Re: [syzbot] [kernel?] possible deadlock in console_flush_all (2)

[Tetsuo Handa]

Hello, bpf developers.

syzbot is reporting circular locking dependency because bpf program is hitting
printk() path. To avoid this dependency, something like
https://syzkaller.appspot.com/text?tag=Patch&x=121c92fe180000 is needed.

Is it possible to call printk_deferred_{enter,exit}() from the bpf side (e.g.
bpf_prog_run()) so that we don't need to add overhead on the scheduler side
when bpf programs are not registered?

printk_deferred_{enter,exit}() requires that printk_deferred_exit() is called
on a CPU which printk_deferred_enter() was called, for percpu counter is used.
Is migrate_{disable,enable}() also needed around printk_deferred_{enter,exit}() ?

On 2023/10/08 15:28, syzbot wrote:
> Hello,
> 
> syzbot found the following issue on:
> 
> HEAD commit:    7d730f1bf6f3 Add linux-next specific files for 20231005
> git tree:       linux-next
> console output: https://syzkaller.appspot.com/x/log.txt?x=172d36e9680000
> kernel config:  https://syzkaller.appspot.com/x/.config?x=f532286be4fff4b5
> dashboard link: https://syzkaller.appspot.com/bug?extid=f78380e4eae53c64125c
> compiler:       gcc (Debian 12.2.0-14) 12.2.0, GNU ld (GNU Binutils for Debian) 2.40