Previously, when scst_pr_send_ua_reg attempted to deliver a UNIT ATTENTION
to a registrant that was on another node in a dlm-based HA cluster, the
unit attention was dropped.
Rectify by adding a pr_reg_queue_rem_ua function to struct scst_cl_ops and
calling it from scst_pr_send_ua_reg.
Each registrant will maintain an incoming 'queue' of unit attentions by
adding next_rem_ua_idx to the registrant data maintained in the DLM. This
will tell the other nodes which PR_REG_UA_LOCK lock to create in the
lockspace in order to 'send' a unit attention to the registrant.
Further, each node will also maintain two lists (pending and sent) for the
outgoing unit attentions from this node to a registrant. When the
recipient has read all the sent unit attentions (and cleared
next_rem_ua_idx), then the sent list may be cleared.
The previous series of patches introduced the use of a const pointer to
scsi_host_template. However, scsi_host_alloc() uses a non-const pointer
prior to kernel version 6.4, causing the build to throw the following
error:
passing argument 1 of ‘scsi_host_alloc’ discards ‘const’
qualifier from pointer target type
Hence, cast away the constness to resolve the error.
Add an error variable to track errors during the loop in
checkpatch_pull.yml, allowing the workflow to run checkpatch for all
commits before exiting with the appropriate status.
Additionally, add COMMIT_LOG_USE_LINK to the ignore list for both
pull and push workflows
Support for the following driver core changes in the Linux kernel v6.4:
- 1aaba11da9aa ("driver core: class: remove module * from class_create()")
- 2243acd50ac4 ("driver core: class: remove struct class_interface * from callbacks")
This patch fixes RPM build errors caused by incorrect file paths:
RPM build errors:
File must begin with "/": %{perl_vendorlib}/SCST
File must begin with "/": %{perl_vendorarch}/auto/SCST_SCST
For some reasons, the perl_vendorlib and perl_vendorarch variables may
not be defined by default on newer RHEL systems.
Therefore, assign the variables explicitly to avoid these errors.
Add a setting to scst_tgt that can prevent scst_gen_aen_or_ua from
generating an AEN, even if the underlying transport is capable of
transmitting them. It will instead generate a UA.
In the previous commit, we introduced the use of RCU protection when
accessing sess_tgt_dev_list in scst_get_max_lun_commands().
As a result, we can now drop the use of scst_mutex when accessing the
list.
We must always protect sess_tgt_dev_list during access and
modification.
There are two mechanisms for that:
- tgt_dev_list_mutex for list modifications.
- RCU for read-only list accesses.
Currently, the codebase doesn't consistently apply protection when
accessing the list. Fix this by adding RCU protection.
See also commit 3e64094b0c ("scst_sysfs: Do not suspend I/O for LUN
management").
There is a memory leak reported by kmemleak:
unreferenced object 0xffffc900003f0000 (size 12288):
comm "modprobe", pid 19117, jiffies 4299751452 (age 42490.264s)
hex dump (first 32 bytes):
00 00 00 00 00 00 00 00 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:
[<00000000629261a8>] __vmalloc_node_range+0xe56/0x1110
[<0000000001906886>] __vmalloc_node+0xbd/0x150
[<000000005bb4dc34>] vmalloc+0x25/0x30
[<00000000a2dc1194>] qla2x00_create_host+0x7a0/0xe30 [qla2xxx]
[<0000000062b14b47>] qla2x00_probe_one+0x2eb8/0xd160 [qla2xxx]
[<00000000641ccc04>] local_pci_probe+0xeb/0x1a0
The root cause is traced to an error-handling path in qla2x00_probe_one()
when the adapter "base_vha" initialize failed. The fab_scan_rp "scan.l" is
used to record the port information and it is allocated in
qla2x00_create_host(). However, it is not released in the error handling
path "probe_failed".
Fix this by freeing the memory of "scan.l" when an error occurs in the
adapter initialization process.
Fixes: a4239945b8ad ("scsi: qla2xxx: Add switch command to simplify fabric discovery")
Signed-off-by: Li Zetao <lizetao1@huawei.com>
Link: https://lore.kernel.org/r/20230325110004.363898-1-lizetao1@huawei.com
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
[ commit 85ade4010e13 upstream ]
Add a new GitHub Actions workflow, checkpatch_pull.yml, which runs
checkpatch.pl for each commit in a pull request targeting the master
branch.
This change ensures proper checkpatch validation for both push
and pull request events.
Commit 9be09fd68b ("scst/include/backport.h: Fix building on
RHEL 8.8") fixed the build for RHEL 8.8, but at the same time, it
broke the build for RHEL 9.0.
set-output is being deprecated:
"Starting 1st June 2023 workflows using save-state or set-output
commands via stdout will fail with an error."
So fix this by using the GITHUB_OUTPUT environment files instead.
While adding and removing the controller, the following call trace was
observed:
WARNING: CPU: 3 PID: 623596 at kernel/dma/mapping.c:532 dma_free_attrs+0x33/0x50
CPU: 3 PID: 623596 Comm: sh Kdump: loaded Not tainted 5.14.0-96.el9.x86_64 #1
RIP: 0010:dma_free_attrs+0x33/0x50
Call Trace:
qla2x00_async_sns_sp_done+0x107/0x1b0 [qla2xxx]
qla2x00_abort_srb+0x8e/0x250 [qla2xxx]
? ql_dbg+0x70/0x100 [qla2xxx]
__qla2x00_abort_all_cmds+0x108/0x190 [qla2xxx]
qla2x00_abort_all_cmds+0x24/0x70 [qla2xxx]
qla2x00_abort_isp_cleanup+0x305/0x3e0 [qla2xxx]
qla2x00_remove_one+0x364/0x400 [qla2xxx]
pci_device_remove+0x36/0xa0
__device_release_driver+0x17a/0x230
device_release_driver+0x24/0x30
pci_stop_bus_device+0x68/0x90
pci_stop_and_remove_bus_device_locked+0x16/0x30
remove_store+0x75/0x90
kernfs_fop_write_iter+0x11c/0x1b0
new_sync_write+0x11f/0x1b0
vfs_write+0x1eb/0x280
ksys_write+0x5f/0xe0
do_syscall_64+0x5c/0x80
? do_user_addr_fault+0x1d8/0x680
? do_syscall_64+0x69/0x80
? exc_page_fault+0x62/0x140
? asm_exc_page_fault+0x8/0x30
entry_SYSCALL_64_after_hwframe+0x44/0xae
The command was completed in the abort path during driver unload with a
lock held, causing the warning in abort path. Hence complete the command
without any lock held.
Reported-by: Lin Li <lilin@redhat.com>
Tested-by: Lin Li <lilin@redhat.com>
Cc: stable@vger.kernel.org
Signed-off-by: Nilesh Javali <njavali@marvell.com>
Link: https://lore.kernel.org/r/20230313043711.13500-2-njavali@marvell.com
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Reviewed-by: John Meneghini <jmeneghi@redhat.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
[ commit 0367076b0817 upstream ]
Since cluster_mode relies upon the t10_dev_id to generate a namespace, once
cluster_mode is set the t10_dev_id can no longer be changed.
However, because cluster_mode is listed as one of the various add_dev_params,
this meant that it would be set earlier than t10_dev_id when scstadmin
processes scst.conf
Rectify by adding t10_dev_id to fileio_add_dev_params, etc and modifying the
SCST.pm openDevice so that cluster_mode is set last.
Smatch reports:
drivers/scsi/qla2xxx/qla_mid.c:1189:6: warning: symbol 'qla_trim_buf' was not declared. Should it be static?
drivers/scsi/qla2xxx/qla_mid.c:1221:6: warning: symbol '__qla_adjust_buf' was not declared. Should it be static?
These functions are only used in qla_mid.c, so they should be static.
Fixes: 1f8f9c34127e ("scsi: qla2xxx: edif: Reduce memory usage during low I/O")
Signed-off-by: Tom Rix <trix@redhat.com>
Link: https://lore.kernel.org/r/20230114013724.3943580-1-trix@redhat.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
[ commit 54c51253b3d5 upstream ]
Printing a size_t value that is the result of the sizeof() operator
requires using the %z format string modifier to avoid a warning on 32-bit
architectures:
drivers/scsi/qla2xxx/qla_mid.c: In function 'qla_create_buf_pool':
drivers/scsi/qla2xxx/qla_mid.c:1094:51: error: format '%ld' expects argument of type 'long int', but argument 5 has type 'unsigned int' [-Werror=format=]
1094 | "Failed to allocate buf_map(%ld).\n", sz * sizeof(unsigned long));
| ~~^ ~~~~~~~~~~~~~~~~~~~~~~~~~~
| | |
| long int unsigned int
| %d
Fixes: 82d8dfd2a238 ("scsi: qla2xxx: edif: Fix performance dip due to lock contention")
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Reviewed-by: Himanshu Madhani <himansnhu.madhani@oracle.com <mailto:himansnhu.madhani@oracle.com>>
Reviewed-by: Nick Desaulniers <ndesaulniers@google.com>
Link: https://lore.kernel.org/r/20230117170029.2387516-1-arnd@kernel.org
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
[ commit d794a23113b1 upstream ]
Commit ccf2e9c8fab7 ("qla2x00t-32gbit: Select qpair depending on which
CPU post_cmd() gets called") introduced an optimization based on the
pci_irq_get_affinity() API.
Due to support for older kernel versions, limit the minimum kernel
version to use this optimization to v4.9.
See also commit ee8d41e53efe ("pci/msi: Retrieve affinity for a vector") # v4.9.
In current I/O path, Tx and Rx may not be processed on same CPU. This may
lead to thrashing and optimum performance may not be achieved.
Pick qpair such that Tx and Rx are processed on same CPU.
Signed-off-by: Shreyas Deodhar <sdeodhar@marvell.com>
Signed-off-by: Nilesh Javali <njavali@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
[ commit 1d201c81d4cc upstream ]
clang warning:
drivers/scsi/qla2xxx/qla_edif_bsg.h:93:12: warning: field remote_pid
within 'struct app_pinfo_req' is less aligned than 'port_id_t' and is
usually due to 'struct app_pinfo_req' being packed, which can lead to
unaligned accesses [-Wunaligned-access]
port_id_t remote_pid;
^
2 warnings generated.
Remove u32 field in remote_pid to silence warning.
Reported-by: kernel test robot <lkp@intel.com>
Fixes: 7ebb336e45ef ("scsi: qla2xxx: edif: Add start + stop bsgs")
Signed-off-by: Quinn Tran <qutran@marvell.com>
Signed-off-by: Nilesh Javali <njavali@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
[ commit 2f5fab1b6c3a upstream ]
For edif, each I/O requires a secondary buffer to carry the FCP
cmnd. During high traffic time, these buffers are cached in the qpair. As
traffic dies down, these buffers will be trimmed as needed. If traffic is
reduced to none over 2 consecutive intervals, then these buffers will be
further trimmed.
Free FCP cmnd buffers to reduce memory usage during slow I/O time.
Signed-off-by: Quinn Tran <qutran@marvell.com>
Signed-off-by: Nilesh Javali <njavali@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
[ commit 1f8f9c34127e upstream ]
For N2N, qla2x00_wait_for_sess_deletion call flushes
a session which accidentally clear the scan_flag and thus prevents
re-login to occur and causes session to stall.
Use session delete to avoid the accidental clearing of scan_flag.
Signed-off-by: Quinn Tran <qutran@marvell.com>
Signed-off-by: Nilesh Javali <njavali@marvell.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
[ commit 129a7c40294f upstream ]
User experienced performance dip on measuring IOPS while EDIF
enabled. During I/O time, driver uses dma_pool_zalloc() call to allocate a
chunk of memory. This call contains a lock behind the scene which
contribute to lock contention. Save the allocated memory for reuse and
avoid the lock.
Signed-off-by: Quinn Tran <qutran@marvell.com>
Signed-off-by: Nilesh Javali <njavali@marvell.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
[ commit 82d8dfd2a238 upstream ]
Building SCST with an old version of GCC (e.g. 4.8.5 on Centos 7.X)
produces a warning about using initializer.
This is GCC bug # 53119:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=53119
Hence, use memset() instead of an initializer to avoid this warning.
Fixes: e8b300ef2a ("usr/fileio/common.c: Fix use of uninitialized
struct field")
There is no functional change in this patch. VP map resource is renamed
and relocated so it is not viewed as just a target mode resource.
Signed-off-by: Quinn Tran <qutran@marvell.com>
Signed-off-by: Nilesh Javali <njavali@marvell.com>
Reviewed-by: Himanshu Madhani <himanshu.madhani@oracle.com>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
[ commit 430eef03a763 upstream ]
