xref: /linux/arch/xtensa/kernel/process.c (revision 4ab5a5d2a4a2289c2af07accbec7170ca5671f41)
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/sched/debug.h>
21 #include <linux/sched/task.h>
22 #include <linux/sched/task_stack.h>
23 #include <linux/kernel.h>
24 #include <linux/mm.h>
25 #include <linux/smp.h>
26 #include <linux/stddef.h>
27 #include <linux/unistd.h>
28 #include <linux/ptrace.h>
29 #include <linux/elf.h>
30 #include <linux/hw_breakpoint.h>
31 #include <linux/init.h>
32 #include <linux/prctl.h>
33 #include <linux/init_task.h>
34 #include <linux/module.h>
35 #include <linux/mqueue.h>
36 #include <linux/fs.h>
37 #include <linux/slab.h>
38 #include <linux/rcupdate.h>
39 
40 #include <asm/pgtable.h>
41 #include <linux/uaccess.h>
42 #include <asm/io.h>
43 #include <asm/processor.h>
44 #include <asm/platform.h>
45 #include <asm/mmu.h>
46 #include <asm/irq.h>
47 #include <linux/atomic.h>
48 #include <asm/asm-offsets.h>
49 #include <asm/regs.h>
50 #include <asm/hw_breakpoint.h>
51 
52 extern void ret_from_fork(void);
53 extern void ret_from_kernel_thread(void);
54 
55 struct task_struct *current_set[NR_CPUS] = {&init_task, };
56 
57 void (*pm_power_off)(void) = NULL;
58 EXPORT_SYMBOL(pm_power_off);
59 
60 
61 #ifdef CONFIG_STACKPROTECTOR
62 #include <linux/stackprotector.h>
63 unsigned long __stack_chk_guard __read_mostly;
64 EXPORT_SYMBOL(__stack_chk_guard);
65 #endif
66 
67 #if XTENSA_HAVE_COPROCESSORS
68 
69 void coprocessor_release_all(struct thread_info *ti)
70 {
71 	unsigned long cpenable;
72 	int i;
73 
74 	/* Make sure we don't switch tasks during this operation. */
75 
76 	preempt_disable();
77 
78 	/* Walk through all cp owners and release it for the requested one. */
79 
80 	cpenable = ti->cpenable;
81 
82 	for (i = 0; i < XCHAL_CP_MAX; i++) {
83 		if (coprocessor_owner[i] == ti) {
84 			coprocessor_owner[i] = 0;
85 			cpenable &= ~(1 << i);
86 		}
87 	}
88 
89 	ti->cpenable = cpenable;
90 	coprocessor_clear_cpenable();
91 
92 	preempt_enable();
93 }
94 
95 void coprocessor_flush_all(struct thread_info *ti)
96 {
97 	unsigned long cpenable, old_cpenable;
98 	int i;
99 
100 	preempt_disable();
101 
102 	RSR_CPENABLE(old_cpenable);
103 	cpenable = ti->cpenable;
104 	WSR_CPENABLE(cpenable);
105 
106 	for (i = 0; i < XCHAL_CP_MAX; i++) {
107 		if ((cpenable & 1) != 0 && coprocessor_owner[i] == ti)
108 			coprocessor_flush(ti, i);
109 		cpenable >>= 1;
110 	}
111 	WSR_CPENABLE(old_cpenable);
112 
113 	preempt_enable();
114 }
115 
116 #endif
117 
118 
119 /*
120  * Powermanagement idle function, if any is provided by the platform.
121  */
122 void arch_cpu_idle(void)
123 {
124 	platform_idle();
125 }
126 
127 /*
128  * This is called when the thread calls exit().
129  */
130 void exit_thread(struct task_struct *tsk)
131 {
132 #if XTENSA_HAVE_COPROCESSORS
133 	coprocessor_release_all(task_thread_info(tsk));
134 #endif
135 }
136 
137 /*
138  * Flush thread state. This is called when a thread does an execve()
139  * Note that we flush coprocessor registers for the case execve fails.
140  */
141 void flush_thread(void)
142 {
143 #if XTENSA_HAVE_COPROCESSORS
144 	struct thread_info *ti = current_thread_info();
145 	coprocessor_flush_all(ti);
146 	coprocessor_release_all(ti);
147 #endif
148 	flush_ptrace_hw_breakpoint(current);
149 }
150 
151 /*
152  * this gets called so that we can store coprocessor state into memory and
153  * copy the current task into the new thread.
154  */
155 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
156 {
157 #if XTENSA_HAVE_COPROCESSORS
158 	coprocessor_flush_all(task_thread_info(src));
159 #endif
160 	*dst = *src;
161 	return 0;
162 }
163 
164 /*
165  * Copy thread.
166  *
167  * There are two modes in which this function is called:
168  * 1) Userspace thread creation,
169  *    regs != NULL, usp_thread_fn is userspace stack pointer.
170  *    It is expected to copy parent regs (in case CLONE_VM is not set
171  *    in the clone_flags) and set up passed usp in the childregs.
172  * 2) Kernel thread creation,
173  *    regs == NULL, usp_thread_fn is the function to run in the new thread
174  *    and thread_fn_arg is its parameter.
175  *    childregs are not used for the kernel threads.
176  *
177  * The stack layout for the new thread looks like this:
178  *
179  *	+------------------------+
180  *	|       childregs        |
181  *	+------------------------+ <- thread.sp = sp in dummy-frame
182  *	|      dummy-frame       |    (saved in dummy-frame spill-area)
183  *	+------------------------+
184  *
185  * We create a dummy frame to return to either ret_from_fork or
186  *   ret_from_kernel_thread:
187  *   a0 points to ret_from_fork/ret_from_kernel_thread (simulating a call4)
188  *   sp points to itself (thread.sp)
189  *   a2, a3 are unused for userspace threads,
190  *   a2 points to thread_fn, a3 holds thread_fn arg for kernel threads.
191  *
192  * Note: This is a pristine frame, so we don't need any spill region on top of
193  *       childregs.
194  *
195  * The fun part:  if we're keeping the same VM (i.e. cloning a thread,
196  * not an entire process), we're normally given a new usp, and we CANNOT share
197  * any live address register windows.  If we just copy those live frames over,
198  * the two threads (parent and child) will overflow the same frames onto the
199  * parent stack at different times, likely corrupting the parent stack (esp.
200  * if the parent returns from functions that called clone() and calls new
201  * ones, before the child overflows its now old copies of its parent windows).
202  * One solution is to spill windows to the parent stack, but that's fairly
203  * involved.  Much simpler to just not copy those live frames across.
204  */
205 
206 int copy_thread(unsigned long clone_flags, unsigned long usp_thread_fn,
207 		unsigned long thread_fn_arg, struct task_struct *p)
208 {
209 	struct pt_regs *childregs = task_pt_regs(p);
210 
211 #if (XTENSA_HAVE_COPROCESSORS || XTENSA_HAVE_IO_PORTS)
212 	struct thread_info *ti;
213 #endif
214 
215 	/* Create a call4 dummy-frame: a0 = 0, a1 = childregs. */
216 	SPILL_SLOT(childregs, 1) = (unsigned long)childregs;
217 	SPILL_SLOT(childregs, 0) = 0;
218 
219 	p->thread.sp = (unsigned long)childregs;
220 
221 	if (!(p->flags & PF_KTHREAD)) {
222 		struct pt_regs *regs = current_pt_regs();
223 		unsigned long usp = usp_thread_fn ?
224 			usp_thread_fn : regs->areg[1];
225 
226 		p->thread.ra = MAKE_RA_FOR_CALL(
227 				(unsigned long)ret_from_fork, 0x1);
228 
229 		/* This does not copy all the regs.
230 		 * In a bout of brilliance or madness,
231 		 * ARs beyond a0-a15 exist past the end of the struct.
232 		 */
233 		*childregs = *regs;
234 		childregs->areg[1] = usp;
235 		childregs->areg[2] = 0;
236 
237 		/* When sharing memory with the parent thread, the child
238 		   usually starts on a pristine stack, so we have to reset
239 		   windowbase, windowstart and wmask.
240 		   (Note that such a new thread is required to always create
241 		   an initial call4 frame)
242 		   The exception is vfork, where the new thread continues to
243 		   run on the parent's stack until it calls execve. This could
244 		   be a call8 or call12, which requires a legal stack frame
245 		   of the previous caller for the overflow handlers to work.
246 		   (Note that it's always legal to overflow live registers).
247 		   In this case, ensure to spill at least the stack pointer
248 		   of that frame. */
249 
250 		if (clone_flags & CLONE_VM) {
251 			/* check that caller window is live and same stack */
252 			int len = childregs->wmask & ~0xf;
253 			if (regs->areg[1] == usp && len != 0) {
254 				int callinc = (regs->areg[0] >> 30) & 3;
255 				int caller_ars = XCHAL_NUM_AREGS - callinc * 4;
256 				put_user(regs->areg[caller_ars+1],
257 					 (unsigned __user*)(usp - 12));
258 			}
259 			childregs->wmask = 1;
260 			childregs->windowstart = 1;
261 			childregs->windowbase = 0;
262 		} else {
263 			int len = childregs->wmask & ~0xf;
264 			memcpy(&childregs->areg[XCHAL_NUM_AREGS - len/4],
265 			       &regs->areg[XCHAL_NUM_AREGS - len/4], len);
266 		}
267 
268 		/* The thread pointer is passed in the '4th argument' (= a5) */
269 		if (clone_flags & CLONE_SETTLS)
270 			childregs->threadptr = childregs->areg[5];
271 	} else {
272 		p->thread.ra = MAKE_RA_FOR_CALL(
273 				(unsigned long)ret_from_kernel_thread, 1);
274 
275 		/* pass parameters to ret_from_kernel_thread:
276 		 * a2 = thread_fn, a3 = thread_fn arg
277 		 */
278 		SPILL_SLOT(childregs, 3) = thread_fn_arg;
279 		SPILL_SLOT(childregs, 2) = usp_thread_fn;
280 
281 		/* Childregs are only used when we're going to userspace
282 		 * in which case start_thread will set them up.
283 		 */
284 	}
285 
286 #if (XTENSA_HAVE_COPROCESSORS || XTENSA_HAVE_IO_PORTS)
287 	ti = task_thread_info(p);
288 	ti->cpenable = 0;
289 #endif
290 
291 	clear_ptrace_hw_breakpoint(p);
292 
293 	return 0;
294 }
295 
296 
297 /*
298  * These bracket the sleeping functions..
299  */
300 
301 unsigned long get_wchan(struct task_struct *p)
302 {
303 	unsigned long sp, pc;
304 	unsigned long stack_page = (unsigned long) task_stack_page(p);
305 	int count = 0;
306 
307 	if (!p || p == current || p->state == TASK_RUNNING)
308 		return 0;
309 
310 	sp = p->thread.sp;
311 	pc = MAKE_PC_FROM_RA(p->thread.ra, p->thread.sp);
312 
313 	do {
314 		if (sp < stack_page + sizeof(struct task_struct) ||
315 		    sp >= (stack_page + THREAD_SIZE) ||
316 		    pc == 0)
317 			return 0;
318 		if (!in_sched_functions(pc))
319 			return pc;
320 
321 		/* Stack layout: sp-4: ra, sp-3: sp' */
322 
323 		pc = MAKE_PC_FROM_RA(*(unsigned long*)sp - 4, sp);
324 		sp = *(unsigned long *)sp - 3;
325 	} while (count++ < 16);
326 	return 0;
327 }
328 
329 /*
330  * xtensa_gregset_t and 'struct pt_regs' are vastly different formats
331  * of processor registers.  Besides different ordering,
332  * xtensa_gregset_t contains non-live register information that
333  * 'struct pt_regs' does not.  Exception handling (primarily) uses
334  * 'struct pt_regs'.  Core files and ptrace use xtensa_gregset_t.
335  *
336  */
337 
338 void xtensa_elf_core_copy_regs (xtensa_gregset_t *elfregs, struct pt_regs *regs)
339 {
340 	unsigned long wb, ws, wm;
341 	int live, last;
342 
343 	wb = regs->windowbase;
344 	ws = regs->windowstart;
345 	wm = regs->wmask;
346 	ws = ((ws >> wb) | (ws << (WSBITS - wb))) & ((1 << WSBITS) - 1);
347 
348 	/* Don't leak any random bits. */
349 
350 	memset(elfregs, 0, sizeof(*elfregs));
351 
352 	/* Note:  PS.EXCM is not set while user task is running; its
353 	 * being set in regs->ps is for exception handling convenience.
354 	 */
355 
356 	elfregs->pc		= regs->pc;
357 	elfregs->ps		= (regs->ps & ~(1 << PS_EXCM_BIT));
358 	elfregs->lbeg		= regs->lbeg;
359 	elfregs->lend		= regs->lend;
360 	elfregs->lcount		= regs->lcount;
361 	elfregs->sar		= regs->sar;
362 	elfregs->windowstart	= ws;
363 
364 	live = (wm & 2) ? 4 : (wm & 4) ? 8 : (wm & 8) ? 12 : 16;
365 	last = XCHAL_NUM_AREGS - (wm >> 4) * 4;
366 	memcpy(elfregs->a, regs->areg, live * 4);
367 	memcpy(elfregs->a + last, regs->areg + last, (wm >> 4) * 16);
368 }
369 
370 int dump_fpu(void)
371 {
372 	return 0;
373 }
374