Re: [PATCH net-next v10 2/4] ethtool: provide customized dim profile management

From: Jakub Kicinski <kuba@kernel.org>
To: Heng Qi <hengqi@linux.alibaba.com>
Cc: netdev@vger.kernel.org, virtualization@lists.linux.dev,
	"David S . Miller" <davem@davemloft.net>,
	Paolo Abeni <pabeni@redhat.com>,
	Eric Dumazet <edumazet@google.com>,
	Jason Wang <jasowang@redhat.com>,
	"Michael S . Tsirkin" <mst@redhat.com>,
	Brett Creeley <bcreeley@amd.com>,
	Ratheesh Kannoth <rkannoth@marvell.com>,
	Alexander Lobakin <aleksander.lobakin@intel.com>,
	Xuan Zhuo <xuanzhuo@linux.alibaba.com>,
	Tal Gilboa <talgi@nvidia.com>, Jonathan Corbet <corbet@lwn.net>,
	linux-doc@vger.kernel.org,
	Maxime Chevallier <maxime.chevallier@bootlin.com>,
	Jiri Pirko <jiri@resnulli.us>,
	Paul Greenwalt <paul.greenwalt@intel.com>,
	Ahmed Zaki <ahmed.zaki@intel.com>,
	Vladimir Oltean <vladimir.oltean@nxp.com>,
	Kory Maincent <kory.maincent@bootlin.com>,
	Andrew Lunn <andrew@lunn.ch>,
	"justinstitt @ google . com" <justinstitt@google.com>
Subject: Re: [PATCH net-next v10 2/4] ethtool: provide customized dim profile management
Date: Mon, 29 Apr 2024 10:47:41 -0700	[thread overview]
Message-ID: <20240429104741.3a628fe6@kernel.org> (raw)
In-Reply-To: <98ea9d4d-1a90-45b9-a4e0-6941969295be@linux.alibaba.com>

On Sun, 28 Apr 2024 22:49:09 +0800 Heng Qi wrote:
> >> +#if IS_ENABLED(CONFIG_DIMLIB)
> >> +	if (dev->irq_moder) {  
> > This may be NULL  
> 
> 
> Do you mean dev may be null or dev->irq_moder may be null?
> The former has been excluded (see const struct ethtool_ops *ops
> 
> = req_info->dev->ethtool_ops;).
> 
> And we are ruling out the latter using 'if (dev->irq_moder)'.
> 
> Or something else?

Hm, I must have misread something here.

> >> +	nla_for_each_nested_type(nest, ETHTOOL_A_PROFILE_IRQ_MODERATION, nests, rem) {
> >> +		ret = nla_parse_nested(moder,
> >> +				       ARRAY_SIZE(coalesce_irq_moderation_policy) - 1,
> >> +				       nest, coalesce_irq_moderation_policy,
> >> +				       extack);
> >> +		if (ret)
> >> +			return ret;
> >> +
> >> +		if (!NL_REQ_ATTR_CHECK(extack, nest, moder, ETHTOOL_A_IRQ_MODERATION_USEC)) {
> >> +			if (irq_moder->coal_flags & DIM_COALESCE_USEC)  
> > There are 3 options here, not 2:
> >
> > 	if (irq_moder->coal_flags & flag) {
> > 		if (NL_REQ_ATTR_CHECK())
> > 			val = nla_get_u32(...);
> > 		else
> > 			return -EINVAL;
> > 	} else {
> > 		if (moder[attr_type)) {
> > 			BAD_ATTR()
> > 			return -EOPNOTSUPP;
> > 		}
> > 	}  
> 
> Maybe we missed something.
> 
> As shown in the commit log, the user is allowed to modify only
> a certain fields in irq-moderation. It is assumed that the driver
> supports modification of usec and pkts, but the user may only
> modify usec and only fill in the usec attr.
> 
> Therefore, the kernel only gets usec attr here. Of course, the user
> may have passed in 5 groups of "n, n, n", which means that nothing
> is modified, and rx_profile and irq_moderation attrs are all empty.

What you describe sounds good, but it's not what the code seems to be
doing. NL_REQ_ATTR_CHECK() will set an error attribute if the attr is
not present.