Add script to replace [0] with [] when used at the end of a struct.
The script also includes an additional struct member to match against so
as to avoid issues with arrays with only a single zero-length element.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Morten Brørup <mb@smartsharesystems.com>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>
Acked-by: Hemant Agrawal <hemant.agrawal@nxp.com>
PAC N6000 is the first OFS platform, its device id is added to ifpga
device support list.
Previous FPGA platform like Intel PAC N3000 and N5000, FME DFL (Device
Feature List) starts from BAR0 by default, port DFL location is indicated
in PORTn_OFFSET register in FME. In OFS implementation, FME DFL and port
DFL location can be defined individually in PCIe VSEC (Vendor Specific
Extended Capabilities). In this patch, DFL definition is searched in VSEC,
the legacy DFL is used only when DFL VSEC is not present.
In original DFL enumeration process, AFU is expected to locate in port DFL,
but this is not the case in OFS implementation. In this patch, enumeration
can search AFU in any PF/VF which has no FME and port.
Signed-off-by: Wei Huang <wei.huang@intel.com>
Acked-by: Tianfei Zhang <tianfei.zhang@intel.com>
Reviewed-by: Rosen Xu <rosen.xu@intel.com>
There is an API rte_pmd_ifpga_cleanup provided by ifpga driver to
free the software resource used by ifpga card. The function call
of rte_pmd_ifpga_cleanup is list below.
rte_pmd_ifpga_cleanup()
ifpga_rawdev_cleanup()
rte_rawdev_pmd_release()
rte_rawdev_close()
ifpga_rawdev_close()
The interrupts are unregistered in ifpga_rawdev_destroy instead of
ifpga_rawdev_close function, so rte_pmd_ifpga_cleanup cannot free
interrupt resource as expected.
To fix such issue, interrupt unregistration is moved from
ifpga_rawdev_destroy to ifpga_rawdev_close function. The change of
function call of ifpga_rawdev_destroy is as below.
ifpga_rawdev_destroy()
ifpga_unregister_msix_irq() // removed
rte_rawdev_pmd_release()
rte_rawdev_close()
ifpga_rawdev_close()
Fixes: e0a1aafe2a ("raw/ifpga: introduce IRQ functions")
Cc: stable@dpdk.org
Signed-off-by: Wei Huang <wei.huang@intel.com>
Acked-by: Tianfei Zhang <tianfei.zhang@intel.com>
Reviewed-by: Rosen Xu <rosen.xu@intel.com>
Virtual devices created on ifpga raw device will not be removed
when ifpga device has closed. To avoid resource leak problem,
this patch introduces an ifpga virtual device remove function,
virtual devices will be destroyed after the ifpga raw device closed.
Fixes: ef1e8ede3d ("raw/ifpga: add Intel FPGA bus rawdev driver")
Cc: stable@dpdk.org
Signed-off-by: Wei Huang <wei.huang@intel.com>
Acked-by: Tianfei Zhang <tianfei.zhang@intel.com>
Reviewed-by: Rosen Xu <rosen.xu@intel.com>
These APIs are introduced in DPDK 21.05 and have been tested in several
release, experimental tag can be formally removed.
Signed-off-by: Wei Huang <wei.huang@intel.com>
Acked-by: Tianfei Zhang <tianfei.zhang@intel.com>
Reviewed-by: Rosen Xu <rosen.xu@intel.com>
Add functions for setting and getting the priority of a thread.
Priorities on multiple platforms are similarly determined by a priority
value and a priority class/policy.
Currently in DPDK most threads operate at the OS-default priority level
but there are cases when increasing the priority is useful. For
example, high performance applications may require elevated priority
levels.
For these reasons, EAL will expose two priority levels which are named
suggestively "normal" and "realtime_critical" and are computed as
follows:
On Linux, the following mapping is created:
RTE_THREAD_PRIORITY_NORMAL corresponds to
* policy SCHED_OTHER
* priority value: (sched_get_priority_min(SCHED_OTHER) +
sched_get_priority_max(SCHED_OTHER))/2;
RTE_THREAD_PRIORITY_REALTIME_CRITICAL corresponds to
* policy SCHED_RR
* priority value: sched_get_priority_max(SCHED_RR);
On Windows, the following mapping is created:
RTE_THREAD_PRIORITY_NORMAL corresponds to
* class NORMAL_PRIORITY_CLASS
* priority THREAD_PRIORITY_NORMAL
RTE_THREAD_PRIORITY_REALTIME_CRITICAL corresponds to
* class REALTIME_PRIORITY_CLASS (when running with privileges)
* class HIGH_PRIORITY_CLASS (when running without privileges)
* priority THREAD_PRIORITY_TIME_CRITICAL
Note that on Linux the resulting priority value will be 0, in
accordance to the documentation that mention the value should be 0 for
SCHED_OTHER policy.
Signed-off-by: Narcisa Vasile <navasile@linux.microsoft.com>
Signed-off-by: Tyler Retzlaff <roretzla@linux.microsoft.com>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>
A sequence lock (seqlock) is a synchronization primitive which allows
for data-race free, low-overhead, high-frequency reads, suitable for
data structures shared across many cores and which are updated
relatively infrequently.
A seqlock permits multiple parallel readers. A spinlock is used to
serialize writers. In cases where there is only a single writer, or
writer-writer synchronization is done by some external means, the
"raw" sequence counter type (and accompanying rte_seqcount_*()
functions) may be used instead.
To avoid resource reclamation and other issues, the data protected by
a seqlock is best off being self-contained (i.e., no pointers [except
to constant data]).
One way to think about seqlocks is that they provide means to perform
atomic operations on data objects larger than what the native atomic
machine instructions allow for.
DPDK seqlocks (and the underlying sequence counters) are not
preemption safe on the writer side. A thread preemption affects
performance, not correctness.
A seqlock contains a sequence number, which can be thought of as the
generation of the data it protects.
A reader will
1. Load the sequence number (sn).
2. Load, in arbitrary order, the seqlock-protected data.
3. Load the sn again.
4. Check if the first and second sn are equal, and even numbered.
If they are not, discard the loaded data, and restart from 1.
The first three steps need to be ordered using suitable memory fences.
A writer will
1. Take the spinlock, to serialize writer access.
2. Load the sn.
3. Store the original sn + 1 as the new sn.
4. Perform load and stores to the seqlock-protected data.
5. Store the original sn + 2 as the new sn.
6. Release the spinlock.
Proper memory fencing is required to make sure the first sn store, the
data stores, and the second sn store appear to the reader in the
mentioned order.
The sn loads and stores must be atomic, but the data loads and stores
need not be.
The original seqlock design and implementation was done by Stephen
Hemminger. This is an independent implementation, using C11 atomics.
For more information on seqlocks, see
https://en.wikipedia.org/wiki/Seqlock
Acked-by: Morten Brørup <mb@smartsharesystems.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Reviewed-by: Ola Liljedahl <ola.liljedahl@arm.com>
Reviewed-by: Chengwen Feng <fengchengwen@huawei.com>
Signed-off-by: Mattias Rönnblom <mattias.ronnblom@ericsson.com>
musl lacks __ppc_get_timebase() but has __builtin_ppc_get_timebase()
Signed-off-by: Duncan Bellamy <dunk@denkimushi.com>
Reviewed-by: David Christensen <drc@linux.vnet.ibm.com>
This patch fixes the following build failure seen on Ubuntu 16.04
with gcc 5.4.0 because of uninitialized variable:
[..]
examples/pipeline/cli.c:2853:9: error: 'session_id' may be used
uninitialized in this function [-Werror=maybe-uninitialized]
[..]
Fixes: 172254555f ("examples/pipeline: support packet mirroring")
Signed-off-by: Ali Alnubani <alialnu@nvidia.com>
Acked-by: Cristian Dumitrescu <cristian.dumitrescu@intel.com>
The device config script currently uses some defaults to configure
devices in a generic way.
With the addition of this option, users have more control over how
queues are configured.
Signed-off-by: Kevin Laatz <kevin.laatz@intel.com>
Reviewed-by: Bruce Richardson <bruce.richardson@intel.com>
Tested-by: Sunil Pai G <sunil.pai.g@intel.com>
The CQ (completion queue) descriptors were updated by hardware, and then
scanned by driver to retrieve hardware completion status.
This patch enhances robustness by following:
1. replace while (true) with a finite loop to avoid potential dead loop.
2. check the csq_head field in CQ descriptor to avoid status array
overflows.
Fixes: 2db4f0b823 ("dma/hisilicon: add data path")
Cc: stable@dpdk.org
Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
If no DMA request is completed, the ring_idx of the last completed
operation need returned by last_idx parameter. This patch adds
testcase for it.
Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
Tested-by: Kevin Laatz <kevin.laatz@intel.com>
If no DMA request is completed, the ring_idx of the last completed
operation need returned by last_idx parameter. This patch fixes it.
Fixes: 2db4f0b823 ("dma/hisilicon: add data path")
Cc: stable@dpdk.org
Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
This patch adds force minimal copy size parameter
(-m/--force-min-copy-size), so when do copy by CPU or DMA, the real copy
size will be the maximum of mbuf's data_len and this parameter.
This parameter was designed to compare the performance between CPU copy
and DMA copy. User could send small packets with a high rate to drive
the performance test.
Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Kevin Laatz <kevin.laatz@intel.com>
The Tx drop statistic was designed to be collected by
rte_eth_dev_tx_buffer mechanism, but the application uses
rte_eth_tx_burst to send packets and this lead the Tx drop statistic
was not collected.
This patch removes rte_eth_dev_tx_buffer mechanism to fix the problem.
Fixes: 632bcd9b5d ("examples/ioat: print statistics")
Cc: stable@dpdk.org
Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Kevin Laatz <kevin.laatz@intel.com>
The MTU in dma App can be configured by 'max_frame_size' parameters which
have a default value(1518). It's not reasonable to use it directly as MTU.
This patch fix it.
Fixes: 1bb4a528c4 ("ethdev: fix max Rx packet length")
Cc: stable@dpdk.org
Signed-off-by: Huisong Li <lihuisong@huawei.com>
Telemetry commands are now registered through the dmadev library
for the gathering of DSA stats. The corresponding callback
functions for listing dmadevs and providing info and stats for a
specific dmadev are implemented in the dmadev library.
An example usage can be seen below:
Connecting to /var/run/dpdk/rte/dpdk_telemetry.v2
{"version": "DPDK 22.03.0-rc2", "pid": 2956551, "max_output_len": 16384}
Connected to application: "dpdk-dma"
--> /
{"/": ["/", "/dmadev/info", "/dmadev/list", "/dmadev/stats", ...]}
--> /dmadev/list
{"/dmadev/list": [0, 1]}
--> /dmadev/info,0
{"/dmadev/info": {"name": "0000:00:01.0", "nb_vchans": 1, "numa_node": 0,
"max_vchans": 1, "max_desc": 4096, "min_desc": 32, "max_sges": 0,
"capabilities": {"mem2mem": 1, "mem2dev": 0, "dev2mem": 0, ...}}}
--> /dmadev/stats,0,0
{"/dmadev/stats": {"submitted": 0, "completed": 0, "errors": 0}}
Signed-off-by: Sean Morrissey <sean.morrissey@intel.com>
Reviewed-by: Bruce Richardson <bruce.richardson@intel.com>
Reviewed-by: Conor Walsh <conor.walsh@intel.com>
Tested-by: Sunil Pai G <sunil.pai.g@intel.com>
Tested-by: Kevin Laatz <kevin.laatz@intel.com>
Acked-by: Chengwen Feng <fengchengwen@huawei.com>
Clarify that once an operation has completed, the output of that
operation is visible to all cores.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Morten Brørup <mb@smartsharesystems.com>
A previous fix had #else instead of #endif.
The error message is:
kernel/linux/kni/kni_net.c: In function ‘kni_net_rx_normal’:
kernel/linux/kni/kni_net.c:448:2: error: #else after #else
Bugzilla ID: 1025
Fixes: c98600d4be ("kni: fix build with Linux 5.18")
Cc: stable@dpdk.org
Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
ESP item is not supported on Windows, yet it is expanded from the
expansion graph when trying to create default flow to RSS all packets.
Support ESP item match (without ability to match on SPI field on Windows).
Split ESP validation per OS.
Signed-off-by: Raja Zidane <rzidane@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
The mlx5_action_construct_data structure memory is managed by ipool
named acts_ipool.
The size of one entry in this ipool is mistakenly defined as size of
rte_flow_hw structure.
This size is used to reset in the allocated part. When the size is
incorrect it resets memory that does not belong to it.
This patch defines the correct size.
Fixes: f13fab2392 ("net/mlx5: add flow jump action")
Cc: stable@dpdk.org
Signed-off-by: Michael Baum <michaelba@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
While non-lcore list operations were supported, non-lcore index will
be converted to MLX5_LIST_NLCORE. In that case, no need to check the
lcore index be -1 or not anymore.
This commit removes the unused lcore check in list.
Fixes: 7e1cf89271 ("common/mlx5: support list non-lcore operations")
Cc: stable@dpdk.org
Signed-off-by: Suanming Mou <suanmingm@nvidia.com>
Acked-by: Matan Azrad <matan@nvidia.com>
Kernel iavf driver has sent patch to increase the completion
wait time to reduce the "Reset never finished" case.
Follow this action in DPDK iavf driver.
Kernel reference commit:
8e3e4b9da7e6 ("iavf: increase reset complete wait time")
Fixes: 22b123a36d ("net/avf: initialize PMD")
Cc: stable@dpdk.org
Signed-off-by: Qiming Yang <qiming.yang@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
When kernel driver does not support quanta size configuration,
it will return error. We do not expect it to occur in default
initialization process.
Fixes: b14e8a57b9 ("net/iavf: support quanta size configuration")
Signed-off-by: Wenjun Wu <wenjun1.wu@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
Added I226 Series device ID in igc driver and updated igc guide
document for new devices.
Signed-off-by: Qiming Yang <qiming.yang@intel.com>
Signed-off-by: Kevin Liu <kevinx.liu@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
Move security context destroy from device stop to device close function.
Deleting the context on device stop can prevent the application from
properly cleaning and releasing resources.
Fixes: 6bc987ecb8 ("net/iavf: support IPsec inline crypto")
Cc: stable@dpdk.org
Signed-off-by: Radu Nicolau <radu.nicolau@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
When the inline crypto feature VF capability flag is set also check if the
feature is enabled, otherwise the initialization will fail even when
the inline crypto is not required.
Fixes: 6bc987ecb8 ("net/iavf: support IPsec inline crypto")
Cc: stable@dpdk.org
Signed-off-by: Radu Nicolau <radu.nicolau@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
In multi-cores cases for Rx timestamp offload, if packets arrive
too fast, aq command to get phc time will be pended.
This patch adds spinlock to fix this issue. To avoid phc time being
frequently overwritten, move related variables to iavf_rx_queue
structure, and each queue will handle timestamp calculation by itself.
Fixes: b5cd735132 ("net/iavf: enable Rx timestamp on flex descriptor")
Fixes: 33db16136e ("net/iavf: improve performance of Rx timestamp offload")
Signed-off-by: Wenjun Wu <wenjun1.wu@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
Enable Protocol Agnostic Flow Offloading for RSS hash in VF. It supports
raw pattern flow rule creation in VF based on Parser Library feature. VF
parses the spec and mask input of raw pattern, and passes it to kernel
driver to create the flow rule. Current rte_flow raw API is utilized.
command example:
RSS hash for ipv4-src-dst:
flow create 0 ingress pattern raw pattern spec
00000000000000000000000008004500001400004000401000000000000000000000
pattern mask
0000000000000000000000000000000000000000000000000000ffffffffffffffff /
end actions rss queues end / end
Signed-off-by: Ting Xu <ting.xu@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
This patch enabled Protocol Agnostic Flow (raw flow) Offloading Flow
Director (FDIR) in AVF, based on the Parser Library feature and the
existing rte_flow `raw` API.
The input spec and mask of raw pattern are first parsed via the
Parser Library, and then passed to the kernel driver to create the
flow rule.
Similar as ice PMD's implemnentation, each raw flow requires:
1. A byte string of raw target packet bits.
2. A byte string contains mask of target packet.
Here is an example:
FDIR matching ipv4 dst addr with 1.2.3.4 and redirect to queue 3:
flow create 0 ingress pattern raw \
pattern spec \
00000000000000000000000008004500001400004000401000000000000001020304 \
pattern mask \
000000000000000000000000000000000000000000000000000000000000ffffffff \
/ end actions queue index 3 / mark id 3 / end
Note that mask of some key bits (e.g., 0x0800 to indicate ipv4 proto)
is optional in our cases. To avoid redundancy, we just omit the mask
of 0x0800 (with 0xFFFF) in the mask byte string example. The prefix
'0x' for the spec and mask byte (hex) strings are also omitted here.
Signed-off-by: Junfeng Guo <junfeng.guo@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
The patch extends existing virtchnl_proto_hdrs structure to allow VF
to pass a pair of buffers as packet data and mask that describe
a match pattern of a filter rule. Then the kernel PF driver is requested
to parse the pair of buffer and figure out low level hardware metadata
(ptype, profile, field vector.. ) to program the expected FDIR or RSS
rules.
Also update the proto_hdrs template init to align the virtchnl changes.
Signed-off-by: Qi Zhang <qi.z.zhang@intel.com>
Signed-off-by: Junfeng Guo <junfeng.guo@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
Add recommended matching list for i40e PMD in DPDK 21.05,
21.08, 21.11 and 22.03. And add a known issue when FW upgrade
to a version 8.4 and higher
Cc: stable@dpdk.org
Signed-off-by: Qiming Yang <qiming.yang@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
For Rx-Queue Interrupt Setting, when VF Rx interrupt
disable (INTENA=0), there are two ways to write back
descriptor to host memory:
1) Set WB_ON_ITR bit 0 to Interrupt Dynamic Control Register:
Completed descriptors are posted to host memory according to
the internal descriptor cache policy (in other words when a
full cache line is available for write-back).
A internal descriptor size is 16 bytes or 32 bytes, a cache
line size is 64 bytes or 128 bytes from datasheet :
PCIe Global Config 2 - GLPCI_CNF2 (0x000BE004; RO)
so the full cache line could contains 4 packets, it means
Network card will send 4 packets to host when a full cache line
is available.
2) Set WB_ON_ITR bit 1 to Interrupt Dynamic Control Register:
Completed descriptors also trigger the ITR. Following ITR
expiration, all leftover completed descriptors are posted to
host memory.
Network card will send packet to host even if only one
descriptor is completed.
Changing 1) to 2) to make sure VF send the packet to host even
if there is only one Rx packet is ready in hardware.
Fixes: d6bde6b5ea ("net/avf: enable Rx interrupt")
Cc: stable@dpdk.org
Signed-off-by: Ke Zhang <ke1x.zhang@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
In the multiple process environment, the subprocess operates on the
shared memory and changes the function pointer of the main process,
resulting in the failure to find the address of the function when main
process releasing, resulting in crash.
Fixes: 319c421f38 ("net/avf: enable SSE Rx Tx")
Cc: stable@dpdk.org
Signed-off-by: Ke Zhang <ke1x.zhang@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
If any queue start fail during dev_start, all started queues
should be stopped.
Fixes: 69dd4c3d08 ("net/avf: enable queue and device")
Cc: stable@dpdk.org
Signed-off-by: Qiming Yang <qiming.yang@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
Previously, max frame size can only be set when link is up, and the wait
time is 1 sec. Startup time of 10G_BASET longer than 1s would result in
failure.
Actually, max frame size of media type I40E_MEDIA_TYPE_BASET can be set
regardless of link status.
This patch omitted the link status check of 10G_MEDIA_TYPE_BASET.
Fixes: a4ba773679 ("net/i40e: enable maximum frame size at port level")
Cc: stable@dpdk.org
Signed-off-by: Wenxuan Wu <wenxuanx.wu@intel.com>
Acked-by: Yuying Zhang <yuying.zhang@intel.com>
Some pointers will be set to NULL when iavf_dev_reset() failed,
for example vf->vf_res, vf->vsi_res vf->rss_key and etc.
APIs access these NULL pointers will trigger segfault.
This patch adds closed flag to indicate that the VF is closed,
and rejects API calls in this state to avoid coredump.
Fixes: e74e1bb628 ("net/iavf: enable port reset")
Cc: stable@dpdk.org
Signed-off-by: Yiding Zhou <yidingx.zhou@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
In the DCF module, Missing maximum and minimum
MTU value settings.
This patch adds the settings of the maximum and
minimum MTU to correctly calculate the MTU value.
Fixes: bf89db4409 ("net/ice: complete device info get in DCF")
Cc: stable@dpdk.org
Signed-off-by: Kevin Liu <kevinx.liu@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
The tx/rx packets were both dropped when creating drop any rule
for ingress direction only, the root cause is the recipe didn't
contain direction flag matching.
This patch adds the packet flag which represents the direction of
source interface to solve the issue.
Fixes: 92317961a7 ("net/ice: support drop any and steer all to queue")
Cc: stable@dpdk.org
Signed-off-by: Yuying Zhang <yuying.zhang@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
Priority configuration is enabled in level 3 and level 4.
Weight configuration is enabled in level 4.
This patch adds warning log for unsupported priority
and weight configuration.
Signed-off-by: Wenjun Wu <wenjun1.wu@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
This patch adds queue and queue group priority configuration
support. The highest priority is 0, and the lowest priority
is 7.
Signed-off-by: Wenjun Wu <wenjun1.wu@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
Enable basic TM API for PF only. Support for adding profiles and queue
nodes. Only max bandwidth is supported in profiles. Profiles can be
assigned to target queues and queue group. To set up the exact queue
group, we need to reconfigure topology by delete and then recreate
queue nodes. Only TC0 is valid.
Signed-off-by: Wenjun Wu <wenjun1.wu@intel.com>
Signed-off-by: Ting Xu <ting.xu@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
This patch adds priority configuration support of the exact
node in the scheduler tree.
This function does not need additional calls to the scheduler
lock.
Signed-off-by: Wenjun Wu <wenjun1.wu@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
This patch adds BW allocation support of queue scheduling node
to support WFQ in queue level.
Signed-off-by: Wenjun Wu <wenjun1.wu@intel.com>
Acked-by: Qi Zhang <qi.z.zhang@intel.com>
