Академический Документы
Профессиональный Документы
Культура Документы
. Exercise RSS,
observe RSS OID's occurring (either with oids.exe to monitor stats o
r via debug output in
checked build of base driver), and observe that the FCoE cpus do *no
t* change when the RSS
cpus change.
Defect
hsd_ladsw 3331784: Low network througput with Jumbo Frames + Lewisville
Sugar Bay & Huron River.
Root Cause:
This issue is caused by a large number of missed packets that could
not be DMA-ed to host
memory, thereby resulting in bad throughput numbers. It is directly
related to the C6 CPU
state on the platform. If C6 is disabled in the BIOS, this issue doe
s not occur. The exit
latency from C6 appears to be more than the time required to get the
packet into Host memory
and hence large numbers of packets are dropped (because there is no
more space in the Packet
Buffer to hold them)
Resolution Notes:
Modified the ITR algorithm function to add a work-around that accoun
ts for missed packets (as
determined by the MPC register) and pegs the CurrentITRMode at Lowes
tLatency. The work-around
will ensure that the LowestLatency mode stays for at least 100 DPCs
before we allow it to
increment to the next order of magnitude (LowLatency)
Build Changes
hsd_ladsw 3332476: E1C driver does not set Host LinkSec Connection Acti
ve (LSECA) bit in the
H2ME register when enabling MACsec
Brief Description:
clears H2ME bits for linksec connections.
Other
Description:
Moved all LSC processing to a work-item thread to improve system per
formance
Changes Made:
Moved all of the Link Status Change (LSC) processing code from the D
PC function to a work-item
function. The advantage gained here is that work item threads run at
IRQL=PASSIVE_LEVEL and
hence do not stall the CPU (or the system) as it would when processe
d in a DPC
Other
Description:
Functions IncrementAdapterIndex() and GetAdapterIndex() need to retu
rn the index of the
adapter (usually AdapterNumber - 1)
Changes Made:
Modified IncrementAdapterIndex() and GetAdapterIndex() to properly r
eturn AdapterIndex of 1
less than AdapterCount.
Defect
hsd_ladsw 3332672: Spring Fountain - NDOS Tester 6.0 - 1c_faulthandling
test causes the system
to bug check
Other
Description:
A number of Klocwork fixes for null pointers, errors with classes, a
nd uninitialized variables
Changes Made:
A number of Klocwork fixes for null pointers, errors with classes, a
nd uninitialized variables
New Feature
Feature Description:
DCR 996 - Event Log Enhancement
Changes Made:
Added code to extract the bus, device and function corresponding to
the instance.
Added code to embed this informatio
n in the event log if the branding string (part of event log) exceeds the MAX li
mit, there by adding identification (which was lost earlier) for corresponding d
river instance.
Other
Description:
AIFS oid fixes got reverted, added back for HSD 3145287
Changes Made:
AIFS oid fixes got reverted, added back for HSD 3145287
Other
Description:
added full support for reading and writing to VF PCI config space
Changes Made:
pass untrapped calls to Ndis to get HW values
Defect
HSD LADSW 3332679 Customized LED value in NVM not applied with LV Beta S
W
Root Cause
The LED writes (and smbus address) were not taking effect because sw
has gated ppw and the
driver is checking for oem write enable bits
Resolution
Removed the check for oem write enable bits for Nahum5
****************************
All Windows Versions 11.8.67.0 (Pci Express) NOT Digitally signed
Date: 08/19/2010
Comments:
Defect
hsd_ladsw 3332339: CERTS: Lewisville fails WoLMagicPacket
New Feature
Feature Description:
Interrupt refactoring to decouple User Priority to Traffic Class map
ping, so that FCOE will
use a fixed TC.
Changes Made:
Base driver reads the UP-TC mapping given by DCB to determine which
TC to use given the UP.
New Feature
Feature Description:
Iteration on new OID format to supply information on mapped queues
in redundant but more
convenient form.
Changes Made:
In addition to supplying a QueueBitmap for each cpu that has one or
more queues mapped,
supply an absolute queue number and a count indicating how many que
ues mapped to the CPU.
Queues>1 are not expected in practice, but the immutable nature of
the OID generally benefits
from fully complete expression.
Testing Hints:
Write FCoE query OID code per the new format. In the FCoE protocol
driver, read back Queue
and QueuesMapped values for CPUs that have queues mapped.
Defect
hsd_ladsw 3332727: Windows FCoE: Basic LUN connectivity issues with refactored RSS changes
Defect
hsd_ladsw 3332794: Lewisville B0 fails to link correctly when forced to
full duplex
Other
Description:
The FCoE driver saw asserts because the single ether type setting fo
r TC3 was incorrect.
Changes Made:
Queues 48-55 are for FCoE. The traffic on these queues is FCOE and ot
her packets for TC 3.
Therefore we cannot set single ether type flag just on a queue just b
ecause we detected an
fcoe packet. The driver now checks the ethertype of every NBL in the
list. If they are all
ethertype FCoE on the given queues, the single ether type flag is set
.
Other
Description:
Defect
hsd_ladsw 3705060: PROMOTE from sighting: AT7 -Power: ME doesn't answer
to ping in S3/S4 when
"wake on pattern match" is set in Host driver.
Defect
hsd_ladsw 3705217: PROMOTE from sighting: AT7 -Power:
lost in S3/M3 -ACDC
after Resume from S3-DC
Root Cause:
The driver did not check for changes to the Offload ME state during
Power Up, only during
ME interrupts and driver enable/disable.
Resolution Notes:
- Always checks the ME valid bit and proxy enable bit after resume,
during MNG interrupts,
and before entering PowerDown flow (to avoid stalling the ME in Sx
).
- Disables ARP/NS offload in S5, there were times where the OS did n
ot remove the patterns,
so I force the disabling of offload in S5.
- Added a new key for keeping track if ARP and/or NS offload is acti
vated by NDIS or ME,
and gave the Reg keys it's own key.
API Change
Brief Description:
Issue found by SWE.
Changes Made:
For PCH2, the LAA MAC address was not offloaded in the ME-based ARP
offload flow. and that
the PHY based solution did offload the LAA address. This was because
updating the NVM
checksum auto-loads the NVM mac address. Writing the current MAC add
ress back into RAR0 in
the MAC and PHY resolves the bug.
Testing Hints:
Verify that the system wakes and offloads ARP from the LAA when pres
ent in the driver
advanced properties tab.
****************************