Ryan

This commit **enhances the `heapprofd` memory profiling system** by **exposing a new `reallocarray` hook** for bionic. It introduces the `heapprofd_reallocarray` function in `src/profiling/memory/malloc_interceptor_bionic_hooks.cc`, allowing the bionic allocator to intercept and profile `reallocarray` calls. Concurrently, the `wrap_reallocarray` function in `src/profiling/memory/wrap_allocators.cc` is **improved to include an overflow check** for `nmemb * size`, ensuring correct allocation size reporting and robust handling of potential integer overflow scenarios. This work provides **more comprehensive and robust memory profiling** for `reallocarray` operations, addressing a reported bug and improving the accuracy of memory usage analysis.

This commit **enhances the UI's track grouping functionality** by ensuring that **process-scoped kernel thread tracks** are correctly aggregated under the "Kernel threads" group, resolving an issue where they were previously overlooked. It **refactors** the `ProcessThreadGroups` plugin, simplifying the kernel thread query by leveraging the `viz.threads` module and removing an always-empty summary track for a cleaner display. This **feature enhancement** provides users with a more comprehensive and organized view of kernel activity within the trace UI, with numerous UI screenshots updated to reflect these visual changes.

This commit performs **code cleanup** by **removing unused files** from the **`traceconv` utility**. These files, located within the `traceconv` module, became obsolete and unreferenced following a prior change (r.android.com/2005680). This **maintenance task** streamlines the codebase, reducing project bloat and improving overall maintainability without impacting any active functionality.

This commit introduces a significant **feature enhancement** to the `traceconv profile` command, enabling **automatic detection of profile types** within a trace. It **refactors** the profile conversion logic into a unified `TraceToProfile` function and `DetectConversionMode`, simplifying the user experience by eliminating the need for explicit flags for common profile types like allocator, perf, and Android Java heap graphs. The `traceconv` tool will now intelligently identify the profile type, prompting the user for clarification only if multiple types are present, and includes **updated documentation** and **integration tests** for this new functionality.

This commit performs a **maintenance update** to the **Ftrace data model and serialization/deserialization logic** for `workqueue` events within Perfetto. It **regenerates protobuf definitions** and **adjusts the Ftrace parser and event information** to align with recent Linux kernel changes, such as the addition of `activate_work.function`, `queue_work.workqueue` becoming a string, and `queue_work.*cpu` fields becoming signed. This ensures correct ingestion of `workqueue` trace data from newer kernels (v5.19+), though it introduces a known limitation where `*cpu` fields might be skipped on kernels v5.18 and below due to encoder constraints. The changes span `protos/perfetto/trace/ftrace/workqueue.proto`, `src/trace_processor/importers/ftrace/`, and `src/traced/probes/ftrace/`.

This commit introduces a **breaking change** to the **multi-machine tracing** behavior in Perfetto, specifically altering how **data sources are matched** across remote producers connected via `traced_relay`. By default, data sources will now only match on the host machine, rather than automatically matching across all connected machines. A **new configuration option**, `trace_all_machines`, has been added to `TraceConfig` to allow users to explicitly enable matching data sources from all machines. This **new capability** and default change prevents unexpected data collection for users accustomed to single-machine tracing, while still allowing explicit multi-machine data collection and indicating the presence of remote machines via `system_info` and `remote_clock_sync` packets.

This commit **introduces a new capability** to the **Trace Processor's ftrace importer** by adding support for the `fwtp_perfetto_counter` ftrace event. It updates `src/trace_processor/importers/ftrace/ftrace_descriptors.cc` with a new `FtraceMessageDescriptor` that defines the event's structure, including its fields and types. This **feature enhancement** allows the trace processor to correctly parse and interpret data from this specific ftrace event, providing more comprehensive insights for **tprobes** and related performance counter analysis. The update was performed by running `tools/run_ftrace_proto_gen`, ensuring the trace processor's ftrace event definitions are up-to-date.

This commit **increases the shared memory allocation** for the **`traced_probes`** component from 1MB to 2MB within `src/traced/probes/probes_producer.cc`. This **bug fix** addresses a resource starvation issue where `traced_probes` was experiencing shared memory exhaustion when many concurrent data sources, particularly from `droidfood` tracing, were active. By doubling the requested shared memory, the change provides critical **headroom** to prevent future starvation, thereby improving the reliability and performance of tracing operations under heavy load.

This commit introduces **new crash keys** within the **`tprobes`** module to track the number of active data sources and tracing sessions. Specifically, `src/traced/probes/probes_producer.cc` is updated to increment and decrement these keys in the `SetupDataSource` and `StopDataSource` functions. This **debugging/tracking enhancement** aims to diagnose shared memory starvation issues in `traced_probes` that can occur when `droidfood tracing` instantiates many concurrent data sources. By providing this telemetry, the change helps understand and resolve performance bottlenecks related to shared memory capacity on Android.

This commit **refactors** the **`trace_processor`'s `perf` data importer** by renaming the `PerfSession` class to `PerfInvocation`. This extensive change, affecting files like `perf_invocation.h`, `perf_invocation.cc`, and various parsing components such as `perf_data_tokenizer.cc` and `record_parser.cc`, aims to avoid future confusion with `perf_session_id` identifiers. It updates class names, member variables, include paths, and build configurations across the entire `perf` importer subsystem. The primary goal is to improve code clarity and maintainability, ensuring better understanding for developers without altering any functional behavior.

This commit **fixes** a semantic issue in the **Trace Processor's perf.data importer** where `perf_session_id` was incorrectly assigned per container file, rather than uniquely identifying each sampling stream. The **bug fix** involves **refactoring** the `PerfSession` builder to assign a distinct `perf_session_id` to each `PerfEventAttr` and updating `InternSample` to use this correct ID. This change ensures that **UI tracks** can properly distinguish between different sampling streams, resolving previous display errors caused by multiple leader counters appearing for the same `perf_session_id`. Consequently, the **perf.data parsing** and **trace visualization** are significantly improved for `traced_perf` packets.

This commit introduces a **feature enhancement** to the **trace processor's system probes parser** to improve thread naming within process trees. It modifies the `ParseProcessTree` logic in `system_probes_parser.cc` to **set the main thread's name from the process's `cmdline`** when the `cmdline_is_comm` flag is true. This change addresses scenarios where `sched_switch` events are absent, preventing anonymous kernel threads in trace visualizations. A new test case in `tests.py` verifies this functionality, ensuring more accurate and complete **process and thread identification** in analyzed traces.

This commit **reverts** the previously introduced support for cloning tracing sessions configured to write directly into a file. This **temporary rollback** affects the **tracing subsystem**, specifically components like `perfetto_cmd`, `tracing_service_impl`, and `consumer_ipc_client_impl`, by removing the logic for handling output file descriptors and the `was_write_into_file` flag during session cloning. The **revert** is necessary because accompanying **SELinux changes** are required for Android integration and are not yet ready, unblocking further development until those security policies are in place. Consequently, the ability to clone `write_into_file` sessions is currently unavailable.

This commit performs a **refactoring** and **cleanup** within the **`ftrace` probe module** to improve terminology accuracy. It systematically renames internal variables and parameters from `ftrace_` to `tracefs_` across core components like `FtraceConfigMuxer`, `predefined_tracepoints`, and `vendor_tracepoints`, as `tracefs` more precisely refers to the virtual filesystem. Additionally, the `ftrace_` member in `probes_producer` is renamed to `ftrace_controller_` to better reflect its specific type. This **internal code improvement** enhances readability and maintainability for developers without altering any runtime behavior or functionality.

This commit **enables `denser_generic_event_encoding` by default** for **Ftrace configurations** within the `traced` Ftrace probe. This **feature rollout** changes the default behavior, causing Ftrace generic events to be encoded using a more self-describing proto format. The `FtraceConfigMuxer` (`src/traced/probes/ftrace/ftrace_config_muxer.cc`) is updated to apply this default, and **documentation** in `ftrace_config.proto` and related proto files now explains the new behavior. This aims to improve the self-describing nature of Ftrace protos, with users able to explicitly set the config field to revert to the old encoding if needed.

This commit introduces a **new capability** to the **`tprobes` ftrace subsystem** by ensuring that the **group name** of generic ftrace events (e.g., "sched" for "sched/sched_switch") is preserved within the Perfetto trace. It adds an optional `group_name` field to the `GenericEventDescriptor` proto message in `ftrace_event_bundle.proto` and `perfetto_trace.proto`. The serialization logic in `proto_translation_table.cc` and `cpu_reader.cc` is updated to populate this field, alongside a **refactoring** of the internal descriptor management using a new `GenericEventProtoDescriptors` struct. This change prevents information loss for self-describing ftrace events, making the trace data more complete and robust for future `trace_processor` analysis.

This commit performs **maintenance refactoring** within the **`protozero` library** by **removing the vestigial `HAS_NONPACKED_REPEATED_FIELDS` template parameter from the `TypedProtoDecoder` class**. This parameter was no longer used after a 2021 implementation change, making it dead code. The change simplifies the `TypedProtoDecoder`'s signature in `include/perfetto/protozero/proto_decoder.h` and updates its generation logic in `src/protozero/protoc_plugin/protozero_plugin.cc`. Consequently, all instantiations of `TypedProtoDecoder` in unit tests and fuzzers are adjusted to reflect the cleaner API, improving code clarity and reducing complexity without altering functionality.

This commit **enhances the Perfetto UI's LinuxPerf plugin** by improving the clarity and usability of performance sample tracks. It introduces **nested perf sample tracks** and **summary tracks**, naming them after their respective counters to better represent the sampled data, which is particularly beneficial for complex traces like Android binder callstacks. The commit also **refactors the track creation logic** into `createPerfCallsitesTrack` for dynamic filtering and adjusts the auto-selection behavior to only open the perf panel if a single process is profiled. This **feature enhancement** provides more informative track labels and a more intuitive UI flow, supported by new Playwright tests.

This commit introduces a **new capability** to the `traced_perf` subsystem, enabling the use of **event modifiers** (e.g., `:u` for userspace, `:k` for kernel) for perf event configurations. These modifiers allow users to specify the execution context (userspace, kernel, hypervisor) in which a perf event should count, providing more granular profiling data. The change involves **schema updates** in `protos/perfetto/common/perf_events.proto` and `protos/perfetto/config/perfetto_config.proto`, **backend implementation** in `src/profiling/perf` to apply these modifiers, and **trace processor integration** in `src/trace_processor/importers/proto/perf_sample_tracker.cc` to append modifier suffixes to counter names. This significantly enhances the flexibility and precision of performance data collection and analysis within Perfetto.

This commit provides a **documentation fix** for the **ftrace documentation**, specifically clarifying the usage of the `__assign_str` and `__string` macros. It updates the `ftrace.md` file to accurately describe how these macros function across different kernel versions. This **maintenance** work improves the clarity and correctness of the **ftrace getting started guide**, helping users understand string handling within ftrace. The change addresses previous attempts to clarify this syntax, ensuring the documentation is precise for various kernel environments.