isci: fix frame received locking
Dan Williams [Fri, 24 Jun 2011 06:44:52 +0000 (23:44 -0700)]
Updates to the frame_rcvd before need to be atomic with respect to when
they are evaluated by libsas.

Signed-off-by: Dan Williams <dan.j.williams@intel.com>

drivers/scsi/isci/phy.c

index 93a401d..c01d762 100644 (file)
@@ -893,6 +893,7 @@ enum sci_status scic_sds_phy_frame_handler(struct scic_sds_phy *sci_phy,
        enum scic_sds_phy_states state = sci_phy->sm.current_state_id;
        struct scic_sds_controller *scic = sci_phy->owning_port->owning_controller;
        enum sci_status result;
+       unsigned long flags;
 
        switch (state) {
        case SCI_PHY_SUB_AWAIT_IAF_UF: {
@@ -911,7 +912,9 @@ enum sci_status scic_sds_phy_frame_handler(struct scic_sds_phy *sci_phy,
                if (iaf.frame_type == 0) {
                        u32 state;
 
+                       spin_lock_irqsave(&iphy->sas_phy.frame_rcvd_lock, flags);
                        memcpy(&iphy->frame_rcvd.iaf, &iaf, sizeof(iaf));
+                       spin_unlock_irqrestore(&iphy->sas_phy.frame_rcvd_lock, flags);
                        if (iaf.smp_tport) {
                                /* We got the IAF for an expander PHY go to the final
                                 * state since there are no power requirements for
@@ -954,9 +957,11 @@ enum sci_status scic_sds_phy_frame_handler(struct scic_sds_phy *sci_phy,
                                                                      frame_index,
                                                                      (void **)&fis_frame_data);
 
+                       spin_lock_irqsave(&iphy->sas_phy.frame_rcvd_lock, flags);
                        scic_sds_controller_copy_sata_response(&iphy->frame_rcvd.fis,
                                                               frame_header,
                                                               fis_frame_data);
+                       spin_unlock_irqrestore(&iphy->sas_phy.frame_rcvd_lock, flags);
 
                        /* got IAF we can now go to the await spinup semaphore state */
                        sci_change_state(&sci_phy->sm, SCI_PHY_SUB_FINAL);