D762 - An interrupting release posts the driver's end-of-pipe event

Status: decided Date: 2026-10-08 known_by: measured (event 0, INT_SEL 2, context ids 0 and 0x101); assumed beyond those

sceAgcDriverAddEqEvent(queue, id, udata) registers the driver's end-of-pipe event id against a guest event queue. Every RELEASE_MEM that raises an interrupt posts one event to each queue registered this way. That is a release whose INT_SEL (bits 26:24 of its selector dword, Mesa sid.h EOP_INT_SEL) is not zero, and a flip's release is one too. The event is what the console delivered (obSCEne REQ-20261008T1200Z-eo03, sweep 20261008-142306, kevent-32b rows):

• ident the registered id
• filter -14 (EVFILT_AGC)
• flags 0x20
• fflags 0
• data the release's context id shifted left sixteen, over 0xff00
• udata the registration's

A release with INT_SEL 0 raises no interrupt and posts nothing. The command processor no longer reports one to the display either.

Measured: event 0 registered, INT_SEL 2, context ids 0 and 0x101, each delivering exactly the bytes above once the release retired.

Assumed: a registered id other than 0 is the event's ident; INT_SEL 1 and 3 post as 2 does; a flip's release posts as well; a context id wider than sixteen bits keeps its upper bits above 0xff00. The one context measured with bit 27 set stalled the probe's queue (-eo02), so what it posts is unknown.

Why: PPSA03416 and PPSA02664 register this event and block on the queue it is posted to. Before, orbistoun answered the registration and posted nothing.