developer | 02e6591 | 2023-08-17 16:33:10 +0800 | [diff] [blame] | 1 | +=============================================================================+ |
| 2 | | Copyright (c) 2012-2020 Rambus, Inc. and/or its subsidiaries. | |
| 3 | | | |
| 4 | | Subject : PCL API Implementation Notes | |
| 5 | | Product : PCL API | |
| 6 | | Date : 18 November, 2020 | |
| 7 | | | |
| 8 | +=============================================================================+ |
| 9 | |
| 10 | The SLAD API is a set of the APIs one of which is the Packet Classification |
| 11 | (PCL) API. The driver implementation specifics of these APIs are |
| 12 | described in short documents that serve as an addendum to the API |
| 13 | specifications. This document describes the PCL API. |
| 14 | |
| 15 | This document uses the phrase "configurable" to indicate that a parameter or |
| 16 | option is build-time configurable in the driver. Please refer to the User Guide |
| 17 | of the driver for details. |
| 18 | |
| 19 | |
| 20 | PCL API |
| 21 | ------- |
| 22 | |
| 23 | The PCL API manages two types of data structures relevant to the |
| 24 | Packet Classification Engine: flow data structures and transform data |
| 25 | structures. |
| 26 | |
| 27 | The PCL API is not used to submit packets (command descriptors) to the |
| 28 | Packet Engine and to retrieve result descriptors (referring to |
| 29 | processed packets). This is performed by the PEC API (see the PEC API |
| 30 | Implementation Notes for details. |
| 31 | |
| 32 | Transform records |
| 33 | Transform records are allocated outside the Driver (using the |
| 34 | DMABuf API) and are initialized outside the Driver by the |
| 35 | Extended SA Builder. The Extended SA Builder is the SA Builder |
| 36 | that is compiled with the "Extended" configuration option |
| 37 | enabled. Transform records are represented by their DMABuf |
| 38 | Handle. They can be registered with PCL_Transform_Register() before |
| 39 | they are used and they are unregistered by |
| 40 | PCL_Transform_Unregister() after use. A subset of the contents |
| 41 | (sequence number and statistics) can be read with |
| 42 | PCL_Transform_Get_ReadOnly(). |
| 43 | |
| 44 | Flow records |
| 45 | Flow Records are represented by an internal data structure |
| 46 | (represented by PCL_FlowHandle_t), which is invisible to the application. |
| 47 | The internal data structure is allocated within the Driver. |
| 48 | A DMA-safe buffer is associated with it and this is also allocated |
| 49 | within the Driver and invisible to the application. |
| 50 | |
| 51 | The Driver makes use of the internal DMAResource API to allocate |
| 52 | the DMA-safe buffer and of an internal memory allocation API |
| 53 | (adapter_alloc.h) to allocate the data structure. |
| 54 | |
| 55 | Allocation of flow records can be a time consuming operation and |
| 56 | it may not be allowed at all in certain situations (e.g. in |
| 57 | interrupt context). Therefore allocating and deallocating the |
| 58 | resources for a flow record is decoupled from adding and removing |
| 59 | flow records in the flow record table. |
| 60 | |
| 61 | When a flow record is allocated with PCL_Flow_Alloc() its |
| 62 | resources (internal data structure and DMA-safe buffer) are |
| 63 | allocated, but the flow record is not initialized, is not part of |
| 64 | the flow table and is inaccessible to the Packet Classification |
| 65 | Engine. PCL_Flow_Add() initializes the record and adds it to a |
| 66 | lookup table. The size of the hash table from which flow lookup |
| 67 | is started has a configurable size. |
| 68 | |
| 69 | PCL_Flow_Get_ReadOnly() reads the variable fields (last used time, |
| 70 | packet statistics) from a flow record. PCL_Flow_Remove() removes it |
| 71 | from the lookup table. At this time the resources are still allocated |
| 72 | and can be reused by another flow record. PCL_Flow_Release() deallocates |
| 73 | all flow resources. |
| 74 | |
| 75 | This implementation does not implement the PCL_Flow_Lookup() function. |
| 76 | |
| 77 | Direct Transform Lookup |
| 78 | Transform records can be looked up directly (without the need for |
| 79 | a flow record). Use the function PCL_DTL_Transform_Add to add an |
| 80 | existing record (already registered with PCL_Transform_Register) |
| 81 | to the lookup table. PCL_DTL_Transform_Remove removes the transform |
| 82 | record from the lookup table again. |
| 83 | |
| 84 | Record invalidation |
| 85 | When flow records or transform records must be removed from the |
| 86 | system, they must also be removed from the record cache of the packet |
| 87 | engine, by means of a special command descriptor. The following |
| 88 | steps are required: |
| 89 | - Prevent the record from being looked up again. Use PCL_Flow_Remove |
| 90 | to remove a flow record from the lookup table. Use |
| 91 | PCL_DTL_Transform_Remove to remove a transform record from the lookup |
| 92 | table. |
| 93 | - Submit a special command descriptor containing a record invalidation |
| 94 | command with PEC_Packet_Put. |
| 95 | - Wait until the corresponding result descriptor is received with |
| 96 | PEC_Packet_Get. |
| 97 | |
| 98 | Byte ordering (endianness) |
| 99 | The transform buffers are considered arrays of 32-bit integers, in host- |
| 100 | native byte ordering. The driver will change the byte order if this is |
| 101 | required (configurable). The data buffers are considered byte arrays and |
| 102 | the driver will not touch these. Flow records are allocated and managed |
| 103 | entirely by the driver and their internal representation is never used by |
| 104 | the application. |
| 105 | |
| 106 | Bounce Buffers |
| 107 | The PCL API does not use bounce buffers. Transform buffers can be |
| 108 | allocated by the application in a DMA-safe way and buffers for the |
| 109 | flow table are allocated entirely within the Driver. |
| 110 | |
| 111 | Banks for DMA resources. |
| 112 | The PCL API requires that all DMA buffers for transform records |
| 113 | are allocated in Bank 1 (Bank parameter in |
| 114 | DMABuf_Properties_t). Internally the PCL adapter takes care of |
| 115 | allocating DMA-safe buffers for flow records in Bank 2. Memory |
| 116 | allocations in Bank 1 and Bank 2 are taken from special-purpose |
| 117 | memory pools. On 64-bit systems these pools are guaranteed to lay |
| 118 | in a 4GB address range. The PCL implementation takes care to convert |
| 119 | 64-bit bus addresses for flow and transform records to 32-bit offsets |
| 120 | with respect to the base address of the pool. |
| 121 | |
| 122 | Concurrent Context Synchronization (CCS) |
| 123 | The PCL API implementation supports one single multi-threaded application, |
| 124 | allowing concurrent and independent use of PCL_Init/UnInit, PCL_Flow and |
| 125 | PCL_Transform functions. Multiple applications for the PCL API are |
| 126 | not supported. |
| 127 | |
| 128 | The PCL API implementation provides synchronization mechanisms for |
| 129 | concurrent contexts invoking the API functions. The API can be used |
| 130 | by multi-threaded user-space or kernel-space applications. The latter |
| 131 | can invoke the API functions from the user process as well as from softirq |
| 132 | contexts. Mixing user process execution with softirq contexts is also |
| 133 | supported. Both Linux Uni-Processor (UP) and Symmetric Multi-Processor |
| 134 | (SMP) kernel configurations are supported. |
| 135 | |
| 136 | All the PCL API functions allow for just one execution context at a time. |
| 137 | Note that although the API functions take the interface ID as an input |
| 138 | parameter it is not used in the current implementation. |
| 139 | |
| 140 | The PCL API implementation serializes the access from concurrent execution |
| 141 | contexts to the classification device so having multiple contexts using |
| 142 | this API will not result in better performance. |
| 143 | |
| 144 | When a function from the PCL API detects that it competes for a resource |
| 145 | already used at the time by another context executing the PCL code it will |
| 146 | not block and return PCL_STATUS_BUSY return code. The caller should try |
| 147 | calling this function again short after. |
| 148 | |
| 149 | |
| 150 | <end of document> |