RGBCube/serenity
1
Fork 0
mirror of https://github.com/RGBCube/serenity synced 2025-07-25 16:27:35 +00:00
Code Issues Activity Actions

Kernel: Use proper Atomic<T> types in CPU

Browse source 
This is needed because Clang's intrinsic atomic functions behave weirdly
if only one of their pointer arguments is volatile.
This commit is contained in:
Daniel Bertalan 2021-06-24 10:03:07 +02:00 • committed by Ali Mohammad Pur
parent f28f00c654
commit 74535628a8
2 changed files with 20 additions and 21 deletions
Download patch file Download diff file Expand all files Collapse all files

10
Kernel/Arch/x86/Processor.h
View file

@ -40,7 +40,7 @@ struct ProcessorMessage {
Callback,
};
Type type;
volatile u32 refs; // atomic
Atomic<u32> refs;
union {
ProcessorMessage* next; // only valid while in the pool
alignas(CallbackFunction) u8 callback_storage[sizeof(CallbackFunction)];
@ -115,7 +115,7 @@ class Processor {
TSS m_tss;
static FPUState s_clean_fpu_state;
CPUFeature m_features;
static volatile u32 g_total_processors; // atomic
static Atomic<u32> g_total_processors;
u8 m_physical_address_bit_width;
ProcessorInfo* m_info;
@ -124,7 +124,7 @@ class Processor {
Thread* m_current_thread;
Thread* m_idle_thread;
volatile ProcessorMessageEntry* m_message_queue; // atomic, LIFO
Atomic<ProcessorMessageEntry*> m_message_queue;
bool m_invoke_scheduler_async;
bool m_scheduler_initialized;
@ -178,8 +178,8 @@ public:
static u32 count()
{
// NOTE: because this value never changes once all APs are booted,
// we don't really need to do an atomic_load() on this variable
return g_total_processors;
// we can safely bypass loading it atomically.
return *g_total_processors.ptr();
}
ALWAYS_INLINE static void wait_check()