Patent Application
20050147117
1. Field of the Invention
This invention relates in general to the field of computer network architecture, and more specifically to an apparatus and method for initializing the polarity of a port in a shared I/O device.
2. Description of the Related Art
Modern computer architecture may be viewed as having three distinct subsystems which, when combined, form what most think of when they hear the term computer. These subsystems are: 1) a processing complex; 2) an interface between the processing complex and I/O (i.e., input/output) controllers or devices; and 3) the I/O controllers or devices themselves.
A processing complex may be as simple as a single processing core, such as a Pentium® microprocessor, or it might be as complex as two or more processing cores. These two or more processing cores may reside on separate devices or integrated circuits, or they may be part of the same integrated circuit. Within the scope of the present invention, a processing core is hardware, microcode (i.e., firmware), or a combination of hardware and microcode that is capable of executing instructions from a particular instruction set architecture (ISA) such as the x86 ISA. Multiple processing cores within a processing complex may execute instances of the same operating system (e.g., multiple instances of Unix), they may run independent operating systems (e.g., one executing Unix and another executing Windows XP®), or they may together execute instructions that are part of a single instance of a symmetrical multi-processing (SMP) operating system. Within a processing complex, multiple processing cores may access a shared memory or they may access independent memory devices.
The interface between the processing complex and I/O is commonly known as the chipset. The chipset interfaces to the processing complex via a bus referred to as the HOST bus. The “side” of the chipset that interfaces to the HOST bus is typically referred to as the “north side” or “north bridge.” The HOST bus is generally a proprietary bus designed to interface to memory, to one or more processing complexes, and to the chipset. On the other side (“south side”) of the chipset are buses which connect the chipset to I/O devices. Examples of such buses include ISA, EISA, PCI, PCI-X, and AGP.
I/O devices allow data to be transferred to or from a processing complex through the chipset on one or more of the busses supported by the chipset. Examples of I/O devices include graphics cards coupled to a computer display; disk controllers (which are coupled to hard disk drives or other data storage systems); network controllers (to interface to networks such as Ethernet); USB and Firewire controllers which interface to a variety of devices from digital cameras to external data storage to digital music systems, etc.; and PS/2 controllers for interfacing to keyboards/mice. I/O devices are designed to connect to the chipset via one of its supported interface buses. For instance, present day computers typically couple graphic cards to the chipset via an AGP bus. Ethernet cards, SATA, Fiber Channel, and SCSI (data storage) cards; and USB and Firewire controllers all connect to the chipset via a Peripheral Component Interconnect (PCI) bus. PS/2 devices are coupled to the chipset via an ISA bus.
The above description is general, yet one skilled in the art will appreciate from the above discussion that, regardless of the type of computer, its configuration will include a processing complex for executing instructions, an interface to I/O controllers, and I/O devices themselves that allow the processing complex to communicate with the outside world. This is true whether the computer is an inexpensive desktop in a home, a high-end workstation used for graphics and video editing, or a clustered server which provides database support or web services to hundreds within a large organization.
A problem that has been recognized by the present inventors is that the requirement to place a processing complex, I/O interface, and I/O devices within every computer is costly and lacks flexibility. That is, once a computer is purchased, all of its subsystems are static from the standpoint of the user. To change a processing complex while still utilizing the same I/O interface and I/O devices is an extremely difficult task. The I/O interface (e.g., the chipset) is typically so closely coupled to the architecture of the processing complex that swapping one without the other doesn't make sense. Furthermore, the I/O devices are typically integrated within the computer, at least for servers and business desktops, such that upgrade or modification of the computer's I/O capabilities ranges in difficulty from extremely cost prohibitive to virtually impossible.
An example of the above limitations is considered helpful. A popular network server produced by Dell Computer Corporation is the Dell PowerEdge 1750®. This server includes a processing core designed by Intel® (a Xeon® microprocessor) along with memory. It has a server-class chipset (i.e., I/O interface) for interfacing the processing complex to I/O controllers/devices. And, it has the following onboard I/O controllers/devices: onboard graphics for connecting to a display, onboard PS/2 for connecting a mouse/keyboard, onboard RAID controller for connecting to data storage, onboard network interface controllers for connecting to 10/100 and 1 gigabit (Gb) Ethernet; and a PCI bus for adding other I/O such as SCSI or Fiber Channel controllers. It is believed that none of the onboard features is upgradeable.
As noted above, one of the problems with a highly integrated architecture is that if another I/O demand emerges, it is difficult and costly to implement the upgrade. For example, 10 Gb Ethernet is on the horizon. How can 10 Gb Ethernet capabilities be easily added to this server? Well, perhaps a 10 Gb Ethernet controller could be purchased and inserted onto an existing PCI bus within the server. But consider a technology infrastructure that includes tens or hundreds of these servers. To move to a faster network architecture requires an upgrade to each of the existing servers. This is an extremely cost prohibitive scenario, which is why it is very difficult to upgrade existing network infrastructures.
The one-to-one correspondence between the processing complex, the interface to the I/O, and the I/O controllers/devices is also costly to the manufacturer. That is, in the example presented above, many of the I/O controllers/devices are manufactured on the motherboard of the server. To include the I/O controllers/devices on the motherboard is costly to the manufacturer, and ultimately to an end user. If the end user utilizes all of the I/O capabilities provided, then a cost-effective situation exists. But if the end user does not wish to utilize, say, the onboard RAID capabilities or the 10/100 Ethernet capabilities, the user is still required to pay for its inclusion. Such one-to-one correspondence is not a cost-effective solution.
Now consider another emerging platform: the blade server. A blade server is essentially a processing complex, an interface to I/O, and I/O controllers/devices that are integrated onto a relatively small printed circuit board that has a backplane connector. The “blade” is configured so that it can be inserted along with other blades into a chassis having a form factor similar to a present day rack server. The benefit of this configuration is that many blade servers can be provided within the same rack space previously required by just one or two rack servers. And while blades have seen growth in market segments where processing density is a real issue, they have yet to gain significant market share for many reasons, one of which is cost. This is because blade servers still must provide all of the features of a pedestal or rack server including a processing complex, an interface to I/O, and the I/O controllers/devices. Furthermore, blade servers must integrate all their I/O controllers/devices onboard because they do not have an external bus which would allow them to interface to other I/O controllers/devices. Consequently, a typical blade server must provide such I/O controllers/devices as Ethernet (e.g., 10/100 and/or 1 Gb) and data storage control (e.g., SCSI, Fiber Channel, etc.)—all onboard.
Infiniband™ is a recent development which was introduced by Intel and other vendors to allow multiple processing complexes to separate themselves from I/O controllers/devices. Infiniband is a high-speed point-to-point serial interconnect designed to provide for multiple, out-of-the-box interconnects. However, it is a switched, channel-based architecture that drastically departs from the load-store architecture of existing processing complexes. That is, Infiniband is based upon a message-passing protocol where a processing complex communicates with a Host-Channel-Adapter (HCA), which then communicates with all downstream Infiniband devices such as I/O devices. The HCA handles all the transport to the Infiniband fabric rather than the processing complex itself. Within an Infiniband architecture, the only device that remains within the load-store domain of the processing complex is the HCA. What this means is that it is necessary to leave the processing complex load-store domain to communicate with I/O controllers/devices. And this departure from the processing complex load-store domain is one of the limitations that contributed to Infiniband's demise as a solution to providing shared I/O. According to one industry analyst referring to Infiniband, “[i]t was over-billed, over-hyped to be the nirvana-for-everything-server, everything I/O, the solution to every problem you can imagine in the data center, . . . , but turned out to be more complex and expensive to deploy, . . . , because it required installing a new cabling system and significant investments in yet another switched high speed serial interconnect.”
Accordingly, the present inventors have recognized that separation of a processing complex, its I/O interface, and the I/O controllers/devices is desirable, yet this separation must not impact either existing operating systems, application software, or existing hardware or hardware infrastructures. By breaking apart the processing complex from its I/O controllers/devices, more cost effective and flexible solutions can be introduced.
In addition, the present inventors have recognized that such a solution must not be a channel-based architecture, performed outside of the box. Rather, the solution should employ a load-store architecture, where the processing complex sends data directly to or receives data directly from (i.e., in an architectural sense by executing loads or stores) an I/O device (e.g., a network controller or data storage controller). This allows the separation to be accomplished without disadvantageously affecting an existing network infrastructure or disrupting the operating system.
Therefore, what is needed is an apparatus and method which separate a processing complex and its interface to I/O from I/O controllers/devices.
In addition, what is needed is an apparatus and method that allow processing complexes and their I/O interfaces to be designed, manufactured, and sold, without requiring I/O controllers/devices to be provided therewith.
Also, what is needed is an apparatus and method that enable an I/O controller/device to be shared by multiple processing complexes.
Furthermore, what is needed is an I/O controller/device that can be shared by two or more processing complexes using a common load-store fabric.
Moreover, what is needed is an apparatus and method that allow multiple processing complexes to share one or more I/O controllers/devices through a common load-store fabric.
Additionally, what is needed is an apparatus and method that provide switching between multiple processing complexes and shared I/O controllers/devices.
Furthermore, what is needed is an apparatus and method that allow multiple processing complexes, each operating independently and executing an operating system independently (i.e., independent operating system domains) to interconnect to shared I/O controllers/devices in such a manner that it appears to each of the multiple processing complexes that the I/O controllers/devices are solely dedicated to a given processing complex from its perspective. That is, from the standpoint of one of the multiple processing complexes, it must appear that the I/O controllers/devices are not shared with any of the other processing complexes.
Moreover, what is needed is an apparatus and method that allow shared I/O controllers/devices to be utilized by different processing complexes without requiring modification to the processing complexes existing operating systems or other application software.
Furthermore, what is needed is an apparatus and method for initializing the polarity of a port in a shared I/O device.
The present invention, among other applications, is directed to solving the above-noted problems and addresses other problems, disadvantages, and limitations of the prior art.
In one embodiment, a switching apparatus for sharing input/output endpoints is provided. The switching apparatus includes a first plurality of I/O ports, a second I/O port, and a plurality of port initialization logic elements. The first plurality of I/O ports is coupled to a plurality of operating system domains through a load-store fabric. Each of the first plurality of I/O ports routes transactions between the plurality of operating system domains and the switching apparatus. The second I/O port is coupled to a first shared input/output endpoint. The first shared input/output endpoint is configured to request/complete the transactions for each of the plurality of operating system domains. One of the plurality of port initialization logic elements is coupled to the second I/O port and remaining ones of the plurality of port initialization logic elements are each coupled to a corresponding one of the first plurality of I/O ports. The plurality of port initialization logic elements is configured to initialize corresponding links between each of the plurality of operating system domains and the switching apparatus, and between the first shared input/output endpoint and the switching apparatus, to support the transactions, where each of the plurality of port initialization logic elements automatically configures a corresponding polarity for each of the first plurality of I/O ports and the second I/O port, and where the corresponding polarity is in a default polarity prior to being configured.
One aspect of the present invention contemplates a shared input/output (I/O) switching mechanism. The shared I/O switching mechanism has a first plurality of I/O ports, a second I/O port, core logic, and a plurality of port initialization logic elements. The first plurality of I/O ports is coupled to a plurality of operating system domains through a load-store fabric, each configured to route transactions between the plurality of operating system domains and the shared input/output (I/O) switching mechanism. The second I/O port is coupled to a first shared input/output endpoint, where the first shared input/output endpoint is configured to request/complete said transactions for each of the plurality of operating system domains. The core logic is coupled to the first plurality of I/O ports and the second I/O port. The core logic routes the transactions between the first plurality of I/O ports and the second I/O port, where the core logic maps said each of the plurality of operating system domains to a corresponding root complex. One of the plurality of port initialization logic elements is coupled to the second I/O port and remaining ones of the plurality of port initialization logic elements are each coupled to a corresponding one of the first plurality of I/O ports. The plurality of port initialization logic elements initializes corresponding links between each of the plurality of operating system domains and the switching mechanism, and between the first shared input/output endpoint and the switching mechanism, to support the transactions, where each of the plurality of port initialization logic elements automatically configures a corresponding polarity for each of the first plurality of I/O ports and the second I/O port, and where the corresponding polarity is in a default polarity prior to being configured.
Another aspect of the present invention comprehends a method for interconnecting independent operating system domains to a shared I/O endpoint within a load-store fabric. The method includes, via first ports, first communicating with each of the independent operating system domains according to a protocol that provides exclusively for a single operating system domain within the load-store fabric; via a second port, second communicating with the shared I/O endpoint according to a variant of the protocol to enable the shared I/O endpoint to associate a prescribed operation with a corresponding one of the independent operating system domains; and initializing links between each of the independent operating system domains and the first ports, and between the shared I/O endpoint and the second port. The initializing includes automatically configuring a corresponding polarity for each of the first ports and the second port, where the corresponding polarity is in a default polarity prior to the configuring.
A further aspect of the present invention is embodied as an apparatus in a shared input/output endpoint. The apparatus has a first I/O port and link training logic. The first I/O port is coupled to a link, and is configured to request/complete said transactions for each of a plurality of operating system domains. The link training logic is coupled to the first I/O port. The link training logic initializes the link in a manner that is transparent to the plurality of operating system domains.
Yet another aspect of the present invention comprehends an apparatus in a root complex, where the root complex corresponds to a plurality of operating system domains. The apparatus has a first I/O port that is coupled to a link, and that is configured to request/complete the transactions for each of the plurality of operating system domains. The apparatus also has link training logic that is coupled to the first I/O port. The link training logic initializes the link in a manner that is transparent to the plurality of operating system domains.
These and other objects, features, and advantages of the present invention will become better understood with regard to the following description, and accompanying drawings where:
The following description is presented to enable one of ordinary skill in the art to make and use the present invention as provided within the context of a particular application and its requirements. Various modifications to the preferred embodiment will, however, be apparent to one skilled in the art, and the general principles defined herein may be applied to other embodiments. Therefore, the present invention is not intended to be limited to the particular embodiments shown and described herein, but is to be accorded the widest scope consistent with the principles and novel features herein disclosed.
Referring to
Each of the servers 102, 104, 106 has a root complex 108. A root complex 108 typically is a chipset which provides the interface between a processing complex, memory, and downstream I/O controllers/devices (e.g., IDE, SATA, Infiniband, Ethernet, Fiber Channel, USB, Firewire, PS/2). However, in the context of the present invention, a root complex 108 may also support more than one processing complexes and/or memories as well as the other functions described above. Furthermore, a root complex 108 may be configured to support a single instance of an operating system executing on multiple processing complexes (e.g., a symmetrical multi-processing (SMP) operating system), multiple processing complexes executing multiple instances of the same operating system, independent operating systems executing on multiple processing complexes, or independent operating systems executing on multiple processing cores within a single processing complex. For example, devices (e.g., microprocessors) are now being contemplated which have multiple processing cores, each of which are independent of the other (i.e., each processing core has its own dedicated resources and executes its own operating system independent of other processing cores within the device). Within the context of the PCI Express architecture (which will be further discussed below), a root complex 108 is a component in a PCI Express hierarchy that connects to the HOST bus segment on the upstream side with one or more PCI Express links on the downstream side. In other words, a PCI Express root complex 108 denotes the device that connects a processing complex to the PCI Express fabric. A root complex 108 need not be provided as a stand-alone integrated circuit, but as logic that performs the root complex function and which can be integrated into a chipset, or into a processing complex itself. Alternatively, root complex logic may be provided according to the present invention which is partially integrated within a processing complex with remaining parts integrated within a chipset. The present invention envisions all of these configurations of a root complex 108. In addition, it is noted that although PCI Express is depicted in the present example of a load-store fabric for interconnecting a multi-server computing environment, one skilled in the art will appreciate that other load-store fabric architectures can be applied as well to include RapidlO, VME, HyperTransport, PCI, VME, etc.
The root complex 108 of each of the servers 102, 104, 106 is connected to three I/O controllers 110, 112, 114. For illustration purposes, the I/O controllers 110, 112, 114 are a presented as a Network Interface Controller (NIC) 110, a Fiber Channel Controller 112, and an Other Controller 114. The three controllers 110, 112, 114 allow the root complex 108 of each of the servers 102, 104, 106 to communicate with networks, and data storage systems such as the Ethernet network 128, the Fiber Channel network 130, and the Other network 132. One skilled in the art will appreciate that these networks 128, 130, 132 may reside within a physical location close in proximity to the servers 102, 104, 106, or they may extend to points anywhere in the world, subject to limitations of the network architecture.
To allow each of the servers 102, 104, 106 to connect to the networks 128, 130, 132, switches 122, 124, 126 are provided between the controllers 110, 112, 114 in each of the servers 102, 104, 106, and the networks 128, 130, 132, respectively. That is, an Ethernet switch 122 is connected to the Network Interface Controllers 110 in each of the servers 102, 104, 106, and to the Ethernet network 128. The Ethernet switch 122 allows data or instructions to be transmitted from any device on the Ethernet network 128 to any of the three servers 102, 104, 106, and vice versa. Thus, whatever the communication channel between the root complex 108 and the Network Interface controller 110 (e.g., ISA, EISA, PCI, PCI-X, PCI Express), the Network Interface controller 110 communicates with the Ethernet network 128 (and the switch 122) utilizing the Ethernet protocol. One skilled in the art will appreciate that the communication channel between the root complex 108 and the network interface controller 110 is within the load-store domain of the root complex 108.
Fiber Channel switch 124 is connected to the Fiber Channel controllers 112 in each of the servers 102, 104, 106, and to the Fiber Channel network 130. The Fiber Channel switch 124 allows data or instructions to be transmitted from any device on the Fiber Channel network 130 to any of the three servers 102, 104, 106, and vice versa.
An Other switch 126 is connected to the Other controllers 114 in each of the servers 102, 104, 106, and to the Other network 132. The Other switch 126 allows data or instructions to be transmitted from any device on the Other network 132 to any of the three servers 102, 104, 106, and vice versa. Examples of Other types of networks include Infiniband, SATA, Serial Attached SCSI, etc. While the above list is not exhaustive, the Other network 132 is illustrated herein to help the reader understand that what will ultimately be described below with respect to the present invention, should not be limited to Ethernet and Fiber Channel networks 128, 130, but rather, can easily be extended to networks that exist today, or that will be defined in the future. Further, the communication speeds of the networks 128, 130, 132 are not discussed because one skilled in the art will appreciate that the interface speed of any network may change over time while still utilizing a preexisting protocol.
To illustrate the operation of the environment 100, if the server 102 wishes to send data or instructions over the Ethernet network 128 to either of the servers 104, 106, or to another device (not shown) on the Ethernet network 128, the root complex 108 of the server 102 will utilize its Ethernet controller 110 within the server's load-store domain to send the data or instructions to the Ethernet switch 122 which will then pass the data or instructions to the other server(s) 104, 106 or to a router (not shown) to get to an external device. One skilled in the art will appreciate that any device connected to the Ethernet network 128 will have its own Network Interface controller 110 to allow its root complex to communicate with the Ethernet network 128.
The present inventors provide the above discussion with reference to
Referring now to
First, it should be appreciated that each of the servers 202, 204, 206 is shown with differing numbers of CPU's 240. Within the scope of the present application, a CPU 240 is equivalent to a processing complex as described above. Server 202 contains one CPU 240. Server 204 contains two CPU's 240. Server 206 contains four CPU's 240. Second, the form factor for each of the servers 202, 204, 206 is approximately the same width, but differing height, to allow servers 202, 204, 206 with different computing capacities and executing different operating systems to physically reside within the same rack or enclosure. Third, the switches 222, 224, 226 also have form factors that allow them to be co-located within the same rack or enclosure as the servers 202, 204, 206. One skilled in the art will appreciate that, as in
Now turning to
In the blade environment 201 of
In addition to showing the servers 250, 252, 254 in a blade form factor along with the switches 222, 224, 226 within a blade chassis 270, the present inventors note that each of the I/O controllers 210, 212, 214 requires logic 211, 213, 215 to interface to the root complex 208 itself and to the specific network media fabric. The logic 211, 213, 215 that provides for interface to the network media fabric is know as Media Access Control (MAC) logic 211, 213, 215. The MAC 211, 213, 215 for each of the I/O controllers 210, 212, 214 typically resides one layer above the physical layer and defines the absolute address of its controller 210, 212, 214 within the media fabric. Corresponding MAC logic is also required on every port of the switches 222, 224, 226 to allow proper routing of data and/or instructions (i.e., usually in packet form) from one port (or device) to another. Thus, within a blade server environment 201, an I/O controller 210, 212, 214 must be supplied on each blade server 250, 252, 254 for each network fabric with which it wishes to communicate. And each I/O controller 210, 212, 214 must include MAC logic 211, 213, 215 to interface the I/O controller 210, 212, 214 to its respective switch 222, 224, 226.
Turning now to
Attention is now directed to
The PCI architecture was developed in the early 1990's by Intel Corporation as a general I/O architecture to enable the transfer of data and instructions much faster than the ISA architecture of the time. PCI has gone thru several improvements since that time, with the latest development being PCI Express. In a nutshell, PCI Express is a replacement of the PCI and PCI-X bus specifications to provide platforms with much greater performance, while utilizing a much lower pin count (Note: PCI and PCI-X are parallel bus architectures; PCI Express is a serial architecture). A complete discussion of PCI Express is beyond the scope of this specification however, it is noted that a thorough background and description can be found in the following books, which are incorporated herein by reference for all intents and purposes: Introduction to PCI Express, A Hardware and Software Developer's Guide, by Adam Wilen, Justin Schade, Ron Thornburg; The Complete PCI Express Reference Design Insights for Hardware and Software Developers, by Edward Solari and Brad Congdon; and PCI Express System Architecture, by Ravi Budruk, Don Anderson, Tom Shanley; all of which are readily available through retail sources such as www.amazon.com. In addition, the PCI Express specification itself is managed and disseminated through the Special Interest Group (SIG) for PCI found at www.pcisig.com.
Referring now to
The server 302 may be either a standalone server, a rack mount server, or a blade server, as shown and discussed above with respect to FIGS. 2A-C, but which includes the PCI Express bus 320 for communication between the root complex 308 and all downstream I/O controllers 324, 326, 328. What should be appreciated at this point is that, even with the advent of PCI Express, a server 302 still requires dedicated I/O controllers 324, 326, 328 to provide the capabilities to interface to network fabrics such as Ethernet, Fiber Channel, etc. In a configuration where the root complex 308 is integrated into one or both of the CPU's 304, 306, the host bus 310 interface from the CPU 304, 306 to the root complex 308 therein may take some other form than that conventionally understood as a host bus 310.
Referring now to
In contrast to server configurations discussed above, and as will be further described below, none of the servers 404, 406, 408 has their own dedicated I/O controller. Rather, the downstream side of each of their respective root complexes 412 is coupled directly to the shared I/O switch 420, thus enabling each of the servers 404, 406, 408 to communicate with the shared I/O controllers 440, 442, 444 while still using the PCI Express load-store fabric for communication. As is more particularly shown, the shared I/O switch 420 includes one or more PCI Express links 422 on its upstream side, a switch core 424 for processing PCI Express data and instructions, and one or more PCI Express+ links 432 on its downstream side for connecting to downstream PCI Express devices 440, 442, 444, and even to additional shared I/O switches 420 for cascading of PCI Express+ links 432. In addition, the present invention envisions the employment of multi-function shared I/O devices. A multi-function shared I/O device according to the present invention comprises a plurality of shared I/O devices. For instance, a shared I/O device consisting of a shared Ethernet NIC and a shared I-SCSI device within the same shared I/O endpoint is but one example of a multi-function shared I/O device according to the present invention. Furthermore, each of the downstream shared I/O devices 440, 442, 444 includes a PCI Express+ interface 441 and Media Access Control (MAC) logic. What should be appreciated by one skilled in the art when comparing
Turning now to
The shared I/O switch 510 is coupled to a shared Ethernet controller 512, a shared Fiber Channel controller 514, and a shared Other controller 516 via PCI Express+ links 511 according to the present invention. The shared Ethernet controller 512 is coupled to an Ethernet fabric 520. The shared Fiber Channel controller 514 is coupled to a Fiber Channel fabric 522. The shared Other controller 516 is coupled to an Other fabric 524. In operation, any of the root complexes 502, 504, 506 may communicate with any of the fabrics 520, 522, 524 via the shared I/O switch 510 and the shared I/O controllers 512, 514, 516. Specifics of how this is accomplished will now be described with reference to
Referring to
In the exemplary environment embodiment 600, each of the ports 603, 605, 607 are coupled to one of 16 ports 640 within a shared I/O switch 610 according to the present invention. In one embodiment, the switch 610 provides 16 ports 640 that support shared I/O transactions via the PCI Express fabric, although other port configurations are contemplated. One skilled in the art will appreciate that these ports 640 may be of different speeds (e.g., 2.5 Gb/sec) and may support multiple PCI Express or PCI Express+ lanes per link 608, 611 (e.g., x1, x2, x4, x8, x12, x16). For example, port 4603 of root complex 1602 may be coupled to port 4 of I/O switch 610, port 7605 of root complex 2604 may be coupled to port 11640 of I/O switch 610, and port 10607 of root complex 3606 may be coupled to port 16640 of switch 610.
On the downstream side of the switch 610, port 9640 of may be coupled to a port (not shown) on a shared I/O controller 650, such as the shared Ethernet controller 650 shown, that supports transactions from one of N different operating system domains (OSDs) via corresponding root complexes 602, 604, 606. Illustrated within the shared I/O controller 650 are four OS resources 651 that are independently supported. That is, the shared I/O controller 650 is capable of transmitting, receiving, isolating, segregating, and processing transactions from up to four distinct root complexes 602, 604, 606 that are associated with four operating system (OS) domains. An OS domain, within the present context, is a system load-store memory map that is associated with one or more processing complexes. Typically, present day operating systems such as Windows, Unix, Linux, VxWorks, etc., must comport with a specific load-store memory map that corresponds to the processing complex upon which they execute. For example, a typical x86 load-store memory map provides for both memory space and I/O space. Conventional memory is mapped to the lower 640 kilobytes (KB) of memory. The next higher 128 KB of memory are employed by legacy video devices. Above that is another 128 KB block of addresses mapped to expansion ROM. And the 128 KB block of addresses below the 1 megabyte (MB) boundary is mapped to boot ROM (i.e., BIOS). Both DRAM space and PCI memory are mapped above the 1 MB boundary. Accordingly, two separate processing complexes may be executing within two distinct OS domains, which typically means that the two processing complexes are executing either two instances of the same operating system or that they are executing two distinct operating systems. However, in a symmetrical multi-processing environment, a plurality of processing complexes may together be executing a single instance of an SMP operating system, in which case the plurality of processing complexes would be associated with a single OS domain. In one embodiment, the link 611 between the shared I/O switch 610 and the shared I/O controller 650 utilizes the PCI Express fabric, but enhances the fabric to allow for identification and segregation of OS domains, as will be further described below. The present inventors refer to the enhanced fabric as “PCI Express+.”
Referring now to
Referring now to
Turning now to
In one embodiment, the packet structure 900 of PCI Express, shown in
Referring now to
In an alternative embodiment, the OS domain number is used to associate a downstream or upstream port with a PCI Express+ packet. That is, where a packet must traverse multiple links between its origination and destination, a different OSD may be employed for routing of a given packet between a port pair on a given link than is employed for routing of the packet between an port pair on another link. Although different OS domain numbers are employed within the packet when traversing multiple links, such an aspect of the present invention still provides for uniquely identifying the packet so that it remains associated with its intended OS domain.
Additionally, within the OS domain header 1100, are a number of reserved (R) bits. It is conceived by the present inventors that the reserved bits have many uses. Accordingly, one embodiment of the present invention employs one or more of the reserved bits to track coherency of messages within a load-store fabric. Other uses of the reserved bits are contemplated as well. For example, one embodiment envisions use of the reserved (R) bits to encode a version number for the PCI Express+ protocol that is associated with one or more corresponding transactions.
In an exemplary embodiment, a two level table lookup is provided. More specifically, an OS domain number is associated with a PCI Express bus hierarchy. The PCI bus hierarchy is then associated with a particular upstream or downstream port. In this embodiment, normal PCI Express discovery and addressing mechanisms are used to communicate with downstream shared I/O switches and/or shared I/O devices. Accordingly, sharing logic within a shared I/O switch 610 (or shared I/O aware root complex or processing complex) maps particular PCI bus hierarchies to particular shared I/O endpoints 650 to keep multiple OS domains from seeing more shared I/O endpoints 650 than have been configured for them by the shared I/O switch 610. All variations which associate a transaction packet with an OS domain are contemplated by the present invention.
In a PCI Express embodiment, the OS domain header 1100 may be the only additional information included within a PCI Express packet 900 to form a PCI Express+ packet 1000. Alternatively, the present invention contemplates other embodiments for associating transactions with a given OS domain. For instance, a “designation” packet may be transmitted to a shared I/O device that associates a specified number of following packets with the given OS domain.
In another embodiment, the contents of the OS domain header 1100 are first established by the shared I/O switch 610 by encapsulating the port number of the shared I/O switch 610 that is coupled to the upstream root complex 602, 604, 606 from which a packet originated, or for which a packet is intended, as the OSD. But other means of associating packets with their origin/destination OS domain are contemplated. One alternative is for each root complex 602, 604, 606 that is coupled to the shared I/O switch 610 to be assigned a unique ID by the shared I/O switch 610 to be used as the OSD. Another alternative is for a root complex 602, 604, 606 to be assigned a unique ID, either by the shared I/O switch 610, or by any other mechanism within or external to the root complex 602, 604, 606, which is then used in packet transfer to the shared I/O switch (or downstream shared I/O controllers).
Turning now to
Turning to
More specifically, the bus interface+ 1304 includes: an interface 1350 to a shared I/O fabric such as PCI Express+; PCI Target logic 1352 such as a table which associates an OS domain with a particular one of N number of operating system domain resources supported by the shared I/O controller 1300; and PCI configuration logic 1354 which, in one embodiment, controls the association of the resources within the shared I/O controller 1300 with particular OS domains. The PCI configuration logic 1354 enables the shared Ethernet Controller 1300 to be enumerated by each supported. OSD. This allows each upstream OS domain that is mapped to the shared I/O controller 1300 to view it as an I/O controller having resources that are dedicated to its OS domain. And, from the viewpoint of the OS domain, no changes to the OS domain application software (e.g., operating system, driver for the controller, etc.) are required because the OS domain communicates transactions directed to the shared I/O controller using its existing load-store protocol (e.g., PCI Express). When these transactions reach a shared I/O aware device, such as a shared I/O aware root complex or shared I/O switch, then encapsulation/decapsulation of the above-described OS domain header is accomplished within the transaction packets to enable association of the transactions with assigned resources within the shared I/O controller 1300. Hence, sharing of the shared I/O controller 1300 between multiple OS domains is essentially transparent to each of the OS domains.
The control register logic+ 1308 includes a number of control register sets 1320-1328, each of which may be independently associated with a distinct OS domain. For example, if the shared I/O controller 1300 supports just three OS domains, then it might have control register sets 1320, 1322, 1324 where each control register set 1320, 1322, 1324 is associated with one of the three OS domains. Thus, transaction packets associated with a first OS domain would be associated with control register set 1320, transaction packets associated with a second OS domain would be associated with control register set 1322, and transaction packets associated with a third OS domain would be associated with control register set 1324. In addition, one skilled in the art will appreciate that while some control registers within a control register set (such as 1320) need to be duplicated within the shared I/O controller 1300 to allow multiple OS domains to share the controller 1300, not all control registers require duplication. That is, some control registers must be duplicated for each OS domain, others can be aliased, while others may be made accessible to each OS domain. What is illustrated in
The transmit logic+ 1312 includes a number of transmit logic elements 1360-1368, each of which may be independently associated with a distinct OS domain for transmission of packets and which are allocated in a substantially similar manner as that described above regarding allocation of the control register sets 1320-1328. In addition, the receive logic+ 1314 includes a number of receive logic elements 1370-1378, each of which may be independently associated with a distinct OS domain for reception of packets and which are allocated in a substantially similar manner as that described above regarding allocation of the control register sets 1320-1328. Although the embodiment of the shared Ethernet Controller 1300 depicts replicated transmit logic elements 1360-1360 and replicated receive logic elements 1370-1378, one skilled in the art will appreciate that there is no requirement to replicate these elements 1360-1368, 1370-1378 in order to embody a shared Ethernet controller 1300 according to the present invention. It is only necessary to provide transmit logic+ 1312 and receive logic+ 1314 that are capable of transmitting and receiving packets according to the present invention in a manner that provides for identification, isolation, segregation, and routing of transactions according to each supported OS domain. Accordingly, one embodiment of the present invention contemplates transmit logic+ 1312 and receive logic+ 1314 that does not comprise replicated transmit or receive logic elements 1360-1368, 1370-1378, but that does provide for the transmission and reception of packets as noted above.
The DMA logic+ 1316 includes N DMA engines 1330, 1332, 1334; N Descriptors 1336, 1338, 1340; and arbitration logic 1342 to arbitrate utilization of the DMA engines 1330-1334. That is, within the context of a shared I/O controller 1300 supporting multiple OS domains, depending on the number of OS domains supported by the shared I/O controller 1300, performance is improved by providing multiple DMA engines 1330-1334, any of which may be utilized at any time by the controller 1300, for any particular packet transfer. Thus, there need not be a direct correspondence between the number of OS domains supported by the shared I/O controller 1300 and the number of DMA engines 1330-1334 provided, or vice versa. Rather, a shared I/O controller manufacturer may support four OS domains with just one DMA engine 1330, or alternatively may support three OS domains with two DMA engines 1330, 1332, depending on the price/performance mix that is desired.
Further, the arbitration logic 1342 may use an algorithm as simple as round-robin, or alternatively may weight processes differently, either utilizing the type of transaction as the weighting factor, or may employ the OS domain associated with the process as the weighting factor. Other arbitration algorithms may be used without departing from the scope of the present invention.
As is noted above, what is illustrated in
Referring now to
More specifically, a bus interface+ 1452 is shown within the controller 1450 which includes a table for associating an OS domain with a resource 1451. In one embodiment, an OSD Header provided by the shared I/O switch 1410 is associated with one of the four resources 1451, where each resource 1451 includes a machine address (MAC). By associating one of N resources 1451 with an OS domain, transaction packets are examined by the bus interface+ 1452 and are assigned to their resource 1451 based on the OSD Header within the transaction packets. Packets that have been processed by the shared I/O Ethernet controller 1450 are transmitted upstream over a PCI Express+ link 1411 by placing its associated OS domain header within the PCI Express+ transaction packet before transmitting it to the shared I/O switch 1410.
In one embodiment, when the multi-OS Ethernet controller 1450 initializes itself with the shared I/O switch 1410, it indicates to the shared I/O switch 1410 that it has resources to support four OS domains (including four MAC addresses). The shared I/O switch 1410 is then aware that it will be binding the three OS domains associated with root complexes 1402, 1404, 1406 to the shared I/O controller 1450, and therefore assigns three OS domain numbers (of the 256 available to it), one associated with each of the root complexes 1402-1406, to each of the OS resources 1451 within the I/O controller 1450. The multi-OS Ethernet controller 1450 receives the “mapping” of OS domain number to MAC address and places the mapping in its table 1452. Then, when transmitting packets to the switch 1410, the shared I/O controller 1450 places the OS domain number corresponding to the packet in the OS domain header of its PCI Express+ packet. Upon receipt, the shared I/O switch 1410 examines the OS domain header to determine a PCI bus hierarchy corresponding to the value of the OS domain header. The shared I/O switch 1410 uses an internal table (not shown) which associates a PCI bus hierarchy with an upstream port 1440 to pass the packet to the appropriate root complex 1402-1406. Alternatively, the specific OSD numbers that are employed within the table 1452 are predetermined according to the maximum number of OS domains that are supported by the multi-OS Ethernet controller 1450. For instance, if the multi-OS Ethernet controller 1450 supports four OS domains, then OSD numbers 0-3 are employed within the table 1452. The shared I/O controller 1450 then associates a unique MAC address to each OSD number within the table 1452.
In an alternative embodiment, the multi-OS Ethernet controller 1450 provides OS domain numbers to the shared I/O switch 1410 for each OS domain that it can support (e.g., 1, 2, 3, or 4 in this illustration). The shared I/O switch 1410 then associates these OS domain numbers with its port that is coupled to the multi-OS controller 1450. When the shared I/O switch 1410 sends/receives packets through this port, it then associates each upstream OS domain that is mapped to the multi-OS controller 1450 to the OS domain numbers provided by the multi-OS controller 1450 according to the PCI bus hierarchy for the packets. In one embodiment, the OS domain numbers provided by the multi-OS controller 1450 index a table (not shown) in the shared I/O switch 1410 which associates the downstream OS domain number with the PCI bus hierarchy of a packet, and determines an upstream OS domain number from the PCI bus hierarchy. The upstream OS domain number is then used to identify the upstream port for transmission of the packet to the appropriate OS domain. One skilled in the art will appreciate that in this embodiment, the OS domain numbers used between the shared I/O switch 1410 and the shared I/O controller 1450 are local to that link 1411. The shared I/O switch 1410 uses the OS domain number on this link 1411 to associate packets with their upstream OS domains to determine the upstream port coupled to the appropriate OS domains. One mechanism for performing this association is a table lookup, but it should be appreciated that the present invention should not be limited association by table lookup.
While not specifically shown for clarity purposes, one skilled in the art will appreciate that for each port 1440 on the switch 1410, resources applicable to PCI bus hierarchies for each port 1440 (such as PCI-to-PCI bridges, buffering logic, etc.) should be presumed available for each port 1440, capable of supporting each of the OS domains on each port 1440. In one embodiment, dedicated resources are provided for each port 1440. In an alternative embodiment, virtual resources are provided for each port 1440 using shared resources within the shared I/O switch 1410. Thus, in a 16-port switch 1410, 16 sets of resources are provided. Or alternatively, one or more sets of resources are provided that are virtually available to each of the ports 1440. In addition, one skilled in the art will appreciate that one aspect of providing resources for each of the OS domains on each port 1440 includes the provision of link level flow control resources for each OS domain. This ensures that the flow of link level packets is independently controlled for each OS domain that is supported by a particular port 1440.
Referring now to
Flow begins at block 1502 and proceeds to decision block 1504.
At decision block 1504, a determination is made at the switch as to whether a request has been made from an OS domain. For clarity purposes, assume that the single OS domain is a associated with a root complex that is not shared I/O aware. That is, does an upstream port within the shared I/O switch contain a packet to be transmitted downstream? If not, flow returns to decision block 1504. Otherwise, flow proceeds to block 1506.
At block 1506, the downstream port for the packet is identified using information within the packet. Flow then proceeds to block 1508.
At block 1508, the shared I/O aware packet is built. If PCI Express is the load-store fabric which is upstream, a PCI Express+ packet is built which includes an OS Header which associates the packet with the OS domain of the packet (or at least with the upstream port associated with the packet). Flow then proceeds to block 1510.
At block 1510, the PCI Express+ packet is sent to the endpoint device, such as a shared I/O Ethernet controller. Flow then proceeds to block 1512.
At block 1512 a process for tracking the PCI Express+ packet is begun. That is, within a PCI Express load-store fabric, many packets require response tracking. This tracking is implemented in the shared I/O switch, for each OS domain for which the port is responsible. Flow then proceeds to block 1514 where packet transmission is completed (from the perspective of the shared I/O switch).
Referring now to
At decision block 1604 a determination is made as to whether a packet has been received on a port within the shared I/O switch that is associated with the shared I/O endpoint. If not, flow returns to decision block 1604. Otherwise, flow proceeds to block 1606.
At block 1606, the OS Header within the PCI Express+ packet is read to determine which OS domain is associated with the packet. Flow then proceeds to block 1608.
At block 1608, a PCI Express packet is built for transmission on the upstream, non-shared I/O aware, PCI Express link. Essentially, the OSD Header is removed (i.e., decapsulated) from the packet and the packet is sent to the port in the shared I/O switch that is associated with the packet (as identified in the OSD Header). Flow then proceeds to block 1610.
At block 1610, the packet is transmitted to the root complex associated with the OS domain designated by the packet. Flow then proceeds to block 1612.
At block 1612 a process is begun, if necessary, to track the upstream packet transmission as described above with reference to block 1512. Flow then proceeds to block 1614 where the flow is completed.
Referring to
At decision block 1704, a determination is made as to whether a packet has been received from the shared I/O switch. If the load-store fabric is PCI Express, then the received packet will be a PCI Express+ packet. If no packet has been received, flow returns to decision block 1704. Otherwise, flow proceeds to block 1706.
At block 1706, the OS domain (or upstream port associated with the packet) is determined. The determination is made using the OSD Header within the PCI Express+ packet. Flow then proceeds to block 1708.
At block 1708, the packet is processed utilizing resources allocated to the OS domain associated with the received packet, as described above with reference to
At block 1710, a process is begun, if necessary to track the packet. As described with reference to block 1512, some packets within the PCI Express architecture require tracking, and ports are tasked with handling the tracking. Within the shared I/O domain on PCI Express+, tracking is provided, per OS domain. Flow then proceeds to block 1712 where transmission is completed.
Referring now to
At decision block 1804, a determination is made as to whether a packet is ready to be transmitted to the shared I/O switch (or other upstream device). If not, flow returns to decision block 1804. Otherwise, flow proceeds to block 1806.
At block 1806, the OS domain (or upstream port) associated with the packet is determined. Flow then proceeds to block 1808.
At block 1808, a PCI Express+ packet is built which identifies the OS domain (or upstream port) associated with the packet. Flow then proceeds to block 1810.
At block 1810, the PCI Express+ packet is transmitted to the shared I/O switch (or other upstream device). Flow then proceeds to block 1812.
At block 1812, tracking for the packet is performed. Flow then proceeds to block 1814 where the transmission is completed.
In operation, three packets “A”, “B”, and “C” are transmitted by root complex 11902 to the shared I/O switch 1910 for downstream delivery. Packet “A” is to be transmitted to the Ethernet controller 1912, packet “B” is to be transmitted to the Fiber Channel controller 1914, and packet “C” is to be transmitted to the Other controller 1916. When the shared I/O switch 1910 receives these packets it identifies the targeted downstream shared I/O device (1912, 1914, or 1916) using information within the packets and performs a table lookup to determine the downstream port associated for transmission of the packets to the targeted downstream shared I/O device (1912, 1914, 1916). The shared I/O switch 1910 then builds PCI Express+ “A”, “B”, and “C” packets which includes encapsulated OSD Header information that associates the packets with root complex 11902 (or with the port (not shown) in the shared I/O switch 1910 that is coupled to root complex 11902). The shared I/O switch 1910 then routes each of the packets to the port coupled to their targeted downstream shared I/O device (1912, 1914, or 1916). Thus, packet “A” is placed on the port coupled to the Ethernet controller 1912, packet “B” is placed on the port coupled to the Fiber Channel controller 1914, and packet “C” is placed on the port coupled to the Other controller 1916. The packets are then transmitted to their respective controller (1912, 1914, or 1916).
From root complex 31906, a packet “G” is transmitted to the shared I/O switch 1910 for delivery to the shared Ethernet controller 1912. Upon receipt, the shared I/O switch 1910 builds a PCI Express+ packet for transmission to the shared Ethernet controller 1912 by encapsulating an OSD header within the PCI Express packet that associates the packet with root complex 31906 (or with the switch port coupled to root complex 31906). The shared I/O switch 1910 then transmits this packet to the shared Ethernet controller 1912.
The Ethernet controller 1912 has one packet “D” for transmission to root complex 21904. This packet is transmitted with an encapsulated OSD Header to the shared I/O switch 1910. The shared I/O switch 1910 receives the “D” packet, examines the OSD Header, and determines that the packet is destined for root complex 21904 (or the upstream port of the switch 1910 coupled to root complex 21904). The switch 1910 strips the OSD Header off (i.e., decapsulation of the OSD header) the “D” packet and transmits the “D” packet to root complex 21904 as a PCI Express packet.
The Fiber Channel controller 1914 has two packets for transmission. Packet “F” is destined for root complex 31906, and packet “E” is destined for root complex 11902. The shared I/O switch 1910 receives these packets over PCI Express+ link 1911. Upon receipt of each of these packets, the encapsulated OSD Header is examined to determine which upstream port is associated with each of the packets. The switch 1910 then builds PCI Express packets “F” and “E” for root complexes 31906, and 11902, respectively, and provides the packets to the ports coupled to root complexes 31906 and 11902 for transmission. The packets are then transmitted to those root complexes 1916, 1902.
The Other controller 1916 has a packet “G” destined for root complex 21904. Packet “G” is transmitted to the shared I/O switch 1910 as a PCI Express+ packet, containing encapsulated OSD header information associating the packet with root complex 21904 (or the upstream port in the shared I/O switch coupled to root complex 21904). The shared I/O switch 1910 decapsulates the OSD header from packet “G” and places the packet on the port coupled to root complex 21904 for transmission. Packet “G” is then transmitted to root complex 21904.
The above discussion of
Referring now to
Also shown in
Now turning to
The core logic 2106 includes transaction arbitration logic 2107 that communicates with the VMACs 2103 via the control bus 2104, and data movement logic 2108 that routes transaction data between the VMACs 2103 via the data bus 2105. The core logic 2106 also has management logic 2111 and global routing logic 2110 that is coupled to the transaction arbitration logic 2107. For purposes of teaching the present invention, an embodiment of the switch 2100 is described herein according to the PCI Express protocol, however, one skilled in the art will appreciate from the foregoing description that the novel concepts and techniques described can be applied to any single load-store domain architecture of which PCI Express is one example.
One of the primary functions of the switch 2100 according to the present invention, as has been alluded to above, is to enable multiple operating system domains (not shown) that are coupled to a plurality of the ports 2101, 2102 to conduct transactions with one or more shared I/O endpoints (not shown) that are coupled to other ports 2101, 2102 over a load-store fabric according to a protocol that provides for transactions exclusively for a single operating system domain. PCI Express is an example of such a load-store fabric and protocol. It is an objective of the switch 2100 according to the present invention to enable the multiple operating system domains to conduct transactions with the one or more shared I/O devices in a manner such that each of the multiple operating system domains only experiences its local load-store domain in terms of transactions with the one or more shared I/O endpoints, when in actuality the switch 2100 is providing for transparent and seamless routing of transactions between each of the multiple operating system domains and the one or more shared I/O endpoints, where transactions for each of the multiple operating system domains are isolated from transactions from the remaining operating system domains. As described above, the switch 2100 provides for 1) mapping of operating system domains to their associated transmit and receive ports 2102, 2101 within the switch 2100 and to particular ones of the one or more shared I/O endpoints, and 2) encapsulation and decapsulation of OSD headers that associate particular transactions with designated operating system domains.
In operation, each of the transmit and receive ports 2102, 2101 perform serializer/deserializer (SERDES) functions that are well known in the art. Deserialized transactions are presented by the receive ports 2101 to the VMACs 2103 over the x4 PCI Express receive buses 2112. Transactions for serialization by the transmit ports 2102 are provided by the VMACs 2103 over the x4 PCI Express transmit buses 2113. A x4 bus 2112, 2113 is capable of being trained to support transactions for up to a x4 PCI Express link, however, one skilled in the art will appreciate that a x4 PCI Express link can also train to a x2 or x1 speed.
Each VMAC 2103 provides PCI Express physical layer, data link layer functions that directly correspond to like layers in the PCI Express Base specification, with the exception of link initialization protocol as will be described in more detail below. Each VMAC 2103 can support operation of two independently configurable PCI Express x4 links, or two x4 links can be combined into a single x8 PCI Express link. In addition, each VMAC 2103 provides PCI Express transaction layer and presentation module functions. The transaction layer and presentation module functions are enhanced according to the present invention to enable identification and isolation of multiple operating system domains.
Upon initialization, the management logic 2111 configures tables within the global routing logic 2110 to map each combination of ingress port number, ingress operating system domain number (numbers are local to each port), and PCI Express traffic class to one or more egress port numbers along with egress operating system domain/virtual channel designations. During PCI Express discovery by each operating system domain, address ranges associated with each shared I/O device connected to the switch 2100 are also placed within the global routing logic 2110 to enable discrimination between egress ports and/or egress operating system domains/virtual channels when more than one shared I/O endpoint is coupled to the switch 2100. In addition, via the control bus 2104, the management logic 2111 configures local routing tables (not shown) within each of the VMACs 2103 with a mapping of operating system domain and traffic class to designated buffer resources for movement of transaction data. The management logic 2111 may comprise hard logic, programmable logic such as EEPROM, or an intelligent device such as a microcontroller or microprocessor that communicates with a management console or one of the operating system domains itself via a management link such as 12C for configuration of the switch 2100. Other forms of management logic are contemplated as well.
In the exemplary embodiment of the switch 2100, each VMAC 2103 can independently route transactions on each of two x4 PCI Express links for a combination of up to 16 operating system domains and virtual channels. For example, 16 independent operating system domains that utilize only one virtual channel each can be mapped. If six virtual channels are employed by one of the operating system domains, then ten remaining combinations of operating system domain/virtual channel are available for mapping. The present inventors note that a maximum number of 16 operating system domains/virtual channels is provided to clearly teach the exemplary embodiment of
The transaction arbitration logic 2107 is configured to ensure fairness of resources within the switch 2100 at two levels: arbitration of receive ports 2101 and arbitration of operating system domains/virtual channels. Fairness of resources is required to ensure that each receive port 2101 is allowed a fair share of a transmit port's bandwidth and that each operating system domain/virtual channel is allowed a fair share of transmit port bandwidth as well. With regard to receive port arbitration, the transaction arbitration logic 2107 employs a fairness sampling technique such as round-robin to ensure that no transmit port 2102 is starved and that bandwidth is balanced. With regard to arbitration of operating system domain/virtual channels, the transaction arbitration logic 2107 employs a second level of arbitration to pick which transaction will be selected as the next one to be transmitted on a given transmit port 2102.
The data movement logic 2108 interfaces to each VMAC 2103 via the data bus 2105 and provides memory resources for storage and movement of transaction data between ports 2101, 2102. A global memory pool, or buffer space is provided therein along with transaction ordering queues for each operating system domain. Transaction buffer space is allocated for each operating system domain from within the global memory pool. Such a configuration allows multiple operating system domains to share transaction buffer space, while still maintaining transaction order. The data movement logic 2108 also performs port arbitration at a final level by selecting which input port 2101 is actually allowed to transfer data to each output port 2102. The data movement logic 2108 also executes an arbitration technique such as round-robin to ensure that each input port 2101 is serviced when more than one input port 2101 has data to send to a given output port 2102.
When a transaction is received by a particular receive port 2101, its VMAC 2103 provides its data to the data movement logic 2108 via the data bus 2105 and routing information (e.g., ingress port number, operating system domain, traffic class, egress port number, and addressing/message ID information) to the transaction arbitration logic 2107 via the control bus 2104. The routing data is provided to the global routing logic 2110 which is configured as described above upon initialization and discovery from which an output port/operating system domain/virtual channel is provided. In accordance with the aforementioned arbitration schemes, the egress routing information and data is routed to an egress VMAC 2103, which then configures an egress transaction packet and transmits it over the designated transmit port 2102. In the case of a transaction packet that is destined for a shared I/O endpoint, the ingress VMAC 2103 performs encapsulation of the OSD header that designates an operating system domain which is associated with the particular transaction into the transaction layer packet. In the case of a transaction packet that is received from a shared I/O endpoint that is destined for a particular operating system domain, the ingress VMAC 2103 performs decapsulation of the OSD header that from within the received transaction layer packet and provides this OSD header along with the aforementioned routing information (e.g., port number, traffic class, address/message ID) to the global routing logic 2110 to determine egress port number and virtual channel.
Now turning to
In operation, the VMAC 2220 is capable of receiving and transmitting data across two transmit/receive port combinations (i.e., T1/R1 and T2/R2) concurrently, wherein each combination can be configured by the link training logic 2208 to operate as a x1, x2, or x4 PCI Express link. In addition, the link training logic 2208 can combine the two transmit/receive port combinations into a single x8 PCI Express link. The receive SERDES logic 2203 is configured to perform well known PCI Express functions to include 8-bit/10-bit decode, clock compensation, and lane polarity inversion. The receive physical layer logic 2204 is configured to perform PCI Express physical layer functions including symbol descrambling, multi-lane deskew, loopback, lane reversal, and symbol deframing. The receive data link layer logic 2205 is configured to execute PCI Express data link layer functions including data link control and management, sequence number checking and CRC checking and stripping. In addition, as alluded to above, the receive data link layer logic 2205 during link initialization performs operating system domain initialization functions and initiation of flow control for each supported operating system domain. The receive transaction layer logic 2206 is configured to execute PCI Express functions and additional functions according to the present invention including parsing of encapsulated OSD headers, generation of flow control for each operating system domain, control of receive buffers, and lookup of address information within the local mapping logic 2209. The receive presentation layer logic 2207 manages and orders transaction queues and received packets and interfaces to core logic via the control and data buses 2221, 2222.
On the transmit side, the transmit presentation layer logic 2214 receives packet data for transmission over the data bus 2222 provided from the data movement logic. The transmit transaction layer logic 2213 performs packet scheduling and retry buffer management. The transmit data link layer logic 2212 performs PCI Express functions including retry buffer management and packet scheduling. In addition, as alluded to above, the transmit data link layer logic 2212 during link initialization performs operating system domain initialization functions and initiation of flow control for each supported operating system domain. The transmit physical layer logic 2211 performs PCI Express functions including symbol framing, and symbol scrambling. The transmit SERDES logic 2210 executes PCI Express functions including 8-bit/10-bit encoding.
During link initialization, the link training logic 2208 directs the receive and transmit data link layer logic 2205, 2212 to operate as though they are coupled over their respective link to a shared I/O-aware device as described herein. Consequently, shared I/O initialization data link layer packets (DLLPs) are initiated for transmission by the transmit data link layer logic 2212 to the link partner. These shared I/O initialization DLLPs are configured, in one embodiment, by employing unused DLLP encodings from the PCI Express Base specification. One skilled in the art will appreciate that if a PCI Express Base device (i.e., a non-shared I/O endpoint) receives a DLLP that employs one of the reserved encodings (i.e., the DLLP is not corrupt, but is has one of the unused DLLP type encodings), it treats the received DLLP as a properly received DLLP, but will discard the received DLLP without further action.. Accordingly, the PCI Express Base device will continue to initialize at the data link layer by transmitting InitFC1 DLLPs until its link partner responds with InitFC1 DLLPs. Consequently, when a VMAC 2200 within a shared I/O switch according to the present invention initializes its link to an I/O endpoint, the link training logic 2208 directs the transmit data link layer logic 2212 via bus TSHARE 2232 to perform link initialization assuming that the I/O endpoint at the opposite end of the link is shared I/O capable. If shared I/O initialization DLLPs are received from the I/O endpoint, then the receive data link layer logic 2205 indicates such to the link training logic 2208 via bus RSHARE 2231, and the link training logic directs the transmit data link layer logic 2212 to initialize flow control for each OSD that the shared I/O endpoint supports. If, on the other hand, InitFC1 DLLPs are received from the I/O endpoint, then the receive data link layer logic 2205 indicates such to the link training logic 2208 via bus RSHARE and the link training logic 2208 directs the transmit data link layer logic 2212 via bus TSHARE 2232 to transmit InitFC1 DLLPs to the shared I/O endpoint and to complete initialization (i.e., transmitting InitFC2 DLLPs, etc.) of the link as a non-shared link. Accordingly, link initialization for both shared and non-shared I/O environments is perfected according to the present invention in a manner that is transparent to all of the operating system domains which are connected to the switch. That is, link initialization—shared or non-shared—is provided for by link training logic 2208 according to the present invention without requiring any intervention whatsoever by operating system resources associated with any of the operating system domains or by any other system software (e.g., management applications, etc.). If an I/O endpoint is not shared I/O capable, then its link is initialized as a PCI Express Base link. In addition, as a feature of the present invention, the link training logic 2208 is configured to retain knowledge of the capabilities of an I/O endpoint (or other link partner device) when a valid InitFC1 DLLP has been received, thus indicating that the I/O endpoint is a non-shared device. This retained knowledge can be employed if events occur over the link that cause the link to be initialized again (e.g., physical layer goes down for a period of time). When the link is initialized again, the link training logic 2208 directs the transmit data link layer logic 2212 via bus TSHARE 2232 to perform link initialization assuming that the I/O endpoint at the opposite end of the link is not shared I/O capable, and thus InitFC1 DLLPs are sent to the link partner. To further illustrate the concept of link initialization in a shared I/O environment, attention is now directed to
In operation, state DL_Inactive 2301 is present when physical layer reports that a corresponding link is non-operational or that nothing is connected to a corresponding port. The DL_Init state 2302 is entered when the physical layer reports that the corresponding link is operational. Flow control is then initialized for the default virtual channel on the corresponding link. When flow control has been established, then state DL-Active 2303 is entered to reflect normal operational mode of the corresponding link. DL_Inactive 2301 is the initial state entered following hot, warm, or cold reset. Upon entry to DL_Inactive 2301, all data link layer state information is set to default values and the contents of the data link layer retry buffer is discarded. In this state, transaction layer packet (TLP) information is discarded when received from both the transaction and physical layers and DLLPs are neither generated nor accepted.
When the physical layer reports that the link is up and if the transaction layer indicates that the link is not disabled, then state transitions to DL_Init 2302. In DL_Init 2302, flow control is initialized for the default virtual channel, VC0, according to flow control initialization protocol described in the PCI Express Base specification. When flow control initialization completes successfully, and if the physical layer reports that the physical link is up, then state transitions to DL_Active 2303. If, during flow control initialization, the physical layer reports that the link is down, then state transitions to DL_Inactive 2301.
In state DL_Init 2302, flow control is initialized for the default virtual channel, VC0, prior to exiting to state DL_Active 2303. Flow control initialization for all remaining virtual channels is accomplished in while in state DL_Active 2303. There are two states to the flow control initialization process for VCO: state FC_INIT1 and state FC_INIT2. While in state FC_INIT1, transmission of TLPs is blocked for the corresponding virtual channel and the following uninterrupted sequence of three successive InitFC1 DLLPs are transmitted: InitFC1-P (first), InitFC1-NP (second), and InitFC1-Cpl (third). This process is repeated until a link partner responds with InitFC1 and InitFC2 DLLPs, thus completing the flow control negotiation process. Following completion of flow control initialization for VC0, state DL_Active 2303 is entered.
In state DL_Active 2303, TLP information is accepted and transferred with the transaction and physical layers and DLLPs are generated and accepted. If the physical layer indicates that the link is down, then state transitions to DL_Inactive 2301.
The state machine 2300 of
Operation of the data link control state machine 2400 according to the present invention is substantially similar to operation of the state machine 2300 for link initialization in a non-shared environment. One skilled in the art will appreciate that according to the PCI Express Base Specification, following a fundamental or hot reset, devices must train their links and initialize flow control status. In a shared I/O environment, however, a shared I/O negotiation step is added after link training and prior to flow control initialization. Physical layer link training is the same in a shared I/O environment as that for a non-shared environment. All of the steps that are required by the PCI Base Specification (e.g., detection of the link partner, bit lock, symbol lock, lane polarity, etc.) are performed for link training in a shared I/O environment. Shared I/O negation occurs after physical layer link training. During negotiation, a shared I/O device determines whether its link partner agrees to use the shared I/O protocol as is herein described, which version of the protocol to use, whether the link partner supports a shared buffer, and how many OS domains are to be supported on the link. Shared I/O devices communicate these parameters via special DLLPs as will be further described below. Upon power up, a default number of operating system domains is established for each port in a shared I/O switch according to the present invention. To negotiate this number with a link partner, the present invention contemplates addition of a link initialization DLLP (described below) that will accomplished operating system domain initialization with a shared I/O-aware link partner. In one embodiment, the link initialization DLLP is an unused DLLP according to the PCI Express Base specification. When auto-negotiation of supported operating system domains is accomplished and when flow control has been is initialized for each default virtual channel for each of the supported operating system domains, then state DL_Active 2403 is entered to reflect normal operational mode of the corresponding link.
DL_Inactive 2401 is the initial state entered following hot, warm, or cold reset. Upon entry to DL_Inactive 2401, all data link layer state information is set to default values and the contents of the data link layer retry buffer is discarded. In this state, transaction layer packet (TLP) information is discarded when received from both the transaction and physical layers and DLLPs are neither generated nor accepted. DL_Inactive 2401 is present when physical layer reports that a corresponding link is non-operational or that nothing is connected to a corresponding port. The DL_Init_OSD1 state 2404 (i.e., InitOSD1 phase) is entered when the physical layer reports that the corresponding link is operational, and auto negotiation for the number of operating system domains that are supported on a corresponding port is begun. DL_Init_OSD12404 allows a device to indicate that it supports the shared I/O protocol as is herein contemplated. In addition, it allows the device to indicate whether it supports shared buffer flow control for virtual channels. Hence, when the physical layer reports that the link is up and if the transaction layer indicates that the link is not disabled, then state transitions to DL_Init_OSD12404. In DL_Init_OSD12404, the data link layer will begin transmitting the shared I/O link initialization DLLP, called InitOSD, continuously, until a valid InitOSD DLLP is received from a corresponding link partner, thus indicating that the link partner is capable of shared I/O operation and optionally indicated whether the link partner supports shared buffer flow control in addition to absolute buffer flow control. When a shared I/O device receives a valid InitOSD DLLP when in the DL_Init_OSD1 state 2404, state transitions to DL_Init_OSD22405 (InitOSD2 phase). While in DL_Init_OSD12404, if the link partner sends an Init_FC1 DLLP, or if a timeout occurs (e.g., 3 microsecond timeout, in one embodiment), then the shared I/O switch transitions state to DL_Init 2402 to complete flow control initialization with a non-shared I/O device.
DL_Init_OSD22405 allows the shared I/O device to set the actual number of OS domains that will be supported on a given link. In this state, as alluded to above, the default number of operating system domains which was established for each port in a shared I/O switch according to the present invention are transmitted to the link partner via InitOSD DLLPs. The link partner also transmits InitOSD DLLPs that indicate its number of supported OS domains. If the supported number of OS domains is received by a shared I/O device is equal to its current number of supported OS domains, then state transitions to DL_Init_OSD32406 (InitOSD3 phase). If not, negotiation of supported OS domains continues according to the rule that a device must decrease its number of supported OS domains to accommodate the capacity of its link partner. This negotiation continues until an InitOSD DLLP is received by a given shared I/O device indicating the same number of supported OS domains as which it is advertising over the link. In one embodiment, a 3 microsecond timer is employed to assure convergence. If supported OS domains are not negotiated within this time frame, then state transitions to DL_Inactive 2401.
The final handshake between two link partners is provided for in state DL_Init_OSD32406. In this state, a shared I/O device transmits valid InitOSD DLLPs that advertising the number of supported OS domains which it received in state DL_Init_OSD22405. When a valid InitOSD DLLP is received from its link partner indicating the very same number of supported OS domains, then state transitions to DL_Sh_Fc_Init 2407. InitOSD DLLPs which are received, but which do not indicated the same number of supported OS domains as are currently being advertised are ignored.
In state DL_Sh_Fc_Init 2407, flow control for each negotiated OSD is initialized in a manner substantially similar to that provided for by the PCI Express Base specification, with the exception that flow control initialization in a shared I/O environment employs a shared I/O flow control initialization DLLP, InitSHFC, to provide for designation of a particular operating system domain to which flow control is being initialized. In one embodiment, the flow control initialization DLLP, InitSHFC, employs unused DLLP types as described in the PCI Express Base specification. When flow control initialization completes successfully for a given operating system domain, and if the physical layer reports that the physical link is up, then state transitions to DL_Active 2403. If, during flow control initialization, the physical layer reports that the link is down, then state transitions to DL_Inactive 2401.
In state DL_Active 2403, TLP information for active operating system domains is accepted and transferred with the transaction and physical layers and DLLPs are generated and accepted. If the physical layer indicates that the link is down, then state transitions to DL_Inactive 2401.
Now referring to
In operation, when a VMAC 2200 according to the present invention performs link initialization, the initialization controller 2503 directs the shared initialization logic 2502 via us 2507 to provide InitOSD DLLPs to the transmit data link layer logic 2212 via bus TSHARE 2505 for auto-negotiation of supported operating system domains. If an InitOSD DLLP is received from a link partner by the receive data link layer logic 2205, then it is provided to the initialization controller 2503 and to the shared initialization logic 2507 via RSHARE 2504. Initialization continues in a shared mode as described above until successful completion when state DL_Active 2403 is entered. If an InitFC DLLP (i.e., flow control initialization DLLP in accordance with PCI Express Base Specification) is received from the link partner by the receive data link layer logic 2205, then it is provided to the initialization controller 2503 and the non-shared initialization logic 2501 over RSHARE 2504. Accordingly, the initialization controller 2503 directs the non-shared initialization logic 2501 to generate InitFC DLLPs for transmission to the link partner. The InitFC DLLPs are provided via bus TSHARE 2505. Non-shared I/O initialization continues as described above until successful initialization of flow control with the non-shared I/O device and state DL_Active 2403 is entered. In one embodiment, the initialization controller 2503 also includes timer logic (not shown) to cause a timeout to occur after transmission begins of InitOSD DLLPs. If this timeout occurs (3 microseconds, in one embodiment), then the non-shared initialization logic 2501 is directed to transmit InitFC DLLPs. In addition, as noted above, the link training logic 2500 is configured to retain knowledge of the capabilities of an I/O endpoint (or other link partner device) when a valid IniFC1 DLLP has been received, thus indicating that the I/O endpoint is a non-shared device. This retained knowledge can be employed if events occur over the link that cause the link to be initialized again (e.g., physical layer goes down for a period of time). In such a case, when the link is initialized again, the initialization controller 2503 directs the non-shared initialization logic 2501 via bus 2506 to generate InitFC DLLPs for transmission to the link partner.
Now turning to
Referring to
Now turning to
Referring to
Now referring to
In a shared I/O environment or in a non-shared I/O environment, as contemplated by the present invention, it is desirable to provide a switching apparatus 3006 and method that automatically detects switching port polarities. This feature allows for the addition of upstream components 3001 (i.e., components associated with an OSD) and the reconfiguration of downstream components 3002 (i.e., shared and non-shared I/O endpoints), without a requiring that the system configuration be known prior to configuration. Accordingly, the block diagram 3000 shows a switch 3006 as described herein. The switch 3006 has four ports 3010 as shown for clarity, although different numbers of ports 3010 (e.g., 8 ports 3010, 16 ports 3010) are contemplated in other embodiments of the switch 3006. One of the ports 3010 is coupled to a root complex 3001 (i.e., an upstream component) via a transmit/receive link 3003, and another port 3010 is coupled to an I/O endpoint 3002 (i.e., a downstream component). In one embodiment, the switch 3006 comprises a PCI Express based shared I/O switch 3006 utilizing PCI Express+ variant of the PCI Express protocol as described herein to allow the root complex 3001 (and other root complexes (not shown) as well) to communicate with the I/O endpoint 3002. In another embodiment, the switch 3006 comprises a PCI Express based non-shared I/O switch 3006 to allow the root complex 3001 to communicate with the shared I/O endpoint 3002. Alternative embodiments contemplate other load-store architectures and protocols and allow for use of combinations of shared and non-shared I/O endpoints 3002 as well. Each port 3010 in the switch 3006 has a port initialization logic element 3011 therein. The port initialization logic element 3011 automatically controls the polarity of a corresponding port 3010 without a requirement for system software intervention.
In operation, a given port initialization logic element 3011 provides for a default polarity. In one embodiment, the default polarity is a downstream port. When the links 3003 are initialized, each port initialization logic element 3011 directs its corresponding port 3010 to initialize as a port having the default polarity. In a PCI Express based shared I/O embodiment, the port initialization logic element 3011 directs its corresponding port 3010 to initially transmit TS1/TS2 ordered sets 3005 having a link number field 3006 set to the PAD character during the Configuration.Linkwidth.Start state of the Configuration Sub-State Machine, which is part of the Physical Layer Link Training and Status State Machine (LTSSM) as defined by the PCI Express Base specification, thus indicating that the ports 3010 have downstream port polarity. That is, transmission of TS1/TS2 ordered sets 3005 with the link number field 3006 set to PAD indicates a downstream port during Configuration.Linkwidth.Start and allows the upstream link partner to set the value of the link number field 3006.
Automatic configuration of a port 3010 is accomplished by the port initialization logic elements 3011 in conjunction with a programmable port polarity event. In one embodiment, the port polarity event comprises a timeout of a timer, upon which the port initialization logic element 3011 directs its corresponding port 3010 to switch polarity and to initialize as an upstream port 3010. In an alternative embodiment, the port polarity event comprises the occurrence of a programmable count of an event or events, upon which the port initialization logic element 3011 directs its corresponding port 3010 to switch polarity and to initialize as an upstream port 3010. In an exemplary PCI Express based embodiment, all the port initialization logic elements 3010 are configured to direct their corresponding ports 3010 to initialize as downstream ports, and to timeout after, for example, 20 milliseconds if communications corresponding to an upstream link partner 3001 are not received. According to the exemplary embodiment, and as shown in the block diagram 3000, since the I/O endpoint 3002 is a downstream link partner, the downstream port 3007 therein initializes with downstream port polarity and begins transmitting TS1/TS2 ordered sets 3005 having link number fields 3006 set to PAD. Thus, the port 3010 on the switch 3006 that is coupled to the I/O endpoint 3002 also initializes by transmitting TS1/TS2 ordered sets 3005 having link number fields 3006 set to PAD, because the port 3010 initializes in the default port polarity (i.e., downstream port polarity). At timeout, however, the port initialization logic element 3011 directs its corresponding port 3010 to automatically configure as an upstream port and begin transmitting TS1/TS2 ordered sets 3005 having link number fields 3006 set to a valid link number L#, thus indicating that it is an upstream port 3010 and thus allowing for completion of initialization without intervention by system software. Alternatively, the port polarity event in the exemplary embodiment could comprise receipt of, say, a count of 200 TS1/TS2 ordered sets 3005 with the link number field 3006 set to PAD. Hence, an advantage of the present invention is that the ports 3010 of a given switch 3006 can be coupled to either upstream components 3001 or downstream components 3002 and the switch 3006 itself automatically accomplishes port polarity configuration upon initialization.
Another embodiment of the present invention contemplates two switches 3006 coupled in tandem, such as is shown in the configuration of
Turning to
While not particularly shown, one skilled in the art will appreciate that many alternative embodiments may be implemented which differ from the above description, while not departing from the spirit and scope of the present invention as claimed. For example, the bulk of the above discussion has concerned itself with removing dedicated I/O from blade servers, and allowing multiple blade servers to share I/O devices though a load-store fabric interface on the blade servers. Such an implementation could easily be installed in rack servers, as well as pedestal servers. Further, blade servers according to the present invention could actually be installed in rack or pedestal servers as the processing complex, while coupling to other hardware typically within rack and pedestal servers such as power supplies, internal hard drives, etc. It is the separation of I/O from the processing complex, and the sharing or partitioning of I/O controllers by disparate complexes that is described herein. And the present inventors also note that employment of a shared I/O fabric according to the present invention does not preclude designers from concurrently employing non-shared I/O fabrics within a particular hybrid configuration. For example, a system designer may chose to employ a non-shared I/O fabric for communications (e.g., Ethernet) within a system while at the same time applying a shared I/O fabric for storage (e.g., Fiber Channel). Such a hybrid configuration is comprehended by the present invention as well.
Additionally, it is noted that the present invention can be utilized in any environment that has at least two processing complexes executing within two independent OS domains that require I/O, whether network, data storage, or other type of I/O is required. To share I/O, at least two operating system domains are required, but the operating system domains can share only one shared I/O endpoint. Thus, the present invention envisions two or more operating system domains which share one or more I/O endpoints.
Furthermore, one skilled in the art will appreciate that many types of shared I/O controllers are envisioned by the present invention. One type, not mentioned above, includes a keyboard, mouse, and/or video controller (KVM). Such a KVM controller would allow blade servers such as those described above, to remove the KVM controller from their board while still allowing an interface to keyboards, video and mouse (or other input devices) from a switch console. That is, a number of blade servers could be plugged into a blade chassis. The blade chassis could incorporate one or more shared devices such as a boot disk, CDROM drive, a management controller, a monitor, a keyboard, etc., and any or all of these devices could be selectively shared by each of the blade servers using the invention described above.
Also, by utilizing the mapping of OS domain to shared I/O controller within a shared I/O switch, it is possible to use the switch to “partition” I/O resources, whether shared or not, to OS domains. For example, given four OS domains (A, B, C, D), and four shared I/O resources (1, 2, 3, 4), three of those resources might be designated as non-shared (1, 2, 3), and one designated as shared (4). Thus, the shared I/O switch could map or partition the fabric as: A-1, B-2, C-3/4, D-4. That is, OS domain A utilizes resource 1 and is not provided access to or visibility of resources 2-4; OS domain B utilizes resource 2 and is not provided access to or visibility of resources 1, 3, or 4; OS domain C utilizes resources 3 and 4 and is not provided access to or visibility of resources 1-2; and OS domain D utilizes resource 4 and shares resource 4 with OS domain C, but is not provided access to or visibility of resources 1-3. In addition, neither OS domain C or D is aware that resource 4 is being shared with another OS domain. In one embodiment, the above partitioning is accomplished within a shared I/O switch according to the present invention.
Furthermore, the present invention has utilized a shared I/O switch to associate and route packets from root complexes associated with one or more OS domains to their associated shared I/O endpoints. As noted several times herein, it is within the scope of the present invention to incorporate features that enable encapsulation and decapsulation, isolation of OS domains and partitioning of shared I/O resources, and routing of transactions across a load-store fabric, within a root complex itself such that everything downstream of the root complex is shared I/O aware (e.g., PCI Express+). If this were the case, shared I/O controllers could be coupled directly to ports on a root complex, as long as the ports on the root complex provided shared I/O information to the I/O controllers, such as OS domain information. What is important is that shared I/O endpoints be able to recognize and associate packets with origin or upstream OS domains, whether or not a shared I/O switch is placed external to the root complexes, or resides within the root complexes themselves.
And, if the shared I/O functions herein described were incorporated within a root complex, the present invention also contemplates incorporation of one or more shared I/O controllers (or other shared I/O endpoints) into the root complex as well. This would allow a single shared I/O aware root complex to support multiple upstream OS domains while packaging everything necessary to talk to fabrics outside of the load-store domain (Ethernet, Fiber Channel, etc.) within the root complex. Furthermore, the present invention also comprehends upstream OS domains that are shared I/O aware, thus allowing for coupling of the OS domains directly to the shared I/O controllers, all within the root complex.
And, it is envisioned that multiple shared I/O switches according to the present invention be cascaded to allow many variations of interconnecting root complexes associated with OS domains with downstream I/O devices, whether the downstream I/O devices shared or not. In such a cascaded scenario, an OSD Header may be employed globally, or it might be employed only locally. That is, it is possible that a local ID be placed within an OSD Header, where the local ID particularly identifies a packet within a given link (e.g., between a root complex and a switch, between a switch and a switch, and/or between a switch and an endpoint). So, a local ID may exist between a downstream shared I/O switch and an endpoint, while a different local ID may be used between an upstream shared I/O switch and the downstream shared I/O switch, and yet another local ID between an upstream shared I/O switch and a root complex. In this scenario, each of the switches would be responsible for mapping packets from one port to another, and rebuilding packets to appropriately identify the packets with their associating upstream/downstream port.
As described above, it is further envisioned that while a root complex within today's nomenclature means a component that interfaces downstream devices (such as I/O) to a host bus that is associated with a single processing complex (and memory), the present invention comprehends a root complex that provides interface between downstream endpoints and multiple upstream processing complexes, where the upstream processing complexes are associated with multiple instances of the same operating system (i.e., multiple OS domains), or where the upstream processing complexes are executing different operating systems (i.e., multiple OS domains), or where the upstream processing complexes are together executing a one instance of a multi-processing operating system (i.e., single OS domain). That is, two or more processing complexes might be coupled to a single root complex, each of which executes their own operating system. Or, a single processing complex might contain multiple processing cores, each executing its own operating system. In either of these contexts, the connection between the processing cores/complexes and the root complex might be shared I/O aware, or it might not. If it is, then the root complex would perform the encapsulation/decapsulation, isolation of OS domain and resource partitioning functions described herein above with particular reference to a shared I/O switch according to the present invention to pass packets from the multiple processing complexes to downstream shared I/O endpoints. Alternatively, if the processing complexes are not shared I/O aware, then the root complexes would add an OS domain association to packets, such as the OSD header, so that downstream shared I/O devices could associate the packets with their originating OS domains.
It is also envisioned that the addition of an OSD header within a load-store fabric, as described above, could be further encapsulated within another load-store fabric yet to be developed, or could be further encapsulated, tunneled, or embedded within a channel-based fabric such as Advanced Switching (AS) or Ethernet. AS is a multi-point, peer-to-peer switched interconnect architecture that is governed by a core AS specification along with a series of companion specifications that define protocol encapsulations that are to be tunneled through AS fabrics. These specifications are controlled by the Advanced Switching Interface Special Interest Group (ASI-SIG), 5440 SW Westgate Drive, Suite 217, Portland, Oreg. 97221 (Phone: 503-291-2566). For example, within an AS embodiment, the present invention contemplates employing an existing AS header that specifically defines a packet path through a I/O switch according to the present invention. Regardless of the fabric used downstream from the OS domain (or root complex), the inventors consider any utilization of the method of associating a shared I/O endpoint with an OS domain to be within the scope of their invention, as long as the shared I/O endpoint is considered to be within the load-store fabric of the OS domain.
Although the present invention and its objects, features and advantages have been described in detail, other embodiments are encompassed by the invention. In addition to implementations of the invention using hardware, the invention can be implemented in computer readable code (e.g., computer readable program code, data, etc.) embodied in a computer usable (e.g., readable) medium. The computer code causes the enablement of the functions or fabrication or both of the invention disclosed herein. For example, this can be accomplished through the use of general programming languages (e.g., C, C++, JAVA, and the like); GDSII databases; hardware description languages (HDL) including Verilog HDL, VHDL, Altera HDL (AHDL), and so on; or other programming and/or circuit (i.e., schematic) capture tools available in the art. The computer code can be disposed in any known computer usable (e.g., readable) medium including semiconductor memory, magnetic disk, optical disk (e.g., CD-ROM, DVD-ROM, and the like), and as a computer data signal embodied in a computer usable (e.g., readable) transmission medium (e.g., carrier wave or any other medium including digital, optical or analog-based medium). As such, the computer code can be transmitted over communication networks, including Internets and intranets. It is understood that the invention can be embodied in computer code (e.g., as part of an IP (intellectual property) core, such as a microprocessor core, or as a system-level design, such as a System on Chip (SOC)) and transformned to hardware as part of the production of integrated circuits. Also, the invention may be embodied as a combination of hardware and computer code.
Finally, those skilled in the art should appreciate that they can readily use the disclosed conception and specific embodiments as a basis for designing or modifying other structures for carrying out the same purposes of the present invention without departing from the spirit and scope of the invention as defined by the appended claims.
This application claims the benefit of the following U.S. Provisional Applications, each of which is herein incorporated by reference for all intents and purposes. SERIALFILINGNUMBERDATETITLE60/541673Feb. 4, 2004PCI SHARED IO WIRE(NEXTIO.0107)LINE PROTOCOL60/555127Mar. 22, 2004PCI EXPRESS SHARED(NEXTIO.0108)IO WIRELINE PROTOCOLSPECIFICATION60/575005May 27, 2004NEXSIS SWITCH(NEXTIO.0109)60/588941Jul. 19, 2004SHARED I/O DEVICE(NEXTIO.0110)60/589174Jul. 19, 2004ARCHITECTURE(NEXTIO.0111)60/615775Oct. 4, 2004PCI EXPRESS(NEXTIO.0112SHARED IO WIRELINEPROTOCOL SPECIFICATION This application is a continuation-in-part of to the following co-pending U.S. Patent Applications, each of which have a common assignee and at least one common inventor, and each of which is herein incorporated by reference for all intents and purposes. SERIALFILINGNUMBERDATETITLE10/827622Apr. 19, 2004SWITCHING APPARATUS AND(NEXTIO.0400)METHOD FOR PROVIDINGSHARED I/O WITHIN ALOAD-STORE FABRIC10/827620Apr. 19, 2004SWITCHING APPARATUS AND(NEXTIO.0401)METHOD FOR PROVIDINGSHARED I/O WITHIN ALOAD-STORE FABRIC10/827117Apr. 19, 2004SWITCHING APPARATUS AND(NEXTIO.0402)METHOD FOR PROVIDINGSHARED I/O WITHIN ALOAD-STORE FABRIC Each of the above-noted co-pending U.S. Patent Applications is a continuation-in-part of U.S. patent application Ser. No. 10/802532 (Docket: NEXTIO.0200), entitled SHARED INPUT/OUTPUT LOAD-STORE ARCHITECTURE, filed on Mar. 16, 2004, which has a common assignee and at least one common inventor, and which is herein incorporated by reference for all intents and purposes. In addition, each of the above-noted co-pending U.S. Patent Applications claim the benefit of the following U.S Provisional Applications. SERIALFILINGNUMBERDATETITLE60/464382Apr. 18, 2003SHARED-IO PCI(NEXTIO.0103)COMPLIANT SWITCH60/491314Jul. 30, 2003SHARED NIC(NEXTIO.0104)BLOCK DIAGRAM60/515558Oct. 29, 2003NEXIS(NEXTIO.0105)60/523522Nov. 19, 2003SWITCH FOR(NEXTIO.0106)SHARED I/O FABRIC60/541673Feb. 4, 2004PCI SHARED I/O(NEXTIO.0107)WIRE LINE PROTOCOL60/555127Mar. 22, 2004PCI EXPRESS SHARED(NEXTIO.0108)IO WIRELINE PROTOCOLSPECIFICATION Co-pending U.S. patent application Ser. No. 10/802532 (Docket: NEXTIO.0200) claims the benefit of the following U.S. Provisional Applications. SERIALFILINGNUMBERDATETITLE60/464382Apr. 18, 2003SHARED-IO PCI(NEXTIO.0103)COMPLIANT SWITCH60/491314Jul. 30, 2003SHARED NIC(NEXTIO.0104)BLOCK DIAGRAM60/515558Oct. 29, 2003NEXIS(NEXTIO.0105)60/523522Nov. 19, 2003SWITCH FOR(NEXTIO.0106)SHARED I/O FABRIC60/541673Feb. 4, 2004PCI SHARED I/O(NEXTIO.0107)WIRE LINE PROTOCOL Co-pending U.S. patent application Ser. No. 10/802532 (Docket: NEXTIO.0200) is a continuation-in-part of the following co-pending U.S. Patent Applications, each of which have a common assignee and at least one common inventor, and each of which are herein incorporated by reference for all intents and purposes. SERIALFILINGNUMBERDATETITLE10/757713Jan. 14, 2004METHOD AND APPARATUS(NEXTIO.0301)FOR SHARED I/O INA LOAD/STORE FABRIC10/757711Jan. 14, 2004METHOD AND APPARATUS(NEXTIO.0302)FOR SHARED I/O INA LOAD/STORE FABRIC10/757714Jan. 14, 2004METHOD AND APPARATUS(NEXTIO.0300FOR SHARED I/O INA LOAD/STORE FABRIC The three aforementioned co-pending U.S. Patent Applications (i.e., Ser. Nos. 10/757713, 10/757711, and 10/757714) claim the benefit of the following U.S. Provisional Applications. SERIALFILINGNUMBERDATETITLE60/440788Jan. 15, 2003SHARED IO ARCHITECTURE(NEXTIO.0101)60/440789Jan. 21, 20033GIO-XAUI COMBINED(NEXTIO.0102)SWITCH60/464382Apr. 18, 2003SHARED-IO PCI (NEXTIO.0103)COMPLIANT SWITCH60/491314Jul. 30, 2003SHARED NIC BLOCK(NEXTIO.0104)DIAGRAM60/515558Oct. 29, 2003NEXIS(NEXTIO.0105)60/523522Nov. 19, 2003SWITCH FOR SHARED(NEXTIO.0106)I/O FABRIC60/541673Feb. 4, 2004PCI SHARED I/O WIRE(NEXTIO.0107)LINE PROTOCOL This application is related to the following co-pending U.S. Patent Applications. SERIALFILINGNUMBERDATETITLE10/864766Jun. 9, 2004METHOD AND APPARATUS(NEXTIO.0310)FOR A SHARED I/O SERIALATA CONTROLLER{overscore ((NEXTIO.0311))}Jan. 27, 2005FIBRE CHANNEL CONTROLLERSHAREABLE BY APLURALITY OFOPERATING SYSTEM DOMAINSWITHIN A LOAD-STOREARCHITECTURE10/909254Jul. 30, 2004METHOD AND APPARATUS(NEXTIO.0312)FOR A SHARED I/ONETWORK INTERFACECONTROLLER10/972669Oct. 25, 2004SWITCHING APPARATUS AND(NEXTIO.0403)METHOD FOR LINKINITIALIZATIONIN A SHAREDI/O ENVIRONMENT{overscore ((NEXTIO.0409))}Jan. 27, 2005NETWORK CONTROLLER FOROBTAINING A PLURALITY OFNETWORK PORT IDENTIFIERSIN RESPONSE TO LOAD-STORE TRANSACTIONS FROM ACORRESPONDING PLURALITYOF OPERATING SYSTEMDOMAINS WITHINA LOAD-STOREARCHITECTURE{overscore ((NEXTIO.0411))}Jan. 27, 2005FIBRE CHANNEL CONTROLLERSHAREABLE BY A PLURALITYOF OPERATING SYSTEMDOMAINS WITHINA LOAD-STOREARCHITECTURE{overscore ((NEXTIO.0412))}Jan. 27, 2005FIBRE CHANNEL CONTROLLERSHAREABLE BY APLURALITY OFOPERATING SYSTEM DOMAINSWITHIN A LOAD-STOREARCHITECTURE
| Number | Date | Country | |
|---|---|---|---|
| 60541673 | Feb 2004 | US | |
| 60555217 | Mar 2004 | US | |
| 60575005 | May 2004 | US | |
| 60588941 | Jul 2004 | US | |
| 60589174 | Jul 2004 | US | |
| 60615775 | Oct 2004 | US | |
| 60515558 | Oct 2003 | US | |
| 60523522 | Nov 2003 | US | |
| 60541673 | Feb 2004 | US | |
| 60555127 | Mar 2004 | US | |
| 60464382 | Apr 2003 | US | |
| 60491314 | Jul 2003 | US | |
| 60515558 | Oct 2003 | US | |
| 60541673 | Feb 2004 | US | |
| 60555127 | Mar 2004 | US | |
| 60491314 | Jul 2003 | US | |
| 60491314 | Jul 2003 | US | |
| 60515558 | Oct 2003 | US | |
| 60523522 | Nov 2003 | US | |
| 60541673 | Feb 2004 | US | |
| 60555127 | Mar 2004 | US | |
| 60464382 | Apr 2003 | US | |
| 60491314 | Jul 2003 | US | |
| 60515558 | Oct 2003 | US | |
| 60523522 | Nov 2003 | US | |
| 60541673 | Feb 2004 | US | |
| 60440788 | Jan 2003 | US | |
| 60440789 | Jan 2003 | US | |
| 60464382 | Apr 2003 | US | |
| 60491314 | Jul 2003 | US | |
| 60515558 | Oct 2003 | US | |
| 60523522 | Nov 2003 | US | |
| 60440788 | Jan 2003 | US | |
| 60440789 | Jan 2003 | US | |
| 60464382 | Apr 2003 | US | |
| 60491314 | Jul 2003 | US | |
| 60515558 | Oct 2003 | US | |
| 60523522 | Nov 2003 | US | |
| 60440788 | Jan 2003 | US | |
| 60440789 | Jan 2003 | US | |
| 60464382 | Apr 2003 | US | |
| 60491314 | Jul 2003 | US | |
| 60515558 | Oct 2003 | US | |
| 60523522 | Nov 2003 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 10827622 | Apr 2004 | US |
| Child | 11048393 | Jan 2005 | US |
| Parent | 10827117 | Apr 2004 | US |
| Child | 11048393 | Jan 2005 | US |
| Parent | 10827620 | Apr 2004 | US |
| Child | 11048393 | Jan 2005 | US |
| Parent | 10802532 | Mar 2004 | US |
| Child | 10827622 | US | |
| Parent | 10802532 | Mar 2004 | US |
| Child | 10827620 | US | |
| Parent | 10802532 | Mar 2004 | US |
| Child | 10827117 | US | |
| Parent | 10757713 | Jan 2004 | US |
| Child | 10827117 | US | |
| Parent | 10757711 | Jan 2004 | US |
| Child | 10827117 | US | |
| Parent | 10757714 | Jan 2004 | US |
| Child | 10827117 | US |
