Agilex PCIe IP Bypass mode

[GS65]

I am trying to use the PCIe Hard IP in bypass mode. I would like to know the following.

1. Does the enable PCIe receive queues (P,NP,C) in bypass modes and handle flow control?

2. If so, is PCIe ordering rules for relaxed ordering implemented at the output of the receive queues before the TLP is passed on to the application layer?

We am trying to decide between Altera and Xilinx for a production device. This will make a difference in device selection.

Thanks for the help

[Wincent_Altera]

Hi,

Thank you for reaching out.

Just to let you know that Intel has received your support request and I am assigned to work on it.

Allow me some time to look into your issue. I shall come back to you with findings.

Thank you for your patience.

Best regards,

Wincent_Intel

[Wincent_Altera]

Hi,

1. Does the enable PCIe receive queues (P,NP,C) in bypass modes and handle flow control?

The credit interface is used to implement flow control for the data movement between the user application interface and each IP block. Each header type (P,NP,CPL) and data type (P,NP,CPL) has an independent credit interface. One data credit consists of 16 bytes. One header credit includes the TLP Header, 1DW prefix (if present) and the digest (if present).The 4.4.1.2.2. Credit Initialization
describes the credit initialization and update flow.

https://www.intel.com/content/www/us/en/docs/programmable/683501/21-4-4-0-0/tlp-bypass-mode.html

2. If so, is PCIe ordering rules for relaxed ordering implemented at the output of the receive queues before the TLP is passed on to the application layer?

the relaxed ordering Implemented on the RX side. This feature is always active. On the TX side, reordering is done by the application.

Let me know if this answering your question.

Regards,

Wincent_Intel

[GS65]

Hi,

Thanks for the quick reply.

  I am referring to F-Tile Avalon® Streaming Intel®
FPGA IP for PCI Express* User Guide Updated for Intel® Quartus® Prime Design Suite: 22.4
IP Version: 8.0.0.

Section 3.3.2.1 says that RX flow control interface is optional and disabled by default in the IP GUI. Does the IP expect the application to maintain separate queues for data and headers for P,NP and C TLPs? 

Is there a single flow control interface? If Relaxed ordering is implemented, NP and C TLPs cannot pass posted TLPs. So, why is the IP expecting separate flow control credits for the TLP types? 

Thanks

[Wincent_Altera]

Hi,

Does the IP expect the application to maintain separate queues for data and headers for P,NP and C TLPs? 
>>The RX flow control interface is optional and disabled by default, meaning that the IP does not require the application to provide flow control credits for incoming TLPs.

Is there a single flow control interface?

>> if the RX flow control interface is optional and disabled by default, it suggests that the IP does not provide a dedicated flow control interface for each TLP type.

If Relaxed ordering is implemented, NP and C TLPs cannot pass posted TLPs. So, why is the IP expecting separate flow control credits for the TLP types? 
>> it is important to note that the PCIe specification requires strict ordering between posted TLPs and completion TLPs. In contrast, non-posted TLPs, such as NP and C TLPs, do not have such strict ordering requirements. This means that the IP may require separate flow control credits for different TLP types to ensure that the ordering requirements are met for posted TLPs, while allowing for more flexibility in the ordering of NP and C TLPs.

Let me know if this answer your question or you need any further clarification.

Regards,

Wincent_Intel

[GS65]

Hi,

     Thanks for the clarification.  The following comment from you is not clear 

>> "it is important to note that the PCIe specification requires strict ordering between posted TLPs and completion TLPs. In contrast, non-posted TLPs, such as NP and C TLPs, do not have such strict ordering requirements. This means that the IP may require separate flow control credits for different TLP types to ensure that the ordering requirements are met for posted TLPs, while allowing for more flexibility in the ordering of NP and C TLPs."

I am implementing relaxed ordering. With RO, non-posted (NP) and completion (C) TLPs cannot pass posted TLPs. So, if a posted TLP is next selected TLP as per RO rules, and the application has no posted credits the output should be blocked. In other words, posted TLPs cannot be bypassed even if I have credits for NP or C TLPs in the application. I agree that separate credits for NP and C TLPs may help in reordering those TLPs. 

[Wincent_Altera · Accepted solution]

Hi,

Your understanding is correct, is there anything else you not understand and I can clarified more ?

Regards,

Wincent_Intel

[GS65]

Hi

    Thanks for the prompt replies. The functionality is pretty clear now. 

[Wincent_Altera]

Hi,

Glad that I am answering your question.

Hence, This thread will be transitioned to community support.

If you have a new question, feel free to open a new thread to get support from Intel experts.

Otherwise, the community users will continue to help you on this thread. Thank you

If you feel your support experience was less than a 9 or 10,

please allow me to correct it before closing or let me know the cause so that I may improve your future support experience.

Regards,

Wincent_Intel

[GS65]

Hi Vincent,

                 I have a related question on AGF014 part. We are trying to build a PCIe switch with a AGF014 development board. (We ordered the DEV-AGF027 but there are supply issues with the board). We were under the impression that any of the 4 x4 ports in the P-tile can be configured as UPSTREAM or DOWNSTREAM ports. However we just found out that the tool only allows all x4's to be either UP or DN. As you can imagine we cannot build a switch will all UP or DN ports. Can we configure each of the 4 x4's in a P-tile to either UP or DN?

Thanks

GS