Currently vfio DMA map function will fail in case the same memory
segment is mapped twice.
This is too strict, as this is not an error to map the same memory
twice.
Instead, use the kernel return value to detect such state and have the
DMA function to return as successful.
For type1 mapping the kernel driver returns EEXISTS.
For spapr mapping EBUSY is returned since kernel 4.10.
Signed-off-by: Shahaf Shuler <shahafs@mellanox.com>
Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
Acked-by: Gaetan Rivet <gaetan.rivet@6wind.com>
Enable users the option to call rte_vfio_dma_map with request to map
to the default vfio fd.
Signed-off-by: Shahaf Shuler <shahafs@mellanox.com>
Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
Acked-by: Gaetan Rivet <gaetan.rivet@6wind.com>
Running in non-legacy mode on a NUMA-enabled system without libnuma
is unsupported, so explicitly print out a warning when trying to
do so.
Running in legacy mode without libnuma is still supported whether or
not we are running with libnuma support enabled, so also fix init to
allow that scenario.
Cc: stable@dpdk.org
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
The memset size for an IPC message is set incorrectly. Fix it to
cover the entire IPC message.
Fixes: 07dcbfe0101f ("malloc: support multiprocess memory hotplug")
Cc: stable@dpdk.org
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Certain failure paths of rte_fbarray_init() will unlock the
mem area lock without locking it first. Fix this by properly
handling the failures.
Fixes: 5b61c62cfd76 ("fbarray: add internal tailq for mapped areas")
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
rte_fbarray_attach() currently locks its internal
spinlock, but never releases it. Secondary processes
won't even start if there is more than one fbarray
to be attached to - the second rte_fbarray_attach()
would be just stuck.
Fix it by releasing the lock at the end of
rte_fbarray_attach(). I believe this was the original
intention.
Fixes: 5b61c62cfd76 ("fbarray: add internal tailq for mapped areas")
Signed-off-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
Reviewed-by: Gavin Hu <gavin.hu@arm.com>
Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
Currently, there is no support for sharing custom VFIO containers
between multiple processes, but it is not documented.
Document this limitation.
Cc: stable@dpdk.org
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Atomic functions are described in doxygen of the file
lib/librte_eal/common/include/generic/rte_atomic.h
The copies in arch-specific files are redundant
and confuse readers about the genericity of the API.
Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
Acked-by: Shahaf Shuler <shahafs@mellanox.com>
From previous patch description: "to improve performance on PPC64,
use light weight sync instruction instead of sync instruction."
Excerpt from IBM doc [1], section "Memory barrier instructions":
"The second form of the sync instruction is light-weight sync,
or lwsync.
This form is used to control ordering for storage accesses to system
memory only. It does not create a memory barrier for accesses to
device memory."
This patch removes the use of lwsync, so calls to rte_wmb() and
rte_rmb() will provide correct memory barrier to ensure order of
accesses to system memory and device memory.
[1] https://www.ibm.com/developerworks/systems/articles/powerpc.html
Fixes: d23a6bd04d72 ("eal/ppc: fix memory barrier for IBM POWER")
Cc: stable@dpdk.org
Signed-off-by: Dekel Peled <dekelp@mellanox.com>
With nr_overcommit_hugepages > 0 application may be able to allocate
hugepages even when free_hugepages == 0. Take this into account when
counting available hugepages.
Signed-off-by: Michał Mirosław <michal.miroslaw@atendesoftware.pl>
Reviewed-by: Anatoly Burakov <anatoly.burakov@intel.com>
When requesting memory with ``-m`` or ``--socket-mem`` flags,
currently the init will fail if the requested memory amount was
bigger than any one memseg list, even if total amount of
available memory was sufficient.
Fix this by making EAL to attempt to allocate pages multiple
times, until we either fulfill our memory requirements, or run
out of hugepages to allocate.
Bugzilla ID: 95
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Previously, when using non-exact allocation, we were requesting
N pages to be allocated, but allowed the memory subsystem to
allocate less than requested. However, we were still expecting
to see N contigous free pages in the memseg list.
This presents a problem because there is no way to try and
allocate as many pages as possible, even if there isn't
enough contiguous free entries in the list.
To address this, use the new "find biggest" fbarray API's when
allocating non-exact number of pages. This way, we will first
check how many entries in the list are actually available, and
then try to allocate up to that number.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Currently, while there is a way to find total amount of used/free
space in an fbarray, there is no way to find biggest contiguous
chunk. Add such API, as well as unit tests to test this API.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Currently, there are numerous reliability issues with fbarray,
such as:
- There is no way to prevent attaching to overlapping memory
areas
- There is no way to prevent double-detach
- Failed destroy leaves fbarray in an invalid state (fbarray
itself is valid, but its backing memory area is already
detached)
In addition, on FreeBSD, doing mmap() on a file descriptor
does not keep the lock, so we also need to store the fd
in order to keep the lock.
This patch improves upon fbarray to address both of these
issues by adding an internal tailq to track allocated areas
and their respective file descriptors.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
The type of value parameter to rte_service_attr_get
should be uint64_t *, since the attributes
are of type uint64_t.
Fixes: 4d55194d76a4 ("service: add attribute get function")
Signed-off-by: Nikhil Rao <nikhil.rao@intel.com>
Reviewed-by: Gage Eads <gage.eads@intel.com>
Reviewed-by: Rami Rosen <ramirose@gmail.com>
Acked-by: Harry van Haaren <harry.van.haaren@intel.com>
Let all architectures use generic ticketlock implementation.
Signed-off-by: Joyce Kong <joyce.kong@arm.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
The spinlock implementation is unfair, some threads may take locks
aggressively while leaving the other threads starving for long time.
This patch introduces ticketlock which gives each waiting thread a
ticket and they can take the lock one by one. First come, first serviced.
This avoids starvation for too long time and is more predictable.
Suggested-by: Jerin Jacob <jerinj@marvell.com>
Signed-off-by: Joyce Kong <joyce.kong@arm.com>
Reviewed-by: Gavin Hu <gavin.hu@arm.com>
Reviewed-by: Ola Liljedahl <ola.liljedahl@arm.com>
Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
The __sync builtin based implementation generates full memory barriers
('dmb ish') on Arm platforms. Using C11 atomic builtins to generate one way
barriers.
Here is the assembly code of __sync_compare_and_swap builtin.
__sync_bool_compare_and_swap(dst, exp, src);
0x000000000090f1b0 <+16>: e0 07 40 f9 ldr x0, [sp, #8]
0x000000000090f1b4 <+20>: e1 0f 40 79 ldrh w1, [sp, #6]
0x000000000090f1b8 <+24>: e2 0b 40 79 ldrh w2, [sp, #4]
0x000000000090f1bc <+28>: 21 3c 00 12 and w1, w1, #0xffff
0x000000000090f1c0 <+32>: 03 7c 5f 48 ldxrh w3, [x0]
0x000000000090f1c4 <+36>: 7f 00 01 6b cmp w3, w1
0x000000000090f1c8 <+40>: 61 00 00 54 b.ne 0x90f1d4
<rte_atomic16_cmpset+52> // b.any
0x000000000090f1cc <+44>: 02 fc 04 48 stlxrh w4, w2, [x0]
0x000000000090f1d0 <+48>: 84 ff ff 35 cbnz w4, 0x90f1c0
<rte_atomic16_cmpset+32>
0x000000000090f1d4 <+52>: bf 3b 03 d5 dmb ish
0x000000000090f1d8 <+56>: e0 17 9f 1a cset w0, eq // eq = none
The benchmarking results showed constant improvements on all available
platforms:
1. Cavium ThunderX2: 126% performance;
2. Hisilicon 1616: 30%;
3. Qualcomm Falkor: 13%;
4. Marvell ARMADA 8040 with A72 cores on macchiatobin: 3.7%
Here is the example test result on TX2:
$sudo ./build/app/test -l 16-27 -- i
RTE>>spinlock_autotest
*** spinlock_autotest without this patch ***
Test with lock on 12 cores...
Core [16] Cost Time = 53886 us
Core [17] Cost Time = 53605 us
Core [18] Cost Time = 53163 us
Core [19] Cost Time = 49419 us
Core [20] Cost Time = 34317 us
Core [21] Cost Time = 53408 us
Core [22] Cost Time = 53970 us
Core [23] Cost Time = 53930 us
Core [24] Cost Time = 53283 us
Core [25] Cost Time = 51504 us
Core [26] Cost Time = 50718 us
Core [27] Cost Time = 51730 us
Total Cost Time = 612933 us
*** spinlock_autotest with this patch ***
Test with lock on 12 cores...
Core [16] Cost Time = 18808 us
Core [17] Cost Time = 29497 us
Core [18] Cost Time = 29132 us
Core [19] Cost Time = 26150 us
Core [20] Cost Time = 21892 us
Core [21] Cost Time = 24377 us
Core [22] Cost Time = 27211 us
Core [23] Cost Time = 11070 us
Core [24] Cost Time = 29802 us
Core [25] Cost Time = 15793 us
Core [26] Cost Time = 7474 us
Core [27] Cost Time = 29550 us
Total Cost Time = 270756 us
In the tests on ThunderX2, with more cores contending, the performance gain
was even higher, indicating the __atomic implementation scales up better
than __sync.
Fixes: af75078fece3 ("first public release")
Cc: stable@dpdk.org
Signed-off-by: Gavin Hu <gavin.hu@arm.com>
Reviewed-by: Phil Yang <phil.yang@arm.com>
Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
Reviewed-by: Ola Liljedahl <ola.liljedahl@arm.com>
Reviewed-by: Steve Capper <steve.capper@arm.com>
Reviewed-by: Jerin Jacob <jerinj@marvell.com>
Acked-by: Nipun Gupta <nipun.gupta@nxp.com>
Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
When estimating tsc frequency using sleep/gettime round it up to the
nearest multiple of 10Mhz for more accuracy.
Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
Reviewed-by: Keith Wiles <keith.wiles@intel.com>
Add macro to align value to the nearest multiple of the given value,
resultant value might be greater than or less than the first parameter
whichever difference is the lowest.
Update unit test to include the new macro.
Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
use case: if callback is used to receive message form socket,
and the message received is disconnect/error, this callback needs
to be unregistered, but cannot because it is still active.
With this patch it is possible to mark the callback to be
unregistered once the interrupt process is done with this
interrupt source.
Signed-off-by: Jakub Grajciar <jgrajcia@cisco.com>
Commit cdc242f260e7 says:
For Linux kernel 4.0 and newer, the ability to obtain
physical page frame numbers for unprivileged users from
/proc/self/pagemap was removed. Instead, when an IOMMU
is present, simply choose our own DMA addresses instead.
In this case the user still sees error messages, so adjust
the log levels. Later, other checks will ensure that errors
are logged in the appropriate cases.
Fixes: cdc242f260e7 ("eal/linux: support running as unprivileged user")
Cc: stable@dpdk.org
Signed-off-by: Kevin Traynor <ktraynor@redhat.com>
Acked-by: Aaron Conole <aconole@redhat.com>
The documentation for rte_realloc claims that the resized area
will always reside on the same NUMA node. This is not actually
the case - while *resized* area will be on the same NUMA node,
if resizing the area is not possible, then the memory will be
reallocated using rte_malloc(), which can allocate memory on
another NUMA node, depending on which lcore rte_realloc() was
called from and which NUMA nodes have memory available.
Fix the API doc to match the actual code of rte_realloc().
Fixes: af75078fece3 ("first public release")
Cc: stable@dpdk.org
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
DPDK malloc library allows broken programs to work because
the semantics of zmalloc and malloc are the same.
This patch enables a more secure model which will catch
(and crash) programs that reuse memory already freed if
RTE_MALLOC_DEBUG is enabled.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Acked-by: Andrew Rybchenko <arybchenko@solarflare.com>
Reviewed-by: Anatoly Burakov <anatoly.burakov@intel.com>
The version number in the DPDK_VERSION file will never have an offset
that needs to be subtracted, so remove that logic from the version
string generation.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Luca Boccassi <bluca@debian.org>
Since we have the version number in a separate file at the root level,
we should not need to duplicate this in rte_version.h too. Best
approach here is to move the macros for specifying the year/month/etc.
parts from the version header file to the build config file - leaving
the other utility macros for e.g. printing the version string, where they
are.
For "make", this is done by having a little bit of awk parse the version
file and pass the results through to the preprocessor for the config
generation stage.
For "meson", this is done by parsing the version and adding it to the
standard dpdk_conf object.
In both cases, we need to append a large number - in this case "99",
previously 16 in original code - to the version number when we want to do
version number comparisons. Without this, the release version e.g. 19.05.0
will compare as less than it's RC's e.g. 19.05.0-rc4. With it, the
comparison is correct as "19.05.0.99 > 19.05.0-rc4.99".
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Acked-by: Luca Boccassi <bluca@debian.org>
Acked-by: Thomas Monjalon <thomas@monjalon.net>
Currently, rte_realloc will not respect original allocation's
NUMA node when memory cannot be resized, and there is no
NUMA-aware equivalent of rte_realloc. This patch adds such a function.
The new API will ensure that reallocated memory stays on
requested NUMA node, as well as allow moving allocated memory
to a different NUMA node.
Signed-off-by: Tomasz Jozwiak <tomaszx.jozwiak@intel.com>
Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
Rather than using linuxapp and bsdapp everywhere, we can change things to
use the, more readable, terms "linux" and "freebsd" in our build configs.
Rather than renaming the configs we can just duplicate the existing ones
with the new names using symlinks, and use the new names exclusively
internally. ["make showconfigs" also only shows the new names to keep the
list short] The result is that backward compatibility is kept fully but any
new builds or development can be done using the newer names, i.e. both
"make config T=x86_64-native-linuxapp-gcc" and "T=x86_64-native-linux-gcc"
work.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Rename the macro and all instances in DPDK code, but keep a copy of
the old macro defined for legacy code linking against DPDK
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Rename the macro to make things shorter and more comprehensible. For
both meson and make builds, keep the old macro around for backward
compatibility.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
The term "linuxapp" is a legacy one, but just calling the subdirectory
"linux" is just clearer for all concerned.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
The term "bsdapp" is a legacy one, but just calling the subdirectory
"freebsd" is just clearer for all concerned.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
-l and -c options are two ways to select the cores used by DPDK.
Their format differs, but the checks on the selected cores are the same.
Use an intermediate array to separate the specific parsing checks from
the common consistency checks.
The parsing functions now concentrate on validating the passed string
and do nothing more.
We can report all invalid core indexes rather than only the first error.
In the error log message, reporting [0, cfg->lcore_count - 1] as a valid
range is then wrong when the core list is not continuous.
Example on my 8 cpus laptop with core 2 and 6 disabled.
echo 0 > /sys/devices/system/cpu/cpu2/online
echo 0 > /sys/devices/system/cpu/cpu6/online
Before:
./master/app/testpmd -l 0-7 --no-huge -m 512 -- --total-num-mbufs 2048
EAL: Detected 6 lcore(s)
EAL: Detected 1 NUMA nodes
EAL: invalid core list, please check core numbers are in [0, 5] range
...
After:
./master/app/testpmd -l 0-7 --no-huge -m 512 -- --total-num-mbufs 2048
EAL: Detected 6 lcore(s)
EAL: Detected 1 NUMA nodes
EAL: lcore 2 unavailable
EAL: lcore 6 unavailable
EAL: invalid core list, please check specified cores are part of 0-1,3-5,7
...
Fixes: d888cb8b9613 ("eal: add core list input format")
Fixes: b38693b612b4 ("eal: fix core number validation")
Cc: stable@dpdk.org
Signed-off-by: David Marchand <david.marchand@redhat.com>
We don't need to look for trailing spaces.
This is a copy/paste block from eal_parse_coremask().
Remove it and the associated comment.
Fixes: d888cb8b9613 ("eal: add core list input format")
Cc: stable@dpdk.org
Signed-off-by: David Marchand <david.marchand@redhat.com>
Spawning the ctrl threads on anything that is not part of the eal
coremask is not that polite to the rest of the system, especially
when you took good care to pin your processes on cpu resources with
tools like taskset (linux) / cpuset (freebsd).
Rather than introduce yet another eal options to control on which cpu
those ctrl threads are created, let's take the startup cpu affinity
as a reference and remove the eal coremask from it.
If no cpu is left, then we default to the master core.
The cpuset is computed once at init before the original cpu affinity
is lost.
Introduced a RTE_CPU_AND macro to abstract the differences between linux
and freebsd respective macros.
Examples in a 4 cores FreeBSD vm:
$ ./build/app/testpmd -l 2,3 --no-huge --no-pci -m 512 \
-- -i --total-num-mbufs=2048
$ procstat -S 1057
PID TID COMM TDNAME CPU CSID CPU MASK
1057 100131 testpmd - 2 1 2
1057 100140 testpmd eal-intr-thread 1 1 0-1
1057 100141 testpmd rte_mp_handle 1 1 0-1
1057 100142 testpmd lcore-slave-3 3 1 3
$ cpuset -l 1,2,3 ./build/app/testpmd -l 2,3 --no-huge --no-pci -m 512 \
-- -i --total-num-mbufs=2048
$ procstat -S 1061
PID TID COMM TDNAME CPU CSID CPU MASK
1061 100131 testpmd - 2 2 2
1061 100144 testpmd eal-intr-thread 1 2 1
1061 100145 testpmd rte_mp_handle 1 2 1
1061 100147 testpmd lcore-slave-3 3 2 3
$ cpuset -l 2,3 ./build/app/testpmd -l 2,3 --no-huge --no-pci -m 512 \
-- -i --total-num-mbufs=2048
$ procstat -S 1065
PID TID COMM TDNAME CPU CSID CPU MASK
1065 100131 testpmd - 2 2 2
1065 100148 testpmd eal-intr-thread 2 2 2
1065 100149 testpmd rte_mp_handle 2 2 2
1065 100150 testpmd lcore-slave-3 3 2 3
Fixes: d651ee4919cd ("eal: set affinity for control threads")
Cc: stable@dpdk.org
Signed-off-by: David Marchand <david.marchand@redhat.com>
Reviewed-by: Anatoly Burakov <anatoly.burakov@intel.com>
Reviewed-by: Olivier Matz <olivier.matz@6wind.com>
pthread_setaffinity_np returns a >0 value on error.
We could end up letting the ctrl threads on the current process cpu
affinity.
Fixes: d651ee4919cd ("eal: set affinity for control threads")
Cc: stable@dpdk.org
Signed-off-by: David Marchand <david.marchand@redhat.com>
Reviewed-by: Anatoly Burakov <anatoly.burakov@intel.com>
Reviewed-by: Olivier Matz <olivier.matz@6wind.com>
pthread_getaffinity_np returns a >0 value when failing.
This is mainly for the sake of correctness.
The only case where it could fail is when passing an incorrect cpuset
size wrt to the kernel.
Fixes: 2eba8d21f3c9 ("eal: restrict cores auto detection")
Cc: stable@dpdk.org
Signed-off-by: David Marchand <david.marchand@redhat.com>
Reviewed-by: Anatoly Burakov <anatoly.burakov@intel.com>
Reviewed-by: Rami Rosen <ramirose@gmail.com>
The RTE_PMD_DEBUG_TRACE was only enabled for EVENTDEV_DEBUG and
that configuration is now handled by RTE_EDEV_LOG macros.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
Use dependency() instead of manual append to ldflags.
Move libbsd inclusion to librte_eal, so that all other libraries and
PMDs will inherit it.
Signed-off-by: Luca Boccassi <bluca@debian.org>
Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Since compat library is only a single header, we can easily move it into
the EAL common headers instead of tracking it separately. The downside of
this is that it becomes a little more difficult to have any libs that are
built before EAL depend on it. Thankfully, this is not a major problem as
the only library which uses rte_compat.h and is built before EAL (kvargs)
already has the path to the compat.h header file explicitly called out as
an include path.
However, to ensure that we don't hit problems later with this, we can add
EAL common headers folder to the global include list in the meson build
which means that all common headers can be safely used by all libraries, no
matter what their build order.
As a side-effect, this patch also fixes an issue with building on BSD using
meson, due to compat lib no longer needing to be listed as a dependency.
Fixes: a8499f65a1d1 ("log: add missing experimental tag")
Cc: stable@dpdk.org
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Reviewed-by: David Marchand <david.marchand@redhat.com>
Tested-by: David Marchand <david.marchand@redhat.com>
Tested-by: Anatoly Burakov <anatoly.burakov@intel.com>
Acked-by: Thomas Monjalon <thomas@monjalon.net>
Add the strlcat function to DPDK to exist alongside the strlcpy one.
While strncat is generally safe for use for concatenation, the API for the
strlcat function is perhaps a little nicer to use, and supports truncation
detection.
See commit 5364de644a4b ("eal: support strlcpy function") for more
details on the function selection logic, since we only should be using the
DPDK-provided version when no system-provided version is present.
Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
Modern memory mode allowes to not reserve any memory by the
'--socket-mem' option. i.e. it could be possible to specify
zero preallocated memory like '--socket-mem 0'.
Also, it should be possible to configure unlimited memory
allocations by '--socket-limit 0'.
Both cases are impossible now and blocks starting the DPDK
application:
./dpdk-app --socket-limit 0 <...>
EAL: invalid parameters for --socket-limit
EAL: Invalid 'command line' arguments.
Unable to initialize DPDK: Invalid argument
Fixes: 6b42f75632f0 ("eal: enable non-legacy memory mode")
Cc: stable@dpdk.org
Signed-off-by: Ilya Maximets <i.maximets@samsung.com>
Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
It is only possible to know IOMMU type of a given VFIO container
by attempting to initialize it. Since secondary process never
attempts to set up VFIO container itself (because they're shared
between primary and secondary), it never knows which IOMMU type
the container is using, and never sets up the appropriate config
structures. This results in inability to perform DMA mappings in
secondary process.
Fix this by allowing secondary process to query IOMMU type of
primary's default container at device initialization.
Note that this fix is assuming we're only interested in default
container.
Bugzilla ID: 174
Fixes: 6bcb7c95fe14 ("vfio: share default container in multi-process")
Cc: stable@dpdk.org
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com>
Reviewed-by: Darek Stojaczyk <dariusz.stojaczyk@intel.com>
This fixes x86_64-native-linuxapp-clang build with
CONFIG_RTE_FORCE_INTRINSICS=y:
include/generic/rte_atomic.h:218:9: error:
implicit declaration of function '__atomic_exchange_2'
is invalid in C99 [-Werror,-Wimplicit-function-declaration]
include/generic/rte_atomic.h:501:9: error:
implicit declaration of function '__atomic_exchange_4'
is invalid in C99 [-Werror,-Wimplicit-function-declaration]
include/generic/rte_atomic.h:783:9: error:
implicit declaration of function '__atomic_exchange_8'
is invalid in C99 [-Werror,-Wimplicit-function-declaration]
We didn't caught this issue previously on other platforms because
CONFIG_RTE_FORCE_INTRINSICS enabled by default only for armv8.
Fixes: 7bdccb93078e ("eal: fix ARM build with clang")
Cc: stable@dpdk.org
Signed-off-by: Ilya Maximets <i.maximets@samsung.com>
