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