Re: [PATCH] octeontx2-pf: Add error handling for cn10k_map_unmap_rq_policer().

[Simon Horman]

On Mon, Apr 07, 2025 at 05:57:14AM +0000, Subbaraya Sundeep wrote:
> On 2025-04-04 at 11:11:38, Simon Horman (horms@xxxxxxxxxx) wrote:
> > On Fri, Apr 04, 2025 at 05:22:16AM +0000, Subbaraya Sundeep Bhatta wrote:
> > > Hi,
> > > 
> > > From: Wentao Liang <vulab@xxxxxxxxxxx> 
> > > Sent: Thursday, April 3, 2025 8:43 PM
> > > To: Sunil Kovvuri Goutham <sgoutham@xxxxxxxxxxx>; Geethasowjanya Akula <gakula@xxxxxxxxxxx>; Subbaraya Sundeep Bhatta <sbhatta@xxxxxxxxxxx>; Hariprasad Kelam <hkelam@xxxxxxxxxxx>; andrew+netdev@xxxxxxx; davem@xxxxxxxxxxxxx; edumazet@xxxxxxxxxx; kuba@xxxxxxxxxx; pabeni@xxxxxxxxxx
> > > Cc: netdev@xxxxxxxxxxxxxxx; linux-kernel@xxxxxxxxxxxxxxx; Wentao Liang <vulab@xxxxxxxxxxx>
> > > Subject: [PATCH] octeontx2-pf: Add error handling for cn10k_map_unmap_rq_policer().
> > > 
> > > The cn10k_free_matchall_ipolicer() calls the cn10k_map_unmap_rq_policer()
> > > for each queue in a for loop without checking for any errors. A proper
> > > implementation can be found in cn10k_set_matchall_ipolicer_rate().
> > > 
> > > Check the return value of the cn10k_map_unmap_rq_policer() function during
> > > each loop. Jump to unlock function and return the error code if the
> > > funciton fails to unmap policer.
> > > 
> > > Fixes: 2ca89a2c3752 ("octeontx2-pf: TC_MATCHALL ingress ratelimiting offload")
> > > Signed-off-by: Wentao Liang <mailto:vulab@xxxxxxxxxxx>
> > > ---
> > >  drivers/net/ethernet/marvell/octeontx2/nic/cn10k.c | 6 ++++--
> > >  1 file changed, 4 insertions(+), 2 deletions(-)
> > > 
> > > diff --git a/drivers/net/ethernet/marvell/octeontx2/nic/cn10k.c b/drivers/net/ethernet/marvell/octeontx2/nic/cn10k.c
> > > index a15cc86635d6..ce58ad61198e 100644
> > > --- a/drivers/net/ethernet/marvell/octeontx2/nic/cn10k.c
> > > +++ b/drivers/net/ethernet/marvell/octeontx2/nic/cn10k.c
> > > @@ -353,11 +353,13 @@ int cn10k_free_matchall_ipolicer(struct otx2_nic *pfvf)
> > >  
> > >  	/* Remove RQ's policer mapping */
> > >  	for (qidx = 0; qidx < hw->rx_queues; qidx++)
> > > -		cn10k_map_unmap_rq_policer(pfvf, qidx,
> > > -					   hw->matchall_ipolicer, false);
> > > +		rc = cn10k_map_unmap_rq_policer(pfvf, qidx, hw->matchall_ipolicer, false);
> > > +		if (rc)
> > > +			goto out;
> > >  
> > > Intentionally we do not bail out when unmapping one of the queues is failed. The reason is during teardown if one of the queues is failed then
> > > we end up not tearing down rest of the queues and those queues cannot be used later which is bad. So leave whatever queues have failed and proceed
> > > with tearing down the rest. Hence all we can do is print an error for the failed queue and continue.
> > 
> > Hi Sundeep,
> > 
> > Sorry that I didn't notice your response before sending my own to Wentao.
> > 
> > I do agree that bailing out here is not a good idea.  But I wonder if there
> > is any value in the function should propagate some error reporting if any
> > call to cn10k_map_unmap_rq_policer fails - e.g. the first failure - while
> > still iterating aver all elements.
> > 
> > Just an idea.
> > 
> Hi Simon,
> 
> We can do but it gets compilcated if more than one queue failed and
> reasons are different. Hence just print error and continue.

Yes, understood. I did think about this some more and I think that
reporting an error is the best we can do, else as you say it gets complex.
So if there is already a log made, then I think that is sufficient.