptrauth_strip_insn_pac(lr));
224 	} else {
225 		printk("pc : %016llx\n", regs->pc);
226 		printk("lr : %016llx\n", lr);
227 	}
228 
229 	printk("sp : %016llx\n", sp);
230 
231 	if (system_uses_irq_prio_masking())
232 		printk("pmr_save: %08llx\n", regs->pmr_save);
233 
234 	i = top_reg;
235 
236 	while (i >= 0) {
237 		printk("x%-2d: %016llx", i, regs->regs[i]);
238 
239 		while (i-- % 3)
240 			pr_cont(" x%-2d: %016llx", i, regs->regs[i]);
241 
242 		pr_cont("\n");
243 	}
244 }
245 
show_regs(struct pt_regs * regs)246 void show_regs(struct pt_regs *regs)
247 {
248 	__show_regs(regs);
249 	dump_backtrace(regs, NULL, KERN_DEFAULT);
250 }
251 EXPORT_SYMBOL_GPL(show_regs);
252 
tls_thread_flush(void)253 static void tls_thread_flush(void)
254 {
255 	write_sysreg(0, tpidr_el0);
256 	if (system_supports_tpidr2())
257 		write_sysreg_s(0, SYS_TPIDR2_EL0);
258 
259 	if (is_compat_task()) {
260 		current->thread.uw.tp_value = 0;
261 
262 		/*
263 		 * We need to ensure ordering between the shadow state and the
264 		 * hardware state, so that we don't corrupt the hardware state
265 		 * with a stale shadow state during context switch.
266 		 */
267 		barrier();
268 		write_sysreg(0, tpidrro_el0);
269 	}
270 }
271 
flush_tagged_addr_state(void)272 static void flush_tagged_addr_state(void)
273 {
274 	if (IS_ENABLED(CONFIG_ARM64_TAGGED_ADDR_ABI))
275 		clear_thread_flag(TIF_TAGGED_ADDR);
276 }
277 
flush_thread(void)278 void flush_thread(void)
279 {
280 	fpsimd_flush_thread();
281 	tls_thread_flush();
282 	flush_ptrace_hw_breakpoint(current);
283 	flush_tagged_addr_state();
284 }
285 
release_thread(struct task_struct * dead_task)286 void release_thread(struct task_struct *dead_task)
287 {
288 }
289 
arch_release_task_struct(struct task_struct * tsk)290 void arch_release_task_struct(struct task_struct *tsk)
291 {
292 	fpsimd_release_task(tsk);
293 }
294 
arch_dup_task_struct(struct task_struct * dst,struct task_struct * src)295 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
296 {
297 	if (current->mm)
298 		fpsimd_preserve_current_state();
299 	*dst = *src;
300 
301 	/* We rely on the above assignment to initialize dst's thread_flags: */
302 	BUILD_BUG_ON(!IS_ENABLED(CONFIG_THREAD_INFO_IN_TASK));
303 
304 	/*
305 	 * Detach src's sve_state (if any) from dst so that it does not
306 	 * get erroneously used or freed prematurely.  dst's copies
307 	 * will be allocated on demand later on if dst uses SVE.
308 	 * For consistency, also clear TIF_SVE here: this could be done
309 	 * later in copy_process(), but to avoid tripping up future
310 	 * maintainers it is best not to leave TIF flags and buffers in
311 	 * an inconsistent state, even temporarily.
312 	 */
313 	dst->thread.sve_state = NULL;
314 	clear_tsk_thread_flag(dst, TIF_SVE);
315 
316 	/*
317 	 * In the unlikely event that we create a new thread with ZA
318 	 * enabled we should retain the ZA state so duplicate it here.
319 	 * This may be shortly freed if we exec() or if CLONE_SETTLS
320 	 * but it's simpler to do it here. To avoid confusing the rest
321 	 * of the code ensure that we have a sve_state allocated
322 	 * whenever za_state is allocated.
323 	 */
324 	if (thread_za_enabled(&src->thread)) {
325 		dst->thread.sve_state = kzalloc(sve_state_size(src),
326 						GFP_KERNEL);
327 		if (!dst->thread.sve_state)
328 			return -ENOMEM;
329 		dst->thread.za_state = kmemdup(src->thread.za_state,
330 					       za_state_size(src),
331 					       GFP_KERNEL);
332 		if (!dst->thread.za_state) {
333 			kfree(dst->thread.sve_state);
334 			dst->thread.sve_state = NULL;
335 			return -ENOMEM;
336 		}
337 	} else {
338 		dst->thread.za_state = NULL;
339 		clear_tsk_thread_flag(dst, TIF_SME);
340 	}
341 
342 	/* clear any pending asynchronous tag fault raised by the parent */
343 	clear_tsk_thread_flag(dst, TIF_MTE_ASYNC_FAULT);
344 
345 	return 0;
346 }
347 
348 asmlinkage void ret_from_fork(void) asm("ret_from_fork");
349 
copy_thread(unsigned long clone_flags,unsigned long stack_start,unsigned long stk_sz,struct task_struct * p,unsigned long tls)350 int copy_thread(unsigned long clone_flags, unsigned long stack_start,
351 		unsigned long stk_sz, struct task_struct *p, unsigned long tls)
352 {
353 	struct pt_regs *childregs = task_pt_regs(p);
354 
355 	memset(&p->thread.cpu_context, 0, sizeof(struct cpu_context));
356 
357 	/*
358 	 * In case p was allocated the same task_struct pointer as some
359 	 * other recently-exited task, make sure p is disassociated from
360 	 * any cpu that may have run that now-exited task recently.
361 	 * Otherwise we could erroneously skip reloading the FPSIMD
362 	 * registers for p.
363 	 */
364 	fpsimd_flush_task_state(p);
365 
366 	ptrauth_thread_init_kernel(p);
367 
368 	if (likely(!(p->flags & (PF_KTHREAD | PF_IO_WORKER)))) {
369 		*childregs = *current_pt_regs();
370 		childregs->regs[0] = 0;
371 
372 		/*
373 		 * Read the current TLS pointer from tpidr_el0 as it may be
374 		 * out-of-sync with the saved value.
375 		 */
376 		*task_user_tls(p) = read_sysreg(tpidr_el0);
377 		if (system_supports_tpidr2())
378 			p->thread.tpidr2_el0 = read_sysreg_s(SYS_TPIDR2_EL0);
379 
380 		if (stack_start) {
381 			if (is_compat_thread(task_thread_info(p)))
382 				childregs->compat_sp = stack_start;
383 			else
384 				childregs->sp = stack_start;
385 		}
386 
387 		/*
388 		 * If a TLS pointer was passed to clone, use it for the new
389 		 * thread.  We also reset TPIDR2 if it's in use.
390 		 */
391 		if (clone_flags & CLONE_SETTLS) {
392 			p->thread.uw.tp_value = tls;
393 			p->thread.tpidr2_el0 = 0;
394 		}
395 	} else {
396 		/*
397 		 * A kthread has no context to ERET to, so ensure any buggy
398 		 * ERET is treated as an illegal exception return.
399 		 *
400 		 * When a user task is created from a kthread, childregs will
401 		 * be initialized by start_thread() or start_compat_thread().
402 		 */
403 		memset(childregs, 0, sizeof(struct pt_regs));
404 		childregs->pstate = PSR_MODE_EL1h | PSR_IL_BIT;
405 
406 		p->thread.cpu_context.x19 = stack_start;
407 		p->thread.cpu_context.x20 = stk_sz;
408 	}
409 	p->thread.cpu_context.pc = (unsigned long)ret_from_fork;
410 	p->thread.cpu_context.sp = (unsigned long)childregs;
411 	/*
412 	 * For the benefit of the unwinder, set up childregs->stackframe
413 	 * as the final frame for the new task.
414 	 */
415 	p->thread.cpu_context.fp = (unsigned long)childregs->stackframe;
416 
417 	ptrace_hw_copy_thread(p);
418 
419 	return 0;
420 }
421 
tls_preserve_current_state(void)422 void tls_preserve_current_state(void)
423 {
424 	*task_user_tls(current) = read_sysreg(tpidr_el0);
425 	if (system_supports_tpidr2() && !is_compat_task())
426 		current->thread.tpidr2_el0 = read_sysreg_s(SYS_TPIDR2_EL0);
427 }
428 
tls_thread_switch(struct task_struct * next)429 static void tls_thread_switch(struct task_struct *next)
430 {
431 	tls_preserve_current_state();
432 
433 	if (is_compat_thread(task_thread_info(next)))
434 		write_sysreg(next->thread.uw.tp_value, tpidrro_el0);
435 	else if (!arm64_kernel_unmapped_at_el0())
436 		write_sysreg(0, tpidrro_el0);
437 
438 	write_sysreg(*task_user_tls(next), tpidr_el0);
439 	if (system_supports_tpidr2())
440 		write_sysreg_s(next->thread.tpidr2_el0, SYS_TPIDR2_EL0);
441 }
442 
443 /*
444  * Force SSBS state on context-switch, since it may be lost after migrating
445  * from a CPU which treats the bit as RES0 in a heterogeneous system.
446  */
ssbs_thread_switch(struct task_struct * next)447 static void ssbs_thread_switch(struct task_struct *next)
448 {
449 	/*
450 	 * Nothing to do for kernel threads, but 'regs' may be junk
451 	 * (e.g. idle task) so check the flags and bail early.
452 	 */
453 	if (unlikely(next->flags & PF_KTHREAD))
454 		return;
455 
456 	/*
457 	 * If all CPUs implement the SSBS extension, then we just need to
458 	 * context-switch the PSTATE field.
459 	 */
460 	if (cpus_have_const_cap(ARM64_SSBS))
461 		return;
462 
463 	spectre_v4_enable_task_mitigation(next);
464 }
465 
466 /*
467  * We store our current task in sp_el0, which is clobbered by userspace. Keep a
468  * shadow copy so that we can restore this upon entry from userspace.
469  *
470  * This is *only* for exception entry from EL0, and is not valid until we
471  * __switch_to() a user task.
472  */
473 DEFINE_PER_CPU(struct task_struct *, __entry_task);
474 
entry_task_switch(struct task_struct * next)475 static void entry_task_switch(struct task_struct *next)
476 {
477 	__this_cpu_write(__entry_task, next);
478 }
479 
480 /*
481  * ARM erratum 1418040 handling, affecting the 32bit view of CNTVCT.
482  * Ensure access is disabled when switching to a 32bit task, ensure
483  * access is enabled when switching to a 64bit task.
484  */
erratum_1418040_thread_switch(struct task_struct * next)485 static void erratum_1418040_thread_switch(struct task_struct *next)
486 {
487 	if (!IS_ENABLED(CONFIG_ARM64_ERRATUM_1418040) ||
488 	    !this_cpu_has_cap(ARM64_WORKAROUND_1418040))
489 		return;
490 
491 	if (is_compat_thread(task_thread_info(next)))
492 		sysreg_clear_set(cntkctl_el1, ARCH_TIMER_USR_VCT_ACCESS_EN, 0);
493 	else
494 		sysreg_clear_set(cntkctl_el1, 0, ARCH_TIMER_USR_VCT_ACCESS_EN);
495 }
496 
erratum_1418040_new_exec(void)497 static void erratum_1418040_new_exec(void)
498 {
499 	preempt_disable();
500 	erratum_1418040_thread_switch(current);
501 	preempt_enable();
502 }
503 
504 /*
505  * __switch_to() checks current->thread.sctlr_user as an optimisation. Therefore
506  * this function must be called with preemption disabled and the update to
507  * sctlr_user must be made in the same preemption disabled block so that
508  * __switch_to() does not see the variable update before the SCTLR_EL1 one.
509  */
update_sctlr_el1(u64 sctlr)510 void update_sctlr_el1(u64 sctlr)
511 {
512 	/*
513 	 * EnIA must not be cleared while in the kernel as this is necessary for
514 	 * in-kernel PAC. It will be cleared on kernel exit if needed.
515 	 */
516 	sysreg_clear_set(sctlr_el1, SCTLR_USER_MASK & ~SCTLR_ELx_ENIA, sctlr);
517 
518 	/* ISB required for the kernel uaccess routines when setting TCF0. */
519 	isb();
520 }
521 
522 /*
523  * Thread switching.
524  */
525 __notrace_funcgraph __sched
__switch_to(struct task_struct * prev,struct task_struct * next)526 struct task_struct *__switch_to(struct task_struct *prev,
527 				struct task_struct *next)
528 {
529 	struct task_struct *last;
530 
531 	fpsimd_thread_switch(next);
532 	tls_thread_switch(next);
533 	hw_breakpoint_thread_switch(next);
534 	contextidr_thread_switch(next);
535 	entry_task_switch(next);
536 	ssbs_thread_switch(next);
537 	erratum_1418040_thread_switch(next);
538 	ptrauth_thread_switch_user(next);
539 
540 	/*
541 	 *  vendor hook is needed before the dsb(),
542 	 *  because MPAM is related to cache maintenance.
543 	 */
544 	trace_android_vh_mpam_set(prev, next);
545 
546 	/*
547 	 * Complete any pending TLB or cache maintenance on this CPU in case
548 	 * the thread migrates to a different CPU.
549 	 * This full barrier is also required by the membarrier system
550 	 * call.
551 	 */
552 	dsb(ish);
553 
554 	/*
555 	 * MTE thread switching must happen after the DSB above to ensure that
556 	 * any asynchronous tag check faults have been logged in the TFSR*_EL1
557 	 * registers.
558 	 */
559 	mte_thread_switch(next);
560 	/* avoid expensive SCTLR_EL1 accesses if no change */
561 	if (prev->thread.sctlr_user != next->thread.sctlr_user)
562 		update_sctlr_el1(next->thread.sctlr_user);
563 
564 	trace_android_vh_is_fpsimd_save(prev, next);
565 
566 	/* the actual thread switch */
567 	last = cpu_switch_to(prev, next);
568 
569 	return last;
570 }
571 
572 struct wchan_info {
573 	unsigned long	pc;
574 	int		count;
575 };
576 
get_wchan_cb(void * arg,unsigned long pc)577 static bool get_wchan_cb(void *arg, unsigned long pc)
578 {
579 	struct wchan_info *wchan_info = arg;
580 
581 	if (!in_sched_functions(pc)) {
582 		wchan_info->pc = pc;
583 		return false;
584 	}
585 	return wchan_info->count++ < 16;
586 }
587 
get_wchan(struct task_struct * p)588 unsigned long get_wchan(struct task_struct *p)
589 {
590 	struct wchan_info wchan_info = {
591 		.pc = 0,
592 		.count = 0,
593 	};
594 
595 	if (!p || p == current || task_is_running(p))
596 		return 0;
597 
598 	if (!try_get_task_stack(p))
599 		return 0;
600 
601 	arch_stack_walk(get_wchan_cb, &wchan_info, p, NULL);
602 
603 	put_task_stack(p);
604 
605 	return wchan_info.pc;
606 }
607 EXPORT_SYMBOL_GPL(get_wchan);
608 
arch_align_stack(unsigned long sp)609 unsigned long arch_align_stack(unsigned long sp)
610 {
611 	if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
612 		sp -= get_random_int() & ~PAGE_MASK;
613 	return sp & ~0xf;
614 }
615 
616 #ifdef CONFIG_COMPAT
compat_elf_check_arch(const struct elf32_hdr * hdr)617 int compat_elf_check_arch(const struct elf32_hdr *hdr)
618 {
619 	if (!system_supports_32bit_el0())
620 		return false;
621 
622 	if ((hdr)->e_machine != EM_ARM)
623 		return false;
624 
625 	if (!((hdr)->e_flags & EF_ARM_EABI_MASK))
626 		return false;
627 
628 	/*
629 	 * Prevent execve() of a 32-bit program from a deadline task
630 	 * if the restricted affinity mask would be inadmissible on an
631 	 * asymmetric system.
632 	 */
633 	return !static_branch_unlikely(&arm64_mismatched_32bit_el0) ||
634 	       !dl_task_check_affinity(current, system_32bit_el0_cpumask());
635 }
636 #endif
637 
638 /*
639  * Called from setup_new_exec() after (COMPAT_)SET_PERSONALITY.
640  */
arch_setup_new_exec(void)641 void arch_setup_new_exec(void)
642 {
643 	unsigned long mmflags = 0;
644 
645 	if (is_compat_task()) {
646 		mmflags = MMCF_AARCH32;
647 
648 		/*
649 		 * Restrict the CPU affinity mask for a 32-bit task so that
650 		 * it contains only 32-bit-capable CPUs.
651 		 *
652 		 * From the perspective of the task, this looks similar to
653 		 * what would happen if the 64-bit-only CPUs were hot-unplugged
654 		 * at the point of execve(), although we try a bit harder to
655 		 * honour the cpuset hierarchy.
656 		 */
657 		if (static_branch_unlikely(&arm64_mismatched_32bit_el0))
658 			force_compatible_cpus_allowed_ptr(current);
659 	} else if (static_branch_unlikely(&arm64_mismatched_32bit_el0)) {
660 		relax_compatible_cpus_allowed_ptr(current);
661 	}
662 
663 	current->mm->context.flags = mmflags;
664 	ptrauth_thread_init_user();
665 	mte_thread_init_user();
666 	erratum_1418040_new_exec();
667 
668 	if (task_spec_ssb_noexec(current)) {
669 		arch_prctl_spec_ctrl_set(current, PR_SPEC_STORE_BYPASS,
670 					 PR_SPEC_ENABLE);
671 	}
672 }
673 
674 #ifdef CONFIG_ARM64_TAGGED_ADDR_ABI
675 /*
676  * Control the relaxed ABI allowing tagged user addresses into the kernel.
677  */
678 static unsigned int tagged_addr_disabled;
679 
set_tagged_addr_ctrl(struct task_struct * task,unsigned long arg)680 long set_tagged_addr_ctrl(struct task_struct *task, unsigned long arg)
681 {
682 	unsigned long valid_mask = PR_TAGGED_ADDR_ENABLE;
683 	struct thread_info *ti = task_thread_info(task);
684 
685 	if (is_compat_thread(ti))
686 		return -EINVAL;
687 
688 	if (system_supports_mte())
689 		valid_mask |= PR_MTE_TCF_SYNC | PR_MTE_TCF_ASYNC \
690 			| PR_MTE_TAG_MASK;
691 
692 	if (arg & ~valid_mask)
693 		return -EINVAL;
694 
695 	/*
696 	 * Do not allow the enabling of the tagged address ABI if globally
697 	 * disabled via sysctl abi.tagged_addr_disabled.
698 	 */
699 	if (arg & PR_TAGGED_ADDR_ENABLE && tagged_addr_disabled)
700 		return -EINVAL;
701 
702 	if (set_mte_ctrl(task, arg) != 0)
703 		return -EINVAL;
704 
705 	update_ti_thread_flag(ti, TIF_TAGGED_ADDR, arg & PR_TAGGED_ADDR_ENABLE);
706 
707 	return 0;
708 }
709 
get_tagged_addr_ctrl(struct task_struct * task)710 long get_tagged_addr_ctrl(struct task_struct *task)
711 {
712 	long ret = 0;
713 	struct thread_info *ti = task_thread_info(task);
714 
715 	if (is_compat_thread(ti))
716 		return -EINVAL;
717 
718 	if (test_ti_thread_flag(ti, TIF_TAGGED_ADDR))
719 		ret = PR_TAGGED_ADDR_ENABLE;
720 
721 	ret |= get_mte_ctrl(task);
722 
723 	return ret;
724 }
725 
726 /*
727  * Global sysctl to disable the tagged user addresses support. This control
728  * only prevents the tagged address ABI enabling via prctl() and does not
729  * disable it for tasks that already opted in to the relaxed ABI.
730  */
731 
732 static struct ctl_table tagged_addr_sysctl_table[] = {
733 	{
734 		.procname	= "tagged_addr_disabled",
735 		.mode		= 0644,
736 		.data		= &tagged_addr_disabled,
737 		.maxlen		= sizeof(int),
738 		.proc_handler	= proc_dointvec_minmax,
739 		.extra1		= SYSCTL_ZERO,
740 		.extra2		= SYSCTL_ONE,
741 	},
742 	{ }
743 };
744 
tagged_addr_init(void)745 static int __init tagged_addr_init(void)
746 {
747 	if (!register_sysctl("abi", tagged_addr_sysctl_table))
748 		return -EINVAL;
749 	return 0;
750 }
751 
752 core_initcall(tagged_addr_init);
753 #endif	/* CONFIG_ARM64_TAGGED_ADDR_ABI */
754 
755 #ifdef CONFIG_BINFMT_ELF
arch_elf_adjust_prot(int prot,const struct arch_elf_state * state,bool has_interp,bool is_interp)756 int arch_elf_adjust_prot(int prot, const struct arch_elf_state *state,
757 			 bool has_interp, bool is_interp)
758 {
759 	/*
760 	 * For dynamically linked executables the interpreter is
761 	 * responsible for setting PROT_BTI on everything except
762 	 * itself.
763 	 */
764 	if (is_interp != has_interp)
765 		return prot;
766 
767 	if (!(state->flags & ARM64_ELF_BTI))
768 		return prot;
769 
770 	if (prot & PROT_EXEC)
771 		prot |= PROT_BTI;
772 
773 	return prot;
774 }
775 #endif
776