Re: [PATCH v2 05/28] coresight: etm4x: Ensure context synchronization is not ignored

[Leo Yan <leo.yan@arm.com>]

Hi Levi,

On Wed, Jul 02, 2025 at 12:10:17PM +0100, Yeoreum Yun wrote:

[...]

> > @@ -579,6 +572,13 @@ static int etm4_enable_hw(struct etmv4_drvdata *drvdata)
> >
> >  	if (!drvdata->paused)
> >  		rc = etm4_enable_trace_unit(drvdata);
> > +
> > +	/*
> > +	 * As recommended by section 4.3.7 (Synchronization of register updates)
> > +	 * of ARM IHI 0064H.b, the self-hosted trace analyzer always executes an
> > +	 * ISB instruction after programming the trace unit registers.
> > +	 */
> > +	isb();
> 
> But according to 4.3.7 ("Synchronization when using memory-mapped
> interface"), doesn't it need to dsb like:
> 
>   if (csa->iomem)
>     dsb(sy);
>   isb();
> 
> Or am I missing something?

Section 4.3.7 suggests using a DSB barrier to ensure that writes have
completed in MMIO mode. It also mentions an alternative:

"If the memory is marked as Device-nGnRE or stronger, read back the
value of any register in the trace unit. This relies on the peripheral
coherence order defined in the Arm architecture."

In the etm4_{enable|disable}_trace_unit() functions, each time the
TRCPRGCTLR register is written, the driver polls bits in TRCSTATR.
This acts as synchronization using read-after-write (RAW), which is
exactly the approach suggested above.

This is why we don't need DSB() anymore.

Thanks,
Leo