Revamp the qdio performance statistics and move them from procfs to
debugfs using the seq_file interface. Since the statistics are not
intended for the general user the removal of /proc/qdio_perf should
not surprise anyone.
The per device statistics are disabled by default, writing 1 to
/<debugfs mountpoint>/qdio/<device bus ID>/statistics enables the
statistics for the given device.
Signed-off-by: Jan Glauber <jang@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Add a counter to the qdio performance statistics that indicates that no
free buffers were left in the input queue. If the counter gets increased
it means that the qdio adapter filled all available buffers and possibly
had more buffers ready but could not transmit them.
Signed-off-by: Jan Glauber <jang@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
* Don't write debug feature log entries for sl, slsb and sbal since these
elements can be located from the qdio_q pointer which is also logged.
* Convert WARN_ON for wrong alignment of sbal to BUG_ON.
* Remove WARN_ON's for wrong alignment of q / qib / slib since these
alignments should be guaranteed by kmem_cache_alloc alignment /
struct aligned attribute / __get_free_page.
Signed-off-by: Jan Glauber <jang@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Channel path vary is currently broken: channel paths which are varied
offline are still used by Linux. The reason for this is that:
* the path mask indicating which paths of an I/O device can be used
is reset by each internal I/O request
* the logic that checks if a path group is already in its designated
target state incorrectly interprets the result "is correctly set"
as "is correctly set and available"
Fix this by resetting the path mask only for internal I/O requests
which affect the path mask and by correcting the pgid check logic.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Memset should be given the size of the structure, not the size of the pointer.
The semantic patch that makes this change is as follows:
(http://coccinelle.lip6.fr/)
// <smpl>
@@
type T;
T *x;
expression E;
@@
memset(x, E, sizeof(
+ *
x))
// </smpl>
Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: Heiko Carstens <heiko.carstens@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Using dev_set_drvdata prior to device_register will force the driver core
to kmalloc its private data. Since we use this for the console subchannel
lets set the drvdata before taking the subchannels spinlock.
Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
* 'timers-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/linux-2.6-tip:
timers, init: Limit the number of per cpu calibration bootup messages
posix-cpu-timers: optimize and document timer_create callback
clockevents: Add missing include to pacify sparse
x86: vmiclock: Fix printk format
x86: Fix printk format due to variable type change
sparc: fix printk for change of variable type
clocksource/events: Fix fallout of generic code changes
nohz: Allow 32-bit machines to sleep for more than 2.15 seconds
nohz: Track last do_timer() cpu
nohz: Prevent clocksource wrapping during idle
nohz: Type cast printk argument
mips: Use generic mult/shift factor calculation for clocks
clocksource: Provide a generic mult/shift factor calculation
clockevents: Use u32 for mult and shift factors
nohz: Introduce arch_needs_cpu
nohz: Reuse ktime in sub-functions of tick_check_idle.
time: Remove xtime_cache
time: Implement logarithmic time accumulation
If we detect a busy subchannel after the driver's set_offline
callback returned in ccw_device_set_offline, the current behavior
is to unregister the device, which may lead to undesired
consequences. Change this to just quiesce the subchannel and go on
with the offline processing.
Note: This is no excuse for not fixing these drivers -
after the set_offline callback they should have no running IO!
Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Improve error recovery for internal I/Os by repeating each I/O
256 times per path to cope with long-running non-permanent error
conditions. Also retry each path twice to cope with link flapping,
i.e. single paths becoming unavailable in the order in which they
are tried.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
IO subchannels are always unregistered in process context, so use
spin_lock_irq in the corresponding remove callback.
Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Ensure that there will be no more interrupts for an
unregistered device by using the same quiesce and disable loop
as in io_subchannel_shutdown.
Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Try to disable the old subchannel before we ask the driver core
to move the attached device to a new parent. This way we can use
the QUIESCE state during shutdown which prevents a possible use
after free situation in some error cases.
Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Handle a failing cio_disable_subchannel at the end of our device
recognition as if the recognition itself failed. This way
subsequent registration steps do not need to handle enabled
subchannels.
Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
DEV_STATE_QUIESCE is used to stop all IO on a busy subchannel.
This patch fixes the following problems related to the QUIESCE
state:
* Fix a potential race condition which could occur when the
resulting state was DEV_STATE_OFFLINE.
* Add missing locking around cio_disable_subchannel,
ccw_device_cancel_halt_clear and the cdev's handler.
* Loop until we know for sure that the subchannel is disabled.
Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
The function ccw_device_unregister has to ensure to remove
all references obtained by device_add and device_initialize.
Unfortunately it gets called for devices which are
1) uninitialized, 2) initialized but unregistered, and
3) registered devices. To distinguish 1) and 2) this patch
introduces a new flag "initialized", which is 1 as long as we
hold the initial device reference.
Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
We used to maintain a "registered" flag in our ccw_device_private
structure. This patch removes the "registered" flag and converts
all users of it to device_is_registered which has the exact same
meaning.
Note: The usage the atomic operation test_and_clear_bit is replaced
by the non-atomic if (device_is_registered()) device_del(). This
will not do harm, since we serialize calls to ccw_device_unregister
with a single-threaded workqueue.
Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
An Unconditional Reserve + Release operation (steal lock) for a
boxed device may fail when encountering special error cases
(e.g. unit checks or path errors). Fix this by using the more
robust ccw_request infrastructure for performing the steal lock
CCW program.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Set-pgid operations fail for some device types under z/VM for which
the hypervisor has already set the pgid. Also reserved devices or
changed pgids are not correctly recognized. Fix these problems by
using a combination of sense-pgid and set-pgid and by also accepting
pre-defined pgid settings.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Split setting (driver wants feature enabled) and status (feature
setup was successful) for PGID related ccw device features so that
setup errors can be detected. Previously, incorrectly handled setup
errors could in rare cases lead to erratic I/O behavior and
permanently unusuable devices.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
After changing all internal I/O functions to use the newly introduced
ccw request infrastructure, retries are handled automatically after a
clear operation. Therefore remove the internal retry flag and
associated code.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Accept a request for setting a not-operational device offline.
This way, users can remove devices from Linux which would otherwise
remain unusable until reboot.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Use the newly introduced ccw request infrastructure to implement
pgid related operations: sense pgid, set pgid and disband pg.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Use the newly introduced ccw request infrastructure to implement
the sense id operation.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Reduce code duplication by introducing a central infrastructure to
perform an internal I/O operation on a CCW device.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Remove the call to BUG() for situations which are unexpected
but do not cause actual problems.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Device recognition needs to be started with the ccw device lock
held to prevent race conditions between I/O starting and interrupt
reception.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Handle verification errors consistently through the existing
callback ccw_device_done to reduce cleanup code duplication.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Remove the return code from ccw_device_recognition and handle
recognition errors through the existing callback
ccw_device_recog_done to reduce cleanup code duplication.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Print a warning message in case a ccw device enters boxed or
not operational state during online/offline processing.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Introduce a central mechanism for performing delayed ccw device work
to ensure that different types of work do not overwrite each other.
Prioritization ensures that the most important work is always
performed while less important tasks are either obsoleted or repeated
later.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Ensure that current and future users of sch->work do not overwrite
each other by introducing a single mechanism for delayed subchannel
work.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Change the initiative to update subchannel-ccw device associations
to the subchannel: when there is an indication that the internal
association no longer reflects the current hardware state, mark
each affected subchannel as requiring attention. Once processing
reaches a subchannel, determine the correct association for that
subchannel at that time and perform the necessary device_move
operations.
This change fixes problems with the previous approach which would
leave devices in an inconsistent state when a new hardware change
occurred while a device_move was already scheduled.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
sch_create_and_recog_new_device() associates a parent subchannel
with its ccw device child even though this is already done by
the subsequently called io_subchannel_recog(). Also make sure
io_subchannel_recog() sets the association under lock.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
io_subchannel_probe() frees memory for sch->private which is later
freed again when io_subchannel_remove() is called. Fix this problem
by removing the cleanup in io_subchannel_probe().
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Allow the architecture to request a normal jiffy tick when the system
goes idle and tick_nohz_stop_sched_tick is called . On s390 the hook is
used to prevent the system going fully idle if there has been an
interrupt other than a clock comparator interrupt since the last wakeup.
On s390 the HiperSockets response time for 1 connection ping-pong goes
down from 42 to 34 microseconds. The CPU cost decreases by 27%.
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
LKML-Reference: <20090929122533.402715150@de.ibm.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Use cio_is_console() in io_subchannel_probe to indicate that the
special handling is console specific. As long as there is no other
subchannel for which this might be true, it is misleading to speak
of "early devices". Should more of these devices be introduced,
a cleanup of all console special handling is in order anyway.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Allow users to set boxed devices offline. After setting them
offline, the device state will still be boxed.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
When a ccw device appears not operational, inform the associated
device driver and act according to the response: if the driver
wants to keep the device, put it into the disconnected state.
If not, or if there is no driver or if the device is not online,
unregister it. This approach is consistent with no-path event
handling.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
When there is no path left to a ccw device, inform the associated
device driver and act according to the response: if the driver
wants to keep the device, put it into the disconnected state.
If not, or if there is no driver or if the device is not online,
unregister it.
Signed-off-by: Peter Oberparleiter <peter.oberparleiter@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
There is a memory leak in /proc/cio_ignore. The iterator is allocated
in cio_ignore_proc_seq_start, but never freed in
cio_ignore_proc_seq_stop, because we cannot use the iterator
that was passed by seqfile. The seqfile interface passes the last
seen iterator to the stop function and not the first one. Since our
next function will return NULL at the end, the iter passed to
cio_ignore_proc_seq_stop is NULL. The original iter has leaked.
The solution is to use seq_open_private.
Found with kmemleak:
unreferenced object 0x1c720580 (size 32):
comm "head", pid 973, jiffies 4294958302
hex dump (first 32 bytes):
00 00 00 00 00 00 00 04 00 00 00 00 00 00 00 00 ................
00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 ................
backtrace:
[<0000000000203154>] kmem_cache_alloc+0x190/0x19c
[<00000000003fb462>] cio_ignore_proc_seq_start+0x5e/0x128
[<0000000000231018>] seq_read+0xc8/0x4bc
[<0000000000273954>] proc_reg_read+0xa8/0xf4
[<000000000020e3d8>] vfs_read+0xac/0x1a4
[<000000000020e5c6>] SyS_read+0x52/0xa8
[<000000000011836e>] sysc_noemu+0x10/0x16
[<0000004690b7936c>] 0x4690b7936c
Signed-off-by: Christian Borntraeger <borntraeger@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Move dev_set_name to when we know that the device will actually be
registered in order to avoid a memory leak if the allocated memory
for the channel path has to be freed.
Signed-off-by: Michael Ernst <mernst@de.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
If input buffers stay in primed state qdio may not receive further interrupts
for the input queue depending on the firmware. That can cause a connection
hang on OSA cards.
Change the state of all primed input buffers that are not acknowledged to
not initialized.
Signed-off-by: Jan Glauber <jang@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Even if turned off the debug message overhead is measurable in the hot path.
Reduce the number of debug message calls in do_QDIO and qdio_kick_handler.
Also use hex numbers to save space in the debug entries.
Signed-off-by: Jan Glauber <jang@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
Previously, there were multiple subchannel scanning mechanisms
which could potentially conflict with each other. Fix this problem
by moving blacklist and ccw driver triggered scanning to the
existing evaluation method.
Signed-off-by: Sebastian Ott <sebott@linux.vnet.ibm.com>
Signed-off-by: Martin Schwidefsky <schwidefsky@de.ibm.com>
