blob: c6421d0bb57089224055b3c227f0b3a7ae5b9f09 [file] [log] [blame]
// Copyright 2023 The Fuchsia Authors
//
// Use of this source code is governed by a MIT-style
// license that can be found in the LICENSE file or at
// https://opensource.org/licenses/MIT
#include <lib/boot-options/boot-options.h>
#include <lib/cbuf.h>
#include <lib/debuglog.h>
#include <lib/uart/all.h>
#include <lib/uart/null.h>
#include <lib/uart/qemu.h>
#include <lib/uart/uart.h>
#include <lib/zbi-format/driver-config.h>
#include <lib/zircon-internal/macros.h>
#include <lib/zircon-internal/thread_annotations.h>
#include <lib/zx/time.h>
#include <stdint.h>
#include <zircon/errors.h>
#include <zircon/time.h>
#include <cassert>
#include <type_traits>
#include <arch/arch_interrupt.h>
#include <dev/init.h>
#include <dev/interrupt.h>
#include <kernel/deadline.h>
#include <kernel/spinlock.h>
#include <kernel/timer.h>
#include <ktl/optional.h>
#include <ktl/type_traits.h>
#include <ktl/variant.h>
#include <lockdep/guard.h>
#include <platform/debug.h>
#include <platform/legacy_debug.h>
#include <platform/uart.h>
#include "platform.h"
namespace {
// No locks will be acquired.
struct NullLockPolicy {};
// Temporary flag for conditionally delegating to legacy_platform_* variants.
bool use_new_serial = false;
bool is_tx_irq_enabled = false;
bool is_serial_enabled = false;
// TODO(https://fxbug.dev/42073424): Switch to lockdep::NullGuard once fbl::NullLock
// and NullGuard are both annotated, and the existing use cases are
// TA compliant.
class TA_SCOPED_CAP NullGuard {
public:
template <typename Lockable>
NullGuard(Lockable* lock, const char*) TA_ACQ(lock) {}
~NullGuard() TA_REL() {}
template <typename T>
void CallUnlocked(T&& op) {
op();
}
};
template <typename LockPolicy, typename Guard>
using GuardSelector =
std::conditional_t<std::is_same_v<LockPolicy, NullLockPolicy>, NullGuard, Guard>;
// Implements SyncPolicy as defined in <lib/uart/sync.h>
struct UartSyncPolicy {
template <typename MemberOf>
using Lock = DECLARE_SPINLOCK_WITH_TYPE(MemberOf, MonitoredSpinLock);
template <typename LockPolicy>
using Guard = GuardSelector<LockPolicy, Guard<MonitoredSpinLock, LockPolicy>>;
class Waiter {
public:
template <typename Guard, typename T>
void Wait(Guard& guard, T&& enable_tx_interrupt) TA_REQ(guard) {
if (is_tx_irq_enabled) {
enable_tx_interrupt();
guard.CallUnlocked([this]() { tx_fifo_not_full_.Wait(); });
} else {
arch::Yield();
}
}
void Wake() { tx_fifo_not_full_.Signal(); }
private:
AutounsignalEvent tx_fifo_not_full_{true};
};
using DefaultLockPolicy = NullLockPolicy;
template <typename LockType>
static void AssertHeld(LockType& lock) TA_ASSERT(lock) TA_ASSERT(lock.lock()) {
lock.lock().AssertHeld();
}
};
uart::all::KernelDriver<PlatformUartIoProvider, UartSyncPolicy> gUart;
// Initialized by UartInitLate, provides buffered output of the uart,
// which helps readers catch up.
// Also provides synchronization mechanisms for character availability.
Cbuf rx_queue;
// Size of the rx queue. The bigger the buffer, the bigger the window for
// the reader to catch up. Useful when the incoming data is bursty.
constexpr size_t kRxQueueSize = 1024;
// When Polling is enabled, this will fire the polling callback for draining UART's RX Queue.
Timer gUartPollTimer;
constexpr zx_duration_t kPollingPeriod = ZX_MSEC(10);
constexpr TimerSlack kPollingSlack = {ZX_MSEC(10), TIMER_SLACK_CENTER};
// Callback used by |gUartPollTimer| when deadline is met. See |Timer| interface for more
// information.
template <bool DrainUart>
void UartPoll(Timer* uart_timer, zx_time_t now, void* arg) {
uart_timer->Set(Deadline(zx_time_add_duration(now, kPollingPeriod), kPollingSlack),
&UartPoll<true>, nullptr);
if constexpr (DrainUart) {
gUart.Visit([&](auto& driver) {
// Drain until there is nothing else in the RX Queue of the device.
while (auto c = driver.Read()) {
rx_queue.WriteChar(*c);
}
});
}
}
} // namespace
bool platform_serial_enabled(void) { return is_serial_enabled; }
void UartDriverHandoffEarly(const uart::all::Driver& serial) {
ktl::visit(
[&](auto& driver) {
is_serial_enabled = !(ktl::is_same_v<ktl::decay_t<decltype(driver)>, uart::null::Driver>);
},
serial);
use_new_serial = gBootOptions->experimental_serial_migration;
if (use_new_serial) {
// Serial driver has been initialized by physboot.
gUart = serial;
}
PlatformUartDriverHandoffEarly(serial);
}
void UartDriverHandoffLate(const uart::all::Driver& serial) {
if (use_new_serial) {
// This buffer is needed even when serial is disabled, to prevent uninitialized
// access to it.
rx_queue.Initialize(kRxQueueSize, malloc(kRxQueueSize));
if (!platform_serial_enabled()) {
return;
}
// Check for interrupt support or explicitly polling uart.
ktl::optional<uint32_t> uart_irq;
bool polling_mode = false;
gUart.Visit([&](auto& driver) {
using cfg_type = ktl::decay_t<decltype(driver.uart().config())>;
if constexpr (ktl::is_same_v<cfg_type, zbi_dcfg_simple_pio_t> ||
ktl::is_same_v<cfg_type, zbi_dcfg_simple_t>) {
uart_irq = PlatformUartGetIrqNumber(driver.uart().config().irq);
} else { // Only |uart::null::Driver| is expected to have a different configuration type.
using driver_type = ktl::decay_t<decltype(driver.uart())>;
constexpr auto kIsNullDriver = ktl::is_same_v<driver_type, uart::null::Driver>;
ZX_ASSERT_MSG(kIsNullDriver, "Unexpected UART Configuration.");
// No IRQ Handler for null driver.
return;
}
// Check for polling mode.
if (!uart_irq || gBootOptions->debug_uart_poll) {
// Start the polling without performing any drain.
UartPoll</*DrainUart=*/false>(&gUartPollTimer, current_time(), nullptr);
printf("UART: POLLING mode enabled.\n");
polling_mode = true;
return;
}
static constexpr auto rx_irq_handler = [](auto& spinlock, auto&& read_char,
auto&& mask_rx_interrupt) {
// This check needs to be performed under a lock, such that we prevent operation
// interleaving that would leave us in a blocked state.
//
// E.g.
// Assume a simple MT scenario with one reader R and one writer R:
//
// * W: Observes the buffer is full.
// * R: Reads a character. The buffer is now empty.
// * R: Unmasks RX.
// * W: Masks RX.
//
// At this point, we have an empty buffer and RX interrupts are masked -
// we're stuck! Thus, to avoid this, we acquire the spinlock before
// checking if the buffer is full, and release after (conditionally)
// masking RX interrupts. By pairing this with the acquisition of the
// same lock around unmasking RX interrupts, we prevent the writer above
// from being interrupted by a read-and-unmask.
{
Guard<MonitoredSpinLock, NoIrqSave> lock(&spinlock, SOURCE_TAG);
if (rx_queue.Full()) {
// disables RX interrupts.
mask_rx_interrupt();
return;
}
}
rx_queue.WriteChar(static_cast<char>(read_char()));
};
static constexpr auto tx_irq_handler = [](auto& spinlock, auto& waiter,
auto&& mask_tx_interrupts) {
// Mask the TX interrupt before signalling any blocked thread as there may
// be a race between masking TX here below and unmasking by the blocked
// thread.
{
Guard<MonitoredSpinLock, NoIrqSave> lock(&spinlock, SOURCE_TAG);
mask_tx_interrupts();
}
// Do not signal the event while holding the sync capability, this could lead
// to invalid lock dependencies.
waiter.Wake();
};
constexpr auto irq_handler = [](void* driver_ptr) {
auto* typed_driver = static_cast<ktl::decay_t<decltype(driver)>*>(driver_ptr);
typed_driver->Interrupt(tx_irq_handler, rx_irq_handler);
};
// Register IRQ Handler.
zx_status_t irq_register_result =
register_permanent_int_handler(*uart_irq, irq_handler, &driver);
DEBUG_ASSERT(irq_register_result == ZX_OK);
// Init Rx Interrupt.
driver.InitInterrupt([uart_irq]() { unmask_interrupt(*uart_irq); });
});
if (!polling_mode) {
printf("UART: IRQ driven RX: enabled\n");
is_tx_irq_enabled = !dlog_bypass();
printf("UART: IRQ driven TX: %s\n", is_tx_irq_enabled ? "enabled" : "disabled");
}
}
PlatformUartDriverHandoffLate(serial);
}
void platform_dputs_thread(const char* str, size_t len) {
if (!platform_serial_enabled()) {
return;
}
if (!use_new_serial) {
legacy_platform_dputs_thread(str, len);
return;
}
gUart.Visit([str, len](auto& driver) { driver.template Write<IrqSave>({str, len}); });
}
void platform_dputs_irq(const char* str, size_t len) {
if (!platform_serial_enabled()) {
return;
}
if (!use_new_serial) {
legacy_platform_dputs_irq(str, len);
return;
}
gUart.Visit([str, len](auto& driver) { driver.template Write<NoIrqSave>({str, len}); });
}
int platform_dgetc(char* c, bool wait) {
if (!platform_serial_enabled()) {
return ZX_ERR_NOT_SUPPORTED;
}
if (!use_new_serial) {
return legacy_platform_dgetc(c, wait);
}
auto read = rx_queue.ReadCharWithContext(wait);
// 1 => Character read.
if (read.is_ok()) {
// This is safe because:
// * The RX IRQ handler is holding the UART lock while the queue is being inspected (Full) and
// the RX IRQ is being disabled.
// * The Read path, which is the only path which can transition the queue from full to not
// full, is not holding the the UART lock while inspecting, but the operations is deferred
// and acquires the lock before enabling interrupts.
//
// As a consequence, the IRQ RX Interrupt cannot be enabled by this path, until the RX IRQ
// Handler, has disabled it and released the lock. This means there is no possible interleaving,
// where both paths observe full queue, and we enable the RX IRQ followed by the IRQ RX Handler
// disabling them.
if (read->transitioned_from_full) {
gUart.Visit([](auto& driver) { driver.template EnableRxInterrupt<IrqSave>(); });
}
*c = read->c;
return 1;
}
// 0 => No character yet.
if (read.status_value() == ZX_ERR_SHOULD_WAIT) {
return 0;
}
// < 0 => Error.
return read.status_value();
}
int platform_pgetc(char* c) {
if (!platform_serial_enabled()) {
return ZX_ERR_NOT_SUPPORTED;
}
if (!use_new_serial) {
return legacy_platform_pgetc(c);
}
ktl::optional<char> read;
gUart.Visit([&](auto& driver) { read = driver.Read(); });
if (read) {
*c = *read;
return 0;
}
return -1;
}
void platform_pputc(char c) {
if (!platform_serial_enabled()) {
return;
}
if (!use_new_serial) {
legacy_platform_pputc(c);
return;
}
gUart.Visit([c](auto& driver) { driver.Write({&c, 1}); });
}