On 6/15/26 09:11, Borah, Chaitanya Kumar wrote:
Hello Mathias,
On 6/3/2026 2:41 PM, Mathias Nyman wrote:
DbC can be enabled and disabled via sysfs, serialize those
with a mutex to make sure everything is done in the correct
order.
remove xhci_do_dbc_stop() and integrate the register write and
dbc->state setting into xhci_do_stop()
This patch seems to cause a regression in our CI. [1]
We could not revert the patch because of merge conflicts but resetting to the
parent commit seems to heal the issue.
Could you please check why the patch causes this regression and provide
a fix if necessary?
Ah, I see, deadlock when enabling DbC via sysfs on runtime suspended xhci
Appears this is an issue mid series in intel-next,
The last patch that enabled runtime pm for DbC should also fix this issue.
The DbC runtime pm support patch was dropped last minute from the upstream
series.
I'll send it as a RFT, can you check it solve the issue in your CI
Details of deadlock issue while mid series:
dbc_store() { // enable dbc via syfs
mutex_lock(&dbc->enable_mutex);
xhci_dbc_start(dbc);
pm_runtime_get_sync(dbc->dev);
xhci_dbc_resume()
mutex_lock(&dbc->enable_mutex);
Thanks
Mathias
