Debug flag changing fuzzer behavior

[Julia Keadey]

Hey there,

I'm currently trying to test some descriptions I wrote. When I run the fuzzer with the -debug flag, everything seems fine, and syzkaller is able to generate good testcases and get proper code coverage. When I take away the debug flag, the debugger stalls and says that that the program repeatedly failed to execute. Originally, I was worried that it was an issue when multiple vms ran at once, but this still seemed to fail when I changed the vm limit to 1.

What exactly does debug do that could cause such weirdness? Does it handle errors differently? Does it change anything about how it sets up the vm?

Thanks,

Julia K

[Aleksandr Nogikh]

Hi,

On Wed, Nov 19, 2025 at 5:25 PM 'Julia Keadey' via syzkaller
<syzk...@googlegroups.com> wrote:
>
> Hey there,
>
> I'm currently trying to test some descriptions I wrote. When I run the fuzzer with the -debug flag, everything seems fine, and syzkaller is able to generate good testcases and get proper code coverage. When I take away the debug flag, the debugger stalls and says that that the program repeatedly failed to execute.

Could you please share some more details on "debugger stalls"? What
debugger and how were you using it?

Regarding the "program repeatedly failed to execute": you mean
"SYZFAIL: repeatedly failed to execute the program", right? Does it
completely stall fuzzing or it just happens from time to time?

> Originally, I was worried that it was an issue when multiple vms ran at once, but this still seemed to fail when I changed the vm limit to 1.
>
> What exactly does debug do that could cause such weirdness? Does it handle errors differently? Does it change anything about how it sets up the vm?

It enables more output and restricts the number of procs (syz-executor
child processes that actually execute the programs) to 1. Otherwise it
should not change things much.

>
> Thanks,
> Julia K
>
> --

[Julia Keadey]

Hey there - I know this is a super late reply, I just wanted to document that I ended up fixing the error. The KCOV remote coverage code I had setup had some bugs. When I resolved those issues everything seemed to work fine. Still not quite sure why there were issues in the first place, but they are gone now. 

-- Julia