xref: /openbmc/linux/arch/sparc/kernel/process_64.c (revision 1fa6ac37)
1 /*  arch/sparc64/kernel/process.c
2  *
3  *  Copyright (C) 1995, 1996, 2008 David S. Miller (davem@davemloft.net)
4  *  Copyright (C) 1996       Eddie C. Dost   (ecd@skynet.be)
5  *  Copyright (C) 1997, 1998 Jakub Jelinek   (jj@sunsite.mff.cuni.cz)
6  */
7 
8 /*
9  * This file handles the architecture-dependent parts of process handling..
10  */
11 
12 #include <stdarg.h>
13 
14 #include <linux/errno.h>
15 #include <linux/module.h>
16 #include <linux/sched.h>
17 #include <linux/kernel.h>
18 #include <linux/mm.h>
19 #include <linux/fs.h>
20 #include <linux/smp.h>
21 #include <linux/stddef.h>
22 #include <linux/ptrace.h>
23 #include <linux/slab.h>
24 #include <linux/user.h>
25 #include <linux/delay.h>
26 #include <linux/compat.h>
27 #include <linux/tick.h>
28 #include <linux/init.h>
29 #include <linux/cpu.h>
30 #include <linux/elfcore.h>
31 #include <linux/sysrq.h>
32 #include <linux/nmi.h>
33 
34 #include <asm/uaccess.h>
35 #include <asm/system.h>
36 #include <asm/page.h>
37 #include <asm/pgalloc.h>
38 #include <asm/pgtable.h>
39 #include <asm/processor.h>
40 #include <asm/pstate.h>
41 #include <asm/elf.h>
42 #include <asm/fpumacro.h>
43 #include <asm/head.h>
44 #include <asm/cpudata.h>
45 #include <asm/mmu_context.h>
46 #include <asm/unistd.h>
47 #include <asm/hypervisor.h>
48 #include <asm/syscalls.h>
49 #include <asm/irq_regs.h>
50 #include <asm/smp.h>
51 
52 #include "kstack.h"
53 
54 static void sparc64_yield(int cpu)
55 {
56 	if (tlb_type != hypervisor) {
57 		touch_nmi_watchdog();
58 		return;
59 	}
60 
61 	clear_thread_flag(TIF_POLLING_NRFLAG);
62 	smp_mb__after_clear_bit();
63 
64 	while (!need_resched() && !cpu_is_offline(cpu)) {
65 		unsigned long pstate;
66 
67 		/* Disable interrupts. */
68 		__asm__ __volatile__(
69 			"rdpr %%pstate, %0\n\t"
70 			"andn %0, %1, %0\n\t"
71 			"wrpr %0, %%g0, %%pstate"
72 			: "=&r" (pstate)
73 			: "i" (PSTATE_IE));
74 
75 		if (!need_resched() && !cpu_is_offline(cpu))
76 			sun4v_cpu_yield();
77 
78 		/* Re-enable interrupts. */
79 		__asm__ __volatile__(
80 			"rdpr %%pstate, %0\n\t"
81 			"or %0, %1, %0\n\t"
82 			"wrpr %0, %%g0, %%pstate"
83 			: "=&r" (pstate)
84 			: "i" (PSTATE_IE));
85 	}
86 
87 	set_thread_flag(TIF_POLLING_NRFLAG);
88 }
89 
90 /* The idle loop on sparc64. */
91 void cpu_idle(void)
92 {
93 	int cpu = smp_processor_id();
94 
95 	set_thread_flag(TIF_POLLING_NRFLAG);
96 
97 	while(1) {
98 		tick_nohz_stop_sched_tick(1);
99 
100 		while (!need_resched() && !cpu_is_offline(cpu))
101 			sparc64_yield(cpu);
102 
103 		tick_nohz_restart_sched_tick();
104 
105 		preempt_enable_no_resched();
106 
107 #ifdef CONFIG_HOTPLUG_CPU
108 		if (cpu_is_offline(cpu))
109 			cpu_play_dead();
110 #endif
111 
112 		schedule();
113 		preempt_disable();
114 	}
115 }
116 
117 #ifdef CONFIG_COMPAT
118 static void show_regwindow32(struct pt_regs *regs)
119 {
120 	struct reg_window32 __user *rw;
121 	struct reg_window32 r_w;
122 	mm_segment_t old_fs;
123 
124 	__asm__ __volatile__ ("flushw");
125 	rw = compat_ptr((unsigned)regs->u_regs[14]);
126 	old_fs = get_fs();
127 	set_fs (USER_DS);
128 	if (copy_from_user (&r_w, rw, sizeof(r_w))) {
129 		set_fs (old_fs);
130 		return;
131 	}
132 
133 	set_fs (old_fs);
134 	printk("l0: %08x l1: %08x l2: %08x l3: %08x "
135 	       "l4: %08x l5: %08x l6: %08x l7: %08x\n",
136 	       r_w.locals[0], r_w.locals[1], r_w.locals[2], r_w.locals[3],
137 	       r_w.locals[4], r_w.locals[5], r_w.locals[6], r_w.locals[7]);
138 	printk("i0: %08x i1: %08x i2: %08x i3: %08x "
139 	       "i4: %08x i5: %08x i6: %08x i7: %08x\n",
140 	       r_w.ins[0], r_w.ins[1], r_w.ins[2], r_w.ins[3],
141 	       r_w.ins[4], r_w.ins[5], r_w.ins[6], r_w.ins[7]);
142 }
143 #else
144 #define show_regwindow32(regs)	do { } while (0)
145 #endif
146 
147 static void show_regwindow(struct pt_regs *regs)
148 {
149 	struct reg_window __user *rw;
150 	struct reg_window *rwk;
151 	struct reg_window r_w;
152 	mm_segment_t old_fs;
153 
154 	if ((regs->tstate & TSTATE_PRIV) || !(test_thread_flag(TIF_32BIT))) {
155 		__asm__ __volatile__ ("flushw");
156 		rw = (struct reg_window __user *)
157 			(regs->u_regs[14] + STACK_BIAS);
158 		rwk = (struct reg_window *)
159 			(regs->u_regs[14] + STACK_BIAS);
160 		if (!(regs->tstate & TSTATE_PRIV)) {
161 			old_fs = get_fs();
162 			set_fs (USER_DS);
163 			if (copy_from_user (&r_w, rw, sizeof(r_w))) {
164 				set_fs (old_fs);
165 				return;
166 			}
167 			rwk = &r_w;
168 			set_fs (old_fs);
169 		}
170 	} else {
171 		show_regwindow32(regs);
172 		return;
173 	}
174 	printk("l0: %016lx l1: %016lx l2: %016lx l3: %016lx\n",
175 	       rwk->locals[0], rwk->locals[1], rwk->locals[2], rwk->locals[3]);
176 	printk("l4: %016lx l5: %016lx l6: %016lx l7: %016lx\n",
177 	       rwk->locals[4], rwk->locals[5], rwk->locals[6], rwk->locals[7]);
178 	printk("i0: %016lx i1: %016lx i2: %016lx i3: %016lx\n",
179 	       rwk->ins[0], rwk->ins[1], rwk->ins[2], rwk->ins[3]);
180 	printk("i4: %016lx i5: %016lx i6: %016lx i7: %016lx\n",
181 	       rwk->ins[4], rwk->ins[5], rwk->ins[6], rwk->ins[7]);
182 	if (regs->tstate & TSTATE_PRIV)
183 		printk("I7: <%pS>\n", (void *) rwk->ins[7]);
184 }
185 
186 void show_regs(struct pt_regs *regs)
187 {
188 	printk("TSTATE: %016lx TPC: %016lx TNPC: %016lx Y: %08x    %s\n", regs->tstate,
189 	       regs->tpc, regs->tnpc, regs->y, print_tainted());
190 	printk("TPC: <%pS>\n", (void *) regs->tpc);
191 	printk("g0: %016lx g1: %016lx g2: %016lx g3: %016lx\n",
192 	       regs->u_regs[0], regs->u_regs[1], regs->u_regs[2],
193 	       regs->u_regs[3]);
194 	printk("g4: %016lx g5: %016lx g6: %016lx g7: %016lx\n",
195 	       regs->u_regs[4], regs->u_regs[5], regs->u_regs[6],
196 	       regs->u_regs[7]);
197 	printk("o0: %016lx o1: %016lx o2: %016lx o3: %016lx\n",
198 	       regs->u_regs[8], regs->u_regs[9], regs->u_regs[10],
199 	       regs->u_regs[11]);
200 	printk("o4: %016lx o5: %016lx sp: %016lx ret_pc: %016lx\n",
201 	       regs->u_regs[12], regs->u_regs[13], regs->u_regs[14],
202 	       regs->u_regs[15]);
203 	printk("RPC: <%pS>\n", (void *) regs->u_regs[15]);
204 	show_regwindow(regs);
205 	show_stack(current, (unsigned long *) regs->u_regs[UREG_FP]);
206 }
207 
208 struct global_reg_snapshot global_reg_snapshot[NR_CPUS];
209 static DEFINE_SPINLOCK(global_reg_snapshot_lock);
210 
211 static void __global_reg_self(struct thread_info *tp, struct pt_regs *regs,
212 			      int this_cpu)
213 {
214 	flushw_all();
215 
216 	global_reg_snapshot[this_cpu].tstate = regs->tstate;
217 	global_reg_snapshot[this_cpu].tpc = regs->tpc;
218 	global_reg_snapshot[this_cpu].tnpc = regs->tnpc;
219 	global_reg_snapshot[this_cpu].o7 = regs->u_regs[UREG_I7];
220 
221 	if (regs->tstate & TSTATE_PRIV) {
222 		struct reg_window *rw;
223 
224 		rw = (struct reg_window *)
225 			(regs->u_regs[UREG_FP] + STACK_BIAS);
226 		if (kstack_valid(tp, (unsigned long) rw)) {
227 			global_reg_snapshot[this_cpu].i7 = rw->ins[7];
228 			rw = (struct reg_window *)
229 				(rw->ins[6] + STACK_BIAS);
230 			if (kstack_valid(tp, (unsigned long) rw))
231 				global_reg_snapshot[this_cpu].rpc = rw->ins[7];
232 		}
233 	} else {
234 		global_reg_snapshot[this_cpu].i7 = 0;
235 		global_reg_snapshot[this_cpu].rpc = 0;
236 	}
237 	global_reg_snapshot[this_cpu].thread = tp;
238 }
239 
240 /* In order to avoid hangs we do not try to synchronize with the
241  * global register dump client cpus.  The last store they make is to
242  * the thread pointer, so do a short poll waiting for that to become
243  * non-NULL.
244  */
245 static void __global_reg_poll(struct global_reg_snapshot *gp)
246 {
247 	int limit = 0;
248 
249 	while (!gp->thread && ++limit < 100) {
250 		barrier();
251 		udelay(1);
252 	}
253 }
254 
255 void arch_trigger_all_cpu_backtrace(void)
256 {
257 	struct thread_info *tp = current_thread_info();
258 	struct pt_regs *regs = get_irq_regs();
259 	unsigned long flags;
260 	int this_cpu, cpu;
261 
262 	if (!regs)
263 		regs = tp->kregs;
264 
265 	spin_lock_irqsave(&global_reg_snapshot_lock, flags);
266 
267 	memset(global_reg_snapshot, 0, sizeof(global_reg_snapshot));
268 
269 	this_cpu = raw_smp_processor_id();
270 
271 	__global_reg_self(tp, regs, this_cpu);
272 
273 	smp_fetch_global_regs();
274 
275 	for_each_online_cpu(cpu) {
276 		struct global_reg_snapshot *gp = &global_reg_snapshot[cpu];
277 
278 		__global_reg_poll(gp);
279 
280 		tp = gp->thread;
281 		printk("%c CPU[%3d]: TSTATE[%016lx] TPC[%016lx] TNPC[%016lx] TASK[%s:%d]\n",
282 		       (cpu == this_cpu ? '*' : ' '), cpu,
283 		       gp->tstate, gp->tpc, gp->tnpc,
284 		       ((tp && tp->task) ? tp->task->comm : "NULL"),
285 		       ((tp && tp->task) ? tp->task->pid : -1));
286 
287 		if (gp->tstate & TSTATE_PRIV) {
288 			printk("             TPC[%pS] O7[%pS] I7[%pS] RPC[%pS]\n",
289 			       (void *) gp->tpc,
290 			       (void *) gp->o7,
291 			       (void *) gp->i7,
292 			       (void *) gp->rpc);
293 		} else {
294 			printk("             TPC[%lx] O7[%lx] I7[%lx] RPC[%lx]\n",
295 			       gp->tpc, gp->o7, gp->i7, gp->rpc);
296 		}
297 	}
298 
299 	memset(global_reg_snapshot, 0, sizeof(global_reg_snapshot));
300 
301 	spin_unlock_irqrestore(&global_reg_snapshot_lock, flags);
302 }
303 
304 #ifdef CONFIG_MAGIC_SYSRQ
305 
306 static void sysrq_handle_globreg(int key, struct tty_struct *tty)
307 {
308 	arch_trigger_all_cpu_backtrace();
309 }
310 
311 static struct sysrq_key_op sparc_globalreg_op = {
312 	.handler	= sysrq_handle_globreg,
313 	.help_msg	= "Globalregs",
314 	.action_msg	= "Show Global CPU Regs",
315 };
316 
317 static int __init sparc_globreg_init(void)
318 {
319 	return register_sysrq_key('y', &sparc_globalreg_op);
320 }
321 
322 core_initcall(sparc_globreg_init);
323 
324 #endif
325 
326 unsigned long thread_saved_pc(struct task_struct *tsk)
327 {
328 	struct thread_info *ti = task_thread_info(tsk);
329 	unsigned long ret = 0xdeadbeefUL;
330 
331 	if (ti && ti->ksp) {
332 		unsigned long *sp;
333 		sp = (unsigned long *)(ti->ksp + STACK_BIAS);
334 		if (((unsigned long)sp & (sizeof(long) - 1)) == 0UL &&
335 		    sp[14]) {
336 			unsigned long *fp;
337 			fp = (unsigned long *)(sp[14] + STACK_BIAS);
338 			if (((unsigned long)fp & (sizeof(long) - 1)) == 0UL)
339 				ret = fp[15];
340 		}
341 	}
342 	return ret;
343 }
344 
345 /* Free current thread data structures etc.. */
346 void exit_thread(void)
347 {
348 	struct thread_info *t = current_thread_info();
349 
350 	if (t->utraps) {
351 		if (t->utraps[0] < 2)
352 			kfree (t->utraps);
353 		else
354 			t->utraps[0]--;
355 	}
356 }
357 
358 void flush_thread(void)
359 {
360 	struct thread_info *t = current_thread_info();
361 	struct mm_struct *mm;
362 
363 	mm = t->task->mm;
364 	if (mm)
365 		tsb_context_switch(mm);
366 
367 	set_thread_wsaved(0);
368 
369 	/* Clear FPU register state. */
370 	t->fpsaved[0] = 0;
371 
372 	if (get_thread_current_ds() != ASI_AIUS)
373 		set_fs(USER_DS);
374 }
375 
376 /* It's a bit more tricky when 64-bit tasks are involved... */
377 static unsigned long clone_stackframe(unsigned long csp, unsigned long psp)
378 {
379 	unsigned long fp, distance, rval;
380 
381 	if (!(test_thread_flag(TIF_32BIT))) {
382 		csp += STACK_BIAS;
383 		psp += STACK_BIAS;
384 		__get_user(fp, &(((struct reg_window __user *)psp)->ins[6]));
385 		fp += STACK_BIAS;
386 	} else
387 		__get_user(fp, &(((struct reg_window32 __user *)psp)->ins[6]));
388 
389 	/* Now align the stack as this is mandatory in the Sparc ABI
390 	 * due to how register windows work.  This hides the
391 	 * restriction from thread libraries etc.
392 	 */
393 	csp &= ~15UL;
394 
395 	distance = fp - psp;
396 	rval = (csp - distance);
397 	if (copy_in_user((void __user *) rval, (void __user *) psp, distance))
398 		rval = 0;
399 	else if (test_thread_flag(TIF_32BIT)) {
400 		if (put_user(((u32)csp),
401 			     &(((struct reg_window32 __user *)rval)->ins[6])))
402 			rval = 0;
403 	} else {
404 		if (put_user(((u64)csp - STACK_BIAS),
405 			     &(((struct reg_window __user *)rval)->ins[6])))
406 			rval = 0;
407 		else
408 			rval = rval - STACK_BIAS;
409 	}
410 
411 	return rval;
412 }
413 
414 /* Standard stuff. */
415 static inline void shift_window_buffer(int first_win, int last_win,
416 				       struct thread_info *t)
417 {
418 	int i;
419 
420 	for (i = first_win; i < last_win; i++) {
421 		t->rwbuf_stkptrs[i] = t->rwbuf_stkptrs[i+1];
422 		memcpy(&t->reg_window[i], &t->reg_window[i+1],
423 		       sizeof(struct reg_window));
424 	}
425 }
426 
427 void synchronize_user_stack(void)
428 {
429 	struct thread_info *t = current_thread_info();
430 	unsigned long window;
431 
432 	flush_user_windows();
433 	if ((window = get_thread_wsaved()) != 0) {
434 		int winsize = sizeof(struct reg_window);
435 		int bias = 0;
436 
437 		if (test_thread_flag(TIF_32BIT))
438 			winsize = sizeof(struct reg_window32);
439 		else
440 			bias = STACK_BIAS;
441 
442 		window -= 1;
443 		do {
444 			unsigned long sp = (t->rwbuf_stkptrs[window] + bias);
445 			struct reg_window *rwin = &t->reg_window[window];
446 
447 			if (!copy_to_user((char __user *)sp, rwin, winsize)) {
448 				shift_window_buffer(window, get_thread_wsaved() - 1, t);
449 				set_thread_wsaved(get_thread_wsaved() - 1);
450 			}
451 		} while (window--);
452 	}
453 }
454 
455 static void stack_unaligned(unsigned long sp)
456 {
457 	siginfo_t info;
458 
459 	info.si_signo = SIGBUS;
460 	info.si_errno = 0;
461 	info.si_code = BUS_ADRALN;
462 	info.si_addr = (void __user *) sp;
463 	info.si_trapno = 0;
464 	force_sig_info(SIGBUS, &info, current);
465 }
466 
467 void fault_in_user_windows(void)
468 {
469 	struct thread_info *t = current_thread_info();
470 	unsigned long window;
471 	int winsize = sizeof(struct reg_window);
472 	int bias = 0;
473 
474 	if (test_thread_flag(TIF_32BIT))
475 		winsize = sizeof(struct reg_window32);
476 	else
477 		bias = STACK_BIAS;
478 
479 	flush_user_windows();
480 	window = get_thread_wsaved();
481 
482 	if (likely(window != 0)) {
483 		window -= 1;
484 		do {
485 			unsigned long sp = (t->rwbuf_stkptrs[window] + bias);
486 			struct reg_window *rwin = &t->reg_window[window];
487 
488 			if (unlikely(sp & 0x7UL))
489 				stack_unaligned(sp);
490 
491 			if (unlikely(copy_to_user((char __user *)sp,
492 						  rwin, winsize)))
493 				goto barf;
494 		} while (window--);
495 	}
496 	set_thread_wsaved(0);
497 	return;
498 
499 barf:
500 	set_thread_wsaved(window + 1);
501 	do_exit(SIGILL);
502 }
503 
504 asmlinkage long sparc_do_fork(unsigned long clone_flags,
505 			      unsigned long stack_start,
506 			      struct pt_regs *regs,
507 			      unsigned long stack_size)
508 {
509 	int __user *parent_tid_ptr, *child_tid_ptr;
510 	unsigned long orig_i1 = regs->u_regs[UREG_I1];
511 	long ret;
512 
513 #ifdef CONFIG_COMPAT
514 	if (test_thread_flag(TIF_32BIT)) {
515 		parent_tid_ptr = compat_ptr(regs->u_regs[UREG_I2]);
516 		child_tid_ptr = compat_ptr(regs->u_regs[UREG_I4]);
517 	} else
518 #endif
519 	{
520 		parent_tid_ptr = (int __user *) regs->u_regs[UREG_I2];
521 		child_tid_ptr = (int __user *) regs->u_regs[UREG_I4];
522 	}
523 
524 	ret = do_fork(clone_flags, stack_start,
525 		      regs, stack_size,
526 		      parent_tid_ptr, child_tid_ptr);
527 
528 	/* If we get an error and potentially restart the system
529 	 * call, we're screwed because copy_thread() clobbered
530 	 * the parent's %o1.  So detect that case and restore it
531 	 * here.
532 	 */
533 	if ((unsigned long)ret >= -ERESTART_RESTARTBLOCK)
534 		regs->u_regs[UREG_I1] = orig_i1;
535 
536 	return ret;
537 }
538 
539 /* Copy a Sparc thread.  The fork() return value conventions
540  * under SunOS are nothing short of bletcherous:
541  * Parent -->  %o0 == childs  pid, %o1 == 0
542  * Child  -->  %o0 == parents pid, %o1 == 1
543  */
544 int copy_thread(unsigned long clone_flags, unsigned long sp,
545 		unsigned long unused,
546 		struct task_struct *p, struct pt_regs *regs)
547 {
548 	struct thread_info *t = task_thread_info(p);
549 	struct sparc_stackf *parent_sf;
550 	unsigned long child_stack_sz;
551 	char *child_trap_frame;
552 	int kernel_thread;
553 
554 	kernel_thread = (regs->tstate & TSTATE_PRIV) ? 1 : 0;
555 	parent_sf = ((struct sparc_stackf *) regs) - 1;
556 
557 	/* Calculate offset to stack_frame & pt_regs */
558 	child_stack_sz = ((STACKFRAME_SZ + TRACEREG_SZ) +
559 			  (kernel_thread ? STACKFRAME_SZ : 0));
560 	child_trap_frame = (task_stack_page(p) +
561 			    (THREAD_SIZE - child_stack_sz));
562 	memcpy(child_trap_frame, parent_sf, child_stack_sz);
563 
564 	t->flags = (t->flags & ~((0xffUL << TI_FLAG_CWP_SHIFT) |
565 				 (0xffUL << TI_FLAG_CURRENT_DS_SHIFT))) |
566 		(((regs->tstate + 1) & TSTATE_CWP) << TI_FLAG_CWP_SHIFT);
567 	t->new_child = 1;
568 	t->ksp = ((unsigned long) child_trap_frame) - STACK_BIAS;
569 	t->kregs = (struct pt_regs *) (child_trap_frame +
570 				       sizeof(struct sparc_stackf));
571 	t->fpsaved[0] = 0;
572 
573 	if (kernel_thread) {
574 		struct sparc_stackf *child_sf = (struct sparc_stackf *)
575 			(child_trap_frame + (STACKFRAME_SZ + TRACEREG_SZ));
576 
577 		/* Zero terminate the stack backtrace.  */
578 		child_sf->fp = NULL;
579 		t->kregs->u_regs[UREG_FP] =
580 		  ((unsigned long) child_sf) - STACK_BIAS;
581 
582 		t->flags |= ((long)ASI_P << TI_FLAG_CURRENT_DS_SHIFT);
583 		t->kregs->u_regs[UREG_G6] = (unsigned long) t;
584 		t->kregs->u_regs[UREG_G4] = (unsigned long) t->task;
585 	} else {
586 		if (t->flags & _TIF_32BIT) {
587 			sp &= 0x00000000ffffffffUL;
588 			regs->u_regs[UREG_FP] &= 0x00000000ffffffffUL;
589 		}
590 		t->kregs->u_regs[UREG_FP] = sp;
591 		t->flags |= ((long)ASI_AIUS << TI_FLAG_CURRENT_DS_SHIFT);
592 		if (sp != regs->u_regs[UREG_FP]) {
593 			unsigned long csp;
594 
595 			csp = clone_stackframe(sp, regs->u_regs[UREG_FP]);
596 			if (!csp)
597 				return -EFAULT;
598 			t->kregs->u_regs[UREG_FP] = csp;
599 		}
600 		if (t->utraps)
601 			t->utraps[0]++;
602 	}
603 
604 	/* Set the return value for the child. */
605 	t->kregs->u_regs[UREG_I0] = current->pid;
606 	t->kregs->u_regs[UREG_I1] = 1;
607 
608 	/* Set the second return value for the parent. */
609 	regs->u_regs[UREG_I1] = 0;
610 
611 	if (clone_flags & CLONE_SETTLS)
612 		t->kregs->u_regs[UREG_G7] = regs->u_regs[UREG_I3];
613 
614 	return 0;
615 }
616 
617 /*
618  * This is the mechanism for creating a new kernel thread.
619  *
620  * NOTE! Only a kernel-only process(ie the swapper or direct descendants
621  * who haven't done an "execve()") should use this: it will work within
622  * a system call from a "real" process, but the process memory space will
623  * not be freed until both the parent and the child have exited.
624  */
625 pid_t kernel_thread(int (*fn)(void *), void * arg, unsigned long flags)
626 {
627 	long retval;
628 
629 	/* If the parent runs before fn(arg) is called by the child,
630 	 * the input registers of this function can be clobbered.
631 	 * So we stash 'fn' and 'arg' into global registers which
632 	 * will not be modified by the parent.
633 	 */
634 	__asm__ __volatile__("mov %4, %%g2\n\t"	   /* Save FN into global */
635 			     "mov %5, %%g3\n\t"	   /* Save ARG into global */
636 			     "mov %1, %%g1\n\t"	   /* Clone syscall nr. */
637 			     "mov %2, %%o0\n\t"	   /* Clone flags. */
638 			     "mov 0, %%o1\n\t"	   /* usp arg == 0 */
639 			     "t 0x6d\n\t"	   /* Linux/Sparc clone(). */
640 			     "brz,a,pn %%o1, 1f\n\t" /* Parent, just return. */
641 			     " mov %%o0, %0\n\t"
642 			     "jmpl %%g2, %%o7\n\t"   /* Call the function. */
643 			     " mov %%g3, %%o0\n\t"   /* Set arg in delay. */
644 			     "mov %3, %%g1\n\t"
645 			     "t 0x6d\n\t"	   /* Linux/Sparc exit(). */
646 			     /* Notreached by child. */
647 			     "1:" :
648 			     "=r" (retval) :
649 			     "i" (__NR_clone), "r" (flags | CLONE_VM | CLONE_UNTRACED),
650 			     "i" (__NR_exit),  "r" (fn), "r" (arg) :
651 			     "g1", "g2", "g3", "o0", "o1", "memory", "cc");
652 	return retval;
653 }
654 EXPORT_SYMBOL(kernel_thread);
655 
656 typedef struct {
657 	union {
658 		unsigned int	pr_regs[32];
659 		unsigned long	pr_dregs[16];
660 	} pr_fr;
661 	unsigned int __unused;
662 	unsigned int	pr_fsr;
663 	unsigned char	pr_qcnt;
664 	unsigned char	pr_q_entrysize;
665 	unsigned char	pr_en;
666 	unsigned int	pr_q[64];
667 } elf_fpregset_t32;
668 
669 /*
670  * fill in the fpu structure for a core dump.
671  */
672 int dump_fpu (struct pt_regs * regs, elf_fpregset_t * fpregs)
673 {
674 	unsigned long *kfpregs = current_thread_info()->fpregs;
675 	unsigned long fprs = current_thread_info()->fpsaved[0];
676 
677 	if (test_thread_flag(TIF_32BIT)) {
678 		elf_fpregset_t32 *fpregs32 = (elf_fpregset_t32 *)fpregs;
679 
680 		if (fprs & FPRS_DL)
681 			memcpy(&fpregs32->pr_fr.pr_regs[0], kfpregs,
682 			       sizeof(unsigned int) * 32);
683 		else
684 			memset(&fpregs32->pr_fr.pr_regs[0], 0,
685 			       sizeof(unsigned int) * 32);
686 		fpregs32->pr_qcnt = 0;
687 		fpregs32->pr_q_entrysize = 8;
688 		memset(&fpregs32->pr_q[0], 0,
689 		       (sizeof(unsigned int) * 64));
690 		if (fprs & FPRS_FEF) {
691 			fpregs32->pr_fsr = (unsigned int) current_thread_info()->xfsr[0];
692 			fpregs32->pr_en = 1;
693 		} else {
694 			fpregs32->pr_fsr = 0;
695 			fpregs32->pr_en = 0;
696 		}
697 	} else {
698 		if(fprs & FPRS_DL)
699 			memcpy(&fpregs->pr_regs[0], kfpregs,
700 			       sizeof(unsigned int) * 32);
701 		else
702 			memset(&fpregs->pr_regs[0], 0,
703 			       sizeof(unsigned int) * 32);
704 		if(fprs & FPRS_DU)
705 			memcpy(&fpregs->pr_regs[16], kfpregs+16,
706 			       sizeof(unsigned int) * 32);
707 		else
708 			memset(&fpregs->pr_regs[16], 0,
709 			       sizeof(unsigned int) * 32);
710 		if(fprs & FPRS_FEF) {
711 			fpregs->pr_fsr = current_thread_info()->xfsr[0];
712 			fpregs->pr_gsr = current_thread_info()->gsr[0];
713 		} else {
714 			fpregs->pr_fsr = fpregs->pr_gsr = 0;
715 		}
716 		fpregs->pr_fprs = fprs;
717 	}
718 	return 1;
719 }
720 EXPORT_SYMBOL(dump_fpu);
721 
722 /*
723  * sparc_execve() executes a new program after the asm stub has set
724  * things up for us.  This should basically do what I want it to.
725  */
726 asmlinkage int sparc_execve(struct pt_regs *regs)
727 {
728 	int error, base = 0;
729 	char *filename;
730 
731 	/* User register window flush is done by entry.S */
732 
733 	/* Check for indirect call. */
734 	if (regs->u_regs[UREG_G1] == 0)
735 		base = 1;
736 
737 	filename = getname((char __user *)regs->u_regs[base + UREG_I0]);
738 	error = PTR_ERR(filename);
739 	if (IS_ERR(filename))
740 		goto out;
741 	error = do_execve(filename,
742 			  (char __user * __user *)
743 			  regs->u_regs[base + UREG_I1],
744 			  (char __user * __user *)
745 			  regs->u_regs[base + UREG_I2], regs);
746 	putname(filename);
747 	if (!error) {
748 		fprs_write(0);
749 		current_thread_info()->xfsr[0] = 0;
750 		current_thread_info()->fpsaved[0] = 0;
751 		regs->tstate &= ~TSTATE_PEF;
752 	}
753 out:
754 	return error;
755 }
756 
757 unsigned long get_wchan(struct task_struct *task)
758 {
759 	unsigned long pc, fp, bias = 0;
760 	struct thread_info *tp;
761 	struct reg_window *rw;
762         unsigned long ret = 0;
763 	int count = 0;
764 
765 	if (!task || task == current ||
766             task->state == TASK_RUNNING)
767 		goto out;
768 
769 	tp = task_thread_info(task);
770 	bias = STACK_BIAS;
771 	fp = task_thread_info(task)->ksp + bias;
772 
773 	do {
774 		if (!kstack_valid(tp, fp))
775 			break;
776 		rw = (struct reg_window *) fp;
777 		pc = rw->ins[7];
778 		if (!in_sched_functions(pc)) {
779 			ret = pc;
780 			goto out;
781 		}
782 		fp = rw->ins[6] + bias;
783 	} while (++count < 16);
784 
785 out:
786 	return ret;
787 }
788