Packet protocol and distributed burst engine

• US 5,978,858 A
• Filed: 09/30/1997
• Issued: 11/02/1999
• Est. Priority Date: 09/30/1997
• Status: Expired due to Term

• Current Assignee: Hewlett-Packard Development Company, L.P. (HP Inc.)
• Original Assignee: Compaq Computer Corporation (HP Inc.)
• Inventors: Medina, Michael P., Bonola, Thomas J., Moriarty, Michael P.
• Primary Examiner(s): Ngo, Chuong Dinh
• Assistant Examiner(s): Mai, Rijue
• Application Number:
  US08/940,911
• Time in Patent Office: 763 Days
• Field of Search: 395/800, 395/821, 395/824, 395/200.03, 395/280, 395/200.2, 395/825, 395/400, 395/685, 395/710, 395/872, 370/60, 370/396
• US Class Current: 710/1
• CPC Class Codes: G06F 12/1081 for peripheral access to main memory, e.g. direct memory access [DMA]
• CPC Class Codes: H04L 12/40123 Interconnection of computers and peripherals printer information exchange with computer G06F3/1293

Abstract

A method and apparatus for decoupling input/output (I/O) from host processing through main memory. A command packet architecture and distributed burst engine for communicating data to an I/O device without using memory mapped I/O or host processor synchronization. The packet architecture includes a header having fields for linking packets in a list with physical and virtual addresses, thereby eliminating address translations. The distributed burst engine includes buffers and controllers for bursting the linked lists of packets between main memory and the I/O device. Doorbell registers are included for the host processor to indicate to the DBE that an event has occurred. The distributed burst engine is versatile enough to be bus independent and located virtually anywhere between main memory and the I/O device.

Claims

1. An apparatus for communicating with a host processor and a device, the host having a memory for storing at least one pool of packets, each pool comprised of a linked list of packets, each packet having a header including a physical address field for containing a physical address of a next packet to form the linked list, the apparatus comprising:

a request head register operable to hold the physical address of a next packet of the linked list;

a free queue register operable to hold the physical address of the last packet of the linked list;

a request queue for holding packets, said request queue operable to provide a fullness indication when a packet is therein;

a request doorbell register operable to receive an indication from the host that a next packet is available to process;

a completion head register operable to hold the physical address of a completion list;

a completion queue for holding packets;

a completion doorbell register operable to receive an indication from the host that said completion queue has been serviced and that the host is ready to receive packets;

a first front end controller coupled to the request queue and responsive to said request doorbell register, said first front end controller operable to read a packet from memory according to the physical address stored in the request head register when said request doorbell register is rung and store the packet in said request queue;

a first back end controller coupled to the request queue and the completion queue and responsive to the request queue fullness indication, said first back end controller operable to read a packet from said request queue when said request queue is not empty and provide the packet to the device, said first back end controller further operable to provide the packet to said completion queue when the device has completed processing the packet; and

a completion controller coupled to said completion queue and responsive to the completion doorbell register and the completion head register, said completion controller operable to complete a packet from the completion queue back to the memory.

2. The apparatus of claim 1, further comprising:

a posted head register operable to hold the physical address of a next packet of the linked list;

a posted queue for holding packets available for use by the device for unsolicited requests;

a posted doorbell register operable to receive an indication from the host that a next packet is available in memory for the device;

a second front end controller coupled to said posted queue and responsive to said posted doorbell register, said second front end controller operable to read posted packets from memory and write the packets into said posted queue;

a second back end controller coupled to the posted queue and the completion queue and responsive to the device, said second back end controller operable to acquire a packet from said posted queue when said posted queue is not empty and provide said packet to the device, said second back end controller further operable to provide said packet to said completion queue when the device has completed processing said packet.

3. The apparatus of claim 2, wherein said second back end controller is further operable to acquire packets from memory in response to the host ringing said posted doorbell register.

4. The apparatus of claim 1, further comprising:

a completion tail register operable to hold the physical address of a last packet of the completion list,

wherein each packet includes a virtual address field for containing a virtual address of a next packet to form the linked list, and wherein if said completion doorbell register is set, said completion controller is further operable to clear said completion doorbell register.

5. The apparatus of claim 4, wherein if said completion head register is null said completion controller is operable to write the physical address of the completed packet in said completion tail register and to write the virtual address of the completed packet in said completion head register.

6. The apparatus of claim 4, wherein if said completion head register is not null said completion controller is operable to write the virtual address of the completed packet in the physical address field of the packet indicated by the completion tail register and to write the physical address of the completed packet in said completion tail register.

8. The apparatus of claim 4, wherein said completion controller is further operable to clear write the physical address of the completed packet into the completion head register.

9. The apparatus of claim 1, wherein each packet includes a virtual address field for containing a virtual address of a next packet to form the linked list, and wherein if said completion doorbell register is not set, said completion controller is further operable to write the virtual address of the completed packet in the packet indicated by the completion head register.

10. The apparatus of claim 1, wherein each packet includes a virtual address field for containing a virtual address of a next packet to form the linked list, wherein the packets are characterized as being asynchronous, polled and interrupt packets, and wherein if the completed packet is an asynchronous packet the completion controller is further operable to write the virtual address of the completed packet in the packet indicated by the free queue register.

11. The apparatus of claim 1, wherein after said first front end controller reads the next packet from memory said first front end controller is operable to replace the physical address stored in the request head register with the physical address contained in the header of the next packet.

12. The apparatus of claim 11 wherein said first front end controller is operable to continue reading packets from memory until the physical address contained in the header of the next packet is null.

13. The apparatus of claim 1, wherein the packet header includes a bit for indicating direct access, and wherein if a packet having the direct access bit set is received by said back end controller, said back end controller is operable to stop processing packets until a special command is received from the host.

14. The apparatus of claim 1, wherein the device is capable of reading data directly from memory, and

wherein the payload provided to the device includes an address range of data in memory for the device to read.

15. The apparatus of claim 1, wherein the device is capable of reading data, and wherein the payload provided to the device includes data.

16. The apparatus of claim 1, wherein the host includes a cache having a certain cacheline, and wherein the payload is aligned to a cacheline.

17. The apparatus of claim 1, wherein the doorbell registers are memory mapped.

7. The apparatus of claim 7, wherein the packets are characterized as being asynchronous, polled and interrupt packets, and wherein if the completed packet is an interrupt packet said completion controller is further operable to cause an interrupt to the host after clearing said completion doorbell register.

18. A computer system having automatic direct memory access, the computer system comprising:

a main memory for storing at least one pool of packets, each pool comprised of a linked list of packets, each packet having a header including a physical address field for containing a physical address of a next packet to form the linked list;

a processor coupled to said main memory;

a mass storage subsystem;

a distributed burst engine coupled between said mass storage subsystem and said main memory, the distributed burst engine comprising;

a request head register operable to hold the physical address of a next packet of the linked list;

a free queue register operable to hold the physical address of the last packet of the linked list;

a request queue for holding packets, said request queue operable to provide a fullness indication when a packet is therein;

a request doorbell register operable to receive an indication from the host that a next packet is available to process;

a completion head register operable to hold the physical address of a completion list;

a completion queue for holding packets;

a completion doorbell register operable to receive an indication from the processor that said completion queue has been serviced and that the processor is ready to receive packets;

a first front end controller coupled to the request queue and responsive to said request doorbell register, said first front end controller operable to read a packet from memory according to the physical address stored in the request head register when said request doorbell register is rung and store the packet in said request queue;

a first back end controller coupled to the request queue and the completion queue and responsive to the request queue fullness indication, said first back end controller operable to read a packet from said request queue when said request queue is not empty and provide the packet to the mass storage subsystem, said first back end controller further operable to provide the packet to said completion queue when the mass storage subsystem has completed processing the packet; and

a completion controller coupled to said completion queue and responsive to the completion doorbell register and the completion head register, said completion controller operable to complete a packet from the completion queue back to the memory.

19. The computer system of claim 18, wherein the distributed burst engine further comprises:

a posted head register operable to hold the physical address of a next packet of the linked list;

a posted queue for holding packets available for use by the mass storage subsystem for unsolicited requests;

a posted doorbell register operable to receive an indication from the processor that a next packet is available in memory for the mass storage subsystem;

a second front end controller coupled to said posted queue and responsive to said posted doorbell register, said second front end controller operable to read posted packets from memory and write the packets into said posted queue;

a second back end controller coupled to the posted queue and the completion queue and responsive to the mass storage subsystem, said second back end controller operable to acquire a packet from said posted queue when said posted queue is not empty and provide said packet to the mass storage subsystem, said second back end controller further operable to provide said packet to said completion queue when the mass storage subsystem has completed processing said packet.

20. The computer system of claim 19, wherein said second back end controller is further operable to acquire packets from memory in response to the processor ringing said posted doorbell register.

21. The computer system of claim 18, wherein the distributed burst engine further comprises:

a completion tail register operable to hold the physical address of a last packet of the completion list,

wherein each packet includes a virtual address field for containing a virtual address of a next packet to form the linked list, and wherein if said completion doorbell register is set, said completion controller is further operable to clear said completion doorbell register.

22. The computer system of claim 21, wherein if said completion head register is null said completion controller is operable to write the physical address of the completed packet in said completion tail register and to write the virtual address of the completed packet in said completion head register.

23. The computer system of claim 21, wherein if said completion head register is not null said completion controller is operable to write the virtual address of the completed packet in the physical address field of the packet indicated by the completion tail register and to write the physical address of the completed packet in said completion tail register.

24. The computer system of claim 21, wherein the packets are characterized as being asynchronous, polled and interrupt packets, and wherein if the completed packet is an interrupt packet said completion controller is further operable to cause an interrupt to the processor after clearing said completion doorbell register.

25. The computer system of claim 21, wherein said completion controller is further operable to clear write the physical address of the completed packet into the completion head register.

26. The computer system of claim 18, wherein each packet includes a virtual address field for containing a virtual address of a next packet to form the linked list, and wherein if said completion doorbell register is not set, said completion controller is further operable to write the virtual address of the completed packet in the packet indicated by the completion head register.

27. The computer system of claim 18, wherein each packet includes a virtual address field for containing a virtual address of a next packet to form the linked list, wherein the packets are characterized as being asynchronous, polled and interrupt packets, and wherein if the completed packet is an asynchronous packet the completion controller is further operable to write the virtual address of the completed packet in the packet indicated by the free queue register.

28. The computer system of claim 18, wherein after said first front end controller reads the next packet from memory said first front end controller is operable to replace the physical address stored in the request head register with the physical address contained in the header of the next packet.

29. The computer system of claim 28 wherein said first front end controller is operable to continue reading packets from memory until the physical address contained in the header of the next packet is null.

30. The computer system of claim 18, wherein the packet header includes a bit for indicating direct access, and wherein if a packet having the direct access bit set is received by said back end controller, said back end controller is operable to stop processing packets until a special command is received from the processor.

31. The computer system of claim 18, wherein the mass storage subsystem is capable of reading data directly from memory, and

wherein the payload provided to the mass storage subsystem includes an address range of data in memory for the mass storage subsystem to read.

32. The computer system of claim 18, wherein the mass storage subsystem is capable of reading data, and wherein the payload provided to the mass storage subsystem includes data.

33. The computer system of claim 18, wherein the processor includes a cache having a certain cacheline, and wherein the payload is aligned to a cacheline.

34. The computer system of claim 18, wherein the doorbell registers are memory mapped.