On Fri, Feb 21, 2025 at 03:42:05AM +0200, Wei Fang wrote:
I'm not sure "correct the statistics" is the best way to describe this change. Maybe "keep track of correct TXBD count in enetc_map_tx_tso_buffs()"?
Hi Vladimir,
Inspired by Michal, I think we don't need to keep the count variable, because we already have index "i", we just need to record the value of the initial i at the beginning. So I plan to do this optimization on the net-next tree in the future. So I don't think it is necessary to modify enetc_map_tx_tso_hdr().
You are saying this in reply to my observation that the title of the change does not correctly represent the change. But I don't see how what you're saying is connected to that. Currently there exists a "count" variable based on which TX BDs are unmapped, and these patches are not changing that fact.
stylistic nitpick: I think this implementation choice obscures the fact, to an unfamiliar reader, that the requirement for an extended TXBD comes from enetc_map_tx_tso_hdr(). This is because you repeat the condition for skb_vlan_tag_present(), but it's not obvious it's correlated to the other one. Something like the change below is more expressive in this regard, in my opinion:
It seems you were objecting to this other change suggestion instead. Sure, I mean, ignore it if you want, but you're saying "well I'm going to change the scheme for net-next, so no point in making the code more obviously correct in stable branches", but the stable branches aren't going to pick up the net-next patch - they are essentially frozen except for bug fixes. I would still recommend writing code that makes the most sense for stable (to the extent that this is logically part of fixing a bug), and then writing code that makes most sense for net-next, even if it implies changing some of it back the way it was.