| 1 | /* @title: Per-CPU structure */ |
| 2 | #pragma once |
| 3 | #include <sch/irql.h> |
| 4 | #include <smp/topology.h> |
| 5 | #include <stdatomic.h> |
| 6 | #include <stdbool.h> |
| 7 | #include <stddef.h> |
| 8 | #include <stdint.h> |
| 9 | #include <thread/dpc.h> |
| 10 | #include <types/types.h> |
| 11 | |
| 12 | #define CPU_FEAT_SSE2 (1ULL << 0) |
| 13 | #define CPU_FEAT_AVX (1ULL << 1) |
| 14 | #define CPU_FEAT_AVX2 (1ULL << 2) |
| 15 | #define CPU_FEAT_AVX512F (1ULL << 3) |
| 16 | |
| 17 | enum cpu_class { |
| 18 | CPU_CLASS_UNKNOWN, |
| 19 | CPU_CLASS_PERFORMANCE, |
| 20 | CPU_CLASS_EFFICIENCY, |
| 21 | }; |
| 22 | |
| 23 | enum { |
| 24 | UARCH_UNKNOWN, |
| 25 | UARCH_GOLDEN_COVE, |
| 26 | UARCH_GRACEMONT, |
| 27 | UARCH_SKYLAKE, |
| 28 | }; |
| 29 | |
| 30 | struct cpu_capability { |
| 31 | enum cpu_class class; |
| 32 | |
| 33 | uint32_t uarch_id; /* e.g. golden cove, gracemont */ |
| 34 | |
| 35 | uint32_t issue_width; |
| 36 | uint32_t retire_width; |
| 37 | |
| 38 | cpu_perf_t perf_score; /* Relative to everyone else on a 0-255 |
| 39 | * scale, how performant are we? The |
| 40 | * higher this number is, the more "performant" |
| 41 | * this CPU currently is, and the more likely |
| 42 | * the scheduler will decide to migrate a |
| 43 | * thread that needs such perf scores onto here. */ |
| 44 | |
| 45 | uint32_t energy_score; /* lower is better */ |
| 46 | |
| 47 | uint64_t feature_bits; /* ISA features, vector width, etc */ |
| 48 | }; |
| 49 | |
| 50 | /* Let's put commonly accessed fields up here |
| 51 | * to make the cache a bit happier */ |
| 52 | struct core { |
| 53 | struct core *self; |
| 54 | cpu_id_t id; |
| 55 | struct thread *current_thread; |
| 56 | struct cpu_capability cap; |
| 57 | |
| 58 | size_t domain_cpu_id; /* what CPU in the domain? */ |
| 59 | |
| 60 | /* array [domain_levels_enabled] -> domain reference */ |
| 61 | struct scheduler_domain *domains[TOPOLOGY_LEVEL_MAX]; |
| 62 | |
| 63 | /* index within each domain's groups */ |
| 64 | int32_t group_index[TOPOLOGY_LEVEL_MAX]; |
| 65 | |
| 66 | atomic_bool executing_dpcs; |
| 67 | atomic_bool idle; |
| 68 | |
| 69 | /* This scratch buffer is stack allocated, and is set upon |
| 70 | * IRQ entry, allowing the top half of the IRQ to modify it |
| 71 | * |
| 72 | * For exception_sync_cb, it is passed into the callback as a parameter */ |
| 73 | uint8_t *irq_stack_scratch_buf; |
| 74 | bool in_interrupt; |
| 75 | enum irql current_irql; |
| 76 | |
| 77 | /* Remains valid in the top half, once the bottom half is |
| 78 | * reached, this becomes IRQL_NONE */ |
| 79 | enum irql irq_entered_irql; /* What IRQL were we at before |
| 80 | * entering an ISR (if !in_interrupt, |
| 81 | * this should be IRQL_NONE */ |
| 82 | |
| 83 | enum dpc_event dpc_event; |
| 84 | |
| 85 | atomic_bool needs_resched; |
| 86 | atomic_bool in_resched; /* in scheduler_yield() */ |
| 87 | uint32_t preempt_disable_depth; |
| 88 | |
| 89 | struct domain *domain; |
| 90 | struct domain_arena *domain_arena; |
| 91 | size_t rr_current_domain; |
| 92 | |
| 93 | struct tss *tss; |
| 94 | |
| 95 | uint32_t lapic_freq; |
| 96 | |
| 97 | struct topology_node *topo_node; |
| 98 | struct topology_cache_info llc; |
| 99 | |
| 100 | numa_node_t numa_node; |
| 101 | uint32_t package_id; |
| 102 | uint32_t smt_mask; |
| 103 | uint32_t smt_id; |
| 104 | uint32_t core_id; |
| 105 | |
| 106 | uint64_t tsc_hz; |
| 107 | uint64_t last_us; |
| 108 | uint64_t last_tsc; /* For time.c */ |
| 109 | |
| 110 | _Atomic uint64_t pt_seen_epoch; |
| 111 | bool reclaiming_page_tables; |
| 112 | }; |
| 113 | |
| 114 | static inline uint64_t smp_core_id() { |
| 115 | uint64_t id; |
| 116 | asm volatile("movq %%gs:%c1, %0" |
| 117 | : "=r" (id) |
| 118 | : "i" (offsetof(struct core, id))); |
| 119 | return id; |
| 120 | } |
| 121 | |
| 122 | static inline struct core *smp_core(void) { |
| 123 | uintptr_t core; |
| 124 | asm volatile("movq %%gs:%c1, %0" |
| 125 | : "=r" (core) |
| 126 | : "i" (offsetof(struct core, self))); |
| 127 | return (struct core *) core; |
| 128 | } |
| 129 | |
| 130 | #define for_each_cpu_struct(__iter) \ |
| 131 | for (size_t __id = 0; \ |
| 132 | ((__iter = global.cores[__id]), __id < global.core_count); __id++) |
| 133 | |
| 134 | #define for_each_cpu_id(__id) for (__id = 0; __id < global.core_count; __id++) |
| 135 | |