1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * Author: Huacai Chen <chenhuacai@loongson.cn> 4 * Copyright (C) 2020-2022 Loongson Technology Corporation Limited 5 * 6 * Derived from MIPS: 7 * Copyright (C) 1994 - 1999, 2000 by Ralf Baechle and others. 8 * Copyright (C) 2005, 2006 by Ralf Baechle (ralf@linux-mips.org) 9 * Copyright (C) 1999, 2000 Silicon Graphics, Inc. 10 * Copyright (C) 2004 Thiemo Seufer 11 * Copyright (C) 2013 Imagination Technologies Ltd. 12 */ 13 #include <linux/cpu.h> 14 #include <linux/init.h> 15 #include <linux/kernel.h> 16 #include <linux/errno.h> 17 #include <linux/sched.h> 18 #include <linux/sched/debug.h> 19 #include <linux/sched/task.h> 20 #include <linux/sched/task_stack.h> 21 #include <linux/hw_breakpoint.h> 22 #include <linux/mm.h> 23 #include <linux/stddef.h> 24 #include <linux/unistd.h> 25 #include <linux/export.h> 26 #include <linux/ptrace.h> 27 #include <linux/mman.h> 28 #include <linux/personality.h> 29 #include <linux/sys.h> 30 #include <linux/completion.h> 31 #include <linux/kallsyms.h> 32 #include <linux/random.h> 33 #include <linux/prctl.h> 34 #include <linux/nmi.h> 35 36 #include <asm/asm.h> 37 #include <asm/bootinfo.h> 38 #include <asm/cpu.h> 39 #include <asm/elf.h> 40 #include <asm/fpu.h> 41 #include <asm/io.h> 42 #include <asm/irq.h> 43 #include <asm/irq_regs.h> 44 #include <asm/loongarch.h> 45 #include <asm/pgtable.h> 46 #include <asm/processor.h> 47 #include <asm/reg.h> 48 #include <asm/unwind.h> 49 #include <asm/vdso.h> 50 51 #ifdef CONFIG_STACKPROTECTOR 52 #include <linux/stackprotector.h> 53 unsigned long __stack_chk_guard __read_mostly; 54 EXPORT_SYMBOL(__stack_chk_guard); 55 #endif 56 57 /* 58 * Idle related variables and functions 59 */ 60 61 unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE; 62 EXPORT_SYMBOL(boot_option_idle_override); 63 64 #ifdef CONFIG_HOTPLUG_CPU 65 void __noreturn arch_cpu_idle_dead(void) 66 { 67 play_dead(); 68 } 69 #endif 70 71 asmlinkage void ret_from_fork(void); 72 asmlinkage void ret_from_kernel_thread(void); 73 74 void start_thread(struct pt_regs *regs, unsigned long pc, unsigned long sp) 75 { 76 unsigned long crmd; 77 unsigned long prmd; 78 unsigned long euen; 79 80 /* New thread loses kernel privileges. */ 81 crmd = regs->csr_crmd & ~(PLV_MASK); 82 crmd |= PLV_USER; 83 regs->csr_crmd = crmd; 84 85 prmd = regs->csr_prmd & ~(PLV_MASK); 86 prmd |= PLV_USER; 87 regs->csr_prmd = prmd; 88 89 euen = regs->csr_euen & ~(CSR_EUEN_FPEN); 90 regs->csr_euen = euen; 91 lose_fpu(0); 92 93 clear_thread_flag(TIF_LSX_CTX_LIVE); 94 clear_thread_flag(TIF_LASX_CTX_LIVE); 95 clear_used_math(); 96 regs->csr_era = pc; 97 regs->regs[3] = sp; 98 } 99 100 void flush_thread(void) 101 { 102 flush_ptrace_hw_breakpoint(current); 103 } 104 105 void exit_thread(struct task_struct *tsk) 106 { 107 } 108 109 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src) 110 { 111 /* 112 * Save any process state which is live in hardware registers to the 113 * parent context prior to duplication. This prevents the new child 114 * state becoming stale if the parent is preempted before copy_thread() 115 * gets a chance to save the parent's live hardware registers to the 116 * child context. 117 */ 118 preempt_disable(); 119 120 if (is_fpu_owner()) 121 save_fp(current); 122 123 preempt_enable(); 124 125 if (used_math()) 126 memcpy(dst, src, sizeof(struct task_struct)); 127 else 128 memcpy(dst, src, offsetof(struct task_struct, thread.fpu.fpr)); 129 130 return 0; 131 } 132 133 /* 134 * Copy architecture-specific thread state 135 */ 136 int copy_thread(struct task_struct *p, const struct kernel_clone_args *args) 137 { 138 unsigned long childksp; 139 unsigned long tls = args->tls; 140 unsigned long usp = args->stack; 141 unsigned long clone_flags = args->flags; 142 struct pt_regs *childregs, *regs = current_pt_regs(); 143 144 childksp = (unsigned long)task_stack_page(p) + THREAD_SIZE; 145 146 /* set up new TSS. */ 147 childregs = (struct pt_regs *) childksp - 1; 148 /* Put the stack after the struct pt_regs. */ 149 childksp = (unsigned long) childregs; 150 p->thread.sched_cfa = 0; 151 p->thread.csr_euen = 0; 152 p->thread.csr_crmd = csr_read32(LOONGARCH_CSR_CRMD); 153 p->thread.csr_prmd = csr_read32(LOONGARCH_CSR_PRMD); 154 p->thread.csr_ecfg = csr_read32(LOONGARCH_CSR_ECFG); 155 if (unlikely(args->fn)) { 156 /* kernel thread */ 157 p->thread.reg03 = childksp; 158 p->thread.reg23 = (unsigned long)args->fn; 159 p->thread.reg24 = (unsigned long)args->fn_arg; 160 p->thread.reg01 = (unsigned long)ret_from_kernel_thread; 161 p->thread.sched_ra = (unsigned long)ret_from_kernel_thread; 162 memset(childregs, 0, sizeof(struct pt_regs)); 163 childregs->csr_euen = p->thread.csr_euen; 164 childregs->csr_crmd = p->thread.csr_crmd; 165 childregs->csr_prmd = p->thread.csr_prmd; 166 childregs->csr_ecfg = p->thread.csr_ecfg; 167 goto out; 168 } 169 170 /* user thread */ 171 *childregs = *regs; 172 childregs->regs[4] = 0; /* Child gets zero as return value */ 173 if (usp) 174 childregs->regs[3] = usp; 175 176 p->thread.reg03 = (unsigned long) childregs; 177 p->thread.reg01 = (unsigned long) ret_from_fork; 178 p->thread.sched_ra = (unsigned long) ret_from_fork; 179 180 /* 181 * New tasks lose permission to use the fpu. This accelerates context 182 * switching for most programs since they don't use the fpu. 183 */ 184 childregs->csr_euen = 0; 185 186 if (clone_flags & CLONE_SETTLS) 187 childregs->regs[2] = tls; 188 189 out: 190 ptrace_hw_copy_thread(p); 191 clear_tsk_thread_flag(p, TIF_USEDFPU); 192 clear_tsk_thread_flag(p, TIF_USEDSIMD); 193 clear_tsk_thread_flag(p, TIF_LSX_CTX_LIVE); 194 clear_tsk_thread_flag(p, TIF_LASX_CTX_LIVE); 195 196 return 0; 197 } 198 199 unsigned long __get_wchan(struct task_struct *task) 200 { 201 unsigned long pc = 0; 202 struct unwind_state state; 203 204 if (!try_get_task_stack(task)) 205 return 0; 206 207 for (unwind_start(&state, task, NULL); 208 !unwind_done(&state); unwind_next_frame(&state)) { 209 pc = unwind_get_return_address(&state); 210 if (!pc) 211 break; 212 if (in_sched_functions(pc)) 213 continue; 214 break; 215 } 216 217 put_task_stack(task); 218 219 return pc; 220 } 221 222 bool in_irq_stack(unsigned long stack, struct stack_info *info) 223 { 224 unsigned long nextsp; 225 unsigned long begin = (unsigned long)this_cpu_read(irq_stack); 226 unsigned long end = begin + IRQ_STACK_START; 227 228 if (stack < begin || stack >= end) 229 return false; 230 231 nextsp = *(unsigned long *)end; 232 if (nextsp & (SZREG - 1)) 233 return false; 234 235 info->begin = begin; 236 info->end = end; 237 info->next_sp = nextsp; 238 info->type = STACK_TYPE_IRQ; 239 240 return true; 241 } 242 243 bool in_task_stack(unsigned long stack, struct task_struct *task, 244 struct stack_info *info) 245 { 246 unsigned long begin = (unsigned long)task_stack_page(task); 247 unsigned long end = begin + THREAD_SIZE; 248 249 if (stack < begin || stack >= end) 250 return false; 251 252 info->begin = begin; 253 info->end = end; 254 info->next_sp = 0; 255 info->type = STACK_TYPE_TASK; 256 257 return true; 258 } 259 260 int get_stack_info(unsigned long stack, struct task_struct *task, 261 struct stack_info *info) 262 { 263 task = task ? : current; 264 265 if (!stack || stack & (SZREG - 1)) 266 goto unknown; 267 268 if (in_task_stack(stack, task, info)) 269 return 0; 270 271 if (task != current) 272 goto unknown; 273 274 if (in_irq_stack(stack, info)) 275 return 0; 276 277 unknown: 278 info->type = STACK_TYPE_UNKNOWN; 279 return -EINVAL; 280 } 281 282 unsigned long stack_top(void) 283 { 284 unsigned long top = TASK_SIZE & PAGE_MASK; 285 286 /* Space for the VDSO & data page */ 287 top -= PAGE_ALIGN(current->thread.vdso->size); 288 top -= PAGE_SIZE; 289 290 /* Space to randomize the VDSO base */ 291 if (current->flags & PF_RANDOMIZE) 292 top -= VDSO_RANDOMIZE_SIZE; 293 294 return top; 295 } 296 297 /* 298 * Don't forget that the stack pointer must be aligned on a 8 bytes 299 * boundary for 32-bits ABI and 16 bytes for 64-bits ABI. 300 */ 301 unsigned long arch_align_stack(unsigned long sp) 302 { 303 if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space) 304 sp -= get_random_u32_below(PAGE_SIZE); 305 306 return sp & STACK_ALIGN; 307 } 308 309 static DEFINE_PER_CPU(call_single_data_t, backtrace_csd); 310 static struct cpumask backtrace_csd_busy; 311 312 static void handle_backtrace(void *info) 313 { 314 nmi_cpu_backtrace(get_irq_regs()); 315 cpumask_clear_cpu(smp_processor_id(), &backtrace_csd_busy); 316 } 317 318 static void raise_backtrace(cpumask_t *mask) 319 { 320 call_single_data_t *csd; 321 int cpu; 322 323 for_each_cpu(cpu, mask) { 324 /* 325 * If we previously sent an IPI to the target CPU & it hasn't 326 * cleared its bit in the busy cpumask then it didn't handle 327 * our previous IPI & it's not safe for us to reuse the 328 * call_single_data_t. 329 */ 330 if (cpumask_test_and_set_cpu(cpu, &backtrace_csd_busy)) { 331 pr_warn("Unable to send backtrace IPI to CPU%u - perhaps it hung?\n", 332 cpu); 333 continue; 334 } 335 336 csd = &per_cpu(backtrace_csd, cpu); 337 csd->func = handle_backtrace; 338 smp_call_function_single_async(cpu, csd); 339 } 340 } 341 342 void arch_trigger_cpumask_backtrace(const cpumask_t *mask, bool exclude_self) 343 { 344 nmi_trigger_cpumask_backtrace(mask, exclude_self, raise_backtrace); 345 } 346 347 #ifdef CONFIG_64BIT 348 void loongarch_dump_regs64(u64 *uregs, const struct pt_regs *regs) 349 { 350 unsigned int i; 351 352 for (i = LOONGARCH_EF_R1; i <= LOONGARCH_EF_R31; i++) { 353 uregs[i] = regs->regs[i - LOONGARCH_EF_R0]; 354 } 355 356 uregs[LOONGARCH_EF_ORIG_A0] = regs->orig_a0; 357 uregs[LOONGARCH_EF_CSR_ERA] = regs->csr_era; 358 uregs[LOONGARCH_EF_CSR_BADV] = regs->csr_badvaddr; 359 uregs[LOONGARCH_EF_CSR_CRMD] = regs->csr_crmd; 360 uregs[LOONGARCH_EF_CSR_PRMD] = regs->csr_prmd; 361 uregs[LOONGARCH_EF_CSR_EUEN] = regs->csr_euen; 362 uregs[LOONGARCH_EF_CSR_ECFG] = regs->csr_ecfg; 363 uregs[LOONGARCH_EF_CSR_ESTAT] = regs->csr_estat; 364 } 365 #endif /* CONFIG_64BIT */ 366