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