Re: [PATCH] bonding: don't force LACPDU tx to ~333 ms boundaries

[Paolo Abeni <pabeni@redhat.com>]

On 6/25/25 7:00 PM, Seth Forshee (DigitalOcean) wrote:
> On Wed, Jun 25, 2025 at 09:30:56AM -0700, Jay Vosburgh wrote:
>> Seth Forshee (DigitalOcean) <sforshee@kernel.org> wrote:
>>
>>> The timer which ensures that no more than 3 LACPDUs are transmitted in
>>> a second rearms itself every 333ms regardless of whether an LACPDU is
>>> transmitted when the timer expires. This causes LACPDU tx to be delayed
>>> until the next expiration of the timer, which effectively aligns LACPDUs
>>> to ~333ms boundaries. This results in a variable amount of jitter in the
>>> timing of periodic LACPDUs.
>>
>> 	To be clear, the "3 per second" limitation that all of this
>> should to conform to is from IEEE 802.1AX-2014, 6.4.16 Transmit machine:
>>
>> 	"When the LACP_Enabled variable is TRUE and the NTT (6.4.7)
>> 	variable is TRUE, the Transmit machine shall ensure that a
>> 	properly formatted LACPDU (6.4.2) is transmitted [i.e., issue a
>> 	CtrlMuxN:M_UNITDATA.Request(LACPDU) service primitive], subject
>> 	to the restriction that no more than three LACPDUs may be
>> 	transmitted in any Fast_Periodic_Time interval. If NTT is set to
>> 	TRUE when this limit is in force, the transmission shall be
>> 	delayed until such a time as the restriction is no longer in
>> 	force. The NTT variable shall be set to FALSE when the Transmit
>> 	machine has transmitted a LACPDU."
>>
>> 	The current implementation conforms to this as you describe: by
>> aligning transmission to 1/3 second boundaries, no more than 3 can ever
>> be sent in one second.
>>
>> 	If, hypothetically, the state machine were to transition, or a
>> user updates port settings (either of which would set NTT each time)
>> more than 3 times in a second, would your patched code obey this
>> restriction?
> 
> As long as the transition doesn't reset sm_tx_timer_counter to something
> smaller than ad_ticks_per_sec/AD_MAX_TX_IN_SECOND, which nothing does
> currently (and if it did it would be at risk of sending more than 3 in a
> second already). The timer is reset on each tx, so no two consecutive
> LACPDUs can be sent less than 300ms apart, therefore no more than 3 can
> be per second. If a state machine transition sets NTT within 300ms of
> the previous tx, it will not send another until the timer expires.

@Jay, I believe the above statement is correct. What I'm missing?

Side note: I'm wondering if this should be considered a fix, and thus
requiring targeting the 'net' tree and a 'fixes' tag.

/P