This invention relates in general to the field of electronics, and more specifically to an InterProcessor Communication (IPC) protocol/network providing intelligent targeting of nodes.
Most electronic systems include a number of networked elements (components) such as hardware and software that form the system. In most systems there is a layer responsible for communication between the different components that form a networked element as well as between the different networked elements themselves. This layer is typically referred to as the InterProcessor Communication (IPC) layer.
Several protocols have been introduced in the last few years to deal with interprocessor communications. One example of an IPC product is PCI AGP Controller (PAC) that integrates a Host-to-PCI bridge, Dynamic Random Access Memory (DRAM) controller and data path and an Accelerated Graphics Port (AGP) interface. Another example of an IPC product is the OMAP™ platforms. Neither of these platforms provide much if any support above the hardware level and provide little design flexibility at the lower level component or channel levels (physical layer).
The PAC platforms for example, are closed architectures and are embedded into the Operating System's TAPI layer, with the IPC code not being accessible to developers. Therefore, these platforms do not extend to the component levels and they also do not allow for dynamic assignment of IPC resources, hardware support capabilities, or multi-node routing, etc. as well as not allowing for the dynamic assignment of the IPC resources. With the need for lower power consumption and less system latencies, a need exists in the art for an IPC network that can provide for intelligent targeting of IPC nodes so that there is less wasted time and less power consumption when looking for a processor in the IPC system that can provide a needed service.
The features of the present invention, which are believed to be novel, are set forth with particularity in the appended claims. The invention may best be understood by reference to the following description, taken in conjunction with the accompanying drawings, in the several figures of which like reference numerals identify like elements, and in which:
While the specification concludes with claims defining the features of the invention that are regarded as novel, it is believed that the invention will be better understood from a consideration of the following description in conjunction with the drawing figures.
The IPC of the present invention provides the support needed for different processors operating in a system to communicate with each other. For example, in a dual processor (or multi-processor) radio architecture for use in a radio communication device that includes an Application Processor (AP) and a Baseband Processor (BP), the IPC provides the support needed for the processors to communicate with each other in an efficient manner. The IPC provides this support without imposing any constrains on the design of the AP or BP.
The IPC allows any processor that adopts the IPC as its inter-processor communication stack to co-exist together and operate as if the two were actually running on the same processor core sharing a common operating system and memory. With the use of multiple processors in communication devices becoming the norm, the IPC of the present invention provides for reliable communications between the different processors.
The IPC hardware provides the physical connection that ties together the different processors to the IPC network. Data packets are preferably transported between the different hosts asynchronously in one embodiment of the invention. Processors that are connected to the IPC network have their physical and logical addresses statistically or dynamically assigned (e.g., IPC addresses). Also, since data packets can flow in any direction within the IPC network in one embodiment of the invention, they need to carry a destination address of the processor that they are trying to reach. Packets are also preferably checked for errors using conventional Cyclic Redundancy Check (CRC) techniques. Although the network activities of the IPC network of the present invention may have some similarities to those found on an internet network that uses IP transport layers such as a Transmission Control Protocol/Internet Protocol (TCP/IP) network, the IPC of the present invention is not divided into smaller networks with gateways as in a TCP/IP network.
Referring now to
In
(1). IPC Presentation Manager (202)—this layer is used to translate different data types between different system components (e.g., software threads).
(2). IPC Session Manager (204)—this layer is a central repository for all incoming/outgoing IPC traffic between the IPC stack and all of the system components. The IPC session manager 204 has several functions: assignment of component IDs for participating IPC components; deciding if the IPC data needs to be encapsulated; routing of IPC data, termination of IPC traffic; place holder for IPC processors; providing IPC addresses, assigning and authenticating IPC clients, etc.
(3). Device Interface Layer (206)—is responsible for managing the IPC physical-to-logical IPC channels. Its main function is to abstract the IPC hardware completely so that the stack IPC becomes hardware independent. The device interface layer 206 manages the physical bandwidth of the IPC link underneath to support all of the IPC logical channels. In the incoming path, the device interface layer 206 picks up data from different physical channels 110-114 and passes them up to the rest of the IPC stack. On the outgoing path, the device interface layer 206 manages the data loading of the IPC logical channels by sending them onto the appropriate physical channels. The device interface layer 206 also handles concatenating IPC packets belonging to the same IPC channel before sending them to the IPC hardware. Channel requirements are pre-negotiated between the IPC session manager 204 and the IPC device interface layer 206. The device interface layer 206 provides for hardware ports which in turn provide a device interface to an IPC client 102-106.
Referring to
Referring now to
In
Referring now to
More detailed steps taken during the IPC client initialization process are shown in
The parameters in the configuration request include the node type and the data type. The session server in response to the configuration request in step 702 assigns the requestor an IPC address. It also sets up a dynamic routing table for the requester if one does not exist. It then sends the requestor a configuration indication as in step 708. The configuration indication parameters include the IPC address of the server and the newly assigned IPC address of the client.
In response to receiving the configuration indication, components attached to the session client can request control/data from the client's session manager. The Session client then sends a configuration indication confirm message to the session server in step 710. The “configuration indication confirm” message has no parameters, upon receiving the configuration indication confirm message in step 710, the session server can initiate IPC streams to the newly configured session client. The session server then sends configuration update messages to the session clients in steps 712 and 714. This causes the both session clients shown in
When a packet is received by an IPC session manager, it comes in the form of data that includes the source component ID, the destination ID, a channel ID and the type of BP or AP. The IPC session manager will add the destination component ID in the event that the destination ID is not inserted. The IPC session manager will also insert an IPC address. It is the IPC session manager that discovers the destination ID based on the message opcode received. The destination ID is based on a lookup table. This lookup table is updated dynamically each time a component subscribes to a new IPC message opcode (e.g., an audio component subscribes to audio messages by sending a request to the IPC session manager).
In
In
The same component initialization process is shown between a component (client) 1002, a session (client) also known as a client session manager 1004 and the session (server) also known as the server session manager 1006 in
Once the client session manager 1004 replies in step 1010 to the configuration request in step 1008, the client session manager 1004 sends a configuration update request in step 1012 to the session server 1006. The parameters for the configuration update request are any new changes that have been made in the dynamic routing table. The session manager updates the dynamic routing table for that IPC address. The server session manager 1006 in step 1016 then sends all the IPC clients a configuration update, while it sends the IPC client a configuration update indication in step 1014. The server's session manager 1006 makes sure the IPC server has updated its routing table with the changes that were sent.
In the configuration update message of step 1016 which includes the dynamic routing tables as a parameter(s), the session server 1006 updates the dynamic routing tables and sends a configuration update confirm message in step 1018. The session server 1006 then makes sure all of the IPC participants have been updated.
The IPC session manager determines the routing path of incoming and outgoing IPC packets. The route of an outgoing packet is determined by the component's IPC address. If the destination address is found to be that of a local processor, a mapping of the IPC to the Operating System (OS) is carried out within the session manager. If the destination address is found to be for a local IPC client, the packet is sent to the IPC stack for further processing (e.g., encapsulation). Note that if the destination component is located on the same processor as the component sending the IPC packet, no encapsulation is required and the packet gets passed over through the normal OS message calling (e.g., Microsoft Message Queue, etc.). In this way components do not have to worry about modifying their message input schemes. They only need to change their message posting methodologies from an OS specific design to an IPC call instead.
For incoming packets, if the destination address of the message is not equal to the IPC server's, the incoming packets gets routed to the proper IPC client. The routing of incoming packets is handled by the session manager of the IPC server. Otherwise, the message is forwarded to the right component or components depending on whether or not the component destination ID is set to a valid component ID or to 0XFF.
The IPC router block transports the IPC data to the destination component. Incoming IPC messages carry among other things, the originator component ID and the IPC message opcodes such as those for Audio, Modem, etc. The IPC session manager relies on its component routing table to send the IPC data to the right component(s). Both the dynamic routing table and the component routing table are updated by the IPC server/client.
During power-up, each component must register itself with its session manager to obtain an IPC component ID. In addition, it must also subscribe to incoming IPC messages such as Audio, Modem, etc. This information is stored in the component routing table for use by the IPC session manager.
When a component 1102, as shown in
In
A typical IPC data request in accordance with an embodiment of the invention is shown in
In step 1304, the component update indication is sent to the component. If the node type is equal to 0xFF, the node tables are returned to the component. If the opcode field is equal to 0xFF, the list of opcodes is returned to the component. However, if the opcode is a specific value, a true or false message is returned. In step 1306, a component data request is made. The parameters for the component data request include the node type, the IPC message opcode, the IPC message data, the channel ID and the component ID. In a component data request, the session manager checks the node type to determine whether the opcode is supported. If the node type does not support the opcode, a component update indication is sent in step 1308. If however, the node type supports the opcode, a data request is sent to the device layer in step 1310. The data request parameters include the IPC message, the channel ID and the IPC header.
The device layer schedules to send the data request message based on the channel ID. The device layer selects the IPC hardware based on the port # header information. Once the data is committed, a data confirm message is sent to the session manager in 1312. In step 1314, the session manager proceeds to send a component data confirm message to the component. The component can wait for the confirmation before sending more IPC messages. Once a data confirm is received, the component can proceed to send the next IPC message.
In step 1316, the device layer sends a data indication message including IPC message data and an IPC header. The session manager checks the destination IPC header of the message, and if different from the local IPC address, the session manager sends (routes) the message to the right IPC node. In step 1310, the session manager sends a data request to the device layer with a reserved channel ID. The session manager checks the destination component ID, and if it is equal to 0xFF, routes the message to all the components subscribed to that opcode. In step 1318, the session manager sends a component data indication message and the component receives the IPC data.
The IPC stack uses a reserved control channel for communication purposes between all participating IPC nodes. On power-up, the IPC server's session manager uses this link to broadcast messages to IPC clients and vice versa. During normal operations, this control channel is used to carry control information between all APs and BPs.
In
IPC Application Program Interfaces (APIs)
Below are listed some of the APIs for the IPC protocol of the present invention.
CreateComponentInst( )
OpenChannelKeep( )
OpenChannel( )
OpenChannelWThru( )
CloseChannel( )
ChannelGrant( )
ChannelError( )
ChannelDataIndication( )
DataChannelRequest( )
ChannelClose( )
OpenChannel( )
CloseChannel( )
ChannelGrant( )
ChannelError( )
ChannelDataIndication( )
DataChannelRequest( )
ChannelClose( )
ChannelDataIndication( )
OpenChannel( )
CloseChannel( )
ChannelGrant( )
ChannelError( )
ChannelDataIndication( )
DataChannelRequest( )
ChannelClose( )
In
The IPC protocol allows for the dynamic addition of any IPC conforming MA into the IPC link for communication. Thus, an IPC network is formed without any compile time dependencies, or any other software assumptions. The IPC of the present invention presents a standard way for software components to communicate with the IPC stack and the hardware below the stack is also abstracted such that components can choose different links to communicate.
Referring now to
In
Intelligent Targeting of IPC Nodes
In order to provide improved overhead efficiency and power management, an IPC protocol/network in accordance with the invention provides for intelligent targeting of nodes (any IPC server or client processor, also referred to an MA in a radio environment) based on the state of the processors operating within the IPC network. For example, when a MA tries to find support for a particular service (e.g., MP3 decoder, etc.) on the IPC network through the use of intelligent targeting, the source application will find a processor or MA that can provide the service in the most efficient fashion. This will provide for improved battery life and less latency to the MA applications.
In one embodiment, the requesting or source MA will send to the IPC server's IPC stack a message with the particular opcode corresponding to the requested service, for example a MP3 service. The IPC stack will then check its opcode table to determine the requested service and then the IPC stack will check its subscription lists to see which MAs/nodes can support the requested service, in this particular example, perform MP3 decoding.
As an illustrative example, if 3 MAs on the IPC network can provide the MP3 decoding service, but one is in a deep sleep mode, another one is busy performing other tasks and does not have the necessary unused processing horsepower to perform MP3 decoding, and the third MA can perform the MP3 processing, the IPC stack will send a message back to the requesting MA and inform it which MA can perform the MP3 decoding based on the current state of the MAs capable of performing the MP3 decoding service.
If the requesting MA does not have a particular need to use a particular MA, it will use the MA that the IPC server's IPC stack has recommended. In this fashion, the resources on the whole IPC network will be best utilized. Located in the IPC server's IPC stack will be a table with the current operational state of each of the MA's (nodes) operating in the IPC network. The table will keep track of which of the MA's/IPC client's are currently in a sleep mode, or are busy performing other tasks. Along with the opcode and subscription lists, the operational state table will let the IPC network know the current operating condition of all of the MAs including their traffic loads and capabilities and whether or not they are currently in sleep mode.
In order to provide the necessary information stored in the operational state table that is found in the IPC server, if an MA in the IPC network wants to go into a sleep mode it will alert the IPC server by sending a control message or API to it that lets it know that is will be going into a sleep mode. The MAs will also alert the IPC server when they wake up and return to normal operation, so that the usage table can be updated. The traffic load information for each of the MAs will also be provided by the MAs to the IPC server, for example prior to a MA commencing a task such as video decoding, it will send a message to the IPC server informing the IPC server that is commencing a task and that a certain amount (e.g., 20%) of its processing capabilities are currently being used. Alternatively, nodes can report to the IPC server when they have reached a certain computational usage threshold level, for example, when they have reached 50% of computational capacity or they can provide their usage levels periodically, every few minutes.
In order to perform the requirements of the present invention, the IPC server such as IPC server 108 shown in
Referring now to
The IPC server will then look through its operational state table which is frequently updated and determine which MA is in the best condition to handle the service request. For example, it could be that out of three MAs that can support the requested service, one may be in sleep mode, one may be fully occupied from a resource standpoint to handle any further service requests, and one may have enough capacity to handle the request. In step 1908, the IPC server sends a message to the service requestor informing it which MA will handle the service request. The IPC server can also update its operational state table at this time noting that the selected MA is performing the service and adjust its operational capabilities accordingly (e.g., note that the MA is using 10% of its computational capability performing the requested service).
In
In row 2006, there is shown that the node 3 processor is currently available to handle any service requests since it may not be performing any services currently. In row 2008, the node 4 processor is shown to be 50% available, but that its MP3 decode capability is not available (e.g., may be currently being used).
So if in this particular example the service requestor had asked for an MP3 decode service, the IPC server would first find all the nodes in the IPC network that can handle the MP3 decode service request (e.g., nodes 1-4), then it would search through the operational state table 2000 to determine which node would be best suited to handle the request. In this example, node 3 would be in the best condition to handle the service request. The IPC server would send a message to the service requestor informing it that node 3 will handle its request.
In an alternative embodiment of the invention, the service requestor can not only send a request for a specific service by sending a message with the opcode corresponding to the requested service, but can also inform the IPC server that it wants a specific node to handle the request. In this particular case, if for example the service requestor wanted node 4 to handle the service request, the IPC server could keep checking its operational state table until node 4 was found to be available, and at that time, assign node 4 to handle the service request.
In
If client 2 (2104) handles a service request from either sub-client 2108 or 2110 and informs it which node to use for the requested service, it must immediately inform the IPC server 2100 so that it can update its operational state table and keep the operational state information of all the nodes in the network current. “Dropping” the operational state table information to lower levels of a daisy-chained IPC network allows for service request decisions to be made quicker, since the service request does not have to travel multiple layers. The operational state table information can be sent to select ones of the clients on a predetermined basis, such as when an update to the operational table in the IPC server 2112 is updated (e.g., one of the nodes (clients) has gone to sleep and informed the IPC server 2112) or after a predetermined period of time has elapsed.
While the preferred embodiments of the invention have been illustrated and described, it will be clear that the invention is not so limited. Numerous modifications, changes, variations, substitutions and equivalents will occur to those skilled in the art without departing from the present invention as defined by the appended claims.
Number | Name | Date | Kind |
---|---|---|---|
5224100 | Lee et al. | Jun 1993 | A |
5623605 | Keshav et al. | Apr 1997 | A |
5926636 | Lam et al. | Jul 1999 | A |
6154785 | Lakhat et al. | Nov 2000 | A |
20020049691 | Majoor | Apr 2002 | A1 |
20020049842 | Huetsch et al. | Apr 2002 | A1 |
20020107962 | Richter et al. | Aug 2002 | A1 |
20030097443 | Gillett et al. | May 2003 | A1 |
20030120782 | Bortoloso et al. | Jun 2003 | A1 |
20030135666 | Kim | Jul 2003 | A1 |
20040205767 | Partanen | Oct 2004 | A1 |
20050044162 | Liang et al. | Feb 2005 | A1 |
Number | Date | Country | |
---|---|---|---|
20050076122 A1 | Apr 2005 | US |