Since the 'mfs' member has been declared as 'u32' in include/scsi/libfc.h,
use the %u format specifier instead of %hu. This patch fixes the following
clang compiler warning:
warning: format specifies type
'unsigned short' but the argument has type 'u32' (aka 'unsigned int')
[-Wformat]
"lport->mfs:%hu\n", mfs, lport->mfs);
~~~ ^~~~~~~~~~
%u
Link: https://lore.kernel.org/r/20210415220826.29438-8-bvanassche@acm.org
Cc: Hannes Reinecke <hare@suse.de>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Improve readability of the code in the SCSI core by introducing an
enumeration type for the values used internally that decide how to continue
processing a SCSI command. The eh_*_handler return values have not been
changed because that would involve modifying all SCSI drivers.
The output of the following command has been inspected to verify that no
out-of-range values are assigned to a variable of type enum
scsi_disposition:
KCFLAGS=-Wassign-enum make CC=clang W=1 drivers/scsi/
Link: https://lore.kernel.org/r/20210415220826.29438-6-bvanassche@acm.org
Cc: Christoph Hellwig <hch@lst.de>
Cc: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Cc: Hannes Reinecke <hare@suse.de>
Cc: Daniel Wagner <dwagner@suse.de>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The comment above scsi_send_eh_cmnd() says: "Returns SUCCESS or FAILED or
NEEDS_RETRY". This patch makes all values returned by scsi_send_eh_cmnd()
match the documentation of this function. This change does not affect the
behavior of scsi_eh_tur() nor of scsi_eh_try_stu() nor of the
scsi_request_sense() callers.
See also commit bbe9fb0d04 ("scsi: Avoid that .queuecommand() gets called
for a blocked SCSI device"; v5.3).
Link: https://lore.kernel.org/r/20210415220826.29438-5-bvanassche@acm.org
Cc: Christoph Hellwig <hch@lst.de>
Cc: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Cc: Hannes Reinecke <hare@suse.de>
Cc: Daniel Wagner <dwagner@suse.de>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Commit 320ae51fee ("blk-mq: new multi-queue block IO queueing mechanism";
v3.13) introduced a code path that calls the blk-mq completion function
from interrupt context. scsi-mq was introduced by commit d285203cf6
("scsi: add support for a blk-mq based I/O path."; v3.17).
Since the introduction of scsi-mq, scsi_softirq_done() can be called from
interrupt context. That made the name of the function misleading, rename it
to scsi_complete().
Link: https://lore.kernel.org/r/20210415220826.29438-4-bvanassche@acm.org
Cc: Christoph Hellwig <hch@lst.de>
Cc: Johannes Thumshirn <johannes.thumshirn@wdc.com>
Cc: Hannes Reinecke <hare@suse.de>
Cc: Daniel Wagner <dwagner@suse.de>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The qdio layer currently provides its own infrastructure to scan for
Request Queue completions & to report them to the device driver. This
comes with several drawbacks - having an async tasklet & timer construct in
qdio introduces additional lifetime complexity, and makes it harder to
integrate them with the rest of the device driver. The timeouts are also
currently hard-coded, and can't be tweaked without affecting other qdio
drivers (ie. qeth).
But due to recent enhancements to the qdio layer, zfcp can actually take
full control of the Request Queue completion processing. It merely needs to
opt-out from the qdio layer mechanisms by setting the scan_threshold to 0,
and then use qdio_inspect_queue() to scan for completions.
So re-implement the tasklet & timer mechanism in zfcp, while initially
copying the scan conditions from qdio's handle_outbound() and
qdio_outbound_tasklet(). One minor behavioural change is that
zfcp_qdio_send() will unconditionally reduce the timeout to 1 HZ, rather
than leaving it at 10 Hz if it was last armed by the tasklet. This just
makes things more consistent. Also note that we can drop a lot of the
accumulated cruft in qdio_outbound_tasklet(), as zfcp doesn't even use PCI
interrupt requests any longer.
This also slightly touches the Response Queue processing, as
qdio_get_next_buffers() will no longer implicitly scan for Request Queue
completions. So complete the migration to qdio_inspect_queue() here as well
and make the tasklet_schedule() visible.
Link: https://lore.kernel.org/r/018d3ddd029f8d6ac00cf4184880288c637c4fd1.1618417667.git.bblock@linux.ibm.com
Reviewed-by: Benjamin Block <bblock@linux.ibm.com>
Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: Benjamin Block <bblock@linux.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
When zfcp_adapter_enqueue() fails to create the zfcp_sysfs_adapter_attrs
group, it calls zfcp_adapter_unregister() to tear down the adapter state
again. This then unconditionally attempts to remove the
zfcp_sysfs_adapter_attrs group, resulting in a "group not found" WARN from
sysfs code.
Avoid this by copying most of zfcp_adapter_unregister() into the error
path, allowing for more fine-granular roll-back. Then skip the sysfs
tear-down steps if we haven't progressed this far in the initialization.
Link: https://lore.kernel.org/r/790922cc3af075795fff9a4b787e6bda19bdb3be.1618417667.git.bblock@linux.ibm.com
Reviewed-by: Benjamin Block <bblock@linux.ibm.com>
Reviewed-by: Steffen Maier <maier@linux.ibm.com>
Signed-off-by: Julian Wiedmann <jwi@linux.ibm.com>
Signed-off-by: Benjamin Block <bblock@linux.ibm.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Commit a6dcfe0848 ("scsi: qla2xxx: Limit interrupt vectors to number of
CPUs") lowers the number of allocated MSI-X vectors to the number of CPUs.
That breaks vector allocation assumptions in qla83xx_iospace_config(),
qla24xx_enable_msix() and qla2x00_iospace_config(). Either of the functions
computes maximum number of qpairs as:
ha->max_qpairs = ha->msix_count - 1 (MB interrupt) - 1 (default
response queue) - 1 (ATIO, in dual or pure target mode)
max_qpairs is set to zero in case of two CPUs and initiator mode. The
number is then used to allocate ha->queue_pair_map inside
qla2x00_alloc_queues(). No allocation happens and ha->queue_pair_map is
left NULL but the driver thinks there are queue pairs available.
qla2xxx_queuecommand() tries to find a qpair in the map and crashes:
if (ha->mqenable) {
uint32_t tag;
uint16_t hwq;
struct qla_qpair *qpair = NULL;
tag = blk_mq_unique_tag(cmd->request);
hwq = blk_mq_unique_tag_to_hwq(tag);
qpair = ha->queue_pair_map[hwq]; # <- HERE
if (qpair)
return qla2xxx_mqueuecommand(host, cmd, qpair);
}
BUG: kernel NULL pointer dereference, address: 0000000000000000
#PF: supervisor read access in kernel mode
#PF: error_code(0x0000) - not-present page
PGD 0 P4D 0
Oops: 0000 [#1] SMP PTI
CPU: 0 PID: 72 Comm: kworker/u4:3 Tainted: G W 5.10.0-rc1+ #25
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS 1.0.0-prebuilt.qemu-project.org 04/01/2014
Workqueue: scsi_wq_7 fc_scsi_scan_rport [scsi_transport_fc]
RIP: 0010:qla2xxx_queuecommand+0x16b/0x3f0 [qla2xxx]
Call Trace:
scsi_queue_rq+0x58c/0xa60
blk_mq_dispatch_rq_list+0x2b7/0x6f0
? __sbitmap_get_word+0x2a/0x80
__blk_mq_sched_dispatch_requests+0xb8/0x170
blk_mq_sched_dispatch_requests+0x2b/0x50
__blk_mq_run_hw_queue+0x49/0xb0
__blk_mq_delay_run_hw_queue+0xfb/0x150
blk_mq_sched_insert_request+0xbe/0x110
blk_execute_rq+0x45/0x70
__scsi_execute+0x10e/0x250
scsi_probe_and_add_lun+0x228/0xda0
__scsi_scan_target+0xf4/0x620
? __pm_runtime_resume+0x4f/0x70
scsi_scan_target+0x100/0x110
fc_scsi_scan_rport+0xa1/0xb0 [scsi_transport_fc]
process_one_work+0x1ea/0x3b0
worker_thread+0x28/0x3b0
? process_one_work+0x3b0/0x3b0
kthread+0x112/0x130
? kthread_park+0x80/0x80
ret_from_fork+0x22/0x30
The driver should allocate enough vectors to provide every CPU it's own HW
queue and still handle reserved (MB, RSP, ATIO) interrupts.
The change fixes the crash on dual core VM and prevents unbalanced QP
allocation where nr_hw_queues is two less than the number of CPUs.
Link: https://lore.kernel.org/r/20210412165740.39318-1-r.bolshakov@yadro.com
Fixes: a6dcfe0848 ("scsi: qla2xxx: Limit interrupt vectors to number of CPUs")
Cc: Daniel Wagner <daniel.wagner@suse.com>
Cc: Himanshu Madhani <himanshu.madhani@oracle.com>
Cc: Quinn Tran <qutran@marvell.com>
Cc: Nilesh Javali <njavali@marvell.com>
Cc: Martin K. Petersen <martin.petersen@oracle.com>
Cc: stable@vger.kernel.org # 5.11+
Reported-by: Aleksandr Volkov <a.y.volkov@yadro.com>
Reported-by: Aleksandr Miloserdov <a.miloserdov@yadro.com>
Reviewed-by: Daniel Wagner <dwagner@suse.de>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Roman Bolshakov <r.bolshakov@yadro.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
This fixes an issue hitting the BUG_ON() in ibmvfc_do_work(). When going
through a host action of IBMVFC_HOST_ACTION_RESET, we change the action to
IBMVFC_HOST_ACTION_TGT_DEL, then drop the host lock, and reset the CRQ,
which changes the host state to IBMVFC_NO_CRQ. If, prior to setting the
host state to IBMVFC_NO_CRQ, ibmvfc_init_host() is called, it can then end
up changing the host action to IBMVFC_HOST_ACTION_INIT. If we then change
the host state to IBMVFC_NO_CRQ, we will then hit the BUG_ON().
Make a couple of changes to avoid this. Leave the host action to be
IBMVFC_HOST_ACTION_RESET or IBMVFC_HOST_ACTION_REENABLE until after we drop
the host lock and reset or reenable the CRQ. Also harden the host state
machine to ensure we cannot leave the reset / reenable state until we've
finished processing the reset or reenable.
Link: https://lore.kernel.org/r/20210413001009.902400-1-tyreld@linux.ibm.com
Fixes: 73ee5d8672 ("[SCSI] ibmvfc: Fix soft lockup on resume")
Signed-off-by: Brian King <brking@linux.vnet.ibm.com>
[tyreld: added fixes tag]
Signed-off-by: Tyrel Datwyler <tyreld@linux.ibm.com>
[mkp: fix comment checkpatch warnings]
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Clean up minor issues spotted by tools and code review:
- Spelling Errors
- Spurious characters and errors in function headers
- nvme_info wqerr and err fields source data reversed
- Extraneous new line in log message 0466
- Spacing error in log message 0109
- Messages 0140 and 0141 have portname and nodename reversed
- Incorrect function labelling in comment
Link: https://lore.kernel.org/r/20210412013127.2387-13-jsmart2021@gmail.com
Co-developed-by: Justin Tee <justin.tee@broadcom.com>
Signed-off-by: Justin Tee <justin.tee@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The lpfc_hdw_queue attribute is to set the number of hardware queues to be
created on the adapter. Normally, the value is set to a default, which
allows the hw queue count to be sized dynamically based on adapter
capabilities, CPU/platform architecture, or CPU type. Currently, when
lpfc_hdw_queue is set to a specific value, is has no effect and the dynamic
sizing occurs.
The routine checking whether parameters are default or not ignores the
lpfc_hdw_queue setting and invokes the dynamic logic.
Fix the routine to additionally check the lpfc_hdw_queue attribute value
before using dynamic scaling. Additionally, SLI-3 supports only a small
number of queues with dedicated functions, thus it needs to be exempted
from the variable scaling and set to the expected values.
Link: https://lore.kernel.org/r/20210412013127.2387-11-jsmart2021@gmail.com
Co-developed-by: Justin Tee <justin.tee@broadcom.com>
Signed-off-by: Justin Tee <justin.tee@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
FDMI registration needs to be performed after every login with the FC Mgmt
service. The flag the driver is using to track registration is cleared on
link up, but never on Mgmt service logout/re-login.
Fix by clearing the flag whenever a new login is completed with the FC Mgmt
service.
While perusing the flag use, logging was performed as if FDMI registration
occurred on vports. However, it is limited to the physical port only.
Revise the logging to reflect physical port based.
Link: https://lore.kernel.org/r/20210412013127.2387-10-jsmart2021@gmail.com
Co-developed-by: Justin Tee <justin.tee@broadcom.com>
Signed-off-by: Justin Tee <justin.tee@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
During target port swap, the swap logic ignores the DROPPED flag in the
nodes. As a node then moves into the UNUSED state, the reference count will
be dropped. If a node is later reused and moved out of the UNUSED state, an
access can result in a use-after-free assert.
Fix by having the port swap logic propagate the DROPPED flag when switching
nodes. This will avoid reference from being dropped.
Link: https://lore.kernel.org/r/20210412013127.2387-8-jsmart2021@gmail.com
Co-developed-by: Justin Tee <justin.tee@broadcom.com>
Signed-off-by: Justin Tee <justin.tee@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
In SLI-4, when performing a mailbox command with MBX_POLL, the driver uses
the BMBX register to send the command rather than the MQ. A flag is set
indicating the BMBX register is active and saves the mailbox job struct
(mboxq) in the mbox_active element of the adapter. The routine then waits
for completion or timeout. The mailbox job struct is not freed by the
routine. In cases of timeout, the adapter will be reset. The
lpfc_sli_mbox_sys_flush() routine will clean up the mbox in preparation for
the reset. It clears the BMBX active flag and marks the job structure as
MBX_NOT_FINISHED. But, it never frees the mboxq job structure. Expectation
in both normal completion and timeout cases is that the issuer of the mbx
command will free the structure. Unfortunately, not all calling paths are
freeing the memory in cases of error.
All calling paths were looked at and updated, if missing, to free the mboxq
memory regardless of completion status.
Link: https://lore.kernel.org/r/20210412013127.2387-7-jsmart2021@gmail.com
Co-developed-by: Justin Tee <justin.tee@broadcom.com>
Signed-off-by: Justin Tee <justin.tee@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
During target port-swap testing with link flips, the initiator could
encounter PRLI errors. If the target node disappears permanently, the ndlp
is found stuck in UNUSED state with ref count of 1. The rmmod of the driver
will hang waiting for this node to be freed.
While handling a link error in PRLI completion path, the code intends to
skip triggering the discovery state machine. However this is causing the
final reference release path to be skipped. This causes the node to be
stuck with ref count of 1
Fix by ensuring the code path triggers the device removal event on the node
state machine.
Link: https://lore.kernel.org/r/20210412013127.2387-6-jsmart2021@gmail.com
Co-developed-by: Justin Tee <justin.tee@broadcom.com>
Signed-off-by: Justin Tee <justin.tee@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Call traces are being seen that result from a nodelist structure ref
counting error. They are typically seen after transmission of an LS_RJT ELS
response.
Aged code in lpfc_cmpl_els_rsp() calls lpfc_nlp_not_used() which, if the
ndlp reference count is exactly 1, will decrement the reference count.
Previously lpfc_nlp_put() was within lpfc_els_free_iocb(), and the 'put'
within the free would only be invoked if cmdiocb->context1 was not NULL.
Since the nodelist structure reference count is decremented when exiting
lpfc_cmpl_els_rsp() the lpfc_nlp_not_used() calls are no longer required.
Calling them is causing the reference count issue.
Fix by removing the lpfc_nlp_not_used() calls.
Link: https://lore.kernel.org/r/20210412013127.2387-4-jsmart2021@gmail.com
Co-developed-by: Justin Tee <justin.tee@broadcom.com>
Signed-off-by: Justin Tee <justin.tee@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Fix a crash caused by a double put on the node when the driver completed an
ACC for an unsolicted abort on the same node. The second put was executed
by lpfc_nlp_not_used() and is wrong because the completion routine executes
the nlp_put when the iocbq was released. Additionally, the driver is
issuing a LOGO then immediately calls lpfc_nlp_set_state to put the node
into NPR. This call does nothing.
Remove the lpfc_nlp_not_used call and additional set_state in the
completion routine. Remove the lpfc_nlp_set_state post issue_logo. Isn't
necessary.
Link: https://lore.kernel.org/r/20210412013127.2387-3-jsmart2021@gmail.com
Co-developed-by: Justin Tee <justin.tee@broadcom.com>
Signed-off-by: Justin Tee <justin.tee@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Rmmod on SLI-4 adapters is sometimes hitting a bad ptr dereference in
lpfc_els_free_iocb().
A prior patch refactored the lpfc_sli_abort_iocb() routine. One of the
changes was to convert from building/sending an abort within the routine to
using a common routine. The reworked routine passes, without modification,
the pring ptr to the new common routine. The older routine had logic to
check SLI-3 vs SLI-4 and adapt the pring ptr if necessary as callers were
passing SLI-3 pointers even when not on an SLI-4 adapter. The new routine
is missing this check and adapt, so the SLI-3 ring pointers are being used
in SLI-4 paths.
Fix by cleaning up the calling routines. In review, there is no need to
pass the ring ptr argument to abort_iocb at all. The routine can look at
the adapter type itself and reference the proper ring.
Link: https://lore.kernel.org/r/20210412013127.2387-2-jsmart2021@gmail.com
Fixes: db7531d2b3 ("scsi: lpfc: Convert abort handling to SLI-3 and SLI-4 handlers")
Cc: <stable@vger.kernel.org> # v5.11+
Co-developed-by: Justin Tee <justin.tee@broadcom.com>
Signed-off-by: Justin Tee <justin.tee@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
