David Marchand [Wed, 17 Sep 2014 13:46:33 +0000 (15:46 +0200)]
ixgbe: use the right debug macro
- We should not use DEBUGOUT*/DEBUGFUNC macros in pmd code.
These macros come as compat wrappers for base driver.
- We should avoid calling RTE_LOG directly as pmd provides a wrapper for logs.
- Replace some PMD_INIT_LOG(DEBUG, "some_func") with PMD_INIT_FUNC_TRACE().
Signed-off-by: David Marchand <david.marchand@6wind.com> Reviewed-by: Jay Rolette <rolette@infiniteio.com> Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Matthew Hall [Sun, 14 Sep 2014 08:34:46 +0000 (01:34 -0700)]
log: add function to retrieve log level
It is helpful when you want outside code to cooperate with and respect
log levels set in DPDK. Then you can avoid using duplicate incompatible
log code in the DPDK and non-DPDK parts of the app.
Signed-off-by: Matthew Hall <mhall@mhcomputing.net> Acked-by: Bruce Richardson <bruce.richardson@intel.com>
[Thomas: add void to fix function signature]
Bruce Richardson [Thu, 18 Sep 2014 10:55:52 +0000 (11:55 +0100)]
ixgbe: fix build with gcc 4.4
The refcnt field is contained within an anonymous union within the mbuf
data structure, and gcc 4.4 gives an error about an unknown field unless
the initialiser for the field is contained within extra braces.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Bruce Richardson [Mon, 15 Sep 2014 16:20:13 +0000 (17:20 +0100)]
ixgbe: fix perf regression due to moved pool ptr
Adjust the fast-path code to fix the regression caused by the pool
pointer moving to the second cache line. This change adjusts the
prefetching and also the way in which the mbufs are freed back to the
mempool.
Note: slow-path e.g. path supporting jumbo frames, is still slower, but
is dealt with by a later commit
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Bruce Richardson [Thu, 11 Sep 2014 13:15:43 +0000 (14:15 +0100)]
ixgbe: rework vector pmd following mbuf changes
The vector PMD expects fields to be in a specific order so that it can
do vector operations on multiple fields at a time. Following mbuf
rework, adjust driver to take account of the new layout and re-enable it
in the config.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Bruce Richardson [Thu, 11 Sep 2014 13:15:44 +0000 (14:15 +0100)]
mbuf: split mbuf across two cache lines.
This change splits the mbuf in two to move the pool and next pointers to
the second cache line. This frees up 16 bytes in first cache line.
The reason for this change is that we believe that there is no possible
way that we can ever fit all the fields we need to fit into a 64-byte
mbuf, and so we need to start looking at a 128-byte mbuf instead. Examples
of new fields that need to fit in, include -
* 32-bits more for filter information for support for the new filters in
the i40e driver (and possibly other future drivers)
* an additional 2-4 bytes for storing info on a second vlan tag to allow
drivers to support double Vlan/QinQ
* 4-bytes for storing a sequence number to enable out of order packet
processing and subsequent packet reordering
as well as potentially a number of other fields or splitting out fields
that are superimposed over each other right now, e.g. for the qos scheduler.
We also want to allow space for use by other non-Intel NIC drivers that may
be open-sourced to dpdk.org in the future too, where they support fields
and offloads that currently supported hardware doesn't.
If we accept the fact of a 2-cache-line mbuf, then the issue becomes
how to rework things so that we spread our fields over the two
cache lines while causing the lowest slow-down possible. The general
approach that we are looking to take is to focus the first cache
line on fields that are updated on RX , so that receive only deals
with one cache line. The second cache line can be used for application
data and information that will only be used on the TX leg. This would
allow us to work on the first cache line in RX as now, and have the
second cache line being prefetched in the background so that it is
available when necessary. Hardware prefetches should help us out
here. We also may move rarely used, or slow-path RX fields e.g. such
as those for chained mbufs with jumbo frames, to the second
cache line, depending upon the performance impact and bytes savings
achieved.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Bruce Richardson [Thu, 11 Sep 2014 13:15:42 +0000 (14:15 +0100)]
mbuf: add named points inside the structure
Add markers or "labels" at given points inside the mbuf which can be
used instead of individual fields to identify the start of logical
sections inside the mbuf.
The use of typedefs and dummy fields was chosen over using unions
because of a couple reasons:
* unions cause an extra level of indentation (more likely two levels as
a union containing a struct for multiple fields would be needed). This
makes the lines longer than they need to be and increases the need for
wrapping. [This was the main reason]
* with markers, you can apply multiple markers at the same point if
wanted.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Olivier Matz <olivier.matz@6wind.com>
Bruce Richardson [Thu, 11 Sep 2014 13:15:40 +0000 (14:15 +0100)]
mbuf: use macros only to access metadata
Removed the explicit zero-sized metadata definition at the end of the
mbuf data structure. Updated the metadata macros to take account of this
change so that all existing code which uses those macros still works.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Olivier Matz <olivier.matz@6wind.com>
Bruce Richardson [Thu, 11 Sep 2014 13:15:38 +0000 (14:15 +0100)]
mbuf: introduce a flag to indicate a control mbuf
Since the flags field is now 64-bits, we can allow one bit to be used to
indicate a control i.e. non-packet mbuf. Dedicate the high bit (bit 63)
for this purpose and add in a utility macro to test if a given mbuf has
the bit set or not.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Olivier Matz <olivier.matz@6wind.com>
Bruce Richardson [Thu, 11 Sep 2014 13:15:37 +0000 (14:15 +0100)]
mbuf: expand ol_flags field to 64-bits
The offload flags field (ol_flags) was 16-bits and had no further room
for expansion. This patch increases the field size to 64-bits, using up
the remaining reserved space in the single-cache-line mbuf.
NOTE: none of the values for existing flags have been changed, i.e. no
new numbers have been explicitly reserved between existing flag
definitions.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Bruce Richardson [Thu, 11 Sep 2014 13:15:36 +0000 (14:15 +0100)]
mbuf: reorder fields by time of use
* Reorder the fields in the mbuf so that we have fields that are used
together side-by-side in the structure. This means that we have a
contiguous block of 8-bytes in the mbuf which are used to reset an mbuf
of descriptor rearm, and a block of 16-bytes of data (excluding flags)
which are set on RX from the received packet descriptor.
* Use dummy fields as appropriate to ensure alignment or to reserve gaps
for later field additions.
* Place most items which are not used by fast-path RX separately at the end
of the structure so they can later be moved to a separate cache line.
[The l2/l3 length fields are not moved at this stage as doing so will
cause overflow to the next cache line].
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Olivier Matz <olivier.matz@6wind.com>
The mbuf structure already contains a pointer to the beginning of the
buffer (m->buf_addr). It is not needed to use 8 bytes again to store
another pointer to the beginning of the data.
Using a 16 bits unsigned integer is enough as we know that a mbuf is
never longer than 64KB. We gain 6 bytes in the structure thanks to
this modification.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
* Updated to apply to latest on mainline.
* Disabled vector PMD in config as it relies heavily on the mbuf layout
This will be re-enabled in a subsequent commit once vPMD has been
reworked to take account of mbuf changes.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Olivier Matz <olivier.matz@6wind.com>
The vlan_macip structure combined a vlan tag id with l2 and l3 headers
lengths for tracking offloads. However, this structure was only used as
a unit by the e1000 and ixgbe drivers, not generally.
This patch removes the structure from the mbuf header and places the
fields into the mbuf structure directly at the required point, without
any net effect on the structure layout. This allows us to treat the vlan
tags and header length fields as separate for future mbuf changes. The
drivers which were written to use the combined structure still do so,
using a driver-local definition of it.
Reduce perf regression caused by splitting vlan_macip field. This is
done by providing a single uint16_t value to allow writing/clearing
the l2 and l3 lengths together. There is still a small perf hit to the
slow path TX due to the reads from vlan_tci and l2/l3 lengths being
separated. (<5% in my tests with testpmd with no extra params).
Unfortunately, this cannot be eliminated, without restoring the vlan
tags and l2/l3 lengths as a combined 32-bit field. This would prevent
us from ever looking to move those fields about and is an artificial tie
that applies only for performance in igb and ixgbe drivers. Therefore,
this patch keeps the vlan_tci field separate from the lengths as the
best solution going forward.
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 [Thu, 28 Aug 2014 15:42:38 +0000 (16:42 +0100)]
mbuf: rename in_port to just port
In some cases we may want to tag a packet for a particular destination
or output port, so rename the "in_port" field in the mbuf to just "port"
so that it can be re-used for this purpose if an application needs it.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com> Acked-by: Olivier Matz <olivier.matz@6wind.com>
Olivier Matz [Thu, 28 Aug 2014 15:42:37 +0000 (16:42 +0100)]
mbuf: remove the rte_pktmbuf structure
The rte_pktmbuf structure was initially included in the rte_mbuf
structure. This was needed when there was 2 types of mbuf (ctrl and
packet). As the control mbuf has been removed, we can merge the
rte_pktmbuf into the rte_mbuf structure.
Advantages of doing this:
- the access to mbuf fields is easier (ex: m->data instead of m->pkt.data)
- make the structure more consistent: for instance, there was no reason
to have the ol_flags field in rte_mbuf
- it will allow a deeper reorganization of the rte_mbuf structure in the
next commits, allowing to gain several bytes in it
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
[Bruce: updated for latest code and new example apps] Signed-off-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Olivier Matz [Thu, 28 Aug 2014 15:42:36 +0000 (16:42 +0100)]
mbuf: remove rte_ctrlmbuf
The initial role of rte_ctrlmbuf is to carry generic messages (data
pointer + data length) but it's not used by the DPDK or it applications.
Keeping it implies:
- loosing 1 byte in the rte_mbuf structure
- having some dead code rte_mbuf.[ch]
This patch removes this feature. Thanks to it, it is now possible to
simplify the rte_mbuf structure by merging the rte_pktmbuf structure
in it. This is done in next commit.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
* Updated patch to HEAD.
* Modified patch to retain the old function names for ctrl mbufs as
macros. This helps with app compatibility, and allows the concept
of a control mbuf to be reintroduced via a single-bit flag in
a future change.
* Updated the packet framework ip_pipeline example application to
work following this change.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com> Acked-by: Olivier Matz <olivier.matz@6wind.com>
Olivier Matz [Thu, 28 Aug 2014 15:42:35 +0000 (16:42 +0100)]
mbuf: rename RTE_MBUF_SCATTER_GATHER into RTE_MBUF_REFCNT
It seems that RTE_MBUF_SCATTER_GATHER is not the proper name for the
feature it provides. "Scatter gather" means that data is stored using
several buffers. RTE_MBUF_REFCNT seems to be a better name for that
feature as it provides a reference counter for mbufs.
The macro RTE_MBUF_SCATTER_GATHER is poisoned to ensure this
modification is seen by drivers or applications using it.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com> Signed-off-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Bruce Richardson [Thu, 28 Aug 2014 15:42:34 +0000 (16:42 +0100)]
ixgbe: keep only non-zero initializer in mbuf definition
Since all unspecified fields in an initializer are assumed to be zero we
can simplify the empty mbuf definition in the vector driver to only use
the fields that are non-zero, i.e. just nb_segs = 1. This makes things
shorter and means that the structure doesn't need as many updates for
other fields being renamed or moved.
The variable itself is never modified and only used by a single function
so it can be made const and local to the using function.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com> Acked-by: Olivier Matz <olivier.matz@6wind.com>
- pci_num_vf() is already defined in RHEL 6
- pci_intx_mask_supported is already defined in RHEL 6.3
- pci_check_and_mask_intx is already defined in RHEL 6.3
Signed-off-by: Guillaume Gaudonville <guillaume.gaudonville@6wind.com> Signed-off-by: David Marchand <david.marchand@6wind.com> Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Thomas Monjalon [Wed, 3 Sep 2014 01:44:39 +0000 (03:44 +0200)]
igb_uio: revert MSI IRQ mode
This reverts commit 399a3f0db8b0c58
"fix IRQ mode handling"
and part of commit 4a5c221f9d9bf641
"fix compability on old kernel"
MSI implementation is using irq_to_desc which is not exported before
kernel 3.4 and commit 3911ff30.
Let's revert it for release 1.7.1, waiting for another solution.
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Make ACL library to build/work on 'default' architecture:
- make rte_acl_classify_scalar really scalar
(make sure it wouldn't use sse4 instrincts through resolve_priority()).
- Provide two versions of rte_acl_classify code path:
rte_acl_classify_sse() - could be build and used only on systems with sse4.2
and upper, return -ENOTSUP on lower arch.
rte_acl_classify_scalar() - a slower version, but could be build and used
on all systems.
- Addition of a new function rte_acl_classify_alg. This function lets you
specify an enum value to override the acl contexts default algorithm when doing
a classification. This allows an application to specify a classification
algorithm without needing to publicize each method. I know there was concern
over keeping those methods public, but we don't have a static ABI at the moment,
so this seems to me a reasonable thing to do, as it gives us less of an ABI
surface to worry about.
- keep common code shared between these two codepaths.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com> Acked-by: Neil Horman <nhorman@tuxdriver.com>
app/testpmd: fix crash in txonly mode and with tx_first
This crash was believed fixed by commit 5886ae07d211e4b5e49806dd183812,
but the actual issue is that the core ID provided to rte_lcore_to_socket_id()
is wrong. It must be looked up in fwd_lcores_cpuids[].
Signed-off-by: Adrien Mazarguil <adrien.mazarguil@6wind.com> Signed-off-by: David Marchand <david.marchand@6wind.com>
Aaro Koskinen [Mon, 18 Aug 2014 12:44:59 +0000 (15:44 +0300)]
kni: fix build with kernel 3.17
Since Linux commit "set name_assign_type in alloc_netdev" (c835a677331495),
the function alloc_netdev takes a new parameter (name_assign_type)
whose default value is NET_NAME_UNKNOWN.
Signed-off-by: Aaro Koskinen <aaro.koskinen@nsn.com> Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Bruce Richardson [Wed, 27 Aug 2014 15:51:17 +0000 (16:51 +0100)]
ixgbe: make vector stores unaligned
When writing to the mbuf array for receiving packets, do not assume
16-byte alignment by using aligned stores. If the pointers are only
8-byte aligned, the program will crash due to incorrect alignment.
Changing "store" to "storeu" fixes this.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
David Marchand [Mon, 18 Aug 2014 11:29:25 +0000 (13:29 +0200)]
app/test: only build what has been selected in config
Avoid building tests if their counterparts are not selected in config.
This has the nice side effect of fixing build errors when disabling parts of
the dpdk.
Signed-off-by: David Marchand <david.marchand@6wind.com> Acked-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Neil Horman <nhorman@tuxdriver.com>
David Marchand [Mon, 18 Aug 2014 11:29:24 +0000 (13:29 +0200)]
app/test: no more builtin commands
The builtin commands list is now empty, we can remove it.
Signed-off-by: David Marchand <david.marchand@6wind.com> Acked-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Neil Horman <nhorman@tuxdriver.com>
David Marchand [Mon, 18 Aug 2014 11:29:23 +0000 (13:29 +0200)]
app/test: convert all tests to register system
Remove all tests from the builtin commands list and use the dynamic commands
list register macro.
Signed-off-by: David Marchand <david.marchand@6wind.com> Acked-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Neil Horman <nhorman@tuxdriver.com>
David Marchand [Mon, 18 Aug 2014 11:29:22 +0000 (13:29 +0200)]
app/test: introduce dynamic commands list
This patch adds a way to dynamically add a test without modifying commands.c.
Move all current tests in a builtin_commands[] list, next patch will convert
them to this new system.
Signed-off-by: David Marchand <david.marchand@6wind.com> Acked-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Neil Horman <nhorman@tuxdriver.com>
Signed-off-by: David Marchand <david.marchand@6wind.com> Acked-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Neil Horman <nhorman@tuxdriver.com>
David Marchand [Mon, 18 Aug 2014 11:29:20 +0000 (13:29 +0200)]
app/test: use accessor to set refcnt field
refcnt field can not be accessed directly as it depends on the RTE_MBUF_REFCNT
build option.
Signed-off-by: David Marchand <david.marchand@6wind.com> Acked-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Neil Horman <nhorman@tuxdriver.com>
David Marchand [Mon, 18 Aug 2014 11:29:19 +0000 (13:29 +0200)]
app: simplify makefiles
No need to test some build option multiple times in a Makefile.
Besides, such option is needed by the associated app, so move it at the
top of the Makefile.
Signed-off-by: David Marchand <david.marchand@6wind.com> Acked-by: Bruce Richardson <bruce.richardson@intel.com> Acked-by: Neil Horman <nhorman@tuxdriver.com>
Ouyang Changchun [Fri, 15 Aug 2014 04:58:01 +0000 (12:58 +0800)]
examples/vhost: support jumbo frame
This patch support mergeable RX feature and thus support jumbo frame RX and TX
in user space vhost(as virtio backend).
On RX, it secures enough room from vring to accommodate one complete scattered
packet which is received by PMD from physical port, and then copy data from
mbuf to vring buffer, possibly across a few vring entries and descriptors.
On TX, it gets a jumbo frame, possibly described by a few vring descriptors which
are chained together with the flags of 'NEXT', and then copy them into one scattered
packet and TX it to physical port through PMD.
Add per-queue RX/TX start/stop function.
Support fields start_rx_per_q and start_tx_per_q.
In the meanwhile, change dev_start/stop to call per-queue RX/TX functions.
Add functions to start/stop specific RX/TX queue.
Support fields start_rx_per_q and start_tx_per_q.
In the meanwhile, change dev_start/stop functions to call per-queue functions.
Ding Heng [Fri, 15 Aug 2014 01:47:32 +0000 (09:47 +0800)]
i40e: enable multicast for promiscuous mode
IPv6 will run NDP with multicast packets, but multicast packets will be
filtered by i40e driver by default. So we need to enable multicast when
promiscuous mode is on, or IPv6 will fail.
Helin Zhang [Thu, 14 Aug 2014 06:12:53 +0000 (14:12 +0800)]
i40e: support xen domain0
i40e was failing to run in XEN domain0, as the physical
memory for adminq DMA should be allocated and translated
in a different way for XEN domain0. So
rte_memzone_reserve_bounded() should be used for DMA
memory allocation, and rte_mem_phy2mch() should be used
for DMA memory address translation to support running
i40e PMD in XEN domain0.
Pawel Wodkowski [Wed, 20 Aug 2014 09:42:53 +0000 (10:42 +0100)]
kni: fix build on Ubuntu 12.04
On Ubuntu 12.04.4 file '/proc/version_signature' contains
'Ubuntu 3.11.0-15.25~precise1-generic 3.11.10'. This introduce compilation
error since '~precise1' will not be discarded. This patch discards
everything after '~' inclusively.
Signed-off-by: Pawel Wodkowski <pawelx.wodkowski@intel.com> Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Neil Horman [Fri, 1 Aug 2014 16:48:57 +0000 (12:48 -0400)]
ixgbe: require only sse3 intrinsics
ixgbe was failing to build in the default configuration because it required
sse4.2 intrinsics, and the default config doesn't support more than sse3.
Modify the pmd so that only sse3 intrinsics are pulled in and used.
Signed-off-by: Neil Horman <nhorman@tuxdriver.com> CC: Konstantin Ananyev <konstantin.ananyev@intel.com> CC: Thomas Monjalon <thomas.monjalon@6wind.com> Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Pablo de Lara [Tue, 5 Aug 2014 10:18:20 +0000 (11:18 +0100)]
pcap: fix Rx crash
Normally, bufs[i] stores the mbuf pointer, the index of buf[i]
is the loop count i, but if header.len > buf_size, DPDK will
free the mbuf, but the loop count i still increases, so some
of the items in bufs[] might be NULL pointer, causing a potential
DPDK core. Using num_rx as the index for bufs[] solves the problem.
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com> Tested-by: Jiajia SunX <sunx.jiajia@intel.com> Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
examples/l3fwd: improve grouping by destination port
Latest changes introduced a small degradation for the corner case
when each input packet is destined to the different port.
For the test-case when 1 core manages 4 ports and packet stream looks like:
IPV4_DSTPORT0, IPV4_DSTPORT1, IPV4_DSTPORT3, IPV4_DSTPORT4, IPV4_DSTPORT0, ...
non-optimised code path outperforms optimised one by 2-3%.
These changes supposed to close that gap.
From my testing: now for the case described above optimised code path
produces same numbers as non-optimised one.
For other test-cases numbers remain about the same.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
- Fix bonding unit test suite which was failing due to a change
in pmd configuration behaviour introduced in commit a130f531187249a88 (add link state interrupt flag)
- Added fixes to allow the ability to re-run test suite from test
application without restarting application
Signed-off-by: Declan Doherty <declan.doherty@intel.com> Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com>
Recent Ubuntu kernel 3.13.0-30.54, although based on Linux kernel 3.13.11,
already provides skb_set_hash() inline function, slightly different than
the one provided by lib/librte_eal/linuxapp/kni/ethtool/igb/kcompat.h
As a result, the implementation provided by kcompat.h must be skipped.
It is not appropriate to test whether LINUX_VERSION_CODE >= KERNEL_VERSION(3,13,11)
because previous Ubuntu kernel 3.13.0-29.53, already based on 3.13.11, needs to
get the implementation provided by kcompat.h
So the full Ubuntu kernel version numbering scheme must be tested:
<base kernel version>-<ABI number>.<upload number>-<flavour>
See "What does a specific Ubuntu kernel version number mean?"
and "How can we determine the version of the running kernel?"
at: https://wiki.ubuntu.com/Kernel/FAQ
Unlike RHEL_RELEASE_CODE, there is no such UBUNTU_RELEASE_CODE available out of
the box, so it needs to be crafted from the Makefile
Similarly, UBUNTU_KERNEL_CODE is generated with ABI and upload numbers.
`lsb_release -si` is first used to check whether we are running Ubuntu
`lsb_release -sr` provides release number 14.04, then converted to integer 1404
/proc/version_signature is parsed to get base kernel version, ABI and upload
numbers, and flavour is dropped
UBUNTU_KERNEL_CODE is indirectly defined using the UBUNTU_KERNEL_VERSION macro,
which in turn is defined in kcompat.h
This makes a single place to define the Ubuntu kernel version numbering scheme,
which is slightly different than the usual "shift by 8" scheme: ABI numbers can
be big (see: https://wiki.ubuntu.com/Kernel/Dev/TopicBranches), so 16-bits have
been reserved for them.
Finally, the implementaion of skb_set_hash is skipped in kcompat.h if we are
running Ubuntu 14.04 with an Ubuntu kernel >= 3.13.0-30.54
Signed-off-by: Patrice Buriez <patrice.buriez@intel.com>
[Thomas: simpler form, use tr instead of subst] Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Fix a couple of issues with my earlier igb_uio stuff:
1. With MSI (like MSI-X) actual IRQ number is not known until
after the pci_enable_msi() is done.
2. If INTX fails, fall back to running without IRQ.
This allows usermode PCI to recover and run without out IRQ
for cases where PCI INTX support is broken (aka VMWare).
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org> Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Add more compatibility wrappers, and split out all the wrapper
code to a separate file. Builds on Debian Squeeze (2.6.32) which
is oldest version of kernel current DPDK supports.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org> Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Anatoly Burakov [Fri, 20 Jun 2014 15:42:25 +0000 (16:42 +0100)]
acl: make tailq fully local
Since the data structures such as rings are shared in their entirety,
those TAILQ pointers are shared as well. Meaning that, after a
successful rte_ring creation, the tailq_next pointer of the last
ring in the TAILQ will be updated with a pointer to a ring which may
not be present in the address space of another process (i.e. a ring
that may be host-local or guest-local, and not shared over IVSHMEM).
Any successive ring create/lookup on the other side of IVSHMEM will
result in trying to dereference an invalid pointer.
This patchset fixes this problem by creating a default tailq entry
that may be used by any data structure that chooses to use TAILQs.
This default TAILQ entry will consist of a tailq_next/tailq_prev
pointers, and an opaque pointer to arbitrary data. All TAILQ
pointers from data structures themselves will be removed and
replaced by those generic TAILQ entries, thus fixing the problem
of potentially exposing local address space to shared structures.
Technically, only rte_ring structure require modification, because
IVSHMEM is only using memzones (which aren't in TAILQs) and rings,
but for consistency's sake other TAILQ-based data structures were
adapted as well.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Anatoly Burakov [Fri, 20 Jun 2014 15:42:23 +0000 (16:42 +0100)]
lpm: make tailq fully local
Since the data structures such as rings are shared in their entirety,
those TAILQ pointers are shared as well. Meaning that, after a
successful rte_ring creation, the tailq_next pointer of the last
ring in the TAILQ will be updated with a pointer to a ring which may
not be present in the address space of another process (i.e. a ring
that may be host-local or guest-local, and not shared over IVSHMEM).
Any successive ring create/lookup on the other side of IVSHMEM will
result in trying to dereference an invalid pointer.
This patchset fixes this problem by creating a default tailq entry
that may be used by any data structure that chooses to use TAILQs.
This default TAILQ entry will consist of a tailq_next/tailq_prev
pointers, and an opaque pointer to arbitrary data. All TAILQ
pointers from data structures themselves will be removed and
replaced by those generic TAILQ entries, thus fixing the problem
of potentially exposing local address space to shared structures.
Technically, only rte_ring structure require modification, because
IVSHMEM is only using memzones (which aren't in TAILQs) and rings,
but for consistency's sake other TAILQ-based data structures were
adapted as well.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Anatoly Burakov [Fri, 20 Jun 2014 15:42:20 +0000 (16:42 +0100)]
hash: make tailq fully local
Since the data structures such as rings are shared in their entirety,
those TAILQ pointers are shared as well. Meaning that, after a
successful rte_ring creation, the tailq_next pointer of the last
ring in the TAILQ will be updated with a pointer to a ring which may
not be present in the address space of another process (i.e. a ring
that may be host-local or guest-local, and not shared over IVSHMEM).
Any successive ring create/lookup on the other side of IVSHMEM will
result in trying to dereference an invalid pointer.
This patchset fixes this problem by creating a default tailq entry
that may be used by any data structure that chooses to use TAILQs.
This default TAILQ entry will consist of a tailq_next/tailq_prev
pointers, and an opaque pointer to arbitrary data. All TAILQ
pointers from data structures themselves will be removed and
replaced by those generic TAILQ entries, thus fixing the problem
of potentially exposing local address space to shared structures.
Technically, only rte_ring structure require modification, because
IVSHMEM is only using memzones (which aren't in TAILQs) and rings,
but for consistency's sake other TAILQ-based data structures were
adapted as well.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Anatoly Burakov [Fri, 20 Jun 2014 15:42:22 +0000 (16:42 +0100)]
mempool: make tailq fully local
Since the data structures such as rings are shared in their entirety,
those TAILQ pointers are shared as well. Meaning that, after a
successful rte_ring creation, the tailq_next pointer of the last
ring in the TAILQ will be updated with a pointer to a ring which may
not be present in the address space of another process (i.e. a ring
that may be host-local or guest-local, and not shared over IVSHMEM).
Any successive ring create/lookup on the other side of IVSHMEM will
result in trying to dereference an invalid pointer.
This patchset fixes this problem by creating a default tailq entry
that may be used by any data structure that chooses to use TAILQs.
This default TAILQ entry will consist of a tailq_next/tailq_prev
pointers, and an opaque pointer to arbitrary data. All TAILQ
pointers from data structures themselves will be removed and
replaced by those generic TAILQ entries, thus fixing the problem
of potentially exposing local address space to shared structures.
Technically, only rte_ring structure require modification, because
IVSHMEM is only using memzones (which aren't in TAILQs) and rings,
but for consistency's sake other TAILQ-based data structures were
adapted as well.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Anatoly Burakov [Fri, 20 Jun 2014 15:42:19 +0000 (16:42 +0100)]
ring: make tailq fully local
Since the data structures such as rings are shared in their entirety,
those TAILQ pointers are shared as well. Meaning that, after a
successful rte_ring creation, the tailq_next pointer of the last
ring in the TAILQ will be updated with a pointer to a ring which may
not be present in the address space of another process (i.e. a ring
that may be host-local or guest-local, and not shared over IVSHMEM).
Any successive ring create/lookup on the other side of IVSHMEM will
result in trying to dereference an invalid pointer.
This patchset fixes this problem by creating a default tailq entry
that may be used by any data structure that chooses to use TAILQs.
This default TAILQ entry will consist of a tailq_next/tailq_prev
pointers, and an opaque pointer to arbitrary data. All TAILQ
pointers from data structures themselves will be removed and
replaced by those generic TAILQ entries, thus fixing the problem
of potentially exposing local address space to shared structures.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Anatoly Burakov [Fri, 20 Jun 2014 15:42:17 +0000 (16:42 +0100)]
eal: use --base-virtaddr for mapping rte_config as well
Use --base-virtaddr to set the address of rte_config file along with
start address of the hugepages. Since the user would likely expect
the hugepages to be starting at the specified address, the specified
address will likely be rounded to either 2M or 1G. So, in order to
not waste space, we subtract the length of the config (and align it
on page boundary) from the base virtual address and map the config
just before the hugepages.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Anatoly Burakov [Fri, 20 Jun 2014 15:42:16 +0000 (16:42 +0100)]
eal: map shared config into exact same address as primary process
Shared config is shared across primary and secondary processes.
However,when using rte_malloc, the malloc elements keep references to
the heap inside themselves. This heap reference might not be referencing
a local heap because the heap reference points to the heap of whatever
process has allocated that malloc element. Therefore, there can be
situations when malloc elements in a given heap actually reference
different addresses for the same heap - depending on which process has
allocated the element. This can lead to segmentation faults when dealing
with malloc elements allocated on the same heap by different processes.
To fix this problem, heaps will now have the same addresses across
processes. In order to achieve that, a new field in a shared mem_config
(a structure that holds the heaps, and which is shared across processes)
was added to keep the address of where this config is mapped in the
primary process.
Secondary process will now map the config in two stages - first, it'll
map it into an arbitrary address and read the address the primary
process has allocated for the shared config. Then, the config is
unmapped and re-mapped using the address previously read.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Pablo de Lara [Wed, 9 Jul 2014 14:35:36 +0000 (15:35 +0100)]
ring: simplify unit tests
As this unit test does not create devices anymore,
and uses devices created by EAL option --vdev,
there were unnecesary tests that were repeated,
so they have been removed.
So now there are three tests:
1 - Test two devices that share a ring, one device
with just one RX queue and the other with one
TX queue.
2 - Test a device connected to itself (loopback) by
a ring, with both RX and TX queue.
3 - Test two devices that share a ring, but both devices
with RX and TX queue, so they can send packets to themselves
and to the other device.
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com> Acked-by: Neil Horman <nhorman@tuxdriver.com>
Pablo de Lara [Wed, 9 Jul 2014 14:35:35 +0000 (15:35 +0100)]
ring: remove extra devices creation with --vdev option
When passing extra arguments in EAL option --vdev, to create
ring ethdevs, API was creating three ethdevs, even if there
was just one argument, such as CREATE.
Signed-off-by: Pablo de Lara <pablo.de.lara.guarch@intel.com> Acked-by: Neil Horman <nhorman@tuxdriver.com>
The adapter struct is just a wrapper around the vmxnet3_hw
structure. Eliminate the wrapper and get rid of the macro
used to access and needlessly cast the private data.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org> Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Update per-queue statistics and add missing multicast into statistics.
Also, no need to zero statistics since they are already cleared
in rte_stats_get.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org> Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
This driver has lots of functions marked always inline which is actually
counterproductive with modern compilers. Better to move the functions to
the one file they are used (proper scope) and let compiler decide.
For trivial functions leave them as static inline.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org> Acked-by: Alan Carew <alan.carew@intel.com>
virtio_net_hdr_mem member within virtqueue structure stores a
physical address and is defined as void ptr. When 32bit pmd is used
with 64bit kernel this leads to truncation of 64bit physical address
and pkt i/o does not work.
Changed virtio_net_hdr_mem to phys_addr_t type and
removed the typecasts
Signed-off-by: Vijayakumar Muthuvel Manickam <mmvijay@gmail.com> Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Alan Carew [Mon, 16 Jun 2014 11:10:26 +0000 (12:10 +0100)]
virtio: fix device specific header offset when MSI-X is disabled
Suggested-by: Neil Horman <nhorman@tuxdriver.com> Suggested-by: Stephen Hemminger <stephen@networkplumber.org> Signed-off-by: Alan Carew <alan.carew@intel.com>
Thomas Monjalon [Tue, 22 Jul 2014 12:27:55 +0000 (14:27 +0200)]
igb_uio: fix build with kernel older than 3.3
Since Linux commit fb51ccbf217 (PCI: Rework config space blocking services),
the functions pci_(un)block_user_cfg_access are replaced by
pci_cfg_access_(un)lock.
The compatibility with older functions was broken since commit 399a3f0db8b0
(igb_uio: fix IRQ mode handling).
Reported-by: Yerden Zhumabekov <e_zhumabekov@sts.kz> Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
[Thomas: isolate MSI code from other patch and don't set info.irq twice] Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
The biggest code change is to use the standard INTX management
code that exists in more recent kernels (and provide backport version).
This also fixes the pci_lock code which was broken, since it was
not protecting against config access, and was doing trylock.
Make this driver behave like other Linux drivers.
Start at MSI-X and degrade to less desireable modes
automatically if the desired type is not available.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org> Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Since only one MSI-X entry is ever defined, there is no need to
put it as an array in the driver private data structure. One msix_entry
can just be put on the stack and initialized there.
Also remove the unused backport defines related to MSI-X.
I suspect this code was just inherited from some other project and
never cleaned up.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org> Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
It is good practice to propogate the return values of failing
functions so that more information can be reported. The failed result
of probe will make it out to errno and get printed by modprobe
and will aid in diagnosis of failures.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org> Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
git.droids-corp.org / dpdk.git / log