Kernel: Optimize networking code
Remove buffering from network layer and rework loopback interface. loopback now has a separate recieve thread to allow concurrent sends and prevent deadlocks
This commit is contained in:
@@ -9,6 +9,7 @@ namespace Kernel
|
||||
{
|
||||
public:
|
||||
static constexpr size_t buffer_size = BAN::numeric_limits<uint16_t>::max() + 1;
|
||||
static constexpr size_t buffer_count = 32;
|
||||
|
||||
public:
|
||||
static BAN::ErrorOr<BAN::RefPtr<LoopbackInterface>> create();
|
||||
@@ -24,8 +25,9 @@ namespace Kernel
|
||||
LoopbackInterface()
|
||||
: NetworkInterface(Type::Loopback)
|
||||
{}
|
||||
~LoopbackInterface();
|
||||
|
||||
BAN::ErrorOr<void> send_bytes(BAN::MACAddress destination, EtherType protocol, BAN::ConstByteSpan) override;
|
||||
BAN::ErrorOr<void> send_bytes(BAN::MACAddress destination, EtherType protocol, BAN::Span<const BAN::ConstByteSpan> payload) override;
|
||||
|
||||
bool can_read_impl() const override { return false; }
|
||||
bool can_write_impl() const override { return false; }
|
||||
@@ -33,8 +35,27 @@ namespace Kernel
|
||||
bool has_hungup_impl() const override { return false; }
|
||||
|
||||
private:
|
||||
SpinLock m_buffer_lock;
|
||||
void receive_thread();
|
||||
|
||||
private:
|
||||
struct Descriptor
|
||||
{
|
||||
uint8_t* addr;
|
||||
uint32_t size;
|
||||
uint8_t state;
|
||||
};
|
||||
|
||||
private:
|
||||
Mutex m_buffer_lock;
|
||||
BAN::UniqPtr<VirtualRange> m_buffer;
|
||||
|
||||
uint32_t m_buffer_tail { 0 };
|
||||
uint32_t m_buffer_head { 0 };
|
||||
Descriptor m_descriptors[buffer_count] {};
|
||||
|
||||
bool m_thread_should_die { false };
|
||||
BAN::Atomic<bool> m_thread_is_dead { true };
|
||||
ThreadBlocker m_thread_blocker;
|
||||
};
|
||||
|
||||
}
|
||||
|
||||
Reference in New Issue
Block a user