Received-SPF: pass (google.com: domain of linux-kernel+bounces-95903-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) client-ip=2604:1380:45e3:2400::1;
Date: Thu, 7 Mar 2024 09:01:45 -0800
From: Jakub Kicinski <kuba@kernel.org>
To: Haiyang Zhang <haiyangz@microsoft.com>
Cc: Shradha Gupta <shradhagupta@linux.microsoft.com>, Shradha Gupta
 <shradhagupta@microsoft.com>, "linux-kernel@vger.kernel.org"
 <linux-kernel@vger.kernel.org>, "linux-hyperv@vger.kernel.org"
 <linux-hyperv@vger.kernel.org>, "linux-rdma@vger.kernel.org"
 <linux-rdma@vger.kernel.org>, "netdev@vger.kernel.org"
 <netdev@vger.kernel.org>, Eric Dumazet <edumazet@google.com>, Paolo Abeni
 <pabeni@redhat.com>, Ajay Sharma <sharmaajay@microsoft.com>, Leon
 Romanovsky <leon@kernel.org>, Thomas Gleixner <tglx@linutronix.de>,
 Sebastian Andrzej Siewior <bigeasy@linutronix.de>, KY Srinivasan
 <kys@microsoft.com>, Wei Liu <wei.liu@kernel.org>, Dexuan Cui
 <decui@microsoft.com>, Long Li <longli@microsoft.com>, Michael Kelley
 <mikelley@microsoft.com>
Subject: Re: [PATCH] net :mana : Add per-cpu stats for MANA device
Message-ID: <20240307090145.2fc7aa2e@kernel.org>
In-Reply-To: <DM6PR21MB14817597567C638DEF020FE3CA202@DM6PR21MB1481.namprd21.prod.outlook.com>
References: <1709823132-22411-1-git-send-email-shradhagupta@linux.microsoft.com>
	<20240307072923.6cc8a2ba@kernel.org>
	<DM6PR21MB14817597567C638DEF020FE3CA202@DM6PR21MB1481.namprd21.prod.outlook.com>
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=US-ASCII
Content-Transfer-Encoding: 7bit

On Thu, 7 Mar 2024 15:49:15 +0000 Haiyang Zhang wrote:
> > > Extend 'ethtool -S' output for mana devices to include per-CPU packet
> > > stats  
> > 
> > But why? You already have per queue stats.  
> Yes. But the q to cpu binding is dynamic, we also want the per-CPU stat 
> to analyze the CPU usage by counting the packets and bytes on each CPU.

Dynamic is a bit of an exaggeration, right? On a well-configured system
each CPU should use a single queue assigned thru XPS. And for manual
debug bpftrace should serve the purpose quite well.

Please note that you can't use num_present_cpus() to size stats in
ethtool -S , you have to use possible_cpus(), because the retrieval
of the stats is done in a multi-syscall fashion and there are no
explicit lengths in the API. So you must always report all possible
stats, not just currently active :(