Change the size of m->port and m->nb_segs to 16 bits. It is now possible
to reference a port identifier larger than 256 and have a mbuf chain
larger than 256 segments.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
To avoid multiple stores on fast path, Ethernet drivers
aggregate the writes to data_off, refcnt, nb_segs and port
to an uint64_t data and write the data in one shot
with uint64_t* at &mbuf->rearm_data address.
Some of the non-IA platforms have store operation overhead
if the store address is not naturally aligned.This patch
fixes the performance issue on those targets.
Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Set the value of m->refcnt to 1, m->nb_segs to 1 and m->next
to NULL when the mbuf is stored inside the mempool (unused).
This is done in rte_pktmbuf_prefree_seg(), before freeing or
recycling a mbuf.
Before this patch, the value of m->refcnt was expected to be 0
while in pool.
The objectives are:
- to avoid drivers to set m->next to NULL in the early Rx path, since
this field is in the second 64B of the mbuf and its access could
trigger a cache miss
- rationalize the behavior of raw_alloc/raw_free: one is now the
symmetric of the other, and refcnt is never changed in these functions.
To optimize the freeing of the segments, we try try to only update
m->refcnt, m->next, and m->nb_segs when it's required (idea from
Konstantin Ananyev <konstantin.ananyev@intel.com>).
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Rename __rte_mbuf_raw_free() as rte_mbuf_raw_free() and make
it public. The old function is kept for compat but is marked as
deprecated.
The next commit changes the behavior of rte_mbuf_raw_free() to
make it more consistent with rte_mbuf_raw_alloc().
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Document the function and make it public, since it is used at several
places in the drivers. The old one is marked as deprecated.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
This commit documents two error return values for the
rte_event_dev_start() function.
-ESTALE indicates not all ports are configured
-ENOLINK indicates that not all queues are linked to ports. If an
application enqueues to such a queue it can lead to deadlock
Suggested-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
This commit adds rte_errno return values to rte_event_enqueue_burst() and
rte_event_dequeue_burst().
These return values allows user software to differentiate between an
invalid argument (such as an invalid queue_id or sched_type in an enqueued
event) and backpressure from the event device.
The port and device ID checks are placed in RTE_LIBRTE_EVENTDEV_DEBUG
header guards to avoid the performance hit in non-debug execution.
Signed-off-by: Gage Eads <gage.eads@intel.com>
Add in APIs for extended stats so that eventdev implementations can report
out information on their internal state. The APIs are based on, but not
identical to, the equivalent ethdev functions.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
PMDs that only do a specific type of scheduling cannot provide
CFG_ALL_TYPES, so the Eventdev infrastructure should not demand
that every PMD supports CFG_ALL_TYPES.
By not overriding the default configuration of the queue as
suggested by the PMD, the eventdev_common unit tests can pass
on all PMDs, regardless of their capabilities.
RTE_EVENT_QUEUE_CFG_DEFAULT is no longer used by the eventdev layer
it can be removed now. Applications should use CFG_ALL_TYPES
if they require enqueue of all types a queue, or specify which
type of queue they require.
The CFG_DEFAULT value is changed to CFG_ALL_TYPES in event/skeleton,
to not break the compile.
A capability flag is added that indicates if the underlying PMD
supports creating queues of ALL_TYPES.
Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
eventdev driver may return error on dequeue timeout tick conversion.
Change the pmd callback interface to address the same.
Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>
This patch initializes the links_map array entries to
EVENT_QUEUE_SERVICE_PRIORITY_INVALID, as expected by
rte_event_port_links_get(). This is necessary for the sw eventdev PMD,
which does not initialize links_map when rte_event_port_setup() calls
rte_event_port_unlink().
Fixes: 4f0804bbdfb9 ("eventdev: implement the northbound APIs")
Signed-off-by: Gage Eads <gage.eads@intel.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
rte_device is a generic device which is available to the applications
and EAL. This patch replaces rte_pci_device in 'struct rte_eventdev'
and in 'struct rte_event_dev_info' with common rte_device.
Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Acked-by: Shreyansh Jain <shreyansh.jain@nxp.com>
Improve the documentation of the return values of the
rte_event_dequeue_timeout_ticks() function, adding a
-ENOTSUP value for eventdevs that do not support waiting.
Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Large port enqueue sizes were not supported as the value
it was stored in was a uint8_t. Using uint8_ts to save
space in config apis makes no sense - increasing the 3
instances of uint8_t enqueue / dequeue depths to more
appropriate values (based on the context around them).
Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
This commit clarifies the usage of nb_links and nb_unlinks when passing
a NULL pointer as the queues argument.
Signed-off-by: Gage Eads <gage.eads@intel.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Updated the comments on 'nb_events_limit' of 'struct rte_event_dev_config'
and 'new_event_threshold' of 'struct rte_event_port_conf'.
Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>
On port_setup, the link_map is updated only
for configured number of event queues.
Limit the port_links_get scan only to configured number
of event queues. Also, Limit the port link and unlink queue
validation to configured number of event queues.
Fixes: 4f0804bbdfb9 ("eventdev: implement the northbound APIs")
Reported-by: Nipun Gupta <nipun.gupta@nxp.com>
Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Acked-by: Nipun Gupta <nipun.gupta@nxp.com>
Added eventdev vdev uninit support to release the resources
allocated in eventdev vdev init.
Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>
- Removed uninitialized max_devs value
- Corrected dev assignment
Fixes: 4f0804bbdfb9 ("eventdev: implement the northbound APIs")
Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>
Since eventdev uses event structures rather than working directly on
mbufs, there is no actual dependencies on the mbuf library. The
inclusion of an mbuf pointer element inside the event itself does not
require the inclusion of the mbuf header file. Similarly the pci
header is not needed, but following their removal, rte_memory.h is
needed for the definition of the __rte_cache_aligned macro.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Signed-off-by: Harry van Haaren <harry.van.haaren@intel.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Added a pointer to the rte_eventdev type in the event port
link and unlink callbacks. This device shall be used by some
of the event drivers to fetch queue related information.
Also, update the skeleton eventdev driver with corresponding changes.
Signed-off-by: Nipun Gupta <nipun.gupta@nxp.com>
Acked-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
This patch adds infrastructure for registering the vdev or
the PCI based event device.
Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
This patch implements northbound eventdev API interface using
southbond driver interface
Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
In a polling model, lcores poll ethdev ports and associated
rx queues directly to look for packet. In an event driven model,
by contrast, lcores call the scheduler that selects packets for
them based on programmer-specified criteria. Eventdev library
adds support for event driven programming model, which offer
applications automatic multicore scaling, dynamic load balancing,
pipelining, packet ingress order maintenance and
synchronization services to simplify application packet processing.
By introducing event driven programming model, DPDK can support
both polling and event driven programming models for packet processing,
and applications are free to choose whatever model
(or combination of the two) that best suits their needs.
This patch adds the eventdev specification header file.
Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
It doesn't make any sense to invoke destroy_device() callback at
while handling SET_MEM_TABLE message.
From the vhost-user spec, it's the GET_VRING_BASE message indicates
the end of a vhost device: the destroy_device() should be invoked
from there (luckily, we already did that).
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
rte_mbuf struct is something more likely will be used only in vhost-user
net driver, while we have made vhost-user generic enough that it can
be used for implementing other drivers (such as vhost-user SCSI), they
have also include <rte_mbuf.h>. Otherwise, the build will be broken.
We could workaround it by using forward declaration, so that other
non-net drivers won't need include <rte_mbuf.h>.
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Rename "rte_virtio_net.h" to "rte_vhost.h", to not let it be virtio
net specific.
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
We used to use rte_vhost_driver_session_start() to trigger the vhost-user
session. It takes no argument, thus it's a global trigger. And it could
be problematic.
The issue is, currently, rte_vhost_driver_register(path, flags) actually
tries to put it into the session loop (by fdset_add). However, it needs
a set of APIs to set a vhost-user driver properly:
* rte_vhost_driver_register(path, flags);
* rte_vhost_driver_set_features(path, features);
* rte_vhost_driver_callback_register(path, vhost_device_ops);
If a new vhost-user driver is registered after the trigger (think OVS-DPDK
that could add a port dynamically from cmdline), the current code will
effectively starts the session for the new driver just after the first
API rte_vhost_driver_register() is invoked, leaving later calls taking
no effect at all.
To handle the case properly, this patch introduce a new API,
rte_vhost_driver_start(path), to trigger a specific vhost-user driver.
To do that, the rte_vhost_driver_register(path, flags) is simplified
to create the socket only and let rte_vhost_driver_start(path) to
actually put it into the session loop.
Meanwhile, the rte_vhost_driver_session_start is removed: we could hide
the session thread internally (create the thread if it has not been
created). This would also simplify the application.
NOTE: the API order in prog guide is slightly adjusted for showing the
correct invoke order.
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Export few APIs for the vhost-user driver to log the guest memory writes,
which is a must for live migration support.
This patch basically moves vhost_log_write() and vhost_log_used_vring()
into vhost.h and then add an wrapper (the public API) to them.
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Features could be changed after the feature negotiation. For example,
VHOST_F_LOG_ALL will be set/cleared at the start/end of live migration,
respecitively. Thus, we need a new callback to inform the application
on such change.
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Rename "virtio-net" to "vhost" in the API comments and vhost prog guide.
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
rename "virtio_net_device_ops" to "vhost_device_ops", to not let it
be virtio-net specific.
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
They are virtio-net specific and should be defined inside the virtio-net
driver.
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Currently, we check vq->desc, vq->kickfd and vq->callfd to know whether
a virtio device is ready or not. However, we only do it when handling
SET_VRING_KICK message, which could be wrong if a vhost-user frontend
send SET_VRING_KICK first and SET_VRING_CALL later.
To work for all possible vhost-user frontend implementations, we could
move the ready check at the end of vhost-user message handler.
Meanwhile, since we do the check more often than before, the "virtio
not ready" message is dropped, to not flood the screen.
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
We used to use rte_vhost_get_queue_num() for telling how many vrings.
However, the return value is the number of "queue pairs", which is
very virtio-net specific. To make it generic, we should return the
number of vrings instead, and let the driver do the proper translation.
Say, virtio-net driver could turn it to the number of queue pairs by
dividing 2.
Meanwhile, mark rte_vhost_get_queue_num as deprecated.
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
The queue pair is very virtio-net specific, other devices don't have
such concept. To make it generic, we should log the number of vrings
instead of the number of queue pairs.
This patch just does a simple convert, a later patch would export the
number of vrings to applications.
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Some vhost-user driver may need this info to setup its own page tables
for GPA (guest physical addr) to HPA (host physical addr) translation.
SPDK (Storage Performance Development Kit) is one example.
Besides, by exporting this memory info, we could also export the
gpa_to_vva() as an inline function, which helps for performance.
Otherwise, it has to be referenced indirectly by a "vid".
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Assume there is an application both support vhost-user net and
vhost-user scsi, the callback should be different. Making notify
ops per vhost driver allow application define different set of
callbacks for different driver.
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Introduce few APIs to set/get/enable/disable driver features.
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
A vhost-user server socket could have many connections, thus many connfd.
However, we currently just use one single int var to store it. Meaning,
it will get overwritten every time a new connection is created.
While this will not create fatal issue as it sounds (since the correct
connfd is closured to the event loop thread by fdset_add), it may cause
fd leaks if a user invokes rte_vhost_driver_unregister before shutting
down all connections: it just closes the recent connfd.
A simple example that should be able to reproduce this leaks issues is,
del the ovs vhost-user port while the connected VMs are still alive. (Note
that it's suggested to use one socket for one VM, which makes the issue
not that fatal as it sounds again).
Since we already use a struct "vhost_user_connection" to track all info
about one connection, it's obvious that we should put the connfd there.
Then we could build a connection list inside the vhost_user_socket struct,
to represent all connections belong that socket file.
Fixes: 164fd396788d ("vhost: fix unregistering in client mode")
Cc: stable@dpdk.org
Cc: Ilya Maximets <i.maximets@samsung.com>
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
A new interrupt type, RTE_INTR_HANDLE_VDEV, is added to support lsc and rxq
interrupt for vdev.
For lsc interrupt, except from original EPOLLIN events, we also listen for
socket peer closed connection event (EPOLLRDHUP and EPOLLHUP).
For rxq interrupt, add a precondition to avoid invoking any vfio and uio
code.
For intr_handle initialization, let each vdev driver to do that.
Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com>
The broadcast_rarp field in the virtio_net struct is checked in the
dequeue datapath regardless of whether descriptors are available or not.
As it is checked with cmpset leading to a write, false sharing on the
virtio_net struct can happen between enqueue and dequeue datapaths
regardless of whether a RARP is requested. In OVS, the issue can cause
a uni-directional performance drop of up to 15%.
Fix that by only performing the cmpset if a read of broadcast_rarp
indicates that the cmpset is likely to succeed.
Fixes: a66bcad32240 ("vhost: arrange struct fields for better cache sharing")
Cc: stable@dpdk.org
Signed-off-by: Kevin Traynor <ktraynor@redhat.com>
Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
This patch implements the function for the application to
get the MTU value.
rte_vhost_get_mtu() fills the mtu parameter with the MTU value
set in QEMU if VIRTIO_NET_F_MTU has been negotiated and returns 0,
-ENOTSUP otherwise.
The function returns -EAGAIN if Virtio feature negotiation
didn't happened yet.
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Acked-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
