Hi,
It took a while but I have finally observed some of the races between
NAPI poll and .ndo_start_xmit in e1000 I was talking about. That is,
e1000_clean VS e1000_xmit_frame. To reproduce the races, I added the
delays in the appropriate places of e1000 and used hardware breakpoints
to track a
000-devel
Sent: Fri, 02 Aug 2013 21:54:41 +0400 (MSD)
Subject: RE: [E1000-devel] Possible races in e1000
Have you seen an actual problem where this has really happened? It's unclear
to me that this can happen with how the lo
3 8:27 AM
> To: e1000-devel
> Subject: [E1000-devel] Possible races in e1000
>
> Hi,
>
> There seem to be several data races in e1000 driver (checked a couple
> of usage scenarios on the vanilla Linux kernel 3.11-rc1). Perhaps, not
> that critical but still worth fixi
Hi,
There seem to be several data races in e1000 driver (checked a couple of
usage scenarios on the vanilla Linux kernel 3.11-rc1). Perhaps, not that
critical but still worth fixing, I suppose.
1.
CPU0 CPU1
e1000_probe()
|
register_netdev()
|
|
