[COMMIT] branch api-next updated. v1.19.0.2_tigermoth-325-ga110685b
by git＠git-us.linaro.org
This is an automated email from the git hooks/post-receive script. It was
generated because a ref change was pushed to the repository containing
the project "".
The branch, api-next has been updated
via a110685b8357276cb4a63ebc6ff421f42f461d94 (commit)
from bc86441b2d02dd518e710f1d9e6936525530c1bb (commit)
Those revisions listed above that are new to this repository have
not appeared on any other notification email; so we list those
revisions in full, below.
- Log -----------------------------------------------------------------
commit a110685b8357276cb4a63ebc6ff421f42f461d94
Author: Dmitry Eremin-Solenikov <dmitry.ereminsolenikov(a)linaro.org>
Date: Fri Nov 23 03:24:39 2018 +0300
linux-gen: event: support flow-awareness API
Signed-off-by: Dmitry Eremin-Solenikov <dmitry.ereminsolenikov(a)linaro.org>
Reviewed-by: Bill Fischofer <bill.fischofer(a)linaro.org>
Reviewed-by: Petri Savolainen <petri.savolainen(a)linaro.org>
Signed-off-by: Maxim Uvarov <maxim.uvarov(a)linaro.org>
diff --git a/platform/linux-generic/odp_event.c b/platform/linux-generic/odp_event.c
index bb378528..bdde93e1 100644
--- a/platform/linux-generic/odp_event.c
+++ b/platform/linux-generic/odp_event.c
@@ -59,6 +59,19 @@ int odp_event_type_multi(const odp_event_t event[], int num,
return i;
}
+/* For now ODP generic does not support flow awareness,
+ * so all flow ids are zero. */
+uint32_t odp_event_flow_id(odp_event_t event ODP_UNUSED)
+{
+ return 0;
+}
+
+void odp_event_flow_id_set(odp_event_t event ODP_UNUSED,
+ uint32_t flow_id ODP_UNUSED)
+{
+ /* Do nothing */
+}
+
void odp_event_free(odp_event_t event)
{
switch (odp_event_type(event)) {
-----------------------------------------------------------------------
Summary of changes:
platform/linux-generic/odp_event.c | 13 +++++++++++++
1 file changed, 13 insertions(+)
hooks/post-receive
--
6 years, 2 months
- 1
- 0
[COMMIT] annotated tag v1.20.0.0 created. v1.20.0.0
by git＠git-us.linaro.org
This is an automated email from the git hooks/post-receive script. It was
generated because a ref change was pushed to the repository containing
the project "".
The annotated tag, v1.20.0.0 has been created
at 503236d6b9058a45ad87881f2d6e7893298b64fc (tag)
tagging ec5066a3430e31a87727ac4aea5793253e5ee843 (commit)
replaces v1.19.0.2_tigermoth
tagged by Maxim Uvarov
on Tue Dec 4 09:52:06 2018 +0300
- Log -----------------------------------------------------------------
== OpenDataPlane (1.20.0.0)
=== Summary of Changes
ODP v1.20.0.0 is a refresh of ODP, incorporating significant configurability
and performance improvements as well as new APIs and API restructures.
==== APIs
===== Symbol `ODP_SHM_NULL` Removed.
An invalid `odp_shm_t` has the value `ODP_SHM_INVALID`, consistent with other
ODP types. The legacy synonym `ODP_SHM_NULL` is now removed for consistency.
===== New 3GPP Crypto Algorithm Support
New support for 3GPP crypto algorithms is added by defining symbols for
* `ODP_CIPHER_ALG_KASUMI_F8`
* `ODP_CIPHER_ALG_SNOW3G_UEA2`
* `ODP_CIPHER_ALG_ZUC_EEA3`
In addition new authentication algorithm symbols are defined for
* `ODP_AUTH_ALG_KASUMI_F9`
* `ODP_AUTH_ALG_SNOW3G_UIA2`
* `ODP_AUTH_ALG_ZUC_EIA3`
These values are returned as ODP capabilities as well as being accepted in
crypto session creation for implementations that indicate support for them.
===== Crypto Capability for Bitwise Operation
The new `bit_mode` capability Boolean is added to the
`odp_crypto_cipher_capability_t` struct to indicate that an implementation
supports operating in bit mode. When operating in bit
mode, field offsets and lengths are expressed in terms of bits rather than
bytes. However, such lengths must always be specified in multiples of 8.
===== Improved Crypto Spec Documentation
The ODP crypto API specification is tightened to specify default values for
cipher and authentication algorithms. Also documented when key and IV
parameters need to be set.
===== IPsec Extensions
IPsec requires "salt" (extra keying material) when the GMAC authentication
algorithm is used. To accommodate this the `auth_key_extra` field is added to
the `odp_ipsec_crypto_param_t` struct and documentation is added clarifying
when this field is needed and how it should be used.
===== Classifier Type Rename
The `odp_pmr_t` type name for an invalid value is renamed from `ODP_PMR_INVAL`
to `ODP_PMR_INVALID` for consistency with the rest of ODP type names. The old
symbol is still available when ODP is configured with
`--enable-deprecated`.
===== New API for Packet Event Subtypes
The `odp_packet_subtype()` API is added that returns the subtype of a packet
event directly.
===== Streamlined Packet Parsing Results
The `odp_packet_parse_result()` API is added that returns the result of
packet parsing as a single `odp_packet_parse_result_t` struct. This can
offer efficiency improvements for applications that need all parse results
rather than making individual parse result calls.
===== PktIO Extensions to Support per-Queue Configuration
PktIO interfaces support multiple input queues to enable increased parallelism
in I/O processing. Previously, all of these input queues were required to
belong to the same scheduler group. The `odp_pktin_queue_param_t` struct is
now extended with an optional `odp_pktin_queue_param_ovr_t` struct that
permits individual pktin queues to be assigned to separate scheduler groups.
This may permit improved performance for advanced application use cases.
===== Timer Pool Capabilities
The `odp_timer_capability_t` struct is extended to return three additional
pieces of information:
`max_pools_combined`::
The total number of timer pools that can be created combining different
clock sources
`max_pools`::
The maximum number of timer pools for a given clock source.
`max_timers`::
The maximum number of timers in a single pool. A zero value means number is
limited only by available memory.
===== Add Scheduler mix/max/default Priority Functions
Three new APIs: `odp_schedule_max_prio()`, `odp_schedule_min_prio()`, and
`odp_schedule_default_prio()` are added that return the min, max, and default
values specified for the `prio` field in the `odp_schedule_param_t` struct.
With the introduction of these scheduling priority functions the previously
defined macros (`ODP_SCHED_PRIO_HIGHEST`, `ODP_SCHED_PRIO_NORMAL`, and
`ODP_SCHED_PRIO_LOWEST`) are now deprecated and should no longer be used.
===== Specification of `odp_schedule_prio_t` as an `int`
Previously, the `odp_schedule_prio_t` type definition was left to each
implementation. With the addition of explicit schedule priority ranges, this
type is now specified to be an `int` to permit efficient implementation
(including inlining) of these functions.
====== New Scheduler APIs
The new scheduler APIs `odp_schedule_multi_wait()` and
`odp_schedule_multi_no_wait()` are added to provide more efficiently
implementable versions of these functions. The existing scheduler APIs remain
unchanged. These new APIs can simply provide a fastpath for some
applications/implementations as an alternative to specifying a parameter on
`odp_schedule_multi()`.
===== Memory Model in `odp_init_global()`
The `odp_init_t` parameter passed to `odp_init_global()` is extended to
add the `mem_model` field. This field is defined by the new `odp_mem_model_t`
struct and is used to specify whether the application will be using a
thread (`ODP_MEM_MODEL_THREAD`) or process (`ODP_MEM_MODEL_PROCESS`)
memory model. The default is a thread model is used for compatibility with
previous levels of ODP.
==== ABI Changes
A number of changes to the ODP ABI have also been made in this release to
improve application binary portability.
===== Strong Typing for Timer Pools
The `odp_timer_pool_t` is now strongly typed.
===== Consistent Initialization
The values of the various `ODP_xxx_INVALID` symbols for ODP abstract types in
the `odp-linux` reference implementation are now consistently zeros. This
reduces errors and improves portability.
=== Implementation Improvements
==== Configuration File
A new configuration file mechanism is introduced that makes use of
https://www.hyperrealm.com/libconfig/libconfig_manual.html[libconfig] to
enable various runtime ODP parameters to be specified dynamically.
Default configuration values for the `odp-linux` reference implementation are
contained in the `config/odp-linux-generic.conf` file. Users may override
these default values by supplying their own configuration file. At
`odp_init_global()` time, if the `ODP_CONFIG_FILE` environment variable is set,
this is used to locate the path to the override configuration file.
==== Process Mode Support
The `odp-linux` reference implementation now supports applications that run in
process mode (`mem_model` = `ODP_MEM_MODEL_PROCESS`) as well as the default
thread mode. This support only applies within a single ODP instance, so any
`fork()` calls must be done only _after_ `odp_init_global()` has been called
to initialize ODP on a root process.
==== Removal of `iQuery` Scheduler
The `iQuery` scheduler is removed from the `odp-linux` reference
implementation, as it offers no performance advantages and has not seen
application use.
==== Number of CPUs
The `odp-linux` reference implementation now supports up to 256 CPUs by
default (increased from 128).
==== Support for Large Burst Sizes
The `odp-linux` reference implementation now supports large burst sizes for
both I/O and non-I/O scheduled events. Large bursts (when available) are
received directly to the application without any stashing for improved
throughput. Burst sizes are configurable via the new configuration file
mechanism, as described above.
==== `--without-openssl` Warnings
When building `odp-linux` using `--without-openssl` a warning will be issued
cautioning that strong cryptography will not be available.
==== Inline Queue Enq/Deq APIs
The various enq/deq APIs are now subject to inlining when `odp-linux` is
built with `--disable-abi-compat`.
==== Configurable Timer Controls
Inline timers are now controlled via a config file option. Timer polling
frequency is similarly controlled via the config file.
==== Huge Page Configuration
The config file is now used to specify the huge page usage limit.
==== Single and Multi-Consumer/Producer Rings
The queue implementation in `odp-linux` now automatically makes use of
optimized single and multi-consumer/producer rings to significantly speed
up enq/deq processing.
==== `odp_shm_print_all()` Improvements
The output from `odp_shm_print_all()` is reworked to provide more useful
and comprehensive shared memory usage information in `odp-linux`.
==== IPsec Improvements
SA lifetime checking is now more scalable to multiple threads. This
significantly reduces overhead for multithreaded IPsec applications.
==== Native Builds
When running in non-ABI compatibility mode, `odp-linux` now enables
native machine-specific optimizations for the CPU architecture of the
local machine.
=== Validation Test Improvements
==== SCTP Test Packets
SCTP test packets are now used in parser testing. SCTP headers are added to
ODP and ODP helpers and SCTP checksums are now inserted and verified as part
of validation testing.
==== `odp_packet_reset()` Test
The packet validation test suite now properly tests `odp_packet_reset()`.
=== Helper Changes
In support of process mode, ODP helper functions have been changed to
better match these new capabilities
==== New `enum`
The `odph_linux_thread_type_t enum` has been replaced with the new
`odp_mem_model_t` type.
==== Helper Options
The new `odph_options()` getter function is added that returns
applicable options in effect via the new `odph_helper_options_t` struct.
This currently includes the memory model (thread or process) that is in use.
==== SCTP Helpers
The new helper APIs `odph_sctp_chksum_set()` and `odph_sctp_chksum_verify()`
are added to facilitate working with SCTP packet checksums.
=== Performance Test Improvements
==== Pool Performance
A new `odp_pool_perf` test has been added that stress-tests ODP pool
functions in a multithreaded environment to generate performance statistics.
==== Scheduler Performance
A new `odp_sched_perf` test has been added that stress-tests the scheduler
in a multithreaded environment.
==== CPU Performance
A new `odp_cpu_bench` performance test has been added that runs
compute-intensive packet operations in a multithreaded environment and prints
the observed maximum throughput for each thread.
=== Example Improvements
==== Classifier Example changes
The `odp_classifier` example program now uses a reduced number of threads by
default to reduce elapsed run time. `ODP_THREAD_COUNT_MAX` is also now used as
the max worker count.
==== Generator Improvements
The `odp_generator` example has numerous cleanups and performance improvements.
==== IPsec Example
The `odp_ipsec` example now properly stops and closes pktio devices on exit.
==== Packet Dumping
A new `odp_packet_dump` example is added that prints received packets to the
terminal. This is useful for debugging packet I/O interfaces.
==== Sysinfo Example
A new `odp_sysinfo` example is provided that prints system information. Useful
for checking the ODP environment during debugging. This includes providing
detailed information about the various crypto facilities supported, as well
as the feature flags used at build time (_e.g.,_ if the binary was built with
ARMv8.0 or ARMv8.1 instructions).
==== Traffic Manager Example
The traffic manager example now properly destroys all TM queues it creates
for improved reliability. It also now always prints a proper termination
summary message.
=== Bug Fixes
==== Numbered Bugs/Issues
===== https://bugs.linaro.org/show_bug.cgi?id=3983[Bug 3983]
Compile fails on OpenSuSE 42.2 Leap with error: negative width in bit field
'__error_if_negative'
===== https://bugs.linaro.org/show_bug.cgi?id=3989[Bug 3989]
odp_system_info_init() issues
===== https://bugs.linaro.org/show_bug.cgi?id=3999[Bug 3999]
IPsec antireplay check drops packets when sequence number jumps.
===== https://bugs.linaro.org/show_bug.cgi?id=4002[Bug 4002]
IPsec SA creation must fail for ESN-enabled SAs
===== https://bugs.linaro.org/show_bug.cgi?id=4013[Bug 4013]
Per-SA IPv4 ID allocation may cause duplicate IDs.
===== https://bugs.linaro.org/show_bug.cgi?id=4017[Bug 4017]
Unexpected IP ID causes IPsec API validation to fail
===== https://github.com/Linaro/odp/issues/662[Issue 662]
rte_mempool_ops_alloc() is not dpdk api
==== Unnumbered Bugs/Issues
* Fixed enq/deq issues encountered on architectures with weak memory ordering.
* Return 0 from `odp_sys_huge_page_size_all()` if hugepages are not
supported/detected. Tests modified to not treat this as an error.
* Set `ODP_CACHE_LINE_SIZE` to 128 on ppc64le systems.
* iplookuptable fix putting values into table
* DPDK pktio support now works properly across multiple ODP instances.
* Zero timer pool memory on reserve (fixes timer failures due to uninitialized
variables).
* `-march=native` disabled for `clang`. This fixes a known issue with recent
levels of clang.
=== Known Issues
==== https://bugs.linaro.org/show_bug.cgi?id=3998[Bug 3998]
IPsec extended sequence number support is missing
==== https://bugs.linaro.org/show_bug.cgi?id=4014[Bug 4014]
Separate IP ID allocation for transport and tunnel mode SAs may cause
duplicate IDs
==== https://bugs.linaro.org/show_bug.cgi?id=4018[Bug 4018]
Unexpected IV causes IPsec API validation to fail
==== https://bugs.linaro.org/show_bug.cgi?id=4040[Bug 4040]
Clang build fails on Ubuntu 18.04
Bill Fischofer (3):
configure: add warning about --without-openssl implications
doc: implementation: add configuration section to implementation guide
changelog: updates for odp v1.20.0.0
Bogdan Pricope (3):
api: pktio: extend odp_pktin_queue_param_t to support per queue configuration
linux-gen: pktio: implement per queue pktin configuration
validation: pktio: test per queue pktin configuration
Dmitry Eremin-Solenikov (47):
travis: stop using deprecated image
travis: pin repositories to architectures
travis: switch iptables policy to let tap driver work
example: sysinfo: print more information about crypto capabilities
linux-gen: crypto: properly sort AES-CMAC capabilities
validation: pktio: check sctp flag in parser tests
validation: pktio: fix checks in pktio tests
test/common: add IPv6 SCTP packet
validation: pktio: parser: add SCTP test
helper: add SCTP protocol header
helper: add SCTP handling functions
linux-gen: add SCTP protocol header
linux-gen: packet: parse SCTP packets and verify checksum
linux-gen: packet: support SCTP packet insertion
linux-gen: pktio: loop: handle SCTP checksums
linux-gen: ipsec: insert checksum into SCTP packets if required
validation: pktio: add SCTP checksum tests
travis: fix DOCKER_NAMESPACE variable setting
travis: switch to unified 16.04 image
m4: odp_dpdk: pass CFLAGS and LDFLAGS to CC when locating libdpdk.so
travis: move netmap to install stage
travis: restore quick build-only testing
ci: rewrite coverage and distcheck scripts to follow other build scripts
ci: move build stage to common script
ci: do compile-after-install test
linux-gen: sysinfo: return 0 if hugepages are not supported
linux-gen: x86: as a last resort parse max cpu freq from bogomips value
validation: system: 0 is valid huge page size
include: abi: set ODP_CACHE_LINE_SIZE to 128 on ppc64le
linux-gen: abi: set ODP_CACHE_LINE_SIZE to 128 on ppc64le
linux-gen: fix pktio private size on 128-byte cache machines
validation: time: fix c&p error
validation: scheduler: print debug diagnostics on test_wait failure
validation: time: be more tolerant wrt delays
validation: timer: be more tolerant wrt delays
validation: scheduler: increase wait tolerance timeout to 150 msec
linux-gen: ipsec: fix sliding window shifts
linux-gen: ipsec: reject SA creation with ESN flag set
configure.ac: set DX environment WITH_PLATFORM from configure.ac
doc: pass generated include files to Doxygen
m4: update ax_prog_doxygen.m4 to latest version
shippable: switch to using official Shippable image
api: crypto: provide definitions for 3GPP crypto algorithms
api: crypto: add bitstring operation capability
validation: crypto: add support for bit mode algorithm testing
validation: crypto: 3GPP crypto algorightms support
travis: install graphviz in doxygen test
Janne Peltonen (8):
linux-gen: ipsec: speed up random IV generation by thread-local buffering
linux-gen: ipsec: remove SA reference counting from outbound processing
linux-gen: ipsec: separate hot r/w data from r/o data in an SA
linux-gen: ipsec: use sequence number counter for counter based IV
validation: ipsec: make output checking accept any IP ID value
linux-gen: ipsec: use global IPv4 ID allocator for all tunnel SAs
linux-gen: ipsec: make IPv4 ID allocator scale better to multiple threads
linux-gen: ipsec: make SA lifetime checking more scalable to multiple threads
Josep Puigdemont (2):
linux-gen: ishm: implement huge page cache
linux-gen: ishm: make huge page cache size dynamic
Matias Elo (105):
example: use min number of workers by default
example: use ODP_THREAD_COUNT_MAX as max worker count
test: performance: add new CPU benchmarking application
linux-gen: pktio: remove unused ring operations
linux-gen: pktio: ring: guarantee enq/deq variable load order
linux-gen: ipsec: fall back to lower odp_random_kind_t if necessary
thread: increase ODP_THREAD_COUNT_MAX to 256
linux-gen: netmap: fix incorrect debug message
linux-gen: dpdk: fix calling internal dpdk function
abi: packet: set ODP_PACKET_SEG_INVALID to zero
abi: classification: set ODP_COS_INVALID to zero
abi: classification: set ODP_PMR_INVAL to zero
abi: ipsec: set ODP_IPSEC_SA_INVALID to zero
abi: pool: set ODP_POOL_INVALID to zero
abi: timer: set ODP_TIMER_POOL_INVALID to zero
abi: timer: set ODP_TIMEOUT_INVALID to zero
abi: timer: set ODP_TIMER_INVALID to zero
linux-gen: shm: add option for allocating internal shm using single VA
linux-gen: shm: increase pre-reserved virtual address space size
linux-gen: shm: reserve internal shms using single VA flag
linux-gen: split global data into RO and RW structs
linux-gen: pcapng: move global data into odp_global_data_rw_t
linux-gen: dpdk: move dpdk_initialized global var into odp_global_data_rw_t
linux-gen: timer: allocate global memory from shm
linux-gen: timer: move global variable 'locks' into timer_global_t
linux-gen: timer: move global variables into timer_global_t
linux-gen: ipsec: allocate memory from shm
linux-gen: hash: allocate crc table memory from shm
linux-gen: pcap: allocate tx buffer from stack
linux-gen: socket_mmap: reserve memory from shm
validation: atomic: allocate test variables from shm
validation: thread: allocate barriers from shm
validation: timer: allocate global data from shm
test: performance: enable testing in process mode
test: example: allocate global data from shm
test: ring: allocate global data from shm
test: mmap_vlan_ins: allocate global data from shm
helper: threads: add ODPH_PROC_MODE environment variable
linux-gen: test: configuration file for running process mode tests
travis: add process mode test
linux-gen: dpdk: fix running multiple odp instances simulaneusly
linux-gen: dpdk: prefix visible internal parse functions with _odp_
linux-gen: dpdk: improved zero-copy implementation
linux-gen: sched scalable: allocate global data from shm
linux-gen: ring: allocate global data from shm
example: stop and close pktio devices on exit
example: generator: use odp_wait_time_ns() instead of timers
example: generator: remove print from packet tx loop
linux-gen: pool: increase minimum packet segment length
linux-gen: queue: add internal interface for adding/removing inline timers
linux-gen: timer: reduce inline timer overhead
linux-gen: timer: run inline timers during queue dequeue operations
linux-gen: ishm: remove _ODP_SHM_NO_HP flag
linux-gen: ishm: remove unused odp_shm_internal header
linux-gen: ishm: add internal _ODP_ISHM_USE_HP flag
linux-gen: ishm: allocate small shm blocks using normal pages
linux-gen: ishm: remove unnecessary _odp_ishm_pool_lookup() function
linux-gen: ishm: remove unused internal functions
linux-gen: queue scalable: remove _ODP_ISHM_SINGLE_VA from pool create
linux-gen: ishm: move block memory mapping into _odp_ishm_address()
validation: pool: add test for creating and using a pool after fork
validation: shmem: reduce the number of workers in single VA alloc test
linux-gen: ishm: read single va size from config
linux-gen: ishm: use pre-reserved single va memory
linux-gen: init: always initialize odp_global_ro.init_param
linux-gen: timer: zero timer pool memory on reserve
linux-gen: timer: enable inline timer implementation using config file
travis: test inline timer implementation
linux-gen: timer: add config option for inline timer poll frequency
linux-gen: timer: decrease inline timer polling interval under load
linux-gen: ishm: add config option for selecting huge page usage limit
api: shm: remove unused ODP_SHM_NULL define
abi: timer: use strong type for odp_timer_pool_t
api: classifier: rename ODP_PMR_INVAL to ODP_PMR_INVALID
api: timer: add timer pool capabilities
linux-gen: timer: implement new timer pool capabilities
validation: timer: use new timer pool capabilities
api: init: add new mem_model member to odp_init_t
linux-gen: init: use new odp_init_t.mem_model parameter
validation: init: add test for odp_init_t.mem_model
helper: threads: replace odph_linux_thread_type_t with odp_mem_model_t
helper: threads: add odph_options() getter function
helper: test: add test for new odph_options() function
test: performance: use mem_model from helper options
example: use mem_model from helper options
validation: use use mem_model from helper options
linux-gen: shm: remove single_va configuration option
example: generator: add signal handler for SIGINT
linux-gen: pool: reduce buffer memory usage
linux-gen: pool: move ODP_CONFIG_BUFFER_ALIGN_MIN out of config header
linux-gen: pool: remove unnecessary align padding from buffers
abi: traffic_mngr: reduce max defines
linux-gen: traffic_mngr: use static array for odp_tm_systems
linux-gen: traffic_mngr: move tm_wred_node_t inside tm_queue_obj_t
linux-gen: traffic_mngr: allocate tm_queue_obj_t from shm
linux-gen: traffic_mngr: allocate tm_system_group_t from shm
linux-gen: traffic_mngr: move tm_schedulers_obj_t inside tm_node_obj_t
linux-gen: traffic_mngr: move tm_wred_node_t inside tm_node_obj_t
linux-gen: traffic_mngr: allocate tm_node_obj_t from shm
linux-gen: traffic_mngr: store tm_queue_obj_t array inside tm_system_t
linux-gen: traffic_mngr: store root node inside tm_system_t
linux-gen: traffic_mngr: store input_work_queue inside tm_system_t
linux-gen: traffic_mngr: allocate profile objects from shm
linux-gen: traffic_mngr: don't reserve memory if tm is disabled
linux-gen: traffic_mngr: allocate all global data from shm
Maxim Uvarov (28):
fix code check warning using CC in code
configure.ac print linker
linux-gen: odp_tm_queue_destroy should not take care about params
example: tm: add tm queues destroy
example: tm: use 0 array index for queues
example: tm: add some message on exit
travis: add docker tests
travis: setup default docker name space
travis: set CI for distcheck
travis: let after_failure task print logs
travis: add ubuntu 18.04 compilation test
linux-gen: odp_shm_print_all: refine output spreadsheet
abi: align ODP_CPUMASK_SIZE with kernel cpu_set_t
helper: iplookuptable fix puting values to table
travis: export CI for first distcheck
linux-gen: shm: do not print map error
linux-gen: do not use huge pages for internal allocations
remove scripts/build-pktio-dpdk
linuxgen: add dumpconfig utility
linux-gen: run without /proc mounted
travis: define compiler for clang test
configure: disable -march=native for clang
travis: check.sh request huge pages at early start
linux-gen: remove performance test for process mode
linux-gen: ishm: add missing cast to calculate max_memory
text: perf odp_sched_pktio: try to terminate on failure
linux-gen: drop performance.m4
configure.ac: update version to v1.20.0.0
Petri Savolainen (129):
test: pool_perf: add new pool performance test
test: sched_perf: add new scheduler performance test
linux-gen: ring: remove unnecessary r_tail synchronization
example: sysinfo: application to print system information
linux-gen: cpu: move num cpu into config header file
linux-gen: config: increase max number of cpus
linux-gen: sysinfo: use cpufreq for max freq by default
linux-gen: sysinfo: parse aarch64 cpuinfo
linux-gen: sched: support large burst sizes
linux-gen: sched: support large burst size from pktin
linux-gen: queue: separate plain and sched dequeues
linux-gen: queue: rename queue basic internal header file
linux-gen: sched: move basic queue scheduler functions
linux-gen: sched: configurable default burst size
test: sched_perf: add num queues option
api: queue: split queue spec header file
linux-gen: queue: inline enq and deq functions
linux-gen: queue: remove internal queue handle conversions
linux-gen: queue: remove enq/deq from queue interface
linux-gen: queue: prepare for separate queue operations
linux-gen: queue: separate plain and sched enqueue functions
linux-gen: queue: fix ordered queue issue
test: queue_perf: handle max queue size capability zero
test: queue_perf: prepare for multiple worker threads
test: queue_perf: add num_cpu option
test: queue_perf: add burst_size option
linux-gen: ring_mpmc: new multi-producer, multi-consumer ring
linux-gen: queue: use mpmc ring in plain queues
linux-gen: ring_st: move ring mask and data pointer
linux-gen: ring_spsc: move ring mask and data pointer
test: sched_perf: total number of queues option
test: sched_perf: add event forward option
test: sched_perf: total events per second
linux-gen: queue: remove extra checks
linux-gen: sched: clean up local data struct
linux-gen: sched: single variable for sync context status
linux-gen: sched: remove queue_destroy_finalize callback
linux-gen: sched: stash ring pointer
linux-gen: ring: change ring_deq return value
linux-gen: config: print config file name
linux-gen: config: add array lookup function
linux-gen: sched: per priority burst size configuration
linux-gen: sched: pack global data struct
linux-gen: pktin: use enqueue multi for dst_queue packets
validation: cls: multiple packet tcp dest port test
linux-gen: queue: fix error print format
test: ipc: reduce interface start poll frequency
example: packet_dump: add new example
example: packet_dump: run during make check
linux-gen: dpdk: prefetch pkt_hdr before packet init
linux-gen: dpdk: prefetch packet data early
linux-gen: pool: decrease minimum segment size to 2k
validation: sched: add queue size test
linux-gen: queue: use queue size parameter
test: sched_pktio: add queue pipeline options
test: sched_pktio: add scheduler sync mode option
test: sched_pktio: add pipeline queue size option
example: build: configure option to disable example build
test: build: configure option to disable test build
validation: packet: add packet reset test
helper: iplookup: check capabilities
test: scheduling: fix script to exit with failure status
test: scheduling: honor pool capability
test: sched_latency: honor pool capability limits
linux-gen: sched: remove unnecessary queue null index
linux-gen: pool: ring size must be larger than num items
linux-gen: ring: add reader tail check
linux-gen: pool: reduce max pool size
validation: pool: add max num pool tests
linux-gen: pktio: fix index calculation of multiple dest_queue
validation: cls: interleave tcp test flows
ci: remove iquery tests
linux-gen: sched: remove iquery from interface
linux-gen: remove iquery scheduler implementation
linux-gen: sched: remove unused schedule interface functions
linux-gen: config: improve config file check error output
linux-gen: config: move queue size config to scalable
linux-gen: pool: output error on pool create
linux-gen: pool: add packet param checks
linux-gen: pool: add max num packets in config file
linux-gen: config: maximum pool size 1M
travis: build ODP on multiple threads
travis: add test cases for optional schedulers
travis: split distcheck test
travis: clean process mode test script
travis: explicit build only tests
travis: change pcapng test to build only
travis: exclude duplicate tests
travis: start coverage test early
travis: explicit netmap test cases
travis: run all test with gcc first
linux-gen: sched: add spread weight config file option
linux-gen: sched: use spread weight from config file
linux-gen: sched: increase max spread weight
test: sched_pktio: add burst size option
linux-gen: config: add schedule group config file options
linux-gen: sched: add config request function to interface
linux-gen: thread: use automatic schedule group configuration
linux-gen: schedule_sp: use sched_cb_pktin_poll
linux-gen: pktio: remove sched_cb_pktin_poll_old
linux-gen: ipsec: check crypto param salt length
validation: crypto: capability call should not fail
test: queue_perf: fix lockfree support check
example: sysinfo: print shm blocks
linux-gen: shm: modify shm print header string
linux-gen: cls: simplify shm usage
linux-gen: sysinfo: print out ARM build time features
linux-gen: arm atomic: fix register numbering with casp
linux-gen: build: enable CPU arch specific optimization
api: packet: add subtype call
linux-gen: packet: implementation subtype api
validation: packet: add packet subtype tests
api: packet: add parse result call
linux-gen: packet: implement packet result calls
validation: packet: add num_pkt parameter to parse_test_alloc
validation: packet: add parse result test
api: crypto: improve odp_crypto_session_param_t documentation
api: ipsec: add auth_key_extra IPSEC crypto param
linux-gen: ipsec: use new auth_key_extra
api: sched: add priority min/max/default functions
api: sched: odp_schedule_prio_t is an integer
api: sched: favor priority functions over macros
linux-gen: sched: implement min/max/default prio functions
validation: sched: convert priority macros to function calls
validation: sched: add priority function tests
validation: queue: default queue parameter values
api: sched: add wait and no_wait schedule functions
linux-gen: sched: implement wait and no_wait functions
validation: sched: add test case for wait and no_wait
Seungha Son (3):
linux-gen: timer: add debug print when running out of timer pools
test: odp_pool_per: fix printf string format
linux-gen: shm: fix wrong spelling
-----------------------------------------------------------------------
hooks/post-receive
--
6 years, 2 months
- 1
- 0
[COMMIT] branch master updated. v1.19.0.2_tigermoth-328-gec5066a3
by git＠git-us.linaro.org
This is an automated email from the git hooks/post-receive script. It was
generated because a ref change was pushed to the repository containing
the project "".
The branch, master has been updated
via ec5066a3430e31a87727ac4aea5793253e5ee843 (commit)
via 487e6bd608a78527809ac7b88f0d3d3ec94cd707 (commit)
from fd5939c3ae2a2a38c0a1f87428a787ee7ae00789 (commit)
Those revisions listed above that are new to this repository have
not appeared on any other notification email; so we list those
revisions in full, below.
- Log -----------------------------------------------------------------
commit ec5066a3430e31a87727ac4aea5793253e5ee843
Author: Maxim Uvarov <maxim.uvarov(a)linaro.org>
Date: Thu Nov 29 21:22:27 2018 +0300
configure.ac: update version to v1.20.0.0
Signed-off-by: Maxim Uvarov <maxim.uvarov(a)linaro.org>
Reviewed-by: Bill Fischofer <bill.fischofer(a)linaro.org>
diff --git a/configure.ac b/configure.ac
index 5dd2e777..8efb8525 100644
--- a/configure.ac
+++ b/configure.ac
@@ -3,9 +3,9 @@ AC_PREREQ([2.5])
# Set correct API version
##########################################################################
m4_define([odpapi_generation_version], [1])
-m4_define([odpapi_major_version], [19])
+m4_define([odpapi_major_version], [20])
m4_define([odpapi_minor_version], [0])
-m4_define([odpapi_point_version], [2])
+m4_define([odpapi_point_version], [0])
m4_define([odpapi_version],
[odpapi_generation_version.odpapi_major_version.odpapi_minor_version.odpapi_point_version])
AC_INIT([OpenDataPlane],[odpapi_version],[lng-odp(a)lists.linaro.org])
commit 487e6bd608a78527809ac7b88f0d3d3ec94cd707
Author: Bill Fischofer <bill.fischofer(a)linaro.org>
Date: Fri Nov 30 08:31:55 2018 -0600
changelog: updates for odp v1.20.0.0
Add updates for ODP v1.20.0.0
Signed-off-by: Bill Fischofer <bill.fischofer(a)linaro.org>
Reviewed-by: Dmitry Eremin-Solenikov <dmitry.ereminsolenikov(a)linaro.org>
Signed-off-by: Maxim Uvarov <maxim.uvarov(a)linaro.org>
diff --git a/CHANGELOG b/CHANGELOG
index 7ee5670d..a0bdf558 100644
--- a/CHANGELOG
+++ b/CHANGELOG
@@ -1,3 +1,315 @@
+== OpenDataPlane (1.20.0.0)
+=== Summary of Changes
+ODP v1.20.0.0 is a refresh of ODP, incorporating significant configurability
+and performance improvements as well as new APIs and API restructures.
+
+==== APIs
+===== Symbol `ODP_SHM_NULL` Removed.
+An invalid `odp_shm_t` has the value `ODP_SHM_INVALID`, consistent with other
+ODP types. The legacy synonym `ODP_SHM_NULL` is now removed for consistency.
+
+===== New 3GPP Crypto Algorithm Support
+New support for 3GPP crypto algorithms is added by defining symbols for
+
+* `ODP_CIPHER_ALG_KASUMI_F8`
+* `ODP_CIPHER_ALG_SNOW3G_UEA2`
+* `ODP_CIPHER_ALG_ZUC_EEA3`
+
+In addition new authentication algorithm symbols are defined for
+
+* `ODP_AUTH_ALG_KASUMI_F9`
+* `ODP_AUTH_ALG_SNOW3G_UIA2`
+* `ODP_AUTH_ALG_ZUC_EIA3`
+
+These values are returned as ODP capabilities as well as being accepted in
+crypto session creation for implementations that indicate support for them.
+
+===== Crypto Capability for Bitwise Operation
+The new `bit_mode` capability Boolean is added to the
+`odp_crypto_cipher_capability_t` struct to indicate that an implementation
+supports operating in bit mode. When operating in bit
+mode, field offsets and lengths are expressed in terms of bits rather than
+bytes. However, such lengths must always be specified in multiples of 8.
+
+===== Improved Crypto Spec Documentation
+The ODP crypto API specification is tightened to specify default values for
+cipher and authentication algorithms. Also documented when key and IV
+parameters need to be set.
+
+===== IPsec Extensions
+IPsec requires "salt" (extra keying material) when the GMAC authentication
+algorithm is used. To accommodate this the `auth_key_extra` field is added to
+the `odp_ipsec_crypto_param_t` struct and documentation is added clarifying
+when this field is needed and how it should be used.
+
+===== Classifier Type Rename
+The `odp_pmr_t` type name for an invalid value is renamed from `ODP_PMR_INVAL`
+to `ODP_PMR_INVALID` for consistency with the rest of ODP type names. The old
+symbol is still available when ODP is configured with
+`--enable-deprecated`.
+
+===== New API for Packet Event Subtypes
+The `odp_packet_subtype()` API is added that returns the subtype of a packet
+event directly.
+
+===== Streamlined Packet Parsing Results
+The `odp_packet_parse_result()` API is added that returns the result of
+packet parsing as a single `odp_packet_parse_result_t` struct. This can
+offer efficiency improvements for applications that need all parse results
+rather than making individual parse result calls.
+
+===== PktIO Extensions to Support per-Queue Configuration
+PktIO interfaces support multiple input queues to enable increased parallelism
+in I/O processing. Previously, all of these input queues were required to
+belong to the same scheduler group. The `odp_pktin_queue_param_t` struct is
+now extended with an optional `odp_pktin_queue_param_ovr_t` struct that
+permits individual pktin queues to be assigned to separate scheduler groups.
+This may permit improved performance for advanced application use cases.
+
+===== Timer Pool Capabilities
+The `odp_timer_capability_t` struct is extended to return three additional
+pieces of information:
+
+`max_pools_combined`::
+The total number of timer pools that can be created combining different
+clock sources
+
+`max_pools`::
+The maximum number of timer pools for a given clock source.
+
+`max_timers`::
+The maximum number of timers in a single pool. A zero value means number is
+limited only by available memory.
+
+===== Add Scheduler mix/max/default Priority Functions
+Three new APIs: `odp_schedule_max_prio()`, `odp_schedule_min_prio()`, and
+`odp_schedule_default_prio()` are added that return the min, max, and default
+values specified for the `prio` field in the `odp_schedule_param_t` struct.
+
+With the introduction of these scheduling priority functions the previously
+defined macros (`ODP_SCHED_PRIO_HIGHEST`, `ODP_SCHED_PRIO_NORMAL`, and
+`ODP_SCHED_PRIO_LOWEST`) are now deprecated and should no longer be used.
+
+===== Specification of `odp_schedule_prio_t` as an `int`
+Previously, the `odp_schedule_prio_t` type definition was left to each
+implementation. With the addition of explicit schedule priority ranges, this
+type is now specified to be an `int` to permit efficient implementation
+(including inlining) of these functions.
+
+====== New Scheduler APIs
+The new scheduler APIs `odp_schedule_multi_wait()` and
+`odp_schedule_multi_no_wait()` are added to provide more efficiently
+implementable versions of these functions. The existing scheduler APIs remain
+unchanged. These new APIs can simply provide a fastpath for some
+applications/implementations as an alternative to specifying a parameter on
+`odp_schedule_multi()`.
+
+===== Memory Model in `odp_init_global()`
+The `odp_init_t` parameter passed to `odp_init_global()` is extended to
+add the `mem_model` field. This field is defined by the new `odp_mem_model_t`
+struct and is used to specify whether the application will be using a
+thread (`ODP_MEM_MODEL_THREAD`) or process (`ODP_MEM_MODEL_PROCESS`)
+memory model. The default is a thread model is used for compatibility with
+previous levels of ODP.
+
+==== ABI Changes
+A number of changes to the ODP ABI have also been made in this release to
+improve application binary portability.
+
+===== Strong Typing for Timer Pools
+The `odp_timer_pool_t` is now strongly typed.
+
+===== Consistent Initialization
+The values of the various `ODP_xxx_INVALID` symbols for ODP abstract types in
+the `odp-linux` reference implementation are now consistently zeros. This
+reduces errors and improves portability.
+
+=== Implementation Improvements
+==== Configuration File
+A new configuration file mechanism is introduced that makes use of
+https://www.hyperrealm.com/libconfig/libconfig_manual.html[libconfig] to
+enable various runtime ODP parameters to be specified dynamically.
+
+Default configuration values for the `odp-linux` reference implementation are
+contained in the `config/odp-linux-generic.conf` file. Users may override
+these default values by supplying their own configuration file. At
+`odp_init_global()` time, if the `ODP_CONFIG_FILE` environment variable is set,
+this is used to locate the path to the override configuration file.
+
+==== Process Mode Support
+The `odp-linux` reference implementation now supports applications that run in
+process mode (`mem_model` = `ODP_MEM_MODEL_PROCESS`) as well as the default
+thread mode. This support only applies within a single ODP instance, so any
+`fork()` calls must be done only _after_ `odp_init_global()` has been called
+to initialize ODP on a root process.
+
+==== Removal of `iQuery` Scheduler
+The `iQuery` scheduler is removed from the `odp-linux` reference
+implementation, as it offers no performance advantages and has not seen
+application use.
+
+==== Number of CPUs
+The `odp-linux` reference implementation now supports up to 256 CPUs by
+default (increased from 128).
+
+==== Support for Large Burst Sizes
+The `odp-linux` reference implementation now supports large burst sizes for
+both I/O and non-I/O scheduled events. Large bursts (when available) are
+received directly to the application without any stashing for improved
+throughput. Burst sizes are configurable via the new configuration file
+mechanism, as described above.
+
+==== `--without-openssl` Warnings
+When building `odp-linux` using `--without-openssl` a warning will be issued
+cautioning that strong cryptography will not be available.
+
+==== Inline Queue Enq/Deq APIs
+The various enq/deq APIs are now subject to inlining when `odp-linux` is
+built with `--disable-abi-compat`.
+
+==== Configurable Timer Controls
+Inline timers are now controlled via a config file option. Timer polling
+frequency is similarly controlled via the config file.
+
+==== Huge Page Configuration
+The config file is now used to specify the huge page usage limit.
+
+==== Single and Multi-Consumer/Producer Rings
+The queue implementation in `odp-linux` now automatically makes use of
+optimized single and multi-consumer/producer rings to significantly speed
+up enq/deq processing.
+
+==== `odp_shm_print_all()` Improvements
+The output from `odp_shm_print_all()` is reworked to provide more useful
+and comprehensive shared memory usage information in `odp-linux`.
+
+==== IPsec Improvements
+SA lifetime checking is now more scalable to multiple threads. This
+significantly reduces overhead for multithreaded IPsec applications.
+
+==== Native Builds
+When running in non-ABI compatibility mode, `odp-linux` now enables
+native machine-specific optimizations for the CPU architecture of the
+local machine.
+
+=== Validation Test Improvements
+==== SCTP Test Packets
+SCTP test packets are now used in parser testing. SCTP headers are added to
+ODP and ODP helpers and SCTP checksums are now inserted and verified as part
+of validation testing.
+
+==== `odp_packet_reset()` Test
+The packet validation test suite now properly tests `odp_packet_reset()`.
+
+=== Helper Changes
+In support of process mode, ODP helper functions have been changed to
+better match these new capabilities
+
+==== New `enum`
+The `odph_linux_thread_type_t enum` has been replaced with the new
+`odp_mem_model_t` type.
+
+==== Helper Options
+The new `odph_options()` getter function is added that returns
+applicable options in effect via the new `odph_helper_options_t` struct.
+This currently includes the memory model (thread or process) that is in use.
+
+==== SCTP Helpers
+The new helper APIs `odph_sctp_chksum_set()` and `odph_sctp_chksum_verify()`
+are added to facilitate working with SCTP packet checksums.
+
+=== Performance Test Improvements
+==== Pool Performance
+A new `odp_pool_perf` test has been added that stress-tests ODP pool
+functions in a multithreaded environment to generate performance statistics.
+
+==== Scheduler Performance
+A new `odp_sched_perf` test has been added that stress-tests the scheduler
+in a multithreaded environment.
+
+==== CPU Performance
+A new `odp_cpu_bench` performance test has been added that runs
+compute-intensive packet operations in a multithreaded environment and prints
+the observed maximum throughput for each thread.
+
+=== Example Improvements
+==== Classifier Example changes
+The `odp_classifier` example program now uses a reduced number of threads by
+default to reduce elapsed run time. `ODP_THREAD_COUNT_MAX` is also now used as
+the max worker count.
+
+==== Generator Improvements
+The `odp_generator` example has numerous cleanups and performance improvements.
+
+==== IPsec Example
+The `odp_ipsec` example now properly stops and closes pktio devices on exit.
+
+==== Packet Dumping
+A new `odp_packet_dump` example is added that prints received packets to the
+terminal. This is useful for debugging packet I/O interfaces.
+
+==== Sysinfo Example
+A new `odp_sysinfo` example is provided that prints system information. Useful
+for checking the ODP environment during debugging. This includes providing
+detailed information about the various crypto facilities supported, as well
+as the feature flags used at build time (_e.g.,_ if the binary was built with
+ARMv8.0 or ARMv8.1 instructions).
+
+==== Traffic Manager Example
+The traffic manager example now properly destroys all TM queues it creates
+for improved reliability. It also now always prints a proper termination
+summary message.
+
+=== Bug Fixes
+==== Numbered Bugs/Issues
+===== https://bugs.linaro.org/show_bug.cgi?id=3983[Bug 3983]
+Compile fails on OpenSuSE 42.2 Leap with error: negative width in bit field
+'__error_if_negative'
+
+===== https://bugs.linaro.org/show_bug.cgi?id=3989[Bug 3989]
+odp_system_info_init() issues
+
+===== https://bugs.linaro.org/show_bug.cgi?id=3999[Bug 3999]
+IPsec antireplay check drops packets when sequence number jumps.
+
+===== https://bugs.linaro.org/show_bug.cgi?id=4002[Bug 4002]
+IPsec SA creation must fail for ESN-enabled SAs
+
+===== https://bugs.linaro.org/show_bug.cgi?id=4013[Bug 4013]
+Per-SA IPv4 ID allocation may cause duplicate IDs.
+
+===== https://bugs.linaro.org/show_bug.cgi?id=4017[Bug 4017]
+Unexpected IP ID causes IPsec API validation to fail
+
+===== https://github.com/Linaro/odp/issues/662[Issue 662]
+rte_mempool_ops_alloc() is not dpdk api
+
+==== Unnumbered Bugs/Issues
+* Fixed enq/deq issues encountered on architectures with weak memory ordering.
+* Return 0 from `odp_sys_huge_page_size_all()` if hugepages are not
+supported/detected. Tests modified to not treat this as an error.
+* Set `ODP_CACHE_LINE_SIZE` to 128 on ppc64le systems.
+* iplookuptable fix putting values into table
+* DPDK pktio support now works properly across multiple ODP instances.
+* Zero timer pool memory on reserve (fixes timer failures due to uninitialized
+variables).
+* `-march=native` disabled for `clang`. This fixes a known issue with recent
+levels of clang.
+
+=== Known Issues
+==== https://bugs.linaro.org/show_bug.cgi?id=3998[Bug 3998]
+IPsec extended sequence number support is missing
+
+==== https://bugs.linaro.org/show_bug.cgi?id=4014[Bug 4014]
+Separate IP ID allocation for transport and tunnel mode SAs may cause
+duplicate IDs
+
+==== https://bugs.linaro.org/show_bug.cgi?id=4018[Bug 4018]
+Unexpected IV causes IPsec API validation to fail
+
+==== https://bugs.linaro.org/show_bug.cgi?id=4040[Bug 4040]
+Clang build fails on Ubuntu 18.04
+
== OpenDataPlane (1.19.0.2)
=== Summary of Changes
ODP v1.19.0.2 is the second service update for the Tiger Moth release. It
-----------------------------------------------------------------------
Summary of changes:
CHANGELOG | 312 +++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
configure.ac | 4 +-
2 files changed, 314 insertions(+), 2 deletions(-)
hooks/post-receive
--
6 years, 2 months
- 1
- 0
[COMMIT] branch master updated. v1.19.0.2_tigermoth-326-gfd5939c3
by git＠git-us.linaro.org
This is an automated email from the git hooks/post-receive script. It was
generated because a ref change was pushed to the repository containing
the project "".
The branch, master has been updated
via fd5939c3ae2a2a38c0a1f87428a787ee7ae00789 (commit)
via 94e47dc62e340818b91c471788c29af3ba167d96 (commit)
via 92e59d9e816a99db318ba24dcb12cb55f2e7392d (commit)
via b1812f17ae652f11ce21f26fd24c8fd27818339b (commit)
via 23102db002f522cc90d1b616e2725d21e525b1fc (commit)
via 049d80427d0145a3c1738d28ba595717ae43d5c2 (commit)
via 46d507adef3902a26b7e311506437211e7417a10 (commit)
via e7ad8003e34195a3900e1dd3d3a93235896d7628 (commit)
via 294856cc30d48d57e12485076bae49da36d346ed (commit)
via 6b79ac4b1640e8050b076ba0ecb590cc297320b0 (commit)
via 3cb35813da911a94eef6e07ae71ce0f5f325ebd8 (commit)
via 3aad0e2ce0e5901fd49e50e26ac7d762c2b9a6aa (commit)
via 98eb7327113fbd33a8e5448406e8f47d8d0ad5fb (commit)
via d64232f45abae8d4f1222313ce44532cc26e2336 (commit)
from 536cce998e84a559e125b4741d00f2a760a0d575 (commit)
Those revisions listed above that are new to this repository have
not appeared on any other notification email; so we list those
revisions in full, below.
- Log -----------------------------------------------------------------
commit fd5939c3ae2a2a38c0a1f87428a787ee7ae00789
Author: Matias Elo <matias.elo(a)nokia.com>
Date: Fri Nov 23 14:03:53 2018 +0200
linux-gen: traffic_mngr: allocate all global data from shm
Signed-off-by: Matias Elo <matias.elo(a)nokia.com>
Reviewed-by: Bill Fischofer <bill.fischofer(a)linaro.org>
Signed-off-by: Maxim Uvarov <maxim.uvarov(a)linaro.org>
diff --git a/platform/linux-generic/odp_traffic_mngr.c b/platform/linux-generic/odp_traffic_mngr.c
index e29cb3cd..23982dce 100644
--- a/platform/linux-generic/odp_traffic_mngr.c
+++ b/platform/linux-generic/odp_traffic_mngr.c
@@ -46,7 +46,7 @@ static const pkt_desc_t EMPTY_PKT_DESC = { .word = 0 };
#define MAX_PRIORITIES ODP_TM_MAX_PRIORITIES
#define NUM_SHAPER_COLORS ODP_NUM_SHAPER_COLORS
-static tm_prop_t basic_prop_tbl[MAX_PRIORITIES][NUM_SHAPER_COLORS] = {
+static const tm_prop_t basic_prop_tbl[MAX_PRIORITIES][NUM_SHAPER_COLORS] = {
[0] = {
[ODP_TM_SHAPER_GREEN] = { 0, DECR_BOTH },
[ODP_TM_SHAPER_YELLOW] = { 0, DECR_BOTH },
@@ -106,6 +106,8 @@ typedef struct {
} profile_tbl_t;
typedef struct {
+ tm_system_t system[ODP_TM_MAX_NUM_SYSTEMS];
+
struct {
tm_system_group_t group[ODP_TM_MAX_NUM_SYSTEMS];
odp_ticketlock_t lock;
@@ -121,20 +123,24 @@ typedef struct {
profile_tbl_t profile_tbl;
- odp_shm_t shm;
-} tm_global_t;
+ odp_ticketlock_t create_lock;
+ odp_ticketlock_t profile_lock;
+ odp_barrier_t first_enq;
-static tm_global_t *tm_glb;
+ int main_thread_cpu;
+ int cpu_num;
-/* TM systems table. */
-static tm_system_t odp_tm_systems[ODP_TM_MAX_NUM_SYSTEMS];
+ /* Service threads */
+ uint64_t busy_wait_counter;
+ odp_bool_t main_loop_running;
+ odp_atomic_u64_t atomic_request_cnt;
+ odp_atomic_u64_t currently_serving_cnt;
+ odp_atomic_u64_t atomic_done_cnt;
-static odp_ticketlock_t tm_create_lock;
-static odp_ticketlock_t tm_profile_lock;
-static odp_barrier_t tm_first_enq;
+ odp_shm_t shm;
+} tm_global_t;
-static int g_main_thread_cpu = -1;
-static int g_tm_cpu_num;
+static tm_global_t *tm_glb;
/* Forward function declarations. */
static void tm_queue_cnts_decrement(tm_system_t *tm_system,
@@ -472,8 +478,8 @@ static tm_system_t *tm_system_alloc(void)
/* Find an open slot in the odp_tm_systems array. */
for (tm_idx = 0; tm_idx < ODP_TM_MAX_NUM_SYSTEMS; tm_idx++) {
- if (odp_tm_systems[tm_idx].status == TM_STATUS_FREE) {
- tm_system = &odp_tm_systems[tm_idx];
+ if (tm_glb->system[tm_idx].status == TM_STATUS_FREE) {
+ tm_system = &tm_glb->system[tm_idx];
memset(tm_system, 0, sizeof(tm_system_t));
tm_system->tm_idx = tm_idx;
tm_system->status = TM_STATUS_RESERVED;
@@ -486,7 +492,7 @@ static tm_system_t *tm_system_alloc(void)
static void tm_system_free(tm_system_t *tm_system)
{
- odp_tm_systems[tm_system->tm_idx].status = TM_STATUS_FREE;
+ tm_glb->system[tm_system->tm_idx].status = TM_STATUS_FREE;
}
static void *tm_common_profile_create(const char *name,
@@ -2373,31 +2379,24 @@ static int tm_process_expired_timers(tm_system_t *tm_system,
return work_done;
}
-static volatile uint64_t busy_wait_counter;
-
-static odp_bool_t main_loop_running;
-static odp_atomic_u64_t atomic_request_cnt;
-static odp_atomic_u64_t currently_serving_cnt;
-static odp_atomic_u64_t atomic_done_cnt;
-
static void busy_wait(uint32_t iterations)
{
uint32_t cnt;
for (cnt = 1; cnt <= iterations; cnt++)
- busy_wait_counter++;
+ tm_glb->busy_wait_counter++;
}
static void signal_request(void)
{
- uint64_t my_request_num, serving_cnt;
+ uint64_t request_num, serving;
- my_request_num = odp_atomic_fetch_inc_u64(&atomic_request_cnt) + 1;
+ request_num = odp_atomic_fetch_inc_u64(&tm_glb->atomic_request_cnt) + 1;
- serving_cnt = odp_atomic_load_u64(¤tly_serving_cnt);
- while (serving_cnt != my_request_num) {
+ serving = odp_atomic_load_u64(&tm_glb->currently_serving_cnt);
+ while (serving != request_num) {
busy_wait(100);
- serving_cnt = odp_atomic_load_u64(¤tly_serving_cnt);
+ serving = odp_atomic_load_u64(&tm_glb->currently_serving_cnt);
}
}
@@ -2405,26 +2404,26 @@ static void check_for_request(void)
{
uint64_t request_num, serving_cnt, done_cnt;
- request_num = odp_atomic_load_u64(&atomic_request_cnt);
- serving_cnt = odp_atomic_load_u64(¤tly_serving_cnt);
+ request_num = odp_atomic_load_u64(&tm_glb->atomic_request_cnt);
+ serving_cnt = odp_atomic_load_u64(&tm_glb->currently_serving_cnt);
if (serving_cnt == request_num)
return;
/* Signal the other requesting thread to proceed and then
* wait for their done indication */
- odp_atomic_inc_u64(¤tly_serving_cnt);
+ odp_atomic_inc_u64(&tm_glb->currently_serving_cnt);
busy_wait(100);
- done_cnt = odp_atomic_load_u64(&atomic_done_cnt);
+ done_cnt = odp_atomic_load_u64(&tm_glb->atomic_done_cnt);
while (done_cnt != request_num) {
busy_wait(100);
- done_cnt = odp_atomic_load_u64(&atomic_done_cnt);
+ done_cnt = odp_atomic_load_u64(&tm_glb->atomic_done_cnt);
}
}
static void signal_request_done(void)
{
- odp_atomic_inc_u64(&atomic_done_cnt);
+ odp_atomic_inc_u64(&tm_glb->atomic_done_cnt);
}
static int thread_affinity_get(odp_cpumask_t *odp_cpu_mask)
@@ -2470,7 +2469,7 @@ static void *tm_system_thread(void *arg)
/* Wait here until we have seen the first enqueue operation. */
odp_barrier_wait(&tm_group->tm_group_barrier);
- main_loop_running = true;
+ tm_glb->main_loop_running = true;
destroying = odp_atomic_load_u64(&tm_system->destroying);
@@ -2674,7 +2673,7 @@ static int affinitize_main_thread(void)
* just record this value and return. */
cpu_count = odp_cpumask_count(&odp_cpu_mask);
if (cpu_count == 1) {
- g_main_thread_cpu = odp_cpumask_first(&odp_cpu_mask);
+ tm_glb->main_thread_cpu = odp_cpumask_first(&odp_cpu_mask);
return 0;
} else if (cpu_count == 0) {
return -1;
@@ -2686,7 +2685,7 @@ static int affinitize_main_thread(void)
CPU_SET(cpu_num, &linux_cpu_set);
rc = sched_setaffinity(0, sizeof(cpu_set_t), &linux_cpu_set);
if (rc == 0)
- g_main_thread_cpu = cpu_num;
+ tm_glb->main_thread_cpu = cpu_num;
else
ODP_DBG("%s sched_setaffinity failed with rc=%d\n",
__func__, rc);
@@ -2699,32 +2698,32 @@ static uint32_t tm_thread_cpu_select(void)
int cpu_count, cpu;
odp_cpumask_default_worker(&odp_cpu_mask, 0);
- if ((g_main_thread_cpu != -1) &&
- odp_cpumask_isset(&odp_cpu_mask, g_main_thread_cpu))
- odp_cpumask_clr(&odp_cpu_mask, g_main_thread_cpu);
+ if ((tm_glb->main_thread_cpu != -1) &&
+ odp_cpumask_isset(&odp_cpu_mask, tm_glb->main_thread_cpu))
+ odp_cpumask_clr(&odp_cpu_mask, tm_glb->main_thread_cpu);
cpu_count = odp_cpumask_count(&odp_cpu_mask);
if (cpu_count < 1) {
odp_cpumask_all_available(&odp_cpu_mask);
- if ((g_main_thread_cpu != -1) &&
- odp_cpumask_isset(&odp_cpu_mask, g_main_thread_cpu))
+ if ((tm_glb->main_thread_cpu != -1) &&
+ odp_cpumask_isset(&odp_cpu_mask, tm_glb->main_thread_cpu))
cpu_count = odp_cpumask_count(&odp_cpu_mask);
if (cpu_count < 1)
odp_cpumask_all_available(&odp_cpu_mask);
}
- if (g_tm_cpu_num == 0) {
+ if (tm_glb->cpu_num == 0) {
cpu = odp_cpumask_first(&odp_cpu_mask);
} else {
- cpu = odp_cpumask_next(&odp_cpu_mask, g_tm_cpu_num);
+ cpu = odp_cpumask_next(&odp_cpu_mask, tm_glb->cpu_num);
if (cpu == -1) {
- g_tm_cpu_num = 0;
+ tm_glb->cpu_num = 0;
cpu = odp_cpumask_first(&odp_cpu_mask);
}
}
- g_tm_cpu_num++;
+ tm_glb->cpu_num++;
return cpu;
}
@@ -2760,8 +2759,8 @@ static void _odp_tm_group_destroy(_odp_tm_group_t odp_tm_group)
rc = pthread_join(tm_group->thread, NULL);
ODP_ASSERT(rc == 0);
pthread_attr_destroy(&tm_group->attr);
- if (g_tm_cpu_num > 0)
- g_tm_cpu_num--;
+ if (tm_glb->cpu_num > 0)
+ tm_glb->cpu_num--;
odp_ticketlock_lock(&tm_glb->system_group.lock);
tm_group->status = TM_STATUS_FREE;
@@ -2931,10 +2930,10 @@ odp_tm_t odp_tm_create(const char *name,
return ODP_TM_INVALID;
/* Allocate tm_system_t record. */
- odp_ticketlock_lock(&tm_create_lock);
+ odp_ticketlock_lock(&tm_glb->create_lock);
tm_system = tm_system_alloc();
if (!tm_system) {
- odp_ticketlock_unlock(&tm_create_lock);
+ odp_ticketlock_unlock(&tm_glb->create_lock);
return ODP_TM_INVALID;
}
@@ -2942,7 +2941,7 @@ odp_tm_t odp_tm_create(const char *name,
name_tbl_id = _odp_int_name_tbl_add(name, ODP_TM_HANDLE, odp_tm);
if (name_tbl_id == ODP_INVALID_NAME) {
tm_system_free(tm_system);
- odp_ticketlock_unlock(&tm_create_lock);
+ odp_ticketlock_unlock(&tm_glb->create_lock);
return ODP_TM_INVALID;
}
@@ -3023,11 +3022,11 @@ odp_tm_t odp_tm_create(const char *name,
tm_system->_odp_int_timer_wheel);
tm_system_free(tm_system);
- odp_ticketlock_unlock(&tm_create_lock);
+ odp_ticketlock_unlock(&tm_glb->create_lock);
return ODP_TM_INVALID;
}
- odp_ticketlock_unlock(&tm_create_lock);
+ odp_ticketlock_unlock(&tm_glb->create_lock);
return odp_tm;
}
@@ -3274,7 +3273,7 @@ int odp_tm_shaper_params_update(odp_tm_shaper_t shaper_profile,
if (!profile_obj)
return -1;
- if (!main_loop_running) {
+ if (!tm_glb->main_loop_running) {
tm_shaper_params_cvt_to(params, profile_obj);
return 0;
}
@@ -3399,7 +3398,7 @@ int odp_tm_sched_params_update(odp_tm_sched_t sched_profile,
if (!profile_obj)
return -1;
- if (!main_loop_running) {
+ if (!tm_glb->main_loop_running) {
tm_sched_params_cvt_to(params, profile_obj);
return 0;
}
@@ -3497,7 +3496,7 @@ int odp_tm_thresholds_params_update(odp_tm_threshold_t threshold_profile,
if (!profile_obj)
return -1;
- if (!main_loop_running) {
+ if (!tm_glb->main_loop_running) {
profile_obj->max_pkts =
params->enable_max_pkts ? params->max_pkts : 0;
profile_obj->max_bytes =
@@ -3615,7 +3614,7 @@ int odp_tm_wred_params_update(odp_tm_wred_t wred_profile,
if (!wred_params)
return -1;
- if (!main_loop_running) {
+ if (!tm_glb->main_loop_running) {
tm_wred_params_cvt_to(params, wred_params);
return 0;
}
@@ -3757,7 +3756,7 @@ int odp_tm_node_destroy(odp_tm_node_t tm_node)
if (!tm_node_obj)
return -1;
- tm_system = &odp_tm_systems[tm_node_obj->tm_idx];
+ tm_system = &tm_glb->system[tm_node_obj->tm_idx];
if (!tm_system)
return -1;
@@ -3820,14 +3819,14 @@ int odp_tm_node_shaper_config(odp_tm_node_t tm_node,
if (!tm_node_obj)
return -1;
- tm_system = &odp_tm_systems[tm_node_obj->tm_idx];
+ tm_system = &tm_glb->system[tm_node_obj->tm_idx];
if (!tm_system)
return -1;
- odp_ticketlock_lock(&tm_profile_lock);
+ odp_ticketlock_lock(&tm_glb->profile_lock);
tm_shaper_config_set(tm_system, shaper_profile,
&tm_node_obj->shaper_obj);
- odp_ticketlock_unlock(&tm_profile_lock);
+ odp_ticketlock_unlock(&tm_glb->profile_lock);
return 0;
}
@@ -3846,10 +3845,10 @@ int odp_tm_node_sched_config(odp_tm_node_t tm_node,
if (!child_tm_node_obj)
return -1;
- odp_ticketlock_lock(&tm_profile_lock);
+ odp_ticketlock_lock(&tm_glb->profile_lock);
child_shaper_obj = &child_tm_node_obj->shaper_obj;
tm_sched_config_set(child_shaper_obj, sched_profile);
- odp_ticketlock_unlock(&tm_profile_lock);
+ odp_ticketlock_unlock(&tm_glb->profile_lock);
return 0;
}
@@ -3862,9 +3861,9 @@ int odp_tm_node_threshold_config(odp_tm_node_t tm_node,
if (!tm_node_obj)
return -1;
- odp_ticketlock_lock(&tm_profile_lock);
+ odp_ticketlock_lock(&tm_glb->profile_lock);
tm_threshold_config_set(&tm_node_obj->tm_wred_node, thresholds_profile);
- odp_ticketlock_unlock(&tm_profile_lock);
+ odp_ticketlock_unlock(&tm_glb->profile_lock);
return 0;
}
@@ -3883,7 +3882,7 @@ int odp_tm_node_wred_config(odp_tm_node_t tm_node,
wred_node = &tm_node_obj->tm_wred_node;
- odp_ticketlock_lock(&tm_profile_lock);
+ odp_ticketlock_lock(&tm_glb->profile_lock);
rc = 0;
if (pkt_color == ODP_PACKET_ALL_COLORS) {
for (color = 0; color < ODP_NUM_PACKET_COLORS; color++)
@@ -3894,7 +3893,7 @@ int odp_tm_node_wred_config(odp_tm_node_t tm_node,
rc = -1;
}
- odp_ticketlock_unlock(&tm_profile_lock);
+ odp_ticketlock_unlock(&tm_glb->profile_lock);
return rc;
}
@@ -4032,7 +4031,7 @@ int odp_tm_queue_destroy(odp_tm_queue_t tm_queue)
if (!tm_queue_obj)
return -1;
- tm_system = &odp_tm_systems[tm_queue_obj->tm_idx];
+ tm_system = &tm_glb->system[tm_queue_obj->tm_idx];
if (!tm_system)
return -1;
@@ -4090,14 +4089,14 @@ int odp_tm_queue_shaper_config(odp_tm_queue_t tm_queue,
if (!tm_queue_obj)
return -1;
- tm_system = &odp_tm_systems[tm_queue_obj->tm_idx];
+ tm_system = &tm_glb->system[tm_queue_obj->tm_idx];
if (!tm_system)
return -1;
- odp_ticketlock_lock(&tm_profile_lock);
+ odp_ticketlock_lock(&tm_glb->profile_lock);
tm_shaper_config_set(tm_system, shaper_profile,
&tm_queue_obj->shaper_obj);
- odp_ticketlock_unlock(&tm_profile_lock);
+ odp_ticketlock_unlock(&tm_glb->profile_lock);
return 0;
}
@@ -4117,10 +4116,10 @@ int odp_tm_queue_sched_config(odp_tm_node_t tm_node,
if (!child_tm_queue_obj)
return -1;
- odp_ticketlock_lock(&tm_profile_lock);
+ odp_ticketlock_lock(&tm_glb->profile_lock);
child_shaper_obj = &child_tm_queue_obj->shaper_obj;
tm_sched_config_set(child_shaper_obj, sched_profile);
- odp_ticketlock_unlock(&tm_profile_lock);
+ odp_ticketlock_unlock(&tm_glb->profile_lock);
return 0;
}
@@ -4134,10 +4133,10 @@ int odp_tm_queue_threshold_config(odp_tm_queue_t tm_queue,
if (!tm_queue_obj)
return -1;
- odp_ticketlock_lock(&tm_profile_lock);
+ odp_ticketlock_lock(&tm_glb->profile_lock);
ret = tm_threshold_config_set(&tm_queue_obj->tm_wred_node,
thresholds_profile);
- odp_ticketlock_unlock(&tm_profile_lock);
+ odp_ticketlock_unlock(&tm_glb->profile_lock);
return ret;
}
@@ -4156,7 +4155,7 @@ int odp_tm_queue_wred_config(odp_tm_queue_t tm_queue,
wred_node = &tm_queue_obj->tm_wred_node;
- odp_ticketlock_lock(&tm_profile_lock);
+ odp_ticketlock_lock(&tm_glb->profile_lock);
rc = 0;
if (pkt_color == ODP_PACKET_ALL_COLORS) {
for (color = 0; color < ODP_NUM_PACKET_COLORS; color++)
@@ -4167,7 +4166,7 @@ int odp_tm_queue_wred_config(odp_tm_queue_t tm_queue,
rc = -1;
}
- odp_ticketlock_unlock(&tm_profile_lock);
+ odp_ticketlock_unlock(&tm_glb->profile_lock);
return rc;
}
@@ -4232,7 +4231,7 @@ int odp_tm_node_connect(odp_tm_node_t src_tm_node, odp_tm_node_t dst_tm_node)
if ((!src_tm_node_obj) || src_tm_node_obj->is_root_node)
return -1;
- tm_system = &odp_tm_systems[src_tm_node_obj->tm_idx];
+ tm_system = &tm_glb->system[src_tm_node_obj->tm_idx];
if (!tm_system)
return -1;
@@ -4302,7 +4301,7 @@ int odp_tm_queue_connect(odp_tm_queue_t tm_queue, odp_tm_node_t dst_tm_node)
if (!src_tm_queue_obj)
return -1;
- tm_system = &odp_tm_systems[src_tm_queue_obj->tm_idx];
+ tm_system = &tm_glb->system[src_tm_queue_obj->tm_idx];
if (!tm_system)
return -1;
@@ -4366,7 +4365,7 @@ int odp_tm_enq(odp_tm_queue_t tm_queue, odp_packet_t pkt)
if (!tm_queue_obj)
return -1;
- tm_system = &odp_tm_systems[tm_queue_obj->tm_idx];
+ tm_system = &tm_glb->system[tm_queue_obj->tm_idx];
if (!tm_system)
return -1;
@@ -4387,7 +4386,7 @@ int odp_tm_enq_with_cnt(odp_tm_queue_t tm_queue, odp_packet_t pkt)
if (!tm_queue_obj)
return -1;
- tm_system = &odp_tm_systems[tm_queue_obj->tm_idx];
+ tm_system = &tm_glb->system[tm_queue_obj->tm_idx];
if (!tm_system)
return -1;
@@ -4653,11 +4652,11 @@ int odp_tm_priority_threshold_config(odp_tm_t odp_tm,
if (thresholds_profile == ODP_TM_INVALID)
return -1;
- odp_ticketlock_lock(&tm_profile_lock);
+ odp_ticketlock_lock(&tm_glb->profile_lock);
tm_system->priority_info[priority].threshold_params =
tm_get_profile_params(thresholds_profile,
TM_THRESHOLD_PROFILE);
- odp_ticketlock_unlock(&tm_profile_lock);
+ odp_ticketlock_unlock(&tm_glb->profile_lock);
return 0;
}
@@ -4670,10 +4669,10 @@ int odp_tm_total_threshold_config(odp_tm_t odp_tm,
if (thresholds_profile == ODP_TM_INVALID)
return -1;
- odp_ticketlock_lock(&tm_profile_lock);
+ odp_ticketlock_lock(&tm_glb->profile_lock);
tm_system->total_info.threshold_params = tm_get_profile_params(
thresholds_profile, TM_THRESHOLD_PROFILE);
- odp_ticketlock_unlock(&tm_profile_lock);
+ odp_ticketlock_unlock(&tm_glb->profile_lock);
return 0;
}
@@ -4737,21 +4736,22 @@ int odp_tm_init_global(void)
memset(tm_glb, 0, sizeof(tm_global_t));
tm_glb->shm = shm;
+ tm_glb->main_thread_cpu = -1;
odp_ticketlock_init(&tm_glb->queue_obj.lock);
odp_ticketlock_init(&tm_glb->node_obj.lock);
odp_ticketlock_init(&tm_glb->system_group.lock);
- odp_ticketlock_init(&tm_create_lock);
- odp_ticketlock_init(&tm_profile_lock);
+ odp_ticketlock_init(&tm_glb->create_lock);
+ odp_ticketlock_init(&tm_glb->profile_lock);
odp_ticketlock_init(&tm_glb->profile_tbl.sched.lock);
odp_ticketlock_init(&tm_glb->profile_tbl.shaper.lock);
odp_ticketlock_init(&tm_glb->profile_tbl.threshold.lock);
odp_ticketlock_init(&tm_glb->profile_tbl.wred.lock);
- odp_barrier_init(&tm_first_enq, 2);
+ odp_barrier_init(&tm_glb->first_enq, 2);
- odp_atomic_init_u64(&atomic_request_cnt, 0);
- odp_atomic_init_u64(¤tly_serving_cnt, 0);
- odp_atomic_init_u64(&atomic_done_cnt, 0);
+ odp_atomic_init_u64(&tm_glb->atomic_request_cnt, 0);
+ odp_atomic_init_u64(&tm_glb->currently_serving_cnt, 0);
+ odp_atomic_init_u64(&tm_glb->atomic_done_cnt, 0);
return 0;
}
commit 94e47dc62e340818b91c471788c29af3ba167d96
Author: Matias Elo <matias.elo(a)nokia.com>
Date: Fri Nov 23 13:55:55 2018 +0200
linux-gen: traffic_mngr: don't reserve memory if tm is disabled
Signed-off-by: Matias Elo <matias.elo(a)nokia.com>
Reviewed-by: Bill Fischofer <bill.fischofer(a)linaro.org>
Signed-off-by: Maxim Uvarov <maxim.uvarov(a)linaro.org>
diff --git a/platform/linux-generic/odp_traffic_mngr.c b/platform/linux-generic/odp_traffic_mngr.c
index 2b7e58f7..e29cb3cd 100644
--- a/platform/linux-generic/odp_traffic_mngr.c
+++ b/platform/linux-generic/odp_traffic_mngr.c
@@ -2918,6 +2918,11 @@ odp_tm_t odp_tm_create(const char *name,
uint32_t max_tm_queues, max_sorted_lists;
int rc;
+ if (odp_global_ro.init_param.not_used.feat.tm) {
+ ODP_ERR("TM has been disabled\n");
+ return ODP_TM_INVALID;
+ }
+
/* If we are using pktio output (usual case) get the first associated
* pktout_queue for this pktio and fail if there isn't one.
*/
@@ -4719,6 +4724,11 @@ int odp_tm_init_global(void)
{
odp_shm_t shm;
+ if (odp_global_ro.init_param.not_used.feat.tm) {
+ ODP_DBG("TM disabled\n");
+ return 0;
+ }
+
shm = odp_shm_reserve("_odp_traffic_mng", sizeof(tm_global_t), 0, 0);
if (shm == ODP_SHM_INVALID)
return -1;
@@ -4747,6 +4757,9 @@ int odp_tm_init_global(void)
int odp_tm_term_global(void)
{
+ if (odp_global_ro.init_param.not_used.feat.tm)
+ return 0;
+
if (odp_shm_free(tm_glb->shm)) {
ODP_ERR("shm free failed\n");
return -1;
commit 92e59d9e816a99db318ba24dcb12cb55f2e7392d
Author: Matias Elo <matias.elo(a)nokia.com>
Date: Fri Nov 23 13:32:28 2018 +0200
linux-gen: traffic_mngr: allocate profile objects from shm
Signed-off-by: Matias Elo <matias.elo(a)nokia.com>
Reviewed-by: Bill Fischofer <bill.fischofer(a)linaro.org>
Signed-off-by: Maxim Uvarov <maxim.uvarov(a)linaro.org>
diff --git a/platform/linux-generic/include/odp_traffic_mngr_internal.h b/platform/linux-generic/include/odp_traffic_mngr_internal.h
index 587f9d1a..040bb117 100644
--- a/platform/linux-generic/include/odp_traffic_mngr_internal.h
+++ b/platform/linux-generic/include/odp_traffic_mngr_internal.h
@@ -83,13 +83,6 @@ typedef uint64_t tm_handle_t;
#define PF_REACHED_EGRESS 0x40
#define PF_ERROR 0x80
-typedef struct {
- uint32_t num_allocd;
- uint32_t num_used;
- uint32_t num_freed;
- void **array_ptrs; /* Ptr to an array of num_allocd void * ptrs. */
-} dynamic_tbl_t;
-
#define ODP_TM_NUM_PROFILES 4
typedef enum {
@@ -115,6 +108,7 @@ typedef struct {
_odp_int_name_t name_tbl_id;
odp_tm_threshold_t thresholds_profile;
uint32_t ref_cnt;
+ tm_status_t status;
} tm_queue_thresholds_t;
typedef struct {
@@ -127,6 +121,7 @@ typedef struct {
odp_tm_percent_t max_drop_prob;
odp_bool_t enable_wred;
odp_bool_t use_byte_fullness;
+ tm_status_t status;
} tm_wred_params_t;
typedef struct {
@@ -165,6 +160,7 @@ typedef struct {
uint32_t ref_cnt;
odp_tm_sched_mode_t sched_modes[ODP_TM_MAX_PRIORITIES];
uint16_t inverted_weights[ODP_TM_MAX_PRIORITIES];
+ tm_status_t status;
} tm_sched_params_t;
typedef enum {
@@ -200,6 +196,7 @@ typedef struct {
int8_t len_adjust;
odp_bool_t dual_rate;
odp_bool_t enabled;
+ tm_status_t status;
} tm_shaper_params_t;
typedef enum { NO_CALLBACK, UNDELAY_PKT } tm_shaper_callback_reason_t;
diff --git a/platform/linux-generic/odp_traffic_mngr.c b/platform/linux-generic/odp_traffic_mngr.c
index 416e7376..2b7e58f7 100644
--- a/platform/linux-generic/odp_traffic_mngr.c
+++ b/platform/linux-generic/odp_traffic_mngr.c
@@ -81,6 +81,30 @@ static tm_prop_t basic_prop_tbl[MAX_PRIORITIES][NUM_SHAPER_COLORS] = {
[ODP_TM_SHAPER_RED] = { 7, DELAY_PKT } }
};
+#define MAX_SHAPER_PROFILES 128
+#define MAX_SCHED_PROFILES 128
+#define MAX_THRESHOLD_PROFILES 128
+#define MAX_WRED_PROFILES 128
+
+typedef struct {
+ struct {
+ tm_shaper_params_t profile[MAX_SHAPER_PROFILES];
+ odp_ticketlock_t lock;
+ } shaper;
+ struct {
+ tm_sched_params_t profile[MAX_SCHED_PROFILES];
+ odp_ticketlock_t lock;
+ } sched;
+ struct {
+ tm_queue_thresholds_t profile[MAX_THRESHOLD_PROFILES];
+ odp_ticketlock_t lock;
+ } threshold;
+ struct {
+ tm_wred_params_t profile[MAX_WRED_PROFILES];
+ odp_ticketlock_t lock;
+ } wred;
+} profile_tbl_t;
+
typedef struct {
struct {
tm_system_group_t group[ODP_TM_MAX_NUM_SYSTEMS];
@@ -95,14 +119,13 @@ typedef struct {
odp_ticketlock_t lock;
} node_obj;
+ profile_tbl_t profile_tbl;
+
odp_shm_t shm;
} tm_global_t;
static tm_global_t *tm_glb;
-/* Profile tables. */
-static dynamic_tbl_t odp_tm_profile_tbls[ODP_TM_NUM_PROFILES];
-
/* TM systems table. */
static tm_system_t odp_tm_systems[ODP_TM_MAX_NUM_SYSTEMS];
@@ -245,62 +268,125 @@ static odp_bool_t tm_random_drop(tm_random_data_t *tm_random_data,
return drop;
}
-static void *alloc_entry_in_dynamic_tbl(dynamic_tbl_t *dynamic_tbl,
- uint32_t record_size,
- uint32_t *dynamic_idx_ptr)
+static void *alloc_entry_in_tbl(profile_tbl_t *profile_tbl,
+ profile_kind_t profile_kind,
+ uint32_t *idx)
{
- uint32_t num_allocd, new_num_allocd, idx;
- void **new_array_ptrs, *new_record;
-
- num_allocd = dynamic_tbl->num_allocd;
- if (num_allocd <= dynamic_tbl->num_used) {
- /* Need to alloc or realloc the array of ptrs. */
- if (num_allocd <= 32)
- new_num_allocd = 64;
- else
- new_num_allocd = 4 * num_allocd;
+ uint32_t i;
- new_array_ptrs = malloc(new_num_allocd * sizeof(void *));
- memset(new_array_ptrs, 0, new_num_allocd * sizeof(void *));
+ switch (profile_kind) {
+ case TM_SHAPER_PROFILE: {
+ tm_shaper_params_t *profile = NULL;
- if (dynamic_tbl->num_used != 0)
- memcpy(new_array_ptrs, dynamic_tbl->array_ptrs,
- dynamic_tbl->num_used * sizeof(void *));
+ odp_ticketlock_lock(&profile_tbl->shaper.lock);
+ for (i = 0; i < MAX_SHAPER_PROFILES; i++) {
+ if (profile_tbl->shaper.profile[i].status !=
+ TM_STATUS_FREE)
+ continue;
- if (dynamic_tbl->array_ptrs)
- free(dynamic_tbl->array_ptrs);
+ profile = &profile_tbl->shaper.profile[i];
+ memset(profile, 0, sizeof(tm_shaper_params_t));
+ profile->status = TM_STATUS_RESERVED;
+ *idx = i;
+ break;
+ }
+ odp_ticketlock_unlock(&profile_tbl->shaper.lock);
+ return profile;
+ }
+ case TM_SCHED_PROFILE: {
+ tm_sched_params_t *profile = NULL;
+
+ odp_ticketlock_lock(&profile_tbl->sched.lock);
+ for (i = 0; i < MAX_SCHED_PROFILES; i++) {
+ if (profile_tbl->sched.profile[i].status !=
+ TM_STATUS_FREE)
+ continue;
+
+ profile = &profile_tbl->sched.profile[i];
+ memset(profile, 0, sizeof(tm_sched_params_t));
+ profile->status = TM_STATUS_RESERVED;
+ *idx = i;
+ break;
+ }
+ odp_ticketlock_unlock(&profile_tbl->sched.lock);
+ return profile;
+ }
+ case TM_THRESHOLD_PROFILE: {
+ tm_queue_thresholds_t *profile = NULL;
+
+ odp_ticketlock_lock(&profile_tbl->threshold.lock);
+ for (i = 0; i < MAX_THRESHOLD_PROFILES; i++) {
+ if (profile_tbl->threshold.profile[i].status !=
+ TM_STATUS_FREE)
+ continue;
+
+ profile = &profile_tbl->threshold.profile[i];
+ memset(profile, 0, sizeof(tm_queue_thresholds_t));
+ profile->status = TM_STATUS_RESERVED;
+ *idx = i;
+ break;
+ }
+ odp_ticketlock_unlock(&profile_tbl->threshold.lock);
+ return profile;
+ }
+ case TM_WRED_PROFILE: {
+ tm_wred_params_t *profile = NULL;
+
+ odp_ticketlock_lock(&profile_tbl->wred.lock);
+ for (i = 0; i < MAX_WRED_PROFILES; i++) {
+ if (profile_tbl->wred.profile[i].status !=
+ TM_STATUS_FREE)
+ continue;
+
+ profile = &profile_tbl->wred.profile[i];
+ memset(profile, 0, sizeof(tm_wred_params_t));
+ profile->status = TM_STATUS_RESERVED;
+ *idx = i;
+ break;
+ }
+ odp_ticketlock_unlock(&profile_tbl->wred.lock);
+ return profile;
+ }
+ default:
+ ODP_ERR("Invalid TM profile\n");
+ return NULL;
- dynamic_tbl->num_allocd = new_num_allocd;
- dynamic_tbl->array_ptrs = new_array_ptrs;
}
+}
- idx = dynamic_tbl->num_used;
- new_record = malloc(record_size);
- memset(new_record, 0, record_size);
+static void free_tbl_entry(profile_tbl_t *profile_tbl,
+ profile_kind_t profile_kind,
+ uint32_t idx)
+{
+ switch (profile_kind) {
+ case TM_SHAPER_PROFILE:
+ odp_ticketlock_lock(&profile_tbl->shaper.lock);
+ profile_tbl->shaper.profile[idx].status = TM_STATUS_RESERVED;
+ odp_ticketlock_unlock(&profile_tbl->shaper.lock);
+ return;
- dynamic_tbl->array_ptrs[idx] = new_record;
- dynamic_tbl->num_used++;
- if (dynamic_idx_ptr)
- *dynamic_idx_ptr = idx;
+ case TM_SCHED_PROFILE:
+ odp_ticketlock_lock(&profile_tbl->sched.lock);
+ profile_tbl->sched.profile[idx].status = TM_STATUS_RESERVED;
+ odp_ticketlock_unlock(&profile_tbl->sched.lock);
+ return;
- return new_record;
-}
+ case TM_THRESHOLD_PROFILE:
+ odp_ticketlock_lock(&profile_tbl->threshold.lock);
+ profile_tbl->threshold.profile[idx].status = TM_STATUS_RESERVED;
+ odp_ticketlock_unlock(&profile_tbl->threshold.lock);
+ return;
-static void free_dynamic_tbl_entry(dynamic_tbl_t *dynamic_tbl,
- uint32_t record_size ODP_UNUSED,
- uint32_t dynamic_idx)
-{
- void *record;
+ case TM_WRED_PROFILE:
+ odp_ticketlock_lock(&profile_tbl->wred.lock);
+ profile_tbl->wred.profile[idx].status = TM_STATUS_RESERVED;
+ odp_ticketlock_unlock(&profile_tbl->wred.lock);
+ return;
+
+ default:
+ ODP_ERR("Invalid TM profile\n");
+ return;
- record = dynamic_tbl->array_ptrs[dynamic_idx];
- if (record) {
- free(record);
- dynamic_tbl->array_ptrs[dynamic_idx] = NULL;
- dynamic_tbl->num_freed++;
- if (dynamic_tbl->num_freed == dynamic_tbl->num_used) {
- free(dynamic_tbl->array_ptrs);
- memset(dynamic_tbl, 0, sizeof(dynamic_tbl_t));
- }
}
}
@@ -405,35 +491,33 @@ static void tm_system_free(tm_system_t *tm_system)
static void *tm_common_profile_create(const char *name,
profile_kind_t profile_kind,
- uint32_t object_size,
tm_handle_t *profile_handle_ptr,
_odp_int_name_t *name_tbl_id_ptr)
{
_odp_int_name_kind_t handle_kind;
_odp_int_name_t name_tbl_id;
- dynamic_tbl_t *dynamic_tbl;
tm_handle_t profile_handle;
- uint32_t dynamic_tbl_idx;
+ uint32_t idx;
void *object_ptr;
- /* Note that alloc_entry_in_dynamic_tbl will zero out all of the memory
- * that it allocates, so an additional memset here is unnnecessary. */
- dynamic_tbl = &odp_tm_profile_tbls[profile_kind];
- object_ptr = alloc_entry_in_dynamic_tbl(dynamic_tbl, object_size,
- &dynamic_tbl_idx);
- if (!object_ptr)
+ /* Note that alloc_entry_in_tbl will zero out all of the memory that it
+ * allocates, so an additional memset here is unnecessary. */
+ object_ptr = alloc_entry_in_tbl(&tm_glb->profile_tbl, profile_kind,
+ &idx);
+ if (!object_ptr) {
+ ODP_ERR("No free profiles left\n");
return NULL;
+ }
handle_kind = PROFILE_TO_HANDLE_KIND[profile_kind];
- profile_handle = MAKE_PROFILE_HANDLE(profile_kind, dynamic_tbl_idx);
+ profile_handle = MAKE_PROFILE_HANDLE(profile_kind, idx);
name_tbl_id = ODP_INVALID_NAME;
if ((name != NULL) && (name[0] != '\0')) {
name_tbl_id = _odp_int_name_tbl_add(name, handle_kind,
profile_handle);
if (name_tbl_id == ODP_INVALID_NAME) {
- free_dynamic_tbl_entry(dynamic_tbl, object_size,
- dynamic_tbl_idx);
+ free_tbl_entry(&tm_glb->profile_tbl, profile_kind, idx);
return NULL;
}
}
@@ -445,20 +529,18 @@ static void *tm_common_profile_create(const char *name,
}
static int tm_common_profile_destroy(tm_handle_t profile_handle,
- uint32_t object_size,
_odp_int_name_t name_tbl_id)
{
profile_kind_t profile_kind;
- dynamic_tbl_t *dynamic_tbl;
- uint32_t dynamic_tbl_idx;
+ uint32_t idx;
if (name_tbl_id != ODP_INVALID_NAME)
_odp_int_name_tbl_delete(name_tbl_id);
- profile_kind = GET_PROFILE_KIND(profile_handle);
- dynamic_tbl = &odp_tm_profile_tbls[profile_kind];
- dynamic_tbl_idx = GET_TBL_IDX(profile_handle);
- free_dynamic_tbl_entry(dynamic_tbl, object_size, dynamic_tbl_idx);
+ profile_kind = GET_PROFILE_KIND(profile_handle);
+ idx = GET_TBL_IDX(profile_handle);
+ free_tbl_entry(&tm_glb->profile_tbl, profile_kind, idx);
+
return 0;
}
@@ -466,16 +548,31 @@ static void *tm_get_profile_params(tm_handle_t profile_handle,
profile_kind_t expected_profile_kind)
{
profile_kind_t profile_kind;
- dynamic_tbl_t *dynamic_tbl;
- uint32_t dynamic_tbl_idx;
+ uint32_t idx;
profile_kind = GET_PROFILE_KIND(profile_handle);
if (profile_kind != expected_profile_kind)
return NULL;
- dynamic_tbl = &odp_tm_profile_tbls[profile_kind];
- dynamic_tbl_idx = GET_TBL_IDX(profile_handle);
- return dynamic_tbl->array_ptrs[dynamic_tbl_idx];
+ idx = GET_TBL_IDX(profile_handle);
+
+ switch (profile_kind) {
+ case TM_SHAPER_PROFILE:
+ return &tm_glb->profile_tbl.shaper.profile[idx];
+
+ case TM_SCHED_PROFILE:
+ return &tm_glb->profile_tbl.sched.profile[idx];
+
+ case TM_THRESHOLD_PROFILE:
+ return &tm_glb->profile_tbl.threshold.profile[idx];
+
+ case TM_WRED_PROFILE:
+ return &tm_glb->profile_tbl.wred.profile[idx];
+
+ default:
+ ODP_ERR("Invalid TM profile\n");
+ return NULL;
+ }
}
static uint64_t tm_bps_to_rate(uint64_t bps)
@@ -3116,7 +3213,6 @@ odp_tm_shaper_t odp_tm_shaper_create(const char *name,
_odp_int_name_t name_tbl_id;
profile_obj = tm_common_profile_create(name, TM_SHAPER_PROFILE,
- sizeof(tm_shaper_params_t),
&shaper_handle, &name_tbl_id);
if (!profile_obj)
return ODP_TM_INVALID;
@@ -3142,7 +3238,6 @@ int odp_tm_shaper_destroy(odp_tm_shaper_t shaper_profile)
return -1;
return tm_common_profile_destroy(shaper_profile,
- sizeof(tm_shaper_params_t),
profile_obj->name_tbl_id);
}
@@ -3243,7 +3338,6 @@ odp_tm_sched_t odp_tm_sched_create(const char *name,
odp_tm_sched_t sched_handle;
profile_obj = tm_common_profile_create(name, TM_SCHED_PROFILE,
- sizeof(tm_sched_params_t),
&sched_handle, &name_tbl_id);
if (!profile_obj)
return ODP_TM_INVALID;
@@ -3269,7 +3363,6 @@ int odp_tm_sched_destroy(odp_tm_sched_t sched_profile)
return -1;
return tm_common_profile_destroy(sched_profile,
- sizeof(tm_sched_params_t),
profile_obj->name_tbl_id);
}
@@ -3335,7 +3428,6 @@ odp_tm_threshold_t odp_tm_threshold_create(const char *name,
_odp_int_name_t name_tbl_id;
profile_obj = tm_common_profile_create(name, TM_THRESHOLD_PROFILE,
- sizeof(tm_queue_thresholds_t),
&threshold_handle, &name_tbl_id);
if (!profile_obj)
return ODP_TM_INVALID;
@@ -3364,7 +3456,6 @@ int odp_tm_threshold_destroy(odp_tm_threshold_t threshold_profile)
return -1;
return tm_common_profile_destroy(threshold_profile,
- sizeof(odp_tm_threshold_params_t),
threshold_params->name_tbl_id);
}
@@ -3462,7 +3553,6 @@ odp_tm_wred_t odp_tm_wred_create(const char *name, odp_tm_wred_params_t *params)
_odp_int_name_t name_tbl_id;
profile_obj = tm_common_profile_create(name, TM_WRED_PROFILE,
- sizeof(tm_wred_params_t),
&wred_handle, &name_tbl_id);
if (!profile_obj)
@@ -3489,7 +3579,6 @@ int odp_tm_wred_destroy(odp_tm_wred_t wred_profile)
return -1;
return tm_common_profile_destroy(wred_profile,
- sizeof(tm_wred_params_t),
ODP_INVALID_NAME);
}
@@ -4644,6 +4733,10 @@ int odp_tm_init_global(void)
odp_ticketlock_init(&tm_glb->system_group.lock);
odp_ticketlock_init(&tm_create_lock);
odp_ticketlock_init(&tm_profile_lock);
+ odp_ticketlock_init(&tm_glb->profile_tbl.sched.lock);
+ odp_ticketlock_init(&tm_glb->profile_tbl.shaper.lock);
+ odp_ticketlock_init(&tm_glb->profile_tbl.threshold.lock);
+ odp_ticketlock_init(&tm_glb->profile_tbl.wred.lock);
odp_barrier_init(&tm_first_enq, 2);
odp_atomic_init_u64(&atomic_request_cnt, 0);
commit b1812f17ae652f11ce21f26fd24c8fd27818339b
Author: Matias Elo <matias.elo(a)nokia.com>
Date: Thu Nov 22 14:37:28 2018 +0200
linux-gen: traffic_mngr: store input_work_queue inside tm_system_t
Signed-off-by: Matias Elo <matias.elo(a)nokia.com>
Reviewed-by: Bill Fischofer <bill.fischofer(a)linaro.org>
Signed-off-by: Maxim Uvarov <maxim.uvarov(a)linaro.org>
diff --git a/platform/linux-generic/include/odp_traffic_mngr_internal.h b/platform/linux-generic/include/odp_traffic_mngr_internal.h
index ca321ed7..587f9d1a 100644
--- a/platform/linux-generic/include/odp_traffic_mngr_internal.h
+++ b/platform/linux-generic/include/odp_traffic_mngr_internal.h
@@ -377,7 +377,7 @@ struct tm_system_s {
void *trace_buffer;
uint32_t next_queue_num;
tm_queue_obj_t *queue_num_tbl[ODP_TM_MAX_TM_QUEUES];
- input_work_queue_t *input_work_queue;
+ input_work_queue_t input_work_queue;
tm_queue_cnts_t priority_queue_cnts;
tm_queue_cnts_t total_queue_cnts;
pkt_desc_t egress_pkt_desc;
diff --git a/platform/linux-generic/odp_traffic_mngr.c b/platform/linux-generic/odp_traffic_mngr.c
index f928d468..416e7376 100644
--- a/platform/linux-generic/odp_traffic_mngr.c
+++ b/platform/linux-generic/odp_traffic_mngr.c
@@ -304,15 +304,11 @@ static void free_dynamic_tbl_entry(dynamic_tbl_t *dynamic_tbl,
}
}
-static input_work_queue_t *input_work_queue_create(void)
+static void input_work_queue_init(input_work_queue_t *input_work_queue)
{
- input_work_queue_t *input_work_queue;
-
- input_work_queue = malloc(sizeof(input_work_queue_t));
memset(input_work_queue, 0, sizeof(input_work_queue_t));
odp_atomic_init_u64(&input_work_queue->queue_cnt, 0);
odp_ticketlock_init(&input_work_queue->lock);
- return input_work_queue;
}
static void input_work_queue_destroy(input_work_queue_t *input_work_queue)
@@ -322,7 +318,7 @@ static void input_work_queue_destroy(input_work_queue_t *input_work_queue)
* stopped new tm_enq() (et al) calls from succeeding.
*/
odp_ticketlock_lock(&input_work_queue->lock);
- free(input_work_queue);
+ memset(input_work_queue, 0, sizeof(input_work_queue_t));
}
static int input_work_queue_append(tm_system_t *tm_system,
@@ -332,7 +328,7 @@ static int input_work_queue_append(tm_system_t *tm_system,
input_work_item_t *entry_ptr;
uint32_t queue_cnt, tail_idx;
- input_work_queue = tm_system->input_work_queue;
+ input_work_queue = &tm_system->input_work_queue;
queue_cnt = odp_atomic_load_u64(&input_work_queue->queue_cnt);
if (INPUT_WORK_RING_SIZE <= queue_cnt) {
input_work_queue->enqueue_fail_cnt++;
@@ -2373,7 +2369,7 @@ static void *tm_system_thread(void *arg)
tm_system = tm_group->first_tm_system;
_odp_int_timer_wheel = tm_system->_odp_int_timer_wheel;
- input_work_queue = tm_system->input_work_queue;
+ input_work_queue = &tm_system->input_work_queue;
/* Wait here until we have seen the first enqueue operation. */
odp_barrier_wait(&tm_group->tm_group_barrier);
@@ -2427,7 +2423,7 @@ static void *tm_system_thread(void *arg)
/* Advance to the next tm_system in the tm_system_group. */
tm_system = tm_system->next;
_odp_int_timer_wheel = tm_system->_odp_int_timer_wheel;
- input_work_queue = tm_system->input_work_queue;
+ input_work_queue = &tm_system->input_work_queue;
}
odp_barrier_wait(&tm_system->tm_system_destroy_barrier);
@@ -2897,10 +2893,7 @@ odp_tm_t odp_tm_create(const char *name,
== _ODP_INT_TIMER_WHEEL_INVALID;
}
- if (create_fail == 0) {
- tm_system->input_work_queue = input_work_queue_create();
- create_fail |= !tm_system->input_work_queue;
- }
+ input_work_queue_init(&tm_system->input_work_queue);
if (create_fail == 0) {
/* Pass any odp_groups or hints to tm_group_attach here. */
@@ -2911,8 +2904,6 @@ odp_tm_t odp_tm_create(const char *name,
if (create_fail) {
_odp_int_name_tbl_delete(name_tbl_id);
- if (tm_system->input_work_queue)
- input_work_queue_destroy(tm_system->input_work_queue);
if (tm_system->_odp_int_sorted_pool
!= _ODP_INT_SORTED_POOL_INVALID)
@@ -2980,7 +2971,7 @@ int odp_tm_destroy(odp_tm_t odp_tm)
* allocated by this group. */
_odp_tm_group_remove(tm_system->odp_tm_group, odp_tm);
- input_work_queue_destroy(tm_system->input_work_queue);
+ input_work_queue_destroy(&tm_system->input_work_queue);
_odp_sorted_pool_destroy(tm_system->_odp_int_sorted_pool);
_odp_queue_pool_destroy(tm_system->_odp_int_queue_pool);
_odp_timer_wheel_destroy(tm_system->_odp_int_timer_wheel);
@@ -4600,7 +4591,7 @@ void odp_tm_stats_print(odp_tm_t odp_tm)
uint32_t queue_num, max_queue_num;
tm_system = GET_TM_SYSTEM(odp_tm);
- input_work_queue = tm_system->input_work_queue;
+ input_work_queue = &tm_system->input_work_queue;
ODP_PRINT("odp_tm_stats_print - tm_system=0x%" PRIX64 " tm_idx=%u\n",
odp_tm, tm_system->tm_idx);
commit 23102db002f522cc90d1b616e2725d21e525b1fc
Author: Matias Elo <matias.elo(a)nokia.com>
Date: Thu Nov 22 14:07:16 2018 +0200
linux-gen: traffic_mngr: store root node inside tm_system_t
Signed-off-by: Matias Elo <matias.elo(a)nokia.com>
Reviewed-by: Bill Fischofer <bill.fischofer(a)linaro.org>
Signed-off-by: Maxim Uvarov <maxim.uvarov(a)linaro.org>
diff --git a/platform/linux-generic/include/odp_traffic_mngr_internal.h b/platform/linux-generic/include/odp_traffic_mngr_internal.h
index 53b5e944..ca321ed7 100644
--- a/platform/linux-generic/include/odp_traffic_mngr_internal.h
+++ b/platform/linux-generic/include/odp_traffic_mngr_internal.h
@@ -386,7 +386,7 @@ struct tm_system_s {
_odp_timer_wheel_t _odp_int_timer_wheel;
_odp_int_sorted_pool_t _odp_int_sorted_pool;
- tm_node_obj_t *root_node;
+ tm_node_obj_t root_node;
odp_tm_egress_t egress;
odp_tm_requirements_t requirements;
odp_tm_capabilities_t capabilities;
diff --git a/platform/linux-generic/odp_traffic_mngr.c b/platform/linux-generic/odp_traffic_mngr.c
index 1c8be7a7..f928d468 100644
--- a/platform/linux-generic/odp_traffic_mngr.c
+++ b/platform/linux-generic/odp_traffic_mngr.c
@@ -404,9 +404,6 @@ static tm_system_t *tm_system_alloc(void)
static void tm_system_free(tm_system_t *tm_system)
{
- if (tm_system->root_node)
- free(tm_system->root_node);
-
odp_tm_systems[tm_system->tm_idx].status = TM_STATUS_FREE;
}
@@ -2456,19 +2453,6 @@ void odp_tm_egress_init(odp_tm_egress_t *egress)
memset(egress, 0, sizeof(odp_tm_egress_t));
}
-static tm_node_obj_t *create_dummy_root_node(void)
-{
- tm_node_obj_t *tm_node_obj;
-
- tm_node_obj = malloc(sizeof(tm_node_obj_t));
- if (!tm_node_obj)
- return NULL;
-
- memset(tm_node_obj, 0, sizeof(tm_node_obj_t));
- tm_node_obj->is_root_node = true;
- return tm_node_obj;
-}
-
int odp_tm_capabilities(odp_tm_capabilities_t capabilities[] ODP_UNUSED,
uint32_t capabilities_size)
{
@@ -2877,6 +2861,7 @@ odp_tm_t odp_tm_create(const char *name,
&tm_system->requirements);
tm_system->next_queue_num = 1;
+ tm_system->root_node.is_root_node = true;
tm_init_random_data(&tm_system->tm_random_data);
@@ -2912,11 +2897,6 @@ odp_tm_t odp_tm_create(const char *name,
== _ODP_INT_TIMER_WHEEL_INVALID;
}
- if (create_fail == 0) {
- tm_system->root_node = create_dummy_root_node();
- create_fail |= tm_system->root_node == NULL;
- }
-
if (create_fail == 0) {
tm_system->input_work_queue = input_work_queue_create();
create_fail |= !tm_system->input_work_queue;
@@ -4173,7 +4153,9 @@ int odp_tm_node_connect(odp_tm_node_t src_tm_node, odp_tm_node_t dst_tm_node)
src_tm_wred_node = &src_tm_node_obj->tm_wred_node;
if (dst_tm_node == ODP_TM_ROOT) {
- src_tm_node_obj->shaper_obj.next_tm_node = tm_system->root_node;
+ tm_node_obj_t *root_node = &tm_system->root_node;
+
+ src_tm_node_obj->shaper_obj.next_tm_node = root_node;
src_tm_wred_node->next_tm_wred_node = NULL;
return 0;
}
@@ -4241,7 +4223,7 @@ int odp_tm_queue_connect(odp_tm_queue_t tm_queue, odp_tm_node_t dst_tm_node)
src_tm_wred_node = &src_tm_queue_obj->tm_wred_node;
if (dst_tm_node == ODP_TM_ROOT) {
- root_node = tm_system->root_node;
+ root_node = &tm_system->root_node;
src_tm_queue_obj->shaper_obj.next_tm_node = root_node;
src_tm_wred_node->next_tm_wred_node = NULL;
return 0;
commit 049d80427d0145a3c1738d28ba595717ae43d5c2
Author: Matias Elo <matias.elo(a)nokia.com>
Date: Thu Nov 22 13:41:55 2018 +0200
linux-gen: traffic_mngr: store tm_queue_obj_t array inside tm_system_t
Signed-off-by: Matias Elo <matias.elo(a)nokia.com>
Reviewed-by: Bill Fischofer <bill.fischofer(a)linaro.org>
Signed-off-by: Maxim Uvarov <maxim.uvarov(a)linaro.org>
diff --git a/platform/linux-generic/include/odp_traffic_mngr_internal.h b/platform/linux-generic/include/odp_traffic_mngr_internal.h
index eb8a3d53..53b5e944 100644
--- a/platform/linux-generic/include/odp_traffic_mngr_internal.h
+++ b/platform/linux-generic/include/odp_traffic_mngr_internal.h
@@ -376,7 +376,7 @@ struct tm_system_s {
void *trace_buffer;
uint32_t next_queue_num;
- tm_queue_obj_t **queue_num_tbl;
+ tm_queue_obj_t *queue_num_tbl[ODP_TM_MAX_TM_QUEUES];
input_work_queue_t *input_work_queue;
tm_queue_cnts_t priority_queue_cnts;
tm_queue_cnts_t total_queue_cnts;
diff --git a/platform/linux-generic/odp_traffic_mngr.c b/platform/linux-generic/odp_traffic_mngr.c
index 0fb6088f..1c8be7a7 100644
--- a/platform/linux-generic/odp_traffic_mngr.c
+++ b/platform/linux-generic/odp_traffic_mngr.c
@@ -407,9 +407,6 @@ static void tm_system_free(tm_system_t *tm_system)
if (tm_system->root_node)
free(tm_system->root_node);
- if (tm_system->queue_num_tbl)
- free(tm_system->queue_num_tbl);
-
odp_tm_systems[tm_system->tm_idx].status = TM_STATUS_FREE;
}
@@ -2840,7 +2837,7 @@ odp_tm_t odp_tm_create(const char *name,
odp_bool_t create_fail;
odp_tm_t odp_tm;
odp_pktout_queue_t pktout;
- uint32_t malloc_len, max_num_queues, max_queued_pkts, max_timers;
+ uint32_t max_num_queues, max_queued_pkts, max_timers;
uint32_t max_tm_queues, max_sorted_lists;
int rc;
@@ -2879,9 +2876,6 @@ odp_tm_t odp_tm_create(const char *name,
tm_system_capabilities_set(&tm_system->capabilities,
&tm_system->requirements);
- malloc_len = max_tm_queues * sizeof(tm_queue_obj_t *);
- tm_system->queue_num_tbl = malloc(malloc_len);
- memset(tm_system->queue_num_tbl, 0, malloc_len);
tm_system->next_queue_num = 1;
tm_init_random_data(&tm_system->tm_random_data);
commit 46d507adef3902a26b7e311506437211e7417a10
Author: Matias Elo <matias.elo(a)nokia.com>
Date: Thu Nov 22 13:28:08 2018 +0200
linux-gen: traffic_mngr: allocate tm_node_obj_t from shm
Signed-off-by: Matias Elo <matias.elo(a)nokia.com>
Reviewed-by: Bill Fischofer <bill.fischofer(a)linaro.org>
Signed-off-by: Maxim Uvarov <maxim.uvarov(a)linaro.org>
diff --git a/platform/linux-generic/include/odp_traffic_mngr_internal.h b/platform/linux-generic/include/odp_traffic_mngr_internal.h
index 4289e5e0..eb8a3d53 100644
--- a/platform/linux-generic/include/odp_traffic_mngr_internal.h
+++ b/platform/linux-generic/include/odp_traffic_mngr_internal.h
@@ -307,6 +307,7 @@ struct tm_node_obj_s {
uint8_t level; /* Primarily for debugging */
uint8_t tm_idx;
uint8_t marked;
+ tm_status_t status;
};
typedef struct {
diff --git a/platform/linux-generic/odp_traffic_mngr.c b/platform/linux-generic/odp_traffic_mngr.c
index 35254096..0fb6088f 100644
--- a/platform/linux-generic/odp_traffic_mngr.c
+++ b/platform/linux-generic/odp_traffic_mngr.c
@@ -90,6 +90,10 @@ typedef struct {
tm_queue_obj_t obj[ODP_TM_MAX_TM_QUEUES];
odp_ticketlock_t lock;
} queue_obj;
+ struct {
+ tm_node_obj_t obj[ODP_TM_MAX_NUM_TM_NODES];
+ odp_ticketlock_t lock;
+ } node_obj;
odp_shm_t shm;
} tm_global_t;
@@ -126,6 +130,11 @@ static inline tm_queue_obj_t *tm_qobj_from_index(uint32_t queue_id)
return &tm_glb->queue_obj.obj[queue_id];
}
+static inline tm_node_obj_t *tm_nobj_from_index(uint32_t node_id)
+{
+ return &tm_glb->node_obj.obj[node_id];
+}
+
static int queue_tm_reenq(odp_queue_t queue, odp_buffer_hdr_t *buf_hdr)
{
odp_tm_queue_t tm_queue = MAKE_ODP_TM_QUEUE(odp_queue_context(queue));
@@ -3582,34 +3591,50 @@ odp_tm_node_t odp_tm_node_create(odp_tm_t odp_tm,
tm_schedulers_obj_t *schedulers_obj;
_odp_int_name_t name_tbl_id;
tm_wred_node_t *tm_wred_node;
- tm_node_obj_t *tm_node_obj;
+ tm_node_obj_t *tm_node_obj = NULL;
odp_tm_node_t odp_tm_node;
odp_tm_wred_t wred_profile;
tm_system_t *tm_system;
uint32_t level, num_priorities, priority, color;
+ uint32_t i;
/* Allocate a tm_node_obj_t record. */
tm_system = GET_TM_SYSTEM(odp_tm);
- tm_node_obj = malloc(sizeof(tm_node_obj_t));
- if (!tm_node_obj)
- return ODP_TM_INVALID;
- level = params->level;
- requirements = &tm_system->requirements.per_level[level];
- num_priorities = requirements->max_priority + 1;
+ odp_ticketlock_lock(&tm_glb->node_obj.lock);
- odp_tm_node = MAKE_ODP_TM_NODE(tm_node_obj);
- name_tbl_id = ODP_INVALID_NAME;
- if ((name) && (name[0] != '\0')) {
- name_tbl_id = _odp_int_name_tbl_add(name, ODP_TM_NODE_HANDLE,
- odp_tm_node);
- if (name_tbl_id == ODP_INVALID_NAME) {
- free(tm_node_obj);
- return ODP_TM_INVALID;
+ for (i = 0; i < ODP_TM_MAX_NUM_TM_NODES; i++) {
+ tm_node_obj_t *cur_node_obj = tm_nobj_from_index(i);
+
+ if (cur_node_obj->status != TM_STATUS_FREE)
+ continue;
+
+ level = params->level;
+ requirements = &tm_system->requirements.per_level[level];
+ num_priorities = requirements->max_priority + 1;
+
+ odp_tm_node = MAKE_ODP_TM_NODE(cur_node_obj);
+ name_tbl_id = ODP_INVALID_NAME;
+ if ((name) && (name[0] != '\0')) {
+ name_tbl_id = _odp_int_name_tbl_add(name,
+ ODP_TM_NODE_HANDLE,
+ odp_tm_node);
+ if (name_tbl_id == ODP_INVALID_NAME)
+ break;
}
+ tm_node_obj = cur_node_obj;
+
+ memset(tm_node_obj, 0, sizeof(tm_node_obj_t));
+ tm_node_obj->status = TM_STATUS_RESERVED;
+
+ break;
}
- memset(tm_node_obj, 0, sizeof(tm_node_obj_t));
+ odp_ticketlock_unlock(&tm_glb->node_obj.lock);
+
+ if (!tm_node_obj)
+ return ODP_TM_INVALID;
+
tm_node_obj->user_context = params->user_context;
tm_node_obj->name_tbl_id = name_tbl_id;
tm_node_obj->max_fanin = params->max_fanin;
@@ -3718,7 +3743,10 @@ int odp_tm_node_destroy(odp_tm_node_t tm_node)
return rc;
}
- free(tm_node_obj);
+ odp_ticketlock_lock(&tm_glb->node_obj.lock);
+ tm_node_obj->status = TM_STATUS_FREE;
+ odp_ticketlock_unlock(&tm_glb->node_obj.lock);
+
odp_ticketlock_unlock(&tm_system->tm_system_lock);
return 0;
}
@@ -4645,6 +4673,7 @@ int odp_tm_init_global(void)
tm_glb->shm = shm;
odp_ticketlock_init(&tm_glb->queue_obj.lock);
+ odp_ticketlock_init(&tm_glb->node_obj.lock);
odp_ticketlock_init(&tm_glb->system_group.lock);
odp_ticketlock_init(&tm_create_lock);
odp_ticketlock_init(&tm_profile_lock);
commit e7ad8003e34195a3900e1dd3d3a93235896d7628
Author: Matias Elo <matias.elo(a)nokia.com>
Date: Thu Nov 22 12:39:26 2018 +0200
linux-gen: traffic_mngr: move tm_wred_node_t inside tm_node_obj_t
Signed-off-by: Matias Elo <matias.elo(a)nokia.com>
Reviewed-by: Bill Fischofer <bill.fischofer(a)linaro.org>
Signed-off-by: Maxim Uvarov <maxim.uvarov(a)linaro.org>
diff --git a/platform/linux-generic/include/odp_traffic_mngr_internal.h b/platform/linux-generic/include/odp_traffic_mngr_internal.h
index 3dc212c2..4289e5e0 100644
--- a/platform/linux-generic/include/odp_traffic_mngr_internal.h
+++ b/platform/linux-generic/include/odp_traffic_mngr_internal.h
@@ -293,7 +293,7 @@ struct tm_queue_obj_s {
struct tm_node_obj_s {
void *user_context;
- tm_wred_node_t *tm_wred_node;
+ tm_wred_node_t tm_wred_node;
tm_schedulers_obj_t schedulers_obj;
tm_shaper_obj_t *fanin_list_head;
tm_shaper_obj_t *fanin_list_tail;
diff --git a/platform/linux-generic/odp_traffic_mngr.c b/platform/linux-generic/odp_traffic_mngr.c
index 363ff7e2..35254096 100644
--- a/platform/linux-generic/odp_traffic_mngr.c
+++ b/platform/linux-generic/odp_traffic_mngr.c
@@ -3594,12 +3594,6 @@ odp_tm_node_t odp_tm_node_create(odp_tm_t odp_tm,
if (!tm_node_obj)
return ODP_TM_INVALID;
- tm_wred_node = malloc(sizeof(tm_wred_node_t));
- if (!tm_wred_node) {
- free(tm_node_obj);
- return ODP_TM_INVALID;
- }
-
level = params->level;
requirements = &tm_system->requirements.per_level[level];
num_priorities = requirements->max_priority + 1;
@@ -3610,21 +3604,20 @@ odp_tm_node_t odp_tm_node_create(odp_tm_t odp_tm,
name_tbl_id = _odp_int_name_tbl_add(name, ODP_TM_NODE_HANDLE,
odp_tm_node);
if (name_tbl_id == ODP_INVALID_NAME) {
- free(tm_wred_node);
free(tm_node_obj);
return ODP_TM_INVALID;
}
}
memset(tm_node_obj, 0, sizeof(tm_node_obj_t));
- memset(tm_wred_node, 0, sizeof(tm_wred_node_t));
tm_node_obj->user_context = params->user_context;
tm_node_obj->name_tbl_id = name_tbl_id;
tm_node_obj->max_fanin = params->max_fanin;
tm_node_obj->is_root_node = false;
tm_node_obj->level = params->level;
tm_node_obj->tm_idx = tm_system->tm_idx;
- tm_node_obj->tm_wred_node = tm_wred_node;
+
+ tm_wred_node = &tm_node_obj->tm_wred_node;
odp_ticketlock_init(&tm_wred_node->tm_wred_node_lock);
schedulers_obj = &tm_node_obj->schedulers_obj;
@@ -3698,16 +3691,14 @@ int odp_tm_node_destroy(odp_tm_node_t tm_node)
if (shaper_obj->shaper_params != NULL)
return -1;
- tm_wred_node = tm_node_obj->tm_wred_node;
- if (tm_wred_node != NULL) {
- if (tm_wred_node->threshold_params != NULL)
- return -1;
+ tm_wred_node = &tm_node_obj->tm_wred_node;
+ if (tm_wred_node->threshold_params != NULL)
+ return -1;
- for (color = 0; color < ODP_NUM_PACKET_COLORS; color++) {
- wred_params = tm_wred_node->wred_params[color];
- if (wred_params != NULL)
- return -1;
- }
+ for (color = 0; color < ODP_NUM_PACKET_COLORS; color++) {
+ wred_params = tm_wred_node->wred_params[color];
+ if (wred_params != NULL)
+ return -1;
}
/* Now that all of the checks are done, time to so some freeing. */
@@ -3715,9 +3706,6 @@ int odp_tm_node_destroy(odp_tm_node_t tm_node)
if (tm_node_obj->name_tbl_id != ODP_INVALID_NAME)
_odp_int_name_tbl_delete(tm_node_obj->name_tbl_id);
- if (tm_node_obj->tm_wred_node != NULL)
- free(tm_node_obj->tm_wred_node);
-
schedulers_obj = &tm_node_obj->schedulers_obj;
num_priorities = schedulers_obj->num_priorities;
for (priority = 0; priority < num_priorities; priority++) {
@@ -3784,11 +3772,11 @@ int odp_tm_node_threshold_config(odp_tm_node_t tm_node,
tm_node_obj_t *tm_node_obj;
tm_node_obj = GET_TM_NODE_OBJ(tm_node);
- if ((!tm_node_obj) || (!tm_node_obj->tm_wred_node))
+ if (!tm_node_obj)
return -1;
odp_ticketlock_lock(&tm_profile_lock);
- tm_threshold_config_set(tm_node_obj->tm_wred_node, thresholds_profile);
+ tm_threshold_config_set(&tm_node_obj->tm_wred_node, thresholds_profile);
odp_ticketlock_unlock(&tm_profile_lock);
return 0;
}
@@ -3806,7 +3794,7 @@ int odp_tm_node_wred_config(odp_tm_node_t tm_node,
if (!tm_node_obj)
return -1;
- wred_node = tm_node_obj->tm_wred_node;
+ wred_node = &tm_node_obj->tm_wred_node;
odp_ticketlock_lock(&tm_profile_lock);
rc = 0;
@@ -4161,7 +4149,7 @@ int odp_tm_node_connect(odp_tm_node_t src_tm_node, odp_tm_node_t dst_tm_node)
if (!tm_system)
return -1;
- src_tm_wred_node = src_tm_node_obj->tm_wred_node;
+ src_tm_wred_node = &src_tm_node_obj->tm_wred_node;
if (dst_tm_node == ODP_TM_ROOT) {
src_tm_node_obj->shaper_obj.next_tm_node = tm_system->root_node;
src_tm_wred_node->next_tm_wred_node = NULL;
@@ -4172,7 +4160,7 @@ int odp_tm_node_connect(odp_tm_node_t src_tm_node, odp_tm_node_t dst_tm_node)
if ((!dst_tm_node_obj) || dst_tm_node_obj->is_root_node)
return -1;
- dst_tm_wred_node = dst_tm_node_obj->tm_wred_node;
+ dst_tm_wred_node = &dst_tm_node_obj->tm_wred_node;
if (src_tm_node_obj->tm_idx != dst_tm_node_obj->tm_idx)
return -1;
@@ -4203,9 +4191,8 @@ int odp_tm_node_disconnect(odp_tm_node_t src_tm_node)
dst_tm_node_obj->current_tm_node_fanin--;
}
- src_tm_wred_node = src_tm_node_obj->tm_wred_node;
- if (src_tm_wred_node != NULL)
- src_tm_wred_node->next_tm_wred_node = NULL;
+ src_tm_wred_node = &src_tm_node_obj->tm_wred_node;
+ src_tm_wred_node->next_tm_wred_node = NULL;
src_tm_node_obj->shaper_obj.next_tm_node = NULL;
return 0;
@@ -4242,7 +4229,7 @@ int odp_tm_queue_connect(odp_tm_queue_t tm_queue, odp_tm_node_t dst_tm_node)
if ((!dst_tm_node_obj) || dst_tm_node_obj->is_root_node)
return -1;
- dst_tm_wred_node = dst_tm_node_obj->tm_wred_node;
+ dst_tm_wred_node = &dst_tm_node_obj->tm_wred_node;
if (src_tm_queue_obj->tm_idx != dst_tm_node_obj->tm_idx)
return -1;
@@ -4275,8 +4262,7 @@ int odp_tm_queue_disconnect(odp_tm_queue_t tm_queue)
}
src_tm_wred_node = &src_tm_queue_obj->tm_wred_node;
- if (src_tm_wred_node != NULL)
- src_tm_wred_node->next_tm_wred_node = NULL;
+ src_tm_wred_node->next_tm_wred_node = NULL;
src_tm_queue_obj->shaper_obj.next_tm_node = NULL;
return 0;
@@ -4362,19 +4348,17 @@ int odp_tm_node_info(odp_tm_node_t tm_node, odp_tm_node_info_t *info)
if (shaper_params != NULL)
info->shaper_profile = shaper_params->shaper_profile;
- tm_wred_node = tm_node_obj->tm_wred_node;
- if (tm_wred_node != NULL) {
- threshold_params = tm_wred_node->threshold_params;
- if (threshold_params != NULL)
- info->threshold_profile =
- threshold_params->thresholds_profile;
+ tm_wred_node = &tm_node_obj->tm_wred_node;
+ threshold_params = tm_wred_node->threshold_params;
+ if (threshold_params != NULL)
+ info->threshold_profile =
+ threshold_params->thresholds_profile;
- for (color = 0; color < ODP_NUM_PACKET_COLORS; color++) {
- wred_params = tm_wred_node->wred_params[color];
- if (wred_params != NULL)
- info->wred_profile[color] =
- wred_params->wred_profile;
- }
+ for (color = 0; color < ODP_NUM_PACKET_COLORS; color++) {
+ wred_params = tm_wred_node->wred_params[color];
+ if (wred_params != NULL)
+ info->wred_profile[color] =
+ wred_params->wred_profile;
}
return 0;
@@ -4538,8 +4522,6 @@ int odp_tm_queue_query(odp_tm_queue_t tm_queue,
return -1;
tm_wred_node = &tm_queue_obj->tm_wred_node;
- if (!tm_wred_node)
- return -1;
/* **TBD** Where do we get the queue_info from. */
queue_info.threshold_params = tm_wred_node->threshold_params;
commit 294856cc30d48d57e12485076bae49da36d346ed
Author: Matias Elo <matias.elo(a)nokia.com>
Date: Thu Nov 22 12:10:23 2018 +0200
linux-gen: traffic_mngr: move tm_schedulers_obj_t inside tm_node_obj_t
Signed-off-by: Matias Elo <matias.elo(a)nokia.com>
Reviewed-by: Bill Fischofer <bill.fischofer(a)linaro.org>
Signed-off-by: Maxim Uvarov <maxim.uvarov(a)linaro.org>
diff --git a/platform/linux-generic/include/odp_traffic_mngr_internal.h b/platform/linux-generic/include/odp_traffic_mngr_internal.h
index 199bfbb6..3dc212c2 100644
--- a/platform/linux-generic/include/odp_traffic_mngr_internal.h
+++ b/platform/linux-generic/include/odp_traffic_mngr_internal.h
@@ -258,7 +258,7 @@ typedef struct {
uint8_t num_priorities;
uint8_t highest_priority;
uint8_t locked;
- tm_sched_state_t sched_states[0];
+ tm_sched_state_t sched_states[ODP_TM_MAX_PRIORITIES];
} tm_schedulers_obj_t;
struct tm_queue_obj_s {
@@ -294,7 +294,7 @@ struct tm_queue_obj_s {
struct tm_node_obj_s {
void *user_context;
tm_wred_node_t *tm_wred_node;
- tm_schedulers_obj_t *schedulers_obj;
+ tm_schedulers_obj_t schedulers_obj;
tm_shaper_obj_t *fanin_list_head;
tm_shaper_obj_t *fanin_list_tail;
tm_shaper_obj_t shaper_obj;
diff --git a/platform/linux-generic/odp_traffic_mngr.c b/platform/linux-generic/odp_traffic_mngr.c
index 630bc42e..363ff7e2 100644
--- a/platform/linux-generic/odp_traffic_mngr.c
+++ b/platform/linux-generic/odp_traffic_mngr.c
@@ -1332,7 +1332,7 @@ static odp_bool_t tm_propagate_pkt_desc(tm_system_t *tm_system,
if (!shaper_change)
return false;
- schedulers_obj = tm_node_obj->schedulers_obj;
+ schedulers_obj = &tm_node_obj->schedulers_obj;
prev_sched_pkt = schedulers_obj->out_pkt_desc;
sched_was_empty = prev_sched_pkt.queue_num == 0;
sched_change = false;
@@ -1426,7 +1426,7 @@ static odp_bool_t tm_demote_pkt_desc(tm_system_t *tm_system,
if ((!blocked_scheduler) && (!timer_shaper))
return false;
- if (tm_node_obj->schedulers_obj == blocked_scheduler)
+ if (&tm_node_obj->schedulers_obj == blocked_scheduler)
return false;
/* See if this first shaper_obj is delaying the demoted_pkt_desc */
@@ -1452,7 +1452,7 @@ static odp_bool_t tm_demote_pkt_desc(tm_system_t *tm_system,
if ((!demoted_pkt_desc) && (!shaper_change))
return false;
- schedulers_obj = tm_node_obj->schedulers_obj;
+ schedulers_obj = &tm_node_obj->schedulers_obj;
prev_sched_pkt = schedulers_obj->out_pkt_desc;
sched_was_empty = prev_sched_pkt.queue_num == 0;
sched_change = false;
@@ -1569,7 +1569,7 @@ static odp_bool_t tm_consume_pkt_desc(tm_system_t *tm_system,
tm_node_obj = shaper_obj->next_tm_node;
while (!tm_node_obj->is_root_node) { /* not at egress */
- schedulers_obj = tm_node_obj->schedulers_obj;
+ schedulers_obj = &tm_node_obj->schedulers_obj;
prev_sched_pkt = schedulers_obj->out_pkt_desc;
sent_priority = schedulers_obj->highest_priority;
@@ -3586,7 +3586,7 @@ odp_tm_node_t odp_tm_node_create(odp_tm_t odp_tm,
odp_tm_node_t odp_tm_node;
odp_tm_wred_t wred_profile;
tm_system_t *tm_system;
- uint32_t level, num_priorities, priority, schedulers_obj_len, color;
+ uint32_t level, num_priorities, priority, color;
/* Allocate a tm_node_obj_t record. */
tm_system = GET_TM_SYSTEM(odp_tm);
@@ -3603,23 +3603,13 @@ odp_tm_node_t odp_tm_node_create(odp_tm_t odp_tm,
level = params->level;
requirements = &tm_system->requirements.per_level[level];
num_priorities = requirements->max_priority + 1;
- schedulers_obj_len = sizeof(tm_schedulers_obj_t)
- + (sizeof(tm_sched_state_t) * num_priorities);
- schedulers_obj = malloc(schedulers_obj_len);
- if (!schedulers_obj) {
- free(tm_wred_node);
- free(tm_node_obj);
- return ODP_TM_INVALID;
- }
- memset(schedulers_obj, 0, schedulers_obj_len);
odp_tm_node = MAKE_ODP_TM_NODE(tm_node_obj);
name_tbl_id = ODP_INVALID_NAME;
if ((name) && (name[0] != '\0')) {
name_tbl_id = _odp_int_name_tbl_add(name, ODP_TM_NODE_HANDLE,
odp_tm_node);
if (name_tbl_id == ODP_INVALID_NAME) {
- free(schedulers_obj);
free(tm_wred_node);
free(tm_node_obj);
return ODP_TM_INVALID;
@@ -3628,7 +3618,6 @@ odp_tm_node_t odp_tm_node_create(odp_tm_t odp_tm,
memset(tm_node_obj, 0, sizeof(tm_node_obj_t));
memset(tm_wred_node, 0, sizeof(tm_wred_node_t));
- memset(schedulers_obj, 0, schedulers_obj_len);
tm_node_obj->user_context = params->user_context;
tm_node_obj->name_tbl_id = name_tbl_id;
tm_node_obj->max_fanin = params->max_fanin;
@@ -3636,9 +3625,9 @@ odp_tm_node_t odp_tm_node_create(odp_tm_t odp_tm,
tm_node_obj->level = params->level;
tm_node_obj->tm_idx = tm_system->tm_idx;
tm_node_obj->tm_wred_node = tm_wred_node;
- tm_node_obj->schedulers_obj = schedulers_obj;
odp_ticketlock_init(&tm_wred_node->tm_wred_node_lock);
+ schedulers_obj = &tm_node_obj->schedulers_obj;
schedulers_obj->num_priorities = num_priorities;
for (priority = 0; priority < num_priorities; priority++) {
sorted_list = _odp_sorted_list_create(
@@ -3666,7 +3655,7 @@ odp_tm_node_t odp_tm_node_create(odp_tm_t odp_tm,
tm_node_obj->magic_num = TM_NODE_MAGIC_NUM;
tm_node_obj->shaper_obj.enclosing_entity = tm_node_obj;
tm_node_obj->shaper_obj.in_tm_node_obj = 1;
- tm_node_obj->schedulers_obj->enclosing_entity = tm_node_obj;
+ tm_node_obj->schedulers_obj.enclosing_entity = tm_node_obj;
odp_ticketlock_unlock(&tm_system->tm_system_lock);
return odp_tm_node;
@@ -3729,21 +3718,18 @@ int odp_tm_node_destroy(odp_tm_node_t tm_node)
if (tm_node_obj->tm_wred_node != NULL)
free(tm_node_obj->tm_wred_node);
- schedulers_obj = tm_node_obj->schedulers_obj;
- if (schedulers_obj != NULL) {
- num_priorities = schedulers_obj->num_priorities;
- for (priority = 0; priority < num_priorities; priority++) {
- sched_state = &schedulers_obj->sched_states[priority];
- sorted_list = sched_state->sorted_list;
- sorted_pool = tm_system->_odp_int_sorted_pool;
- rc = _odp_sorted_list_destroy(sorted_pool,
- sorted_list);
- if (rc != 0)
- return rc;
- }
+ schedulers_obj = &tm_node_obj->schedulers_obj;
+ num_priorities = schedulers_obj->num_priorities;
+ for (priority = 0; priority < num_priorities; priority++) {
+ sched_state = &schedulers_obj->sched_states[priority];
+ sorted_list = sched_state->sorted_list;
+ sorted_pool = tm_system->_odp_int_sorted_pool;
+ rc = _odp_sorted_list_destroy(sorted_pool,
+ sorted_list);
+ if (rc != 0)
+ return rc;
}
- free(schedulers_obj);
free(tm_node_obj);
odp_ticketlock_unlock(&tm_system->tm_system_lock);
return 0;
commit 6b79ac4b1640e8050b076ba0ecb590cc297320b0
Author: Matias Elo <matias.elo(a)nokia.com>
Date: Wed Nov 21 13:46:46 2018 +0200
linux-gen: traffic_mngr: allocate tm_system_group_t from shm
Signed-off-by: Matias Elo <matias.elo(a)nokia.com>
Reviewed-by: Bill Fischofer <bill.fischofer(a)linaro.org>
Signed-off-by: Maxim Uvarov <maxim.uvarov(a)linaro.org>
diff --git a/platform/linux-generic/include/odp_traffic_mngr_internal.h b/platform/linux-generic/include/odp_traffic_mngr_internal.h
index 21d95273..199bfbb6 100644
--- a/platform/linux-generic/include/odp_traffic_mngr_internal.h
+++ b/platform/linux-generic/include/odp_traffic_mngr_internal.h
@@ -415,15 +415,13 @@ struct tm_system_s {
* while the input work queue is shared - timers are not. */
struct tm_system_group_s {
- tm_system_group_t *prev;
- tm_system_group_t *next;
-
odp_barrier_t tm_group_barrier;
tm_system_t *first_tm_system;
uint32_t num_tm_systems;
uint32_t first_enq;
pthread_t thread;
pthread_attr_t attr;
+ tm_status_t status;
};
#ifdef __cplusplus
diff --git a/platform/linux-generic/odp_traffic_mngr.c b/platform/linux-generic/odp_traffic_mngr.c
index cfa755ee..630bc42e 100644
--- a/platform/linux-generic/odp_traffic_mngr.c
+++ b/platform/linux-generic/odp_traffic_mngr.c
@@ -82,6 +82,10 @@ static tm_prop_t basic_prop_tbl[MAX_PRIORITIES][NUM_SHAPER_COLORS] = {
};
typedef struct {
+ struct {
+ tm_system_group_t group[ODP_TM_MAX_NUM_SYSTEMS];
+ odp_ticketlock_t lock;
+ } system_group;
struct {
tm_queue_obj_t obj[ODP_TM_MAX_TM_QUEUES];
odp_ticketlock_t lock;
@@ -98,8 +102,6 @@ static dynamic_tbl_t odp_tm_profile_tbls[ODP_TM_NUM_PROFILES];
/* TM systems table. */
static tm_system_t odp_tm_systems[ODP_TM_MAX_NUM_SYSTEMS];
-static tm_system_group_t *tm_group_list;
-
static odp_ticketlock_t tm_create_lock;
static odp_ticketlock_t tm_profile_lock;
static odp_barrier_t tm_first_enq;
@@ -2663,35 +2665,9 @@ static int tm_thread_create(tm_system_group_t *tm_group)
return rc;
}
-
-static _odp_tm_group_t _odp_tm_group_create(const char *name ODP_UNUSED)
-{
- tm_system_group_t *tm_group, *first_tm_group, *second_tm_group;
-
- tm_group = malloc(sizeof(tm_system_group_t));
- memset(tm_group, 0, sizeof(tm_system_group_t));
- odp_barrier_init(&tm_group->tm_group_barrier, 2);
-
- /* Add this group to the tm_group_list linked list. */
- if (tm_group_list == NULL) {
- tm_group_list = tm_group;
- tm_group->next = tm_group;
- tm_group->prev = tm_group;
- } else {
- first_tm_group = tm_group_list;
- second_tm_group = first_tm_group->next;
- first_tm_group->next = tm_group;
- second_tm_group->prev = tm_group;
- tm_group->next = second_tm_group;
- tm_group->prev = first_tm_group;
- }
-
- return MAKE_ODP_TM_SYSTEM_GROUP(tm_group);
-}
-
static void _odp_tm_group_destroy(_odp_tm_group_t odp_tm_group)
{
- tm_system_group_t *tm_group, *prev_tm_group, *next_tm_group;
+ tm_system_group_t *tm_group;
int rc;
tm_group = GET_TM_GROUP(odp_tm_group);
@@ -2704,23 +2680,9 @@ static void _odp_tm_group_destroy(_odp_tm_group_t odp_tm_group)
if (g_tm_cpu_num > 0)
g_tm_cpu_num--;
- /* Remove this group from the tm_group_list linked list. Special case
- * when this is the last tm_group in the linked list. */
- prev_tm_group = tm_group->prev;
- next_tm_group = tm_group->next;
- if (prev_tm_group == next_tm_group) {
- ODP_ASSERT(tm_group_list == tm_group);
- tm_group_list = NULL;
- } else {
- prev_tm_group->next = next_tm_group;
- next_tm_group->prev = prev_tm_group;
- if (tm_group_list == tm_group)
- tm_group_list = next_tm_group;
- }
-
- tm_group->prev = NULL;
- tm_group->next = NULL;
- free(tm_group);
+ odp_ticketlock_lock(&tm_glb->system_group.lock);
+ tm_group->status = TM_STATUS_FREE;
+ odp_ticketlock_unlock(&tm_glb->system_group.lock);
}
static int _odp_tm_group_add(_odp_tm_group_t odp_tm_group, odp_tm_t odp_tm)
@@ -2792,12 +2754,21 @@ static int _odp_tm_group_remove(_odp_tm_group_t odp_tm_group, odp_tm_t odp_tm)
return 0;
}
+static void _odp_tm_init_tm_group(tm_system_group_t *tm_group)
+{
+ memset(tm_group, 0, sizeof(tm_system_group_t));
+
+ tm_group->status = TM_STATUS_RESERVED;
+ odp_barrier_init(&tm_group->tm_group_barrier, 2);
+}
+
static int tm_group_attach(odp_tm_t odp_tm)
{
tm_system_group_t *tm_group, *min_tm_group;
_odp_tm_group_t odp_tm_group;
odp_cpumask_t all_cpus, worker_cpus;
uint32_t total_cpus, avail_cpus;
+ uint32_t i;
/* If this platform has a small number of cpu's then allocate one
* tm_group and assign all tm_system's to this tm_group. Otherwise in
@@ -2811,34 +2782,37 @@ static int tm_group_attach(odp_tm_t odp_tm)
avail_cpus = odp_cpumask_count(&worker_cpus);
if (total_cpus < 24) {
- tm_group = tm_group_list;
- odp_tm_group = MAKE_ODP_TM_SYSTEM_GROUP(tm_group);
- if (tm_group == NULL)
- odp_tm_group = _odp_tm_group_create("");
+ tm_group = &tm_glb->system_group.group[0];
- _odp_tm_group_add(odp_tm_group, odp_tm);
- return 0;
- }
+ odp_ticketlock_lock(&tm_glb->system_group.lock);
+ if (tm_group->status == TM_STATUS_FREE)
+ _odp_tm_init_tm_group(tm_group);
+ odp_ticketlock_unlock(&tm_glb->system_group.lock);
- /* Manycore case. */
- if ((tm_group_list == NULL) || (avail_cpus > 1)) {
- odp_tm_group = _odp_tm_group_create("");
+ odp_tm_group = MAKE_ODP_TM_SYSTEM_GROUP(tm_group);
_odp_tm_group_add(odp_tm_group, odp_tm);
return 0;
}
/* Pick a tm_group according to the smallest number of tm_systems. */
- tm_group = tm_group_list;
min_tm_group = NULL;
- while (tm_group != NULL) {
+ odp_ticketlock_lock(&tm_glb->system_group.lock);
+ for (i = 0; i < ODP_TM_MAX_NUM_SYSTEMS && i < avail_cpus; i++) {
+ tm_group = &tm_glb->system_group.group[i];
+
+ if (tm_group->status == TM_STATUS_FREE) {
+ _odp_tm_init_tm_group(tm_group);
+ min_tm_group = tm_group;
+ break;
+ }
+
if (min_tm_group == NULL)
min_tm_group = tm_group;
else if (tm_group->num_tm_systems <
min_tm_group->num_tm_systems)
min_tm_group = tm_group;
-
- tm_group = tm_group->next;
}
+ odp_ticketlock_unlock(&tm_glb->system_group.lock);
if (min_tm_group == NULL)
return -1;
@@ -4703,6 +4677,7 @@ int odp_tm_init_global(void)
tm_glb->shm = shm;
odp_ticketlock_init(&tm_glb->queue_obj.lock);
+ odp_ticketlock_init(&tm_glb->system_group.lock);
odp_ticketlock_init(&tm_create_lock);
odp_ticketlock_init(&tm_profile_lock);
odp_barrier_init(&tm_first_enq, 2);
commit 3cb35813da911a94eef6e07ae71ce0f5f325ebd8
Author: Matias Elo <matias.elo(a)nokia.com>
Date: Thu Nov 22 09:57:19 2018 +0200
linux-gen: traffic_mngr: allocate tm_queue_obj_t from shm
Signed-off-by: Matias Elo <matias.elo(a)nokia.com>
Reviewed-by: Bill Fischofer <bill.fischofer(a)linaro.org>
Signed-off-by: Maxim Uvarov <maxim.uvarov(a)linaro.org>
diff --git a/platform/linux-generic/include/odp_traffic_mngr_internal.h b/platform/linux-generic/include/odp_traffic_mngr_internal.h
index 463759c1..21d95273 100644
--- a/platform/linux-generic/include/odp_traffic_mngr_internal.h
+++ b/platform/linux-generic/include/odp_traffic_mngr_internal.h
@@ -287,6 +287,7 @@ struct tm_queue_obj_s {
uint8_t tm_idx;
uint8_t delayed_cnt;
uint8_t blocked_cnt;
+ tm_status_t status;
odp_queue_t queue;
};
diff --git a/platform/linux-generic/odp_traffic_mngr.c b/platform/linux-generic/odp_traffic_mngr.c
index 1c80a840..cfa755ee 100644
--- a/platform/linux-generic/odp_traffic_mngr.c
+++ b/platform/linux-generic/odp_traffic_mngr.c
@@ -81,6 +81,17 @@ static tm_prop_t basic_prop_tbl[MAX_PRIORITIES][NUM_SHAPER_COLORS] = {
[ODP_TM_SHAPER_RED] = { 7, DELAY_PKT } }
};
+typedef struct {
+ struct {
+ tm_queue_obj_t obj[ODP_TM_MAX_TM_QUEUES];
+ odp_ticketlock_t lock;
+ } queue_obj;
+
+ odp_shm_t shm;
+} tm_global_t;
+
+static tm_global_t *tm_glb;
+
/* Profile tables. */
static dynamic_tbl_t odp_tm_profile_tbls[ODP_TM_NUM_PROFILES];
@@ -108,6 +119,11 @@ static odp_bool_t tm_demote_pkt_desc(tm_system_t *tm_system,
tm_shaper_obj_t *timer_shaper,
pkt_desc_t *demoted_pkt_desc);
+static inline tm_queue_obj_t *tm_qobj_from_index(uint32_t queue_id)
+{
+ return &tm_glb->queue_obj.obj[queue_id];
+}
+
static int queue_tm_reenq(odp_queue_t queue, odp_buffer_hdr_t *buf_hdr)
{
odp_tm_queue_t tm_queue = MAKE_ODP_TM_QUEUE(odp_queue_context(queue));
@@ -3889,70 +3905,84 @@ odp_tm_queue_t odp_tm_queue_create(odp_tm_t odp_tm,
odp_tm_queue_params_t *params)
{
_odp_int_pkt_queue_t _odp_int_pkt_queue;
- tm_queue_obj_t *tm_queue_obj;
- odp_tm_queue_t odp_tm_queue;
+ tm_queue_obj_t *queue_obj;
+ odp_tm_queue_t odp_tm_queue = ODP_TM_INVALID;
odp_queue_t queue;
odp_tm_wred_t wred_profile;
tm_system_t *tm_system;
uint32_t color;
+ uint32_t i;
/* Allocate a tm_queue_obj_t record. */
tm_system = GET_TM_SYSTEM(odp_tm);
- tm_queue_obj = malloc(sizeof(tm_queue_obj_t));
- if (!tm_queue_obj)
- return ODP_TM_INVALID;
- _odp_int_pkt_queue = _odp_pkt_queue_create(
- tm_system->_odp_int_queue_pool);
- if (_odp_int_pkt_queue == _ODP_INT_PKT_QUEUE_INVALID) {
- free(tm_queue_obj);
- return ODP_TM_INVALID;
- }
+ odp_ticketlock_lock(&tm_glb->queue_obj.lock);
- odp_tm_queue = MAKE_ODP_TM_QUEUE(tm_queue_obj);
- memset(tm_queue_obj, 0, sizeof(tm_queue_obj_t));
- tm_queue_obj->user_context = params->user_context;
- tm_queue_obj->priority = params->priority;
- tm_queue_obj->tm_idx = tm_system->tm_idx;
- tm_queue_obj->queue_num = tm_system->next_queue_num++;
- tm_queue_obj->_odp_int_pkt_queue = _odp_int_pkt_queue;
- tm_queue_obj->pkt = ODP_PACKET_INVALID;
- odp_ticketlock_init(&tm_queue_obj->tm_wred_node.tm_wred_node_lock);
-
- queue = odp_queue_create(NULL, NULL);
- if (queue == ODP_QUEUE_INVALID) {
- free(tm_queue_obj);
- return ODP_TM_INVALID;
- }
+ for (i = 0; i < ODP_TM_MAX_TM_QUEUES; i++) {
+ _odp_int_queue_pool_t int_queue_pool;
- tm_queue_obj->queue = queue;
- odp_queue_context_set(queue, tm_queue_obj, sizeof(tm_queue_obj_t));
- queue_fn->set_enq_deq_fn(queue,
- queue_tm_reenq, queue_tm_reenq_multi,
- NULL, NULL);
+ queue_obj = tm_qobj_from_index(i);
- tm_system->queue_num_tbl[tm_queue_obj->queue_num - 1] = tm_queue_obj;
- odp_ticketlock_lock(&tm_system->tm_system_lock);
- if (params->shaper_profile != ODP_TM_INVALID)
- tm_shaper_config_set(tm_system, params->shaper_profile,
- &tm_queue_obj->shaper_obj);
+ if (queue_obj->status != TM_STATUS_FREE)
+ continue;
- if (params->threshold_profile != ODP_TM_INVALID)
- tm_threshold_config_set(&tm_queue_obj->tm_wred_node,
- params->threshold_profile);
+ int_queue_pool = tm_system->_odp_int_queue_pool;
+ _odp_int_pkt_queue = _odp_pkt_queue_create(int_queue_pool);
+ if (_odp_int_pkt_queue == _ODP_INT_PKT_QUEUE_INVALID)
+ continue;
- for (color = 0; color < ODP_NUM_PACKET_COLORS; color++) {
- wred_profile = params->wred_profile[color];
- if (wred_profile != ODP_TM_INVALID)
- tm_wred_config_set(&tm_queue_obj->tm_wred_node, color,
- wred_profile);
+ odp_tm_queue = MAKE_ODP_TM_QUEUE(queue_obj);
+ memset(queue_obj, 0, sizeof(tm_queue_obj_t));
+ queue_obj->user_context = params->user_context;
+ queue_obj->priority = params->priority;
+ queue_obj->tm_idx = tm_system->tm_idx;
+ queue_obj->queue_num = tm_system->next_queue_num++;
+ queue_obj->_odp_int_pkt_queue = _odp_int_pkt_queue;
+ queue_obj->pkt = ODP_PACKET_INVALID;
+ odp_ticketlock_init(&queue_obj->tm_wred_node.tm_wred_node_lock);
+
+ queue = odp_queue_create(NULL, NULL);
+ if (queue == ODP_QUEUE_INVALID) {
+ odp_tm_queue = ODP_TM_INVALID;
+ continue;
+ }
+
+ queue_obj->queue = queue;
+ odp_queue_context_set(queue, queue_obj, sizeof(tm_queue_obj_t));
+ queue_fn->set_enq_deq_fn(queue, queue_tm_reenq,
+ queue_tm_reenq_multi, NULL, NULL);
+
+ tm_system->queue_num_tbl[queue_obj->queue_num - 1] = queue_obj;
+
+ odp_ticketlock_lock(&tm_system->tm_system_lock);
+
+ if (params->shaper_profile != ODP_TM_INVALID)
+ tm_shaper_config_set(tm_system, params->shaper_profile,
+ &queue_obj->shaper_obj);
+
+ if (params->threshold_profile != ODP_TM_INVALID)
+ tm_threshold_config_set(&queue_obj->tm_wred_node,
+ params->threshold_profile);
+
+ for (color = 0; color < ODP_NUM_PACKET_COLORS; color++) {
+ wred_profile = params->wred_profile[color];
+ if (wred_profile != ODP_TM_INVALID)
+ tm_wred_config_set(&queue_obj->tm_wred_node,
+ color, wred_profile);
+ }
+
+ queue_obj->magic_num = TM_QUEUE_MAGIC_NUM;
+ queue_obj->shaper_obj.enclosing_entity = queue_obj;
+ queue_obj->shaper_obj.in_tm_node_obj = 0;
+
+ odp_ticketlock_unlock(&tm_system->tm_system_lock);
+
+ queue_obj->status = TM_STATUS_RESERVED;
+ break;
}
- tm_queue_obj->magic_num = TM_QUEUE_MAGIC_NUM;
- tm_queue_obj->shaper_obj.enclosing_entity = tm_queue_obj;
- tm_queue_obj->shaper_obj.in_tm_node_obj = 0;
+ odp_ticketlock_unlock(&tm_glb->queue_obj.lock);
- odp_ticketlock_unlock(&tm_system->tm_system_lock);
return odp_tm_queue;
}
@@ -3984,9 +4014,10 @@ int odp_tm_queue_destroy(odp_tm_queue_t tm_queue)
odp_queue_destroy(tm_queue_obj->queue);
- /* First delete any associated tm_wred_node and then the tm_queue_obj
- * itself */
- free(tm_queue_obj);
+ odp_ticketlock_lock(&tm_glb->queue_obj.lock);
+ tm_queue_obj->status = TM_STATUS_FREE;
+ odp_ticketlock_unlock(&tm_glb->queue_obj.lock);
+
odp_ticketlock_unlock(&tm_system->tm_system_lock);
return 0;
}
@@ -4660,6 +4691,18 @@ void odp_tm_stats_print(odp_tm_t odp_tm)
int odp_tm_init_global(void)
{
+ odp_shm_t shm;
+
+ shm = odp_shm_reserve("_odp_traffic_mng", sizeof(tm_global_t), 0, 0);
+ if (shm == ODP_SHM_INVALID)
+ return -1;
+
+ tm_glb = odp_shm_addr(shm);
+ memset(tm_glb, 0, sizeof(tm_global_t));
+
+ tm_glb->shm = shm;
+
+ odp_ticketlock_init(&tm_glb->queue_obj.lock);
odp_ticketlock_init(&tm_create_lock);
odp_ticketlock_init(&tm_profile_lock);
odp_barrier_init(&tm_first_enq, 2);
@@ -4672,5 +4715,9 @@ int odp_tm_init_global(void)
int odp_tm_term_global(void)
{
+ if (odp_shm_free(tm_glb->shm)) {
+ ODP_ERR("shm free failed\n");
+ return -1;
+ }
return 0;
}
commit 3aad0e2ce0e5901fd49e50e26ac7d762c2b9a6aa
Author: Matias Elo <matias.elo(a)nokia.com>
Date: Wed Nov 21 14:14:32 2018 +0200
linux-gen: traffic_mngr: move tm_wred_node_t inside tm_queue_obj_t
Signed-off-by: Matias Elo <matias.elo(a)nokia.com>
Reviewed-by: Bill Fischofer <bill.fischofer(a)linaro.org>
Signed-off-by: Maxim Uvarov <maxim.uvarov(a)linaro.org>
diff --git a/platform/linux-generic/include/odp_traffic_mngr_internal.h b/platform/linux-generic/include/odp_traffic_mngr_internal.h
index eb1e5775..463759c1 100644
--- a/platform/linux-generic/include/odp_traffic_mngr_internal.h
+++ b/platform/linux-generic/include/odp_traffic_mngr_internal.h
@@ -269,7 +269,7 @@ struct tm_queue_obj_s {
uint32_t pkts_dequeued_cnt;
uint32_t pkts_consumed_cnt;
_odp_int_pkt_queue_t _odp_int_pkt_queue;
- tm_wred_node_t *tm_wred_node;
+ tm_wred_node_t tm_wred_node;
odp_packet_t pkt;
odp_packet_t sent_pkt;
uint32_t timer_seq;
diff --git a/platform/linux-generic/odp_traffic_mngr.c b/platform/linux-generic/odp_traffic_mngr.c
index 81117ce9..1c80a840 100644
--- a/platform/linux-generic/odp_traffic_mngr.c
+++ b/platform/linux-generic/odp_traffic_mngr.c
@@ -1635,7 +1635,7 @@ static odp_bool_t tm_consume_sent_pkt(tm_system_t *tm_system,
pkt_len = sent_pkt_desc->pkt_len;
tm_queue_obj->pkts_consumed_cnt++;
- tm_queue_cnts_decrement(tm_system, tm_queue_obj->tm_wred_node,
+ tm_queue_cnts_decrement(tm_system, &tm_queue_obj->tm_wred_node,
tm_queue_obj->priority, pkt_len);
/* Get the next pkt in the tm_queue, if there is one. */
@@ -1901,7 +1901,7 @@ static int tm_enqueue(tm_system_t *tm_system,
pkt_color = odp_packet_color(pkt);
drop_eligible = odp_packet_drop_eligible(pkt);
- initial_tm_wred_node = tm_queue_obj->tm_wred_node;
+ initial_tm_wred_node = &tm_queue_obj->tm_wred_node;
if (drop_eligible) {
drop = random_early_discard(tm_system, tm_queue_obj,
initial_tm_wred_node, pkt_color);
@@ -3890,7 +3890,6 @@ odp_tm_queue_t odp_tm_queue_create(odp_tm_t odp_tm,
{
_odp_int_pkt_queue_t _odp_int_pkt_queue;
tm_queue_obj_t *tm_queue_obj;
- tm_wred_node_t *tm_wred_node;
odp_tm_queue_t odp_tm_queue;
odp_queue_t queue;
odp_tm_wred_t wred_profile;
@@ -3903,35 +3902,25 @@ odp_tm_queue_t odp_tm_queue_create(odp_tm_t odp_tm,
if (!tm_queue_obj)
return ODP_TM_INVALID;
- tm_wred_node = malloc(sizeof(tm_wred_node_t));
- if (!tm_wred_node) {
- free(tm_queue_obj);
- return ODP_TM_INVALID;
- }
-
_odp_int_pkt_queue = _odp_pkt_queue_create(
tm_system->_odp_int_queue_pool);
if (_odp_int_pkt_queue == _ODP_INT_PKT_QUEUE_INVALID) {
- free(tm_wred_node);
free(tm_queue_obj);
return ODP_TM_INVALID;
}
odp_tm_queue = MAKE_ODP_TM_QUEUE(tm_queue_obj);
memset(tm_queue_obj, 0, sizeof(tm_queue_obj_t));
- memset(tm_wred_node, 0, sizeof(tm_wred_node_t));
tm_queue_obj->user_context = params->user_context;
tm_queue_obj->priority = params->priority;
tm_queue_obj->tm_idx = tm_system->tm_idx;
tm_queue_obj->queue_num = tm_system->next_queue_num++;
- tm_queue_obj->tm_wred_node = tm_wred_node;
tm_queue_obj->_odp_int_pkt_queue = _odp_int_pkt_queue;
tm_queue_obj->pkt = ODP_PACKET_INVALID;
- odp_ticketlock_init(&tm_wred_node->tm_wred_node_lock);
+ odp_ticketlock_init(&tm_queue_obj->tm_wred_node.tm_wred_node_lock);
queue = odp_queue_create(NULL, NULL);
if (queue == ODP_QUEUE_INVALID) {
- free(tm_wred_node);
free(tm_queue_obj);
return ODP_TM_INVALID;
}
@@ -3949,13 +3938,14 @@ odp_tm_queue_t odp_tm_queue_create(odp_tm_t odp_tm,
&tm_queue_obj->shaper_obj);
if (params->threshold_profile != ODP_TM_INVALID)
- tm_threshold_config_set(tm_wred_node,
+ tm_threshold_config_set(&tm_queue_obj->tm_wred_node,
params->threshold_profile);
for (color = 0; color < ODP_NUM_PACKET_COLORS; color++) {
wred_profile = params->wred_profile[color];
if (wred_profile != ODP_TM_INVALID)
- tm_wred_config_set(tm_wred_node, color, wred_profile);
+ tm_wred_config_set(&tm_queue_obj->tm_wred_node, color,
+ wred_profile);
}
tm_queue_obj->magic_num = TM_QUEUE_MAGIC_NUM;
@@ -3996,7 +3986,6 @@ int odp_tm_queue_destroy(odp_tm_queue_t tm_queue)
/* First delete any associated tm_wred_node and then the tm_queue_obj
* itself */
- free(tm_queue_obj->tm_wred_node);
free(tm_queue_obj);
odp_ticketlock_unlock(&tm_system->tm_system_lock);
return 0;
@@ -4080,7 +4069,7 @@ int odp_tm_queue_threshold_config(odp_tm_queue_t tm_queue,
return -1;
odp_ticketlock_lock(&tm_profile_lock);
- ret = tm_threshold_config_set(tm_queue_obj->tm_wred_node,
+ ret = tm_threshold_config_set(&tm_queue_obj->tm_wred_node,
thresholds_profile);
odp_ticketlock_unlock(&tm_profile_lock);
return ret;
@@ -4099,7 +4088,7 @@ int odp_tm_queue_wred_config(odp_tm_queue_t tm_queue,
if (!tm_queue_obj)
return -1;
- wred_node = tm_queue_obj->tm_wred_node;
+ wred_node = &tm_queue_obj->tm_wred_node;
odp_ticketlock_lock(&tm_profile_lock);
rc = 0;
@@ -4250,7 +4239,7 @@ int odp_tm_queue_connect(odp_tm_queue_t tm_queue, odp_tm_node_t dst_tm_node)
if (!tm_system)
return -1;
- src_tm_wred_node = src_tm_queue_obj->tm_wred_node;
+ src_tm_wred_node = &src_tm_queue_obj->tm_wred_node;
if (dst_tm_node == ODP_TM_ROOT) {
root_node = tm_system->root_node;
src_tm_queue_obj->shaper_obj.next_tm_node = root_node;
@@ -4294,7 +4283,7 @@ int odp_tm_queue_disconnect(odp_tm_queue_t tm_queue)
dst_tm_node_obj->current_tm_queue_fanin--;
}
- src_tm_wred_node = src_tm_queue_obj->tm_wred_node;
+ src_tm_wred_node = &src_tm_queue_obj->tm_wred_node;
if (src_tm_wred_node != NULL)
src_tm_wred_node->next_tm_wred_node = NULL;
@@ -4498,7 +4487,7 @@ int odp_tm_queue_info(odp_tm_queue_t tm_queue, odp_tm_queue_info_t *info)
if (shaper_params != NULL)
info->shaper_profile = shaper_params->shaper_profile;
- tm_wred_node = tm_queue_obj->tm_wred_node;
+ tm_wred_node = &tm_queue_obj->tm_wred_node;
if (tm_wred_node != NULL) {
threshold_params = tm_wred_node->threshold_params;
if (threshold_params != NULL)
@@ -4557,7 +4546,7 @@ int odp_tm_queue_query(odp_tm_queue_t tm_queue,
if (!tm_queue_obj)
return -1;
- tm_wred_node = tm_queue_obj->tm_wred_node;
+ tm_wred_node = &tm_queue_obj->tm_wred_node;
if (!tm_wred_node)
return -1;
commit 98eb7327113fbd33a8e5448406e8f47d8d0ad5fb
Author: Matias Elo <matias.elo(a)nokia.com>
Date: Fri Nov 16 15:53:01 2018 +0200
linux-gen: traffic_mngr: use static array for odp_tm_systems
Signed-off-by: Matias Elo <matias.elo(a)nokia.com>
Reviewed-by: Bill Fischofer <bill.fischofer(a)linaro.org>
Signed-off-by: Maxim Uvarov <maxim.uvarov(a)linaro.org>
diff --git a/platform/linux-generic/include/odp_traffic_mngr_internal.h b/platform/linux-generic/include/odp_traffic_mngr_internal.h
index 770a6490..eb1e5775 100644
--- a/platform/linux-generic/include/odp_traffic_mngr_internal.h
+++ b/platform/linux-generic/include/odp_traffic_mngr_internal.h
@@ -99,6 +99,11 @@ typedef enum {
TM_WRED_PROFILE
} profile_kind_t;
+typedef enum {
+ TM_STATUS_FREE = 0,
+ TM_STATUS_RESERVED
+} tm_status_t;
+
typedef struct tm_queue_obj_s tm_queue_obj_t;
typedef struct tm_node_obj_s tm_node_obj_t;
@@ -396,6 +401,7 @@ struct tm_system_s {
uint8_t tm_idx;
uint8_t first_enq;
odp_bool_t is_idle;
+ tm_status_t status;
uint64_t shaper_green_cnt;
uint64_t shaper_yellow_cnt;
diff --git a/platform/linux-generic/odp_traffic_mngr.c b/platform/linux-generic/odp_traffic_mngr.c
index cf2ce83e..81117ce9 100644
--- a/platform/linux-generic/odp_traffic_mngr.c
+++ b/platform/linux-generic/odp_traffic_mngr.c
@@ -85,7 +85,7 @@ static tm_prop_t basic_prop_tbl[MAX_PRIORITIES][NUM_SHAPER_COLORS] = {
static dynamic_tbl_t odp_tm_profile_tbls[ODP_TM_NUM_PROFILES];
/* TM systems table. */
-static tm_system_t *odp_tm_systems[ODP_TM_MAX_NUM_SYSTEMS];
+static tm_system_t odp_tm_systems[ODP_TM_MAX_NUM_SYSTEMS];
static tm_system_group_t *tm_group_list;
@@ -363,11 +363,11 @@ static tm_system_t *tm_system_alloc(void)
/* Find an open slot in the odp_tm_systems array. */
for (tm_idx = 0; tm_idx < ODP_TM_MAX_NUM_SYSTEMS; tm_idx++) {
- if (!odp_tm_systems[tm_idx]) {
- tm_system = malloc(sizeof(tm_system_t));
+ if (odp_tm_systems[tm_idx].status == TM_STATUS_FREE) {
+ tm_system = &odp_tm_systems[tm_idx];
memset(tm_system, 0, sizeof(tm_system_t));
- odp_tm_systems[tm_idx] = tm_system;
tm_system->tm_idx = tm_idx;
+ tm_system->status = TM_STATUS_RESERVED;
return tm_system;
}
}
@@ -383,8 +383,7 @@ static void tm_system_free(tm_system_t *tm_system)
if (tm_system->queue_num_tbl)
free(tm_system->queue_num_tbl);
- odp_tm_systems[tm_system->tm_idx] = NULL;
- free(tm_system);
+ odp_tm_systems[tm_system->tm_idx].status = TM_STATUS_FREE;
}
static void *tm_common_profile_create(const char *name,
@@ -3702,7 +3701,7 @@ int odp_tm_node_destroy(odp_tm_node_t tm_node)
if (!tm_node_obj)
return -1;
- tm_system = odp_tm_systems[tm_node_obj->tm_idx];
+ tm_system = &odp_tm_systems[tm_node_obj->tm_idx];
if (!tm_system)
return -1;
@@ -3770,7 +3769,7 @@ int odp_tm_node_shaper_config(odp_tm_node_t tm_node,
if (!tm_node_obj)
return -1;
- tm_system = odp_tm_systems[tm_node_obj->tm_idx];
+ tm_system = &odp_tm_systems[tm_node_obj->tm_idx];
if (!tm_system)
return -1;
@@ -3978,7 +3977,7 @@ int odp_tm_queue_destroy(odp_tm_queue_t tm_queue)
if (!tm_queue_obj)
return -1;
- tm_system = odp_tm_systems[tm_queue_obj->tm_idx];
+ tm_system = &odp_tm_systems[tm_queue_obj->tm_idx];
if (!tm_system)
return -1;
@@ -4036,7 +4035,7 @@ int odp_tm_queue_shaper_config(odp_tm_queue_t tm_queue,
if (!tm_queue_obj)
return -1;
- tm_system = odp_tm_systems[tm_queue_obj->tm_idx];
+ tm_system = &odp_tm_systems[tm_queue_obj->tm_idx];
if (!tm_system)
return -1;
@@ -4178,7 +4177,7 @@ int odp_tm_node_connect(odp_tm_node_t src_tm_node, odp_tm_node_t dst_tm_node)
if ((!src_tm_node_obj) || src_tm_node_obj->is_root_node)
return -1;
- tm_system = odp_tm_systems[src_tm_node_obj->tm_idx];
+ tm_system = &odp_tm_systems[src_tm_node_obj->tm_idx];
if (!tm_system)
return -1;
@@ -4247,7 +4246,7 @@ int odp_tm_queue_connect(odp_tm_queue_t tm_queue, odp_tm_node_t dst_tm_node)
if (!src_tm_queue_obj)
return -1;
- tm_system = odp_tm_systems[src_tm_queue_obj->tm_idx];
+ tm_system = &odp_tm_systems[src_tm_queue_obj->tm_idx];
if (!tm_system)
return -1;
@@ -4312,7 +4311,7 @@ int odp_tm_enq(odp_tm_queue_t tm_queue, odp_packet_t pkt)
if (!tm_queue_obj)
return -1;
- tm_system = odp_tm_systems[tm_queue_obj->tm_idx];
+ tm_system = &odp_tm_systems[tm_queue_obj->tm_idx];
if (!tm_system)
return -1;
@@ -4333,7 +4332,7 @@ int odp_tm_enq_with_cnt(odp_tm_queue_t tm_queue, odp_packet_t pkt)
if (!tm_queue_obj)
return -1;
- tm_system = odp_tm_systems[tm_queue_obj->tm_idx];
+ tm_system = &odp_tm_systems[tm_queue_obj->tm_idx];
if (!tm_system)
return -1;
commit d64232f45abae8d4f1222313ce44532cc26e2336
Author: Matias Elo <matias.elo(a)nokia.com>
Date: Wed Nov 21 15:25:38 2018 +0200
abi: traffic_mngr: reduce max defines
Decrease TM maximum defines to reduce memory usage. With the selected
values TM SHM usage stays under 10MB.
Signed-off-by: Matias Elo <matias.elo(a)nokia.com>
Reviewed-by: Bill Fischofer <bill.fischofer(a)linaro.org>
Signed-off-by: Maxim Uvarov <maxim.uvarov(a)linaro.org>
diff --git a/include/odp/api/abi-default/traffic_mngr.h b/include/odp/api/abi-default/traffic_mngr.h
index b7b04b83..9c01ef98 100644
--- a/include/odp/api/abi-default/traffic_mngr.h
+++ b/include/odp/api/abi-default/traffic_mngr.h
@@ -28,7 +28,7 @@ extern "C" {
* systems that may be created. On some platforms this might be much more
* limited to as little as one hardware TM system.
*/
-#define ODP_TM_MAX_NUM_SYSTEMS 64
+#define ODP_TM_MAX_NUM_SYSTEMS 8
/** The ODP_TM_MAX_PRIORITIES constant specifies the largest range of
* priorities that any TM system can support. All strict priority values MUST
@@ -57,7 +57,7 @@ extern "C" {
/** The ODP_TM_MAX_TM_QUEUES constant is the largest number of tm_queues
* that can be handled by any one TM system.
*/
-#define ODP_TM_MAX_TM_QUEUES (16 * 1024 * 1024)
+#define ODP_TM_MAX_TM_QUEUES (4 * 1024)
/** The ODP_TM_MAX_NUM_OUTPUTS constant is the largest number of outputs that
* can be configured for any one TM system.
@@ -67,13 +67,13 @@ extern "C" {
/** The ODP_TM_MAX_NUM_TM_NODES constant is the largest number of tm_nodes that
* can be in existence for any one TM system.
*/
-#define ODP_TM_MAX_NUM_TM_NODES (1024 * 1024)
+#define ODP_TM_MAX_NUM_TM_NODES (4 * 1024)
/** The ODP_TM_MAX_TM_NODE_FANIN constant is the largest number of fan-in
* "inputs" that can be simultaneously connected to a single tm_node.
* *TBD* Does this need to be as large as ODP_TM_MAX_TM_QUEUES? *TBD*
*/
-#define ODP_TM_MAX_TM_NODE_FANIN (1024 * 1024)
+#define ODP_TM_MAX_TM_NODE_FANIN (4 * 1024)
/** The ODP_TM_MIN_SHAPER_BW constant is the smallest amount of bandwidth that
* can a shaper's peak or commit rate can be set to. It is in units of
-----------------------------------------------------------------------
Summary of changes:
include/odp/api/abi-default/traffic_mngr.h | 8 +-
.../include/odp_traffic_mngr_internal.h | 37 +-
platform/linux-generic/odp_traffic_mngr.c | 958 +++++++++++----------
3 files changed, 543 insertions(+), 460 deletions(-)
hooks/post-receive
--
6 years, 2 months
- 1
- 0