There are several scenarios where the ibmvfc driver needs to
try to log back into a target on the fabric. Today when these events
occur, we simply go through re-discovery for all attached targets,
assuming that either the query of the name server or an ADISC will
indicate we might need to log back into the target, which doesn't
work for all scenarios. Fix this by taking note of the affected target(s)
in these conditions and ensuring we try to PLOGI back into the target.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
For certain scenarios during device rediscovery, we detect we need
to log back into a target. Currently we do just that - PLOGI/PRLI
back into the target. Change the code to delete and add the target
from the FC transport layer as well, to ensure we handle any cases
where the target may have changed.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
The virtual I/O server controlling the NPIV adapter associated with
a virtual fibre channel adapter can send a HALT event to the client.
When this occurs, the client can no longer send commands until a RESUME
is received. By adding support for flush on halt, we will get all of
our outstanding commands flushed back before the Virtual I/O server
enters the halt state, eliminating potential command timeouts for
outstanding commands which might occur if we did not support this feature.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
This patch adds support for a new command supported by the Virtual I/O
Server, NPIV Logout. The command will abort all outstanding commands
and log out of the fabric. Currently, the only way to do this is
by breaking the CRQ, which can take a fairly long time when lots of
commands are outstanding. The NPIV Logout commands provides a mechanism
to accomplish virtually the same function, but is much faster.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
The ibmvfc driver currently logs errors during discovery for several
transient fabric errors, which generally get retried. If retries
do not work, we see multiple errors in the log. If retries do work,
we see errors in the log which may be confusing since the retry worked.
This patch enhances the discovery time error logging to only log errors
for command failures during discovery if all allowed retries have been
used up. The existing behavior of logging all failures can be restored
by setting the hosts log_level to a value of 3 or greater.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Use DEVICE_ATTR macro for defining device sysfs attributes.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Since target allocations can occur while resetting the virtual adapter,
we shouldn't be using GFP_KERNEL for them as it could hang. Switch to
use GFP_NOIO.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Fix an obvious bug in processing error responses for SCSI commands
which can result in successful responses being incorrectly returned
with DID_ERROR.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Signed-off-by: Roel Kluin <roel.kluin@gmail.com>
Acked by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Bump driver version to 1.0.5.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
The ibmvfc driver currently breaks the CRQ and essentially
resets the entire virtual FC adapter, killing all outstanding
ops to all attached targets, if an ADISC times out during target
discover/rediscovery. This patch adds some code to cancel the
ADISC if it times out, which prevents a single ADISC timeout from
affecting the other devices attached to the fabric.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Set show_host_maxframe_size so that maxframe_size gets exported in
sysfs for the host.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
The ibmvfc driver really does not handle dynamically changing disc_threads.
To change this dynamically would cause confusion in the driver regarding
the number of event structs allocated. Fix this by simply not allowing
disc_threads to be changed at runtime.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
This patch fixes a problem of possible dropped interrupts. Currently,
the ibmvfc driver has a race condition where after ibmvfc_interrupt
gets run, the platform code clears the interrupt. This can result in
lost interrupts and, in worst case scenarios, result in command
timeouts. Fix this by implementing a tasklet similar to what the
ibmvscsi driver does so that interrupt processing is no longer done in
the actual interrupt handler, which eliminates the race.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
The ibmvfc driver currently always sets the role of all rports
to FC_PORT_ROLE_FCP_TARGET, which is not correct for other initiators.
This can cause problems if other initiators are on the fabric
when we then try to scan the rport for LUNs. Fix this by looking
at the service parameters returned in the PRLI to set the roles
appropriately. Also look at the returned service parameters to
decide whether or not we were actually able to successfully log into
the target.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
During cancel testing it has been shown that 15 seconds is not
nearly long enough for the VIOS to respond to a cancel under
loaded situations. Increasing this timeout to 60 seconds allows
time for the VIOS to cancel the outstanding commands and prevents
us from escalating to a full host reset, which can take much longer.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
The ibmvfc driver has a bug in its SCN handling. If it receives
an ELS event such asn an N-Port SCN event or an unsolicited PLOGI,
or any other SCN event which causes ibmvfc_reinit_host to be called,
it is possible that we will call fc_remote_port_add for a target
that already has an rport added, which can result in duplicate
rports getting created for the same targets. Fix this by calling
fc_remote_port_rolechg in this scenario instead to report any possible
role change that may have occurred.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Currently the ibmvfc driver sets the IBMVFC_CLASS_3_ERR flag
in the VFC Frame if both the adapter and the device claim support
for Class 3. However, this bit actually refers to Class 3 Error
Recovery, which is currently not supported by the VIOS. Setting this
bit can cause lots of command timeout responses from the VIOS resulting
in general instability. Fix this by never setting this bit.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
The ibmvscsi client driver is not unmapping the SCSI command after
encountering a DMA mapping error while trying to map an indirect
scattergather list for the event pool. This leads to a leak of DMA
entitlement that could result in the device failing future DMA operations
in a CMO environment.
Signed-off-by: Robert Jennings <rcj@linux.vnet.ibm.com>
Acked-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
There is currently a DMA mapping leak that can occur in the ibmvfc
driver if we fail to allocate a scatterlist. Fix this by unmapping
the scatterlist in the failure path. Additionally, only log an error
for a scatterlist allocation failure if the log level is greater
than the default, since this can occur when running Active Memory
Sharing and this is not considered an error.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
This is a powerpc specific driver.
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Convert arch/powerpc/ over to long long based u64:
-#ifdef __powerpc64__
-# include <asm-generic/int-l64.h>
-#else
-# include <asm-generic/int-ll64.h>
-#endif
+#include <asm-generic/int-ll64.h>
This will avoid reoccuring spurious warnings in core kernel code that
comes when people test on their own hardware. (i.e. x86 in ~98% of the
cases) This is what x86 uses and it generally helps keep 64-bit code
32-bit clean too.
[Adjusted to not impact user mode (from paulus) - sfr]
Signed-off-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Signed-off-by: Benjamin Herrenschmidt <benh@kernel.crashing.org>
In a previous patch to fix an issue with error recovery,
the behavior of the max_requests module paramater was also
changed. If, for some reason, max_requests is set to one by
the user, we will end up with a negative number for can_queue.
Fix this by making max_requests not include the two event structs
needed to do error recovery.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
If a link down event is received, outstanding commands may get
returned to the ibmvfc driver with a "transaction cancelled implicit"
response. This is currently translated to DID_ABORT, which does
not get retried by SCSI core, but rather passes the failure up
the stack. This can result in I/O errors at the filesystem level.
Fix up this response a well as a few other error responses.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
[jejb: limit ioctl to returning 20 characters to avoid overrun
on long device names and add a few more conversions]
Signed-off-by: Kay Sievers <kay.sievers@vrfy.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
While doing various error injection testing, such as cable
pulls and target moves, some issues were observed in handling
these events. This patch improves the way these events are handled
by increasing the delay waiting for the fabric to settle and also
changes the behavior of Link Up to break the CRQ to ensure everything
gets cleaned up properly on the VIOS.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
The ibmvscsi driver currently has a bug in it which can result
in it using up all its event structs for commands. If something
results in all those commands timing out, we won't have any resources
left to send aborts or resets. This results in escalating to a host reset
in order to recover, which is a bit heavy handed. This fixes it
by reducing can_queue by two in order to have resources to do EH.
It also changes the max_requests module parameter so that it is not
writable at runtime, since the code really does not handle it changing
at runtime.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
In order to ensure the VIOS sees a consistent command buffer, we
need to add a memory barrier after building the command buffer
but before sending the command.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Adds a delay prior to retrying a failed NPIV login. This fixes
a scenario if the backing fibre channel adapter is getting reset
due to an EEH event, NPIV login will fail. Currently, ibmvfc
retries three times very quickly, resets the CRQ and tries one
more time. If the adapter is getting reset due to EEH, this isn't
enough time. This adds a delay prior to retrying a failed NPIV
login and also increments the number of retries.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
The virtual fibre channel stack can return a failure response for a command
indicating the port login has been invalidated without sending the client
an async event. Add code to handle this response and initiate a PLOGI.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
The CRQs used by the ibmvfc driver are read and written by both
the client and the server. Therefore, we need to mark them volatile
so that we do not cache their contents when handling an interrupt.
This fixes a problem which can surface as occasional command timeouts.
No commands were actually timing out, but due to accessing cached data
for the CRQ in the interrupt handler, the interrupt was not processing
all command completions as it should.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Fixes an oops that can occur in the interrupt handler
if we get a lot of async events.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Stops gcc from complaining about a possible uninitialized
variable being used in ibmvfc_reset_device.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
If the ibmvfc driver is in discovery attempting to log into a target
and it encounters an error, the command may get retried one or more
times, depending on the error received. If the retries are
unsuccessful such that the discovery thread gives up on discovery to
that target, the target ends up in a state where, if SCSI core had
previously known about the device, the host will get unblocked but the
host will not be logged into the target, causing any commands sent to
the target to fail. This patch fixes this so that if this occurs, the
target is deleted such that the normal dev_loss processing can occur
instead.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Due to an ambiguity in the VIOS VFC interface specification,
abort/cancel handling is not done correctly and can result in double
completion of commands. In order to cancel all outstanding commands to
a device, a cancel must be sent, followed by an abort task set. After
the responses are received for these commands, there may still be
commands outstanding, in the process of getting flushed back, in which
case, we need to wait for them. This patch removes the assumption that
if the abort and the cancel both complete successfully that the device
queue has been flushed and waits for all the responses to come back.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
If either a "transport fault" or a "general transport" error is received
and no other error information is available, the command is improperly
returned as successful. Fix this to return DID_ERROR in this case.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
The ibmvfc log level filtering logic was reversed. The log_level scsi
host parameter should result in more verbose logs when log_level is
larger, not smaller.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Calling crq_queue_create could lead to the creation of a rport. We
need to set up everything before creating a rport. This moves
crq_queue_create to the end of initialization to avoid a race which
causes an oops if lost.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Reported-by: Olaf Hering <olh@suse.de>
Cc: stable@kernel.org
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
ibmvscsi sets the timeout in its slave configure routine for disk
devices. This now needs to update the request queue timeout in block.
Cc: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
The fc class now calls scsi_target_unblock after calling the
terminate callback, so this patch removes the calls from the
drivers.
Signed-off-by: Mike Christie <michaelc@cs.wisc.edu>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Right now SCSI and others do their own command timeout handling.
Move those bits to the block layer.
Instead of having a timer per command, we try to be a bit more clever
and simply have one per-queue. This avoids the overhead of having to
tear down and setup a timer for each command, so it will result in a lot
less timer fiddling.
Signed-off-by: Mike Anderson <andmike@linux.vnet.ibm.com>
Signed-off-by: Jens Axboe <jens.axboe@oracle.com>
Bump driver version to 1.0.2.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
When logging async events, also print the payload in addition to the
event received.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Sanitize the response lengths in order to prevent possible oopses
in the command response path.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
If the client virtual fibre channel adapter is already logged into the
server and does an NPIV Login again, the async queue, which is used for
reporting Link Up/Link Down type of events, does not get reset on the
server side. Fix up the client driver so that we also do not reset it.
This fixes a problem of lost async events following relogins.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
If an ELS is received while the virtual fibre channel adapter is going
through its discovery, a flag is set which causes discovery to get
re-driven. However, the hosts's state does not get set back to
IBMVFC_INITIALIZING and scsi_block_requests does not get called again,
which can result in queuecommand ops getting sent during
discovery. This should not occur and may cause problems. One example
is that we may no longer be logged into the target we send the command
to, resulting in a failure which should not have occurred.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
This fixes a hang on module removal. The module removal code was setting
the hosts's state to IBMVFC_HOST_OFFLINE before tearing down the kernel
thread, but, due to a bug in ibmvfc_wait_while_resetting, was not waiting
for the kernel thread's offlining work to be done prior to destroying
the kernel thread, which left the scsi host in a blocked state which we
never got out of.
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
