Hi Martin, > With the recent additions of new features, non-split phy dumps exceed the > maximum Netlink size of 4096 bytes. Depending on their feature set, not all > radios are affected, but for example hwsim is. Hmm, ok. I wonder which features in particular? > While userspace dumping phys can request a split dump as a work around, > legacy tools without split support may get oversized dumps. Much more > problematic are phy new/del notifications, which are always limited to > a page size: Generating these messages fails, and Netlink notifications > are silently dropped. This breaks userspace relying on these events. Right, not good. > There is no single commit that broke these events, as their size highly > depends on the phy feature set. Finding attributes to strip is challenging: > The largest attributes are the frequency lists, but they are already > reduced to the minimum and have rather useful information. Right. > HE information > probably could be stripped, Right, bad. We should strip that entire stuff. > but the saving is about 200 bytes, not enough > to fix dumps for many cases. :( > The next larger attributes are usually the > RX/TX frame attributes. Removing these attributes can reduce a hwsim phy > dump from 4576 to 3288 bytes, and it seems to be the most reasonable > approach. Well the problem with that approach is that it kinda results in a different regression because the RX/TX stuff *was* present before split wiphys were even introduced. OTOH, I just checked, and back then wpa_supplicant wasn't using that information, so it seems fair game to remove. I've ended up with a patch that's a bit different - I'll post it in a minute. johannes
