eth_tsn_tas_demo Directory Reference

Directory dependency graph for eth_tsn_tas_demo:

Directories
	src

Detailed Description

Programs the RA8D2 Ethernet Agent (ETHA) time-sensitive-networking shapers that no other example referenced (recon gap #134): the time-aware shaper (TAS, 802.1Qbv scheduled traffic) and the credit-based shaper (CBS, 802.1Qav).

ra8_tsn is the temperature sensor, not TSN networking

Recon #134 named ra8_tsn as the time-sensitive-networking driver, but libs/ra8_hal/inc/ra8_tsn.h is the on-die temperature sensor (demonstrated by adc_diag_tsn_demo). The real TSN networking surface on this part is the ETHA shaper block, which is what this example drives.

A TAS entry on this part carries a single gate-state bit for the queue whose TAS RAM block holds it (HUM Table 32.6 p 1691) – not an interleaved per-class gate vector. The app programs a short gate list on the PTP/control descriptor queue, one window open and one shut, then reads every entry back out of TAS RAM and compares.

TAS times its gate-control list against the gPTP counter, so the app starts that time base for real first and only then runs ra8_etha_init in CONFIG mode – the only mode in which the shaper registers are writable. Before #498 the time base was never started at all: the app programmed a gate-control list against a counter fixed at zero.

What the verdict proves, and what it does not

It requires three things:

1. A real hardware assertion. The 78-bit gPTP counter is sampled either side of a SysTick-timed window and must have advanced by that interval to within 10 %. If the time base is not running, the app fails.
2. A second real hardware assertion. Every TAS entry programmed is read back through EATASGR / EATASGRR and must match the gate state and gate time that were written. Added with #539, when the driver turned out to be writing gate states into EATASGL0 – whose field is the entry ADDRESS – while returning k_ra8_ok on every call.
3. That every shaper call returned k_ra8_ok. This half proves only that the arguments were accepted and the writes were issued. ETHA stays in CONFIG mode, so no frame is ever transmitted and nothing about shaped egress is measured.

The shaper values (window duration, cycle time, CBS increment and upper limit) are illustrative round numbers; the point is exercising the programming path, not a specific traffic profile. HUM references: Ch 32 "Ethernet Agent (ETHA)" – EATASC / EATASGL (TAS), EACAEC / EACAIVC / EACAULC (CBS). The example adds no raw MMIO of its own.

Blocked on

Nothing off-target models the ETHA shapers or the GPTP timer – both windows fall to the sparse config-reflect fallback, where a counter cannot advance – so the app would correctly report failure under emulation. The EK-RA8D2 Ethernet wire is also marginal (#21).

A measurement peer is provisioned on the bench (linuxptp on the HIL Pi's built-in Ethernet port, which carries a real PTP hardware clock), and it still does not unblock this app. #292 recorded two firmware blockers; one is closed and one is not. Closed: the schedule reference is now started, because #498 rewrote ra8_eth_gptp_init off an invented register window and onto the real HUM Ch 35 map. Open: nothing is transmitted – the port never leaves CONFIG mode, no queue is opened, and no frame is queued, and a shaper with no egress produces nothing to measure. Promotion needs this app extended to actually transmit, which is not a bench change.