| 1 Jun 2025 |
 emily | I remember you saying a few days ago that SGMII encompasses 2.5G. am I right to think that "2.5G SGMII" is that HSGMII or whatever you were talking about and that's distinct from 2500BASE-X? | 12:45:46 |
 K900 | Yes | 12:45:55 |
| K900 | Though SGMII doesn' | 12:46:00 |
| K900 | t | 12:46:00 |
| K900 | Ugh | 12:46:05 |
| K900 | I think what I was saying is that there's two PCS | 12:46:11 |
| K900 | One for SGMII and one for USXGMII | 12:46:20 |
| K900 | But | 12:46:21 |
| K900 | 2500base-x is also handled by the "SGMII" PCS | 12:46:30 |
| emily | I'm glad that this is confusing enough that even you don't understand what you say :D | 12:46:41 |
| K900 | Which technically makes it an SGMII/2500base-x combo PCS | 12:46:44 |
| K900 | Though it can also do 1000base-x | 12:46:59 |
| K900 | So it's technically an SGMII/2500base-x/1000base-x combo PCS | 12:47:08 |
| emily | okay so here's what the BCM84891L datasheet (https://datasheet4u.com/pdf-down/B/C/M/BCM84891L-Broadcom.pdf) says: "The BCM84891L supports the XFI/10GBASE-KR, USXGMII, 5000BASE-R, 2500BASE-R, 5000BASE-X, 2500BASE-X, and 1000BASE-X (SGMII) interface for connection to a MAC." | 12:47:19 |
| emily | do you know what I should take that to imply about how it's going to interface? | 12:47:29 |
| emily | "2.5G rate over USXGMII/XFI/2500BASE-R/2500BASE-X MAC interface" | 12:47:38 |
| emily | this is implying to me that it should be able to do the "dumb" passthrough? | 12:47:59 |
| emily | where the kernel can talk to the MAC directly? | 12:48:10 |
| K900 | I assume it's going to depend on some magic preset configuration | 12:48:21 |
| K900 | For a specific module | 12:48:25 |
| emily | right I'm not talking about the modules themselves | 12:48:37 |
| emily | I've sorta accepted I just have to keep gambling on those | 12:48:43 |
| emily | I'm just trying to establish what the actual chip can do | 12:48:48 |
| emily | "The BCM84891L can communicate with Ethernet MACs that support 2500BASE-X interface through XFI at 3.125 Gbaud (Gb/s) through the RDN[x]/RDP[x] and TDN[x]/TDP[x] pins." | 12:49:20 |
| emily | that seems promising | 12:49:22 |
| emily | that would mean I could use the on-SoC MAC in the R4 and the BCM84891L would just handle the actual 2.5GBASE-T physical layer conversion? | 12:50:16 |
 magic_rb | cant you buy a 5g copper module and then plug 2.5g copper into it? wouldnt it figure out the autoneg with the other end? | 12:50:46 |
| emily | there are 2.5GBASE-T modules that will talk 10G to the host with pause frames or whatever | 12:51:23 |
| emily | but I think that implies the kernel not being able to directly talk to the MAC | 12:51:52 |
| emily | because it means stuff has to happen on the transceiver end | 12:52:04 |
