Commit 9bfefde7 authored by Stefan Haberland's avatar Stefan Haberland Committed by Martin Schwidefsky

s390/dasd: fix incorrect locking order for LCU device add/remove

The correct lock order for LCU lock and cdev lock is to take the cdev
lock first and afterwards the LCU lock. This is caused by the fact
that LCU functions are called in an interrupt context with the cdev
lock implicitly hold by CIO.

To assure the right locking order but also be able to iterate over
devices in a LCU introduce a trylock block that can be called with
the device lock for one device hold and then takes the LCU lock and
try to lock all devices accounted to this LCU. Afterwards all devices
and the LCU itself are locked.
Signed-off-by: default avatarStefan Haberland <stefan.haberland@de.ibm.com>
Signed-off-by: default avatarMartin Schwidefsky <schwidefsky@de.ibm.com>
parent 1f021ea0
......@@ -319,22 +319,14 @@ static int _add_device_to_lcu(struct alias_lcu *lcu,
struct dasd_eckd_private *private;
struct alias_pav_group *group;
struct dasd_uid uid;
unsigned long flags;
private = (struct dasd_eckd_private *) device->private;
/* only lock if not already locked */
if (device != pos)
spin_lock_irqsave_nested(get_ccwdev_lock(device->cdev), flags,
CDEV_NESTED_SECOND);
private->uid.type = lcu->uac->unit[private->uid.real_unit_addr].ua_type;
private->uid.base_unit_addr =
lcu->uac->unit[private->uid.real_unit_addr].base_ua;
uid = private->uid;
if (device != pos)
spin_unlock_irqrestore(get_ccwdev_lock(device->cdev), flags);
/* if we have no PAV anyway, we don't need to bother with PAV groups */
if (lcu->pav == NO_PAV) {
list_move(&device->alias_list, &lcu->active_devices);
......@@ -411,6 +403,130 @@ suborder_not_supported(struct dasd_ccw_req *cqr)
return 0;
}
/*
* This function tries to lock all devices on an lcu via trylock
* return NULL on success otherwise return first failed device
*/
static struct dasd_device *_trylock_all_devices_on_lcu(struct alias_lcu *lcu,
struct dasd_device *pos)
{
struct alias_pav_group *pavgroup;
struct dasd_device *device;
list_for_each_entry(device, &lcu->active_devices, alias_list) {
if (device == pos)
continue;
if (!spin_trylock(get_ccwdev_lock(device->cdev)))
return device;
}
list_for_each_entry(device, &lcu->inactive_devices, alias_list) {
if (device == pos)
continue;
if (!spin_trylock(get_ccwdev_lock(device->cdev)))
return device;
}
list_for_each_entry(pavgroup, &lcu->grouplist, group) {
list_for_each_entry(device, &pavgroup->baselist, alias_list) {
if (device == pos)
continue;
if (!spin_trylock(get_ccwdev_lock(device->cdev)))
return device;
}
list_for_each_entry(device, &pavgroup->aliaslist, alias_list) {
if (device == pos)
continue;
if (!spin_trylock(get_ccwdev_lock(device->cdev)))
return device;
}
}
return NULL;
}
/*
* unlock all devices except the one that is specified as pos
* stop if enddev is specified and reached
*/
static void _unlock_all_devices_on_lcu(struct alias_lcu *lcu,
struct dasd_device *pos,
struct dasd_device *enddev)
{
struct alias_pav_group *pavgroup;
struct dasd_device *device;
list_for_each_entry(device, &lcu->active_devices, alias_list) {
if (device == pos)
continue;
if (device == enddev)
return;
spin_unlock(get_ccwdev_lock(device->cdev));
}
list_for_each_entry(device, &lcu->inactive_devices, alias_list) {
if (device == pos)
continue;
if (device == enddev)
return;
spin_unlock(get_ccwdev_lock(device->cdev));
}
list_for_each_entry(pavgroup, &lcu->grouplist, group) {
list_for_each_entry(device, &pavgroup->baselist, alias_list) {
if (device == pos)
continue;
if (device == enddev)
return;
spin_unlock(get_ccwdev_lock(device->cdev));
}
list_for_each_entry(device, &pavgroup->aliaslist, alias_list) {
if (device == pos)
continue;
if (device == enddev)
return;
spin_unlock(get_ccwdev_lock(device->cdev));
}
}
}
/*
* this function is needed because the locking order
* device lock -> lcu lock
* needs to be assured when iterating over devices in an LCU
*
* if a device is specified in pos then the device lock is already hold
*/
static void _trylock_and_lock_lcu_irqsave(struct alias_lcu *lcu,
struct dasd_device *pos,
unsigned long *flags)
{
struct dasd_device *failed;
do {
spin_lock_irqsave(&lcu->lock, *flags);
failed = _trylock_all_devices_on_lcu(lcu, pos);
if (failed) {
_unlock_all_devices_on_lcu(lcu, pos, failed);
spin_unlock_irqrestore(&lcu->lock, *flags);
cpu_relax();
}
} while (failed);
}
static void _trylock_and_lock_lcu(struct alias_lcu *lcu,
struct dasd_device *pos)
{
struct dasd_device *failed;
do {
spin_lock(&lcu->lock);
failed = _trylock_all_devices_on_lcu(lcu, pos);
if (failed) {
_unlock_all_devices_on_lcu(lcu, pos, failed);
spin_unlock(&lcu->lock);
cpu_relax();
}
} while (failed);
}
static int read_unit_address_configuration(struct dasd_device *device,
struct alias_lcu *lcu)
{
......@@ -505,10 +621,7 @@ static int _lcu_update(struct dasd_device *refdev, struct alias_lcu *lcu)
if (rc)
return rc;
/* need to take cdev lock before lcu lock */
spin_lock_irqsave_nested(get_ccwdev_lock(refdev->cdev), flags,
CDEV_NESTED_FIRST);
spin_lock(&lcu->lock);
_trylock_and_lock_lcu_irqsave(lcu, NULL, &flags);
lcu->pav = NO_PAV;
for (i = 0; i < MAX_DEVICES_PER_LCU; ++i) {
switch (lcu->uac->unit[i].ua_type) {
......@@ -527,8 +640,8 @@ static int _lcu_update(struct dasd_device *refdev, struct alias_lcu *lcu)
alias_list) {
_add_device_to_lcu(lcu, device, refdev);
}
spin_unlock(&lcu->lock);
spin_unlock_irqrestore(get_ccwdev_lock(refdev->cdev), flags);
_unlock_all_devices_on_lcu(lcu, NULL, NULL);
spin_unlock_irqrestore(&lcu->lock, flags);
return 0;
}
......@@ -616,8 +729,6 @@ int dasd_alias_add_device(struct dasd_device *device)
private = (struct dasd_eckd_private *) device->private;
lcu = private->lcu;
rc = 0;
/* need to take cdev lock before lcu lock */
spin_lock_irqsave(get_ccwdev_lock(device->cdev), flags);
spin_lock(&lcu->lock);
if (!(lcu->flags & UPDATE_PENDING)) {
......@@ -754,30 +865,19 @@ static void _restart_all_base_devices_on_lcu(struct alias_lcu *lcu)
struct alias_pav_group *pavgroup;
struct dasd_device *device;
struct dasd_eckd_private *private;
unsigned long flags;
/* active and inactive list can contain alias as well as base devices */
list_for_each_entry(device, &lcu->active_devices, alias_list) {
private = (struct dasd_eckd_private *) device->private;
spin_lock_irqsave(get_ccwdev_lock(device->cdev), flags);
if (private->uid.type != UA_BASE_DEVICE) {
spin_unlock_irqrestore(get_ccwdev_lock(device->cdev),
flags);
if (private->uid.type != UA_BASE_DEVICE)
continue;
}
spin_unlock_irqrestore(get_ccwdev_lock(device->cdev), flags);
dasd_schedule_block_bh(device->block);
dasd_schedule_device_bh(device);
}
list_for_each_entry(device, &lcu->inactive_devices, alias_list) {
private = (struct dasd_eckd_private *) device->private;
spin_lock_irqsave(get_ccwdev_lock(device->cdev), flags);
if (private->uid.type != UA_BASE_DEVICE) {
spin_unlock_irqrestore(get_ccwdev_lock(device->cdev),
flags);
if (private->uid.type != UA_BASE_DEVICE)
continue;
}
spin_unlock_irqrestore(get_ccwdev_lock(device->cdev), flags);
dasd_schedule_block_bh(device->block);
dasd_schedule_device_bh(device);
}
......@@ -841,38 +941,20 @@ static void flush_all_alias_devices_on_lcu(struct alias_lcu *lcu)
spin_unlock_irqrestore(&lcu->lock, flags);
}
static void __stop_device_on_lcu(struct dasd_device *device,
struct dasd_device *pos)
{
/* If pos == device then device is already locked! */
if (pos == device) {
dasd_device_set_stop_bits(pos, DASD_STOPPED_SU);
return;
}
spin_lock(get_ccwdev_lock(pos->cdev));
dasd_device_set_stop_bits(pos, DASD_STOPPED_SU);
spin_unlock(get_ccwdev_lock(pos->cdev));
}
/*
* This function is called in interrupt context, so the
* cdev lock for device is already locked!
*/
static void _stop_all_devices_on_lcu(struct alias_lcu *lcu,
struct dasd_device *device)
static void _stop_all_devices_on_lcu(struct alias_lcu *lcu)
{
struct alias_pav_group *pavgroup;
struct dasd_device *pos;
struct dasd_device *device;
list_for_each_entry(pos, &lcu->active_devices, alias_list)
__stop_device_on_lcu(device, pos);
list_for_each_entry(pos, &lcu->inactive_devices, alias_list)
__stop_device_on_lcu(device, pos);
list_for_each_entry(device, &lcu->active_devices, alias_list)
dasd_device_set_stop_bits(device, DASD_STOPPED_SU);
list_for_each_entry(device, &lcu->inactive_devices, alias_list)
dasd_device_set_stop_bits(device, DASD_STOPPED_SU);
list_for_each_entry(pavgroup, &lcu->grouplist, group) {
list_for_each_entry(pos, &pavgroup->baselist, alias_list)
__stop_device_on_lcu(device, pos);
list_for_each_entry(pos, &pavgroup->aliaslist, alias_list)
__stop_device_on_lcu(device, pos);
list_for_each_entry(device, &pavgroup->baselist, alias_list)
dasd_device_set_stop_bits(device, DASD_STOPPED_SU);
list_for_each_entry(device, &pavgroup->aliaslist, alias_list)
dasd_device_set_stop_bits(device, DASD_STOPPED_SU);
}
}
......@@ -880,33 +962,16 @@ static void _unstop_all_devices_on_lcu(struct alias_lcu *lcu)
{
struct alias_pav_group *pavgroup;
struct dasd_device *device;
unsigned long flags;
list_for_each_entry(device, &lcu->active_devices, alias_list) {
spin_lock_irqsave(get_ccwdev_lock(device->cdev), flags);
list_for_each_entry(device, &lcu->active_devices, alias_list)
dasd_device_remove_stop_bits(device, DASD_STOPPED_SU);
spin_unlock_irqrestore(get_ccwdev_lock(device->cdev), flags);
}
list_for_each_entry(device, &lcu->inactive_devices, alias_list) {
spin_lock_irqsave(get_ccwdev_lock(device->cdev), flags);
list_for_each_entry(device, &lcu->inactive_devices, alias_list)
dasd_device_remove_stop_bits(device, DASD_STOPPED_SU);
spin_unlock_irqrestore(get_ccwdev_lock(device->cdev), flags);
}
list_for_each_entry(pavgroup, &lcu->grouplist, group) {
list_for_each_entry(device, &pavgroup->baselist, alias_list) {
spin_lock_irqsave(get_ccwdev_lock(device->cdev), flags);
list_for_each_entry(device, &pavgroup->baselist, alias_list)
dasd_device_remove_stop_bits(device, DASD_STOPPED_SU);
spin_unlock_irqrestore(get_ccwdev_lock(device->cdev),
flags);
}
list_for_each_entry(device, &pavgroup->aliaslist, alias_list) {
spin_lock_irqsave(get_ccwdev_lock(device->cdev), flags);
list_for_each_entry(device, &pavgroup->aliaslist, alias_list)
dasd_device_remove_stop_bits(device, DASD_STOPPED_SU);
spin_unlock_irqrestore(get_ccwdev_lock(device->cdev),
flags);
}
}
}
......@@ -932,13 +997,14 @@ static void summary_unit_check_handling_work(struct work_struct *work)
spin_unlock_irqrestore(get_ccwdev_lock(device->cdev), flags);
reset_summary_unit_check(lcu, device, suc_data->reason);
spin_lock_irqsave(&lcu->lock, flags);
_trylock_and_lock_lcu_irqsave(lcu, NULL, &flags);
_unstop_all_devices_on_lcu(lcu);
_restart_all_base_devices_on_lcu(lcu);
/* 3. read new alias configuration */
_schedule_lcu_update(lcu, device);
lcu->suc_data.device = NULL;
dasd_put_device(device);
_unlock_all_devices_on_lcu(lcu, NULL, NULL);
spin_unlock_irqrestore(&lcu->lock, flags);
}
......@@ -974,10 +1040,7 @@ void dasd_alias_handle_summary_unit_check(struct dasd_device *device,
" unit check (no lcu structure)");
return;
}
spin_lock(&lcu->lock);
_stop_all_devices_on_lcu(lcu, device);
/* prepare for lcu_update */
private->lcu->flags |= NEED_UAC_UPDATE | UPDATE_PENDING;
_trylock_and_lock_lcu(lcu, device);
/* If this device is about to be removed just return and wait for
* the next interrupt on a different device
*/
......@@ -985,6 +1048,7 @@ void dasd_alias_handle_summary_unit_check(struct dasd_device *device,
DBF_DEV_EVENT(DBF_WARNING, device, "%s",
"device is in offline processing,"
" don't do summary unit check handling");
_unlock_all_devices_on_lcu(lcu, device, NULL);
spin_unlock(&lcu->lock);
return;
}
......@@ -993,12 +1057,17 @@ void dasd_alias_handle_summary_unit_check(struct dasd_device *device,
DBF_DEV_EVENT(DBF_WARNING, device, "%s",
"previous instance of summary unit check worker"
" still pending");
_unlock_all_devices_on_lcu(lcu, device, NULL);
spin_unlock(&lcu->lock);
return ;
}
_stop_all_devices_on_lcu(lcu);
/* prepare for lcu_update */
private->lcu->flags |= NEED_UAC_UPDATE | UPDATE_PENDING;
lcu->suc_data.reason = reason;
lcu->suc_data.device = device;
dasd_get_device(device);
_unlock_all_devices_on_lcu(lcu, device, NULL);
spin_unlock(&lcu->lock);
if (!schedule_work(&lcu->suc_data.worker))
dasd_put_device(device);
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment