xref: /openbmc/linux/arch/xtensa/kernel/process.c (revision 0edbfea5)
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/hw_breakpoint.h>
28 #include <linux/init.h>
29 #include <linux/prctl.h>
30 #include <linux/init_task.h>
31 #include <linux/module.h>
32 #include <linux/mqueue.h>
33 #include <linux/fs.h>
34 #include <linux/slab.h>
35 #include <linux/rcupdate.h>
36 
37 #include <asm/pgtable.h>
38 #include <asm/uaccess.h>
39 #include <asm/io.h>
40 #include <asm/processor.h>
41 #include <asm/platform.h>
42 #include <asm/mmu.h>
43 #include <asm/irq.h>
44 #include <linux/atomic.h>
45 #include <asm/asm-offsets.h>
46 #include <asm/regs.h>
47 #include <asm/hw_breakpoint.h>
48 
49 extern void ret_from_fork(void);
50 extern void ret_from_kernel_thread(void);
51 
52 struct task_struct *current_set[NR_CPUS] = {&init_task, };
53 
54 void (*pm_power_off)(void) = NULL;
55 EXPORT_SYMBOL(pm_power_off);
56 
57 
58 #if XTENSA_HAVE_COPROCESSORS
59 
60 void coprocessor_release_all(struct thread_info *ti)
61 {
62 	unsigned long cpenable;
63 	int i;
64 
65 	/* Make sure we don't switch tasks during this operation. */
66 
67 	preempt_disable();
68 
69 	/* Walk through all cp owners and release it for the requested one. */
70 
71 	cpenable = ti->cpenable;
72 
73 	for (i = 0; i < XCHAL_CP_MAX; i++) {
74 		if (coprocessor_owner[i] == ti) {
75 			coprocessor_owner[i] = 0;
76 			cpenable &= ~(1 << i);
77 		}
78 	}
79 
80 	ti->cpenable = cpenable;
81 	coprocessor_clear_cpenable();
82 
83 	preempt_enable();
84 }
85 
86 void coprocessor_flush_all(struct thread_info *ti)
87 {
88 	unsigned long cpenable;
89 	int i;
90 
91 	preempt_disable();
92 
93 	cpenable = ti->cpenable;
94 
95 	for (i = 0; i < XCHAL_CP_MAX; i++) {
96 		if ((cpenable & 1) != 0 && coprocessor_owner[i] == ti)
97 			coprocessor_flush(ti, i);
98 		cpenable >>= 1;
99 	}
100 
101 	preempt_enable();
102 }
103 
104 #endif
105 
106 
107 /*
108  * Powermanagement idle function, if any is provided by the platform.
109  */
110 void arch_cpu_idle(void)
111 {
112 	platform_idle();
113 }
114 
115 /*
116  * This is called when the thread calls exit().
117  */
118 void exit_thread(struct task_struct *tsk)
119 {
120 #if XTENSA_HAVE_COPROCESSORS
121 	coprocessor_release_all(task_thread_info(tsk));
122 #endif
123 }
124 
125 /*
126  * Flush thread state. This is called when a thread does an execve()
127  * Note that we flush coprocessor registers for the case execve fails.
128  */
129 void flush_thread(void)
130 {
131 #if XTENSA_HAVE_COPROCESSORS
132 	struct thread_info *ti = current_thread_info();
133 	coprocessor_flush_all(ti);
134 	coprocessor_release_all(ti);
135 #endif
136 	flush_ptrace_hw_breakpoint(current);
137 }
138 
139 /*
140  * this gets called so that we can store coprocessor state into memory and
141  * copy the current task into the new thread.
142  */
143 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
144 {
145 #if XTENSA_HAVE_COPROCESSORS
146 	coprocessor_flush_all(task_thread_info(src));
147 #endif
148 	*dst = *src;
149 	return 0;
150 }
151 
152 /*
153  * Copy thread.
154  *
155  * There are two modes in which this function is called:
156  * 1) Userspace thread creation,
157  *    regs != NULL, usp_thread_fn is userspace stack pointer.
158  *    It is expected to copy parent regs (in case CLONE_VM is not set
159  *    in the clone_flags) and set up passed usp in the childregs.
160  * 2) Kernel thread creation,
161  *    regs == NULL, usp_thread_fn is the function to run in the new thread
162  *    and thread_fn_arg is its parameter.
163  *    childregs are not used for the kernel threads.
164  *
165  * The stack layout for the new thread looks like this:
166  *
167  *	+------------------------+
168  *	|       childregs        |
169  *	+------------------------+ <- thread.sp = sp in dummy-frame
170  *	|      dummy-frame       |    (saved in dummy-frame spill-area)
171  *	+------------------------+
172  *
173  * We create a dummy frame to return to either ret_from_fork or
174  *   ret_from_kernel_thread:
175  *   a0 points to ret_from_fork/ret_from_kernel_thread (simulating a call4)
176  *   sp points to itself (thread.sp)
177  *   a2, a3 are unused for userspace threads,
178  *   a2 points to thread_fn, a3 holds thread_fn arg for kernel threads.
179  *
180  * Note: This is a pristine frame, so we don't need any spill region on top of
181  *       childregs.
182  *
183  * The fun part:  if we're keeping the same VM (i.e. cloning a thread,
184  * not an entire process), we're normally given a new usp, and we CANNOT share
185  * any live address register windows.  If we just copy those live frames over,
186  * the two threads (parent and child) will overflow the same frames onto the
187  * parent stack at different times, likely corrupting the parent stack (esp.
188  * if the parent returns from functions that called clone() and calls new
189  * ones, before the child overflows its now old copies of its parent windows).
190  * One solution is to spill windows to the parent stack, but that's fairly
191  * involved.  Much simpler to just not copy those live frames across.
192  */
193 
194 int copy_thread(unsigned long clone_flags, unsigned long usp_thread_fn,
195 		unsigned long thread_fn_arg, struct task_struct *p)
196 {
197 	struct pt_regs *childregs = task_pt_regs(p);
198 
199 #if (XTENSA_HAVE_COPROCESSORS || XTENSA_HAVE_IO_PORTS)
200 	struct thread_info *ti;
201 #endif
202 
203 	/* Create a call4 dummy-frame: a0 = 0, a1 = childregs. */
204 	*((int*)childregs - 3) = (unsigned long)childregs;
205 	*((int*)childregs - 4) = 0;
206 
207 	p->thread.sp = (unsigned long)childregs;
208 
209 	if (!(p->flags & PF_KTHREAD)) {
210 		struct pt_regs *regs = current_pt_regs();
211 		unsigned long usp = usp_thread_fn ?
212 			usp_thread_fn : regs->areg[1];
213 
214 		p->thread.ra = MAKE_RA_FOR_CALL(
215 				(unsigned long)ret_from_fork, 0x1);
216 
217 		/* This does not copy all the regs.
218 		 * In a bout of brilliance or madness,
219 		 * ARs beyond a0-a15 exist past the end of the struct.
220 		 */
221 		*childregs = *regs;
222 		childregs->areg[1] = usp;
223 		childregs->areg[2] = 0;
224 
225 		/* When sharing memory with the parent thread, the child
226 		   usually starts on a pristine stack, so we have to reset
227 		   windowbase, windowstart and wmask.
228 		   (Note that such a new thread is required to always create
229 		   an initial call4 frame)
230 		   The exception is vfork, where the new thread continues to
231 		   run on the parent's stack until it calls execve. This could
232 		   be a call8 or call12, which requires a legal stack frame
233 		   of the previous caller for the overflow handlers to work.
234 		   (Note that it's always legal to overflow live registers).
235 		   In this case, ensure to spill at least the stack pointer
236 		   of that frame. */
237 
238 		if (clone_flags & CLONE_VM) {
239 			/* check that caller window is live and same stack */
240 			int len = childregs->wmask & ~0xf;
241 			if (regs->areg[1] == usp && len != 0) {
242 				int callinc = (regs->areg[0] >> 30) & 3;
243 				int caller_ars = XCHAL_NUM_AREGS - callinc * 4;
244 				put_user(regs->areg[caller_ars+1],
245 					 (unsigned __user*)(usp - 12));
246 			}
247 			childregs->wmask = 1;
248 			childregs->windowstart = 1;
249 			childregs->windowbase = 0;
250 		} else {
251 			int len = childregs->wmask & ~0xf;
252 			memcpy(&childregs->areg[XCHAL_NUM_AREGS - len/4],
253 			       &regs->areg[XCHAL_NUM_AREGS - len/4], len);
254 		}
255 
256 		/* The thread pointer is passed in the '4th argument' (= a5) */
257 		if (clone_flags & CLONE_SETTLS)
258 			childregs->threadptr = childregs->areg[5];
259 	} else {
260 		p->thread.ra = MAKE_RA_FOR_CALL(
261 				(unsigned long)ret_from_kernel_thread, 1);
262 
263 		/* pass parameters to ret_from_kernel_thread:
264 		 * a2 = thread_fn, a3 = thread_fn arg
265 		 */
266 		*((int *)childregs - 1) = thread_fn_arg;
267 		*((int *)childregs - 2) = usp_thread_fn;
268 
269 		/* Childregs are only used when we're going to userspace
270 		 * in which case start_thread will set them up.
271 		 */
272 	}
273 
274 #if (XTENSA_HAVE_COPROCESSORS || XTENSA_HAVE_IO_PORTS)
275 	ti = task_thread_info(p);
276 	ti->cpenable = 0;
277 #endif
278 
279 	clear_ptrace_hw_breakpoint(p);
280 
281 	return 0;
282 }
283 
284 
285 /*
286  * These bracket the sleeping functions..
287  */
288 
289 unsigned long get_wchan(struct task_struct *p)
290 {
291 	unsigned long sp, pc;
292 	unsigned long stack_page = (unsigned long) task_stack_page(p);
293 	int count = 0;
294 
295 	if (!p || p == current || p->state == TASK_RUNNING)
296 		return 0;
297 
298 	sp = p->thread.sp;
299 	pc = MAKE_PC_FROM_RA(p->thread.ra, p->thread.sp);
300 
301 	do {
302 		if (sp < stack_page + sizeof(struct task_struct) ||
303 		    sp >= (stack_page + THREAD_SIZE) ||
304 		    pc == 0)
305 			return 0;
306 		if (!in_sched_functions(pc))
307 			return pc;
308 
309 		/* Stack layout: sp-4: ra, sp-3: sp' */
310 
311 		pc = MAKE_PC_FROM_RA(*(unsigned long*)sp - 4, sp);
312 		sp = *(unsigned long *)sp - 3;
313 	} while (count++ < 16);
314 	return 0;
315 }
316 
317 /*
318  * xtensa_gregset_t and 'struct pt_regs' are vastly different formats
319  * of processor registers.  Besides different ordering,
320  * xtensa_gregset_t contains non-live register information that
321  * 'struct pt_regs' does not.  Exception handling (primarily) uses
322  * 'struct pt_regs'.  Core files and ptrace use xtensa_gregset_t.
323  *
324  */
325 
326 void xtensa_elf_core_copy_regs (xtensa_gregset_t *elfregs, struct pt_regs *regs)
327 {
328 	unsigned long wb, ws, wm;
329 	int live, last;
330 
331 	wb = regs->windowbase;
332 	ws = regs->windowstart;
333 	wm = regs->wmask;
334 	ws = ((ws >> wb) | (ws << (WSBITS - wb))) & ((1 << WSBITS) - 1);
335 
336 	/* Don't leak any random bits. */
337 
338 	memset(elfregs, 0, sizeof(*elfregs));
339 
340 	/* Note:  PS.EXCM is not set while user task is running; its
341 	 * being set in regs->ps is for exception handling convenience.
342 	 */
343 
344 	elfregs->pc		= regs->pc;
345 	elfregs->ps		= (regs->ps & ~(1 << PS_EXCM_BIT));
346 	elfregs->lbeg		= regs->lbeg;
347 	elfregs->lend		= regs->lend;
348 	elfregs->lcount		= regs->lcount;
349 	elfregs->sar		= regs->sar;
350 	elfregs->windowstart	= ws;
351 
352 	live = (wm & 2) ? 4 : (wm & 4) ? 8 : (wm & 8) ? 12 : 16;
353 	last = XCHAL_NUM_AREGS - (wm >> 4) * 4;
354 	memcpy(elfregs->a, regs->areg, live * 4);
355 	memcpy(elfregs->a + last, regs->areg + last, (wm >> 4) * 16);
356 }
357 
358 int dump_fpu(void)
359 {
360 	return 0;
361 }
362