Re: [PATCH RFC] firmware: arm_scmi: Allow for better protocol extensibility

From: Sudeep Holla <sudeep.holla@arm.com>
To: Florian Fainelli <f.fainelli@gmail.com>
Cc: linux-arm-kernel@lists.infradead.org, james.quinlan@broadcom.com,
	bcm-kernel-feedback-list@broadcom.com,
	Michael Turquette <mturquette@baylibre.com>,
	Stephen Boyd <sboyd@kernel.org>,
	"Rafael J. Wysocki" <rjw@rjwysocki.net>,
	Viresh Kumar <viresh.kumar@linaro.org>,
	Jean Delvare <jdelvare@suse.com>,
	Guenter Roeck <linux@roeck-us.net>,
	"open list:COMMON CLK FRAMEWORK" <linux-clk@vger.kernel.org>,
	open list <linux-kernel@vger.kernel.org>,
	"open list:CPU FREQUENCY SCALING FRAMEWORK" 
	<linux-pm@vger.kernel.org>,
	"open list:HARDWARE MONITORING" <linux-hwmon@vger.kernel.org>,
	Sudeep Holla <sudeep.holla@arm.com>
Subject: Re: [PATCH RFC] firmware: arm_scmi: Allow for better protocol extensibility
Date: Fri, 24 May 2019 15:19:19 +0100	[thread overview]
Message-ID: <20190524141918.GA4408@e107155-lin> (raw)
In-Reply-To: <f9403e7e-1b87-dc46-dfc5-62227c659e7c@gmail.com>

On Thu, May 23, 2019 at 10:17:50AM -0700, Florian Fainelli wrote:
> On 5/21/19 1:01 PM, Florian Fainelli wrote:
> > The SCMI specific allows implementors to define their custom protocols
> > in the 0x80-0xFF space. The current scmi_handle structure requires us to
> > extend the structure with a set of operations and associated private
> > data in a way that is not quite scaling well.
> > 
> > Create a 255 bytes structure that contains an opaque pointer to a set of
> > operations and private data and create two helper functions to retrieve
> > those based on the protocol identifier. Several options were considered,
> > like using a linked list but since we could be performance sensitive in
> > some paths, using an array was faster and simpler.
> > 
> > Convert all call sites to use either scmi_protocol_get_ops() or
> > scmi_protocol_get_info().
> > 
> > Signed-off-by: Florian Fainelli <f.fainelli@gmail.com>
> 
> On second thought, what I really need is private storage to the scmi_dev
> (the consumer side), and not so much the protocol (provider) side.
> Therefore using dev_{set,get}_drvadata() against scmi_device::dev should
> be working just fine, and if we are concerned about another part of the
> SCMI stack making use of that storage, we can always extend struct
> scmi_device with a private cookie.

Sorry, I haven't looked into the original patch in detail yet. But I
always have rejected to add support for just infrastructure to add
vendor specific protocols both internally @ARM and to some private
emails I have received. I prefer to merge it with the first reference
vendor specific protocol so that the users of this infrastructure gets
a fair idea on how to use the same.

I will look at the RFC next week.

--
Regards,
Sudeep