kvargs: fix crash when parsing an invalid token on freebsd
The behavior of strtok_r() is not the same between GNU libc and FreeBSD
libc: in the first case, the context is set to "" when the last token is
returned, while in the second case it is set to NULL.
On FreeBSD, the current code crashes because we are dereferencing a NULL
pointer (ctx1). Fix it by first checking if it is NULL. This works with
both GNU and FreeBSD libc.
David Marchand [Mon, 27 Apr 2020 13:23:41 +0000 (15:23 +0200)]
ethdev: fix build warning on 64-bit value
Building OVS with dpdk, sparse complains about 64-bit constant being
passed as a normal integer that can't fit it:
error: constant 0xffffffffffffffff is so big it is unsigned long
Fixes: ecbc8570131d ("ethdev: add PFCP header to flow API") Signed-off-by: David Marchand <david.marchand@redhat.com> Reviewed-by: Andrew Rybchenko <arybchenko@solarflare.com> Acked-by: Thomas Monjalon <thomas@monjalon.net>
David Marchand [Mon, 27 Apr 2020 13:23:39 +0000 (15:23 +0200)]
ring: fix build with -Wswitch-enum
Some popular vswitch implementation might use a gcc option that
complains about missing enums in switch statements.
Fix this by listing all possible values.
Fixes: 664ff4b1729b ("ring: introduce peek style API") Signed-off-by: David Marchand <david.marchand@redhat.com> Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Currently, shutdown for l3fwd-power application is all over the place
and may or may not happen either in the signal handler or in the main()
function. Fix this so that the signal handler will only set the exit
variable, thereby allowing all of the loops to end properly and proceed
to deinitialize everything.
examples/l3fwd-power: exit on power lib init failure
Currently, if power library initialization fails, only a log message is
displayed. This is suboptimal for a number of reasons, but the main one
is that telemetry mode does not depend on the power library and can
therefore run in environments where l3fwd-power would normally not run
correctly (such as inside a VM). This will lead to attempts to
deinitialize the power library on exit, with a subsequent forced unclean
shutdown of DPDK.
Fix this by only initializing the power library in modes that actually
need it, and change a log message to a failure to initialize.
Phil Yang [Sun, 26 Apr 2020 14:45:15 +0000 (22:45 +0800)]
timer: relax barrier for status update
Volatile has no ordering semantics. The rte_timer structure defines
timer status as a volatile variable and uses the rte_r/wmb barrier
to guarantee inter-thread visibility.
This patch optimized the volatile operation with c11 atomic operations
and one-way barrier to save the performance penalty. According to the
timer_perf_autotest benchmarking results, this patch can uplift 10%~16%
timer appending performance, 3%~20% timer resetting performance and 45%
timer callbacks scheduling performance on aarch64 and no loss in
performance for x86.
Suggested-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com> Signed-off-by: Phil Yang <phil.yang@arm.com> Reviewed-by: Gavin Hu <gavin.hu@arm.com> Acked-by: Erik Gabriel Carrillo <erik.g.carrillo@intel.com>
Kalesh AP [Thu, 23 Apr 2020 15:02:24 +0000 (20:32 +0530)]
net/bnxt: fix VNIC Rx queue count on VNIC free
bnxt_free_one_vnic and bnxt_setup_one_vnic are called on configuring
port vlan stripping. bnxt_setup_one_vnic keeps incrementing the
vnic rx_queue_cnt. Fix to reset vnic rx_queue_cnt in bnxt_free_one_vnic.
During port 0 rxq 1 start ie queue start,
bnxt_free_hwrm_rx_ring() we are clearing the pointers to mbuf array.
Due to this we overwrite the queue with fresh mbuf allocations
causing previously allocated mbufs to leak.
Add a check before allocating mbuf to replenish only empty mbuf slots
in the RxQ.
Luca Boccassi [Thu, 12 Mar 2020 16:30:27 +0000 (16:30 +0000)]
usertools: check for pci.ids in /usr/share/misc
Debian and Ubuntu switched years ago from /usr/share/hwdata to
/usr/share/misc, and the former is just a compat symlink now.
We are starting to get bug reports to nudge us into changing.
So check the new path first, and the old one as a fallback.
Cc: stable@dpdk.org Signed-off-by: Luca Boccassi <luca.boccassi@microsoft.com> Acked-by: David Marchand <david.marchand@redhat.com>
rte_timer_subsystem_initialized is a global variable that can be
accessed by multiple processes simultaneously. Hence, any access
to rte_timer_subsystem_initialized should be protected by
rte_mcfg_timer_lock.
Fixes: f9d6cd8bfe9e ("timer: fix resource leak in finalize") Cc: stable@dpdk.org Signed-off-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com> Reviewed-by: Gavin Hu <gavin.hu@arm.com> Reviewed-by: Phil Yang <phil.yang@arm.com> Acked-by: Erik Gabriel Carrillo <erik.g.carrillo@intel.com>
ipsec: use hash lookup with signatures in SAD lookup
Change hash function from jhash to crc.
Precalculate hash signatures for a bulk of keys and then
use rte_hash_lookup_with_hash_bulk_data() to speed up sad lookup
Also use rte_hash_add_key_with_hash and _del_key_with_hash with
precalculated hash signature for a key in rte_ipsec_sad_add and
rte_ipsec_sad_del
Signed-off-by: Vladimir Medvedkin <vladimir.medvedkin@intel.com> Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com>
Implement rte_hash_lookup_with_hash_bulk_data() and
rte_hash_lookup_with_hash_bulk() - bulk lookup
functions with precomputed hash signatures.
Add these two functions into performance tests.
Signed-off-by: Vladimir Medvedkin <vladimir.medvedkin@intel.com> Acked-by: Yipeng Wang <yipeng1.wang@intel.com>
All recent POWER systems, Power 8 and 9 specifically, support an IOMMU
(it can't be disabled). The functionality of the IOMMU is different
depending on whether it's running on a bare metal PowerNV system or in
a virtual environment (PowerVM LPAR or KVM/QEMU). DPDK currently
supports the IOMMU found on PowerNV platforms, sPAPRv2, so IOVA=VA
mode can be enabled when the correct platform is detected.
The POWER IOMMU type can't be detected through mechanisms such as
parsing files in the /sys hierarchy like x86_64 systems so the
/proc/cpuinfo file is parsed to determine whether Linux is running
on bare metal (i.e. PowerNV) or in a virtual environment (KVM/QEMU).
Signed-off-by: David Christensen <drc@linux.vnet.ibm.com>
Haiyue Wang [Sat, 25 Apr 2020 10:56:17 +0000 (18:56 +0800)]
build: add internal tag check
Add checks during build to ensure that all symbols in the INTERNAL
version map section have __internal tags on their definitions, and
enable the warnings needed to announce their use.
Signed-off-by: Haiyue Wang <haiyue.wang@intel.com> Signed-off-by: David Marchand <david.marchand@redhat.com>
Haiyue Wang [Sat, 25 Apr 2020 10:56:15 +0000 (18:56 +0800)]
eal: add internal ABI tag definition
Introduce the __rte_internal tag to mark internal ABI function which is
used only by the drivers or other libraries.
Like for __rte_experimental, this tag must be on a separate line before
function proprotypes.
Signed-off-by: Haiyue Wang <haiyue.wang@intel.com>
Li Feng [Fri, 24 Apr 2020 13:19:05 +0000 (21:19 +0800)]
mem: exclude unused memory from core dump
Currently, even though memory is mapped with PROT_NONE, this does not
cause it to be excluded from core dumps. This is counter-productive,
because in a lot of cases, this memory will go unused (e.g. when the
memory subsystem preallocates VA space but hasn't yet mapped physical
pages into it).
Use `madvise()` call with MADV_DONTDUMP/MADV_NOCORE to exclude the
unused memory from being dumped.
Signed-off-by: Li Feng <fengli@smartx.com> Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
Li Feng [Fri, 24 Apr 2020 10:42:05 +0000 (18:42 +0800)]
mem: mark pages as not accessed when freeing memory
Commit 8a4baf06c17a ("mem: mark pages as not accessed when reserving VA")
has mapped the initialized memory with PROT_NONE, and when it's unmapped,
eal_memalloc.c should remmap the anonymous memory with PROT_NONE too.
Fixes: 8a4baf06c17a ("mem: mark pages as not accessed when reserving VA") Cc: stable@dpdk.org Signed-off-by: Li Feng <fengli@smartx.com> Acked-by: Anatoly Burakov <anatoly.burakov@intel.com>
Fix spelling errors in comments (found with codespell).
Note that "inbetween" is not correct in English and should
either be two words or better yet, the in can be dropped.
https://www.grammarly.com/blog/in-between-or-inbetween/
Fixes: 12f45fa7e29b ("eal/arm: read timer from PMU if enabled") Fixes: 096ffd811fe2 ("eal/x86: use lock-prefixed instructions for SMP barrier") Fixes: 1d406458db47 ("mem: make segment preallocation OS-specific") Fixes: bb372060dad4 ("malloc: make heap a doubly-linked list") Fixes: 7353ee7344b4 ("fbarray: add API to find biggest used or free chunks") Cc: stable@dpdk.org Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
The code didn't compile when using exported mempool functions
under Windows.
compilation error logs:
rte_mempool_exports.def : error LNK2001:
unresolved external symbol rte_mempool_cache_flush
rte_mempool_exports.def : error LNK2001:
unresolved external symbol rte_mempool_default_cache
rte_mempool_exports.def : error LNK2001:
unresolved external symbol rte_mempool_generic_get
rte_mempool_exports.def : error LNK2001:
unresolved external symbol rte_mempool_generic_put
lib\librte_mempool.dll.a : fatal error LNK1120: 4 unresolved externals
clang: error: linker command failed with exit code 1120 (use -v to see invocation)
The cause was that there were some inline functions that were included
in the export list.
To solve this the functions, which are implemented in the header
and shouldn't be exported, were removed from rte_mempool_version.map
export list.
Fixes: 4b5062755aa7 ("mempool: allow user-owned cache") Fixes: 656f2d3ede96 ("mempool: deprecate specific get and put functions") Cc: stable@dpdk.org Signed-off-by: Fady Bader <fady@mellanox.com> Acked-by: Andrew Rybchenko <arybchenko@solarflare.com>
Thomas Monjalon [Fri, 24 Apr 2020 09:32:24 +0000 (11:32 +0200)]
devtools: fix tracepoint symbols check
The tracepoint symbols __rte_*_trace_* are defined via a macro,
adding prefix "__", so they cannot be found by map checker.
Those symbols defined by RTE_TRACE_POINT and RTE_TRACE_POINT_FP
are checked in source code without the generated prefix.
The same logic is applied to per core variables, previously skipped.
Kalesh AP [Tue, 21 Apr 2020 21:33:50 +0000 (14:33 -0700)]
net/bnxt: log firmware status on early init failure
Firmware reports any fatal error (either ASIC or Firmware related) via a
new status register. This status register can provide more detailed
information about the firmware errors, especially if error occurs before
HWRM_VER_GET is issued. Attempt to map this register if it is present
and check for firmware status when VER_GET command fails.
Refactored the code to allocate the "bp->recovery_info" structure
in bnxt_init_fw() instead of doing in bnxt_hwrm_error_recovery_qcfg().
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com> Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Kalesh AP [Tue, 21 Apr 2020 21:33:49 +0000 (14:33 -0700)]
net/bnxt: fix VLAN add when port is stopped
Driver destroys the vnic when the port is brought down.
When user tries to add a vlan when port is stopped, driver
issues HWRM command to FW with invalid vnic_id and it fails.
Fixed to return an error while setting vlan when port is
not started.
Kalesh AP [Tue, 21 Apr 2020 21:33:46 +0000 (14:33 -0700)]
net/bnxt: fix HWRM command during FW reset
FW returns HWRM_ERR_CODE_HOT_RESET_PROGRESS(0xa) when it is
unable to process a specific cmd while hot reset is in progress.
Host driver is expected to keep retrying the cmd for 2s with
a gap of 50ms between each retrial.
Also, fixed to fail port start if the HWRM_FUNC_DRV_IF_CHANGE
still returns error after 2 seconds.
Fixes: 0b533591238f ("net/bnxt: inform firmware about IF state changes") Cc: stable@dpdk.org Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com> Reviewed-by: Lance Richardson <lance.richardson@broadcom.com> Reviewed-by: Somnath Kotur <somnath.kotur@broadcom.com> Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
Kalesh AP [Tue, 21 Apr 2020 21:33:45 +0000 (14:33 -0700)]
net/bnxt: support speed capabilities query
Added information about supported speeds for the port in the
"dev_infos_get". As other PMDs are returning the speed capabilities,
apps may expect this behavior from bnxt PMD.
Signed-off-by: Kalesh AP <kalesh-anakkur.purayil@broadcom.com> Reviewed-by: Ajit Khaparde <ajit.khaparde@broadcom.com>
If there's VF reset, the kernel PF will remove rules
associated with the reset VF no matter the HW VSI ID
is changed or not. So DCF should redirector all rules
associated with the reset VF no matter the HW VSI ID
is changed or not.
Fixes: 3b3757bda3c3 ("net/ice: get VF hardware index in DCF") Fixes: c8183dd8e059 ("net/ice: redirect switch rule to new VSI") Signed-off-by: Beilei Xing <beilei.xing@intel.com> Acked-by: Haiyue Wang <haiyue.wang@intel.com>
The meson build file does not enable i40e vectorization support for
PPC/altivec systems, even though the existing Makefile does enable the
support. Add the required architecture check and sources line.
Signed-off-by: David Christensen <drc@linux.vnet.ibm.com> Acked-by: Bruce Richardson <bruce.richardson@intel.com>
Shougang Wang [Mon, 13 Apr 2020 01:38:39 +0000 (01:38 +0000)]
net/ixgbe: fix link status after port reset
It's a normal behavior to change the link status to up after
resetting the port. So it is unnecessary to set link down before
starting port, and changing the link state(link up/down) frequently
will cause link speed unstable.
Fixes: c3f2fbff78cf ("net/ixgbe: fix link status") Cc: stable@dpdk.org Signed-off-by: Shougang Wang <shougangx.wang@intel.com> Acked-by: Qiming Yang <qiming.yang@intel.com> Tested-by: Xueming Zhang <xuemingx.zhang@intel.com>
Tao Zhu [Wed, 22 Apr 2020 12:37:36 +0000 (12:37 +0000)]
net/ixgbe: fix resource leak after thread exits normally
When the thread exits normally, pthread_join() is not called, which can
result in a resource leak. Therefore, the thread is set to separation
mode using function pthread_detach(), so that no program call
pthread_join() is required to recycle, and when the thread exits,
the system automatically reclaims resources.
Wait for the thread to finish with timeout argument(0 means that it will
not return until link complete), wait until the thread finishes before
returning. Normally, the thread will finish in a shorter time, and give
a warning message if it hasn't finished in a longer time.
Fixes: 819d0d1d57f1 ("net/ixgbe: fix blocking system events") Cc: stable@dpdk.org Signed-off-by: Tao Zhu <taox.zhu@intel.com> Acked-by: Konstantin Ananyev <konstantin.ananyev@intel.com> Reviewed-by: Xiaolong Ye <xiaolong.ye@intel.com>
MPRQ is silently turned off in case there is not enough
Rx queues configured. Improve the logging to show a
warning in this case to notify a user about the Rx burst
function selected.
Using a global mbuf dynamic field for metadata incurs some
performance penalty on a datapath. Store this information in
the Rx queue descriptor for a better cache locality.
Fixes: a18ac6113331 ("net/mlx5: add metadata support to Rx datapath") Cc: stable@dpdk.org Signed-off-by: Alexander Kozyrev <akozyrev@mellanox.com> Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
The header modify actions number supported now has some limitation,
and it is decided by both driver and hardware. If the configuration
is different or the table to insert the flow is different, the result
might be different if the flow contains header modify actions.
Currently, the actual action number could only be calculated in the
later stage called translate, from user specified value to the driver
format. And the action numbers checking is missed in the flow
validation. So PMD will return incorrect result to indicate the
flow actions are valid by rte_flow_validate but then it will fail
when calling rte_flow_create.
Adding some simple checking in the validation will help to get rid
of this incorrect checking. Most of the actions will only consume 1
SW action field except the MAC address and IPv6 address. And from
SW POV, the maximal action fields for these will be consumed even if
only part of such field will be modified because that there is no
mask in the flow actions and the mask will always be all ONEs.
The metering or extra metadata supports will cost one more action.
The meters of ports share the same meter table on
the port. When releasing meters, don't check value returned
using assert. Because other meters may reference to it.
Lijun Ou [Sat, 18 Apr 2020 03:44:38 +0000 (11:44 +0800)]
app/testpmd: remove hardcoded descriptors limit
When users configure rxds and txds by used port config cmd based
on testpmd application, it will not be able to configure rxd and
txd according to the max capability range supported by the actual
NIC hardware. Due testpmd defects, it can only configure a fixed
range to 0 to 2048.
The final result is that an incorrect printing prompt appears and
cannot be applied using rxd && txd according to the actual
capabilities supported by the device.
In order to solve the above problems, we modify the testpmd. First
by calling the rte_eth_dev_info_get api to obtain the max and min
rx/tx capability supported by the hns3, and then use this range
to compare with the actual value by users configured and make
reasonable limitation.
Signed-off-by: Lijun Ou <oulijun@huawei.com> Signed-off-by: Huisong Li <lihuisong@huawei.com> Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com> Acked-by: Bernard Iremonger <bernard.iremonger@intel.com>
This patch adds getting Rx/Tx queue fbd information in extended device
statistics. The upper level application can get them by calling the
rte_eth_xstats_get API function.
The fbd registers of every Rx/Tx queue are very useful to identify the
Rx/Tx bottleneck.
1. The Rx queue fbd register is the number of the unprocessed buffer
descriptors which are waiting for driver to process;
2. The Tx queue fbd register is the number of the unprocessed buffer
descriptors which are waiting for network engine hardware to process.
As a result, we get the following output information in testpmd
application by using the command "show port xstats" as below:
rx_q0RX_QUEUE_FBD: 19
rx_q1RX_QUEUE_FBD: 18
tx_q0TX_QUEUE_FBD: 0
tx_q1TX_QUEUE_FBD: 0
Signed-off-by: Chengwen Feng <fengchengwen@huawei.com> Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com> Signed-off-by: Huisong Li <lihuisong@huawei.com>
This patch modifies the print format for firmware version in the log, It
replaces "0x%08x" with "%lu.%lu.%lu.%lu" in the format control string.
By the way, this patch adds ".fw_version_get" ops implemation for hns3
VF PMD driver.
Signed-off-by: Wei Hu (Xavier) <xavier.huwei@huawei.com>
Jeff Guo [Wed, 22 Apr 2020 01:02:55 +0000 (21:02 -0400)]
app/testpmd: add new types to RSS hash commands
Add some new types, such as eth/l2-src-only/l2-dst-only/svlan/cvlan/
l2tpv3/esp/ah/pfcp types into RSS hash commands, it could be used
to configure these rss input set by cmdline.
Example flow commands was:
testpmd>flow create 0 ingress pattern eth / ipv4 / l2tpv3oip / end \
actions rss types l2tpv3 end key_len 0 queues end / end
port config commands was:
testpmd>port config all rss l2tpv3
Signed-off-by: Jeff Guo <jia.guo@intel.com> Acked-by: Ori Kam <orika@mellanox.com> Acked-by: Bernard Iremonger <bernard.iremonger@intel.com>
Jeff Guo [Wed, 22 Apr 2020 01:02:54 +0000 (21:02 -0400)]
net/iavf: add RSS configuration for VF
The VF must be capable of configuring RSS. Add a virtchnl handler to
parse a specific RSS configuration, and process the configuration for
VFs, such as add or delete a RSS rule.
Signed-off-by: Jeff Guo <jia.guo@intel.com> Reviewed-by: Qi Zhang <qi.z.zhang@intel.com>
Jeff Guo [Wed, 22 Apr 2020 01:02:53 +0000 (21:02 -0400)]
ethdev: add RSS offload types
Defines some new RSS offload types for ETH/S_VLAN/C_VLAN/L2TPV3/
/PFCP/L2_SRC_ONLY/L2_DST_ONLY.
Signed-off-by: Jeff Guo <jia.guo@intel.com> Reviewed-by: Qi Zhang <qi.z.zhang@intel.com> Reviewed-by: Bernard Iremonger <bernard.iremonger@intel.com> Acked-by: Ori Kam <orika@mellanox.com> Reviewed-by: Ferruh Yigit <ferruh.yigit@intel.com>
One of the reasons to destroy a flow is the fact that no packet matches
the flow for "timeout" time.
For example, when TCP\UDP sessions are suddenly closed.
Currently, there is not any DPDK mechanism for flow aging and the
applications use their own ways to detect and destroy aged-out flows.
The flow aging implementation need include:
- A new rte_flow action: RTE_FLOW_ACTION_TYPE_AGE to set the timeout and
the application flow context for each flow.
- A new ethdev event: RTE_ETH_EVENT_FLOW_AGED for the driver to report
that there are new aged-out flows.
- A new rte_flow API: rte_flow_get_aged_flows to get the aged-out flows
contexts from the port.
- Support input flow aging command line in Testpmd.
The new event type addition in the enum is flagged as an ABI breakage,
so an ignore rule is added for these reasons:
- It is not changing value of existing types (except MAX)
- The new value is not used by existing API if the event is not
registered
In general, it is safe adding new ethdev event types at the end of the
enum, because of event callback registration mechanism.
Signed-off-by: Dong Zhou <dongz@mellanox.com> Acked-by: Ori Kam <orika@mellanox.com> Acked-by: Andrew Rybchenko <arybchenko@solarflare.com> Acked-by: Jerin Jacob <jerinj@marvell.com> Acked-by: Matan Azrad <matan@mellanox.com>
Thomas Monjalon [Thu, 16 Apr 2020 15:59:04 +0000 (17:59 +0200)]
common/mlx5: fix build with rdma-core 21
drivers/common/mlx5/mlx5_glue.c: In function 'mlx5_glue_devx_qp_query':
drivers/common/mlx5/mlx5_glue.c:1010:9: error:
implicit declaration of function 'mlx5dv_devx_qp_query'
The function mlx5dv_devx_qp_query() was added in rdma-core 22.
This function is compiled only if HAVE_IBV_DEVX_OBJ,
which is true if the function mlx5dv_devx_obj_create is found.
Unfortunately mlx5dv_devx_obj_create() was introduced in rdma-core 21,
when mlx5dv_devx_qp_query() did not exist yet.
A new flag HAVE_IBV_DEVX_QP is added when mlx5dv_devx_qp_query()
is found.
Fixes: 62d6f70f30f4 ("net/mlx5: add glue for queue query via DevX") Cc: stable@dpdk.org Signed-off-by: Thomas Monjalon <thomas@monjalon.net> Acked-by: Viacheslav Ovsiienko <viacheslavo@mellanox.com>
Remove references to struct rte_fdir_conf which is deprecated,
in i40e_flow_parse_fdir_filter(), i40e_flow_destroy() and
i40e_flow_add_del_fdir_filter().
Haiyue Wang [Thu, 16 Apr 2020 03:11:22 +0000 (11:11 +0800)]
net/ice: check DSN package file firstly
Align the package file search sequence with PF only for DCF mode. Get
the DSN through the virtual channel firstly to check the accessibility
of the package file.
Signed-off-by: Haiyue Wang <haiyue.wang@intel.com> Acked-by: Qi Zhang <qi.z.zhang@intel.com>
Leyi Rong [Mon, 20 Apr 2020 06:16:20 +0000 (14:16 +0800)]
net/iavf: add flow director enabled switch value
The commit adds fdir_enabled flag into iavf_rx_queue structure
to identify if fdir id is active. Rx data path can be benefit if
fdir id parsing is not needed, especially in vector path.
Signed-off-by: Leyi Rong <leyi.rong@intel.com> Reviewed-by: Qi Zhang <qi.z.zhang@intel.com>
When destroy the flow with RSS, flow can invoke the queues information
from hrxq index table object, since the queue number and list are both
saved to the index table object. No need to save the duplicated data in
rte flow.
Save the RSS description information to the intermediate private data
when create the flow with RSS action helps to save the memory for rte
flow.
Wentao Cui [Thu, 16 Apr 2020 08:34:28 +0000 (16:34 +0800)]
net/mlx5: optimize flow director filter memory
This commit is for mlx5 fdir flow memory optimization.
Currently for the fdir member in rte_flow structure. It saves the fdir
memory pointer directly. As fdir is fading away, use one bit help to
indicate the function in the flow and add the content to an extra list
save the memory for the other widely usage cases.
Allocate mark copy resource from indexed pool helps rte flow saves the 4
bytes index instead of 8 bytes pointer. For mark copy resource itself,
it helps save MALLOC_ELEM_OVERHEAD bytes from rte_malloc().
While flow attaches the meter handle, the meter id can be the unique tag
for the flow to get the meter handle. It's no need for flow to save the
pointer of the meter handle.
Save the meter id instead of pointer helps reduce the size for rte flow
structure.
As the supported maximum meter rule is 4K, uint16_t type is selected for
the meter id.
Currently, the mlx5_flow_handle struct is not fully aligned and has some
bits wasted. The members can be optimized and reorganized to save memory.
1. As metadata and meter is sharing the same flow match id, now the flow
id is limited to 24 bits due to the 8 MSBs are used as for the meter
color. Align the flow id to other bit members to 32 bits to save the
mlx5 flow handle memory.
2. The vlan_vf in struct mlx5_flow_handle_dv was already moved to struct
mlx5_flow_handle. Remove the legacy vlan_vf in struct
mlx5_flow_handle_dv.
3. Reorganize the vlan_vf in mlx5_flow_handle with member SILIST_ENTRY
next to make it align with 8 bytes.
4. Reorganize the header modify in mlx5_flow_handle_dv to ILIST_ENTRY
next to make it align to with bytes.
5. Introduce __rte_pack attribute to make the struct tightly organized.
It will totally save 20 bytes memory for mlx5_flow_handle struct.
For the resource objects which are converted to indexed, align the names
with the prefix of rix_.
As only limited bits is used in act_flags for flow destroy, it's a bit
expensive to save the whole 64 bits. Move the act_flags out of flow
handle and save the needed bits for flow destroy to save some bytes for
the flow handle data struct.
The fate action type and mark bits are reserved as they will be used in
flow destroy.
git.droids-corp.org / dpdk.git / log