ti_edma3 matches ti,edma3, a compatible string no device tree has used since the binding was split. The current binding describes a channel controller (ti,edma3-tpcc) and three separate transfer controllers (ti,edma3-tptc), so the driver never attached on am335x.

Match the channel controller node, and add a small driver for the transfer controllers. They need no programming, but their target modules stay clock gated until something claims them, and a gated transfer controller accepts a request and then moves nothing — indistinguishable from a peripheral that never raised a request.

Also expose the interfaces a driver owning a channel needs: whether the controller attached, whether a transfer completed, whether an event was missed, and how to link PaRAM sets.

This is a prerequisite for offloading the MMC data phase. On its own it attaches the engine and removes three unclaimed device tree nodes from the boot output:

ti_sysc48: <dma@0> compat ti,edma3-tptc (no driver attached)
ti_sysc49: <dma@0> compat ti,edma3-tptc (no driver attached)
ti_sysc50: <dma@0> compat ti,edma3-tptc (no driver attached)

Boots on a BeagleBone Black Rev B3 with the channel controller and all three transfer controllers attaching.

Nothing in-tree builds these files today, since AM335x was removed from armv7 GENERIC in 3416e102c4e9, so compile coverage requires an AM335x kernel config — see D46703.

Full evidence: https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=297800