On 10/19/2026 3:51 AM, Aaron Ma wrote:
> ice_suspend() stops the main service task but leaves PTP work active.
> The PTP kworker is not freezable, so it can continue issuing MMIO
> accesses after the PCI function has entered D3hot. Both the periodic
> work updating the cached PHC time and the E822 offset calibration work,
> which reschedules itself while calibration is pending, are affected.
>
> Synchronously cancel both works when PTP is ready, after disabling the
> VSIs and before shutting down the control queues.
>
> Fixes: 77a781155a65 ("ice: enable receive hardware timestamping")
> Fixes: a69f1cb62aec ("ice: exit bypass mode once hardware finishes timestamp
> calibration")
> Signed-off-by: Aaron Ma <[email protected]>
> ---
> drivers/net/ethernet/intel/ice/ice_main.c | 6 ++++++
> 1 file changed, 6 insertions(+)
>
> diff --git a/drivers/net/ethernet/intel/ice/ice_main.c
> b/drivers/net/ethernet/intel/ice/ice_main.c
> index ddebb28b2454f..3962910759067 100644
> --- a/drivers/net/ethernet/intel/ice/ice_main.c
> +++ b/drivers/net/ethernet/intel/ice/ice_main.c
> @@ -5450,6 +5450,12 @@ static void ice_prepare_for_shutdown(struct ice_pf *pf)
> /* disable the VSIs and their queues that are not already DOWN */
> ice_pf_dis_all_vsi(pf, false);
>
> + if (test_bit(ICE_FLAG_PTP_SUPPORTED, pf->flags) &&
> + pf->ptp.state == ICE_PTP_READY) {
> + kthread_cancel_delayed_work_sync(&pf->ptp.work);
> + kthread_cancel_delayed_work_sync(&pf->ptp.port.ov_work);
> + }
> +
> ice_for_each_vsi(pf, v)
> if (pf->vsi[v])
> pf->vsi[v]->vsi_num = 0;
I don't think kthread has the same API as workers to fully disable
re-queueing, but at least the periodic work is only requeueud within its
own run or by rebuild flows. ov_work I am not fully sure about, I guess
it depends on if this can race with another thread that would
re-schedule. I believe the PTP teardown also cancel the tasks but may
happen later than needed? But maybe we lack a PTP flow for
prepare_for_shutdown. Hmm.
This is at least better than what we do now so:
Reviewed-by: Jacob Keller <[email protected]>
