In case some CPUs don't support AVX512. Enable AVX2 for them to
get better per-core performance.
Signed-off-by: Wenzhuo Lu
---
doc/guides/rel_notes/release_24_03.rst | 3 +
drivers/common/idpf/idpf_common_device.h| 1 +
drivers/common/idpf/idpf_common_rxtx.h | 4 +
drivers/co
In case some CPUs don't support AVX512. Enable AVX2 for them to
get better per-core performance.
Signed-off-by: Wenzhuo Lu
---
drivers/common/idpf/idpf_common_device.h| 1 +
drivers/common/idpf/idpf_common_rxtx.h | 4 +
drivers/common/idpf/idpf_common_rxtx_avx2.c | 590 +
In case some CPUs don't support AVX512. Enable AVX2 for them to
get better per-core performance.
---
v2:
- Removed unused code.
*** BLURB HERE ***
Wenzhuo Lu (2):
common/idpf: enable AVX2 for single queue Rx
common/idpf: enable AVX2 for single queue Tx
doc/guides/rel_notes/release_24_03.r
From: Sunil Kumar Kori
Use the dpdk-cmdline-gen script to autogenerate all the boilerplate
structs and defines for the commandline part of the application.
Depends-on: series-30459 ("enhancements for dpdk-cmdline-gen script")
Signed-off-by: Sunil Kumar Kori
---
app/graph/cli.c
From: Pavan Nikhilesh
Update base code.
Signed-off-by: Pavan Nikhilesh
---
drivers/common/cnxk/hw/ssow.h | 4 +
drivers/common/cnxk/hw/tim.h| 5 +-
drivers/common/cnxk/roc_mbox.h | 11 +++
drivers/common/cnxk/roc_sso.c | 123 ++--
drivers/c
From: Pavan Nikhilesh
Allow dequeue timeout to be configured as zero, when
RTE_EVENT_DEV_CFG_PER_DEQUEUE_TIMEOUT is disabled.
Fixes: 5512c7de85cd ("event/cnxk: add common configuration validation")
Cc: sta...@dpdk.org
Signed-off-by: Pavan Nikhilesh
---
drivers/event/cnxk/cnxk_eventdev.c | 13
From: Pavan Nikhilesh
Optimize Rx routine to use AVX2 instructions when underlying
architecture supports it.
Signed-off-by: Pavan Nikhilesh
---
doc/guides/rel_notes/release_24_03.rst | 5 +
drivers/net/octeon_ep/cnxk_ep_rx_avx.c | 123 +
drivers/net/octeon_ep/meson.bu
From: Pavan Nikhilesh
Optimize Rx routine to use SSE instructions.
Signed-off-by: Pavan Nikhilesh
---
drivers/net/octeon_ep/cnxk_ep_rx.c | 159 +--
drivers/net/octeon_ep/cnxk_ep_rx.h | 167 +
drivers/net/octeon_ep/cnxk_ep_rx_sse.c | 130 +
From: Pavan Nikhilesh
Preset rearm data to avoid writing multiple fields in fastpath,
Increase maximum outstanding Tx instructions from 128 to 256.
Signed-off-by: Pavan Nikhilesh
---
v4 Changes:
- Fix checkpatch.
- Update release notes.
v3 Chnages:
- Add more comments to the code.
- Re-enable 3
On Tue, Dec 5, 2023 at 3:16 PM Rakesh Kudurumalla
wrote:
>
> Add support for eth_rx_descriptor_dump for cn9k and cn10k.
> This patch dumps contents of receviced packet descriptor from CQ
> for debug to file
>
> Signed-off-by: Rakesh Kudurumalla
> ---
> drivers/net/cnxk/cn10k_ethdev.c | 67 ++
On Wed, Dec 6, 2023 at 10:54 PM wrote:
>
> From: Pavan Nikhilesh
>
> Optimize Rx routine to use SSE instructions.
>
> Signed-off-by: Pavan Nikhilesh
1)Please fix the following typo
### [PATCH] net/octeon_ep: use SSE instructions for Rx routine
WARNING:TYPO_SPELLING: 'littel-endian' may be miss
On Tue, Dec 5, 2023 at 3:39 PM Rakesh Kudurumalla
wrote:
>
> Invoked newly added roc API to disable
> SQB aura FC during TX queue start and TX queue
> stop. This fix reduces ethdev teardown time
>
> Signed-off-by: Rakesh Kudurumalla
> ---
Updated the git commit as follows and Series applied to
In case some CPUs don't support AVX512. Enable AVX2 for them to
get better per-core performance.
Signed-off-by: Wenzhuo Lu
---
doc/guides/rel_notes/release_24_03.rst | 3 +
drivers/common/idpf/idpf_common_device.h| 1 +
drivers/common/idpf/idpf_common_rxtx.h | 4 +
drivers/co
In case some CPUs don't support AVX512. Enable AVX2 for them to
get better per-core performance.
Signed-off-by: Wenzhuo Lu
---
drivers/common/idpf/idpf_common_device.h| 1 +
drivers/common/idpf/idpf_common_rxtx.h | 4 +
drivers/common/idpf/idpf_common_rxtx_avx2.c | 609 +
In case some CPUs don't support AVX512. Enable AVX2 for them to
get better per-core performance.
Wenzhuo Lu (2):
common/idpf: enable AVX2 for single queue Rx
common/idpf: enable AVX2 for single queue Tx
doc/guides/rel_notes/release_24_03.rst | 3 +
drivers/common/idpf/idpf_common_devi
>
> Add external mbuf test for AES GCM aead algo.
>
> Signed-off-by: Aakash Sasidharan
Acked-by: Anoob Joseph
The Tx offload capabilities of ICE ethdev doesn't include
tso tunnel, which will result in tso tunnel setting to
not take effect.
The patch adds tunnel tso offload to ICE_TX_NO_VECTOR_FLAGS.
This commit will add tso tunnel capabilities in ice_dev_info_get().
Bugzilla ID: 1327
Fixes: d852fec1be63
Replace the use of gcc builtin __atomic_xxx intrinsics with
corresponding rte_atomic_xxx optional stdatomic API.
Signed-off-by: Jie Hai
---
drivers/net/hns3/hns3_cmd.c | 24 ++-
drivers/net/hns3/hns3_dcb.c | 3 +-
drivers/net/hns3/hns3_ethdev.c| 50 ++
From: Dengdui Huang
Currently, there are two problems in hns3vf_is_reset_pending():
1. The new detect reset level is not HNS3_NONE_RESET, but the
last reset level is HNS3_NONE_RESET, this function returns false.
2. Comparison between last_req and new_req is opposite.
In addition, the reset le
From: Dengdui Huang
Disable command only when need to delay handle reset.
This patch fixes it.
Fixes: 5be38fc6c0fc ("net/hns3: fix multiple reset detected log")
Cc: sta...@dpdk.org
Signed-off-by: Dengdui Huang
Signed-off-by: Jie Hai
---
drivers/net/hns3/hns3_ethdev.c | 8 +++-
1 file cha
From: Dengdui Huang
There are two ways for the hns3 VF driver to know reset event, namely,
interrupt task and periodic detection task. For the latter, the real
reset process will delay several microseconds to execute. Both tasks
cause the count to increase by 1.
However, the periodic detection t
This patchset fixes some bugs on reset, and replaces all
gcc builtin __atomic_xxx with rte_atomic_xxx.
--
v2:
1. remove mailbox bugfix which has been applied.
2. add new bugfixes on reset.
3. Add RTE_ATOMIC to related variable.
--
Dengdui Huang (3):
net/hns3: fix VF multiple count on one reset
From: Dengdui Huang
The 'hns3_send_mbx_msg' function has following problem:
1. the name is vague, missing caller indication.
2. too many input parameters because the filling messages
are placed in commands the send command.
Therefore, a common interface is encapsulated to fill in
the mailbox
From: Dengdui Huang
The mailbox messages of the PF and VF are processed in
the same function. The PF and VF call the same function
to process the messages. This code is excessive coupling
and isn't good for maintenance. Therefore, this patch
separates the interfaces that handle PF mailbox message
From: Dengdui Huang
The data region in PF to VF mbx message command is used
to communicate with VF driver. And this data region exists
as an array. As a result, some complicated feature commands,
like mailbox response, link change event, close promisc mode,
reset request and update pvid state, ha
From: Dengdui Huang
The data region in VF to PF mbx message command is
used to communicate with PF driver. And this data
region exists as an array. As a result, some complicated
feature commands, like setting promisc mode, map/unmap
ring vector and setting VLAN id, have to use magic number
to set
This patchset refactors mailbox codes.
--
v3:
1. fix the checkpatch warning on __atomic_xxx.
--
Dengdui Huang (4):
net/hns3: refactor VF mailbox message struct
net/hns3: refactor PF mailbox message struct
net/hns3: refactor send mailbox function
net/hns3: refactor handle mailbox function
On 12/5/2023 6:33 PM, Stephen Hemminger wrote:
> On Tue, 5 Dec 2023 04:05:24 +
> Joyce Kong wrote:
>
>> There will be a segfault when tx burst size is larger than
>> 256. This is because eth_memif_tx uses an index i which is
>> uint8_t to count transmitted nb_pkts. Extend i to uint16_t,
>> t
All drivers should be using dynamic logtype.
This should have been caught during initial driver review.
Since this driver requires non-standard external library this
patch can not be tested by me.
Fixes: 6435f9a0ac22 ("net/nfb: add new netcope driver")
Signed-off-by: Stephen Hemminger
---
drive
From: Pavan Nikhilesh
Optimize Rx routine to use AVX2 instructions when underlying
architecture supports it.
Signed-off-by: Pavan Nikhilesh
---
drivers/net/octeon_ep/cnxk_ep_rx_avx.c | 123 +
drivers/net/octeon_ep/meson.build | 12 +++
drivers/net/octeon_ep/otx_ep
From: Pavan Nikhilesh
Optimize Rx routine to use SSE instructions.
Signed-off-by: Pavan Nikhilesh
---
drivers/net/octeon_ep/cnxk_ep_rx.c | 159 +--
drivers/net/octeon_ep/cnxk_ep_rx.h | 167 +
drivers/net/octeon_ep/cnxk_ep_rx_sse.c | 130 +
From: Pavan Nikhilesh
Preset rearm data to avoid writing multiple fields in fastpath,
Increase maximum outstanding Tx instructions from 128 to 256.
Signed-off-by: Pavan Nikhilesh
---
v3 Chnages:
- Add more comments to the code.
- Re-enable 32b build to prevent ABI break.
v2 Changes:
- Skip comp
[AMD Official Use Only - General]
From: Bruce Richardson
Sent: 06 December 2023 21:20
To: Varghese, Vipin
Cc: Dumitrescu, Cristian ; dev@dpdk.org
; Yigit, Ferruh ; Parikh, Neerav
Subject: Re: [PATCH] cfgfile: increase value length
Caution: This message originated from an External Source. Use
> -Original Message-
> From: Bruce Richardson
> Sent: Wednesday, December 6, 2023 7:00 PM
> To: Sunil Kumar Kori
> Cc: dev@dpdk.org; david.march...@redhat.com
> Subject: Re: [EXT] [PATCH 1/3] buildtools/dpdk-cmdline-gen: support
> optional parameters
>
> On Wed, Dec 06, 2023 at 07:23:51A
On Wed, Dec 06, 2023 at 03:22:41PM +, Varghese, Vipin wrote:
>[AMD Official Use Only - General]
>
>Thanks Bruce & Cristian for the comments.
>An increase seems ok to me, but is an 8x increase really necessary? If
>lines in the config files are over 1k in size, then it implies t
[AMD Official Use Only - General]
Thanks Bruce & Cristian for the comments.
An increase seems ok to me, but is an 8x increase really necessary? If
lines in the config files are over 1k in size, then it implies that some
other mechanism would surely be better for configuration.
Can we make do with
On 05/12/2023 18:30, Stephen Hemminger wrote:
On Mon, 4 Dec 2023 05:31:01 -0500
Maryam Tahhan wrote:
With the original 'use_cni' implementation, (using a
hardcoded socket rather than a configurable one),
if a single pod is requesting multiple net devices
and these devices are from different p
On Mon, Nov 27, 2023 at 11:42 PM Tomasz Duszynski
wrote:
>
> Dynamically allocated log type is a standard approach among all drivers.
> Switch to it.
>
> Signed-off-by: Tomasz Duszynski
> ---
> v2:
> - cleanup remaining plt_ log calls
Applied to dpdk-next-net-mrvl/for-main. Thanks
> -Original Message-
> From: Richardson, Bruce
> Sent: Wednesday, December 6, 2023 1:22 PM
> To: Vipin Varghese
> Cc: dev@dpdk.org; Dumitrescu, Cristian ;
> Ferruh Yigit
> Subject: Re: [PATCH] cfgfile: increase value length
>
> On Wed, Dec 06, 2023 at 04:59:52PM +0530, Vipin Varghese
On Wed, Dec 06, 2023 at 07:23:51AM +, Sunil Kumar Kori wrote:
> > -Original Message-
> > From: Bruce Richardson
> > Sent: Tuesday, December 5, 2023 8:21 PM
> > To: dev@dpdk.org
> > Cc: Sunil Kumar Kori ; david.march...@redhat.com;
> > Bruce Richardson
> > Subject: [EXT] [PATCH 1/3] bu
On Wed, Dec 06, 2023 at 04:59:52PM +0530, Vipin Varghese wrote:
> The default value for CFG_VALUE_LEN is set to 256 characters.
> This limits the parsing for longer strings in configuration file.
> Setting the default to 2048 characters increases the value array
> size in `struct rte_cfgfile_entry`
On 12/5/2023 1:16 PM, Loftus, Ciara wrote:
>>
>> In xdp_umem_configure() allocated memzone for the 'umem', we should
>> free it when xsk_umem__create() call fails, otherwise it will lead to
>> memory zone leak. To fix it move 'umem->mz = mz;' assignment after
>> 'mz == NULL' check.
>>
>> Fixes: f1d
On Sat, Nov 25, 2023 at 10:52 PM wrote:
>
> From: Pavan Nikhilesh
>
> Optimize Rx routine to use SSE instructions.
>
> Signed-off-by: Pavan Nikhilesh
> ---
> diff --git a/drivers/net/octeon_ep/cnxk_ep_rx_sse.c
> b/drivers/net/octeon_ep/cnxk_ep_rx_sse.c
> new file mode 100644
> index 00
On Mon, Nov 27, 2023 at 7:32 AM wrote:
>
> From: Pavan Nikhilesh
>
> Preset rearm data to avoid writing multiple fields in fastpath,
> Increase maximum outstanding Tx instructions from 128 to 256.
>
> Signed-off-by: Pavan Nikhilesh
> ---
> v2 Changes:
> - Skip compiling for 32b x86 targets.
A
Sorry, forgot the subject, add it.
On Wed, Dec 6, 2023 at 7:30 PM Nini Gu wrote:
> Hello,
>
> I met the following issue when running dpdk-testpmd under the OpenShift
> Virtualization environment recently. We did the test on 2 VFs, each passed
> through to a VM, and they come from the same PF(XXV
The default value for CFG_VALUE_LEN is set to 256 characters.
This limits the parsing for longer strings in configuration file.
Setting the default to 2048 characters increases the value array
size in `struct rte_cfgfile_entry`.
Files using cfgfile library are
1. drivers/net/mvpp2/
2. app/test-dma
Triggered by another discussion, I have identified a potential bug related to
this patch.
> From: dev [mailto:dev-boun...@dpdk.org] On Behalf Of Viacheslav
> Ovsiienko
> Sent: Monday, 20 January 2020 20.16
>
> Update detach routine to check the mbuf pool type.
> Introduce the special internal ve
> > > >
> > > > NAK.
> > > >
> > > > This patch is not race safe.
> > >
> > > +1, It is a bad idea.
> >
> > The patch does raise a couple of issues that could be addressed by
> > rearranging. There is duplicate code, and there are no comments
> > to explain the rationale.
> >
> > Maybe something
On Tue, Dec 05, 2023 at 10:50:32AM -0800, Stephen Hemminger wrote:
> On Mon, 4 Dec 2023 11:07:08 +
> Konstantin Ananyev wrote:
>
> > > > 2.25.1
> > >
> > > NAK.
> > >
> > > This patch is not race safe.
> >
> > +1, It is a bad idea.
>
> The patch does raise a couple of issues that cou
On Tue, Dec 5, 2023 at 3:13 AM Stephen Hemminger
wrote:
>
> This patchset removes most of the uses of static LOGTYPE's in DPDK
> libraries. It starts with the easy one and goes on to the more complex ones.
>
> There are several options on how to treat the old static types:
> leave them there, mark
50 matches
Mail list logo
