On Monday, October 11, 2021 10:21:34 AM CEST Dan Carpenter wrote: > On Sun, Oct 10, 2021 at 11:21:49AM +0200, Fabio M. De Francesco wrote: > > On Saturday, October 9, 2021 6:31:12 PM CEST Fabio M. De Francesco wrote: > > > On Thursday, August 26, 2021 3:54:13 PM CEST Dan Carpenter wrote: > > > > Another thing to fix are some of the sleeping in atomic bugs. > > > > > > > > drivers/staging/r8188eu/core/rtw_ap.c:139 update_BCNTIM() warn: sleeping > > > > in atomic context > > > > drivers/staging/r8188eu/core/rtw_ap.c:1296 update_bcn_wps_ie() warn: > > > > sleeping in atomic context > > > > > > > > [...] > > > > > > > Hello Dan, > > > > > > I'd like to address these kind of bugs, but I have a couple of questions > > > about them. > > > > > > 1) You've listed what looks like the output of a compiler or static > > > analyzer. > > > How did you get the warnings you copy-pasted above? > > > > > > 2) I know that both the execution of interrupt handlers (ISRs) as well as > > > any > > > code blocks that are executed holding spinlocks are "atomic contexts". In > > > these cases, "sleeping" is not allowed (for obvious reasons). Besides the > > > two > > > mentioned above, are there any further cases of "atomic contexts" in the > > > kernel? > > > > After some research, I've found that Softirqs and Tasklets are also executed > > in "atomic context", as hardware interrupt service routines are. > > > > Furthermore, I've also found a .config option named DEBUG_ATOMIC_SLEEP > > that should warn if some code is sleeping in "atomic context". However, the > > documentation of that option does not explain where the output of these > > checks can be read. > > > > I would appreciate any help on this matter. > > These are a new Smatch warning. > > https://lkml.org/lkml/2021/9/1/950 > > You would need to rebuild the Smatch database probably around five times > to see the warnings. It takes a long time to build... It's probably > not that hard to figure out just from looking at the code without the > generating the warning. I must confess that, since I started to submit patches to the Kernel this year in April and until now, I have not ever used Smatch. I thought that building with GCC and setting C=2 and W=1 were enough. Sometimes I've also used Coccinelle. That's all. Now I admit that I was plainly wrong. My _big_ fault, sorry :( This morning I have taken a quick look at your code at https://github.com/ error27/smatch. Obviously, what I could see is only the overall design and it looks quite impressive. I'll start using Smatch soon. > So spin locks can't sleep. Mutexes can. There are read/write locks, > built on both mutexes and spin locks. Rcu_read/write cannot sleep. Oh, right. RCU were missing from my lists. Thanks very much for Smatch and for your kind replies. Regards, Fabio M. De Francesco P.S.: Yesterday I read the first one third of a paper co-authored by Julia Lawall (Effective Detection of Sleep-in-atomic-context Bugs in the Linux Kernel - https://doi.org/10.1145/3381990). It talks about a practical static approach named DSAC. It looks really interesting, but it seems that their tool is not yet available to the public. Do you know something about it? I've found an old message by Greg K-H that asked for where to find that tool but, as far as I know, he have not yet had that information. With this message I've Cc'd Julia in case she has time to reply. > regards, > dan carpenter > >
