Symptom
-------
With NDR traffic blasting at VPP, bringing up a new VM with vhost
connection to VPP causes packet drops. I am able to recreate this
problem easily using a simple setup like this.
TREX-------------- switch ---- VPP
|---------------| |-------|
Cause
-----
The reason for the packet drops is due to vhost holding onto the worker
barrier lock for too long in vhost_user_socket_read(). There are quite a
few of system calls inside the routine. At the end of the routine, it
unconditionally calls vhost_user_update_iface_state() for all message
types. vhost_user_update_iface_state() also unconditionally calls
vhost_user_rx_thread_placement() and vhost_user_tx_thread_placement().
vhost_user_rx_thread_placement scraps out all existing cpu/queue mappings
for the interface and creates brand new cpu/queue mappings for the
interface. This process is very disruptive and very expensive. In my
opinion, this area of code needs a makeover.
Fixes
-----
* vhost_user_socket_read() is rewritten that it should not hold
onto the worker barrier lock for system calls, or at least minimize the
need for doing it.
* Remove the call to vhost_user_update_iface_state as a default route at
the end of vhost_user_socket_read(). There is only a couple of message
types which really need to call vhost_user_update_iface_state(). We put
the call to those message types which need it.
* Remove vhost_user_rx_thread_placement() and
vhost_user_tx_thread_placement from vhost_user_update_iface_state().
There is no need to repetatively change the cpu/queue mappings.
* vhost_user_rx_thread_placement() is actually quite expensive. It should
be called only once per queue for the interface. There is no need to
scrap the existing cpu/queue mappings and create new cpu/queue mappings
when the additional queues becomes active/enable.
* Change to create the cpu/queue mappings for the first RX when the
interface is created. Dont remove the cpu/queue mapping when the
interface is disconnected. Remove the cpu/queue mapping only when the
interface is deleted.
The create vhost user interface CLI also has some very expensive system
calls if the command is entered with the optional keyword "server"
As a bonus, This patch makes the create vhost user interface binary-api and
CLI thread safe. Do the protection for the small amount of code which is
thread unsafe.
Change-Id: I4a19cbf7e9cc37ea01286169882e5603e6d7eb77
Signed-off-by: Steven Luong <sluong@cisco.com>
Coverity complains about dead code as shown below and it is right.
The fix is to simply remove the dead code.
503 if (v_indicator != indicator)
CID 190173 (#3 of 3): Logically dead code (DEADCODE)
dead_error_line: Execution cannot reach this statement: return -1;.
504 return -1;
Change-Id: Ibca9e10451a4459db099bef5ecc6939474bdb903
Signed-off-by: Steven Luong <sluong@cisco.com>
During CSIT testing we discovered that LACP tests were failing and
producing coredumps. Reverting this patch fix the problem with VPP
crashing.
This reverts commit f23890138e.
Change-Id: Icf97053ce1473350add885cbebe591f7f3efcbea
Signed-off-by: Peter Mikus <pmikus@cisco.com>
Some x86 CPUs have IOMMU capable dealing only with 39-bit address space
This patch also adds option to specify physmem base address from
startup.conf
Change-Id: I9e8abd26efb60e9c4ad54c035fb1751a4a61f4dc
Signed-off-by: Damjan Marion <damarion@cisco.com>
Create a common pg_send method for cases when results are just discarded.
Change-Id: I786960d2d7bbb96dcb407f6e59aa96951b7b19e7
Signed-off-by: Paul Vinciguerra <pvinci@vinciconsulting.com>
Some tracked files were ignored due to gitignore rules that date back to
autotools based buildsystem. Update gitignore accordingly.
~# git ls-files -i --exclude-standard
Makefile
build-root/Makefile
build/external/Makefile
build/external/deb/debian/dkms/Makefile
doxygen/Makefile
extras/rpm/Makefile
src/vnet/config.h
test/Makefile
test/doc/Makefile
test/ext/Makefile
Change-Id: I3326c932cd174434211b744bf5e26bc4da450339
Signed-off-by: Benoît Ganne <bganne@cisco.com>
Only with debug image and when next node is not ethernet-input...
Change-Id: Iaa404b5d35d5c04996ff48cd16877858092b78d7
Signed-off-by: Damjan Marion <damarion@cisco.com>
The main thread squirrels away vlib_time_now (&vlib_global_main),
worker threads use it to calculate an offset in f64 seconds from their
own vlib_time_now(vm) value. We use that offset until the next barrier
sync.
Thanks to Damjan for the suggestion.
Change-Id: If56cdfe68e5ad8ac3b0d0fc885dc3ba556cd1215
Signed-off-by: Dave Barach <dave@barachs.net>
vom library was missing version apended to libvom.so
This patch fixes this issue.
Change-Id: I99a6c282d651e1882574af999f04610d7a2c4d44
Signed-off-by: Mohsin Kazmi <sykazmi@cisco.com>
Mock additional 3rd party modules and replace wildcard imports that
were confusing sphinx.
Change-Id: Ia9e599901a2a3fad854d3631dd0991183ae6d715
Signed-off-by: Paul Vinciguerra <pvinci@vinciconsulting.com>
This covers both naming option for opensuse leap15
dep package naming
Change-Id: I2ade004e44e75c08afde0f3af42b33cd97ce0ebc
Signed-off-by: Ed Kern <ejk@cisco.com>
Unless there's an error, recvfrom should return the number of bytes
received. The current code returns zero on success due to
ldp_copy_ep_to_sockaddr returning zero.
Change-Id: Ib843b99a08c44b5ee2d16efe9edab0cf5df65170
Signed-off-by: Haggai Eran <haggai.eran@gmail.com>
As part of JVPP migration this removes JVPP completely from VPP.
Change-Id: I6b74e7961aa474ae471e63fe43a624cd9fc3659b
Signed-off-by: Michal Cmarada <mcmarada@cisco.com>
Recent VIC models can support 25, 50, and 100Gbps links. Use the
helper (port_type_from_link_speed) to set the port type as it supports
all possible link speeds.
Change-Id: I748d8ac716a6393d116a9db8a599151c70a9000a
Signed-off-by: Hyong Youb Kim <hyonkim@cisco.com>
This commit adds a "gso" parameter to existing "create tap..." CLI,
and a "no-gso" parameter for the compatibility with the future,
when/if defaults change.
It makes use of the lowest bit of the "tap_flags" field in the API call
in order to allow creation of GSO interfaces via API as well.
It does the necessary syscalls to enable the GSO
and checksum offload support on the kernel side and sets two flags
on the interface: virtio-specific virtio_if_t.gso_enabled,
and vnet_hw_interface_t.flags & VNET_HW_INTERFACE_FLAG_SUPPORTS_GSO.
The first one, if enabled, triggers the marking of the GSO-encapsulated
packets on ingress with VNET_BUFFER_F_GSO flag, and
setting vnet_buffer2(b)->gso_size to the desired L4 payload size.
VNET_HW_INTERFACE_FLAG_SUPPORTS_GSO determines the egress packet
processing in interface-output for such packets:
When the flag is set, they are sent out almost as usual (just taking
care to set the vnet header for virtio).
When the flag is not enabled (the case for most interfaces),
the egress path performs the re-segmentation such that
the L4 payload of the transmitted packets equals gso_size.
The operations in the datapath are enabled only when there is at least
one GSO-compatible interface in the system - this is done by tracking
the count in interface_main.gso_interface_count. This way the impact
of conditional checks for the setups that do not use GSO is minimized.
"show tap" CLI shows the state of the GSO flag on the interface, and
the total count of GSO-enabled interfaces (which is used to enable
the GSO-related processing in the packet path).
This commit lacks IPv6 extension header traversal support of any kind -
the L4 payload is assumed to follow the IPv6 header. Also it performs
the offloads only for TCP (TSO - TCP segmentation offload).
The UDP fragmentation offload (UFO) is not part of it.
For debug purposes it also adds the debug CLI:
"set tap gso {<interface> | sw_if_index <sw_idx>} <enable|disable>"
Change-Id: Ifd562db89adcc2208094b3d1032cee8c307aaef9
Signed-off-by: Andrew Yourtchenko <ayourtch@gmail.com>
/vpp/src/tools/vppapigen/vppapigen:823: DeprecationWarning:
the imp module is deprecated in favour of importlib;
see the module's documentation for alternative uses
Change-Id: If7729778374e9193f6381c8bd2ed34c875db3f1e
Signed-off-by: Paul Vinciguerra <pvinci@vinciconsulting.com>
- Better handle buffer starvation scenarios
- Handle case when both peers enter recovery due to packet loss.
- Fix passive open establish cleanup
Change-Id: I2f28baa2ff0383bb8f5f6d2452b49aa38ce69bce
Signed-off-by: Florin Coras <fcoras@cisco.com>
Makes ip6-local node dual-loop explicit. This is only a style change.
Change-Id: Ic8e7cecb3f51e98b8a069b501f5c338156934a6d
Signed-off-by: Benoît Ganne <bganne@cisco.com>
1) stats are accessed via the stat segment which is more condusive to
monitoring
2) stats are accurate in the presence of multiple threads. There's no
guarantee that an SA is access from only one worker.
Change-Id: Id5e217ea253ddfc9480aaedb0d008dea031b1148
Signed-off-by: Neale Ranns <nranns@cisco.com>
Optimize IPv6 ip6-local node by rewriting the dual/single loop with
prefetch and simpler unrolling.
My local, unrepresentative tests for GRE4 termination over IPv6 show a
performance improvement of ~40% for ip6-local node alone and ~5%
globally.
Change-Id: I11e1e86d3838dd3c081aa6be5e25dae16ed6e2d8
Signed-off-by: Benoît Ganne <bganne@cisco.com>
Steven Luong