Error handling in an interconnect
A system level error detection and handling of the network IO in a multi-chip-package (MCP) die is provided. The error detection and handling mechanism conceived may be used between a system-on-chip (SoC) die and a different type of die, such as a die manufactured by a third-party (e.g., a high-bandwidth network IO die). To provide a timely indication in case of any part of the network is at fault, a control unit on the SoC die handles error detection on the network IO links using various indicators. After errors are detected, the control unit groups the errors into two categories: a link failure and a virtual channel failure. Such an error handling mechanism may consolidate the actions and provide consistency in hardware behavior.
1. A tangible, non-transitory, computer-readable medium having instructions stored thereon, wherein the instructions are configured to cause a processor to:
receive an indication of an error for a link between a plurality of dies of a multi-die system, wherein the link is to enable network communications between the plurality of dies;
determine whether the error is a network error associated with the network communications;
when the error is not a network error, logically switch the link off; and
send a not-acknowledged message in response to incoming packets with an error code indicating a type of link error.
2. The tangible, non-transitory, computer-readable medium of claim 1 , wherein the network error occurs during operation of the multi-die system and is not related to an initial boot or restart of the multi-die system, and wherein the processor is configured to invoke an interrupt handler to interrupt the link in response to the network error.
3. The tangible, non-transitory, computer-readable medium of claim 1 , wherein logically switching the link off comprises clearing in-flight packets designated to be transferred over the link.
4. The tangible, non-transitory, computer-readable medium of claim 1 , wherein the link comprises an advanced interface bus link.
5. The tangible, non-transitory, computer-readable medium of claim 4 , wherein the error comprises an advanced interface bus error.
6. The tangible, non-transitory, computer-readable medium of claim 5 , wherein the processor is configured to start a timer after an initial boot or restart of the multi-die system, and wherein the indication of the error comprises a duration of the timer elapsing without receiving a transmitter enable bit or a receiver enable bit over the advanced interface bus link.
7. The tangible, non-transitory, computer-readable medium of claim 1 , wherein the link comprises a silicon photonics link.
8. The tangible, non-transitory, computer-readable medium of claim 7 , wherein the error comprises an optical link training timeout error of the silicon photonics link.
9. The tangible, non-transitory, computer-readable medium of claim 8 , wherein the optical link training timeout error of the silicon photonics link comprises a lack of receiving a clock and data recovery lock status bit associated with the silicon photonics link within a threshold of time after a boot of the multi-die system.
10. The tangible, non-transitory, computer-readable medium of claim 8 , wherein the optical link training timeout error of the silicon photonics link comprises a lack of receiving a phase-locked loop lock status bit within a threshold of time after a boot of the multi-die system.
11. The tangible, non-transitory, computer-readable medium of claim 1 , wherein the error comprises a channel alignment error that occurs when alignment training is not complete within a specific time of a boot of the multi-die system.
12. The tangible, non-transitory, computer-readable medium of claim 1 , wherein the link comprises an optical link, and the error comprises an optical link failure corresponding to a threshold number of virtual channels of the optical link have timed out.
13. The tangible, non-transitory, computer-readable medium of claim 12 , wherein the threshold number comprises all of the virtual channels of the optical link.
14. A method, comprising:
determining that a timeout for a virtual channel of a link between a plurality of dies of a multi-die system;
determining that packets have been replayed over the virtual channel more than a threshold number of attempts and continues to timeout; and
in response to the packets being replayed more than the threshold number of attempts, disabling the virtual channel.
15. The method of claim 14 , wherein the threshold number is a specified number of replays of the packets, ranging from 4 to 8.
16. The method of claim 14 , comprising:
determining that a threshold number of virtual channels of the link have failed; and
in response to a determination that more than the threshold number of the virtual channels of the link have failed, disable the link.
17. The method of claim 14 , comprising:
determining that fewer than a threshold number of virtual channels of the link have failed; and
continue using the link while disabling the virtual channel.
18. The method of claim 14 , wherein disabling the virtual channel comprises responding to incoming requests to transfer data over the virtual channel with non-acknowledged message indicating that the virtual channel has failed using a corresponding error code.
19. The method of claim 14 , wherein disabling the virtual channel comprises clearing in-flight packets that are designated to be transferred over the virtual channel.
20. A method, comprising:
determining that a timeout for a virtual channel of a link between a plurality of dies of a multi-die system;
determining that packets have been replayed over the virtual channel more than a threshold number of attempts and continues to timeout;
in response to the packets being replayed more than the threshold number of attempts, disable the virtual channel;
responding to requests on the disabled virtual channel with a first not-acknowledged response indicating that the virtual channel has failed;
determining that more virtual channels of the link are disabled than a link threshold;
responsive to a determination that more virtual channels of the link are disabled than the link threshold, disabling the link; and
responding to requests on the disabled link with a first not-acknowledged response indicating that the link has failed.