Bug 18757
| Summary: | [PATCH] fxp driver doesn't enable flow control | ||||||
|---|---|---|---|---|---|---|---|
| Product: | Base System | Reporter: | ken <ken> | ||||
| Component: | kern | Assignee: | Andre Oppermann <andre> | ||||
| Status: | Closed FIXED | ||||||
| Severity: | Affects Only Me | ||||||
| Priority: | Normal | ||||||
| Version: | 5.0-CURRENT | ||||||
| Hardware: | Any | ||||||
| OS: | Any | ||||||
| Attachments: |
|
||||||
Responsible Changed From-To: freebsd-bugs->dg Over to the driver author. State Changed From-To: open->feedback Is this PR still applicable? It looks to me like jlemon committed such support (in a different way) in March, 2001. Can this PR now be closed? Responsible Changed From-To: dg->jlemon State Changed From-To: feedback->closed I'm closing this PR because it open for too long and this issue seems to fixed, at least together with my Cisco Catalyst switch flow control seems to work. I don't know which exact commit fixed it. Responsible Changed From-To: jlemon->andre I'm closing this PR because it open for too long and this issue seems to fixed, at least together with my Cisco Catalyst switch flow control seems to work. I don't know which exact commit fixed it. |
By default, the fxp driver doesn't enable flow control. The 82559 chip on my ethernet card, however, negotiates flow control with my Intel 410T switch. The switch proceeds to send flow control ethernet packets to the card, but since flow control isn't enabled on the card, the flow control packets are routed up to the kernel. They look like this from tcpdump: 22:15:53.313742 0:0:0:0:0:10 1:80:c2:0:0:1 8808 60: 0001 0000 0000 8808 0001 0000 0000 8808 0001 0000 0000 8808 0001 0000 0000 8808 0001 0000 0000 8808 0001 0000 0000 22:15:53.355673 0:0:0:0:0:10 1:80:c2:0:0:1 8808 60: 0001 0000 0000 8808 0001 0000 0000 8808 0001 0000 0000 8808 0001 0000 0000 8808 0001 0000 0000 8808 0001 0000 0000 The last octet of the source hardware address changes based on what port in the switch I'm plugged into. (It's in port 10 now. If I plug it into say port 11, it'll be 0x20, 12, 0x30, and so on.) There are several screens full of the above packets in tcpdump every second without the patch below. netstat -i shows packet counts increasing significantly every second. Fix: The patch below is also located here: http://people.FreeBSD.ORG/~ken/if_fxp.flow_control.20000522 I figured out how to enable flow control from the Linux driver for the 8255x boards. Intel's documentation is rather un-helpful. I have been running with this patch on -current since early February 2000. There are a couple of caveats for this patch: 1. I don't know what the bits do, or whether all of them are necessary for flow control. Someone with access to the (real) documentation for the 8225x boards should take a look at this. The documentation for these boards is only available under NDA, from what I have heard. The documentation available on thier web site isn't sufficient to investigate or fix this problem. 2. I have only tested this on my 82559 "InBusiness" board, so I don't know what it will do to say a vanilla 82557. ==== //depot/FreeBSD-ken/src/sys/pci/if_fxp.c#4 - /a/ken/perforce/FreeBSD-ken/src/sys/pci/if_fxp.c ==== How-To-Repeat: Get an Intel 410T switch (haven't tried any other switch) and an Intel 82559 board, watch packet counts and run tcpdump.