Re: panic: syncache: mbuf too small
- Reply: Ryan Stone : "Re: panic: syncache: mbuf too small"
- In reply to: Drew Gallatin : "Re: panic: syncache: mbuf too small"
- Go to: [ bottom of page ] [ top of archives ] [ this month ]
Date: Tue, 08 Feb 2022 22:38:15 UTC
On Tue, 8 Feb 2022 at 14:34, Drew Gallatin <gallatin@netflix.com> wrote: > > I don't think the size has changed recently. However, there is a size difference for pkthdrs (and hence MHLEN) on 32-bit platforms vs 64-bit platforms. > > There are a number of bad ways to handle this. Eg, don't permit Ipv6 on these interfaces, make these interfaces chain their headers, assuming they can do s/g dma, make them copy to a contiguous buffer. Make mbufs bigger. All of the things I can think of are ugly. (a) yeah, some of the wifi firmware based things (bcom, realtek) have some weird ass requirements where the FW/PHY/MAC/802.11 headers end up needing to be in a contig buffer due to hardware/firmware requirements. You can sometimes chain things afterwards, but eg on the bwi/bwn firmware NICs you really need all of that data in the first sg buffer the firmware sees. (b) i think honestly we can just linearise the mbufs for the wifi drivers that need a big headroom nowdays, it's not a huge deal and in a lot of cases the linux drivers do? did? end up doing this to simplify handling. -adrian