1Design 2====== 3 4Environment or Architecture-specific Sources 5-------------------------------------------- 6 7In DPDK and DPDK applications, some code is specific to an architecture (i686, x86_64) or to an executive environment (bsdapp or linuxapp) and so on. 8As far as is possible, all such instances of architecture or env-specific code should be provided via standard APIs in the EAL. 9 10By convention, a file is common if it is not located in a directory indicating that it is specific. 11For instance, a file located in a subdir of "x86_64" directory is specific to this architecture. 12A file located in a subdir of "linuxapp" is specific to this execution environment. 13 14.. note:: 15 16 Code in DPDK libraries and applications should be generic. 17 The correct location for architecture or executive environment specific code is in the EAL. 18 19When absolutely necessary, there are several ways to handle specific code: 20 21* Use a ``#ifdef`` with the CONFIG option in the C code. 22 This can be done when the differences are small and they can be embedded in the same C file: 23 24 .. code-block:: c 25 26 #ifdef RTE_ARCH_I686 27 toto(); 28 #else 29 titi(); 30 #endif 31 32* Use the CONFIG option in the Makefile. This is done when the differences are more significant. 33 In this case, the code is split into two separate files that are architecture or environment specific. 34 This should only apply inside the EAL library. 35 36.. note:: 37 38 As in the linux kernel, the ``CONFIG_`` prefix is not used in C code. 39 This is only needed in Makefiles or shell scripts. 40 41Per Architecture Sources 42~~~~~~~~~~~~~~~~~~~~~~~~ 43 44The following config options can be used: 45 46* ``CONFIG_RTE_ARCH`` is a string that contains the name of the architecture. 47* ``CONFIG_RTE_ARCH_I686``, ``CONFIG_RTE_ARCH_X86_64``, ``CONFIG_RTE_ARCH_X86_64_32`` or ``CONFIG_RTE_ARCH_PPC_64`` are defined only if we are building for those architectures. 48 49Per Execution Environment Sources 50~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 51 52The following config options can be used: 53 54* ``CONFIG_RTE_EXEC_ENV`` is a string that contains the name of the executive environment. 55* ``CONFIG_RTE_EXEC_ENV_BSDAPP`` or ``CONFIG_RTE_EXEC_ENV_LINUXAPP`` are defined only if we are building for this execution environment. 56 57Library Statistics 58------------------ 59 60Description 61~~~~~~~~~~~ 62 63This document describes the guidelines for DPDK library-level statistics counter 64support. This includes guidelines for turning library statistics on and off and 65requirements for preventing ABI changes when implementing statistics. 66 67 68Mechanism to allow the application to turn library statistics on and off 69~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 70 71Each library that maintains statistics counters should provide a single build 72time flag that decides whether the statistics counter collection is enabled or 73not. This flag should be exposed as a variable within the DPDK configuration 74file. When this flag is set, all the counters supported by current library are 75collected for all the instances of every object type provided by the library. 76When this flag is cleared, none of the counters supported by the current library 77are collected for any instance of any object type provided by the library: 78 79.. code-block:: console 80 81 # DPDK file config/common_linuxapp, config/common_bsdapp, etc. 82 CONFIG_RTE_<LIBRARY_NAME>_STATS_COLLECT=y/n 83 84The default value for this DPDK configuration file variable (either "yes" or 85"no") is decided by each library. 86 87 88Prevention of ABI changes due to library statistics support 89~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 90 91The layout of data structures and prototype of functions that are part of the 92library API should not be affected by whether the collection of statistics 93counters is turned on or off for the current library. In practical terms, this 94means that space should always be allocated in the API data structures for 95statistics counters and the statistics related API functions are always built 96into the code, regardless of whether the statistics counter collection is turned 97on or off for the current library. 98 99When the collection of statistics counters for the current library is turned 100off, the counters retrieved through the statistics related API functions should 101have a default value of zero. 102 103 104Motivation to allow the application to turn library statistics on and off 105~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 106 107It is highly recommended that each library provides statistics counters to allow 108an application to monitor the library-level run-time events. Typical counters 109are: number of packets received/dropped/transmitted, number of buffers 110allocated/freed, number of occurrences for specific events, etc. 111 112However, the resources consumed for library-level statistics counter collection 113have to be spent out of the application budget and the counters collected by 114some libraries might not be relevant to the current application. In order to 115avoid any unwanted waste of resources and/or performance impacts, the 116application should decide at build time whether the collection of library-level 117statistics counters should be turned on or off for each library individually. 118 119Library-level statistics counters can be relevant or not for specific 120applications: 121 122* For Application A, counters maintained by Library X are always relevant and 123 the application needs to use them to implement certain features, such as traffic 124 accounting, logging, application-level statistics, etc. In this case, 125 the application requires that collection of statistics counters for Library X is 126 always turned on. 127 128* For Application B, counters maintained by Library X are only useful during the 129 application debug stage and are not relevant once debug phase is over. In this 130 case, the application may decide to turn on the collection of Library X 131 statistics counters during the debug phase and at a later stage turn them off. 132 133* For Application C, counters maintained by Library X are not relevant at all. 134 It might be that the application maintains its own set of statistics counters 135 that monitor a different set of run-time events (e.g. number of connection 136 requests, number of active users, etc). It might also be that the application 137 uses multiple libraries (Library X, Library Y, etc) and it is interested in the 138 statistics counters of Library Y, but not in those of Library X. In this case, 139 the application may decide to turn the collection of statistics counters off for 140 Library X and on for Library Y. 141 142The statistics collection consumes a certain amount of CPU resources (cycles, 143cache bandwidth, memory bandwidth, etc) that depends on: 144 145* Number of libraries used by the current application that have statistics 146 counters collection turned on. 147 148* Number of statistics counters maintained by each library per object type 149 instance (e.g. per port, table, pipeline, thread, etc). 150 151* Number of instances created for each object type supported by each library. 152 153* Complexity of the statistics logic collection for each counter: when only 154 some occurrences of a specific event are valid, additional logic is typically 155 needed to decide whether the current occurrence of the event should be counted 156 or not. For example, in the event of packet reception, when only TCP packets 157 with destination port within a certain range should be recorded, conditional 158 branches are usually required. When processing a burst of packets that have been 159 validated for header integrity, counting the number of bits set in a bitmask 160 might be needed. 161 162PF and VF Considerations 163------------------------ 164 165The primary goal of DPDK is to provide a userspace dataplane. Managing VFs from 166a PF driver is a control plane feature and developers should generally rely on 167the Linux Kernel for that. 168 169Developers should work with the Linux Kernel community to get the required 170functionality upstream. PF functionality should only be added to DPDK for 171testing and prototyping purposes while the kernel work is ongoing. It should 172also be marked with an "EXPERIMENTAL" tag. If the functionality isn't 173upstreamable then a case can be made to maintain the PF functionality in DPDK 174without the EXPERIMENTAL tag. 175