process.c

  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 <linux/uaccess.h>
 41#include <asm/io.h>
 42#include <asm/processor.h>
 43#include <asm/platform.h>
 44#include <asm/mmu.h>
 45#include <asm/irq.h>
 46#include <linux/atomic.h>
 47#include <asm/asm-offsets.h>
 48#include <asm/regs.h>
 49#include <asm/hw_breakpoint.h>
 50#include <asm/sections.h>
 51#include <asm/traps.h>
 52
 53extern void ret_from_fork(void);
 54extern void ret_from_kernel_thread(void);
 55
 56void (*pm_power_off)(void) = NULL;
 57EXPORT_SYMBOL(pm_power_off);
 58
 59
 60#ifdef CONFIG_STACKPROTECTOR
 61#include <linux/stackprotector.h>
 62unsigned long __stack_chk_guard __read_mostly;
 63EXPORT_SYMBOL(__stack_chk_guard);
 64#endif
 65
 66#if XTENSA_HAVE_COPROCESSORS
 67
 68void local_coprocessors_flush_release_all(void)
 69{
 70	struct thread_info **coprocessor_owner;
 71	struct thread_info *unique_owner[XCHAL_CP_MAX];
 72	int n = 0;
 73	int i, j;
 74
 75	coprocessor_owner = this_cpu_ptr(&exc_table)->coprocessor_owner;
 76	xtensa_set_sr(XCHAL_CP_MASK, cpenable);
 77
 78	for (i = 0; i < XCHAL_CP_MAX; i++) {
 79		struct thread_info *ti = coprocessor_owner[i];
 80
 81		if (ti) {
 82			coprocessor_flush(ti, i);
 83
 84			for (j = 0; j < n; j++)
 85				if (unique_owner[j] == ti)
 86					break;
 87			if (j == n)
 88				unique_owner[n++] = ti;
 89
 90			coprocessor_owner[i] = NULL;
 91		}
 92	}
 93	for (i = 0; i < n; i++) {
 94		/* pairs with memw (1) in fast_coprocessor and memw in switch_to */
 95		smp_wmb();
 96		unique_owner[i]->cpenable = 0;
 97	}
 98	xtensa_set_sr(0, cpenable);
 99}
100
101static void local_coprocessor_release_all(void *info)
102{
103	struct thread_info *ti = info;
104	struct thread_info **coprocessor_owner;
105	int i;
106
107	coprocessor_owner = this_cpu_ptr(&exc_table)->coprocessor_owner;
108
109	/* Walk through all cp owners and release it for the requested one. */
110
111	for (i = 0; i < XCHAL_CP_MAX; i++) {
112		if (coprocessor_owner[i] == ti)
113			coprocessor_owner[i] = NULL;
114	}
115	/* pairs with memw (1) in fast_coprocessor and memw in switch_to */
116	smp_wmb();
117	ti->cpenable = 0;
118	if (ti == current_thread_info())
119		xtensa_set_sr(0, cpenable);
120}
121
122void coprocessor_release_all(struct thread_info *ti)
123{
124	if (ti->cpenable) {
125		/* pairs with memw (2) in fast_coprocessor */
126		smp_rmb();
127		smp_call_function_single(ti->cp_owner_cpu,
128					 local_coprocessor_release_all,
129					 ti, true);
130	}
131}
132
133static void local_coprocessor_flush_all(void *info)
134{
135	struct thread_info *ti = info;
136	struct thread_info **coprocessor_owner;
137	unsigned long old_cpenable;
138	int i;
139
140	coprocessor_owner = this_cpu_ptr(&exc_table)->coprocessor_owner;
141	old_cpenable = xtensa_xsr(ti->cpenable, cpenable);
142
143	for (i = 0; i < XCHAL_CP_MAX; i++) {
144		if (coprocessor_owner[i] == ti)
145			coprocessor_flush(ti, i);
146	}
147	xtensa_set_sr(old_cpenable, cpenable);
148}
149
150void coprocessor_flush_all(struct thread_info *ti)
151{
152	if (ti->cpenable) {
153		/* pairs with memw (2) in fast_coprocessor */
154		smp_rmb();
155		smp_call_function_single(ti->cp_owner_cpu,
156					 local_coprocessor_flush_all,
157					 ti, true);
158	}
159}
160
161static void local_coprocessor_flush_release_all(void *info)
162{
163	local_coprocessor_flush_all(info);
164	local_coprocessor_release_all(info);
165}
166
167void coprocessor_flush_release_all(struct thread_info *ti)
168{
169	if (ti->cpenable) {
170		/* pairs with memw (2) in fast_coprocessor */
171		smp_rmb();
172		smp_call_function_single(ti->cp_owner_cpu,
173					 local_coprocessor_flush_release_all,
174					 ti, true);
175	}
176}
177
178#endif
179
180
181/*
182 * Powermanagement idle function, if any is provided by the platform.
183 */
184void arch_cpu_idle(void)
185{
186	platform_idle();
187	raw_local_irq_disable();
188}
189
190/*
191 * This is called when the thread calls exit().
192 */
193void exit_thread(struct task_struct *tsk)
194{
195#if XTENSA_HAVE_COPROCESSORS
196	coprocessor_release_all(task_thread_info(tsk));
197#endif
198}
199
200/*
201 * Flush thread state. This is called when a thread does an execve()
202 * Note that we flush coprocessor registers for the case execve fails.
203 */
204void flush_thread(void)
205{
206#if XTENSA_HAVE_COPROCESSORS
207	struct thread_info *ti = current_thread_info();
208	coprocessor_flush_release_all(ti);
209#endif
210	flush_ptrace_hw_breakpoint(current);
211}
212
213/*
214 * this gets called so that we can store coprocessor state into memory and
215 * copy the current task into the new thread.
216 */
217int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
218{
219#if XTENSA_HAVE_COPROCESSORS
220	coprocessor_flush_all(task_thread_info(src));
221#endif
222	*dst = *src;
223	return 0;
224}
225
226/*
227 * Copy thread.
228 *
229 * There are two modes in which this function is called:
230 * 1) Userspace thread creation,
231 *    regs != NULL, usp_thread_fn is userspace stack pointer.
232 *    It is expected to copy parent regs (in case CLONE_VM is not set
233 *    in the clone_flags) and set up passed usp in the childregs.
234 * 2) Kernel thread creation,
235 *    regs == NULL, usp_thread_fn is the function to run in the new thread
236 *    and thread_fn_arg is its parameter.
237 *    childregs are not used for the kernel threads.
238 *
239 * The stack layout for the new thread looks like this:
240 *
241 *	+------------------------+
242 *	|       childregs        |
243 *	+------------------------+ <- thread.sp = sp in dummy-frame
244 *	|      dummy-frame       |    (saved in dummy-frame spill-area)
245 *	+------------------------+
246 *
247 * We create a dummy frame to return to either ret_from_fork or
248 *   ret_from_kernel_thread:
249 *   a0 points to ret_from_fork/ret_from_kernel_thread (simulating a call4)
250 *   sp points to itself (thread.sp)
251 *   a2, a3 are unused for userspace threads,
252 *   a2 points to thread_fn, a3 holds thread_fn arg for kernel threads.
253 *
254 * Note: This is a pristine frame, so we don't need any spill region on top of
255 *       childregs.
256 *
257 * The fun part:  if we're keeping the same VM (i.e. cloning a thread,
258 * not an entire process), we're normally given a new usp, and we CANNOT share
259 * any live address register windows.  If we just copy those live frames over,
260 * the two threads (parent and child) will overflow the same frames onto the
261 * parent stack at different times, likely corrupting the parent stack (esp.
262 * if the parent returns from functions that called clone() and calls new
263 * ones, before the child overflows its now old copies of its parent windows).
264 * One solution is to spill windows to the parent stack, but that's fairly
265 * involved.  Much simpler to just not copy those live frames across.
266 */
267
268int copy_thread(struct task_struct *p, const struct kernel_clone_args *args)
269{
270	u64 clone_flags = args->flags;
271	unsigned long usp_thread_fn = args->stack;
272	unsigned long tls = args->tls;
273	struct pt_regs *childregs = task_pt_regs(p);
274
275#if (XTENSA_HAVE_COPROCESSORS || XTENSA_HAVE_IO_PORTS)
276	struct thread_info *ti;
277#endif
278
279#if defined(__XTENSA_WINDOWED_ABI__)
280	/* Create a call4 dummy-frame: a0 = 0, a1 = childregs. */
281	SPILL_SLOT(childregs, 1) = (unsigned long)childregs;
282	SPILL_SLOT(childregs, 0) = 0;
283
284	p->thread.sp = (unsigned long)childregs;
285#elif defined(__XTENSA_CALL0_ABI__)
286	/* Reserve 16 bytes for the _switch_to stack frame. */
287	p->thread.sp = (unsigned long)childregs - 16;
288#else
289#error Unsupported Xtensa ABI
290#endif
291
292	if (!args->fn) {
293		struct pt_regs *regs = current_pt_regs();
294		unsigned long usp = usp_thread_fn ?
295			usp_thread_fn : regs->areg[1];
296
297		p->thread.ra = MAKE_RA_FOR_CALL(
298				(unsigned long)ret_from_fork, 0x1);
299
300		*childregs = *regs;
301		childregs->areg[1] = usp;
302		childregs->areg[2] = 0;
303
304		/* When sharing memory with the parent thread, the child
305		   usually starts on a pristine stack, so we have to reset
306		   windowbase, windowstart and wmask.
307		   (Note that such a new thread is required to always create
308		   an initial call4 frame)
309		   The exception is vfork, where the new thread continues to
310		   run on the parent's stack until it calls execve. This could
311		   be a call8 or call12, which requires a legal stack frame
312		   of the previous caller for the overflow handlers to work.
313		   (Note that it's always legal to overflow live registers).
314		   In this case, ensure to spill at least the stack pointer
315		   of that frame. */
316
317		if (clone_flags & CLONE_VM) {
318			/* check that caller window is live and same stack */
319			int len = childregs->wmask & ~0xf;
320			if (regs->areg[1] == usp && len != 0) {
321				int callinc = (regs->areg[0] >> 30) & 3;
322				int caller_ars = XCHAL_NUM_AREGS - callinc * 4;
323				put_user(regs->areg[caller_ars+1],
324					 (unsigned __user*)(usp - 12));
325			}
326			childregs->wmask = 1;
327			childregs->windowstart = 1;
328			childregs->windowbase = 0;
329		}
330
331		if (clone_flags & CLONE_SETTLS)
332			childregs->threadptr = tls;
333	} else {
334		p->thread.ra = MAKE_RA_FOR_CALL(
335				(unsigned long)ret_from_kernel_thread, 1);
336
337		/* pass parameters to ret_from_kernel_thread: */
338#if defined(__XTENSA_WINDOWED_ABI__)
339		/*
340		 * a2 = thread_fn, a3 = thread_fn arg.
341		 * Window underflow will load registers from the
342		 * spill slots on the stack on return from _switch_to.
343		 */
344		SPILL_SLOT(childregs, 2) = (unsigned long)args->fn;
345		SPILL_SLOT(childregs, 3) = (unsigned long)args->fn_arg;
346#elif defined(__XTENSA_CALL0_ABI__)
347		/*
348		 * a12 = thread_fn, a13 = thread_fn arg.
349		 * _switch_to epilogue will load registers from the stack.
350		 */
351		((unsigned long *)p->thread.sp)[0] = (unsigned long)args->fn;
352		((unsigned long *)p->thread.sp)[1] = (unsigned long)args->fn_arg;
353#else
354#error Unsupported Xtensa ABI
355#endif
356
357		/* Childregs are only used when we're going to userspace
358		 * in which case start_thread will set them up.
359		 */
360	}
361
362#if (XTENSA_HAVE_COPROCESSORS || XTENSA_HAVE_IO_PORTS)
363	ti = task_thread_info(p);
364	ti->cpenable = 0;
365#endif
366
367	clear_ptrace_hw_breakpoint(p);
368
369	return 0;
370}
371
372
373/*
374 * These bracket the sleeping functions..
375 */
376
377unsigned long __get_wchan(struct task_struct *p)
378{
379	unsigned long sp, pc;
380	unsigned long stack_page = (unsigned long) task_stack_page(p);
381	int count = 0;
382
383	sp = p->thread.sp;
384	pc = MAKE_PC_FROM_RA(p->thread.ra, _text);
385
386	do {
387		if (sp < stack_page + sizeof(struct task_struct) ||
388		    sp >= (stack_page + THREAD_SIZE) ||
389		    pc == 0)
390			return 0;
391		if (!in_sched_functions(pc))
392			return pc;
393
394		/* Stack layout: sp-4: ra, sp-3: sp' */
395
396		pc = MAKE_PC_FROM_RA(SPILL_SLOT(sp, 0), _text);
397		sp = SPILL_SLOT(sp, 1);
398	} while (count++ < 16);
399	return 0;
400}