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/exec.h>
  41 #include <asm/fpu.h>
  42 #include <asm/lbt.h>
  43 #include <asm/io.h>
  44 #include <asm/irq.h>
  45 #include <asm/irq_regs.h>
  46 #include <asm/loongarch.h>
  47 #include <asm/pgtable.h>
  48 #include <asm/processor.h>
  49 #include <asm/reg.h>
  50 #include <asm/unwind.h>
  51 #include <asm/vdso.h>
  52
  53 #ifdef CONFIG_STACKPROTECTOR
  54 #include <linux/stackprotector.h>
  55 unsigned long __stack_chk_guard __read_mostly;
  56 EXPORT_SYMBOL(__stack_chk_guard);
  57 #endif
  58
  59 /*
  60  * Idle related variables and functions
  61  */
  62
  63 unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
  64 EXPORT_SYMBOL(boot_option_idle_override);
  65
  66 asmlinkage void ret_from_fork(void);
  67 asmlinkage void ret_from_kernel_thread(void);
  68
  69 void start_thread(struct pt_regs *regs, unsigned long pc, unsigned long sp)
  70 {
  71         unsigned long crmd;
  72         unsigned long prmd;
  73         unsigned long euen;
  74
  75         /* New thread loses kernel privileges. */
  76         crmd = regs->csr_crmd & ~(PLV_MASK);
  77         crmd |= PLV_USER;
  78         regs->csr_crmd = crmd;
  79
  80         prmd = regs->csr_prmd & ~(PLV_MASK);
  81         prmd |= PLV_USER;
  82         regs->csr_prmd = prmd;
  83
  84         euen = regs->csr_euen & ~(CSR_EUEN_FPEN);
  85         regs->csr_euen = euen;
  86         lose_fpu(0);
  87         lose_lbt(0);
  88         current->thread.fpu.fcsr = boot_cpu_data.fpu_csr0;
  89
  90         clear_thread_flag(TIF_LSX_CTX_LIVE);
  91         clear_thread_flag(TIF_LASX_CTX_LIVE);
  92         clear_thread_flag(TIF_LBT_CTX_LIVE);
  93         clear_used_math();
  94         regs->csr_era = pc;
  95         regs->regs[3] = sp;
  96 }
  97
  98 void flush_thread(void)
  99 {
 100         flush_ptrace_hw_breakpoint(current);
 101 }
 102
 103 void exit_thread(struct task_struct *tsk)
 104 {
 105 }
 106
 107 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
 108 {
 109         /*
 110          * Save any process state which is live in hardware registers to the
 111          * parent context prior to duplication. This prevents the new child
 112          * state becoming stale if the parent is preempted before copy_thread()
 113          * gets a chance to save the parent's live hardware registers to the
 114          * child context.
 115          */
 116         preempt_disable();
 117
 118         if (is_fpu_owner()) {
 119                 if (is_lasx_enabled())
 120                         save_lasx(current);
 121                 else if (is_lsx_enabled())
 122                         save_lsx(current);
 123                 else
 124                         save_fp(current);
 125         }
 126
 127         preempt_enable();
 128
 129         if (!used_math())
 130                 memcpy(dst, src, offsetof(struct task_struct, thread.fpu.fpr));
 131         else
 132                 memcpy(dst, src, offsetof(struct task_struct, thread.lbt.scr0));
 133
 134 #ifdef CONFIG_CPU_HAS_LBT
 135         memcpy(&dst->thread.lbt, &src->thread.lbt, sizeof(struct loongarch_lbt));
 136 #endif
 137
 138         return 0;
 139 }
 140
 141 /*
 142  * Copy architecture-specific thread state
 143  */
 144 int copy_thread(struct task_struct *p, const struct kernel_clone_args *args)
 145 {
 146         unsigned long childksp;
 147         unsigned long tls = args->tls;
 148         unsigned long usp = args->stack;
 149         unsigned long clone_flags = args->flags;
 150         struct pt_regs *childregs, *regs = current_pt_regs();
 151
 152         childksp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
 153
 154         /* set up new TSS. */
 155         childregs = (struct pt_regs *) childksp - 1;
 156         /*  Put the stack after the struct pt_regs.  */
 157         childksp = (unsigned long) childregs;
 158         p->thread.sched_cfa = 0;
 159         p->thread.csr_euen = 0;
 160         p->thread.csr_crmd = csr_read32(LOONGARCH_CSR_CRMD);
 161         p->thread.csr_prmd = csr_read32(LOONGARCH_CSR_PRMD);
 162         p->thread.csr_ecfg = csr_read32(LOONGARCH_CSR_ECFG);
 163         if (unlikely(args->fn)) {
 164                 /* kernel thread */
 165                 p->thread.reg03 = childksp;
 166                 p->thread.reg23 = (unsigned long)args->fn;
 167                 p->thread.reg24 = (unsigned long)args->fn_arg;
 168                 p->thread.reg01 = (unsigned long)ret_from_kernel_thread;
 169                 p->thread.sched_ra = (unsigned long)ret_from_kernel_thread;
 170                 memset(childregs, 0, sizeof(struct pt_regs));
 171                 childregs->csr_euen = p->thread.csr_euen;
 172                 childregs->csr_crmd = p->thread.csr_crmd;
 173                 childregs->csr_prmd = p->thread.csr_prmd;
 174                 childregs->csr_ecfg = p->thread.csr_ecfg;
 175                 goto out;
 176         }
 177
 178         /* user thread */
 179         *childregs = *regs;
 180         childregs->regs[4] = 0; /* Child gets zero as return value */
 181         if (usp)
 182                 childregs->regs[3] = usp;
 183
 184         p->thread.reg03 = (unsigned long) childregs;
 185         p->thread.reg01 = (unsigned long) ret_from_fork;
 186         p->thread.sched_ra = (unsigned long) ret_from_fork;
 187
 188         /*
 189          * New tasks lose permission to use the fpu. This accelerates context
 190          * switching for most programs since they don't use the fpu.
 191          */
 192         childregs->csr_euen = 0;
 193
 194         if (clone_flags & CLONE_SETTLS)
 195                 childregs->regs[2] = tls;
 196
 197 out:
 198         ptrace_hw_copy_thread(p);
 199         clear_tsk_thread_flag(p, TIF_USEDFPU);
 200         clear_tsk_thread_flag(p, TIF_USEDSIMD);
 201         clear_tsk_thread_flag(p, TIF_USEDLBT);
 202         clear_tsk_thread_flag(p, TIF_LSX_CTX_LIVE);
 203         clear_tsk_thread_flag(p, TIF_LASX_CTX_LIVE);
 204         clear_tsk_thread_flag(p, TIF_LBT_CTX_LIVE);
 205
 206         return 0;
 207 }
 208
 209 unsigned long __get_wchan(struct task_struct *task)
 210 {
 211         unsigned long pc = 0;
 212         struct unwind_state state;
 213
 214         if (!try_get_task_stack(task))
 215                 return 0;
 216
 217         for (unwind_start(&state, task, NULL);
 218              !unwind_done(&state); unwind_next_frame(&state)) {
 219                 pc = unwind_get_return_address(&state);
 220                 if (!pc)
 221                         break;
 222                 if (in_sched_functions(pc))
 223                         continue;
 224                 break;
 225         }
 226
 227         put_task_stack(task);
 228
 229         return pc;
 230 }
 231
 232 bool in_irq_stack(unsigned long stack, struct stack_info *info)
 233 {
 234         unsigned long nextsp;
 235         unsigned long begin = (unsigned long)this_cpu_read(irq_stack);
 236         unsigned long end = begin + IRQ_STACK_START;
 237
 238         if (stack < begin || stack >= end)
 239                 return false;
 240
 241         nextsp = *(unsigned long *)end;
 242         if (nextsp & (SZREG - 1))
 243                 return false;
 244
 245         info->begin = begin;
 246         info->end = end;
 247         info->next_sp = nextsp;
 248         info->type = STACK_TYPE_IRQ;
 249
 250         return true;
 251 }
 252
 253 bool in_task_stack(unsigned long stack, struct task_struct *task,
 254                         struct stack_info *info)
 255 {
 256         unsigned long begin = (unsigned long)task_stack_page(task);
 257         unsigned long end = begin + THREAD_SIZE;
 258
 259         if (stack < begin || stack >= end)
 260                 return false;
 261
 262         info->begin = begin;
 263         info->end = end;
 264         info->next_sp = 0;
 265         info->type = STACK_TYPE_TASK;
 266
 267         return true;
 268 }
 269
 270 int get_stack_info(unsigned long stack, struct task_struct *task,
 271                    struct stack_info *info)
 272 {
 273         task = task ? : current;
 274
 275         if (!stack || stack & (SZREG - 1))
 276                 goto unknown;
 277
 278         if (in_task_stack(stack, task, info))
 279                 return 0;
 280
 281         if (task != current)
 282                 goto unknown;
 283
 284         if (in_irq_stack(stack, info))
 285                 return 0;
 286
 287 unknown:
 288         info->type = STACK_TYPE_UNKNOWN;
 289         return -EINVAL;
 290 }
 291
 292 unsigned long stack_top(void)
 293 {
 294         unsigned long top = TASK_SIZE & PAGE_MASK;
 295
 296         /* Space for the VDSO & data page */
 297         top -= PAGE_ALIGN(current->thread.vdso->size);
 298         top -= VVAR_SIZE;
 299
 300         /* Space to randomize the VDSO base */
 301         if (current->flags & PF_RANDOMIZE)
 302                 top -= VDSO_RANDOMIZE_SIZE;
 303
 304         return top;
 305 }
 306
 307 /*
 308  * Don't forget that the stack pointer must be aligned on a 8 bytes
 309  * boundary for 32-bits ABI and 16 bytes for 64-bits ABI.
 310  */
 311 unsigned long arch_align_stack(unsigned long sp)
 312 {
 313         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 314                 sp -= get_random_u32_below(PAGE_SIZE);
 315
 316         return sp & STACK_ALIGN;
 317 }
 318
 319 static DEFINE_PER_CPU(call_single_data_t, backtrace_csd);
 320 static struct cpumask backtrace_csd_busy;
 321
 322 static void handle_backtrace(void *info)
 323 {
 324         nmi_cpu_backtrace(get_irq_regs());
 325         cpumask_clear_cpu(smp_processor_id(), &backtrace_csd_busy);
 326 }
 327
 328 static void raise_backtrace(cpumask_t *mask)
 329 {
 330         call_single_data_t *csd;
 331         int cpu;
 332
 333         for_each_cpu(cpu, mask) {
 334                 /*
 335                  * If we previously sent an IPI to the target CPU & it hasn't
 336                  * cleared its bit in the busy cpumask then it didn't handle
 337                  * our previous IPI & it's not safe for us to reuse the
 338                  * call_single_data_t.
 339                  */
 340                 if (cpumask_test_and_set_cpu(cpu, &backtrace_csd_busy)) {
 341                         pr_warn("Unable to send backtrace IPI to CPU%u - perhaps it hung?\n",
 342                                 cpu);
 343                         continue;
 344                 }
 345
 346                 csd = &per_cpu(backtrace_csd, cpu);
 347                 csd->func = handle_backtrace;
 348                 smp_call_function_single_async(cpu, csd);
 349         }
 350 }
 351
 352 void arch_trigger_cpumask_backtrace(const cpumask_t *mask, int exclude_cpu)
 353 {
 354         nmi_trigger_cpumask_backtrace(mask, exclude_cpu, raise_backtrace);
 355 }
 356
 357 #ifdef CONFIG_64BIT
 358 void loongarch_dump_regs64(u64 *uregs, const struct pt_regs *regs)
 359 {
 360         unsigned int i;
 361
 362         for (i = LOONGARCH_EF_R1; i <= LOONGARCH_EF_R31; i++) {
 363                 uregs[i] = regs->regs[i - LOONGARCH_EF_R0];
 364         }
 365
 366         uregs[LOONGARCH_EF_ORIG_A0] = regs->orig_a0;
 367         uregs[LOONGARCH_EF_CSR_ERA] = regs->csr_era;
 368         uregs[LOONGARCH_EF_CSR_BADV] = regs->csr_badvaddr;
 369         uregs[LOONGARCH_EF_CSR_CRMD] = regs->csr_crmd;
 370         uregs[LOONGARCH_EF_CSR_PRMD] = regs->csr_prmd;
 371         uregs[LOONGARCH_EF_CSR_EUEN] = regs->csr_euen;
 372         uregs[LOONGARCH_EF_CSR_ECFG] = regs->csr_ecfg;
 373         uregs[LOONGARCH_EF_CSR_ESTAT] = regs->csr_estat;
 374 }
 375 #endif /* CONFIG_64BIT */