Add a timed recovery procedure to reactivate ccw devices in cases
where HW/VM events are not sufficient to allow for proper recovery
of reappearing channel paths.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Make sure pending timers are always deleted, even if we don't go
through ccw_device_done().
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Also define helpers sch_{g,s}et_cdev() to make the intention more
clear.
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Some fields may be !0 only for I/O subchannels. Add some checks
where required. Also adapt cio_enable_subchannel() to make the
caller specify the intparm, which makes it more generic.
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Introduce a private pointer in struct subchannel to store
per-subchannel type data (cannot use dev->priv since this
is already used for something else).
Create a new header io_sch.h for I/O subchannel specific structures
and instructions.
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Cleanup cio_debug.h.
Also make CIO_DEBUG add the "cio:" prefix to the printk string
so that it isn't needed for the debug feature.
Fix outdated comments for cio_debug_init() and clean it up.
Enlarge cio_crw to the same size as cio_msg so we may actually
find some relevant information there.
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Make all callbacks in css_driver take a struct subchannel (and not
a struct device).
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Information about a ccw device will be dumped in
case of a ccw timeout. This can be enabled with
the kernel parameter ccw_timeout_log.
Signed-off-by: Sebastian Ott <sebott@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Make sure all subchannel handling is done on the slow path workqueue
so that we don't have races between an old subchannel unregistering
and a new subchannel with the same name registering.
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
There was the possibilty that an action like ccw_device_set_offline()
triggered by a device gone machine check might trigger a not oper
event. Unfortunately, this could lead to the situation that we tried
to unregister a subchannel twice: Once from the slow path evaluation,
and once via the not oper event.
Fix this by always using the same mechanism (css_schedule_eval()) for
triggering the unregister. This makes sure that unregistration will
only be done once. As an added bonus, it also simplyfies the code.
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Rename css[] to channel_subsystems[] to avoid name clashes.
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
- Remove unneeded messages.
- Move some messages into the debug feature.
- Use dev_* where appropriate.
- Use "cio: " prefix consistently.
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Deregister ccw device when device failure is detected during offline-
processing (e.g. when no last-path-gone indication was presented by
the hardware) to prevent the device from entering a non-recoverable
not-operational state.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Ensure that channel-path related subchannel data is only retrieved and
used when it is valid and that it is updated when it may have changed.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Use a bitmap for indicating which subchannels require evaluation
instead of allocating memory for each evaluation request. This
approach reduces memory consumption during recovery in case of
massive evaluation request occurrence and removes the need for
memory allocation failure handling.
Cc: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Add a new attribute to the channel-path sysfs directory through which
channel-path configure operations can be triggered. Also listen for
hardware events requesting channel-path configure operations and
process them accordingly.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Introduce data type for channel-path IDs.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
The subchannel may just be status pending, even with actl == 0. We
must go through the cancel_halt_clear procedure to put the subchannel
into a defined state.
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
After I/O has been killed by the common I/O layer, trigger path
verification which will queue cio_device_nopath_notify itself if it
finds a device to be without paths.
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Make sure we hold the device lock when we modify the ccw device
structure but always call the notify function without the lock held.
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
If we try to start path verification when an unsolicited interrupt
is already pending, stctl shows status pending and we delay path
verification again. We need to check for the doverify bit when the
unsolicited interrupt comes in and then do path verification.
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
- create a 'pseudo_subchannel' per channel subsystem (the 'orphanage')
- use the orphanage as a shelter for ccw_devices that can't remain on the same
subchannel
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Cc: Greg KH <greg@kroah.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
If I/O was running on a just varied off chpid, it will be terminated.
If this was a common I/O layer internal I/O, it needs to be retried.
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
If the last path to a device is gone after a chpid has been varied
off, putting it on the slow queue doesn't prevent a device driver
from still attempting to use it (it may stay on the slow queue for a
long time). Instead, trigger a verify event which will prevent I/O
attempts from the device driver immediately.
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Reset device operational notification flag when channel paths become
unavailable during path verification.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
The grace period handling introduced needless complexity. It didn't
help the dasd driver (which can handle terminated I/O just well),
and it doesn't help for long running channel programs (which won't
complete during the grace period anyway). Terminating I/O using a
path that just disappeared immediately is much more consistent with
what the user expects.
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Use the proper structures to identify device and subchannel. Change
get_disc_ccwdev_by_devno() to get_disc_ccwdev_by_dev_id().
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
While the machine owns us an interrupt in these cases (and we should get
one), reality isn't always like that...
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Reappearing channel paths are sometimes not utilized by CCW devices
because path verification incorrectly relies on path-operational-mask
information which is not updated until a channel path has been used
again.
Modify path verification procedure to always query all available paths
to a device.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Previous patch that was intended to reduce stack usage within common
i/o layer didn't consider implicit memset(..., 0, ...) used with the
initializations used before.
Add these missing memsets wherever it's not obvious that the
concerned memory region is zeroed. This should give the same semantics
as before.
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Use different kind of assignment to make sure gcc doesn't create code
that creates temp variables on the stack, assigns values to it and
copies the content of the whole temp variable to the destination.
This reduces stack usage of e.g. ccwgroup_driver_register from 976
to 48 bytes instead.
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
I/O on a CCW device may stall if a channel path to that device is
logicaly varied off/on. A user I/O interrupt can get misinterpreted
as interrupt for an internal path verification operation due to a
missing check and is therefore never reported to the device driver.
Correct check for pending interruptions before starting path
verification.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
In special conditions where a subchannel rejects the HALT I/O-
instruction with a busy indication (cc 2), I/O may stall.
I/O request termination logic retries HALT I/O indefinitely
because it expects HALT I/O to alter the subchannel status which
is not true when cc 2 is returned.
In case of a busy indication, try CLEAR I/O instruction immediately.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
1. Multipath devices for which SetPGID is not supported are not handled well.
Use NOP ccws for path verification (sans path grouping) when SetPGID is not
supported.
2. Check for PGIDs already set with SensePGID on _all_ paths (not just the
first one) and try to find a common one. Moan if no common PGID can be
found (and use NOP verification). If no PGIDs have been set, use the css
global PGID (as before). (Rationale: SetPGID will get a command reject if
the PGID it tries to set does not match the already set PGID.)
3. Immediately before reboot, issue RESET CHANNEL PATH (rcp) on all chpids. This
will remove the old PGIDs. rcp will generate solicited CRWs which can be
savely ignored by the machine check handler (all other actions create
unsolicited CRWs).
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Fixes for several channel measurement facility bugs:
* Blocks copied from the hardware might not be consistent. Solve this
by moving the copying into idle state and repeating the copying.
* avg_sample_interval changed with every read, even though no new block
was available. Solve this by storing a timestamp when the last new
block was received.
* Several locking issues.
* Measurements were not reenabled after a disconnected device became
available again.
* Remove #defines for ioctls that were never implemented.
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
After setting a path to a dasd offline at the SE, I/O hangs on that
dasd for 5 minutes, then continues.
I/O for which an interrupt will not be reported after the channel
path has been disabled was not terminated by the common I/O layer,
causing the dasd MIH to hit after 5 minutes.
Be more aggressive in terminating I/O after setting a channel path
offline. Also make sure to generate a fake irb if the device
driver issues an I/O request after being notified of the killed
I/O and clear residual information from the irb before trying to
start the delayed verification.
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
When cio waits for the interrupt for a basic sense, interrupts for hsch() or
csch() issued in the meantime are wrongly counted as interrupts for the basic
sense and the accumulated irb is passed to the device driver. In
ccw_device_w4sense(), check for clear or halt function in the irb and pass the
irb for the csch() or hsch() to the device driver.
Signed-off-by: Cornelia Huck <cornelia.huck@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Add support for multiple subchannel sets. Works with arbitrary devices in
subchannel set 1 and is transparent to device drivers. Although currently
only two subchannel sets are available, this will work with the architectured
maximum number of subchannel sets as well.
Signed-off-by: Cornelia Huck <cohuck@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
This patch introduces a struct subchannel_id containing the subchannel number
(formerly referred to as "irq") and switches code formerly relying on the
subchannel number over to it.
While we're touching inline assemblies anyway, make sure they have correct
memory constraints.
Signed-off-by: Cornelia Huck <cohuck@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
