From 1f08be80bef6f7a3faaa728db836b47ff742f41f Mon Sep 17 00:00:00 2001 From: Sebastian Ott Date: Wed, 5 Sep 2012 14:18:22 +0200 Subject: s390/dasd: move wake_up call Ensure that all work is done when the process waiting for a dasd state change is woken up. With this change it is save to assume that after a userspace triggered state change and a udev settle invocation there are no unexpected users of a dasd device. Acked-by: Stefan Weinhuber Signed-off-by: Sebastian Ott Signed-off-by: Martin Schwidefsky --- drivers/s390/block/dasd.c | 6 +++--- 1 file changed, 3 insertions(+), 3 deletions(-) (limited to 'drivers') diff --git a/drivers/s390/block/dasd.c b/drivers/s390/block/dasd.c index 15370a2c5ff0..6498d15f874f 100644 --- a/drivers/s390/block/dasd.c +++ b/drivers/s390/block/dasd.c @@ -534,11 +534,11 @@ static void dasd_change_state(struct dasd_device *device) if (rc) device->target = device->state; - if (device->state == device->target) - wake_up(&dasd_init_waitq); - /* let user-space know that the device status changed */ kobject_uevent(&device->cdev->dev.kobj, KOBJ_CHANGE); + + if (device->state == device->target) + wake_up(&dasd_init_waitq); } /* -- cgit v1.3-18-gd494 From 43d0be75af8d05654b88d1da494cf292714fbdec Mon Sep 17 00:00:00 2001 From: Sebastian Ott Date: Wed, 5 Sep 2012 14:19:42 +0200 Subject: s390/cio: fix IO subchannel event race If the subchannel event function is called from IRQ context and we observe that the subchannel in question is gone we flag the attached device as not operational and schedule the event function to be called again from process context where the subchannel gets deregistered. However if the subchannel reappeared at the time the event function gets called from process context we would do nothing and leave the device in not operational state. Recognize this case in sch_get_action and trigger reexamination of the subchannel/device. Acked-by: Peter Oberparleiter Signed-off-by: Sebastian Ott Signed-off-by: Martin Schwidefsky --- drivers/s390/cio/device.c | 2 ++ 1 file changed, 2 insertions(+) (limited to 'drivers') diff --git a/drivers/s390/cio/device.c b/drivers/s390/cio/device.c index ed25c8740a9c..e8e1a108cdf8 100644 --- a/drivers/s390/cio/device.c +++ b/drivers/s390/cio/device.c @@ -1426,6 +1426,8 @@ static enum io_sch_action sch_get_action(struct subchannel *sch) return IO_SCH_REPROBE; if (cdev->online) return IO_SCH_VERIFY; + if (cdev->private->state == DEV_STATE_NOT_OPER) + return IO_SCH_UNREG_ATTACH; return IO_SCH_NOP; } -- cgit v1.3-18-gd494 From 3368ba25bf387109b24732c1e270c628f20e345d Mon Sep 17 00:00:00 2001 From: Sebastian Ott Date: Wed, 5 Sep 2012 14:20:41 +0200 Subject: s390/cio: invalidate cdev pointer before deregistration Make sure that the cdev pointer for IO subchannels is set to NULL when we deregister the device (and release its last reference). This will fix a bug were another process operates on an already freed ccw device. Acked-by: Peter Oberparleiter Signed-off-by: Sebastian Ott Signed-off-by: Martin Schwidefsky --- drivers/s390/cio/device.c | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) (limited to 'drivers') diff --git a/drivers/s390/cio/device.c b/drivers/s390/cio/device.c index e8e1a108cdf8..fc916f5d7314 100644 --- a/drivers/s390/cio/device.c +++ b/drivers/s390/cio/device.c @@ -1521,11 +1521,14 @@ static int io_subchannel_sch_event(struct subchannel *sch, int process) goto out; break; case IO_SCH_UNREG_ATTACH: + spin_lock_irqsave(sch->lock, flags); if (cdev->private->flags.resuming) { /* Device will be handled later. */ rc = 0; - goto out; + goto out_unlock; } + sch_set_cdev(sch, NULL); + spin_unlock_irqrestore(sch->lock, flags); /* Unregister ccw device. */ ccw_device_unregister(cdev); break; -- cgit v1.3-18-gd494 From 12d7b1078bb374fc3e2955b9f2815415a66157b6 Mon Sep 17 00:00:00 2001 From: Stefan Haberland Date: Tue, 11 Sep 2012 15:10:58 +0200 Subject: s390/dasd: fix pathgroup race If a new path is available we need to verify the path data. If it is the first path for a device the stop bits are removed after path verification. If a pathgroup is established we need to set system characteristics for the lcu. Therefore I/O has to be started. If the device is stopped the set system characteristics worker may block the path verification worker and the device is blocked. Turn on failfast for set system characteristics CQR to prevent a deadlock with the path verification worker. If a pathgroup is established on a device that is not in use trigger path verification. Maybe we were not informed about a working path. Signed-off-by: Stefan Haberland Reviewed-by: Stefan Weinhuber Signed-off-by: Martin Schwidefsky --- drivers/s390/block/dasd.c | 11 +++++++++++ drivers/s390/block/dasd_eckd.c | 32 +++++++++++++++++++++++++------- 2 files changed, 36 insertions(+), 7 deletions(-) (limited to 'drivers') diff --git a/drivers/s390/block/dasd.c b/drivers/s390/block/dasd.c index 6498d15f874f..0595c763dafd 100644 --- a/drivers/s390/block/dasd.c +++ b/drivers/s390/block/dasd.c @@ -2157,6 +2157,7 @@ static int _dasd_sleep_on(struct dasd_ccw_req *maincqr, int interruptible) test_bit(DASD_CQR_FLAGS_FAILFAST, &cqr->flags) && (!dasd_eer_enabled(device))) { cqr->status = DASD_CQR_FAILED; + cqr->intrc = -EAGAIN; continue; } /* Don't try to start requests if device is stopped */ @@ -3270,6 +3271,16 @@ void dasd_generic_path_event(struct ccw_device *cdev, int *path_event) dasd_schedule_device_bh(device); } if (path_event[chp] & PE_PATHGROUP_ESTABLISHED) { + if (!(device->path_data.opm & eventlpm) && + !(device->path_data.tbvpm & eventlpm)) { + /* + * we can not establish a pathgroup on an + * unavailable path, so trigger a path + * verification first + */ + device->path_data.tbvpm |= eventlpm; + dasd_schedule_device_bh(device); + } DBF_DEV_EVENT(DBF_WARNING, device, "%s", "Pathgroup re-established\n"); if (device->discipline->kick_validate) diff --git a/drivers/s390/block/dasd_eckd.c b/drivers/s390/block/dasd_eckd.c index 2fb2b9ea97ec..c48c72abbefc 100644 --- a/drivers/s390/block/dasd_eckd.c +++ b/drivers/s390/block/dasd_eckd.c @@ -1507,7 +1507,8 @@ static struct dasd_ccw_req *dasd_eckd_build_psf_ssc(struct dasd_device *device, * call might change behaviour of DASD devices. */ static int -dasd_eckd_psf_ssc(struct dasd_device *device, int enable_pav) +dasd_eckd_psf_ssc(struct dasd_device *device, int enable_pav, + unsigned long flags) { struct dasd_ccw_req *cqr; int rc; @@ -1516,10 +1517,19 @@ dasd_eckd_psf_ssc(struct dasd_device *device, int enable_pav) if (IS_ERR(cqr)) return PTR_ERR(cqr); + /* + * set flags e.g. turn on failfast, to prevent blocking + * the calling function should handle failed requests + */ + cqr->flags |= flags; + rc = dasd_sleep_on(cqr); if (!rc) /* trigger CIO to reprobe devices */ css_schedule_reprobe(); + else if (cqr->intrc == -EAGAIN) + rc = -EAGAIN; + dasd_sfree_request(cqr, cqr->memdev); return rc; } @@ -1527,7 +1537,8 @@ dasd_eckd_psf_ssc(struct dasd_device *device, int enable_pav) /* * Valide storage server of current device. */ -static void dasd_eckd_validate_server(struct dasd_device *device) +static int dasd_eckd_validate_server(struct dasd_device *device, + unsigned long flags) { int rc; struct dasd_eckd_private *private; @@ -1536,17 +1547,18 @@ static void dasd_eckd_validate_server(struct dasd_device *device) private = (struct dasd_eckd_private *) device->private; if (private->uid.type == UA_BASE_PAV_ALIAS || private->uid.type == UA_HYPER_PAV_ALIAS) - return; + return 0; if (dasd_nopav || MACHINE_IS_VM) enable_pav = 0; else enable_pav = 1; - rc = dasd_eckd_psf_ssc(device, enable_pav); + rc = dasd_eckd_psf_ssc(device, enable_pav, flags); /* may be requested feature is not available on server, * therefore just report error and go ahead */ DBF_EVENT_DEVID(DBF_WARNING, device->cdev, "PSF-SSC for SSID %04x " "returned rc=%d", private->uid.ssid, rc); + return rc; } /* @@ -1556,7 +1568,13 @@ static void dasd_eckd_do_validate_server(struct work_struct *work) { struct dasd_device *device = container_of(work, struct dasd_device, kick_validate); - dasd_eckd_validate_server(device); + if (dasd_eckd_validate_server(device, DASD_CQR_FLAGS_FAILFAST) + == -EAGAIN) { + /* schedule worker again if failed */ + schedule_work(&device->kick_validate); + return; + } + dasd_put_device(device); } @@ -1685,7 +1703,7 @@ dasd_eckd_check_characteristics(struct dasd_device *device) if (rc) goto out_err2; - dasd_eckd_validate_server(device); + dasd_eckd_validate_server(device, 0); /* device may report different configuration data after LCU setup */ rc = dasd_eckd_read_conf(device); @@ -4153,7 +4171,7 @@ static int dasd_eckd_restore_device(struct dasd_device *device) rc = dasd_alias_make_device_known_to_lcu(device); if (rc) return rc; - dasd_eckd_validate_server(device); + dasd_eckd_validate_server(device, DASD_CQR_FLAGS_FAILFAST); /* RE-Read Configuration Data */ rc = dasd_eckd_read_conf(device); -- cgit v1.3-18-gd494 From 03429f34c22cedc3300a81e13d5790450cc468af Mon Sep 17 00:00:00 2001 From: Stefan Haberland Date: Tue, 11 Sep 2012 17:19:12 +0200 Subject: s390/dasd: fix read unit address configuration loop Read unit address is done for all devices during online processing to read out LCU features. This is also done after disconnect/connect a LCU. Some older storage hardware does not provide the capability to read unit address configuration. This leads to a loop trying to read unit address configuration every 30 seconds. The device is still operational but logs are flooded with error messages. Fix the loop by recognizing a command reject saying that the suborder for ruac is not supported. Signed-off-by: Stefan Haberland Signed-off-by: Martin Schwidefsky --- drivers/s390/block/dasd_alias.c | 27 ++++++++++++++++++++++++++- 1 file changed, 26 insertions(+), 1 deletion(-) (limited to 'drivers') diff --git a/drivers/s390/block/dasd_alias.c b/drivers/s390/block/dasd_alias.c index 157defe5e069..6b556995bb33 100644 --- a/drivers/s390/block/dasd_alias.c +++ b/drivers/s390/block/dasd_alias.c @@ -384,6 +384,29 @@ static void _remove_device_from_lcu(struct alias_lcu *lcu, group->next = NULL; }; +static int +suborder_not_supported(struct dasd_ccw_req *cqr) +{ + char *sense; + char reason; + char msg_format; + char msg_no; + + sense = dasd_get_sense(&cqr->irb); + if (!sense) + return 0; + + reason = sense[0]; + msg_format = (sense[7] & 0xF0); + msg_no = (sense[7] & 0x0F); + + /* command reject, Format 0 MSG 4 - invalid parameter */ + if ((reason == 0x80) && (msg_format == 0x00) && (msg_no == 0x04)) + return 1; + + return 0; +} + static int read_unit_address_configuration(struct dasd_device *device, struct alias_lcu *lcu) { @@ -435,6 +458,8 @@ static int read_unit_address_configuration(struct dasd_device *device, do { rc = dasd_sleep_on(cqr); + if (rc && suborder_not_supported(cqr)) + return -EOPNOTSUPP; } while (rc && (cqr->retries > 0)); if (rc) { spin_lock_irqsave(&lcu->lock, flags); @@ -521,7 +546,7 @@ static void lcu_update_work(struct work_struct *work) * processing the data */ spin_lock_irqsave(&lcu->lock, flags); - if (rc || (lcu->flags & NEED_UAC_UPDATE)) { + if ((rc && (rc != -EOPNOTSUPP)) || (lcu->flags & NEED_UAC_UPDATE)) { DBF_DEV_EVENT(DBF_WARNING, device, "could not update" " alias data in lcu (rc = %d), retry later", rc); schedule_delayed_work(&lcu->ruac_data.dwork, 30*HZ); -- cgit v1.3-18-gd494