|Publication number||US20070165596 A1|
|Application number||US 11/334,678|
|Publication date||Jul 19, 2007|
|Filing date||Jan 18, 2006|
|Priority date||Jan 18, 2006|
|Also published as||CN101013989A, CN101013989B, US7907604, US20080235430|
|Publication number||11334678, 334678, US 2007/0165596 A1, US 2007/165596 A1, US 20070165596 A1, US 20070165596A1, US 2007165596 A1, US 2007165596A1, US-A1-20070165596, US-A1-2007165596, US2007/0165596A1, US2007/165596A1, US20070165596 A1, US20070165596A1, US2007165596 A1, US2007165596A1|
|Inventors||William Boyd, Douglas Freimuth, William Holland, Steven Hunter, Renato Recio, Steven Thurber, Madeline Vega|
|Original Assignee||Boyd William T, Freimuth Douglas M, Holland William G, Hunter Steven W, Recio Renato J, Thurber Steven M, Madeline Vega|
|Export Citation||BiBTeX, EndNote, RefMan|
|Referenced by (14), Classifications (9), Legal Events (2)|
|External Links: USPTO, USPTO Assignment, Espacenet|
1. Field of the Invention
The invention disclosed and claimed herein generally pertains to a method and related apparatus for routing PCIe transaction packets between multiple hosts and adapters, through a PCIe switched-fabric. More particularly, the invention pertains to a method for creating and managing the structures needed for routing PCI transaction packets between multiple hosts and adapters when using a Destination Identification (DID) that is integrated into the PBA.
2. Description of the Related Art
As is well known by those of skill in the art, PCI Express (PCIe) is widely used in computer systems to interconnect host units to adapters or other components, by means of a PCI switched-fabric bus or the like. However, PCIe currently does not permit the sharing of input/output (I/O) adapters in topologies where there are multiple hosts with multiple shared PCIe links. As a result, even though such sharing capability could be very valuable when using blade clusters or other clustered servers, adapters for PCIe and secondary networks (e.g., FC, IB, Enet) are at present generally placed only into individual blades and server systems. Thus, such adapters cannot be shared between clustered blades, or even between multiple roots within a clustered system.
In an environment containing multiple blades or blade clusters, it can be very costly to dedicate a PCI adapter for use with only a single blade. For example, a 10 Gigabit Ethernet (10 GigE) adapter currently costs on the order of $6,000. The inability to share these expensive adapters between blades has, in fact, contributed to the slow adoption rate of certain new network technologies such as 10 GigE. Moreover, there is a constraint imposed by the limited space available in blades to accommodate I/O adapters. This problem of limited space could be overcome if a PC network was able to support attachment of multiple hosts to a single PCI adapter, so that virtual PCIe I/O adapters could be shared between the multiple hosts.
In order to allow virtualization of PCIe adapters in the above environment, a mechanism is required for creating and managing the structures needed for routing PCI transaction packets between multiple hosts and adapters. The mechanism must be designed so that it protects memory and data in the system image of one host from being accessed by unauthorized applications in system images of other hosts. Access by other adapters in the same PCI tree must also be prevented. Moreover, implementation of the mechanism should minimize changes that must be made to currently used PCI hardware.
The invention is generally directed to the provision and management of tables for routing packets through an environment that includes multiple hosts and shared PCIe switches and adapters. The invention features modification of a conventional PCI Bus Address (PBA) by including a Destination Identification (DID) field in the PBA. Thus, the DID field is embedded in a transaction packet dispatched through the PCIe switches, and is integrated into the PCI address. A particular DID is associated with a particular host or system image, and thus identifies the physical or virtual end point of its packet. One useful embodiment of the invention is directed to a method for creating and managing the structures needed for routing PCIe transaction packets through PCIe switches in a distributed computer system comprising multiple root nodes, wherein each root node includes one or more hosts. The system further includes one or more PCI adapters. A physical tree that is indicative of a physical configuration of the distributed computing system is determined, and a virtual tree is created from the physical tree. The virtual tree is then modified to change an association between at least one source device and at least one target device in the virtual tree. A validation mechanism validates the changed association between the at least one source device and the at least one target device to enable routing of data from the at least one source device to the at least one target device.
The RCs 110, 120, and 130 are integral components of RN 160, 162 and 164, respectively. There may be more than one RC in an RN, such as RCs 140 and 142 which are both integral components of RN 166. In addition to the RCs, each RN consists of one or more Central Processing Units (CPUs) 102-104, 112-114, 122-124 and 132-134, memories 106, 116, 126 and 136, and memory controllers 108, 118, 128 and 138. The memory controllers respectively interconnect the CPUS, memory, and I/O RCs of their corresponding RNs, and perform such functions as handling the coherency traffic for respective memories.
RN's may be connected together at their memory controllers, such as by a link 146 extending between memory controllers 108 and 118 of RNs 160 and 162. This forms one coherency domain which may act as a single Symmetric Multi-Processing (SMP) system. Alternatively, nodes may be independent from one another with separate coherency domains as in RNs 164 and 166.
Distributed computing system 100 may be implemented using various commercially available computer systems. For example, distributed computing system 100 may be implemented using an IBM eServer iSeries Model 840 system available from International Business Machines Corporation. Such a system may support logical partitioning using an OS/400 operating system, which is also available from International Business Machines Corporation.
Those of ordinary skill in the art will appreciate that the hardware depicted in
With reference to
Partitioned hardware 230 includes a plurality of processors 232-238, a plurality of system memory units 240-246, a plurality of input/output (I/O) adapters 248-262, and a storage unit 270. Partition hardware 230 also includes service processor 290, which may be used to provide various services, such as processing of errors in the partitions. Each of the processors 232-238, memory units 240-246, NVRAM 298, and I/O adapters 248-262 may be assigned to one of multiple partitions within logically partitioned platform 200, each of which corresponds to one of operating systems 202, 204, 206 and 208.
Partition management firmware (hypervisor) 210 performs a number of functions and services for partitions 212, 214, 216 and 218 to create and enforce the partitioning of logically partitioned platform 200. Hypervisor 210 is a firmware implemented virtual machine identical to the underlying hardware. Hypervisor software is available from International Business Machines Corporation. Firmware is “software” stored in a memory chip that holds its content without electrical power, such as, for example, read-only memory (ROM), programmable ROM (PROM), electrically erasable programmable ROM (EEPROM), and non-volatile random access memory (NVRAM). Thus, hypervisor 210 allows the simultaneous execution of independent OS images 202, 204, 206 and 208 by virtualizing all the hardware resources of logically partitioned platform 200.
Operation of the different partitions may be controlled through a hardware management console, such as hardware management console 280. Hardware management console 280 is a separate distributed computing system from which a system administrator may perform various functions including reallocation of resources to different partitions.
In an environment of the type shown in
Accordingly, some functionality is needed in the bridges that connect IOAs to the I/O bus so as to be able to assign resources, such as individual IOAs or parts of IOAs to separate partitions; and, at the same time, prevent the assigned resources from affecting other partitions such as by obtaining access to resources of the other partitions.
Referring further to
Each of the host CPU sets has an associated root complex as described above, through which the system images of respective hosts interface with or access the I/O fabric 144. More particularly, host sets 332-336 are interconnected to RCs 338-342, respectively. Root complex 338 has ports 344 and 346, and root complexes 340 and 342 each has only a single port, i.e. ports 348 and 350, respectively. Each of the host CPU sets, together with its corresponding root complex, comprises an example or instance of a root node, such as RNs 160-166 shown in
Respective ports of a multi-root aware switch, such as switches 302 and 304, can be used as upstream ports, downstream ports, or both upstream and downstream ports. Generally, upstream ports are closer to a source of data and receive a data stream. Downstream ports are further from the data source and send out a data stream. Upstream/downstream ports can have characteristics of both upstream and downstream ports. In
The ports configured as downstream ports are to be attached or connected to adapters or to the upstream port of another switch. In
Each of the ports configured as an upstream port is used to connect to one of the root complexes 338-342. Thus,
The ports configured as upstream/downstream ports are used to connect to the upstream/downstream port of another switch. Thus,
I/O adapter 352 is shown as a virtualized I/O adapter, having its function 0 (F0) assigned and accessible to the system image SI 1, and its function 1 (F1) assigned and accessible to the system image SI 2. Similarly, I/O adapter 358 is shown as a virtualized I/O adapter, having its function 0 (F0) assigned and assessible to SI 3, its function 1 (F1) assigned and accessible to SI 4 and its function 3 (F3) assigned to SI 5. I/O adapter 366 is shown as a virtualized I/O adapter with its function F0 assigned and accessible to SI 2 and its function F1 assigned and accessible to SI 4. I/O adapter 368 is shown as a single function I/O adapter assigned and accessible to SI 5.
In a system such as distributed computer system 300, the PCM must query a PCI switch, to determine whether or not the switch supports use of integrated DID for routing packets. In system 300, switches 302 and 304 support integrated DID as described herein, but switch 306 does not.
More specifically, it is essential to understand that in connection with the IDIRT, the higher order bits in the PCI address space (selected to be the highest 16 bits in this embodiment) are used to identify a destination. Thus, a switch receiving a PCIe Packet uses the high order bits, for example the upper 16 bits, of the address to select the port that routes to the correct destination. The remaining 48 bits of the address base will then be addresses that are used by that destination.
When a particular host connects to a switch that supports integrated DID, the PCM configures the switch so that one of the PBA address spaces of the IDIRT is assigned to the particular host. The PCM carries this out by creating an entry in the IDIRT for each connected host. Thus, an entry could be made that, as an example, assigns address space 402 of
As stated above, when a PBA address space is assigned to a host, the highest 16 bits of the address space are thereafter used as a destination identifier or DID that is associated with the host. For example, the bits x0000 of space 402 could be the assigned DID to root complex 338. The switch would then report to the host that the lower 48 bits of the address space 402 are available for use with packets pertaining to root complex 338. Each root complex, such as root complexes 338, 340, and 342, is identified by the destination identifier and can use host virtualization to route incoming PCIe transactions to the appropriate host SI. In this arrangement, when an virtual end point, such as 354, initiates a PCIe memory transaction the adapter places the integrated DID in the upper 16 bits of the PCIe memory transaction's address field. The switches then use the IDIRT to route PCIe transaction to the root complex associated with the integrated DID.
When an adapter is connected to a switch capable of supporting integrated DID, the switch reports this event to the PCM. The PCM then places an entry in the switch IDIRT for each virtual end point and communicates to each root complex the set of virtual end points that are associated to that root complex, along with the integrated DID for each of those virtual end points. As a result of this action, the virtual end points adapter are “made visible” to each of the associated hosts, and can be accessed thereby. For example, the bits x0001 of space 408 could be the assigned DID to virtual end point 354. Each virtual end point, such as virtual end points 354, 356, 360, 362, 364, 350, 351, and 352, is identified by the destination identifier and can use host virtualization to route incoming PCIe transactions to the appropriate virtual end point. In this arrangement, when a root complex, such as 338, initiates a PCIe memory transaction the root complex places the integrated DID in the upper 16 bits of the PCIe memory transaction's address field. The switches then use the IDIRT to route PCIe transaction to the virtual end point associated with the integrated DID.
The PCM can query the IDIRT of a switch to determine what is in the switch configuration. Also, the PCM can modify entries in a switch IDIRT or can destroy or delete entries therein when those entries are no longer valid. Embodiments of the invention thus combine or aggregate multiple devices with a single DID number, to simplify routing lookup. Moreover, each host can only communicate to PCI addresses within its PCI address space segment. This is enforced at the switch containing the IDIRT, which is also referred to herein as a root switch. All PCIe component trees below a root switch are joined at the switch to form a single tree.
The Integrated DID number 542 of the packet is used by the switch to look up an entry in the IDIRT 500 that contains the switch port number to emit the packet out of. For example, if the Integrated DID number 542 points to IDIRT entry 1 548, then Port A 556 on the switch is used to emit the packet.
Before an outbound PCIe packet can be emitted from a port, the switch checks if the port can accept PCIe packets from the BDF# contained in the inbound PCIe packet 540. The switch performs this function by using the Integrated DID 542 to look up an entry in the Integrated DID-to-BDF# Validation Table (IDIVT) 570 and comparing the BDF# 544 from the incoming packet 540 to the list of BDFs 590 in the IDIVT 570. IDID numbers 584 and 588 respectively correspond to BDF numbers 595 and 598.
The present invention is directed to a method and system for managing the routing of data in a distributed computing system, for example, a distributed computing system that uses PCI Express protocol to communicate over an I/O fabric, to reflect modifications made to the distributed computing system. In particular, the present invention provides a mechanism for managing the Integrated Destination ID field included in the above-described data routing mechanism to ensure that the routing mechanism properly reflects modifications made in the distributed computing system that affects the routing of data through the system such as transferring IOAs from one host to another, or adding or removing hosts and/or IOAs from the system.
As shown in diagram 702, the PCI Configuration Manager (PCM) first creates an Integrated DID Routing Table (IDIDRT) representing a tree indicative of the current physical configuration of the distributed computing system. The PCM creates this table by discovering the current configuration of the I/O fabric so that it will have a full view of the physical configuration of the fabric, and then creates the IDIDRT from this information. The manner in which this may be accomplished is described in detail in commonly assigned, copending U.S. patent application entitled ______, Ser. No. ______, Attorney Docket No. AUS920050367US1, filed on ______, the disclosure of which is hereby incorporated by reference. In the physical tree shown in diagram 702, it is assumed that End Point 1 (EP 1) and EP 3 be assigned to RC 1, and that EP 2 be assigned to RC 2. The PCM then creates a virtual tree from the physical tree to be presented to an administrator or agent for RC 1 as shown in diagram 704. It will be noted that this configuration is the same as the physical configuration shown in diagram 702, but is now virtual.
The system administrator or agent for RC 1 then modifies the virtual tree by deleting EP 2 so that it cannot communicate with RC 1 as shown in diagram 706. The PCM then creates a new IDID Validation Table (IDIDVT) to reflect the modification of the virtual tree.
The procedure illustrated in diagrams 704 and 706 is then repeated for RC 2. In particular, the PCM presents a virtual tree to the system administrator or agent for RC 2, and the system administrator or agent modifies the virtual tree by deleting EP 1 and EP 3 so that they cannot communicate with RC 2 as shown in diagram 708.
When the above-described process has been completed for all RCs in the physical tree, the IDIDVT in the switch will be as shown in diagram 710 wherein the IDIDVT validates RC 1 to communicate with EP 1 and EP 3 and vice versa, and validates RC 2 to communicate with EP 2 and vice versa. It should be understood that although only two RCs and three EPs are included in the physical tree in
If the switch is multi-root aware (Yes output of Step 902), the PCM begins at Port AP (AP=Active Port) of the switch, and starts with Bus#=0 (Step 906). The PCM then queries the PCIe Configuration Space of the component attached to port AP (Step 908). A determination is made whether the component is a switch (Step 910). If the component is a switch (Yes output of Step 910), a determination is made whether a Bus# has been assigned to port AP (Step 912). If a Bus# has been assigned to port AP (Yes output of Step 912), port AP is set equal to port AP−1 (Step 914), and the method returns to Step 908 to repeat the method with the next port.
If a Bus# has not been assigned to port AP (No output of Step 912), a Bus# of AP=BN is assigned on current; BN=BN+1 (Step 916), and Bus#s are assigned to the I/O fabric below the switch by re-entering this method for the switch below the switch (Step 918). Port AP is then set equal to port AP−1 (Step 914), and the method returns to Step 908 to repeat the method with the next port.
If the component is determined not to be a switch (No output to Step 910), a determination is made whether the component is an RC (Step 920). If the component is an RC (Yes output of Step 920), a BDF# is assigned (Step 922) and a determination is made whether the RC supports the IDID (Step 924). If the RC does support the IDID (Yes output of Step 924), the IDID is assigned to the RC (Step 926). The AP is then set to be equal to AP−1 (Step 928), and a determination is made whether the AP is greater than 0 (Step 930). If the AP is not greater than 0 (No output of Step 930), the method ends. If the AP is greater than 0 (Yes output of Step 930), the method returns to Step 908 to query the PCIe configuration Space of the component attached to the next port.
If the RC does not support IDID (No output of Step 924), the AP is set=AP−1 (Step 928), and the process continues as described above.
Meanwhile, if the component is determined not to be an RC (No output of Step 920), a BDF# is assigned (Step 932), and a determination is made whether the EP supports IDID (Step 934). If the EP supports IDID (Yes output of Step 934), the IDID is assigned to each Virtual EP (Step 936). The AP is set=AP−1 (Step 928), and the process continues from there as described above.
If the EP does not support IDID (No output of Step 934), the AP is set=AP−1 (Step 928), and the process continues as described above.
Returning back to
A IDIDVT is then created on each switch showing the RC IDID# associated with the list of EP BDFs, and EP IDID# associated with the list of EP BDF#s (Step 816). The RCN is then made equal to RCN−1 (Step 818), and a determination is made whether RCN=0 (Step 820). If the RCN=0 (Yes output of Step 820), the method ends. If RCN does not equal 0 (No output of Step 820), the method returns to Step 810, and a virtual tree is created by copying the next physical tree and repeating the subsequent steps for the next virtual tree.
The present invention thus provides a method and system for managing the routing of data in a distributed computing system, such as a distributed computing system that uses PCI Express protocol to communicate over an I/O fabric. A physical tree that is indicative of a physical configuration of the distributed computing system is determined, and a virtual tree is created from the physical tree. The virtual tree is then modified to change an association between at least one source device and at least one target device in the virtual tree. A validation mechanism validates the changed association between the at least one source device and the at least one target device to enable routing of data from the at least one source device to the at least one target device.
The invention can take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment containing both hardware and software elements. In a preferred embodiment, the invention is implemented in software, which includes but is not limited to firmware, resident software, microcode, etc.
Furthermore, the invention can take the form of a computer program product accessible from a computer-usable or computer-readable medium providing program code for use by or in connection with a computer or any instruction execution system. For the purposes of this description, a computer-usable or computer readable medium can be any apparatus that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.
The medium can be an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system (or apparatus or device) or a propagation medium. Examples of a computer-readable medium include a semiconductor or solid state memory, magnetic tape, a removable computer diskette, a random access memory (RAM), a read-only memory (ROM), a rigid magnetic disk and an optical disk. Current examples of optical disks include compact disk—read only memory (CD-ROM), compact disk—read/write (CD-R/W) and DVD.
A data processing system suitable for storing and/or executing program code will include at least one processor coupled directly or indirectly to memory elements through a system bus. The memory elements can include local memory employed during actual execution of the program code, bulk storage, and cache memories which provide temporary storage of at least some program code in order to reduce the number of times code must be retrieved from bulk storage during execution.
Input/output or I/O devices (including but not limited to keyboards, displays, pointing devices, etc.) can be coupled to the system either directly or through intervening I/O controllers.
Network adapters may also be coupled to the system to enable the data processing system to become coupled to other data processing systems or remote printers or storage devices through intervening private or public networks. Modems, cable modem and Ethernet cards are just a few of the currently available types of network adapters.
The description of the present invention has been presented for purposes of illustration and description, and is not intended to be exhaustive or limited to the invention in the form disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiment was chosen and described in order to best explain the principles of the invention, the practical application, and to enable others of ordinary skill in the art to understand the invention for various embodiments with various modifications as are suited to the particular use contemplated.
|Citing Patent||Filing date||Publication date||Applicant||Title|
|US7506094||Jun 9, 2008||Mar 17, 2009||International Business Machines Corporation||Method using a master node to control I/O fabric configuration in a multi-host environment|
|US7549003||Feb 18, 2008||Jun 16, 2009||International Business Machines Corporation||Creation and management of destination ID routing structures in multi-host PCI topologies|
|US7831759||May 1, 2008||Nov 9, 2010||International Business Machines Corporation||Method, apparatus, and computer program product for routing packets utilizing a unique identifier, included within a standard address, that identifies the destination host computer system|
|US7930598||Jan 19, 2009||Apr 19, 2011||International Business Machines Corporation||Broadcast of shared I/O fabric error messages in a multi-host environment to all affected root nodes|
|US7937518||Dec 22, 2008||May 3, 2011||International Business Machines Corporation||Method, apparatus, and computer usable program code for migrating virtual adapters from source physical adapters to destination physical adapters|
|US8341327||Aug 24, 2009||Dec 25, 2012||Hitachi, Ltd.||Computer system and method for sharing PCI devices thereof|
|US8352665||Jun 18, 2009||Jan 8, 2013||Hitachi, Ltd.||Computer system and bus assignment method|
|US8407391||May 28, 2010||Mar 26, 2013||Hitachi, Ltd.||Computer system managing I/O path and port|
|US8533381||Feb 19, 2010||Sep 10, 2013||Hitachi, Ltd.||Computer system, method of managing PCI switch, and management server|
|US8683109||Jan 3, 2013||Mar 25, 2014||Hitachi, Ltd.||Computer system and bus assignment method|
|US8725926||Nov 19, 2012||May 13, 2014||Hitachi, Ltd.||Computer system and method for sharing PCI devices thereof|
|US9110865||Jul 1, 2011||Aug 18, 2015||International Business Machines Corporation||Virtual machine dynamic routing|
|US20140310430 *||Apr 9, 2014||Oct 16, 2014||Marvell World Trade Ltd.||Tunneling Transaction Packets|
|EP2568389A2 *||Sep 7, 2012||Mar 13, 2013||Apple Inc.||Coherence switch for I/O traffic|
|U.S. Classification||370/351, 370/389|
|International Classification||H04L12/56, H04L12/28|
|Cooperative Classification||H04L45/54, G06F13/4022, H04L45/00|
|European Classification||H04L45/54, G06F13/40D2|
|Mar 8, 2006||AS||Assignment|
Owner name: MACHINES CORPORATION, INTERNATIONAL BUSINESS, NEW
Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:BOYD, WILLIAM T.;FREIMUTH, DOUGLAS M.;HOLLAND, WILLIAM G.;AND OTHERS;REEL/FRAME:017271/0260;SIGNING DATES FROM 20051122 TO 20051213
|Apr 22, 2013||AS||Assignment|
Free format text: CORRECTIVE ASSIGNMENT TO CORRECT THE ASSIGNEE NAME TO READ: INTERNATIONAL BUSINESS MACHINES CORPORATION PREVIOUSLY RECORDED ON REEL 017271 FRAME 0260. ASSIGNOR(S) HEREBY CONFIRMS THE ASSIGNMENT OF ASSIGNOR S INTEREST;ASSIGNORS:BOYD, WILLIAM T.;FREIMUTH, DOUGLAS M.;HOLLAND, WILLIAM G.;AND OTHERS;SIGNING DATES FROM 20051122 TO 20051213;REEL/FRAME:030259/0525
Owner name: INTERNATIONAL BUSINESS MACHINES CORPORATION, NEW Y