The write back throttling (WBT) code checks if REQ_SYNC | REQ_IDLE is set
to determine if a write is O_DIRECT vs buffered. If the bits are not set
then it assumes it's a buffered write and will throttle LIO if we hit
certain metrics. LIO itself is not using the buffer cache and is doing
direct I/O, so this has us set the direct bits so we are not throttled.
When the initiator application is doing direct I/O this can greatly improve
performance. It depends on the backend device but we have seen where the
WBT code is throttling writes to only 20K IOPs with 4K I/Os when the device
can support 100K+.
Signed-off-by: Mike Christie <michael.christie@oracle.com>
Link: https://lore.kernel.org/r/20230817192902.346791-1-michael.christie@oracle.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
There is a long call chain that &fip->ctlr_lock is acquired by isr
fnic_isr_msix_wq_copy() under hard IRQ context. Thus other process context
code acquiring the lock should disable IRQ, otherwise deadlock could happen
if the IRQ preempts the execution while the lock is held in process context
on the same CPU.
[ISR]
fnic_isr_msix_wq_copy()
-> fnic_wq_copy_cmpl_handler()
-> fnic_fcpio_cmpl_handler()
-> fnic_fcpio_flogi_reg_cmpl_handler()
-> fnic_flush_tx()
-> fnic_send_frame()
-> fcoe_ctlr_els_send()
-> spin_lock_bh(&fip->ctlr_lock)
[Process Context]
1. fcoe_ctlr_timer_work()
-> fcoe_ctlr_flogi_send()
-> spin_lock_bh(&fip->ctlr_lock)
2. fcoe_ctlr_recv_work()
-> fcoe_ctlr_recv_handler()
-> fcoe_ctlr_recv_els()
-> fcoe_ctlr_announce()
-> spin_lock_bh(&fip->ctlr_lock)
3. fcoe_ctlr_recv_work()
-> fcoe_ctlr_recv_handler()
-> fcoe_ctlr_recv_els()
-> fcoe_ctlr_flogi_retry()
-> spin_lock_bh(&fip->ctlr_lock)
4. -> fcoe_xmit()
-> fcoe_ctlr_els_send()
-> spin_lock_bh(&fip->ctlr_lock)
spin_lock_bh() is not enough since fnic_isr_msix_wq_copy() is a
hardirq.
These flaws were found by an experimental static analysis tool I am
developing for irq-related deadlock.
The patch fix the potential deadlocks by spin_lock_irqsave() to disable
hard irq.
Fixes: 794d98e77f ("[SCSI] libfcoe: retry rejected FLOGI to another FCF if possible")
Signed-off-by: Chengfeng Ye <dg573847474@gmail.com>
Link: https://lore.kernel.org/r/20230817074708.7509-1-dg573847474@gmail.com
Reviewed-by: Davidlohr Bueso <dave@stgolabs.net>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
In the function sli_xmit_bls_rsp64_wqe(), the 'if' and 'else' conditions
evaluates the same expression and give the same output. Also, params->s_id
shall not be equal to U32_MAX. Remove the unused code.
This fixes coccinelle warning such as:
drivers/scsi/elx/libefc_sli/sli4.c:2320:2-4: WARNING: possible
condition with no effect (if == else)
Signed-off-by: Rajeshwar R Shinde <coolrrsh@gmail.com>
Link: https://lore.kernel.org/r/20230817114301.17601-1-coolrrsh@gmail.com
Reviewed-by: Ram Vegesna <ram.vegesna@broadcom.com>
Acked-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
I have an Iomega Z100P2 zip drive, but it does not work with my StarTech
PEX1P2 AX99100 PCIe parallel port, which evidently does not support 16-bit
or 32-bit EPP. Currently the only way to tell the PPA driver to use 8-bit
EPP is to write 'mode=3' to /proc/scsi/ppa/*, but the driver doesn't
actually distinguish between the three EPP modes and still tries to use
16-bit or 32-bit EPP. And even if writing to that file did make the driver
use 8-bit EPP, it still wouldn't do me any good because by the time that
file exists, the drive has already failed to initialize.
Add a new parameter /sys/module/ppa/mode to set the transfer mode before
initializing the drive. This parameter replaces the use of
CONFIG_SCSI_IZIP_EPP16 in the PPA driver.
At the same time, default to 8-bit EPP. 16-bit and 32-bit EPP are not
necessary for the drive to function, nor are they part of the IEEE 1284
standard, so the driver should not assume that they are available.
Signed-off-by: Alex Henrie <alexhenrie24@gmail.com>
Link: https://lore.kernel.org/r/20230807155856.362864-2-alexhenrie24@gmail.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
When a timestamp update or an event acknowledgment command times out, the
driver invokes the soft reset handler to recover the controller while
holding a mutex lock. The soft reset handler also tries to acquire the same
mutex to send initialization commands to the controller which leads to a
deadlock scenario.
To resolve the issue the driver will check thestatus and if this indicates
the controller is operational, the driver will issue a diagnostic fault
reset and exit out of the command processing function. If the controller is
already faulted or asynchronously reset, then the driver will just exit the
command processing function.
Signed-off-by: Ranjan Kumar <ranjan.kumar@broadcom.com>
Link: https://lore.kernel.org/r/20230804104248.118924-2-ranjan.kumar@broadcom.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Only nodes whose state is at least past a PLOGI issue and strictly less
than a PRLI issue should be put into device recovery mode upon RSCN
receipt. Previously, the allowance of LOGO and PRLI completion states did
not make sense because those nodes should be allowed to flow through and
marked as NPort dissappeared as is normally done. A follow up RSCN GID_FT
would recover those nodes in such cases.
Signed-off-by: Justin Tee <justin.tee@broadcom.com>
Link: https://lore.kernel.org/r/20230804195546.157839-1-justintee8345@gmail.com
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Bart Van Assche <bvanassche@acm.org> says:
Hi Martin,
This patch includes the following changes, none of which should change the
functionality of the UFS host controller driver:
- Improve the kernel-doc headers further.
- Fix multiple W=2 compiler warnings.
- Simplify ufshcd_abort_all().
- Simplify the code for creating and parsing UFS Transport Protocol (UTP)
headers.
Please consider this patch series for the next merge window.
Thanks,
Bart.
Link: https://lore.kernel.org/r/20230727194457.3152309-1-bvanassche@acm.org
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
Qcom SoCs require scaling the interconnect paths for proper working of the
peripherals connected through interconnects. Even for accessing the UFS
controller, someone should setup the interconnect paths. So far, the
bootloaders used to setup the interconnect paths before booting Linux as
they need to access the UFS storage for things like fetching boot firmware.
But with the advent of multi boot options, bootloader nowadays like in
SA8540p SoC do not setup the interconnect paths at all.
So trying to configure UFS in the absence of the interconnect path
configuration results in a boot crash.
To fix this issue, and also to dynamically scale the interconnects (UFS-DDR
and CPU-UFS), interconnect API support is added to the Qcom UFS driver.
With this support, the interconnect paths are scaled dynamically based on
the gear configuration.
During the early stage of ufs_qcom_init(), ufs_qcom_icc_init() will setup
the paths to max bandwidth to allow configuring the UFS registers. Touching
the registers without configuring the icc paths would result in a crash.
However, we don't really need to set max vote for the icc paths as any
minimal vote would suffice. But the max value would allow initialization to
be done faster. After init, the bandwidth will get updated using
ufs_qcom_icc_update_bw() based on the gear and lane configuration.
The bandwidth values defined in ufs_qcom_bw_table struct are taken from
Qcom downstream vendor devicetree source and are calculated as per the
UFS3.1 Spec, Section 6.4.1, HS Gear Rates. So it is fixed across platforms.
Cc: Brian Masney <bmasney@redhat.com>
Signed-off-by: Manivannan Sadhasivam <manivannan.sadhasivam@linaro.org>
Link: https://lore.kernel.org/r/20230731145020.41262-3-manivannan.sadhasivam@linaro.org
Signed-off-by: Martin K. Petersen <martin.petersen@oracle.com>
