RE: [PATCH v2 0/4] hv_utils: Add the support of hibernation

[Dexuan Cui]

> From: Dexuan Cui
> Sent: Sunday, January 12, 2020 10:29 PM
> 
> Hi,
> This is an updated version of the v1 patchset:
> https://lkml.org/lkml/2019/9/11/861
> 
> Patch #1 is a new patch that makes the daemons more robust.
> 
> Patch #2 is the same as v1.
> 
> Patch #3 sends the host-initiated hibernation request to the user space via
> udev.
> (v1 used call_usermodehelper() and "/sbin/hyperv-hibernate".)
> 
> Patch #4 handles fcopy/vss specially to avoid possible inconsistent states.
> 
> Please review.
> 
> Thanks!
> 
> Dexuan Cui (4):
>   Patch #1: Tools: hv: Reopen the devices if read() or write() returns errors
>   Patch #2: hv_utils: Support host-initiated restart request
>   Patch #3: hv_utils: Support host-initiated hibernation request
>   Patch #4: hv_utils: Add the support of hibernation

Hi Vitaly,
I forgot to mention this for patch #4: IMO we don't need to add a new
HVUTIL_SUSPENDED state to the hvutil state machine, because:

When we reach util_suspend(), all the userspace processes have been
frozen: see kernel/power/hibernate.c: hibernate() -> freeze_processes() ->
try_to_freeze_tasks(true) -> freeze_task() -> fake_signal_wake_up(). When
try_to_freeze_tasks(true) returns 0, all the user-space processes must be 
frozen in do_signal() -> get_signal() -> try_to_freeze() -> ... -> __refrigerator().

hibernate () -> hibernation_snapshot () -> dpm_suspend() -> ... -> 
util_suspend() only runs after hibernate() -> freeze_processes(), so I'm
pretty sure we have no race condition with the user space daemon.

util_suspend() -> srv->util_pre_suspend() disables the tasklet and cancels any
pening work items, so there is no race in the kernel space, either.

Thanks,
-- Dexuan