RE: [EXTERNAL] Re: [PATCH V2,net-next, 1/2] net: mana: Add support for coalesced RX packets on CQE

[Haiyang Zhang]

> -----Original Message-----
> From: Jakub Kicinski <kuba@xxxxxxxxxx>
> Sent: Monday, January 12, 2026 8:22 PM
> To: Haiyang Zhang <haiyangz@xxxxxxxxxxxxx>
> Cc: Haiyang Zhang <haiyangz@xxxxxxxxxxxxxxxxxxx>; linux-
> hyperv@xxxxxxxxxxxxxxx; netdev@xxxxxxxxxxxxxxx; KY Srinivasan
> <kys@xxxxxxxxxxxxx>; Wei Liu <wei.liu@xxxxxxxxxx>; Dexuan Cui
> <DECUI@xxxxxxxxxxxxx>; Long Li <longli@xxxxxxxxxxxxx>; Andrew Lunn
> <andrew+netdev@xxxxxxx>; David S. Miller <davem@xxxxxxxxxxxxx>; Eric
> Dumazet <edumazet@xxxxxxxxxx>; Paolo Abeni <pabeni@xxxxxxxxxx>; Konstantin
> Taranov <kotaranov@xxxxxxxxxxxxx>; Simon Horman <horms@xxxxxxxxxx>; Erni
> Sri Satya Vennela <ernis@xxxxxxxxxxxxxxxxxxx>; Shradha Gupta
> <shradhagupta@xxxxxxxxxxxxxxxxxxx>; Saurabh Sengar
> <ssengar@xxxxxxxxxxxxxxxxxxx>; Aditya Garg
> <gargaditya@xxxxxxxxxxxxxxxxxxx>; Dipayaan Roy
> <dipayanroy@xxxxxxxxxxxxxxxxxxx>; Shiraz Saleem
> <shirazsaleem@xxxxxxxxxxxxx>; linux-kernel@xxxxxxxxxxxxxxx; linux-
> rdma@xxxxxxxxxxxxxxx; Paul Rosswurm <paulros@xxxxxxxxxxxxx>
> Subject: Re: [EXTERNAL] Re: [PATCH V2,net-next, 1/2] net: mana: Add
> support for coalesced RX packets on CQE
> 
> On Mon, 12 Jan 2026 21:01:59 +0000 Haiyang Zhang wrote:
> > > > Our NIC can have up to 4 RX packets on 1 CQE. To support this
> feature,
> > > > check and process the type CQE_RX_COALESCED_4. The default setting
> is
> > > > disabled, to avoid possible regression on latency.
> > > >
> > > > And add ethtool handler to switch this feature. To turn it on, run:
> > > >   ethtool -C <nic> rx-frames 4
> > > > To turn it off:
> > > >   ethtool -C <nic> rx-frames 1
> > >
> > > Exposing just rx frame count, and only two values is quite unusual.
> > > Please explain in more detail the coalescing logic of the device.
> > Our NIC device only supports coalescing on RX. And when it's disabled
> each
> > RX CQE indicates 1 RX packet; when enabled each RX CQE indicates up to 4
> packets.
> 
> I get that. What is the logic for combining 4 packets into a single
> completion? How does it work? Your commit message mentions "regression
> on latency" - what is the bound on that regression?

When we received CQE type CQE_RX_COALESCED_4, it's a coalesced CQE. And in the CQE
OOB, there is an array with 4 PPI elements, with each pkt's length:
oob->ppi[i].pkt_len.

So we read the related WQE and the DMA buffers for the RX pkt payloads, up to 4.
But, if the coalesced pkts <4, the pkt_len will be 0 after the last pkt, so we
know when to stop reading the WQEs.

Thanks,
- Haiyang