Re: [EXT] Re: [PATCH] arm64: smccc: Add trace events to SMC calls.

[Sudeep Holla]

On Wed, Mar 08, 2023 at 12:25:15PM +0000, Rakesh Babu Saladi wrote:
> Hi Mark, 
> 
> 
>   https://urldefense.proofpoint.com/v2/url?u=https-3A__lore.kernel.org_lkml_20210923112058.GA14893-40C02TD0UTHF1T.local_&d=DwIBAg&c=nKjWec2b6R0mOyPaz7xtfQ&r=OrNjoyjPmqlDfiSr6RMDtAJU4gxiGfgzEtgbqDmoQPA&m=fj6aQCqc0xOYgRMw7BGCVRrOEabi2taRuKl63gCdHras-JfpWCGhjp6HqaLZYj9I&s=1_EJt65j5EoyHhcroJxJFMIkTqQY6jKtWypPIzmgYK8&e= 
>

(Nit: You need to fix your email setup, it is spoiling the links and also
flow for reading in general)


> I don't think anything has changed, and this has all the same problems as
> before, so I do not think we should do this.
>
> Rakesh >> Can you please be more specific why the changes are not getting accepted?
>

Have you gone through the link Mark pointed at ?
For starters, you have new v1.2 specific calls that has x0-x17 as the
input and output registers which neither this patch or the earlier one
addressed if we agree to allow tracing SMCCC calls this way.

> Which SMC calls do you want to trace, and why?
> Rakesh >> These traces we would like to have for debugging purposes while calling any SMC call.
>

The main point was why would you want to enable universal tracing of all SMCCC
calls. Remember that means you would trace the most common and fundamental
for boot PSCI calls which cover CPU_SUSPEND and it gets called a *lot* for
obvious reasons. So it is not a good idea to do so.

So, what are you using this trace for ? Is that a specific driver using
SMCCC ? Can't the trace be added in that driver as you could add specifics
there ?

Taking v1.2 SMCCC as example, though the spec and the API itself allows
18 registers to be input and output, it may not be used in most of the
cases and adding them to the trace is just useless. Hence the suggestion
of making it specific to your use-case.

-- 
Regards,
Sudeep