Re: [PATCH v10 10/14] Revert "crypto: qce - Mark QCE as BROKEN"

[Birk Skyum]

On Tue, 13 Oct 2026 11:57:39 +0200, Bartosz Golaszewski <bartosz.golaszewski@xxxxxxxxxxxxxxxx> wrote:
> The self-tests for this driver have been fixed. The crypto algorithms
> are only registered if the user explicitly requests it with the provided
> module parameter defaulting to false. There's no reason to further keep
> it unbuildable with the BROKEN label.

On X1E80100 the BROKEN label also costs deep sleep in suspend. With
nothing bound to the crypto node, aggre2_noc and mc_virt never reach
sync_state() ("sync_state() pending due to 1dfa000.crypto"), and on a
Lenovo Yoga Slim 7x a deep suspend on 7.3 never counts aosd, cxsd or
ddr in qcom_stats.

I built qcrypto out of tree from 7.3 with 09 applied, which is what
this patch makes possible in tree, and loaded it on 7.3 (with distro
patches, none in QCE, the interconnect drivers or the driver core) and
the crypto node as in mainline, with ALWAYS tags on its paths. It bound
without registering any algorithms, went to runtime suspend with both
crypto paths at 0, and both providers synced. The next deep suspend
counted aosd, cxsd and ddr, with about 13.7 s in them during an 18 s
suspend.

Herbert asked for the series to be split, and the resends since then
don't carry 09 and this one. Both should come back as fixes, for 7.3.y
as well.

Tested-by: Birk Skyum <birk.skyum@xxxxx> # X1E80100, Lenovo Yoga Slim 7x