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/mm.h>
  22 #include <linux/stddef.h>
  23 #include <linux/unistd.h>
  24 #include <linux/export.h>
  25 #include <linux/ptrace.h>
  26 #include <linux/mman.h>
  27 #include <linux/personality.h>
  28 #include <linux/sys.h>
  29 #include <linux/completion.h>
  30 #include <linux/kallsyms.h>
  31 #include <linux/random.h>
  32 #include <linux/prctl.h>
  33 #include <linux/nmi.h>
  34
  35 #include <asm/asm.h>
  36 #include <asm/bootinfo.h>
  37 #include <asm/cpu.h>
  38 #include <asm/elf.h>
  39 #include <asm/fpu.h>
  40 #include <asm/io.h>
  41 #include <asm/irq.h>
  42 #include <asm/irq_regs.h>
  43 #include <asm/loongarch.h>
  44 #include <asm/pgtable.h>
  45 #include <asm/processor.h>
  46 #include <asm/reg.h>
  47 #include <asm/unwind.h>
  48 #include <asm/vdso.h>
  49
  50 /*
  51  * Idle related variables and functions
  52  */
  53
  54 unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
  55 EXPORT_SYMBOL(boot_option_idle_override);
  56
  57 #ifdef CONFIG_HOTPLUG_CPU
  58 void arch_cpu_idle_dead(void)
  59 {
  60         play_dead();
  61 }
  62 #endif
  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         current->thread.fpu.fcsr = boot_cpu_data.fpu_csr0;
  86
  87         clear_thread_flag(TIF_LSX_CTX_LIVE);
  88         clear_thread_flag(TIF_LASX_CTX_LIVE);
  89         clear_used_math();
  90         regs->csr_era = pc;
  91         regs->regs[3] = sp;
  92 }
  93
  94 void exit_thread(struct task_struct *tsk)
  95 {
  96 }
  97
  98 int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
  99 {
 100         /*
 101          * Save any process state which is live in hardware registers to the
 102          * parent context prior to duplication. This prevents the new child
 103          * state becoming stale if the parent is preempted before copy_thread()
 104          * gets a chance to save the parent's live hardware registers to the
 105          * child context.
 106          */
 107         preempt_disable();
 108
 109         if (is_fpu_owner())
 110                 save_fp(current);
 111
 112         preempt_enable();
 113
 114         if (used_math())
 115                 memcpy(dst, src, sizeof(struct task_struct));
 116         else
 117                 memcpy(dst, src, offsetof(struct task_struct, thread.fpu.fpr));
 118
 119         return 0;
 120 }
 121
 122 /*
 123  * Copy architecture-specific thread state
 124  */
 125 int copy_thread(struct task_struct *p, const struct kernel_clone_args *args)
 126 {
 127         unsigned long childksp;
 128         unsigned long tls = args->tls;
 129         unsigned long usp = args->stack;
 130         unsigned long clone_flags = args->flags;
 131         struct pt_regs *childregs, *regs = current_pt_regs();
 132
 133         childksp = (unsigned long)task_stack_page(p) + THREAD_SIZE;
 134
 135         /* set up new TSS. */
 136         childregs = (struct pt_regs *) childksp - 1;
 137         /*  Put the stack after the struct pt_regs.  */
 138         childksp = (unsigned long) childregs;
 139         p->thread.sched_cfa = 0;
 140         p->thread.csr_euen = 0;
 141         p->thread.csr_crmd = csr_read32(LOONGARCH_CSR_CRMD);
 142         p->thread.csr_prmd = csr_read32(LOONGARCH_CSR_PRMD);
 143         p->thread.csr_ecfg = csr_read32(LOONGARCH_CSR_ECFG);
 144         if (unlikely(args->fn)) {
 145                 /* kernel thread */
 146                 p->thread.reg03 = childksp;
 147                 p->thread.reg23 = (unsigned long)args->fn;
 148                 p->thread.reg24 = (unsigned long)args->fn_arg;
 149                 p->thread.reg01 = (unsigned long)ret_from_kernel_thread;
 150                 p->thread.sched_ra = (unsigned long)ret_from_kernel_thread;
 151                 memset(childregs, 0, sizeof(struct pt_regs));
 152                 childregs->csr_euen = p->thread.csr_euen;
 153                 childregs->csr_crmd = p->thread.csr_crmd;
 154                 childregs->csr_prmd = p->thread.csr_prmd;
 155                 childregs->csr_ecfg = p->thread.csr_ecfg;
 156                 goto out;
 157         }
 158
 159         /* user thread */
 160         *childregs = *regs;
 161         childregs->regs[4] = 0; /* Child gets zero as return value */
 162         if (usp)
 163                 childregs->regs[3] = usp;
 164
 165         p->thread.reg03 = (unsigned long) childregs;
 166         p->thread.reg01 = (unsigned long) ret_from_fork;
 167         p->thread.sched_ra = (unsigned long) ret_from_fork;
 168
 169         /*
 170          * New tasks lose permission to use the fpu. This accelerates context
 171          * switching for most programs since they don't use the fpu.
 172          */
 173         childregs->csr_euen = 0;
 174
 175         if (clone_flags & CLONE_SETTLS)
 176                 childregs->regs[2] = tls;
 177
 178 out:
 179         clear_tsk_thread_flag(p, TIF_USEDFPU);
 180         clear_tsk_thread_flag(p, TIF_USEDSIMD);
 181         clear_tsk_thread_flag(p, TIF_LSX_CTX_LIVE);
 182         clear_tsk_thread_flag(p, TIF_LASX_CTX_LIVE);
 183
 184         return 0;
 185 }
 186
 187 unsigned long __get_wchan(struct task_struct *task)
 188 {
 189         unsigned long pc = 0;
 190         struct unwind_state state;
 191
 192         if (!try_get_task_stack(task))
 193                 return 0;
 194
 195         for (unwind_start(&state, task, NULL);
 196              !unwind_done(&state); unwind_next_frame(&state)) {
 197                 pc = unwind_get_return_address(&state);
 198                 if (!pc)
 199                         break;
 200                 if (in_sched_functions(pc))
 201                         continue;
 202                 break;
 203         }
 204
 205         put_task_stack(task);
 206
 207         return pc;
 208 }
 209
 210 bool in_irq_stack(unsigned long stack, struct stack_info *info)
 211 {
 212         unsigned long nextsp;
 213         unsigned long begin = (unsigned long)this_cpu_read(irq_stack);
 214         unsigned long end = begin + IRQ_STACK_START;
 215
 216         if (stack < begin || stack >= end)
 217                 return false;
 218
 219         nextsp = *(unsigned long *)end;
 220         if (nextsp & (SZREG - 1))
 221                 return false;
 222
 223         info->begin = begin;
 224         info->end = end;
 225         info->next_sp = nextsp;
 226         info->type = STACK_TYPE_IRQ;
 227
 228         return true;
 229 }
 230
 231 bool in_task_stack(unsigned long stack, struct task_struct *task,
 232                         struct stack_info *info)
 233 {
 234         unsigned long begin = (unsigned long)task_stack_page(task);
 235         unsigned long end = begin + THREAD_SIZE;
 236
 237         if (stack < begin || stack >= end)
 238                 return false;
 239
 240         info->begin = begin;
 241         info->end = end;
 242         info->next_sp = 0;
 243         info->type = STACK_TYPE_TASK;
 244
 245         return true;
 246 }
 247
 248 int get_stack_info(unsigned long stack, struct task_struct *task,
 249                    struct stack_info *info)
 250 {
 251         task = task ? : current;
 252
 253         if (!stack || stack & (SZREG - 1))
 254                 goto unknown;
 255
 256         if (in_task_stack(stack, task, info))
 257                 return 0;
 258
 259         if (task != current)
 260                 goto unknown;
 261
 262         if (in_irq_stack(stack, info))
 263                 return 0;
 264
 265 unknown:
 266         info->type = STACK_TYPE_UNKNOWN;
 267         return -EINVAL;
 268 }
 269
 270 unsigned long stack_top(void)
 271 {
 272         unsigned long top = TASK_SIZE & PAGE_MASK;
 273
 274         /* Space for the VDSO & data page */
 275         top -= PAGE_ALIGN(current->thread.vdso->size);
 276         top -= PAGE_SIZE;
 277
 278         /* Space to randomize the VDSO base */
 279         if (current->flags & PF_RANDOMIZE)
 280                 top -= VDSO_RANDOMIZE_SIZE;
 281
 282         return top;
 283 }
 284
 285 /*
 286  * Don't forget that the stack pointer must be aligned on a 8 bytes
 287  * boundary for 32-bits ABI and 16 bytes for 64-bits ABI.
 288  */
 289 unsigned long arch_align_stack(unsigned long sp)
 290 {
 291         if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
 292                 sp -= prandom_u32_max(PAGE_SIZE);
 293
 294         return sp & STACK_ALIGN;
 295 }
 296
 297 static DEFINE_PER_CPU(call_single_data_t, backtrace_csd);
 298 static struct cpumask backtrace_csd_busy;
 299
 300 static void handle_backtrace(void *info)
 301 {
 302         nmi_cpu_backtrace(get_irq_regs());
 303         cpumask_clear_cpu(smp_processor_id(), &backtrace_csd_busy);
 304 }
 305
 306 static void raise_backtrace(cpumask_t *mask)
 307 {
 308         call_single_data_t *csd;
 309         int cpu;
 310
 311         for_each_cpu(cpu, mask) {
 312                 /*
 313                  * If we previously sent an IPI to the target CPU & it hasn't
 314                  * cleared its bit in the busy cpumask then it didn't handle
 315                  * our previous IPI & it's not safe for us to reuse the
 316                  * call_single_data_t.
 317                  */
 318                 if (cpumask_test_and_set_cpu(cpu, &backtrace_csd_busy)) {
 319                         pr_warn("Unable to send backtrace IPI to CPU%u - perhaps it hung?\n",
 320                                 cpu);
 321                         continue;
 322                 }
 323
 324                 csd = &per_cpu(backtrace_csd, cpu);
 325                 csd->func = handle_backtrace;
 326                 smp_call_function_single_async(cpu, csd);
 327         }
 328 }
 329
 330 void arch_trigger_cpumask_backtrace(const cpumask_t *mask, bool exclude_self)
 331 {
 332         nmi_trigger_cpumask_backtrace(mask, exclude_self, raise_backtrace);
 333 }
 334
 335 #ifdef CONFIG_64BIT
 336 void loongarch_dump_regs64(u64 *uregs, const struct pt_regs *regs)
 337 {
 338         unsigned int i;
 339
 340         for (i = LOONGARCH_EF_R1; i <= LOONGARCH_EF_R31; i++) {
 341                 uregs[i] = regs->regs[i - LOONGARCH_EF_R0];
 342         }
 343
 344         uregs[LOONGARCH_EF_ORIG_A0] = regs->orig_a0;
 345         uregs[LOONGARCH_EF_CSR_ERA] = regs->csr_era;
 346         uregs[LOONGARCH_EF_CSR_BADV] = regs->csr_badvaddr;
 347         uregs[LOONGARCH_EF_CSR_CRMD] = regs->csr_crmd;
 348         uregs[LOONGARCH_EF_CSR_PRMD] = regs->csr_prmd;
 349         uregs[LOONGARCH_EF_CSR_EUEN] = regs->csr_euen;
 350         uregs[LOONGARCH_EF_CSR_ECFG] = regs->csr_ecfg;
 351         uregs[LOONGARCH_EF_CSR_ESTAT] = regs->csr_estat;
 352 }
 353 #endif /* CONFIG_64BIT */