Skip to content

Commit

Permalink
scsi: qla2xxx: Add cleanup for PCI EEH recovery
Browse files Browse the repository at this point in the history 
During EEH error recovery testing it was discovered that driver's reset()
callback partially frees resources used by driver, leaving some stale
memory.  After reset() is done and when resume() callback in driver uses
old data which results into error leaving adapter disabled due to PCIe
error.

This patch does cleanup for EEH recovery code path and prevents adapter
from getting disabled.

Signed-off-by: Quinn Tran <qutran@marvell.com>
Signed-off-by: Himanshu Madhani <hmadhani@marvell.com>
Reviewed-by: Ewan D. Milne <emilne@redhat.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
  • Loading branch information
@martinkpetersen
Quinn Tran authored and martinkpetersen committed May 14, 2019
1 parent d4023db commit 5386a4e
Showing 1 changed file with 82 additions and 139 deletions.
221 changes: 82 additions & 139 deletions drivers/scsi/qla2xxx/qla_os.c
View file
Original file line number Diff line number Diff line change
Expand Up @@ -6826,6 +6826,78 @@ qla2x00_release_firmware(void)
mutex_unlock(&qla_fw_lock);
}

static void qla_pci_error_cleanup(scsi_qla_host_t *vha)
{
struct qla_hw_data *ha = vha->hw;
scsi_qla_host_t *base_vha = pci_get_drvdata(ha->pdev);
struct qla_qpair *qpair = NULL;
struct scsi_qla_host *vp;
fc_port_t *fcport;
int i;
unsigned long flags;

ha->chip_reset++;

ha->base_qpair->chip_reset = ha->chip_reset;
for (i = 0; i < ha->max_qpairs; i++) {
if (ha->queue_pair_map[i])
ha->queue_pair_map[i]->chip_reset =
ha->base_qpair->chip_reset;
}

/* purge MBox commands */
if (atomic_read(&ha->num_pend_mbx_stage3)) {
clear_bit(MBX_INTR_WAIT, &ha->mbx_cmd_flags);
complete(&ha->mbx_intr_comp);
}

i = 0;

while (atomic_read(&ha->num_pend_mbx_stage3) ||
atomic_read(&ha->num_pend_mbx_stage2) ||
atomic_read(&ha->num_pend_mbx_stage1)) {
msleep(20);
i++;
if (i > 50)
break;
}

ha->flags.purge_mbox = 0;

mutex_lock(&ha->mq_lock);
list_for_each_entry(qpair, &base_vha->qp_list, qp_list_elem)
qpair->online = 0;
mutex_unlock(&ha->mq_lock);

qla2x00_mark_all_devices_lost(vha, 0);

spin_lock_irqsave(&ha->vport_slock, flags);
list_for_each_entry(vp, &ha->vp_list, list) {
atomic_inc(&vp->vref_count);
spin_unlock_irqrestore(&ha->vport_slock, flags);
qla2x00_mark_all_devices_lost(vp, 0);
spin_lock_irqsave(&ha->vport_slock, flags);
atomic_dec(&vp->vref_count);
}
spin_unlock_irqrestore(&ha->vport_slock, flags);

/* Clear all async request states across all VPs. */
list_for_each_entry(fcport, &vha->vp_fcports, list)
fcport->flags &= ~(FCF_LOGIN_NEEDED | FCF_ASYNC_SENT);

spin_lock_irqsave(&ha->vport_slock, flags);
list_for_each_entry(vp, &ha->vp_list, list) {
atomic_inc(&vp->vref_count);
spin_unlock_irqrestore(&ha->vport_slock, flags);
list_for_each_entry(fcport, &vp->vp_fcports, list)
fcport->flags &= ~(FCF_LOGIN_NEEDED | FCF_ASYNC_SENT);
spin_lock_irqsave(&ha->vport_slock, flags);
atomic_dec(&vp->vref_count);
}
spin_unlock_irqrestore(&ha->vport_slock, flags);
}


static pci_ers_result_t
qla2xxx_pci_error_detected(struct pci_dev *pdev, pci_channel_state_t state)
{
Expand All @@ -6851,20 +6923,7 @@ qla2xxx_pci_error_detected(struct pci_dev *pdev, pci_channel_state_t state)
return PCI_ERS_RESULT_CAN_RECOVER;
case pci_channel_io_frozen:
ha->flags.eeh_busy = 1;
/* For ISP82XX complete any pending mailbox cmd */
if (IS_QLA82XX(ha)) {
ha->flags.isp82xx_fw_hung = 1;
ql_dbg(ql_dbg_aer, vha, 0x9001, "Pci channel io frozen\n");
qla82xx_clear_pending_mbx(vha);
}
qla2x00_free_irqs(vha);
pci_disable_device(pdev);
/* Return back all IOs */
qla2x00_abort_all_cmds(vha, DID_RESET << 16);
if (ql2xmqsupport || ql2xnvmeenable) {
set_bit(QPAIR_ONLINE_CHECK_NEEDED, &vha->dpc_flags);
qla2xxx_wake_dpc(vha);
}
qla_pci_error_cleanup(vha);
return PCI_ERS_RESULT_NEED_RESET;
case pci_channel_io_perm_failure:
ha->flags.pci_channel_io_perm_failure = 1;
Expand Down Expand Up @@ -6918,122 +6977,14 @@ qla2xxx_pci_mmio_enabled(struct pci_dev *pdev)
return PCI_ERS_RESULT_RECOVERED;
}

static uint32_t
qla82xx_error_recovery(scsi_qla_host_t *base_vha)
{
uint32_t rval = QLA_FUNCTION_FAILED;
uint32_t drv_active = 0;
struct qla_hw_data *ha = base_vha->hw;
int fn;
struct pci_dev *other_pdev = NULL;

ql_dbg(ql_dbg_aer, base_vha, 0x9006,
"Entered %s.\n", __func__);

set_bit(ABORT_ISP_ACTIVE, &base_vha->dpc_flags);

if (base_vha->flags.online) {
/* Abort all outstanding commands,
* so as to be requeued later */
qla2x00_abort_isp_cleanup(base_vha);
}


fn = PCI_FUNC(ha->pdev->devfn);
while (fn > 0) {
fn--;
ql_dbg(ql_dbg_aer, base_vha, 0x9007,
"Finding pci device at function = 0x%x.\n", fn);
other_pdev =
pci_get_domain_bus_and_slot(pci_domain_nr(ha->pdev->bus),
ha->pdev->bus->number, PCI_DEVFN(PCI_SLOT(ha->pdev->devfn),
fn));

if (!other_pdev)
continue;
if (atomic_read(&other_pdev->enable_cnt)) {
ql_dbg(ql_dbg_aer, base_vha, 0x9008,
"Found PCI func available and enable at 0x%x.\n",
fn);
pci_dev_put(other_pdev);
break;
}
pci_dev_put(other_pdev);
}

if (!fn) {
/* Reset owner */
ql_dbg(ql_dbg_aer, base_vha, 0x9009,
"This devfn is reset owner = 0x%x.\n",
ha->pdev->devfn);
qla82xx_idc_lock(ha);

qla82xx_wr_32(ha, QLA82XX_CRB_DEV_STATE,
QLA8XXX_DEV_INITIALIZING);

qla82xx_wr_32(ha, QLA82XX_CRB_DRV_IDC_VERSION,
QLA82XX_IDC_VERSION);

drv_active = qla82xx_rd_32(ha, QLA82XX_CRB_DRV_ACTIVE);
ql_dbg(ql_dbg_aer, base_vha, 0x900a,
"drv_active = 0x%x.\n", drv_active);

qla82xx_idc_unlock(ha);
/* Reset if device is not already reset
* drv_active would be 0 if a reset has already been done
*/
if (drv_active)
rval = qla82xx_start_firmware(base_vha);
else
rval = QLA_SUCCESS;
qla82xx_idc_lock(ha);

if (rval != QLA_SUCCESS) {
ql_log(ql_log_info, base_vha, 0x900b,
"HW State: FAILED.\n");
qla82xx_clear_drv_active(ha);
qla82xx_wr_32(ha, QLA82XX_CRB_DEV_STATE,
QLA8XXX_DEV_FAILED);
} else {
ql_log(ql_log_info, base_vha, 0x900c,
"HW State: READY.\n");
qla82xx_wr_32(ha, QLA82XX_CRB_DEV_STATE,
QLA8XXX_DEV_READY);
qla82xx_idc_unlock(ha);
ha->flags.isp82xx_fw_hung = 0;
rval = qla82xx_restart_isp(base_vha);
qla82xx_idc_lock(ha);
/* Clear driver state register */
qla82xx_wr_32(ha, QLA82XX_CRB_DRV_STATE, 0);
qla82xx_set_drv_active(base_vha);
}
qla82xx_idc_unlock(ha);
} else {
ql_dbg(ql_dbg_aer, base_vha, 0x900d,
"This devfn is not reset owner = 0x%x.\n",
ha->pdev->devfn);
if ((qla82xx_rd_32(ha, QLA82XX_CRB_DEV_STATE) ==
QLA8XXX_DEV_READY)) {
ha->flags.isp82xx_fw_hung = 0;
rval = qla82xx_restart_isp(base_vha);
qla82xx_idc_lock(ha);
qla82xx_set_drv_active(base_vha);
qla82xx_idc_unlock(ha);
}
}
clear_bit(ABORT_ISP_ACTIVE, &base_vha->dpc_flags);

return rval;
}

static pci_ers_result_t
qla2xxx_pci_slot_reset(struct pci_dev *pdev)
{
pci_ers_result_t ret = PCI_ERS_RESULT_DISCONNECT;
scsi_qla_host_t *base_vha = pci_get_drvdata(pdev);
struct qla_hw_data *ha = base_vha->hw;
struct rsp_que *rsp;
int rc, retries = 10;
int rc;
struct qla_qpair *qpair = NULL;

ql_dbg(ql_dbg_aer, base_vha, 0x9004,
"Slot Reset.\n");
Expand Down Expand Up @@ -7062,24 +7013,16 @@ qla2xxx_pci_slot_reset(struct pci_dev *pdev)
goto exit_slot_reset;
}

rsp = ha->rsp_q_map[0];
if (qla2x00_request_irqs(ha, rsp))
goto exit_slot_reset;

if (ha->isp_ops->pci_config(base_vha))
goto exit_slot_reset;

if (IS_QLA82XX(ha)) {
if (qla82xx_error_recovery(base_vha) == QLA_SUCCESS) {
ret = PCI_ERS_RESULT_RECOVERED;
goto exit_slot_reset;
} else
goto exit_slot_reset;
}

while (ha->flags.mbox_busy && retries--)
msleep(1000);
mutex_lock(&ha->mq_lock);
list_for_each_entry(qpair, &base_vha->qp_list, qp_list_elem)
qpair->online = 1;
mutex_unlock(&ha->mq_lock);

base_vha->flags.online = 1;
set_bit(ABORT_ISP_ACTIVE, &base_vha->dpc_flags);
if (ha->isp_ops->abort_isp(base_vha) == QLA_SUCCESS)
ret = PCI_ERS_RESULT_RECOVERED;
Expand All @@ -7103,13 +7046,13 @@ qla2xxx_pci_resume(struct pci_dev *pdev)
ql_dbg(ql_dbg_aer, base_vha, 0x900f,
"pci_resume.\n");

ha->flags.eeh_busy = 0;

ret = qla2x00_wait_for_hba_online(base_vha);
if (ret != QLA_SUCCESS) {
ql_log(ql_log_fatal, base_vha, 0x9002,
"The device failed to resume I/O from slot/link_reset.\n");
}

ha->flags.eeh_busy = 0;
}

static void
Expand Down

0 comments on commit 5386a4e

Please sign in to comment.