Re: [Xen-devel] [PATCH net] xen-netback: use jiffies_64 value to calculate credit timeout

[jianhai luan <jianhai.luan@xxxxxxxxxx>]

On 2013-10-28 10:58, annie li wrote:
> On 2013-10-28 10:36, annie li wrote:
> > On 2013-10-27 19:11, Wei Liu wrote:
> > > time_after_eq() only works if the delta is < MAX_ULONG/2.
> > >
> > > For a 32bit Dom0, if netfront sends packets at a very low rate, the 
> > > time
> > > between subsequent calls to tx_credit_exceeded() may exceed MAX_ULONG/2
> > > and the test for timer_after_eq() will be incorrect. Credit will not be
> > > replenished and the guest may become unable to send packets (e.g., if
> > > prior to the long gap, all credit was exhausted).
> > >
> > > Use jiffies_64 variant to mitigate this problem for 32bit Dom0.
> > >
> > > Suggested-by: Jan Beulich <jbeulich@xxxxxxxx>
> > > Suggested-by: David Vrabel <david.vrabel@xxxxxxxxxx>
> > > Signed-off-by: Wei Liu <wei.liu2@xxxxxxxxxx>
> > > Cc: Ian Campbell <ian.campbell@xxxxxxxxxx>
> > > Cc: Jason Luan <jianhai.luan@xxxxxxxxxx>
> > > ---
> > >   drivers/net/xen-netback/common.h    |    1 +
> > >   drivers/net/xen-netback/interface.c |    4 ++--
> > >   drivers/net/xen-netback/netback.c   |   13 ++++++-------
> > >   3 files changed, 9 insertions(+), 9 deletions(-)
> > >
> > > diff --git a/drivers/net/xen-netback/common.h 
> > > b/drivers/net/xen-netback/common.h
> > > index 5715318..400fea1 100644
> > > --- a/drivers/net/xen-netback/common.h
> > > +++ b/drivers/net/xen-netback/common.h
> > > @@ -163,6 +163,7 @@ struct xenvif {
> > >       unsigned long   credit_usec;
> > >       unsigned long   remaining_credit;
> > >       struct timer_list credit_timeout;
> > > +    u64 credit_window_start;
> > >         /* Statistics */
> > >       unsigned long rx_gso_checksum_fixup;
> > > diff --git a/drivers/net/xen-netback/interface.c 
> > > b/drivers/net/xen-netback/interface.c
> > > index 01bb854..8c45b63 100644
> > > --- a/drivers/net/xen-netback/interface.c
> > > +++ b/drivers/net/xen-netback/interface.c
> > > @@ -312,8 +312,8 @@ struct xenvif *xenvif_alloc(struct device 
> > > *parent, domid_t domid,
> > >       vif->credit_bytes = vif->remaining_credit = ~0UL;
> > >       vif->credit_usec  = 0UL;
> > >       init_timer(&vif->credit_timeout);
> > > -    /* Initialize 'expires' now: it's used to track the credit 
> > > window. */
> > > -    vif->credit_timeout.expires = jiffies;
> > > +    /* credit window is tracked in credit_window_start */
> > > +    vif->credit_window_start = get_jiffies_64();
> > >         dev->netdev_ops    = &xenvif_netdev_ops;
> > >       dev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_TSO;
> > > diff --git a/drivers/net/xen-netback/netback.c 
> > > b/drivers/net/xen-netback/netback.c
> > > index f3e591c..1bc0688 100644
> > > --- a/drivers/net/xen-netback/netback.c
> > > +++ b/drivers/net/xen-netback/netback.c
> > > @@ -1185,18 +1185,17 @@ out:
> > >     static bool tx_credit_exceeded(struct xenvif *vif, unsigned size)
> > >   {
> > > -    unsigned long now = jiffies;
> > > -    unsigned long next_credit =
> > > -        vif->credit_timeout.expires +
> > > -        msecs_to_jiffies(vif->credit_usec / 1000);
> > > +    u64 now = get_jiffies_64();
> > > +    u64 next_credit = vif->credit_window_start +
> > > +        (u64)msecs_to_jiffies(vif->credit_usec / 1000);
> > You simply replace "credit_timeout.expires" with 
> > "vif->credit_window_start" here, and never update 
> > "vif->credit_window_start" in following code.
> > >         /* Timer could already be pending in rare cases. */
> > >       if (timer_pending(&vif->credit_timeout))
> > >           return true;
> > >         /* Passed the point where we can replenish credit? */
> > > -    if (time_after_eq(now, next_credit)) {
> > > -        vif->credit_timeout.expires = now;
> > > +    if (time_after_eq64(now, next_credit)) {
> > > +        vif->credit_timeout.expires = (unsigned long)now;
> > updates credit_window_start as following,
> > vif->credit_window_start = (unsigned long)now;
> both credit_window_start and credit_timeout.expires need to be updated 
> here,
> vif->credit_window_start = (unsigned long)now;
> vif->credit_timeout.expires = (unsigned long)now;
Annie should be correct.
It is good to me but i afraid that get_jiffies_64() and 
time_after_eq64() will bring some  performance's load when transmit is 
busy in 32bit.  I will do some test about the patch.
No matter what, i prefer own patch.

Thanks,
Jason
> > >           tx_add_credit(vif);
> > >       }
> > >   @@ -1207,7 +1206,7 @@ static bool tx_credit_exceeded(struct xenvif 
> > > *vif, unsigned size)
> > >           vif->credit_timeout.function =
> > >               tx_credit_callback;
> > >           mod_timer(&vif->credit_timeout,
> > > -              next_credit);
> > > +              (unsigned long)next_credit);
> > vif->credit_timeout.expires is unsigned long, and this still causes 
> > original issue on 32bit system, which works well on 64bit. Or 
> > rewriting code to avoid uses of vif->credit_timeout.expires, but it 
> > is complex.
> My understanding here is wrong, please ignore this.
>
> Thanks
> Annie

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel