xref: /openbmc/linux/arch/xtensa/kernel/process.c (revision 05bcf503)
1 /*
2  * arch/xtensa/kernel/process.c
3  *
4  * Xtensa Processor version.
5  *
6  * This file is subject to the terms and conditions of the GNU General Public
7  * License.  See the file "COPYING" in the main directory of this archive
8  * for more details.
9  *
10  * Copyright (C) 2001 - 2005 Tensilica Inc.
11  *
12  * Joe Taylor <joe@tensilica.com, joetylr@yahoo.com>
13  * Chris Zankel <chris@zankel.net>
14  * Marc Gauthier <marc@tensilica.com, marc@alumni.uwaterloo.ca>
15  * Kevin Chea
16  */
17 
18 #include <linux/errno.h>
19 #include <linux/sched.h>
20 #include <linux/kernel.h>
21 #include <linux/mm.h>
22 #include <linux/smp.h>
23 #include <linux/stddef.h>
24 #include <linux/unistd.h>
25 #include <linux/ptrace.h>
26 #include <linux/elf.h>
27 #include <linux/init.h>
28 #include <linux/prctl.h>
29 #include <linux/init_task.h>
30 #include <linux/module.h>
31 #include <linux/mqueue.h>
32 #include <linux/fs.h>
33 #include <linux/slab.h>
34 #include <linux/rcupdate.h>
35 
36 #include <asm/pgtable.h>
37 #include <asm/uaccess.h>
38 #include <asm/io.h>
39 #include <asm/processor.h>
40 #include <asm/platform.h>
41 #include <asm/mmu.h>
42 #include <asm/irq.h>
43 #include <linux/atomic.h>
44 #include <asm/asm-offsets.h>
45 #include <asm/regs.h>
46 
47 extern void ret_from_fork(void);
48 extern void ret_from_kernel_thread(void);
49 
50 struct task_struct *current_set[NR_CPUS] = {&init_task, };
51 
52 void (*pm_power_off)(void) = NULL;
53 EXPORT_SYMBOL(pm_power_off);
54 
55 
56 #if XTENSA_HAVE_COPROCESSORS
57 
58 void coprocessor_release_all(struct thread_info *ti)
59 {
60 	unsigned long cpenable;
61 	int i;
62 
63 	/* Make sure we don't switch tasks during this operation. */
64 
65 	preempt_disable();
66 
67 	/* Walk through all cp owners and release it for the requested one. */
68 
69 	cpenable = ti->cpenable;
70 
71 	for (i = 0; i < XCHAL_CP_MAX; i++) {
72 		if (coprocessor_owner[i] == ti) {
73 			coprocessor_owner[i] = 0;
74 			cpenable &= ~(1 << i);
75 		}
76 	}
77 
78 	ti->cpenable = cpenable;
79 	coprocessor_clear_cpenable();
80 
81 	preempt_enable();
82 }
83 
84 void coprocessor_flush_all(struct thread_info *ti)
85 {
86 	unsigned long cpenable;
87 	int i;
88 
89 	preempt_disable();
90 
91 	cpenable = ti->cpenable;
92 
93 	for (i = 0; i < XCHAL_CP_MAX; i++) {
94 		if ((cpenable & 1) != 0 && coprocessor_owner[i] == ti)
95 			coprocessor_flush(ti, i);
96 		cpenable >>= 1;
97 	}
98 
99 	preempt_enable();
100 }
101 
102 #endif
103 
104 
105 /*
106  * Powermanagement idle function, if any is provided by the platform.
107  */
108 
109 void cpu_idle(void)
110 {
111   	local_irq_enable();
112 
113 	/* endless idle loop with no priority at all */
114 	while (1) {
115 		rcu_idle_enter();
116 		while (!need_resched())
117 			platform_idle();
118 		rcu_idle_exit();
119 		schedule_preempt_disabled();
120 	}
121 }
122 
123 /*
124  * This is called when the thread calls exit().
125  */
126 void exit_thread(void)
127 {
128 #if XTENSA_HAVE_COPROCESSORS
129 	coprocessor_release_all(current_thread_info());
130 #endif
131 }
132 
133 /*
134  * Flush thread state. This is called when a thread does an execve()
135  * Note that we flush coprocessor registers for the case execve fails.
136  */
137 void flush_thread(void)
138 {
139 #if XTENSA_HAVE_COPROCESSORS
140 	struct thread_info *ti = current_thread_info();
141 	coprocessor_flush_all(ti);
142 	coprocessor_release_all(ti);
143 #endif
144 }
145 
146 /*
147  * this gets called so that we can store coprocessor state into memory and
148  * copy the current task into the new thread.
149  */
150 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
151 {
152 #if XTENSA_HAVE_COPROCESSORS
153 	coprocessor_flush_all(task_thread_info(src));
154 #endif
155 	*dst = *src;
156 	return 0;
157 }
158 
159 /*
160  * Copy thread.
161  *
162  * There are two modes in which this function is called:
163  * 1) Userspace thread creation,
164  *    regs != NULL, usp_thread_fn is userspace stack pointer.
165  *    It is expected to copy parent regs (in case CLONE_VM is not set
166  *    in the clone_flags) and set up passed usp in the childregs.
167  * 2) Kernel thread creation,
168  *    regs == NULL, usp_thread_fn is the function to run in the new thread
169  *    and thread_fn_arg is its parameter.
170  *    childregs are not used for the kernel threads.
171  *
172  * The stack layout for the new thread looks like this:
173  *
174  *	+------------------------+
175  *	|       childregs        |
176  *	+------------------------+ <- thread.sp = sp in dummy-frame
177  *	|      dummy-frame       |    (saved in dummy-frame spill-area)
178  *	+------------------------+
179  *
180  * We create a dummy frame to return to either ret_from_fork or
181  *   ret_from_kernel_thread:
182  *   a0 points to ret_from_fork/ret_from_kernel_thread (simulating a call4)
183  *   sp points to itself (thread.sp)
184  *   a2, a3 are unused for userspace threads,
185  *   a2 points to thread_fn, a3 holds thread_fn arg for kernel threads.
186  *
187  * Note: This is a pristine frame, so we don't need any spill region on top of
188  *       childregs.
189  *
190  * The fun part:  if we're keeping the same VM (i.e. cloning a thread,
191  * not an entire process), we're normally given a new usp, and we CANNOT share
192  * any live address register windows.  If we just copy those live frames over,
193  * the two threads (parent and child) will overflow the same frames onto the
194  * parent stack at different times, likely corrupting the parent stack (esp.
195  * if the parent returns from functions that called clone() and calls new
196  * ones, before the child overflows its now old copies of its parent windows).
197  * One solution is to spill windows to the parent stack, but that's fairly
198  * involved.  Much simpler to just not copy those live frames across.
199  */
200 
201 int copy_thread(unsigned long clone_flags, unsigned long usp_thread_fn,
202 		unsigned long thread_fn_arg,
203 		struct task_struct *p, struct pt_regs *unused)
204 {
205 	struct pt_regs *childregs = task_pt_regs(p);
206 
207 #if (XTENSA_HAVE_COPROCESSORS || XTENSA_HAVE_IO_PORTS)
208 	struct thread_info *ti;
209 #endif
210 
211 	/* Create a call4 dummy-frame: a0 = 0, a1 = childregs. */
212 	*((int*)childregs - 3) = (unsigned long)childregs;
213 	*((int*)childregs - 4) = 0;
214 
215 	p->thread.sp = (unsigned long)childregs;
216 
217 	if (!(p->flags & PF_KTHREAD)) {
218 		struct pt_regs *regs = current_pt_regs();
219 		unsigned long usp = usp_thread_fn ?
220 			usp_thread_fn : regs->areg[1];
221 
222 		p->thread.ra = MAKE_RA_FOR_CALL(
223 				(unsigned long)ret_from_fork, 0x1);
224 
225 		/* This does not copy all the regs.
226 		 * In a bout of brilliance or madness,
227 		 * ARs beyond a0-a15 exist past the end of the struct.
228 		 */
229 		*childregs = *regs;
230 		childregs->areg[1] = usp;
231 		childregs->areg[2] = 0;
232 
233 		/* When sharing memory with the parent thread, the child
234 		   usually starts on a pristine stack, so we have to reset
235 		   windowbase, windowstart and wmask.
236 		   (Note that such a new thread is required to always create
237 		   an initial call4 frame)
238 		   The exception is vfork, where the new thread continues to
239 		   run on the parent's stack until it calls execve. This could
240 		   be a call8 or call12, which requires a legal stack frame
241 		   of the previous caller for the overflow handlers to work.
242 		   (Note that it's always legal to overflow live registers).
243 		   In this case, ensure to spill at least the stack pointer
244 		   of that frame. */
245 
246 		if (clone_flags & CLONE_VM) {
247 			/* check that caller window is live and same stack */
248 			int len = childregs->wmask & ~0xf;
249 			if (regs->areg[1] == usp && len != 0) {
250 				int callinc = (regs->areg[0] >> 30) & 3;
251 				int caller_ars = XCHAL_NUM_AREGS - callinc * 4;
252 				put_user(regs->areg[caller_ars+1],
253 					 (unsigned __user*)(usp - 12));
254 			}
255 			childregs->wmask = 1;
256 			childregs->windowstart = 1;
257 			childregs->windowbase = 0;
258 		} else {
259 			int len = childregs->wmask & ~0xf;
260 			memcpy(&childregs->areg[XCHAL_NUM_AREGS - len/4],
261 			       &regs->areg[XCHAL_NUM_AREGS - len/4], len);
262 		}
263 // FIXME: we need to set THREADPTR in thread_info...
264 		if (clone_flags & CLONE_SETTLS)
265 			childregs->areg[2] = childregs->areg[6];
266 	} else {
267 		p->thread.ra = MAKE_RA_FOR_CALL(
268 				(unsigned long)ret_from_kernel_thread, 1);
269 
270 		/* pass parameters to ret_from_kernel_thread:
271 		 * a2 = thread_fn, a3 = thread_fn arg
272 		 */
273 		*((int *)childregs - 1) = thread_fn_arg;
274 		*((int *)childregs - 2) = usp_thread_fn;
275 
276 		/* Childregs are only used when we're going to userspace
277 		 * in which case start_thread will set them up.
278 		 */
279 	}
280 
281 #if (XTENSA_HAVE_COPROCESSORS || XTENSA_HAVE_IO_PORTS)
282 	ti = task_thread_info(p);
283 	ti->cpenable = 0;
284 #endif
285 
286 	return 0;
287 }
288 
289 
290 /*
291  * These bracket the sleeping functions..
292  */
293 
294 unsigned long get_wchan(struct task_struct *p)
295 {
296 	unsigned long sp, pc;
297 	unsigned long stack_page = (unsigned long) task_stack_page(p);
298 	int count = 0;
299 
300 	if (!p || p == current || p->state == TASK_RUNNING)
301 		return 0;
302 
303 	sp = p->thread.sp;
304 	pc = MAKE_PC_FROM_RA(p->thread.ra, p->thread.sp);
305 
306 	do {
307 		if (sp < stack_page + sizeof(struct task_struct) ||
308 		    sp >= (stack_page + THREAD_SIZE) ||
309 		    pc == 0)
310 			return 0;
311 		if (!in_sched_functions(pc))
312 			return pc;
313 
314 		/* Stack layout: sp-4: ra, sp-3: sp' */
315 
316 		pc = MAKE_PC_FROM_RA(*(unsigned long*)sp - 4, sp);
317 		sp = *(unsigned long *)sp - 3;
318 	} while (count++ < 16);
319 	return 0;
320 }
321 
322 /*
323  * xtensa_gregset_t and 'struct pt_regs' are vastly different formats
324  * of processor registers.  Besides different ordering,
325  * xtensa_gregset_t contains non-live register information that
326  * 'struct pt_regs' does not.  Exception handling (primarily) uses
327  * 'struct pt_regs'.  Core files and ptrace use xtensa_gregset_t.
328  *
329  */
330 
331 void xtensa_elf_core_copy_regs (xtensa_gregset_t *elfregs, struct pt_regs *regs)
332 {
333 	unsigned long wb, ws, wm;
334 	int live, last;
335 
336 	wb = regs->windowbase;
337 	ws = regs->windowstart;
338 	wm = regs->wmask;
339 	ws = ((ws >> wb) | (ws << (WSBITS - wb))) & ((1 << WSBITS) - 1);
340 
341 	/* Don't leak any random bits. */
342 
343 	memset(elfregs, 0, sizeof(*elfregs));
344 
345 	/* Note:  PS.EXCM is not set while user task is running; its
346 	 * being set in regs->ps is for exception handling convenience.
347 	 */
348 
349 	elfregs->pc		= regs->pc;
350 	elfregs->ps		= (regs->ps & ~(1 << PS_EXCM_BIT));
351 	elfregs->lbeg		= regs->lbeg;
352 	elfregs->lend		= regs->lend;
353 	elfregs->lcount		= regs->lcount;
354 	elfregs->sar		= regs->sar;
355 	elfregs->windowstart	= ws;
356 
357 	live = (wm & 2) ? 4 : (wm & 4) ? 8 : (wm & 8) ? 12 : 16;
358 	last = XCHAL_NUM_AREGS - (wm >> 4) * 4;
359 	memcpy(elfregs->a, regs->areg, live * 4);
360 	memcpy(elfregs->a + last, regs->areg + last, (wm >> 4) * 16);
361 }
362 
363 int dump_fpu(void)
364 {
365 	return 0;
366 }
367 
368 asmlinkage
369 long xtensa_clone(unsigned long clone_flags, unsigned long newsp,
370                   void __user *parent_tid, void *child_tls,
371                   void __user *child_tid, long a5,
372                   struct pt_regs *regs)
373 {
374         return do_fork(clone_flags, newsp, regs, 0, parent_tid, child_tid);
375 }
376