WO2022160307A1

WO2022160307A1 - Router and system on chip

Info

Publication number: WO2022160307A1
Application number: PCT/CN2021/074543
Authority: WO
Inventors: 郭鑫; 徐度; 朱强
Original assignee: 华为技术有限公司
Priority date: 2021-01-30
Filing date: 2021-01-30
Publication date: 2022-08-04
Also published as: CN116686269A

Abstract

A router and a system on chip, which are used for improving the utilization rate of bandwidths when scheduling packets. The router comprises a plurality of inports and a plurality of outports, wherein each inport is used to send, with regard to a plurality of input packets, a request message to the outports corresponding the respective packets; and each outport is used to arbitrate the received request message, and send a permission message to the inport which passes the arbitration. By means of the described solution, when sending, to outports, a request message requesting for the scheduling of packets, each inport can send a request to a plurality of outports respectively with regard to a plurality of input packets, and the sending of the request by the inport is not limited by a service type. Compared with the prior art, in the case where a request message sent by an inport to a certain outport has not passed the arbitration, the inport can also be arbitrated by other outports, and the utilization rate of the bandwidths of the outport is high.

Description

A router and system-on-chip

technical field

The present application relates to the field of chip technology, and in particular, to a router and a system-on-chip.

Background technique

A network on chip (NoC) is a communication method for a system on chip (SoC). FIG. 1 shows a schematic diagram of an internal interconnection bus of an SoC. Among them, the on-chip network router (router) is responsible for realizing the routing of the host (master) accessing the slave (slave): the router decides to schedule the corresponding message to an output node (another router or a slave) according to the access address sent by each host. machine).

Since the host has different service access requirements for the slave, there are multiple virtual channels (virtual channels, VC) inside the router, and each virtual channel is used to transmit packets of one service type. FIG. 2 shows a possible schematic diagram of the internal structure of the router. The router consists of two parts: an input port (inport) and an output port (outport). The input port can be divided into multiple VCs (two VCs are divided as an example in Figure 2), and each VC corresponds to a first input first output (FIFO) VC queue. The VC queue will send an arbitration request to the corresponding output port according to the currently scheduled packets. In the case of receiving multiple arbitration requests, the arbitor in the output port arbitrates out a certain VC queue of a certain input port to schedule packets according to the priority of each input port that sends the arbitration request.

In some scenarios, bandwidth loss may occur when the above packet scheduling scheme is used. As shown in FIG. 3 , when the same VC (such as VC0) of input port 0 and input port 1 both needs to schedule packets to output port 0, both input port 0 and input port 1 send arbitration requests to output port 0. Assuming that input port 0 obtains arbitration, input port 0 can dispatch packets to output port 0, but input port 1 cannot temporarily dispatch packets to output port 0. Since the VC0 queue in input port 1 is a FIFO structure, if the message scheduled to output port 1 (referred to as message 0) is not sent out on input port 1, even if there is still a need for scheduling in the VC0 queue of input port 1 For packets sent to output port 1 (referred to as packet 1), input port 1 cannot apply to schedule packets to output port 1 due to the head of the VC0 queue. 1 is not used, so bandwidth is wasted.

To sum up, there is a problem of wasting bandwidth in the packet scheduling solution provided by the prior art.

SUMMARY OF THE INVENTION

Embodiments of the present application provide a router and a system-on-chip, which are used to improve bandwidth utilization when scheduling packets.

In a first aspect, an embodiment of the present application provides a router, where the router includes multiple input ports and multiple output ports. Wherein, each input port of the multiple input ports is used to send a request message to the corresponding output port of each message for the multiple input packets; each output port of the multiple output ports is used to respond to the received request The message is arbitrated, sending a grant message to the arbitrated input port.

With the above solution, when each input port sends a request message requesting scheduling packets to the output port, it can send requests to multiple output ports for multiple input packets, and the input port sends requests without any service type. limit. Compared with the prior art scheme in which each input port can only send one request message to one output port for one service type, in the case where the request message sent by the input port to an output port is not arbitrated, the input port The port can also obtain arbitration of other output ports, and the phenomenon of head blocking of the VC queue described in the prior art will not occur, so the bandwidth utilization rate of the output port is high.

In a possible design, each input port is configured with a shared buffer, the shared buffer stores packets of multiple service types, and each input port is used to simultaneously send multiple packets stored in the shared buffer to each report. The corresponding output port sends a request message, and after receiving the permission message sent by the corresponding output port, the corresponding message is scheduled and sent to the corresponding output port.

Among them, the shared cache can be implemented by random access memory RAM or registers.

With the above solution, in each input port, the packets received by the input port are stored in the shared buffer. The input port can simultaneously send request messages to the corresponding output ports for multiple messages in the shared buffer, so that multiple output ports can arbitrate the received request messages at the same time and determine the request message that has obtained the arbitration, which improves the message scheduling efficiency.

Further, in the shared cache, the storage space of the packets of the first service type may be set to be less than or equal to the first threshold value, and the storage space of the packets of the second service type may be set to be less than or equal to the second threshold value.

With the above solution, in the shared cache, different storage space upper limits (waterlines) are set for packets of different service types, so as to avoid the storage space of the shared cache being occupied by packets of some service types, resulting in packets of other service types. storage space is limited.

In addition, each input port can maintain multiple linked lists, each linked list corresponds to the corresponding service type supported by the input port and the corresponding output port, and each linked list is used to record the corresponding service type that needs to be scheduled to the corresponding output port 's message.

Further, each input port can respectively send a request message to the corresponding output port for each linked list in which the message to be scheduled is recorded, and after receiving the permission message sent by the corresponding output port, schedule the message recorded in the linked list and Send the message to the corresponding output port.

In a possible design, the router provided by the first aspect includes a first input port, a second input port, a first output port and a second output port.

The first input port may send a first request message to the first output port, where the first request message is used to instruct the first input port to request to schedule the first packet of the first service type to the first output port.

The second input port may send a second request message to the first output port, where the second request message is used to instruct the second input port to request to schedule a second packet of the first service type to the first output port; the second input port may also A third request message is sent to the second output port, where the third request message is used to instruct the second input port to request to schedule a third packet of the first service type to the second output port.

The first output port may arbitrate the first request message and the second request message, and send a first permission message to the first input port, where the first permission message is used to indicate that the first input port is allowed to schedule the first message to the first output port. arts.

The second output port may arbitrate the third request message, and send a second permission message to the second input port, where the second permission message is used to indicate that the second input port is allowed to schedule the third packet to the second output port.

With the above solution, if the second input port does not obtain the arbitration of the first output port, the second input port can also obtain the arbitration of the second output port. Compared with the prior art, with the above solution, since the input port can send request messages to the corresponding different output ports for messages that need to be scheduled to different output ports, the second input port schedules the second output port to the first output port. In the case where the request for the message does not obtain arbitration, the second input port can obtain the arbitration of the second output port and schedule the third message to the second output port, that is, the VC queue occurrence head mentioned in the prior art will not appear. Therefore, the bandwidth utilization rate of the output port is high.

In a possible design, the first input port may also send a first accept message to the first output port, where the first accept message is used to instruct the first input port to accept the permission of the first output port; schedule the first message to the first output port; sending a first release message to the first output port, where the first release message is used to indicate that the first packet scheduling is completed.

With the above solution, after the first input port sends the first permission message to the first output port, the transmission relationship between the first input port and the first output port can be locked. The first input port sends a first release message to the first output port after the first packet scheduling is completed, the first output port is released, and the transmission relationship between the first input port and the first output port is released. The first input port may resend the request message for the packet to be scheduled in the first shared buffer. The first output port can receive the request message sent by the input port again, and arbitrate the received request message to start the next packet scheduling.

In a possible design, the second input port may also send a second accept message to the second output port, where the second accept message is used to instruct the second input port to accept the permission of the second output port; schedule the third message to the second output port; sending a second release message to the second output port, where the second release message is used to indicate that the scheduling of the third packet is completed.

With the above solution, after the second input port sends the second permission message to the second output port, the transmission relationship between the second input port and the second output port can be locked. The second input port sends a second release message to the second output port after the third packet scheduling is completed, the second output port is released, and the transmission relationship between the second input port and the second output port is released. The second input port may resend the request message for the packet to be scheduled in the second shared buffer. The second output port can receive the request message sent by the input port again, and arbitrate the received request message to start the next packet scheduling.

In addition, the first input port may also send a fourth request message to the third output port among the plurality of output ports, where the fourth request message is used to instruct the first input port to request the third output port to schedule the first service type of the first service type to the third output port. Four packets; the third output port is used to arbitrate the fourth request message and send a third permission message to the first input port, and the third permission message is used to indicate that the first input port is allowed to schedule the fourth request message to the third output port. message.

With the above solution, both the first output port and the third output port decide that the first input port obtains the arbitration, and respectively send the first permission message and the third permission message to the first input port, and the first input port selects the first output port first. The port schedules the first message, and sends the first accept message to the first output port. Since the third output port does not receive the accept message, the third output port can subsequently receive requests from other input ports for scheduling other packets, thereby improving the bandwidth utilization of the output port.

In a possible design, the first input port may also send a fifth request message to the first output port, where the fifth request message is used to instruct the first input port to request to schedule a fifth message of the second service type to the first output port The first output port is further configured to arbitrate the fifth request message, and determine that the first input port is not allowed to schedule the fifth message.

In the above solution, the first input port sends request messages of two service types to the first output port at the same time, and the first output port arbitrates the two request messages (the first request message and the fifth request message) to determine the first request message. A request message obtains arbitration, and a fifth request message does not obtain arbitration.

In a possible design, the first input port may also send a sixth request message to the second output port, where the sixth request message is used to instruct the first input port to request to schedule a sixth request of the first service type to the second output port message; the second output port is further configured to arbitrate the sixth request message to determine that the first input port is not allowed to schedule the sixth message.

In the above solution, the first input port also requests to schedule the sixth message of the first service type to the second output port, and the second output port arbitrates the received third request message and the sixth request message, and determines to allow the third request message and the sixth request message. The second input port schedules the third message, and the first input port is not allowed to schedule the sixth message. Therefore, the second output port sends the second permission message to the second input port and does not send the permission message to the first input port.

In a second aspect, an embodiment of the present application further provides a system-on-chip, including a host, a slave, and the router provided in the first aspect and any possible design thereof, where the router is configured to route a packet sent by the host to the slave.

In addition, for the technical effects brought by any possible design manner in the second aspect, reference may be made to the technical effects brought about by different design manners in the first aspect, which will not be repeated here.

Description of drawings

1 is a schematic diagram of a SoC internal interconnection bus provided by the prior art;

Fig. 2 is a kind of internal structure schematic diagram of router provided by prior art;

3 is a schematic diagram of head blocking at an input port provided by the prior art;

4 is a schematic diagram of a shared cache provided by an embodiment of the present application;

FIG. 5 is a schematic structural diagram of a router according to an embodiment of the present application;

6 is a schematic diagram of an interaction flow between a first input port and a first output port according to an embodiment of the present application;

7 is a schematic diagram of an interaction flow between a second input port and a second output port according to an embodiment of the present application;

FIG. 8 is a schematic structural diagram of another router according to an embodiment of the present application;

FIG. 9 is a schematic structural diagram of a system-on-chip provided by an embodiment of the present application.

Detailed ways

In the following, the application scenarios of the embodiments of the present application are first introduced.

The embodiments of the present application can be applied to the system-on-chip shown in FIG. 1 . The system-on-chip includes a master, a router, and a slave. The router is responsible for implementing the route for the host to access the slave: the router decides to route the corresponding message to an output node (the next-level router or the slave) according to the access address sent by each host.

It should be noted that the SoC shown in FIG. 1 includes two levels of routing: router 0 and router 1 are located at the first level, and router 2 and router 3 are located at the second level. The first-level routing is used to schedule packets from the master to the second-level routing; the second-level routing is used to schedule packets from the first-level routing to the slave. In practical applications, the SoC may include one-level routing or multiple-level routing, and is not limited to the two-level routing shown in FIG. 1 . In addition, the number of masters, slaves and routers is not limited to that shown in FIG. 1 .

In the SoC shown in Figure 1, each router includes multiple input ports and multiple output ports. When an input port needs to dispatch packets to an output port, the input port sends an arbitration request to the output port, and the output port receives multiple arbitration requests according to the priority of each input port that sends the arbitration request , arbitrate out an input port for packet scheduling.

In the prior art, as shown in FIG. 2 , the input port includes multiple FIFO queues, and each FIFO queue is used to store a service type sent by an upper-level port (such as a host or an upper-level router) of the input port. message.

Different from the prior art, in this embodiment of the present application, the input port stores the input message (for example, the message sent by the upper level) in the shared buffer. The shared buffer can be used to store packets of various service types. The packets are not stored in the form of queues in the shared buffer. Therefore, the packets in the shared buffer are not restricted by the FIFO rule during scheduling.

Specifically, as shown in FIG. 4 , the shared buffer in each input port can be divided into a plurality of storage areas, including a plurality of first buffer areas and a second buffer area. Wherein, each first buffer area is used to store the message of a single service type sent by the upper level, and the second buffer area can store the message of multiple service types.

Among them, the function of configuring the first buffer area is to enable the input port to receive messages of any service type sent by the upper level, that is, to reserve a certain storage space (also referred to as pre-processing) for the messages of each service type. keep credit). For example, if there are N types of services that can schedule packets on the input port, N≥1, then N first buffer areas can be configured in the shared buffer, and each first buffer area is used to store packets of a single service type. In the case that the storage of the second buffer area is full, since the first buffer area is also configured for each service type, the input port can still receive the packets of any service type sent by the upper level.

It should be noted that, in the shared cache of the input port, the first cache area and the second cache area may be divided from the hardware level or the software level. The division method is not specifically limited. Assuming that random access memory (RAM) is used to instantiate the shared cache, then, in a possible example, part of the storage array in the RAM can be set to correspond to the first cache area, and another part of the storage array to correspond to the first cache area. Second cache area. In another possible example, part of the addressing space may be set to correspond to the first cache area, and another part of the addressing space may be set to correspond to the second cache area. In another possible example, the first cache area and the second cache area are dynamically changed, and which arrays or which addresses correspond to a certain cache area are not strictly specified, as long as the shared cache is used for each service type Reserve a storage space (with the same size as the first cache area); for example, if the input port receives the message 1 of service type 1 sent by the upper level and stores it in the shared cache, then the storage area of the message 1 It can be regarded as the first buffer area reserved for service type 1; the input port receives the message 2 of service type 1 sent by the upper level and stores it in the shared cache, then the storage area of the message 2 can be regarded as The first buffer area reserved for service type 1, and the storage area of packet 1 is regarded as the second buffer area.

In a shared cache, packets are not subject to the FIFO rule when scheduling. For example, the input port first receives the message 1 of service type 1 sent by the upper level and stores it in the shared buffer, and the message 1 needs to be dispatched to the output port 1; then, the input port receives the service type sent by the upper level. The message 2 of 1 is stored in the shared buffer, and the message 2 needs to be dispatched to the output port 2. The input port can send arbitration requests to output port 1 and output port 2 at the same time, and request to schedule message 1 and message 2 respectively. That is to say, for multiple packets that need to be dispatched to different output ports, the input port can send arbitration requests to corresponding different output ports for multiple packets. If the input port obtains arbitration at output port 1, the The input port can schedule the message 1 to the output port 1; if the input port obtains arbitration on the output port 2, the input port can schedule the corresponding message 2 to the output port 2; if the input port is on the output port 1 and the output port 2. If both output port 2 obtains arbitration, the input port can select either scheduling packet 1 or scheduling packet 2.

In addition, in this embodiment of the present application, there may be multiple manners for the division of service types. For example, packets that are sensitive to transmission delay are classified into one type of service, and packets that are not sensitive to transmission delay are classified into another type of service; for another example, packets that access host 0 are classified as one type of service. Service type, the packets accessing host 1 are classified into another service type. This embodiment of the present application does not specifically limit the division manner of service types.

The embodiments of the present application will be described in further detail below with reference to the accompanying drawings.

It should be noted that, in the embodiments of the present application, multiple refers to two or more. In addition, in the description of this application, words such as "first" and "second" are only used for the purpose of distinguishing and describing, and cannot be understood as indicating or implying relative importance, nor can they be understood as indicating or implying order.

Embodiments of the present application provide a router. The router includes multiple input ports and multiple output ports. Wherein, each input port is used to send a request message to the corresponding output port of each message for multiple input messages; each output port is used to arbitrate the received request message, and send permission to the input port that has obtained the arbitration information.

That is to say, when each input port sends a request message requesting a scheduling packet to an output port, it can send requests to multiple output ports for multiple input packets. The input port sends requests without any service type. limit. Compared with the prior art scheme in which each input port can only send one request message to one output port for one service type, in the case where the request message sent by the input port to an output port is not arbitrated, the input port The port can also obtain arbitration of other output ports, and the phenomenon of head blocking of the VC queue described in the prior art will not occur, so the bandwidth utilization rate of the output port is high.

In addition, each input port is configured with a shared buffer, and the shared buffer stores packets of various service types. Each input port is used to send multiple packets stored in the shared buffer to the corresponding output port of each packet at the same time. In the request message, after receiving the permission message sent by the corresponding output port, schedule the corresponding message and send the message to the corresponding output port.

That is, in each input port, the packets received by the input port are stored in the shared buffer. The input port can simultaneously send a request message to the corresponding output port for multiple packets in the shared buffer. For example, the input port may send a request message requesting to schedule packets of the first service type to the first output port, and also send a request message requesting to schedule packets of the first service type to the second output port.

Referring to FIG. 5 , in a specific example, the router 500 includes a first input port 501 , a second input port 502 , a first output port 503 and a second output port 504 . The first input port 501 may send a first request message to the first output port 503, where the first request message is used to instruct the first input port 501 to request the first output port 503 to schedule a first packet of the first service type; The second input port 502 may send a second request message to the first output port 503, where the second request message is used to instruct the second input port 502 to request to schedule a second packet of the first service type to the first output port 503; The input port 502 may also send a third request message to the second output port 504, where the third request message is used to instruct the second input port 502 to request to schedule a third packet of the first service type to the second output port 504; the first output The port 503 can arbitrate the first request message and the second request message, and send the first permission message to the first input port 501, where the first permission message is used to indicate that the first input port 501 is allowed to schedule the first output port 503. message; the second output port 504 can arbitrate the third request message, and send a second permission message to the second input port 502, the second permission message is used to indicate that the second input port 502 is allowed to schedule the second output port 504. Three messages.

In the router 500 provided in this embodiment of the present application, for a message of the first service type, the first input port 501 sends a first request message to the first output port 503 to request to schedule the first message; the second input port 502 The second request message and the third request message are sent to the first output port 503 and the second output port 504, respectively, to request scheduling of the second packet and the third packet. That is to say, in this embodiment of the present application, the input port may send request messages to the corresponding different output ports for packets that need to be dispatched to different output ports. The second output port 504 sends the second request message and the third request message.

After receiving the first request message and the second request message, the first output port 503 performs arbitration according to the priorities of the first input port 501 and the second input port 502, and determines that the first input port 501 obtains the arbitration and the second input port 502 obtains arbitration. Without obtaining arbitration, a transmission relationship may be established between the first input port 501 and the first output port 503 . Since the second input port 502 also sends a third request message to the second output port 504, after the second output port 504 performs arbitration, it is determined that the second input port 502 has obtained the arbitration, and there is no connection between the second input port 502 and the second output port 504. A transfer relationship can be established.

In the router 500 , if the second input port 502 does not obtain the arbitration of the first output port 503 , the second input port 502 can also obtain the arbitration of the second output port 504 . Compared with the prior art, in this embodiment of the present application, since the input port can send request messages to corresponding different output ports for packets that need to be dispatched to different output ports, the second input port 502 sends a request message to the first output port. 503 In the case where the request for scheduling the second packet has not been arbitrated, the second input port 502 can obtain the arbitration of the second output port 504 and schedule the third packet to the second output port 504, that is, the prior art does not occur. The phenomenon of head blocking occurs in the mentioned VC queue, so the bandwidth utilization rate of the output port is high.

It should be noted that, for convenience of description, FIG. 5 only shows that the router 500 includes a first input port 501 , a second input port 502 , a first output port 503 and a second output port 504 . In practical applications, the router 500 may include multiple input ports and multiple output ports, and the embodiments of the present application do not specifically limit the number of input ports and output ports.

In the router 500 provided in this embodiment of the present application, the input port may send request messages to corresponding different output ports respectively for packets that need to be scheduled to different output ports. In this embodiment of the present application, multiple packets to be scheduled in the first input port 501 may be stored in the first shared buffer, and the scheduling order of the packets stored in the first shared buffer is not limited by the FIFO rule. Similarly, multiple packets that need to be scheduled in the second input port 502 can be stored in the second shared buffer, and the scheduling order of the packets stored in the second shared buffer is also not limited by the FIFO rule.

Taking the second shared cache as an example, the second shared cache may be used to store packets of multiple service types. The second shared buffer may include a plurality of third buffer areas and a fourth buffer area, and each third buffer area in the plurality of third buffer areas is used to For messages of service types (ie, a certain storage space is reserved for messages of each service type), the fourth cache area is used to store messages of multiple service types.

Likewise, the first shared cache may include a plurality of first cache areas and a second cache area.

It should be understood that, in this embodiment of the present application, the router 500 may include multiple input ports, and each input port is configured with a shared cache. As mentioned above, in the shared cache, the packets are not stored in the form of queues, and the scheduling order of each packet is not restricted by the FIFO rule.

In addition, in the shared cache, different storage space upper limits (watermarks) can be set for packets of different service types, so as to avoid the storage space of the shared cache being occupied by packets of some service types, resulting in the loss of packets of other service types. Storage space is limited. The following description takes setting the upper limit of the storage space for the first shared cache as an example for description.

For example, in the first shared cache, the storage space of the packets of the first service type can be set to be less than or equal to the first threshold value, and the storage space of the packets of the second service type can be set to be less than or equal to the second threshold value. The first threshold value and the second threshold value may be the same or different.

For another example, in the second buffer area, the storage space of the packets of the first service type can be set to be less than or equal to the third threshold value, and the storage space of the packets of the second service type can be set to be less than or equal to the fourth threshold value. The third threshold value and the fourth threshold value may be the same or different.

Exemplarily, the first shared cache includes 16 storage spaces, and the router 500 supports two service types, service type 1 and service type 2. If 2 storage spaces are reserved for each service type, the second cache The area includes 12 (16-2*2=12) storage spaces. Then, the upper limit of the storage space of service type 1 in the second cache area can be set to 8, and the upper limit of storage space of service type 2 can be set to 8; 2 has a storage cap of 8.

It should be noted that, in this embodiment of the present application, an upper limit of storage space may be set for messages of each service type, or an upper limit of storage space may be set only for messages of some service types. For packets of different service types, the upper limit of the storage space can be the same or different.

In practical applications, the shared cache (such as the first shared cache or the second shared cache) configured in the input port may be implemented by RAM or registers, that is, RAM or registers may be used to instantiate the shared cache. When there are many types of services and a large amount of data in the input port scheduling message, compared with the solution of using registers to instantiate the shared cache, the use of RAM to instantiate the shared cache can reduce the loss of area and power consumption.

In addition, in the embodiment of the present application, in the shared buffer of the input port, different linked lists (states) may be maintained for packets of different service types that need to be scheduled to different output ports. Each linked list corresponds to the corresponding service type supported by the input port and the corresponding output port, and each linked list is used to record the messages of the corresponding service type that need to be scheduled to the corresponding output port.

Then, each input port can send a request message to the corresponding output port for each linked list in which the message to be scheduled is recorded, and after receiving the permission message sent by the corresponding output port, schedule the message recorded in the linked list and send it. The message is sent to the corresponding output port.

Specifically, in each linked list, a scheduling sequence of packets of a certain service type that needs to be scheduled to a certain output port can be maintained. The number of the maintained linked list for each input port is equal to the product of the number of service types supported by the input port and the number of output ports included in the router.

Illustratively, router 500 includes four input ports (input port 1, input port 2, input port 3, and input port 4) and four output ports (output port 1, output port 2, output port 3, and output port 4). ), assuming that each input port handles two service types, service type 1 and service type 2. Then, taking input port 1 as an example, assuming that 14 packets are stored in the shared buffer of input port 1, the service types and corresponding output ports of the 14 packets can be shown in Table 1 below.

Table 1

	业务类型1business type 1	业务类型2business type 2
输出端口1output port 1	报文1message 1	报文2、报文3Message 2, Message 3
输出端口2output port 2	报文4、报文5、报文6Message 4, Message 5, Message 6	报文7message 7
输出端口3output port 3	报文8、报文9Message 8, Message 9	报文10、报文11、报文12Packet 10, Packet 11, Packet 12
输出端口4output port 4	报文13Message 13	报文14Message 14

In the shared buffer of input port 1, 8 (4*2=8) linked lists can be maintained. Each linked list records the scheduling sequence of multiple packets of one service type that need to be scheduled to one output port. For example, in the linked list maintained for the packets of service type 2 that need to be scheduled to output port 1, the scheduling sequence of packet 2 and packet 3 is recorded, that is, packet 2 is scheduled first, and then packet 3; , in the linked list maintained for the packets of service type 1 that need to be scheduled to output port 2, the scheduling sequence of packet 4, packet 5 and packet 6 is recorded, that is, packet 4 is scheduled first, and then packet 5 is scheduled. , and finally schedule packet 6.

It should be noted that, in another possible example, the multiple linked lists maintained in the input port may also only record multiple packets of a certain service type that need to be scheduled to an output port, but for multiple The scheduling order of packets is not limited.

In addition, it should be noted that the specific content of the linked list maintained in the shared cache will be updated according to the reception and transmission of packets in the input port.

For example, after packet 1 has been scheduled, the content in the linked list of output port 1 - service type 1 in Table 1 is empty; after that, if input port 1 receives a message of service type 1 that needs to be scheduled to output port 1 again Packet 15, the content in the linked list of output port 1-service type 1 is modified to Packet 15.

For another example, after the message 10 has been scheduled, the content in the linked list of output port 3-service type 2 in Table 1 is modified to message 11 and message 12; after that, if the input port 1 receives it again, it needs to be scheduled to If packets 16 and 17 of service type 2 of port 3 are output, the contents in the linked list of output port 3 - service type 2 are modified to packet 11 , packet 12 , packet 16 , and packet 17 .

In the prior art, each input port maintains a first-in, first-out queue for messages of each service type. The packets that need to be dispatched to other output ports are blocked, resulting in other output ports not being used. However, in the embodiment of the present application, different linked lists are maintained in the shared cache for messages of different service types scheduled to different output ports. Then, the transmission of packets that need to be scheduled to different output ports does not affect each other, so that the bandwidth utilization rate of the output ports is improved.

In addition, in this embodiment of the present application, the first input port 501 may also: send a first accept message to the first output port 503, where the first accept message is used to instruct the first input port 501 to accept the permission of the first output port 503; The first packet is scheduled to the first output port 503; the first release message is sent to the first output port 503, where the first release message is used to indicate that the first packet is scheduled to be completed. Specifically, the interaction flow between the first input port 501 and the first output port 503 may be as shown in FIG. 6 .

After the first input port 501 sends the first permission message to the first output port 503, the transmission relationship between the first input port 501 and the first output port 503 is locked. The first input port 501 can schedule the first packet from the first shared buffer to the first output port 503; after the first packet scheduling is completed, the first input port 501 can send the first release to the first output port 503 message to indicate the completion of the first packet scheduling.

After the first release message is sent to the first output port 503, the first output port 503 is released, and the transmission relationship between the first input port 501 and the first output port 503 is released. The first input port 501 may resend the request message (ie, start the next handshake application) for the packet to be scheduled in the first shared buffer. The first output port 503 may receive the request message sent by the input port again, and perform arbitration on the received request message to start the next packet scheduling.

In the process of performing a message scheduling between the first input port 501 and the first output port 503, the message interaction between the two can be realized through four steps: 1. The first input port 501 sends the first output port 503 the first message; A request message; 2. The first output port 503 sends a first permission message to the first input port 501; 3. The first output port 503 sends a first accept message to the first input port 501; 4. The first output port 503 sends a first accept message to the first input port 501 The first input port 501 sends a first release message. In this embodiment of the present application, the above interaction process is referred to as a "four-step handshake mechanism", that is, a handshake between the first input port 501 and the first output port 503 is completed through the interaction of four messages.

In addition, in this embodiment of the present application, the second input port 502 may further: send a second accept message to the second output port 504, where the second accept message is used to instruct the second input port 502 to accept the permission of the second output port 504; The third packet is scheduled to the second output port 504; the second release message is sent to the second output port 504, where the second release message is used to indicate that the scheduling of the third packet is completed. Specifically, the interaction flow between the second input port 502 and the second output port 504 may be as shown in FIG. 7 .

After the second input port 502 sends the second permission message to the second output port 504, the transmission relationship between the second input port 502 and the second output port 504 is locked. The second input port 502 can schedule the third packet from the second shared buffer to the second output port 504 ; after the scheduling of the third packet is completed, the second input port 502 can send a second release to the second output port 504 message to indicate that the scheduling of the third packet is complete.

After the second release message is sent to the second output port 504, the second output port 504 is released, and the transmission relationship between the second input port 502 and the second output port 504 is released. The second input port 502 may resend the request message (ie, start the next handshake application) for the packet to be scheduled in the second shared buffer. The second output port 504 may receive the request message sent by the input port again, and perform arbitration on the received request message to start the next packet scheduling.

Similarly, the interaction process between the second input port 502 and the second output port 504 also adopts the aforementioned "four-step handshake mechanism", which will not be repeated here.

Specifically, in router 500, the interaction between input ports and output ports may be based on clock cycles. In each clock cycle, the input port can send multiple request messages for the packets stored in the shared buffer according to the linked list maintained in the shared buffer. Specifically, the number of request messages sent by the input port in a certain clock cycle is the same as the number of linked lists maintained in the shared cache.

As mentioned above, the input port can maintain different linked lists for messages of different service types scheduled to different output ports. For example, in the example of Table 1, in the current clock cycle, the shared buffer of input port 1 maintains a Eight linked lists, each of which records the scheduling sequence of multiple packets of one service type that need to be scheduled to one output port. Then, in the current clock cycle, the input port 1 can send eight request messages, and the eight request messages are: 1. A request message requesting to dispatch the message 1 of service type 1 to the output port 1; 2. A request to the output port 1 A request message for port 1 to schedule message 2 of service type 2; 3. A request message for requesting to schedule message 4 of service type 1 to output port 2; 4. A request for scheduling message 7 of service type 2 to output port 2 5. A request message requesting to dispatch the message 8 of service type 1 to output port 3; 6. A request message requesting to dispatch the message 10 of service type 2 to output port 3; 7. Requesting to dispatch the service type to output port 4 1. A request message for the message 13; 8. A request message for requesting to dispatch the message 14 of the service type 2 to the output port 4.

After the input port and the output port establish a transmission relationship through the aforementioned arbitration mechanism, the input port can schedule packets to the output port. Specifically, taking the second input port 502 and the second output port 504 as an example, after the second input port 502 sends the second accept message to the second output port 504, the second input port 502 may send the second accept message at the clock The cycle starts to schedule the third packet to the second output port 504 immediately, and the third packet may also start to be scheduled to the second output port 504 in the next clock cycle. The scheduling of the third packet may be completed within one clock cycle, or may be completed within multiple clock cycles, depending on the size of the third packet.

In addition, the update of the linked list maintained in the shared cache can also be done on a clock cycle basis. That is, in each clock cycle, the specific content of the linked list maintained in the shared cache will be updated according to the reception and transmission of packets in the input port in the previous clock cycle.

Further, after the linked list is updated, the input port may send a request message requesting a scheduling message according to the updated linked list. That is to say, in each clock cycle, the linked list in the shared cache is updated, and the input port sends request messages to corresponding different output ports according to the updated linked list in the current clock cycle.

In the previous example, the interaction process of packet transmission between the first input port 501 and the first output port 503 and the interaction of packet transmission between the second input port 502 and the second output port 504 are mainly introduced process. In practical applications, the router 500 may include more input ports and more output ports. Other scenarios are described below with a few specific examples.

Case 1

The router 500 further includes a third output port; the first input port 501 is further configured to send a fourth request message to the third output port, and the fourth request message is used to instruct the first input port 501 to request to schedule the first service to the third output port Type of fourth message; the third output port is used to arbitrate the fourth request message, and send the third permission message to the first input port 501, and the third permission message is used to indicate that the first input port 501 is allowed to output to the third The port schedules the fourth packet.

In case 1, the first input port 501 sends a request message for scheduling packets to both the first output port 503 and the third output port, and both the first output port 503 and the third output port rule that the first input port 501 obtains arbitration, And send the first permission message and the third permission message to the first input port 501 respectively, that is, the first input port 501 receives two permission messages. In this embodiment of the present application, the first input port 501 chooses to schedule the first message to the first output port 503 first, then the first input port 501 sends the first accept message to the first output port 503, but does not send the first message to the first output port 503. Three output ports send accept messages.

Since the third output port does not receive the accept message, the third output port can subsequently receive requests from other input ports for scheduling other packets, thereby improving the bandwidth utilization of the output port.

Case 2

The first input port 501 is further configured to send a fifth request message to the first output port 503, where the fifth request message is used to instruct the first input port 501 to request the first output port 503 to schedule a fifth message of the second service type; The first output port 503 is further configured to arbitrate the fifth request message, and determine that the first input port 501 is not allowed to schedule the fifth message.

Wherein, the second service type and the first service type are different service types.

That is to say, the first input port 501 can send request messages of two service types to the first output port 503 at the same time, and the first output port 503 arbitrates the two request messages (the first request message and the fifth request message) , it is determined that the first request message has been arbitrated and the fifth request message has not been arbitrated, that is, it is determined that the first input port 501 is allowed to schedule messages of the first service type, and the first input port 501 is not allowed to schedule messages of the second service type.

Case three

The first input port 501 is further configured to: send a sixth request message to the second output port 504, where the sixth request message is used to instruct the first input port 501 to request to the second output port 504 to schedule a sixth packet of the first service type ; The second output port 504 is further configured to arbitrate the sixth request message, and determine that the first input port 501 is not allowed to schedule the sixth message.

That is to say, the first input port 501 also requests to schedule the sixth message of the first service type to the second output port 504, and the second output port 504 arbitrates the received third request message and the sixth request message, and determines The second input port 502 is allowed to schedule the third message, and the first input port 501 is not allowed to schedule the sixth message. Therefore, the second output port 504 sends the second permission message to the second input port 502 and does not send the permission message to the first input port 501 .

In an example, using the router 500 provided in this embodiment of the present application, for a packet of the first service type, the first input port 501 sends a first request message to the first output port 503 to request scheduling of the first packet; The second input port 502 sends a second request message and a third request message to the first output port 503 and the second output port 504 respectively, so as to request to schedule the second packet and the third packet. After receiving the first request message and the second request message, the first output port 503 performs arbitration according to the priorities of the first input port 501 and the second input port 502, and determines that the first input port 501 obtains the arbitration and the second input port 502 obtains arbitration. Without obtaining arbitration, a transmission relationship is established between the first input port 501 and the first output port 503 . Since the second input port 502 also sends a third request message to the second output port 504, after the second output port 504 performs arbitration, it is determined that the second input port 502 has obtained the arbitration, and the second input port 502 and the second output port 504 can also be established. transfer relationship.

In the router 500, when the second input port 502 does not obtain arbitration from the first output port 503, because the second input port 502 also sends a third packet to the second output port 504 for the third packet of the first service type. request message, so that the second input port 502 can also obtain arbitration for the second output port 504 . Compared with the prior art, in the router 500, since the input port can send request messages to the corresponding different output ports for packets that need to be scheduled to different output ports, the second input port 502 schedules the first output port 503. If the request for the second packet has not been arbitrated, the second input port 502 can obtain arbitration from the second output port 504, and the second input port 502 can schedule the third packet to the second output port 504, that is, no The phenomenon of head blocking occurs in the VC queue mentioned in the prior art, so the bandwidth utilization rate of the output port is high.

An embodiment of the present application further provides a router, and the router may be as shown in FIG. 8 . The router includes input port 0, input port 1, output port 0, and output port 1. It can be understood that, for convenience of description, only two input ports and two output ports are shown here, but there may be more input ports and more output ports, depending on actual needs and actual application scenarios. There are two service types to be routed by the router, service type 0 (VC0) and service type 1 (VC1). It can be understood that the business types here are not limited to two, and there can be more, depending on the actual application scenario.

A shared buffer is configured in each input port. The shared buffer includes a VC0 reserved buffer (VC0credit), a VC1 reserved buffer (VC1credit), and a common buffer (common buffer). The VC0 reserved buffer and the VC1 reserved buffer in input port 0 can be regarded as a specific example of the aforementioned first buffer areas, and the general buffer in input port 0 can be regarded as a specific example of the aforementioned second buffer area . The VC0 reserved buffer and the VC1 reserved buffer in input port 1 can be regarded as a specific example of the foregoing multiple third buffer areas, and the general buffer in input port 1 can be regarded as a specific example of the foregoing fourth buffer area. The shared cache maintains different linked lists (states) for different service types and output ports, that is, four linked lists of VC0-output port 0, VC1-output port 0, VC0-output port 1 and VC1-output port 1 are maintained. The input port can schedule the packets in the shared buffer according to the arbitration result of the output port and the linked list maintained by itself.

Each output port is configured with an arbiter, which is used to arbitrate the request messages sent by the input port for scheduling packets of different service types. Specifically, each output port includes arbiter 0, arbiter 1 and arbiter 2. Arbiter 0 is used to arbitrate requests for scheduling packets of service type 0, arbiter 1 is used to arbitrate requests for scheduling packets of service type 1; Arbitration results are arbitrated to finally determine which request gets the arbitration. For example, arbiter 0 in output port 0 receives request 1 and request 2 of service type 0, and judges that request 1 obtains arbitration; arbiter 1 receives request 3 and request 4 of service type 1, judges that request 4 obtains arbitration; Arbiter 2 arbitrates the arbitration results of Arbiter 0 and Arbiter 1, and finally determines that Request 4 is arbitrated. Then, output port 0 determines that request 4 is arbitrated, and sends a permission message to the input port that sent request 4.

Wherein, in this embodiment of the present application, the interaction process between the input port and the output port may be as follows.

1. Input port 0 sends REQ to two output ports according to the scheduling requirement of each VC (a specific example of the foregoing request message). Specifically, input port 0 sends REQ1 for scheduling VC0 type packets and REQ2 for scheduling VC1 type packets to output port 0; input port 0 sends REQ3 for scheduling VC0 type packets and REQ4 for scheduling VC1 type packets to output port 1 .

At the same time, input port 1 sends REQ to the two output ports according to the scheduling requirements of each VC. Specifically, input port 1 sends REQ5 for scheduling VC0 type packets and REQ6 for scheduling VC1 type packets to output port 0; input port 1 sends REQ7 for scheduling VC0 type packets and REQ8 for scheduling VC1 type packets to output port 1 .

2. After the arbiter 0 in the output port 0 receives REQ1 and REQ5, it arbitrates according to the priority and other conditions, and determines that REQ1 obtains the arbitration; after the arbiter 1 in the output port 0 receives REQ2 and REQ6, according to the priority and other conditions Arbitration is performed to determine that REQ2 has obtained the arbitration; arbiter 2 in output port 2 arbitrates the arbitration result of arbiter 0 (REQ1) and the arbitration result of arbiter 1 (REQ2), determines that REQ1 has obtained arbitration, and then returns to input port 0 GRANT (a specific example of the aforementioned first grant message), instructs REQ1 to obtain arbitration.

At the same time, after receiving REQ3 and REQ7, the arbiter 0 in output port 1 conducts arbitration according to the priority and other conditions, and determines that REQ7 wins the arbitration; after receiving REQ4 and REQ8, the arbiter 1 in the output port 1 performs arbitration according to the priority and other conditions. Arbitration is performed to determine that REQ8 has obtained the arbitration; arbiter 2 in output port 1 arbitrates the arbitration result of arbiter 0 (REQ7) and the arbitration result of arbiter 1 (REQ8), determines that REQ7 has obtained arbitration, and then returns to input port 1 GRANT (a specific example of the aforementioned second grant message), instructing REQ7 to obtain arbitration.

3. After input port 0 receives GRANT, it returns ACCEPT (a specific example of the aforementioned first acceptance message) to output port 0; after input port 1 receives GRANT, it returns ACCEPT (one of the aforementioned second acceptance messages) to output port 1. specific example).

4. After input port 0 returns ACCEPT to output port 0, the transmission relationship between input port 0 and output port 0 is locked. At this time, the input port 0 starts to schedule the VC0 type message to the output port 1. After sending the message that needs to be scheduled, it returns the SWITCH signal (a specific example of the aforementioned first release message) to release the arbitration lock of the output port 0. , to start the next handshake application.

After input port 1 returns ACCEPT to output port 1, the transmission relationship between input port 1 and output port 1 is locked. At this time, the input port 1 starts to schedule the VC0 type message to the output port 1. After sending the message that needs to be scheduled, it returns the SWITCH signal (a specific example of the aforementioned second release message) to release the arbitration lock of the output port 1. , to start the next handshake application.

In the router shown in Figure 8, input port 0 dispatches REQ1 of VC0 type packets to output port 0 to obtain arbitration, input port 0 and output port 0 establish a transmission relationship, and input port 1 dispatches VC0 type packets to output port 0 REQ5 did not receive arbitration. Since input port 1 also sends REQ7 to output port 1 to request scheduling of VC0 type messages, REQ7 can obtain arbitration for output port 1. Therefore, the scheduling of VC0 type packets in input port 1 will not be blocked, that is, input port 1 can establish a transmission relationship with output port 1, and schedule VC0 type packets to output port 1. Therefore, using the router shown in Figure 8 can improve the bandwidth utilization of the output port.

It should be noted that, for the implementation manners and technical effects that are not described in detail in the router shown in FIG. 8 , reference may be made to the relevant descriptions in the router 500 , which will not be repeated here.

Based on the same inventive concept, an embodiment of the present application also provides a system-on-chip. Referring to FIG. 9 , the system-on-a-chip 900 includes a host 901 , a slave 902 and the aforementioned router 500 .

The router 500 is used to route the message sent by the master 901 to the slave 902 .

Obviously, those skilled in the art can make various changes and modifications to the embodiments of the present application without departing from the scope of the embodiments of the present application. Thus, if these modifications and variations of the embodiments of the present application fall within the scope of the claims of the present application and their equivalents, the present application is also intended to include these modifications and variations.

Claims

A router, characterized in that it includes:

a plurality of input ports, each input port of the plurality of input ports is used for sending a request message to the output port corresponding to each message for the plurality of inputted messages;

A plurality of output ports, each output port of the plurality of output ports is used to arbitrate the received request message, and send a permission message to the input port that has obtained the arbitration.
The router according to claim 1, wherein each input port is configured with a shared cache, the shared cache stores packets of multiple service types, and each input port is used for multiple data stored in the shared cache. Each message sends a request message to the corresponding output port of each message at the same time. After receiving the permission message sent by the corresponding output port, the corresponding message is scheduled and sent to the corresponding output port.
The router according to claim 1 or 2, wherein each input port maintains a plurality of linked lists, each linked list corresponds to a corresponding service type supported by the input port and a corresponding output port, each The linked list is used to record the packets of the corresponding service type that need to be scheduled to the corresponding output port.
The router according to claim 3, wherein each input port is used to send a request message to a corresponding output port for each linked list in which the message to be scheduled is recorded, and after receiving the permission message sent by the corresponding output port After that, schedule the message recorded in the linked list and send the message to the corresponding output port.
The router according to any one of claims 1 to 4, wherein the first input port among the plurality of input ports is used to send a first request message to the first output port, the first request message The message is used to indicate that the first input port requests to schedule a first packet of the first service type to the first output port;

a second input port among the plurality of input ports, configured to send a second request message to the first output port, where the second request message is used to instruct the second input port to request the first output port The port schedules the second packet of the first service type; sends a third request message to the second output port, where the third request message is used to instruct the second input port to request the second output port to schedule all describe the third message of the first service type;

a first output port among the plurality of output ports, configured to arbitrate the first request message and the second request message, and send a first permission message to the first input port, the first permission message The message is used to indicate that the first input port is allowed to schedule the first packet to the first output port;

a second output port among the plurality of output ports, configured to arbitrate the third request message, and send a second permission message to the second input port, where the second permission message is used to indicate that the The second input port schedules the third packet to the second output port.
The router of claim 5, wherein the first input port is further used for:

sending a first accept message to the first output port, where the first accept message is used to instruct the first input port to accept the permission of the first output port;

scheduling the first message to the first output port;

A first release message is sent to the first output port, where the first release message is used to indicate that the first packet scheduling is completed.
The router according to claim 5 or 6, wherein the first input port is further used for:

Send a fourth request message to a third output port of the plurality of output ports, where the fourth request message is used to instruct the first input port to request scheduling of the first service type to the third output port the fourth message;

The third output port is configured to arbitrate the fourth request message, and send a third permission message to the first input port, where the third permission message is used to indicate that the first input port is allowed to send the request to the first input port. The third output port schedules the fourth packet.
The router according to any one of claims 5 to 7, wherein the second input port is further used for:

sending a second accept message to the second output port, where the second accept message is used to instruct the second input port to accept the permission of the second output port;

scheduling the third message to the second output port;

A second release message is sent to the second output port, where the second release message is used to indicate that the scheduling of the third packet is completed.
The router according to any one of claims 5 to 8, wherein the first input port is further used for:

sending a fifth request message to the first output port, where the fifth request message is used to instruct the first input port to request to schedule a fifth packet of the second service type to the first output port;

The first output port is further configured to arbitrate the fifth request message, and determine that the first input port is not allowed to schedule the fifth message.
The router according to any one of claims 5 to 9, wherein the first input port is further used for:

sending a sixth request message to the second output port, where the sixth request message is used to instruct the first input port to request to schedule a sixth packet of the first service type to the second output port;

The second output port is further configured to arbitrate the sixth request message, and determine that the first input port is not allowed to schedule the sixth message.
The router according to any one of claims 2 to 10, wherein, in the shared cache, the storage space of the packets of the first service type is less than or equal to a first threshold value, and the second service The storage space of the type of message is less than or equal to the second threshold value.
The router according to any one of claims 2 to 11, wherein the shared cache is implemented by a random access memory RAM or a register.
A system-on-a-chip, comprising a host, a slave, and the router according to any one of claims 1 to 12, wherein the router is configured to route a packet sent by the host to the slave.