[memnic PATCH 1/5] pmd: fix race condition

[memnic PATCH 1/5] pmd: fix race condition

[Hiroshi Shimamoto]

From: Hiroshi Shimamoto <h-shimamoto-ehU+Cx/zZe18UrSeD/g0lQ@public.gmane.org>

There is a race condition, on transmit to vSwitch.

Guest PMD                 Host
Thread-A     Thread-B     vSwitch
   |idx=0       |idx=0       |p[0] st!=2
   |cmpxchg     |            |
   |p[0] st->1  |            |
   |idx=1       |            |
   |fill data   |            |
   |p[0] st->2  |            |p[0] st==2
   |            |            |receive data
   |            |            |p[0] st->0
   |            |cmpxchg     |
   |            |success     |p[1] st!=2
   |            |p[0] st->1  |
                  This is BAD

That causes traffic stop.

We have to take care about that race condition with checking
whether current index is correct.

Signed-off-by: Hiroshi Shimamoto <h-shimamoto-ehU+Cx/zZe18UrSeD/g0lQ@public.gmane.org>
Reviewed-by: Hayato Momma <h-momma-JhyGz2TFV9J8UrSeD/g0lQ@public.gmane.org>
---
 pmd/pmd_memnic.c | 9 +++++++++
 1 file changed, 9 insertions(+)

diff --git a/pmd/pmd_memnic.c b/pmd/pmd_memnic.c
index 30d5a1b..805f0b2 100644
--- a/pmd/pmd_memnic.c
+++ b/pmd/pmd_memnic.c
@@ -278,6 +278,15 @@ retry:
 			goto retry;
 		}
 
+		if (idx != ACCESS_ONCE(adapter->down_idx)) {
+			/*
+			 * vSwitch freed this and got false positive,
+			 * need to recover the status and retry.
+			 */
+			p->status = MEMNIC_PKT_ST_FREE;
+			goto retry;
+		}
+
 		if (++idx >= MEMNIC_NR_PACKET)
 			idx = 0;
 		adapter->down_idx = idx;
-- 
1.8.4

Re: [memnic PATCH 1/5] pmd: fix race condition

[Olivier MATZ]

Hi Hiroshi-san,

Please see my comments below.

On 03/11/2014 06:37 AM, Hiroshi Shimamoto wrote:
> From: Hiroshi Shimamoto <h-shimamoto-ehU+Cx/zZe18UrSeD/g0lQ@public.gmane.org>
> 
> There is a race condition, on transmit to vSwitch.

I think we should not talk specifically about vSwitch, as
another implementation of host memnic is possible. Maybe using
the term "host" is more appropriate?

> +		if (idx != ACCESS_ONCE(adapter->down_idx)) {
> +			/*
> +			 * vSwitch freed this and got false positive,
> +			 * need to recover the status and retry.
> +			 */
> +			p->status = MEMNIC_PKT_ST_FREE;
> +			goto retry;
> +		}
> +

The patch indeed looks to improve reliability, even if it's
difficult to me to be sure that there is no other race condition.
Again, I would replace "vSwitch" by "host".

By the way, I guess the Linux code in linux/memnic_net.c should be
modified in the same way.

Regards,
Olivier

Re: [memnic PATCH 1/5] pmd: fix race condition

[Hiroshi Shimamoto]

Hi,

> Subject: Re: [dpdk-dev] [memnic PATCH 1/5] pmd: fix race condition
> 
> Hi Hiroshi-san,
> 
> Please see my comments below.
> 
> On 03/11/2014 06:37 AM, Hiroshi Shimamoto wrote:
> > From: Hiroshi Shimamoto <h-shimamoto-ehU+Cx/zZe18UrSeD/g0lQ@public.gmane.org>
> >
> > There is a race condition, on transmit to vSwitch.
> 
> I think we should not talk specifically about vSwitch, as
> another implementation of host memnic is possible. Maybe using
> the term "host" is more appropriate?
> 
> > +		if (idx != ACCESS_ONCE(adapter->down_idx)) {
> > +			/*
> > +			 * vSwitch freed this and got false positive,
> > +			 * need to recover the status and retry.
> > +			 */
> > +			p->status = MEMNIC_PKT_ST_FREE;
> > +			goto retry;
> > +		}
> > +
> 
> The patch indeed looks to improve reliability, even if it's
> difficult to me to be sure that there is no other race condition.
> Again, I would replace "vSwitch" by "host".

okay, I'm fine with that.
Do you want me resubmit update one?
If so, will do next week.

> 
> By the way, I guess the Linux code in linux/memnic_net.c should be
> modified in the same way.

Hm, yes, we should check kernel driver too.


thanks,
Hiroshi

> 
> Regards,
> Olivier

Re: [memnic PATCH 1/5] pmd: fix race condition

[Thomas Monjalon]

2014-03-28 09:49, Hiroshi Shimamoto:
> Do you want me resubmit update one?
> If so, will do next week.

Yes, please submit a v2.
Thank you
-- 
Thomas