Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Architecture-specific setup.
4 *
5 * Copyright (C) 1998-2003 Hewlett-Packard Co
6 * David Mosberger-Tang <davidm@hpl.hp.com>
7 * 04/11/17 Ashok Raj <ashok.raj@intel.com> Added CPU Hotplug Support
8 *
9 * 2005-10-07 Keith Owens <kaos@sgi.com>
10 * Add notify_die() hooks.
11 */
12#include <linux/cpu.h>
13#include <linux/pm.h>
14#include <linux/elf.h>
15#include <linux/errno.h>
16#include <linux/kernel.h>
17#include <linux/mm.h>
18#include <linux/slab.h>
19#include <linux/module.h>
20#include <linux/notifier.h>
21#include <linux/personality.h>
22#include <linux/sched.h>
23#include <linux/sched/debug.h>
24#include <linux/sched/hotplug.h>
25#include <linux/sched/task.h>
26#include <linux/sched/task_stack.h>
27#include <linux/stddef.h>
28#include <linux/thread_info.h>
29#include <linux/unistd.h>
30#include <linux/efi.h>
31#include <linux/interrupt.h>
32#include <linux/delay.h>
33#include <linux/kdebug.h>
34#include <linux/utsname.h>
35#include <linux/tracehook.h>
36#include <linux/rcupdate.h>
37
38#include <asm/cpu.h>
39#include <asm/delay.h>
40#include <asm/elf.h>
41#include <asm/irq.h>
42#include <asm/kexec.h>
43#include <asm/pgalloc.h>
44#include <asm/processor.h>
45#include <asm/sal.h>
46#include <asm/switch_to.h>
47#include <asm/tlbflush.h>
48#include <linux/uaccess.h>
49#include <asm/unwind.h>
50#include <asm/user.h>
51
52#include "entry.h"
53
54#ifdef CONFIG_PERFMON
55# include <asm/perfmon.h>
56#endif
57
58#include "sigframe.h"
59
60void (*ia64_mark_idle)(int);
61
62unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
63EXPORT_SYMBOL(boot_option_idle_override);
64void (*pm_power_off) (void);
65EXPORT_SYMBOL(pm_power_off);
66
67void
68ia64_do_show_stack (struct unw_frame_info *info, void *arg)
69{
70 unsigned long ip, sp, bsp;
71
72 printk("\nCall Trace:\n");
73 do {
74 unw_get_ip(info, &ip);
75 if (ip == 0)
76 break;
77
78 unw_get_sp(info, &sp);
79 unw_get_bsp(info, &bsp);
80 printk(" [<%016lx>] %pS\n"
81 " sp=%016lx bsp=%016lx\n",
82 ip, (void *)ip, sp, bsp);
83 } while (unw_unwind(info) >= 0);
84}
85
86void
87show_stack (struct task_struct *task, unsigned long *sp)
88{
89 if (!task)
90 unw_init_running(ia64_do_show_stack, NULL);
91 else {
92 struct unw_frame_info info;
93
94 unw_init_from_blocked_task(&info, task);
95 ia64_do_show_stack(&info, NULL);
96 }
97}
98
99void
100show_regs (struct pt_regs *regs)
101{
102 unsigned long ip = regs->cr_iip + ia64_psr(regs)->ri;
103
104 print_modules();
105 printk("\n");
106 show_regs_print_info(KERN_DEFAULT);
107 printk("psr : %016lx ifs : %016lx ip : [<%016lx>] %s (%s)\n",
108 regs->cr_ipsr, regs->cr_ifs, ip, print_tainted(),
109 init_utsname()->release);
110 printk("ip is at %pS\n", (void *)ip);
111 printk("unat: %016lx pfs : %016lx rsc : %016lx\n",
112 regs->ar_unat, regs->ar_pfs, regs->ar_rsc);
113 printk("rnat: %016lx bsps: %016lx pr : %016lx\n",
114 regs->ar_rnat, regs->ar_bspstore, regs->pr);
115 printk("ldrs: %016lx ccv : %016lx fpsr: %016lx\n",
116 regs->loadrs, regs->ar_ccv, regs->ar_fpsr);
117 printk("csd : %016lx ssd : %016lx\n", regs->ar_csd, regs->ar_ssd);
118 printk("b0 : %016lx b6 : %016lx b7 : %016lx\n", regs->b0, regs->b6, regs->b7);
119 printk("f6 : %05lx%016lx f7 : %05lx%016lx\n",
120 regs->f6.u.bits[1], regs->f6.u.bits[0],
121 regs->f7.u.bits[1], regs->f7.u.bits[0]);
122 printk("f8 : %05lx%016lx f9 : %05lx%016lx\n",
123 regs->f8.u.bits[1], regs->f8.u.bits[0],
124 regs->f9.u.bits[1], regs->f9.u.bits[0]);
125 printk("f10 : %05lx%016lx f11 : %05lx%016lx\n",
126 regs->f10.u.bits[1], regs->f10.u.bits[0],
127 regs->f11.u.bits[1], regs->f11.u.bits[0]);
128
129 printk("r1 : %016lx r2 : %016lx r3 : %016lx\n", regs->r1, regs->r2, regs->r3);
130 printk("r8 : %016lx r9 : %016lx r10 : %016lx\n", regs->r8, regs->r9, regs->r10);
131 printk("r11 : %016lx r12 : %016lx r13 : %016lx\n", regs->r11, regs->r12, regs->r13);
132 printk("r14 : %016lx r15 : %016lx r16 : %016lx\n", regs->r14, regs->r15, regs->r16);
133 printk("r17 : %016lx r18 : %016lx r19 : %016lx\n", regs->r17, regs->r18, regs->r19);
134 printk("r20 : %016lx r21 : %016lx r22 : %016lx\n", regs->r20, regs->r21, regs->r22);
135 printk("r23 : %016lx r24 : %016lx r25 : %016lx\n", regs->r23, regs->r24, regs->r25);
136 printk("r26 : %016lx r27 : %016lx r28 : %016lx\n", regs->r26, regs->r27, regs->r28);
137 printk("r29 : %016lx r30 : %016lx r31 : %016lx\n", regs->r29, regs->r30, regs->r31);
138
139 if (user_mode(regs)) {
140 /* print the stacked registers */
141 unsigned long val, *bsp, ndirty;
142 int i, sof, is_nat = 0;
143
144 sof = regs->cr_ifs & 0x7f; /* size of frame */
145 ndirty = (regs->loadrs >> 19);
146 bsp = ia64_rse_skip_regs((unsigned long *) regs->ar_bspstore, ndirty);
147 for (i = 0; i < sof; ++i) {
148 get_user(val, (unsigned long __user *) ia64_rse_skip_regs(bsp, i));
149 printk("r%-3u:%c%016lx%s", 32 + i, is_nat ? '*' : ' ', val,
150 ((i == sof - 1) || (i % 3) == 2) ? "\n" : " ");
151 }
152 } else
153 show_stack(NULL, NULL);
154}
155
156/* local support for deprecated console_print */
157void
158console_print(const char *s)
159{
160 printk(KERN_EMERG "%s", s);
161}
162
163void
164do_notify_resume_user(sigset_t *unused, struct sigscratch *scr, long in_syscall)
165{
166 if (fsys_mode(current, &scr->pt)) {
167 /*
168 * defer signal-handling etc. until we return to
169 * privilege-level 0.
170 */
171 if (!ia64_psr(&scr->pt)->lp)
172 ia64_psr(&scr->pt)->lp = 1;
173 return;
174 }
175
176#ifdef CONFIG_PERFMON
177 if (current->thread.pfm_needs_checking)
178 /*
179 * Note: pfm_handle_work() allow us to call it with interrupts
180 * disabled, and may enable interrupts within the function.
181 */
182 pfm_handle_work();
183#endif
184
185 /* deal with pending signal delivery */
186 if (test_thread_flag(TIF_SIGPENDING)) {
187 local_irq_enable(); /* force interrupt enable */
188 ia64_do_signal(scr, in_syscall);
189 }
190
191 if (test_and_clear_thread_flag(TIF_NOTIFY_RESUME)) {
192 local_irq_enable(); /* force interrupt enable */
193 tracehook_notify_resume(&scr->pt);
194 }
195
196 /* copy user rbs to kernel rbs */
197 if (unlikely(test_thread_flag(TIF_RESTORE_RSE))) {
198 local_irq_enable(); /* force interrupt enable */
199 ia64_sync_krbs();
200 }
201
202 local_irq_disable(); /* force interrupt disable */
203}
204
205static int __init nohalt_setup(char * str)
206{
207 cpu_idle_poll_ctrl(true);
208 return 1;
209}
210__setup("nohalt", nohalt_setup);
211
212#ifdef CONFIG_HOTPLUG_CPU
213/* We don't actually take CPU down, just spin without interrupts. */
214static inline void play_dead(void)
215{
216 unsigned int this_cpu = smp_processor_id();
217
218 /* Ack it */
219 __this_cpu_write(cpu_state, CPU_DEAD);
220
221 max_xtp();
222 local_irq_disable();
223 idle_task_exit();
224 ia64_jump_to_sal(&sal_boot_rendez_state[this_cpu]);
225 /*
226 * The above is a point of no-return, the processor is
227 * expected to be in SAL loop now.
228 */
229 BUG();
230}
231#else
232static inline void play_dead(void)
233{
234 BUG();
235}
236#endif /* CONFIG_HOTPLUG_CPU */
237
238void arch_cpu_idle_dead(void)
239{
240 play_dead();
241}
242
243void arch_cpu_idle(void)
244{
245 void (*mark_idle)(int) = ia64_mark_idle;
246
247#ifdef CONFIG_SMP
248 min_xtp();
249#endif
250 rmb();
251 if (mark_idle)
252 (*mark_idle)(1);
253
254 safe_halt();
255
256 if (mark_idle)
257 (*mark_idle)(0);
258#ifdef CONFIG_SMP
259 normal_xtp();
260#endif
261}
262
263void
264ia64_save_extra (struct task_struct *task)
265{
266#ifdef CONFIG_PERFMON
267 unsigned long info;
268#endif
269
270 if ((task->thread.flags & IA64_THREAD_DBG_VALID) != 0)
271 ia64_save_debug_regs(&task->thread.dbr[0]);
272
273#ifdef CONFIG_PERFMON
274 if ((task->thread.flags & IA64_THREAD_PM_VALID) != 0)
275 pfm_save_regs(task);
276
277 info = __this_cpu_read(pfm_syst_info);
278 if (info & PFM_CPUINFO_SYST_WIDE)
279 pfm_syst_wide_update_task(task, info, 0);
280#endif
281}
282
283void
284ia64_load_extra (struct task_struct *task)
285{
286#ifdef CONFIG_PERFMON
287 unsigned long info;
288#endif
289
290 if ((task->thread.flags & IA64_THREAD_DBG_VALID) != 0)
291 ia64_load_debug_regs(&task->thread.dbr[0]);
292
293#ifdef CONFIG_PERFMON
294 if ((task->thread.flags & IA64_THREAD_PM_VALID) != 0)
295 pfm_load_regs(task);
296
297 info = __this_cpu_read(pfm_syst_info);
298 if (info & PFM_CPUINFO_SYST_WIDE)
299 pfm_syst_wide_update_task(task, info, 1);
300#endif
301}
302
303/*
304 * Copy the state of an ia-64 thread.
305 *
306 * We get here through the following call chain:
307 *
308 * from user-level: from kernel:
309 *
310 * <clone syscall> <some kernel call frames>
311 * sys_clone :
312 * do_fork do_fork
313 * copy_thread copy_thread
314 *
315 * This means that the stack layout is as follows:
316 *
317 * +---------------------+ (highest addr)
318 * | struct pt_regs |
319 * +---------------------+
320 * | struct switch_stack |
321 * +---------------------+
322 * | |
323 * | memory stack |
324 * | | <-- sp (lowest addr)
325 * +---------------------+
326 *
327 * Observe that we copy the unat values that are in pt_regs and switch_stack. Spilling an
328 * integer to address X causes bit N in ar.unat to be set to the NaT bit of the register,
329 * with N=(X & 0x1ff)/8. Thus, copying the unat value preserves the NaT bits ONLY if the
330 * pt_regs structure in the parent is congruent to that of the child, modulo 512. Since
331 * the stack is page aligned and the page size is at least 4KB, this is always the case,
332 * so there is nothing to worry about.
333 */
334int
335copy_thread(unsigned long clone_flags,
336 unsigned long user_stack_base, unsigned long user_stack_size,
337 struct task_struct *p)
338{
339 extern char ia64_ret_from_clone;
340 struct switch_stack *child_stack, *stack;
341 unsigned long rbs, child_rbs, rbs_size;
342 struct pt_regs *child_ptregs;
343 struct pt_regs *regs = current_pt_regs();
344 int retval = 0;
345
346 child_ptregs = (struct pt_regs *) ((unsigned long) p + IA64_STK_OFFSET) - 1;
347 child_stack = (struct switch_stack *) child_ptregs - 1;
348
349 rbs = (unsigned long) current + IA64_RBS_OFFSET;
350 child_rbs = (unsigned long) p + IA64_RBS_OFFSET;
351
352 /* copy parts of thread_struct: */
353 p->thread.ksp = (unsigned long) child_stack - 16;
354
355 /*
356 * NOTE: The calling convention considers all floating point
357 * registers in the high partition (fph) to be scratch. Since
358 * the only way to get to this point is through a system call,
359 * we know that the values in fph are all dead. Hence, there
360 * is no need to inherit the fph state from the parent to the
361 * child and all we have to do is to make sure that
362 * IA64_THREAD_FPH_VALID is cleared in the child.
363 *
364 * XXX We could push this optimization a bit further by
365 * clearing IA64_THREAD_FPH_VALID on ANY system call.
366 * However, it's not clear this is worth doing. Also, it
367 * would be a slight deviation from the normal Linux system
368 * call behavior where scratch registers are preserved across
369 * system calls (unless used by the system call itself).
370 */
371# define THREAD_FLAGS_TO_CLEAR (IA64_THREAD_FPH_VALID | IA64_THREAD_DBG_VALID \
372 | IA64_THREAD_PM_VALID)
373# define THREAD_FLAGS_TO_SET 0
374 p->thread.flags = ((current->thread.flags & ~THREAD_FLAGS_TO_CLEAR)
375 | THREAD_FLAGS_TO_SET);
376
377 ia64_drop_fpu(p); /* don't pick up stale state from a CPU's fph */
378
379 if (unlikely(p->flags & PF_KTHREAD)) {
380 if (unlikely(!user_stack_base)) {
381 /* fork_idle() called us */
382 return 0;
383 }
384 memset(child_stack, 0, sizeof(*child_ptregs) + sizeof(*child_stack));
385 child_stack->r4 = user_stack_base; /* payload */
386 child_stack->r5 = user_stack_size; /* argument */
387 /*
388 * Preserve PSR bits, except for bits 32-34 and 37-45,
389 * which we can't read.
390 */
391 child_ptregs->cr_ipsr = ia64_getreg(_IA64_REG_PSR) | IA64_PSR_BN;
392 /* mark as valid, empty frame */
393 child_ptregs->cr_ifs = 1UL << 63;
394 child_stack->ar_fpsr = child_ptregs->ar_fpsr
395 = ia64_getreg(_IA64_REG_AR_FPSR);
396 child_stack->pr = (1 << PRED_KERNEL_STACK);
397 child_stack->ar_bspstore = child_rbs;
398 child_stack->b0 = (unsigned long) &ia64_ret_from_clone;
399
400 /* stop some PSR bits from being inherited.
401 * the psr.up/psr.pp bits must be cleared on fork but inherited on execve()
402 * therefore we must specify them explicitly here and not include them in
403 * IA64_PSR_BITS_TO_CLEAR.
404 */
405 child_ptregs->cr_ipsr = ((child_ptregs->cr_ipsr | IA64_PSR_BITS_TO_SET)
406 & ~(IA64_PSR_BITS_TO_CLEAR | IA64_PSR_PP | IA64_PSR_UP));
407
408 return 0;
409 }
410 stack = ((struct switch_stack *) regs) - 1;
411 /* copy parent's switch_stack & pt_regs to child: */
412 memcpy(child_stack, stack, sizeof(*child_ptregs) + sizeof(*child_stack));
413
414 /* copy the parent's register backing store to the child: */
415 rbs_size = stack->ar_bspstore - rbs;
416 memcpy((void *) child_rbs, (void *) rbs, rbs_size);
417 if (clone_flags & CLONE_SETTLS)
418 child_ptregs->r13 = regs->r16; /* see sys_clone2() in entry.S */
419 if (user_stack_base) {
420 child_ptregs->r12 = user_stack_base + user_stack_size - 16;
421 child_ptregs->ar_bspstore = user_stack_base;
422 child_ptregs->ar_rnat = 0;
423 child_ptregs->loadrs = 0;
424 }
425 child_stack->ar_bspstore = child_rbs + rbs_size;
426 child_stack->b0 = (unsigned long) &ia64_ret_from_clone;
427
428 /* stop some PSR bits from being inherited.
429 * the psr.up/psr.pp bits must be cleared on fork but inherited on execve()
430 * therefore we must specify them explicitly here and not include them in
431 * IA64_PSR_BITS_TO_CLEAR.
432 */
433 child_ptregs->cr_ipsr = ((child_ptregs->cr_ipsr | IA64_PSR_BITS_TO_SET)
434 & ~(IA64_PSR_BITS_TO_CLEAR | IA64_PSR_PP | IA64_PSR_UP));
435
436#ifdef CONFIG_PERFMON
437 if (current->thread.pfm_context)
438 pfm_inherit(p, child_ptregs);
439#endif
440 return retval;
441}
442
443static void
444do_copy_task_regs (struct task_struct *task, struct unw_frame_info *info, void *arg)
445{
446 unsigned long mask, sp, nat_bits = 0, ar_rnat, urbs_end, cfm;
447 unsigned long uninitialized_var(ip); /* GCC be quiet */
448 elf_greg_t *dst = arg;
449 struct pt_regs *pt;
450 char nat;
451 int i;
452
453 memset(dst, 0, sizeof(elf_gregset_t)); /* don't leak any kernel bits to user-level */
454
455 if (unw_unwind_to_user(info) < 0)
456 return;
457
458 unw_get_sp(info, &sp);
459 pt = (struct pt_regs *) (sp + 16);
460
461 urbs_end = ia64_get_user_rbs_end(task, pt, &cfm);
462
463 if (ia64_sync_user_rbs(task, info->sw, pt->ar_bspstore, urbs_end) < 0)
464 return;
465
466 ia64_peek(task, info->sw, urbs_end, (long) ia64_rse_rnat_addr((long *) urbs_end),
467 &ar_rnat);
468
469 /*
470 * coredump format:
471 * r0-r31
472 * NaT bits (for r0-r31; bit N == 1 iff rN is a NaT)
473 * predicate registers (p0-p63)
474 * b0-b7
475 * ip cfm user-mask
476 * ar.rsc ar.bsp ar.bspstore ar.rnat
477 * ar.ccv ar.unat ar.fpsr ar.pfs ar.lc ar.ec
478 */
479
480 /* r0 is zero */
481 for (i = 1, mask = (1UL << i); i < 32; ++i) {
482 unw_get_gr(info, i, &dst[i], &nat);
483 if (nat)
484 nat_bits |= mask;
485 mask <<= 1;
486 }
487 dst[32] = nat_bits;
488 unw_get_pr(info, &dst[33]);
489
490 for (i = 0; i < 8; ++i)
491 unw_get_br(info, i, &dst[34 + i]);
492
493 unw_get_rp(info, &ip);
494 dst[42] = ip + ia64_psr(pt)->ri;
495 dst[43] = cfm;
496 dst[44] = pt->cr_ipsr & IA64_PSR_UM;
497
498 unw_get_ar(info, UNW_AR_RSC, &dst[45]);
499 /*
500 * For bsp and bspstore, unw_get_ar() would return the kernel
501 * addresses, but we need the user-level addresses instead:
502 */
503 dst[46] = urbs_end; /* note: by convention PT_AR_BSP points to the end of the urbs! */
504 dst[47] = pt->ar_bspstore;
505 dst[48] = ar_rnat;
506 unw_get_ar(info, UNW_AR_CCV, &dst[49]);
507 unw_get_ar(info, UNW_AR_UNAT, &dst[50]);
508 unw_get_ar(info, UNW_AR_FPSR, &dst[51]);
509 dst[52] = pt->ar_pfs; /* UNW_AR_PFS is == to pt->cr_ifs for interrupt frames */
510 unw_get_ar(info, UNW_AR_LC, &dst[53]);
511 unw_get_ar(info, UNW_AR_EC, &dst[54]);
512 unw_get_ar(info, UNW_AR_CSD, &dst[55]);
513 unw_get_ar(info, UNW_AR_SSD, &dst[56]);
514}
515
516void
517do_dump_task_fpu (struct task_struct *task, struct unw_frame_info *info, void *arg)
518{
519 elf_fpreg_t *dst = arg;
520 int i;
521
522 memset(dst, 0, sizeof(elf_fpregset_t)); /* don't leak any "random" bits */
523
524 if (unw_unwind_to_user(info) < 0)
525 return;
526
527 /* f0 is 0.0, f1 is 1.0 */
528
529 for (i = 2; i < 32; ++i)
530 unw_get_fr(info, i, dst + i);
531
532 ia64_flush_fph(task);
533 if ((task->thread.flags & IA64_THREAD_FPH_VALID) != 0)
534 memcpy(dst + 32, task->thread.fph, 96*16);
535}
536
537void
538do_copy_regs (struct unw_frame_info *info, void *arg)
539{
540 do_copy_task_regs(current, info, arg);
541}
542
543void
544do_dump_fpu (struct unw_frame_info *info, void *arg)
545{
546 do_dump_task_fpu(current, info, arg);
547}
548
549void
550ia64_elf_core_copy_regs (struct pt_regs *pt, elf_gregset_t dst)
551{
552 unw_init_running(do_copy_regs, dst);
553}
554
555int
556dump_fpu (struct pt_regs *pt, elf_fpregset_t dst)
557{
558 unw_init_running(do_dump_fpu, dst);
559 return 1; /* f0-f31 are always valid so we always return 1 */
560}
561
562/*
563 * Flush thread state. This is called when a thread does an execve().
564 */
565void
566flush_thread (void)
567{
568 /* drop floating-point and debug-register state if it exists: */
569 current->thread.flags &= ~(IA64_THREAD_FPH_VALID | IA64_THREAD_DBG_VALID);
570 ia64_drop_fpu(current);
571}
572
573/*
574 * Clean up state associated with a thread. This is called when
575 * the thread calls exit().
576 */
577void
578exit_thread (struct task_struct *tsk)
579{
580
581 ia64_drop_fpu(tsk);
582#ifdef CONFIG_PERFMON
583 /* if needed, stop monitoring and flush state to perfmon context */
584 if (tsk->thread.pfm_context)
585 pfm_exit_thread(tsk);
586
587 /* free debug register resources */
588 if (tsk->thread.flags & IA64_THREAD_DBG_VALID)
589 pfm_release_debug_registers(tsk);
590#endif
591}
592
593unsigned long
594get_wchan (struct task_struct *p)
595{
596 struct unw_frame_info info;
597 unsigned long ip;
598 int count = 0;
599
600 if (!p || p == current || p->state == TASK_RUNNING)
601 return 0;
602
603 /*
604 * Note: p may not be a blocked task (it could be current or
605 * another process running on some other CPU. Rather than
606 * trying to determine if p is really blocked, we just assume
607 * it's blocked and rely on the unwind routines to fail
608 * gracefully if the process wasn't really blocked after all.
609 * --davidm 99/12/15
610 */
611 unw_init_from_blocked_task(&info, p);
612 do {
613 if (p->state == TASK_RUNNING)
614 return 0;
615 if (unw_unwind(&info) < 0)
616 return 0;
617 unw_get_ip(&info, &ip);
618 if (!in_sched_functions(ip))
619 return ip;
620 } while (count++ < 16);
621 return 0;
622}
623
624void
625cpu_halt (void)
626{
627 pal_power_mgmt_info_u_t power_info[8];
628 unsigned long min_power;
629 int i, min_power_state;
630
631 if (ia64_pal_halt_info(power_info) != 0)
632 return;
633
634 min_power_state = 0;
635 min_power = power_info[0].pal_power_mgmt_info_s.power_consumption;
636 for (i = 1; i < 8; ++i)
637 if (power_info[i].pal_power_mgmt_info_s.im
638 && power_info[i].pal_power_mgmt_info_s.power_consumption < min_power) {
639 min_power = power_info[i].pal_power_mgmt_info_s.power_consumption;
640 min_power_state = i;
641 }
642
643 while (1)
644 ia64_pal_halt(min_power_state);
645}
646
647void machine_shutdown(void)
648{
649#ifdef CONFIG_HOTPLUG_CPU
650 int cpu;
651
652 for_each_online_cpu(cpu) {
653 if (cpu != smp_processor_id())
654 cpu_down(cpu);
655 }
656#endif
657#ifdef CONFIG_KEXEC
658 kexec_disable_iosapic();
659#endif
660}
661
662void
663machine_restart (char *restart_cmd)
664{
665 (void) notify_die(DIE_MACHINE_RESTART, restart_cmd, NULL, 0, 0, 0);
666 efi_reboot(REBOOT_WARM, NULL);
667}
668
669void
670machine_halt (void)
671{
672 (void) notify_die(DIE_MACHINE_HALT, "", NULL, 0, 0, 0);
673 cpu_halt();
674}
675
676void
677machine_power_off (void)
678{
679 if (pm_power_off)
680 pm_power_off();
681 machine_halt();
682}
683
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Architecture-specific setup.
4 *
5 * Copyright (C) 1998-2003 Hewlett-Packard Co
6 * David Mosberger-Tang <davidm@hpl.hp.com>
7 * 04/11/17 Ashok Raj <ashok.raj@intel.com> Added CPU Hotplug Support
8 *
9 * 2005-10-07 Keith Owens <kaos@sgi.com>
10 * Add notify_die() hooks.
11 */
12#include <linux/cpu.h>
13#include <linux/pm.h>
14#include <linux/elf.h>
15#include <linux/errno.h>
16#include <linux/kernel.h>
17#include <linux/mm.h>
18#include <linux/slab.h>
19#include <linux/module.h>
20#include <linux/notifier.h>
21#include <linux/personality.h>
22#include <linux/sched.h>
23#include <linux/sched/debug.h>
24#include <linux/sched/hotplug.h>
25#include <linux/sched/task.h>
26#include <linux/sched/task_stack.h>
27#include <linux/stddef.h>
28#include <linux/thread_info.h>
29#include <linux/unistd.h>
30#include <linux/efi.h>
31#include <linux/interrupt.h>
32#include <linux/delay.h>
33#include <linux/kdebug.h>
34#include <linux/utsname.h>
35#include <linux/tracehook.h>
36#include <linux/rcupdate.h>
37
38#include <asm/cpu.h>
39#include <asm/delay.h>
40#include <asm/elf.h>
41#include <asm/irq.h>
42#include <asm/kexec.h>
43#include <asm/processor.h>
44#include <asm/sal.h>
45#include <asm/switch_to.h>
46#include <asm/tlbflush.h>
47#include <linux/uaccess.h>
48#include <asm/unwind.h>
49#include <asm/user.h>
50#include <asm/xtp.h>
51
52#include "entry.h"
53
54#include "sigframe.h"
55
56void (*ia64_mark_idle)(int);
57
58unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
59EXPORT_SYMBOL(boot_option_idle_override);
60void (*pm_power_off) (void);
61EXPORT_SYMBOL(pm_power_off);
62
63static void
64ia64_do_show_stack (struct unw_frame_info *info, void *arg)
65{
66 unsigned long ip, sp, bsp;
67 const char *loglvl = arg;
68
69 printk("%s\nCall Trace:\n", loglvl);
70 do {
71 unw_get_ip(info, &ip);
72 if (ip == 0)
73 break;
74
75 unw_get_sp(info, &sp);
76 unw_get_bsp(info, &bsp);
77 printk("%s [<%016lx>] %pS\n"
78 " sp=%016lx bsp=%016lx\n",
79 loglvl, ip, (void *)ip, sp, bsp);
80 } while (unw_unwind(info) >= 0);
81}
82
83void
84show_stack (struct task_struct *task, unsigned long *sp, const char *loglvl)
85{
86 if (!task)
87 unw_init_running(ia64_do_show_stack, (void *)loglvl);
88 else {
89 struct unw_frame_info info;
90
91 unw_init_from_blocked_task(&info, task);
92 ia64_do_show_stack(&info, (void *)loglvl);
93 }
94}
95
96void
97show_regs (struct pt_regs *regs)
98{
99 unsigned long ip = regs->cr_iip + ia64_psr(regs)->ri;
100
101 print_modules();
102 printk("\n");
103 show_regs_print_info(KERN_DEFAULT);
104 printk("psr : %016lx ifs : %016lx ip : [<%016lx>] %s (%s)\n",
105 regs->cr_ipsr, regs->cr_ifs, ip, print_tainted(),
106 init_utsname()->release);
107 printk("ip is at %pS\n", (void *)ip);
108 printk("unat: %016lx pfs : %016lx rsc : %016lx\n",
109 regs->ar_unat, regs->ar_pfs, regs->ar_rsc);
110 printk("rnat: %016lx bsps: %016lx pr : %016lx\n",
111 regs->ar_rnat, regs->ar_bspstore, regs->pr);
112 printk("ldrs: %016lx ccv : %016lx fpsr: %016lx\n",
113 regs->loadrs, regs->ar_ccv, regs->ar_fpsr);
114 printk("csd : %016lx ssd : %016lx\n", regs->ar_csd, regs->ar_ssd);
115 printk("b0 : %016lx b6 : %016lx b7 : %016lx\n", regs->b0, regs->b6, regs->b7);
116 printk("f6 : %05lx%016lx f7 : %05lx%016lx\n",
117 regs->f6.u.bits[1], regs->f6.u.bits[0],
118 regs->f7.u.bits[1], regs->f7.u.bits[0]);
119 printk("f8 : %05lx%016lx f9 : %05lx%016lx\n",
120 regs->f8.u.bits[1], regs->f8.u.bits[0],
121 regs->f9.u.bits[1], regs->f9.u.bits[0]);
122 printk("f10 : %05lx%016lx f11 : %05lx%016lx\n",
123 regs->f10.u.bits[1], regs->f10.u.bits[0],
124 regs->f11.u.bits[1], regs->f11.u.bits[0]);
125
126 printk("r1 : %016lx r2 : %016lx r3 : %016lx\n", regs->r1, regs->r2, regs->r3);
127 printk("r8 : %016lx r9 : %016lx r10 : %016lx\n", regs->r8, regs->r9, regs->r10);
128 printk("r11 : %016lx r12 : %016lx r13 : %016lx\n", regs->r11, regs->r12, regs->r13);
129 printk("r14 : %016lx r15 : %016lx r16 : %016lx\n", regs->r14, regs->r15, regs->r16);
130 printk("r17 : %016lx r18 : %016lx r19 : %016lx\n", regs->r17, regs->r18, regs->r19);
131 printk("r20 : %016lx r21 : %016lx r22 : %016lx\n", regs->r20, regs->r21, regs->r22);
132 printk("r23 : %016lx r24 : %016lx r25 : %016lx\n", regs->r23, regs->r24, regs->r25);
133 printk("r26 : %016lx r27 : %016lx r28 : %016lx\n", regs->r26, regs->r27, regs->r28);
134 printk("r29 : %016lx r30 : %016lx r31 : %016lx\n", regs->r29, regs->r30, regs->r31);
135
136 if (user_mode(regs)) {
137 /* print the stacked registers */
138 unsigned long val, *bsp, ndirty;
139 int i, sof, is_nat = 0;
140
141 sof = regs->cr_ifs & 0x7f; /* size of frame */
142 ndirty = (regs->loadrs >> 19);
143 bsp = ia64_rse_skip_regs((unsigned long *) regs->ar_bspstore, ndirty);
144 for (i = 0; i < sof; ++i) {
145 get_user(val, (unsigned long __user *) ia64_rse_skip_regs(bsp, i));
146 printk("r%-3u:%c%016lx%s", 32 + i, is_nat ? '*' : ' ', val,
147 ((i == sof - 1) || (i % 3) == 2) ? "\n" : " ");
148 }
149 } else
150 show_stack(NULL, NULL, KERN_DEFAULT);
151}
152
153/* local support for deprecated console_print */
154void
155console_print(const char *s)
156{
157 printk(KERN_EMERG "%s", s);
158}
159
160void
161do_notify_resume_user(sigset_t *unused, struct sigscratch *scr, long in_syscall)
162{
163 if (fsys_mode(current, &scr->pt)) {
164 /*
165 * defer signal-handling etc. until we return to
166 * privilege-level 0.
167 */
168 if (!ia64_psr(&scr->pt)->lp)
169 ia64_psr(&scr->pt)->lp = 1;
170 return;
171 }
172
173 /* deal with pending signal delivery */
174 if (test_thread_flag(TIF_SIGPENDING) ||
175 test_thread_flag(TIF_NOTIFY_SIGNAL)) {
176 local_irq_enable(); /* force interrupt enable */
177 ia64_do_signal(scr, in_syscall);
178 }
179
180 if (test_thread_flag(TIF_NOTIFY_RESUME)) {
181 local_irq_enable(); /* force interrupt enable */
182 tracehook_notify_resume(&scr->pt);
183 }
184
185 /* copy user rbs to kernel rbs */
186 if (unlikely(test_thread_flag(TIF_RESTORE_RSE))) {
187 local_irq_enable(); /* force interrupt enable */
188 ia64_sync_krbs();
189 }
190
191 local_irq_disable(); /* force interrupt disable */
192}
193
194static int __init nohalt_setup(char * str)
195{
196 cpu_idle_poll_ctrl(true);
197 return 1;
198}
199__setup("nohalt", nohalt_setup);
200
201#ifdef CONFIG_HOTPLUG_CPU
202/* We don't actually take CPU down, just spin without interrupts. */
203static inline void play_dead(void)
204{
205 unsigned int this_cpu = smp_processor_id();
206
207 /* Ack it */
208 __this_cpu_write(cpu_state, CPU_DEAD);
209
210 max_xtp();
211 local_irq_disable();
212 idle_task_exit();
213 ia64_jump_to_sal(&sal_boot_rendez_state[this_cpu]);
214 /*
215 * The above is a point of no-return, the processor is
216 * expected to be in SAL loop now.
217 */
218 BUG();
219}
220#else
221static inline void play_dead(void)
222{
223 BUG();
224}
225#endif /* CONFIG_HOTPLUG_CPU */
226
227void arch_cpu_idle_dead(void)
228{
229 play_dead();
230}
231
232void arch_cpu_idle(void)
233{
234 void (*mark_idle)(int) = ia64_mark_idle;
235
236#ifdef CONFIG_SMP
237 min_xtp();
238#endif
239 rmb();
240 if (mark_idle)
241 (*mark_idle)(1);
242
243 raw_safe_halt();
244
245 if (mark_idle)
246 (*mark_idle)(0);
247#ifdef CONFIG_SMP
248 normal_xtp();
249#endif
250}
251
252void
253ia64_save_extra (struct task_struct *task)
254{
255 if ((task->thread.flags & IA64_THREAD_DBG_VALID) != 0)
256 ia64_save_debug_regs(&task->thread.dbr[0]);
257}
258
259void
260ia64_load_extra (struct task_struct *task)
261{
262 if ((task->thread.flags & IA64_THREAD_DBG_VALID) != 0)
263 ia64_load_debug_regs(&task->thread.dbr[0]);
264}
265
266/*
267 * Copy the state of an ia-64 thread.
268 *
269 * We get here through the following call chain:
270 *
271 * from user-level: from kernel:
272 *
273 * <clone syscall> <some kernel call frames>
274 * sys_clone :
275 * kernel_clone kernel_clone
276 * copy_thread copy_thread
277 *
278 * This means that the stack layout is as follows:
279 *
280 * +---------------------+ (highest addr)
281 * | struct pt_regs |
282 * +---------------------+
283 * | struct switch_stack |
284 * +---------------------+
285 * | |
286 * | memory stack |
287 * | | <-- sp (lowest addr)
288 * +---------------------+
289 *
290 * Observe that we copy the unat values that are in pt_regs and switch_stack. Spilling an
291 * integer to address X causes bit N in ar.unat to be set to the NaT bit of the register,
292 * with N=(X & 0x1ff)/8. Thus, copying the unat value preserves the NaT bits ONLY if the
293 * pt_regs structure in the parent is congruent to that of the child, modulo 512. Since
294 * the stack is page aligned and the page size is at least 4KB, this is always the case,
295 * so there is nothing to worry about.
296 */
297int
298copy_thread(unsigned long clone_flags, unsigned long user_stack_base,
299 unsigned long user_stack_size, struct task_struct *p, unsigned long tls)
300{
301 extern char ia64_ret_from_clone;
302 struct switch_stack *child_stack, *stack;
303 unsigned long rbs, child_rbs, rbs_size;
304 struct pt_regs *child_ptregs;
305 struct pt_regs *regs = current_pt_regs();
306 int retval = 0;
307
308 child_ptregs = (struct pt_regs *) ((unsigned long) p + IA64_STK_OFFSET) - 1;
309 child_stack = (struct switch_stack *) child_ptregs - 1;
310
311 rbs = (unsigned long) current + IA64_RBS_OFFSET;
312 child_rbs = (unsigned long) p + IA64_RBS_OFFSET;
313
314 /* copy parts of thread_struct: */
315 p->thread.ksp = (unsigned long) child_stack - 16;
316
317 /*
318 * NOTE: The calling convention considers all floating point
319 * registers in the high partition (fph) to be scratch. Since
320 * the only way to get to this point is through a system call,
321 * we know that the values in fph are all dead. Hence, there
322 * is no need to inherit the fph state from the parent to the
323 * child and all we have to do is to make sure that
324 * IA64_THREAD_FPH_VALID is cleared in the child.
325 *
326 * XXX We could push this optimization a bit further by
327 * clearing IA64_THREAD_FPH_VALID on ANY system call.
328 * However, it's not clear this is worth doing. Also, it
329 * would be a slight deviation from the normal Linux system
330 * call behavior where scratch registers are preserved across
331 * system calls (unless used by the system call itself).
332 */
333# define THREAD_FLAGS_TO_CLEAR (IA64_THREAD_FPH_VALID | IA64_THREAD_DBG_VALID \
334 | IA64_THREAD_PM_VALID)
335# define THREAD_FLAGS_TO_SET 0
336 p->thread.flags = ((current->thread.flags & ~THREAD_FLAGS_TO_CLEAR)
337 | THREAD_FLAGS_TO_SET);
338
339 ia64_drop_fpu(p); /* don't pick up stale state from a CPU's fph */
340
341 if (unlikely(p->flags & (PF_KTHREAD | PF_IO_WORKER))) {
342 if (unlikely(!user_stack_base)) {
343 /* fork_idle() called us */
344 return 0;
345 }
346 memset(child_stack, 0, sizeof(*child_ptregs) + sizeof(*child_stack));
347 child_stack->r4 = user_stack_base; /* payload */
348 child_stack->r5 = user_stack_size; /* argument */
349 /*
350 * Preserve PSR bits, except for bits 32-34 and 37-45,
351 * which we can't read.
352 */
353 child_ptregs->cr_ipsr = ia64_getreg(_IA64_REG_PSR) | IA64_PSR_BN;
354 /* mark as valid, empty frame */
355 child_ptregs->cr_ifs = 1UL << 63;
356 child_stack->ar_fpsr = child_ptregs->ar_fpsr
357 = ia64_getreg(_IA64_REG_AR_FPSR);
358 child_stack->pr = (1 << PRED_KERNEL_STACK);
359 child_stack->ar_bspstore = child_rbs;
360 child_stack->b0 = (unsigned long) &ia64_ret_from_clone;
361
362 /* stop some PSR bits from being inherited.
363 * the psr.up/psr.pp bits must be cleared on fork but inherited on execve()
364 * therefore we must specify them explicitly here and not include them in
365 * IA64_PSR_BITS_TO_CLEAR.
366 */
367 child_ptregs->cr_ipsr = ((child_ptregs->cr_ipsr | IA64_PSR_BITS_TO_SET)
368 & ~(IA64_PSR_BITS_TO_CLEAR | IA64_PSR_PP | IA64_PSR_UP));
369
370 return 0;
371 }
372 stack = ((struct switch_stack *) regs) - 1;
373 /* copy parent's switch_stack & pt_regs to child: */
374 memcpy(child_stack, stack, sizeof(*child_ptregs) + sizeof(*child_stack));
375
376 /* copy the parent's register backing store to the child: */
377 rbs_size = stack->ar_bspstore - rbs;
378 memcpy((void *) child_rbs, (void *) rbs, rbs_size);
379 if (clone_flags & CLONE_SETTLS)
380 child_ptregs->r13 = tls;
381 if (user_stack_base) {
382 child_ptregs->r12 = user_stack_base + user_stack_size - 16;
383 child_ptregs->ar_bspstore = user_stack_base;
384 child_ptregs->ar_rnat = 0;
385 child_ptregs->loadrs = 0;
386 }
387 child_stack->ar_bspstore = child_rbs + rbs_size;
388 child_stack->b0 = (unsigned long) &ia64_ret_from_clone;
389
390 /* stop some PSR bits from being inherited.
391 * the psr.up/psr.pp bits must be cleared on fork but inherited on execve()
392 * therefore we must specify them explicitly here and not include them in
393 * IA64_PSR_BITS_TO_CLEAR.
394 */
395 child_ptregs->cr_ipsr = ((child_ptregs->cr_ipsr | IA64_PSR_BITS_TO_SET)
396 & ~(IA64_PSR_BITS_TO_CLEAR | IA64_PSR_PP | IA64_PSR_UP));
397 return retval;
398}
399
400asmlinkage long ia64_clone(unsigned long clone_flags, unsigned long stack_start,
401 unsigned long stack_size, unsigned long parent_tidptr,
402 unsigned long child_tidptr, unsigned long tls)
403{
404 struct kernel_clone_args args = {
405 .flags = (lower_32_bits(clone_flags) & ~CSIGNAL),
406 .pidfd = (int __user *)parent_tidptr,
407 .child_tid = (int __user *)child_tidptr,
408 .parent_tid = (int __user *)parent_tidptr,
409 .exit_signal = (lower_32_bits(clone_flags) & CSIGNAL),
410 .stack = stack_start,
411 .stack_size = stack_size,
412 .tls = tls,
413 };
414
415 return kernel_clone(&args);
416}
417
418static void
419do_copy_task_regs (struct task_struct *task, struct unw_frame_info *info, void *arg)
420{
421 unsigned long mask, sp, nat_bits = 0, ar_rnat, urbs_end, cfm;
422 unsigned long ip;
423 elf_greg_t *dst = arg;
424 struct pt_regs *pt;
425 char nat;
426 int i;
427
428 memset(dst, 0, sizeof(elf_gregset_t)); /* don't leak any kernel bits to user-level */
429
430 if (unw_unwind_to_user(info) < 0)
431 return;
432
433 unw_get_sp(info, &sp);
434 pt = (struct pt_regs *) (sp + 16);
435
436 urbs_end = ia64_get_user_rbs_end(task, pt, &cfm);
437
438 if (ia64_sync_user_rbs(task, info->sw, pt->ar_bspstore, urbs_end) < 0)
439 return;
440
441 ia64_peek(task, info->sw, urbs_end, (long) ia64_rse_rnat_addr((long *) urbs_end),
442 &ar_rnat);
443
444 /*
445 * coredump format:
446 * r0-r31
447 * NaT bits (for r0-r31; bit N == 1 iff rN is a NaT)
448 * predicate registers (p0-p63)
449 * b0-b7
450 * ip cfm user-mask
451 * ar.rsc ar.bsp ar.bspstore ar.rnat
452 * ar.ccv ar.unat ar.fpsr ar.pfs ar.lc ar.ec
453 */
454
455 /* r0 is zero */
456 for (i = 1, mask = (1UL << i); i < 32; ++i) {
457 unw_get_gr(info, i, &dst[i], &nat);
458 if (nat)
459 nat_bits |= mask;
460 mask <<= 1;
461 }
462 dst[32] = nat_bits;
463 unw_get_pr(info, &dst[33]);
464
465 for (i = 0; i < 8; ++i)
466 unw_get_br(info, i, &dst[34 + i]);
467
468 unw_get_rp(info, &ip);
469 dst[42] = ip + ia64_psr(pt)->ri;
470 dst[43] = cfm;
471 dst[44] = pt->cr_ipsr & IA64_PSR_UM;
472
473 unw_get_ar(info, UNW_AR_RSC, &dst[45]);
474 /*
475 * For bsp and bspstore, unw_get_ar() would return the kernel
476 * addresses, but we need the user-level addresses instead:
477 */
478 dst[46] = urbs_end; /* note: by convention PT_AR_BSP points to the end of the urbs! */
479 dst[47] = pt->ar_bspstore;
480 dst[48] = ar_rnat;
481 unw_get_ar(info, UNW_AR_CCV, &dst[49]);
482 unw_get_ar(info, UNW_AR_UNAT, &dst[50]);
483 unw_get_ar(info, UNW_AR_FPSR, &dst[51]);
484 dst[52] = pt->ar_pfs; /* UNW_AR_PFS is == to pt->cr_ifs for interrupt frames */
485 unw_get_ar(info, UNW_AR_LC, &dst[53]);
486 unw_get_ar(info, UNW_AR_EC, &dst[54]);
487 unw_get_ar(info, UNW_AR_CSD, &dst[55]);
488 unw_get_ar(info, UNW_AR_SSD, &dst[56]);
489}
490
491static void
492do_copy_regs (struct unw_frame_info *info, void *arg)
493{
494 do_copy_task_regs(current, info, arg);
495}
496
497void
498ia64_elf_core_copy_regs (struct pt_regs *pt, elf_gregset_t dst)
499{
500 unw_init_running(do_copy_regs, dst);
501}
502
503/*
504 * Flush thread state. This is called when a thread does an execve().
505 */
506void
507flush_thread (void)
508{
509 /* drop floating-point and debug-register state if it exists: */
510 current->thread.flags &= ~(IA64_THREAD_FPH_VALID | IA64_THREAD_DBG_VALID);
511 ia64_drop_fpu(current);
512}
513
514/*
515 * Clean up state associated with a thread. This is called when
516 * the thread calls exit().
517 */
518void
519exit_thread (struct task_struct *tsk)
520{
521
522 ia64_drop_fpu(tsk);
523}
524
525unsigned long
526get_wchan (struct task_struct *p)
527{
528 struct unw_frame_info info;
529 unsigned long ip;
530 int count = 0;
531
532 if (!p || p == current || task_is_running(p))
533 return 0;
534
535 /*
536 * Note: p may not be a blocked task (it could be current or
537 * another process running on some other CPU. Rather than
538 * trying to determine if p is really blocked, we just assume
539 * it's blocked and rely on the unwind routines to fail
540 * gracefully if the process wasn't really blocked after all.
541 * --davidm 99/12/15
542 */
543 unw_init_from_blocked_task(&info, p);
544 do {
545 if (task_is_running(p))
546 return 0;
547 if (unw_unwind(&info) < 0)
548 return 0;
549 unw_get_ip(&info, &ip);
550 if (!in_sched_functions(ip))
551 return ip;
552 } while (count++ < 16);
553 return 0;
554}
555
556void
557cpu_halt (void)
558{
559 pal_power_mgmt_info_u_t power_info[8];
560 unsigned long min_power;
561 int i, min_power_state;
562
563 if (ia64_pal_halt_info(power_info) != 0)
564 return;
565
566 min_power_state = 0;
567 min_power = power_info[0].pal_power_mgmt_info_s.power_consumption;
568 for (i = 1; i < 8; ++i)
569 if (power_info[i].pal_power_mgmt_info_s.im
570 && power_info[i].pal_power_mgmt_info_s.power_consumption < min_power) {
571 min_power = power_info[i].pal_power_mgmt_info_s.power_consumption;
572 min_power_state = i;
573 }
574
575 while (1)
576 ia64_pal_halt(min_power_state);
577}
578
579void machine_shutdown(void)
580{
581 smp_shutdown_nonboot_cpus(reboot_cpu);
582
583#ifdef CONFIG_KEXEC
584 kexec_disable_iosapic();
585#endif
586}
587
588void
589machine_restart (char *restart_cmd)
590{
591 (void) notify_die(DIE_MACHINE_RESTART, restart_cmd, NULL, 0, 0, 0);
592 efi_reboot(REBOOT_WARM, NULL);
593}
594
595void
596machine_halt (void)
597{
598 (void) notify_die(DIE_MACHINE_HALT, "", NULL, 0, 0, 0);
599 cpu_halt();
600}
601
602void
603machine_power_off (void)
604{
605 if (pm_power_off)
606 pm_power_off();
607 machine_halt();
608}
609
610EXPORT_SYMBOL(ia64_delay_loop);