xref: /openbmc/linux/arch/sparc/kernel/process_64.c (revision 12109610)
1 // SPDX-License-Identifier: GPL-2.0
2 /*  arch/sparc64/kernel/process.c
3  *
4  *  Copyright (C) 1995, 1996, 2008 David S. Miller (davem@davemloft.net)
5  *  Copyright (C) 1996       Eddie C. Dost   (ecd@skynet.be)
6  *  Copyright (C) 1997, 1998 Jakub Jelinek   (jj@sunsite.mff.cuni.cz)
7  */
8 
9 /*
10  * This file handles the architecture-dependent parts of process handling..
11  */
12 #include <linux/errno.h>
13 #include <linux/export.h>
14 #include <linux/sched.h>
15 #include <linux/sched/debug.h>
16 #include <linux/sched/task.h>
17 #include <linux/sched/task_stack.h>
18 #include <linux/kernel.h>
19 #include <linux/mm.h>
20 #include <linux/fs.h>
21 #include <linux/smp.h>
22 #include <linux/stddef.h>
23 #include <linux/ptrace.h>
24 #include <linux/slab.h>
25 #include <linux/user.h>
26 #include <linux/delay.h>
27 #include <linux/compat.h>
28 #include <linux/tick.h>
29 #include <linux/init.h>
30 #include <linux/cpu.h>
31 #include <linux/perf_event.h>
32 #include <linux/elfcore.h>
33 #include <linux/sysrq.h>
34 #include <linux/nmi.h>
35 #include <linux/context_tracking.h>
36 #include <linux/signal.h>
37 
38 #include <linux/uaccess.h>
39 #include <asm/page.h>
40 #include <asm/pgalloc.h>
41 #include <asm/processor.h>
42 #include <asm/pstate.h>
43 #include <asm/elf.h>
44 #include <asm/fpumacro.h>
45 #include <asm/head.h>
46 #include <asm/cpudata.h>
47 #include <asm/mmu_context.h>
48 #include <asm/unistd.h>
49 #include <asm/hypervisor.h>
50 #include <asm/syscalls.h>
51 #include <asm/irq_regs.h>
52 #include <asm/smp.h>
53 #include <asm/pcr.h>
54 
55 #include "kstack.h"
56 
57 /* Idle loop support on sparc64. */
58 void arch_cpu_idle(void)
59 {
60 	if (tlb_type != hypervisor) {
61 		touch_nmi_watchdog();
62 	} else {
63 		unsigned long pstate;
64 
65 		raw_local_irq_enable();
66 
67                 /* The sun4v sleeping code requires that we have PSTATE.IE cleared over
68                  * the cpu sleep hypervisor call.
69                  */
70 		__asm__ __volatile__(
71 			"rdpr %%pstate, %0\n\t"
72 			"andn %0, %1, %0\n\t"
73 			"wrpr %0, %%g0, %%pstate"
74 			: "=&r" (pstate)
75 			: "i" (PSTATE_IE));
76 
77 		if (!need_resched() && !cpu_is_offline(smp_processor_id())) {
78 			sun4v_cpu_yield();
79 			/* If resumed by cpu_poke then we need to explicitly
80 			 * call scheduler_ipi().
81 			 */
82 			scheduler_poke();
83 		}
84 
85 		/* Re-enable interrupts. */
86 		__asm__ __volatile__(
87 			"rdpr %%pstate, %0\n\t"
88 			"or %0, %1, %0\n\t"
89 			"wrpr %0, %%g0, %%pstate"
90 			: "=&r" (pstate)
91 			: "i" (PSTATE_IE));
92 
93 		raw_local_irq_disable();
94 	}
95 }
96 
97 #ifdef CONFIG_HOTPLUG_CPU
98 void __noreturn arch_cpu_idle_dead(void)
99 {
100 	sched_preempt_enable_no_resched();
101 	cpu_play_dead();
102 }
103 #endif
104 
105 #ifdef CONFIG_COMPAT
106 static void show_regwindow32(struct pt_regs *regs)
107 {
108 	struct reg_window32 __user *rw;
109 	struct reg_window32 r_w;
110 
111 	__asm__ __volatile__ ("flushw");
112 	rw = compat_ptr((unsigned int)regs->u_regs[14]);
113 	if (copy_from_user (&r_w, rw, sizeof(r_w))) {
114 		return;
115 	}
116 
117 	printk("l0: %08x l1: %08x l2: %08x l3: %08x "
118 	       "l4: %08x l5: %08x l6: %08x l7: %08x\n",
119 	       r_w.locals[0], r_w.locals[1], r_w.locals[2], r_w.locals[3],
120 	       r_w.locals[4], r_w.locals[5], r_w.locals[6], r_w.locals[7]);
121 	printk("i0: %08x i1: %08x i2: %08x i3: %08x "
122 	       "i4: %08x i5: %08x i6: %08x i7: %08x\n",
123 	       r_w.ins[0], r_w.ins[1], r_w.ins[2], r_w.ins[3],
124 	       r_w.ins[4], r_w.ins[5], r_w.ins[6], r_w.ins[7]);
125 }
126 #else
127 #define show_regwindow32(regs)	do { } while (0)
128 #endif
129 
130 static void show_regwindow(struct pt_regs *regs)
131 {
132 	struct reg_window __user *rw;
133 	struct reg_window *rwk;
134 	struct reg_window r_w;
135 
136 	if ((regs->tstate & TSTATE_PRIV) || !(test_thread_flag(TIF_32BIT))) {
137 		__asm__ __volatile__ ("flushw");
138 		rw = (struct reg_window __user *)
139 			(regs->u_regs[14] + STACK_BIAS);
140 		rwk = (struct reg_window *)
141 			(regs->u_regs[14] + STACK_BIAS);
142 		if (!(regs->tstate & TSTATE_PRIV)) {
143 			if (copy_from_user (&r_w, rw, sizeof(r_w))) {
144 				return;
145 			}
146 			rwk = &r_w;
147 		}
148 	} else {
149 		show_regwindow32(regs);
150 		return;
151 	}
152 	printk("l0: %016lx l1: %016lx l2: %016lx l3: %016lx\n",
153 	       rwk->locals[0], rwk->locals[1], rwk->locals[2], rwk->locals[3]);
154 	printk("l4: %016lx l5: %016lx l6: %016lx l7: %016lx\n",
155 	       rwk->locals[4], rwk->locals[5], rwk->locals[6], rwk->locals[7]);
156 	printk("i0: %016lx i1: %016lx i2: %016lx i3: %016lx\n",
157 	       rwk->ins[0], rwk->ins[1], rwk->ins[2], rwk->ins[3]);
158 	printk("i4: %016lx i5: %016lx i6: %016lx i7: %016lx\n",
159 	       rwk->ins[4], rwk->ins[5], rwk->ins[6], rwk->ins[7]);
160 	if (regs->tstate & TSTATE_PRIV)
161 		printk("I7: <%pS>\n", (void *) rwk->ins[7]);
162 }
163 
164 void show_regs(struct pt_regs *regs)
165 {
166 	show_regs_print_info(KERN_DEFAULT);
167 
168 	printk("TSTATE: %016lx TPC: %016lx TNPC: %016lx Y: %08x    %s\n", regs->tstate,
169 	       regs->tpc, regs->tnpc, regs->y, print_tainted());
170 	printk("TPC: <%pS>\n", (void *) regs->tpc);
171 	printk("g0: %016lx g1: %016lx g2: %016lx g3: %016lx\n",
172 	       regs->u_regs[0], regs->u_regs[1], regs->u_regs[2],
173 	       regs->u_regs[3]);
174 	printk("g4: %016lx g5: %016lx g6: %016lx g7: %016lx\n",
175 	       regs->u_regs[4], regs->u_regs[5], regs->u_regs[6],
176 	       regs->u_regs[7]);
177 	printk("o0: %016lx o1: %016lx o2: %016lx o3: %016lx\n",
178 	       regs->u_regs[8], regs->u_regs[9], regs->u_regs[10],
179 	       regs->u_regs[11]);
180 	printk("o4: %016lx o5: %016lx sp: %016lx ret_pc: %016lx\n",
181 	       regs->u_regs[12], regs->u_regs[13], regs->u_regs[14],
182 	       regs->u_regs[15]);
183 	printk("RPC: <%pS>\n", (void *) regs->u_regs[15]);
184 	show_regwindow(regs);
185 	show_stack(current, (unsigned long *)regs->u_regs[UREG_FP], KERN_DEFAULT);
186 }
187 
188 union global_cpu_snapshot global_cpu_snapshot[NR_CPUS];
189 static DEFINE_SPINLOCK(global_cpu_snapshot_lock);
190 
191 static void __global_reg_self(struct thread_info *tp, struct pt_regs *regs,
192 			      int this_cpu)
193 {
194 	struct global_reg_snapshot *rp;
195 
196 	flushw_all();
197 
198 	rp = &global_cpu_snapshot[this_cpu].reg;
199 
200 	rp->tstate = regs->tstate;
201 	rp->tpc = regs->tpc;
202 	rp->tnpc = regs->tnpc;
203 	rp->o7 = regs->u_regs[UREG_I7];
204 
205 	if (regs->tstate & TSTATE_PRIV) {
206 		struct reg_window *rw;
207 
208 		rw = (struct reg_window *)
209 			(regs->u_regs[UREG_FP] + STACK_BIAS);
210 		if (kstack_valid(tp, (unsigned long) rw)) {
211 			rp->i7 = rw->ins[7];
212 			rw = (struct reg_window *)
213 				(rw->ins[6] + STACK_BIAS);
214 			if (kstack_valid(tp, (unsigned long) rw))
215 				rp->rpc = rw->ins[7];
216 		}
217 	} else {
218 		rp->i7 = 0;
219 		rp->rpc = 0;
220 	}
221 	rp->thread = tp;
222 }
223 
224 /* In order to avoid hangs we do not try to synchronize with the
225  * global register dump client cpus.  The last store they make is to
226  * the thread pointer, so do a short poll waiting for that to become
227  * non-NULL.
228  */
229 static void __global_reg_poll(struct global_reg_snapshot *gp)
230 {
231 	int limit = 0;
232 
233 	while (!gp->thread && ++limit < 100) {
234 		barrier();
235 		udelay(1);
236 	}
237 }
238 
239 void arch_trigger_cpumask_backtrace(const cpumask_t *mask, bool exclude_self)
240 {
241 	struct thread_info *tp = current_thread_info();
242 	struct pt_regs *regs = get_irq_regs();
243 	unsigned long flags;
244 	int this_cpu, cpu;
245 
246 	if (!regs)
247 		regs = tp->kregs;
248 
249 	spin_lock_irqsave(&global_cpu_snapshot_lock, flags);
250 
251 	this_cpu = raw_smp_processor_id();
252 
253 	memset(global_cpu_snapshot, 0, sizeof(global_cpu_snapshot));
254 
255 	if (cpumask_test_cpu(this_cpu, mask) && !exclude_self)
256 		__global_reg_self(tp, regs, this_cpu);
257 
258 	smp_fetch_global_regs();
259 
260 	for_each_cpu(cpu, mask) {
261 		struct global_reg_snapshot *gp;
262 
263 		if (exclude_self && cpu == this_cpu)
264 			continue;
265 
266 		gp = &global_cpu_snapshot[cpu].reg;
267 
268 		__global_reg_poll(gp);
269 
270 		tp = gp->thread;
271 		printk("%c CPU[%3d]: TSTATE[%016lx] TPC[%016lx] TNPC[%016lx] TASK[%s:%d]\n",
272 		       (cpu == this_cpu ? '*' : ' '), cpu,
273 		       gp->tstate, gp->tpc, gp->tnpc,
274 		       ((tp && tp->task) ? tp->task->comm : "NULL"),
275 		       ((tp && tp->task) ? tp->task->pid : -1));
276 
277 		if (gp->tstate & TSTATE_PRIV) {
278 			printk("             TPC[%pS] O7[%pS] I7[%pS] RPC[%pS]\n",
279 			       (void *) gp->tpc,
280 			       (void *) gp->o7,
281 			       (void *) gp->i7,
282 			       (void *) gp->rpc);
283 		} else {
284 			printk("             TPC[%lx] O7[%lx] I7[%lx] RPC[%lx]\n",
285 			       gp->tpc, gp->o7, gp->i7, gp->rpc);
286 		}
287 
288 		touch_nmi_watchdog();
289 	}
290 
291 	memset(global_cpu_snapshot, 0, sizeof(global_cpu_snapshot));
292 
293 	spin_unlock_irqrestore(&global_cpu_snapshot_lock, flags);
294 }
295 
296 #ifdef CONFIG_MAGIC_SYSRQ
297 
298 static void sysrq_handle_globreg(int key)
299 {
300 	trigger_all_cpu_backtrace();
301 }
302 
303 static const struct sysrq_key_op sparc_globalreg_op = {
304 	.handler	= sysrq_handle_globreg,
305 	.help_msg	= "global-regs(y)",
306 	.action_msg	= "Show Global CPU Regs",
307 };
308 
309 static void __global_pmu_self(int this_cpu)
310 {
311 	struct global_pmu_snapshot *pp;
312 	int i, num;
313 
314 	if (!pcr_ops)
315 		return;
316 
317 	pp = &global_cpu_snapshot[this_cpu].pmu;
318 
319 	num = 1;
320 	if (tlb_type == hypervisor &&
321 	    sun4v_chip_type >= SUN4V_CHIP_NIAGARA4)
322 		num = 4;
323 
324 	for (i = 0; i < num; i++) {
325 		pp->pcr[i] = pcr_ops->read_pcr(i);
326 		pp->pic[i] = pcr_ops->read_pic(i);
327 	}
328 }
329 
330 static void __global_pmu_poll(struct global_pmu_snapshot *pp)
331 {
332 	int limit = 0;
333 
334 	while (!pp->pcr[0] && ++limit < 100) {
335 		barrier();
336 		udelay(1);
337 	}
338 }
339 
340 static void pmu_snapshot_all_cpus(void)
341 {
342 	unsigned long flags;
343 	int this_cpu, cpu;
344 
345 	spin_lock_irqsave(&global_cpu_snapshot_lock, flags);
346 
347 	memset(global_cpu_snapshot, 0, sizeof(global_cpu_snapshot));
348 
349 	this_cpu = raw_smp_processor_id();
350 
351 	__global_pmu_self(this_cpu);
352 
353 	smp_fetch_global_pmu();
354 
355 	for_each_online_cpu(cpu) {
356 		struct global_pmu_snapshot *pp = &global_cpu_snapshot[cpu].pmu;
357 
358 		__global_pmu_poll(pp);
359 
360 		printk("%c CPU[%3d]: PCR[%08lx:%08lx:%08lx:%08lx] PIC[%08lx:%08lx:%08lx:%08lx]\n",
361 		       (cpu == this_cpu ? '*' : ' '), cpu,
362 		       pp->pcr[0], pp->pcr[1], pp->pcr[2], pp->pcr[3],
363 		       pp->pic[0], pp->pic[1], pp->pic[2], pp->pic[3]);
364 
365 		touch_nmi_watchdog();
366 	}
367 
368 	memset(global_cpu_snapshot, 0, sizeof(global_cpu_snapshot));
369 
370 	spin_unlock_irqrestore(&global_cpu_snapshot_lock, flags);
371 }
372 
373 static void sysrq_handle_globpmu(int key)
374 {
375 	pmu_snapshot_all_cpus();
376 }
377 
378 static const struct sysrq_key_op sparc_globalpmu_op = {
379 	.handler	= sysrq_handle_globpmu,
380 	.help_msg	= "global-pmu(x)",
381 	.action_msg	= "Show Global PMU Regs",
382 };
383 
384 static int __init sparc_sysrq_init(void)
385 {
386 	int ret = register_sysrq_key('y', &sparc_globalreg_op);
387 
388 	if (!ret)
389 		ret = register_sysrq_key('x', &sparc_globalpmu_op);
390 	return ret;
391 }
392 
393 core_initcall(sparc_sysrq_init);
394 
395 #endif
396 
397 /* Free current thread data structures etc.. */
398 void exit_thread(struct task_struct *tsk)
399 {
400 	struct thread_info *t = task_thread_info(tsk);
401 
402 	if (t->utraps) {
403 		if (t->utraps[0] < 2)
404 			kfree (t->utraps);
405 		else
406 			t->utraps[0]--;
407 	}
408 }
409 
410 void flush_thread(void)
411 {
412 	struct thread_info *t = current_thread_info();
413 	struct mm_struct *mm;
414 
415 	mm = t->task->mm;
416 	if (mm)
417 		tsb_context_switch(mm);
418 
419 	set_thread_wsaved(0);
420 
421 	/* Clear FPU register state. */
422 	t->fpsaved[0] = 0;
423 }
424 
425 /* It's a bit more tricky when 64-bit tasks are involved... */
426 static unsigned long clone_stackframe(unsigned long csp, unsigned long psp)
427 {
428 	bool stack_64bit = test_thread_64bit_stack(psp);
429 	unsigned long fp, distance, rval;
430 
431 	if (stack_64bit) {
432 		csp += STACK_BIAS;
433 		psp += STACK_BIAS;
434 		__get_user(fp, &(((struct reg_window __user *)psp)->ins[6]));
435 		fp += STACK_BIAS;
436 		if (test_thread_flag(TIF_32BIT))
437 			fp &= 0xffffffff;
438 	} else
439 		__get_user(fp, &(((struct reg_window32 __user *)psp)->ins[6]));
440 
441 	/* Now align the stack as this is mandatory in the Sparc ABI
442 	 * due to how register windows work.  This hides the
443 	 * restriction from thread libraries etc.
444 	 */
445 	csp &= ~15UL;
446 
447 	distance = fp - psp;
448 	rval = (csp - distance);
449 	if (raw_copy_in_user((void __user *)rval, (void __user *)psp, distance))
450 		rval = 0;
451 	else if (!stack_64bit) {
452 		if (put_user(((u32)csp),
453 			     &(((struct reg_window32 __user *)rval)->ins[6])))
454 			rval = 0;
455 	} else {
456 		if (put_user(((u64)csp - STACK_BIAS),
457 			     &(((struct reg_window __user *)rval)->ins[6])))
458 			rval = 0;
459 		else
460 			rval = rval - STACK_BIAS;
461 	}
462 
463 	return rval;
464 }
465 
466 /* Standard stuff. */
467 static inline void shift_window_buffer(int first_win, int last_win,
468 				       struct thread_info *t)
469 {
470 	int i;
471 
472 	for (i = first_win; i < last_win; i++) {
473 		t->rwbuf_stkptrs[i] = t->rwbuf_stkptrs[i+1];
474 		memcpy(&t->reg_window[i], &t->reg_window[i+1],
475 		       sizeof(struct reg_window));
476 	}
477 }
478 
479 void synchronize_user_stack(void)
480 {
481 	struct thread_info *t = current_thread_info();
482 	unsigned long window;
483 
484 	flush_user_windows();
485 	if ((window = get_thread_wsaved()) != 0) {
486 		window -= 1;
487 		do {
488 			struct reg_window *rwin = &t->reg_window[window];
489 			int winsize = sizeof(struct reg_window);
490 			unsigned long sp;
491 
492 			sp = t->rwbuf_stkptrs[window];
493 
494 			if (test_thread_64bit_stack(sp))
495 				sp += STACK_BIAS;
496 			else
497 				winsize = sizeof(struct reg_window32);
498 
499 			if (!copy_to_user((char __user *)sp, rwin, winsize)) {
500 				shift_window_buffer(window, get_thread_wsaved() - 1, t);
501 				set_thread_wsaved(get_thread_wsaved() - 1);
502 			}
503 		} while (window--);
504 	}
505 }
506 
507 static void stack_unaligned(unsigned long sp)
508 {
509 	force_sig_fault(SIGBUS, BUS_ADRALN, (void __user *) sp);
510 }
511 
512 static const char uwfault32[] = KERN_INFO \
513 	"%s[%d]: bad register window fault: SP %08lx (orig_sp %08lx) TPC %08lx O7 %08lx\n";
514 static const char uwfault64[] = KERN_INFO \
515 	"%s[%d]: bad register window fault: SP %016lx (orig_sp %016lx) TPC %08lx O7 %016lx\n";
516 
517 void fault_in_user_windows(struct pt_regs *regs)
518 {
519 	struct thread_info *t = current_thread_info();
520 	unsigned long window;
521 
522 	flush_user_windows();
523 	window = get_thread_wsaved();
524 
525 	if (likely(window != 0)) {
526 		window -= 1;
527 		do {
528 			struct reg_window *rwin = &t->reg_window[window];
529 			int winsize = sizeof(struct reg_window);
530 			unsigned long sp, orig_sp;
531 
532 			orig_sp = sp = t->rwbuf_stkptrs[window];
533 
534 			if (test_thread_64bit_stack(sp))
535 				sp += STACK_BIAS;
536 			else
537 				winsize = sizeof(struct reg_window32);
538 
539 			if (unlikely(sp & 0x7UL))
540 				stack_unaligned(sp);
541 
542 			if (unlikely(copy_to_user((char __user *)sp,
543 						  rwin, winsize))) {
544 				if (show_unhandled_signals)
545 					printk_ratelimited(is_compat_task() ?
546 							   uwfault32 : uwfault64,
547 							   current->comm, current->pid,
548 							   sp, orig_sp,
549 							   regs->tpc,
550 							   regs->u_regs[UREG_I7]);
551 				goto barf;
552 			}
553 		} while (window--);
554 	}
555 	set_thread_wsaved(0);
556 	return;
557 
558 barf:
559 	set_thread_wsaved(window + 1);
560 	force_sig(SIGSEGV);
561 }
562 
563 /* Copy a Sparc thread.  The fork() return value conventions
564  * under SunOS are nothing short of bletcherous:
565  * Parent -->  %o0 == childs  pid, %o1 == 0
566  * Child  -->  %o0 == parents pid, %o1 == 1
567  */
568 int copy_thread(struct task_struct *p, const struct kernel_clone_args *args)
569 {
570 	unsigned long clone_flags = args->flags;
571 	unsigned long sp = args->stack;
572 	unsigned long tls = args->tls;
573 	struct thread_info *t = task_thread_info(p);
574 	struct pt_regs *regs = current_pt_regs();
575 	struct sparc_stackf *parent_sf;
576 	unsigned long child_stack_sz;
577 	char *child_trap_frame;
578 
579 	/* Calculate offset to stack_frame & pt_regs */
580 	child_stack_sz = (STACKFRAME_SZ + TRACEREG_SZ);
581 	child_trap_frame = (task_stack_page(p) +
582 			    (THREAD_SIZE - child_stack_sz));
583 
584 	t->new_child = 1;
585 	t->ksp = ((unsigned long) child_trap_frame) - STACK_BIAS;
586 	t->kregs = (struct pt_regs *) (child_trap_frame +
587 				       sizeof(struct sparc_stackf));
588 	t->fpsaved[0] = 0;
589 
590 	if (unlikely(args->fn)) {
591 		memset(child_trap_frame, 0, child_stack_sz);
592 		__thread_flag_byte_ptr(t)[TI_FLAG_BYTE_CWP] =
593 			(current_pt_regs()->tstate + 1) & TSTATE_CWP;
594 		t->kregs->u_regs[UREG_G1] = (unsigned long) args->fn;
595 		t->kregs->u_regs[UREG_G2] = (unsigned long) args->fn_arg;
596 		return 0;
597 	}
598 
599 	parent_sf = ((struct sparc_stackf *) regs) - 1;
600 	memcpy(child_trap_frame, parent_sf, child_stack_sz);
601 	if (t->flags & _TIF_32BIT) {
602 		sp &= 0x00000000ffffffffUL;
603 		regs->u_regs[UREG_FP] &= 0x00000000ffffffffUL;
604 	}
605 	t->kregs->u_regs[UREG_FP] = sp;
606 	__thread_flag_byte_ptr(t)[TI_FLAG_BYTE_CWP] =
607 		(regs->tstate + 1) & TSTATE_CWP;
608 	if (sp != regs->u_regs[UREG_FP]) {
609 		unsigned long csp;
610 
611 		csp = clone_stackframe(sp, regs->u_regs[UREG_FP]);
612 		if (!csp)
613 			return -EFAULT;
614 		t->kregs->u_regs[UREG_FP] = csp;
615 	}
616 	if (t->utraps)
617 		t->utraps[0]++;
618 
619 	/* Set the return value for the child. */
620 	t->kregs->u_regs[UREG_I0] = current->pid;
621 	t->kregs->u_regs[UREG_I1] = 1;
622 
623 	/* Set the second return value for the parent. */
624 	regs->u_regs[UREG_I1] = 0;
625 
626 	if (clone_flags & CLONE_SETTLS)
627 		t->kregs->u_regs[UREG_G7] = tls;
628 
629 	return 0;
630 }
631 
632 /* TIF_MCDPER in thread info flags for current task is updated lazily upon
633  * a context switch. Update this flag in current task's thread flags
634  * before dup so the dup'd task will inherit the current TIF_MCDPER flag.
635  */
636 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
637 {
638 	if (adi_capable()) {
639 		register unsigned long tmp_mcdper;
640 
641 		__asm__ __volatile__(
642 			".word 0x83438000\n\t"	/* rd  %mcdper, %g1 */
643 			"mov %%g1, %0\n\t"
644 			: "=r" (tmp_mcdper)
645 			:
646 			: "g1");
647 		if (tmp_mcdper)
648 			set_thread_flag(TIF_MCDPER);
649 		else
650 			clear_thread_flag(TIF_MCDPER);
651 	}
652 
653 	*dst = *src;
654 	return 0;
655 }
656 
657 unsigned long __get_wchan(struct task_struct *task)
658 {
659 	unsigned long pc, fp, bias = 0;
660 	struct thread_info *tp;
661 	struct reg_window *rw;
662         unsigned long ret = 0;
663 	int count = 0;
664 
665 	tp = task_thread_info(task);
666 	bias = STACK_BIAS;
667 	fp = task_thread_info(task)->ksp + bias;
668 
669 	do {
670 		if (!kstack_valid(tp, fp))
671 			break;
672 		rw = (struct reg_window *) fp;
673 		pc = rw->ins[7];
674 		if (!in_sched_functions(pc)) {
675 			ret = pc;
676 			goto out;
677 		}
678 		fp = rw->ins[6] + bias;
679 	} while (++count < 16);
680 
681 out:
682 	return ret;
683 }
684