Vu Pham [Mon, 13 Apr 2020 21:17:48 +0000 (14:17 -0700)]
common/mlx5: refactor memory management
Refactor common memory btree and cache management to common driver.
Replace some input parameters of MR APIs to more common data structure
like PD, port_id, share_cache,... so that multiple PMD drivers can
use those MR APIs.
Modify mlx5 net pmd driver to use MR management APIs from common driver.
Signed-off-by: Vu Pham <vuhuong@mellanox.com> Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
Vu Pham [Mon, 13 Apr 2020 21:17:47 +0000 (14:17 -0700)]
common/mlx5: refactor IPC handling from net driver
Refactor common multi-process handling codes from net PMD to common
driver. Using tuple mp_id{name, port_id} as standard input parameter
for all multi-process IPC APIs instead of using rte_eth_dev.
Modify net PMD to use multi-process APIs from common driver.
Signed-off-by: Vu Pham <vuhuong@mellanox.com> Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
Currently, when translate jump action, the table reference will be
increased all the time. But when release the jump action, the table
resource reference will only be decreased when jump action is released.
It means for jump action which was referenced more than one time, the
increased table reference only decrease one time when jump action is
released.
Add table release when the jump action was not new created.
Currently, the meter suffix table is created and saved in the mlx5
shared struct. It causes the suffix table will never be released
even without any meter rules.
Move the suffix table to meter domain struct to help the suffix table
be released when all the meter rules are destroyed.
The multi-stride operations now allow to reduce a stride size
while supporting Jumbo frames. That means that it is possible
to have mbufs configured with a size smaller than the whole
packet received. It is not an issue during normal MPRQ operations
since we attach external buffers instead of copying the data
into the mbuf itself. But it is not the case in "emergency mode"
when we have to copy every packet because of no more external
mbufs are available. Assemble a multi-segment packet to overcome
this issue in case scatter mode is enabled, drop a packet if not.
MPRQ feature should be updated to allow a packet to be received
into multiple strides in order to support the MTU exceeding 8KB.
Special care is needed to prevent the headroom corruption in the
multi-stride mode since the headroom space is borrowed by the PMD
from the tail of the preceding stride. Copy the whole packet into
a separate mbuf in this case or just the overlapping data if the
Rx scattering is supported by an application.
net/mlx5: add device parameter for MPRQ stride size
Define a device parameter to configure log 2 of a stride size for MPRQ
- mprq_log_stride_size. User is able to specify a stride size in a range
allowed by an underlying hardware. The default stride size is defined as
2048 bytes to encompass most commonly used packet sizes in the Internet
(MTU 1518 and less) and will be used in case a maximum configured packet
size cannot fit into the largest possible stride size. Otherwise a
stride size is set to a large enough value to encompass a whole packet.
net/ice/base: force switch to use different recipe
When we use profile rule as switch rule to download, if
we download 2 different rules one by one, there will be
rejection from function ice_aq_sw_rules(), for example:
"flow create 0 priority 0 ingress pattern eth / ipv6 / ah
/ end actions queue index 3 / end"
"flow create 0 priority 0 ingress pattern eth / ipv6 / esp
/ end actions queue index 2 / end"
That is because the 2 rules has the same s_rule input set
except action queue index, so it will be rejected by
hardware. So we have to use different recipes for them.
Also, we need to add recipe_id to keep record of recipe
index, which will be used in rule remove, if not, there
will be error when search recipe in function
ice_rem_adv_rule() if we create 2 or more profile rule.
For example:
"flow create 0 priority 0 ingress pattern eth / ipv4 / udp
/ pfcp s_field is 1 / end actions queue index 4 / end"
"flow create 0 priority 0 ingress pattern eth / ipv4 / udp
/ pfcp s_field is 0 / end actions queue index 5 / end"
then,
"flow flush 0"
you will find only the first rule will be delete,
because ice_find_recp() will always return recipe
id of the first rule.
net/ice/base: add mask check to find switch recipe
In order to find accurate recipe for switch filter, we
need to add mask as an element when searching for recipe.
If we create different rules with the same input set, but
using different masks, then proper recipes should use
those different mask.
When we add some long switch rule, we need check the
number of final recipe number, if it is large than
ICE_MAX_CHAIN_RECIPE, we should refuse this rule.
For example:
"flow create 0 ingress pattern eth / ipv6
src is CDCD:910A:2222:5498:8475:1111:3900:1536
dst is CDCD:910A:2222:5498:8475:1111:3900:2022
tc is 3 / udp dst is 45 / end actions queue index 2 / end"
This rule will consume 6 recipe, if it is not refused, it
will cause the following code over write of lkup_indx and mask.
Gavin Hu [Mon, 13 Apr 2020 16:40:25 +0000 (00:40 +0800)]
net/i40e: restrict pointer aliasing for NEON
Restrict pointer aliasing to optimize the code generated.
The patch showed ~3% performance uplift on Arm N1SDP platform, and no
degradation on ThunderX2. The tet case is RFC2544 zero-loss L2
forwarding running testpmd.
Disable CQ_DISABLED error interrupt in NIX_LF_ERR_INT
to fix spurious interrupts in event dev mode. Also skip
configuring RSS when RQ count is '0' because
RSS table initialization is done incorrectly due to
divide-by-zero error and it is leading to RQ_OOR error
in NIX_LF_ERR_INT.
Fixes: 83ce2880e22e ("net/octeontx2: support RSS") Cc: stable@dpdk.org Signed-off-by: Nithin Dabilpuram <ndabilpuram@marvell.com> Acked-by: Jerin Jacob <jerinj@marvell.com>
In case of bnx2xvf pmd, tx packets can support vland id in 2 ways:
1. Setting the mbuf ol_flags=PKT_TX_VLAN_PKT and passing the
vlanid in mbuf->vlan_tci.
2. The tx packet itself has the vlan id included in the packet.
The first case is working as expected but the second case where
the vlan id is included in thetx packets itself was found not
working as expected. To handle that we need to properly set the
start_bd bitfield and the vlan_or_ethertype instead of setting it
to just the ethertype in case of VF.
Add support the set_mc_addr_list device operation in the bnx2xvf PMD.
The configured addresses are stored in the device private area, so
they can be flushed before adding new ones.
Without this v6 multicast packets were properly forwarded to the
Guest VF.
fgets(3)/fread(3)/fscanf(3) etc. use mmap(2)/munmap(2) which leads
to TLB shutdown interrupts to all DPDK app cores including RX cores.
This can cause packet drops. Use read(2)/write(2) instead.
When creating a flow, usually the creating routine is called in
serial. No parallel execution is supported right now. The same
function will be called only once for a single flow creation.
But there is a special case that the creating routine will be called
nested. If the xmeta feature is enabled and there is FLAG / MARK in
the actions list, some metadata reg copy flow needs to be created
before the original flow is applied to the hardware.
In the flow non-cached mode, resources only for flow creation will
not be saved anymore. The memory space is pre-allocated and reused
for each flow. A global index for each device is used to indicate
the memory address of the resources. If the function is called in a
nested mode, then the index will be reset and make everything get
corrupted.
To solve this, a nested index is introduced to save the position for
the original flow creation. Currently, only one level nested call
of the flow creating routine is supported.
Fixes: e7bfa3596a0a ("net/mlx5: separate the flow handle resource") Signed-off-by: Bing Zhao <bingz@mellanox.com> Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
Thomas Monjalon [Wed, 8 Apr 2020 00:09:00 +0000 (02:09 +0200)]
net/mlx4: fix build with -fno-common
The variable storages of the same name are merged together
if compiled with -fcommon. This is the default.
This default behaviour allows to declare a variable in a header file and
share the variable in every .o binaries thanks to merge at link-time.
In the case of dlopen linking of the glue library, the pointer mlx4_glue
is referencing the glue functions struct and is set after calling
dlopen.
If compiling with -fno-common (default in GCC 10), the variables must be
declared as extern to avoid multiple re-definitions.
In case the glue layer is split in glue library, the variable mlx4_glue
needs to have its own storage for the rest of the PMD.
Cc: stable@dpdk.org Signed-off-by: Thomas Monjalon <thomas@monjalon.net> Acked-by: Matan Azrad <matan@mellanox.com>
Thomas Monjalon [Wed, 8 Apr 2020 00:08:59 +0000 (02:08 +0200)]
common/mlx5: fix build with -fno-common
The variable storages of the same name are merged together
if compiled with -fcommon. This is the default.
This default behaviour allows to declare a variable in a header file and
share the variable in every .o binaries thanks to merge at link-time.
In the case of dlopen linking of the glue library, the pointer mlx5_glue
is referencing the glue functions struct and is set after calling
dlopen.
If compiling with -fno-common (default in GCC 10), the variable must be
declared as extern to avoid multiple re-definitions.
In case the glue layer is split in glue library, the variable mlx5_glue
needs to have its own storage for the rest of the PMD.
Cc: stable@dpdk.org Signed-off-by: Thomas Monjalon <thomas@monjalon.net> Acked-by: Matan Azrad <matan@mellanox.com>
Thomas Monjalon [Wed, 8 Apr 2020 00:08:58 +0000 (02:08 +0200)]
common/mlx5: split glue initialization
The function mlx5_glue_init was doing three things:
- initialize logs
- load glue library if in dlopen mode
- initialize glue layer
They are split in three functions for clarity.
The config option RTE_IBVERBS_LINK_DLOPEN is not used anymore
outside of make and meson files. It is replaced with MLX5_GLUE,
which is defined in the same condition and is already used with dlopen.
Signed-off-by: Thomas Monjalon <thomas@monjalon.net> Acked-by: Matan Azrad <matan@mellanox.com>
common/iavf: support flow director in virtual channel
Adds new ops and structures to support VF to add/delete/validate/
query flow director.
ADD and VALIDATE FDIR share one ops: VIRTCHNL_OP_ADD_FDIR_FILTER.
VF sends this request to PF by filling out the related field in
virtchnl_fdir_add. If the rule is created successfully, PF
will return flow id and program status to VF. If the rule is
validated successfully, the PF will only return program status
to VF.
DELETE FDIR uses ops: VIRTCHNL_OP_DEL_FDIR_FILTER.
VF sends this request to PF by filling out the related field in
virtchnl_fdir_del. If the rule is deleted successfully, PF
will return program status to VF.
Query FDIR uses ops: VIRTCHNL_OP_QUERY_FDIR_FILTER.
VF sends this request to PF by filling out the related field in
virtchnl_fdir_query. If the request is successfully done by PF,
PF will return program status and query info to VF.
Signed-off-by: Simei Su <simei.su@intel.com> Signed-off-by: Paul M Stillwell Jr <paul.m.stillwell.jr@intel.com> Signed-off-by: Qi Zhang <qi.z.zhang@intel.com> Acked-by: Beilei Xing <beilei.xing@intel.com>
common/iavf: support advanced RSS input set change
Add new ops and a new VIRTCHNL_VF_OFFLOAD_ADV_RSS_PF flag to support
VF to add or del a specific rss configuration by virtchnl.
Signed-off-by: Jeff Guo <jia.guo@intel.com> Signed-off-by: Paul M Stillwell Jr <paul.m.stillwell.jr@intel.com> Signed-off-by: Qi Zhang <qi.z.zhang@intel.com> Acked-by: Beilei Xing <beilei.xing@intel.com>
To support advanced AVF's FDIR and RSS feature, we need to figure out
what kind of data structure should be passed from VF to PF to describe
an FDIR rule or RSS config rule. The common part of the requirement is
we need a data structure to represent the input set selection of a rule's
hash key.
An input set selection is a group of fields be selected from one or more
network protocol layers that could be identified as a specific flow.
For example, select dst IP address from an IPv4 header combined with
dst port from the TCP header as the input set for an IPv4/TCP flow.
The patch adds a new data structure virtchnl_proto_hdrs to abstract
a network protocol headers group which is composed of layers of network
protocol header(virtchnl_proto_hdr).
A protocol header contains a 32 bits mask (field_selector) to describe
which fields are selected as input sets, as well as a header type
(enum virtchnl_proto_hdr_type). Each bit is mapped to a field in
enum virtchnl_proto_hdr_field guided by its header type.
+------------+-----------+------------------------------+
| | Proto Hdr | Header Type A |
| | +------------------------------+
| | | BIT 31 | ... | BIT 1 | BIT 0 |
| |-----------+------------------------------+
|Proto Hdrs | Proto Hdr | Header Type B |
| | +------------------------------+
| | | BIT 31 | ... | BIT 1 | BIT 0 |
| |-----------+------------------------------+
| | Proto Hdr | Header Type C |
| | +------------------------------+
| | | BIT 31 | ... | BIT 1 | BIT 0 |
| |-----------+------------------------------+
| | .... |
+-------------------------------------------------------+
All fields in enum virtchnl_proto_hdr_fields are grouped with header type
and the value of the first field of a header type is always 32 aligned.
So we have:
proto_hdr_type = proto_hdr_field / 32
bit offset = proto_hdr_field % 32
To simply the protocol header's operations, couple help macros are added.
For example, to select src IP and dst port as input set for an IPv4/UDP
flow.
A protocol header also contains a byte array, this field should only
be used by an FDIR rule and should be ignored by RSS. For an FDIR rule,
the byte array is used to store the protocol header of a training
package. The byte array must be network order.
Signed-off-by: Jeff Guo <jia.guo@intel.com> Signed-off-by: Paul M Stillwell Jr <paul.m.stillwell.jr@intel.com> Signed-off-by: Qi Zhang <qi.z.zhang@intel.com> Acked-by: Beilei Xing <beilei.xing@intel.com>
Add new VIRTCHNL_VF_OFFLOAD_RX_FLEX_DESC flag, opcode
VIRTCHNL_OP_GET_SUPPORTED_RXDIDS and add member rxdid
in struct virtchnl_rxq_info to support AVF Flex RXD
extension.
Signed-off-by: Leyi Rong <leyi.rong@intel.com> Signed-off-by: Paul M Stillwell Jr <paul.m.stillwell.jr@intel.com> Signed-off-by: Qi Zhang <qi.z.zhang@intel.com> Acked-by: Beilei Xing <beilei.xing@intel.com>
common/iavf: add DDP package query in virtual channel
Add VIRTCHNL_OP_DCF_GET_PKG_INFO to query DDP package identification.
Signed-off-by: Leyi Rong <leyi.rong@intel.com> Signed-off-by: Paul M Stillwell Jr <paul.m.stillwell.jr@intel.com> Signed-off-by: Qi Zhang <qi.z.zhang@intel.com> Acked-by: Beilei Xing <beilei.xing@intel.com>
Add IAVF_RX_PTYPE_PARSER_ABORTED definition, so iavf driver will know
opcode for parser aborted packets.
Without this definition driver would have to rely on magic numbers.
Signed-off-by: Przemyslaw Patynowski <przemyslawx.patynowski@intel.com> Signed-off-by: Paul M Stillwell Jr <paul.m.stillwell.jr@intel.com> Signed-off-by: Qi Zhang <qi.z.zhang@intel.com> Acked-by: Beilei Xing <beilei.xing@intel.com>
net/hns3: fix VLAN filter when setting promisucous mode
Currently, when upper level application call the API function named
rte_eth_dev_set_vlan_offload to configure the hardware vlan filter
offload and call the rte_eth_promiscuous_enable API to enable
promiscuous mode based on hns3 PF device, driver can't receive the
packets with a vlan tag which has not been added by calling the API
function named rte_eth_dev_vlan_filter.
This patch fixes it by disabling the vlan filter when setting the
promiscuous mode and enabling the vlan filter again after the
promiscuous mode are disabled.
Fixes: 19a3ca4c99cf ("net/hns3: add start/stop and configure operations") Cc: stable@dpdk.org Signed-off-by: Chengchang Tang <tangchengchang@huawei.com> Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
net/hns3: fix default VLAN filter configuration for PF
Currently, By default VLAN filter is enabled during initialization and
couldn't be turned off based on hns3 PF device. If upper applications
don't call rte_eth_dev_vlan_filter API function to set vlan based on
hns3 PF device, hns3 PF PMD driver will can't receive the packets with
vlan tag. It will leads to some compatibility issues, the behaviors of
using hns3 network engine and other NICs are different.
This patch disables the VLAN filter during initialization and allows the
upper level applications to enable or disable the VLAN filter.
Fixes: 411d23b9eafb ("net/hns3: support VLAN") Cc: stable@dpdk.org Signed-off-by: Chengchang Tang <tangchengchang@huawei.com> Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Lijun Ou [Fri, 10 Apr 2020 11:09:28 +0000 (19:09 +0800)]
net/hns3: fix RSS key length
When upper application calls the rte_eth_dev_rss_hash_conf_get API
function to get the RSS key parameters, the function should return the
RSS key length supported by the device. Otherwise, an error will occur
when the upper application needs to use the RSS key length supported
by this specified hardware for judgment and configure the specified
key into hardware.
For example, in the following scenario:
When users want to use their own RSS key, but the length of the key is
bigger than the one of the supported by hardware.
As a result, users need to get the RSS key length supported by hardware
according to the above API firstly, and then compare the actual obtained
RSS key length with the length of their own RSS key.
If the driver does not return the actual value, error may occur when
user calls the rte_eth_dev_rss_hash_update API function to configure
their own key into hardware.
Besides, this patch fixes the problem of stepping on memory when the RSS
key array configured by users are less than the RSS key length supported
by the driver at the same time.
Fixes: c37ca66f2b27 ("net/hns3: support RSS") Cc: stable@dpdk.org Signed-off-by: Lijun Ou <oulijun@huawei.com> Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com> Signed-off-by: Chengwen Feng <fengchengwen@huawei.com> Signed-off-by: Chengchang Tang <tangchengchang@huawei.com>
Lijun Ou [Fri, 10 Apr 2020 11:09:27 +0000 (19:09 +0800)]
net/hns3: add RSS hash offload to capabilities
Currently, when upper application calls rte_eth_dev_info_get API
function to query the Rx offload capability based on hns3 network
engine, RSS hash offload capacity is missing.
This patch fixes it by adding the related capacity in the
'.dev_infos_get' ops implementation function named hns3_dev_infos_get
and hns3vf_dev_infos_get for hns3 PF/VF PMD driver.
Fixes: c37ca66f2b27 ("net/hns3: support RSS") Cc: stable@dpdk.org Signed-off-by: Lijun Ou <oulijun@huawei.com> Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
This patch fixes that the flow director rules are not cleared during
initialization, which lead to remaining flow director rules after upper
application (such as testpmd) restarted.
Fixes: fcba820d9b9e ("net/hns3: support flow director") Cc: stable@dpdk.org Signed-off-by: Chengwen Feng <fengchengwen@huawei.com> Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Currently, Rx interrupt cannot work normally after reset (such as FLR,
global reset and IMP reset), when running l3fwd-power application based
on hns3 network engine.
The root cause is that the hardware configuration about Rx interrupt
does not recover after reset.
This patch fixes it with the following modification.
1. The internal static function named hns3(vf)_init_ring_with_vector is
moved from hns3_init_pf to hns3(vf)_init_hardware because
hns3(vf)_init_hardware is called both in the initialization and the
RESET_STAGE_DEV_INIT stage of the reset process.
2. The internal static function named hns3(vf)_restore_rx_interrupt is
added in hns3(vf)_restore_conf, it is used to recover hardware
configuration about interrupt vectors of rx queues in the
RESET_STAGE_DEV_INIT stage of the reset process.
3. The internal static function named hns3_dev_all_rx_queue_intr_enable
and hns3_enable_all_queues are added in hns3(vf)_dev_start(which
called in the initialization, so after calling the rte_eth_dev_start
API successfully, the driver is ready to work.
4. The function named hns3_dev_all_rx_queue_intr_enable and
hns3_enable_all_queues are also added in hns3(vf)_start_service(which
called in the RESET_STAGE_DEV_INIT stage of the reset process), so
after start_service, the driver is ready to work.
Note:
1. Because FLR will clear queue's interrupt enable bit hardware
configuration, so we add calling hns3_dev_all_rx_queue_intr_enable to
enable interrupt before enabling queues.
2. After finished the initialization, we can enable queues to work by
calling the internal function named hns3_enable_all_queues.
Currently, when upper application calls the rte_eth_dev_mac_addr_add API
function to add a MC mac address based on hns3 PF/VF device, it will
fail.
In hns3 network engine adding UC and MC mac address with different
commands with firmware. We need to determine whether the input address
is a UC or a MC address to call different commands in the
'.mac_addr_add' and '.mac_addr_remove' ops implementation functions in
hns3 PF and VF driver as below:
hns3_add_mac_addr
hns3vf_add_uc_mac_addr
hns3_remove_mac_addr
hns3vf_remove_mac_addr
By the way, it is recommended calling the rte_eth_dev_set_mc_addr_list API
function to set the MC mac address, because using the
rte_eth_dev_mac_addr_add API function to set MC mac address may affect the
specifications of UC mac addresses.
Fixes: 7d7f9f80bbfb ("net/hns3: support MAC address related operations") Cc: stable@dpdk.org Signed-off-by: Chengchang Tang <tangchengchang@huawei.com> Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com> Signed-off-by: Lijun Ou <oulijun@huawei.com> Signed-off-by: Chengwen Feng <fengchengwen@huawei.com>
Itsuro Oda [Wed, 11 Mar 2020 23:19:18 +0000 (08:19 +0900)]
vhost: make IOTLB cache name unique among processes
Currently, iotlb cache name is comprised of vid and virtqueue
index. For example, "iotlb_cache_0_0". Because vid is assigned
per process, iotlb cache name is not unique among multi processes.
For example a secondary process uses a vhost
(ex. eth_vhost0,iface=/tmp/sock0) and another secondary process
uses a vhost (ex. eth_vhost1,iface=/tmp/sock1), iotlb cache
name of both vhost ("iotlb_cache_0_0") are same and as a result
iotlb cache is broken.
This patch makes iotlb cache name unique among milti processes
by adding process id to the iotlb cache name.
The prefix of the name is shortened to "iotlb_" since the maximum
length of pool name is 25 bytes (RTE_MEMPOOL_NAMESIZE is 26).
Note that it is just 25 characters in maximum at the moment.
Here,
* pid_t == int: max 10 digits.
* vid < MAX_VHOST_DECICE(1024): max 4 digits.
* vq_index < VHOST_MAX_VRING(256): max 3 digits.
Fixes: d012d1f293f4 ("vhost: add IOTLB helper functions") Cc: stable@dpdk.org Signed-off-by: Itsuro Oda <oda@valinux.co.jp> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Xiaolong Ye [Sat, 7 Mar 2020 13:22:35 +0000 (21:22 +0800)]
vhost: remove unused variable
VHOST_FEATURES has been removed in previous refactoring.
Fixes: 0917f9d1f059 ("vhost: use new APIs to handle features") Cc: stable@dpdk.org Signed-off-by: Xiaolong Ye <xiaolong.ye@intel.com> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Xiaolong Ye [Sat, 7 Mar 2020 13:22:34 +0000 (21:22 +0800)]
net/virtio: fix outdated comment
Fix comment that is no more correct as the code evolved.
Fixes: 9470427c88e1 ("net/virtio: do not store PCI device pointer at shared memory") Cc: stable@dpdk.org Signed-off-by: Xiaolong Ye <xiaolong.ye@intel.com> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Itsuro Oda [Thu, 5 Mar 2020 02:54:50 +0000 (11:54 +0900)]
net/vhost: fix potential memory leak on close
If a vhost device is closed before eth_dev_configure is done
to the device, internal resources allocated to the device
would not be freed. This patch fixes it.
Fixes: 3d01b759d267 ("net/vhost: delay driver setup") Cc: stable@dpdk.org Signed-off-by: Itsuro Oda <oda@valinux.co.jp> Reviewed-by: Xiaolong Ye <xiaolong.ye@intel.com>
Xiaolong Ye [Wed, 26 Feb 2020 13:45:34 +0000 (21:45 +0800)]
net/vhost: enable promiscuous and multicast by default
With this patch, the promiscuous and multicast fields are initialized as
enabled for vhost PMD by default, this allows the devices to be used
when running applications that attempt to enable promiscuous or
multicast mode.
Similar things have done for other virtual PMDs by commit f165210321c4
("drivers/net: enable promiscuous and multicast by default")
Signed-off-by: Xiaolong Ye <xiaolong.ye@intel.com> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
net/vhost: add options for linear and external buffer
Added vHost PMD arguments 'linear-buffer' and 'ext-buffer'
to configure 'RTE_VHOST_USER_LINEARBUF_SUPPORT' and
'RTE_VHOST_USER_EXTBUF_SUPPORT' flags in the vhost library
Signed-off-by: Sivaprasad Tummala <sivaprasad.tummala@intel.com> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Marvin Liu [Mon, 24 Feb 2020 15:14:19 +0000 (23:14 +0800)]
vhost: fix packed ring zero-copy
Available buffer ID should be stored in the zmbuf in the packed-ring
dequeue path. There's no guarantee that local queue avail index is
equal to buffer ID.
Fixes: d1eafb532268 ("vhost: add packed ring zcopy batch and single dequeue") Cc: stable@dpdk.org Signed-off-by: Marvin Liu <yong.liu@intel.com> Reported-by: Yinan Wang <yinan.wang@intel.com> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Fan Zhang [Wed, 29 Jan 2020 10:19:37 +0000 (10:19 +0000)]
vhost/crypto: add missing user protocol flag
This patch fixes the vhost crypto missed
"VHOST_USER_PROTOCOL_F_CONFIG" flag problem during initialization.
Newer Qemu version requires this feature enabled.
Fixes: 939066d96563 ("vhost/crypto: add public function implementation") Cc: stable@dpdk.org Signed-off-by: Fan Zhang <roy.fan.zhang@intel.com> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Xiaoyun Wang [Fri, 10 Apr 2020 09:21:47 +0000 (17:21 +0800)]
net/hinic: adds Tx queue xstats members
Because some apps may pass illegal parameters, driver increases
checks on illegal parameters and DFX statistics, which includes
sge_len0 and mbuf_null txq xstats member.
Signed-off-by: Xiaoyun Wang <cloud.wangxiaoyun@huawei.com>
Xiaoyun Wang [Fri, 10 Apr 2020 09:21:45 +0000 (17:21 +0800)]
net/hinic/base: fix PF firmware hot-active problem
When FW is hotactive which means updating the FW but not needs
to reboot OS, FW returns HINIC_DEV_BUSY_ACTIVE_FW for pf driver
because firmware is being reinitialized, at which point the cmdq
initialization that relies on the fw channel will fail, so driver
should reinit the cmdq when port start.
Fixes: 0194313b2df6 ("net/hinic/base: fix port start during FW hot update") Cc: stable@dpdk.org Signed-off-by: Xiaoyun Wang <cloud.wangxiaoyun@huawei.com>
Ferruh Yigit [Mon, 2 Mar 2020 17:36:40 +0000 (17:36 +0000)]
net/null: fix secondary burst function selection
Secondary process uses the primary process device and while setting the
Rx/Tx functions it uses the device arguments from the secondary process
instead of the primary ones.
This may cause primary and secondary process use different Rx/Tx
functions unintentionally.
Jiaqi Min [Wed, 8 Apr 2020 10:05:22 +0000 (10:05 +0000)]
net/i40e/base: add constants for PTP pins
Introduce constants for handling PTP pins used for external
clock source.
Signed-off-by: Piotr Kwapulinski <piotr.kwapulinski@intel.com> Signed-off-by: Jiaqi Min <jiaqix.min@intel.com> Acked-by: Piotr Kwapulinski <piotr.kwapulinski@intel.com> Acked-by: Xiaolong Ye <xiaolong.ye@intel.com> Acked-by: Beilei Xing <beilei.xing@intel.com>
Jiaqi Min [Wed, 8 Apr 2020 10:05:21 +0000 (10:05 +0000)]
net/i40e/base: introduce device ID for V710-TL 5G
This change is adding new device ID and handling it in the same way as
X710-T*L head of family. A new device ID is for new V710-T*L adapter
supporting speeds up to 5G.
Signed-off-by: Zalfresso-Jundzillo <marekx.zalfresso-jundzillo@intel.com> Signed-off-by: Jiaqi Min <jiaqix.min@intel.com> Acked-by: Xiaolong Ye <xiaolong.ye@intel.com> Acked-by: Beilei Xing <beilei.xing@intel.com>
Qiming Yang [Fri, 3 Apr 2020 05:42:41 +0000 (13:42 +0800)]
net/iavf: support generic flow API
This patch added iavf_flow_create, iavf_flow_destroy,
iavf_flow_flush and iavf_flow_validate support,
these are used to handle all the generic filters.
This patch supported basic L2, L3, L4 and GTPU patterns.
Signed-off-by: Qiming Yang <qiming.yang@intel.com> Acked-by: Qi Zhang <qi.z.zhang@intel.com>
Yunjian Wang [Thu, 9 Apr 2020 01:59:00 +0000 (09:59 +0800)]
net/pfe: fix double free of MAC address
The 'mac_addrs' freeing has been moved to rte_eth_dev_release_port(),
so freeing 'mac_addrs' like this in pfe_eth_exit() is unnecessary and
will cause double free.
Currently, the counter struct saves both the members used by batch
counters and none batch counters. The members which are only used
by none batch counters cost 16 bytes extra memory for batch counters.
As normally there will be limited none batch counters, mix the none
batch counter and batch counter members becomes quite expensive for
batch counter. If 1 million batch counters are created, it means 16 MB
memory which will not be used by the batch counters are allocated.
Split the mlx5_flow_counter struct for batch and none batch counters
helps save the memory.
Currently, DV and verbs counters are both changed to indexed. It means
while creating the flow with counter, flow can save the indexed value to
address the counter.
Save the 4 bytes indexed value in the rte_flow instead of 8 bytes
pointer helps to save memory with millions of flows.
This part of the counter optimize change the DV counter to indexed as
what have already done in verbs. In this case, all the mlx5 flow counter
can be addressed by index.
The counter index is composed of pool index and the counter offset in
the pool counter array. The batch and none batch counter dcs ID offset
0x800000 is used to avoid the mix up for the index. As batch counter dcs
ID starts from 0x800000 and none batch counter dcs starts from 0, the
0x800000 offset is added to the batch counter index to indicate the
index of batch counter.
The counter pointer in rte_flow struct will be aligned to index instead
of pointer. It will save 4 bytes memory for every rte_flow. With
millions of rte_flow, it will save MBytes memory.
This commit is a part for the DV counter optimization.
The batch counter dcs id starts from 0x800000 and none batch counter
starts from 0. As currently, the counter is changed to be indexed by
pool index and the offset of the counter in the pool counters_raw array.
It means now the counter index is same for batch and none batch counter.
Add the 0x800000 batch counter offset to the batch counter index helps
indicate the counter index is from batch or none batch container pool.
Query generation was introduced to avoid counter to be reallocated
before the counter statistics be fully updated. Since the counters
be released between query trigger and query handler may miss the
packets arrived in the trigger and handler gap period. In this case,
user can only allocate the counter while pool query_gen is greater
than the counter query_gen + 1 which indicates a new round of query
finished, the statistic is fully updated.
Split the pool query_gen to start_query_gen and end_query_gen helps
to have a better identify for the counter released in the gap period.
And it helps the counter released before query trigger or after query
handler can be reallocated more efficiently.
As none-batch counter pool allocates only one counter every time, after
the new allocated counter pop out, the pool will be empty and moved to
the end of the container list in the container.
Currently, the new non-batch counter allocation maybe happened with new
counter pool allocated, it means the new counter comes from a new pool.
While new pool is allocated, the container resize and switch happens.
In this case, after the pool becomes empty, it should be added to the
new container pool list as the pool belongs.
Update the container pointer accordingly with pool allocation to avoid
add the pool to the incorrect container.
Michal Krawczyk [Wed, 8 Apr 2020 08:29:21 +0000 (10:29 +0200)]
net/ena: update driver version to v2.1.0
The v2.1.0 is refactoring Tx and Rx paths, including few bug fixes and
is also adding a new features which are going to be available with the
newest hardware.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
Michal Krawczyk [Wed, 8 Apr 2020 08:29:20 +0000 (10:29 +0200)]
doc: add notes on ENA usage on metal instances
As AWS metal instances are supporting IOMMU, the usage of igb_uio or
vfio-pci can lead to a problems (when to use which module), especially
that the vfio-pci isn't supporting SMMU on arm64.
To clear up the problem of using those modules in various setup
conditions (with or without IOMMU) on metal instances, more detailed
explanation was added.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
Michal Krawczyk [Wed, 8 Apr 2020 08:29:19 +0000 (10:29 +0200)]
net/ena: reuse zero length Rx descriptor
Some ENA devices can pass to the driver descriptor with length 0. To
avoid extra allocation, the descriptor can be reused by simply putting
it back to the device.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
Michal Krawczyk [Wed, 8 Apr 2020 08:29:18 +0000 (10:29 +0200)]
net/ena: refactor Tx
The original Tx function was very long and was containing both cleanup
and the sending sections. Because of that it was having a lot of local
variables, big indentation and was hard to read.
This function was split into 2 sections:
* Sending - which is responsible for preparing the mbuf, mapping it
to the device descriptors and finally, sending packet to the HW
* Cleanup - which is releasing packets sent by the HW. Loop which was
releasing packets was reworked a bit, to make intention more visible
and aligned with other parts of the driver.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
Michal Krawczyk [Wed, 8 Apr 2020 08:29:17 +0000 (10:29 +0200)]
net/ena: use macros for ring index operations
To improve code readability, abstraction was added for operating on IO
rings indexes.
Driver was defining local variable for ring mask in each function that
needed to operate on the ring indexes. Now it is being stored in the
ring as this value won't change unless size of the ring will change and
macros for advancing indexes using the mask has been added.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
Michal Krawczyk [Wed, 8 Apr 2020 08:29:16 +0000 (10:29 +0200)]
net/ena: limit refill threshold by fixed value
Divider used for both Tx and Rx cleanup/refill threshold can cause too
big delay in case of the really big rings - for example if the 8k Rx
ring will be used, the refill won't trigger unless 1024 threshold will
be reached. It will also cause driver to try to allocate that much
descriptors.
Limiting it by fixed value - 256 in that case, would limit maximum
time spent in repopulate function.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
Michal Krawczyk [Wed, 8 Apr 2020 08:29:15 +0000 (10:29 +0200)]
net/ena: rework getting number of available descriptors
ena_com API should be preferred for getting number of used/available
descriptors unless extra calculation needs to be performed.
Some helper variables were added for storing values that are later
reused. Moreover, for limiting the value of sent/received packets to
the number of available descriptors, the RTE_MIN is used instead of
if function, which was doing similar thing but was less descriptive.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
Michal Krawczyk [Wed, 8 Apr 2020 08:29:14 +0000 (10:29 +0200)]
net/ena: refactor Rx
* Split main Rx function into multiple ones - the body of the main
was very big and further there were 2 nested loops, which were
making the code hard to read
* Rework how the Rx mbuf chains are being created - Instead of having
while loop which has conditional check if it's first segment, handle
this segment outside the loop and if more fragments are existing,
process them inside.
* Initialize Rx mbuf using simple function - it's the common thing for
the 1st and next segments.
* Create structure for Rx buffer to align it with Tx path, other ENA
drivers and to make the variable name more descriptive - on DPDK, Rx
buffer must hold only mbuf, so initially array of mbufs was used as
the buffers. However, it was misleading, as it was named
"rx_buffer_info". To make it more clear, the structure holding mbuf
pointer was added and now there is possibility to expand it in the
future without reworking the driver.
* Remove redundant variables and conditional checks.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
Michal Krawczyk [Wed, 8 Apr 2020 08:29:13 +0000 (10:29 +0200)]
net/ena: disable meta caching
In the LLQ (Low-latency queue) mode, the device can indicate that meta
data descriptor caching is disabled. In that case the driver should send
valid meta descriptor on every Tx packet.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
Michal Krawczyk [Wed, 8 Apr 2020 08:29:12 +0000 (10:29 +0200)]
net/ena: add Tx drops statistic
ENA device can report in the AENQ handler amount of Tx packets that were
dropped and not sent.
This statistic is showing global value for the device and because
rte_eth_stats is missing field that could indicate this value (it
isn't the Tx error), it is being presented as a extended statistic.
As the current design of extended statistics prevents tx_drops from
being an atomic variable and both tx_drops and rx_drops are only updated
from the AENQ handler, both were set as non-atomic for the alignment.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
Michal Krawczyk [Wed, 8 Apr 2020 08:29:10 +0000 (10:29 +0200)]
net/ena: support large LLQ headers
Default LLQ (Low-latency queue) maximum header size is 96 bytes and can
be too small for some types of packets - like IPv6 packets with multiple
extension. This can be fixed, by using large LLQ headers.
If the device supports larger LLQ headers, the user can activate them by
using device argument 'large_llq_hdr' with value '1'.
If the device isn't supporting this feature, the default value (96B)
will be used.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
Michal Krawczyk [Wed, 8 Apr 2020 08:29:09 +0000 (10:29 +0200)]
net/ena: refactor getting IO queues capabilities
Reading values from the device is about the maximum capabilities of the
device. Because of that, the names of the fields storing those values,
functions and temporary variables, should be more descriptive in order
to improve self documentation of the code.
In connection with this, the way of getting maximum queue size could be
simplified - no hardcoded values are needed, as the device is going to
send it's capabilities anyway.
Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
Michal Krawczyk [Wed, 8 Apr 2020 08:29:08 +0000 (10:29 +0200)]
net/ena: set IO ring size to valid value
IO rings were configured with the maximum allowed size for the Tx/Rx
rings. However, the application could decide to create smaller rings.
This patch is using value stored in the ring instead of the value from
the adapter which is indicating the maximum allowed value.
Fixes: df238f84c0a2 ("net/ena: recreate HW IO rings on start and stop") Cc: stable@dpdk.org Signed-off-by: Michal Krawczyk <mk@semihalf.com> Reviewed-by: Igor Chauskin <igorch@amazon.com> Reviewed-by: Guy Tzalik <gtzalik@amazon.com>
git.droids-corp.org / dpdk.git / log