Typical SLI-4 hardware supports up to 2 4KB pages to be registered per XRI
to contain the exchanges Scatter/Gather List. This caps the number of SGL
elements that can be in the SGL. There are not extensions to extend the
list out of the 2 pages.
The G7 hardware adds a SGE type that allows the SGL to be vectored to a
different scatter/gather list segment. And that segment can contain a SGE
to go to another segment and so on. The initial segment must still be
pre-registered for the XRI, but it can be a much smaller amount (256Bytes)
as it can now be dynamically grown. This much smaller allocation can
handle the SG list for most normal I/O, and the dynamic aspect allows it to
support many MB's if needed.
The implementation creates a pool which contains "segments" and which is
initially sized to hold the initial small segment per xri. If an I/O
requires additional segments, they are allocated from the pool. If the
pool has no more segments, the pool is grown based on what is now
needed. After the I/O completes, the additional segments are returned to
the pool for use by other I/Os. Once allocated, the additional segments are
not released under the assumption of "if needed once, it will be needed
again". Pools are kept on a per-hardware queue basis, which is typically
1:1 per cpu, but may be shared by multiple cpus.
The switch to the smaller initial allocation significantly reduces the
memory footprint of the driver (which only grows if large ios are
issued). Based on the several K of XRIs for the adapter, the 8KB->256B
reduction can conserve 32MBs or more.
It has been observed with per-cpu resource pools that allocating a resource
on CPU A, may be put back on CPU B. While the get routines are distributed
evenly, only a limited subset of CPUs may be handling the put routines.
This can put a strain on the lpfc_put_cmd_rsp_buf_per_cpu routine because
all the resources are being put on a limited subset of CPUs.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Added code to support driver loopback with MDS Diagnostics. This style of
diagnostics passes frames from the fabric to the driver who then echo them
back out the link. SEND_FRAME WQEs are used to transmit the frames. Added
the SOF and EOF field location definitions for use by SEND_FRAME.
Also ensure that enable_mds_diags is a RW parameter.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
In order to see real addresses, convert %p with %px for kernel addresses
and replace %p with %pf for functions.
While converting, standardize on "x%px" throughout (not %px or 0x%px).
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Running on Coverity produced the following errors:
- coding style (indentation)
- memset size mismatch errors
note: comment cases where it is purposely a mismatch
Fix the errors.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
As part of firmware download, the adapter is reset. On the adapter the
reset causes the function to stop and all outstanding io is terminated
(without responses). The reset path then starts teardown of the adapter,
starting with deregistration of the remote ports with the nvme-fc
transport. The local port is then deregistered and the driver waits for
local port deregistration. This never finishes.
The remote port deregistrations terminated the nvme controllers, causing
them to send aborts for all the outstanding io. The aborts were serviced in
the driver, but stalled due to its state. The nvme layer then stops to
reclaim it's outstanding io before continuing. The io must be returned
before the reset on the controller is deemed complete and the controller
delete performed. The remote port deregistration won't complete until all
the controllers are terminated. And the local port deregistration won't
complete until all controllers and remote ports are terminated. Thus things
hang.
The issue is the reset which stopped the adapter also stopped all the
responses that would drive i/o completions, and the aborts were also
stopped that stopped i/o completions. The driver, when resetting the
adapter like this, needs to be generating the completions as part of the
adapter reset so that I/O complete (in error), and any aborts are not
queued.
Fix by adding flush routines whenever the adapter port has been reset or
discovered in error. The flush routines will generate the completions for
the scsi and nvme outstanding io. The abort ios, if waiting, will be caught
and flushed as well.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
If the adapter encounters a condition which causes the adapter to fail
(driver must detect the failure) simultaneously to a request to the driver
to reset the adapter (such as a host_reset), the reset path will be racing
with the asynchronously-detect adapter failure path. In the failing
situation, one path has started to tear down the adapter data structures
(io_wq's) while the other path has initiated a repeat of the teardown and
is in the lpfc_sli_flush_xxx_rings path and attempting to access the
just-freed data structures.
Fix by the following:
- In cases where an adapter failure is detected, rather than explicitly
calling offline_eratt() to start the teardown, change the adapter state
and let the later calls of posted work to the slowpath thread invoke the
adapter recovery. In essence, this means all requests to reset are
serialized on the slowpath thread.
- Clean up the routine that restarts the adapter. If there is a failure
from brdreset, don't immediately error and leave things in a partial
state. Instead, ensure the adapter state is set and finish the teardown
of structures before returning.
- If in the scsi host reset handler and the board fails to reset and
restart (which can be due to parallel reset/recovery paths), instead of
hard failing and explicitly calling offline_eratt() (which gets into the
redundant path), just fail out and let the asynchronous path resolve the
adapter state.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
In tests with remote ports contantly logging out/logging coupled with
occassional local link bounce, if a remote port is disocnnected for longer
than devloss_tmo and then subsequently reconnected, eventually the test
will fail to login with the remote port and remote port connectivity is
lost.
When devloss_tmo expires, the driver does not free the node struct until
the port or npiv instances is being deleted. The node is left allocated but
the state set to UNUSED. If the node was in the process of logging in when
the local link drop occurred, meaning the RPI was allocated for the node in
order to send the ELS, but not yet registered which comes after successful
login, the node is moved to the NPR state, and if devloss expires, to
UNUSED state. If the remote port comes back, the node associated with it
is restarted and this path happens to allocate a new RPI and overwrites the
prior RPI value. In the cases where the port was logged in and loggs out,
the path did release the RPI but did not set the node rpi value. In the
cases where the remote port never finished logging in, the path never did
the call to release the rpi. In this latter case, when the node is
subsequently restore, the new rpi allocation overwrites the rpi that was
not released, and the rpi is now leaked. Eventually the port will run out
of RPI resources to log into new remote ports.
Fix by following changes:
- When an rpi is released, do so under locks and ensure the node rpi value
is set to a non-allocated value (LPFC_RPI_ALLOC_ERROR). Note:
refactored to a small service routine to avoid indentation issues.
- When re-enabling a node, check the rpi value to determine if a new
allocation is necessary. If already set, use the prior rpi.
Enhanced logging to help in the future.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The adapter reset path (lpfc_sli_hba_down) is taking/releasing a lock with
irq. But, the path is already under the hbalock which raised irq so it's
unnecessary.
Convert to simple lock/unlock.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
If the adapter is reset while there are outstanding ELS's, subsequent
reinitialization of the adapter will fail as it has not recovered all of
the io contexts relative to the ELS's.
If an ELS timed out or otherwise failed and an the ELS was attempted to be
aborted (which changes the ELS completion context), in causes where the
driver generates completions for the outstanding IO as the adapter would
not due to being reset, the driver released only the ELS context and failed
to release the abort context. When the adapter went to reinit, as it had
not received all of the contexts, it failed to reinit.
Fix by having the ELS completion handler identify the driver-generated
completion status and release the abort context.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
When connected to a high number of remote ports, the driver is encountering
PLOGI errors. The errors are due to adapter detected failures indicating
illegal field values.
Turns out the driver was prematurely clearing an RPI bitmask before waiting
for an UNREG_RPI mailbox completion. This allowed the RPI to be reused
before it was actually available.
Fix by clearing RPI bitmask only after UNREG_RPI mailbox completion.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Remove the redundant initialization code.
Signed-off-by: Fuqian Huang <huangfq.daxian@gmail.com>
Reviewed-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
This is mostly update of the usual drivers: qla2xxx, hpsa, lpfc, ufs,
mpt3sas, ibmvscsi, megaraid_sas, bnx2fc and hisi_sas as well as the
removal of the osst driver (I heard from Willem privately that he
would like the driver removed because all his test hardware has
failed). Plus number of minor changes, spelling fixes and other
trivia.
Signed-off-by: James E.J. Bottomley <jejb@linux.ibm.com>
-----BEGIN PGP SIGNATURE-----
iJwEABMIAEQWIQTnYEDbdso9F2cI+arnQslM7pishQUCXSTl4yYcamFtZXMuYm90
dG9tbGV5QGhhbnNlbnBhcnRuZXJzaGlwLmNvbQAKCRDnQslM7pishdcxAQDCJVbd
fPUX76/V1ldupunF97+3DTharxxbst+VnkOnCwD8D4c0KFFFOI9+F36cnMGCPegE
fjy17dQLvsJ4GsidHy8=
=aS5B
-----END PGP SIGNATURE-----
Merge tag 'scsi-misc' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
Pull SCSI updates from James Bottomley:
"This is mostly update of the usual drivers: qla2xxx, hpsa, lpfc, ufs,
mpt3sas, ibmvscsi, megaraid_sas, bnx2fc and hisi_sas as well as the
removal of the osst driver (I heard from Willem privately that he
would like the driver removed because all his test hardware has
failed). Plus number of minor changes, spelling fixes and other
trivia.
The big merge conflict this time around is the SPDX licence tags.
Following discussion on linux-next, we believe our version to be more
accurate than the one in the tree, so the resolution is to take our
version for all the SPDX conflicts"
Note on the SPDX license tag conversion conflicts: the SCSI tree had
done its own SPDX conversion, which in some cases conflicted with the
treewide ones done by Thomas & co.
In almost all cases, the conflicts were purely syntactic: the SCSI tree
used the old-style SPDX tags ("GPL-2.0" and "GPL-2.0+") while the
treewide conversion had used the new-style ones ("GPL-2.0-only" and
"GPL-2.0-or-later").
In these cases I picked the new-style one.
In a few cases, the SPDX conversion was actually different, though. As
explained by James above, and in more detail in a pre-pull-request
thread:
"The other problem is actually substantive: In the libsas code Luben
Tuikov originally specified gpl 2.0 only by dint of stating:
* This file is licensed under GPLv2.
In all the libsas files, but then muddied the water by quoting GPLv2
verbatim (which includes the or later than language). So for these
files Christoph did the conversion to v2 only SPDX tags and Thomas
converted to v2 or later tags"
So in those cases, where the spdx tag substantially mattered, I took the
SCSI tree conversion of it, but then also took the opportunity to turn
the old-style "GPL-2.0" into a new-style "GPL-2.0-only" tag.
Similarly, when there were whitespace differences or other differences
to the comments around the copyright notices, I took the version from
the SCSI tree as being the more specific conversion.
Finally, in the spdx conversions that had no conflicts (because the
treewide ones hadn't been done for those files), I just took the SCSI
tree version as-is, even if it was old-style. The old-style conversions
are perfectly valid, even if the "-only" and "-or-later" versions are
perhaps more descriptive.
* tag 'scsi-misc' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi: (185 commits)
scsi: qla2xxx: move IO flush to the front of NVME rport unregistration
scsi: qla2xxx: Fix NVME cmd and LS cmd timeout race condition
scsi: qla2xxx: on session delete, return nvme cmd
scsi: qla2xxx: Fix kernel crash after disconnecting NVMe devices
scsi: megaraid_sas: Update driver version to 07.710.06.00-rc1
scsi: megaraid_sas: Introduce various Aero performance modes
scsi: megaraid_sas: Use high IOPS queues based on IO workload
scsi: megaraid_sas: Set affinity for high IOPS reply queues
scsi: megaraid_sas: Enable coalescing for high IOPS queues
scsi: megaraid_sas: Add support for High IOPS queues
scsi: megaraid_sas: Add support for MPI toolbox commands
scsi: megaraid_sas: Offload Aero RAID5/6 division calculations to driver
scsi: megaraid_sas: RAID1 PCI bandwidth limit algorithm is applicable for only Ventura
scsi: megaraid_sas: megaraid_sas: Add check for count returned by HOST_DEVICE_LIST DCMD
scsi: megaraid_sas: Handle sequence JBOD map failure at driver level
scsi: megaraid_sas: Don't send FPIO to RL Bypass queue
scsi: megaraid_sas: In probe context, retry IOC INIT once if firmware is in fault
scsi: megaraid_sas: Release Mutex lock before OCR in case of DCMD timeout
scsi: megaraid_sas: Call disable_irq from process IRQ poll
scsi: megaraid_sas: Remove few debug counters from IO path
...
This patch adds general RSCN support:
- The ability to transmit an RSCN to the port on the other end of
the link (regular port if pt2pt, or fabric controller if fabric).
- And general recognition of an RSCN ELS when an ELS is received.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Reviewed-by: Arun Easi <aeasi@marvell.com>
Signed-off-by: Sagi Grimberg <sagi@grimberg.me>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Fix sparse warnings:
drivers/scsi/lpfc/lpfc_sli.c:115:1: warning: symbol 'lpfc_sli4_pcimem_bcopy' was not declared. Should it be static?
drivers/scsi/lpfc/lpfc_sli.c:7854:1: warning: symbol 'lpfc_sli4_process_missed_mbox_completions' was not declared. Should it be static?
drivers/scsi/lpfc/lpfc_nvmet.c:223:27: warning: symbol 'lpfc_nvmet_get_ctx_for_xri' was not declared. Should it be static?
drivers/scsi/lpfc/lpfc_nvmet.c:245:27: warning: symbol 'lpfc_nvmet_get_ctx_for_oxid' was not declared. Should it be static?
drivers/scsi/lpfc/lpfc_init.c:75:10: warning: symbol 'lpfc_present_cpu' was not declared. Should it be static?
Reported-by: Hulk Robot <hulkci@huawei.com>
Signed-off-by: YueHaibing <yuehaibing@huawei.com>
Acked-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
While fixing the resources per socket, realized the driver was not using
hardware queues (up to 1 per cpu) if there were fewer interrupt
vectors. The driver was only using the hardware queue assigned to the cpu
with the vector.
Rework the affinity map check to use the additional hardware queue elements
that had been allocated. If the cpu count exceeds the hardware queue count
- share, but choose what is shared with by: hyperthread peer, core peer,
socket peer, or finally similar cpu in a different socket.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
eq create is leaking mailbox memory if it encounters an error.
rework error path to free the memory.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Currently the driver is notified of new command frame receipt by CQEs. As
part of the CQE processing, the driver upcalls the nvmet_fc transport to
deliver the command. nvmet_fc, as part of receiving the command builds out
a context for it, where one of the first steps is to allocate memory for
the io.
When running with tests that do large ios (1MB), it was found on some
systems, the total number of outstanding I/O's, at 1MB per, completely
consumed the system's memory. Thus additional ios were getting blocked in
the memory allocator. Given that this blocked the lpfc thread processing
CQEs, there were lots of other commands that were received and which are
then held up, and given CQEs are serially processed, the aggregate delays
for an IO waiting behind the others became cummulative - enough so that the
initiator hit timeouts for the ios.
The basic fix is to avoid the direct upcall and instead schedule a work
item for each io as it is received. This allows the cq processing to
complete very quickly, and each io can then run or block on it's own.
However, this general solution hurts latency when there are few ios. As
such, implemented the fix such that the driver watches how many CQEs it has
processed sequentially in one run. As long as the count is below a
threshold, the direct nvmet_fc upcall will be made. Only when the count is
exceeded will it revert to work scheduling.
Given that debug of this showed a surprisingly long delay in cq processing,
the io timer stats were updated to better reflect the processing of the
different points.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
There were a number of erroneous comments and incorrect older lockdep
checks that were causing a number of warnings.
Resolve the following:
- Inconsistent lock state warnings in lpfc_nvme_info_show().
- Fixed comments and code on sequences where ring lock is now held instead
of hbalock.
- Reworked calling sequences around lpfc_sli_iocbq_lookup(). Rather than
locking prior to the routine and have routine guess on what lock, take
the lock within the routine. The lockdep check becomes unnecessary.
- Fixed comments and removed erroneous hbalock checks.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
CC: Bart Van Assche <bvanassche@acm.org>
Tested-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
This patch avoids that a kernel warning appears when smp_processor_id() is
called with preempt debugging enabled.
Cc: James Smart <james.smart@broadcom.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Acked-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Remove those functions that are not called from outside the removed
functions.
Cc: James Smart <james.smart@broadcom.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Acked-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
This patch avoid that smatch complains about misleading indentation.
Cc: James Smart <james.smart@broadcom.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Acked-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
This patch avoids that the compiler complains about missing declarations
when building with W=1.
Cc: James Smart <james.smart@broadcom.com>
Signed-off-by: Bart Van Assche <bvanassche@acm.org>
Acked-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Declaring interrupt clear routines as inline is bogus as they are used as
an indirect pointer.
Remove the inline references.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
You can't declare a function inline in a header if it doesn't have a body
available to the compiler. So realistically you either don't declare it
inline or you make it a static inline in the header. I think the latter
applies in this case, so this should be the fix
Signed-off-by: James Bottomley <James.Bottomley@HansenPartnership.com>
Acked-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Change the SLI4 queue creation code to use NUMA node based memory
allocation based on the cpu the queues will be related to.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Currently the driver maintains a sideband structure which has a pointer for
each queue element. However, at 8 bytes per pointer, and up to 4k elements
per queue, and 100s of queues, this can take up a lot of memory.
Convert the driver to using an access routine that calculates the element
address based on its index rather than using the pointer table.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The driver is currently reporting the firmware revision not the actual boot
bios version in FDMI data.
Modify the driver to obtain the boot bios version from the adapter and use
that data in the FMDI data sent to the switch.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The adapter initialization sequence enables interrupts, initializes the
adapter link_state to LINK_DOWN, then issues commands to initialize the
adapter. The interrupt handler on the adapter validates the link_state (has
to be at least LINK_DOWN) and if invalid, will discard the interrupting
event.
In most cases, there is not a command completion, thus an interrupt until
the initialization commands have been sent which is post the setting of
state to LINK_DOWN. However, in cases of firmware reset, the reset will
modify the link_state to an invalid value (indicating a reset of the
adapter) and there occasionally are cases where the adapter will generate
an asynchronous event which shares the eq/cq used for mailbox commands. In
the failure case, an interrupt is generated immediately after enabling them
due to the async event. As link_state is invalid, the eq is list and the
CQ not serviced. At this point link_state is initialized and the mailbox
command sent. As the CQ has not been serviced, it is not armed, so no
interrupt event is generated when the mailbox command completes.
Modify the initialization sequence so that interrupts are enabled after
link_state is properly initialized, which avoids the race condition with
the async event.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Current code is using msleep when polling for hw ready. Unfortunately the
msleep routine isn't very accurate on rescheduling. In fact, on a busy
systems which reset the adapter, it became 10s of seconds before it was
rescheduled.
Fix by busy waiting using udelay. As we're now busy waiting, significantly
reduce the wait time so that we can exit the pool loop as soon as possible.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The driver periodically checks for adapter error in a background thread. If
the thread detects an error, the adapter will be reset including the
deletion and reallocation of workqueues on the adapter. Simultaneously,
there may be a user-space request to offline the adapter which may try to
do many of the same steps, in parallel, on a different thread. As memory
was deallocated while unexpected, the parallel offline request hit a bad
pointer.
Add coordination between the two threads. The error recovery thread has
precedence. So, when an error is detected, a flag is set on the adapter to
indicate the error thread is terminating the adapter. But, before doing
that work, it will look for a flag that is set by the offline flow, and if
set, will wait for it to complete before then processing the error handling
path. Similarly, in the offline thread, it first checks for whether the
error thread is resetting the adapter, and if so, will then wait for the
error thread to finish. Only after it has finished, will it set its flag
and offline the adapter.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
In a couple of cases, the driver detected a pci error (via pci device state
or via failed register reads) but didn't take any action to disable the
device. Additionally, the driver is ignoring the status of pci
configuration space reads.
Having the driver take the adapter offline whenever the pci error is
detected. Pay attention to pci_config_space_read status and return failure
if an error is seen.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
With negative test injection, the driver is receiving a command with first
burst enabled, meaning Sequence initiative is not passed with the command
frame. The driver notes the condition and discards the frame. However the
driver calls the incorrect buffer free routine, resulting in a NULL pointer
reference.
For hbq buffer free, convert to using lpfc_rq_buf_free().
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
When unloading the driver, mailbox commands may be sent without holding a
reference on the ndlp. By the time the mailbox command completes, the ndlp
may have reduced its ref counts and been freed. The problem was reported
by KASAN.
While unregistering due to driver unload, have the completion noop'd by
setting the ndlp context NULL'd. Due to the unload, no further action was
necessary. Also, while reviewing this path, the generic nulling of the
context after handling should be slightly moved.
Reported by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
This is the final round of mostly small fixes and performance
improvements to our initial submit. The main regression fix is the
ia64 simscsi build failure which was missed in the serial number
elimination conversion.
Signed-off-by: James E.J. Bottomley <jejb@linux.ibm.com>
-----BEGIN PGP SIGNATURE-----
iJwEABMIAEQWIQTnYEDbdso9F2cI+arnQslM7pishQUCXIxBayYcamFtZXMuYm90
dG9tbGV5QGhhbnNlbnBhcnRuZXJzaGlwLmNvbQAKCRDnQslM7pisherpAP4rxLpX
bcUnQnEsvoxys/JyoK08Qfv1JebZo1B2MAZ62wD/VZ7LpOuzVLhsM2KhLFGRrs1/
7D2K4tgtO2dQsFix7H0=
=pcHl
-----END PGP SIGNATURE-----
Merge tag 'scsi-misc' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
Pull more SCSI updates from James Bottomley:
"This is the final round of mostly small fixes and performance
improvements to our initial submit.
The main regression fix is the ia64 simscsi build failure which was
missed in the serial number elimination conversion"
* tag 'scsi-misc' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi: (24 commits)
scsi: ia64: simscsi: use request tag instead of serial_number
scsi: aacraid: Fix performance issue on logical drives
scsi: lpfc: Fix error codes in lpfc_sli4_pci_mem_setup()
scsi: libiscsi: Hold back_lock when calling iscsi_complete_task
scsi: hisi_sas: Change SERDES_CFG init value to increase reliability of HiLink
scsi: hisi_sas: Send HARD RESET to clear the previous affiliation of STP target port
scsi: hisi_sas: Set PHY linkrate when disconnected
scsi: hisi_sas: print PHY RX errors count for later revision of v3 hw
scsi: hisi_sas: Fix a timeout race of driver internal and SMP IO
scsi: hisi_sas: Change return variable type in phy_up_v3_hw()
scsi: qla2xxx: check for kstrtol() failure
scsi: lpfc: fix 32-bit format string warning
scsi: lpfc: fix unused variable warning
scsi: target: tcmu: Switch to bitmap_zalloc()
scsi: libiscsi: fall back to sendmsg for slab pages
scsi: qla2xxx: avoid printf format warning
scsi: lpfc: resolve static checker warning in lpfc_sli4_hba_unset
scsi: lpfc: Correct __lpfc_sli_issue_iocb_s4 lockdep check
scsi: ufs: hisi: fix ufs_hba_variant_ops passing
scsi: qla2xxx: Fix panic in qla_dfs_tgt_counters_show
...
This is mostly update of the usual drivers: arcmsr, qla2xxx, lpfc,
hisi_sas, target/iscsi and target/core. Additionally Christoph
refactored gdth as part of the dma changes. The major mid-layer
change this time is the removal of bidi commands and with them the
whole of the osd/exofs driver and filesystem.
Signed-off-by: James E.J. Bottomley <jejb@linux.ibm.com>
-----BEGIN PGP SIGNATURE-----
iJwEABMIAEQWIQTnYEDbdso9F2cI+arnQslM7pishQUCXIC54SYcamFtZXMuYm90
dG9tbGV5QGhhbnNlbnBhcnRuZXJzaGlwLmNvbQAKCRDnQslM7pishT1GAPwJEV23
ExPiPsnuVgKj49nLTagZ3rILRQcYNbL+MNYqxQEA0cT8FHzSDBfWY5OKPNE+RQ8z
f69LpXGmMpuagKGvvd4=
=Fhy1
-----END PGP SIGNATURE-----
Merge tag 'scsi-misc' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi
Pull SCSI updates from James Bottomley:
"This is mostly update of the usual drivers: arcmsr, qla2xxx, lpfc,
hisi_sas, target/iscsi and target/core.
Additionally Christoph refactored gdth as part of the dma changes. The
major mid-layer change this time is the removal of bidi commands and
with them the whole of the osd/exofs driver and filesystem. This is a
major simplification for block and mq in particular"
* tag 'scsi-misc' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi: (240 commits)
scsi: cxgb4i: validate tcp sequence number only if chip version <= T5
scsi: cxgb4i: get pf number from lldi->pf
scsi: core: replace GFP_ATOMIC with GFP_KERNEL in scsi_scan.c
scsi: mpt3sas: Add missing breaks in switch statements
scsi: aacraid: Fix missing break in switch statement
scsi: kill command serial number
scsi: csiostor: drop serial_number usage
scsi: mvumi: use request tag instead of serial_number
scsi: dpt_i2o: remove serial number usage
scsi: st: osst: Remove negative constant left-shifts
scsi: ufs-bsg: Allow reading descriptors
scsi: ufs: Allow reading descriptor via raw upiu
scsi: ufs-bsg: Change the calling convention for write descriptor
scsi: ufs: Remove unused device quirks
Revert "scsi: ufs: disable vccq if it's not needed by UFS device"
scsi: megaraid_sas: Remove a bunch of set but not used variables
scsi: clean obsolete return values of eh_timed_out
scsi: sd: Optimal I/O size should be a multiple of physical block size
scsi: MAINTAINERS: SCSI initiator and target tweaks
scsi: fcoe: make use of fip_mode enum complete
...
The outer routine lpfc_sli_issue_iocb(), which decomposes into the
SLI3 (s3) or SLI4 (s4) subroutines takes out the locks. For s3, it takes
out the hbalock. For s4, it takes out the ring_lock. The lockdep check in
the s3 and s4 subroutines both check hbalock, which is incorrect for s4.
Revise the s4 subroutine to lockdep check the ring_lock.
Reported-by: Bart Van Assche <bvanassche@acm.org>
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
There are a handful of statements that are indented incorrectly. Fix these.
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Acked-by: James Smart <james.smart@broadcom.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
For files modified as part of 12.2.0.0 patches, update copyright to 2019
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
A scsi host lock is taken on every io completion to check whether the abort
handler is waiting on the io completion. This is an expensive lock to take
on all completion when rarely in an abort condition.
Replace scsi host lock with command-specific lock. Synchronize completion
and abort paths by new cmd lock. Ensure all flag changing and nulling of
context pointers taken under lock. When adding lock to task management
abort, realized it was missing other synchronization locks. Added that
synchronization to match normal paths.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
When driving high iop counts, auto_imax coalescing kicks in and drives the
performance to extremely small iops levels.
There are two issues:
1) auto_imax is enabled by default. The auto algorithm, when iops gets
high, divides the iops by the hdwq count and uses that value to
calculate EQ_Delay. The EQ_Delay is set uniformly on all EQs whether
they have load or not. The EQ_delay is only manipulated every 5s (a
long time). Thus there were large 5s swings of no interrupt delay
followed by large/maximum delay, before repeating.
2) When processing a CQ, the driver got mixed up on the rate of when
to ring the doorbell to keep the chip appraised of the eqe or cqe
consumption as well as how how long to sit in the thread and
process queue entries. Currently, the driver capped its work at
64 entries (very small) and exited/rearmed the CQ. Thus, on heavy
loads, additional overheads were taken to exit and re-enter the
interrupt handler. Worse, if in the large/maximum coalescing
windows,k it could be a while before getting back to servicing.
The issues are corrected by the following:
- A change in defaults. Auto_imax is turned OFF and fcp_imax is set
to 0. Thus all interrupts are immediate.
- Cleanup of field names and their meanings. Existing names were
non-intuitive or used for duplicate things.
- Added max_proc_limit field, to control the length of time the
handlers would service completions.
- Reworked EQ handling:
Added common routine that walks eq, applying notify interval and max
processing limits. Use queue_claimed to claim ownership of the queue
while processing. Always rearm the queue whenever the common routine
is called.
Rework queue element processing, namely to eliminate hba_index vs
host_index. Only one index is necessary. The queue entry can be
marked invalid and the host_index updated immediately after eqe
processing.
After rework, xx_release routines are now DB write functions. Renamed
the routines as such.
Moved lpfc_sli4_eq_flush(), which does similar action, to same area.
Replaced the 2 individual loops that walk an eq with a call to the
common routine.
Slightly revised lpfc_sli4_hba_handle_eqe() calling syntax.
Added per-cpu counters to detect interrupt rates and scale
interrupt coalescing values.
- Reworked CQ handling:
Added common routine that walks cq, applying notify interval and max
processing limits. Use queue_claimed to claim ownership of the queue
while processing. Always rearm the queue whenever the common routine
is called.
Rework queue element processing, namely to eliminate hba_index vs
host_index. Only one index is necessary. The queue entry can be
marked invalid and the host_index updated immediately after cqe
processing.
After rework, xx_release routines are now DB write functions. Renamed
the routines as such.
Replaced the 3 individual loops that walk a cq with a call to the
common routine.
Redefined lpfc_sli4_sp_handle_mcqe() to commong handler definition with
queue reference. Add increment for mbox completion to handler.
- Added a new module/sysfs attribute: lpfc_cq_max_proc_limit To allow
dynamic changing of the CQ max_proc_limit value being used.
Although this leaves an EQ as an immediate interrupt, that interrupt will
only occur if a CQ bound to it is in an armed state and has cqe's to
process. By staying in the cq processing routine longer, high loads will
avoid generating more interrupts as they will only rearm as the processing
thread exits. The immediately interrupt is also beneficial to idle or
lower-processing CQ's as they get serviced immediately without being
penalized by sharing an EQ with a more loaded CQ.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Review of the eq coalescing logic showed the code was a bit fragmented.
Sometimes it would save/set via an interrupt max value, while in others it
would do so via a usdelay. There were also two places changing eq delay,
one place that issued mailbox commands, and another that changed via
register writes if supported.
Clean this up by:
- Standardizing the operation of lpfc_modify_hba_eq_delay() routine so
that it is always told of a us delay to impose. The routine then chooses
the best way to set that - via register or via mbx.
- Rather than two value types stored in eq->q_mode (usdelay if change via
register, imax if change via mbox) - q_mode always contains usdelay.
Before any value change, old vs new value is compared and only if
different is a change done.
- Revised the dmult calculation. dmult is not set based on overall imax
divided by hardware queues - instead imax applies to a single cpu and
the value will be replicated to all cpus.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
So far MSIX vector allocation assumed it would be 1:1 with hardware
queues. However, there are several reasons why fewer MSIX vectors may be
allocated than hardware queues such as the platform being out of vectors or
adapter limits being less than cpu count.
This patch reworks the MSIX/EQ relationships with the per-cpu hardware
queues so they can function independently. MSIX vectors will be equitably
split been cpu sockets/cores and then the per-cpu hardware queues will be
mapped to the vectors most efficient for them.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Default behavior is to use the information from the upper IO stacks to
select the hardware queue to use for IO submission. Which typically has
good cpu affinity.
However, the driver, when used on some variants of the upstream kernel, has
found queuing information to be suboptimal for FCP or IO completion locked
on particular cpus.
For command submission situations, the lpfc_fcp_io_sched module parameter
can be set to specify a hardware queue selection policy that overrides the
os stack information.
For IO completion situations, rather than queing cq processing based on the
cpu servicing the interrupting event, schedule the cq processing on the cpu
associated with the hardware queue's cq.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
The XRI get/put lists were partitioned per hardware queue. However, the
adapter rarely had sufficient resources to give a large number of resources
per queue. As such, it became common for a cpu to encounter a lack of XRI
resource and request the upper io stack to retry after returning a BUSY
condition. This occurred even though other cpus were idle and not using
their resources.
Create as efficient a scheme as possible to move resources to the cpus that
need them. Each cpu maintains a small private pool which it allocates from
for io. There is a watermark that the cpu attempts to keep in the private
pool. The private pool, when empty, pulls from a global pool from the
cpu. When the cpu's global pool is empty it will pull from other cpu's
global pool. As there many cpu global pools (1 per cpu or hardware queue
count) and as each cpu selects what cpu to pull from at different rates and
at different times, it creates a radomizing effect that minimizes the
number of cpu's that will contend with each other when the steal XRI's from
another cpu's global pool.
On io completion, a cpu will push the XRI back on to its private pool. A
watermark level is maintained for the private pool such that when it is
exceeded it will move XRI's to the CPU global pool so that other cpu's may
allocate them.
On NVME, as heartbeat commands are critical to get placed on the wire, a
single expedite pool is maintained. When a heartbeat is to be sent, it will
allocate an XRI from the expedite pool rather than the normal cpu
private/global pools. On any io completion, if a reduction in the expedite
pools is seen, it will be replenished before the XRI is placed on the cpu
private pool.
Statistics are added to aid understanding the XRI levels on each cpu and
their behaviors.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
SLI4 nvme functions are passing the SLI3 ring number when posting wqe to
hardware. This should be indicating the hardware queue to use, not the ring
number.
Replace ring number with the hardware queue that should be used.
Note: SCSI avoided this issue as it utilized an older lfpc_issue_iocb
routine that properly adapts.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Both NVME and SCSI aborts are now processed off the CQ workqueue and do not
generate events for the slowpath any more.
Remove the unused event code.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Once the IO buff allocations were made shared, there was a single XRI
buffer list shared by all hardware queues. A single list isn't great for
performance when shared across the per-cpu hardware queues.
Create a separate XRI IO buffer get/put list for each Hardware Queue. As
SGLs and associated IO buffers get allocated/posted to the firmware; round
robin their assignment across all available hardware Queues so that there
is an equitable assignment.
Modify SCSI and NVME IO submit code paths to use the Hardware Queue logic
for XRI allocation.
Add a debugfs interface to display hardware queue statistics
Added new empty_io_bufs counter to track if a cpu runs out of XRIs.
Replace common_ variables/names with io_ to make meanings clearer.
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Currently, both nvme and fcp each have their own concept of an io_channel,
which is a combination wq/cq and associated msix. Different cpus would
share an io_channel.
The driver is now moving to per-cpu wq/cq pairs and msix vectors. The
driver will still use separate wq/cq pairs per protocol on each cpu, but
the protocols will share the msix vector.
Given the elimination of the nvme and fcp io channels, the module
parameters will be removed. A new parameter, lpfc_hdw_queue is added which
allows the wq/cq pair allocation per cpu to be overridden and allocated to
lesser value. If lpfc_hdw_queue is zero, the number of pairs allocated will
be based on the number of cpus. If non-zero, the parameter specifies the
number of queues to allocate. At this time, the maximum non-zero value is
64.
To manage this new paradigm, a new hardware queue structure is created to
track queue activity and relationships.
As MSIX vector allocation must be known before setting up the
relationships, msix allocation now occurs before queue datastructures are
allocated. If the number of vectors allocated is less than the desired
hardware queues, the hardware queue counts will be reduced to the number of
vectors
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
There is a extra queue and msix vector for expresslane. Now that the driver
will be doing queues per cpu, this oddball queue is no longer needed.
Expresslane will utilize the normal per-cpu queues.
Updated debugfs sli4 queue output to go along with the change
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Currently, both NVME and SCSI get their IO buffers from separate
pools. XRI's are associated 1:1 with IO buffers, so XRI's are also split
between protocols.
Eliminate the independent pools and use a single pool. Each buffer
structure now has a common section and a protocol section. Per protocol
routines for SGL initialization are removed and replaced by common
routines. Initialization of the buffers is only done on the common area.
All other fields, which are protocol specific, are initialized when the
buffer is allocated for use in the per-protocol allocation routine.
In the past, the SCSI side allocated IO buffers as part of slave_alloc
calls until the maximum XRIs for SCSI was reached. As all XRIs are now
common and may be used for either protocol, allocation for everything is
done as part of adapter initialization and the scsi side has no action in
slave alloc.
As XRI's are no longer split, the lpfc_xri_split module parameter is
removed.
Adapters based on SLI3 will continue to use the older scsi_buf_list_get/put
routines. All SLI4 adapters utilize the new IO buffer scheme
Signed-off-by: Dick Kennedy <dick.kennedy@broadcom.com>
Signed-off-by: James Smart <jsmart2021@gmail.com>
Reviewed-by: Hannes Reinecke <hare@suse.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
