xref: /openbmc/linux/arch/xtensa/kernel/process.c (revision b9b77222)
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;
98 	int i;
99 
100 	preempt_disable();
101 
102 	cpenable = ti->cpenable;
103 
104 	for (i = 0; i < XCHAL_CP_MAX; i++) {
105 		if ((cpenable & 1) != 0 && coprocessor_owner[i] == ti)
106 			coprocessor_flush(ti, i);
107 		cpenable >>= 1;
108 	}
109 
110 	preempt_enable();
111 }
112 
113 #endif
114 
115 
116 /*
117  * Powermanagement idle function, if any is provided by the platform.
118  */
119 void arch_cpu_idle(void)
120 {
121 	platform_idle();
122 }
123 
124 /*
125  * This is called when the thread calls exit().
126  */
127 void exit_thread(struct task_struct *tsk)
128 {
129 #if XTENSA_HAVE_COPROCESSORS
130 	coprocessor_release_all(task_thread_info(tsk));
131 #endif
132 }
133 
134 /*
135  * Flush thread state. This is called when a thread does an execve()
136  * Note that we flush coprocessor registers for the case execve fails.
137  */
138 void flush_thread(void)
139 {
140 #if XTENSA_HAVE_COPROCESSORS
141 	struct thread_info *ti = current_thread_info();
142 	coprocessor_flush_all(ti);
143 	coprocessor_release_all(ti);
144 #endif
145 	flush_ptrace_hw_breakpoint(current);
146 }
147 
148 /*
149  * this gets called so that we can store coprocessor state into memory and
150  * copy the current task into the new thread.
151  */
152 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
153 {
154 #if XTENSA_HAVE_COPROCESSORS
155 	coprocessor_flush_all(task_thread_info(src));
156 #endif
157 	*dst = *src;
158 	return 0;
159 }
160 
161 /*
162  * Copy thread.
163  *
164  * There are two modes in which this function is called:
165  * 1) Userspace thread creation,
166  *    regs != NULL, usp_thread_fn is userspace stack pointer.
167  *    It is expected to copy parent regs (in case CLONE_VM is not set
168  *    in the clone_flags) and set up passed usp in the childregs.
169  * 2) Kernel thread creation,
170  *    regs == NULL, usp_thread_fn is the function to run in the new thread
171  *    and thread_fn_arg is its parameter.
172  *    childregs are not used for the kernel threads.
173  *
174  * The stack layout for the new thread looks like this:
175  *
176  *	+------------------------+
177  *	|       childregs        |
178  *	+------------------------+ <- thread.sp = sp in dummy-frame
179  *	|      dummy-frame       |    (saved in dummy-frame spill-area)
180  *	+------------------------+
181  *
182  * We create a dummy frame to return to either ret_from_fork or
183  *   ret_from_kernel_thread:
184  *   a0 points to ret_from_fork/ret_from_kernel_thread (simulating a call4)
185  *   sp points to itself (thread.sp)
186  *   a2, a3 are unused for userspace threads,
187  *   a2 points to thread_fn, a3 holds thread_fn arg for kernel threads.
188  *
189  * Note: This is a pristine frame, so we don't need any spill region on top of
190  *       childregs.
191  *
192  * The fun part:  if we're keeping the same VM (i.e. cloning a thread,
193  * not an entire process), we're normally given a new usp, and we CANNOT share
194  * any live address register windows.  If we just copy those live frames over,
195  * the two threads (parent and child) will overflow the same frames onto the
196  * parent stack at different times, likely corrupting the parent stack (esp.
197  * if the parent returns from functions that called clone() and calls new
198  * ones, before the child overflows its now old copies of its parent windows).
199  * One solution is to spill windows to the parent stack, but that's fairly
200  * involved.  Much simpler to just not copy those live frames across.
201  */
202 
203 int copy_thread(unsigned long clone_flags, unsigned long usp_thread_fn,
204 		unsigned long thread_fn_arg, struct task_struct *p)
205 {
206 	struct pt_regs *childregs = task_pt_regs(p);
207 
208 #if (XTENSA_HAVE_COPROCESSORS || XTENSA_HAVE_IO_PORTS)
209 	struct thread_info *ti;
210 #endif
211 
212 	/* Create a call4 dummy-frame: a0 = 0, a1 = childregs. */
213 	SPILL_SLOT(childregs, 1) = (unsigned long)childregs;
214 	SPILL_SLOT(childregs, 0) = 0;
215 
216 	p->thread.sp = (unsigned long)childregs;
217 
218 	if (!(p->flags & PF_KTHREAD)) {
219 		struct pt_regs *regs = current_pt_regs();
220 		unsigned long usp = usp_thread_fn ?
221 			usp_thread_fn : regs->areg[1];
222 
223 		p->thread.ra = MAKE_RA_FOR_CALL(
224 				(unsigned long)ret_from_fork, 0x1);
225 
226 		/* This does not copy all the regs.
227 		 * In a bout of brilliance or madness,
228 		 * ARs beyond a0-a15 exist past the end of the struct.
229 		 */
230 		*childregs = *regs;
231 		childregs->areg[1] = usp;
232 		childregs->areg[2] = 0;
233 
234 		/* When sharing memory with the parent thread, the child
235 		   usually starts on a pristine stack, so we have to reset
236 		   windowbase, windowstart and wmask.
237 		   (Note that such a new thread is required to always create
238 		   an initial call4 frame)
239 		   The exception is vfork, where the new thread continues to
240 		   run on the parent's stack until it calls execve. This could
241 		   be a call8 or call12, which requires a legal stack frame
242 		   of the previous caller for the overflow handlers to work.
243 		   (Note that it's always legal to overflow live registers).
244 		   In this case, ensure to spill at least the stack pointer
245 		   of that frame. */
246 
247 		if (clone_flags & CLONE_VM) {
248 			/* check that caller window is live and same stack */
249 			int len = childregs->wmask & ~0xf;
250 			if (regs->areg[1] == usp && len != 0) {
251 				int callinc = (regs->areg[0] >> 30) & 3;
252 				int caller_ars = XCHAL_NUM_AREGS - callinc * 4;
253 				put_user(regs->areg[caller_ars+1],
254 					 (unsigned __user*)(usp - 12));
255 			}
256 			childregs->wmask = 1;
257 			childregs->windowstart = 1;
258 			childregs->windowbase = 0;
259 		} else {
260 			int len = childregs->wmask & ~0xf;
261 			memcpy(&childregs->areg[XCHAL_NUM_AREGS - len/4],
262 			       &regs->areg[XCHAL_NUM_AREGS - len/4], len);
263 		}
264 
265 		/* The thread pointer is passed in the '4th argument' (= a5) */
266 		if (clone_flags & CLONE_SETTLS)
267 			childregs->threadptr = childregs->areg[5];
268 	} else {
269 		p->thread.ra = MAKE_RA_FOR_CALL(
270 				(unsigned long)ret_from_kernel_thread, 1);
271 
272 		/* pass parameters to ret_from_kernel_thread:
273 		 * a2 = thread_fn, a3 = thread_fn arg
274 		 */
275 		SPILL_SLOT(childregs, 3) = thread_fn_arg;
276 		SPILL_SLOT(childregs, 2) = usp_thread_fn;
277 
278 		/* Childregs are only used when we're going to userspace
279 		 * in which case start_thread will set them up.
280 		 */
281 	}
282 
283 #if (XTENSA_HAVE_COPROCESSORS || XTENSA_HAVE_IO_PORTS)
284 	ti = task_thread_info(p);
285 	ti->cpenable = 0;
286 #endif
287 
288 	clear_ptrace_hw_breakpoint(p);
289 
290 	return 0;
291 }
292 
293 
294 /*
295  * These bracket the sleeping functions..
296  */
297 
298 unsigned long get_wchan(struct task_struct *p)
299 {
300 	unsigned long sp, pc;
301 	unsigned long stack_page = (unsigned long) task_stack_page(p);
302 	int count = 0;
303 
304 	if (!p || p == current || p->state == TASK_RUNNING)
305 		return 0;
306 
307 	sp = p->thread.sp;
308 	pc = MAKE_PC_FROM_RA(p->thread.ra, p->thread.sp);
309 
310 	do {
311 		if (sp < stack_page + sizeof(struct task_struct) ||
312 		    sp >= (stack_page + THREAD_SIZE) ||
313 		    pc == 0)
314 			return 0;
315 		if (!in_sched_functions(pc))
316 			return pc;
317 
318 		/* Stack layout: sp-4: ra, sp-3: sp' */
319 
320 		pc = MAKE_PC_FROM_RA(*(unsigned long*)sp - 4, sp);
321 		sp = *(unsigned long *)sp - 3;
322 	} while (count++ < 16);
323 	return 0;
324 }
325 
326 /*
327  * xtensa_gregset_t and 'struct pt_regs' are vastly different formats
328  * of processor registers.  Besides different ordering,
329  * xtensa_gregset_t contains non-live register information that
330  * 'struct pt_regs' does not.  Exception handling (primarily) uses
331  * 'struct pt_regs'.  Core files and ptrace use xtensa_gregset_t.
332  *
333  */
334 
335 void xtensa_elf_core_copy_regs (xtensa_gregset_t *elfregs, struct pt_regs *regs)
336 {
337 	unsigned long wb, ws, wm;
338 	int live, last;
339 
340 	wb = regs->windowbase;
341 	ws = regs->windowstart;
342 	wm = regs->wmask;
343 	ws = ((ws >> wb) | (ws << (WSBITS - wb))) & ((1 << WSBITS) - 1);
344 
345 	/* Don't leak any random bits. */
346 
347 	memset(elfregs, 0, sizeof(*elfregs));
348 
349 	/* Note:  PS.EXCM is not set while user task is running; its
350 	 * being set in regs->ps is for exception handling convenience.
351 	 */
352 
353 	elfregs->pc		= regs->pc;
354 	elfregs->ps		= (regs->ps & ~(1 << PS_EXCM_BIT));
355 	elfregs->lbeg		= regs->lbeg;
356 	elfregs->lend		= regs->lend;
357 	elfregs->lcount		= regs->lcount;
358 	elfregs->sar		= regs->sar;
359 	elfregs->windowstart	= ws;
360 
361 	live = (wm & 2) ? 4 : (wm & 4) ? 8 : (wm & 8) ? 12 : 16;
362 	last = XCHAL_NUM_AREGS - (wm >> 4) * 4;
363 	memcpy(elfregs->a, regs->areg, live * 4);
364 	memcpy(elfregs->a + last, regs->areg + last, (wm >> 4) * 16);
365 }
366 
367 int dump_fpu(void)
368 {
369 	return 0;
370 }
371