ignoring xenbus state transitions during shutdown

ignoring xenbus state transitions during shutdown

[David Lively]

Hi Folks -
  In otherend_changed() in xenbus_probe.c, the driver is trying to 
ignore xenbus state transitions while the guest OS is shutting down.  It 
currently does this by looking for system_state > SYSTEM_RUNNING.  But 
system_state is not exported by some kernels that use it (e.g., SLES9 
running in a HVM guest), so the xenbus driver can't be loaded. 
  The comments indicate we're trying to avoid a failure that occurs when 
there is no rootfs.  I haven't yet seen this failure (after removing the 
code my shutdowns with PV drivers seemed to work fine, though I haven't 
pushed hard yet), but I suspect I shouldn't be ignoring it.  I was 
thinking it might be better to test for the screw condition (no rootfs) 
directly, perhaps something like try vfs_stat("/", &stat) and see 
whether it fails (??).  (These state transitions are relatively rare, so 
it doesn't need to be a particularly cheap test.)
  Any other suggestions?  I've heard of the old global 'system_running', 
but that's not used SLES9, at least.

Thanks in Advance,
Dave

Re: ignoring xenbus state transitions during shutdown

[Keir Fraser]

Unless you care about domU kexec, you can pretty safely remove that
system_state() check, so long as you remove the .shutdown callback hooks (in
the same file) too.

 -- Keir

On 25/6/07 19:41, "David Lively" <dlively@virtualiron.com> wrote:

> Hi Folks -
>   In otherend_changed() in xenbus_probe.c, the driver is trying to
> ignore xenbus state transitions while the guest OS is shutting down.  It
> currently does this by looking for system_state > SYSTEM_RUNNING.  But
> system_state is not exported by some kernels that use it (e.g., SLES9
> running in a HVM guest), so the xenbus driver can't be loaded.
>   The comments indicate we're trying to avoid a failure that occurs when
> there is no rootfs.  I haven't yet seen this failure (after removing the
> code my shutdowns with PV drivers seemed to work fine, though I haven't
> pushed hard yet), but I suspect I shouldn't be ignoring it.  I was
> thinking it might be better to test for the screw condition (no rootfs)
> directly, perhaps something like try vfs_stat("/", &stat) and see
> whether it fails (??).  (These state transitions are relatively rare, so
> it doesn't need to be a particularly cheap test.)
>   Any other suggestions?  I've heard of the old global 'system_running',
> but that's not used SLES9, at least.
> 
> Thanks in Advance,
> Dave
> 
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel

Re: ignoring xenbus state transitions during shutdown

[David Lively]

Ok --  the .shutdown callbacks are there for Linux kernels >= 2.6.16.  
So I think I want to protect this (ignore transitions) code with:
#if defined(CONFIG_KEXEC) || LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,16)

(That is, ignore state transitions if kexec is enabled or if we're 
defining the shutdown callbacks.)

Right?  (Or should I instead test only for CONFIG_KEXEC there and leave 
the shutdown callbacks NULL unless KEXEC is enabled??  Either way works 
for the cases I care about.)

Dave

Keir Fraser wrote:
> Unless you care about domU kexec, you can pretty safely remove that
> system_state() check, so long as you remove the .shutdown callback hooks (in
> the same file) too.
>
>  -- Keir
>
> On 25/6/07 19:41, "David Lively" <dlively@virtualiron.com> wrote:
>
>   
>> Hi Folks -
>>   In otherend_changed() in xenbus_probe.c, the driver is trying to
>> ignore xenbus state transitions while the guest OS is shutting down.  It
>> currently does this by looking for system_state > SYSTEM_RUNNING.  But
>> system_state is not exported by some kernels that use it (e.g., SLES9
>> running in a HVM guest), so the xenbus driver can't be loaded.
>>   The comments indicate we're trying to avoid a failure that occurs when
>> there is no rootfs.  I haven't yet seen this failure (after removing the
>> code my shutdowns with PV drivers seemed to work fine, though I haven't
>> pushed hard yet), but I suspect I shouldn't be ignoring it.  I was
>> thinking it might be better to test for the screw condition (no rootfs)
>> directly, perhaps something like try vfs_stat("/", &stat) and see
>> whether it fails (??).  (These state transitions are relatively rare, so
>> it doesn't need to be a particularly cheap test.)
>>   Any other suggestions?  I've heard of the old global 'system_running',
>> but that's not used SLES9, at least.
>>
>> Thanks in Advance,
>> Dave
>>
>>
>> _______________________________________________
>> Xen-devel mailing list
>> Xen-devel@lists.xensource.com
>> http://lists.xensource.com/xen-devel
>>     
>
>   

Re: ignoring xenbus state transitions during shutdown

[Keir Fraser]

Only need #if LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,16). If the shutdown
hooks aren't defined then we won't be causing xenbus state transitions when
we shut down, kexec or not.

 -- Keir

On 25/6/07 21:28, "David Lively" <dlively@virtualiron.com> wrote:

> Ok --  the .shutdown callbacks are there for Linux kernels >= 2.6.16.
> So I think I want to protect this (ignore transitions) code with:
> #if defined(CONFIG_KEXEC) || LINUX_VERSION_CODE >= KERNEL_VERSION(2,6,16)
> 
> (That is, ignore state transitions if kexec is enabled or if we're
> defining the shutdown callbacks.)
> 
> Right?  (Or should I instead test only for CONFIG_KEXEC there and leave
> the shutdown callbacks NULL unless KEXEC is enabled??  Either way works
> for the cases I care about.)
> 
> Dave
> 
> Keir Fraser wrote:
>> Unless you care about domU kexec, you can pretty safely remove that
>> system_state() check, so long as you remove the .shutdown callback hooks (in
>> the same file) too.
>> 
>>  -- Keir
>> 
>> On 25/6/07 19:41, "David Lively" <dlively@virtualiron.com> wrote:
>> 
>>   
>>> Hi Folks -
>>>   In otherend_changed() in xenbus_probe.c, the driver is trying to
>>> ignore xenbus state transitions while the guest OS is shutting down.  It
>>> currently does this by looking for system_state > SYSTEM_RUNNING.  But
>>> system_state is not exported by some kernels that use it (e.g., SLES9
>>> running in a HVM guest), so the xenbus driver can't be loaded.
>>>   The comments indicate we're trying to avoid a failure that occurs when
>>> there is no rootfs.  I haven't yet seen this failure (after removing the
>>> code my shutdowns with PV drivers seemed to work fine, though I haven't
>>> pushed hard yet), but I suspect I shouldn't be ignoring it.  I was
>>> thinking it might be better to test for the screw condition (no rootfs)
>>> directly, perhaps something like try vfs_stat("/", &stat) and see
>>> whether it fails (??).  (These state transitions are relatively rare, so
>>> it doesn't need to be a particularly cheap test.)
>>>   Any other suggestions?  I've heard of the old global 'system_running',
>>> but that's not used SLES9, at least.
>>> 
>>> Thanks in Advance,
>>> Dave
>>> 
>>> 
>>> _______________________________________________
>>> Xen-devel mailing list
>>> Xen-devel@lists.xensource.com
>>> http://lists.xensource.com/xen-devel
>>>     
>> 
>>   
> 
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel@lists.xensource.com
> http://lists.xensource.com/xen-devel