Enabling 'Extended Tag' and resetting 'Max Read Request Size' in PCI
config space have big impacts to i40e performance. They cannot be
changed on some BIOS implementations, though can on others. Two sys
files of 'extended_tag' and 'max_read_request_size' are added to
support changing them by 'echo' in user space.
Signed-off-by: Helin Zhang <helin.zhang@intel.com>
Signed-off-by: Jing Chen <jing.d.chen@intel.com>
Acked-by: Cunming Liang <cunming.liang@intel.com>
Acked-by: Jijiang Liu <jijiang.liu@intel.com>
Acked-by: Jingjing Wu <jingjing.wu@intel.com>
Acked-by: Heqing Zhu <heqing.zhu@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Unit tests for Packet Framework libraries.
Signed-off-by: Cristian Dumitrescu <cristian.dumitrescu@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked-by: Pablo de Lara Guarch <pablo.de.lara.guarch@intel.com>
Acked by: Ivan Boule <ivan.boule@6wind.com>
The Packet Framework pipeline library provides a standard methodology
(logically similar to OpenFlow) for rapid development of complex packet
processing pipelines out of ports, tables and actions.
A pipeline is constructed by connecting its input ports to its output ports
through a chain of lookup tables. As result of lookup operation into the
current table, one of the table entries (or the default table entry, in case
of lookup miss) is identified to provide the actions to be executed on the
current packet and the associated action meta-data.
The behavior of user actions is defined through the configurable table action
handler, while the reserved actions define the next hop for the current packet
(either another table, an output port or packet drop) and are handled
transparently by the framework.
Signed-off-by: Cristian Dumitrescu <cristian.dumitrescu@intel.com>
Acked-by: Pablo de Lara Guarch <pablo.de.lara.guarch@intel.com>
Acked by: Ivan Boule <ivan.boule@6wind.com>
This file defines the operations to be implemented by
any Packet Framework table.
Signed-off-by: Cristian Dumitrescu <cristian.dumitrescu@intel.com>
Acked-by: Pablo de Lara Guarch <pablo.de.lara.guarch@intel.com>
Acked by: Ivan Boule <ivan.boule@6wind.com>
This file defines the port operations that have to be implemented
by Packet Framework ports.
Signed-off-by: Cristian Dumitrescu <cristian.dumitrescu@intel.com>
Acked-by: Pablo de Lara Guarch <pablo.de.lara.guarch@intel.com>
Acked by: Ivan Boule <ivan.boule@6wind.com>
Removing PCI ID list to make igb_uio more similar to a generic driver
like vfio-pci or pci_uio_generic. This is done to make it easier for
the binding script to support multiple drivers.
Note that since igb_uio no longer has a PCI ID list, it can now be
bound to any device, not just those explicitly supported by DPDK. In
other words, it now behaves similar to PCI stub, VFIO and other generic
PCI drivers.
Therefore to bind a new device to igb_uio, the user will now have to
first write its PCI ID to "new_id" file inside the igb_uio driver
directory, and only then write the PCI ID to "bind". This is reflected
in changes to PCI binding script as well.
There's a weird behaviour of sysfs when a new device ID is added to
new_id. Subsequent writing to "bind" will result in IOError on
closing the file. This error is harmless but it triggers the
exception anyway, so in order to work around that, we check if the
device was actually bound to the driver before raising an error.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Tested-by: HuilongX Xu <huilongx.xu@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Unlike igb_uio, VFIO interrupt type is not set by kernel module
parameters but is set up via ioctl() calls at runtime. This warrants
a new EAL command-line parameter. It will have no effect if VFIO is
not compiled, but will set VFIO interrupt type to either "legacy", "msi"
or "msix" if VFIO support is compiled. Note that VFIO initialization
will fail if the interrupt type selected is not supported by the system.
If the interrupt type parameter wasn't specified, VFIO will try all
interrupt types (starting with MSI-X).
In unit tests, we don't know if VFIO is compiled (eal_vfio.h header is
internal to Linuxapp EAL), so we check this flag regardless.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Add support for binding VFIO devices if RTE_PCI_DRV_NEED_MAPPING is set
for this driver. Try VFIO first, if not mapped then try IGB_UIO too.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Tested-by: HuilongX Xu <huilongx.xu@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Since VFIO cannot be used to map the same device twice, secondary
processes receive the device/group fd's by means of communicating over a
local socket. Only group and container fd's should be sent, as device
fd's can be obtained via ioctl() calls' on the group fd.
For multiprocess, VFIO distinguishes between existing but unused groups
(e.g. grups that aren't bound to VFIO driver) and non-existing groups in
order to know if the secondary process requests a valid group, or if
secondary process requests something that doesn't exist.
VFIO multiprocess sync communicates over a simple protocol. It defines
two requests - request for group fd, and request for container fd.
Possible replies are: SOCKET_OK (an OK signal), SOCKET_ERR (error
signal) and SOCKET_NO_FD (a signal that indicates that the requested
VFIO group is valid, but no fd is present for that group - indicating
that the respective group is simply not bound to VFIO driver).
Here is the logic in a nutshell:
1. secondary process sends SOCKET_REQ_CONTAINER or SOCKET_REQ_GROUP
1a. in case of SOCKET_REQ_GROUP, client also then sends group number
2. primary process receives message
2a. in case of invalid group, SOCKET_ERR is sent back to secondary
2b. in case of unbound group, SOCKET_NO_FD is sent back to secondary
2c. in case of valid group, SOCKET_OK is sent and followed by fd
3. socket is closed
in case of any error, socket is closed and SOCKET_ERR is sent.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Adding code to support VFIO mapping (primary processes only). Most of
the things are done via ioctl() calls on either /dev/vfio/vfio (the
container) or a /dev/vfio/$GROUP_NR (IOMMU group).
In a nutshell, the code does the following:
1. creates a VFIO container (an entity that allows sharing IOMMU DMA
mappings between devices)
2. checks if a given PCI device is a member of an IOMMU group (if it's
not, this indicates that the device isn't bound to VFIO)
3. calls open() the group file to obtain a group fd
4. checks if the group is viable (that is, if all the devices in the
same IOMMU group are either bound to VFIO or not bound to anything)
5. adds the group to a container
6. sets up DMA mappings (only done once, mapping whole DPDK hugepage
memory for DMA, with a 1:1 correspondence of IOVA to PA)
7. gets the actual PCI device fd from the group fd (can fail, which
simply means that this particular device is not bound to VFIO)
8. maps BARs (MSI-X BAR cannot be mmaped, so skipping it)
9. sets up interrupt structures (but not enables them!)
10. enables PCI bus mastering
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Tested-by: HuilongX Xu <huilongx.xu@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Creating code to handle VFIO interrupts in EAL interrupts (supports all
types of interrupts).
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Tested-by: HuilongX Xu <huilongx.xu@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Add VFIO compilation option to linuxapp config.
Adding a header that will determine if VFIO support should be compiled
in. If VFIO is enabled in config (and it's enabled by default), then the
header will also check for kernel version. If VFIO is enabled in config
and if the kernel version is 3.6+, then VFIO_PRESENT will be defined.
This is the macro that should be used to determine if VFIO support is
being compiled in.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Tested-by: HuilongX Xu <huilongx.xu@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Moving interrupt type enum out of igb_uio and renaming it to be more
generic. Such a strange header naming and separation is done mostly to
make coming virtio patches easier to port to dpdk.org tree.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Currently, igb_uio is always compiled. Some Linux distributions may not
want to include igb_uio with DPDK, so we need to make sure that igb_uio
compilation for Linuxapp targets can be optional.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Tested-by: HuilongX Xu <huilongx.xu@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Rename the RTE_PCI_DRV_NEED_IGB_UIO to be more generic.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Currently, EAL does not distinguish between actual failures and expected
initialization errors. E.g. sometimes the driver fails to initialize
because it was not supposed to be initialized in the first place, such
as device not being managed by said driver.
This patch makes EAL fail on actual initialization errors while still
skipping over expected initialization errors.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Separating mapping code and calls to open. This is a preparatory work
for VFIO patch since it'll need to map BARs too but it doesn't use path
in mapped_pci_resource. Also, renaming structs to be more generic.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
This makes it possible to run DPDK without hugepage memory when VFIO
is used, as VFIO uses virtual addresses to set up DMA mappings.
Technically, malloc is just fine, but we want to guarantee that
memory will be page-aligned, so using mmap to be safe.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
eal_hpet.c was renamed to eal_timer.c and, thanks to code changes, does
not need the -Wno-return-type any more.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
This adds the code for a new Intel DPDK library for packet distribution.
The distributor is a component which is designed to pass packets
one-at-a-time to workers, with dynamic load balancing. Using the RSS
field in the mbuf as a tag, the distributor tracks what packet tag is
being processed by what worker and then ensures that no two packets with
the same tag are in-flight simultaneously. Once a tag is not in-flight,
then the next packet with that tag will be sent to the next available
core.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
[Thomas: add doxygen @file comment]
Allows to lookup four IP addresses in an LPM table.
Uses SSE instrincts.
Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked-by: Pablo de Lara Guarch <pablo.de.lara.guarch@intel.com>
This commit removes trailing whitespace from lines in files. Almost all
files are affected, as the BSD license copyright header had trailing
whitespace on 4 lines in it [hence the number of files reporting 8 lines
changed in the diffstat].
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
[Thomas: remove spaces before tabs in libs]
[Thomas: remove more trailing spaces in non-C files]
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Add __rte_unused to
pci_unbind_kernel_driver(struct rte_pci_device *dev)
Signed-off-by: Alan Carew <alan.carew@intel.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Recent change to rte_dump_tailq (commit 591a9d7985c1230652),
which now uses a FILE parameter causes compilation to fail under FreeBSD
and sourced to a missing include of stdio.h.
Errors:
rte_tailq.h: unknown type name 'FILE' void rte_dump_tailq(FILE *f);
rte_memory.h: unknown type name 'FILE' void rte_dump_physmem_layout(FILE *f);
Signed-off-by: Alan Carew <alan.carew@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
The unit of allocated_size is MB, so the change below is made.
Otherwise, it will fail to free memory when available memory is not enough.
Signed-off-by: Jijiang Liu <jijiang.liu@intel.com>
Acked-by: Huawei Xie <huawei.xie@intel.com>
Tested-by: Heng Ding <hengx.ding@intel.com>
Since Linux kernel version 3.13.0,
the xen_create/destroy_contiguous_region() API has been changed,
and the first parameter is physical address in the API.
Signed-off-by: Jijiang Liu <jijiang.liu@intel.com>
Acked-by: Huawei Xie <huawei.xie@intel.com>
Tested-by: Heng Ding <hengx.ding@intel.com>
The patch changes the way of reserving memory in Dom0 driver.
It will reserve memory at installing rte_dom0_mm.ko kernel module
instead of requesting memory dynamically during DPDK application startup.
Meanwhile, now driver requests memory size of 4M once first,
if it failed, and request memory size of 2M once.
The main reasons for these changes are as follows:
First, to reduce the impact of increasing in memory fragment
after system run a long time.
Second, to reduce number of memory segment.
Signed-off-by: Jijiang Liu <jijiang.liu@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
This patch adds API to support queue start and stop functionality for RX/TX.
It allows RX and TX queue is started or stopped one by one, instead of starting
and stopping all of them at the same time.
Signed-off-by: Ouyang Changchun <changchun.ouyang@intel.com>
Tested-by: Waterman Cao <waterman.cao@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Error of "implicit-function-declaration" can be seen when building KNI
kernel module on Linux kernel 3.6.10 platform, as follows.
lib/librte_eal/linuxapp/kni/igb_ethtool.c:
In function igb_get_eee:
lib/librte_eal/linuxapp/kni/igb_ethtool.c:
2441:4: error: implicit declaration of function
mmd_eee_adv_to_ethtool_adv_t
lib/librte_eal/linuxapp/kni/igb_ethtool.c:
In function igb_set_eee:
lib/librte_eal/linuxapp/kni/igb_ethtool.c:
2551:2: error: implicit declaration of function
ethtool_adv_to_mmd_eee_adv_t
The root cause is as follows.
On Fedora 18 with kernel 3.6.10, ETHTOOL_GEEE is defined in Linux
header file of "linux/ethtool.h", while is not defined in most of other
linux kernel versions.
mmd_eee_cap_to_ethtool_sup_t(), mmd_eee_adv_to_ethtool_adv_t() and
ethtool_adv_to_mmd_eee_adv_t() in kcompat.h are disabled by "#if
!defined(ETHTOOL_GEEE) || (RHEL_RELEASE_CODE && RHEL_RELEASE_CODE <=
RHEL_RELEASE_VERSION(6,4))", while are called in igb_get_eee() in
igb_ethtool.c which is enabled by "#ifdef ETHTOOL_GEEE".
Reported-by: Prashant Upadhyaya <prashant.upadhyaya@aricent.com>
Signed-off-by: Helin Zhang <helin.zhang@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
There was an indentation error in commit e57f20e051770861377ff
"make vdev init path generic for both virtual and pci devices"
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Currently, physical device pmds use a separate initalization path
(rte_pmd_init_all) while virtual devices use a constructor registration and
rte_eal_dev_init. Theres no reason to have them be separate. This patch
removes the vdev specific nomenclature from the vdev init path and makes it more
generic for use with all pmds. This is the first step in converting the
physical device pmds to using the same constructor based registration path that
the virtual devices use.
Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Rather than have each driver have to remember to add a constructor to it to make
sure its gets registered properly, wrap that process up in a macro to make
registration a one line affair. This also sets the stage for us to make
registration of vdev pmds and physical pmds a uniform process
Signed-off-by: Neil Horman <nhorman@tuxdriver.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
When doing diagnostic function, it is useful to have a ability
to iterate over all memzones.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
It makes no sense to inline string functions, in fact snprintf
can't be inlined because the function supports variable number of
arguments.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
[Thomas: update includes]
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
The DPDK dump functions are useful for remote debugging of an
applications. But when application runs as a daemon, stdout
is typically routed to /dev/null.
Instead change all these functions to take a stdio FILE * handle
instead. An application can then use open_memstream() to capture
the output.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
[Thomas: fix quota_watermark example]
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Currently, if there is more memory in hugepages than the amount
requested by dpdk application, the memory is allocated by taking as much
memory as possible from each socket, starting from first one.
For example if a system is configured with 8 GB in 2 sockets (4 GB per
socket), and dpdk is requesting only 4GB of memory, all memory will be
taken in socket 0 (that have exactly 4GB of free hugepages) even if some
cores are configured on socket 1, and there are free hugepages on socket
1...
Change this behaviour to allocate memory on all sockets where some cores
are configured, spreading the memory amongst sockets using following
ratio per socket:
N° of cores configured on the socket / Total number of configured cores
* requested memory
If this new algorithm fails, it defaults to previous behaviour.
This algorithm is used when memory amount is specified globally using
-m option. Per socket memory allocation can always be done using
--socket-mem option.
It is implemented only for Linux as BSD part looks not to be ready for NUMA.
Signed-off-by: Didier Pallard <didier.pallard@6wind.com>
Signed-off-by: David Marchand <david.marchand@6wind.com>
Acked-by: Venky Venkatesan <venky.venkatesan@intel.com>
RTE_EAL_UNBIND_PORTS was deprecated in DPDK 1.4.0 and removed in 1.6.0, but the
code was not removed.
The bind/unbind operations should not be handled by the eal.
These operations should be either done outside of dpdk or inside the PMDs
themselves as these are their problems.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Signed-off-by: David Marchand <david.marchand@6wind.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Move RTE_PCI_DRV_FORCE_UNBIND flag handling out of RTE_EAL_UNBIND_PORTS section.
This had nothing to do with RTE_EAL_UNBIND_PORTS anyway.
Signed-off-by: David Marchand <david.marchand@6wind.com>
Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
The pci_switch_module() function should only do what its name tells: unbind pci
devices and rebind them on the specified kernel driver.
Hence, it can not call pci_uio_map_resource().
Call to pci_uio_map_resource() should be moved to rte_eal_pci_probe_one_driver()
so that we can factorize code.
Signed-off-by: David Marchand <david.marchand@6wind.com>
Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
A fd leak happens in pci_map_resource when multiple bars are mapped.
Fix this by closing fd unconditionnally in this function and open the
intr_handle fd in pci_uio_map_resource instead.
Signed-off-by: David Marchand <david.marchand@6wind.com>
Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
virtio-uio does not need eal to map bars from uio device, so remove flag
RTE_PCI_DRV_NEED_IGB_UIO.
Then, move virtio-uio workaround out of generic eal_pci.c for linux
implementation.
Signed-off-by: David Marchand <david.marchand@6wind.com>
Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
bsd implementation lacks check on driver flags, fix this.
Besides, check on BAR0 is not needed and could cause trouble for devices that
have no BAR0.
Signed-off-by: David Marchand <david.marchand@6wind.com>
Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Looking at bsd implementation, we can see that there are some potential mem
leaks in linux implementation. Fix them.
Signed-off-by: David Marchand <david.marchand@6wind.com>
Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
Acked-by: Neil Horman <nhorman@tuxdriver.com>
Some applications reserve hugepages for later use,
but DPDK doesn't take reserved pages into account
when calculating number of available number of hugepages.
This patch adds reading from "resv_hugepages" file
in addition to "free_hugepages".
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Acked-by: Thomas Monjalon <thomas.monjalon@6wind.com>
