Received-SPF: pass (google.com: domain of linux-kernel+bounces-196120-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.80.249 as permitted sender) client-ip=147.75.80.249;
Date: Thu, 30 May 2024 18:07:31 -0700
From: Joe Damato <jdamato@fastly.com>
To: Jakub Kicinski <kuba@kernel.org>, linux-kernel@vger.kernel.org,
	netdev@vger.kernel.org, nalramli@fastly.com,
	"David S. Miller" <davem@davemloft.net>,
	Eric Dumazet <edumazet@google.com>,
	Leon Romanovsky <leon@kernel.org>,
	"open list:MELLANOX MLX5 core VPI driver" <linux-rdma@vger.kernel.org>,
	Paolo Abeni <pabeni@redhat.com>,
	Richard Cochran <richardcochran@gmail.com>,
	Saeed Mahameed <saeedm@nvidia.com>,
	Tariq Toukan <tariqt@nvidia.com>
Subject: Re: [RFC net-next v3 0/2] mlx5: Add netdev-genl queue stats
Message-ID: <Zlki09qJi4h4l5xS@LQ3V64L9R2>
Mail-Followup-To: Joe Damato <jdamato@fastly.com>,
	Jakub Kicinski <kuba@kernel.org>, linux-kernel@vger.kernel.org,
	netdev@vger.kernel.org, nalramli@fastly.com,
	"David S. Miller" <davem@davemloft.net>,
	Eric Dumazet <edumazet@google.com>,
	Leon Romanovsky <leon@kernel.org>,
	"open list:MELLANOX MLX5 core VPI driver" <linux-rdma@vger.kernel.org>,
	Paolo Abeni <pabeni@redhat.com>,
	Richard Cochran <richardcochran@gmail.com>,
	Saeed Mahameed <saeedm@nvidia.com>,
	Tariq Toukan <tariqt@nvidia.com>
References: <20240529031628.324117-1-jdamato@fastly.com>
 <20240530171128.35bd0ee2@kernel.org>
 <ZlkWnXirc-NhQERA@LQ3V64L9R2>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <ZlkWnXirc-NhQERA@LQ3V64L9R2>

On Thu, May 30, 2024 at 05:15:25PM -0700, Joe Damato wrote:
> On Thu, May 30, 2024 at 05:11:28PM -0700, Jakub Kicinski wrote:
> > On Wed, 29 May 2024 03:16:25 +0000 Joe Damato wrote:
> > > Worth noting that Tariq suggested I also export HTB/QOS stats in
> > > mlx5e_get_base_stats.
> > 
> > Why to base, and not report them as queue stats?
> > 
> > Judging by mlx5e_update_tx_netdev_queues() calls sprinkled in
> > ../mlx5/core/en/htb.c it seems that the driver will update the
> > real_num_tx_queues accordingly. And from mlx5e_qid_from_qos()
> > it seems like the inverse calculation is:
> > 
> > i - (chs->params.num_channels + is_ptp)*mlx5e_get_dcb_num_tc(&chs->params)
> > 
> > But really, isn't it enough to use priv->txq2sq[i] for the active
> > queues, and not active ones you've already covered?
> 
> This is what I proposed in the thread for the v2, but Tariq
> suggested a different approach he liked more, please see this
> message for more details:
> 
>   https://lore.kernel.org/netdev/68225941-f3c3-4335-8f3d-edee43f59033@gmail.com/
> 
> I attempted to implement option 1 as he described in his message.
>  
> > > I am open to doing this, but I think if I were to do that, HTB/QOS queue
> > > stats should also be exported by rtnl so that the script above will
> > > continue to show that the output is correct.
> > > 
> > > I'd like to propose: adding HTB/QOS to both rtnl *and* the netdev-genl
> > > code together at the same time, but a later time, separate from this
> > > change.
> > 
> > Hm, are HTB queues really not counted in rtnl? That'd be pretty wrong.
> 
> As far as I can tell (and I could be wrong) I didn't see them
> included in the rtnl stats, but I'll take another look now.

I looked and it seems like the htb stats are included in ethtool's
output if you look at mlx5/core/en_stats.c, it appears that
mlx5e_stats_grp_sw_update_stats_qos rolls up the htb stats.

However, the rtnl stats, seem to be computed in mlx5/core/en_main.c
via mlx5e_get_stats calling mlx5e_fold_sw_stats64, which appears to
gather stats for rx, tx, and ptp, but it doesn't seem like qos/htb
is handled.

Unless I am missing something, I think mlx5e_fold_sw_stats64 would
need code similar to mlx5e_stats_grp_sw_update_stats_qos and then
rtnl would account for htb stats.

That said: since it seems the htb numbers are not included right
now, I was proposing adding that in later both to rtnl and
netdev-genl together, hoping that would keep the proposed
simpler/easier to get accepted.

LMK what you think.