Re: [PATCH net-next 5/5] net: dsa: mv88e6xxx: implementation of dynamic ATU entries

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



On Sat, Feb 04, 2023 at 09:48:24AM +0100, netdev@xxxxxxxxxxxxxxxxxxxx wrote:
> On 2023-02-04 09:12, Simon Horman wrote:
> > On Fri, Feb 03, 2023 at 10:44:22PM +0200, Vladimir Oltean wrote:
> > > On Fri, Feb 03, 2023 at 09:20:22AM +0100, Simon Horman wrote:
> > > > > else if (someflag)
> > > > >         dosomething();
> > > > >
> > > > > For now only one flag will actually be set and they are mutually exclusive,
> > > > > as they will not make sense together with the potential flags I know, but
> > > > > that can change at some time of course.
> > > >
> > > > Yes, I see that is workable. I do feel that checking for other flags would
> > > > be a bit more robust. But as you say, there are none. So whichever
> > > > approach you prefer is fine by me.
> > > 
> > > The model we have for unsupported bits in the
> > > SWITCHDEV_ATTR_ID_PORT_PRE_BRIDGE_FLAGS
> > > and SWITCHDEV_ATTR_ID_PORT_BRIDGE_FLAGS handlers is essentially this:
> > > 
> > > 	if (flags & ~(supported_flag_mask))
> > > 		return -EOPNOTSUPP;
> > > 
> > > 	if (flags & supported_flag_1)
> > > 		...
> > > 
> > > 	if (flags & supported_flag_2)
> > > 		...
> > > 
> > > I suppose applying this model here would address Simon's
> > > extensibility concern.
> > 
> > Yes, that is the model I had in mind.
> 
> The only thing is that we actually need to return both 0 and -EOPNOTSUPP for
> unsupported flags. The dynamic flag requires 0 when not supported (and
> supported) AFAICS.
> Setting a mask as 'supported' for a feature that is not really supported
> defeats the notion of 'supported' IMHO.

Just to clarify my suggestion one last time, it would be along the lines
of the following (completely untested!). I feel that it robustly covers
all cases for fdb_flags. And as a bonus doesn't need to be modified
if other (unsupported) flags are added in future.

	if (fdb_flags & ~(DSA_FDB_FLAG_DYNAMIC))
		return -EOPNOTSUPP;

	is_dynamic = !!(fdb_flags & DSA_FDB_FLAG_DYNAMIC)
	if (is_dynamic)
		state = MV88E6XXX_G1_ATU_DATA_STATE_UC_AGE_7_NEWEST;


And perhaps for other drivers:

	if (fdb_flags & ~(DSA_FDB_FLAG_DYNAMIC))
		return -EOPNOTSUPP;
	if (fdb_flags)
		return 0;

Perhaps a helper would be warranted for the above.

But in writing this I think that, perhaps drivers could return -EOPNOTSUPP
for the DSA_FDB_FLAG_DYNAMIC case and the caller can handle, rather tha
propagate, -EOPNOTSUPP.

Returning -EOPNOTSUPP is the normal way to drivers to respond to requests
for unsupported hardware offloads. Sticking to that may be clearner
in the long run. That said, I do agree your current patch is correct
given the flag that is defined (by your patchset).



[Index of Archives]     [Linux Samsung SOC]     [Linux Wireless]     [Linux Kernel]     [ATH6KL]     [Linux Bluetooth]     [Linux Netdev]     [Kernel Newbies]     [IDE]     [Security]     [Git]     [Netfilter]     [Bugtraq]     [Yosemite News]     [MIPS Linux]     [ARM Linux]     [Linux Security]     [Linux RAID]     [Linux ATA RAID]     [Samba]     [Device Mapper]

  Powered by Linux