On PPC64, the ioports are mapped in memory. Implement the missing part
of ioport API for PPC64 when using uio. This may also work on other
architectures but it has not been tested.
Signed-off-by: David Marchand <david.marchand@6wind.com>
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Split pci_parse_sysfs_resource() and introduce
pci_parse_one_sysfs_resource() that parses one line of sysfs resource
file.
This new function will be exported and used in next commits when
mapping the ioports resources.
No functional change.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
In a previous commit, the file used to map the PCI resources changed
from "/dev/uio<x>" to "/sys/bus/pci/devices/<busaddr>/resource", making
the comment wrong. Remove it.
Fixes: 9e67561acd1a ("eal/linux: mmap uio resources using resourceX files")
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
From iopl(2) man page: "This call is mostly for the x86 architecture. On
many other architectures it does not exist or will always return an
error".
This patch removes the call to iopl() in rte_eal_iopl_init() for
architectures other than x86, and always return 0 (success). This was
already done for ARM in
commit 0291476ae364 ("eal/linux: never check iopl for arm")
Next patches will introduce the support of memory mapped IO resources
for architectures != x86.
On BSD, there is nothing to do as open("/dev/io") already does the
proper thing. See man IO(4).
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Fix some typos and add missing comments related to ioports API in
rte_pci.h.
Fixes: 756ce64b1 ("eal: introduce PCI ioport API")
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
The behavior of PKT_RX_VLAN_PKT was not very well defined, resulting in
PMDs not advertising the same flags in similar conditions.
Following discussion in [1], introduce 2 new flags PKT_RX_VLAN_STRIPPED
and PKT_RX_QINQ_STRIPPED that are better defined:
PKT_RX_VLAN_STRIPPED: a vlan has been stripped by the hardware and its
tci is saved in mbuf->vlan_tci. This can only happen if vlan stripping
is enabled in the RX configuration of the PMD.
For now, the old flag PKT_RX_VLAN_PKT is kept but marked as deprecated.
It should be removed from applications and PMDs in a future revision.
This patch also updates the drivers. For PKT_RX_VLAN_PKT:
- e1000, enic, i40e, mlx5, nfp, vmxnet3: done, PKT_RX_VLAN_PKT already
had the same meaning than PKT_RX_VLAN_STRIPPED, minor update is
required.
- fm10k: done, PKT_RX_VLAN_PKT already had the same meaning than
PKT_RX_VLAN_STRIPPED, and vlan stripping is always enabled on fm10k.
- ixgbe: modification done (vector and normal), the old flag was set
when a vlan was recognized, even if vlan stripping was disabled.
- the other drivers do not support vlan stripping.
For PKT_RX_QINQ_PKT, it was only supported on i40e, and the behavior was
already correct, so we can reuse the same bit value for
PKT_RX_QINQ_STRIPPED.
[1] http://dpdk.org/ml/archives/dev/2016-April/037837.html,
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Following the discussion on dpdk-users [1], enhance the API
documentation of rte_eth_tx_burst() to specify that the mbufs have to be
allocated from a pool.
[1] http://dpdk.org/ml/archives/users/2016-June/000618.html
Reported-by: Xiaoban Wu <xiaoban_wu@student.uml.edu>
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
This patch fixes rte_memcpy performance in Haswell and Broadwell for
vhost when copy size larger than 256 bytes.
It is observed that for large copies like 1024/1518 ones, rte_memcpy
suffers high ratio of store buffer full issue which causes pipeline
to stall in scenarios like vhost enqueue. This can be alleviated by
adjusting instruction layout. Note that this issue may not be visible
in micro test.
How to reproduce?
PHY-VM-PHY using vhost/virtio or vhost/virtio loop back, with large
packets like 1024/1518 bytes ones. Make sure packet generation rate
is not the bottleneck if PHY-VM-PHY is used.
Test report: http://dpdk.org/ml/archives/dev/2016-May/039716.html
Signed-off-by: Zhihong Wang <zhihong.wang@intel.com>
Tested-by: Qian Xu <qian.q.xu@intel.com>
The mempool local cache was not initialized properly leading to
undefined behavior in cases where the allocated memory was used
previously and left with data.
Fixes: 213af31e0960 ("mempool: reduce structure size if no cache needed")
Signed-off-by: Sergio Gonzalez Monroy <sergio.gonzalez.monroy@intel.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
Function like 'rte_kni_rx_burst()' keeps
allocating 'MAX_MBUF_BURST_NUM' mbufs to
kni fifo queue unless the queue's capacity
('KNI_FIFO_COUNT_MAX') is reached. So, if
the mempool is under-provisioned, user may
run into "Out of Memory" logs from KNI code.
This commit documents the need to provision
mempool capacity of more than
"2 x KNI_FIFO_COUNT_MAX" for each KNI interface.
Signed-off-by: Alex Wang <alex@awakenetworks.com>
Acked-by: Ferruh Yigit <ferruh.yigit@intel.com>
Acked-by: John McNamara <john.mcnamara@intel.com>
The 'mbufs' alloc/free descriptions for
'rte_kni_tx_burst()' and 'rte_kni_rx_burst()'
should be inverted.
Fixes: 3fc5ca2 ("kni: initial import")
Signed-off-by: Alex Wang <alex@awakenetworks.com>
Acked-by: Ferruh Yigit <ferruh.yigit@intel.com>
This patch is used to add the class_id (class_code,
subclass_code, programming_interface) support for
pci_device probe. With this patch, it will be
flexible for users to probe a class of devices
by class_id.
Signed-off-by: Ziye Yang <ziye.yang@intel.com>
The SYSFS_PCI_DEVICES is a constant that makes the PCI testing
difficult as it points to an absolute path. We remove using this
constant and introducing a function pci_get_sysfs_path that gives
the same value. However, the user can pass a SYSFS_PCI_DEVICES env
variable to override the path. It is now possible to create a fake
sysfs hierarchy for testing.
Signed-off-by: Jan Viktorin <viktorin@rehivetech.com>
This patch adds missing DEPDIRS to avoid any library referring to
symbols they are not linked against.
Signed-off-by: Christian Ehrhardt <christian.ehrhardt@canonical.com>
The libraries rte_mempool and rte_ring are not used in EAL,
except for the ivshmem part (CONFIG_RTE_LIBRTE_IVSHMEM).
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Acked-by: David Marchand <david.marchand@6wind.com>
The log history uses rte_mempool. In order to remove the mempool
dependency in EAL (and improve the build), this feature is deprecated.
The ABI is kept but the behaviour is now voided because it seems this
function was not used. The history can be read from syslog.
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Acked-by: David Marchand <david.marchand@6wind.com>
Removes rte_acl_dump() call from rte_table_acl_build () as it invokes
number of printf messages.
Signed-off-by: Jasvinder Singh <jasvinder.singh@intel.com>
Acked-by: Cristian Dumitrescu <cristian.dumitrescu@intel.com>
Fix an error with computation of physical address of
content descriptor in the symmetric operations session
Fixes: 1703e94ac5ce ("qat: add driver for QuickAssist devices")
Signed-off-by: Arek Kusztal <arkadiuszx.kusztal@intel.com>
Acked-by: Deepak Kumar Jain <deepak.k.jain@intel.com>
Mempool consist of multiple memzones, at least from two of them.
ivshmem assumes mempool and elements are all in same memzone.
Updating code to add all memzones when a mempool added.
Fixes: d1d914ebbc25 ("mempool: allocate in several memory chunks by default")
Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
Partial revert of an earlier ill-conceived "fix".
Adjacent segments can never be considered overlapping because we
are not comparing ends to starts, but rather starts to starts.
Therefore the earlier fix was wrong (plus it also had a typo).
Fixes: d6cf31419e51 ("ivshmem: avoid infinite loop when concatenating segments")
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Tested-by: Ferruh Yigit <ferruh.yigit@intel.com>
Acked-by: Ferruh Yigit <ferruh.yigit@intel.com>
Fix compile error because of Linux API change, 'trans_start' field
removed from 'struct net_device'.
Linux: 9b36627acecd ("net: remove dev->trans_start")
Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com>
The $(comma) variable is not defined in this Makefile, nor in
any included Makefile. Seen while doing a "make clean" on ubuntu:
$ make clean
== Clean lib
== Clean lib/librte_compat
== Clean lib/librte_eal
== Clean lib/librte_eal/common
== Clean lib/librte_eal/linuxapp
== Clean lib/librte_eal/linuxapp/eal
== Clean lib/librte_eal/linuxapp/igb_uio
== Clean lib/librte_eal/linuxapp/kni
tr: missing operand after ‘.-’
Two strings must be given when translating.
Try 'tr --help' for more information.
This commit replaces $(comma) by a ',' character, it's not a problem in
that case since we are inside antiquotes.
Fixes: a09b359daca3 ("kni: fix build on Ubuntu 14.04")
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Acked-by Ferruh Yigit <ferruh.yigit@intel.com>
Removed comparison against $CC in Makefiles as
in cross-compiling mode CC can be a different string
instead of string "gcc"
Suggested-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Clang seems to have a bug with asm inside inline function rte_xabort():
rte_rtm.h:56:15: error: invalid operand for inline asm constraint 'i'
asm volatile(".byte 0xc6,0xf8,%P0" :: "i" (status) : "memory");
^
It is seen only when building with EXTRA_CFLAGS=-O0.
The workaround is to replace the inline function by a macro.
Fixes: ba7468997ea6 ("spinlock: add HTM lock elision for x86")
Signed-off-by: Damjan Marion <damarion@cisco.com>
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
The RTE_ETH_VALID_PORTID_OR_ERR_RET macro is used in some places
to check if a port id is valid or not. This commit makes use of it in
some new parts of the code.
Signed-off-by: Mauricio Vasquez B <mauricio.vasquezbernal@studenti.polito.it>
Signed-off-by: Thomas Monjalon <thomas.monjalon@6wind.com>
Some architectures (ex: Power8) have a cache line size of 128 bytes,
so the drivers should not expect that prefetching the second part of
the mbuf with rte_prefetch0(&m->cacheline1) is valid.
This commit add helpers that can be used by drivers to prefetch the
rx or tx part of the mbuf, whatever the cache line size.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Reviewed-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
In function call rte_mempool_xmem_size, division by expression total_size
which may be zero has undefined behavior.
Coverity issue: 13243
Fixes: 148f963fb532 ("xen: core library changes")
Signed-off-by: Slawomir Mrozowicz <slawomirx.mrozowicz@intel.com>
Acked-by: Olivier Matz <olivier.matz@6wind.com>
Add a new flag to remove the constraint of having physically contiguous
objects inside a mempool.
Add this flag to the log history mempool to start, but we could add
it in most cases where objects are not mbufs.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Check the return value of snprintf to ensure that the name of
the object is not truncated.
By the way, update the test to avoid to trigger an error in
that case.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Now that we can populate a mempool with any virtual memory,
it is easier to introduce a function to populate a mempool
with memory coming from an anonymous mapping, as it's done
in test-pmd.
The next commit will replace test-pmd anonymous mapping by
this function.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Instead of creating the internal ring at mempool creation, do
it when populating the mempool with the first memory chunk. The
objective here is to simplify the change of external handler
when it will be introduced.
For instance, this will be possible:
mp = rte_mempool_create_empty(...)
rte_mempool_set_ext_handler(mp, my_handler)
rte_mempool_populate_default()
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Avoid to have a specific file for that, and remove #ifdefs.
Now that we have introduced a function to populate a mempool
with a virtual area, the support of xen dom0 is much easier.
The only thing we need to do is to convert the guest physical
address into the machine physical address using rte_mem_phy2mch().
This function does nothing when not running xen.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
The conversion from guest physical address to machine physical address
is fast when the caller knows the memseg corresponding to the gpa.
But in case the user does not know this information, just find it
by browsing the segments. This feature will be used by next commit.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Introduce a new function rte_mempool_create_empty()
that allocates a mempool that is not populated.
The functions rte_mempool_create() and rte_mempool_xmem_create()
now make use of it, making their code much easier to read.
Currently, they are the only users of rte_mempool_create_empty()
but the function will be made public in next commits.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Introduce rte_mempool_free() that:
- unlink the mempool from the global list if it is found
- free all the memory chunks using their free callbacks
- free the internal ring
- free the memzone containing the mempool
Currently this function is only used in error cases when
creating a new mempool, but it will be made public later
in the patch series.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Storing the pointer to the memzone instead of the physical address
provides more information than just the physical address: for instance,
the memzone flags.
Moreover, keeping the memzone pointer will allow us to free the mempool
(this is done later in the series).
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Introduce a new function rte_mempool_populate_virt() that is now called
by default when hugepages are not supported. This function populate the
mempool with several physically contiguous chunks whose minimum size is
the page size of the system.
Thanks to this, rte_mempool_create() will work properly in without
hugepages (if the object size is smaller than a page size), and 2
specific workarouds can be removed:
- trailer_size was artificially extended to a page size
- rte_mempool_virt2phy() did not rely on object physical address
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Although the physical address won't be correct in memory segment,
this allows at least to retrieve the physical address using
rte_mem_virt2phy(). Indeed, if the page is not locked, the page
may not be present in physical memory.
With next commit, it allows a mempool to have properly filled physical
addresses when using --no-huge option.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Introduce rte_mempool_populate_default() which allocates
mempool objects in several memzones.
The mempool header is now always allocated in a specific memzone
(not with its objects). Thanks to this modification, we can remove
many specific behavior that was required when hugepages are not
enabled in case we are using rte_mempool_xmem_create().
This change requires to update how kni and mellanox drivers lookup for
mbuf memory. For now, this will only work if there is only one memory
chunk (like today), but we could make use of rte_mempool_mem_iter() to
support more memory chunks.
We can also remove RTE_MEMPOOL_OBJ_NAME that is not required anymore for
the lookup, as memory chunks are referenced by the mempool.
Note that rte_mempool_create() is still broken (it was the case before)
when there is no hugepages support (rte_mempool_create_xmem() has to be
used). This is fixed in next commit.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Update rte_mempool_xmem_size() so that when the page_shift argument is
set to 0, assume that memory is physically contiguous, allowing to
ignore page boundaries. This will be used in the next commits.
By the way, rename the variable 'n' as 'obj_per_page' and avoid the
affectation inside the if().
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Introduce a free callback that is passed to the populate* functions,
which is used when freeing a mempool. This is unused now, but as next
commits will populate the mempool with several chunks of memory, we
need a way to free them properly on error.
Later in the series, we will also introduce a public rte_mempool_free()
and the ability for the user to populate a mempool with its own memory.
For that, we also need a free callback.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
This commit simplifies rte_mempool_xmem_usage().
Since previous commit, the function rte_mempool_xmem_usage() is
now the last user of rte_mempool_obj_mem_iter(). This complex
code can now be moved inside the function. We can get rid of the
callback and do some simplification to make the code more readable.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
In the same model than rte_mempool_obj_iter(), introduce
rte_mempool_mem_iter() to iterate the memory chunks attached
to the mempool.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Do not use paddr table to store the mempool memory chunks.
This will allow to have several chunks with different virtual addresses.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
This commit removes MEMPOOL_IS_CONTIG().
The next commits will change the behavior of the mempool library so that
the objects will never be allocated in the same memzone than the mempool
header. Therefore, there is no reason to keep this macro that would
always return 0.
This macro was only used in app/test.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
Store the physical address of the object in its header. It simplifies
rte_mempool_virt2phy() and prepares the removing of the paddr[] table
in the mempool header.
Signed-off-by: Olivier Matz <olivier.matz@6wind.com>
