1 // SPDX-License-Identifier: GPL-2.0 2 3 #include <linux/irq-entry-common.h> 4 #include <linux/resume_user_mode.h> 5 #include <linux/highmem.h> 6 #include <linux/jump_label.h> 7 #include <linux/kmsan.h> 8 #include <linux/livepatch.h> 9 #include <linux/tick.h> 10 11 /* Workaround to allow gradual conversion of architecture code */ 12 void __weak arch_do_signal_or_restart(struct pt_regs *regs) { } 13 14 /** 15 * exit_to_user_mode_loop - do any pending work before leaving to user space 16 * @regs: Pointer to pt_regs on entry stack 17 * @ti_work: TIF work flags as read by the caller 18 */ 19 __always_inline unsigned long exit_to_user_mode_loop(struct pt_regs *regs, 20 unsigned long ti_work) 21 { 22 /* 23 * Before returning to user space ensure that all pending work 24 * items have been completed. 25 */ 26 while (ti_work & EXIT_TO_USER_MODE_WORK) { 27 28 local_irq_enable_exit_to_user(ti_work); 29 30 if (ti_work & (_TIF_NEED_RESCHED | _TIF_NEED_RESCHED_LAZY)) 31 schedule(); 32 33 if (ti_work & _TIF_UPROBE) 34 uprobe_notify_resume(regs); 35 36 if (ti_work & _TIF_PATCH_PENDING) 37 klp_update_patch_state(current); 38 39 if (ti_work & (_TIF_SIGPENDING | _TIF_NOTIFY_SIGNAL)) 40 arch_do_signal_or_restart(regs); 41 42 if (ti_work & _TIF_NOTIFY_RESUME) 43 resume_user_mode_work(regs); 44 45 /* Architecture specific TIF work */ 46 arch_exit_to_user_mode_work(regs, ti_work); 47 48 /* 49 * Disable interrupts and reevaluate the work flags as they 50 * might have changed while interrupts and preemption was 51 * enabled above. 52 */ 53 local_irq_disable_exit_to_user(); 54 55 /* Check if any of the above work has queued a deferred wakeup */ 56 tick_nohz_user_enter_prepare(); 57 58 ti_work = read_thread_flags(); 59 } 60 61 /* Return the latest work state for arch_exit_to_user_mode() */ 62 return ti_work; 63 } 64 65 noinstr void irqentry_enter_from_user_mode(struct pt_regs *regs) 66 { 67 enter_from_user_mode(regs); 68 } 69 70 noinstr void irqentry_exit_to_user_mode(struct pt_regs *regs) 71 { 72 instrumentation_begin(); 73 exit_to_user_mode_prepare(regs); 74 instrumentation_end(); 75 exit_to_user_mode(); 76 } 77 78 noinstr irqentry_state_t irqentry_enter(struct pt_regs *regs) 79 { 80 irqentry_state_t ret = { 81 .exit_rcu = false, 82 }; 83 84 if (user_mode(regs)) { 85 irqentry_enter_from_user_mode(regs); 86 return ret; 87 } 88 89 /* 90 * If this entry hit the idle task invoke ct_irq_enter() whether 91 * RCU is watching or not. 92 * 93 * Interrupts can nest when the first interrupt invokes softirq 94 * processing on return which enables interrupts. 95 * 96 * Scheduler ticks in the idle task can mark quiescent state and 97 * terminate a grace period, if and only if the timer interrupt is 98 * not nested into another interrupt. 99 * 100 * Checking for rcu_is_watching() here would prevent the nesting 101 * interrupt to invoke ct_irq_enter(). If that nested interrupt is 102 * the tick then rcu_flavor_sched_clock_irq() would wrongfully 103 * assume that it is the first interrupt and eventually claim 104 * quiescent state and end grace periods prematurely. 105 * 106 * Unconditionally invoke ct_irq_enter() so RCU state stays 107 * consistent. 108 * 109 * TINY_RCU does not support EQS, so let the compiler eliminate 110 * this part when enabled. 111 */ 112 if (!IS_ENABLED(CONFIG_TINY_RCU) && is_idle_task(current)) { 113 /* 114 * If RCU is not watching then the same careful 115 * sequence vs. lockdep and tracing is required 116 * as in irqentry_enter_from_user_mode(). 117 */ 118 lockdep_hardirqs_off(CALLER_ADDR0); 119 ct_irq_enter(); 120 instrumentation_begin(); 121 kmsan_unpoison_entry_regs(regs); 122 trace_hardirqs_off_finish(); 123 instrumentation_end(); 124 125 ret.exit_rcu = true; 126 return ret; 127 } 128 129 /* 130 * If RCU is watching then RCU only wants to check whether it needs 131 * to restart the tick in NOHZ mode. rcu_irq_enter_check_tick() 132 * already contains a warning when RCU is not watching, so no point 133 * in having another one here. 134 */ 135 lockdep_hardirqs_off(CALLER_ADDR0); 136 instrumentation_begin(); 137 kmsan_unpoison_entry_regs(regs); 138 rcu_irq_enter_check_tick(); 139 trace_hardirqs_off_finish(); 140 instrumentation_end(); 141 142 return ret; 143 } 144 145 void raw_irqentry_exit_cond_resched(void) 146 { 147 if (!preempt_count()) { 148 /* Sanity check RCU and thread stack */ 149 rcu_irq_exit_check_preempt(); 150 if (IS_ENABLED(CONFIG_DEBUG_ENTRY)) 151 WARN_ON_ONCE(!on_thread_stack()); 152 if (need_resched()) 153 preempt_schedule_irq(); 154 } 155 } 156 #ifdef CONFIG_PREEMPT_DYNAMIC 157 #if defined(CONFIG_HAVE_PREEMPT_DYNAMIC_CALL) 158 DEFINE_STATIC_CALL(irqentry_exit_cond_resched, raw_irqentry_exit_cond_resched); 159 #elif defined(CONFIG_HAVE_PREEMPT_DYNAMIC_KEY) 160 DEFINE_STATIC_KEY_TRUE(sk_dynamic_irqentry_exit_cond_resched); 161 void dynamic_irqentry_exit_cond_resched(void) 162 { 163 if (!static_branch_unlikely(&sk_dynamic_irqentry_exit_cond_resched)) 164 return; 165 raw_irqentry_exit_cond_resched(); 166 } 167 #endif 168 #endif 169 170 noinstr void irqentry_exit(struct pt_regs *regs, irqentry_state_t state) 171 { 172 lockdep_assert_irqs_disabled(); 173 174 /* Check whether this returns to user mode */ 175 if (user_mode(regs)) { 176 irqentry_exit_to_user_mode(regs); 177 } else if (!regs_irqs_disabled(regs)) { 178 /* 179 * If RCU was not watching on entry this needs to be done 180 * carefully and needs the same ordering of lockdep/tracing 181 * and RCU as the return to user mode path. 182 */ 183 if (state.exit_rcu) { 184 instrumentation_begin(); 185 /* Tell the tracer that IRET will enable interrupts */ 186 trace_hardirqs_on_prepare(); 187 lockdep_hardirqs_on_prepare(); 188 instrumentation_end(); 189 ct_irq_exit(); 190 lockdep_hardirqs_on(CALLER_ADDR0); 191 return; 192 } 193 194 instrumentation_begin(); 195 if (IS_ENABLED(CONFIG_PREEMPTION)) 196 irqentry_exit_cond_resched(); 197 198 /* Covers both tracing and lockdep */ 199 trace_hardirqs_on(); 200 instrumentation_end(); 201 } else { 202 /* 203 * IRQ flags state is correct already. Just tell RCU if it 204 * was not watching on entry. 205 */ 206 if (state.exit_rcu) 207 ct_irq_exit(); 208 } 209 } 210 211 irqentry_state_t noinstr irqentry_nmi_enter(struct pt_regs *regs) 212 { 213 irqentry_state_t irq_state; 214 215 irq_state.lockdep = lockdep_hardirqs_enabled(); 216 217 __nmi_enter(); 218 lockdep_hardirqs_off(CALLER_ADDR0); 219 lockdep_hardirq_enter(); 220 ct_nmi_enter(); 221 222 instrumentation_begin(); 223 kmsan_unpoison_entry_regs(regs); 224 trace_hardirqs_off_finish(); 225 ftrace_nmi_enter(); 226 instrumentation_end(); 227 228 return irq_state; 229 } 230 231 void noinstr irqentry_nmi_exit(struct pt_regs *regs, irqentry_state_t irq_state) 232 { 233 instrumentation_begin(); 234 ftrace_nmi_exit(); 235 if (irq_state.lockdep) { 236 trace_hardirqs_on_prepare(); 237 lockdep_hardirqs_on_prepare(); 238 } 239 instrumentation_end(); 240 241 ct_nmi_exit(); 242 lockdep_hardirq_exit(); 243 if (irq_state.lockdep) 244 lockdep_hardirqs_on(CALLER_ADDR0); 245 __nmi_exit(); 246 } 247