Re: [PATCH v3 0/3] clk: qcom: gcc-glymur: Tie the CX power domain to controller

[Abel Vesa]

On 26-08-25 18:37:53, Abel Vesa wrote:
> On 26-08-17 12:10:06, Konrad Dybcio wrote:
> > On 8/14/26 8:30 AM, Pankaj Patil wrote:
> > > On 7/15/2026 6:59 PM, Abel Vesa wrote:
> > >> The suggestion of having the CX power domain tied up to the GCC
> > >> controller started here:
> > >>
> > >> https://lore.kernel.org/lkml/fe210f0b-692b-4c45-afc4-fc2bc5e57854@xxxxxxxxxxxxxxxx/
> > >>
> > >> and then, for Glymur, it has been brought up here as well:
> > >>
> > >> https://lore.kernel.org/all/0248dc51-1036-426c-b1de-dbc71696e2c1@xxxxxxxxxxxxxxxx/
> > >>
> > >> These plus a discussion off-list led to this patchset being done.
> > >>
> > > 
> > > This series prevents cx collapse during suspend on recent linux-next tag: next-20260806
> > > Reproducible on latest tag next-20260813 as well
> > 
> > Yes and the ""fix"" is to disable the GPU nodes because the custom
> > runtime pm dance done in the gxclkctl driver keeps a vote online
> > 
> 
> Actually, what happens is that a sync_state dependency is never met.
> 
> GMU doesn't sync state because there is no driver to bind to. No driver,
> no sync_state.
> 
> Then GPUCC doesn't sync_state because of GMU.
> 
> Then GCC doesn't sync_state because of GPUCC.
> 
> Forcing sync_state with timeout, allows the GCC to drop the vote on CX.
> 
> This is not only a problem for Glymur. I have the same issue on Hamoa
> (XPS 13). I'm almost sure this is going to be a problem for all
> platforms.
> 
> I'm looking into possible ways to get the GMU to sync_state.

This does exactly that:

https://lore.kernel.org/all/20260513-gmu-sync-state-fix-v1-0-6e33e6aa9b4f@xxxxxxxxxxxxxxxx/#t