转载-PCIE

时间:2024-01-31 14:31:16

http://www.hudong.com/wiki/PCI-E

summary:

基于高速序列构架产生了很多传输标准。包括HyperTransport,InfiniBand,RapidIO和StarFabric等等。这些均有业 界的不同企业支持,背后也都有大量的资金投入标准的研究开发,所以每一标准都声称自己与众不同,独占优势。主要的差异在于可扩展性、灵活性与反应时间、单 位成本的取舍平衡各不相同。其中的一个例子是在传输包上增加一个复杂的头信息以支持复杂路由传输(PCI Express不支持这种方式)。这样的信息增加降低了接口的有效带宽也使传输更复杂,但是相应创造了新的软件支持此功能。这种架构下需要软件追踪网络拓 扑结构的变化以实现系统支持热插拔。InfiniBand 和 StarFabric 标准即能实现这以功能。另一个例子是缩小信息包以减少反应时间。较小的信息包意味着包头占用了包的更大百分比,这样又降低了有效带宽。能实现此功能的标准 是RapidIO 和HyperTransport。PCI Express取中庸之道,定位于设计成一种系统互连接口(总线)而非一种设备接口或路由网络协议。另外为了针对软件透明,它的设计目标限制了它作为协 议,也在某种程度上增加了它的反应时间

 

 

============================

5.1 TLP的格式

(2011-08-08 16:21:32)
转载
标签:

杂谈

分类: 浅谈PCIe体系结构

当处理器或者其他PCIe设备访问PCIe设备时,所传送的数据报文首先通过事务层被封装为一个或者多个TLP,之后才能通过PCIe总线的各个层次发送出去。TLP的基本格式如图5‑1所示。

 

 

 

一个完整的TLP由1个或者多个TLP Prefix、TLP头、Data Payload(数据有效负载)和TLP Digest组成。TLP头是TLP最重要的标志,不同的TLP其头的定义并不相同。TLP头包含了当前TLP的总线事务类型、路由信息等一系列信息。在一个TLP中,Data Payload的长度可变,最小为0,最大为1024DW。

TLP Digest是一个可选项, 一个TLP是否需要TLP Digest由TLP头决定。Data Payload也是一个可选项,有些TLP并不需要Data Payload,如存储器读请求、配置和I/O写完成TLP并不需要Data Payload。

TLP Prefix由PCIe V2.1总线规范引入,分为Local TLP Prefix和EP-EP TLP Prefix两类。其中Local TLP Prefix的主要作用是在PCIe链路的两端传递消息,而EP-EP TLP Prefix的主要作用是在发送设备和接收设备之间传递消息。设置TLP Prefix的主要目的是为了扩展TLP头,并以此支持PCIe V2.1规范的一些新的功能。

TLP头由3个或者4个双字(DW)组成。其中第一个双字中保存通用TLP头,其他字段与通用TLP头的Type字段相关。一个通用TLP头由Fmt、Type、TC、Length等字段组成,如图5‑2所示。

 

如果存储器读写TLP支持64位地址模式时,TLP头的长度为4DW,否则为3DW。而完成报文的TLP头不含有地址信息,使用的TLP头长度为3DW。其中Byte 4~Byte 15的格式与TLP相关,下文将结合具体的TLP介绍这些字段。

5.1.1 通用TLP头的Fmt字段和Type字段

Fmt和Type字段确认当前TLP使用的总线事务,TLP头的大小是由3个双字还是4个双字组成,当前TLP是否包含有效负载。其具体含义如表5‑1所示。

 表5‑1 Fmt[1:0]字段

Fmt[2:0]

TLP的格式

0b000

TLP大小为3个双字,不带数据。

0b001

TLP大小为4个双字,不带数据。

0b010

TLP大小为3个双字,带数据。

0b011

TLP大小为4个双字,带数据。

0b100

TLP Prefix

其他

PCIe总线保留

 

其中所有读请求TLP都不带数据,而写请求TLP带数据,而其他TLP可能带数据也可能不带数据,如完成报文可能含有数据,也可能仅含有完成标志而并不携带数据。在TLP的Type字段中存放TLP的类型,即PCIe总线支持的总线事务。该字段共由5位组成,其含义如表5‑2所示。

表5‑2 Type[4:0]字段

TLP类型

Fmt[2:0]

Type[4:0]

描述

MRd

0b000

0b001

0b0 0000

存储器读请求;TLP头大小为3个或者4个双字,不带数据。

MRdLk

0b000

0b001

0b0 0001

带锁的存储器读请求;TLP头大小为3个或者4个双字,不带数据。

MWr

0b010

0b011

0b0 0000

存储器写请求;TLP头大小为3个或者4个双字,带数据。

IORd

0b000

0b0 0010

IO读请求;TLP头大小为3个双字,不带数据。

IOWr

0b010

0b0 0010

IO写请求;TLP头大小为3个双字,带数据。

CfgRd0

0b000

0b0 0100

配置0读请求;TLP头大小为3个双字,不带数据。

CfgWr0

0b010

0b0 0100

配置0写请求;TLP头大小为3个双字,带数据。

CfgRd1

0b000

0b0 0101

配置1读请求;不带数据。

CfgWr1

0b010

0b0 0101

配置1写请求;带数据。

TCfgRd

0b010

0b1 1011

本书对这两种总线事务不做介绍。

TCfgWr

0b001

0b1 1011

Msg

0b001

0b1 0r2r1r0

消息请求;TLP头大小为4个双字,不带数据。“rrr”字段是消息请求报文的Route字段,下文将详细介绍该字段。

MsgD

0b011

0b1 0r2r1r0

消息请求;TLP头大小为4个双字,带数据。

Cpl

0b000

0b0 1010

完成报文;TLP头大小为3个双字,不带数据。包括存储器、配置和I/O写完成。

CplD

0b001

0b0 1010

带数据的完成报文,TLP头大小为3个双字,包括存储器读、I/O读、配置读和原子操作读完成。

CplLk

0b000

0b0 1011

锁定的完成报文,TLP头大小为3个双字,不带数据。

CplDLk

0b010

0b0 1011

带数据的锁定完成报文,TLP头大小为3个双字,带数据。

FetchAdd

0b010

0b011

0b0 1100

Fetch and Add原子操作。

Swap

0b010

0b011

0b0 1101

Swap原子操作。

CAS

0b010

0b011

0b0 1110

CAS原子操作。

LPrfx

0b100

0b0 L3L2L1L0

Local TLP Prefix

EPrfx

0b100

0b1 E3E2E1E0

End-End TLP Prefix

由上表所示,存储器读和写请求,IO读和写请求,及配置读和写请求的type字段相同,如存储器读和写请求的Type字段都为0b0 0000。此时PCIe总线规范使用Fmt字段区分读写请求,当Fmt字段是“带数据”的报文,一定是“写报文”;当Fmt字段是“不带数据”的报文,一定是“读报文”。

PCIe总线的数据报文传送方式与PCI总线数据传送有类似之处。其中存储器写TLP使用Posted方式进行传送,而其他总线事务使用Non-Posted方式。

PCIe总线规定所有Non-Posted存储器请求使用Split总线方式进行数据传递。当PCIe设备进行存储器读、I/O读写或者配置读写请求时,首先向目标设备发送数据读写请求TLP,当目标设备收到这些读写请求TLP后,将数据和完成信息通过完成报文(Cpl或者CplD)发送给源设备。

其中存储器读、I/O读和配置读需要使用CplD报文,因为目标设备需要将数据传递给源设备;而I/O写和配置写需要使用Cpl报文,因为目标设备不需要将任何数据传递给源设备,但是需要通知源设备,写操作已经完成,数据已经成功地传递给目标设备。

在PCIe总线中,进行存储器或者I/O写操作时,数据与数据包头一起传递;而进行存储器或者I/O读操作时,源设备首先向目标设备发送读请求TLP,而目标设备在准备好数据后,向源设备发出完成报文。

PCIe总线规范还定义了MRdLk报文,该报文的主要作用是与PCI总线的锁操作相兼容,但是PCIe总线规范并不建议用户使用这种功能,因为使用这种功能将极大影响PCIe总线的数据传送效率。

与PCI总线并不相同,PCIe总线规范定义了Msg报文,即消息报文。分别为Msg和MsgD,这两种报文的区别在于一个报文可以传递数据,一个不能传递数据。

PCIe V2.1总线规范还补充了一些总线事务,如FetchAdd、Swap、CAS、LPrfx和EPrfx。其中LPrfx和EPrfx总线事务分别与Local TLP Prefix和EP-EP TLP Prefix对应。在PCIe总线规范V2.0中,TLP头的大小为1DW,而使用LPrfx和EPrfx总线事务可以对TLP头进行扩展,本节不对这些TLP Prefix做进一步介绍。PCIe设备可以使用FetchAdd、Swap和CAS总线事务进行原子操作,本篇将在第5.3.5节详细介绍该类总线事务。

5.1.2 TC字段

TC字段表示当前TLP的传送类型,PCIe总线规定了8种传输类型,分别为TC0~TC7,缺省值为TC0,该字段与PCIe的QoS相关。PCIe设备使用TC区分不同类型的数据传递,而多数EP中只含有一个VC,因此这些EP在发送TLP时,也仅仅使用TC0,但是有些对实时性要求较高的EP中,含有可以设置TC字段的寄存器。

在Intel的高精度声卡控制器(High Definition Audio Controller)的扩展配置空间中含有一个TCSEL寄存器。系统软件可以设置该寄存器,使声卡控制器发出的TLP使用合适的TC。声卡控制器可以使用TC7传送一些对实时性要求较强的控制信息,而使用TC0传送一般的数据信息。在具体实现中,一个EP也可以将控制TC字段的寄存器放入到设备的BAR空间中,而不必和Intel的高精度声卡控制器相同,存放在PCI配置空间中。

目前许多处理器系统的RC仅支持一个VC通路,此时EP使用不同的TC进行传递数据的意义不大。x86处理器的MCH中一般支持两个VC通路,而多数PowerPC处理器仅支持一个VC通路。PLX公司的多数Switch也仅支持两个VC通路。

有些RC,如MPC8572处理器,也能决定其发出TLP使用的TC。在该处理器的PCIe Outbound窗口寄存器(PEXOWARn)中,含有一个TC字段,通过设置该字段可以确定RC发出的TLP使用的TC字段。不同的TC可以使用PCIe链路中的不同VC,而不同的VC的仲裁级别并不相同。EP或者RC通过调整其发出TLP的TC字段,可以调整TLP使用的VC,从而调整TLP的优先级。

5.1.3 Attr字段

Attr字段由3位组成,其中第2位表示该TLP是否支持PCIe总线的ID-based Ordering;第1位表示是否支持Relaxed Ordering;而第0位表示该TLP在经过RC到达存储器时,是否需要进行Cache共享一致性处理。Attr字段如图5‑3所示。

 

 

 

一个TLP可以同时支持ID-based Ordering和Relaxed Ordering两种位序。Relaxed Ordering最早在PCI-X总线规范中提出,用来提高PCI-X总线的数据传送效率;而ID-based Ordering由PCIe V2.1总线规范提出。TLP支持的序如表5‑3所示。

 表5‑3 TLP支持的序

Attr[2]

Attr[1]

类型

0

0

缺省序,即强序模型

0

1

PCI-X Relaxed Ordering模型

1

0

ID-Based Ordering(IDO)模型

1

1

同时支持Relaxed Ordering和IDO模型

 

当使用标准的强序模型时,在数据的整个传送路径中,PCIe设备在处理相同类型的TLP时,如PCIe设备发送两个存储器写TLP时,后面的存储器写TLP必须等待前一个存储器写TLP完成后才能被处理,即便当前报文在传送过程中被阻塞,后一个报文也必须等待。

如果使用Relaxed Ordering模型,后一个存储器写TLP可以穿越前一个存储器写TLP,提前执行,从而提高了PCIe总线的利用率。有时一个PCIe设备发出的TLP,其目的地址并不相同,可能先进入发送队列的TLP,在某种情况下无法发送,但这并不影响后续TLP的发送,因为这两个TLP的目的地址并不相同,发送条件也并不相同。

值得注意的是,在使用PCI总线强序模型时,不同种类的TLP间也可以乱序通过同一条PCIe链路,比如存储器写TLP可以超越存储器读请求TLP提前进行。而PCIe总线支持Relaxed Ordering模型之后,在TLP的传递过程中出现乱序种类更多,但是这些乱序仍然是有条件限制的。在PCIe总线规范中为了避免死锁,还规定了不同报文的传送数据规则,即Ordering Rules。

PCIe V2.1总线规范引入了一种新的“序”模型,即IDO(ID-Based Ordering)模型,IDO模型与数据传送的数据流相关,是PCIe V2.1规范引入的序模型。

Attr字段的第0位是“No Snoop Attribute”位。当该位为0时表示当前TLP所传送的数据在通过FSB时,需要与Cache保持一致,这种一致性由FSB通过总线监听自动完成而不需要软件干预;如果为1,表示FSB并不会将TLP中的数据与Cache进行一致,在这种情况下,进行数据传送时,必须使用软件保证Cache的一致性。

在PCI总线中没有与这个“No Snoop Attribute”位对应的概念,因此一个PCI设备对存储器进行DMA操作时会进行Cache一致性操作[1]。这种“自动的”Cache一致性行为在某些特殊情况下并不能带来更高的效率。

当一个PCIe设备对存储器进行DMA读操作时,如果传送的数据非常大,比如512MB,Cache的一致性操作不但不会提高DMA写的效率,反而会降低。因为这个DMA读访问的数据在绝大多数情况下,并不会在Cache中命中,但是FSB依然需要使用Snoop Phase进行总线监听。而处理器在进行Cache一致性操作时仍然需要占用一定的时钟周期,即在Snoop Phase中占用的时钟周期,Snoop Phase是FSB总线事务的一个阶段,如图3‑6所示。

对于这类情况,一个较好的做法是,首先使用软件指令保证Cache与主存储器的一致性,并置“No Snoop Attribute”位为1[2],然后再进行DMA读操作。同理使用这种方法对一段较大的数据区域进行DMA写时,也可以提高效率。

除此之外,当PCIe设备访问的存储器,不是“可Cache空间”时,也可以通过设置“No Snoop Attribute”位,避免FSB的Cache共享一致性操作,从而提高FSB的效率。“No Snoop Attribute”位是PCIe总线针对PCI总线的不足,所作出的重要改动。

5.1.4 通用TLP头中的其他字段

除了Fmt和Type字段外,通用TLP头还含有以下字段。

1 TH位、TD位和EP位

TH位为1表示当前TLP中含有TPH(TLP Processing Hint)信息,TPH是PCIe V2.1总线规范引入的一个重要功能。TLP的发送端可以使用TPH信息,通知接收端即将访问数据的特性,以便接收端合理地预读和管理数据,TPH的详细介绍见第5.3.6节。

TD位表示TLP中的TLP Digest是否有效,为1表示有效,为0表示无效。而EP位表示当前TLP中的数据是否有效,为1表示无效,为0表示有效。

2 AT字段

AT字段与PCIe总线的地址转换相关。在一些PCIe设备中设置了ATC(Address Translation Cache)部件,这个部件的主要功能是进行地址转换。只有在支持IOMMU技术的处理器系统中,PCIe设备才能使用该字段。

AT字段可以用作存储器域与PCI总线域之间的地址转换,但是设置这个字段的主要目的是为了方便多个虚拟主机共享同一个PCIe设备。对这个字段有兴趣的读者可以参考Address Translation Sevices规范,这个规范是PCI的IO Virtualization规范的重要组成部分。对虚拟化技术有兴趣的读者可以参考清华大学出版社的《系统虚拟化——原理与实现》,以获得基本的关于虚拟化的入门知识。

3 Length字段

Length字段用来描述TLP的有效负载(Data Payload)大小[3]。PCIe总线规范规定一个TLP的Data Payload的大小在1B~4096B之间。PCIe总线设置Length字段的目的是提高总线的传送效率。

当PCI设备在进行数据传送时,其目标设备并不知道实际的数据传送大小,这在一定程度上影响了PCI总线的数据传送效率。而在PCIe总线中,目标设备可以通过Length字段提前获知源设备需要发送或者请求的数据长度,从而合理地管理接收缓冲,并根据实际情况进行Cache一致性操作。

当PCI设备进行DMA写操作,将PCI设备中4KB大小的数据传送到主存储器时,这个PCI设备的DMA控制器将存放传送的目的地址和传送大小,然后启动DMA写操作,将数据写入到主存储器。由于PCI总线是一条共享总线,因此传送4KB大小的数据,可能会使用若干个PCI总线写事务才能完成[4],而每一个PCI总线写事务都不知道DMA控制器何时才能将数据传送完毕。

如果这些总线写事务还通过一系列PCI桥才能到达存储器,在这个路径上的每一个PCI桥也无法预知,何时这个DMA操作才能结束。这种“不可预知”将导致PCI总线的带宽不能被充分利用,而且极易造成PCI桥数据缓冲的浪费。

而PCIe总线通过TLP的Length字段,可以有效避免PCIe链路带宽的浪费。值得注意的是,Length字段以DW为单位,其最小单位为1个DW。如果PCIe主设备传送的单位小于1个DW或者传送的数据并不以DW对界时,需要使用字节使能字段,即“DW BE”字段。有关“DW BE”字段的详细说明见第5.3.1节。



[1] PowerPC处理器通过设置Inbound寄存器,也可以避免这个Cache一致性操作。

[2] FSB收到这类TLP后,不进行Cache一致性操作。

[3] 存储器读请求TLP没有DataPayload字段,此时该TLP使用Length字段表示需要读取多少数据。

[4] 当多个PCI设备共享一条PCI总线时,一个设备不会长时间占用PCI总线,这个设备在使用这条PCI总线一定的时间后,将让出PCI总线的使用权。

 

5.4 TLP中与数据负载相关的参数

(2011-08-09 11:36:53)
转载
标签:

杂谈

分类: 浅谈PCIe体系结构

在PCIe总线中,有些TLP含有Data Payload,如存储器写请求、存储器读完成TLP等。在PCIe总线中,TLP含有的Data Payload大小与Max_Payload_Size、Max_Read_Request_Size和RCB参数相关。下文将分别介绍这些参数的使用。

5.4.1 Max_Payload_Size参数

PCIe总线规定在TLP报文中,数据有效负载的最大值为4KB,但是PCIe设备并不一定能够发送这么大的数据报文。PCIe设备含有“Max_Payload_Size”和“Max_Payload_Size Supported”参数,这两个参数分别在Device Capability寄存器和Device Control寄存器中定义。

“Max_Payload_Size Supported”参数存放在一个PCIe设备中,TLP有效负载的最大值,该参数由PCIe设备的硬件逻辑确定,系统软件不能改写该参数。而Max_Payload_Size参数存放PCIe设备实际使用的,TLP有效负载的最大值。该参数由PCIe链路两端的设备协商决定,是PCIe设备进行数据传送时,实际使用的参数。

PCIe设备发送数据报文时,使用Max_Payload_Size参数决定TLP的最大有效负载。当PCIe设备的所传送的数据大小超过Max_Payload_Size参数时,这段数据将被分割为多个TLP进行发送。当PCIe设备接收TLP时,该TLP的最大有效负载也不能超过Max_Payload_Size参数,如果接收的TLP,其Length字段超过Max_Payload_Size参数,该PCIe设备将认为该TLP非法。

RC或者EP在发送存储器读完成TLP时,这个存储器读完成TLP的最大Payload也不能超过Max_Payload_Size参数,如果超过该参数,PCIe设备需要发送多个读完成报文。值得注意的是,这些读完成报文需要满足RCB参数的要求,有关RCB参数的详细说明见下文。

在实际应用中,尽管有些PCIe设备的Max_Payload_Size Supported参数可以为256B、512B、1024B或者更高,但是如果PCIe链路的对端设备可以支持的Max_Payload_Size参数为128B时,系统软件将使用对端设备的Max_Payload_Size Supported参数,初始化该设备的Max_Payload_Size参数,即选用PCIe链路两端最小的Max_Payload_Size Supported参数初始化Max_Payload_Size参数。

在多数x86处理器系统的MCH或者ICH中,Max_Payload_Size Supported参数为128B。这也意味着在x86处理器中,与MCH或者ICH直接相连的PCIe设备进行DMA读写时,数据的有效负载不能超过128B。而在PowerPC处理器系统中,该参数大多为256B。

目前在大多数EP中,Max_Payload_Size Supported参数不大于512B,因为在大多数处理器系统的RC中,Max_Payload_Size Supported参数也不大于512B。因此即便EP支持较大的Max_Payload_Size Supported参数,并不会提高数据传送效率。

而Max_Payload_Size参数的大小与PCIe链路的传送效率成正比,该参数越大,PCIe链路带宽的利用率越高,该参数越小,PCIe链路带宽的利用率越低。

PCIe总线规范规定,对于实时性要求较高的PCIe设备,Max_Payload_Size参数不应设置过大,因此这个参数有时会低于PCIe链路允许使用的最大值。

5.4.2 Max_Read_Request_Size参数

Max_Read_Request_Size参数由PCIe设备决定,该参数规定了PCIe设备一次能从目标设备读取多少数据。

Max_Read_Request_Size参数在Device Control寄存器中定义。该参数与存储器读请求TLP的Length字段相关,其中Length字段不能大于Max_Read_Request_Size参数。在存储器读请求TLP中,Length字段表示需要从目标设备读取多少数据。

值得注意的是,Max_Read_Request_Size参数与Max_Payload_Size参数间没有直接联系,Max_Payload_Size参数仅与存储器写请求和存储器读完成报文相关。

PCIe总线规定存储器读请求,其读取的数据长度不能超过Max_Read_Request_Size参数,即存储器读TLP中的Length字段不能大于这个参数。如果一次存储器读操作需要读取的数据范围大于Max_Read_Request_Size参数时,该PCIe设备需要向目标设备发送多个存储器读请求TLP。

PCIe总线规定Max_Read_Request_Size参数的最大值为4KB,但是系统软件需要根据硬件特性决定该参数的值。因为PCIe总线规定EP在进行存储器读请求时,需要具有足够大的缓冲接收来自目标设备的数据。

如果一个EP的Max_Read_Request_Size参数被设置为4KB,而且这个EP每发出一个4KB大小存储器读请求时,EP都需要准备一个4KB大小的缓冲[1]。这对于绝大多数EP,这都是一个相当苛刻的条件。为此在实际设计中,一个EP会对Max_Read_Request_Size参数的大小进行限制。

5.4.3 RCB参数

RCB位在Link Control寄存器中定义。RCB位决定了RCB参数的值,在PCIe总线中,RCB参数的大小为64B或者128B,如果一个PCIe设备没有设置RCB的大小[2],则RC的RCB参数缺省值为64B,而其他PCIe设备的RCB参数的缺省值为128B。PCIe总线规定RC的RCB参数的值为64B或者128B,其他PCIe设备的RCB参数为128B。

在PCIe总线中,一个存储器读请求TLP可能收到目标设备发出的多个完成报文后,才能完成一次存储器读操作。因为在PCIe总线中,一个存储器读请求最多可以请求4KB大小的数据报文,而目标设备可能会使用多个存储器读完成TLP才能将数据传递完毕。

当一个EP向RC或者其他EP读取数据时,这个EP首先向RC或者其他EP发送存储器读请求TLP;之后由RC或者其他EP发送存储器读完成TLP,将数据传递给这个EP。

如果存储器读完成报文所传递数据的地址范围没有跨越RCB参数的边界,那么数据发送端只能使用一个存储器完成报文将数据传递给请求方,否则可以使用多个存储器读完成TLP。

假定一个EP向地址范围为0xFFFF-0000~0xFFFF-0010这段区域进行DMA读操作,RC收到这个存储器读请求TLP后,将组织存储器读完成TLP,由于这段区域并没有跨越RCB边界,因此RC只能使用一个存储器读完成TLP完成数据传递。

如果存储器读完成报文所传递数据的地址范围跨越了RCB边界,那么数据发送端(目标设备)可以使用一个或者多个完成报文进行数据传递。数据发送端使用多个存储器读完成报文完成数据传递时,需要遵循以下原则。

  • 第一个完成报文所传送的数据,其起始地址与要求的起始地址相同。其结束地址或者为要求的结束地址(使用一个完成报文传递所有数据),或者为RCB参数的整数倍(使用多个完成报文传递数据)。
  • 最后一个完成报文的起始地址或者为要求的起始地址(使用一个完成报文传递所有数据),或者为RCB参数的整数倍(使用多个完成报文传递数据)。其结束地址必须为要求的结束地址。
  • 中间的完成报文的起始地址和结束地址必须为RCB参数的整数倍。

当RC或者EP需要使用多个存储器读完成报文将0xFFFE-FFF0~0xFFFF-00C7之间的数据发送给数据请求方时,可以将这些完成报文按照表5‑9方式组织。

 表5‑9 存储器读完成报文的拆分方法

方式1

方式2

方式3

0xFFFE-FFF0~0xFFFE-FFFF

0xFFFE-FFF0~0xFFFE-FFFF

0xFFFE-FFF0~0xFFFE-FFFF

0xFFFF-0000~0xFFFF-003F

0xFFFF-0000~0xFFFF-007F

0xFFFF-0000~0xFFFF-00C7

0xFFFF-0040~0xFFFF-007F

0xFFFF-0080~0xFFFF-00C7

 

0xFFFF-0080~0xFFFF-00BF

 

 

0xFFFF-00C0~0xFFFF-00C7

 

 

 

上表提供的方式仅供参考,目标设备还可以使用其他拆分方法发送存储器读完成TLP。PCIe总线使用多个完成报文实现一次数据读请求的主要原因是考虑Cache行长度和流量控制。在多数x86处理器系统中,存储器读完成报文的数据长度为一个Cache行,即一次传送64B。除此之外,较短的数据完成报文占用流量控制的资源较少,而且可以有效避免数据拥塞。

5.5 小结

   本章重点介绍PCIe总线的事务层。在PCIe总线层次结构中,事务层最易理解,同时也与系统软件直接相关。


[1] 这是流量控制Infinite FC Unit的要求,详见第9.3.2节。

[2] 有些PCIe设备可能没有Link Control寄存器。

 

 

PCI Express

From Wikipedia, the free encyclopedia
  (Redirected from PCIe)
PCI Express
}}
Year created 2004
Created by Intel · Dell · IBM · HP
Supersedes AGP · PCI · PCI-X
Width in bits 1–32
Number of devices

One device each on each endpoint of each connection.

PCI Express switches can create multiple endpoints out of one endpoint to allow sharing one endpoint with multiple devices.
Capacity

Per lane (each direction):

  • v1.x: 250 MB/s (2.5 GT/s)
  • v2.x: 500 MB/s (5 GT/s)
  • v3.0: 1 GB/s (8 GT/s)

16 lane slot (each direction):

  • v1.x: 4 GB/s (40 GT/s)
  • v2.x: 8 GB/s (80 GT/s)
  • v3.0: 16 GB/s (128 GT/s)
Style Serial
Hotplugging interface Yes, if ExpressCard or PCI Express ExpressModule
External interface Yes, with PCI Express External Cabling such as Intel Thunderbolt.

PCI Express (Peripheral Component Interconnect Express), officially abbreviated as PCIe, is a computer expansion card standard designed to replace the older PCI, PCI-X, and AGP bus standards. PCIe has numerous improvements over the aforementioned bus standards, including higher maximum system bus throughput, lower I/O pin count and smaller physical footprint, better performance-scaling for bus devices, a more detailed error detection and reporting mechanism, and native hot plug functionality. More recent revisions of the PCIe standard support hardware I/O virtualization.

The PCIe electrical interface is also used in a variety of other standards, most notably ExpressCard, a laptop expansion card interface.

Format specifications are maintained and developed by the PCI-SIG (PCI Special Interest Group), a group of more than 900 companies that also maintain the Conventional PCI specifications. PCIe 3.0 is the latest standard for expansion cards that is available on mainstream personal computers.[1][2]

Contents

 [hide

[edit] Applications

PCI Express is used in consumer, server, and industrial applications, as a motherboard-level interconnect (to link motherboard-mounted peripherals), a passive backplane interconnect and as an expansion card interface for add-in boards.

In virtually all modern PCs, from consumer laptops and desktops to enterprise data servers, the PCIe bus serves as the primary motherboard-level interconnect, connecting the host system processor with both integrated-peripherals (surface mounted ICs) and add-on peripherals (expansion cards.) In most of these systems, the PCIe bus co-exists with 1 or more legacy PCI busses, for backward compatibility with the large body of legacy PCI peripherals.

[edit] Architecture

Conceptually, the PCIe bus is like a high-speed serial replacement of the older PCI/PCI-X bus,[3] an interconnect bus using shared address/data lines.

A key difference between PCIe bus and the older PCI is the bus topology. PCI uses a shared parallel bus architecture, where the PCI host and all devices share a common set of address/data/control lines. In contrast, PCIe is based on point-to-point topology, with separate serial links connecting every device to the root complex (host). Due to its shared bus topology, access to the older PCI bus is arbitrated (in the case of multiple masters), and limited to 1 master at a time, in a single direction. Furthermore, the older PCI\'s clocking scheme limits the bus clock to the slowest peripheral on the bus (regardless of the devices involved in the bus transaction). In contrast, a PCIe bus link supports full-duplex communication between any two endpoints, with no inherent limitation on concurrent access across multiple endpoints.

In terms of bus protocol, PCIe communication is encapsulated in packets. The work of packetizing and de-packetizing data and status-message traffic is handled by the transaction layer of the PCIe port (described later). Radical differences in electrical signaling and bus protocol require the use of a different mechanical form factor and expansion connectors (and thus, new motherboards and new adapter boards); PCI slots and PCIe slots are not interchangeable. At the software level, PCIe preserves backward compatibility with PCI; legacy PCI system software can detect and configure newer PCIe devices without explicit support for the PCIe standard, though PCIe\'s new features are inaccessible.

The PCIe link between 2 devices can consist of anywhere from 1 to 32 lanes. In a multi-lane link, the packet data is striped across lanes, and peak data-throughput scales with the overall link width. The lane count is automatically negotiated during device initialization, and can be restricted by either endpoint. For example, a single-lane PCIe (x1) card can be inserted into a multi-lane slot (x4, x8, etc.), and the initialization cycle auto-negotiates the highest mutually supported lane count. The link can dynamically down-configure the link to use fewer lanes, thus providing some measure of failure tolerance in the presence of bad or unreliable lanes. The PCIe standard defines slots and connectors for multiple widths: x1, x4, x8, x16, x32. This allows PCIe bus to serve both cost-sensitive applications where high throughput is not needed, as well as performance-critical applications such as 3D graphics, network (10 Gigabit Ethernet, multiport Gigabit Ethernet), and enterprise storage (SAS, Fibre Channel.)

As a point of reference, a PCI-X (133 MHz 64 bit) device and PCIe device at 4-lanes (x4), Gen1 speed have roughly the same peak transfer rate in a single-direction: 1064MB/sec. The PCIe bus has the potential to perform better than the PCI-X bus in cases where multiple devices are transferring data communicating simultaneously, or if communication with the PCIe peripheral is bidirectional.

[edit] Interconnect

PCIe devices communicate via a logical connection called an interconnect[4] or link. A link is a point-to-point communication channel between 2 PCIe ports, allowing both to send/receive ordinary PCI-requests (configuration read/write, I/O read/write, memory read/write) and interrupts (INTx, MSI, MSI-X). At the physical level, a link is composed of 1 or more lanes.[4] Low-speed peripherals (such as an 802.11 Wi-Fi card) use a single-lane (×1) link, while a graphics adapter typically uses a much wider (and thus, faster) 16-lane link.

[edit] Lane

A lane is composed of a transmit and receive pair of differential lines. Each lane is composed of 4 wires or signal paths, meaning conceptually, each lane is a full-duplex byte stream, transporting data packets in 8 bit \'byte\' format, between endpoints of a link, in both directions simultaneously.[5] Physical PCIe slots may contain from one to thirty-two lanes, in powers of two (1, 2, 4, 8, 16 and 32).[4] Lane counts are written with an x prefix (e.g., x16 represents a sixteen-lane card or slot), with x16 being the largest size in common use.[6]

[edit] Serial bus

The bonded serial format was chosen over a traditional parallel bus format due to the latter\'s inherent limitations, including single-duplex operation, excess signal count and an inherently lower bandwidth due to timing skew. Timing skew results from separate electrical signals within a parallel interface traveling down different-length conductors, on potentially different printed circuit board layers, at possibly different signal velocities. Despite being transmitted simultaneously as a single word, signals on a parallel interface experience different travel times and arrive at their destinations at different moments. When the interface clock rate is increased to a point where its inverse (i.e., its clock period) is shorter than the largest possible time between signal arrivals, the signals no longer arrive with sufficient coincidence to make recovery of the transmitted word possible. Since timing skew over a parallel bus can amount to a few nanoseconds, the resulting bandwidth limitation is in the range of hundreds of megahertz.

A serial interface does not exhibit timing skew because there is only one differential signal in each direction within each lane, and there is no external clock signal since clocking information is embedded within the serial signal. As such, typical bandwidth limitations on serial signals are in the multi-gigahertz range. PCIe is just one example of a general trend away from parallel buses to serial interconnects. Other examples include Serial ATA, USB, SAS, FireWire and RapidIO.

Multichannel serial design increases flexibility by allocating slow devices to fewer lanes than fast devices.

[edit] Form factors

[edit] PCI Express (standard)

Various PCI slots. From top to bottom:
  • PCI Express ×4
  • PCI Express ×16
  • PCI Express ×1
  • PCI Express ×16
  • Conventional PCI (32-bit)

A PCIe card fits into a slot of its physical size or larger (maximum x16), but may not fit into a smaller PCIe slot (x16 in a x8 slot). Some slots use open-ended sockets to permit physically longer cards and negotiates the best available electrical connection. The number of lanes actually connected to a slot may also be less than the number supported by the physical slot size.

An example is a ×8 slot that actually only runs at ×1. These slots allow any ×1, ×2, ×4 or ×8 card, though only running at ×1 speed. This type of socket is called a ×8 (×1 mode) slot, meaning it physically accepts up to ×8 cards but only runs at ×1 speed. The advantage is that it can accommodate a larger range of PCIe cards without requiring motherboard hardware to support the full transfer rate. This keeps design and implementation costs down.

[edit] Pinout

The following table identifies the conductors on each side of the edge connector on a x4 PCI Express card. The solder side of the printed circuit board (PCB) is the A side, and the component side is the B side.[7]

PCI express ×4 connector pinout
PinSide BSide AComments
1 +12V PRSNT1# Pulled low to indicate card inserted
2 +12V +12V  
3 Reserved +12V
4 Ground Ground
5 SMCLK TCK SMBus and JTAG port pins
6 SMDAT TDI
7 Ground TDO
8 +3.3V TMS
9 TRST# +3.3V
10 +3.3Vaux +3.3V Standby power
11 WAKE# PWRGD Link reactivation, power good.
Key notch
12 Reserved Ground  
13 Ground REFCLK+ Reference clock differential pair
14 HSOp(0) REFCLK- Lane 0 transmit data, + and −
15 HSOn(0) Ground
16 Ground HSIp(0) Lane 0 receive data, + and −
17 PRSNT2# HSIn(0)
18 Ground Ground  
 
19 HSOp(1) Reserved Lane 1 transmit data, + and −
20 HSOn(1) Ground
21 Ground HSIp(1) Lane 1 receive data, + and −
22 Ground HSIn(1)
23 HSOp(2) Ground Lane 2 transmit data, + and −
24 HSOn(2) Ground
25 Ground HSIp(2) Lane 2 receive data, + and −
26 Ground HSIn(2)
27 HSOp(3) Ground Lane 3 transmit data, + and −
28 HSOn(3) Ground
29 Ground HSIp(3) Lane 3 receive data, + and −
30 Reserved HSIn(3)
31 PRSNT2# Ground  
32 Ground Reserved

An ×1 slot is a shorter version of this, ending after pin 18. ×8 and ×16 slots extend the pattern.

Legend
Ground pin Zero volt reference
Power pin Supplies power to the PCIe card
Output pin Signal from the card to the motherboard
Input pin Signal from the motherboard to the card
Open drain May be pulled low and/or sensed by multiple cards
Sense pin Tied together on card
Reserved Not presently used, do not connect

[edit] Power

PCI Express cards are allowed a maximum power consumption of 25W (×1: 10W for power-up). Low profile cards are limited to 10W (×16 to 25W). PCI Express Graphics (PEG) cards may increase power (from slot) to 75W after configuration (3.3V/3A + 12V/5.5A).[8] Optional connectors add 75W (6-pin) or 150W (8-pin) power for up to 300W total.

[edit] PCI Express Mini Card

A WLAN PCI Express Mini Card and its connector.
MiniPCI and MiniPCI Express cards in comparison

PCI Express Mini Card (also known as Mini PCI Express, Mini PCIe, and Mini PCI-E) is a replacement for the Mini PCI form factor, based on PCI Express. It is developed by the PCI-SIG. The host device supports both PCI Express and USB 2.0 connectivity, and each card may use either standard. Most laptop computers built after 2005 are based on PCI Express and can have several Mini Card slots.[citation needed]

[edit] Physical dimensions

PCI Express Mini Cards are 30×50.95 mm. There is a 52 pin edge connector, consisting of two staggered rows on a 0.8 mm pitch. Each row has 8 contacts, a gap equivalent to 4 contacts, then a further 18 contacts. A half-length card is also specified 30×26.8 mm. Cards have a thickness of 1.0 mm (excluding components).

[edit] Electrical interface

PCI Express Mini Card edge connector provide multiple connections and buses:

  • PCIe x1
  • USB 2.0
  • SMBus
  • Wires to diagnostics LEDs for wireless network (i.e., WiFi) status on computer\'s chassis
  • SIM card for GSM and WCDMA applications. (UIM signals on spec)
  • Future extension for another PCIe lane
  • 1.5 and 3.3 volt power

[edit] Mini PCI Express & mSATA

Despite the mini-PCI Express form factor, a mini-PCI Express slot must have support for the electrical connections an mSATA drive requires. For this reason, only certain notebooks are compatible with mSATA drives. Most compatible systems are based on Intel\'s newest Sandy Bridge processor architecture, using the new Huron River platform.

Notebooks like Lenovo\'s newest T-Series, W-Series, and X-Series ThinkPads released in March-April of 2011 have support for an mSATA SSD card in their WWAN card slot. The ThinkPad Edge E220s/E420s, and the Lenovo IdeaPad Y460/Y560 also support mSATA.[9]

Some notebooks (notably the Asus Eee PC, the MacBook Air, and the Dell mini9 and mini10) use a variant of the PCI Express Mini Card as an SSD. This variant uses the reserved and several non-reserved pins to implement SATA and IDE interface passthrough, keeping only USB, ground lines, and sometimes the core PCIe 1x bus intact.[10] This makes the \'miniPCIe\' flash and solid state drives sold for netbooks largely incompatible with true PCI Express Mini implementations.

Also, the typical Asus miniPCIe SSD is 71mm long, causing the Dell 51mm model to often be (incorrectly) referred to as half length. A true 51mm Mini PCIe SSD was announced in 2009, with two stacked PCB layers, which allows for higher storage capacity. The announced design preserves the PCIe interface, making it compatible with the standard mini PCIe slot. No working product has yet been developed, likely as a result of the popularity of the alternative variant.

[edit] PCI Express External Cabling

PCI Express External Cabling (also known as External PCI Express, Cabled PCI Express, or ePCIe) specifications were released by the PCI-SIG in February 2007.[11][12]

Standard cables and connectors have been defined for x1, x4, x8, and x16 link widths, with a transfer rate of 250 MB/s per lane. The PCI-SIG also expects the norm will evolve to reach the 500 MB/s, as in PCI Express 2.0. The maximum cable length remains undetermined. An example of the uses of Cabled PCI Express is a metal enclosure, containing a number of PCI slots and PCI-to-ePCIe adapter circuitry. This device would not be possible had it not been for the ePCIe spec.

[edit] Derivative forms

There are several other expansion card types derived from PCIe. These include:

  • Low height card
  • ExpressCard: successor to the PC card form factor (with ×1 PCIe and USB 2.0; hot-pluggable)
  • PCI Express ExpressModule: a hot-pluggable modular form factor defined for servers and workstations
  • XMC: similar to the CMC/PMC form factor (with ×4 PCIe or Serial RapidI/O)
  • AdvancedTCA: a complement to CompactPCI for larger applications; supports serial based backplane topologies
  • AMC: a complement to the AdvancedTCA specification; supports processor and I/O modules on ATCA boards (×1, ×2, ×4 or ×8 PCIe).
  • FeaturePak: a tiny expansion card format (43 x 65 mm) for embedded and small form factor applications; it implements two x1 PCIe links on a high-density connector along with USB, I2C, and up to 100 points of I/O.
  • Universal IO: A variant from Super Micro Computer Inc designed for use in low profile rack mounted chassis. It has the connector bracket reversed so it cannot fit in a normal PCI Express socket, but is pin compatible and may be inserted if the bracket is removed.
  • Thunderbolt: A variant from Intel that combines DisplayPort and PCIe protocols in a form factor compatible with Mini DisplayPort.

[edit] History

While in early development, PCIe was initially referred to as HSI (for High Speed Interconnect), and underwent a name change to 3GIO (for 3rd Generation I/O) before finally settling on its PCI-SIG name PCI Express. It was first drawn up by a technical working group named the Arapaho Work Group (AWG) that, for initial drafts, consisted only of Intel engineers. Subsequently the AWG expanded to include industry partners.

PCIe is a technology under constant development and improvement. The current PCI Express implementation is version 3.0.

[edit] PCI Express 1.0a

In 2003, PCI-SIG introduced PCIe 1.0a, with a data rate of 250 MB/s and a transfer rate of 2.5 GT/s.

[edit] PCI Express 1.1

In 2005, PCI-SIG introduced PCIe 1.1. This updated specification includes clarifications and several improvements, but is fully compatible with PCI Express 1.0a. No changes were made to the data rate.

[edit] PCI Express 2.0

PCI-SIG announced the availability of the PCI Express Base 2.0 specification on 15 January 2007.[13] The PCIe 2.0 standard doubles the per-lane throughput from the PCIe 1.0 standard\'s 250 MB/s to 500 MB/s. This means a 32-lane PCI connector (x32) can support throughput up to 16 GB/s aggregate. The PCIe 2.0 standard uses a base clock speed of 2.5 GHz, while the first version operates at 1.25 GHz.

PCIe 2.0 motherboard slots are fully backward compatible with PCIe v1.x cards. PCIe 2.0 cards are also generally backward compatible with PCIe 1.x motherboards, using the available bandwidth of PCI Express 1.1. Overall, graphic cards or motherboards designed for v2.0 will work with the other being v1.1 or v1.0.

The PCI-SIG also said that PCIe 2.0 features improvements to the point-to-point data transfer protocol and its software architecture.[14]

Intel\'s first PCIe 2.0 capable chipset was the X38 and boards began to ship from various vendors (Abit, Asus, Gigabyte) as of October 21, 2007.[15] AMD started supporting PCIe 2.0 with its AMD 700 chipset series and nVidia started with the MCP72.[16] All of Intel\'s prior chipsets, including the Intel P35 chipset, supported PCIe 1.1 or 1.0a.[17]

[edit] PCI Express 2.1

PCI Express 2.1 supports a large proportion of the management, support, and troubleshooting systems planned for full implementation in PCI Express 3.0. However, the speed is the same as PCI Express 2.0. Most motherboards sold currently come with PCI Express 2.1 connectors.

[edit] PCI Express 3.0

PCI Express 3.0 Base specification revision 3.0 was made available in November 2010, after multiple delays. In August 2007, PCI-SIG announced that PCI Express 3.0 would carry a bit rate of 8 gigatransfers per second, and that it would be backwards compatible with existing PCIe implementations. At that time, it was also announced that the final specification for PCI Express 3.0 would be delayed until 2011.[18] New features for the PCIe 3.0 specification include a number of optimizations for enhanced signaling and data integrity, including transmitter and receiver equalization, PLL improvements, clock data recovery, and channel enhancements for currently supported topologies.[19]

Following a six-month technical analysis of the feasibility of scaling the PCIe interconnect bandwidth, PCI-SIG\'s analysis found out that 8 gigatransfers per second can be manufactured in mainstream silicon process technology, and can be deployed with existing low-cost materials and infrastructure, while maintaining full compatibility (with negligible impact) to the PCIe protocol stack.

PCIe 2.0 delivers 5 GT/s, but uses an 8b/10b encoding scheme that results in a 20 percent ((10-8)/10) overhead on the raw bit rate. PCIe 3.0 removes the requirement for 8b/10b encoding, and instead uses a technique called "scrambling" that applies a known binary polynomial to a data stream in a feedback topology. Because the scrambling polynomial is known, the data can be recovered by running it through a feedback topology using the inverse polynomial.[20] and also uses a 128b/130b encoding scheme, reducing the overhead to approximately 1.5% ((130-128)/130), as opposed to the 20% overhead of 8b/10b encoding used by PCIe 2.0. PCIe 3.0\'s 8 GT/s bit rate effectively delivers double PCIe 2.0 bandwidth. PCI-SIG expects the PCIe 3.0 specifications to undergo rigorous technical vetting and validation before being released to the industry. This process, which was followed in the development of prior generations of the PCIe Base and various form factor specifications, includes the corroboration of the final electrical parameters with data derived from test silicon and other simulations conducted by multiple members of the PCI-SIG.

On November 18, 2010, the PCI Special Interest Group officially published the finalized PCI Express 3.0 specification to its members to build devices based on this new version of PCI Express.[21]

[edit] Current status

PCI Express has replaced AGP as the default interface for graphics cards on new systems. With a few exceptions, all graphics cards being released as of 2009 and 2010 from AMD (ATI) and NVIDIA use PCI Express. NVIDIA uses the high bandwidth data transfer of PCIe for its Scalable Link Interface (SLI) technology, which allows multiple graphics cards of the same chipset and model number to run in tandem, allowing increased performance. ATI has also developed a multi-GPU system based on PCIe called CrossFire. AMD and NVIDIA have released motherboard chipsets that support up to four PCIe ×16 slots, allowing tri-GPU and quad-GPU card configurations.

PCI Express has displaced a major portion of the add-in card market. PCI Express was originally only common in disk array controllers, onboard gigabit Ethernet, Wi-Fi and graphics cards. Most sound cards, TV/capture-cards, modems, serial port/USB/Firewire cards, network/WiFi cards that would have used the conventional PCI in the past have moved to PCI Express x8, x4, or x1. While some motherboards have conventional PCI slots, these are primarily for legacy cards and are being phased out.

[edit] Hardware protocol summary

The PCIe link is built around dedicated unidirectional couples of serial (1-bit), point-to-point connections known as lanes. This is in sharp contrast to the earlier PCI connection, which is a bus-based system where all the devices share the same bidirectional, 32-bit or 64-bit parallel bus.

PCI Express is a layered protocol, consisting of a transaction layer, a data link layer, and a physical layer. The Data Link Layer is subdivided to include a media access control (MAC) sublayer. The Physical Layer is subdivided into logical and electrical sublayers. The Physical logical-sublayer contains a physical coding sublayer (PCS). The terms are borrowed from the IEEE 802 networking protocol model.

[edit] Physical layer

The PCIe Physical Layer (PHY, PCIEPHY, PCI Express PHY, or PCIe PHY) specification is divided into two sub-layers, corresponding to electrical and logical specifications. The logical sublayer is sometimes further divided into a MAC sublayer and a PCS, although this division is not formally part of the PCIe specification. A specification published by Intel, the PHY Interface for PCI Express (PIPE),[22] defines the MAC/PCS functional partitioning and the interface between these two sub-layers. The PIPE specification also identifies the physical media attachment (PMA) layer, which includes the serializer/deserializer (SerDes) and other analog circuitry; however, since SerDes implementations vary greatly among ASIC vendors, PIPE does not specify an interface between the PCS and PMA.

At the electrical level, each lane consists of two unidirectional LVDS or PCML pairs at 2.525 Gbit/s. Transmit and receive are separate differential pairs, for a total of 4 data wires per lane.

A connection between any two PCIe devices is known as a link, and is built up from a collection of 1 or more lanes. All devices must minimally support single-lane (x1) link. Devices may optionally support wider links composed of 2, 4, 8, 12, 16, or 32 lanes. This allows for very good compatibility in two ways:

  • A PCIe card physically fits (and works correctly) in any slot that is at least as large as it is (e.g., an ×1 sized card will work in any sized slot);
  • A slot of a large physical size (e.g., ×16) can be wired electrically with fewer lanes (e.g., ×1, ×4, ×8, or ×12) as long as it provides the ground connections required by the larger physical slot size.

In both cases, PCIe negotiates the highest mutually supported number of lanes. Many graphics cards, motherboards and bios versions are verified to support ×1, ×4, ×8 and ×16 connectivity on the same connection.

Even though the two would be signal-compatible, it is not usually possible to place a physically larger PCIe card (e.g., a ×16 sized card) into a smaller slot —though if the PCIe slots are open-ended, by design or by hack, some motherboards will allow this.[citation needed]

The width of a PCIe connector is 8.8 mm, while the height is 11.25 mm, and the length is variable. The fixed section of the connector is 11.65 mm in length and contains 2 rows of 11 (22 pins total), while the length of the other section is variable depending on the number of lanes. The pins are spaced at 1 mm intervals, and the thickness of the card going into the connector is 1.8 mm.[23][24]

LanesPinsLength
TotalVariableTotalVariable
×1 2×18 = 36[25] 2×7 = 14 25 mm 7.65 mm
×4 2×32 = 64 2×21 = 42 39 mm 21.65 mm
×8 2×49 = 98 2×38 = 76 56 mm 38.65 mm
×16 2×82 = 164 2×71 = 142 89 mm 71.65 mm

[edit] Data transmission

PCIe sends all control messages, including interrupts, over the same links used for data. The serial protocol can never be blocked, so latency is still comparable to conventional PCI, which has dedicated interrupt lines.

Data transmitted on multiple-lane links is interleaved, meaning that each successive byte is sent down successive lanes. The PCIe specification refers to this interleaving as data striping. While requiring significant hardware complexity to synchronize (or deskew) the incoming striped data, striping can significantly reduce the latency of the nth byte on a link. Due to padding requirements, striping may not necessarily reduce the latency of small data packets on a link.

As with other high data rate serial transmission protocols, clocking information is embedded in the signal. At the physical level, PCI Express 2.0 utilizes the 8b/10b encoding scheme[20] to ensure that strings of consecutive ones or consecutive zeros are limited in length. This was used to prevent the receiver from losing track of where the bit edges are. In this coding scheme every 8 (uncoded) payload bits of data are replaced with 10 (encoded) bits of transmit data, causing a 20% overhead in the electrical bandwidth. To improve the available bandwidth, PCI Express version 3.0 employs 128b/130b encoding instead: similar but with much lower overhead.

Many other protocols (such as SONET) use a different form of encoding known as scrambling to embed clock information into data streams. The PCIe specification also defines a scrambling algorithm, but it is used to reduce electromagnetic interference (EMI) by preventing repeating data patterns in the transmitted data stream.

[edit] Data link layer

The Data Link Layer performs three vital services for the PCIe express link: (1) sequence the transaction layer packets (TLPs) that are generated by the transaction layer, (2) ensure reliable delivery of TLPs between two endpoints via an acknowledgement protocol (ACK and NAK signaling) that explicitly requires replay of unacknowledged/bad TLPs, (3) initialize and manage flow control credits

On the transmit side, the data link layer generates an incrementing sequence number for each outgoing TLP. It serves as a unique identification tag for each transmitted TLP, and is inserted into the header of the outgoing TLP. A 32-bit cyclic redundancy check code (known in this context as Link CRC or LCRC) is also appended to the end of each outgoing TLP.

On the receive side, the received TLP\'s LCRC and sequence number are both validated in the link layer. If either the LCRC check fails (indicating a data error), or the sequence-number is out of range (non-consecutive from the last valid received TLP), then the bad TLP, as well as any TLPs received after the bad TLP, are considered invalid and discarded. The receiver sends a negative acknowledgement message (NAK) with the sequence-number of the invalid TLP, requesting re-transmission of all TLPs forward of that sequence-number. If the received TLP passes the LCRC check and has the correct sequence number, it is treated as valid. The link receiver increments the sequence-number (which tracks the last received good TLP), and forwards the valid TLP to the receiver\'s transaction layer. An ACK message is sent to remote transmitter, indicating the TLP was successfully received (and by extension, all TLPs with past sequence-numbers.)

If the transmitter receives a NAK message, or no acknowledgement (NAK or ACK) is received until a timeout period expires, the transmitter must retransmit all TLPs that lack a positive acknowledgement (ACK). Barring a persistent malfunction of the device or transmission medium, the link-layer presents a reliable connection to the transaction layer, since the transmission protocol ensures delivery of TLPs over an unreliable medium.

In addition to sending and receiving TLPs generated by the transaction layer, the data-link layer also generates and consumes DLLPs, data link layer packets. ACK and NAK signals are communicated via (DLLP), as are flow control credit information, some power management messages and flow control credit information (on behalf of the transaction layer.)

In practice, the number of in-flight, unacknowledged TLPs on the link is limited by two factors: the size of the transmitter\'s replay buffer (which must store a copy of all transmitted TLPs until they the remote receiver ACKs them), and the flow control credits issued by the receiver to a transmitter. PCI Express requires all receivers to issue a minimum number of credits, to guarantee a link allows sending PCIConfig TLPs and message TLPs.

[edit] Transaction layer

PCI Express implements split transactions (transactions with request and response separated by time), allowing the link to carry other traffic while the target device gathers data for the response.

PCI Express uses credit-based flow control. In this scheme, a device advertises an initial amount of credit for each received buffer in its transaction layer. The device at the opposite end of the link, when sending transactions to this device, counts the number of credits each TLP consumes from its account. The sending device may only transmit a TLP when doing so does not make its consumed credit count exceed its credit limit. When the receiving device finishes processing the TLP from its buffer, it signals a return of credits to the sending device, which increases the credit limit by the restored amount. The credit counters are modular counters, and the comparison of consumed credits to credit limit requires modular arithmetic. The advantage of this scheme (compared to other methods such as wait states or handshake-based transfer protocols) is that the latency of credit return does not affect performance, provided that the credit limit is not encountered. This assumption is generally met if each device is designed with adequate buffer sizes.

PCIe 1.x is often quoted to support a data rate of 250 MB/s in each direction, per lane. This figure is a calculation from the physical signaling rate (2.5 Gbaud) divided by the encoding overhead (10 bits per byte.) This means a sixteen lane (x16) PCIe card would then be theoretically capable of 16×250 MB/s = 4 GB/s in each direction. While this is correct in terms of data bytes, more meaningful calculations are based on the usable data payload rate, which depends on the profile of the traffic, which is a function of the high-level (software) application and intermediate protocol levels.

Like other high data rate serial interconnect systems, PCIe has a protocol and processing overhead due to the additional transfer robustness (CRC and acknowledgements). Long continuous unidirectional transfers (such as those typical in high-performance storage controllers) can approach >95% of PCIe\'s raw (lane) data rate. These transfers also benefit the most from increased number of lanes (×2, ×4, etc.) But in more typical applications (such as a USB or Ethernet controller), the traffic profile is characterized as short data packets with frequent enforced acknowledgements.[26] This type of traffic reduces the efficiency of the link, due to overhead from packet parsing and forced interrupts (either in the device\'s host interface or the PC\'s CPU.) Being a protocol for devices connected to the same printed circuit board, it does not require the same tolerance for transmission errors as a protocol for communication over longer distances, and thus, this loss of efficiency is not particular to PCIe.

[edit] Uses

[edit] External PCIe cards

Theoretically, external PCIe could give a notebook the graphics power of a desktop, by connecting a notebook with any PCIe desktop video card (enclosed in its own external housing, with strong power supply and cooling); This is possible with an ExpressCard interface, which provides single lane v1.1 performance.

[27][28][29][30][31]

IBM/Lenovo has also included a PCI-Express slot in their Advanced Docking Station 250310U. It provides a half sized slot with an x16 length slot, but only x1 connectivity.[32] However, docking stations with expansion slots are becoming less common as the laptops are getting more advanced video cards and either DVI-D interfaces, or DVI-D pass through for port replicators and docking stations.

Additionally, Nvidia has developed Quadro Plex external PCIe Video Cards that can be used for advanced graphic applications. These video cards require a PCI Express x8 or x16 slot for the interconnection cable.[33] In 2008, AMD announced the ATI XGP technology, based on a proprietary cabling solution that is compatible with PCIe x8 signal transmissions.[34] This connector is available on the Fujitsu Amilo and the Acer Ferrari One notebooks. Only Fujitsu has an actual external box available, which also works on the Ferrari One. Recently Acer launched the Dynavivid graphics dock for XGP. Shuttle introduced their own external graphics solutions, GXT.

There are now card hubs in development that one can connect to a laptop through an ExpressCard slot, though they are currently rare, obscure, or unavailable on the open market. These hubs can have full-sized cards placed in them.

Magma and ViDock also makes use of ExpressCard and implements the usage of External graphic solutions .ViDock are expansion chassis tailored specifically for adapting PCI Express graphics cards for use with ExpressCard equipped laptop PCs. This enables user to make use of connecting PCIe cards externally. Although, the developments in these technologies are still ongoing. Other examples that underwent are - MSI GUS, Asus XG Station.

Recently, Intel and Apple introduced Thunderbolt, which allows for external PCI(e) devices.

[edit] Competing protocols

Several communications standards have emerged based on high bandwidth serial architectures. These include InfiniBand, RapidIO, HyperTransport, QPI and StarFabric. The differences are based on the tradeoffs between flexibility and extensibility vs latency and overhead. An example of such a tradeoff is adding complex header information to a transmitted packet to allow for complex routing (PCI Express is not capable of this). The additional overhead reduces the effective bandwidth of the interface and complicates bus discovery and initialization software. Also making the system hot-pluggable requires that software track network topology changes. Examples of buses suited for this purpose are InfiniBand and StarFabric.

Another example is making the packets shorter to decrease latency (as is required if a bus must operate as a memory interface). Smaller packets mean packet headers consume a higher percentage of the packet, thus decreasing the effective bandwidth. Examples of bus protocols designed for this purpose are RapidIO and HyperTransport.

PCI Express falls somewhere in the middle, targeted by design as a system interconnect (local bus) rather than a device interconnect or routed network protocol. Additionally, its design goal of software transparency constrains the protocol and raises its latency somewhat.

[edit] Development tools

When developing and/or troubleshooting the PCI Express bus, examination of hardware signals can be very important to find the problems. Logic analyzers and bus analyzers are tools that collect, analyze, decode, store signals so people can view the high-speed waveforms at their leisure.

[edit] See also

[edit] References

  1. ^ "PCI Express 2.0 (Training)". MindShare. Retrieved 2009-12-07.
  2. ^ "PCI Express Base specification". PCI_SIG. Retrieved 2010-10-18.
  3. ^ "HowStuffWorks "How PCI Express Works"". Computer.howstuffworks.com. Retrieved 2009-12-07.
  4. ^ a b c "PCI Express Architecture Frequently Asked Questions". PCI-SIG. Retrieved 23 November 2008.
  5. ^ "PCI Express Bus". Retrieved 2010-06-12.
  6. ^ "PCI Express – An Overview of the PCI Express Standard - Developer Zone - National Instruments". Zone.ni.com. 2009-08-13. Retrieved 2009-12-07.
  7. ^ "What is the A side, B side configuration of PCI cards". Frequently Asked Questions. Adex Electronics. 1998. Retrieved 2011 Oct 24.
  8. ^ PCI-SIG: Board Design Guidelines for PCI Express Architecture 2004 p. 19
  9. ^ http://forum.notebookreview.com/lenovo-ibm/574993-msata-faq-basic-primer.html
  10. ^ "Eee PC Research". Retrieved 26 October 2009.
  11. ^ "PCI Express External Cabling 1.0 Specification". Retrieved 9 February 2007.
  12. ^ "February 7, 2007". Pci-Sig. 2007-02-07. Retrieved 2010-09-11.
  13. ^ "PCI Express Base 2.0 specification announced" (PDF) (Press release). PCI-SIG. 15 January 2007. Retrieved 9 February 2007. — note that in this press release the term aggregate bandwidth refers to the sum of incoming and outgoing bandwidth; using this terminology the aggregate bandwidth of full duplex 100BASE-TX is 200 Mbit/s
  14. ^ Tony Smith (11 October 2006). "PCI Express 2.0 final draft spec published". The Register. Retrieved 9 February 2007.
  15. ^ Gary Key & Wesley Fink (21 May 2007). "Intel P35: Intel\'s Mainstream Chipset Grows Up". AnandTech. Retrieved 21 May 2007.
  16. ^ Anh Huynh (8 February 2007). "NVIDIA "MCP72" Details Unveiled". AnandTech. Retrieved 9 February 2007.
  17. ^ "Intel P35 Express Chipset Product Brief" (PDF). Intel. Retrieved 5 September 2007.
  18. ^ Hachman, Mark. "PC Magazine". Pcmag.com. Retrieved 2010-09-11.
  19. ^ "PCI Express 3.0 Bandwidth: 8.0 Gigatransfers/s". ExtremeTech. 9 August 2007. Retrieved 5 September 2007.
  20. ^ a b "PCI Express 3.0 Frequently Asked Questions". PCI-SIG. Retrieved 23 November 2010.
  21. ^ "PCI Special Interest Group Publishes PCI Express 3.0 Standard.". 18 November 2010. Retrieved 18 November 2010.
  22. ^ "PHY Interface for the PCI Express Architecture, version 2.00" (PDF). Retrieved 21 May 2008.
  23. ^ "Mechanical Drawing for PCI Express Connector". Retrieved 7 December 2007.
  24. ^ "FCi schematic for PCIe connectors". Retrieved 7 December 2007.[dead link]
  25. ^ "PCI Express 1x, 4x, 8x, 16x bus pinout and wiring @". Pinouts.ru. Retrieved 2009-12-07.
  26. ^ "Computer Peripherals And Interfaces". Technical Publications Pune. Retrieved 23 July 2009.
  27. ^ "Magma ExpressBox1: Cabled PCI Express for Desktops and Laptops". Magma.com. Retrieved 2010-09-11.
  28. ^ "TheInquirer". TheInquirer. Retrieved 2010-09-11.
  29. ^ "Custompcmag.co.uk". Custompcmag.co.uk. Retrieved 2010-09-11.
  30. ^ ASUSTeK Computer
  31. ^ "Technology Beats. News and Reviews". VR-Zone. 1995-09-09. Retrieved 2010-09-11.
  32. ^ "IBM/Lenovo Thinkpad Advanced Dock Overview". 307.ibm.com. 2010-03-07. Retrieved 2010-09-11.
  33. ^ "NVIDIA Quadro Plex VCS – Advanced visualization and remote graphics". Nvidia.com. Retrieved 2010-09-11.
  34. ^ "Advanced Micro Devices, AMD – Global Provider of Innovative Microprocessor, Graphics and Media Solutions". Ati.amd.com. Retrieved 2010-09-11.

[edit] Further reading

  • PCI Express System Architecture; 1st Ed; Ravi Budruk / Don Anderson / Tom Shanley; 1120 pages; 2003; ISBN 9780321156303.
  • Introduction to PCI Express : A Hardware and Software Developer\'s Guide; 1st Ed; 325 pages; 2003; ISBN 9780970284693.
  • Complete PCI Express Reference : Design Implications for Hardware and Software Developers; 1st Ed; 1056 pages; 2003; ISBN 9780971786196.