| 1 | // SPDX-License-Identifier: GPL-2.0 |
| 2 | /* |
| 3 | * This file handles the architecture dependent parts of process handling. |
| 4 | * |
| 5 | * Copyright IBM Corp. 1999, 2009 |
| 6 | * Author(s): Martin Schwidefsky <schwidefsky@de.ibm.com>, |
| 7 | * Hartmut Penner <hp@de.ibm.com>, |
| 8 | * Denis Joseph Barrow, |
| 9 | */ |
| 10 | |
| 11 | #include <linux/elf-randomize.h> |
| 12 | #include <linux/compiler.h> |
| 13 | #include <linux/cpu.h> |
| 14 | #include <linux/sched.h> |
| 15 | #include <linux/sched/debug.h> |
| 16 | #include <linux/sched/task.h> |
| 17 | #include <linux/sched/task_stack.h> |
| 18 | #include <linux/kernel.h> |
| 19 | #include <linux/mm.h> |
| 20 | #include <linux/elfcore.h> |
| 21 | #include <linux/smp.h> |
| 22 | #include <linux/slab.h> |
| 23 | #include <linux/interrupt.h> |
| 24 | #include <linux/tick.h> |
| 25 | #include <linux/personality.h> |
| 26 | #include <linux/syscalls.h> |
| 27 | #include <linux/kprobes.h> |
| 28 | #include <linux/random.h> |
| 29 | #include <linux/init_task.h> |
| 30 | #include <linux/entry-common.h> |
| 31 | #include <linux/io.h> |
| 32 | #include <asm/guarded_storage.h> |
| 33 | #include <asm/access-regs.h> |
| 34 | #include <asm/switch_to.h> |
| 35 | #include <asm/cpu_mf.h> |
| 36 | #include <asm/processor.h> |
| 37 | #include <asm/ptrace.h> |
| 38 | #include <asm/vtimer.h> |
| 39 | #include <asm/exec.h> |
| 40 | #include <asm/fpu.h> |
| 41 | #include <asm/irq.h> |
| 42 | #include <asm/nmi.h> |
| 43 | #include <asm/smp.h> |
| 44 | #include <asm/stacktrace.h> |
| 45 | #include <asm/runtime_instr.h> |
| 46 | #include <asm/unwind.h> |
| 47 | #include "entry.h" |
| 48 | |
| 49 | void ret_from_fork(void) asm("ret_from_fork" ); |
| 50 | |
| 51 | void __ret_from_fork(struct task_struct *prev, struct pt_regs *regs) |
| 52 | { |
| 53 | void (*func)(void *arg); |
| 54 | |
| 55 | schedule_tail(prev); |
| 56 | |
| 57 | if (!user_mode(regs)) { |
| 58 | /* Kernel thread */ |
| 59 | func = (void *)regs->gprs[9]; |
| 60 | func((void *)regs->gprs[10]); |
| 61 | } |
| 62 | clear_pt_regs_flag(regs, PIF_SYSCALL); |
| 63 | syscall_exit_to_user_mode(regs); |
| 64 | } |
| 65 | |
| 66 | void flush_thread(void) |
| 67 | { |
| 68 | } |
| 69 | |
| 70 | void arch_setup_new_exec(void) |
| 71 | { |
| 72 | if (get_lowcore()->current_pid != current->pid) { |
| 73 | get_lowcore()->current_pid = current->pid; |
| 74 | if (test_facility(40)) |
| 75 | lpp(&get_lowcore()->lpp); |
| 76 | } |
| 77 | } |
| 78 | |
| 79 | void arch_release_task_struct(struct task_struct *tsk) |
| 80 | { |
| 81 | runtime_instr_release(tsk); |
| 82 | guarded_storage_release(tsk); |
| 83 | } |
| 84 | |
| 85 | int arch_dup_task_struct(struct task_struct *dst, struct task_struct *src) |
| 86 | { |
| 87 | save_user_fpu_regs(); |
| 88 | |
| 89 | *dst = *src; |
| 90 | dst->thread.kfpu_flags = 0; |
| 91 | |
| 92 | /* |
| 93 | * Don't transfer over the runtime instrumentation or the guarded |
| 94 | * storage control block pointers. These fields are cleared here instead |
| 95 | * of in copy_thread() to avoid premature freeing of associated memory |
| 96 | * on fork() failure. Wait to clear the RI flag because ->stack still |
| 97 | * refers to the source thread. |
| 98 | */ |
| 99 | dst->thread.ri_cb = NULL; |
| 100 | dst->thread.gs_cb = NULL; |
| 101 | dst->thread.gs_bc_cb = NULL; |
| 102 | |
| 103 | return 0; |
| 104 | } |
| 105 | |
| 106 | int copy_thread(struct task_struct *p, const struct kernel_clone_args *args) |
| 107 | { |
| 108 | u64 clone_flags = args->flags; |
| 109 | unsigned long new_stackp = args->stack; |
| 110 | unsigned long tls = args->tls; |
| 111 | struct fake_frame |
| 112 | { |
| 113 | struct stack_frame sf; |
| 114 | struct pt_regs childregs; |
| 115 | } *frame; |
| 116 | |
| 117 | frame = container_of(task_pt_regs(p), struct fake_frame, childregs); |
| 118 | p->thread.ksp = (unsigned long) frame; |
| 119 | /* Save access registers to new thread structure. */ |
| 120 | save_access_regs(&p->thread.acrs[0]); |
| 121 | /* start new process with ar4 pointing to the correct address space */ |
| 122 | /* Don't copy debug registers */ |
| 123 | memset(&p->thread.per_user, 0, sizeof(p->thread.per_user)); |
| 124 | memset(&p->thread.per_event, 0, sizeof(p->thread.per_event)); |
| 125 | clear_tsk_thread_flag(tsk: p, flag: TIF_SINGLE_STEP); |
| 126 | p->thread.per_flags = 0; |
| 127 | /* Initialize per thread user and system timer values */ |
| 128 | p->thread.user_timer = 0; |
| 129 | p->thread.guest_timer = 0; |
| 130 | p->thread.system_timer = 0; |
| 131 | p->thread.hardirq_timer = 0; |
| 132 | p->thread.softirq_timer = 0; |
| 133 | p->thread.last_break = 1; |
| 134 | |
| 135 | frame->sf.back_chain = 0; |
| 136 | frame->sf.gprs[11 - 6] = (unsigned long)&frame->childregs; |
| 137 | frame->sf.gprs[12 - 6] = (unsigned long)p; |
| 138 | /* new return point is ret_from_fork */ |
| 139 | frame->sf.gprs[14 - 6] = (unsigned long)ret_from_fork; |
| 140 | /* fake return stack for resume(), don't go back to schedule */ |
| 141 | frame->sf.gprs[15 - 6] = (unsigned long)frame; |
| 142 | |
| 143 | /* Store access registers to kernel stack of new process. */ |
| 144 | if (unlikely(args->fn)) { |
| 145 | /* kernel thread */ |
| 146 | memset(&frame->childregs, 0, sizeof(struct pt_regs)); |
| 147 | frame->childregs.psw.mask = PSW_KERNEL_BITS | PSW_MASK_IO | |
| 148 | PSW_MASK_EXT | PSW_MASK_MCHECK; |
| 149 | frame->childregs.gprs[9] = (unsigned long)args->fn; |
| 150 | frame->childregs.gprs[10] = (unsigned long)args->fn_arg; |
| 151 | frame->childregs.orig_gpr2 = -1; |
| 152 | frame->childregs.last_break = 1; |
| 153 | return 0; |
| 154 | } |
| 155 | frame->childregs = *current_pt_regs(); |
| 156 | frame->childregs.gprs[2] = 0; /* child returns 0 on fork. */ |
| 157 | frame->childregs.flags = 0; |
| 158 | if (new_stackp) |
| 159 | frame->childregs.gprs[15] = new_stackp; |
| 160 | /* |
| 161 | * Clear the runtime instrumentation flag after the above childregs |
| 162 | * copy. The CB pointer was already cleared in arch_dup_task_struct(). |
| 163 | */ |
| 164 | frame->childregs.psw.mask &= ~PSW_MASK_RI; |
| 165 | |
| 166 | /* Set a new TLS ? */ |
| 167 | if (clone_flags & CLONE_SETTLS) { |
| 168 | p->thread.acrs[0] = (unsigned int)(tls >> 32); |
| 169 | p->thread.acrs[1] = (unsigned int)tls; |
| 170 | } |
| 171 | /* |
| 172 | * s390 stores the svc return address in arch_data when calling |
| 173 | * sigreturn()/restart_syscall() via vdso. 1 means no valid address |
| 174 | * stored. |
| 175 | */ |
| 176 | p->restart_block.arch_data = 1; |
| 177 | return 0; |
| 178 | } |
| 179 | |
| 180 | void execve_tail(void) |
| 181 | { |
| 182 | current->thread.ufpu.fpc = 0; |
| 183 | fpu_sfpc(0); |
| 184 | } |
| 185 | |
| 186 | struct task_struct *__switch_to(struct task_struct *prev, struct task_struct *next) |
| 187 | { |
| 188 | save_user_fpu_regs(); |
| 189 | save_kernel_fpu_regs(&prev->thread); |
| 190 | save_access_regs(&prev->thread.acrs[0]); |
| 191 | save_ri_cb(prev->thread.ri_cb); |
| 192 | save_gs_cb(prev->thread.gs_cb); |
| 193 | update_cr_regs(next); |
| 194 | restore_kernel_fpu_regs(&next->thread); |
| 195 | restore_access_regs(&next->thread.acrs[0]); |
| 196 | restore_ri_cb(next->thread.ri_cb, prev->thread.ri_cb); |
| 197 | restore_gs_cb(next->thread.gs_cb); |
| 198 | return __switch_to_asm(prev, next); |
| 199 | } |
| 200 | |
| 201 | unsigned long __get_wchan(struct task_struct *p) |
| 202 | { |
| 203 | struct unwind_state state; |
| 204 | unsigned long ip = 0; |
| 205 | |
| 206 | if (!task_stack_page(task: p)) |
| 207 | return 0; |
| 208 | |
| 209 | if (!try_get_task_stack(tsk: p)) |
| 210 | return 0; |
| 211 | |
| 212 | unwind_for_each_frame(&state, p, NULL, 0) { |
| 213 | if (state.stack_info.type != STACK_TYPE_TASK) { |
| 214 | ip = 0; |
| 215 | break; |
| 216 | } |
| 217 | |
| 218 | ip = unwind_get_return_address(state: &state); |
| 219 | if (!ip) |
| 220 | break; |
| 221 | |
| 222 | if (!in_sched_functions(ip)) |
| 223 | break; |
| 224 | } |
| 225 | |
| 226 | put_task_stack(tsk: p); |
| 227 | return ip; |
| 228 | } |
| 229 | |
| 230 | unsigned long arch_align_stack(unsigned long sp) |
| 231 | { |
| 232 | if (!(current->personality & ADDR_NO_RANDOMIZE) && randomize_va_space) |
| 233 | sp -= get_random_u32_below(PAGE_SIZE); |
| 234 | return sp & ~0xf; |
| 235 | } |
| 236 | |
| 237 | static inline unsigned long brk_rnd(void) |
| 238 | { |
| 239 | return (get_random_u16() & BRK_RND_MASK) << PAGE_SHIFT; |
| 240 | } |
| 241 | |
| 242 | unsigned long arch_randomize_brk(struct mm_struct *mm) |
| 243 | { |
| 244 | unsigned long ret; |
| 245 | |
| 246 | ret = PAGE_ALIGN(mm->brk + brk_rnd()); |
| 247 | return (ret > mm->brk) ? ret : mm->brk; |
| 248 | } |
| 249 | |