Ruifeng Wang [Thu, 18 Jul 2019 06:22:30 +0000 (14:22 +0800)]
lpm: use atomic store to avoid partial update
Compiler could generate non-atomic stores for whole table entry
updating. This may cause incorrect nexthop to be returned, if
the byte with valid flag is updated prior to the byte with nexthop
is updated.
Besides, field by field updating of table entries follow
read-modify-write sequences. The operations are not atomic,
nor efficient. And could cause entries out of synchronization.
Changed to use atomic store to update whole table entry.
Suggested-by: Vladimir Medvedkin <vladimir.medvedkin@intel.com> Suggested-by: Gavin Hu <gavin.hu@arm.com> Signed-off-by: Ruifeng Wang <ruifeng.wang@arm.com> Reviewed-by: Gavin Hu <gavin.hu@arm.com> Acked-by: Vladimir Medvedkin <vladimir.medvedkin@intel.com>
Ruifeng Wang [Thu, 18 Jul 2019 06:22:29 +0000 (14:22 +0800)]
lpm: avoid race conditions for v20
When a tbl8 group is getting attached to a tbl24 entry, lookup
might fail even though the entry is configured in the table.
For ex: consider a LPM table configured with 10.10.10.1/24.
When a new entry 10.10.10.32/28 is being added, a new tbl8
group is allocated and tbl24 entry is changed to point to
the tbl8 group. If the tbl24 entry is written without the tbl8
group entries updated, a lookup on 10.10.10.9 will return
failure.
Correct memory orderings are required to ensure that the
store to tbl24 does not happen before the stores to tbl8 group
entries complete.
Besides, explicit structure alignment is used to address atomic
operation building issue with older version clang.
Suggested-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com> Signed-off-by: Ruifeng Wang <ruifeng.wang@arm.com> Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com> Reviewed-by: Gavin Hu <gavin.hu@arm.com> Acked-by: Vladimir Medvedkin <vladimir.medvedkin@intel.com>
Ruifeng Wang [Thu, 18 Jul 2019 06:22:28 +0000 (14:22 +0800)]
lpm: avoid race conditions for v1604
When a tbl8 group is getting attached to a tbl24 entry, lookup
might fail even though the entry is configured in the table.
For ex: consider a LPM table configured with 10.10.10.1/24.
When a new entry 10.10.10.32/28 is being added, a new tbl8
group is allocated and tbl24 entry is changed to point to
the tbl8 group. If the tbl24 entry is written without the tbl8
group entries updated, a lookup on 10.10.10.9 will return
failure.
Correct memory orderings are required to ensure that the
store to tbl24 does not happen before the stores to tbl8 group
entries complete.
The ordering patches in general have no notable impact on LPM
performance test on both Arm A72 platform and x86 E5 platform.
Signed-off-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com> Signed-off-by: Ruifeng Wang <ruifeng.wang@arm.com> Reviewed-by: Gavin Hu <gavin.hu@arm.com> Acked-by: Vladimir Medvedkin <vladimir.medvedkin@intel.com>
Ruifeng Wang [Thu, 18 Jul 2019 06:22:27 +0000 (14:22 +0800)]
lpm: uninline some functions
Tests showed that the function inlining caused performance drop
on some x86 platforms with the memory ordering patches applied.
By force no-inline functions, the performance was better than
before on x86 and no impact to arm64 platforms.
Besides inlines of other functions are removed to let compiler
to decide whether to inline.
Suggested-by: Vladimir Medvedkin <vladimir.medvedkin@intel.com> Signed-off-by: Ruifeng Wang <ruifeng.wang@arm.com> Reviewed-by: Gavin Hu <gavin.hu@arm.com> Acked-by: Vladimir Medvedkin <vladimir.medvedkin@intel.com>
In general, DPDK libraries to not print error messages to
stdout because that is often redirected to /dev/null for daemons.
This patch changes cfgfile library to use RTE_LOG with its
own type.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
No need to initialize variable if it is immediately overwritten.
It is better style not do unnecessary initialization with modern
tools since it lets compiler and other static checkers detect
uninitialized data.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
Since direct register access is used in npa_lf_aura_op_alloc_bulk()
use __rte_noinline instead of __rte_always_inline to preserve ABI.
Based on the compiler npa_lf_aura_op_alloc_bulk might be inlined
differently which may lead to undefined behaviour due to handcoded
asm.
Fixes: 29893042c29d ("mempool/octeontx2: fix clang build for arm64") Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com> Acked-by: Jerin Jacob <jerinj@marvell.com>
'kni_net_rx_lo_fifo()' can get segmented buffers, using 'pkt_len' for
that case will be wrong and some values can cause buffer overflow
in destination mbuf data.
Fixes: d89a58dfe90b ("kni: support chained mbufs") Cc: stable@dpdk.org Signed-off-by: Ferruh Yigit <ferruh.yigit@intel.com> Acked-by: Stephen Hemminger <stephen@networkplumber.org>
If the timer subsystem is not initialized before rte_timer_manage (for
example) is invoked, a pointer to a shared hugepage memory region will
still be null and dereferenced when it is checked for validity; handle
this case.
Fixes: c0749f7096c7 ("timer: allow management in shared memory") Cc: stable@dpdk.org Signed-off-by: Erik Gabriel Carrillo <erik.g.carrillo@intel.com>
With latest changes l3fwd_simple_forward() blindly set
(PKT_TX_IPV4 | PKT_TX_IP_CKSUM) ol_flags for all IPv4 packets.
Though for un-fragmented packets we also do have to set l3_len
to make HW IP cksum offload to work properly.
That causes HW/PMD to drop or generate invalid packets.
Though for un-fragmented packets we don't need to regenerate
IPv4 cksum, as L3 header is not modified.
Fix by setting ol_flags only when required.
Fixes: 16863bbb4a41 ("examples/ip_fragmentation: enable IP checksum offload") Signed-off-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Marcin Zapolski [Wed, 17 Jul 2019 07:58:05 +0000 (09:58 +0200)]
examples/ip_frag: fix use of ethdev internal device array
Modify ip_fragmentation example app to use rte_eth_info_get instead
of global rte_eth_devices structure.
Apps should not be using internal DPDK data structures directly.
Fixes: 9758b956dcf4 ("examples/ip_fragmentation: fix Tx queues init") Cc: stable@dpdk.org Signed-off-by: Marcin Zapolski <marcinx.a.zapolski@intel.com> Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Marcin Zapolski [Wed, 17 Jul 2019 07:58:04 +0000 (09:58 +0200)]
examples: fix use of ethdev internal device array
Modify l3fwd and related example apps to use locally defined port_conf
instead of global rte_eth_devices which is not a part of public API.
Apps should not be using internal DPDK data structures directly.
Fixes: 1ef9600b2d20 ("examples/l3fwd: convert to ethdev offloads API") Fixes: ba8c103d2455 ("examples/l3fwd-acl: convert to new ethdev offloads API") Fixes: 40df1d7a695d ("examples/l3fwd-power: convert to new ethdev offloads API") Fixes: 43fc038262b8 ("examples/l3fwd-vf: convert to new ethdev offloads API") Fixes: 373149c631fe ("examples/performance-thread: convert to new offloads API") Cc: stable@dpdk.org Signed-off-by: Marcin Zapolski <marcinx.a.zapolski@intel.com>
David Hunt [Tue, 16 Jul 2019 11:11:12 +0000 (12:11 +0100)]
examples/power: fix policy handling for FIFO
While the core frequency scale-up and scale down all works fine with the
fifo-per-core functionality, there was a gap for policy handling. When
creating or destroying a policy, the core associated with the FIFO
needs to be automatically added to the core list, so the policy is
associated with the correct core.
Fixes: 221e7026d521 ("examples/power: add FIFO per core for JSON interface") Signed-off-by: David Hunt <david.hunt@intel.com> Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
David Hunt [Mon, 15 Jul 2019 14:07:04 +0000 (15:07 +0100)]
examples/l3fwd-power: fix busyness number
Current implementation only outputs 3 numbers for busyness, 0, 50 and 100.
Fix this so that the 50% is replaced by a curve, more meaningful.
This can be replaced in each use case by a suitable calculation for
that use case.
Gavin Hu [Tue, 16 Jul 2019 19:23:56 +0000 (03:23 +0800)]
hash: flush rings instead of dequeuing one by one
Within rte_hash_reset, calling a while loop to dequeue one by
one from the ring, while not using them at all, is wasting cycles,
The patch just flush the ring by resetting the indices can save CPU
cycles.
Signed-off-by: Gavin Hu <gavin.hu@arm.com> Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com> Acked-by: Yipeng Wang <yipeng1.wang@intel.com>
When initializing EAL with "-w 0:0.0", this error is blocking:
munmap_chunk(): invalid pointer
ElectricFence reports this root cause:
free(7fffeec25a11): address not from malloc()
Fixes: e67a61614d0b ("bus/fslmc: support device iteration") Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com> Tested-by: Thomas Monjalon <thomas@monjalon.net>
David Hunt [Tue, 16 Jul 2019 11:19:00 +0000 (12:19 +0100)]
examples/power: fix strcpy buffer overrun
replace strcpy with strlcpy to prevent buffer overrun
With fix, attempting to use a VERY lonng vm name results in a nicely
truncated 32 character name rather than a segfault:
Setting VM Name to [sdfdsfsfsdffdsdsasdsadasdakjshd]
Using strlcpy rather than rte_strlcpy, as the rte_ version is only a
fallback.
As well as the fix in main.c, this patch also changes an occurrence of
rte_strlcpy in channel_manager.c and channel_monitor.c to strlcpy.
Fixes: 59287933a0bb ("examples/vm_power: add options to guest app") Cc: stable@dpdk.org Signed-off-by: David Hunt <david.hunt@intel.com> Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
Currently PKT_TX_IP_CKSUM is being set into mbuf->ol_flags during
fragmentation operation implicitly by the library. Because of this,
application is forced to use checksum offload whether it is supported
by platform or not.
Also documentation does not provide any expected value of ol_flags in
returned fragmented mbufs so application will never come to know that which
offloads are enabled. So transmission may be failed for the platforms which
does not support checksum offload.
So removing mentioned flag from the library.
Mentioned change is part of http://patches.dpdk.org/patch/53475.
Changes for reassembly operation is already accepted. This patch set
implements the similar change for fragmentation operation.
Fixes: e29fc44370c2 ("ip_frag: remove IP checkum offload flag") Signed-off-by: Sunil Kumar Kori <skori@marvell.com>
test: add procfs error message for multi-process launch
Currently, if there is no procfs mounted, test application will
fail to run any multiprocess-related autotests (EAL flags etc.)
without a clear explanation as to why this happens.
Add a check specifically for that condition, as well as add a
general stringified error message to rte_panic.
In ppc64le, expanding DMA areas always fail because we cannot remove
a DMA window. As a result, we cannot allocate more than one memseg in
ppc64le. This is because vfio_spapr_dma_mem_map() doesn't unmap all
the mapped DMA before removing the window. This patch fixes this
incorrect behavior.
I also fixed the order of ioctl for unregister and unmap. The ioctl
for unregister sometimes report device busy errors due to the
existence of mapped area.
Signed-off-by: Takeshi Yoshimura <tyos@jp.ibm.com> Acked-by: David Christensen <drc@linux.vnet.ibm.com>
Issue is observed while running 'metrics_autotest' continuously
without quiting. During first execution all test cases pass but
second run onwards first test case fails as library is already
initialized.
To resolve, introduced a new API to deinitialise the library
after all test cases are executed.
Once the library usage is over, it must be deinitialized which
will free the shared memory reserved during initialization.
Observed an issue while running 'metrics_autotest' continuously
without quiting. For the first run 'metrics_autotest' passes
all test cases but second run onwards first test case fails
because metrics library is already initialized during first run.
Gavin Hu [Fri, 12 Jul 2019 09:58:34 +0000 (17:58 +0800)]
config: fix meson build on ThunderX2
Since gcc-8.3(I tried 8.3 and 9.1), the meson build failed on ThunderX2.
It got the following compiling errors:
/usr/lib/gcc/aarch64-linux-gnu/9/include/arm_neon.h:26493:1: error:
inlining failed in call to always_inline ‘vmull_p64’:
target specific option mismatch 26493 | vmull_p64 (poly64_t a, poly64_t b)
Fixes: 7286c9d7234f ("config: add thunderx2 machine") Cc: stable@dpdk.org Signed-off-by: Gavin Hu <gavin.hu@arm.com> Reviewed-by: Phil Yang <phil.yang@arm.com> Reviewed-by: Jingzhao Ni <jingzhao.ni@arm.com> Acked-by: Jerin Jacob <jerinj@marvell.com>
Putting color escape sequences in the log look pretty for the
developer but fails in real world DPDK usage. A real application
will put DPDK log to syslog, and syslog does not handle escape
sequences.
Fixes: dd543124cd93 ("common/octeontx2: add runtime log infra") Signed-off-by: Stephen Hemminger <stephen@networkplumber.org> Acked-by: Jerin Jacob <jerinj@marvell.com>
This patch adds following:
1. 'g_container' variable name is not right way to represent the
FSLMC container. Renaming it to fslmc_container.
2. dynamic selection of IOMMU mode based on run environment
Recent (18.11+), devargs structure was changed and so was DPDK port
usage in applications like OVS. Applications are now allowed to
plug/unplug ports (eth) using APIs (hotplug) based on device
arguments.
This patch enables the plug/unplug function (which are dummy for
FSLMC) and the iterator function for rte_dev_probe() and similar
API support.
Yangchao Zhou [Tue, 25 Jun 2019 15:04:14 +0000 (23:04 +0800)]
kni: fix kernel crash with multi-segments
va2pa depends on the physical address and virtual address offset of
current mbuf. It may get the wrong physical address of next mbuf which
allocated in another hugepage segment.
In rte_mempool_populate_default(), trying to allocate whole block of
contiguous memory could be failed. Then, it would reserve memory in
several memzones that have different physical address and virtual address
offsets. The rte_mempool_populate_default() is used by
rte_pktmbuf_pool_create().
Since kernel 2.6.28 the network subsystem has provided
dev->stats for devices to use statistics handling and is the
default if no ndo_get_stats is provided.
This allow allows for 64 bit (rather than just 32 bit)
statistics with KNI.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org> Acked-by: Ferruh Yigit <ferruh.yigit@intel.com>
netdev_alloc_skb is optimized to any alignment or setup
of skb->dev that is required. The kernel has chosen to not pad
packets on x86 (for many years), because it is faster.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org> Acked-by: Ferruh Yigit <ferruh.yigit@intel.com>
Update platform support of CNF95xx in documentation and
also, update the HW cap based on PCI subsystem id and revision id.
This patch also changes HW capability handling to be based on
PCI Revision ID. PCI Revision ID contains a unique identifier
to identify chip, major and minor revisions.
Signed-off-by: Nithin Dabilpuram <ndabilpuram@marvell.com> Acked-by: Jerin Jacob <jerinj@marvell.com>
The config create function did not store the mem config address in
the shared memconfig structure, so the secondary processes couldn't
map it at the required address.
Fixes: b149a7064261 ("eal/freebsd: add config reattach in secondary process") Cc: stable@dpdk.org Signed-off-by: Anatoly Burakov <anatoly.burakov@intel.com> Acked-by: Bruce Richardson <bruce.richardson@intel.com>
David Hunt [Fri, 12 Jul 2019 09:06:51 +0000 (10:06 +0100)]
examples/power: fix FreeBSD meson lib dependency
Samples apps that depend on the power library should not build if
the power library is not present. So now the following apps will
check to see if it's built, else skip building.
* l3fwd-pwer
* vm_power_manager
* guest_cli
Fixes: e0130788576f ("examples/power: support meson/ninja build") Fixes: 89f0711f9ddf ("examples: build some samples with meson") Cc: stable@dpdk.org Signed-off-by: David Hunt <david.hunt@intel.com> Acked-by: Bruce Richardson <bruce.richardson@intel.com>
The commit db90b4969e2e ("vfio: retry creating sPAPR DMA window")
introduced a build breakage on old Linux. Linux <4.2 does not define ddw in
struct vfio_iommu_spapr_tce_info. Without ddw, we cannot change window size
and so should give up the creation. I just exculuded the retrying code if
ddw is not supported.
Suanming Mou [Wed, 15 May 2019 05:10:39 +0000 (05:10 +0000)]
app/pdump: exit with primary process
The pdump tool works as the secondary process. When the primary process
exits and the residual secondary process keeps running, it will make the
primary process can't start up again. Since the ex-fbarry files are still
attached by the secondary process pdump, the 'new' primary process can't
get these files locked.
The patch is to set up an alarm which runs every 0.5s periodically
to monitor the primary process in the pdump. Once the primary exits,
so will the pdump.
David Hunt [Wed, 10 Jul 2019 15:26:20 +0000 (16:26 +0100)]
examples/l3fwd-power: fix metrics divisions
6 issues caught by Coverity 343465
* Possible divide by zero on 3 lines
* Convert to float then back to int, losing precision on 3 lines
This patch modifies the code so that it only assigns calculated
values if the divisor is > 0, otherwise sets metrics to zero.
Also removes the un-needed round() function.
examples/power: add FIFO per core for JSON interface
This patch implements a separate FIFO for each cpu core to improve the
previous functionality where anyone with access to the FIFO could affect
any core on the system. By using appropriate permissions, FIFO interfaces
can be configured to only affect the particular cores.
Because each FIFO is per core, the following fields have been removed
from the command JSON format: core_list, resource_id, name.
Signed-off-by: Lukasz Krakowiak <lukaszx.krakowiak@intel.com> Signed-off-by: Lukasz Gosiewski <lukaszx.gosiewski@intel.com> Signed-off-by: Marcin Hajkowski <marcinx.hajkowski@intel.com> Tested-by: David Hunt <david.hunt@intel.com> Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
This patch fixes the out-of-bounds coverity issue by removing the
offending line of code at line 107 in rte_flow_classify_parse.c
which is never executed.
Currently, when fbarray is destroyed, the fbarray structure is not
zeroed out, which leads to stale data being there and confusing
secondary process init in legacy mem mode. Fix it by always
memsetting the fbarray to zero when destroying it.
David Marchand [Wed, 10 Jul 2019 12:33:40 +0000 (14:33 +0200)]
vfio: fix interrupts race condition
Populating the eventfd in rte_intr_enable in each request to vfio
triggers a reconfiguration of the interrupt handler on the kernel side.
The problem is that rte_intr_enable is often used to re-enable masked
interrupts from drivers interrupt handlers.
This reconfiguration leaves a window during which a device could send
an interrupt and then the kernel logs this (unsolicited from the kernel
point of view) interrupt:
[158764.159833] do_IRQ: 9.34 No irq handler for vector
VFIO api makes it possible to set the fd at setup time.
Make use of this and then we only need to ask for masking/unmasking
legacy interrupts and we have nothing to do for MSI/MSIX.
"rxtx" interrupts are left untouched but are most likely subject to the
same issue.
Thomas Monjalon [Tue, 9 Jul 2019 13:11:17 +0000 (15:11 +0200)]
doc: fix PDF build
The command "make doc-guides-pdf" is failing because
there are more than 1500 lines in the file MAINTAINERS
which is included in the contributing guide.
We are facing the issue mentioned in this comment:
https://github.com/sphinx-doc/sphinx/issues/3099#issuecomment-256440704
Anyway the file MAINTAINERS is mentioned several times in the guide.
So the "literalinclude" is removed from the guide to fix the build
of the PDF.
Signed-off-by: Thomas Monjalon <thomas@monjalon.net> Tested-by: Ferruh Yigit <ferruh.yigit@intel.com>
The cmdline_parse_ether_addr does not need to be used everywhere
in testpmd. Can use rte_ether_unformat_addr instead.
As an added bonus it eliminates some code for copying.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org> Acked-by: Bernard Iremonger <bernard.iremonger@intel.com>
Now that there is a version of ether_aton in rte_ether, it can
be used by the cmdline ethernet address parser.
Note: ether_aton_r can not be used in cmdline because
the old code would accept either bytes XX:XX:XX:XX:XX:XX
or words XXXX:XXXX:XXXX and we need to keep compatibility.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org> Acked-by: Olivier Matz <olivier.matz@6wind.com>
Use rte_eth_unformat_addr, so that ethdev can be built and work
without the cmdline library. The dependency on cmdline was
an arrangement of convenience anyway.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org> Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
net: add function to convert string to ethernet address
Make a function that can be used in place of eth_aton_r
to convert a string to rte_ether_addr. This function
allows both byte (xx:xx:xx:xx:xx:xx) and word (XXXX:XXXX:XXXX)
format and has the same lack of error handling as the original.
This also allows ethdev to no longer have a hard dependency
on the cmdline library.
Signed-off-by: Stephen Hemminger <stephen@networkplumber.org> Reviewed-by: Andrew Rybchenko <arybchenko@solarflare.com>
Xiaolong Ye [Sun, 7 Jul 2019 15:57:05 +0000 (23:57 +0800)]
net/af_xdp: correct XDP frame size
XSK_UMEM__DEFAULT_FRAME_SIZE has been changed to 4096 in kernel commit 123e8da1d330 (xsk: Change the default frame size to 4096 and allow
controlling it),
but we still need to keep ETH_AF_XDP_FRAME_SIZE as 2048 to fit most
dpdk apps.
Signed-off-by: Xiaolong Ye <xiaolong.ye@intel.com> Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
Xiaoyu Min [Fri, 5 Jul 2019 09:54:26 +0000 (17:54 +0800)]
app/testpmd: match GRE key and present bits
Support matching on GRE key and present bits (C,K,S)
Example testpmd command could be:
testpmd>flow create 0 ingress group 1 pattern eth / ipv4 /
gre / gre_key value is 0x12345678 / end
actions rss queues 1 0 end / mark id 196 / end
Which will match GRE packet with k present bit set and key value is
0x12345678.
Signed-off-by: Xiaoyu Min <jackmin@mellanox.com> Acked-by: Ori Kam <orika@mellanox.com> Acked-by: Adrien Mazarguil <adrien.mazarguil@6wind.com>
Xiaoyu Min [Wed, 3 Jul 2019 07:22:49 +0000 (15:22 +0800)]
net/mlx5: support matching on ICMP/ICMP6
On DV/DR flow engine, MLX5 can match on ICMP/ICMP6's code and type field
via FLEX Parser, which can be enabled by config FW using FLEX Parser
profile 2:
mlxconfig -d <mst device> -y set FLEX_PARSER_PROFILE_ENABLE=2
The testpmd commands could be:
testpmd> flow create 0 ingress pattern eth / ipv4 /
icmp type is 8 code is 0 / end
actions rss queues 0 1 end / end
testpmd> flow create 0 ingress pattern eth / ipv6 /
icmp6 type is 128 code is 0 / end
actions rss queues 0 1 end / end
Signed-off-by: Xiaoyu Min <jackmin@mellanox.com> Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
Eli Britstein [Thu, 4 Jul 2019 11:16:28 +0000 (11:16 +0000)]
net/mlx5: zero out UDP checksum in encapsulation
Mellanox NICs do not support UDP checksum hardware tx offload over IPv6.
This limitation becomes critical for UDP based tunnels like VXLAN.
Beside the UDP checksum validity is required by IPv6 there is an option
in Linux to allow accepting UDP zero sum (see udp6zerocsumrx in iproute2
package).
This patch zeroes out the UDP checksum field for encapsulation headers
in raw encap action.
Signed-off-by: Eli Britstein <elibr@mellanox.com> Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
Since 18.11, it is suggested that driver should release all its private
resources at the dev_close routine. So all resources previously released
in remove routine are now released at the dev_close routine, and the
dev_close routine will be called in driver remove routine in order to
support removing a device without closing its ports.
Above behavior changes are supported by setting RTE_ETH_DEV_CLOSE_REMOVE
flag during probe stage.
Xiao Wang [Wed, 3 Jul 2019 02:53:32 +0000 (10:53 +0800)]
net/fm10k: fix descriptor filling in vector Tx
The shift left operation "pkt->vlan_tci << 16" gets vlan_tci extended
to signed type and may cause invalid descriptor. Also the same issue for
the "data_len" field. This patch fixes it by casting them to uint64_t.
Andy Pei [Mon, 1 Jul 2019 10:36:25 +0000 (18:36 +0800)]
net/ipn3ke: add stats register and clearing
ipn3ke can work on 10G mode and 25G mode.
10G mode and 25G mode has different MAC register address for statistics.
This patch implements statistics registers for 10G mode and 25G mode.
Also implements different stats clearing per mode.
Fixes: c01c748e4ae6 ("net/ipn3ke: add new driver") Cc: stable@dpdk.org Signed-off-by: Andy Pei <andy.pei@intel.com> Acked-by: Rosen Xu <rosen.xu@intel.com>
Andy Pei [Mon, 1 Jul 2019 10:36:26 +0000 (18:36 +0800)]
net/ipn3ke: delete MAC register address mask
Original code is compatible with older device, whose mac register
address is no more than 10 bits. Now we have mac register address
longer than 10 bits, so we just delete the mask here.
Fixes: c01c748e4ae6 ("net/ipn3ke: add new driver") Cc: stable@dpdk.org Signed-off-by: Andy Pei <andy.pei@intel.com> Acked-by: Rosen Xu <rosen.xu@intel.com>
When a hash entry is added, there are 2 sets of stores.
1) The application writes its data to memory (whose address
is provided in rte_hash_add_key_with_hash_data API (or NULL))
2) The rte_hash library writes to its own internal data structures;
key store entry and the hash table.
The only ordering requirement between these 2 is that - store
to the application data must complete before the store to key_index.
There are no ordering requirements between the stores to
key/signature and store to application data. The synchronization
point for application data can be any point between the 'store to
application data' and 'store to the key_index'. So, 'pdata' should not
be a guard variable for the data in hash table. It should be a guard
variable only for the application data written to the memory location
pointed by 'pdata'. Hence, in the lookup functions, 'pdata' can be
loaded after full key comparison succeeds.
The synchronization point for the application data (store-release
to 'pdata' in key store) is changed to be consistent with the order
of loads in lookup function. However, this change is cosmetic and
does not affect the functionality.
Fixes: e605a1d36 ("hash: add lock-free r/w concurrency") Cc: stable@dpdk.org Signed-off-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com> Reviewed-by: Gavin Hu <gavin.hu@arm.com> Tested-by: Ruifeng Wang <ruifeng.wang@arm.com> Acked-by: Yipeng Wang <yipeng1.wang@intel.com>
Relaxed signature comparison is done first. Further ordered loads
are done only if the signature matches. Any false positives are
caught by the full key comparison. This provides performance
benefits as load-acquire is executed only when required.
Fixes: e605a1d36 ("hash: add lock-free r/w concurrency") Cc: stable@dpdk.org Signed-off-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com> Reviewed-by: Gavin Hu <gavin.hu@arm.com> Tested-by: Ruifeng Wang <ruifeng.wang@arm.com> Acked-by: Yipeng Wang <yipeng1.wang@intel.com>
git.droids-corp.org / dpdk.git / log