On Fri, Oct 20, 2023 at 1:31 PM Daniel Berlin <dberlin@xxxxxxxxxxx> wrote: > > So, at least in the example code I have, all variants of the 109 and > 112 structures both have bcnflags in that place - it was always > missing here. > For example, see > https://android.googlesource.com/kernel/google-modules/wlan/bcmdhd/bcm4398/+/refs/heads/android-gs-shusky-5.15-u-qpr1-beta2/include/wlioctl.h > and compare v109 and v112 of bssinfo. > > As such, the 109 and 112 structures are compatible given these definitions. > > I don't know if what is there right now is wrong, or it is "yet > another variant" of the 109 structure that we need to handle. > Any idea what the ground truth is? Circling back to this - i have checked other sources as well - infineon also has u8 (though it's marked as padding) there, etc. As far as i can tell, the structure should have that u8 there in all versions i can find. Nevertheless, I think I can make it not matter ;) I'm going to post an RFC of some patches that handle this and other structure versioning things through function pointer structures that we set based on interface versions available. So instead of setting feature flags, we query the various iovars/firmware info for the right interface versions, and set up the structures/function pointers to handle the versions we will get from the firmware. This also makes the common code cleaner as they no longer have to deal with the structure differences - for example, brcmf_cfg80211_connect now just calls something like drvr->join_params_handler.get_struct_from_connect to get an extended join params struct of the right version and doesn't look inside the result anymore. It's an RFC so we can argue about the approach and APIs, i just did what seemed easy/obvious. I have converted bss info, join params, scan params, and netinfo versions to use this approach, and all now support all versions of the structures available. If we go that route, it wont matter if we overlay bss info structures or not. As an aside, while doing the function pointer work, i discovered some other structure bugs (missing fields) that seem to be causing failures/fallbacks on every chip i tested (no matter who the vendor was). So for example, some of the extended join parameter substructs are missing fields, and every chip i tried (5 variants, representing 2 vendors and 3 join_param versions) all gave BUFTOSHORT when trying to use the join iovar as a result, and all fell back to using the SET_SSID method. With the v0/v1 structure fixed, they all use the join iovar successfully and never fall back. --Dan
