Nipun Gupta [Fri, 4 May 2018 10:11:25 +0000 (15:41 +0530)]
bus/fslmc: keep Tx queues information for DPCI devices
The DPCI devices have both Tx and Rx queues. Event devices use
DPCI Rx queues only, but CMDIF (AIOP) uses both Tx and Rx queues.
This patch enables Tx queues configuration too.
Declan Doherty [Wed, 2 May 2018 15:59:40 +0000 (16:59 +0100)]
net/ixgbe: fix probe with no devargs
Initialise rte_ethdev_args parameters to zero to handle
the case where no devargs are passed to the IXGBE PF on
device probe, so that there is no invalid attempts to create
representor ports.
Coverity Issue: 277231 Fixes: cf80ba6e2038 ("net/ixgbe: add support for representor ports") Signed-off-by: Declan Doherty <declan.doherty@intel.com> Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
Declan Doherty [Wed, 2 May 2018 15:59:39 +0000 (16:59 +0100)]
net/ixgbe: revert default PF device name
Changes introduced by cf80ba6e2038 modified the default name generated
for the IXGBE PF PMD, this patch reverts the default name to the
original PCI BDBF.
Fixes: cf80ba6e2038 ("net/ixgbe: add support for representor ports") Signed-off-by: Declan Doherty <declan.doherty@intel.com> Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
net/i40e: fix Tx function selection with new offloads
The Tx function selection code in the driver only used the older txq
flags values to check whether the scalar or vector functions should be
used. This caused performance regressions with testpmd io-fwd as the
scalar path rather than the vector one was being used in the default
case. Fix this by changing the code to take account of new offloads and
deleting the defines used for the old ones.
Fixes: 7497d3e2f777 ("net/i40e: convert to new Tx offloads API") Signed-off-by: Bruce Richardson <bruce.richardson@intel.com> Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
Changes introduced by e0cb96204b71 modified the default name generated
for the i40e PF PMD, this patch reverts the default name to the
original PCI BDF.
Fixes: e0cb96204b71 ("net/i40e: add support for representor ports") Signed-off-by: Declan Doherty <declan.doherty@intel.com> Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
Using special types value -1 with mlx4_conv_rss_types() is supposed to
return a supported set of Verbs RSS hash fields, that is, priv->hw_rss_sup
unmodified.
Due to the way this function is written and because it is also used to
initially populate priv->hw_rss_sup however, this special value works
properly only once and fails with ENOTSUP errors afterward.
This problem can be seen when re-creating default flows (e.g. by entering
and leaving isolated mode).
When creation of a flow rule fails during dev_start(), the usage count of
the common RSS context is not decremented, which triggers an assertion
failure in debug mode during dev_close().
This is addressed by tracking the initialization status of the common RSS
context in order to add missing cleanup code.
A similar issue exists in mlx4_rxq_attach(), where usage count is
incremented on a Rx queue but not released in case of error. This may lead
to the above issue since RSS contexts created by flow rules attach
themselves to Rx queues, incrementing their usage count.
Flow director rules matching traffic properties above layer 2 do not
target a fixed hash Rx queue (HASH_RXQ_ETH), it actually depends on the
highest protocol layer specified by each flow rule.
mlx5_fdir_filter_delete() makes this wrong assumption and causes a crash
when attempting to destroy flow rules with L3/L4 specifications.
Anatoly Burakov [Wed, 2 May 2018 15:38:16 +0000 (16:38 +0100)]
malloc: avoid padding elements on page deallocation
Currently, when deallocating pages, malloc will fixup other
elements' headers if there is not enough space to store a full
element in leftover space. This leads to race conditions because
there are some functions that check for pad size with an unlocked
heap, expecting pad size to be constant.
Fix it by being more conservative and only freeing pages when
there is enough space before and after the page to store a free
element.
Jianfeng Tan [Wed, 2 May 2018 13:52:14 +0000 (13:52 +0000)]
eal: fix use-after-free on control thread creation
After below commit, we encounter some strange issue:
1) Dead lock as described here:
http://dpdk.org/ml/archives/dev/2018-April/099806.html
2) SIGSEGV issue when starting a testpmd in VM.
Considering below commit changes to use dynamic memory instead of
stack for memory barrier, we doubt it's caused by use-after-free.
Fixes: 3d09a6e26d8b ("eal: fix threads block on barrier") Reported-by: Maxime Coquelin <maxime.coquelin@redhat.com> Reported-by: Lei Yao <lei.a.yao@intel.com> Suggested-by: Stephen Hemminger <stephen@networkplumber.org> Suggested-by: Olivier Matz <olivier.matz@6wind.com> Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com> Reviewed-by: Olivier Matz <olivier.matz@6wind.com>
Luca Boccassi [Tue, 1 May 2018 13:50:33 +0000 (14:50 +0100)]
igb_uio: pass MODULE_CFLAGS in Kbuild
With the legacy build system MODULE_CFLAGS can be set to pass compiler
flags specific for the kernel modules builds.
This is used currently by Ubuntu and Debian.
Set ccflags-y in the Kbuild to achieve the same result with Meson, and
to keep backward compatbility with older scripts.
Fixes regression in Ubuntu/Debian when the Kbuild is included in the
DKMS source package, as DKMS will pick it up silently by default if
present, causing the MODULE_CFLAGS to be ignored.
Fixes: a52f4574f798 ("igb_uio: build with meson") Cc: stable@dpdk.org Signed-off-by: Luca Boccassi <bluca@debian.org> Acked-by: Bruce Richardson <bruce.richardson@intel.com>
examples/flow_classify: fix validation in port init
The port_init function calls the rte_eth_dev_is_valid_port function.
This function now returns 1 if the port state is attached.
A return value of 1 now means a valid port.
Fixes: a9dbe1802226 ("fix ethdev port id validation") Signed-off-by: Bernard Iremonger <bernard.iremonger@intel.com>
When heap initializes, we need to add already allocated segments
onto the heap. However, in doing that, we never increased total
heap size. Fix it by adding segment length to total heap length
when initializing the heap.
At hugepage info initialization, EAL takes out a write lock on
hugetlbfs directories, and drops it after the memory init is
finished. However, in non-legacy mode, if "-m" or "--socket-mem"
switches are passed, this leads to a deadlock because EAL tries
to allocate pages (and thus take out a write lock on hugedir)
while still holding a separate hugedir write lock in EAL.
Fix it by checking if write lock in hugepage info is active, and
not trying to lock the directory if the hugedir fd is valid.
Fixes: 1a7dc2252f28 ("mem: revert to using flock and add per-segment lockfiles") Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Tested-by: Maxime Coquelin <maxime.coquelin@redhat.com> Tested-by: Shahaf Shuler <shahafs@mellanox.com> Tested-by: Andrew Rybchenko <arybchenko@solarflare.com>
Signed-off-by: Akhil Goyal <akhil.goyal@nxp.com> Acked-by: Pablo de Lara <pablo.de.lara.guarch@intel.com> Acked-by: Thomas Monjalon <thomas@monjalon.net>
Claim maintainership of all areas of EAL memory init, including
OS-specific parts of it.
Also, claim maintainership of fbarray, since although it's not
related to memory allocation, it is heavily used by it and its
primary purpose is to serve memory allocation functions, and
thus will appear under "memory allocation" banner.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Acked-by: Thomas Monjalon <thomas@monjalon.net>
mem: revert to using flock and add per-segment lockfiles
The original implementation used flock() locks, but was later
switched to using fcntl() locks for page locking, because
fcntl() locks allow locking parts of a file, which is useful
for single-file segments mode, where locking the entire file
isn't as useful because we still need to grow and shrink it.
However, according to fcntl()'s Ubuntu manpage [1], semantics of
fcntl() locks have a giant oversight:
This interface follows the completely stupid semantics of System
V and IEEE Std 1003.1-1988 (“POSIX.1”) that require that all
locks associated with a file for a given process are removed
when any file descriptor for that file is closed by that process.
This semantic means that applications must be aware of any files
that a subroutine library may access.
Basically, closing *any* fd with an fcntl() lock (which we do because
we don't want to leak fd's) will drop the lock completely.
So, in this commit, we will be reverting back to using flock() locks
everywhere. However, that still leaves the problem of locking parts
of a memseg list file in single file segments mode, and we will be
solving it with creating separate lock files per each page, and
tracking those with flock().
We will also be removing all of this tailq business and replacing it
with a simple array - saving a few bytes is not worth the extra
hassle of dealing with pointers and potential memory allocation
failures. Also, remove the tailq lock since it is not needed - these
fd lists are per-process, and within a given process, it is always
only one thread handling access to hugetlbfs.
So, first one to allocate a segment will create a lockfile, and put
a shared lock on it. When we're shrinking the page file, we will be
trying to take out a write lock on that lockfile, which would fail if
any other process is holding onto the lockfile as well. This way, we
can know if we can shrink the segment file. Also, if no other locks
are found in the lock list for a given memseg list, the memseg list
fd is automatically closed.
One other thing to note is, according to flock() Ubuntu manpage [2],
upgrading the lock from shared to exclusive is implemented by dropping
and reacquiring the lock, which is not atomic and thus would have
created race conditions. So, on attempting to perform operations in
hugetlbfs, we will take out a writelock on hugetlbfs directory, so
that only one process could perform hugetlbfs operations concurrently.
Currently, memseg lists for secondary process are allocated on
sync (triggered by init), when they are accessed for the first
time. Move this initialization to a separate init stage for
memalloc.
Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Acked-by: Bruce Richardson <bruce.richardson@intel.com>
mem: improve autodetection of hugepage counts on 32-bit
For non-legacy mode, we are preallocating space for hugepages, so
we know in advance which pages we will be able to allocate, and
which we won't. However, the init procedure was using hugepage
counts gathered from sysfs and paid no attention to hugepage
sizes that were actually available for reservation, and failed
on attempts to reserve unavailable pages.
Fix this by limiting total page counts by number of pages
actually preallocated.
Also, VA preallocate procedure only looks at mountpoints that are
available, and expects pages to exist if a mountpoint exists. That
might not necessarily be the case, so also check if there are
hugepages available for a particular page size on a particular
NUMA node.
Previously, if we couldn't preallocate VA space on 32-bit for
one page size, we simply bailed out, even though we could've
tried allocating VA space with other page sizes.
For example, if user had both 1G and 2M pages enabled, and
has asked DPDK to allocate memory on both sockets, DPDK
would've tried to allocate VA space for 1x1G page on both
sockets, failed and never tried again, even though it
could've allocated the same 1G of VA space for 512x2M pages.
Fix this by retrying with different page sizes if VA space
reservation failed.
mem: fix 32-bit memory upper limit for non-legacy mode
32-bit mode has an upper limit on amount of VA space it can preallocate,
but the original implementation used the wrong constant, resulting in
failure to initialize due to integer overflow. Fix it by using the
correct constant.
Previous code checked for both first/last elements being NULL,
but if they weren't, the expectation was that they're both
non-NULL, which will be the case under normal conditions, but
may not be the case due to heap structure corruption.
Coverity issue: 272566 Fixes: bb372060dad4 ("malloc: make heap a doubly-linked list") Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Acked-by: Harry van Haaren <harry.van.haaren@intel.com>
Technically, while the pointer would've been invalid if msl_idx
were invalid, we wouldn't have actually attempted to access the
pointer until verifying the index. Fix it by moving array access
to after we've verified validity of the index.
Coverity issue: 272574 Fixes: 66cc45e293ed ("mem: replace memseg with memseg lists") Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Acked-by: Harry van Haaren <harry.van.haaren@intel.com>
If user has specified a flag to unmap the area right after mapping it,
we were passing an already-unmapped pointer to RTE_LOG. This is not an
issue since RTE_LOG doesn't actually dereference the pointer, but fix
it anyway by moving call to RTE_LOG to before unmap.
Coverity issue: 272584 Fixes: b7cc54187ea4 ("mem: move virtual area function in common directory") Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Coverity reports these lines as having no effect. Technically, we do
want for those lines to have no effect, however they would've likely
been optimized out. Add volatile qualifiers to ensure the code has
effects.
Coverity issue: 272608 Fixes: 582bed1e1d1d ("mem: support mapping hugepages at runtime") Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Previously, if mmap failed to map page address at requested
address, we were attempting to unmap the wrong address. Fix it
by unmapping our actual mapped address, and jump further to
avoid unmapping memory that is not allocated.
Coverity issue: 272602 Fixes: 582bed1e1d1d ("mem: support mapping hugepages at runtime") Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Previous code had an old rebase leftover from the time when
oldpolicy was an actual int, instead of a pointer. Fix it to
do comparison with dereferencing the pointer.
Coverity issue: 272589 Fixes: 582bed1e1d1d ("mem: support mapping hugepages at runtime") Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Acked-by: Bruce Richardson <bruce.richardson@intel.com>
We close fd if we managed to find it in the list of allocated
segment lists (which should always be the case under normal
conditions), but if we didn't, the fd was leaking. Close it if
we couldn't find it in the segment list. This is not an issue
as if the segment is zero length, we're getting rid of it
anyway, so there's no harm in not storing the fd anywhere.
Coverity issue: 272568 Fixes: 2a04139f66b4 ("eal: add single file segments option") Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Acked-by: Bruce Richardson <bruce.richardson@intel.com>
We were closing descriptor before checking if mapping has
failed, but if it did, we did a second close afterwards. Fix
it by moving closing descriptor to after we've done all error
checks.
Jianfeng Tan [Thu, 26 Apr 2018 08:06:53 +0000 (08:06 +0000)]
mem: fix resize return handling for --single-file-segments
resize_hugefile() returns either 0 (which indicates success) or -1
(which indicates failure). We failed to check the success as we
use --single-file-segments option.
Fixes: 2a04139f66b4 ("eal: add single file segments option") Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com> Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
Jianfeng Tan [Fri, 27 Apr 2018 16:41:42 +0000 (16:41 +0000)]
eal: fix threads block on barrier
Below commit introduced pthread barrier for synchronization.
But two IPC threads block on the barrier, and never wake up.
(gdb) bt
#0 futex_wait (private=0, expected=0, futex_word=0x7fffffffcff4)
at ../sysdeps/unix/sysv/linux/futex-internal.h:61
#1 futex_wait_simple (private=0, expected=0, futex_word=0x7fffffffcff4)
at ../sysdeps/nptl/futex-internal.h:135
#2 __pthread_barrier_wait (barrier=0x7fffffffcff0) at pthread_barrier_wait.c:184
#3 rte_thread_init (arg=0x7fffffffcfe0)
at ../dpdk/lib/librte_eal/common/eal_common_thread.c:160
#4 start_thread (arg=0x7ffff6ecf700) at pthread_create.c:333
#5 clone () at ../sysdeps/unix/sysv/linux/x86_64/clone.S:109
Through analysis, we find the barrier defined on the stack could be the
root cause. This patch will change to use heap memory as the barrier.
Fixes: d651ee4919cd ("eal: set affinity for control threads") Signed-off-by: Jianfeng Tan <jianfeng.tan@intel.com> Acked-by: Shreyansh Jain <shreyansh.jain@nxp.com>
bus/dpaa: optimize physical to virtual address search
With Hotplugging memory support, the order of memseg has been changed
from physically contiguous to virtual contiguous. DPAA bus and drivers
depend on PA to VA address conversion for I/O.
This patch creates a list of blocks requested to be pinned to the
DPAA mempool. For searching physical addresses, it is expected that
it would belong to this list (from hardware pool) and hence it is
less expensive than memseg walks. Though, there is a marginal drop
in performance vis-a-vis the legacy mode with physically contiguous
memsegs.
Signed-off-by: Shreyansh Jain <shreyansh.jain@nxp.com> Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
bus/fslmc: optimize physical to virtual address search
With Hotplugging memory support, the order of memseg has been changed
from physically contiguous to virtual contiguous. FSLMC bus and dpaa2
drivers depend on PA to VA address conversion when in Physical
addressing mode.
This patch creates a list of blocks requested to be pinned to the
DPAA2 mempool. For searching physical addresses, it is expected that
it would belong to this list (from hardware pool) and hence it is
less expensive than memseg walks. Though, this has marginal impact on
performance vis-a-vis legacy mode with physically contiguous memsegs.
Signed-off-by: Shreyansh Jain <shreyansh.jain@nxp.com> Signed-off-by: Thomas Monjalon <thomas@monjalon.net>
crypto/dpaa_sec: remove ctx based offset for PA-VA conversion
Crypto requires physical to virtual address conversion for
descriptors. Prior to memory hotplugging this was based on memseg
iteration assuming memsegs are all physical contiguous and using
cached start address fast calculations can be done. This
assumption now stands invalid with memory hotplugging support.
In preparation for supporting hotplugging change to memory,
this patchset removes the optimized pool context stored physical
address offset based PA-VA conversion.
This adversely affects the performance as complete memsegs now need
to be parsed, but a rework containing necessary optimization would be
posted over this.
This patch is to accommodate an experimental feature of mbuf - external
buffer attachment. If mbuf is attached to an external buffer, its ol_flags
will have EXT_ATTACHED_MBUF set. Without enabling/using the feature,
everything remains same.
If PMD delivers Rx packets with non-direct mbuf, ol_flags should not be
overwritten. For mlx5 PMD, if Multi-Packet RQ is enabled, Rx packets could
be carried with externally attached mbufs.
Signed-off-by: Yongseok Koh <yskoh@mellanox.com> Acked-by: Andrew Rybchenko <arybchenko@solarflare.com>
This patch introduces a new way of attaching an external buffer to a mbuf.
Attaching an external buffer is quite similar to mbuf indirection in
replacing buffer addresses and length of a mbuf, but a few differences:
- When an indirect mbuf is attached, refcnt of the direct mbuf would be
2 as long as the direct mbuf itself isn't freed after the attachment.
In such cases, the buffer area of a direct mbuf must be read-only. But
external buffer has its own refcnt and it starts from 1. Unless
multiple mbufs are attached to a mbuf having an external buffer, the
external buffer is writable.
- There's no need to allocate buffer from a mempool. Any buffer can be
attached with appropriate free callback.
- Smaller metadata is required to maintain shared data such as refcnt.
Signed-off-by: Yongseok Koh <yskoh@mellanox.com> Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com> Acked-by: Olivier Matz <olivier.matz@6wind.com> Acked-by: Andrew Rybchenko <arybchenko@solarflare.com>
Xiao Wang [Wed, 25 Apr 2018 02:18:27 +0000 (10:18 +0800)]
vhost: fix vDPA set features
We should call set_features callback after setting features in virtio_net
structure, otherwise vDPA driver cannot get the right features.
Fixes: 07718b4f87aa ("vhost: adapt library for selective datapath") Signed-off-by: Xiao Wang <xiao.w.wang@intel.com> Acked-by: Zhihong Wang <zhihong.wang@intel.com> Reviewed-by: Maxime Coquelin <maxime.coquelin@redhat.com>
While the patch did solve concurrency issue, it induces more
pages copies as some clean pages are marked as dirty for
performance reasons. Moreover, as there is no more contention
doing the logging, the rate of packets than can be processed is
higher, leading to even more pages to be dirtied.
It has been reported that with more than one queue pair, and
with a relatively low packet rate (1Mpps), the live migration
never converges until the flow is stopped.
While a better solution is found, it is better to reset to the
old behaviour, i.e. using atomic operation for dirty pages
logging.
Signed-off-by: Maxime Coquelin <maxime.coquelin@redhat.com>
Matej Vido [Fri, 27 Apr 2018 08:57:05 +0000 (10:57 +0200)]
doc: update doc and release notes for szedata2 driver
New version of the packages with dependencies for the szedata2
driver is needed due to the new API of the libsze2 library which
is used in the driver.
The documentation and the release notes are updated to contain
the information about the required versions.
Signed-off-by: Matej Vido <vido@cesnet.cz> Acked-by: Jan Remes <remes@netcope.com>
Roman Zhukov [Thu, 19 Apr 2018 11:37:03 +0000 (12:37 +0100)]
net/sfc/base: get max supported value for action MARK
The mark value for MATCH_ACTION_MARK has a maximum value.
Requesting a value larger than the maximum will cause the
filter insertion to fail with EINVAL. This patch allows the
driver to check the value at the filter validation.
Signed-off-by: Roman Zhukov <roman.zhukov@oktetlabs.ru> Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
Andrew Rybchenko [Thu, 19 Apr 2018 11:37:00 +0000 (12:37 +0100)]
net/sfc: support flow marks in equal stride super-buffer Rx
Equal stride super-buffer Rx mode allows to mark packets in HW
using filters. Process the data on datapath and advertise
corresponding features to allow flow API support to implement it.
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com>
Andrew Rybchenko [Thu, 19 Apr 2018 11:36:59 +0000 (12:36 +0100)]
net/sfc: add Rx descriptor wait timeout
Add device argument to customize Rx descriptor wait timeout which
is supported in DPDK firmware variant only in equal stride super-buffer
Rx mode only.
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com> Reviewed-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Andrew Rybchenko [Thu, 19 Apr 2018 11:36:58 +0000 (12:36 +0100)]
net/sfc: support DPDK firmware variant
DPDK firmware variant supports equal stride super-buffer Rx mode which
provides higher packet rate and packet marks but requires dedicated
mempool manager with contiguous object block allocation (e.g. bucket).
Also the firmware supports subvariant without checksumming on Tx which
allows to reach higher packet rates on transmit if checksumming is not
required.
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com> Reviewed-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Andrew Rybchenko [Thu, 19 Apr 2018 11:36:55 +0000 (12:36 +0100)]
net/sfc: support equal stride super-buffer Rx mode
HW Rx descriptor represents many contiguous packet buffers which
follow each other. Number of buffers, stride and maximum DMA
length are setup-time configurable per Rx queue based on provided
mempool. The mempool must support contiguous block allocation and
get info API to retrieve number of objects in the block.
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com> Reviewed-by: Ivan Malov <ivan.malov@oktetlabs.ru>
Andrew Rybchenko [Thu, 19 Apr 2018 11:36:53 +0000 (12:36 +0100)]
net/sfc: allow one Rx queue entry carry many packet buffers
One HW Rx descriptor has many packet buffers in the case of equal
stride super-buffer Rx modes. Each packet buffer is still treated
as separate SW Rx descriptor. rxq_entries is the size of HW Rx ring
whereas nb_rx_desc is the number of SW Rx descriptors.
Signed-off-by: Andrew Rybchenko <arybchenko@solarflare.com> Reviewed-by: Ivan Malov <ivan.malov@oktetlabs.ru>
git.droids-corp.org / dpdk.git / log