Each object stored in mempools are suffixed by a trailer, storing
a cookie in debug mode which help to detect memory corruptions.
Like for headers, introduce a structure that materializes the content of
this trailer.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Each object stored in mempools are prefixed by a header, allowing for
instance to retrieve the mempool pointer from the object. When debug is
enabled, a cookie is also added in this header that helps to detect
corruptions and double-frees.
Introduce a structure that materializes the content of this header,
and will simplify future patches adding things in this header.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
This patch adds methods that use hardware memory transactions (HTM) on
fast-path for rwlock (a.k.a. lock elision). Here the methods are implemented
for x86 using Restricted Transactional Memory instructions (Intel(r)
Transactional Synchronization Extensions). The implementation fall-backs to
the normal rwlock if HTM is not available or memory transactions fail. This is
not a replacement for all rwlock usages since not all critical sections
protected by locks are friendly to HTM. For example, an attempt to perform
a HW I/O operation inside a hardware memory transaction always aborts
the transaction since the CPU is not able to roll-back should the transaction
fail. Therefore, hardware transactional locks are not advised to be used around
rte_eth_rx_burst() and rte_eth_tx_burst() calls.
Signed-off-by: Roman Dementiev <roman.dementiev@intel.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
This patch adds methods that use hardware memory transactions (HTM) on fast-path
for spinlocks (a.k.a. lock elision). Here the methods are implemented for x86
using Restricted Transactional Memory instructions (Intel(r) Transactional
Synchronization Extensions). The implementation fall-backs to the normal
spinlock if HTM is not available or memory transactions fail. This is not
a replacement for all spinlock usages since not all critical sections protected
by spinlocks are friendly to HTM. For example, an attempt to perform a HW I/O
operation inside a hardware memory transaction always aborts the transaction
since the CPU is not able to roll-back should the transaction fail.
Therefore, hardware transactional locks are not advised to be used around
rte_eth_rx_burst() and rte_eth_tx_burst() calls.
Signed-off-by: Roman Dementiev <roman.dementiev@intel.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Introduce new RTE_ACL_MASKLEN_TO_BITMASK macro, that will be used
in several places inside librte_acl and it's UT.
Simplify and cleanup build_trie() code a bit.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
When rebuilding a trie for limited rule-set,
don't try to split the rule-set even further.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Move check for build confg parameter into a separate function.
Simplify acl_calc_wildness() function.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
When ring_writer_nodrop port fails to send data, it tries to resend.
Operation is aborted when maximum number of retries is reached.
Signed-off-by: Maciej Gajdzica <maciejx.t.gajdzica@intel.com>
Acked-by: Cristian Dumitrescu <cristian.dumitrescu@intel.com>
When ethdev_writer_nodrop port fails to send data, it tries to resend.
Operation is aborted when maximum number of retries is reached.
Signed-off-by: Maciej Gajdzica <maciejx.t.gajdzica@intel.com>
Acked-by: Cristian Dumitrescu <cristian.dumitrescu@intel.com>
New implementation sends burst without copying data to internal buffer
if it is possible. It is similar to tx_bulk function in ethdev_writer
port.
Signed-off-by: Maciej Gajdzica <maciejx.t.gajdzica@intel.com>
Acked-by: Cristian Dumitrescu <cristian.dumitrescu@intel.com>
There was two implementations of tx_bulk function in ethdev_writer port.
The function to run is chosen with WRITER_APPROACH define. This patch
removes WRITER_APPROACH = 0 implementation, as it seems to be slower.
Signed-off-by: Maciej Gajdzica <maciejx.t.gajdzica@intel.com>
Acked-by: Cristian Dumitrescu <cristian.dumitrescu@intel.com>
This patch fix doxygen warnings when generating documentation
for qos_meter and qos_sched.
Signed-off-by: Michal Jastrzebski <michalx.k.jastrzebski@intel.com>
Acked-by: Cristian Dumitrescu <cristian.dumitrescu@intel.com>
Remove these unnecessary vring descriptor length updating, vhost should
not change them.
virtio in front end should assign value to desc.len for both rx and tx.
Test report: http://dpdk.org/ml/archives/dev/2015-June/018610.html
Signed-off-by: Changchun Ouyang <changchun.ouyang@intel.com>
Acked-by: Huawei Xie <huawei.xie@intel.com>
Extract codes into a function:
update_secure_len which is used to accumulate the buffer len in the
vring descriptors and to fill struct buf_vec.
Signed-off-by: Changchun Ouyang <changchun.ouyang@intel.com>
Acked-by: Huawei Xie <huawei.xie@intel.com>
Vring enqueue need consider the 2 cases:
1. use separate descriptors to contain virtio header and actual data,
e.g. the first descriptor is for virtio header, and then followed
by descriptors for actual data.
2. virtio header and some data are put together in one descriptor,
e.g. the first descriptor contain both virtio header and part of
actual data, and then followed by more descriptors for rest of packet
data, current DPDK based virtio-net pmd implementation is this case;
So does vring dequeue, it should not assume vring descriptor is chained
or not chained, it should use desc->flags to check whether it is chained
or not. This patch also fixes TX corrupt issue when vhost co-work with
virtio-net driver which uses one single vring descriptor (header and data
are in one descriptor) for virtio tx process on default.
Test report: http://dpdk.org/ml/archives/dev/2015-June/018610.html
Signed-off-by: Changchun Ouyang <changchun.ouyang@intel.com>
Acked-by: Huawei Xie <huawei.xie@intel.com>
In containers like docker, current->pid returns current process's global
PID instead of its own PID under containers's PID namespace, and
get_net_ns_by_pid() suppose to accept a virtual PID under its own
namespace, so we should use task_pid_vnr(current) to get current process's
virtual PID instead of current->pid.
Signed-off-by: Wenfeng Liu <liuwf@arraynetworks.com.cn>
Acked-by: Helin Zhang <helin.zhang@intel.com>
We did some (very basic) tests with IGMP, which involves adding
multicast addresses to ETH interfaces. This is done via the ip tool,
an example can be found on e.g.,
http://superuser.com/questions/324824/linux-built-in-or-open-source-program-to-join-multicast-group
and this will fail on KNI interfaces because of an unimplemented ioctl
SIOCADDMULTI. The patch simply adds an empty callback for set_rx_mode
(typically used for setting up hardware) so that the ioctl succeeds.
This is the same thing as the Linux tap interface does.
Signed-off-by: Simon Kagstrom <simon.kagstrom@netinsight.net>
Signed-off-by: Johan Faltstrom <johan.faltstrom@netinsight.net>
Reviewed-by: Stephen Hemminger <stephen@networkplumber.org>
Acked-by: Helin Zhang <helin.zhang@intel.com>
Loop processing packets dequeued from rx_q was using the number of
packets requested, not how many it actually received.
Variable rename to make code a little more clear
Signed-off-by: Jay Rolette <rolette@infiniteio.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Helin Zhang <helin.zhang@intel.com>
No reason to check out many entries are in kni->rx_q prior to
actually pulling them from the fifo. You can't dequeue more than
are there anyway. Max entries to dequeue is either the max batch
size or however much space is available on kni->free_q (lesser of the two).
Signed-off-by: Jay Rolette <rolette@infiniteio.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Helin Zhang <helin.zhang@intel.com>
Do not need the 'safe' version of list_for_each_entry() if you are
not deleting from the list as you iterate over it.
Signed-off-by: Jay Rolette <rolette@infiniteio.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Helin Zhang <helin.zhang@intel.com>
Implement .ndo_change_carrier to enable
DPDK applications to propagate link state changes to
kni virtual interfaces through sysfs
Signed-off-by: Vijayakumar Muthuvel Manickam <mmvijay@gmail.com>
Acked-by: Helin Zhang <helin.zhang@intel.com>
When a KNI object is created, a name is assigned to it which is stored
internally. There is also an API function to look up a KNI object by
name, but there is no API to query the current name of an existing
KNI object. This patch adds just such an API.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Helin Zhang <helin.zhang@intel.com>
Changed name to something more meaningful,
and mark rte_jhash2 as deprecated.
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
rte_jhash is basically like __rte_jhash_2hashes but
it returns only 1 hash, instead of 2.
In order to remove duplicated code, rte_jhash calls __rte_jhash_2hashes,
passing 0 as the second seed and returning just the first hash value.
(performance penalty is negligible)
The same is done with rte_jhash2. Also, rte_jhash2 is just an specific case
where keys are multiple of 32 bits, and where no key alignment check is required.
So,to avoid duplicated code, the function calls __rte_jhash_2hashes
with check_align = 0 (to use the optimal path)
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
With the jhash update, two new functions were introduced:
- rte_jhash_2hashes: Same as rte_jhash, but takes two seeds
and return two hashes (uint32_ts)
- rte_jhash2_2hashes: Same as rte_jhash2, but takes two seeds
and return two hashes (uint32_ts)
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
[Thomas: fix doxygen typos]
Jenkins hash function was developed originally in 1996,
and was integrated in first versions of DPDK.
The function has been improved in 2006,
achieving up to 35% better performance, compared to the original one.
This patch integrates that code into the rte_jhash library.
It also updates the precalculated hash values in the unit test,
as the code now returns different values (expected).
A final note has been added in release notes for stating
the changes made.
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Needed to run as non-root but with higher memory allocations, and
removes a constraint on no-huge mode being limited to 64M. A usage
example is if running with file input with the pcap PMD, which can be
done as non-root after this patch via e.g.,
./test-dpdk --no-huge -m 1024 -l 0,1 -n3
--vdev 'eth_pcap0,rx_pcap=eth-rx.pcap,tx_pcap=eth-tx.pcap'
Signed-off-by: Simon Kagstrom <simon.kagstrom@netinsight.net>
Signed-off-by: Johan Faltstrom <johan.faltstrom@netinsight.net>
Acked-by: David Marchand <david.marchand@6wind.com>
When we migrate VM, without this feature, qemu will report error:
"migrate: Migration disabled: vhost lacks VHOST_F_LOG_ALL feature".
Signed-off-by: Krishna Murthy <krishna.j.murthy@intel.com>
In __rte_pktmbuf_prefree_seg(), there was an optimization to avoid using
a costly atomic operation when updating the mbuf reference counter if
its value is 1. Indeed, it means that we are the only owner of the mbuf,
and therefore nobody can change it at the same time.
We can generalize this optimization directly in rte_mbuf_refcnt_update()
so the other callers of this function, like rte_pktmbuf_attach(), can
also take advantage of this optimization.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
With the current PMD API, the receipt of multicast packets on a given
port can only be enabled by invoking the "rte_eth_allmulticast_enable"
function.
This method may not work on Virtual Functions in SR-IOV architectures
when the host PF driver does not allow such operation on VFs.
In such cases, joined multicast addresses must be individually added
in the set of multicast addresses that are filtered by the [VF] port.
For this purpose, a new function "set_mc_addr_list" is introduced
into the set of functions that are exported by a Poll Mode Driver.
Signed-off-by: Ivan Boule <ivan.boule@6wind.com>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>
[Thomas: export new function in .map]
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Ran this code base through a script which:
- removes trailing whitespace
- removes space before tabs
- removes blank lines at end of file
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Acked-by: Helin Zhang <helin.zhang@intel.com>
As now subtree_id is not used acl_merge_trie() any more,
there is no point to calculate and maintain that information.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Reported by Zi Hu:
"
cat test_data/rule1
@192.168.0.0/24 192.168.0.0/24 400 : 500 0 : 52 6/0xff
@192.168.0.0/24 192.168.0.0/24 400 : 500 54 : 65280 6/0xff
@192.168.0.0/24 192.168.0.0/24 400 : 500 0 : 65535 6/0xff
cat test_data/trace1
0xc0a80005 0xc0a80009 450 53 0x06
I run the test by:
sudo ./testacl -n 2 -c 4 -- --rulesf=./test_data/rule1
--tracef=./test_data/trace1
The result shows that the packet matches the second rule, which is wrong.
The dest port of the pkt is 53, so it should match the third rule.
"
Indeed there is problem at ACL build stage.
Sometimes acl_merge_trie() is too aggressive in trying to conserve
space at build time.
So it takes a wrong assumptions and didn't duplicate a node,
even when it should.
The easiest and safest fix seems to always duplicate a left non-root/non-leaf
node first, and let the further code to destroy the node, if it is not needed.
Reported-by: Zi Hu <huzilucky@gmail.com>
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
eth_stats is already cleared by rte_eth_stats_get
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Using the "physical_package_id" as a fallback for determining the
numa node of a core tends to be unreliable. Fix this by using a
detection routine which reads the numa information from
/sys/devices/system/node and just returns a numa node of 0 on
failure.
Reported-by: Wang Sheng-Hui <shhuiw@gmail.com>
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Stephen Hemminger <stephen@networkplumber.org>
On Fedora 22, with GCC 5.1, errors are reported due to array accesses
being potentially out of bounds. This commit fixes this by adding in an
extra bounds check to the loop counter.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
On Fedora 22, with GCC 5.1, errors are reported due to array accesses
being potentially out of bounds. This commit fixes this by ensuring the
bounds check in the loop takes account of the array size.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
The file rte_kni.h depends upon a number of other headers, some of which
are missing from the #include lines. The following #includes are added:
* rte_memory.h - for the definition of phys_addr_t
* rte_mempool.h - for the definition of mempool struct and the mempool
create function.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Marc Sune <marc.sune@bisdn.de>
rte_pci.h depends upon stdio.h for the definition of the FILE type. Add
in #include <stdio.h> to the file to satisfy this dependency in cases
where the including C file does not already include stdio.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Marc Sune <marc.sune@bisdn.de>
In rte_mempool_obj_iter(), when element boundary coincides with page boundary,
even if a single page is required per object, a loop checks that the next page
is contiguous and drops the first one otherwise.
This commit checks subsequent pages only when several are required per object.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Reviewed-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
