]> git.baikalelectronics.ru Git - kernel.git/commitdiff
LoongArch: Add process management
authorHuacai Chen <chenhuacai@loongson.cn>
Tue, 31 May 2022 10:04:11 +0000 (18:04 +0800)
committerHuacai Chen <chenhuacai@loongson.cn>
Fri, 3 Jun 2022 12:09:28 +0000 (20:09 +0800)
Add process management support for LoongArch, including: thread info
definition, context switch and process tracing.

Reviewed-by: WANG Xuerui <git@xen0n.name>
Reviewed-by: Jiaxun Yang <jiaxun.yang@flygoat.com>
Signed-off-by: Huacai Chen <chenhuacai@loongson.cn>
14 files changed:
arch/loongarch/include/asm/fpu.h [new file with mode: 0644]
arch/loongarch/include/asm/idle.h [new file with mode: 0644]
arch/loongarch/include/asm/mmu.h [new file with mode: 0644]
arch/loongarch/include/asm/mmu_context.h [new file with mode: 0644]
arch/loongarch/include/asm/processor.h [new file with mode: 0644]
arch/loongarch/include/asm/ptrace.h [new file with mode: 0644]
arch/loongarch/include/asm/switch_to.h [new file with mode: 0644]
arch/loongarch/include/asm/thread_info.h [new file with mode: 0644]
arch/loongarch/include/uapi/asm/ptrace.h [new file with mode: 0644]
arch/loongarch/kernel/fpu.S [new file with mode: 0644]
arch/loongarch/kernel/idle.c [new file with mode: 0644]
arch/loongarch/kernel/process.c [new file with mode: 0644]
arch/loongarch/kernel/ptrace.c [new file with mode: 0644]
arch/loongarch/kernel/switch.S [new file with mode: 0644]

diff --git a/arch/loongarch/include/asm/fpu.h b/arch/loongarch/include/asm/fpu.h
new file mode 100644 (file)
index 0000000..358b254
--- /dev/null
@@ -0,0 +1,129 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Author: Huacai Chen <chenhuacai@loongson.cn>
+ * Copyright (C) 2020-2022 Loongson Technology Corporation Limited
+ */
+#ifndef _ASM_FPU_H
+#define _ASM_FPU_H
+
+#include <linux/sched.h>
+#include <linux/sched/task_stack.h>
+#include <linux/ptrace.h>
+#include <linux/thread_info.h>
+#include <linux/bitops.h>
+
+#include <asm/cpu.h>
+#include <asm/cpu-features.h>
+#include <asm/current.h>
+#include <asm/loongarch.h>
+#include <asm/processor.h>
+#include <asm/ptrace.h>
+
+struct sigcontext;
+
+extern void _init_fpu(unsigned int);
+extern void _save_fp(struct loongarch_fpu *);
+extern void _restore_fp(struct loongarch_fpu *);
+
+/*
+ * Mask the FCSR Cause bits according to the Enable bits, observing
+ * that Unimplemented is always enabled.
+ */
+static inline unsigned long mask_fcsr_x(unsigned long fcsr)
+{
+       return fcsr & ((fcsr & FPU_CSR_ALL_E) <<
+                       (ffs(FPU_CSR_ALL_X) - ffs(FPU_CSR_ALL_E)));
+}
+
+static inline int is_fp_enabled(void)
+{
+       return (csr_read32(LOONGARCH_CSR_EUEN) & CSR_EUEN_FPEN) ?
+               1 : 0;
+}
+
+#define enable_fpu()           set_csr_euen(CSR_EUEN_FPEN)
+
+#define disable_fpu()          clear_csr_euen(CSR_EUEN_FPEN)
+
+#define clear_fpu_owner()      clear_thread_flag(TIF_USEDFPU)
+
+static inline int is_fpu_owner(void)
+{
+       return test_thread_flag(TIF_USEDFPU);
+}
+
+static inline void __own_fpu(void)
+{
+       enable_fpu();
+       set_thread_flag(TIF_USEDFPU);
+       KSTK_EUEN(current) |= CSR_EUEN_FPEN;
+}
+
+static inline void own_fpu_inatomic(int restore)
+{
+       if (cpu_has_fpu && !is_fpu_owner()) {
+               __own_fpu();
+               if (restore)
+                       _restore_fp(&current->thread.fpu);
+       }
+}
+
+static inline void own_fpu(int restore)
+{
+       preempt_disable();
+       own_fpu_inatomic(restore);
+       preempt_enable();
+}
+
+static inline void lose_fpu_inatomic(int save, struct task_struct *tsk)
+{
+       if (is_fpu_owner()) {
+               if (save)
+                       _save_fp(&tsk->thread.fpu);
+               disable_fpu();
+               clear_tsk_thread_flag(tsk, TIF_USEDFPU);
+       }
+       KSTK_EUEN(tsk) &= ~(CSR_EUEN_FPEN | CSR_EUEN_LSXEN | CSR_EUEN_LASXEN);
+}
+
+static inline void lose_fpu(int save)
+{
+       preempt_disable();
+       lose_fpu_inatomic(save, current);
+       preempt_enable();
+}
+
+static inline void init_fpu(void)
+{
+       unsigned int fcsr = current->thread.fpu.fcsr;
+
+       __own_fpu();
+       _init_fpu(fcsr);
+       set_used_math();
+}
+
+static inline void save_fp(struct task_struct *tsk)
+{
+       if (cpu_has_fpu)
+               _save_fp(&tsk->thread.fpu);
+}
+
+static inline void restore_fp(struct task_struct *tsk)
+{
+       if (cpu_has_fpu)
+               _restore_fp(&tsk->thread.fpu);
+}
+
+static inline union fpureg *get_fpu_regs(struct task_struct *tsk)
+{
+       if (tsk == current) {
+               preempt_disable();
+               if (is_fpu_owner())
+                       _save_fp(&current->thread.fpu);
+               preempt_enable();
+       }
+
+       return tsk->thread.fpu.fpr;
+}
+
+#endif /* _ASM_FPU_H */
diff --git a/arch/loongarch/include/asm/idle.h b/arch/loongarch/include/asm/idle.h
new file mode 100644 (file)
index 0000000..f7f2b7d
--- /dev/null
@@ -0,0 +1,9 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+#ifndef __ASM_IDLE_H
+#define __ASM_IDLE_H
+
+#include <linux/linkage.h>
+
+extern asmlinkage void __arch_cpu_idle(void);
+
+#endif /* __ASM_IDLE_H  */
diff --git a/arch/loongarch/include/asm/mmu.h b/arch/loongarch/include/asm/mmu.h
new file mode 100644 (file)
index 0000000..0cc2d08
--- /dev/null
@@ -0,0 +1,16 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020-2022 Loongson Technology Corporation Limited
+ */
+#ifndef __ASM_MMU_H
+#define __ASM_MMU_H
+
+#include <linux/atomic.h>
+#include <linux/spinlock.h>
+
+typedef struct {
+       u64 asid[NR_CPUS];
+       void *vdso;
+} mm_context_t;
+
+#endif /* __ASM_MMU_H */
diff --git a/arch/loongarch/include/asm/mmu_context.h b/arch/loongarch/include/asm/mmu_context.h
new file mode 100644 (file)
index 0000000..9f97c34
--- /dev/null
@@ -0,0 +1,152 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Switch a MMU context.
+ *
+ * Copyright (C) 2020-2022 Loongson Technology Corporation Limited
+ */
+#ifndef _ASM_MMU_CONTEXT_H
+#define _ASM_MMU_CONTEXT_H
+
+#include <linux/errno.h>
+#include <linux/sched.h>
+#include <linux/mm_types.h>
+#include <linux/smp.h>
+#include <linux/slab.h>
+
+#include <asm/cacheflush.h>
+#include <asm/tlbflush.h>
+#include <asm-generic/mm_hooks.h>
+
+/*
+ *  All unused by hardware upper bits will be considered
+ *  as a software asid extension.
+ */
+static inline u64 asid_version_mask(unsigned int cpu)
+{
+       return ~(u64)(cpu_asid_mask(&cpu_data[cpu]));
+}
+
+static inline u64 asid_first_version(unsigned int cpu)
+{
+       return cpu_asid_mask(&cpu_data[cpu]) + 1;
+}
+
+#define cpu_context(cpu, mm)   ((mm)->context.asid[cpu])
+#define asid_cache(cpu)                (cpu_data[cpu].asid_cache)
+#define cpu_asid(cpu, mm)      (cpu_context((cpu), (mm)) & cpu_asid_mask(&cpu_data[cpu]))
+
+static inline int asid_valid(struct mm_struct *mm, unsigned int cpu)
+{
+       if ((cpu_context(cpu, mm) ^ asid_cache(cpu)) & asid_version_mask(cpu))
+               return 0;
+
+       return 1;
+}
+
+static inline void enter_lazy_tlb(struct mm_struct *mm, struct task_struct *tsk)
+{
+}
+
+/* Normal, classic get_new_mmu_context */
+static inline void
+get_new_mmu_context(struct mm_struct *mm, unsigned long cpu)
+{
+       u64 asid = asid_cache(cpu);
+
+       if (!((++asid) & cpu_asid_mask(&cpu_data[cpu])))
+               local_flush_tlb_user(); /* start new asid cycle */
+
+       cpu_context(cpu, mm) = asid_cache(cpu) = asid;
+}
+
+/*
+ * Initialize the context related info for a new mm_struct
+ * instance.
+ */
+static inline int
+init_new_context(struct task_struct *tsk, struct mm_struct *mm)
+{
+       int i;
+
+       for_each_possible_cpu(i)
+               cpu_context(i, mm) = 0;
+
+       return 0;
+}
+
+static inline void switch_mm_irqs_off(struct mm_struct *prev, struct mm_struct *next,
+                                     struct task_struct *tsk)
+{
+       unsigned int cpu = smp_processor_id();
+
+       /* Check if our ASID is of an older version and thus invalid */
+       if (!asid_valid(next, cpu))
+               get_new_mmu_context(next, cpu);
+
+       write_csr_asid(cpu_asid(cpu, next));
+
+       if (next != &init_mm)
+               csr_write64((unsigned long)next->pgd, LOONGARCH_CSR_PGDL);
+       else
+               csr_write64((unsigned long)invalid_pg_dir, LOONGARCH_CSR_PGDL);
+
+       /*
+        * Mark current->active_mm as not "active" anymore.
+        * We don't want to mislead possible IPI tlb flush routines.
+        */
+       cpumask_set_cpu(cpu, mm_cpumask(next));
+}
+
+#define switch_mm_irqs_off switch_mm_irqs_off
+
+static inline void switch_mm(struct mm_struct *prev, struct mm_struct *next,
+                            struct task_struct *tsk)
+{
+       unsigned long flags;
+
+       local_irq_save(flags);
+       switch_mm_irqs_off(prev, next, tsk);
+       local_irq_restore(flags);
+}
+
+/*
+ * Destroy context related info for an mm_struct that is about
+ * to be put to rest.
+ */
+static inline void destroy_context(struct mm_struct *mm)
+{
+}
+
+#define activate_mm(prev, next)        switch_mm(prev, next, current)
+#define deactivate_mm(task, mm)        do { } while (0)
+
+/*
+ * If mm is currently active, we can't really drop it.
+ * Instead, we will get a new one for it.
+ */
+static inline void
+drop_mmu_context(struct mm_struct *mm, unsigned int cpu)
+{
+       int asid;
+       unsigned long flags;
+
+       local_irq_save(flags);
+
+       asid = read_csr_asid() & cpu_asid_mask(&current_cpu_data);
+
+       if (asid == cpu_asid(cpu, mm)) {
+               if (!current->mm || (current->mm == mm)) {
+                       get_new_mmu_context(mm, cpu);
+                       write_csr_asid(cpu_asid(cpu, mm));
+                       goto out;
+               }
+       }
+
+       /* Will get a new context next time */
+       cpu_context(cpu, mm) = 0;
+       cpumask_clear_cpu(cpu, mm_cpumask(mm));
+out:
+       local_irq_restore(flags);
+}
+
+#endif /* _ASM_MMU_CONTEXT_H */
diff --git a/arch/loongarch/include/asm/processor.h b/arch/loongarch/include/asm/processor.h
new file mode 100644 (file)
index 0000000..1d63c93
--- /dev/null
@@ -0,0 +1,209 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020-2022 Loongson Technology Corporation Limited
+ */
+#ifndef _ASM_PROCESSOR_H
+#define _ASM_PROCESSOR_H
+
+#include <linux/atomic.h>
+#include <linux/cpumask.h>
+#include <linux/sizes.h>
+
+#include <asm/cpu.h>
+#include <asm/cpu-info.h>
+#include <asm/loongarch.h>
+#include <asm/vdso/processor.h>
+#include <uapi/asm/ptrace.h>
+#include <uapi/asm/sigcontext.h>
+
+#ifdef CONFIG_32BIT
+
+#define TASK_SIZE      0x80000000UL
+#define TASK_SIZE_MIN  TASK_SIZE
+#define STACK_TOP_MAX  TASK_SIZE
+
+#define TASK_IS_32BIT_ADDR 1
+
+#endif
+
+#ifdef CONFIG_64BIT
+
+#define TASK_SIZE32    0x100000000UL
+#define TASK_SIZE64     (0x1UL << ((cpu_vabits > VA_BITS) ? VA_BITS : cpu_vabits))
+
+#define TASK_SIZE      (test_thread_flag(TIF_32BIT_ADDR) ? TASK_SIZE32 : TASK_SIZE64)
+#define TASK_SIZE_MIN  TASK_SIZE32
+#define STACK_TOP_MAX  TASK_SIZE64
+
+#define TASK_SIZE_OF(tsk)                                              \
+       (test_tsk_thread_flag(tsk, TIF_32BIT_ADDR) ? TASK_SIZE32 : TASK_SIZE64)
+
+#define TASK_IS_32BIT_ADDR test_thread_flag(TIF_32BIT_ADDR)
+
+#endif
+
+#define VDSO_RANDOMIZE_SIZE    (TASK_IS_32BIT_ADDR ? SZ_1M : SZ_64M)
+
+unsigned long stack_top(void);
+#define STACK_TOP stack_top()
+
+/*
+ * This decides where the kernel will search for a free chunk of vm
+ * space during mmap's.
+ */
+#define TASK_UNMAPPED_BASE PAGE_ALIGN(TASK_SIZE / 3)
+
+#define FPU_REG_WIDTH          256
+#define FPU_ALIGN              __attribute__((aligned(32)))
+
+union fpureg {
+       __u32   val32[FPU_REG_WIDTH / 32];
+       __u64   val64[FPU_REG_WIDTH / 64];
+};
+
+#define FPR_IDX(width, idx)    (idx)
+
+#define BUILD_FPR_ACCESS(width) \
+static inline u##width get_fpr##width(union fpureg *fpr, unsigned idx) \
+{                                                                      \
+       return fpr->val##width[FPR_IDX(width, idx)];                    \
+}                                                                      \
+                                                                       \
+static inline void set_fpr##width(union fpureg *fpr, unsigned int idx, \
+                                 u##width val)                         \
+{                                                                      \
+       fpr->val##width[FPR_IDX(width, idx)] = val;                     \
+}
+
+BUILD_FPR_ACCESS(32)
+BUILD_FPR_ACCESS(64)
+
+struct loongarch_fpu {
+       unsigned int    fcsr;
+       unsigned int    vcsr;
+       uint64_t        fcc;    /* 8x8 */
+       union fpureg    fpr[NUM_FPU_REGS];
+};
+
+#define INIT_CPUMASK { \
+       {0,} \
+}
+
+#define ARCH_MIN_TASKALIGN     32
+
+struct loongarch_vdso_info;
+
+/*
+ * If you change thread_struct remember to change the #defines below too!
+ */
+struct thread_struct {
+       /* Main processor registers. */
+       unsigned long reg01, reg03, reg22; /* ra sp fp */
+       unsigned long reg23, reg24, reg25, reg26; /* s0-s3 */
+       unsigned long reg27, reg28, reg29, reg30, reg31; /* s4-s8 */
+
+       /* CSR registers */
+       unsigned long csr_prmd;
+       unsigned long csr_crmd;
+       unsigned long csr_euen;
+       unsigned long csr_ecfg;
+       unsigned long csr_badvaddr;     /* Last user fault */
+
+       /* Scratch registers */
+       unsigned long scr0;
+       unsigned long scr1;
+       unsigned long scr2;
+       unsigned long scr3;
+
+       /* Eflags register */
+       unsigned long eflags;
+
+       /* Other stuff associated with the thread. */
+       unsigned long trap_nr;
+       unsigned long error_code;
+       struct loongarch_vdso_info *vdso;
+
+       /*
+        * FPU & vector registers, must be at last because
+        * they are conditionally copied at fork().
+        */
+       struct loongarch_fpu fpu FPU_ALIGN;
+};
+
+#define INIT_THREAD  {                                         \
+       /*                                                      \
+        * Main processor registers                             \
+        */                                                     \
+       .reg01                  = 0,                            \
+       .reg03                  = 0,                            \
+       .reg22                  = 0,                            \
+       .reg23                  = 0,                            \
+       .reg24                  = 0,                            \
+       .reg25                  = 0,                            \
+       .reg26                  = 0,                            \
+       .reg27                  = 0,                            \
+       .reg28                  = 0,                            \
+       .reg29                  = 0,                            \
+       .reg30                  = 0,                            \
+       .reg31                  = 0,                            \
+       .csr_crmd               = 0,                            \
+       .csr_prmd               = 0,                            \
+       .csr_euen               = 0,                            \
+       .csr_ecfg               = 0,                            \
+       .csr_badvaddr           = 0,                            \
+       /*                                                      \
+        * Other stuff associated with the process              \
+        */                                                     \
+       .trap_nr                = 0,                            \
+       .error_code             = 0,                            \
+       /*                                                      \
+        * FPU & vector registers                               \
+        */                                                     \
+       .fpu                    = {                             \
+               .fcsr           = 0,                            \
+               .vcsr           = 0,                            \
+               .fcc            = 0,                            \
+               .fpr            = {{{0,},},},                   \
+       },                                                      \
+}
+
+struct task_struct;
+
+/* Free all resources held by a thread. */
+#define release_thread(thread) do { } while (0)
+
+enum idle_boot_override {IDLE_NO_OVERRIDE = 0, IDLE_HALT, IDLE_NOMWAIT, IDLE_POLL};
+
+extern unsigned long           boot_option_idle_override;
+/*
+ * Do necessary setup to start up a newly executed thread.
+ */
+extern void start_thread(struct pt_regs *regs, unsigned long pc, unsigned long sp);
+
+static inline void flush_thread(void)
+{
+}
+
+unsigned long __get_wchan(struct task_struct *p);
+
+#define __KSTK_TOS(tsk) ((unsigned long)task_stack_page(tsk) + \
+                        THREAD_SIZE - 32 - sizeof(struct pt_regs))
+#define task_pt_regs(tsk) ((struct pt_regs *)__KSTK_TOS(tsk))
+#define KSTK_EIP(tsk) (task_pt_regs(tsk)->csr_era)
+#define KSTK_ESP(tsk) (task_pt_regs(tsk)->regs[3])
+#define KSTK_EUEN(tsk) (task_pt_regs(tsk)->csr_euen)
+#define KSTK_ECFG(tsk) (task_pt_regs(tsk)->csr_ecfg)
+
+#define return_address() ({__asm__ __volatile__("":::"$1"); __builtin_return_address(0);})
+
+#ifdef CONFIG_CPU_HAS_PREFETCH
+
+#define ARCH_HAS_PREFETCH
+#define prefetch(x) __builtin_prefetch((x), 0, 1)
+
+#define ARCH_HAS_PREFETCHW
+#define prefetchw(x) __builtin_prefetch((x), 1, 1)
+
+#endif
+
+#endif /* _ASM_PROCESSOR_H */
diff --git a/arch/loongarch/include/asm/ptrace.h b/arch/loongarch/include/asm/ptrace.h
new file mode 100644 (file)
index 0000000..17838c6
--- /dev/null
@@ -0,0 +1,152 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020-2022 Loongson Technology Corporation Limited
+ */
+#ifndef _ASM_PTRACE_H
+#define _ASM_PTRACE_H
+
+#include <asm/page.h>
+#include <asm/thread_info.h>
+#include <uapi/asm/ptrace.h>
+
+/*
+ * This struct defines the way the registers are stored on the stack during
+ * a system call/exception. If you add a register here, please also add it to
+ * regoffset_table[] in arch/loongarch/kernel/ptrace.c.
+ */
+struct pt_regs {
+       /* Main processor registers. */
+       unsigned long regs[32];
+
+       /* Original syscall arg0. */
+       unsigned long orig_a0;
+
+       /* Special CSR registers. */
+       unsigned long csr_era;
+       unsigned long csr_badvaddr;
+       unsigned long csr_crmd;
+       unsigned long csr_prmd;
+       unsigned long csr_euen;
+       unsigned long csr_ecfg;
+       unsigned long csr_estat;
+       unsigned long __last[0];
+} __aligned(8);
+
+static inline int regs_irqs_disabled(struct pt_regs *regs)
+{
+       return arch_irqs_disabled_flags(regs->csr_prmd);
+}
+
+static inline unsigned long kernel_stack_pointer(struct pt_regs *regs)
+{
+       return regs->regs[3];
+}
+
+/*
+ * Don't use asm-generic/ptrace.h it defines FP accessors that don't make
+ * sense on LoongArch.  We rather want an error if they get invoked.
+ */
+
+static inline void instruction_pointer_set(struct pt_regs *regs, unsigned long val)
+{
+       regs->csr_era = val;
+}
+
+/* Query offset/name of register from its name/offset */
+extern int regs_query_register_offset(const char *name);
+#define MAX_REG_OFFSET (offsetof(struct pt_regs, __last))
+
+/**
+ * regs_get_register() - get register value from its offset
+ * @regs:       pt_regs from which register value is gotten.
+ * @offset:     offset number of the register.
+ *
+ * regs_get_register returns the value of a register. The @offset is the
+ * offset of the register in struct pt_regs address which specified by @regs.
+ * If @offset is bigger than MAX_REG_OFFSET, this returns 0.
+ */
+static inline unsigned long regs_get_register(struct pt_regs *regs, unsigned int offset)
+{
+       if (unlikely(offset > MAX_REG_OFFSET))
+               return 0;
+
+       return *(unsigned long *)((unsigned long)regs + offset);
+}
+
+/**
+ * regs_within_kernel_stack() - check the address in the stack
+ * @regs:       pt_regs which contains kernel stack pointer.
+ * @addr:       address which is checked.
+ *
+ * regs_within_kernel_stack() checks @addr is within the kernel stack page(s).
+ * If @addr is within the kernel stack, it returns true. If not, returns false.
+ */
+static inline int regs_within_kernel_stack(struct pt_regs *regs, unsigned long addr)
+{
+       return ((addr & ~(THREAD_SIZE - 1))  ==
+               (kernel_stack_pointer(regs) & ~(THREAD_SIZE - 1)));
+}
+
+/**
+ * regs_get_kernel_stack_nth() - get Nth entry of the stack
+ * @regs:       pt_regs which contains kernel stack pointer.
+ * @n:          stack entry number.
+ *
+ * regs_get_kernel_stack_nth() returns @n th entry of the kernel stack which
+ * is specified by @regs. If the @n th entry is NOT in the kernel stack,
+ * this returns 0.
+ */
+static inline unsigned long regs_get_kernel_stack_nth(struct pt_regs *regs, unsigned int n)
+{
+       unsigned long *addr = (unsigned long *)kernel_stack_pointer(regs);
+
+       addr += n;
+       if (regs_within_kernel_stack(regs, (unsigned long)addr))
+               return *addr;
+       else
+               return 0;
+}
+
+struct task_struct;
+
+/*
+ * Does the process account for user or for system time?
+ */
+#define user_mode(regs) (((regs)->csr_prmd & PLV_MASK) == PLV_USER)
+
+static inline long regs_return_value(struct pt_regs *regs)
+{
+       return regs->regs[4];
+}
+
+#define instruction_pointer(regs) ((regs)->csr_era)
+#define profile_pc(regs) instruction_pointer(regs)
+
+extern void die(const char *, struct pt_regs *) __noreturn;
+
+static inline void die_if_kernel(const char *str, struct pt_regs *regs)
+{
+       if (unlikely(!user_mode(regs)))
+               die(str, regs);
+}
+
+#define current_pt_regs()                                              \
+({                                                                     \
+       unsigned long sp = (unsigned long)__builtin_frame_address(0);   \
+       (struct pt_regs *)((sp | (THREAD_SIZE - 1)) + 1 - 32) - 1;      \
+})
+
+/* Helpers for working with the user stack pointer */
+
+static inline unsigned long user_stack_pointer(struct pt_regs *regs)
+{
+       return regs->regs[3];
+}
+
+static inline void user_stack_pointer_set(struct pt_regs *regs,
+       unsigned long val)
+{
+       regs->regs[3] = val;
+}
+
+#endif /* _ASM_PTRACE_H */
diff --git a/arch/loongarch/include/asm/switch_to.h b/arch/loongarch/include/asm/switch_to.h
new file mode 100644 (file)
index 0000000..2a8d043
--- /dev/null
@@ -0,0 +1,37 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020-2022 Loongson Technology Corporation Limited
+ */
+#ifndef _ASM_SWITCH_TO_H
+#define _ASM_SWITCH_TO_H
+
+#include <asm/cpu-features.h>
+#include <asm/fpu.h>
+
+struct task_struct;
+
+/**
+ * __switch_to - switch execution of a task
+ * @prev:      The task previously executed.
+ * @next:      The task to begin executing.
+ * @next_ti:   task_thread_info(next).
+ *
+ * This function is used whilst scheduling to save the context of prev & load
+ * the context of next. Returns prev.
+ */
+extern asmlinkage struct task_struct *__switch_to(struct task_struct *prev,
+                       struct task_struct *next, struct thread_info *next_ti);
+
+/*
+ * For newly created kernel threads switch_to() will return to
+ * ret_from_kernel_thread, newly created user threads to ret_from_fork.
+ * That is, everything following __switch_to() will be skipped for new threads.
+ * So everything that matters to new threads should be placed before __switch_to().
+ */
+#define switch_to(prev, next, last)                                    \
+do {                                                                   \
+       lose_fpu_inatomic(1, prev);                                     \
+       (last) = __switch_to(prev, next, task_thread_info(next));       \
+} while (0)
+
+#endif /* _ASM_SWITCH_TO_H */
diff --git a/arch/loongarch/include/asm/thread_info.h b/arch/loongarch/include/asm/thread_info.h
new file mode 100644 (file)
index 0000000..99beb11
--- /dev/null
@@ -0,0 +1,106 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * thread_info.h: LoongArch low-level thread information
+ *
+ * Copyright (C) 2020-2022 Loongson Technology Corporation Limited
+ */
+
+#ifndef _ASM_THREAD_INFO_H
+#define _ASM_THREAD_INFO_H
+
+#ifdef __KERNEL__
+
+#ifndef __ASSEMBLY__
+
+#include <asm/processor.h>
+
+/*
+ * low level task data that entry.S needs immediate access to
+ * - this struct should fit entirely inside of one cache line
+ * - this struct shares the supervisor stack pages
+ * - if the contents of this structure are changed, the assembly constants
+ *   must also be changed
+ */
+struct thread_info {
+       struct task_struct      *task;          /* main task structure */
+       unsigned long           flags;          /* low level flags */
+       unsigned long           tp_value;       /* thread pointer */
+       __u32                   cpu;            /* current CPU */
+       int                     preempt_count;  /* 0 => preemptible, <0 => BUG */
+       struct pt_regs          *regs;
+       unsigned long           syscall;        /* syscall number */
+       unsigned long           syscall_work;   /* SYSCALL_WORK_ flags */
+};
+
+/*
+ * macros/functions for gaining access to the thread information structure
+ */
+#define INIT_THREAD_INFO(tsk)                  \
+{                                              \
+       .task           = &tsk,                 \
+       .flags          = 0,                    \
+       .cpu            = 0,                    \
+       .preempt_count  = INIT_PREEMPT_COUNT,   \
+}
+
+/* How to get the thread information struct from C. */
+register struct thread_info *__current_thread_info __asm__("$r2");
+
+static inline struct thread_info *current_thread_info(void)
+{
+       return __current_thread_info;
+}
+
+register unsigned long current_stack_pointer __asm__("$r3");
+
+#endif /* !__ASSEMBLY__ */
+
+/* thread information allocation */
+#define THREAD_SIZE            SZ_16K
+#define THREAD_MASK            (THREAD_SIZE - 1UL)
+#define THREAD_SIZE_ORDER      ilog2(THREAD_SIZE / PAGE_SIZE)
+/*
+ * thread information flags
+ * - these are process state flags that various assembly files may need to
+ *   access
+ * - pending work-to-be-done flags are in LSW
+ * - other flags in MSW
+ */
+#define TIF_SIGPENDING         1       /* signal pending */
+#define TIF_NEED_RESCHED       2       /* rescheduling necessary */
+#define TIF_NOTIFY_RESUME      3       /* callback before returning to user */
+#define TIF_NOTIFY_SIGNAL      4       /* signal notifications exist */
+#define TIF_RESTORE_SIGMASK    5       /* restore signal mask in do_signal() */
+#define TIF_NOHZ               6       /* in adaptive nohz mode */
+#define TIF_UPROBE             7       /* breakpointed or singlestepping */
+#define TIF_USEDFPU            8       /* FPU was used by this task this quantum (SMP) */
+#define TIF_USEDSIMD           9       /* SIMD has been used this quantum */
+#define TIF_MEMDIE             10      /* is terminating due to OOM killer */
+#define TIF_FIXADE             11      /* Fix address errors in software */
+#define TIF_LOGADE             12      /* Log address errors to syslog */
+#define TIF_32BIT_REGS         13      /* 32-bit general purpose registers */
+#define TIF_32BIT_ADDR         14      /* 32-bit address space */
+#define TIF_LOAD_WATCH         15      /* If set, load watch registers */
+#define TIF_SINGLESTEP         16      /* Single Step */
+#define TIF_LSX_CTX_LIVE       17      /* LSX context must be preserved */
+#define TIF_LASX_CTX_LIVE      18      /* LASX context must be preserved */
+
+#define _TIF_SIGPENDING                (1<<TIF_SIGPENDING)
+#define _TIF_NEED_RESCHED      (1<<TIF_NEED_RESCHED)
+#define _TIF_NOTIFY_RESUME     (1<<TIF_NOTIFY_RESUME)
+#define _TIF_NOTIFY_SIGNAL     (1<<TIF_NOTIFY_SIGNAL)
+#define _TIF_NOHZ              (1<<TIF_NOHZ)
+#define _TIF_UPROBE            (1<<TIF_UPROBE)
+#define _TIF_USEDFPU           (1<<TIF_USEDFPU)
+#define _TIF_USEDSIMD          (1<<TIF_USEDSIMD)
+#define _TIF_FIXADE            (1<<TIF_FIXADE)
+#define _TIF_LOGADE            (1<<TIF_LOGADE)
+#define _TIF_32BIT_REGS                (1<<TIF_32BIT_REGS)
+#define _TIF_32BIT_ADDR                (1<<TIF_32BIT_ADDR)
+#define _TIF_LOAD_WATCH                (1<<TIF_LOAD_WATCH)
+#define _TIF_SINGLESTEP                (1<<TIF_SINGLESTEP)
+#define _TIF_LSX_CTX_LIVE      (1<<TIF_LSX_CTX_LIVE)
+#define _TIF_LASX_CTX_LIVE     (1<<TIF_LASX_CTX_LIVE)
+
+#endif /* __KERNEL__ */
+#endif /* _ASM_THREAD_INFO_H */
diff --git a/arch/loongarch/include/uapi/asm/ptrace.h b/arch/loongarch/include/uapi/asm/ptrace.h
new file mode 100644 (file)
index 0000000..083193f
--- /dev/null
@@ -0,0 +1,52 @@
+/* SPDX-License-Identifier: GPL-2.0+ WITH Linux-syscall-note */
+/*
+ * Author: Hanlu Li <lihanlu@loongson.cn>
+ *         Huacai Chen <chenhuacai@loongson.cn>
+ *
+ * Copyright (C) 2020-2022 Loongson Technology Corporation Limited
+ */
+#ifndef _UAPI_ASM_PTRACE_H
+#define _UAPI_ASM_PTRACE_H
+
+#include <linux/types.h>
+
+#ifndef __KERNEL__
+#include <stdint.h>
+#endif
+
+/*
+ * For PTRACE_{POKE,PEEK}USR. 0 - 31 are GPRs,
+ * 32 is syscall's original ARG0, 33 is PC, 34 is BADVADDR.
+ */
+#define GPR_BASE       0
+#define GPR_NUM                32
+#define GPR_END                (GPR_BASE + GPR_NUM - 1)
+#define ARG0           (GPR_END + 1)
+#define PC             (GPR_END + 2)
+#define BADVADDR       (GPR_END + 3)
+
+#define NUM_FPU_REGS   32
+
+struct user_pt_regs {
+       /* Main processor registers. */
+       unsigned long regs[32];
+
+       /* Original syscall arg0. */
+       unsigned long orig_a0;
+
+       /* Special CSR registers. */
+       unsigned long csr_era;
+       unsigned long csr_badv;
+       unsigned long reserved[10];
+} __attribute__((aligned(8)));
+
+struct user_fp_state {
+       uint64_t    fpr[32];
+       uint64_t    fcc;
+       uint32_t    fcsr;
+};
+
+#define PTRACE_SYSEMU                  0x1f
+#define PTRACE_SYSEMU_SINGLESTEP       0x20
+
+#endif /* _UAPI_ASM_PTRACE_H */
diff --git a/arch/loongarch/kernel/fpu.S b/arch/loongarch/kernel/fpu.S
new file mode 100644 (file)
index 0000000..75c6ce0
--- /dev/null
@@ -0,0 +1,261 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Author: Lu Zeng <zenglu@loongson.cn>
+ *         Pei Huang <huangpei@loongson.cn>
+ *         Huacai Chen <chenhuacai@loongson.cn>
+ *
+ * Copyright (C) 2020-2022 Loongson Technology Corporation Limited
+ */
+#include <asm/asm.h>
+#include <asm/asmmacro.h>
+#include <asm/asm-offsets.h>
+#include <asm/errno.h>
+#include <asm/export.h>
+#include <asm/fpregdef.h>
+#include <asm/loongarch.h>
+#include <asm/regdef.h>
+
+#define FPU_REG_WIDTH          8
+#define LSX_REG_WIDTH          16
+#define LASX_REG_WIDTH         32
+
+       .macro  EX insn, reg, src, offs
+.ex\@: \insn   \reg, \src, \offs
+       .section __ex_table,"a"
+       PTR     .ex\@, fault
+       .previous
+       .endm
+
+       .macro sc_save_fp base
+       EX      fst.d $f0,  \base, (0 * FPU_REG_WIDTH)
+       EX      fst.d $f1,  \base, (1 * FPU_REG_WIDTH)
+       EX      fst.d $f2,  \base, (2 * FPU_REG_WIDTH)
+       EX      fst.d $f3,  \base, (3 * FPU_REG_WIDTH)
+       EX      fst.d $f4,  \base, (4 * FPU_REG_WIDTH)
+       EX      fst.d $f5,  \base, (5 * FPU_REG_WIDTH)
+       EX      fst.d $f6,  \base, (6 * FPU_REG_WIDTH)
+       EX      fst.d $f7,  \base, (7 * FPU_REG_WIDTH)
+       EX      fst.d $f8,  \base, (8 * FPU_REG_WIDTH)
+       EX      fst.d $f9,  \base, (9 * FPU_REG_WIDTH)
+       EX      fst.d $f10, \base, (10 * FPU_REG_WIDTH)
+       EX      fst.d $f11, \base, (11 * FPU_REG_WIDTH)
+       EX      fst.d $f12, \base, (12 * FPU_REG_WIDTH)
+       EX      fst.d $f13, \base, (13 * FPU_REG_WIDTH)
+       EX      fst.d $f14, \base, (14 * FPU_REG_WIDTH)
+       EX      fst.d $f15, \base, (15 * FPU_REG_WIDTH)
+       EX      fst.d $f16, \base, (16 * FPU_REG_WIDTH)
+       EX      fst.d $f17, \base, (17 * FPU_REG_WIDTH)
+       EX      fst.d $f18, \base, (18 * FPU_REG_WIDTH)
+       EX      fst.d $f19, \base, (19 * FPU_REG_WIDTH)
+       EX      fst.d $f20, \base, (20 * FPU_REG_WIDTH)
+       EX      fst.d $f21, \base, (21 * FPU_REG_WIDTH)
+       EX      fst.d $f22, \base, (22 * FPU_REG_WIDTH)
+       EX      fst.d $f23, \base, (23 * FPU_REG_WIDTH)
+       EX      fst.d $f24, \base, (24 * FPU_REG_WIDTH)
+       EX      fst.d $f25, \base, (25 * FPU_REG_WIDTH)
+       EX      fst.d $f26, \base, (26 * FPU_REG_WIDTH)
+       EX      fst.d $f27, \base, (27 * FPU_REG_WIDTH)
+       EX      fst.d $f28, \base, (28 * FPU_REG_WIDTH)
+       EX      fst.d $f29, \base, (29 * FPU_REG_WIDTH)
+       EX      fst.d $f30, \base, (30 * FPU_REG_WIDTH)
+       EX      fst.d $f31, \base, (31 * FPU_REG_WIDTH)
+       .endm
+
+       .macro sc_restore_fp base
+       EX      fld.d $f0,  \base, (0 * FPU_REG_WIDTH)
+       EX      fld.d $f1,  \base, (1 * FPU_REG_WIDTH)
+       EX      fld.d $f2,  \base, (2 * FPU_REG_WIDTH)
+       EX      fld.d $f3,  \base, (3 * FPU_REG_WIDTH)
+       EX      fld.d $f4,  \base, (4 * FPU_REG_WIDTH)
+       EX      fld.d $f5,  \base, (5 * FPU_REG_WIDTH)
+       EX      fld.d $f6,  \base, (6 * FPU_REG_WIDTH)
+       EX      fld.d $f7,  \base, (7 * FPU_REG_WIDTH)
+       EX      fld.d $f8,  \base, (8 * FPU_REG_WIDTH)
+       EX      fld.d $f9,  \base, (9 * FPU_REG_WIDTH)
+       EX      fld.d $f10, \base, (10 * FPU_REG_WIDTH)
+       EX      fld.d $f11, \base, (11 * FPU_REG_WIDTH)
+       EX      fld.d $f12, \base, (12 * FPU_REG_WIDTH)
+       EX      fld.d $f13, \base, (13 * FPU_REG_WIDTH)
+       EX      fld.d $f14, \base, (14 * FPU_REG_WIDTH)
+       EX      fld.d $f15, \base, (15 * FPU_REG_WIDTH)
+       EX      fld.d $f16, \base, (16 * FPU_REG_WIDTH)
+       EX      fld.d $f17, \base, (17 * FPU_REG_WIDTH)
+       EX      fld.d $f18, \base, (18 * FPU_REG_WIDTH)
+       EX      fld.d $f19, \base, (19 * FPU_REG_WIDTH)
+       EX      fld.d $f20, \base, (20 * FPU_REG_WIDTH)
+       EX      fld.d $f21, \base, (21 * FPU_REG_WIDTH)
+       EX      fld.d $f22, \base, (22 * FPU_REG_WIDTH)
+       EX      fld.d $f23, \base, (23 * FPU_REG_WIDTH)
+       EX      fld.d $f24, \base, (24 * FPU_REG_WIDTH)
+       EX      fld.d $f25, \base, (25 * FPU_REG_WIDTH)
+       EX      fld.d $f26, \base, (26 * FPU_REG_WIDTH)
+       EX      fld.d $f27, \base, (27 * FPU_REG_WIDTH)
+       EX      fld.d $f28, \base, (28 * FPU_REG_WIDTH)
+       EX      fld.d $f29, \base, (29 * FPU_REG_WIDTH)
+       EX      fld.d $f30, \base, (30 * FPU_REG_WIDTH)
+       EX      fld.d $f31, \base, (31 * FPU_REG_WIDTH)
+       .endm
+
+       .macro sc_save_fcc base, tmp0, tmp1
+       movcf2gr        \tmp0, $fcc0
+       move    \tmp1, \tmp0
+       movcf2gr        \tmp0, $fcc1
+       bstrins.d       \tmp1, \tmp0, 15, 8
+       movcf2gr        \tmp0, $fcc2
+       bstrins.d       \tmp1, \tmp0, 23, 16
+       movcf2gr        \tmp0, $fcc3
+       bstrins.d       \tmp1, \tmp0, 31, 24
+       movcf2gr        \tmp0, $fcc4
+       bstrins.d       \tmp1, \tmp0, 39, 32
+       movcf2gr        \tmp0, $fcc5
+       bstrins.d       \tmp1, \tmp0, 47, 40
+       movcf2gr        \tmp0, $fcc6
+       bstrins.d       \tmp1, \tmp0, 55, 48
+       movcf2gr        \tmp0, $fcc7
+       bstrins.d       \tmp1, \tmp0, 63, 56
+       EX      st.d \tmp1, \base, 0
+       .endm
+
+       .macro sc_restore_fcc base, tmp0, tmp1
+       EX      ld.d \tmp0, \base, 0
+       bstrpick.d      \tmp1, \tmp0, 7, 0
+       movgr2cf        $fcc0, \tmp1
+       bstrpick.d      \tmp1, \tmp0, 15, 8
+       movgr2cf        $fcc1, \tmp1
+       bstrpick.d      \tmp1, \tmp0, 23, 16
+       movgr2cf        $fcc2, \tmp1
+       bstrpick.d      \tmp1, \tmp0, 31, 24
+       movgr2cf        $fcc3, \tmp1
+       bstrpick.d      \tmp1, \tmp0, 39, 32
+       movgr2cf        $fcc4, \tmp1
+       bstrpick.d      \tmp1, \tmp0, 47, 40
+       movgr2cf        $fcc5, \tmp1
+       bstrpick.d      \tmp1, \tmp0, 55, 48
+       movgr2cf        $fcc6, \tmp1
+       bstrpick.d      \tmp1, \tmp0, 63, 56
+       movgr2cf        $fcc7, \tmp1
+       .endm
+
+       .macro sc_save_fcsr base, tmp0
+       movfcsr2gr      \tmp0, fcsr0
+       EX      st.w \tmp0, \base, 0
+       .endm
+
+       .macro sc_restore_fcsr base, tmp0
+       EX      ld.w \tmp0, \base, 0
+       movgr2fcsr      fcsr0, \tmp0
+       .endm
+
+       .macro sc_save_vcsr base, tmp0
+       movfcsr2gr      \tmp0, vcsr16
+       EX      st.w \tmp0, \base, 0
+       .endm
+
+       .macro sc_restore_vcsr base, tmp0
+       EX      ld.w \tmp0, \base, 0
+       movgr2fcsr      vcsr16, \tmp0
+       .endm
+
+/*
+ * Save a thread's fp context.
+ */
+SYM_FUNC_START(_save_fp)
+       fpu_save_csr    a0 t1
+       fpu_save_double a0 t1                   # clobbers t1
+       fpu_save_cc     a0 t1 t2                # clobbers t1, t2
+       jirl zero, ra, 0
+SYM_FUNC_END(_save_fp)
+EXPORT_SYMBOL(_save_fp)
+
+/*
+ * Restore a thread's fp context.
+ */
+SYM_FUNC_START(_restore_fp)
+       fpu_restore_double a0 t1                # clobbers t1
+       fpu_restore_csr a0 t1
+       fpu_restore_cc  a0 t1 t2                # clobbers t1, t2
+       jirl zero, ra, 0
+SYM_FUNC_END(_restore_fp)
+
+/*
+ * Load the FPU with signalling NANS.  This bit pattern we're using has
+ * the property that no matter whether considered as single or as double
+ * precision represents signaling NANS.
+ *
+ * The value to initialize fcsr0 to comes in $a0.
+ */
+
+SYM_FUNC_START(_init_fpu)
+       li.w    t1, CSR_EUEN_FPEN
+       csrxchg t1, t1, LOONGARCH_CSR_EUEN
+
+       movgr2fcsr      fcsr0, a0
+
+       li.w    t1, -1                          # SNaN
+
+       movgr2fr.d      $f0, t1
+       movgr2fr.d      $f1, t1
+       movgr2fr.d      $f2, t1
+       movgr2fr.d      $f3, t1
+       movgr2fr.d      $f4, t1
+       movgr2fr.d      $f5, t1
+       movgr2fr.d      $f6, t1
+       movgr2fr.d      $f7, t1
+       movgr2fr.d      $f8, t1
+       movgr2fr.d      $f9, t1
+       movgr2fr.d      $f10, t1
+       movgr2fr.d      $f11, t1
+       movgr2fr.d      $f12, t1
+       movgr2fr.d      $f13, t1
+       movgr2fr.d      $f14, t1
+       movgr2fr.d      $f15, t1
+       movgr2fr.d      $f16, t1
+       movgr2fr.d      $f17, t1
+       movgr2fr.d      $f18, t1
+       movgr2fr.d      $f19, t1
+       movgr2fr.d      $f20, t1
+       movgr2fr.d      $f21, t1
+       movgr2fr.d      $f22, t1
+       movgr2fr.d      $f23, t1
+       movgr2fr.d      $f24, t1
+       movgr2fr.d      $f25, t1
+       movgr2fr.d      $f26, t1
+       movgr2fr.d      $f27, t1
+       movgr2fr.d      $f28, t1
+       movgr2fr.d      $f29, t1
+       movgr2fr.d      $f30, t1
+       movgr2fr.d      $f31, t1
+
+       jirl zero, ra, 0
+SYM_FUNC_END(_init_fpu)
+
+/*
+ * a0: fpregs
+ * a1: fcc
+ * a2: fcsr
+ */
+SYM_FUNC_START(_save_fp_context)
+       sc_save_fcc a1 t1 t2
+       sc_save_fcsr a2 t1
+       sc_save_fp a0
+       li.w    a0, 0                                   # success
+       jirl zero, ra, 0
+SYM_FUNC_END(_save_fp_context)
+
+/*
+ * a0: fpregs
+ * a1: fcc
+ * a2: fcsr
+ */
+SYM_FUNC_START(_restore_fp_context)
+       sc_restore_fp a0
+       sc_restore_fcc a1 t1 t2
+       sc_restore_fcsr a2 t1
+       li.w    a0, 0                                   # success
+       jirl zero, ra, 0
+SYM_FUNC_END(_restore_fp_context)
+
+SYM_FUNC_START(fault)
+       li.w    a0, -EFAULT                             # failure
+       jirl zero, ra, 0
+SYM_FUNC_END(fault)
diff --git a/arch/loongarch/kernel/idle.c b/arch/loongarch/kernel/idle.c
new file mode 100644 (file)
index 0000000..1a65d05
--- /dev/null
@@ -0,0 +1,16 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * LoongArch idle loop support.
+ *
+ * Copyright (C) 2020-2022 Loongson Technology Corporation Limited
+ */
+#include <linux/cpu.h>
+#include <linux/irqflags.h>
+#include <asm/cpu.h>
+#include <asm/idle.h>
+
+void __cpuidle arch_cpu_idle(void)
+{
+       raw_local_irq_enable();
+       __arch_cpu_idle(); /* idle instruction needs irq enabled */
+}
diff --git a/arch/loongarch/kernel/process.c b/arch/loongarch/kernel/process.c
new file mode 100644 (file)
index 0000000..5402022
--- /dev/null
@@ -0,0 +1,260 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Author: Huacai Chen <chenhuacai@loongson.cn>
+ * Copyright (C) 2020-2022 Loongson Technology Corporation Limited
+ *
+ * Derived from MIPS:
+ * Copyright (C) 1994 - 1999, 2000 by Ralf Baechle and others.
+ * Copyright (C) 2005, 2006 by Ralf Baechle (ralf@linux-mips.org)
+ * Copyright (C) 1999, 2000 Silicon Graphics, Inc.
+ * Copyright (C) 2004 Thiemo Seufer
+ * Copyright (C) 2013  Imagination Technologies Ltd.
+ */
+#include <linux/cpu.h>
+#include <linux/init.h>
+#include <linux/kernel.h>
+#include <linux/errno.h>
+#include <linux/sched.h>
+#include <linux/sched/debug.h>
+#include <linux/sched/task.h>
+#include <linux/sched/task_stack.h>
+#include <linux/mm.h>
+#include <linux/stddef.h>
+#include <linux/unistd.h>
+#include <linux/export.h>
+#include <linux/ptrace.h>
+#include <linux/mman.h>
+#include <linux/personality.h>
+#include <linux/sys.h>
+#include <linux/completion.h>
+#include <linux/kallsyms.h>
+#include <linux/random.h>
+#include <linux/prctl.h>
+#include <linux/nmi.h>
+
+#include <asm/asm.h>
+#include <asm/bootinfo.h>
+#include <asm/cpu.h>
+#include <asm/elf.h>
+#include <asm/fpu.h>
+#include <asm/io.h>
+#include <asm/irq.h>
+#include <asm/irq_regs.h>
+#include <asm/loongarch.h>
+#include <asm/pgtable.h>
+#include <asm/processor.h>
+#include <asm/reg.h>
+#include <asm/vdso.h>
+
+/*
+ * Idle related variables and functions
+ */
+
+unsigned long boot_option_idle_override = IDLE_NO_OVERRIDE;
+EXPORT_SYMBOL(boot_option_idle_override);
+
+asmlinkage void ret_from_fork(void);
+asmlinkage void ret_from_kernel_thread(void);
+
+void start_thread(struct pt_regs *regs, unsigned long pc, unsigned long sp)
+{
+       unsigned long crmd;
+       unsigned long prmd;
+       unsigned long euen;
+
+       /* New thread loses kernel privileges. */
+       crmd = regs->csr_crmd & ~(PLV_MASK);
+       crmd |= PLV_USER;
+       regs->csr_crmd = crmd;
+
+       prmd = regs->csr_prmd & ~(PLV_MASK);
+       prmd |= PLV_USER;
+       regs->csr_prmd = prmd;
+
+       euen = regs->csr_euen & ~(CSR_EUEN_FPEN);
+       regs->csr_euen = euen;
+       lose_fpu(0);
+
+       clear_thread_flag(TIF_LSX_CTX_LIVE);
+       clear_thread_flag(TIF_LASX_CTX_LIVE);
+       clear_used_math();
+       regs->csr_era = pc;
+       regs->regs[3] = sp;
+}
+
+void exit_thread(struct task_struct *tsk)
+{
+}
+
+int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src)
+{
+       /*
+        * Save any process state which is live in hardware registers to the
+        * parent context prior to duplication. This prevents the new child
+        * state becoming stale if the parent is preempted before copy_thread()
+        * gets a chance to save the parent's live hardware registers to the
+        * child context.
+        */
+       preempt_disable();
+
+       if (is_fpu_owner())
+               save_fp(current);
+
+       preempt_enable();
+
+       if (used_math())
+               memcpy(dst, src, sizeof(struct task_struct));
+       else
+               memcpy(dst, src, offsetof(struct task_struct, thread.fpu.fpr));
+
+       return 0;
+}
+
+/*
+ * Copy architecture-specific thread state
+ */
+int copy_thread(unsigned long clone_flags, unsigned long usp,
+       unsigned long kthread_arg, struct task_struct *p, unsigned long tls)
+{
+       unsigned long childksp;
+       struct pt_regs *childregs, *regs = current_pt_regs();
+
+       childksp = (unsigned long)task_stack_page(p) + THREAD_SIZE - 32;
+
+       /* set up new TSS. */
+       childregs = (struct pt_regs *) childksp - 1;
+       /*  Put the stack after the struct pt_regs.  */
+       childksp = (unsigned long) childregs;
+       p->thread.csr_euen = 0;
+       p->thread.csr_crmd = csr_read32(LOONGARCH_CSR_CRMD);
+       p->thread.csr_prmd = csr_read32(LOONGARCH_CSR_PRMD);
+       p->thread.csr_ecfg = csr_read32(LOONGARCH_CSR_ECFG);
+       if (unlikely(p->flags & (PF_KTHREAD | PF_IO_WORKER))) {
+               /* kernel thread */
+               p->thread.reg23 = usp; /* fn */
+               p->thread.reg24 = kthread_arg;
+               p->thread.reg03 = childksp;
+               p->thread.reg01 = (unsigned long) ret_from_kernel_thread;
+               memset(childregs, 0, sizeof(struct pt_regs));
+               childregs->csr_euen = p->thread.csr_euen;
+               childregs->csr_crmd = p->thread.csr_crmd;
+               childregs->csr_prmd = p->thread.csr_prmd;
+               childregs->csr_ecfg = p->thread.csr_ecfg;
+               return 0;
+       }
+
+       /* user thread */
+       *childregs = *regs;
+       childregs->regs[4] = 0; /* Child gets zero as return value */
+       if (usp)
+               childregs->regs[3] = usp;
+
+       p->thread.reg03 = (unsigned long) childregs;
+       p->thread.reg01 = (unsigned long) ret_from_fork;
+
+       /*
+        * New tasks lose permission to use the fpu. This accelerates context
+        * switching for most programs since they don't use the fpu.
+        */
+       childregs->csr_euen = 0;
+
+       clear_tsk_thread_flag(p, TIF_USEDFPU);
+       clear_tsk_thread_flag(p, TIF_USEDSIMD);
+       clear_tsk_thread_flag(p, TIF_LSX_CTX_LIVE);
+       clear_tsk_thread_flag(p, TIF_LASX_CTX_LIVE);
+
+       if (clone_flags & CLONE_SETTLS)
+               childregs->regs[2] = tls;
+
+       return 0;
+}
+
+unsigned long __get_wchan(struct task_struct *task)
+{
+       return 0;
+}
+
+unsigned long stack_top(void)
+{
+       unsigned long top = TASK_SIZE & PAGE_MASK;
+
+       /* Space for the VDSO & data page */
+       top -= PAGE_ALIGN(current->thread.vdso->size);
+       top -= PAGE_SIZE;
+
+       /* Space to randomize the VDSO base */
+       if (current->flags & PF_RANDOMIZE)
+               top -= VDSO_RANDOMIZE_SIZE;
+
+       return top;
+}
+
+/*
+ * Don't forget that the stack pointer must be aligned on a 8 bytes
+ * boundary for 32-bits ABI and 16 bytes for 64-bits ABI.
+ */
+unsigned long arch_align_stack(unsigned long sp)
+{
+       if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space)
+               sp -= get_random_int() & ~PAGE_MASK;
+
+       return sp & STACK_ALIGN;
+}
+
+static DEFINE_PER_CPU(call_single_data_t, backtrace_csd);
+static struct cpumask backtrace_csd_busy;
+
+static void handle_backtrace(void *info)
+{
+       nmi_cpu_backtrace(get_irq_regs());
+       cpumask_clear_cpu(smp_processor_id(), &backtrace_csd_busy);
+}
+
+static void raise_backtrace(cpumask_t *mask)
+{
+       call_single_data_t *csd;
+       int cpu;
+
+       for_each_cpu(cpu, mask) {
+               /*
+                * If we previously sent an IPI to the target CPU & it hasn't
+                * cleared its bit in the busy cpumask then it didn't handle
+                * our previous IPI & it's not safe for us to reuse the
+                * call_single_data_t.
+                */
+               if (cpumask_test_and_set_cpu(cpu, &backtrace_csd_busy)) {
+                       pr_warn("Unable to send backtrace IPI to CPU%u - perhaps it hung?\n",
+                               cpu);
+                       continue;
+               }
+
+               csd = &per_cpu(backtrace_csd, cpu);
+               csd->func = handle_backtrace;
+               smp_call_function_single_async(cpu, csd);
+       }
+}
+
+void arch_trigger_cpumask_backtrace(const cpumask_t *mask, bool exclude_self)
+{
+       nmi_trigger_cpumask_backtrace(mask, exclude_self, raise_backtrace);
+}
+
+#ifdef CONFIG_64BIT
+void loongarch_dump_regs64(u64 *uregs, const struct pt_regs *regs)
+{
+       unsigned int i;
+
+       for (i = LOONGARCH_EF_R1; i <= LOONGARCH_EF_R31; i++) {
+               uregs[i] = regs->regs[i - LOONGARCH_EF_R0];
+       }
+
+       uregs[LOONGARCH_EF_ORIG_A0] = regs->orig_a0;
+       uregs[LOONGARCH_EF_CSR_ERA] = regs->csr_era;
+       uregs[LOONGARCH_EF_CSR_BADV] = regs->csr_badvaddr;
+       uregs[LOONGARCH_EF_CSR_CRMD] = regs->csr_crmd;
+       uregs[LOONGARCH_EF_CSR_PRMD] = regs->csr_prmd;
+       uregs[LOONGARCH_EF_CSR_EUEN] = regs->csr_euen;
+       uregs[LOONGARCH_EF_CSR_ECFG] = regs->csr_ecfg;
+       uregs[LOONGARCH_EF_CSR_ESTAT] = regs->csr_estat;
+}
+#endif /* CONFIG_64BIT */
diff --git a/arch/loongarch/kernel/ptrace.c b/arch/loongarch/kernel/ptrace.c
new file mode 100644 (file)
index 0000000..e6ab879
--- /dev/null
@@ -0,0 +1,431 @@
+// SPDX-License-Identifier: GPL-2.0
+/*
+ * Author: Hanlu Li <lihanlu@loongson.cn>
+ *         Huacai Chen <chenhuacai@loongson.cn>
+ *
+ * Copyright (C) 2020-2022 Loongson Technology Corporation Limited
+ *
+ * Derived from MIPS:
+ * Copyright (C) 1992 Ross Biro
+ * Copyright (C) Linus Torvalds
+ * Copyright (C) 1994, 95, 96, 97, 98, 2000 Ralf Baechle
+ * Copyright (C) 1996 David S. Miller
+ * Kevin D. Kissell, kevink@mips.com and Carsten Langgaard, carstenl@mips.com
+ * Copyright (C) 1999 MIPS Technologies, Inc.
+ * Copyright (C) 2000 Ulf Carlsson
+ */
+#include <linux/kernel.h>
+#include <linux/audit.h>
+#include <linux/compiler.h>
+#include <linux/context_tracking.h>
+#include <linux/elf.h>
+#include <linux/errno.h>
+#include <linux/mm.h>
+#include <linux/ptrace.h>
+#include <linux/regset.h>
+#include <linux/sched.h>
+#include <linux/sched/task_stack.h>
+#include <linux/security.h>
+#include <linux/smp.h>
+#include <linux/stddef.h>
+#include <linux/seccomp.h>
+#include <linux/uaccess.h>
+
+#include <asm/byteorder.h>
+#include <asm/cpu.h>
+#include <asm/cpu-info.h>
+#include <asm/fpu.h>
+#include <asm/loongarch.h>
+#include <asm/page.h>
+#include <asm/pgtable.h>
+#include <asm/processor.h>
+#include <asm/reg.h>
+#include <asm/syscall.h>
+
+static void init_fp_ctx(struct task_struct *target)
+{
+       /* The target already has context */
+       if (tsk_used_math(target))
+               return;
+
+       /* Begin with data registers set to all 1s... */
+       memset(&target->thread.fpu.fpr, ~0, sizeof(target->thread.fpu.fpr));
+       set_stopped_child_used_math(target);
+}
+
+/*
+ * Called by kernel/ptrace.c when detaching..
+ *
+ * Make sure single step bits etc are not set.
+ */
+void ptrace_disable(struct task_struct *child)
+{
+       /* Don't load the watchpoint registers for the ex-child. */
+       clear_tsk_thread_flag(child, TIF_LOAD_WATCH);
+       clear_tsk_thread_flag(child, TIF_SINGLESTEP);
+}
+
+/* regset get/set implementations */
+
+static int gpr_get(struct task_struct *target,
+                  const struct user_regset *regset,
+                  struct membuf to)
+{
+       int r;
+       struct pt_regs *regs = task_pt_regs(target);
+
+       r = membuf_write(&to, &regs->regs, sizeof(u64) * GPR_NUM);
+       r = membuf_write(&to, &regs->orig_a0, sizeof(u64));
+       r = membuf_write(&to, &regs->csr_era, sizeof(u64));
+       r = membuf_write(&to, &regs->csr_badvaddr, sizeof(u64));
+
+       return r;
+}
+
+static int gpr_set(struct task_struct *target,
+                  const struct user_regset *regset,
+                  unsigned int pos, unsigned int count,
+                  const void *kbuf, const void __user *ubuf)
+{
+       int err;
+       int a0_start = sizeof(u64) * GPR_NUM;
+       int era_start = a0_start + sizeof(u64);
+       int badvaddr_start = era_start + sizeof(u64);
+       struct pt_regs *regs = task_pt_regs(target);
+
+       err = user_regset_copyin(&pos, &count, &kbuf, &ubuf,
+                                &regs->regs,
+                                0, a0_start);
+       err |= user_regset_copyin(&pos, &count, &kbuf, &ubuf,
+                                &regs->orig_a0,
+                                a0_start, a0_start + sizeof(u64));
+       err |= user_regset_copyin(&pos, &count, &kbuf, &ubuf,
+                                &regs->csr_era,
+                                era_start, era_start + sizeof(u64));
+       err |= user_regset_copyin(&pos, &count, &kbuf, &ubuf,
+                                &regs->csr_badvaddr,
+                                badvaddr_start, badvaddr_start + sizeof(u64));
+
+       return err;
+}
+
+
+/*
+ * Get the general floating-point registers.
+ */
+static int gfpr_get(struct task_struct *target, struct membuf *to)
+{
+       return membuf_write(to, &target->thread.fpu.fpr,
+                           sizeof(elf_fpreg_t) * NUM_FPU_REGS);
+}
+
+static int gfpr_get_simd(struct task_struct *target, struct membuf *to)
+{
+       int i, r;
+       u64 fpr_val;
+
+       BUILD_BUG_ON(sizeof(fpr_val) != sizeof(elf_fpreg_t));
+       for (i = 0; i < NUM_FPU_REGS; i++) {
+               fpr_val = get_fpr64(&target->thread.fpu.fpr[i], 0);
+               r = membuf_write(to, &fpr_val, sizeof(elf_fpreg_t));
+       }
+
+       return r;
+}
+
+/*
+ * Choose the appropriate helper for general registers, and then copy
+ * the FCC and FCSR registers separately.
+ */
+static int fpr_get(struct task_struct *target,
+                  const struct user_regset *regset,
+                  struct membuf to)
+{
+       int r;
+
+       if (sizeof(target->thread.fpu.fpr[0]) == sizeof(elf_fpreg_t))
+               r = gfpr_get(target, &to);
+       else
+               r = gfpr_get_simd(target, &to);
+
+       r = membuf_write(&to, &target->thread.fpu.fcc, sizeof(target->thread.fpu.fcc));
+       r = membuf_write(&to, &target->thread.fpu.fcsr, sizeof(target->thread.fpu.fcsr));
+
+       return r;
+}
+
+static int gfpr_set(struct task_struct *target,
+                   unsigned int *pos, unsigned int *count,
+                   const void **kbuf, const void __user **ubuf)
+{
+       return user_regset_copyin(pos, count, kbuf, ubuf,
+                                 &target->thread.fpu.fpr,
+                                 0, NUM_FPU_REGS * sizeof(elf_fpreg_t));
+}
+
+static int gfpr_set_simd(struct task_struct *target,
+                      unsigned int *pos, unsigned int *count,
+                      const void **kbuf, const void __user **ubuf)
+{
+       int i, err;
+       u64 fpr_val;
+
+       BUILD_BUG_ON(sizeof(fpr_val) != sizeof(elf_fpreg_t));
+       for (i = 0; i < NUM_FPU_REGS && *count > 0; i++) {
+               err = user_regset_copyin(pos, count, kbuf, ubuf,
+                                        &fpr_val, i * sizeof(elf_fpreg_t),
+                                        (i + 1) * sizeof(elf_fpreg_t));
+               if (err)
+                       return err;
+               set_fpr64(&target->thread.fpu.fpr[i], 0, fpr_val);
+       }
+
+       return 0;
+}
+
+/*
+ * Choose the appropriate helper for general registers, and then copy
+ * the FCC register separately.
+ */
+static int fpr_set(struct task_struct *target,
+                  const struct user_regset *regset,
+                  unsigned int pos, unsigned int count,
+                  const void *kbuf, const void __user *ubuf)
+{
+       const int fcc_start = NUM_FPU_REGS * sizeof(elf_fpreg_t);
+       const int fcc_end = fcc_start + sizeof(u64);
+       int err;
+
+       BUG_ON(count % sizeof(elf_fpreg_t));
+       if (pos + count > sizeof(elf_fpregset_t))
+               return -EIO;
+
+       init_fp_ctx(target);
+
+       if (sizeof(target->thread.fpu.fpr[0]) == sizeof(elf_fpreg_t))
+               err = gfpr_set(target, &pos, &count, &kbuf, &ubuf);
+       else
+               err = gfpr_set_simd(target, &pos, &count, &kbuf, &ubuf);
+       if (err)
+               return err;
+
+       if (count > 0)
+               err |= user_regset_copyin(&pos, &count, &kbuf, &ubuf,
+                                         &target->thread.fpu.fcc,
+                                         fcc_start, fcc_end);
+
+       return err;
+}
+
+static int cfg_get(struct task_struct *target,
+                  const struct user_regset *regset,
+                  struct membuf to)
+{
+       int i, r;
+       u32 cfg_val;
+
+       i = 0;
+       while (to.left > 0) {
+               cfg_val = read_cpucfg(i++);
+               r = membuf_write(&to, &cfg_val, sizeof(u32));
+       }
+
+       return r;
+}
+
+/*
+ * CFG registers are read-only.
+ */
+static int cfg_set(struct task_struct *target,
+                  const struct user_regset *regset,
+                  unsigned int pos, unsigned int count,
+                  const void *kbuf, const void __user *ubuf)
+{
+       return 0;
+}
+
+struct pt_regs_offset {
+       const char *name;
+       int offset;
+};
+
+#define REG_OFFSET_NAME(n, r) {.name = #n, .offset = offsetof(struct pt_regs, r)}
+#define REG_OFFSET_END {.name = NULL, .offset = 0}
+
+static const struct pt_regs_offset regoffset_table[] = {
+       REG_OFFSET_NAME(r0, regs[0]),
+       REG_OFFSET_NAME(r1, regs[1]),
+       REG_OFFSET_NAME(r2, regs[2]),
+       REG_OFFSET_NAME(r3, regs[3]),
+       REG_OFFSET_NAME(r4, regs[4]),
+       REG_OFFSET_NAME(r5, regs[5]),
+       REG_OFFSET_NAME(r6, regs[6]),
+       REG_OFFSET_NAME(r7, regs[7]),
+       REG_OFFSET_NAME(r8, regs[8]),
+       REG_OFFSET_NAME(r9, regs[9]),
+       REG_OFFSET_NAME(r10, regs[10]),
+       REG_OFFSET_NAME(r11, regs[11]),
+       REG_OFFSET_NAME(r12, regs[12]),
+       REG_OFFSET_NAME(r13, regs[13]),
+       REG_OFFSET_NAME(r14, regs[14]),
+       REG_OFFSET_NAME(r15, regs[15]),
+       REG_OFFSET_NAME(r16, regs[16]),
+       REG_OFFSET_NAME(r17, regs[17]),
+       REG_OFFSET_NAME(r18, regs[18]),
+       REG_OFFSET_NAME(r19, regs[19]),
+       REG_OFFSET_NAME(r20, regs[20]),
+       REG_OFFSET_NAME(r21, regs[21]),
+       REG_OFFSET_NAME(r22, regs[22]),
+       REG_OFFSET_NAME(r23, regs[23]),
+       REG_OFFSET_NAME(r24, regs[24]),
+       REG_OFFSET_NAME(r25, regs[25]),
+       REG_OFFSET_NAME(r26, regs[26]),
+       REG_OFFSET_NAME(r27, regs[27]),
+       REG_OFFSET_NAME(r28, regs[28]),
+       REG_OFFSET_NAME(r29, regs[29]),
+       REG_OFFSET_NAME(r30, regs[30]),
+       REG_OFFSET_NAME(r31, regs[31]),
+       REG_OFFSET_NAME(orig_a0, orig_a0),
+       REG_OFFSET_NAME(csr_era, csr_era),
+       REG_OFFSET_NAME(csr_badvaddr, csr_badvaddr),
+       REG_OFFSET_NAME(csr_crmd, csr_crmd),
+       REG_OFFSET_NAME(csr_prmd, csr_prmd),
+       REG_OFFSET_NAME(csr_euen, csr_euen),
+       REG_OFFSET_NAME(csr_ecfg, csr_ecfg),
+       REG_OFFSET_NAME(csr_estat, csr_estat),
+       REG_OFFSET_END,
+};
+
+/**
+ * regs_query_register_offset() - query register offset from its name
+ * @name:       the name of a register
+ *
+ * regs_query_register_offset() returns the offset of a register in struct
+ * pt_regs from its name. If the name is invalid, this returns -EINVAL;
+ */
+int regs_query_register_offset(const char *name)
+{
+       const struct pt_regs_offset *roff;
+
+       for (roff = regoffset_table; roff->name != NULL; roff++)
+               if (!strcmp(roff->name, name))
+                       return roff->offset;
+       return -EINVAL;
+}
+
+enum loongarch_regset {
+       REGSET_GPR,
+       REGSET_FPR,
+       REGSET_CPUCFG,
+};
+
+static const struct user_regset loongarch64_regsets[] = {
+       [REGSET_GPR] = {
+               .core_note_type = NT_PRSTATUS,
+               .n              = ELF_NGREG,
+               .size           = sizeof(elf_greg_t),
+               .align          = sizeof(elf_greg_t),
+               .regset_get     = gpr_get,
+               .set            = gpr_set,
+       },
+       [REGSET_FPR] = {
+               .core_note_type = NT_PRFPREG,
+               .n              = ELF_NFPREG,
+               .size           = sizeof(elf_fpreg_t),
+               .align          = sizeof(elf_fpreg_t),
+               .regset_get     = fpr_get,
+               .set            = fpr_set,
+       },
+       [REGSET_CPUCFG] = {
+               .core_note_type = NT_LOONGARCH_CPUCFG,
+               .n              = 64,
+               .size           = sizeof(u32),
+               .align          = sizeof(u32),
+               .regset_get     = cfg_get,
+               .set            = cfg_set,
+       },
+};
+
+static const struct user_regset_view user_loongarch64_view = {
+       .name           = "loongarch64",
+       .e_machine      = ELF_ARCH,
+       .regsets        = loongarch64_regsets,
+       .n              = ARRAY_SIZE(loongarch64_regsets),
+};
+
+
+const struct user_regset_view *task_user_regset_view(struct task_struct *task)
+{
+       return &user_loongarch64_view;
+}
+
+static inline int read_user(struct task_struct *target, unsigned long addr,
+                           unsigned long __user *data)
+{
+       unsigned long tmp = 0;
+
+       switch (addr) {
+       case 0 ... 31:
+               tmp = task_pt_regs(target)->regs[addr];
+               break;
+       case ARG0:
+               tmp = task_pt_regs(target)->orig_a0;
+               break;
+       case PC:
+               tmp = task_pt_regs(target)->csr_era;
+               break;
+       case BADVADDR:
+               tmp = task_pt_regs(target)->csr_badvaddr;
+               break;
+       default:
+               return -EIO;
+       }
+
+       return put_user(tmp, data);
+}
+
+static inline int write_user(struct task_struct *target, unsigned long addr,
+                           unsigned long data)
+{
+       switch (addr) {
+       case 0 ... 31:
+               task_pt_regs(target)->regs[addr] = data;
+               break;
+       case ARG0:
+               task_pt_regs(target)->orig_a0 = data;
+               break;
+       case PC:
+               task_pt_regs(target)->csr_era = data;
+               break;
+       case BADVADDR:
+               task_pt_regs(target)->csr_badvaddr = data;
+               break;
+       default:
+               return -EIO;
+       }
+
+       return 0;
+}
+
+long arch_ptrace(struct task_struct *child, long request,
+                unsigned long addr, unsigned long data)
+{
+       int ret;
+       unsigned long __user *datap = (void __user *) data;
+
+       switch (request) {
+       case PTRACE_PEEKUSR:
+               ret = read_user(child, addr, datap);
+               break;
+
+       case PTRACE_POKEUSR:
+               ret = write_user(child, addr, data);
+               break;
+
+       default:
+               ret = ptrace_request(child, request, addr, data);
+               break;
+       }
+
+       return ret;
+}
diff --git a/arch/loongarch/kernel/switch.S b/arch/loongarch/kernel/switch.S
new file mode 100644 (file)
index 0000000..53e2fa8
--- /dev/null
@@ -0,0 +1,35 @@
+/* SPDX-License-Identifier: GPL-2.0 */
+/*
+ * Copyright (C) 2020-2022 Loongson Technology Corporation Limited
+ */
+#include <asm/asm.h>
+#include <asm/asmmacro.h>
+#include <asm/asm-offsets.h>
+#include <asm/loongarch.h>
+#include <asm/regdef.h>
+#include <asm/stackframe.h>
+#include <asm/thread_info.h>
+
+/*
+ * task_struct *__switch_to(task_struct *prev, task_struct *next,
+ *                         struct thread_info *next_ti)
+ */
+       .align  5
+SYM_FUNC_START(__switch_to)
+       csrrd   t1, LOONGARCH_CSR_PRMD
+       stptr.d t1, a0, THREAD_CSRPRMD
+
+       cpu_save_nonscratch a0
+       stptr.d ra, a0, THREAD_REG01
+       move    tp, a2
+       cpu_restore_nonscratch a1
+
+       li.w    t0, _THREAD_SIZE - 32
+       PTR_ADD t0, t0, tp
+       set_saved_sp    t0, t1, t2
+
+       ldptr.d t1, a1, THREAD_CSRPRMD
+       csrwr   t1, LOONGARCH_CSR_PRMD
+
+       jr      ra
+SYM_FUNC_END(__switch_to)