Techniques for Coalescing Doorbells in a Request Message

• US 20180181530A1
• Filed: 12/28/2016
• Published: 06/28/2018
• Est. Priority Date: 12/28/2016
• Status: Active Grant

• Current Assignee: Intel Corporation
• Original Assignee: Intel Corporation
• Inventors: Kantecki, Tomasz, Friedman, Ben-Zion, McDonnell, Niall D., Richardson, Bruce
• Granted Patent: US 10,884,970 B2
• Time in Patent Office: Days
• CPC Class Codes: G06F 12/0875 with dedicated cache, e.g. instruction or stack
• CPC Class Codes: G06F 13/126 and has means for transferring I/O instructions and statuses between control unit and main processor
• CPC Class Codes: G06F 13/4022 using switching circuits, e.g. switching matrix, connection or expansion network G06F13/4009 takes precedence
• CPC Class Codes: G06F 13/4282 on a serial bus, e.g. I2C bus, SPI bus on daisy chain buses G06F13/4247
• CPC Class Codes: G06F 2212/452 Instruction code
• CPC Class Codes: G06F 2213/0026 PCI express

Abstract

Examples include techniques for coalescing doorbells in a request message. Example techniques include gathering doorbells to access a device. The gathered are combined in a cache line structure and the cache line structure is written to a cache or buffer for a central processing unit in a single write operation.

Claims

1. An apparatus comprising:

circuitry for a computing platform, the circuitry to include a central processing unit (CPU) cache;

doorbell logic for execution by the circuitry to generate a plurality of doorbells targeted to a respective plurality of memory rings or queues for a device on the computing platform communicatively coupled with the circuitry via a link arranged to operate according to a peripheral component interconnect express (PCIe) specification;

combine logic for execution by the circuitry to combine the plurality of doorbells in a cache line structure that includes separate information for each doorbell; and

write logic for execution by the circuitry to write the cache line structure to the CPU cache in a single write operation to cause the plurality of doorbells to be posted to the respective plurality of memory rings or queues.

2. The apparatus of claim 1, comprising:

notify logic for execution by the circuitry to cause a write PCIe transaction layer packet (TLP) request message to be generated and sent to the device, the write PCIe TLP request message to include the separate information for each doorbell.

3. The apparatus of claim 1, comprising the separate information for each doorbell to include respective identifier information for each respective memory ring or queue to uniquely identify each respective memory ring and to include a tail pointer to indicate respective address ranges for each respective memory ring or queue having valid requests for the device.

4. The apparatus of claim 1, the single write operation comprises the write logic to write the cache line structure to the CPU cache via use of write combine memory.

5. The apparatus of claim 1, the single write operation comprises the write logic to write the cache line structure to the CPU cache via use of write back memory.

6. The apparatus of claim 1, comprising the cache line structure to have a size of 64 bytes and the separate information for each doorbell has a size of 8 bytes.

7. The apparatus of claim 1, the device comprises a network interface card (NIC) having switch capabilities, a cryptographic accelerator or a non-volatile memory controller.

8. The apparatus of claim 1, comprising a digital display coupled to the circuitry to present a user interface view.

9. A method comprising:

generating a plurality of doorbells targeted to a respective plurality of memory rings or queues for a device on a computing platform operating according to a peripheral component interconnect express (PCIe) specification;

combining the plurality of doorbells in a cache line structure that includes separate information for each doorbell; and

writing the cache line structure to a cache or buffer for a central processing unit (CPU) of the computing platform in a single write operation to cause the plurality of doorbells to be posted to the respective plurality of memory rings or queues.

10. The method of claim 9, comprising:

causing a write PCIe transaction layer packet (TLP) request message to be generated and sent to the device, the write PCIe TLP request message including the separate information for each doorbell.

11. The method of claim 9, comprising the separate information for each doorbell including respective identifier information for each respective memory ring or queue to uniquely identify each respective memory ring and including a tail pointer to indicate respective address ranges for each respective memory ring or queue having valid requests for the device.

12. The method of claim 9, the single write operation comprises writing the cache line structure to the cache or buffer using write combine memory.

13. The method of claim 9, the single write operation comprises writing the cache line structure to the cache or buffer using write back memory.

14. The method of claim 9, comprising the cache line structure having a size of 64 bytes and the separate information for each doorbell has a size of 8 bytes.

15. The method of claim 9, the device comprises a network interface card (NIC) having switch capabilities, a cryptographic accelerator or a non-volatile memory controller.

16. At least one machine readable medium comprising a plurality of instructions that in response to being executed by a system at a computing platform cause the system to:

generate a plurality of doorbells targeted to a respective plurality of memory rings or queues for a device on the computing platform, the device to operate according to a peripheral component interconnect express (PCIe) specification;

combine the plurality of doorbells in a cache line structure that includes separate information for each doorbell; and

write the cache line structure to a cache or buffer for a central processing unit (CPU) of the computing platform in a single write operation to cause the plurality of doorbells to be posted to the respective plurality of memory rings or queues.

17. The at least one machine readable medium of claim 16, comprising the instructions to further cause the system to:

cause a write PCIe transaction layer packet (TLP) request message to be generated and sent to the device, the write PCIe TLP request message to include the separate information for each doorbell.

18. The at least one machine readable medium of claim 16, comprising the separate information for each doorbell to include respective identifier information for each respective memory ring or queue to uniquely identify each respective memory ring and to include a tail pointer to indicate respective address ranges for each respective memory ring or queue having valid requests for the device.

19. The at least one machine readable medium of claim 16, the single write operation comprises the system to write the cache line structure to the cache or buffer via use of write combine memory.

20. The at least one machine readable medium of claim 16, the single write operation comprises the system to write the cache line structure to the cache or buffer via use of write back memory.

21. The at least one machine readable medium of claim 16, comprising the cache line structure to have a size of 64 bytes and the separate information for each doorbell has a size of 8 bytes.

22. The at least one machine readable medium of claim 16, the device comprises a network interface card (NIC) having switch capabilities, a cryptographic accelerator or a non-volatile memory controller.