Kernel: Fix RTL8169 driver

This now works properly on real hardware. Main issue was a race
condition with now ISR was handled in the interrupt handler. We now loop
until we read back ISR as zero
This commit is contained in:
2026-06-27 20:46:00 +03:00
parent 924576cf0d
commit 6339044e4c
3 changed files with 127 additions and 114 deletions

View File

@@ -7,25 +7,25 @@ namespace Kernel
enum RTL8169_IO_REGS : uint16_t enum RTL8169_IO_REGS : uint16_t
{ {
RTL8169_IO_IDR0 = 0x00, RTL8169_IO_IDR0 = 0x00,
RTL8169_IO_IDR1 = 0x01, RTL8169_IO_IDR1 = 0x01,
RTL8169_IO_IDR2 = 0x02, RTL8169_IO_IDR2 = 0x02,
RTL8169_IO_IDR3 = 0x03, RTL8169_IO_IDR3 = 0x03,
RTL8169_IO_IDR4 = 0x04, RTL8169_IO_IDR4 = 0x04,
RTL8169_IO_IDR5 = 0x05, RTL8169_IO_IDR5 = 0x05,
RTL8169_IO_TNPDS = 0x20, RTL8169_IO_TNPDS = 0x20,
RTL8169_IO_CR = 0x37, RTL8169_IO_CR = 0x37,
RTL8169_IO_TPPoll = 0x38, RTL8169_IO_TPPoll = 0x38,
RTL8169_IO_IMR = 0x3C, RTL8169_IO_IMR = 0x3C,
RTL8169_IO_ISR = 0x3E, RTL8169_IO_ISR = 0x3E,
RTL8169_IO_TCR = 0x40, RTL8169_IO_TCR = 0x40,
RTL8169_IO_RCR = 0x44, RTL8169_IO_RCR = 0x44,
RTL8169_IO_9346CR = 0x50, RTL8169_IO_PHYSts = 0x6C,
RTL8169_IO_PHYSts = 0x6C, RTL8169_IO_RMS = 0xDA,
RTL8169_IO_RMS = 0xDA, RTL8169_IO_CPlusCR = 0xE0,
RTL8169_IO_RDSAR = 0xE4, RTL8169_IO_RDSAR = 0xE4,
RTL8169_IO_MTPS = 0xEC, RTL8169_IO_MTPS = 0xEC,
}; };
enum RTL8169_CR : uint8_t enum RTL8169_CR : uint8_t

View File

@@ -64,14 +64,17 @@ namespace Kernel
BAN::UniqPtr<DMARegion> m_rx_descriptor_region; BAN::UniqPtr<DMARegion> m_rx_descriptor_region;
BAN::UniqPtr<DMARegion> m_tx_descriptor_region; BAN::UniqPtr<DMARegion> m_tx_descriptor_region;
SpinLock m_lock; BAN::Atomic<bool> m_rx_thread_should_die { false };
BAN::Atomic<bool> m_rx_thread_is_dead { true };
bool m_thread_should_die { false }; uint32_t m_rx_head { 0 };
BAN::Atomic<bool> m_thread_is_dead { true }; SpinLock m_rx_lock;
ThreadBlocker m_thread_blocker; ThreadBlocker m_rx_blocker;
uint32_t m_rx_current { 0 }; BAN::Atomic<uint32_t> m_tx_head { 0 };
size_t m_tx_current { 0 }; BAN::Atomic<uint32_t> m_tx_commit { 0 };
SpinLock m_tx_lock;
ThreadBlocker m_tx_blocker;
BAN::MACAddress m_mac_address {}; BAN::MACAddress m_mac_address {};
BAN::Atomic<bool> m_link_up { false }; BAN::Atomic<bool> m_link_up { false };

View File

@@ -47,30 +47,29 @@ namespace Kernel
TRY(reset()); TRY(reset());
// disable rx, tx, checksum offload
m_io_bar_region->write16(RTL8169_IO_CPlusCR, 0x0000);
m_io_bar_region->write8 (RTL8169_IO_CR, 0x00);
dprintln(" reset done"); dprintln(" reset done");
for (size_t i = 0; i < 6; i++) for (size_t i = 0; i < 6; i++)
m_mac_address.address[i] = m_io_bar_region->read8(RTL8169_IO_IDR0 + i); m_mac_address.address[i] = m_io_bar_region->read8(RTL8169_IO_IDR0 + i);
dprintln(" MAC {}", m_mac_address); dprintln(" MAC {}", m_mac_address);
// unlock config registers
m_io_bar_region->write8(RTL8169_IO_9346CR, RTL8169_9346CR_MODE_CONFIG);
TRY(initialize_rx()); TRY(initialize_rx());
TRY(initialize_tx()); TRY(initialize_tx());
m_io_bar_region->write8(RTL8169_IO_CR, RTL8169_CR_RE | RTL8169_CR_TE); m_io_bar_region->write8(RTL8169_IO_CR, RTL8169_CR_RE | RTL8169_CR_TE);
dprintln(" descriptors initialized"); dprintln(" descriptors initialized");
m_link_up = m_io_bar_region->read8(RTL8169_IO_PHYSts) & RTL8169_PHYSts_LinkSts; m_link_up = m_io_bar_region->read8(RTL8169_IO_PHYSts) & RTL8169_PHYSts_LinkSts;
if (m_link_up) dprintln(" link status {}", link_up() ? "UP" : "DOWN");
if (link_up())
dprintln(" link speed {}", link_speed()); dprintln(" link speed {}", link_speed());
TRY(enable_interrupt()); TRY(enable_interrupt());
dprintln(" interrupts enabled"); dprintln(" interrupts enabled");
// lock config registers
m_io_bar_region->write8(RTL8169_IO_9346CR, RTL8169_9346CR_MODE_NORMAL);
auto* thread = TRY(Thread::create_kernel([](void* rtl8169_ptr) { auto* thread = TRY(Thread::create_kernel([](void* rtl8169_ptr) {
static_cast<RTL8169*>(rtl8169_ptr)->receive_thread(); static_cast<RTL8169*>(rtl8169_ptr)->receive_thread();
}, this)); }, this));
@@ -79,17 +78,17 @@ namespace Kernel
delete thread; delete thread;
return ret.release_error(); return ret.release_error();
} }
m_thread_is_dead = false; m_rx_thread_is_dead = false;
return {}; return {};
} }
RTL8169::~RTL8169() RTL8169::~RTL8169()
{ {
m_thread_should_die = true; m_rx_thread_should_die = true;
m_thread_blocker.unblock(); m_rx_blocker.unblock();
while (!m_thread_is_dead) while (!m_rx_thread_is_dead)
Processor::yield(); Processor::yield();
} }
@@ -110,32 +109,26 @@ namespace Kernel
m_rx_buffer_region = TRY(DMARegion::create(m_rx_descriptor_count * s_buffer_size, PageTable::MemoryType::Normal)); m_rx_buffer_region = TRY(DMARegion::create(m_rx_descriptor_count * s_buffer_size, PageTable::MemoryType::Normal));
m_rx_descriptor_region = TRY(DMARegion::create(m_rx_descriptor_count * sizeof(RTL8169Descriptor))); m_rx_descriptor_region = TRY(DMARegion::create(m_rx_descriptor_count * sizeof(RTL8169Descriptor)));
auto* rx_descriptors = reinterpret_cast<volatile RTL8169Descriptor*>(m_rx_descriptor_region->vaddr());
for (size_t i = 0; i < m_rx_descriptor_count; i++) for (size_t i = 0; i < m_rx_descriptor_count; i++)
{ {
const paddr_t rx_buffer_paddr = m_rx_buffer_region->paddr() + i * s_buffer_size; const paddr_t rx_buffer_paddr = m_rx_buffer_region->paddr() + i * s_buffer_size;
rx_descriptors[i].command = 0x1FF8 | RTL8169_DESC_CMD_OWN;
uint32_t command = 0x1FF8 | RTL8169_DESC_CMD_OWN; rx_descriptors[i].vlan = 0;
if (i == m_rx_descriptor_count - 1) rx_descriptors[i].buffer_low = rx_buffer_paddr & 0xFFFFFFFF;
command |= RTL8169_DESC_CMD_EOR; rx_descriptors[i].buffer_high = rx_buffer_paddr >> 32;
auto& rx_descriptor = reinterpret_cast<volatile RTL8169Descriptor*>(m_rx_descriptor_region->vaddr())[i];
rx_descriptor.command = command;
rx_descriptor.vlan = 0;
rx_descriptor.buffer_low = rx_buffer_paddr & 0xFFFFFFFF;
rx_descriptor.buffer_high = rx_buffer_paddr >> 32;
} }
rx_descriptors[m_rx_descriptor_count - 1].command |= RTL8169_DESC_CMD_EOR;
// configure rx descriptor addresses // configure rx descriptor addresses
m_io_bar_region->write32(RTL8169_IO_RDSAR + 0, m_rx_descriptor_region->paddr() & 0xFFFFFFFF);
m_io_bar_region->write32(RTL8169_IO_RDSAR + 4, m_rx_descriptor_region->paddr() >> 32); m_io_bar_region->write32(RTL8169_IO_RDSAR + 4, m_rx_descriptor_region->paddr() >> 32);
m_io_bar_region->write32(RTL8169_IO_RDSAR + 0, m_rx_descriptor_region->paddr() & 0xFFFFFFFF);
// configure receibe control (no fifo threshold, max dma burst 1024, accept physical match, broadcast, multicast) // configure receive control (no fifo threshold, max dma burst unlimited, broadcast, multicast, accept physical match)
m_io_bar_region->write32(RTL8169_IO_RCR, m_io_bar_region->write32(RTL8169_IO_RCR,
RTL8169_RCR_RXFTH_NO | RTL8169_RCR_MXDMA_1024 | RTL8169_RCR_AB | RTL8169_RCR_AM | RTL8169_RCR_APM RTL8169_RCR_RXFTH_NO | RTL8169_RCR_MXDMA_UNLIMITED | RTL8169_RCR_AB | RTL8169_RCR_AM | RTL8169_RCR_APM
); );
m_io_bar_region->write32(0x44, 0b111111 | (0b111 << 8) | (0b111 << 13));
// configure max rx packet size // configure max rx packet size
m_io_bar_region->write16(RTL8169_IO_RMS, RTL8169_RMS_MAX); m_io_bar_region->write16(RTL8169_IO_RMS, RTL8169_RMS_MAX);
@@ -147,27 +140,23 @@ namespace Kernel
m_tx_buffer_region = TRY(DMARegion::create(m_tx_descriptor_count * s_buffer_size, PageTable::MemoryType::Normal)); m_tx_buffer_region = TRY(DMARegion::create(m_tx_descriptor_count * s_buffer_size, PageTable::MemoryType::Normal));
m_tx_descriptor_region = TRY(DMARegion::create(m_tx_descriptor_count * sizeof(RTL8169Descriptor))); m_tx_descriptor_region = TRY(DMARegion::create(m_tx_descriptor_count * sizeof(RTL8169Descriptor)));
auto* tx_descriptors = reinterpret_cast<volatile RTL8169Descriptor*>(m_tx_descriptor_region->vaddr());
for (size_t i = 0; i < m_tx_descriptor_count; i++) for (size_t i = 0; i < m_tx_descriptor_count; i++)
{ {
const paddr_t tx_buffer_paddr = m_tx_buffer_region->paddr() + i * s_buffer_size; const paddr_t tx_buffer_paddr = m_tx_buffer_region->paddr() + i * s_buffer_size;
tx_descriptors[i].command = 0;
uint32_t command = 0; tx_descriptors[i].vlan = 0;
if (i == m_tx_descriptor_count - 1) tx_descriptors[i].buffer_low = tx_buffer_paddr & 0xFFFFFFFF;
command |= RTL8169_DESC_CMD_EOR; tx_descriptors[i].buffer_high = tx_buffer_paddr >> 32;
auto& tx_descriptor = reinterpret_cast<volatile RTL8169Descriptor*>(m_tx_descriptor_region->vaddr())[i];
tx_descriptor.command = command;
tx_descriptor.vlan = 0;
tx_descriptor.buffer_low = tx_buffer_paddr & 0xFFFFFFFF;
tx_descriptor.buffer_high = tx_buffer_paddr >> 32;
} }
tx_descriptors[m_tx_descriptor_count - 1].command |= RTL8169_DESC_CMD_EOR;
// configure tx descriptor addresses // configure tx descriptor addresses
m_io_bar_region->write32(RTL8169_IO_TNPDS + 0, m_tx_descriptor_region->paddr() & 0xFFFFFFFF);
m_io_bar_region->write32(RTL8169_IO_TNPDS + 4, m_tx_descriptor_region->paddr() >> 32); m_io_bar_region->write32(RTL8169_IO_TNPDS + 4, m_tx_descriptor_region->paddr() >> 32);
m_io_bar_region->write32(RTL8169_IO_TNPDS + 0, m_tx_descriptor_region->paddr() & 0xFFFFFFFF);
// configure transmit control (standard ifg, max dma burst 1024) // configure transmit control (standard ifg, max dma burst unlimited)
m_io_bar_region->write32(RTL8169_IO_TCR, RTL8169_TCR_IFG_0 | RTL8169_TCR_MXDMA_1024); m_io_bar_region->write32(RTL8169_IO_TCR, RTL8169_TCR_IFG_0 | RTL8169_TCR_MXDMA_UNLIMITED);
// configure max tx packet size // configure max tx packet size
m_io_bar_region->write8(RTL8169_IO_MTPS, RTL8169_MTPS_MAX); m_io_bar_region->write8(RTL8169_IO_MTPS, RTL8169_MTPS_MAX);
@@ -181,14 +170,13 @@ namespace Kernel
m_pci_device.enable_interrupt(0, *this); m_pci_device.enable_interrupt(0, *this);
m_io_bar_region->write16(RTL8169_IO_IMR, m_io_bar_region->write16(RTL8169_IO_IMR,
RTL8169_IR_ROK RTL8169_IR_ROK |
| RTL8169_IR_RER RTL8169_IR_RER |
| RTL8169_IR_TOK RTL8169_IR_TOK |
| RTL8169_IR_TER RTL8169_IR_TER |
| RTL8169_IR_RDU RTL8169_IR_RDU |
| RTL8169_IR_LinkChg RTL8169_IR_LinkChg |
| RTL8169_IR_FVOW RTL8169_IR_FVOW
| RTL8169_IR_TDU
); );
m_io_bar_region->write16(RTL8169_IO_ISR, 0xFFFF); m_io_bar_region->write16(RTL8169_IO_ISR, 0xFFFF);
@@ -214,20 +202,23 @@ namespace Kernel
if (!link_up()) if (!link_up())
return BAN::Error::from_errno(EADDRNOTAVAIL); return BAN::Error::from_errno(EADDRNOTAVAIL);
auto state = m_lock.lock(); const auto interrupt_state = Processor::get_interrupt_state();
Processor::set_interrupt_state(InterruptState::Disabled);
const uint32_t tx_current = m_tx_current; const uint32_t tx_current_nowrap = m_tx_head.fetch_add(1);
m_tx_current = (m_tx_current + 1) % m_tx_descriptor_count; const uint32_t tx_current = tx_current_nowrap % m_tx_descriptor_count;
auto& descriptor = reinterpret_cast<volatile RTL8169Descriptor*>(m_tx_descriptor_region->vaddr())[tx_current]; auto& descriptor = reinterpret_cast<volatile RTL8169Descriptor*>(m_tx_descriptor_region->vaddr())[tx_current];
while (descriptor.command & RTL8169_DESC_CMD_OWN) if (descriptor.command & RTL8169_DESC_CMD_OWN)
{ {
SpinLockAsMutex smutex(m_lock, state); SpinLockGuard guard(m_tx_lock);
m_thread_blocker.block_indefinite(&smutex); while (descriptor.command & RTL8169_DESC_CMD_OWN)
{
SpinLockGuardAsMutex smutex(guard);
m_tx_blocker.block_indefinite(&smutex);
}
} }
m_lock.unlock(state);
auto* tx_buffer = reinterpret_cast<uint8_t*>(m_tx_buffer_region->vaddr() + tx_current * s_buffer_size); auto* tx_buffer = reinterpret_cast<uint8_t*>(m_tx_buffer_region->vaddr() + tx_current * s_buffer_size);
// write packet // write packet
@@ -245,25 +236,31 @@ namespace Kernel
// give packet ownership to NIC // give packet ownership to NIC
uint32_t command = packet_size | RTL8169_DESC_CMD_OWN | RTL8169_DESC_CMD_LS | RTL8169_DESC_CMD_FS; uint32_t command = packet_size | RTL8169_DESC_CMD_OWN | RTL8169_DESC_CMD_LS | RTL8169_DESC_CMD_FS;
if (tx_current >= m_tx_descriptor_count - 1) if (tx_current == m_tx_descriptor_count - 1)
command |= RTL8169_DESC_CMD_EOR; command |= RTL8169_DESC_CMD_EOR;
descriptor.command = command; descriptor.command = command;
// notify NIC about new packet // ring tx queue doorbell
m_io_bar_region->write8(RTL8169_IO_TPPoll, RTL8169_TPPoll_NPQ); if (tx_current_nowrap == m_tx_commit.load())
m_io_bar_region->write8(RTL8169_IO_TPPoll, RTL8169_TPPoll_NPQ);
while (tx_current_nowrap != m_tx_commit.load())
Processor::pause();
m_tx_commit.add_fetch(1);
Processor::set_interrupt_state(interrupt_state);
return {}; return {};
} }
void RTL8169::receive_thread() void RTL8169::receive_thread()
{ {
SpinLockGuard _(m_lock); SpinLockGuard rx_lock_guard(m_rx_lock);
while (!m_thread_should_die) while (!m_rx_thread_should_die)
{ {
for (;;) for (;;)
{ {
auto& descriptor = reinterpret_cast<volatile RTL8169Descriptor*>(m_rx_descriptor_region->vaddr())[m_rx_current]; auto& descriptor = reinterpret_cast<volatile RTL8169Descriptor*>(m_rx_descriptor_region->vaddr())[m_rx_head];
if (descriptor.command & RTL8169_DESC_CMD_OWN) if (descriptor.command & RTL8169_DESC_CMD_OWN)
break; break;
@@ -277,54 +274,67 @@ namespace Kernel
; // descriptor has an error ; // descriptor has an error
else else
{ {
m_lock.unlock(InterruptState::Enabled); m_rx_lock.unlock(InterruptState::Enabled);
NetworkManager::get().on_receive(*this, BAN::ConstByteSpan { NetworkManager::get().on_receive(*this, BAN::ConstByteSpan {
reinterpret_cast<const uint8_t*>(m_rx_buffer_region->vaddr() + m_rx_current * s_buffer_size), reinterpret_cast<const uint8_t*>(m_rx_buffer_region->vaddr() + m_rx_head * s_buffer_size),
packet_length packet_length
}); });
m_lock.lock(); m_rx_lock.lock();
} }
m_rx_current = (m_rx_current + 1) % m_rx_descriptor_count; uint32_t command = 0x1FF8 | RTL8169_DESC_CMD_OWN;
if (m_rx_head == m_rx_descriptor_count - 1)
command |= RTL8169_DESC_CMD_EOR;
descriptor.command = command;
descriptor.command = descriptor.command | RTL8169_DESC_CMD_OWN; m_rx_head = (m_rx_head + 1) % m_rx_descriptor_count;
} }
SpinLockAsMutex smutex(m_lock, InterruptState::Enabled); SpinLockGuardAsMutex smutex(rx_lock_guard);
m_thread_blocker.block_indefinite(&smutex); m_rx_blocker.block_indefinite(&smutex);
} }
m_thread_is_dead = true; m_rx_thread_is_dead = true;
} }
void RTL8169::handle_irq() void RTL8169::handle_irq()
{ {
const uint16_t interrupt_status = m_io_bar_region->read16(RTL8169_IO_ISR); uint16_t isr;
m_io_bar_region->write16(RTL8169_IO_ISR, interrupt_status); while ((isr = m_io_bar_region->read16(RTL8169_IO_ISR)))
if (interrupt_status & RTL8169_IR_LinkChg)
{ {
m_link_up = m_io_bar_region->read8(RTL8169_IO_PHYSts) & RTL8169_PHYSts_LinkSts; m_io_bar_region->write16(RTL8169_IO_ISR, isr);
dprintln("link status -> {}", m_link_up.load());
}
if (interrupt_status & (RTL8169_IR_TOK | RTL8169_IR_ROK)) if (isr & RTL8169_IR_LinkChg)
{ {
SpinLockGuard _(m_lock); m_link_up = m_io_bar_region->read8(RTL8169_IO_PHYSts) & RTL8169_PHYSts_LinkSts;
m_thread_blocker.unblock(); dprintln("link status {}", link_up() ? "UP" : "DOWN");
} if (link_up())
dprintln("link speed {}", link_speed());
}
if (interrupt_status & RTL8169_IR_RER) if (isr & (RTL8169_IR_TER | RTL8169_IR_TOK))
dwarnln("Rx error"); {
if (interrupt_status & RTL8169_IR_TER) SpinLockGuard _(m_tx_lock);
dwarnln("Tx error"); m_tx_blocker.unblock();
if (interrupt_status & RTL8169_IR_RDU) }
dwarnln("Rx descriptor not available");
if (interrupt_status & RTL8169_IR_FVOW) if (isr & (RTL8169_IR_RER | RTL8169_IR_ROK))
dwarnln("Rx FIFO overflow"); {
// dont log TDU is sent after each sent packet SpinLockGuard _(m_rx_lock);
m_rx_blocker.unblock();
}
if (isr & RTL8169_IR_RER)
dwarnln("Rx error");
if (isr & RTL8169_IR_TER)
dwarnln("Tx error");
if (isr & RTL8169_IR_RDU)
dwarnln("Rx descriptor not available");
if (isr & RTL8169_IR_FVOW)
dwarnln("Rx FIFO overflow");
}
} }
} }