xref: /openbmc/linux/arch/xtensa/kernel/process.c (revision b35565bb)
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 #if XTENSA_HAVE_COPROCESSORS
62 
63 void coprocessor_release_all(struct thread_info *ti)
64 {
65 	unsigned long cpenable;
66 	int i;
67 
68 	/* Make sure we don't switch tasks during this operation. */
69 
70 	preempt_disable();
71 
72 	/* Walk through all cp owners and release it for the requested one. */
73 
74 	cpenable = ti->cpenable;
75 
76 	for (i = 0; i < XCHAL_CP_MAX; i++) {
77 		if (coprocessor_owner[i] == ti) {
78 			coprocessor_owner[i] = 0;
79 			cpenable &= ~(1 << i);
80 		}
81 	}
82 
83 	ti->cpenable = cpenable;
84 	coprocessor_clear_cpenable();
85 
86 	preempt_enable();
87 }
88 
89 void coprocessor_flush_all(struct thread_info *ti)
90 {
91 	unsigned long cpenable;
92 	int i;
93 
94 	preempt_disable();
95 
96 	cpenable = ti->cpenable;
97 
98 	for (i = 0; i < XCHAL_CP_MAX; i++) {
99 		if ((cpenable & 1) != 0 && coprocessor_owner[i] == ti)
100 			coprocessor_flush(ti, i);
101 		cpenable >>= 1;
102 	}
103 
104 	preempt_enable();
105 }
106 
107 #endif
108 
109 
110 /*
111  * Powermanagement idle function, if any is provided by the platform.
112  */
113 void arch_cpu_idle(void)
114 {
115 	platform_idle();
116 }
117 
118 /*
119  * This is called when the thread calls exit().
120  */
121 void exit_thread(struct task_struct *tsk)
122 {
123 #if XTENSA_HAVE_COPROCESSORS
124 	coprocessor_release_all(task_thread_info(tsk));
125 #endif
126 }
127 
128 /*
129  * Flush thread state. This is called when a thread does an execve()
130  * Note that we flush coprocessor registers for the case execve fails.
131  */
132 void flush_thread(void)
133 {
134 #if XTENSA_HAVE_COPROCESSORS
135 	struct thread_info *ti = current_thread_info();
136 	coprocessor_flush_all(ti);
137 	coprocessor_release_all(ti);
138 #endif
139 	flush_ptrace_hw_breakpoint(current);
140 }
141 
142 /*
143  * this gets called so that we can store coprocessor state into memory and
144  * copy the current task into the new thread.
145  */
146 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
147 {
148 #if XTENSA_HAVE_COPROCESSORS
149 	coprocessor_flush_all(task_thread_info(src));
150 #endif
151 	*dst = *src;
152 	return 0;
153 }
154 
155 /*
156  * Copy thread.
157  *
158  * There are two modes in which this function is called:
159  * 1) Userspace thread creation,
160  *    regs != NULL, usp_thread_fn is userspace stack pointer.
161  *    It is expected to copy parent regs (in case CLONE_VM is not set
162  *    in the clone_flags) and set up passed usp in the childregs.
163  * 2) Kernel thread creation,
164  *    regs == NULL, usp_thread_fn is the function to run in the new thread
165  *    and thread_fn_arg is its parameter.
166  *    childregs are not used for the kernel threads.
167  *
168  * The stack layout for the new thread looks like this:
169  *
170  *	+------------------------+
171  *	|       childregs        |
172  *	+------------------------+ <- thread.sp = sp in dummy-frame
173  *	|      dummy-frame       |    (saved in dummy-frame spill-area)
174  *	+------------------------+
175  *
176  * We create a dummy frame to return to either ret_from_fork or
177  *   ret_from_kernel_thread:
178  *   a0 points to ret_from_fork/ret_from_kernel_thread (simulating a call4)
179  *   sp points to itself (thread.sp)
180  *   a2, a3 are unused for userspace threads,
181  *   a2 points to thread_fn, a3 holds thread_fn arg for kernel threads.
182  *
183  * Note: This is a pristine frame, so we don't need any spill region on top of
184  *       childregs.
185  *
186  * The fun part:  if we're keeping the same VM (i.e. cloning a thread,
187  * not an entire process), we're normally given a new usp, and we CANNOT share
188  * any live address register windows.  If we just copy those live frames over,
189  * the two threads (parent and child) will overflow the same frames onto the
190  * parent stack at different times, likely corrupting the parent stack (esp.
191  * if the parent returns from functions that called clone() and calls new
192  * ones, before the child overflows its now old copies of its parent windows).
193  * One solution is to spill windows to the parent stack, but that's fairly
194  * involved.  Much simpler to just not copy those live frames across.
195  */
196 
197 int copy_thread(unsigned long clone_flags, unsigned long usp_thread_fn,
198 		unsigned long thread_fn_arg, struct task_struct *p)
199 {
200 	struct pt_regs *childregs = task_pt_regs(p);
201 
202 #if (XTENSA_HAVE_COPROCESSORS || XTENSA_HAVE_IO_PORTS)
203 	struct thread_info *ti;
204 #endif
205 
206 	/* Create a call4 dummy-frame: a0 = 0, a1 = childregs. */
207 	SPILL_SLOT(childregs, 1) = (unsigned long)childregs;
208 	SPILL_SLOT(childregs, 0) = 0;
209 
210 	p->thread.sp = (unsigned long)childregs;
211 
212 	if (!(p->flags & PF_KTHREAD)) {
213 		struct pt_regs *regs = current_pt_regs();
214 		unsigned long usp = usp_thread_fn ?
215 			usp_thread_fn : regs->areg[1];
216 
217 		p->thread.ra = MAKE_RA_FOR_CALL(
218 				(unsigned long)ret_from_fork, 0x1);
219 
220 		/* This does not copy all the regs.
221 		 * In a bout of brilliance or madness,
222 		 * ARs beyond a0-a15 exist past the end of the struct.
223 		 */
224 		*childregs = *regs;
225 		childregs->areg[1] = usp;
226 		childregs->areg[2] = 0;
227 
228 		/* When sharing memory with the parent thread, the child
229 		   usually starts on a pristine stack, so we have to reset
230 		   windowbase, windowstart and wmask.
231 		   (Note that such a new thread is required to always create
232 		   an initial call4 frame)
233 		   The exception is vfork, where the new thread continues to
234 		   run on the parent's stack until it calls execve. This could
235 		   be a call8 or call12, which requires a legal stack frame
236 		   of the previous caller for the overflow handlers to work.
237 		   (Note that it's always legal to overflow live registers).
238 		   In this case, ensure to spill at least the stack pointer
239 		   of that frame. */
240 
241 		if (clone_flags & CLONE_VM) {
242 			/* check that caller window is live and same stack */
243 			int len = childregs->wmask & ~0xf;
244 			if (regs->areg[1] == usp && len != 0) {
245 				int callinc = (regs->areg[0] >> 30) & 3;
246 				int caller_ars = XCHAL_NUM_AREGS - callinc * 4;
247 				put_user(regs->areg[caller_ars+1],
248 					 (unsigned __user*)(usp - 12));
249 			}
250 			childregs->wmask = 1;
251 			childregs->windowstart = 1;
252 			childregs->windowbase = 0;
253 		} else {
254 			int len = childregs->wmask & ~0xf;
255 			memcpy(&childregs->areg[XCHAL_NUM_AREGS - len/4],
256 			       &regs->areg[XCHAL_NUM_AREGS - len/4], len);
257 		}
258 
259 		/* The thread pointer is passed in the '4th argument' (= a5) */
260 		if (clone_flags & CLONE_SETTLS)
261 			childregs->threadptr = childregs->areg[5];
262 	} else {
263 		p->thread.ra = MAKE_RA_FOR_CALL(
264 				(unsigned long)ret_from_kernel_thread, 1);
265 
266 		/* pass parameters to ret_from_kernel_thread:
267 		 * a2 = thread_fn, a3 = thread_fn arg
268 		 */
269 		SPILL_SLOT(childregs, 3) = thread_fn_arg;
270 		SPILL_SLOT(childregs, 2) = usp_thread_fn;
271 
272 		/* Childregs are only used when we're going to userspace
273 		 * in which case start_thread will set them up.
274 		 */
275 	}
276 
277 #if (XTENSA_HAVE_COPROCESSORS || XTENSA_HAVE_IO_PORTS)
278 	ti = task_thread_info(p);
279 	ti->cpenable = 0;
280 #endif
281 
282 	clear_ptrace_hw_breakpoint(p);
283 
284 	return 0;
285 }
286 
287 
288 /*
289  * These bracket the sleeping functions..
290  */
291 
292 unsigned long get_wchan(struct task_struct *p)
293 {
294 	unsigned long sp, pc;
295 	unsigned long stack_page = (unsigned long) task_stack_page(p);
296 	int count = 0;
297 
298 	if (!p || p == current || p->state == TASK_RUNNING)
299 		return 0;
300 
301 	sp = p->thread.sp;
302 	pc = MAKE_PC_FROM_RA(p->thread.ra, p->thread.sp);
303 
304 	do {
305 		if (sp < stack_page + sizeof(struct task_struct) ||
306 		    sp >= (stack_page + THREAD_SIZE) ||
307 		    pc == 0)
308 			return 0;
309 		if (!in_sched_functions(pc))
310 			return pc;
311 
312 		/* Stack layout: sp-4: ra, sp-3: sp' */
313 
314 		pc = MAKE_PC_FROM_RA(*(unsigned long*)sp - 4, sp);
315 		sp = *(unsigned long *)sp - 3;
316 	} while (count++ < 16);
317 	return 0;
318 }
319 
320 /*
321  * xtensa_gregset_t and 'struct pt_regs' are vastly different formats
322  * of processor registers.  Besides different ordering,
323  * xtensa_gregset_t contains non-live register information that
324  * 'struct pt_regs' does not.  Exception handling (primarily) uses
325  * 'struct pt_regs'.  Core files and ptrace use xtensa_gregset_t.
326  *
327  */
328 
329 void xtensa_elf_core_copy_regs (xtensa_gregset_t *elfregs, struct pt_regs *regs)
330 {
331 	unsigned long wb, ws, wm;
332 	int live, last;
333 
334 	wb = regs->windowbase;
335 	ws = regs->windowstart;
336 	wm = regs->wmask;
337 	ws = ((ws >> wb) | (ws << (WSBITS - wb))) & ((1 << WSBITS) - 1);
338 
339 	/* Don't leak any random bits. */
340 
341 	memset(elfregs, 0, sizeof(*elfregs));
342 
343 	/* Note:  PS.EXCM is not set while user task is running; its
344 	 * being set in regs->ps is for exception handling convenience.
345 	 */
346 
347 	elfregs->pc		= regs->pc;
348 	elfregs->ps		= (regs->ps & ~(1 << PS_EXCM_BIT));
349 	elfregs->lbeg		= regs->lbeg;
350 	elfregs->lend		= regs->lend;
351 	elfregs->lcount		= regs->lcount;
352 	elfregs->sar		= regs->sar;
353 	elfregs->windowstart	= ws;
354 
355 	live = (wm & 2) ? 4 : (wm & 4) ? 8 : (wm & 8) ? 12 : 16;
356 	last = XCHAL_NUM_AREGS - (wm >> 4) * 4;
357 	memcpy(elfregs->a, regs->areg, live * 4);
358 	memcpy(elfregs->a + last, regs->areg + last, (wm >> 4) * 16);
359 }
360 
361 int dump_fpu(void)
362 {
363 	return 0;
364 }
365