Re: [syzbot] [kvm?] INFO: task hung in kvm_gmem_release

[Hillf Danton <hdanton@sina.com>]

On Tue,  1 Sep 2026 15:33:42 +0800 Hillf Danton wrote:
>On Mon, 31 Aug 2026 08:15:44 -0700 Sean Christopherson wrote:
>>
>> Marking this invalid, as I'm 99.9% certain guest_memfd is an innocent bystander.
>> CPU1 is stuck on an IN instruction (the actual port I/O), and there are a handful
>> of identical failures in random subsystems that pop up a few times a month[*].
>> My guess is the host side of the serial port is going out to lunch intermittently.
>> 
>> [*] https://lore.kernel.org/all/?q=%22io_serial_in%22%20syzkaller
>> 
>> #syz invalid
>>
>The 0.1% uncertainty hints this invalid mark is NOT true or at least earlier
>than thought, because a) the IN instruction is not kvm specific [11] and b) the
>i2c hang [11] is i2c specific [12].
>
>[11] Subject: [syzbot] [i2c?] [usb?] INFO: task hung in em28xx_i2c_unregister
>https://lore.kernel.org/lkml/6a95ef74.764875ec.b345.0003.GAE@google.com/
>
>[12] Subject: [syzbot] [i2c?] [usb?] INFO: task hung in em28xx_i2c_unregister
>https://lore.kernel.org/lkml/6a9676c7.764875ec.b345.0007.GAE@google.com/
>
>That said, extra evidence is needed to prove that the serial port is going out
>to lunch on the host side.
>
For archive purpose, the sleeper with nfsd_mutex held [15] explains the nfsd_mutex
waiter [16], another case that defies the lunch guess.

[15] Subject: [syzbot] [nfs?] INFO: task hung in nfsd4_cld_tracking_init
https://lore.kernel.org/lkml/6a99f6f0.2e567484.299ffc.000c.GAE@google.com/

[16] Subject: Re: [syzbot] [nfs?] INFO: task hung in nfsd_nrthreads
https://lore.kernel.org/lkml/6a98a3c5.27a413cd.1e878c.0010.GAE@google.com/