AMAB interconnector PL301(二)

1)Frequency Conversion Components:包含三种component。

   AXI-AXI async bridge:拥有两种mode:bypass mode 和 async mode。所以需要外加一个independent clock来管理这个独立

                                    的 clock domain。内部有两个信号SYNCMODEREQ(SI 端),SYNCMODEACK(MI 端)。

                                      其中SYNCMODEACK信号由AWSYNCMODEACK,WSYNCMODEACK,BSYNCMODEACK,

                                                                          ARSYNCMODEACK,RSYNCMODEACK相与来实现。

                                      bypass mode中SYNCMODEREQ是"1"请求,async mode中SYNCMODEREQ是"0"请求。

                                      在async实现中,所有的axi signal都会有register,为了跨时钟域会有3-4个时钟的buffer。

                                                            所有的transaction都会有FIFO。

   Upwards-sync bridge:SI 端是slower external clock domain,MI 端是faster external clock domain。

                                    global 信号有ACLKS,ACLKM,ACLKEN,ARESETn。其中ACLKEN信号由外部的clock controller来

                                    决定。控制SI和MI端口的clock。

                                    其他的signal,加register来同步。

   Downwards-sync bridge:类似。

   

2)Data Bus Width Conversion Components:包含两种component。Downsizer,Expander两种component类似。

   Downsizer:拥有两种模式:Pass-through mode 和 Downsizer mode。

                    在Downsizer mode中,该component 主要是处理data channel的multiplexing。将一个transaction根据

                                           AxBURST,AxLEN,AxADDR分为多个transaction。其中涉及到address buffer/transition,

                                           Transaction storage,Response signal,Atomic/Exclusive access。

                     address buffer/transition:AxSIZE需要根据MI接口定义来修改,FIXED/INCR burst变为 INCR burst,多个max length  

                                                           与一个short length的burst的和,来保证传递的字节数一致。     

                                                           对于WRAP burst,not-align时,第一个burst先变为一个align address INCR在MI端

                                                           口,align时,直接变为INCR burst但是address的处理与上一种FIXED/INCR不一样。   

                     Response signal:对于在MI端口分开的transaction,收到的response是与transaction数目相同的。所以返回

                                               到SI端口的response是经过处理的。详见相册。

                     Atomic/Exclusive access:这两种transaction再分为多个transaction时,情况比较特殊。

                                                           Atomic分成的多个transaction,除了最后一个子transaction是normal,其他所有的

                                                                     transaction都是lock的。

                                                           Exclusive在read/write操作时,可以分为多个transaction仍保持exclusive不变。并各个

                                                                     response都返回exokay。但是必须保证address不会overlapping。针对WRAP burst。

3)Interface Conversion Components:包含4中bridge,AHB-Lite到AXI,AHB-Lite到AXI for memory optimized.

                                                                             AXI到AHB-Lite,AXI到APB。  

                     AHB-Lite到AXI:Fixed-length AHB burst转换为等长的AXI burst。

                                            Undefined-length burst转换为single transfer。  

                                            由于AHB不支持secure,所以AxPORT[1] tie to "0",所有的access都允许。  

                                            生成的AXI中不支持write strobe。

                                          其他信号,如HWDATA,HRDATA,WDATA,RDATA,HWUSER,HRUSER,WUSER,RUSER都处于bypass mode

                     AHB-Lite到AXI memory optimized:应用HPROT信号,实现buffer writes,来提高总线性能。

                                                                        支持broken bursts,

                                                                        并且undefined-length INCR burst转换为INCR4 burst来提高通信速率。

                                                                                  (拥有一个4depth的buffer,并且address 不超过4KB)

                                                                        register AHB signals来提高timing。 

                                             对于AHB侧的lock access,IDLE cycle不可以分隔一个locked access,但是IDLE transfer可以。

                                                  主要区别是HREADY是否会有效,这样AHB才会重新采样HMASTLOCK的值。

                    AXI到APB:因为生产的APB会有PSEL信号,所以bridge内部需要实现一个multiplexor来选择不同slave的PRDATA,PREADY,

                                     PSLVERR信号。因APB总线速率较慢,所以bridge支持PREADY来表示slave插入wait states。

 

4)Register slice:有三种模式,Fully registerred, Registered forward path, Static bypass。

                         Fully registerred是一个default configeration,完全隔离Valid and AXI payload和 AXI ready信号。

                         Registered forward path,valid and payload signals被isolated,ready信号是一个combination path。 

                         Static bypass:no timing isolated。

posted @ 2015-08-26 10:54  _9_8  阅读(1420)  评论(0编辑  收藏  举报