Marvell Driver Release Notes

		       Marvell, Inc.
		       All rights reserved


Table of Contents

1. Change History
2. Known Issues
3. Notices
4. Contacting Support


1. Change History

Current version: 3.1.87.0 - August 30, 2021

This section contains:

 * 1.1 Hardware Support
 * 1.2 Software Component
 * 1.3 Bug Fixes


1.1 Hardware Support

Initial Drop 3.1.1.0:

 * Support for 2400/2500/2600/2700/2800 Series Fibre Channel adapters
 * Support for 8100/8200/8300 Series Converged Network Adapters
 * Mt. Rainier support.

Between versions 3.1.1.0 and 3.1.9.0:

 * None

Between versions 3.1.9.0 and 3.1.10.0:

* Remove support for Fabric Cache (Mt. Rainier) adapter

Between versions 3.1.10.0 and 3.1.87.0:

* None


1.2 Software Components

Between versions 3.1.1.0 and 3.1.3.0:

 * None

Between versions 3.1.3.0 and 3.1.4.0:

 * End-to-End QoS fabric priority support
 * Update ISP25XX fw version to 8.08.01.

Between versions 3.1.4.0 and 3.1.6.0:

 * None

Between versions 3.1.6.0 and 3.1.7.0:

 * Revert update ISP25XX fw version to 8.08.01.
   FW version is now back to 8.07.00

Between versions 3.1.7.0 and 3.1.10.0:

 * None

Between versions 3.1.10.0 and 3.1.11.0:

 * Update ISP25XX fw version to 8.08.01.
 * Add support for Mach Auxiliary Image Status and
suppporting flash-related functionality improvements

Between versions 3.1.11.0 and 3.1.12.0:

 * SmartSAN features disabled by default 

Between versions 3.1.12.0 and 3.1.16.0:

 * None

Between versions 3.1.16.0 and 3.1.17.0:

 * Secure flash update support for ISP28XX 

Between versions 3.1.17.0 and 3.1.18.0:

 * None
 
Between versions 3.1.18.0 and 3.1.19.0:

 * Simplified Fabric Discovery support 

Between versions 3.1.19.0 and 3.1.27.0:

 * None

Between versions 3.1.27.0 and 3.1.28.0:

 * Update ISP25XX fw to version 8.08.207

Between versions 3.1.28.0 and 3.1.29.0:

 * Revert ISP25XX fw to version 8.08.206

Between versions 3.1.29.0 and 3.1.31.0:

 * None 

Between versions 3.1.31.0 and 3.1.32.0:

 * SAN Congestion Management support

Between versions 3.1.32.0 and 3.1.33.0:

 * Update ISP25XX fw to version 8.08.207

Between versions 3.1.33.0 and 3.1.39.0:

 * None

Between versions 3.1.39.0 and 3.1.40.0:

 * Statistic collection for error detection
 * Disable FW command timer by default.

Between versions 3.1.40.0 and 3.1.41.0:

ERXXXXXX:	SAN Congestion Management support, phase 2
Change:		Enhancements to SAN Congestion Management feature
Relevance:	27XX and 28XX adapters 

Between versions 3.1.41.0 and 3.1.42.0:

ERXXXXXX:	Lockdown support
Change:		Added support for management security lockdown feature
Relevance:	27XX and 28XX adapters

Between versions 3.1.42.0 and 3.1.74.0:

 * None

Between versions 3.1.74.0 and 3.1.82.0:

 * Encryption support

Between versions 3.1.82.0 and 3.1.83.0:

 * API support for encryption

Between versions 3.1.83.0 and 3.1.87.0:

 * None


1.3 Bug Fixes

Between versions 3.1.1.0 and 3.1.2.0:

Problem Description:
Improvements
Solution:
Various Mach-related fixes uncovered during testing.

Problem Description:
Failing during unload testing for ESX 6.7
Solution:
Proper cleanup during detach function. 

Between versions 3.1.2.0 and 3.1.3.0:

Problem Description:
Issue with Mach adapter fw dump
Solution:
Fix firmware dump template entry to not
insert extra dword of info.

Between versions 3.1.3.0 and 3.1.4.0:

Problem Description:
Secure flash requires that the driver only writes to flash through
MB Cmds for Qlipper and newer adapters.
Solution:
Patch up holes in write flash routine that could cause the driver to use register access.

Problem Description:
Driver was advertising Application Services support in the RFT_ID command
regardless if enabled or not.
Solution:
Only enable Application Services if ql2xvmidsupport is turned on.

Problem Description:
Target devices are temporariry not accessible when the link toggle occurs on one of the target device paths.
Solution:
Clear the "login needed" bit when the initial login attempt fails. This login will get
retried in the relogin dpc call. This prevents target ports remaining in a "login needed"
state despite disappearing from the fabric entirely.

Problem Description:
Driver was not handling ABTS received IOCB in response to a PUREX IOCB failure.
Solution:
Code was implemented to handle ABTS received.

Problem Description:
Driver would not send the full RDP response with a switch port that was not in the logged in state.
Solution:
Remove the limitation as the new firmware now supports splitting up the RDP response payload over multiple frames
This will be based on a check of the supported firmware versions.

Problem Description:
Code improvement.
Solution:
Move ql2xvmidsupport check to module_init routine

Problem Description:
VMID information was not getting displayed for the NPIV key value in the vmkmgmt interface.
Solution:
Added VMID information to NPIV vmkmgmt output

Problem Description:
Code improvement
Solution:
DPORT improvement - return mb1 and mb2 contents after execution.
Also added more info in DPORT AEN messages.

Problem Description:
ESX 6.7 IOVP DDV testing was failing.
Solution:
Use proper cleanup of driver resources in error paths

Between versions 3.1.4.0 and 3.1.5.0:

Problem Description:
Flash writes were failing.
Solution:
Ensure dword count and byte count were used correctly in flash write routine.

Between versions 3.1.5.0 and 3.1.6.0:

Problem Description:
Driver read incorrectly from flash for E2E QoS values,
leading to invalid table readings and disabling of feature.
Solution:
Pass in byte count instead of dword count to flash read function.

Problem Descriptoin:
RDP response payload was not formatted correctly, leading to missing
contents or response frame not transmitted.
Solution:
Properly populate link services attribute and implement correct check
of firmware version and login state when building payload.

Problem Description: 
Driver was advertising 1G speed support for 8G adapters.
Solution:
Remove 1G speed support for 8G adapters.
Also, cleaned up supported speed code for various features and adapters.

Problem Description:
FDMI info showing incorrect supported speeds for 16G mezz adapters in FDMI.
Solution
Remove 4G supported speed if NVRAM parameter indicates to do so.

Between versions 3.1.6.0 and 3.1.7.0:

None 

Between versions 3.1.7.0 and 3.1.8.0:

Problem Description:
When E2E QoS solution was disabled on a LUN, driver was still populating
the priority field in the CS_CTL for I/O to that LUN.
Solution:
Don't fall back to standard fabric priority QoS unless the fabric priority
bit in the NVRAM is set.

Between versions 3.1.8.0 and 3.1.9.0:

Problem Description:
qlnativefc driver would fail to load on a system with NPAR enabled on NIC adapter.
Solution:
Dump callback registration has a hard limit of 40 (ESX limit), so on a system
with many adapters/functions registering a callback, this call could fail.
Solution was to treat a failed dump callback registration as non-fatal
and continue with driver load.

Between versions 3.1.9.0 and 3.1.10.0:

Problem Description:
New compiler in ESX engineering drop uncovered coding issues
Solution:
Fix existent coding issues uncovered.

Problem Description:
Code inspection - driver was ignoring potential retry delay returned in
FCP Response frame from target.
Solution:
Handle retry delay by returning VMK_SCSI_HOST_RETRY for specified time.

Problem Description:
FDMI info registered on switch was inconsistent across different OSes.
Solution:
Sync up information registered to switch for FDMI across all drivers.

Problem Description:
Errors encountered during DDV testing.
Solution:
Fix issues uncovered during DDV inbox testing at VMware.

Improvement: Cleanup various log messages

Problem Description:
Set Params mailbox command was failing during initialization.
Solution:
Only send CS_CTL map info using Set Params MB Cmd when GFO successful.

Problem Description:
Flash access was failing during Mach adapter bring-up
Solution:
Flash access routines were converting flash addresses incorrectly.

Problem Description:
Problems with Mach bring-up with new flash signature.
Description:
Add check for new Mach flash signature during initialization

Problem Description:
Targets would fail to get logged into if a target on the fabric
list sent a REJECT for a given login.
Solution:
Continue with the logins in the scan loop despite seeing a login failure

Between versions 3.1.10.0 and 3.1.11.0:

Problem Description:
Certain HBAs full marketing name was being trucated when
executing the "esxcli storage san fc list" command.
Solution:
Copy over full contents of VPD product description field to API.

Between versions 3.1.11.0 and 3.1.12.0:

Problem Description:
Fabric Discovery could take a long time, which could result in path loss or in the case of NPIV, 
vports could not get created.
Solution:
GFO sent to an unsupported switch would take 20 seconds to timeout.
Fix was to send the GFO command in a separate thread then the fabric discovery.

Problem Description:
A management application doing periodic small reads of flash would cause
the system to experience I/O failures with "RETRY" status.
Solution:
Eliminate the code to block I/O during small read operations of the flash.
This code is no longer needed.

Between versions 3.1.12.0 and 3.1.13.0:

Code improvement:
Turn on ql2xattemptdumponpanic module parameter by default.

Code improvement:
Remove uncessary RISC parity disable/enable sequence during ISP abort
- only needed on obsoleted adapters.

Problem Description:
Incorrect supported speeds were being reported by driver on Mach adapter.
Solution:
Correctly interpret supported speeds returned by FW after execution.

Problem Description:
FW dump template entry T262 was showing "bad range" in the logs.
Solution:
If start and end address are equal, this is a valid dump entry.

Problem Description:
NPIV port creation would fail with "Physical uid does not match
VPORT uid, NPIV Disabled for this VM"
Solution:
Ensure the target ID assigned to the WWPN on the physical port is the
same on the NPIV port.

Between versions 3.1.13.0 and 3.1.14.0:

Code improvement:
Various log messages improvement.

Problem Description:
PSOD observed while running DDV test
Solution:
Check for NULL pointer during memory alloc call before using buffer.

Between versions 3.1.14.0 and 3.1.15.0:

Problem Description:
Fabric registrations were not taking place with Fabric Priority support enabled.
Solution:
Don't unecessrily clear REGISTER_FC4_NEEDED flag in fabric scan code

Problem Description:
Hazard C16 test was failing with input/output error.
Solution:
Virtual Reset code needs to abort all I/Os on a given world, whereas
the old code would stop searching for outstanding I/O in the event
of one abort failure.  Fix was to not exit I/O loop until all oustanding
I/O have had an abort attempt.

Problem Description:
Dell-EMC VMAX/PMAX setup would not report any LUNs
Solution:
Omit check for same domain and area as adapter's during "local loop"
device discovery

Between versions 3.1.15.0 and 3.1.16.0:

Code Review:
Add brackets missing for secondary VPD location assignment.

Problem Description:
If ql2xattemptdumponpanic is set and dump partition is on SAN LUN,
saving off of coredump during PSOD will fail
Solution:
Reset and re-init adapter after FW dump attempt so paths to LUNS come
back online.  Set ql2xattemptdumponpanic to default to 0 as well.

Between versions 3.1.16.0 and 3.1.17.0:

Problem Description:
PSOD observed during NPIV failover testing.
Solution:
scsiDoneList only initialized and used in the base vha, so fix was to
ensure the cmd done code path used the base vha scsiDoneList

Problem Description:
Incorrect supported speed values were being reported in management app.
Solution: 
Return correct supported speed bit map values in vmkmgt interface

Problem Description:
FW dump callback was only saving off one 27xx/28xx dump
Solution:
Improve dump callback logic to save off and/or take fw dumps
on all adapters available.


Between versions 3.1.17.0 and 3.1.18.0:

Problem Description:
Flash update failing on Mach adapter
Solution:
Fix non-secure flash updates on Mach. Also, improved the code
to handle multiple chunks for a FW update. Added support
for MPI and PEP FW updates.

Problem Description:
Loss of paths when target sends ASC/ASCQ of 0x3F/0x3 (Inquiry Data has changed)
Solution:
Vendor unique T10 DIF code existed to intercept this status and handle the change
internally, however it was not within a check to see if LUN supported this feature
or not. Fix was to put this handling within the T10 DIF supported code path only.

Problem Description:
Sending SCSI pass thru commands to an NPIV port was failing
Solution: 
Correctly pass in NPIV vha structure to pass thru function, instead of
the base vha which would not have the corret vport index to pass to firmware.


Between versions 3.1.18.0 and 3.1.19.0:

Problem Description:
Secure Flash Update MB Cmd was failing during testing with operational firmware. 
Solution:
Driver was passing down incorrect SFUB length in bytes - should be in DWORD count.

Problem Description:
Code improvement
Solution:
Secure adapter and fw support displayed in vmkmgmt interface

Problem Description:
Code improvement
Solution:
Correctly indicate default values for module parameter

Between versions 3.1.19.0 and 3.1.20.0:

Problem Description:
DDV testing for vSphere next inbox was failing.
Solution:
Re-work some of the memory cleanup in different failure paths
to avoid driver reload issues during DDV testing.

Problem Description:
Driver vmkmgmt shows the VMID info under the NPIV section, when NPIV is not enabled. 
Solution:
This info was reprinted in the NPIV section unecessarily,
already produced in the HBA section, so it was removed.

Problem Description:
Various issues introduced Simplified Fabric Discovery support.
Solution:
Fixed VMID, application and unload issues introduced with this new feature.

Between versions 3.1.20.0 and 3.1.21.0:

Problem Description:
ql2xt10difvendor was disabled and associated with ql2xenablesmartsan
Solution:
Re-enable ql2xt10difvendor and separate the two features out.a

Problem Description:
During ESX 6.7U2 beta testing, a PSOD was encounterd during driver load.
Solution: 
Recent changes for Simplified Fabric Discovery introduced deferred work threads
to handle further IOCB processing after interrupts. workLock protected the
deferred work list and posting however the hardwareLock with the same lock rank
could be held at the same time. Solution was to lower the workLock rank below
that of the hardwareLock.

Between versions 3.1.21.0 and 3.1.22.0:

Problem Description:
PSODs during ESX 6.5 U2 BETA testing.
Solution:
Adjust spinlock usage and ranking for locks introduced for
Simplified Fabric Discovery feature.

Between versions 3.1.22.0 and 3.1.23.0:

Problem Description:
zdump fails to get saved off on BFS configuration
Solution:
Properly initialize code to execute I/O for dump callback - this was 
was missing due to new logic introduced with Simplified Fabric Discovery

Problem Description:
PSODs encountered during open zone testing and utilizing NPIV.
Solution:
Various improvments and fixes to Simplified Fabric Discovery handling when
NPIV is enabled.

Problem Description:
DDV testing failure encountered during VMWare inbox driver testing
Solution:
Memory cleanup fix in the IOCTL memory alloc logic

Problem Description:
'show fdmi’ command does not list “OS Name and Version” and
"CT Payload Length" field on Cisco switches
Solution:
Added "OS Name and Version" and "CT Payload Length" attributes to original FDMI implementation.

Between versions 3.1.23.0 and 3.1.24.0:

Problem Descrption:
System would PSOD during driver unload/load testing.
Solution:
Release hardware lock before vmk_WorldYield calls in fw dump rouintes.

Problem Description:
I/O were not getting completed properly when FW would fail an IOCB
with invalid entry status.  This lead to continous aborts observed.
Solution:
Populate handle field in srb when command is submitted to firmware.

Problem Description:
ISL cable pulls were resulting in path loss and PSODs.
Solution:
RSCN handling logic made the same regardless of type.  The code to
handle domain change was old and removed with this fix.

Between versions 3.1.24.0 and 3.1.25.0:

Proble Description:
During ISL disable/enable testing, some target ports not coming back online
Solution:
Handle Port Down I/O status so that the discovery state gets set to DELETED.

Problem Description:
PSOD occurs during ISL disable/enable testing.
Solution:
When cleaning up I/O, don't decrement refcount twice.


Between versions 3.1.25.0 and 3.1.26.0:

Proble Description:
Code improvement
Solution:
Unecessary sp free call made in GPNFT done path

Problme Description:
Code improvement
Solution:
Print out all mailboxes up to 7 for AEN 8002 per FW team's request

Problem Description:
Code Improvement
Solution:
Cleanup pDIF log messages

Problem Description:
Used fc-esxcli version 1.0.20 to check negotiated PCI width and speed for
the FC adapters but it is not displayed. Same is true with VI plugin.
Solution:
Return 16-bit value for PCIe Link Status register.  Previous code
was returning a 32-bit value and shifting the PCIe width and speed
values up 16 bits.

Problem Description:
PSOD encountered during open zone testing with cable pulls and I/O.
Solution:
Initialization of work item for simplified fabric discovery needs to be
done at the time of allocation, not each time a work item is queued.
This could lead to link lists being broken.

Problem Description:
Driver lock-ups could occur with MPI heartbeat stop async event
Solution
For MPI heartbeat stop async event, this patch would capture MPI FW dump and chip reset.
FW will tell which function to capture FW dump.  Also, added handling for timeout
occuring during fw dump where the dump procedure would be shortened

Problem Description:
Flash update failures can occur on single port 28XX adapters.
Solution:
For single port 28XX adapter, the second core can still runs in
the back ground.  The flash semaphore can be held by the non-active
core.  This patch tell MPI FW to check for this case and clear the
semaphore from the non-active core.

Problem Description:
Code Improvement
Solution:
Added MPI dump handling when firmware panic occurs

Problem Description:
FDMI 2 RHBA command was getting rejected by the switch.
Solution:
Correctly formwat version strings in attribute list.  Also,
bring formatting inline with other OS drivers.

Problem Description:
GetActiveRegions API not supported on Baker/Qlipper
Solution:
Added support for this API call

Between versions 3.1.26.0 and 3.1.27.0:

Problem Description:
PSOD occurs during flash update with multi-queue disabled.
Solution:
Only execute qpair code logic in ISP abort handler if mq is enabled.

Problem Descrption:
PSOD occurs during Boot From SAN (BFS) and RSCN storm testing
Solution:
SRB timer code was not being canceled correctly if still active during
driver shutdown

Problem Description:
PSOD occurs during VMID and VVOL testing
Solution:
gblDsdList for Command Type 6 IOCBs was not fully protected 
during command processing

Between versions 3.1.27.0 and 3.1.28.0:

Problem Description:
DDV tests during IOVP were failing
Solution:
Various fixes and improvments to the driver detach routine as well
as memory allocation failure paths.

Problem Description:
FW Dump on PSOD was failing when ql2xattemptdumponpanic was set.
Solution:
Don't do an ISP abort after fw dump.

Between versions 3.1.28.0 and 3.1.29.0:

Problem Description:
On systems with ql2xmqcpuaffinity disabled OR if the number of NUMA
nodes is 1, a PSOD could occur during command processing.
Solution:
Properly initialize qpair pointer when ql2xcpuaffinity is not enabled in
the command submission path.

Between versions 3.1.29.0 and 3.1.30.0:

Problem Description:
Target remained OFFLINE during port toggle testing after a FW panic (8002).
Solution:
During target session deletion, set FCF_ASYNC_SENT flag to prevent
other session management commands from being executed.  Also, other improvements
to the fcport deletion state handling. ER146424

Problem Description:
I/O errors were seen when driver encountered a FW panic.
Solution:
Properly clean up outstanding I/O and populate with correct host
status when aborted during 8002 Async Event processing. ER146425

Problem Description:
vmkmgmt output was showing incorrect number of response queues when multi-queue
was disabled
Solution:
Calculate proper response queue count to display; also show CPU
affinity disabled/enabled. ER146367

Problem Description:
Hilda mezz adapters showed incorrect supported speeds
Solution:
Do runtime check for Synergy Quartz mezz card - show only 16G supported speed. ER146323

Problem Description:
Mach adapter driver strings showed 64G FC in qlnativefc_devices.py file.
Solution:
Base string off of subsystem device ID for better clarity and improve generic Mach string
to show 32G/64G support. ER146372


Between versions 3.1.30.0 and 3.1.31.0:

Problem Description:
PSOD was seen during testing on blade servers with 8G and Hilda adapters
Solution:
Removed reads from request queue in pointers during normal operation.
Also, correctly initialize request queue in pointer address during non
multi-queue operation. ER146525 

Problem Description:
PSOD during driver load on one NUMA node server.
Solution:
Do not attempt to set fw_started bit on qpairs using the queue pair map
if it hasn't bee allocated.  ER146639


Between versions 3.1.31.0 and 3.1.32.0:

Problem Description:
MPI fw dump and reset was done with ISP fw dump and reset
which was unecessary.
Solution:
Improve MPI and ISP fw dump and reset handling to the agreed
upon specification.  Allow for either dump to be performed
separately. ER146608

Problem Description:
Application Services, when enabled (ql2xvmidsupport) was not getting
registered on the Brocade switch.
Solution:
Enable Application Services FC4 type when sending RFT_ID command. ER146749


Between versions 3.1.32.0 and 3.1.33.0:

Problem Description:
A PSOD would occur when encountering an MPI pause during MPI fw dump
handling
Solution:
Defer MPI dump collection to DPC thread.  ER146769


Between versions 3.1.33.0 and 3.1.34.0:

Problem Description:
FCP targets don't come back online after first cable pull with NVMe
enabled.
Solution:
Don't re-use IFWCB buffer for N2N PLOGI template.  Also, some
improvements to N2N handling with NVMe enabled. ER146763

Problem Description:
Prevent panic in vmkmgmt I/O timeout path with verbose logging turned on.
Solution:
Check cmd pointer in qlnativefcPrintScsiCmd function before printing to log.

Problem Description:
Namespace discovery and path claiming for NVMe storage was not
completing and so paths would always show up dead.
Solution:
Data underruns from FW with "good" NVMe status should be returned
"good" status to the NVMe layer. ER146708

Problem Description:
I/O timeouts and aborts observed when 8G HBA FW loaded from flash
that did not support multiqueue.
Solution:
When CPU affinity setup fails, queue pointers need to be re-initialized
to correct non-MQ location on 8G adapters.

Problem Description:
FCP_RSP frame status qualifier field not supported.
Solution:
FCP-4 (referred FCP-4 rev-2b) identifies the earlier known "retry delay
timer" field as "status qualifier", which is described in SAM-5 and
later specs. This fix makes appropriate driver side modifications to
honor the new definition. The SAM document referred was SAM-6 rev-5.

Problem Description:
MPI dump execution taking a long time.
Solution:
Check MPI dump flag in timer function when attempting to
wake up DPC thread. Also, made improvments to the functions
involved in collectin the MPI dump. ER146608

Problem Description:
vmkmgmt key-value info wasn't showing pDIF support on LUNs
Solution:
Added pDIF info to the vmkmgmt output for debugging purposes.

Problem Description:
It may takes more than twenty minutes to unload the driver - ESX7.0
inbox testing.
Solution:
Check WaitForHbaReady every 10ms as opposed to every sec (10ms). DCPN58478


Between versions 3.1.34.0 and 3.1.35.0:

Problem Description:
Support for NVMe target reporting to apps in ESX 7.0
Solution:
Create functions out of code that assigns target IDs

Problem Description:
DDV test failing with PSOD when forcing slab alloc errors on RFT_ID
command
Solution:
Retries done on CT passthru commands need timer to be restarted.
Also, attempt retry even if timer not active.

Problem Description:
"Mid-layer underflow" error messages seen in logs as a result of pDIF I/O sent
to a non-pDIF LUN.
Solution:
Properly protect the data buffer(s) used to parse the Inquiry response for each LUN
before determining if pDIF is supported or not.  This buffer
was not protected which could cause incorrect Inquiry data read leading to incorrectly
enabling pDIF. ER146982

Problem Description:
Code review - improvement needed with debugging issues when handling errors
in response queue processing.
Solution:
Force reset and fw dump collection on various response queue error conditions


Between versions 3.1.35.0 and 3.1.36.0:

Problem Description:
DDV failures in VMWare inbox testing
Solution:
Various fixes in DMA alloc failure path and driver detach routine.

Problem Description:
SCM support was not getting enabled in FW on Qlipper/Baker adapters.
Solution:
FW team had defined bit 10 in FW extended attributes lower to indicate
SCM support, which confliced with secure flash support in Mach adapters.
New bit 12 is now used to indicate SCM support. ER147063

Problem Description:
Fabric priority QoS was not working with ql2xfabricpriority module parameter
Solution:
Fix initialization of GFO work thread when ql2xfabricpriority is turned on. ER147121


Between versions 3.1.36.0 and 3.1.37.0:

Problem Description:
DDV EILoading test was failing with non-zero heap.
Solution:
During driver unload, short circuit RFT_ID/RFF_ID/RNN_ID/RSNN_ID retries
in the "done" handler and just clean-up the DMA buffers.

Problem Description:
Debug print statement in LUN reset path was printing incorrect target and LUN Ids
Solution:
Print correct values.  No impact for LUN reset handling - cosmetic.

Problem Description:
Peg core dump would fail with invalid address error.
Solution:
Populate RAM ID field in MB Register 10

Between versions 3.1.37.0 and 3.1.39.0:

Code improvement : Various code improvements after code review.
Resolution   : Code improvement when setting up qpair in I/O path.
Relevance   : all adapters


Between versions 3.1.39.0 and 3.1.40.0:

DCPN22882 : Use function number for logical device creation 
Resolution : This will esnure persistent logical device names
	across driver reload and system reboot.
Relevance   : all adapters


Between versions 3.1.40.0 and 3.1.41.0:

ER147415:	Driver not relogging in during storage bounce test
Resolution:	When the driver receives a login status of "Port ID used", logout
		the nport handle before attempting a relogin.
Relevance:	All adapters

ERXXXXXX:	MPI and FW dump processing improvements
Resolution:	Improvements to dump processing to conform with additions
		to specification
Relevance:	27XX adapters and later

ERXXXXXX:	Code review - vmkmgmt calls were missing cleanup function
Resolution:	Ensure all exits paths call the qlnativefcExitApi routine
Relevance:	All adapters

ERXXXXXX: 	Improvements to enhanced error detection stats collection feature
Resolution:	Additions required by customer; also put API into separate source
		and header files
Relevance:	27XX adapters and later


Between versions 3.1.41.0 and 3.1.42.0:

ER147525: 	vmkmgmt interface for NVMe hosts shows FCP targets in EED and SCM stats
Resolution:	Display the NVMe or FCP target lists based on host vmhba type.
Relevance: 	27xx adapters and later

ER147526:	ql2xenhancedabort support shows available on ESX 6.7 driver
Resolution:	Enhanced abort only supported with NVMe supported drivers - move
		ql2xenhancedabortsupport to NVME compile switch. 
Relevance:	27XX adapters and later

ER147524:	PSOD seen when handling SCM peer congestion event
Resolution:	System did not have multi-queue enabled so check for using slow
		queue to throttle I/O needs to have a check for multi-queue as well.
Relevance:	27XX adapters and later

ER147606:	CPU affinity multi-queue disabled with SCM phase 2 support
Resolution:	Slow queue pair implementation was improved to use unique MSI-X vector when
		initializing
Relevance:	27XX adapters and later

ER147415:	Driver not relogging in during storage bounce test, part 2
Resolution:	When the driver receives a port logout 8014 Async Event, logout
		the nport handle before attempting a relogin.
Relevance:	All adapters

ERXXXXXX: 	Use MPI hang trigger to do PEGTUNE halt.
Resolution:	Utilize vmkmgmt interface for MPI pause to do a PEGTUNE halt on 83XX adapters
Relevance:	83XX adapters

DCPN65467:	PSOD occurrs during DDV testing when forcing queue creation failures
		in CPU affinity setup.
Resolution:	Don't disable mqenable flag when CPU affinity qpair creation fails.
Relevance:	All adapters

ER147603:	Enhanced Error Detection API would return 0 for entry count when
		retrieving host stats.
Resolution:	Populate entry count correctly before returning stat values.
Relevance:	27XX adapters and later

ERXXXXXX:	Code review - enhanced abort MB Cmd timeout value.
Reslution:	Increase MB Cmd 0x54 timeout value to be greater than FW timeout value.
Relevance:	27XX adapters and later


Between versions 3.1.42.0 and 3.1.43.0:

ER147561:	Host side stats were not being updated for FPIN warnings.
Resolution:	FPIN descriptor structure had been modified incorrectly for phase 2
		of SCM feature.  Fix was to revert to correct endianess of original
		structure
Relevance:	27XX adapters and later

ERXXXXXX:	Code review - vmkmgmt MPI flags need updating after MPI reset
Resolution:	Capture MPI flags for display after MPI reset.  Also, include MPI
		Reporting support in output.
Relevance:	27XX adapters and later

ER147621:	PSOD when NVMe I/O is returned with UNDERRUN.
Resolution:	Ensure good NVMe completion status from target before handling
		UNDERRUN.  Also, only return SUCCESS status if data was transferred.
Relevance:	27XX adapters and later

ERXXXXXX:	Enhanced abort logging improvements for NVMe.
Resolution:	When enhanced abort is used, print Abort IOCB completion and
		NVMe completion status by default
Relevance:	27XX adapters and later

ER147533:	ESX CLI doesn't show UCSCM stats.
Resolution:	SCM Phase 2 broke backward compatibility with Phase 1.
		interface; bring back old interface and create new API calls for Phase 2
Relevance:	27XX adapters and later

ER146879:	Support to clear SCM/SCMR stats.
Resolution:	Added an interface in the driver to clear stats via
		qaucli.
Relevance:	27XX adapters and later

ER147630:	SCM throttling increase did not happen as expected after congestion event.
Resolution:	Ensure throttle change value is at least 1 after calculating the new
		value.
Relevance:	27XX adapters and later

ERXXXXXX:	Various SCM improvements
Resolution:	API structure additions to bring inline with latest spec
		SCMR algorithm improvements
		Clear peer congestion after removal from fabric
		SCMR support for NVMe
		Log message improvments
Relevance:	27XX adapters and later


Between versions 3.1.43.0 and 3.1.44.0:

ER147576:	PSOD and data inconsistencies preceeded by numerous invalid parameter
		in response IOCBs - "Process error entry. type/count/sys/status/comp = 18:2:0:8:1000"
Resolution:	Prevent data issues by passing back correct error status when invalid entry error
		seen on response queue.
Relevance: 	All adapters

ER147689:	MPI reset was timing out and status not showing up with link down..
Resolution:	Driver was only grabbing MPI Status info if FW state was ready, i.e. link up.  Change
		was to get this info regardless of FW state.
Relevance:	27xx adapters and later

ERXXXXXX:	SCM: Remove call to notify FW of slow device
Resolution:	Prevent driver from sending MB Cmd 0x1A (Set Port Params)
		to notify FW of slow device until this feature is fully supported
Relevance:	27xx adapters and later

ER147733:	Enabling ql2xvmidsupport and ql2xenablesmartsan at the same time on a
		boot from SAN setup caused driver load failures.
Resolution:	Prevent both module parameters from being set at the same time - disable
    		ql2xvmidsupport in that event.
Relevance:	27xx adapters and later


Between versions 3.1.44.0 and 3.1.45.0:

ER147768:	SCMR: When ql2x_scmr_flow_ctl_tgt and ql2x_scmr_flow_ctl_host set to 0, congestion
		state for both target and host would not get cleared.
Resolution:	Allow congestion clearing to be called even when not actively
		throttling.
Relevance:	27xx adapters and later

ERXXXXXX:	SCMR: Remove remaining code to notify FW of slow device
Resolution:	Some code had been left in from previous attempt.  Also, some various
		log message improvments.
Relevance:	27xx adapters and later


Between versions 3.1.45.0 and 3.1.46.0:

ER146879:	Alarm and warning counters were not getting cleared in QL_SET_PORT_SCM
		vmkmgmt callback.
Resolution:	Clear counters correctly in callback.
Relevance:	27xx adapters and later

ER147795:	Host side congestion was not honoring event period sent in FPIN for
		when to clear congestion state.
Resolution:	Copy over correct FPIN field when determining event period for a host
		congestion scenario.  Also print out correct value of event period.
Relevance: 	27xx adapters and later.


Between versions 3.1.45.0 and 3.1.61.0:

ERXXXXXX:	Various improvements to Enhanced Error Detection support
Resolution:	Short link down counter fix, logging improvments and API callback
		wait time improvements
Relevance:	27XX adapters and later

ER148040:	Performance regression with ESX 7.0 inbox testing (DPCN64008)
Resolution:	GPSC failures were leading to IIDMA for target being set to 1GB/s.
		Fix was to initialize target port speed to "unknown" so GPSC failure
		path would not lead to IIDMA being incorrectly set.
Relevance:	All adapters

ER147303:	Fix response queue handler reading stale packets
Resolution:	Two module parameters are introduced (qlx2rsq_follow_inptr and
		ql2xrspq_follow_inptr_legacy) to control response queue processing
		logic to follow in pointer rather than signature (on by default)
Relevance:	All adapters

ERXXXXXX:	Improvement to SAN Congestion Managment algorithm
Resolution:	Added queue depth based throttling and turned on by default
		(ql2x_scmr_throttle_mode = 2)
Relevancce:	27xx adapters and later

ER148010:	Output of "esxcli qlfc qcc port scmchk get" for "Seconds Since Last
		Event" always shows 0.
Resolution:	Keep track of and return elapsed time for each congestion event.
Relevance:	27xx adapter and later

ER147961:	Target VP failover/failback shows "Process error entry" messages
Resolution:	Mark target port as "OFFLINE" immediately before modifying port ID in
		driver database
Relevance:	All adapters


Between versions 3.1.61.0 and 3.1.62.0:

ER148066:	Host side throttling - qdepth value shows negative
Resolution:	Call qdepth increment function regardless if throttle attempt was
		made.
Relevance: 	27xx and later

ER148045:	CPU lockup occurs during FPIN processing
Resolution:	Problem was due to the FPIN processing loop potentially spinning indefinitely.
		Fix was to wait only for 2 seconds for extra FPIN packets before failing.
Relevance: 	27xx adapters and later

ER148082:	Host side throttling: seeing continous FPIN error messages in the logs
Resolution:	A new snapshot of the response queue in pointer needed to be read
		after processing FPIN entries to avoid getting out of sync with firmware
		and reading stale entries.
Relevance:	27xx adapters and later

ER148097:	Host stays in congested state with SCM signals
Resolution:	Set correct event and throttle period values when receiving signals.
		Also, check for cleared congestion only after handling throttling state.
Relevance:	28xx adapters

Between versions 3.1.62.0 and 3.1.63.0:

ER148066:	Q depth throttling messages show negative values
Resolution:	Manage qdepth using SRB flag to account for commands that are not
		counted towards qdepth calculation.
Relevance:	27xx adapters and later

ER148138:	Congestion Severity shows None for Signals in esxcli
Resolution:	Populate congestion severity when receiving signals based on
		warning or alarms.
Relevance:	27xx adapters and later

ER148140:	Congestion Severity doesn't reset after clear event.
Resolution:	Clear congestion severity either when congestion clears naturally or
		through an FPIN event.
Relevance:	27xx adapters and later

ERXXXXXX:	Certain Enhanced Error Detection API calls need to be allowed to
		execute when port isolated.
Resolution:	Changed API interface to allow or not allow EED calls to go through
		based on requirements.  Also, increment target short link down timeout
		when target logs out from the initiator.
Relevance:	27xx adapters and later

ERXXXXXX:	ISP Abort was taking a long time to trigger when MB Cmds were active
Resolution:	Reduce MB Cmd timeout to 5 seconds when an ISP abort is needed
Relevance:	All adapters


Between versions 3.1.63.0 and 3.1.64.0:

ERXXXXXX:	MB Cmd timeouts seen during Enhanced Error Detection port disable/enable testing
Resolution:	Don't return enable API call until FW is initialized.  Also, clear
		initDone flag with port disabled to allow MB Cmd polling instead of
		waiting for interrupt.
Relevance:	27xx adapters and later

ER148213:	Disable Host and Peer throttling using module parameter system
		don't show congested state.
Resolution:	Make calls to set/clear congestion separate from the throttling
		functions and put in appropriate module parameter check only for
		throttling
Relevance:	27xx adapters and later

ER148079: 	DDV EILoading runs for an indefinite of time if the datastores
		are mapped
Resolution:	Don't destory DPC world until after the priority workqueues have been
    		destroyed
Relevance:	All adapters

ER148145:	Cable unplugged messages are not seen on vmkernel when SFP is
		present but no cable
Resolution:	In timer task, print cable unplugged after loop down time if FW never
		became ready
Relevance:	All adapters

ERXXXXXX:	Adapters older than 27xx do not have shadow registers so need to
		directly read chip registers to determine response in pointer
Resolution:	Turn off ql2xrspq_follow_inptr_legacy by default for now to avoid
		reading response queue in register during interrupt processing on
		older chips.
Relevance:	25xx and Hilda adapters


Between versions 3.1.64.0 and 3.1.65.0:

ERXXXXXX:	Failures seen during DDV testing with 7.0 inbox driver	
Resolution:	Improvements to handle DDV testing at VMWare
Relevance:	All adapters

ERXXXXXX:	Fabric zone deletion with 1 target did not increment target short
		link down counter
Resolution:	Logout all target ports when GPN_FT fails after 5 attempts.
Relevance:	All adapters

ER148223:	Logs are continuously flooded with ADISC failure messages during
		IOM disruptive activities
Resolution:	Clear logoutCompleted flag before sending logout after target is
		removed from fabric - this prevents subsequent logout and ADISC failures
		in the event of target logging into adapter first.
Relevance:	All adapters


Between versions 3.1.65.0 and 3.1.66.0:

ERXXXXXX:	The NVMe event list spinlock was at a lower rank than the hardware lock which
		was used when assigning IDs from the targetIds array, resulting
		in PSOD during inbox testing at VMWare.
Resolution:	Create new target ids spinlock to handle target ID assignment with a
		lower rank than the event list lock.
Relevance:	27xx adapters and later

ERXXXXXX:	Enhanced Error Detection port isolation operation could be delayed
		and/or return busy when driver is performing an ISP abort until
		after link up occurs
Resolution:	Allow disable port API call to be made regardless of DPC thread
		activity but wait for activity to quiesce.  Set port isolated flags
		early in the call so ISP abort handler can be short Short circuited.
Relevance:	27xx adapters and later


Between versions 3.1.66.0 and 3.1.67.0:

DCPN73306:	DDV::EILoading test hit a PSOD - ASSERT bora/vmkernel/main/fastslab.c:2230
Resolution:	Prevent various work threads when driver is unloading to prevent heap
		memory not getting completely freed.
Relevance:	All adapters

ERXXXXXX:	DDV testing resulted in PSOD with driver worlds still active
Resolution:	Check status on vmk_WorldWaitForDeath to prevent driver unload with
		world still active
Relevance:	All adapters

ERXXXXXX:	ql2xuseshadowregisters wasn't used consistently throughout the code.
Resolution:	Use qpair->useShadowReg as the main flag to check if the adapter
		is shadow register capable AND module parameter is enabled.
Relevance:	27xx adapters and later	

ERXXXXXX:	Driver was not using proper Slab ID check before destroying slabs
		during driver unload - found during code review.
Resolution:	Add proper check to Slab ID before freeing up slab.
Relevance:	All adapters

Between versions 3.1.67.0 and 3.1.68.0:

ERXXXXXX: 	USCM: Firmware will fail to go ready while waiting for EDC and RDF ELS commands
		to complete to switch.
Resolution:	Feature that enables driver to control the sending of the EDC and
		RDF commands.  Controlled by module parameter ql2xcontrol_edc_rdf,
		enabled by default.
Relevance:	27xx adapters and later


Between versions 3.1.68.0 and 3.1.69.0:

FCD-92:		USCM: ESX Cli and VI-plug scmstats, scmchk failing.
Resolution:	When driver controlled RDF completes successfully, set SCM_FLAG_RDF_COMPLETED
		in scmFabricConnectionFlags.
Relevance:	27xx adapters and later

FCD-93:		UCSM: Driver controlled EDC commands failing on Qlipper adapters
Resolution:	Driver was populating TX buffer length in EDC ELS passthrough IOCB incorrectly
		when only CSC descriptor was present.  Initialize buffer length to
		include 8 bytes of opcode and descriptor length regardless of which
		descriptors present.
Relevance:	27xx adapters and later


Between versions 3.1.69.0 and 3.1.70.0:

ER148141:	USCM: Driver vmkmgt interface doesn't differentiate FPIN vs signal
		alarm/warnings in the Host Stats counters display.
Resolution:	Introduced separate internal counters to track FPIN vs signal
		alarms/warnings.
Relevance:	27xx and later

FCD-109:	USCM: EDC commands failing on Qlipper adapters
Resolution:	Buffer byte count was incorrectly set for adapters that supported
		LFC descriptor in EDC command.  Also, made some improvements to the error
		handling code.
Relevance:	27xx adapters

FCD-XX:		USCM: ql2xcontrol_edc_rdf was missing from vmkmgmt list
Resolution:	Add ql2xcontrol_edc_rdf to vmkmgmt module parameter list
Relevance:	27xx adapters and later

FCD-XX:		USCM: Log messages show throttling state being cleared whenever target relogin occurred
		regardless of throttling state
Resolution:	Avoid clearing throttling state when already cleared during device resync
Relevance:	27xx adapters and later

DCPN73768:	qpair spinlock consumes more CPU cycles on VVOL workload compared to VMFS
Resolution: 	Various improvements made in the performance path to optimize usage of qpair lock
		as well as avoiding non-slab memory allocs.
Relevance:	All adapters

ER148590:	PSOD while powering on VM with NPIV enabled
Resolution:	Check vmk_WorldWaitForDeath status on scsiScan world before continuing
		to execute vport deletion
Relevance:	All adapters


Between versions 3.1.70.0 and 3.1.71.0:

FCD-XXX: 	USCM: Capability to get accurate signal count not utilized.
Resolution:	Retrieve signal count from FW using MB Cmd; display in logs
		for now - later will be incorporated in UCSM main scheme.
Relevance:	27xx adapters and later

FCD-153:	USCM: Internal congestion alarm/warning counters displayed in vmkmgmt don't get cleared
Resolution:	Clear internal alarm and warning counters when application requests.
Relevance:	27xx adapters and later

FCD-137/
FCD-117:	PSOD during DDV testing
Resolution:	Check memory buffer allocation was successful before calling qlnativefcGetParamsMbx
		during driver initializaion.
Relevance:	27xx adapters and later

DCPN80597:	DDV eiloading test hanging for a long time
Resolution:	A new flags is introduced that will trigger a world exit when a world
		is killed during driver unload.  This is done to prevent missed
		VMK_DEATH_PENDING status when the world function contains multiple wait
		related system calls.
Relevance:	All adapters

FCD-143:	PSOD on VM powering on VMs
Resolution:	Removed one aspect of changes made for DCPN73768 as this was leading
		to two threads accessing the oustanding command array.
Relevance:	All adapters

Between versions 3.1.71.0 and 3.1.72.0:

FCD-181:	8G adapter failng login when fw loaded without multi-queue support
Resolution:	Error path when driver encounters non-MQ fw did not handle resetting
		request and response queue pointers to point to proper address.
		Fix was to ensure if firmware does not support multi-queue
		to re-init pointer registers properlya
Relevance:	25xx adapters

DCPN78849:	PSOD encountered during DDV testing in qlnativefcDmaUnmap call
Resolution:	Perform DMA buffer NULL check before deallocating the buffer.
Relevance:	All adapters

DCPN78850:	PSOD in qlnativefcFreeSpPool during DDV eiloading testing
Resolution:	Ensure all SRB queued up for deferred work are free before destroying
		the memory slab
Relevance:	All adapters

FCD-XXX:	UCSM: Remove signal count message in log added in last driver version.
Resolution:	Deferred until next phase of UCSM support
Relevance:	27xx adapters and later

FCD-243: 	USCM: EDC/RDF reject response code shows 1 for all error
		conditions
Resolution:	Print out the correct field in the response frame.  Also, made some log
		message improvements.
Relevance:	27xx adapters and later

FCD-236:	vmhba info not in stats section of vmkmgmt output
Resolution:	Added vmhba info to SCM_STATS and EED_STATS output
Relevance:	All adapters

ER148589:	PSOD seen during loopback testing
Resolution:	Buffer used for command passthrough and loopback testing was shared
		without protection from multiple threads making API calls.  Fix
		was to protect the buffer using a sempahore.
Relevance:	All adapters

Between versions 3.1.72.0 and 3.1.73.0:

FCD-XXX:	Debug code caused 8G adapter to fail setting up CPU affinity qpairs
Resolution:	Remove debug code in driver
Relevance:	25xx adapters

Between versions 3.1.73.0 and 3.1.74.0:

FCD-278:	PSOD observed while running DDV EILoading
Resolution:	Semaphore to protect vmk API buffer needs to be checked for NULL before destroying
		in detach routine.
Relevance:	All adapters

FCD-277:	PSOD observed while running DeviceStateChange IOVP test
Resolution:	Ensure all DMA buffers used in work objects get cleaned up in detach function
Relevance:	All adapters

Between versions 3.1.73.0 and 3.1.74.0:

FCD-298:	PSOD observed while running DDV EILoading
Resolution:	Ensure all DMA buffers used in work objects get cleaned up in detach function,
		key off of type field instead of buffer pointer values
Relevance:	All adapters

Between versions 3.1.74.0 and 3.1.83.0:

None.


Between versions 3.1.83.0 and 3.1.84.0:


FCD-396:	Spinlock rank PSOD with ESX beta build
Resolution:	Ensure proper lock rank when creating various spinlocks
Relevance:	All adapters

FCD-439:	No info about encrypted target ports in FC Target list in vmkmgmt output.
Resolution:	Added StorCryption target port info in list
Relevance:	28xx adapters and later

FCD-438:	StorCryption management port shows up in vmkmgmt output in Non-Target FC
		Port list
Resolution:	Only print out whether StorCryption management port is enabled/disabled; remove
		from non-target list
Relevance:	28xx adapters and later

FCD-440:	StorCryption debug logging messages showing up in standard trace
Resolution:	Remove StorCryption deubg messages - no longer needed with stable driver
Relevance:	28xx adapters and later

FCD-425:	Unable to disable StorCryption support in driver from module parameter
Resolution:	Added ql2xstorcryption module parameter (enabled by default)
Relevance:	28xx adapters and later

FCD-432:	Not able to see any encrypted Lun/Lun information from esxcli qlfc
Resolution:	Allow SCSI Test Unit Ready commands from application to execute without
		enabling encryption in IOCB.
Relevance:	All adapters

FCD-251:	Loopback failures with NVMe targets connected
Resolution:	Due to NVMe exchanges being persistently active for Async Event
		Requests, the FW requires the driver to build the Echo payload
		themselves for the Diagnostic Echo MB Cmd rather than having the
		FW send a vendor-unique Echo command.
Relevance:	All adapters


Between versions 3.1.84.0 and 3.1.85.0:

FCD-236:	MBI version info missing from vmkmgmt output
Resolution:	Driver records MBI version during start of day and
		prints MBI version in vmkmgmt output
Relevance:	26xx adapters and later

FCD-421: 	PSOD while running Driver Load Unload IOVP Test
Resolution:	Properly cleanup buffers used for Storcryption in driver unload and buffer
		allocation failure paths.
Relevance:	27xx adapters and later


Between versions 3.1.85.0 and 3.1.86.0:

FCD-XXX:	"-debug" string shown in driver version
Resolution:	ql2xextended_error_logging was inadvertently enabled by default.
Relevance:	All adapters

FCD-612:	PSOD during driver unload with ESX 6.7 beta build.
Resolution:	Remove extra deallocation of Storcryption bit vectors.
Relevance:	All adapters


Between versions 3.1.86.0 and 3.1.87.0:

FCD-641:	Debug messages printed in FCP-SCSI I/O path for non-standard
		SCSI read/write opcodes.
Resolution:	Remove debug messages
Relevance:	All adapters.




2. Known Issues

* Setting ql2xattemptdumponpanic=1 not supported if dump partition is on SAN


3. Notices

Information furnished in this document is believed to be accurate and
reliable. However, Marvell, Inc. assumes no responsibility for
its use, nor for any infringements of patents or other rights of
third parties which may result from its use.  Marvell, Inc. 
reserves the right to change product specifications at any time
without notice. Applications described in this document for any of
these products are only for illustrative purposes.  Marvell, Inc. 
makes no representation nor warranty that such applications are
suitable for the specified use without further testing or
modification. Marvell, Inc. assumes no responsibility for any
errors that may appear in this document.


4. Contacting Support

   For further assistance, contact Marvell Technical Support at:
   http://support.marvell.com


(c) Copyright 2021. All rights reserved worldwide. Marvell, Inc, the Marvell 
logo, and the Powered by Marvell logo are registered trademarks of
Marvell, Inc. All other brand and product names are trademarks
or registered trademarks of their respective owners.