Re: [PATCH 27/41]bnx2x: smp_mb and not just smp_rmb

["Eilon Greenstein" <eilong@broadcom.com>]

On Thu, 2009-02-12 at 11:21 -0800, Ben Hutchings wrote:
> On Thu, 2009-02-12 at 20:37 +0200, Eilon Greenstein wrote:
> > Subject: [PATCH 27/41]bnx2x: smp_mb and not just smp_rmb
> 
> This could do with some more explanation (and comments in the code).
Sorry about that, I got lazy at this part.
Since bnx2x_tx_int is also writing, the smb_rmb should be replaced with
smp_mb.

> 
> > Signed-off-by: Eilon Greenstein <eilong@broadcom.com>
> > ---
> >  drivers/net/bnx2x_main.c |    4 ++--
> >  1 files changed, 2 insertions(+), 2 deletions(-)
> > 
> > diff --git a/drivers/net/bnx2x_main.c b/drivers/net/bnx2x_main.c
> > index 6bd9219..2e95799 100644
> > --- a/drivers/net/bnx2x_main.c
> > +++ b/drivers/net/bnx2x_main.c
> > @@ -7173,7 +7173,7 @@ static int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode)
> >  		struct bnx2x_fastpath *fp = &bp->fp[i];
> >  
> >  		cnt = 1000;
> > -		smp_rmb();
> > +		smp_mb();
> >  		while (bnx2x_has_tx_work_unload(fp)) {
> >  
> >  			bnx2x_tx_int(fp, 1000);
> > @@ -7189,7 +7189,7 @@ static int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode)
> >  			}
> >  			cnt--;
> >  			msleep(1);
> > -			smp_rmb();
> > +			smp_mb();
> 
> I suspect that smp_*mb() is redundant after a schedule().
Hmmm... It seems so... I will have to run it on few different systems in
the lab to make sure - if I won't run into any problems I will send a
new patch on top of this one with the removal of this alleged redundant
smb_mb.

Thanks Ben!

> 
> >  		}
> >  	}
> >  	/* Give HW time to discard old tx messages */
> 
> Ben.
>