Page MenuHomeFreeBSD

LinuxKPI: 802.11: Always lock around lkpi_{ic_vap_delete,mo_stop}()
ClosedPublic

Authored by temcbun_gmail.com on Sat, Sep 12, 9:16 AM.
Referenced Files
F172616538: D59608.diff
Sat, Sep 19, 5:27 PM
F172543840: D59608.id.diff
Sat, Sep 19, 3:49 AM
F172502895: D59608.id186503.diff
Fri, Sep 18, 9:53 PM
Unknown Object (File)
Thu, Sep 17, 9:54 AM
Unknown Object (File)
Wed, Sep 16, 9:46 PM
Unknown Object (File)
Wed, Sep 16, 2:31 PM
Unknown Object (File)
Tue, Sep 15, 6:15 PM
Unknown Object (File)
Tue, Sep 15, 8:04 AM
Subscribers

Details

Summary
Add the missing locking as drivers always expect it (by assertion) and
cannot distinguish between state.

Add the assertions to the downcalls as we need both of them locked
and both of them can sleep.

PR: 298417

I tried to implement the changes to fix the bug, introduced in the PR, based on 5d479b75a894, which fixed the similar problem.

Test Plan

With this patch applied and the INVARIANTS kernel, destroying the rtw89(4) interface should not freeze the system:

$ ifconfig wlan create wlandev rtw890
wlan0
$ ifconfig wlan0 destroy

Diff Detail

Repository
rG FreeBSD src repository
Lint
Lint Skipped
Unit
Tests Skipped

Event Timeline

I'll commit this with my next batch.

Thanks for this! I thought (twice now) that I pushed updates for all of them but it seems I last time also only pushed the one to fix the PR.

I have a further issue in that I currently cannot even load rtw89 after a rebase of main; I am not sure what changed to cause that yet :( *sigh*

This revision is now accepted and ready to land.Sun, Sep 13, 4:28 PM

@bz Could you please MFC this? :)

It's on the list with 16 others for, likely, Sunday evening. I am trying to batch the MFCs.