Re: [RFC PATCH 1/2] arm64/tracing: add cntvct based trace clock

Re: [RFC PATCH 1/2] arm64/tracing: add cntvct based trace clock

[Steven Rostedt]

On Fri, 19 Nov 2021 11:21:17 +0100
Nicolas Saenz Julienne <nsaenzju@redhat.com> wrote:

> +/* SPDX-License-Identifier: GPL-2.0 */
> +#ifndef _ASM_ARM64_TRACE_CLOCK_H
> +#define _ASM_ARM64_TRACE_CLOCK_H
> +
> +#include <linux/types.h>
> +
> +extern u64 notrace trace_clock_arm64_cntvct(void);
> +
> +# define ARCH_TRACE_CLOCKS \
> +	{ trace_clock_arm64_cntvct, "cntvct", .in_ns = 0 },
> +
> +#endif  /* _ASM_ARM64_TRACE_CLOCK_H */

So this will appear as a usable clock in trace-cmd.

And since this will be used to synchronize between host and guest like the
x86_tsc is used, that means that trace-cmd needs to know that this is the
an arch "CPU" clock. I wonder if we should rename x86_clock (or at least
make it an alias) to "kvm_clock". Then we can have trace-cmd use
"kvm_clock" as the clock for synchronization between host and guests for
all architectures?

Thinking about this, instead of renaming it, I'll add code to create an
alias to these clocks. Then every arch can pick what clock is used that is
the same between hosts and guests such that user space tooling doesn't have
to keep a database of what clocks are used for synchronization between
hosts and guests for each arch.

I'll go add some code ;-)

-- Steve

Re: [RFC PATCH 1/2] arm64/tracing: add cntvct based trace clock

[Nicolas Saenz Julienne]

On Mon, 2021-11-22 at 09:57 -0500, Steven Rostedt wrote:
> On Fri, 19 Nov 2021 11:21:17 +0100
> Nicolas Saenz Julienne <nsaenzju@redhat.com> wrote:
> 
> > +/* SPDX-License-Identifier: GPL-2.0 */
> > +#ifndef _ASM_ARM64_TRACE_CLOCK_H
> > +#define _ASM_ARM64_TRACE_CLOCK_H
> > +
> > +#include <linux/types.h>
> > +
> > +extern u64 notrace trace_clock_arm64_cntvct(void);
> > +
> > +# define ARCH_TRACE_CLOCKS \
> > +	{ trace_clock_arm64_cntvct, "cntvct", .in_ns = 0 },
> > +
> > +#endif  /* _ASM_ARM64_TRACE_CLOCK_H */
> 
> So this will appear as a usable clock in trace-cmd.
> 
> And since this will be used to synchronize between host and guest like the
> x86_tsc is used, that means that trace-cmd needs to know that this is the
> an arch "CPU" clock. I wonder if we should rename x86_clock (or at least
> make it an alias) to "kvm_clock". Then we can have trace-cmd use
> "kvm_clock" as the clock for synchronization between host and guests for
> all architectures?
>
> Thinking about this, instead of renaming it, I'll add code to create an
> alias to these clocks. Then every arch can pick what clock is used that is
> the same between hosts and guests such that user space tooling doesn't have
> to keep a database of what clocks are used for synchronization between
> hosts and guests for each arch.
> 
> I'll go add some code ;-)

I really like the idea, please keep me in the loop if you send something
upstream.

-- 
Nicolás Sáenz