aboutsummaryrefslogtreecommitdiff
path: root/sys/dev/acpi/uacpi/event.c
diff options
context:
space:
mode:
Diffstat (limited to 'sys/dev/acpi/uacpi/event.c')
-rw-r--r--sys/dev/acpi/uacpi/event.c2449
1 files changed, 2449 insertions, 0 deletions
diff --git a/sys/dev/acpi/uacpi/event.c b/sys/dev/acpi/uacpi/event.c
new file mode 100644
index 0000000..0c58372
--- /dev/null
+++ b/sys/dev/acpi/uacpi/event.c
@@ -0,0 +1,2449 @@
+#include <uacpi/internal/event.h>
+#include <uacpi/internal/registers.h>
+#include <uacpi/internal/context.h>
+#include <uacpi/internal/io.h>
+#include <uacpi/internal/log.h>
+#include <uacpi/internal/namespace.h>
+#include <uacpi/internal/interpreter.h>
+#include <uacpi/internal/notify.h>
+#include <uacpi/internal/utilities.h>
+#include <uacpi/internal/mutex.h>
+#include <uacpi/internal/stdlib.h>
+#include <uacpi/acpi.h>
+
+#define UACPI_EVENT_DISABLED 0
+#define UACPI_EVENT_ENABLED 1
+
+#if !defined(UACPI_REDUCED_HARDWARE) && !defined(UACPI_BAREBONES_MODE)
+
+static uacpi_handle g_gpe_state_slock;
+static struct uacpi_recursive_lock g_event_lock;
+static uacpi_bool g_gpes_finalized;
+
+struct fixed_event {
+ uacpi_u8 enable_field;
+ uacpi_u8 status_field;
+ uacpi_u16 enable_mask;
+ uacpi_u16 status_mask;
+};
+
+struct fixed_event_handler {
+ uacpi_interrupt_handler handler;
+ uacpi_handle ctx;
+};
+
+static const struct fixed_event fixed_events[UACPI_FIXED_EVENT_MAX + 1] = {
+ [UACPI_FIXED_EVENT_GLOBAL_LOCK] = {
+ .status_field = UACPI_REGISTER_FIELD_GBL_STS,
+ .enable_field = UACPI_REGISTER_FIELD_GBL_EN,
+ .enable_mask = ACPI_PM1_EN_GBL_EN_MASK,
+ .status_mask = ACPI_PM1_STS_GBL_STS_MASK,
+ },
+ [UACPI_FIXED_EVENT_TIMER_STATUS] = {
+ .status_field = UACPI_REGISTER_FIELD_TMR_STS,
+ .enable_field = UACPI_REGISTER_FIELD_TMR_EN,
+ .enable_mask = ACPI_PM1_EN_TMR_EN_MASK,
+ .status_mask = ACPI_PM1_STS_TMR_STS_MASK,
+ },
+ [UACPI_FIXED_EVENT_POWER_BUTTON] = {
+ .status_field = UACPI_REGISTER_FIELD_PWRBTN_STS,
+ .enable_field = UACPI_REGISTER_FIELD_PWRBTN_EN,
+ .enable_mask = ACPI_PM1_EN_PWRBTN_EN_MASK,
+ .status_mask = ACPI_PM1_STS_PWRBTN_STS_MASK,
+ },
+ [UACPI_FIXED_EVENT_SLEEP_BUTTON] = {
+ .status_field = UACPI_REGISTER_FIELD_SLPBTN_STS,
+ .enable_field = UACPI_REGISTER_FIELD_SLPBTN_EN,
+ .enable_mask = ACPI_PM1_EN_SLPBTN_EN_MASK,
+ .status_mask = ACPI_PM1_STS_SLPBTN_STS_MASK,
+ },
+ [UACPI_FIXED_EVENT_RTC] = {
+ .status_field = UACPI_REGISTER_FIELD_RTC_STS,
+ .enable_field = UACPI_REGISTER_FIELD_RTC_EN,
+ .enable_mask = ACPI_PM1_EN_RTC_EN_MASK,
+ .status_mask = ACPI_PM1_STS_RTC_STS_MASK,
+ },
+};
+
+static struct fixed_event_handler
+fixed_event_handlers[UACPI_FIXED_EVENT_MAX + 1];
+
+static uacpi_status initialize_fixed_events(void)
+{
+ uacpi_size i;
+
+ for (i = 0; i < UACPI_FIXED_EVENT_MAX; ++i) {
+ uacpi_write_register_field(
+ fixed_events[i].enable_field, UACPI_EVENT_DISABLED
+ );
+ }
+
+ return UACPI_STATUS_OK;
+}
+
+static uacpi_status set_event(uacpi_u8 event, uacpi_u8 value)
+{
+ uacpi_status ret;
+ uacpi_u64 raw_value;
+ const struct fixed_event *ev = &fixed_events[event];
+
+ ret = uacpi_write_register_field(ev->enable_field, value);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+
+ ret = uacpi_read_register_field(ev->enable_field, &raw_value);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+
+ if (raw_value != value) {
+ uacpi_error("failed to %sable fixed event %d\n",
+ value ? "en" : "dis", event);
+ return UACPI_STATUS_HARDWARE_TIMEOUT;
+ }
+
+ uacpi_trace("fixed event %d %sabled successfully\n",
+ event, value ? "en" : "dis");
+ return UACPI_STATUS_OK;
+}
+
+uacpi_status uacpi_enable_fixed_event(uacpi_fixed_event event)
+{
+ uacpi_status ret;
+
+ UACPI_ENSURE_INIT_LEVEL_AT_LEAST(UACPI_INIT_LEVEL_SUBSYSTEM_INITIALIZED);
+
+ if (uacpi_unlikely(event < 0 || event > UACPI_FIXED_EVENT_MAX))
+ return UACPI_STATUS_INVALID_ARGUMENT;
+ if (uacpi_is_hardware_reduced())
+ return UACPI_STATUS_OK;
+
+ ret = uacpi_recursive_lock_acquire(&g_event_lock);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+
+ /*
+ * Attempting to enable an event that doesn't have a handler is most likely
+ * an error, don't allow it.
+ */
+ if (uacpi_unlikely(fixed_event_handlers[event].handler == UACPI_NULL)) {
+ ret = UACPI_STATUS_NO_HANDLER;
+ goto out;
+ }
+
+ ret = set_event(event, UACPI_EVENT_ENABLED);
+
+out:
+ uacpi_recursive_lock_release(&g_event_lock);
+ return ret;
+}
+
+uacpi_status uacpi_disable_fixed_event(uacpi_fixed_event event)
+{
+ uacpi_status ret;
+
+ UACPI_ENSURE_INIT_LEVEL_AT_LEAST(UACPI_INIT_LEVEL_SUBSYSTEM_INITIALIZED);
+
+ if (uacpi_unlikely(event < 0 || event > UACPI_FIXED_EVENT_MAX))
+ return UACPI_STATUS_INVALID_ARGUMENT;
+ if (uacpi_is_hardware_reduced())
+ return UACPI_STATUS_OK;
+
+ ret = uacpi_recursive_lock_acquire(&g_event_lock);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+
+ ret = set_event(event, UACPI_EVENT_DISABLED);
+
+ uacpi_recursive_lock_release(&g_event_lock);
+ return ret;
+}
+
+uacpi_status uacpi_clear_fixed_event(uacpi_fixed_event event)
+{
+ UACPI_ENSURE_INIT_LEVEL_AT_LEAST(UACPI_INIT_LEVEL_SUBSYSTEM_INITIALIZED);
+
+ if (uacpi_unlikely(event < 0 || event > UACPI_FIXED_EVENT_MAX))
+ return UACPI_STATUS_INVALID_ARGUMENT;
+ if (uacpi_is_hardware_reduced())
+ return UACPI_STATUS_OK;
+
+ return uacpi_write_register_field(
+ fixed_events[event].status_field, ACPI_PM1_STS_CLEAR
+ );
+}
+
+static uacpi_interrupt_ret dispatch_fixed_event(
+ const struct fixed_event *ev, uacpi_fixed_event event
+)
+{
+ uacpi_status ret;
+ struct fixed_event_handler *evh = &fixed_event_handlers[event];
+
+ ret = uacpi_write_register_field(ev->status_field, ACPI_PM1_STS_CLEAR);
+ if (uacpi_unlikely_error(ret))
+ return UACPI_INTERRUPT_NOT_HANDLED;
+
+ if (uacpi_unlikely(evh->handler == UACPI_NULL)) {
+ uacpi_warn(
+ "fixed event %d fired but no handler installed, disabling...\n",
+ event
+ );
+ uacpi_write_register_field(ev->enable_field, UACPI_EVENT_DISABLED);
+ return UACPI_INTERRUPT_NOT_HANDLED;
+ }
+
+ return evh->handler(evh->ctx);
+}
+
+static uacpi_interrupt_ret handle_fixed_events(void)
+{
+ uacpi_interrupt_ret int_ret = UACPI_INTERRUPT_NOT_HANDLED;
+ uacpi_status ret;
+ uacpi_u64 enable_mask, status_mask;
+ uacpi_size i;
+
+ ret = uacpi_read_register(UACPI_REGISTER_PM1_STS, &status_mask);
+ if (uacpi_unlikely_error(ret))
+ return int_ret;
+
+ ret = uacpi_read_register(UACPI_REGISTER_PM1_EN, &enable_mask);
+ if (uacpi_unlikely_error(ret))
+ return int_ret;
+
+ for (i = 0; i < UACPI_FIXED_EVENT_MAX; ++i)
+ {
+ const struct fixed_event *ev = &fixed_events[i];
+
+ if (!(status_mask & ev->status_mask) ||
+ !(enable_mask & ev->enable_mask))
+ continue;
+
+ int_ret |= dispatch_fixed_event(ev, i);
+ }
+
+ return int_ret;
+}
+
+struct gpe_native_handler {
+ uacpi_gpe_handler cb;
+ uacpi_handle ctx;
+
+ /*
+ * Preserved values to be used for state restoration if this handler is
+ * removed at any point.
+ */
+ uacpi_handle previous_handler;
+ uacpi_u8 previous_triggering : 1;
+ uacpi_u8 previous_handler_type : 3;
+ uacpi_u8 previously_enabled : 1;
+};
+
+struct gpe_implicit_notify_handler {
+ struct gpe_implicit_notify_handler *next;
+ uacpi_namespace_node *device;
+};
+
+#define EVENTS_PER_GPE_REGISTER 8
+
+/*
+ * NOTE:
+ * This API and handler types are inspired by ACPICA, let's not reinvent the
+ * wheel and follow a similar path that people ended up finding useful after
+ * years of dealing with ACPI. Obviously credit goes to them for inventing
+ * "implicit notify" and other neat API.
+ */
+enum gpe_handler_type {
+ GPE_HANDLER_TYPE_NONE = 0,
+ GPE_HANDLER_TYPE_AML_HANDLER = 1,
+ GPE_HANDLER_TYPE_NATIVE_HANDLER = 2,
+ GPE_HANDLER_TYPE_NATIVE_HANDLER_RAW = 3,
+ GPE_HANDLER_TYPE_IMPLICIT_NOTIFY = 4,
+};
+
+struct gp_event {
+ union {
+ struct gpe_native_handler *native_handler;
+ struct gpe_implicit_notify_handler *implicit_handler;
+ uacpi_namespace_node *aml_handler;
+ uacpi_handle *any_handler;
+ };
+
+ struct gpe_register *reg;
+ uacpi_u16 idx;
+
+ // "reference count" of the number of times this event has been enabled
+ uacpi_u8 num_users;
+
+ uacpi_u8 handler_type : 3;
+ uacpi_u8 triggering : 1;
+ uacpi_u8 wake : 1;
+ uacpi_u8 block_interrupts : 1;
+};
+
+struct gpe_register {
+ uacpi_mapped_gas status;
+ uacpi_mapped_gas enable;
+
+ uacpi_u8 runtime_mask;
+ uacpi_u8 wake_mask;
+ uacpi_u8 masked_mask;
+ uacpi_u8 current_mask;
+
+ uacpi_u16 base_idx;
+};
+
+struct gpe_block {
+ struct gpe_block *prev, *next;
+
+ /*
+ * Technically this can only refer to \_GPE, but there's also apparently a
+ * "GPE Block Device" with id "ACPI0006", which is not used by anyone. We
+ * still keep it as a possibility that someone might eventually use it, so
+ * it is supported here.
+ */
+ uacpi_namespace_node *device_node;
+
+ struct gpe_register *registers;
+ struct gp_event *events;
+ struct gpe_interrupt_ctx *irq_ctx;
+
+ uacpi_u16 num_registers;
+ uacpi_u16 num_events;
+ uacpi_u16 base_idx;
+};
+
+struct gpe_interrupt_ctx {
+ struct gpe_interrupt_ctx *prev, *next;
+
+ struct gpe_block *gpe_head;
+ uacpi_handle irq_handle;
+ uacpi_u32 irq;
+};
+static struct gpe_interrupt_ctx *g_gpe_interrupt_head;
+
+static uacpi_u8 gpe_get_mask(struct gp_event *event)
+{
+ return 1 << (event->idx - event->reg->base_idx);
+}
+
+enum gpe_state {
+ GPE_STATE_ENABLED,
+ GPE_STATE_ENABLED_CONDITIONALLY,
+ GPE_STATE_DISABLED,
+};
+
+static uacpi_status set_gpe_state(struct gp_event *event, enum gpe_state state)
+{
+ uacpi_status ret;
+ struct gpe_register *reg = event->reg;
+ uacpi_u64 enable_mask;
+ uacpi_u8 event_bit;
+ uacpi_cpu_flags flags;
+
+ event_bit = gpe_get_mask(event);
+ if (state != GPE_STATE_DISABLED && (reg->masked_mask & event_bit))
+ return UACPI_STATUS_OK;
+
+ if (state == GPE_STATE_ENABLED_CONDITIONALLY) {
+ if (!(reg->current_mask & event_bit))
+ return UACPI_STATUS_OK;
+
+ state = GPE_STATE_ENABLED;
+ }
+
+ flags = uacpi_kernel_lock_spinlock(g_gpe_state_slock);
+
+ ret = uacpi_gas_read_mapped(&reg->enable, &enable_mask);
+ if (uacpi_unlikely_error(ret))
+ goto out;
+
+ switch (state) {
+ case GPE_STATE_ENABLED:
+ enable_mask |= event_bit;
+ break;
+ case GPE_STATE_DISABLED:
+ enable_mask &= ~event_bit;
+ break;
+ default:
+ ret = UACPI_STATUS_INVALID_ARGUMENT;
+ goto out;
+ }
+
+ ret = uacpi_gas_write_mapped(&reg->enable, enable_mask);
+out:
+ uacpi_kernel_unlock_spinlock(g_gpe_state_slock, flags);
+ return ret;
+}
+
+static uacpi_status clear_gpe(struct gp_event *event)
+{
+ struct gpe_register *reg = event->reg;
+
+ return uacpi_gas_write_mapped(&reg->status, gpe_get_mask(event));
+}
+
+static uacpi_status restore_gpe(struct gp_event *event)
+{
+ uacpi_status ret;
+
+ if (event->triggering == UACPI_GPE_TRIGGERING_LEVEL) {
+ ret = clear_gpe(event);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+ }
+
+ ret = set_gpe_state(event, GPE_STATE_ENABLED_CONDITIONALLY);
+ event->block_interrupts = UACPI_FALSE;
+
+ return ret;
+}
+
+static void async_restore_gpe(uacpi_handle opaque)
+{
+ uacpi_status ret;
+ struct gp_event *event = opaque;
+
+ ret = restore_gpe(event);
+ if (uacpi_unlikely_error(ret)) {
+ uacpi_error("unable to restore GPE(%02X): %s\n",
+ event->idx, uacpi_status_to_string(ret));
+ }
+}
+
+static void async_run_gpe_handler(uacpi_handle opaque)
+{
+ uacpi_status ret;
+ struct gp_event *event = opaque;
+
+ ret = uacpi_namespace_write_lock();
+ if (uacpi_unlikely_error(ret))
+ goto out_no_unlock;
+
+ switch (event->handler_type) {
+ case GPE_HANDLER_TYPE_AML_HANDLER: {
+ uacpi_object *method_obj;
+ uacpi_object_name name;
+
+ method_obj = uacpi_namespace_node_get_object_typed(
+ event->aml_handler, UACPI_OBJECT_METHOD_BIT
+ );
+ if (uacpi_unlikely(method_obj == UACPI_NULL)) {
+ uacpi_error("GPE(%02X) AML handler gone\n", event->idx);
+ break;
+ }
+
+ name = uacpi_namespace_node_name(event->aml_handler);
+ uacpi_trace(
+ "executing GPE(%02X) handler %.4s\n",
+ event->idx, name.text
+ );
+
+ ret = uacpi_execute_control_method(
+ event->aml_handler, method_obj->method, UACPI_NULL, UACPI_NULL
+ );
+ if (uacpi_unlikely_error(ret)) {
+ uacpi_error(
+ "error while executing GPE(%02X) handler %.4s: %s\n",
+ event->idx, event->aml_handler->name.text,
+ uacpi_status_to_string(ret)
+ );
+ }
+ break;
+ }
+
+ case GPE_HANDLER_TYPE_IMPLICIT_NOTIFY: {
+ struct gpe_implicit_notify_handler *handler;
+
+ handler = event->implicit_handler;
+ while (handler) {
+ /*
+ * 2 - Device Wake. Used to notify OSPM that the device has signaled
+ * its wake event, and that OSPM needs to notify OSPM native device
+ * driver for the device.
+ */
+ uacpi_notify_all(handler->device, 2);
+ handler = handler->next;
+ }
+ break;
+ }
+
+ default:
+ break;
+ }
+
+ uacpi_namespace_write_unlock();
+
+out_no_unlock:
+ /*
+ * We schedule the work as NOTIFICATION to make sure all other notifications
+ * finish before this GPE is re-enabled.
+ */
+ ret = uacpi_kernel_schedule_work(
+ UACPI_WORK_NOTIFICATION, async_restore_gpe, event
+ );
+ if (uacpi_unlikely_error(ret)) {
+ uacpi_error("unable to schedule GPE(%02X) restore: %s\n",
+ event->idx, uacpi_status_to_string(ret));
+ async_restore_gpe(event);
+ }
+}
+
+static uacpi_interrupt_ret dispatch_gpe(
+ uacpi_namespace_node *device_node, struct gp_event *event
+)
+{
+ uacpi_status ret;
+ uacpi_interrupt_ret int_ret = UACPI_INTERRUPT_NOT_HANDLED;
+
+ /*
+ * For raw handlers we don't do any management whatsoever, we just let the
+ * handler know a GPE has triggered and let it handle disable/enable as
+ * well as clearing.
+ */
+ if (event->handler_type == GPE_HANDLER_TYPE_NATIVE_HANDLER_RAW) {
+ return event->native_handler->cb(
+ event->native_handler->ctx, device_node, event->idx
+ );
+ }
+
+ ret = set_gpe_state(event, GPE_STATE_DISABLED);
+ if (uacpi_unlikely_error(ret)) {
+ uacpi_error("failed to disable GPE(%02X): %s\n",
+ event->idx, uacpi_status_to_string(ret));
+ return int_ret;
+ }
+
+ event->block_interrupts = UACPI_TRUE;
+
+ if (event->triggering == UACPI_GPE_TRIGGERING_EDGE) {
+ ret = clear_gpe(event);
+ if (uacpi_unlikely_error(ret)) {
+ uacpi_error("unable to clear GPE(%02X): %s\n",
+ event->idx, uacpi_status_to_string(ret));
+ set_gpe_state(event, GPE_STATE_ENABLED_CONDITIONALLY);
+ return int_ret;
+ }
+ }
+
+ switch (event->handler_type) {
+ case GPE_HANDLER_TYPE_NATIVE_HANDLER:
+ int_ret = event->native_handler->cb(
+ event->native_handler->ctx, device_node, event->idx
+ );
+ if (!(int_ret & UACPI_GPE_REENABLE))
+ break;
+
+ ret = restore_gpe(event);
+ if (uacpi_unlikely_error(ret)) {
+ uacpi_error("unable to restore GPE(%02X): %s\n",
+ event->idx, uacpi_status_to_string(ret));
+ }
+ break;
+
+ case GPE_HANDLER_TYPE_AML_HANDLER:
+ case GPE_HANDLER_TYPE_IMPLICIT_NOTIFY:
+ ret = uacpi_kernel_schedule_work(
+ UACPI_WORK_GPE_EXECUTION, async_run_gpe_handler, event
+ );
+ if (uacpi_unlikely_error(ret)) {
+ uacpi_warn(
+ "unable to schedule GPE(%02X) for execution: %s\n",
+ event->idx, uacpi_status_to_string(ret)
+ );
+ }
+ break;
+
+ default:
+ uacpi_warn("GPE(%02X) fired but no handler, keeping disabled\n",
+ event->idx);
+ break;
+ }
+
+ return UACPI_INTERRUPT_HANDLED;
+}
+
+static uacpi_interrupt_ret detect_gpes(struct gpe_block *block)
+{
+ uacpi_status ret;
+ uacpi_interrupt_ret int_ret = UACPI_INTERRUPT_NOT_HANDLED;
+ struct gpe_register *reg;
+ struct gp_event *event;
+ uacpi_u64 status, enable;
+ uacpi_size i, j;
+
+ while (block) {
+ for (i = 0; i < block->num_registers; ++i) {
+ reg = &block->registers[i];
+
+ if (!reg->runtime_mask && !reg->wake_mask)
+ continue;
+
+ ret = uacpi_gas_read_mapped(&reg->status, &status);
+ if (uacpi_unlikely_error(ret))
+ return int_ret;
+
+ ret = uacpi_gas_read_mapped(&reg->enable, &enable);
+ if (uacpi_unlikely_error(ret))
+ return int_ret;
+
+ if (status == 0)
+ continue;
+
+ for (j = 0; j < EVENTS_PER_GPE_REGISTER; ++j) {
+ if (!((status & enable) & (1ull << j)))
+ continue;
+
+ event = &block->events[j + i * EVENTS_PER_GPE_REGISTER];
+ int_ret |= dispatch_gpe(block->device_node, event);
+ }
+ }
+
+ block = block->next;
+ }
+
+ return int_ret;
+}
+
+static uacpi_status maybe_dispatch_gpe(
+ uacpi_namespace_node *gpe_device, struct gp_event *event
+)
+{
+ uacpi_status ret;
+ struct gpe_register *reg = event->reg;
+ uacpi_u64 status;
+
+ ret = uacpi_gas_read_mapped(&reg->status, &status);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+
+ if (!(status & gpe_get_mask(event)))
+ return ret;
+
+ dispatch_gpe(gpe_device, event);
+ return ret;
+}
+
+static uacpi_interrupt_ret handle_gpes(uacpi_handle opaque)
+{
+ struct gpe_interrupt_ctx *ctx = opaque;
+
+ if (uacpi_unlikely(ctx == UACPI_NULL))
+ return UACPI_INTERRUPT_NOT_HANDLED;
+
+ return detect_gpes(ctx->gpe_head);
+}
+
+static uacpi_status find_or_create_gpe_interrupt_ctx(
+ uacpi_u32 irq, struct gpe_interrupt_ctx **out_ctx
+)
+{
+ uacpi_status ret;
+ struct gpe_interrupt_ctx *entry = g_gpe_interrupt_head;
+
+ while (entry) {
+ if (entry->irq == irq) {
+ *out_ctx = entry;
+ return UACPI_STATUS_OK;
+ }
+
+ entry = entry->next;
+ }
+
+ entry = uacpi_kernel_alloc_zeroed(sizeof(*entry));
+ if (uacpi_unlikely(entry == UACPI_NULL))
+ return UACPI_STATUS_OUT_OF_MEMORY;
+
+ /*
+ * SCI interrupt is installed by other code and is responsible for more
+ * things than just the GPE handling. Don't install it here.
+ */
+ if (irq != g_uacpi_rt_ctx.fadt.sci_int) {
+ ret = uacpi_kernel_install_interrupt_handler(
+ irq, handle_gpes, entry, &entry->irq_handle
+ );
+ if (uacpi_unlikely_error(ret)) {
+ uacpi_free(entry, sizeof(*entry));
+ return ret;
+ }
+ }
+
+ entry->irq = irq;
+ entry->next = g_gpe_interrupt_head;
+ g_gpe_interrupt_head = entry;
+
+ *out_ctx = entry;
+ return UACPI_STATUS_OK;
+}
+
+static void gpe_release_implicit_notify_handlers(struct gp_event *event)
+{
+ struct gpe_implicit_notify_handler *handler, *next_handler;
+
+ handler = event->implicit_handler;
+ while (handler) {
+ next_handler = handler->next;
+ uacpi_free(handler, sizeof(*handler));
+ handler = next_handler;
+ }
+
+ event->implicit_handler = UACPI_NULL;
+}
+
+enum gpe_block_action
+{
+ GPE_BLOCK_ACTION_DISABLE_ALL,
+ GPE_BLOCK_ACTION_ENABLE_ALL_FOR_RUNTIME,
+ GPE_BLOCK_ACTION_ENABLE_ALL_FOR_WAKE,
+ GPE_BLOCK_ACTION_CLEAR_ALL,
+};
+
+static uacpi_status gpe_block_apply_action(
+ struct gpe_block *block, enum gpe_block_action action
+)
+{
+ uacpi_status ret;
+ uacpi_size i;
+ uacpi_u8 value;
+ struct gpe_register *reg;
+
+ for (i = 0; i < block->num_registers; ++i) {
+ reg = &block->registers[i];
+
+ switch (action) {
+ case GPE_BLOCK_ACTION_DISABLE_ALL:
+ value = 0;
+ break;
+ case GPE_BLOCK_ACTION_ENABLE_ALL_FOR_RUNTIME:
+ value = reg->runtime_mask & ~reg->masked_mask;
+ break;
+ case GPE_BLOCK_ACTION_ENABLE_ALL_FOR_WAKE:
+ value = reg->wake_mask;
+ break;
+ case GPE_BLOCK_ACTION_CLEAR_ALL:
+ ret = uacpi_gas_write_mapped(&reg->status, 0xFF);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+ continue;
+ default:
+ return UACPI_STATUS_INVALID_ARGUMENT;
+ }
+
+ reg->current_mask = value;
+ ret = uacpi_gas_write_mapped(&reg->enable, value);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+ }
+
+ return UACPI_STATUS_OK;
+}
+
+static void gpe_block_mask_safe(struct gpe_block *block)
+{
+ uacpi_size i;
+ struct gpe_register *reg;
+
+ for (i = 0; i < block->num_registers; ++i) {
+ reg = &block->registers[i];
+
+ // No need to flush or do anything if it's not currently enabled
+ if (!reg->current_mask)
+ continue;
+
+ // 1. Mask the GPEs, this makes sure their state is no longer modifyable
+ reg->masked_mask = 0xFF;
+
+ /*
+ * 2. Wait for in-flight work & IRQs to finish, these might already
+ * be past the respective "if (masked)" check and therefore may
+ * try to re-enable a masked GPE.
+ */
+ uacpi_kernel_wait_for_work_completion();
+
+ /*
+ * 3. Now that this GPE's state is unmodifyable and we know that
+ * currently in-flight IRQs will see the masked state, we can
+ * safely disable all events knowing they won't be re-enabled by
+ * a racing IRQ.
+ */
+ uacpi_gas_write_mapped(&reg->enable, 0x00);
+
+ /*
+ * 4. Wait for the last possible IRQ to finish, now that this event is
+ * disabled.
+ */
+ uacpi_kernel_wait_for_work_completion();
+ }
+}
+
+static void uninstall_gpe_block(struct gpe_block *block)
+{
+ if (block->registers != UACPI_NULL) {
+ struct gpe_register *reg;
+ uacpi_size i;
+
+ gpe_block_mask_safe(block);
+
+ for (i = 0; i < block->num_registers; ++i) {
+ reg = &block->registers[i];
+
+ if (reg->enable.total_bit_width)
+ uacpi_unmap_gas_nofree(&reg->enable);
+ if (reg->status.total_bit_width)
+ uacpi_unmap_gas_nofree(&reg->status);
+ }
+ }
+
+ if (block->prev)
+ block->prev->next = block->next;
+
+ if (block->irq_ctx) {
+ struct gpe_interrupt_ctx *ctx = block->irq_ctx;
+
+ // Are we the first GPE block?
+ if (block == ctx->gpe_head) {
+ ctx->gpe_head = ctx->gpe_head->next;
+ } else {
+ struct gpe_block *prev_block = ctx->gpe_head;
+
+ // We're not, do a search
+ while (prev_block) {
+ if (prev_block->next == block) {
+ prev_block->next = block->next;
+ break;
+ }
+
+ prev_block = prev_block->next;
+ }
+ }
+
+ // This GPE block was the last user of this interrupt context, remove it
+ if (ctx->gpe_head == UACPI_NULL) {
+ if (ctx->prev)
+ ctx->prev->next = ctx->next;
+
+ if (ctx->irq != g_uacpi_rt_ctx.fadt.sci_int) {
+ uacpi_kernel_uninstall_interrupt_handler(
+ handle_gpes, ctx->irq_handle
+ );
+ }
+
+ uacpi_free(block->irq_ctx, sizeof(*block->irq_ctx));
+ }
+ }
+
+ if (block->events != UACPI_NULL) {
+ uacpi_size i;
+ struct gp_event *event;
+
+ for (i = 0; i < block->num_events; ++i) {
+ event = &block->events[i];
+
+ switch (event->handler_type) {
+ case GPE_HANDLER_TYPE_NONE:
+ case GPE_HANDLER_TYPE_AML_HANDLER:
+ break;
+
+ case GPE_HANDLER_TYPE_NATIVE_HANDLER:
+ case GPE_HANDLER_TYPE_NATIVE_HANDLER_RAW:
+ uacpi_free(event->native_handler,
+ sizeof(*event->native_handler));
+ break;
+
+ case GPE_HANDLER_TYPE_IMPLICIT_NOTIFY: {
+ gpe_release_implicit_notify_handlers(event);
+ break;
+ }
+
+ default:
+ break;
+ }
+ }
+
+ }
+
+ uacpi_free(block->registers,
+ sizeof(*block->registers) * block->num_registers);
+ uacpi_free(block->events,
+ sizeof(*block->events) * block->num_events);
+ uacpi_free(block, sizeof(*block));
+}
+
+static struct gp_event *gpe_from_block(struct gpe_block *block, uacpi_u16 idx)
+{
+ uacpi_u16 offset;
+
+ if (idx < block->base_idx)
+ return UACPI_NULL;
+
+ offset = idx - block->base_idx;
+ if (offset > block->num_events)
+ return UACPI_NULL;
+
+ return &block->events[offset];
+}
+
+struct gpe_match_ctx {
+ struct gpe_block *block;
+ uacpi_u32 matched_count;
+ uacpi_bool post_dynamic_table_load;
+};
+
+static uacpi_iteration_decision do_match_gpe_methods(
+ uacpi_handle opaque, uacpi_namespace_node *node, uacpi_u32 depth
+)
+{
+ uacpi_status ret;
+ struct gpe_match_ctx *ctx = opaque;
+ struct gp_event *event;
+ uacpi_u8 triggering;
+ uacpi_u64 idx;
+
+ UACPI_UNUSED(depth);
+
+ if (node->name.text[0] != '_')
+ return UACPI_ITERATION_DECISION_CONTINUE;
+
+ switch (node->name.text[1]) {
+ case 'L':
+ triggering = UACPI_GPE_TRIGGERING_LEVEL;
+ break;
+ case 'E':
+ triggering = UACPI_GPE_TRIGGERING_EDGE;
+ break;
+ default:
+ return UACPI_ITERATION_DECISION_CONTINUE;
+ }
+
+ ret = uacpi_string_to_integer(&node->name.text[2], 2, UACPI_BASE_HEX, &idx);
+ if (uacpi_unlikely_error(ret)) {
+ uacpi_trace("invalid GPE method name %.4s, ignored\n", node->name.text);
+ return UACPI_ITERATION_DECISION_CONTINUE;
+ }
+
+ event = gpe_from_block(ctx->block, idx);
+ if (event == UACPI_NULL)
+ return UACPI_ITERATION_DECISION_CONTINUE;
+
+ switch (event->handler_type) {
+ /*
+ * This had implicit notify configured but this is no longer needed as we
+ * now have an actual AML handler. Free the implicit notify list and switch
+ * this handler to AML mode.
+ */
+ case GPE_HANDLER_TYPE_IMPLICIT_NOTIFY:
+ gpe_release_implicit_notify_handlers(event);
+ UACPI_FALLTHROUGH;
+ case GPE_HANDLER_TYPE_NONE:
+ event->aml_handler = node;
+ event->handler_type = GPE_HANDLER_TYPE_AML_HANDLER;
+ break;
+
+ case GPE_HANDLER_TYPE_AML_HANDLER:
+ // This is okay, since we're re-running the detection code
+ if (!ctx->post_dynamic_table_load) {
+ uacpi_warn(
+ "GPE(%02X) already matched %.4s, skipping %.4s\n",
+ (uacpi_u32)idx, event->aml_handler->name.text, node->name.text
+ );
+ }
+ return UACPI_ITERATION_DECISION_CONTINUE;
+
+ case GPE_HANDLER_TYPE_NATIVE_HANDLER:
+ case GPE_HANDLER_TYPE_NATIVE_HANDLER_RAW:
+ uacpi_trace(
+ "not assigning GPE(%02X) to %.4s, override "
+ "installed by user\n", (uacpi_u32)idx, node->name.text
+ );
+ UACPI_FALLTHROUGH;
+ default:
+ return UACPI_ITERATION_DECISION_CONTINUE;
+ }
+
+ uacpi_trace("assigned GPE(%02X) -> %.4s\n",
+ (uacpi_u32)idx, node->name.text);
+ event->triggering = triggering;
+ ctx->matched_count++;
+
+ return UACPI_ITERATION_DECISION_CONTINUE;
+}
+
+void uacpi_events_match_post_dynamic_table_load(void)
+{
+ struct gpe_match_ctx match_ctx = {
+ .post_dynamic_table_load = UACPI_TRUE,
+ };
+ struct gpe_interrupt_ctx *irq_ctx;
+
+ uacpi_namespace_write_unlock();
+
+ if (uacpi_unlikely_error(uacpi_recursive_lock_acquire(&g_event_lock)))
+ goto out;
+
+ irq_ctx = g_gpe_interrupt_head;
+
+ while (irq_ctx) {
+ match_ctx.block = irq_ctx->gpe_head;
+
+ while (match_ctx.block) {
+ uacpi_namespace_do_for_each_child(
+ match_ctx.block->device_node, do_match_gpe_methods, UACPI_NULL,
+ UACPI_OBJECT_METHOD_BIT, UACPI_MAX_DEPTH_ANY,
+ UACPI_SHOULD_LOCK_YES, UACPI_PERMANENT_ONLY_YES, &match_ctx
+ );
+ match_ctx.block = match_ctx.block->next;
+ }
+
+ irq_ctx = irq_ctx->next;
+ }
+
+ if (match_ctx.matched_count) {
+ uacpi_info("matched %u additional GPEs post dynamic table load\n",
+ match_ctx.matched_count);
+ }
+
+out:
+ uacpi_recursive_lock_release(&g_event_lock);
+ uacpi_namespace_write_lock();
+}
+
+static uacpi_status create_gpe_block(
+ uacpi_namespace_node *device_node, uacpi_u32 irq, uacpi_u16 base_idx,
+ uacpi_u64 address, uacpi_u8 address_space_id, uacpi_u16 num_registers
+)
+{
+ uacpi_status ret = UACPI_STATUS_OUT_OF_MEMORY;
+ struct gpe_match_ctx match_ctx = { 0 };
+ struct gpe_block *block;
+ struct gpe_register *reg;
+ struct gp_event *event;
+ struct acpi_gas tmp_gas = { 0 };
+ uacpi_size i, j;
+
+ tmp_gas.address_space_id = address_space_id;
+ tmp_gas.register_bit_width = 8;
+
+ block = uacpi_kernel_alloc_zeroed(sizeof(*block));
+ if (uacpi_unlikely(block == UACPI_NULL))
+ return ret;
+
+ block->device_node = device_node;
+ block->base_idx = base_idx;
+
+ block->num_registers = num_registers;
+ block->registers = uacpi_kernel_alloc_zeroed(
+ num_registers * sizeof(*block->registers)
+ );
+ if (uacpi_unlikely(block->registers == UACPI_NULL))
+ goto error_out;
+
+ block->num_events = num_registers * EVENTS_PER_GPE_REGISTER;
+ block->events = uacpi_kernel_alloc_zeroed(
+ block->num_events * sizeof(*block->events)
+ );
+ if (uacpi_unlikely(block->events == UACPI_NULL))
+ goto error_out;
+
+ for (reg = block->registers, event = block->events, i = 0;
+ i < num_registers; ++i, ++reg) {
+
+ /*
+ * Initialize this register pair as well as all the events within it.
+ *
+ * Each register has two sub registers: status & enable, 8 bits each.
+ * Each bit corresponds to one event that we initialize below.
+ */
+ reg->base_idx = base_idx + (i * EVENTS_PER_GPE_REGISTER);
+
+
+ tmp_gas.address = address + i;
+ ret = uacpi_map_gas_noalloc(&tmp_gas, &reg->status);
+ if (uacpi_unlikely_error(ret))
+ goto error_out;
+
+ tmp_gas.address += num_registers;
+ ret = uacpi_map_gas_noalloc(&tmp_gas, &reg->enable);
+ if (uacpi_unlikely_error(ret))
+ goto error_out;
+
+ for (j = 0; j < EVENTS_PER_GPE_REGISTER; ++j, ++event) {
+ event->idx = reg->base_idx + j;
+ event->reg = reg;
+ }
+
+ /*
+ * Disable all GPEs in this register & clear anything that might be
+ * pending from earlier.
+ */
+ ret = uacpi_gas_write_mapped(&reg->enable, 0x00);
+ if (uacpi_unlikely_error(ret))
+ goto error_out;
+
+ ret = uacpi_gas_write_mapped(&reg->status, 0xFF);
+ if (uacpi_unlikely_error(ret))
+ goto error_out;
+ }
+
+ ret = find_or_create_gpe_interrupt_ctx(irq, &block->irq_ctx);
+ if (uacpi_unlikely_error(ret))
+ goto error_out;
+
+ block->next = block->irq_ctx->gpe_head;
+ block->irq_ctx->gpe_head = block;
+ match_ctx.block = block;
+
+ uacpi_namespace_do_for_each_child(
+ device_node, do_match_gpe_methods, UACPI_NULL,
+ UACPI_OBJECT_METHOD_BIT, UACPI_MAX_DEPTH_ANY,
+ UACPI_SHOULD_LOCK_YES, UACPI_PERMANENT_ONLY_YES, &match_ctx
+ );
+
+ uacpi_trace("initialized GPE block %.4s[%d->%d], %d AML handlers (IRQ %d)\n",
+ device_node->name.text, base_idx, base_idx + block->num_events,
+ match_ctx.matched_count, irq);
+ return UACPI_STATUS_OK;
+
+error_out:
+ uninstall_gpe_block(block);
+ return ret;
+}
+
+typedef uacpi_iteration_decision (*gpe_block_iteration_callback)
+ (struct gpe_block*, uacpi_handle);
+
+static void for_each_gpe_block(
+ gpe_block_iteration_callback cb, uacpi_handle handle
+)
+{
+ uacpi_iteration_decision decision;
+ struct gpe_interrupt_ctx *irq_ctx = g_gpe_interrupt_head;
+ struct gpe_block *block;
+
+ while (irq_ctx) {
+ block = irq_ctx->gpe_head;
+
+ while (block) {
+ decision = cb(block, handle);
+ if (decision == UACPI_ITERATION_DECISION_BREAK)
+ return;
+
+ block = block->next;
+ }
+
+ irq_ctx = irq_ctx->next;
+ }
+}
+
+struct gpe_search_ctx {
+ uacpi_namespace_node *gpe_device;
+ uacpi_u16 idx;
+ struct gpe_block *out_block;
+ struct gp_event *out_event;
+};
+
+static uacpi_iteration_decision do_find_gpe(
+ struct gpe_block *block, uacpi_handle opaque
+)
+{
+ struct gpe_search_ctx *ctx = opaque;
+
+ if (block->device_node != ctx->gpe_device)
+ return UACPI_ITERATION_DECISION_CONTINUE;
+
+ ctx->out_block = block;
+ ctx->out_event = gpe_from_block(block, ctx->idx);
+ if (ctx->out_event == UACPI_NULL)
+ return UACPI_ITERATION_DECISION_CONTINUE;
+
+ return UACPI_ITERATION_DECISION_BREAK;
+}
+
+static struct gp_event *get_gpe(
+ uacpi_namespace_node *gpe_device, uacpi_u16 idx
+)
+{
+ struct gpe_search_ctx ctx = { 0 };
+
+ ctx.gpe_device = gpe_device;
+ ctx.idx = idx;
+
+ for_each_gpe_block(do_find_gpe, &ctx);
+ return ctx.out_event;
+}
+
+static void gp_event_toggle_masks(struct gp_event *event, uacpi_bool set_on)
+{
+ uacpi_u8 this_mask;
+ struct gpe_register *reg = event->reg;
+
+ this_mask = gpe_get_mask(event);
+
+ if (set_on) {
+ reg->runtime_mask |= this_mask;
+ reg->current_mask = reg->runtime_mask;
+ return;
+ }
+
+ reg->runtime_mask &= ~this_mask;
+ reg->current_mask = reg->runtime_mask;
+}
+
+static uacpi_status gpe_remove_user(struct gp_event *event)
+{
+ uacpi_status ret = UACPI_STATUS_OK;
+
+ if (uacpi_unlikely(event->num_users == 0))
+ return UACPI_STATUS_INVALID_ARGUMENT;
+
+ if (--event->num_users == 0) {
+ gp_event_toggle_masks(event, UACPI_FALSE);
+
+ ret = set_gpe_state(event, GPE_STATE_DISABLED);
+ if (uacpi_unlikely_error(ret)) {
+ gp_event_toggle_masks(event, UACPI_TRUE);
+ event->num_users++;
+ }
+ }
+
+ return ret;
+}
+
+enum event_clear_if_first {
+ EVENT_CLEAR_IF_FIRST_YES,
+ EVENT_CLEAR_IF_FIRST_NO,
+};
+
+static uacpi_status gpe_add_user(
+ struct gp_event *event, enum event_clear_if_first clear_if_first
+)
+{
+ uacpi_status ret = UACPI_STATUS_OK;
+
+ if (uacpi_unlikely(event->num_users == 0xFF))
+ return UACPI_STATUS_INVALID_ARGUMENT;
+
+ if (++event->num_users == 1) {
+ if (clear_if_first == EVENT_CLEAR_IF_FIRST_YES)
+ clear_gpe(event);
+
+ gp_event_toggle_masks(event, UACPI_TRUE);
+
+ ret = set_gpe_state(event, GPE_STATE_ENABLED);
+ if (uacpi_unlikely_error(ret)) {
+ gp_event_toggle_masks(event, UACPI_FALSE);
+ event->num_users--;
+ }
+ }
+
+ return ret;
+}
+
+const uacpi_char *uacpi_gpe_triggering_to_string(
+ uacpi_gpe_triggering triggering
+)
+{
+ switch (triggering) {
+ case UACPI_GPE_TRIGGERING_EDGE:
+ return "edge";
+ case UACPI_GPE_TRIGGERING_LEVEL:
+ return "level";
+ default:
+ return "invalid";
+ }
+}
+
+static uacpi_bool gpe_needs_polling(struct gp_event *event)
+{
+ return event->num_users && event->triggering == UACPI_GPE_TRIGGERING_EDGE;
+}
+
+static uacpi_status gpe_mask_unmask(
+ struct gp_event *event, uacpi_bool should_mask
+)
+{
+ struct gpe_register *reg;
+ uacpi_u8 mask;
+
+ reg = event->reg;
+ mask = gpe_get_mask(event);
+
+ if (should_mask) {
+ if (reg->masked_mask & mask)
+ return UACPI_STATUS_INVALID_ARGUMENT;
+
+ // 1. Mask the GPE, this makes sure its state is no longer modifyable
+ reg->masked_mask |= mask;
+
+ /*
+ * 2. Wait for in-flight work & IRQs to finish, these might already
+ * be past the respective "if (masked)" check and therefore may
+ * try to re-enable a masked GPE.
+ */
+ uacpi_kernel_wait_for_work_completion();
+
+ /*
+ * 3. Now that this GPE's state is unmodifyable and we know that currently
+ * in-flight IRQs will see the masked state, we can safely disable this
+ * event knowing it won't be re-enabled by a racing IRQ.
+ */
+ set_gpe_state(event, GPE_STATE_DISABLED);
+
+ /*
+ * 4. Wait for the last possible IRQ to finish, now that this event is
+ * disabled.
+ */
+ uacpi_kernel_wait_for_work_completion();
+
+ return UACPI_STATUS_OK;
+ }
+
+ if (!(reg->masked_mask & mask))
+ return UACPI_STATUS_INVALID_ARGUMENT;
+
+ reg->masked_mask &= ~mask;
+ if (!event->block_interrupts && event->num_users)
+ set_gpe_state(event, GPE_STATE_ENABLED_CONDITIONALLY);
+
+ return UACPI_STATUS_OK;
+}
+
+/*
+ * Safely mask the event before we modify its handlers.
+ *
+ * This makes sure we can't get an IRQ in the middle of modifying this
+ * event's structures.
+ */
+static uacpi_bool gpe_mask_safe(struct gp_event *event)
+{
+ // No need to flush or do anything if it's not currently enabled
+ if (!(event->reg->current_mask & gpe_get_mask(event)))
+ return UACPI_FALSE;
+
+ gpe_mask_unmask(event, UACPI_TRUE);
+ return UACPI_TRUE;
+}
+
+static uacpi_iteration_decision do_initialize_gpe_block(
+ struct gpe_block *block, uacpi_handle opaque
+)
+{
+ uacpi_status ret;
+ uacpi_bool *poll_blocks = opaque;
+ uacpi_size i, j, count_enabled = 0;
+ struct gp_event *event;
+
+ for (i = 0; i < block->num_registers; ++i) {
+ for (j = 0; j < EVENTS_PER_GPE_REGISTER; ++j) {
+ event = &block->events[j + i * EVENTS_PER_GPE_REGISTER];
+
+ if (event->wake ||
+ event->handler_type != GPE_HANDLER_TYPE_AML_HANDLER)
+ continue;
+
+ ret = gpe_add_user(event, EVENT_CLEAR_IF_FIRST_NO);
+ if (uacpi_unlikely_error(ret)) {
+ uacpi_warn("failed to enable GPE(%02X): %s\n",
+ event->idx, uacpi_status_to_string(ret));
+ continue;
+ }
+
+ *poll_blocks |= gpe_needs_polling(event);
+ count_enabled++;
+ }
+ }
+
+ if (count_enabled) {
+ uacpi_info(
+ "enabled %zu GPEs in block %.4s@[%d->%d]\n",
+ count_enabled, block->device_node->name.text,
+ block->base_idx, block->base_idx + block->num_events
+ );
+ }
+ return UACPI_ITERATION_DECISION_CONTINUE;
+}
+
+uacpi_status uacpi_finalize_gpe_initialization(void)
+{
+ uacpi_status ret;
+ uacpi_bool poll_blocks = UACPI_FALSE;
+
+ UACPI_ENSURE_INIT_LEVEL_AT_LEAST(UACPI_INIT_LEVEL_NAMESPACE_LOADED);
+
+ ret = uacpi_recursive_lock_acquire(&g_event_lock);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+
+ if (g_gpes_finalized)
+ goto out;
+
+ g_gpes_finalized = UACPI_TRUE;
+
+ for_each_gpe_block(do_initialize_gpe_block, &poll_blocks);
+ if (poll_blocks)
+ detect_gpes(g_gpe_interrupt_head->gpe_head);
+
+out:
+ uacpi_recursive_lock_release(&g_event_lock);
+ return ret;
+}
+
+static uacpi_status sanitize_device_and_find_gpe(
+ uacpi_namespace_node **gpe_device, uacpi_u16 idx,
+ struct gp_event **out_event
+)
+{
+ if (*gpe_device == UACPI_NULL) {
+ *gpe_device = uacpi_namespace_get_predefined(
+ UACPI_PREDEFINED_NAMESPACE_GPE
+ );
+ }
+
+ *out_event = get_gpe(*gpe_device, idx);
+ if (*out_event == UACPI_NULL)
+ return UACPI_STATUS_NOT_FOUND;
+
+ return UACPI_STATUS_OK;
+}
+
+static uacpi_status do_install_gpe_handler(
+ uacpi_namespace_node *gpe_device, uacpi_u16 idx,
+ uacpi_gpe_triggering triggering, enum gpe_handler_type type,
+ uacpi_gpe_handler handler, uacpi_handle ctx
+)
+{
+ uacpi_status ret;
+ struct gp_event *event;
+ struct gpe_native_handler *native_handler;
+ uacpi_bool did_mask;
+
+ UACPI_ENSURE_INIT_LEVEL_AT_LEAST(UACPI_INIT_LEVEL_NAMESPACE_LOADED);
+
+ if (uacpi_unlikely(triggering > UACPI_GPE_TRIGGERING_MAX))
+ return UACPI_STATUS_INVALID_ARGUMENT;
+
+ ret = uacpi_recursive_lock_acquire(&g_event_lock);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+
+ ret = sanitize_device_and_find_gpe(&gpe_device, idx, &event);
+ if (uacpi_unlikely_error(ret))
+ goto out;
+
+ if (event->handler_type == GPE_HANDLER_TYPE_NATIVE_HANDLER ||
+ event->handler_type == GPE_HANDLER_TYPE_NATIVE_HANDLER_RAW) {
+ ret = UACPI_STATUS_ALREADY_EXISTS;
+ goto out;
+ }
+
+ native_handler = uacpi_kernel_alloc(sizeof(*native_handler));
+ if (uacpi_unlikely(native_handler == UACPI_NULL)) {
+ ret = UACPI_STATUS_OUT_OF_MEMORY;
+ goto out;
+ }
+
+ native_handler->cb = handler;
+ native_handler->ctx = ctx;
+ native_handler->previous_handler = event->any_handler;
+ native_handler->previous_handler_type = event->handler_type;
+ native_handler->previous_triggering = event->triggering;
+ native_handler->previously_enabled = UACPI_FALSE;
+
+ did_mask = gpe_mask_safe(event);
+
+ if ((event->handler_type == GPE_HANDLER_TYPE_AML_HANDLER ||
+ event->handler_type == GPE_HANDLER_TYPE_IMPLICIT_NOTIFY) &&
+ event->num_users != 0) {
+ native_handler->previously_enabled = UACPI_TRUE;
+ gpe_remove_user(event);
+
+ if (uacpi_unlikely(event->triggering != triggering)) {
+ uacpi_warn(
+ "GPE(%02X) user handler claims %s triggering, originally "
+ "configured as %s\n", idx,
+ uacpi_gpe_triggering_to_string(triggering),
+ uacpi_gpe_triggering_to_string(event->triggering)
+ );
+ }
+ }
+
+ event->native_handler = native_handler;
+ event->handler_type = type;
+ event->triggering = triggering;
+
+ if (did_mask)
+ gpe_mask_unmask(event, UACPI_FALSE);
+out:
+ uacpi_recursive_lock_release(&g_event_lock);
+ return ret;
+}
+
+uacpi_status uacpi_install_gpe_handler(
+ uacpi_namespace_node *gpe_device, uacpi_u16 idx,
+ uacpi_gpe_triggering triggering, uacpi_gpe_handler handler,
+ uacpi_handle ctx
+)
+{
+ return do_install_gpe_handler(
+ gpe_device, idx, triggering, GPE_HANDLER_TYPE_NATIVE_HANDLER,
+ handler, ctx
+ );
+}
+
+uacpi_status uacpi_install_gpe_handler_raw(
+ uacpi_namespace_node *gpe_device, uacpi_u16 idx,
+ uacpi_gpe_triggering triggering, uacpi_gpe_handler handler,
+ uacpi_handle ctx
+)
+{
+ return do_install_gpe_handler(
+ gpe_device, idx, triggering, GPE_HANDLER_TYPE_NATIVE_HANDLER_RAW,
+ handler, ctx
+ );
+}
+
+uacpi_status uacpi_uninstall_gpe_handler(
+ uacpi_namespace_node *gpe_device, uacpi_u16 idx,
+ uacpi_gpe_handler handler
+)
+{
+ uacpi_status ret;
+ struct gp_event *event;
+ struct gpe_native_handler *native_handler;
+ uacpi_bool did_mask;
+
+ UACPI_ENSURE_INIT_LEVEL_AT_LEAST(UACPI_INIT_LEVEL_NAMESPACE_LOADED);
+
+ ret = uacpi_recursive_lock_acquire(&g_event_lock);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+
+ ret = sanitize_device_and_find_gpe(&gpe_device, idx, &event);
+ if (uacpi_unlikely_error(ret))
+ goto out;
+
+ if (event->handler_type != GPE_HANDLER_TYPE_NATIVE_HANDLER &&
+ event->handler_type != GPE_HANDLER_TYPE_NATIVE_HANDLER_RAW) {
+ ret = UACPI_STATUS_NOT_FOUND;
+ goto out;
+ }
+
+ native_handler = event->native_handler;
+ if (uacpi_unlikely(native_handler->cb != handler)) {
+ ret = UACPI_STATUS_INVALID_ARGUMENT;
+ goto out;
+ }
+
+ did_mask = gpe_mask_safe(event);
+
+ event->aml_handler = native_handler->previous_handler;
+ event->triggering = native_handler->previous_triggering;
+ event->handler_type = native_handler->previous_handler_type;
+
+ if ((event->handler_type == GPE_HANDLER_TYPE_AML_HANDLER ||
+ event->handler_type == GPE_HANDLER_TYPE_IMPLICIT_NOTIFY) &&
+ native_handler->previously_enabled) {
+ gpe_add_user(event, EVENT_CLEAR_IF_FIRST_NO);
+ }
+
+ uacpi_free(native_handler, sizeof(*native_handler));
+
+ if (did_mask)
+ gpe_mask_unmask(event, UACPI_FALSE);
+
+ if (gpe_needs_polling(event))
+ maybe_dispatch_gpe(gpe_device, event);
+out:
+ uacpi_recursive_lock_release(&g_event_lock);
+ return ret;
+}
+
+uacpi_status uacpi_enable_gpe(
+ uacpi_namespace_node *gpe_device, uacpi_u16 idx
+)
+{
+ uacpi_status ret;
+ struct gp_event *event;
+
+ UACPI_ENSURE_INIT_LEVEL_AT_LEAST(UACPI_INIT_LEVEL_NAMESPACE_LOADED);
+
+ ret = uacpi_recursive_lock_acquire(&g_event_lock);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+
+ ret = sanitize_device_and_find_gpe(&gpe_device, idx, &event);
+ if (uacpi_unlikely_error(ret))
+ goto out;
+
+ if (uacpi_unlikely(event->handler_type == GPE_HANDLER_TYPE_NONE)) {
+ ret = UACPI_STATUS_NO_HANDLER;
+ goto out;
+ }
+
+ ret = gpe_add_user(event, EVENT_CLEAR_IF_FIRST_YES);
+ if (uacpi_unlikely_error(ret))
+ goto out;
+
+ if (gpe_needs_polling(event))
+ maybe_dispatch_gpe(gpe_device, event);
+
+out:
+ uacpi_recursive_lock_release(&g_event_lock);
+ return ret;
+}
+
+uacpi_status uacpi_disable_gpe(
+ uacpi_namespace_node *gpe_device, uacpi_u16 idx
+)
+{
+ uacpi_status ret;
+ struct gp_event *event;
+
+ UACPI_ENSURE_INIT_LEVEL_AT_LEAST(UACPI_INIT_LEVEL_NAMESPACE_LOADED);
+
+ ret = uacpi_recursive_lock_acquire(&g_event_lock);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+
+ ret = sanitize_device_and_find_gpe(&gpe_device, idx, &event);
+ if (uacpi_unlikely_error(ret))
+ goto out;
+
+ ret = gpe_remove_user(event);
+out:
+ uacpi_recursive_lock_release(&g_event_lock);
+ return ret;
+}
+
+uacpi_status uacpi_clear_gpe(
+ uacpi_namespace_node *gpe_device, uacpi_u16 idx
+)
+{
+ uacpi_status ret;
+ struct gp_event *event;
+
+ UACPI_ENSURE_INIT_LEVEL_AT_LEAST(UACPI_INIT_LEVEL_NAMESPACE_LOADED);
+
+ ret = uacpi_recursive_lock_acquire(&g_event_lock);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+
+ ret = sanitize_device_and_find_gpe(&gpe_device, idx, &event);
+ if (uacpi_unlikely_error(ret))
+ goto out;
+
+ ret = clear_gpe(event);
+out:
+ uacpi_recursive_lock_release(&g_event_lock);
+ return ret;
+}
+
+static uacpi_status gpe_suspend_resume(
+ uacpi_namespace_node *gpe_device, uacpi_u16 idx, enum gpe_state state
+)
+{
+ uacpi_status ret;
+ struct gp_event *event;
+
+ UACPI_ENSURE_INIT_LEVEL_AT_LEAST(UACPI_INIT_LEVEL_NAMESPACE_LOADED);
+
+ ret = uacpi_recursive_lock_acquire(&g_event_lock);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+
+ ret = sanitize_device_and_find_gpe(&gpe_device, idx, &event);
+ if (uacpi_unlikely_error(ret))
+ goto out;
+
+ event->block_interrupts = state == GPE_STATE_DISABLED;
+ ret = set_gpe_state(event, state);
+out:
+ uacpi_recursive_lock_release(&g_event_lock);
+ return ret;
+}
+
+uacpi_status uacpi_suspend_gpe(
+ uacpi_namespace_node *gpe_device, uacpi_u16 idx
+)
+{
+ return gpe_suspend_resume(gpe_device, idx, GPE_STATE_DISABLED);
+}
+
+uacpi_status uacpi_resume_gpe(
+ uacpi_namespace_node *gpe_device, uacpi_u16 idx
+)
+{
+ return gpe_suspend_resume(gpe_device, idx, GPE_STATE_ENABLED);
+}
+
+uacpi_status uacpi_finish_handling_gpe(
+ uacpi_namespace_node *gpe_device, uacpi_u16 idx
+)
+{
+ uacpi_status ret;
+ struct gp_event *event;
+
+ UACPI_ENSURE_INIT_LEVEL_AT_LEAST(UACPI_INIT_LEVEL_NAMESPACE_LOADED);
+
+ ret = uacpi_recursive_lock_acquire(&g_event_lock);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+
+ ret = sanitize_device_and_find_gpe(&gpe_device, idx, &event);
+ if (uacpi_unlikely_error(ret))
+ goto out;
+
+ event = get_gpe(gpe_device, idx);
+ if (uacpi_unlikely(event == UACPI_NULL)) {
+ ret = UACPI_STATUS_NOT_FOUND;
+ goto out;
+ }
+
+ ret = restore_gpe(event);
+out:
+ uacpi_recursive_lock_release(&g_event_lock);
+ return ret;
+
+}
+
+static uacpi_status gpe_get_mask_unmask(
+ uacpi_namespace_node *gpe_device, uacpi_u16 idx, uacpi_bool should_mask
+)
+{
+ uacpi_status ret;
+ struct gp_event *event;
+
+ UACPI_ENSURE_INIT_LEVEL_AT_LEAST(UACPI_INIT_LEVEL_NAMESPACE_LOADED);
+
+ ret = uacpi_recursive_lock_acquire(&g_event_lock);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+
+ ret = sanitize_device_and_find_gpe(&gpe_device, idx, &event);
+ if (uacpi_unlikely_error(ret))
+ goto out;
+
+ ret = gpe_mask_unmask(event, should_mask);
+
+out:
+ uacpi_recursive_lock_release(&g_event_lock);
+ return ret;
+}
+
+uacpi_status uacpi_mask_gpe(
+ uacpi_namespace_node *gpe_device, uacpi_u16 idx
+)
+{
+ return gpe_get_mask_unmask(gpe_device, idx, UACPI_TRUE);
+}
+
+uacpi_status uacpi_unmask_gpe(
+ uacpi_namespace_node *gpe_device, uacpi_u16 idx
+)
+{
+ return gpe_get_mask_unmask(gpe_device, idx, UACPI_FALSE);
+}
+
+uacpi_status uacpi_setup_gpe_for_wake(
+ uacpi_namespace_node *gpe_device, uacpi_u16 idx,
+ uacpi_namespace_node *wake_device
+)
+{
+ uacpi_status ret;
+ struct gp_event *event;
+ uacpi_bool did_mask;
+
+ UACPI_ENSURE_INIT_LEVEL_AT_LEAST(UACPI_INIT_LEVEL_NAMESPACE_LOADED);
+
+ if (wake_device != UACPI_NULL) {
+ uacpi_bool is_dev = wake_device == uacpi_namespace_root();
+
+ if (!is_dev) {
+ ret = uacpi_namespace_node_is(wake_device, UACPI_OBJECT_DEVICE, &is_dev);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+ }
+
+ if (!is_dev)
+ return UACPI_STATUS_INVALID_ARGUMENT;
+ }
+
+ ret = uacpi_recursive_lock_acquire(&g_event_lock);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+
+ ret = sanitize_device_and_find_gpe(&gpe_device, idx, &event);
+ if (uacpi_unlikely_error(ret))
+ goto out;
+
+ did_mask = gpe_mask_safe(event);
+
+ if (wake_device != UACPI_NULL) {
+ switch (event->handler_type) {
+ case GPE_HANDLER_TYPE_NONE:
+ event->handler_type = GPE_HANDLER_TYPE_IMPLICIT_NOTIFY;
+ event->triggering = UACPI_GPE_TRIGGERING_LEVEL;
+ break;
+
+ case GPE_HANDLER_TYPE_AML_HANDLER:
+ /*
+ * An AML handler already exists, we expect it to call Notify() as
+ * it sees fit. For now just make sure this event is disabled if it
+ * had been enabled automatically previously during initialization.
+ */
+ gpe_remove_user(event);
+ break;
+
+ case GPE_HANDLER_TYPE_NATIVE_HANDLER_RAW:
+ case GPE_HANDLER_TYPE_NATIVE_HANDLER:
+ uacpi_warn(
+ "not configuring implicit notify for GPE(%02X) -> %.4s: "
+ " a user handler already installed\n", event->idx,
+ wake_device->name.text
+ );
+ break;
+
+ // We will re-check this below
+ case GPE_HANDLER_TYPE_IMPLICIT_NOTIFY:
+ break;
+
+ default:
+ uacpi_warn("invalid GPE(%02X) handler type: %d\n",
+ event->idx, event->handler_type);
+ ret = UACPI_STATUS_INTERNAL_ERROR;
+ goto out_unmask;
+ }
+
+ /*
+ * This GPE has no known AML handler, so we configure it to receive
+ * implicit notifications for wake devices when we get a corresponding
+ * GPE triggered. Usually it's the job of a matching AML handler, but
+ * we didn't find any.
+ */
+ if (event->handler_type == GPE_HANDLER_TYPE_IMPLICIT_NOTIFY) {
+ struct gpe_implicit_notify_handler *implicit_handler;
+
+ implicit_handler = event->implicit_handler;
+ while (implicit_handler) {
+ if (implicit_handler->device == wake_device) {
+ ret = UACPI_STATUS_ALREADY_EXISTS;
+ goto out_unmask;
+ }
+
+ implicit_handler = implicit_handler->next;
+ }
+
+ implicit_handler = uacpi_kernel_alloc(sizeof(*implicit_handler));
+ if (uacpi_likely(implicit_handler != UACPI_NULL)) {
+ implicit_handler->device = wake_device;
+ implicit_handler->next = event->implicit_handler;
+ event->implicit_handler = implicit_handler;
+ } else {
+ uacpi_warn(
+ "unable to configure implicit wake for GPE(%02X) -> %.4s: "
+ "out of memory\n", event->idx, wake_device->name.text
+ );
+ }
+ }
+ }
+
+ event->wake = UACPI_TRUE;
+
+out_unmask:
+ if (did_mask)
+ gpe_mask_unmask(event, UACPI_FALSE);
+out:
+ uacpi_recursive_lock_release(&g_event_lock);
+ return ret;
+}
+
+static uacpi_status gpe_enable_disable_for_wake(
+ uacpi_namespace_node *gpe_device, uacpi_u16 idx, uacpi_bool enabled
+)
+{
+ uacpi_status ret;
+ struct gp_event *event;
+ struct gpe_register *reg;
+ uacpi_u8 mask;
+
+ UACPI_ENSURE_INIT_LEVEL_AT_LEAST(UACPI_INIT_LEVEL_NAMESPACE_LOADED);
+
+ ret = uacpi_recursive_lock_acquire(&g_event_lock);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+
+ ret = sanitize_device_and_find_gpe(&gpe_device, idx, &event);
+ if (uacpi_unlikely_error(ret))
+ goto out;
+
+ if (!event->wake) {
+ ret = UACPI_STATUS_INVALID_ARGUMENT;
+ goto out;
+ }
+
+ reg = event->reg;
+ mask = gpe_get_mask(event);
+
+ if (enabled)
+ reg->wake_mask |= mask;
+ else
+ reg->wake_mask &= mask;
+
+out:
+ uacpi_recursive_lock_release(&g_event_lock);
+ return ret;
+}
+
+uacpi_status uacpi_enable_gpe_for_wake(
+ uacpi_namespace_node *gpe_device, uacpi_u16 idx
+)
+{
+ return gpe_enable_disable_for_wake(gpe_device, idx, UACPI_TRUE);
+}
+
+uacpi_status uacpi_disable_gpe_for_wake(
+ uacpi_namespace_node *gpe_device, uacpi_u16 idx
+)
+{
+ return gpe_enable_disable_for_wake(gpe_device, idx, UACPI_FALSE);
+}
+
+struct do_for_all_gpes_ctx {
+ enum gpe_block_action action;
+ uacpi_status ret;
+};
+
+static uacpi_iteration_decision do_for_all_gpes(
+ struct gpe_block *block, uacpi_handle opaque
+)
+{
+ struct do_for_all_gpes_ctx *ctx = opaque;
+
+ ctx->ret = gpe_block_apply_action(block, ctx->action);
+ if (uacpi_unlikely_error(ctx->ret))
+ return UACPI_ITERATION_DECISION_BREAK;
+
+ return UACPI_ITERATION_DECISION_CONTINUE;
+}
+
+static uacpi_status for_all_gpes_locked(struct do_for_all_gpes_ctx *ctx)
+{
+ uacpi_status ret;
+
+ UACPI_ENSURE_INIT_LEVEL_AT_LEAST(UACPI_INIT_LEVEL_NAMESPACE_LOADED);
+
+ ret = uacpi_recursive_lock_acquire(&g_event_lock);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+
+ for_each_gpe_block(do_for_all_gpes, ctx);
+
+ uacpi_recursive_lock_release(&g_event_lock);
+ return ctx->ret;
+}
+
+uacpi_status uacpi_disable_all_gpes(void)
+{
+ struct do_for_all_gpes_ctx ctx = {
+ .action = GPE_BLOCK_ACTION_DISABLE_ALL,
+ };
+ return for_all_gpes_locked(&ctx);
+}
+
+uacpi_status uacpi_enable_all_runtime_gpes(void)
+{
+ struct do_for_all_gpes_ctx ctx = {
+ .action = GPE_BLOCK_ACTION_ENABLE_ALL_FOR_RUNTIME,
+ };
+ return for_all_gpes_locked(&ctx);
+}
+
+uacpi_status uacpi_enable_all_wake_gpes(void)
+{
+ struct do_for_all_gpes_ctx ctx = {
+ .action = GPE_BLOCK_ACTION_ENABLE_ALL_FOR_WAKE,
+ };
+ return for_all_gpes_locked(&ctx);
+}
+
+static uacpi_status initialize_gpes(void)
+{
+ uacpi_status ret;
+ uacpi_namespace_node *gpe_node;
+ struct acpi_fadt *fadt = &g_uacpi_rt_ctx.fadt;
+ uacpi_u8 gpe0_regs = 0, gpe1_regs = 0;
+
+ gpe_node = uacpi_namespace_get_predefined(UACPI_PREDEFINED_NAMESPACE_GPE);
+
+ if (fadt->x_gpe0_blk.address && fadt->gpe0_blk_len) {
+ gpe0_regs = fadt->gpe0_blk_len / 2;
+
+ ret = create_gpe_block(
+ gpe_node, fadt->sci_int, 0, fadt->x_gpe0_blk.address,
+ fadt->x_gpe0_blk.address_space_id, gpe0_regs
+ );
+ if (uacpi_unlikely_error(ret)) {
+ uacpi_error("unable to create FADT GPE block 0: %s\n",
+ uacpi_status_to_string(ret));
+ }
+ }
+
+ if (fadt->x_gpe1_blk.address && fadt->gpe1_blk_len) {
+ gpe1_regs = fadt->gpe1_blk_len / 2;
+
+ if (uacpi_unlikely((gpe0_regs * EVENTS_PER_GPE_REGISTER) >
+ fadt->gpe1_base)) {
+ uacpi_error(
+ "FADT GPE block 1 [%d->%d] collides with GPE block 0 "
+ "[%d->%d], ignoring\n",
+ 0, gpe0_regs * EVENTS_PER_GPE_REGISTER, fadt->gpe1_base,
+ gpe1_regs * EVENTS_PER_GPE_REGISTER
+ );
+ gpe1_regs = 0;
+ goto out;
+ }
+
+ ret = create_gpe_block(
+ gpe_node, fadt->sci_int, fadt->gpe1_base, fadt->x_gpe1_blk.address,
+ fadt->x_gpe1_blk.address_space_id, gpe1_regs
+ );
+ if (uacpi_unlikely_error(ret)) {
+ uacpi_error("unable to create FADT GPE block 1: %s\n",
+ uacpi_status_to_string(ret));
+ }
+ }
+
+ if (gpe0_regs == 0 && gpe1_regs == 0)
+ uacpi_trace("platform has no FADT GPE events\n");
+
+out:
+ return UACPI_STATUS_OK;
+}
+
+uacpi_status uacpi_install_gpe_block(
+ uacpi_namespace_node *gpe_device, uacpi_u64 address,
+ uacpi_address_space address_space, uacpi_u16 num_registers, uacpi_u32 irq
+)
+{
+ uacpi_status ret;
+ uacpi_bool is_dev;
+
+ UACPI_ENSURE_INIT_LEVEL_AT_LEAST(UACPI_INIT_LEVEL_NAMESPACE_LOADED);
+
+ ret = uacpi_namespace_node_is(gpe_device, UACPI_OBJECT_DEVICE, &is_dev);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+ if (!is_dev)
+ return UACPI_STATUS_INVALID_ARGUMENT;
+
+ ret = uacpi_recursive_lock_acquire(&g_event_lock);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+
+ if (uacpi_unlikely(get_gpe(gpe_device, 0) != UACPI_NULL)) {
+ ret = UACPI_STATUS_ALREADY_EXISTS;
+ goto out;
+ }
+
+ ret = create_gpe_block(
+ gpe_device, irq, 0, address, address_space, num_registers
+ );
+
+out:
+ uacpi_recursive_lock_release(&g_event_lock);
+ return ret;
+}
+
+uacpi_status uacpi_uninstall_gpe_block(
+ uacpi_namespace_node *gpe_device
+)
+{
+ uacpi_status ret;
+ uacpi_bool is_dev;
+ struct gpe_search_ctx search_ctx = { 0 };
+
+ search_ctx.idx = 0;
+ search_ctx.gpe_device = gpe_device;
+
+ UACPI_ENSURE_INIT_LEVEL_AT_LEAST(UACPI_INIT_LEVEL_NAMESPACE_LOADED);
+
+ ret = uacpi_namespace_node_is(gpe_device, UACPI_OBJECT_DEVICE, &is_dev);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+ if (!is_dev)
+ return UACPI_STATUS_INVALID_ARGUMENT;
+
+ ret = uacpi_recursive_lock_acquire(&g_event_lock);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+
+ for_each_gpe_block(do_find_gpe, &search_ctx);
+ if (search_ctx.out_block == UACPI_NULL) {
+ ret = UACPI_STATUS_NOT_FOUND;
+ goto out;
+ }
+
+ uninstall_gpe_block(search_ctx.out_block);
+
+out:
+ uacpi_recursive_lock_release(&g_event_lock);
+ return ret;
+}
+
+static uacpi_interrupt_ret handle_global_lock(uacpi_handle ctx)
+{
+ uacpi_cpu_flags flags;
+ UACPI_UNUSED(ctx);
+
+ if (uacpi_unlikely(!g_uacpi_rt_ctx.has_global_lock)) {
+ uacpi_warn("platform has no global lock but a release event "
+ "was fired anyway?\n");
+ return UACPI_INTERRUPT_HANDLED;
+ }
+
+ flags = uacpi_kernel_lock_spinlock(g_uacpi_rt_ctx.global_lock_spinlock);
+ if (!g_uacpi_rt_ctx.global_lock_pending) {
+ uacpi_trace("spurious firmware global lock release notification\n");
+ goto out;
+ }
+
+ uacpi_trace("received a firmware global lock release notification\n");
+
+ uacpi_kernel_signal_event(g_uacpi_rt_ctx.global_lock_event);
+ g_uacpi_rt_ctx.global_lock_pending = UACPI_FALSE;
+
+out:
+ uacpi_kernel_unlock_spinlock(g_uacpi_rt_ctx.global_lock_spinlock, flags);
+ return UACPI_INTERRUPT_HANDLED;
+}
+
+static uacpi_interrupt_ret handle_sci(uacpi_handle ctx)
+{
+ uacpi_interrupt_ret int_ret = UACPI_INTERRUPT_NOT_HANDLED;
+
+ int_ret |= handle_fixed_events();
+ int_ret |= handle_gpes(ctx);
+
+ return int_ret;
+}
+
+uacpi_status uacpi_initialize_events_early(void)
+{
+ uacpi_status ret;
+
+ if (uacpi_is_hardware_reduced())
+ return UACPI_STATUS_OK;
+
+ g_gpe_state_slock = uacpi_kernel_create_spinlock();
+ if (uacpi_unlikely(g_gpe_state_slock == UACPI_NULL))
+ return UACPI_STATUS_OUT_OF_MEMORY;
+
+ ret = uacpi_recursive_lock_init(&g_event_lock);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+
+ ret = initialize_fixed_events();
+ if (uacpi_unlikely_error(ret))
+ return ret;
+
+ return UACPI_STATUS_OK;
+}
+
+uacpi_status uacpi_initialize_events(void)
+{
+ uacpi_status ret;
+
+ if (uacpi_is_hardware_reduced())
+ return UACPI_STATUS_OK;
+
+ ret = initialize_gpes();
+ if (uacpi_unlikely_error(ret))
+ return ret;
+
+ ret = uacpi_kernel_install_interrupt_handler(
+ g_uacpi_rt_ctx.fadt.sci_int, handle_sci, g_gpe_interrupt_head,
+ &g_uacpi_rt_ctx.sci_handle
+ );
+ if (uacpi_unlikely_error(ret)) {
+ uacpi_error(
+ "unable to install SCI interrupt handler: %s\n",
+ uacpi_status_to_string(ret)
+ );
+ return ret;
+ }
+ g_uacpi_rt_ctx.sci_handle_valid = UACPI_TRUE;
+
+ g_uacpi_rt_ctx.global_lock_event = uacpi_kernel_create_event();
+ if (uacpi_unlikely(g_uacpi_rt_ctx.global_lock_event == UACPI_NULL))
+ return UACPI_STATUS_OUT_OF_MEMORY;
+
+ g_uacpi_rt_ctx.global_lock_spinlock = uacpi_kernel_create_spinlock();
+ if (uacpi_unlikely(g_uacpi_rt_ctx.global_lock_spinlock == UACPI_NULL))
+ return UACPI_STATUS_OUT_OF_MEMORY;
+
+ ret = uacpi_install_fixed_event_handler(
+ UACPI_FIXED_EVENT_GLOBAL_LOCK, handle_global_lock, UACPI_NULL
+ );
+ if (uacpi_likely_success(ret)) {
+ if (uacpi_unlikely(g_uacpi_rt_ctx.facs == UACPI_NULL)) {
+ uacpi_uninstall_fixed_event_handler(UACPI_FIXED_EVENT_GLOBAL_LOCK);
+ uacpi_warn("platform has global lock but no FACS was provided\n");
+ return ret;
+ }
+ g_uacpi_rt_ctx.has_global_lock = UACPI_TRUE;
+ } else if (ret == UACPI_STATUS_HARDWARE_TIMEOUT) {
+ // has_global_lock remains set to false
+ uacpi_trace("platform has no global lock\n");
+ ret = UACPI_STATUS_OK;
+ }
+
+ return ret;
+}
+
+void uacpi_deinitialize_events(void)
+{
+ struct gpe_interrupt_ctx *ctx, *next_ctx = g_gpe_interrupt_head;
+ uacpi_size i;
+
+ g_gpes_finalized = UACPI_FALSE;
+
+ if (g_uacpi_rt_ctx.sci_handle_valid) {
+ uacpi_kernel_uninstall_interrupt_handler(
+ handle_sci, g_uacpi_rt_ctx.sci_handle
+ );
+ g_uacpi_rt_ctx.sci_handle_valid = UACPI_FALSE;
+ }
+
+ while (next_ctx) {
+ struct gpe_block *block, *next_block;
+
+ ctx = next_ctx;
+ next_ctx = ctx->next;
+
+ next_block = ctx->gpe_head;
+ while (next_block) {
+ block = next_block;
+ next_block = block->next;
+ uninstall_gpe_block(block);
+ }
+ }
+
+ for (i = 0; i < UACPI_FIXED_EVENT_MAX; ++i) {
+ if (fixed_event_handlers[i].handler)
+ uacpi_uninstall_fixed_event_handler(i);
+ }
+
+ if (g_gpe_state_slock != UACPI_NULL) {
+ uacpi_kernel_free_spinlock(g_gpe_state_slock);
+ g_gpe_state_slock = UACPI_NULL;
+ }
+
+ uacpi_recursive_lock_deinit(&g_event_lock);
+
+ g_gpe_interrupt_head = UACPI_NULL;
+}
+
+uacpi_status uacpi_install_fixed_event_handler(
+ uacpi_fixed_event event, uacpi_interrupt_handler handler,
+ uacpi_handle user
+)
+{
+ uacpi_status ret;
+ struct fixed_event_handler *ev;
+
+ UACPI_ENSURE_INIT_LEVEL_AT_LEAST(UACPI_INIT_LEVEL_SUBSYSTEM_INITIALIZED);
+
+ if (uacpi_unlikely(event < 0 || event > UACPI_FIXED_EVENT_MAX))
+ return UACPI_STATUS_INVALID_ARGUMENT;
+ if (uacpi_is_hardware_reduced())
+ return UACPI_STATUS_OK;
+
+ ret = uacpi_recursive_lock_acquire(&g_event_lock);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+
+ ev = &fixed_event_handlers[event];
+
+ if (ev->handler != UACPI_NULL) {
+ ret = UACPI_STATUS_ALREADY_EXISTS;
+ goto out;
+ }
+
+ ev->handler = handler;
+ ev->ctx = user;
+
+ ret = set_event(event, UACPI_EVENT_ENABLED);
+ if (uacpi_unlikely_error(ret)) {
+ ev->handler = UACPI_NULL;
+ ev->ctx = UACPI_NULL;
+ }
+
+out:
+ uacpi_recursive_lock_release(&g_event_lock);
+ return ret;
+}
+
+uacpi_status uacpi_uninstall_fixed_event_handler(
+ uacpi_fixed_event event
+)
+{
+ uacpi_status ret;
+ struct fixed_event_handler *ev;
+
+ UACPI_ENSURE_INIT_LEVEL_AT_LEAST(UACPI_INIT_LEVEL_SUBSYSTEM_INITIALIZED);
+
+ if (uacpi_unlikely(event < 0 || event > UACPI_FIXED_EVENT_MAX))
+ return UACPI_STATUS_INVALID_ARGUMENT;
+ if (uacpi_is_hardware_reduced())
+ return UACPI_STATUS_OK;
+
+ ret = uacpi_recursive_lock_acquire(&g_event_lock);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+
+ ev = &fixed_event_handlers[event];
+
+ ret = set_event(event, UACPI_EVENT_DISABLED);
+ if (uacpi_unlikely_error(ret))
+ goto out;
+
+ uacpi_kernel_wait_for_work_completion();
+
+ ev->handler = UACPI_NULL;
+ ev->ctx = UACPI_NULL;
+
+out:
+ uacpi_recursive_lock_release(&g_event_lock);
+ return ret;
+}
+
+uacpi_status uacpi_fixed_event_info(
+ uacpi_fixed_event event, uacpi_event_info *out_info
+)
+{
+ uacpi_status ret;
+ const struct fixed_event *ev;
+ uacpi_u64 raw_value;
+ uacpi_event_info info = 0;
+
+ UACPI_ENSURE_INIT_LEVEL_AT_LEAST(UACPI_INIT_LEVEL_SUBSYSTEM_INITIALIZED);
+
+ if (uacpi_unlikely(event < 0 || event > UACPI_FIXED_EVENT_MAX))
+ return UACPI_STATUS_INVALID_ARGUMENT;
+ if (uacpi_is_hardware_reduced())
+ return UACPI_STATUS_NOT_FOUND;
+
+ ret = uacpi_recursive_lock_acquire(&g_event_lock);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+
+ if (fixed_event_handlers[event].handler != UACPI_NULL)
+ info |= UACPI_EVENT_INFO_HAS_HANDLER;
+
+ ev = &fixed_events[event];
+
+ ret = uacpi_read_register_field(ev->enable_field, &raw_value);
+ if (uacpi_unlikely_error(ret))
+ goto out;
+ if (raw_value)
+ info |= UACPI_EVENT_INFO_ENABLED | UACPI_EVENT_INFO_HW_ENABLED;
+
+ ret = uacpi_read_register_field(ev->status_field, &raw_value);
+ if (uacpi_unlikely_error(ret))
+ goto out;
+ if (raw_value)
+ info |= UACPI_EVENT_INFO_HW_STATUS;
+
+ *out_info = info;
+out:
+ uacpi_recursive_lock_release(&g_event_lock);
+ return ret;
+}
+
+uacpi_status uacpi_gpe_info(
+ uacpi_namespace_node *gpe_device, uacpi_u16 idx, uacpi_event_info *out_info
+)
+{
+ uacpi_status ret;
+ struct gp_event *event;
+ struct gpe_register *reg;
+ uacpi_u8 mask;
+ uacpi_u64 raw_value;
+ uacpi_event_info info = 0;
+
+ UACPI_ENSURE_INIT_LEVEL_AT_LEAST(UACPI_INIT_LEVEL_NAMESPACE_LOADED);
+
+ ret = uacpi_recursive_lock_acquire(&g_event_lock);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+
+ ret = sanitize_device_and_find_gpe(&gpe_device, idx, &event);
+ if (uacpi_unlikely_error(ret))
+ goto out;
+
+ if (event->handler_type != GPE_HANDLER_TYPE_NONE)
+ info |= UACPI_EVENT_INFO_HAS_HANDLER;
+
+ mask = gpe_get_mask(event);
+ reg = event->reg;
+
+ if (reg->runtime_mask & mask)
+ info |= UACPI_EVENT_INFO_ENABLED;
+ if (reg->masked_mask & mask)
+ info |= UACPI_EVENT_INFO_MASKED;
+ if (reg->wake_mask & mask)
+ info |= UACPI_EVENT_INFO_ENABLED_FOR_WAKE;
+
+ ret = uacpi_gas_read_mapped(&reg->enable, &raw_value);
+ if (uacpi_unlikely_error(ret))
+ goto out;
+ if (raw_value & mask)
+ info |= UACPI_EVENT_INFO_HW_ENABLED;
+
+ ret = uacpi_gas_read_mapped(&reg->status, &raw_value);
+ if (uacpi_unlikely_error(ret))
+ goto out;
+ if (raw_value & mask)
+ info |= UACPI_EVENT_INFO_HW_STATUS;
+
+ *out_info = info;
+out:
+ uacpi_recursive_lock_release(&g_event_lock);
+ return ret;
+}
+
+#define PM1_STATUS_BITS ( \
+ ACPI_PM1_STS_TMR_STS_MASK | \
+ ACPI_PM1_STS_BM_STS_MASK | \
+ ACPI_PM1_STS_GBL_STS_MASK | \
+ ACPI_PM1_STS_PWRBTN_STS_MASK | \
+ ACPI_PM1_STS_SLPBTN_STS_MASK | \
+ ACPI_PM1_STS_RTC_STS_MASK | \
+ ACPI_PM1_STS_PCIEXP_WAKE_STS_MASK | \
+ ACPI_PM1_STS_WAKE_STS_MASK \
+)
+
+uacpi_status uacpi_clear_all_events(void)
+{
+ uacpi_status ret;
+ struct do_for_all_gpes_ctx ctx = {
+ .action = GPE_BLOCK_ACTION_CLEAR_ALL,
+ };
+
+ UACPI_ENSURE_INIT_LEVEL_AT_LEAST(UACPI_INIT_LEVEL_NAMESPACE_LOADED);
+
+ ret = uacpi_recursive_lock_acquire(&g_event_lock);
+ if (uacpi_unlikely_error(ret))
+ return ret;
+
+ ret = uacpi_write_register(UACPI_REGISTER_PM1_STS, PM1_STATUS_BITS);
+ if (uacpi_unlikely_error(ret))
+ goto out;
+
+ for_each_gpe_block(do_for_all_gpes, &ctx);
+ ret = ctx.ret;
+
+out:
+ uacpi_recursive_lock_release(&g_event_lock);
+ return ret;
+}
+
+#endif // !UACPI_REDUCED_HARDWARE && !UACPI_BAREBONES_MODE