[PATCH] coresight: etm-perf: Add support for PID tracing for kernel at EL2

[Leo Yan]

On Sun, Jan 10, 2021 at 10:34:57PM +0000, Suzuki Kuruppassery Poulose wrote:
> On 1/10/21 1:41 AM, Leo Yan wrote:
> > Hi Suzuki,
> > 
> > On Sun, Jan 10, 2021 at 01:24:30AM +0000, Suzuki Kuruppassery Poulose wrote:
> > 
> > [...]
> > 
> > > +static ssize_t format_attr_pid_show(struct device *dev,
> > > +				    struct device_attribute *attr,
> > > +				    char *page)
> > > +{
> > > +	int pid_fmt = ETM_OPT_CTXTID;
> > > +
> > > +#ifdef CONFIG_CORESIGHT_SOURCE_ETM4X
> > > +	if (is_kernel_in_hyp_mode())
> > > +		pid_fmt  = ETM_OPT_CTXTID_IN_VMID;
> > > +#endif
> > 
> > I have a different idea for this.  As Armv8 Arm describes: "The
> > Virtualization Host Extensions only apply to an implementation that
> > includes EL2 using AArch64.", seems to me it's more reasable to use
> > the condition "#ifdef CONFIG_64BIT", like below:
> > 
> > #ifdef CONFIG_64BIT
> >          if (is_kernel_in_hyp_mode())
> >                  pid_fmt  = ETM_OPT_CTXTID_IN_VMID;
> > #endif
> > 
> > Please let me know if this is better or not.  I will resend the patch
> > series after agreement.
> 
> I did think of that. But CORESIGHT_SOURCE_ETM4X already depends on CONFIG_ARM64.
> So, that is more restrictive in terms of usage. And this wouldn't be the only
> build failure if someone tried enable ETM4X for arm32. So I wouldn't worry about
> someone trying to use ETM4X on arm32 accidentally.

Makes sense, will send v2 with this change.

Thanks for confirmation,
Leo