WARNING - OLD ARCHIVES

This is an archived copy of the Xen.org mailing list, which we have preserved to ensure that existing links to archives are not broken. The live archive, which contains the latest emails, can be found at http://lists.xen.org/
   
 
 
Xen 
 
Home Products Support Community News
 
   
 

xen-devel

[Xen-devel] Re: netback commit history

>>> On 20.09.11 at 14:10, "Jan Beulich" <JBeulich@xxxxxxxx> wrote:
>>>> On 20.09.11 at 13:26, Ian Campbell <Ian.Campbell@xxxxxxxxxxxxx> wrote:
>> On Tue, 2011-09-20 at 12:04 +0100, Jan Beulich wrote:
>>> One thing I wonder about in this context is whether the
>>> netif_stop_queue() call from xenvif_close() shouldn't happen before
>>> xenvif_down() (not the least for reasons of symmetry with
>>> xenvif_open()).
>> 
>> I seem to recall looking at that too, it was the same in the old kernels
>> too and I didn't know why so I avoided touching it (I was doing too much
>> other cleanup at the time to risk it).
> 
> After looking further I don't think that would help (though I still think
> it would be more correct), as netif_stop_queue() basically evaluates
> to a set_bit() without any other locks taken. So it's completely
> asynchronous wrt dev_hard_start_xmit() (and its callers, which are
> the ones looking at the bit with HARD_TX_LOCK() carried out).

Which in turn suggests that the upstream driver isn't safe either:
There's nothing preventing vif->netbk from becoming NULL between
the early check in xenvif_start_xmit() and its actual use in
xen_netbk_queue_tx_skb(). I think vif->netbk needs to be latched
into a local variable, checked against NULL, and passed instead of
vif to xen_netbk_queue_tx_skb().

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-devel

<Prev in Thread] Current Thread [Next in Thread>