process.c revision 4c94cada48f7c660eca582be6032427a5e367117
1/*
2 *  linux/arch/alpha/kernel/process.c
3 *
4 *  Copyright (C) 1995  Linus Torvalds
5 */
6
7/*
8 * This file handles the architecture-dependent parts of process handling.
9 */
10
11#include <linux/errno.h>
12#include <linux/module.h>
13#include <linux/sched.h>
14#include <linux/kernel.h>
15#include <linux/mm.h>
16#include <linux/smp.h>
17#include <linux/stddef.h>
18#include <linux/unistd.h>
19#include <linux/ptrace.h>
20#include <linux/user.h>
21#include <linux/time.h>
22#include <linux/major.h>
23#include <linux/stat.h>
24#include <linux/vt.h>
25#include <linux/mman.h>
26#include <linux/elfcore.h>
27#include <linux/reboot.h>
28#include <linux/tty.h>
29#include <linux/console.h>
30#include <linux/slab.h>
31#include <linux/rcupdate.h>
32
33#include <asm/reg.h>
34#include <asm/uaccess.h>
35#include <asm/io.h>
36#include <asm/pgtable.h>
37#include <asm/hwrpb.h>
38#include <asm/fpu.h>
39
40#include "proto.h"
41#include "pci_impl.h"
42
43/*
44 * Power off function, if any
45 */
46void (*pm_power_off)(void) = machine_power_off;
47EXPORT_SYMBOL(pm_power_off);
48
49void
50cpu_idle(void)
51{
52	set_thread_flag(TIF_POLLING_NRFLAG);
53
54	while (1) {
55		/* FIXME -- EV6 and LCA45 know how to power down
56		   the CPU.  */
57
58		rcu_idle_enter();
59		while (!need_resched())
60			cpu_relax();
61
62		rcu_idle_exit();
63		schedule_preempt_disabled();
64	}
65}
66
67
68struct halt_info {
69	int mode;
70	char *restart_cmd;
71};
72
73static void
74common_shutdown_1(void *generic_ptr)
75{
76	struct halt_info *how = (struct halt_info *)generic_ptr;
77	struct percpu_struct *cpup;
78	unsigned long *pflags, flags;
79	int cpuid = smp_processor_id();
80
81	/* No point in taking interrupts anymore. */
82	local_irq_disable();
83
84	cpup = (struct percpu_struct *)
85			((unsigned long)hwrpb + hwrpb->processor_offset
86			 + hwrpb->processor_size * cpuid);
87	pflags = &cpup->flags;
88	flags = *pflags;
89
90	/* Clear reason to "default"; clear "bootstrap in progress". */
91	flags &= ~0x00ff0001UL;
92
93#ifdef CONFIG_SMP
94	/* Secondaries halt here. */
95	if (cpuid != boot_cpuid) {
96		flags |= 0x00040000UL; /* "remain halted" */
97		*pflags = flags;
98		set_cpu_present(cpuid, false);
99		set_cpu_possible(cpuid, false);
100		halt();
101	}
102#endif
103
104	if (how->mode == LINUX_REBOOT_CMD_RESTART) {
105		if (!how->restart_cmd) {
106			flags |= 0x00020000UL; /* "cold bootstrap" */
107		} else {
108			/* For SRM, we could probably set environment
109			   variables to get this to work.  We'd have to
110			   delay this until after srm_paging_stop unless
111			   we ever got srm_fixup working.
112
113			   At the moment, SRM will use the last boot device,
114			   but the file and flags will be the defaults, when
115			   doing a "warm" bootstrap.  */
116			flags |= 0x00030000UL; /* "warm bootstrap" */
117		}
118	} else {
119		flags |= 0x00040000UL; /* "remain halted" */
120	}
121	*pflags = flags;
122
123#ifdef CONFIG_SMP
124	/* Wait for the secondaries to halt. */
125	set_cpu_present(boot_cpuid, false);
126	set_cpu_possible(boot_cpuid, false);
127	while (cpumask_weight(cpu_present_mask))
128		barrier();
129#endif
130
131	/* If booted from SRM, reset some of the original environment. */
132	if (alpha_using_srm) {
133#ifdef CONFIG_DUMMY_CONSOLE
134		/* If we've gotten here after SysRq-b, leave interrupt
135		   context before taking over the console. */
136		if (in_interrupt())
137			irq_exit();
138		/* This has the effect of resetting the VGA video origin.  */
139		take_over_console(&dummy_con, 0, MAX_NR_CONSOLES-1, 1);
140#endif
141		pci_restore_srm_config();
142		set_hae(srm_hae);
143	}
144
145	if (alpha_mv.kill_arch)
146		alpha_mv.kill_arch(how->mode);
147
148	if (! alpha_using_srm && how->mode != LINUX_REBOOT_CMD_RESTART) {
149		/* Unfortunately, since MILO doesn't currently understand
150		   the hwrpb bits above, we can't reliably halt the
151		   processor and keep it halted.  So just loop.  */
152		return;
153	}
154
155	if (alpha_using_srm)
156		srm_paging_stop();
157
158	halt();
159}
160
161static void
162common_shutdown(int mode, char *restart_cmd)
163{
164	struct halt_info args;
165	args.mode = mode;
166	args.restart_cmd = restart_cmd;
167	on_each_cpu(common_shutdown_1, &args, 0);
168}
169
170void
171machine_restart(char *restart_cmd)
172{
173	common_shutdown(LINUX_REBOOT_CMD_RESTART, restart_cmd);
174}
175
176
177void
178machine_halt(void)
179{
180	common_shutdown(LINUX_REBOOT_CMD_HALT, NULL);
181}
182
183
184void
185machine_power_off(void)
186{
187	common_shutdown(LINUX_REBOOT_CMD_POWER_OFF, NULL);
188}
189
190
191/* Used by sysrq-p, among others.  I don't believe r9-r15 are ever
192   saved in the context it's used.  */
193
194void
195show_regs(struct pt_regs *regs)
196{
197	dik_show_regs(regs, NULL);
198}
199
200/*
201 * Re-start a thread when doing execve()
202 */
203void
204start_thread(struct pt_regs * regs, unsigned long pc, unsigned long sp)
205{
206	regs->pc = pc;
207	regs->ps = 8;
208	wrusp(sp);
209}
210EXPORT_SYMBOL(start_thread);
211
212/*
213 * Free current thread data structures etc..
214 */
215void
216exit_thread(void)
217{
218}
219
220void
221flush_thread(void)
222{
223	/* Arrange for each exec'ed process to start off with a clean slate
224	   with respect to the FPU.  This is all exceptions disabled.  */
225	current_thread_info()->ieee_state = 0;
226	wrfpcr(FPCR_DYN_NORMAL | ieee_swcr_to_fpcr(0));
227
228	/* Clean slate for TLS.  */
229	current_thread_info()->pcb.unique = 0;
230}
231
232void
233release_thread(struct task_struct *dead_task)
234{
235}
236
237/*
238 * "alpha_clone()".. By the time we get here, the
239 * non-volatile registers have also been saved on the
240 * stack. We do some ugly pointer stuff here.. (see
241 * also copy_thread)
242 *
243 * Notice that "fork()" is implemented in terms of clone,
244 * with parameters (SIGCHLD, 0).
245 */
246int
247alpha_clone(unsigned long clone_flags, unsigned long usp,
248	    int __user *parent_tid, int __user *child_tid,
249	    unsigned long tls_value, struct pt_regs *regs)
250{
251	if (!usp)
252		usp = rdusp();
253
254	return do_fork(clone_flags, usp, regs, 0, parent_tid, child_tid);
255}
256
257int
258alpha_vfork(struct pt_regs *regs)
259{
260	return do_fork(CLONE_VFORK | CLONE_VM | SIGCHLD, rdusp(),
261		       regs, 0, NULL, NULL);
262}
263
264/*
265 * Copy an alpha thread..
266 *
267 * Note the "stack_offset" stuff: when returning to kernel mode, we need
268 * to have some extra stack-space for the kernel stack that still exists
269 * after the "ret_from_fork".  When returning to user mode, we only want
270 * the space needed by the syscall stack frame (ie "struct pt_regs").
271 * Use the passed "regs" pointer to determine how much space we need
272 * for a kernel fork().
273 */
274
275int
276copy_thread(unsigned long clone_flags, unsigned long usp,
277	    unsigned long unused,
278	    struct task_struct * p, struct pt_regs * regs)
279{
280	extern void ret_from_fork(void);
281
282	struct thread_info *childti = task_thread_info(p);
283	struct pt_regs * childregs;
284	struct switch_stack * childstack, *stack;
285	unsigned long stack_offset, settls;
286
287	stack_offset = PAGE_SIZE - sizeof(struct pt_regs);
288	if (!(regs->ps & 8))
289		stack_offset = (PAGE_SIZE-1) & (unsigned long) regs;
290	childregs = (struct pt_regs *)
291	  (stack_offset + PAGE_SIZE + task_stack_page(p));
292
293	*childregs = *regs;
294	settls = regs->r20;
295	childregs->r0 = 0;
296	childregs->r19 = 0;
297	childregs->r20 = 1;	/* OSF/1 has some strange fork() semantics.  */
298	regs->r20 = 0;
299	stack = ((struct switch_stack *) regs) - 1;
300	childstack = ((struct switch_stack *) childregs) - 1;
301	*childstack = *stack;
302	childstack->r26 = (unsigned long) ret_from_fork;
303	childti->pcb.usp = usp;
304	childti->pcb.ksp = (unsigned long) childstack;
305	childti->pcb.flags = 1;	/* set FEN, clear everything else */
306
307	/* Set a new TLS for the child thread?  Peek back into the
308	   syscall arguments that we saved on syscall entry.  Oops,
309	   except we'd have clobbered it with the parent/child set
310	   of r20.  Read the saved copy.  */
311	/* Note: if CLONE_SETTLS is not set, then we must inherit the
312	   value from the parent, which will have been set by the block
313	   copy in dup_task_struct.  This is non-intuitive, but is
314	   required for proper operation in the case of a threaded
315	   application calling fork.  */
316	if (clone_flags & CLONE_SETTLS)
317		childti->pcb.unique = settls;
318
319	return 0;
320}
321
322/*
323 * Fill in the user structure for a ELF core dump.
324 */
325void
326dump_elf_thread(elf_greg_t *dest, struct pt_regs *pt, struct thread_info *ti)
327{
328	/* switch stack follows right below pt_regs: */
329	struct switch_stack * sw = ((struct switch_stack *) pt) - 1;
330
331	dest[ 0] = pt->r0;
332	dest[ 1] = pt->r1;
333	dest[ 2] = pt->r2;
334	dest[ 3] = pt->r3;
335	dest[ 4] = pt->r4;
336	dest[ 5] = pt->r5;
337	dest[ 6] = pt->r6;
338	dest[ 7] = pt->r7;
339	dest[ 8] = pt->r8;
340	dest[ 9] = sw->r9;
341	dest[10] = sw->r10;
342	dest[11] = sw->r11;
343	dest[12] = sw->r12;
344	dest[13] = sw->r13;
345	dest[14] = sw->r14;
346	dest[15] = sw->r15;
347	dest[16] = pt->r16;
348	dest[17] = pt->r17;
349	dest[18] = pt->r18;
350	dest[19] = pt->r19;
351	dest[20] = pt->r20;
352	dest[21] = pt->r21;
353	dest[22] = pt->r22;
354	dest[23] = pt->r23;
355	dest[24] = pt->r24;
356	dest[25] = pt->r25;
357	dest[26] = pt->r26;
358	dest[27] = pt->r27;
359	dest[28] = pt->r28;
360	dest[29] = pt->gp;
361	dest[30] = ti == current_thread_info() ? rdusp() : ti->pcb.usp;
362	dest[31] = pt->pc;
363
364	/* Once upon a time this was the PS value.  Which is stupid
365	   since that is always 8 for usermode.  Usurped for the more
366	   useful value of the thread's UNIQUE field.  */
367	dest[32] = ti->pcb.unique;
368}
369EXPORT_SYMBOL(dump_elf_thread);
370
371int
372dump_elf_task(elf_greg_t *dest, struct task_struct *task)
373{
374	dump_elf_thread(dest, task_pt_regs(task), task_thread_info(task));
375	return 1;
376}
377EXPORT_SYMBOL(dump_elf_task);
378
379int
380dump_elf_task_fp(elf_fpreg_t *dest, struct task_struct *task)
381{
382	struct switch_stack *sw = (struct switch_stack *)task_pt_regs(task) - 1;
383	memcpy(dest, sw->fp, 32 * 8);
384	return 1;
385}
386EXPORT_SYMBOL(dump_elf_task_fp);
387
388/*
389 * sys_execve() executes a new program.
390 */
391asmlinkage int
392do_sys_execve(const char __user *ufilename,
393	      const char __user *const __user *argv,
394	      const char __user *const __user *envp, struct pt_regs *regs)
395{
396	int error;
397	char *filename;
398
399	filename = getname(ufilename);
400	error = PTR_ERR(filename);
401	if (IS_ERR(filename))
402		goto out;
403	error = do_execve(filename, argv, envp, regs);
404	putname(filename);
405out:
406	return error;
407}
408
409/*
410 * Return saved PC of a blocked thread.  This assumes the frame
411 * pointer is the 6th saved long on the kernel stack and that the
412 * saved return address is the first long in the frame.  This all
413 * holds provided the thread blocked through a call to schedule() ($15
414 * is the frame pointer in schedule() and $15 is saved at offset 48 by
415 * entry.S:do_switch_stack).
416 *
417 * Under heavy swap load I've seen this lose in an ugly way.  So do
418 * some extra sanity checking on the ranges we expect these pointers
419 * to be in so that we can fail gracefully.  This is just for ps after
420 * all.  -- r~
421 */
422
423unsigned long
424thread_saved_pc(struct task_struct *t)
425{
426	unsigned long base = (unsigned long)task_stack_page(t);
427	unsigned long fp, sp = task_thread_info(t)->pcb.ksp;
428
429	if (sp > base && sp+6*8 < base + 16*1024) {
430		fp = ((unsigned long*)sp)[6];
431		if (fp > sp && fp < base + 16*1024)
432			return *(unsigned long *)fp;
433	}
434
435	return 0;
436}
437
438unsigned long
439get_wchan(struct task_struct *p)
440{
441	unsigned long schedule_frame;
442	unsigned long pc;
443	if (!p || p == current || p->state == TASK_RUNNING)
444		return 0;
445	/*
446	 * This one depends on the frame size of schedule().  Do a
447	 * "disass schedule" in gdb to find the frame size.  Also, the
448	 * code assumes that sleep_on() follows immediately after
449	 * interruptible_sleep_on() and that add_timer() follows
450	 * immediately after interruptible_sleep().  Ugly, isn't it?
451	 * Maybe adding a wchan field to task_struct would be better,
452	 * after all...
453	 */
454
455	pc = thread_saved_pc(p);
456	if (in_sched_functions(pc)) {
457		schedule_frame = ((unsigned long *)task_thread_info(p)->pcb.ksp)[6];
458		return ((unsigned long *)schedule_frame)[12];
459	}
460	return pc;
461}
462
463int kernel_execve(const char *path, const char *const argv[], const char *const envp[])
464{
465	/* Avoid the HAE being gratuitously wrong, which would cause us
466	   to do the whole turn off interrupts thing and restore it.  */
467	struct pt_regs regs = {.hae = alpha_mv.hae_cache};
468	int err = do_execve(path, argv, envp, &regs);
469	if (!err) {
470		struct pt_regs *p = current_pt_regs();
471		/* copy regs to normal position and off to userland we go... */
472		*p = regs;
473		__asm__ __volatile__ (
474			"mov	%0, $sp;"
475			"br	$31, ret_from_sys_call"
476			: : "r"(p));
477	}
478	return err;
479}
480EXPORT_SYMBOL(kernel_execve);
481