Bruce Richardson [Wed, 30 Sep 2015 12:12:20 +0000 (13:12 +0100)]
ring: store memzone pointer
Add a new field to the rte_ring structure to store the memzone pointer which
contains the ring. For rings created using rte_ring_create(), the field will
be set automatically.
This new field will allow users of the ring to query the numa node a ring is
allocated on, or to get the physical address of the ring, if so needed.
The rte_ring structure will also maintain ABI compatibility, as the
structure members, after the new one, are set to be cache line aligned,
so leaving a space.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Olivier Matz <olivier.matz@6wind.com> Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Bruce Richardson [Wed, 30 Sep 2015 12:12:19 +0000 (13:12 +0100)]
ring: enhance device setup from rings
The ring ethdev creation function creates an ethdev, but does not
actually set it up for use. Even if it's just a single ring, the user
still needs to create a mempool, call rte_eth_dev_configure, then call
rx and tx setup functions before the ethdev can be used.
This patch changes things so that the ethdev is fully set up after the
call to create the ethdev. The above-mentionned functions can still be
called - as will be the case, for instance, if the NIC is created via
commandline parameters - but they no longer are essential.
The function now also sets rte_errno appropriately on error, so the
caller can get a better indication of why a call may have failed.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Maryam Tahhan [Tue, 20 Oct 2015 10:34:18 +0000 (11:34 +0100)]
ethdev: do not deprecate imissed counter
Remove the deprecation tag and notice for imissed as it is a generic
register that accounts for packets that were dropped by the HW,
because there are no available mbufs (RX queues are full). imissed is
different to ierrors and can help with general debug.
This patch removes the mac local fault count and
mac remote fault count from rx errors. The mac
fault count registers count faults, not packets,
and hence should not be added to packet counters.
Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
Update the strings used for presenting stats to adhere
to the scheme previously presented. Updated xstats_get()
function to handle Q information only if xstats() is not
implemented in the PMD, providing the PMD with the needed
flexibility to expose its extended Q stats.
Add extended statistic section to the programmers
guide, poll mode driver section. This section describes
how the strings stats are formatted, and how the client
code can use this to gather information about the stat.
Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com> Acked-by: Maryam Tahhan <maryam.tahhan@intel.com>
Marcel Apfelbaum [Thu, 15 Oct 2015 11:08:39 +0000 (14:08 +0300)]
vhost: enable virtio 1.0
Make vhost-user virtio 1.0 compatible by adding it to the
supported features and keeping the header length
the same as for mergeable RX buffers.
Signed-off-by: Marcel Apfelbaum <marcel@redhat.com> Acked-by: Michael S. Tsirkin <mst@redhat.com> Reviewed-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Huawei Xie [Thu, 29 Oct 2015 14:53:26 +0000 (22:53 +0800)]
virtio: add vector Rx
With fixed avail ring, we don't need to get desc idx from avail ring.
virtio driver only has to deal with desc ring.
This patch uses vector instruction to accelerate processing desc ring.
Signed-off-by: Huawei Xie <huawei.xie@intel.com> Acked-by: Jianfeng Tan <jianfeng.tan@intel.com>
Huawei Xie [Thu, 29 Oct 2015 14:53:24 +0000 (22:53 +0800)]
virtio: optimize ring layout
In DPDK based switching environment, mostly vhost runs on a dedicated core
while virtio processing in guest VMs runs on different cores.
Take RX for example, with generic implementation, for each guest buffer,
a) virtio driver allocates a descriptor from free descriptor list
b) modify the entry of avail ring to point to allocated descriptor
c) after packet is received, free the descriptor
When vhost fetches the avail ring, it need to fetch the modified L1 cache from
virtio core, which is a heavy cost in current CPU implementation.
This idea of this optimization is:
allocate the fixed descriptor for each entry of avail ring, so avail ring will
always be the same during the run.
This removes L1M cache transfer from virtio core to vhost core for avail ring.
(Note we couldn't avoid the cache transfer for descriptors).
Besides, descriptor allocation and free operation is eliminated.
This also makes vector procesing possible to further accelerate the processing.
This is the layout for the avail ring(take 256 ring entries for example), with
each entry pointing to the descriptor with the same index.
avail
idx
+
|
+----+----+---+-------------+------+
| 0 | 1 | 2 | ... | 254 | 255 | avail ring
+-+--+-+--+-+-+---------+---+--+---+
| | | | | |
| | | | | |
v v v | v v
+-+--+-+--+-+-+---------+---+--+---+
| 0 | 1 | 2 | ... | 254 | 255 | desc ring
+----+----+---+-------------+------+
|
|
+----+----+---+-------------+------+
| 0 | 1 | 2 | | 254 | 255 | used ring
+----+----+---+-------------+------+
|
+
This is the ring layout for TX.
As we need one virtio header for each xmit packet, we have 128 slots available.
++
||
||
+-----+-----+-----+--------------+------+------+------+
| 0 | 1 | ... | 127 || 128 | 129 | ... | 255 | avail ring
+--+--+--+--+-----+---+------+---+--+---+------+--+---+
| | | || | | |
v v v || v v v
+--+--+--+--+-----+---+------+---+--+---+------+--+---+
| 128 | 129 | ... | 255 || 128 | 129 | ... | 255 | desc ring for virtio_net_hdr
+--+--+--+--+-----+---+------+---+--+---+------+--+---+
| | | || | | |
v v v || v v v
+--+--+--+--+-----+---+------+---+--+---+------+--+---+
| 0 | 1 | ... | 127 || 0 | 1 | ... | 127 | desc ring for tx dat
+-----+-----+-----+--------------+------+------+------+
||
||
++
Signed-off-by: Huawei Xie <huawei.xie@intel.com> Acked-by: Jianfeng Tan <jianfeng.tan@intel.com>
Vector RX function will process 4 packets at a time. When the RX
ring wrapps to the tail and the left descriptor size is not multiple
of 4, SW will overwrite memory that not belongs to it and cause crash.
The fix will allocate additional 4 HW/SW spaces at the tail to avoid
overwrite.
Vector TX use different way to manage TX queue, it's necessary
to use different functions to reset TX queue and release mbuf
in TX queue. So, introduce 2 function pointers to do such ops.
This patch add below functions:
1. Add function fm10k_rxq_rearm to re-allocate mbuf for used desc
in RX HW ring.
2. Add 2 functions, in which using SSE instructions to parse RX desc
to get pkt_type and ol_flags in mbuf.
3. Add func fm10k_recv_raw_pkts_vec to parse raw packets, in which
includes possible chained packets.
4. Add func fm10k_recv_pkts_vec to receive single mbuf packet.
Add condition check in rx_queue_setup func. If number of RX desc
can't satisfy vPMD requirement, record it into a variable. Or
call fm10k_rxq_vec_setup to initialize Vector RX.
Add the ability for the upper layer to query RX/TX queue information.
Add into rte_eth_dev_info new fields to represent information about
RX/TX descriptors min/max/alig nnumbers per queue for the device.
Add new structures:
struct rte_eth_rxq_info
struct rte_eth_txq_info
new functions:
rte_eth_rx_queue_info_get
rte_eth_tx_queue_info_get
into rte_etdev API.
Left extra free space in the queue info structures,
so extra fields could be added later without ABI breakage.
Add new fields:
rx_desc_lim
tx_desc_lim
into rte_eth_dev_info.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com> Acked-by: Remy Horton <remy.horton@intel.com>
According to XL710 datasheet:
RX QLEN restrictions: When the PXE_MODE flag in the GLLAN_RCTL_0
register is cleared, the QLEN must be whole number of 32
descriptors.
TX QLEN restrictions: When the PXE_MODE flag in the GLLAN_RCTL_0
register is cleared, the QLEN must be whole number of 32
descriptors.
So make sure that for both RX and TX queues number of HW descriptors is
a multiple of 32.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com> Acked-by: Remy Horton <remy.horton@intel.com>
Tomasz Kulasek [Fri, 30 Oct 2015 14:25:52 +0000 (15:25 +0100)]
null: add virtual RSS configuration
This implementation allows to set and read RSS configuration for null
device, and is used to validate right values propagation over the slaves,
in test units for dynamic RSS configuration for bonding.
Signed-off-by: Tomasz Kulasek <tomaszx.kulasek@intel.com> Acked-by: Tetsuya Mukawa <mukawa@igel.co.jp>
Tomasz Kulasek [Fri, 30 Oct 2015 14:25:48 +0000 (15:25 +0100)]
bonding: support RSS dynamic configuration
Bonding device implements independent management of RSS settings. It
stores its own copies of settings i.e. RETA, RSS hash function and RSS
key. It’s required to ensure consistency.
1) RSS hash function set for bonding device is maximal set of RSS hash
functions supported by all bonded devices. That mean, to have RSS support
for bonding, all slaves should be RSS-capable.
2) RSS key is propagated over the slaves "as is".
3) RETA for bonding is an internal table managed by bonding API, and is
used as a pattern to set up slaves. Its size is GCD of all RETA sizes, so
it can be easily used as a pattern providing expected behavior, even if
slaves RETA sizes are different.
Signed-off-by: Tomasz Kulasek <tomaszx.kulasek@intel.com> Acked-by: Declan Doherty <declan.doherty@intel.com>
Shaopeng He [Sat, 31 Oct 2015 02:44:42 +0000 (10:44 +0800)]
fm10k: support VMDQ in MAC/VLAN filter
The patch does below things for fm10k MAC/VLAN filter:
- Add separate functions for VMDQ and main VSI to change
MAC filter.
- Disable modification to VLAN filter in VMDQ mode.
- In device close phase, delete logic ports to remove all
MAC/VLAN filters belonging to those ports.
Signed-off-by: Shaopeng He <shaopeng.he@intel.com> Acked-by: Jingjing Wu <jingjing.wu@intel.com> Acked-by: Michael Qiu <michael.qiu@intel.com>
Jingjing Wu [Sat, 31 Oct 2015 15:57:25 +0000 (23:57 +0800)]
i40e: support DCB mode
This patch enables DCB feature on Intel XL710/X710 NICs. It includes:
Receive queue classification based on traffic class
Round Robin ETS schedule (rx and tx)
Priority flow control
Adrien Mazarguil [Fri, 30 Oct 2015 18:55:14 +0000 (19:55 +0100)]
app/testpmd: fix missing init in RSS hash show command
The "show port X rss-hash" command sometimes displays garbage instead of the
expected RSS hash key because the maximum key length is undefined. When the
requested key is too large to fit in the buffer,
rte_eth_dev_rss_hash_conf_get() does not update it.
Nelio Laranjeiro [Fri, 30 Oct 2015 18:55:13 +0000 (19:55 +0100)]
app/testpmd: add missing type to RSS hash commands
DPDK uses a structure to get or set a new hash key (see
eth_rte_rss_hash_conf). rss_hf field from this structure is used in
rss_hash_get_conf to retrieve the hash key and in rss_hash_update uses
it to verify the key exists before trying to update it.
Nelio Laranjeiro [Fri, 30 Oct 2015 18:55:12 +0000 (19:55 +0100)]
mlx5: use one RSS hash key per flow type
DPDK expects to have an RSS hash key per flow type (IPv4, IPv6, UDPv4,
etc.), to handle this the PMD must keep a table of hash keys to be able
to reconfigure the queues at each start/stop call.
Nelio Laranjeiro [Fri, 30 Oct 2015 18:55:11 +0000 (19:55 +0100)]
mlx5: support RSS hash update and get
First implementation of rss_hash_update and rss_hash_conf_get, those
functions still lack in functionality but are usable to change the RSS
hash key. For now, the PMD does not handle an indirection table for
each kind of flow (IPv4, IPv6, etc.), the same RSS hash key is used
for all protocols. This situation explains why the rss_hash_conf_get
returns the RSS hash key for all DPDK supported protocols and why the
hash key is set for all of them too.
Olga Shern [Fri, 30 Oct 2015 18:55:10 +0000 (19:55 +0100)]
mlx5: use alternate method to configure promisc and allmulti modes
Promiscuous and allmulticast modes were historically enabled by adding
specific flows with types IBV_FLOW_ATTR_ALL_DEFAULT or
IBV_EXP_FLOW_ATTR_MC_DEFAULT to each hash RX queue, but this method is
deprecated.
- Promiscuous mode is now enabled by omitting destination MAC addresses from
basic flow specifications.
- Allmulticast mode is now enabled by using flow specifications that match
the broadcast bit in destination MAC addresses.
Olga Shern [Fri, 30 Oct 2015 18:55:09 +0000 (19:55 +0100)]
mlx5: define specific flow steering rules for each hash Rx QP
All hash RX QPs currently use the same flow steering rule (L2 MAC filtering)
regardless of their type (TCP, UDP, IPv4, IPv6), which prevents them from
being dispatched properly. This is fixed by adding flow information to the
hash RX queue initialization data and generating specific flow steering
rules for each of them.
Olga Shern [Fri, 30 Oct 2015 18:55:07 +0000 (19:55 +0100)]
mlx5: use separate indirection table for default hash Rx queue
The default hash RX queue handles packets that are not matched by more
specific types and requires its own indirection table of size 1 to work
properly.
This commit implements support for multiple indirection tables by grouping
their layout and properties in a static initialization table.
Adrien Mazarguil [Fri, 30 Oct 2015 18:55:06 +0000 (19:55 +0100)]
mlx5: refactor Rx code for the new verbs RSS API
The new Verbs RSS API is lower-level than the previous one and much more
flexible but requires RX queues to use Work Queues (WQs) internally instead
of Queue Pairs (QPs), which are grouped in an indirection table used by a
new kind of hash RX QPs.
Hash RX QPs and the indirection table together replace the parent RSS QP
while WQs are mostly similar to child QPs.
RSS hash key is not configurable yet.
Summary of changes:
- Individual DPDK RX queues do not store flow properties anymore, this info
is now part of the hash RX queues.
- All functions affecting the parent queue when RSS is enabled or the basic
queues otherwise are modified to affect hash RX queues instead.
- Hash RX queues are also used when a single DPDK RX queue is configured (no
RSS) to remove that special case.
- Hash RX queues and indirection table are created/destroyed when device
is started/stopped in addition to create/destroy flows.
- Contrary to QPs, WQs are moved to the "ready" state before posting RX
buffers, otherwise they are ignored.
- Resource domain information is added to WQs for better performance.
- CQs are not resized anymore when switching between non-SG and SG modes as
it does not work correctly with WQs. Use the largest possible size
instead, since CQ size does not have to be the same as the number of
elements in the RX queue. This also applies to the maximum number of
outstanding WRs in a WQ (max_recv_wr).
Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com> Signed-off-by: Olga Shern <olgas@mellanox.com> Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com> Signed-off-by: Or Ami <ora@mellanox.com> Signed-off-by: Yaacov Hazan <yaacovh@mellanox.com>
Adrien Mazarguil [Fri, 30 Oct 2015 18:55:04 +0000 (19:55 +0100)]
mlx5: use fast verbs interface for scattered Rx
This commit updates mlx5_rx_burst_sp() to use the fast verbs interface for
posting RX buffers just like mlx5_rx_burst(). Doing so avoids a loop in
libmlx5 and an indirect function call through libibverbs.
Note: recv_sg_list() is not implemented in the QP burst API, this commit is
only to prepare transition to the WQ-based API.
Adrien Mazarguil [Fri, 30 Oct 2015 19:00:08 +0000 (20:00 +0100)]
mlx4: do not complain about port state
Port is considering inactive when the related netdevice is down. There is no
reason to warn about it (and confuse users) since it's automatically brought
up later by the PMD.
Nelio Laranjeiro [Fri, 30 Oct 2015 19:00:07 +0000 (20:00 +0100)]
mlx4: improve Rx performance with better prefetching
Prefetching initial bytes of mbuf structures earlier and in two cache lines
instead of one improves performance of mlx4_rx_burst(), which accesses the
mbuf->next field not present in the first 128 bytes.
Adrien Mazarguil [Fri, 30 Oct 2015 18:52:34 +0000 (19:52 +0100)]
mlx5: support scattered Rx and Tx
A dedicated RX callback is added to handle scattered buffers. For better
performance, it is only used when jumbo frames are enabled and MTU is larger
than a single mbuf.
On the TX path, scattered buffers are also handled in a separate function.
When there are more than MLX5_PMD_SGE_WR_N segments in a given mbuf, the
remaining segments are linearized in the last SGE entry.
Adrien Mazarguil [Fri, 30 Oct 2015 18:52:30 +0000 (19:52 +0100)]
mlx5: introduce new driver for Mellanox ConnectX-4 adapters
In its current state, this driver implements the bare minimum to initialize
itself and Mellanox ConnectX-4 adapters without doing anything else
(no RX/TX for instance). It is disabled by default since it is based on the
mlx4 driver and also depends on libibverbs.
Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com> Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro@6wind.com> Signed-off-by: Or Ami <ora@mellanox.com>
John Daley [Thu, 29 Oct 2015 18:45:16 +0000 (11:45 -0700)]
enic: improve Tx packet rate
For every packet sent, a completion was being requested and the
posted_index register on the nic was being updated. Instead, request a
completion and update the posted index once per burst after all
packets have been sent by the burst function.
Signed-off-by: John Daley <johndale@cisco.com> Acked-by: Sujith Sankar <ssujith@cisco.com>
git.droids-corp.org / dpdk.git / log