Public API for the EtherCAT master runtime.
EtherCAT is the entry point for a master-owned session lifecycle:
Masterowns startup, activation-blocked startup, and recovery policyDomainowns cyclic LRW exchange and the logical PDO imageSlaveowns ESM transitions and slave-local configurationDCowns distributed-clock initialization and runtime maintenanceBusserializes all frame I/O
Runtime footprint is intentionally small: no NIFs, no kernel module, and a
minimal runtime dependency surface. The bus uses raw sockets, sysfs, and OTP
directly, with :telemetry as the only runtime Hex dependency.
Public session lifecycle is exposed through state/0:
:idle:discovering:awaiting_preop:preop_ready:deactivated:operational:activation_blocked:recovering
await_running/1 waits for a usable session. Startup/activation paths drain
startup bus traffic before reporting ready, so the first mailbox or OP
exchange starts from a clean transport state. await_operational/1 waits for
cyclic OP. Per-slave health is exposed through slaves/0.
Runtime Lifecycle
This is the user-facing state/0 lifecycle. It shows the main session flow
without trying to encode every exact Master state transition in one dense
graph.
flowchart TD
A[start/1] --> B[discovering]
B --> C[awaiting_preop]
B -->|startup fails or stop/0| Z[idle]
C -->|configured slaves become usable in PREOP| D{activation requested?}
C -->|timeout, fatal startup failure, or stop/0| Z
D -->|no| E[preop_ready]
D -->|yes, target reached| F[operational]
D -->|yes, transition incomplete| G[activation_blocked]
E -->|activate/0 succeeds| F
E -->|activate/0 is incomplete| G
F -->|deactivate/0 to SAFEOP| H[deactivated]
F -->|deactivate/0 to PREOP| E
E -->|critical runtime fault| I[recovering]
H -->|critical runtime fault| I
F -->|critical runtime fault| I
G -->|runtime faults remain after activation retry| I
G -->|retry reaches OP| F
G -->|retry settles in SAFEOP| H
G -->|retry settles in PREOP| E
I -->|faults cleared, target OP| F
I -->|faults cleared, target SAFEOP| H
I -->|faults cleared, target PREOP| E
E -->|stop/0 or fatal exit| Z
H -->|stop/0 or fatal exit| Z
F -->|stop/0 or fatal exit| Z
G -->|stop/0 or fatal exit| Z
I -->|recovery fails or stop/0| ZPhysical link loss normally appears here as a runtime :recovering transition,
not an immediate return to :idle. :idle is reserved for explicit stop,
startup failure, bus-process exit, or fatal policy. For the exact master-side
state semantics, read EtherCAT.Master.
Startup Sequence
sequenceDiagram
autonumber
participant App
participant EtherCAT
participant Bus
participant Slaves
participant DC
App->>EtherCAT: start/1
EtherCAT->>Bus: discover ring, assign stations, verify link
opt DC is configured
EtherCAT->>DC: initialize clocks
end
EtherCAT->>Slaves: start configured slaves
Slaves->>Bus: reach PREOP through SII, mailbox, and PDO setup
Slaves-->>EtherCAT: report ready at PREOP
opt activation is requested and possible
EtherCAT->>Bus: start cyclic domains
opt DC runtime is available
EtherCAT->>DC: start runtime maintenance
opt DC lock is required
EtherCAT->>DC: wait for lock
end
end
EtherCAT->>Slaves: request SAFEOP then OP
end
EtherCAT-->>App: preop_ready, activation_blocked, or operationalUsage
EtherCAT.start(
interface: "eth0",
dc: %EtherCAT.DC.Config{
cycle_ns: 1_000_000,
await_lock?: true,
lock_policy: :recovering
},
domains: [
%EtherCAT.Domain.Config{id: :main, cycle_time_us: 1_000}
],
slaves: [
%EtherCAT.Slave.Config{name: :coupler},
%EtherCAT.Slave.Config{
name: :sensor,
driver: MyApp.EL1809,
process_data: {:all, :main}
},
%EtherCAT.Slave.Config{
name: :valve,
driver: MyApp.EL2809,
process_data: {:all, :main}
}
]
)
:ok = EtherCAT.await_running()
EtherCAT.subscribe(:sensor, :ch1) # receive {:ethercat, :signal, :sensor, :ch1, value}
EtherCAT.write_output(:valve, :ch1, 1)
EtherCAT.deactivate()
EtherCAT.stop()Dynamic PREOP Configuration
EtherCAT.start(
interface: "eth0",
domains: [%EtherCAT.Domain.Config{id: :main, cycle_time_us: 1_000}]
)
:ok = EtherCAT.await_running()
:ok =
EtherCAT.configure_slave(
:slave_1,
driver: MyApp.EL1809,
process_data: {:all, :main},
target_state: :op
)
:ok = EtherCAT.activate()
:ok = EtherCAT.await_operational()Sub-modules
EtherCAT.Master, EtherCAT.Slave, EtherCAT.Domain, EtherCAT.DC,
EtherCAT.Bus — public runtime boundaries plus direct frame transactions.
Summary
Types
Cycle health reported by domain_info/1.
Domain freshness snapshot reported in domain_info/1.
Detailed snapshot returned by domain_info/1.
Runtime state reported in domain_info/1.
Compact domain summary returned by domains/0.
Local wrapper errors returned when a synchronous master query cannot complete.
Successful query value wrapped with :ok, or a local master query error.
Public master session states returned by state/0 once the local query
succeeds.
Direction of a registered process-data signal.
AL state reported in slave_info/1.
SyncManager attachment summary reported in slave_info/1.
ESC capability snapshot reported in slave_info/1.
Identity snapshot reported in slave_info/1.
Detailed snapshot returned by slave_info/1.
Signal registration summary reported in slave_info/1.
Compact slave summary returned by slaves/0.
Functions
Start cyclic operation after dynamic PREOP configuration.
Wait for DC lock.
Block until the master reaches operational cyclic runtime, then return :ok.
Block until the master reaches a usable session state, then return :ok.
Return the stable bus server reference for direct frame transactions.
Configure a discovered slave while the session is still in PREOP.
Return a Distributed Clocks status snapshot for the current session.
Leave OP while keeping the session alive.
Return a diagnostic snapshot for a domain.
Return {:ok, [{id, cycle_time_us, pid}]} for all running domains.
Download a CoE SDO value to a slave mailbox object entry.
Return the last terminal startup/runtime failure retained after the master
returned to :idle.
Read the latest decoded input sample for a slave input signal.
Return the current DC reference clock as %{name, station}.
Return a diagnostic snapshot for a slave.
Return {:ok, [%{name:, station:, server:, pid:, fault:}]} for all running
slaves.
Start the master: open the interface, discover slaves, and begin self-driving configuration.
Return the current public session state.
Stop the master: tear the session down completely.
Subscribe to named slave events.
Update the live cycle period of a running domain.
Upload a CoE SDO value from a slave mailbox object entry.
Stage value into a slave output PDO for the next domain cycle.
Types
@type domain_cycle_health() :: :healthy | {:invalid, term()}
Cycle health reported by domain_info/1.
@type domain_freshness_info() :: %{ state: :not_ready | :fresh | :stale, refreshed_at_us: integer() | nil, age_us: non_neg_integer() | nil, stale_after_us: pos_integer() }
Domain freshness snapshot reported in domain_info/1.
@type domain_info() :: %{ id: EtherCAT.Domain.domain_id(), cycle_time_us: pos_integer(), state: domain_runtime_state(), cycle_count: non_neg_integer(), miss_count: non_neg_integer(), total_miss_count: non_neg_integer(), cycle_health: domain_cycle_health(), logical_base: non_neg_integer(), image_size: non_neg_integer(), expected_wkc: non_neg_integer(), freshness: domain_freshness_info(), last_cycle_started_at_us: integer() | nil, last_cycle_completed_at_us: integer() | nil, last_valid_cycle_at_us: integer() | nil, last_invalid_cycle_at_us: integer() | nil, last_invalid_reason: term() | nil }
Detailed snapshot returned by domain_info/1.
@type domain_runtime_state() :: :open | :cycling | :stopped
Runtime state reported in domain_info/1.
@type domain_summary() :: {EtherCAT.Domain.domain_id(), pos_integer(), pid()}
Compact domain summary returned by domains/0.
@type master_query_error() :: {:error, :not_started | :timeout | {:server_exit, term()}}
Local wrapper errors returned when a synchronous master query cannot complete.
These are transport-level API failures, not session lifecycle states.
@type master_query_result(value) :: {:ok, value} | master_query_error()
Successful query value wrapped with :ok, or a local master query error.
@type session_state() ::
:idle
| :discovering
| :awaiting_preop
| :preop_ready
| :deactivated
| :operational
| :activation_blocked
| :recovering
Public master session states returned by state/0 once the local query
succeeds.
@type signal_direction() :: :input | :output
Direction of a registered process-data signal.
@type slave_al_state() :: :init | :preop | :safeop | :op
AL state reported in slave_info/1.
@type slave_attachment_summary() :: %{ domain: EtherCAT.Domain.domain_id(), sm_index: non_neg_integer(), direction: signal_direction(), logical_address: non_neg_integer() | nil, sm_size: non_neg_integer() | nil, signal_count: non_neg_integer(), signals: [atom()] }
SyncManager attachment summary reported in slave_info/1.
@type slave_esc_info() :: %{ fmmu_count: non_neg_integer(), sm_count: non_neg_integer() }
ESC capability snapshot reported in slave_info/1.
@type slave_identity() :: %{ vendor_id: non_neg_integer(), product_code: non_neg_integer(), revision: non_neg_integer(), serial_number: non_neg_integer() }
Identity snapshot reported in slave_info/1.
@type slave_info() :: %{ name: atom(), station: non_neg_integer(), al_state: slave_al_state(), identity: slave_identity() | nil, esc: slave_esc_info() | nil, driver: module() | nil, coe: boolean(), available_fmmus: non_neg_integer() | nil, used_fmmus: non_neg_integer(), attachments: [slave_attachment_summary()], pdo_health: map(), signals: [slave_signal_summary()], configuration_error: term() | nil }
Detailed snapshot returned by slave_info/1.
@type slave_signal_summary() :: %{ name: atom(), domain: EtherCAT.Domain.domain_id(), direction: signal_direction(), sm_index: non_neg_integer(), bit_offset: non_neg_integer(), bit_size: pos_integer() }
Signal registration summary reported in slave_info/1.
@type slave_summary() :: %{ name: atom(), station: non_neg_integer(), server: :gen_statem.server_ref(), pid: pid() | nil, fault: term() | nil }
Compact slave summary returned by slaves/0.
Functions
@spec activate() :: :ok | {:error, term()}
Start cyclic operation after dynamic PREOP configuration.
This starts the DC runtime, starts all domains cycling, and advances all slaves
whose target_state is :op.
@spec await_dc_locked(timeout_ms :: pos_integer()) :: :ok | {:error, term()}
Wait for DC lock.
Returns :ok once the active DC runtime reports :locked.
@spec await_operational(timeout_ms :: pos_integer()) :: :ok | {:error, term()}
Block until the master reaches operational cyclic runtime, then return :ok.
This is stricter than await_running/1: :preop_ready and :deactivated
are not enough.
@spec await_running(timeout_ms :: pos_integer()) :: :ok | {:error, term()}
Block until the master reaches a usable session state, then return :ok.
Returns {:error, :timeout} if startup does not complete within timeout_ms ms.
Returns {:error, :not_started} if start/1 has not been called.
Returns startup degradation or runtime recovery errors if the session is not
currently usable.
@spec bus() :: master_query_result(EtherCAT.Bus.server() | nil)
Return the stable bus server reference for direct frame transactions.
Returns {:ok, bus} while the session owns a running bus process.
Returns {:ok, nil} if the master process exists but the bus subsystem is
not currently running, such as after the session has settled back to :idle.
Returns {:error, :not_started} if the master does not exist,
{:error, :timeout} if the local master call itself times out,
and {:error, {:server_exit, reason}} if the local master dies mid-call.
@spec configure_slave(atom(), keyword() | EtherCAT.Slave.Config.t()) :: :ok | {:error, term()}
Configure a discovered slave while the session is still in PREOP.
This is the dynamic counterpart to providing %EtherCAT.Slave.Config{} entries
up front in start/1.
@spec dc_status() :: master_query_result(EtherCAT.DC.Status.t())
Return a Distributed Clocks status snapshot for the current session.
Returns {:ok, status} on success.
Returns {:error, :not_started} if the master process does not exist.
Returns {:error, :timeout} if the local master call itself times out.
@spec deactivate(:safeop | :preop) :: :ok | {:error, term()}
Leave OP while keeping the session alive.
deactivate/0 settles the runtime in SAFEOP by default. Use
deactivate(:preop) when you need to re-enter PREOP for reconfiguration.
@spec domain_info(atom()) :: {:ok, domain_info()} | {:error, :not_found | :timeout | {:server_exit, term()}}
Return a diagnostic snapshot for a domain.
Keys:
:id— domain atom identifier:cycle_time_us— current cycle period in microseconds:state—:open | :cycling | :stopped:cycle_count— successful LRW cycles since last start:miss_count— consecutive missed cycles (resets on success):total_miss_count— cumulative missed cycles since last start:logical_base— current LRW logical start address for this domain image:image_size— PDO image size in bytes:expected_wkc— expected working counter for a healthy bus:freshness— cached-input freshness window derived from the domain cycle time
Example
iex> EtherCAT.domain_info(:main)
{:ok, %{
id: :main,
cycle_time_us: 1_000,
state: :cycling,
cycle_count: 12345,
miss_count: 0,
total_miss_count: 2,
logical_base: 0,
image_size: 4,
expected_wkc: 3,
freshness: %{state: :fresh, refreshed_at_us: 1_234_000, age_us: 250, stale_after_us: 3_000}
}}
@spec domains() :: master_query_result([domain_summary()])
Return {:ok, [{id, cycle_time_us, pid}]} for all running domains.
Returns {:error, :not_started} if the master does not exist and
{:error, :timeout} if the local master call itself times out.
@spec download_sdo(atom(), non_neg_integer(), non_neg_integer(), binary()) :: :ok | {:error, term()}
Download a CoE SDO value to a slave mailbox object entry.
This is a blocking acyclic mailbox transfer and is only valid once the slave mailbox is configured, typically from PREOP onward.
@spec last_failure() :: master_query_result(map() | nil)
Return the last terminal startup/runtime failure retained after the master
returned to :idle.
Returns {:ok, failure} on success, where failure may be nil.
Returns {:error, :not_started} if the master does not exist and
{:error, :timeout} if the local master call itself times out.
Read the latest decoded input sample for a slave input signal.
Returns {value, refreshed_at_us} where refreshed_at_us is the last valid
domain refresh time for the cached process-image sample, not a hardware-edge
timestamp.
Returns {:error, :not_ready} until the first domain cycle completes and
{:error, {:stale, details}} once the cached sample is older than the
domain freshness window.
@spec reference_clock() :: {:ok, %{name: atom() | nil, station: non_neg_integer()}} | {:error, term()}
Return the current DC reference clock as %{name, station}.
@spec slave_info(atom()) :: {:ok, slave_info()} | {:error, :not_found | :timeout | {:server_exit, term()}}
Return a diagnostic snapshot for a slave.
Keys:
:name— slave atom name:station— assigned bus station address:al_state— current ESM state::init | :preop | :safeop | :op:identity—%{vendor_id, product_code, revision, serial_number}from SII, ornil:esc—%{fmmu_count, sm_count}from ESC base registers, ornil:driver— driver module in use:coe—trueif the slave has a mailbox (CoE-capable):available_fmmus— FMMUs supported by the ESC, ornil:used_fmmus— count of active{domain, SyncManager}attachments:attachments— list of%{domain, sm_index, direction, logical_address, sm_size, signal_count, signals}:signals— list of%{name, domain, direction, bit_offset, bit_size}for registered signals:configuration_error— last configuration failure term, ornilCommon values are structured tuples such as{:mailbox_config_failed, index, subindex, reason}.
Example
iex> EtherCAT.slave_info(:sensor)
{:ok, %{
name: :sensor,
station: 0x1001,
al_state: :op,
identity: %{vendor_id: 0x2, product_code: 0x07111389, revision: 0x00190000, serial_number: 0},
esc: %{fmmu_count: 3, sm_count: 4},
driver: MyApp.EL1809,
coe: false,
available_fmmus: 3,
used_fmmus: 1,
attachments: [
%{domain: :main, sm_index: 3, direction: :input, logical_address: 0x0000, sm_size: 2, signal_count: 2, signals: [:ch1, :ch2]}
],
pdo_health: %{state: :fresh, domains: [%{id: :main, state: :fresh, refreshed_at_us: 1_234_000, age_us: 250, stale_after_us: 3_000}]},
signals: [
%{name: :ch1, domain: :main, direction: :input, bit_offset: 0, bit_size: 1},
...
],
configuration_error: nil
}}
@spec slaves() :: master_query_result([slave_summary()])
Return {:ok, [%{name:, station:, server:, pid:, fault:}]} for all running
slaves.
Returns {:error, :not_started} if the master does not exist and
{:error, :timeout} if the local master call itself times out.
Start the master: open the interface, discover slaves, and begin self-driving configuration.
Returns :ok once discovery has started. Call await_running/1 to block
until startup finishes. If the master falls back to :idle, inspect
last_failure/0 for the retained reason. For dynamic PREOP workflows, call
activate/0 afterwards to enter cyclic operation.
Options:
:interface(required) — network interface, e.g."eth0":domains— list of%EtherCAT.Domain.Config{}structs describing domain intent. High-level domain configs do not take:logical_base; the master allocates logical windows automatically.:slaves— list of%EtherCAT.Slave.Config{}structs (position matters — station address =base_station + index).nilentries are rejected; use%EtherCAT.Slave.Config{name: :coupler}for unnamed couplers. If omitted or empty, one default slave process is started per discovered station and held in:preopfor dynamic configuration.:base_station— first station address, default0x1000:dc—%EtherCAT.DC.Config{}for master-wide Distributed Clocks, ornilto disable DC.await_lock?gates startup activation;lock_policycontrols the runtime reaction to DC lock loss.:frame_timeout_ms— optional fixed bus frame response timeout in ms (otherwise auto-tuned from slave count and cycle time):scan_poll_ms— optional discovery poll interval in ms:scan_stable_ms— optional identical-count stability window in ms before startup begins
@spec state() :: master_query_result(session_state())
Return the current public session state.
Values:
:idle:discovering:awaiting_preop:preop_ready:deactivated— session is live but intentionally settled below OP, typically SAFEOP:operational— cyclic OP path is healthy; inspectslaves/0for non-critical per-slave faults:activation_blocked— the transition to the desired runtime target is blocked:recovering— runtime fault recovery in progress
Returns {:ok, state} on success.
{:ok, :idle} means the master process exists and the session is idle.
{:error, :not_started} means there is no local master process at all.
Returns {:error, :not_started} if the master does not exist and
{:error, :timeout} if the local master call itself times out.
@spec stop() :: :ok | {:error, :already_stopped | :timeout | {:server_exit, term()}}
Stop the master: tear the session down completely.
Returns {:error, :already_stopped} if not running.
@spec subscribe(atom(), atom(), pid()) :: :ok | {:error, {:not_registered, atom()} | :not_found | :timeout | {:server_exit, term()}}
Subscribe to named slave events.
name may refer to:
- a registered process-data signal, delivered as
{:ethercat, :signal, slave_name, name, value} - a named latch configured through
sync.latches, delivered as{:ethercat, :latch, slave_name, name, timestamp_ns}
Unknown names are rejected with {:error, {:not_registered, name}}.
@spec update_domain_cycle_time(atom(), pos_integer()) :: :ok | {:error, term()}
Update the live cycle period of a running domain.
This changes the Domain runtime directly. The master keeps its initial
domain plan; domains/0 and domain_info/1 reflect the live period owned by
the Domain process.
@spec upload_sdo(atom(), non_neg_integer(), non_neg_integer()) :: {:ok, binary()} | {:error, term()}
Upload a CoE SDO value from a slave mailbox object entry.
This is a blocking acyclic mailbox transfer and is only valid once the slave mailbox is configured, typically from PREOP onward.
Stage value into a slave output PDO for the next domain cycle.
This confirms the value was staged into the master's domain output buffer for the next cycle. It does not prove the slave has already applied the value on hardware.