Skip to content
Porthole organization iconPorthole

A (created) MVM carries call audio: what matters is whether the modem had a call, not the joined flag

scope: soc:msm8998 · severity: finding · confidence: proven · subsystem: audio

The question — the driver logs voice session up: mvm=0x0026 (created) and the earlier notes say an audible call is the one that reads (joined). Is a (created) MVM a regression to chase before trusting anything else in the run?

The answer — no. On 2026-08-27 a carrier call came up with

voice session up: mvm=0x0026 (created) cvs=0x0100 (created) cvp=0x0100,
                  rx=0x1006 tx=0x4001

and it was audible in both directions – the tester heard the carrier, and the carrier heard the microphone. The (joined) label is not the thing that decides.

What decides is what the older trap actually measured: whether the modem had a call of its own when the session was made. A session forced open while the modem is idle gets a handle from a different range (0x0020) and is silent for the rest of the boot; a session made during a real call gets 0x0026 and carries audio, whether the ADSP answered the CREATE with EALREADY (joined) or allocated it (created).

So the useful check on a call result is the handle and the modem’s state, not the flag. taimen-voicehold gating on an ACTIVE call is still exactly right, and is why this call got 0x0026 at all.

What this rules out —

  • “(created) means no audio, stop and reboot.” It does not. Two hours were nearly spent treating a working call as a regression on the strength of that word.
  • “The flag is still lying, as it did before patch 0177.” No: 0177 made only a CREATE able to set it, and it is accurate here. A CREATE that really did allocate a fresh session reports (created), correctly – the session is simply not silent for that reason.

How it was established — one live carrier call, both directions confirmed by ear (the tester’s, and the far end’s report). The trace either side of it is in logs/2026-08-27-slow/ and the session line above is from dmesg. What would overturn it: a 0x0026 (created) call that IS silent, which would mean the handle range is not the discriminator either.