Struct TracePerformanceEventManager 

pub struct TracePerformanceEventManager { /* private fields */ }

Kernel-wide manager of pid/tid to koid mappings, shared by all recording clients (system tracing and CPU profiling). Stored in kernel.trace_event_manager.

The manager lives for the lifetime of the Kernel; whether recording is active is tracked by the session count inside it, not by its presence. Clients start recording by obtaining a PidKoidSession from TracePerformanceEventManager::open and stop by dropping it, so overlapping sessions from independent clients compose correctly: the map is seeded when the first session opens and released when the last one drops, and cleanup runs on every exit path because it is driven by Drop.

When no session is active, task creation fast-paths out on a single relaxed atomic load with no locks and no allocations.

Implementations

impl TracePerformanceEventManager

pub fn new(weak_kernel: Weak<Kernel>) -> Self

Creates a new manager holding a weak reference to the enclosing Kernel.

pub fn is_recording(&self) -> bool

Returns true if at least one session is actively recording mappings.

pub fn open(self: &Arc<Self>) -> PidKoidSession

Opens a session. When the first session opens, the map is seeded with all currently running Tasks; tasks created afterwards record themselves via Task::record_pid_koid_mapping. Recording continues until all active sessions have been dropped.

pub fn record(&self, pid: pid_t, tid: tid_t, identity: ZirconIdentity)

Records the mapping for one Task if recording is active. Called from Task creation, after the task’s Zircon thread handle is attached, so the identity is always complete.