| Message ID | 1544212602-28085-1-git-send-email-glagolig@gmail.com (mailing list archive) |
|---|---|
| State | Changes Requested |
| Headers | show |
| Series | scsi: qla2xxx: disabling pci error handler early | expand |
On 12/07/2018 01:56 PM, Anatoliy Glagolev wrote: > qla2x00_disable_board_on_pci_error and pcie error handlers may run > in parallel. Specifically, I observed qla2xxx_pci_slot_reset running > at around the same moment as qla2x00_disable_board_on_pci_error. > If scsi_qla_host_t or qla_hw_data structs are removed before an error > handler completes, the handler crashes. > > This patch disables pcie error handling early in > qla2x00_disable_board_on_pci_error and in other paths that remove > those structs. While this may fix this issue for PCIe AER, I think you'll still have the exposure for EEH errors on a Power system, since we don't have the pci_enable_pcie_error_reporting API wired up to do anything, nor do we have much ability to do anything with it since its an attribute of the hardware. Is there a way to fix this that doesn't break Power? Thanks, Brian > > Signed-off-by: Anatoliy Glagolev <glagolig@gmail.com> > --- > drivers/scsi/qla2xxx/qla_os.c | 15 +++++++++------ > 1 file changed, 9 insertions(+), 6 deletions(-) > > diff --git a/drivers/scsi/qla2xxx/qla_os.c b/drivers/scsi/qla2xxx/qla_os.c > index e881fce..b8f277a 100644 > --- a/drivers/scsi/qla2xxx/qla_os.c > +++ b/drivers/scsi/qla2xxx/qla_os.c > @@ -2775,9 +2775,6 @@ static void qla2x00_iocb_work_fn(struct work_struct *work) > return ret; > } > > - /* This may fail but that's ok */ > - pci_enable_pcie_error_reporting(pdev); > - > ha = kzalloc(sizeof(struct qla_hw_data), GFP_KERNEL); > if (!ha) { > ql_log_pci(ql_log_fatal, pdev, 0x0009, > @@ -3039,6 +3036,9 @@ static void qla2x00_iocb_work_fn(struct work_struct *work) > goto probe_hw_failed; > } > > + /* This may fail but that's ok */ > + pci_enable_pcie_error_reporting(pdev); > + > pci_set_drvdata(pdev, base_vha); > set_bit(PFLG_DRIVER_PROBING, &base_vha->pci_flags); > > @@ -3400,6 +3400,8 @@ static void qla2x00_iocb_work_fn(struct work_struct *work) > kthread_stop(t); > } > > + pci_disable_pcie_error_reporting(); > + > qla2x00_free_device(base_vha); > scsi_host_put(base_vha->host); > /* > @@ -3625,6 +3627,8 @@ static void qla2x00_iocb_work_fn(struct work_struct *work) > } > qla2x00_wait_for_hba_ready(base_vha); > > + pci_disable_pcie_error_reporting(pdev); > + > qla2x00_wait_for_sess_deletion(base_vha); > > /* > @@ -3698,8 +3702,6 @@ static void qla2x00_iocb_work_fn(struct work_struct *work) > pci_release_selected_regions(ha->pdev, ha->bars); > kfree(ha); > > - pci_disable_pcie_error_reporting(pdev); > - > pci_disable_device(pdev); > } > > @@ -5826,6 +5828,8 @@ void qla2x00_relogin(struct scsi_qla_host *vha) > return; > } > > + pci_disable_pcie_error_reporting(pdev); > + > qla2x00_wait_for_sess_deletion(base_vha); > > set_bit(UNLOADING, &base_vha->dpc_flags); > @@ -5866,7 +5870,6 @@ void qla2x00_relogin(struct scsi_qla_host *vha) > qla2x00_unmap_iobases(ha); > > pci_release_selected_regions(ha->pdev, ha->bars); > - pci_disable_pcie_error_reporting(pdev); > pci_disable_device(pdev); > > /* >
Thanks, Brian. Great point. Even for AER, it looks like in-flight error handler completion is not guaranteed on pci_disable_pcie_error_reporting call, so the crash is still possible. It looks like we need to maintain per-pci_dev context and keep track of in-flight callbacks to make a clean fix. I will send a new patch. On Fri, Dec 07, 2018 at 04:00:27PM -0600, Brian King wrote: > On 12/07/2018 01:56 PM, Anatoliy Glagolev wrote: > > qla2x00_disable_board_on_pci_error and pcie error handlers may run > > in parallel. Specifically, I observed qla2xxx_pci_slot_reset running > > at around the same moment as qla2x00_disable_board_on_pci_error. > > If scsi_qla_host_t or qla_hw_data structs are removed before an error > > handler completes, the handler crashes. > > > > This patch disables pcie error handling early in > > qla2x00_disable_board_on_pci_error and in other paths that remove > > those structs. > > While this may fix this issue for PCIe AER, I think you'll still have > the exposure for EEH errors on a Power system, since we don't have > the pci_enable_pcie_error_reporting API wired up to do anything, > nor do we have much ability to do anything with it since its an > attribute of the hardware. > > Is there a way to fix this that doesn't break Power? > > Thanks, > > Brian > > > > > Signed-off-by: Anatoliy Glagolev <glagolig@gmail.com> > > --- > > drivers/scsi/qla2xxx/qla_os.c | 15 +++++++++------ > > 1 file changed, 9 insertions(+), 6 deletions(-) > > > > diff --git a/drivers/scsi/qla2xxx/qla_os.c b/drivers/scsi/qla2xxx/qla_os.c > > index e881fce..b8f277a 100644 > > --- a/drivers/scsi/qla2xxx/qla_os.c > > +++ b/drivers/scsi/qla2xxx/qla_os.c > > @@ -2775,9 +2775,6 @@ static void qla2x00_iocb_work_fn(struct work_struct *work) > > return ret; > > } > > > > - /* This may fail but that's ok */ > > - pci_enable_pcie_error_reporting(pdev); > > - > > ha = kzalloc(sizeof(struct qla_hw_data), GFP_KERNEL); > > if (!ha) { > > ql_log_pci(ql_log_fatal, pdev, 0x0009, > > @@ -3039,6 +3036,9 @@ static void qla2x00_iocb_work_fn(struct work_struct *work) > > goto probe_hw_failed; > > } > > > > + /* This may fail but that's ok */ > > + pci_enable_pcie_error_reporting(pdev); > > + > > pci_set_drvdata(pdev, base_vha); > > set_bit(PFLG_DRIVER_PROBING, &base_vha->pci_flags); > > > > @@ -3400,6 +3400,8 @@ static void qla2x00_iocb_work_fn(struct work_struct *work) > > kthread_stop(t); > > } > > > > + pci_disable_pcie_error_reporting(); > > + > > qla2x00_free_device(base_vha); > > scsi_host_put(base_vha->host); > > /* > > @@ -3625,6 +3627,8 @@ static void qla2x00_iocb_work_fn(struct work_struct *work) > > } > > qla2x00_wait_for_hba_ready(base_vha); > > > > + pci_disable_pcie_error_reporting(pdev); > > + > > qla2x00_wait_for_sess_deletion(base_vha); > > > > /* > > @@ -3698,8 +3702,6 @@ static void qla2x00_iocb_work_fn(struct work_struct *work) > > pci_release_selected_regions(ha->pdev, ha->bars); > > kfree(ha); > > > > - pci_disable_pcie_error_reporting(pdev); > > - > > pci_disable_device(pdev); > > } > > > > @@ -5826,6 +5828,8 @@ void qla2x00_relogin(struct scsi_qla_host *vha) > > return; > > } > > > > + pci_disable_pcie_error_reporting(pdev); > > + > > qla2x00_wait_for_sess_deletion(base_vha); > > > > set_bit(UNLOADING, &base_vha->dpc_flags); > > @@ -5866,7 +5870,6 @@ void qla2x00_relogin(struct scsi_qla_host *vha) > > qla2x00_unmap_iobases(ha); > > > > pci_release_selected_regions(ha->pdev, ha->bars); > > - pci_disable_pcie_error_reporting(pdev); > > pci_disable_device(pdev); > > > > /* > > > > > -- > Brian King > Power Linux I/O > IBM Linux Technology Center >
diff --git a/drivers/scsi/qla2xxx/qla_os.c b/drivers/scsi/qla2xxx/qla_os.c index e881fce..b8f277a 100644 --- a/drivers/scsi/qla2xxx/qla_os.c +++ b/drivers/scsi/qla2xxx/qla_os.c @@ -2775,9 +2775,6 @@ static void qla2x00_iocb_work_fn(struct work_struct *work) return ret; } - /* This may fail but that's ok */ - pci_enable_pcie_error_reporting(pdev); - ha = kzalloc(sizeof(struct qla_hw_data), GFP_KERNEL); if (!ha) { ql_log_pci(ql_log_fatal, pdev, 0x0009, @@ -3039,6 +3036,9 @@ static void qla2x00_iocb_work_fn(struct work_struct *work) goto probe_hw_failed; } + /* This may fail but that's ok */ + pci_enable_pcie_error_reporting(pdev); + pci_set_drvdata(pdev, base_vha); set_bit(PFLG_DRIVER_PROBING, &base_vha->pci_flags); @@ -3400,6 +3400,8 @@ static void qla2x00_iocb_work_fn(struct work_struct *work) kthread_stop(t); } + pci_disable_pcie_error_reporting(); + qla2x00_free_device(base_vha); scsi_host_put(base_vha->host); /* @@ -3625,6 +3627,8 @@ static void qla2x00_iocb_work_fn(struct work_struct *work) } qla2x00_wait_for_hba_ready(base_vha); + pci_disable_pcie_error_reporting(pdev); + qla2x00_wait_for_sess_deletion(base_vha); /* @@ -3698,8 +3702,6 @@ static void qla2x00_iocb_work_fn(struct work_struct *work) pci_release_selected_regions(ha->pdev, ha->bars); kfree(ha); - pci_disable_pcie_error_reporting(pdev); - pci_disable_device(pdev); } @@ -5826,6 +5828,8 @@ void qla2x00_relogin(struct scsi_qla_host *vha) return; } + pci_disable_pcie_error_reporting(pdev); + qla2x00_wait_for_sess_deletion(base_vha); set_bit(UNLOADING, &base_vha->dpc_flags); @@ -5866,7 +5870,6 @@ void qla2x00_relogin(struct scsi_qla_host *vha) qla2x00_unmap_iobases(ha); pci_release_selected_regions(ha->pdev, ha->bars); - pci_disable_pcie_error_reporting(pdev); pci_disable_device(pdev); /*
qla2x00_disable_board_on_pci_error and pcie error handlers may run in parallel. Specifically, I observed qla2xxx_pci_slot_reset running at around the same moment as qla2x00_disable_board_on_pci_error. If scsi_qla_host_t or qla_hw_data structs are removed before an error handler completes, the handler crashes. This patch disables pcie error handling early in qla2x00_disable_board_on_pci_error and in other paths that remove those structs. Signed-off-by: Anatoliy Glagolev <glagolig@gmail.com> --- drivers/scsi/qla2xxx/qla_os.c | 15 +++++++++------ 1 file changed, 9 insertions(+), 6 deletions(-)