1 // SPDX-License-Identifier: GPL-2.0-or-later
2
3 #include <linux/context_tracking.h>
4 #include <linux/err.h>
5 #include <linux/compat.h>
6 #include <linux/sched/debug.h> /* for show_regs */
7
8 #include <asm/kup.h>
9 #include <asm/cputime.h>
10 #include <asm/hw_irq.h>
11 #include <asm/interrupt.h>
12 #include <asm/kprobes.h>
13 #include <asm/paca.h>
14 #include <asm/ptrace.h>
15 #include <asm/reg.h>
16 #include <asm/signal.h>
17 #include <asm/switch_to.h>
18 #include <asm/syscall.h>
19 #include <asm/time.h>
20 #include <asm/tm.h>
21 #include <asm/unistd.h>
22
23 #if defined(CONFIG_PPC_ADV_DEBUG_REGS) && defined(CONFIG_PPC32)
24 unsigned long global_dbcr0[NR_CPUS];
25 #endif
26
27 #ifdef CONFIG_PPC_BOOK3S_64
28 DEFINE_STATIC_KEY_FALSE(interrupt_exit_not_reentrant);
exit_must_hard_disable(void)29 static inline bool exit_must_hard_disable(void)
30 {
31 return static_branch_unlikely(&interrupt_exit_not_reentrant);
32 }
33 #else
exit_must_hard_disable(void)34 static inline bool exit_must_hard_disable(void)
35 {
36 return true;
37 }
38 #endif
39
40 /*
41 * local irqs must be disabled. Returns false if the caller must re-enable
42 * them, check for new work, and try again.
43 *
44 * This should be called with local irqs disabled, but if they were previously
45 * enabled when the interrupt handler returns (indicating a process-context /
46 * synchronous interrupt) then irqs_enabled should be true.
47 *
48 * restartable is true then EE/RI can be left on because interrupts are handled
49 * with a restart sequence.
50 */
prep_irq_for_enabled_exit(bool restartable)51 static notrace __always_inline bool prep_irq_for_enabled_exit(bool restartable)
52 {
53 /* This must be done with RI=1 because tracing may touch vmaps */
54 trace_hardirqs_on();
55
56 if (exit_must_hard_disable() || !restartable)
57 __hard_EE_RI_disable();
58
59 #ifdef CONFIG_PPC64
60 /* This pattern matches prep_irq_for_idle */
61 if (unlikely(lazy_irq_pending_nocheck())) {
62 if (exit_must_hard_disable() || !restartable) {
63 local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
64 __hard_RI_enable();
65 }
66 trace_hardirqs_off();
67
68 return false;
69 }
70 #endif
71 return true;
72 }
73
booke_load_dbcr0(void)74 static notrace void booke_load_dbcr0(void)
75 {
76 #ifdef CONFIG_PPC_ADV_DEBUG_REGS
77 unsigned long dbcr0 = current->thread.debug.dbcr0;
78
79 if (likely(!(dbcr0 & DBCR0_IDM)))
80 return;
81
82 /*
83 * Check to see if the dbcr0 register is set up to debug.
84 * Use the internal debug mode bit to do this.
85 */
86 mtmsr(mfmsr() & ~MSR_DE);
87 if (IS_ENABLED(CONFIG_PPC32)) {
88 isync();
89 global_dbcr0[smp_processor_id()] = mfspr(SPRN_DBCR0);
90 }
91 mtspr(SPRN_DBCR0, dbcr0);
92 mtspr(SPRN_DBSR, -1);
93 #endif
94 }
95
check_return_regs_valid(struct pt_regs * regs)96 static void check_return_regs_valid(struct pt_regs *regs)
97 {
98 #ifdef CONFIG_PPC_BOOK3S_64
99 unsigned long trap, srr0, srr1;
100 static bool warned;
101 u8 *validp;
102 char *h;
103
104 if (trap_is_scv(regs))
105 return;
106
107 trap = TRAP(regs);
108 // EE in HV mode sets HSRRs like 0xea0
109 if (cpu_has_feature(CPU_FTR_HVMODE) && trap == INTERRUPT_EXTERNAL)
110 trap = 0xea0;
111
112 switch (trap) {
113 case 0x980:
114 case INTERRUPT_H_DATA_STORAGE:
115 case 0xe20:
116 case 0xe40:
117 case INTERRUPT_HMI:
118 case 0xe80:
119 case 0xea0:
120 case INTERRUPT_H_FAC_UNAVAIL:
121 case 0x1200:
122 case 0x1500:
123 case 0x1600:
124 case 0x1800:
125 validp = &local_paca->hsrr_valid;
126 if (!*validp)
127 return;
128
129 srr0 = mfspr(SPRN_HSRR0);
130 srr1 = mfspr(SPRN_HSRR1);
131 h = "H";
132
133 break;
134 default:
135 validp = &local_paca->srr_valid;
136 if (!*validp)
137 return;
138
139 srr0 = mfspr(SPRN_SRR0);
140 srr1 = mfspr(SPRN_SRR1);
141 h = "";
142 break;
143 }
144
145 if (srr0 == regs->nip && srr1 == regs->msr)
146 return;
147
148 /*
149 * A NMI / soft-NMI interrupt may have come in after we found
150 * srr_valid and before the SRRs are loaded. The interrupt then
151 * comes in and clobbers SRRs and clears srr_valid. Then we load
152 * the SRRs here and test them above and find they don't match.
153 *
154 * Test validity again after that, to catch such false positives.
155 *
156 * This test in general will have some window for false negatives
157 * and may not catch and fix all such cases if an NMI comes in
158 * later and clobbers SRRs without clearing srr_valid, but hopefully
159 * such things will get caught most of the time, statistically
160 * enough to be able to get a warning out.
161 */
162 barrier();
163
164 if (!*validp)
165 return;
166
167 if (!warned) {
168 warned = true;
169 printk("%sSRR0 was: %lx should be: %lx\n", h, srr0, regs->nip);
170 printk("%sSRR1 was: %lx should be: %lx\n", h, srr1, regs->msr);
171 show_regs(regs);
172 }
173
174 *validp = 0; /* fixup */
175 #endif
176 }
177
178 static notrace unsigned long
interrupt_exit_user_prepare_main(unsigned long ret,struct pt_regs * regs)179 interrupt_exit_user_prepare_main(unsigned long ret, struct pt_regs *regs)
180 {
181 unsigned long ti_flags;
182
183 again:
184 ti_flags = read_thread_flags();
185 while (unlikely(ti_flags & (_TIF_USER_WORK_MASK & ~_TIF_RESTORE_TM))) {
186 local_irq_enable();
187 if (ti_flags & _TIF_NEED_RESCHED) {
188 schedule();
189 } else {
190 /*
191 * SIGPENDING must restore signal handler function
192 * argument GPRs, and some non-volatiles (e.g., r1).
193 * Restore all for now. This could be made lighter.
194 */
195 if (ti_flags & _TIF_SIGPENDING)
196 ret |= _TIF_RESTOREALL;
197 do_notify_resume(regs, ti_flags);
198 }
199 local_irq_disable();
200 ti_flags = read_thread_flags();
201 }
202
203 if (IS_ENABLED(CONFIG_PPC_BOOK3S_64) && IS_ENABLED(CONFIG_PPC_FPU)) {
204 if (IS_ENABLED(CONFIG_PPC_TRANSACTIONAL_MEM) &&
205 unlikely((ti_flags & _TIF_RESTORE_TM))) {
206 restore_tm_state(regs);
207 } else {
208 unsigned long mathflags = MSR_FP;
209
210 if (cpu_has_feature(CPU_FTR_VSX))
211 mathflags |= MSR_VEC | MSR_VSX;
212 else if (cpu_has_feature(CPU_FTR_ALTIVEC))
213 mathflags |= MSR_VEC;
214
215 /*
216 * If userspace MSR has all available FP bits set,
217 * then they are live and no need to restore. If not,
218 * it means the regs were given up and restore_math
219 * may decide to restore them (to avoid taking an FP
220 * fault).
221 */
222 if ((regs->msr & mathflags) != mathflags)
223 restore_math(regs);
224 }
225 }
226
227 check_return_regs_valid(regs);
228
229 user_enter_irqoff();
230 if (!prep_irq_for_enabled_exit(true)) {
231 user_exit_irqoff();
232 local_irq_enable();
233 local_irq_disable();
234 goto again;
235 }
236
237 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
238 local_paca->tm_scratch = regs->msr;
239 #endif
240
241 booke_load_dbcr0();
242
243 account_cpu_user_exit();
244
245 /* Restore user access locks last */
246 kuap_user_restore(regs);
247
248 return ret;
249 }
250
251 /*
252 * This should be called after a syscall returns, with r3 the return value
253 * from the syscall. If this function returns non-zero, the system call
254 * exit assembly should additionally load all GPR registers and CTR and XER
255 * from the interrupt frame.
256 *
257 * The function graph tracer can not trace the return side of this function,
258 * because RI=0 and soft mask state is "unreconciled", so it is marked notrace.
259 */
syscall_exit_prepare(unsigned long r3,struct pt_regs * regs,long scv)260 notrace unsigned long syscall_exit_prepare(unsigned long r3,
261 struct pt_regs *regs,
262 long scv)
263 {
264 unsigned long ti_flags;
265 unsigned long ret = 0;
266 bool is_not_scv = !IS_ENABLED(CONFIG_PPC_BOOK3S_64) || !scv;
267
268 CT_WARN_ON(ct_state() == CONTEXT_USER);
269
270 kuap_assert_locked();
271
272 regs->result = r3;
273
274 /* Check whether the syscall is issued inside a restartable sequence */
275 rseq_syscall(regs);
276
277 ti_flags = read_thread_flags();
278
279 if (unlikely(r3 >= (unsigned long)-MAX_ERRNO) && is_not_scv) {
280 if (likely(!(ti_flags & (_TIF_NOERROR | _TIF_RESTOREALL)))) {
281 r3 = -r3;
282 regs->ccr |= 0x10000000; /* Set SO bit in CR */
283 }
284 }
285
286 if (unlikely(ti_flags & _TIF_PERSYSCALL_MASK)) {
287 if (ti_flags & _TIF_RESTOREALL)
288 ret = _TIF_RESTOREALL;
289 else
290 regs->gpr[3] = r3;
291 clear_bits(_TIF_PERSYSCALL_MASK, ¤t_thread_info()->flags);
292 } else {
293 regs->gpr[3] = r3;
294 }
295
296 if (unlikely(ti_flags & _TIF_SYSCALL_DOTRACE)) {
297 do_syscall_trace_leave(regs);
298 ret |= _TIF_RESTOREALL;
299 }
300
301 local_irq_disable();
302 ret = interrupt_exit_user_prepare_main(ret, regs);
303
304 #ifdef CONFIG_PPC64
305 regs->exit_result = ret;
306 #endif
307
308 return ret;
309 }
310
311 #ifdef CONFIG_PPC64
syscall_exit_restart(unsigned long r3,struct pt_regs * regs)312 notrace unsigned long syscall_exit_restart(unsigned long r3, struct pt_regs *regs)
313 {
314 /*
315 * This is called when detecting a soft-pending interrupt as well as
316 * an alternate-return interrupt. So we can't just have the alternate
317 * return path clear SRR1[MSR] and set PACA_IRQ_HARD_DIS (unless
318 * the soft-pending case were to fix things up as well). RI might be
319 * disabled, in which case it gets re-enabled by __hard_irq_disable().
320 */
321 __hard_irq_disable();
322 local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
323
324 #ifdef CONFIG_PPC_BOOK3S_64
325 set_kuap(AMR_KUAP_BLOCKED);
326 #endif
327
328 trace_hardirqs_off();
329 user_exit_irqoff();
330 account_cpu_user_entry();
331
332 BUG_ON(!user_mode(regs));
333
334 regs->exit_result = interrupt_exit_user_prepare_main(regs->exit_result, regs);
335
336 return regs->exit_result;
337 }
338 #endif
339
interrupt_exit_user_prepare(struct pt_regs * regs)340 notrace unsigned long interrupt_exit_user_prepare(struct pt_regs *regs)
341 {
342 unsigned long ret;
343
344 BUG_ON(regs_is_unrecoverable(regs));
345 BUG_ON(arch_irq_disabled_regs(regs));
346 CT_WARN_ON(ct_state() == CONTEXT_USER);
347
348 /*
349 * We don't need to restore AMR on the way back to userspace for KUAP.
350 * AMR can only have been unlocked if we interrupted the kernel.
351 */
352 kuap_assert_locked();
353
354 local_irq_disable();
355
356 ret = interrupt_exit_user_prepare_main(0, regs);
357
358 #ifdef CONFIG_PPC64
359 regs->exit_result = ret;
360 #endif
361
362 return ret;
363 }
364
365 void preempt_schedule_irq(void);
366
interrupt_exit_kernel_prepare(struct pt_regs * regs)367 notrace unsigned long interrupt_exit_kernel_prepare(struct pt_regs *regs)
368 {
369 unsigned long flags;
370 unsigned long ret = 0;
371 unsigned long kuap;
372 bool stack_store = read_thread_flags() & _TIF_EMULATE_STACK_STORE;
373
374 if (regs_is_unrecoverable(regs))
375 unrecoverable_exception(regs);
376 /*
377 * CT_WARN_ON comes here via program_check_exception, so avoid
378 * recursion.
379 *
380 * Skip the assertion on PMIs on 64e to work around a problem caused
381 * by NMI PMIs incorrectly taking this interrupt return path, it's
382 * possible for this to hit after interrupt exit to user switches
383 * context to user. See also the comment in the performance monitor
384 * handler in exceptions-64e.S
385 */
386 if (!IS_ENABLED(CONFIG_PPC_BOOK3E_64) &&
387 TRAP(regs) != INTERRUPT_PROGRAM &&
388 TRAP(regs) != INTERRUPT_PERFMON)
389 CT_WARN_ON(ct_state() == CONTEXT_USER);
390
391 kuap = kuap_get_and_assert_locked();
392
393 local_irq_save(flags);
394
395 if (!arch_irq_disabled_regs(regs)) {
396 /* Returning to a kernel context with local irqs enabled. */
397 WARN_ON_ONCE(!(regs->msr & MSR_EE));
398 again:
399 if (IS_ENABLED(CONFIG_PREEMPT)) {
400 /* Return to preemptible kernel context */
401 if (unlikely(read_thread_flags() & _TIF_NEED_RESCHED)) {
402 if (preempt_count() == 0)
403 preempt_schedule_irq();
404 }
405 }
406
407 check_return_regs_valid(regs);
408
409 /*
410 * Stack store exit can't be restarted because the interrupt
411 * stack frame might have been clobbered.
412 */
413 if (!prep_irq_for_enabled_exit(unlikely(stack_store))) {
414 /*
415 * Replay pending soft-masked interrupts now. Don't
416 * just local_irq_enabe(); local_irq_disable(); because
417 * if we are returning from an asynchronous interrupt
418 * here, another one might hit after irqs are enabled,
419 * and it would exit via this same path allowing
420 * another to fire, and so on unbounded.
421 */
422 hard_irq_disable();
423 replay_soft_interrupts();
424 /* Took an interrupt, may have more exit work to do. */
425 goto again;
426 }
427 #ifdef CONFIG_PPC64
428 /*
429 * An interrupt may clear MSR[EE] and set this concurrently,
430 * but it will be marked pending and the exit will be retried.
431 * This leaves a racy window where MSR[EE]=0 and HARD_DIS is
432 * clear, until interrupt_exit_kernel_restart() calls
433 * hard_irq_disable(), which will set HARD_DIS again.
434 */
435 local_paca->irq_happened &= ~PACA_IRQ_HARD_DIS;
436
437 } else {
438 check_return_regs_valid(regs);
439
440 if (unlikely(stack_store))
441 __hard_EE_RI_disable();
442 #endif /* CONFIG_PPC64 */
443 }
444
445 if (unlikely(stack_store)) {
446 clear_bits(_TIF_EMULATE_STACK_STORE, ¤t_thread_info()->flags);
447 ret = 1;
448 }
449
450 #ifdef CONFIG_PPC_TRANSACTIONAL_MEM
451 local_paca->tm_scratch = regs->msr;
452 #endif
453
454 /*
455 * 64s does not want to mfspr(SPRN_AMR) here, because this comes after
456 * mtmsr, which would cause Read-After-Write stalls. Hence, take the
457 * AMR value from the check above.
458 */
459 kuap_kernel_restore(regs, kuap);
460
461 return ret;
462 }
463
464 #ifdef CONFIG_PPC64
interrupt_exit_user_restart(struct pt_regs * regs)465 notrace unsigned long interrupt_exit_user_restart(struct pt_regs *regs)
466 {
467 __hard_irq_disable();
468 local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
469
470 #ifdef CONFIG_PPC_BOOK3S_64
471 set_kuap(AMR_KUAP_BLOCKED);
472 #endif
473
474 trace_hardirqs_off();
475 user_exit_irqoff();
476 account_cpu_user_entry();
477
478 BUG_ON(!user_mode(regs));
479
480 regs->exit_result |= interrupt_exit_user_prepare(regs);
481
482 return regs->exit_result;
483 }
484
485 /*
486 * No real need to return a value here because the stack store case does not
487 * get restarted.
488 */
interrupt_exit_kernel_restart(struct pt_regs * regs)489 notrace unsigned long interrupt_exit_kernel_restart(struct pt_regs *regs)
490 {
491 __hard_irq_disable();
492 local_paca->irq_happened |= PACA_IRQ_HARD_DIS;
493
494 #ifdef CONFIG_PPC_BOOK3S_64
495 set_kuap(AMR_KUAP_BLOCKED);
496 #endif
497
498 if (regs->softe == IRQS_ENABLED)
499 trace_hardirqs_off();
500
501 BUG_ON(user_mode(regs));
502
503 return interrupt_exit_kernel_prepare(regs);
504 }
505 #endif
506