Document adapter->ctrl_pf_lock as the outer lifetime lock for TX clock state and PTP hardware semaphore operations. This makes the existing ordering constraints visible to future callers and helps avoid nesting the rwsem below a DPLL lock.
Signed-off-by: Sergey Temerkhanov <[email protected]> Reviewed-by: Aleksandr Loktionov <[email protected]> --- drivers/net/ethernet/intel/ice/ice_dpll.h | 18 +++++++++++++++--- 1 file changed, 15 insertions(+), 3 deletions(-) diff --git a/drivers/net/ethernet/intel/ice/ice_dpll.h b/drivers/net/ethernet/intel/ice/ice_dpll.h index 641b02b0296a..30767f209cd2 100644 --- a/drivers/net/ethernet/intel/ice/ice_dpll.h +++ b/drivers/net/ethernet/intel/ice/ice_dpll.h @@ -164,13 +164,25 @@ struct ice_dpll { * Locking: * Acquisition order (top to bottom): * - * txclk_notify_rwsem (read) - * -> pf->dplls.lock - * -> ctrl_pf->dplls.lock + * adapter->ctrl_pf_lock (read) + * -> txclk_notify_rwsem (read) + * -> pf->dplls.lock + * -> ctrl_pf->dplls.lock * * health_notify_rwsem (read) * -> pf->dplls.lock * + * PTP hardware semaphore path: + * + * adapter->ps_lock + * -> adapter->ctrl_pf_lock (read) + * -> PFTSYN_SEM + * + * - ice_ptp_port_phy_restart() runs with adapter->ps_lock held by its + * caller and reaches ice_ptp_lock() via ice_start_phy_timer_eth56g() + * or ice_start_phy_timer_e82x(), so ctrl_pf_lock is not the outermost + * lock on that path. Never take adapter->ps_lock with ctrl_pf_lock + * held. * - @lock serializes all DPLL state mutations on this PF. When the * controlling PF's lock must also be taken (e.g. updating the shared * tx_refclks usage map), acquire pf->dplls.lock first, then -- 2.53.0
