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
49#ifdef CONFIG_ALPHA_WTINT
50/*
51 * Sleep the CPU.
52 * EV6, LCA45 and QEMU know how to power down, skipping N timer interrupts.
53 */
54void arch_cpu_idle(void)
55{
56	wtint(0);
57	local_irq_enable();
58}
59
60void arch_cpu_idle_dead(void)
61{
62	wtint(INT_MAX);
63}
64#endif /* ALPHA_WTINT */
65
66struct halt_info {
67	int mode;
68	char *restart_cmd;
69};
70
71static void
72common_shutdown_1(void *generic_ptr)
73{
74	struct halt_info *how = (struct halt_info *)generic_ptr;
75	struct percpu_struct *cpup;
76	unsigned long *pflags, flags;
77	int cpuid = smp_processor_id();
78
79	/* No point in taking interrupts anymore. */
80	local_irq_disable();
81
82	cpup = (struct percpu_struct *)
83			((unsigned long)hwrpb + hwrpb->processor_offset
84			 + hwrpb->processor_size * cpuid);
85	pflags = &cpup->flags;
86	flags = *pflags;
87
88	/* Clear reason to "default"; clear "bootstrap in progress". */
89	flags &= ~0x00ff0001UL;
90
91#ifdef CONFIG_SMP
92	/* Secondaries halt here. */
93	if (cpuid != boot_cpuid) {
94		flags |= 0x00040000UL; /* "remain halted" */
95		*pflags = flags;
96		set_cpu_present(cpuid, false);
97		set_cpu_possible(cpuid, false);
98		halt();
99	}
100#endif
101
102	if (how->mode == LINUX_REBOOT_CMD_RESTART) {
103		if (!how->restart_cmd) {
104			flags |= 0x00020000UL; /* "cold bootstrap" */
105		} else {
106			/* For SRM, we could probably set environment
107			   variables to get this to work.  We'd have to
108			   delay this until after srm_paging_stop unless
109			   we ever got srm_fixup working.
110
111			   At the moment, SRM will use the last boot device,
112			   but the file and flags will be the defaults, when
113			   doing a "warm" bootstrap.  */
114			flags |= 0x00030000UL; /* "warm bootstrap" */
115		}
116	} else {
117		flags |= 0x00040000UL; /* "remain halted" */
118	}
119	*pflags = flags;
120
121#ifdef CONFIG_SMP
122	/* Wait for the secondaries to halt. */
123	set_cpu_present(boot_cpuid, false);
124	set_cpu_possible(boot_cpuid, false);
125	while (cpumask_weight(cpu_present_mask))
126		barrier();
127#endif
128
129	/* If booted from SRM, reset some of the original environment. */
130	if (alpha_using_srm) {
131#ifdef CONFIG_DUMMY_CONSOLE
132		/* If we've gotten here after SysRq-b, leave interrupt
133		   context before taking over the console. */
134		if (in_interrupt())
135			irq_exit();
136		/* This has the effect of resetting the VGA video origin.  */
137		console_lock();
138		do_take_over_console(&dummy_con, 0, MAX_NR_CONSOLES-1, 1);
139		console_unlock();
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	show_regs_print_info(KERN_DEFAULT);
198	dik_show_regs(regs, NULL);
199}
200
201/*
202 * Re-start a thread when doing execve()
203 */
204void
205start_thread(struct pt_regs * regs, unsigned long pc, unsigned long sp)
206{
207	regs->pc = pc;
208	regs->ps = 8;
209	wrusp(sp);
210}
211EXPORT_SYMBOL(start_thread);
212
213/*
214 * Free current thread data structures etc..
215 */
216void
217exit_thread(void)
218{
219}
220
221void
222flush_thread(void)
223{
224	/* Arrange for each exec'ed process to start off with a clean slate
225	   with respect to the FPU.  This is all exceptions disabled.  */
226	current_thread_info()->ieee_state = 0;
227	wrfpcr(FPCR_DYN_NORMAL | ieee_swcr_to_fpcr(0));
228
229	/* Clean slate for TLS.  */
230	current_thread_info()->pcb.unique = 0;
231}
232
233void
234release_thread(struct task_struct *dead_task)
235{
236}
237
238/*
239 * Copy architecture-specific thread state
240 */
241int
242copy_thread(unsigned long clone_flags, unsigned long usp,
243	    unsigned long kthread_arg,
244	    struct task_struct *p)
245{
246	extern void ret_from_fork(void);
247	extern void ret_from_kernel_thread(void);
248
249	struct thread_info *childti = task_thread_info(p);
250	struct pt_regs *childregs = task_pt_regs(p);
251	struct pt_regs *regs = current_pt_regs();
252	struct switch_stack *childstack, *stack;
253
254	childstack = ((struct switch_stack *) childregs) - 1;
255	childti->pcb.ksp = (unsigned long) childstack;
256	childti->pcb.flags = 1;	/* set FEN, clear everything else */
257
258	if (unlikely(p->flags & PF_KTHREAD)) {
259		/* kernel thread */
260		memset(childstack, 0,
261			sizeof(struct switch_stack) + sizeof(struct pt_regs));
262		childstack->r26 = (unsigned long) ret_from_kernel_thread;
263		childstack->r9 = usp;	/* function */
264		childstack->r10 = kthread_arg;
265		childregs->hae = alpha_mv.hae_cache,
266		childti->pcb.usp = 0;
267		return 0;
268	}
269	/* Note: if CLONE_SETTLS is not set, then we must inherit the
270	   value from the parent, which will have been set by the block
271	   copy in dup_task_struct.  This is non-intuitive, but is
272	   required for proper operation in the case of a threaded
273	   application calling fork.  */
274	if (clone_flags & CLONE_SETTLS)
275		childti->pcb.unique = regs->r20;
276	childti->pcb.usp = usp ?: rdusp();
277	*childregs = *regs;
278	childregs->r0 = 0;
279	childregs->r19 = 0;
280	childregs->r20 = 1;	/* OSF/1 has some strange fork() semantics.  */
281	regs->r20 = 0;
282	stack = ((struct switch_stack *) regs) - 1;
283	*childstack = *stack;
284	childstack->r26 = (unsigned long) ret_from_fork;
285	return 0;
286}
287
288/*
289 * Fill in the user structure for a ELF core dump.
290 */
291void
292dump_elf_thread(elf_greg_t *dest, struct pt_regs *pt, struct thread_info *ti)
293{
294	/* switch stack follows right below pt_regs: */
295	struct switch_stack * sw = ((struct switch_stack *) pt) - 1;
296
297	dest[ 0] = pt->r0;
298	dest[ 1] = pt->r1;
299	dest[ 2] = pt->r2;
300	dest[ 3] = pt->r3;
301	dest[ 4] = pt->r4;
302	dest[ 5] = pt->r5;
303	dest[ 6] = pt->r6;
304	dest[ 7] = pt->r7;
305	dest[ 8] = pt->r8;
306	dest[ 9] = sw->r9;
307	dest[10] = sw->r10;
308	dest[11] = sw->r11;
309	dest[12] = sw->r12;
310	dest[13] = sw->r13;
311	dest[14] = sw->r14;
312	dest[15] = sw->r15;
313	dest[16] = pt->r16;
314	dest[17] = pt->r17;
315	dest[18] = pt->r18;
316	dest[19] = pt->r19;
317	dest[20] = pt->r20;
318	dest[21] = pt->r21;
319	dest[22] = pt->r22;
320	dest[23] = pt->r23;
321	dest[24] = pt->r24;
322	dest[25] = pt->r25;
323	dest[26] = pt->r26;
324	dest[27] = pt->r27;
325	dest[28] = pt->r28;
326	dest[29] = pt->gp;
327	dest[30] = ti == current_thread_info() ? rdusp() : ti->pcb.usp;
328	dest[31] = pt->pc;
329
330	/* Once upon a time this was the PS value.  Which is stupid
331	   since that is always 8 for usermode.  Usurped for the more
332	   useful value of the thread's UNIQUE field.  */
333	dest[32] = ti->pcb.unique;
334}
335EXPORT_SYMBOL(dump_elf_thread);
336
337int
338dump_elf_task(elf_greg_t *dest, struct task_struct *task)
339{
340	dump_elf_thread(dest, task_pt_regs(task), task_thread_info(task));
341	return 1;
342}
343EXPORT_SYMBOL(dump_elf_task);
344
345int
346dump_elf_task_fp(elf_fpreg_t *dest, struct task_struct *task)
347{
348	struct switch_stack *sw = (struct switch_stack *)task_pt_regs(task) - 1;
349	memcpy(dest, sw->fp, 32 * 8);
350	return 1;
351}
352EXPORT_SYMBOL(dump_elf_task_fp);
353
354/*
355 * Return saved PC of a blocked thread.  This assumes the frame
356 * pointer is the 6th saved long on the kernel stack and that the
357 * saved return address is the first long in the frame.  This all
358 * holds provided the thread blocked through a call to schedule() ($15
359 * is the frame pointer in schedule() and $15 is saved at offset 48 by
360 * entry.S:do_switch_stack).
361 *
362 * Under heavy swap load I've seen this lose in an ugly way.  So do
363 * some extra sanity checking on the ranges we expect these pointers
364 * to be in so that we can fail gracefully.  This is just for ps after
365 * all.  -- r~
366 */
367
368unsigned long
369thread_saved_pc(struct task_struct *t)
370{
371	unsigned long base = (unsigned long)task_stack_page(t);
372	unsigned long fp, sp = task_thread_info(t)->pcb.ksp;
373
374	if (sp > base && sp+6*8 < base + 16*1024) {
375		fp = ((unsigned long*)sp)[6];
376		if (fp > sp && fp < base + 16*1024)
377			return *(unsigned long *)fp;
378	}
379
380	return 0;
381}
382
383unsigned long
384get_wchan(struct task_struct *p)
385{
386	unsigned long schedule_frame;
387	unsigned long pc;
388	if (!p || p == current || p->state == TASK_RUNNING)
389		return 0;
390	/*
391	 * This one depends on the frame size of schedule().  Do a
392	 * "disass schedule" in gdb to find the frame size.  Also, the
393	 * code assumes that sleep_on() follows immediately after
394	 * interruptible_sleep_on() and that add_timer() follows
395	 * immediately after interruptible_sleep().  Ugly, isn't it?
396	 * Maybe adding a wchan field to task_struct would be better,
397	 * after all...
398	 */
399
400	pc = thread_saved_pc(p);
401	if (in_sched_functions(pc)) {
402		schedule_frame = ((unsigned long *)task_thread_info(p)->pcb.ksp)[6];
403		return ((unsigned long *)schedule_frame)[12];
404	}
405	return pc;
406}
407