Allow the nic bind/unbind script to print out its status messages even
if the igb_uio driver is not loaded. For binding and unbinding NICs, the
behaviour is the same, and the igb_uio driver still needs to be loaded.
Signed-off-by: Bruce richardson <bruce.richardson@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
When binding devices to a generic driver (i.e. one that doesn't have a
PCI ID table, some devices that are not bound to any other driver could
be bound even if no one has asked them to. hence, we check the list of
drivers again, and see if some of the previously-unbound devices were
erroneously bound. if such devices are found, they are unbound back.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
We might want to only change a parameter rather than have to set all possible
parameters, so add "partial" commands.
These commands only change the specified parameter.
To avoid duplicating code all around, a unique parser is kept. This parser uses
the .data parameter to select the right behavior.
Signed-off-by: David Marchand <david.marchand@6wind.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Move parser after declarations to prepare rework in next commit.
Signed-off-by: David Marchand <david.marchand@6wind.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Following commit 2d95b84aaa, rte_eth_fc_conf
struct contains a autoneg field that must be set by callers.
Add this parameter to testpmd.
Signed-off-by: David Marchand <david.marchand@6wind.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Valgrind reports this issue:
==29880== Invalid read of size 1
==29880== at 0x56FF9A5: cpu_socket_id (eal_lcore.c:101)
==29880== by 0x56FFAE9: rte_eal_cpu_init (eal_lcore.c:168)
==29880== by 0x56F944A: rte_eal_init (eal.c:975)
The problem is that endptr points to memory allocated underneath the DIR
handle, which has already been freed. So move the closedir() call lower.
Signed-off-by: Aaron Campbell <aaron@arbor.net>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
The function rte_snprintf serves no useful purpose. It is the
same as snprintf() for all valid inputs. Deprecate it and
replace all uses in current code.
Leave the tests for the deprecated function in place.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Mark the rte_log, cmdline_printf and rte_snprintf functions as
being printf-style functions. This causes compilation errors
due to mis-matched parameter types, so the parameter types are
fixed where appropriate.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
When overriding the CC variable on the commandline, the HOSTCC value
was remaining at "gcc", which caused errors on Free BSD10 when using a gcc
as e.g. gcc48, without a binary just called "gcc". This change overrides
HOSTCC when CC is overridden, which means that setting CC=gcc48, will
now also set HOSTCC=gcc48, allowing the "testhost" app to compile on
BSD10.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Using gcc 4.8 on FreeBSD 10, support for SSE4.x is not detected by
the compiler, meaning that the ACL library, which depends on SSE4.2
cannot compile. Disable this library for the native target allows
compiles to succeed on FreeBSD 10 using gcc.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Since commit "add FILE argument to debug functions" (591a9d7985),
application which includes rte_memory.h without stdio.h will be hit
compilation failure:
/path/to/include/rte_memory.h:146:30: error: unknown type name ‘FILE’
void rte_dump_physmem_layout(FILE *f);
Signed-off-by: Hiroshi Shimamoto <h-shimamoto@ct.jp.nec.com>
Reviewed-by: Hayato Momma <h-momma@ce.jp.nec.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Only some devices support the link state interrupt configuration option.
Link state control does not work in virtual drivers
(virtio, vmxnet3, igbvf, and ixgbevf). Instead of having the application
try and guess whether it will work or not provide a driver flag that
can be checked instead.
Note: if device driver doesn't support link state control, what
would happen previously is that the code would never detect link
transitions. This prevents that.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
[Thomas: rename flag]
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
It is now possible to read link status updated by interrupt without
having manual link_update() service provided by the PMD.
Indeed link_update() is useless in interrupt case. So check of this
function pointer must be done in the interrupt case only.
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
Make everything NUMA-related depend on lcore sockets, not device
sockets. This is because the init_mem() function allocates all data
structures based on NUMA nodes of the lcores in the coremask. Therefore,
when no cores are on socket 0, but there are devices on socket 0, it may
lead to segmentation faults.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Some implementations of memmove may make a copy of src before writing to
dst. We avoid that by explicitly writing from src to dst backwards.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
When using key compare function, it uses key length of the first
argument to determine how long should be the keys that are compared.
However, currently we are passing a key from the fragmentation table as
first argument. the problem with this is that this key is potentially
uninitialized (i.e. contains all zeroes, including key length). this
leads to a nasty bug of comparing only the key id's and not keys
themselves.
Of course, a safer way would be to do RTE_MAX between key lengths, but
since this compare is done per-packet, every cycle counts, so we just
use the key whose length is guaranteed to be correct because it comes
from an actual packet.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Fix typos and false assumptions in malloc unit tests.
Without enhancements to lib rte_malloc, malloc autotest fails every
second (2nd) run. With enhancements, malloc autotest fails in
function test_multi_alloc_statistics, because we compare the wrong
sets of statistics.
Signed-off-by: Robert Sanford <rsanford2@gmail.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Problems with lib rte_malloc:
1. Rte_malloc searches a heap's entire free list looking for the best
fit, resulting in linear complexity.
2. Heaps store free blocks in a singly-linked list, resulting in
linear complexity when rte_free needs to remove an adjacent block.
3. The library inserts and removes free blocks with ad hoc, in-line
code, rather than using linked-list functions or macros.
4. The library wastes potential small blocks of size 64 and 128 bytes
(plus overhead of 64 bytes) as padding when reusing free blocks or
resizing allocated blocks.
This patch addresses those problems as follows:
1. Replace single free list with a handful of free lists. Each free
list contains blocks of a specified size range, for example:
list[0]: (0 , 2^8]
list[1]: (2^8 , 2^10]
list[2]: (2^10, 2^12]
list[3]: (2^12, 2^14]
list[4]: (2^14, MAX_SIZE]
When allocating a block, start at the first list that can contain
a big enough block. Search subsequent lists, if necessary.
Terminate the search as soon as we find a block that is big enough.
2. Use doubly-linked lists, so that we can remove free blocks in
constant time.
3. Use BSD LIST macros, as defined in sys/queue.h and the QUEUE(3)
man page.
4. Change code to utilize small blocks of data size 64 and 128, when
splitting larger blocks.
Signed-off-by: Robert Sanford <rsanford2@gmail.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
We usually use testpmd fwd to demonstrate IO forwarding throughput.
For best throughput, it has to assign special parameters to testpmd.
To make it easier to run, now set it as defalut value.
Such parameters are MBUF Mempool Cache and RX/TX threshold registers.
MBCACHE: 250
RX threshold registers: pthresh=8 hthresh=8 wthresh=0
TX threshold registers: pthresh=32 hthresh=0 wthresh=0
Signed-off-by: Cunming Liang <cunming.liang@intel.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Yong Liu <yong.liu@intel.com>
Tested-by: Zhaochen Zhan <zhaochen.zhan@intel.com>
The vpmd RX don't accept burst size less than 32.
As vPMD is set =y by default, while default testpmd burst size is 16.
Which will cause RX nothing if not assign burst size correctly.
Signed-off-by: Cunming Liang <cunming.liang@intel.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Yong Liu <yong.liu@intel.com>
Tested-by: Zhaochen Zhan <zhaochen.zhan@intel.com>
RTE_IXGBE_RX_OLFLAGS_ENABLE gives a hint whick keeping packet type
in RX ol_flags or not.
By default it is set to update ol_flags in RX mbuf header.
If unset it, will gain addtional performance, but will lose packet
type information.
Signed-off-by: Cunming Liang <cunming.liang@intel.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Yong Liu <yong.liu@intel.com>
Tested-by: Zhaochen Zhan <zhaochen.zhan@intel.com>
Vectorized Rx cannot be enabled if RTE_LIBRTE_IEEE1588=y.
So variable dev is not used in this case.
Signed-off-by: Cunming Liang <cunming.liang@intel.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Yong Liu <yong.liu@intel.com>
Tested-by: Zhaochen Zhan <zhaochen.zhan@intel.com>
The vpmd RX routine won't be used if RTE_LIBRTE_IXGBE_RX_ALLOW_BULK_ALLOC
is not defined or its condition check fail.
The case RTE_LIBRTE_IXGBE_RX_ALLOC_BULK_ALLOC=n and RTE_IXGBE_INC_VECTOR=y
may exist when choose to use standard RX and optimized TX
(ixgbe_xmit_pkts_vec/ixgbe_xmit_pkts_simple).
Signed-off-by: Cunming Liang <cunming.liang@intel.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Yong Liu <yong.liu@intel.com>
Tested-by: Zhaochen Zhan <zhaochen.zhan@intel.com>
This was previously solved in commit 60a70d4e04
(fix link status interrupt of bypass device),
but this alternative fix reverts the previous one and solves the
same issue without modifying the base driver (also named "shared code").
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
There is a warning in base driver (shared code) on 32-bits RHEL6.3/6.5:
lib/librte_pmd_i40e/i40e/i40e_lan_hmc.c:917:
error: integer constant is too large for ‘long’ type
As we don't modify base driver, this warning must be ignored.
Signed-off-by: Chen Jing D(Mark) <jing.d.chen@intel.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Acked-by: Helin Zhang <helin.zhang@intel.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Intel PMDs are built on top of the base drivers which are provided by Intel
and shouldn't be modified to allow easy batch upgrade from Intel.
The base driver is a "shared code" between many projects. But in DPDK,
the "base driver" naming makes more sense.
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
i40e supports 16 and 32 bytes RX descriptors which can be configured.
It needs to check the driver type and the configuration to determine
if 16 or 32 bytes RX descriptors is being used, for reading and
displaying the different sizes of RX descriptors.
Signed-off-by: Helin Zhang <helin.zhang@intel.com>
Acked-by: Jing Chen <jing.d.chen@intel.com>
Acked-by: Cunming Liang <cunming.liang@intel.com>
For NVM4.2.2 or after, the firmware has the correct configurations
and load the macvlan filter as expected. It is not needed to
Update the default macvlan filter which cannot be removed at
all during initialization.
Signed-off-by: Helin Zhang <helin.zhang@intel.com>
Acked-by: Jing Chen <jing.d.chen@intel.com>
Acked-by: Cunming Liang <cunming.liang@intel.com>
Double vlan should be specifically disabled by default during
port initialization which is expected.
Signed-off-by: Helin Zhang <helin.zhang@intel.com>
Acked-by: Jing Chen <jing.d.chen@intel.com>
Acked-by: Cunming Liang <cunming.liang@intel.com>
The bit shifting were written wrongly in '0x1 < j',
the correct one should be '0x1 << j'.
Signed-off-by: Helin Zhang <helin.zhang@intel.com>
Acked-by: Jing Chen <jing.d.chen@intel.com>
Acked-by: Cunming Liang <cunming.liang@intel.com>
It wrongly gets the RSS hash result from the RX descriptor which
has been modified for receiving new packet. The fix is to get the
RSS hash result from the buffer which saves the RX descriptor.
Signed-off-by: Helin Zhang <helin.zhang@intel.com>
Acked-by: Jing Chen <jing.d.chen@intel.com>
Acked-by: Cunming Liang <cunming.liang@intel.com>
Fix a bug of copying wrong size of link information in the
function i40evf_get_link_status().
Signed-off-by: Helin Zhang <helin.zhang@intel.com>
Acked-by: Jing Chen <jing.d.chen@intel.com>
Acked-by: Cunming Liang <cunming.liang@intel.com>
i40evf_dev_atomic_read_link_status() was defined but not used.
To avoid possible warnings by some compilers, it needs to
delete the whole function.
Signed-off-by: Helin Zhang <helin.zhang@intel.com>
Acked-by: Jing Chen <jing.d.chen@intel.com>
Acked-by: Cunming Liang <cunming.liang@intel.com>
The compile errors are copied as follows. The fixes came from
Linux drivers of ixgbe-3.21.2 and igb-5.1.2 with modifications.
The idea is to use self-defined functions no matter they have
already been defined somewhere or not.
* Oracle Linux6.4
lib/librte_eal/linuxapp/kni/ethtool/ixgbe/kcompat.h:3111:
error: redefinition of 'ether_addr_equal'
include/linux/etherdevice.h:180: note: previous definition
of 'ether_addr_equal' was here
* RHEL6.5
lib/librte_eal/linuxapp/kni/ethtool/igb/kcompat.h:3597:
error: redefinition of 'mmd_eee_cap_to_ethtool_sup_t'
include/linux/mdio.h:387: note: previous definition of
'mmd_eee_cap_to_ethtool_sup_t' was here
lib/librte_eal/linuxapp/kni/ethtool/igb/kcompat.h:3625:
error: redefinition of 'mmd_eee_adv_to_ethtool_adv_t'
include/linux/mdio.h:415: note: previous definition of
'mmd_eee_adv_to_ethtool_adv_t' was here
lib/librte_eal/linuxapp/kni/ethtool/igb/kcompat.h:3653:
error: redefinition of 'ethtool_adv_to_mmd_eee_adv_t'
include/linux/mdio.h:443: note: previous definition of
'ethtool_adv_to_mmd_eee_adv_t' was here
Signed-off-by: Helin Zhang <helin.zhang@intel.com>
Acked-by: Cunming Liang <cunming.liang@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Cast constant to fix the ICC compilation error
introduced in commit 2b12431b53
(add vlan stripping and insertion to VF)
Signed-off-by: Chen Jing D(Mark) <jing.d.chen@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Take the existing get_pkt API and split out the parts for requesting a new
packet from the part to poll for arrival of a new packet. These individual
functions can then be used independently of the get function, which still acts
as before.
The split functions for request and poll will allow a worker to pull
packets from multiple distributors, or to act as multiple workers with a
single distributor if needed for better performance.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
VF driver add support to configure vlan offload and pvid set/clear
when host driver is DPDK. For Linux driver, it's not supported yet
since it doesn't support from host side.
Signed-off-by: Chen Jing D(Mark) <jing.d.chen@intel.com>
Acked-by: Helin Zhang <helin.zhang@intel.com>
Acked-by: Cunming Liang <cunming.liang@intel.com>
Acked-by: Jingjing Wu <jingjing.wu@intel.com>
Change original 2 functions to common ones that can take effect on
both VF and PF VSIs.
Signed-off-by: Chen Jing D(Mark) <jing.d.chen@intel.com>
Acked-by: Helin Zhang <helin.zhang@intel.com>
Acked-by: Cunming Liang <cunming.liang@intel.com>
Acked-by: Jingjing Wu <jingjing.wu@intel.com>
In old firmware versions, the default mac vlan filter setting is
not the one that we expected, so we'll try to remove it and add
new one to change default setting. After firmware updated, it
change default setting to the one that we expected and don't
allow to remove the setting. In this case, we should add the
perenant mac address into mac list and then return.
Signed-off-by: Chen Jing D(Mark) <jing.d.chen@intel.com>
Acked-by: Helin Zhang <helin.zhang@intel.com>
Acked-by: Cunming Liang <cunming.liang@intel.com>
Acked-by: Jingjing Wu <jingjing.wu@intel.com>
