arch/loongarch/kernel/process.c

   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/lbt.h>
  42 #include <asm/io.h>
  43 #include <asm/irq.h>
  44 #include <asm/irq_regs.h>
  45 #include <asm/loongarch.h>
  46 #include <asm/pgtable.h>
  47 #include <asm/processor.h>
  48 #include <asm/reg.h>
  49 #include <asm/unwind.h>
  50 #include <asm/vdso.h>
  51
  52 #ifdef CONFIG_STACKPROTECTOR
  53 #include <linux/stackprotector.h>
  54 unsigned long __stack_chk_guard __read_mostly;
  55 EXPORT_SYMBOL(__stack_chk_guard);
  56 #endif
  57
  58 /*
  59  * Idle related variables and functions
  60  */
  61
  62 unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
  63 EXPORT_SYMBOL(boot_option_idle_override);
  64
  65 asmlinkage void ret_from_fork(void);
  66 asmlinkage void ret_from_kernel_thread(void);
  67
  68 void start_thread(struct pt_regs *regs, unsigned long pc, unsigned long sp)
  69 {
  70         unsigned long crmd;
  71         unsigned long prmd;
  72         unsigned long euen;
  73
  74         /* New thread loses kernel privileges. */
  75         crmd = regs->csr_crmd & ~(PLV_MASK);
  76         crmd |= PLV_USER;
  77         regs->csr_crmd = crmd;
  78
  79         prmd = regs->csr_prmd & ~(PLV_MASK);
  80         prmd |= PLV_USER;
  81         regs->csr_prmd = prmd;
  82
  83         euen = regs->csr_euen & ~(CSR_EUEN_FPEN);
  84         regs->csr_euen = euen;
  85         lose_fpu(0);
  86         lose_lbt(0);
  87
  88         clear_thread_flag(TIF_LSX_CTX_LIVE);
  89         clear_thread_flag(TIF_LASX_CTX_LIVE);
  90         clear_thread_flag(TIF_LBT_CTX_LIVE);
  91         clear_used_math();
  92         regs->csr_era = pc;
  93         regs->regs[3] = sp;
  94 }
  95
  96 void flush_thread(void)
  97 {
  98         flush_ptrace_hw_breakpoint(current);
  99 }
 100
 101 void exit_thread(struct task_struct *tsk)
 102 {
 103 }
 104
 105 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
 106 {
 107         /*
 108          * Save any process state which is live in hardware registers to the
 109          * parent context prior to duplication. This prevents the new child
 110          * state becoming stale if the parent is preempted before copy_thread()
 111          * gets a chance to save the parent's live hardware registers to the
 112          * child context.
 113          */
 114         preempt_disable();
 115
 116         if (is_fpu_owner()) {
 117                 if (is_lasx_enabled())
 118                         save_lasx(current);
 119                 else if (is_lsx_enabled())
 120                         save_lsx(current);
 121                 else
 122                         save_fp(current);
 123         }
 124
 125         preempt_enable();
 126
 127         if (!used_math())
 128                 memcpy(dst, src, offsetof(struct task_struct, thread.fpu.fpr));
 129         else
 130                 memcpy(dst, src, offsetof(struct task_struct, thread.lbt.scr0));
 131
 132 #ifdef CONFIG_CPU_HAS_LBT
 133         memcpy(&dst->thread.lbt, &src->thread.lbt, sizeof(struct loongarch_lbt));
 134 #endif
 135
 136         return 0;
 137 }
 138
 139 /*
 140  * Copy architecture-specific thread state
 141  */
 142 int copy_thread(struct task_struct *p, const struct kernel_clone_args *args)
 143 {
 144         unsigned long childksp;
 145         unsigned long tls = args->tls;
 146         unsigned long usp = args->stack;
 147         unsigned long clone_flags = args->flags;
 148         struct pt_regs *childregs, *regs = current_pt_regs();
 149
 150         childksp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
 151
 152         /* set up new TSS. */
 153         childregs = (struct pt_regs *) childksp - 1;
 154         /*  Put the stack after the struct pt_regs.  */
 155         childksp = (unsigned long) childregs;
 156         p->thread.sched_cfa = 0;
 157         p->thread.csr_euen = 0;
 158         p->thread.csr_crmd = csr_read32(LOONGARCH_CSR_CRMD);
 159         p->thread.csr_prmd = csr_read32(LOONGARCH_CSR_PRMD);
 160         p->thread.csr_ecfg = csr_read32(LOONGARCH_CSR_ECFG);
 161         if (unlikely(args->fn)) {
 162                 /* kernel thread */
 163                 p->thread.reg03 = childksp;
 164                 p->thread.reg23 = (unsigned long)args->fn;
 165                 p->thread.reg24 = (unsigned long)args->fn_arg;
 166                 p->thread.reg01 = (unsigned long)ret_from_kernel_thread;
 167                 p->thread.sched_ra = (unsigned long)ret_from_kernel_thread;
 168                 memset(childregs, 0, sizeof(struct pt_regs));
 169                 childregs->csr_euen = p->thread.csr_euen;
 170                 childregs->csr_crmd = p->thread.csr_crmd;
 171                 childregs->csr_prmd = p->thread.csr_prmd;
 172                 childregs->csr_ecfg = p->thread.csr_ecfg;
 173                 goto out;
 174         }
 175
 176         /* user thread */
 177         *childregs = *regs;
 178         childregs->regs[4] = 0; /* Child gets zero as return value */
 179         if (usp)
 180                 childregs->regs[3] = usp;
 181
 182         p->thread.reg03 = (unsigned long) childregs;
 183         p->thread.reg01 = (unsigned long) ret_from_fork;
 184         p->thread.sched_ra = (unsigned long) ret_from_fork;
 185
 186         /*
 187          * New tasks lose permission to use the fpu. This accelerates context
 188          * switching for most programs since they don't use the fpu.
 189          */
 190         childregs->csr_euen = 0;
 191
 192         if (clone_flags & CLONE_SETTLS)
 193                 childregs->regs[2] = tls;
 194
 195 out:
 196         ptrace_hw_copy_thread(p);
 197         clear_tsk_thread_flag(p, TIF_USEDFPU);
 198         clear_tsk_thread_flag(p, TIF_USEDSIMD);
 199         clear_tsk_thread_flag(p, TIF_USEDLBT);
 200         clear_tsk_thread_flag(p, TIF_LSX_CTX_LIVE);
 201         clear_tsk_thread_flag(p, TIF_LASX_CTX_LIVE);
 202         clear_tsk_thread_flag(p, TIF_LBT_CTX_LIVE);
 203
 204         return 0;
 205 }
 206
 207 unsigned long __get_wchan(struct task_struct *task)
 208 {
 209         unsigned long pc = 0;
 210         struct unwind_state state;
 211
 212         if (!try_get_task_stack(task))
 213                 return 0;
 214
 215         for (unwind_start(&state, task, NULL);
 216              !unwind_done(&state); unwind_next_frame(&state)) {
 217                 pc = unwind_get_return_address(&state);
 218                 if (!pc)
 219                         break;
 220                 if (in_sched_functions(pc))
 221                         continue;
 222                 break;
 223         }
 224
 225         put_task_stack(task);
 226
 227         return pc;
 228 }
 229
 230 bool in_irq_stack(unsigned long stack, struct stack_info *info)
 231 {
 232         unsigned long nextsp;
 233         unsigned long begin = (unsigned long)this_cpu_read(irq_stack);
 234         unsigned long end = begin + IRQ_STACK_START;
 235
 236         if (stack < begin || stack >= end)
 237                 return false;
 238
 239         nextsp = *(unsigned long *)end;
 240         if (nextsp & (SZREG - 1))
 241                 return false;
 242
 243         info->begin = begin;
 244         info->end = end;
 245         info->next_sp = nextsp;
 246         info->type = STACK_TYPE_IRQ;
 247
 248         return true;
 249 }
 250
 251 bool in_task_stack(unsigned long stack, struct task_struct *task,
 252                         struct stack_info *info)
 253 {
 254         unsigned long begin = (unsigned long)task_stack_page(task);
 255         unsigned long end = begin + THREAD_SIZE;
 256
 257         if (stack < begin || stack >= end)
 258                 return false;
 259
 260         info->begin = begin;
 261         info->end = end;
 262         info->next_sp = 0;
 263         info->type = STACK_TYPE_TASK;
 264
 265         return true;
 266 }
 267
 268 int get_stack_info(unsigned long stack, struct task_struct *task,
 269                    struct stack_info *info)
 270 {
 271         task = task ? : current;
 272
 273         if (!stack || stack & (SZREG - 1))
 274                 goto unknown;
 275
 276         if (in_task_stack(stack, task, info))
 277                 return 0;
 278
 279         if (task != current)
 280                 goto unknown;
 281
 282         if (in_irq_stack(stack, info))
 283                 return 0;
 284
 285 unknown:
 286         info->type = STACK_TYPE_UNKNOWN;
 287         return -EINVAL;
 288 }
 289
 290 unsigned long stack_top(void)
 291 {
 292         unsigned long top = TASK_SIZE & PAGE_MASK;
 293
 294         /* Space for the VDSO & data page */
 295         top -= PAGE_ALIGN(current->thread.vdso->size);
 296         top -= VVAR_SIZE;
 297
 298         /* Space to randomize the VDSO base */
 299         if (current->flags & PF_RANDOMIZE)
 300                 top -= VDSO_RANDOMIZE_SIZE;
 301
 302         return top;
 303 }
 304
 305 /*
 306  * Don't forget that the stack pointer must be aligned on a 8 bytes
 307  * boundary for 32-bits ABI and 16 bytes for 64-bits ABI.
 308  */
 309 unsigned long arch_align_stack(unsigned long sp)
 310 {
 311         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 312                 sp -= get_random_u32_below(PAGE_SIZE);
 313
 314         return sp & STACK_ALIGN;
 315 }
 316
 317 static DEFINE_PER_CPU(call_single_data_t, backtrace_csd);
 318 static struct cpumask backtrace_csd_busy;
 319
 320 static void handle_backtrace(void *info)
 321 {
 322         nmi_cpu_backtrace(get_irq_regs());
 323         cpumask_clear_cpu(smp_processor_id(), &backtrace_csd_busy);
 324 }
 325
 326 static void raise_backtrace(cpumask_t *mask)
 327 {
 328         call_single_data_t *csd;
 329         int cpu;
 330
 331         for_each_cpu(cpu, mask) {
 332                 /*
 333                  * If we previously sent an IPI to the target CPU & it hasn't
 334                  * cleared its bit in the busy cpumask then it didn't handle
 335                  * our previous IPI & it's not safe for us to reuse the
 336                  * call_single_data_t.
 337                  */
 338                 if (cpumask_test_and_set_cpu(cpu, &backtrace_csd_busy)) {
 339                         pr_warn("Unable to send backtrace IPI to CPU%u - perhaps it hung?\n",
 340                                 cpu);
 341                         continue;
 342                 }
 343
 344                 csd = &per_cpu(backtrace_csd, cpu);
 345                 csd->func = handle_backtrace;
 346                 smp_call_function_single_async(cpu, csd);
 347         }
 348 }
 349
 350 void arch_trigger_cpumask_backtrace(const cpumask_t *mask, int exclude_cpu)
 351 {
 352         nmi_trigger_cpumask_backtrace(mask, exclude_cpu, raise_backtrace);
 353 }
 354
 355 #ifdef CONFIG_64BIT
 356 void loongarch_dump_regs64(u64 *uregs, const struct pt_regs *regs)
 357 {
 358         unsigned int i;
 359
 360         for (i = LOONGARCH_EF_R1; i <= LOONGARCH_EF_R31; i++) {
 361                 uregs[i] = regs->regs[i - LOONGARCH_EF_R0];
 362         }
 363
 364         uregs[LOONGARCH_EF_ORIG_A0] = regs->orig_a0;
 365         uregs[LOONGARCH_EF_CSR_ERA] = regs->csr_era;
 366         uregs[LOONGARCH_EF_CSR_BADV] = regs->csr_badvaddr;
 367         uregs[LOONGARCH_EF_CSR_CRMD] = regs->csr_crmd;
 368         uregs[LOONGARCH_EF_CSR_PRMD] = regs->csr_prmd;
 369         uregs[LOONGARCH_EF_CSR_EUEN] = regs->csr_euen;
 370         uregs[LOONGARCH_EF_CSR_ECFG] = regs->csr_ecfg;
 371         uregs[LOONGARCH_EF_CSR_ESTAT] = regs->csr_estat;
 372 }
 373 #endif /* CONFIG_64BIT */