Re: [PATCH v6 2/2] clocksource: Add Intel Keem Bay timer support

On 11/11/2021 11:42, Sanil, Shruthi wrote:

quoted

-----Original Message-----
From: Thomas Gleixner <redacted>
Sent: Monday, September 27, 2021 3:11 AM
To: Sanil, Shruthi <redacted>; daniel.lezcano@linaro.org;
robh+dt@kernel.org; linux-kernel@vger.kernel.org;
devicetree@vger.kernel.org
Cc: andriy.shevchenko@linux.intel.com; kris.pan@linux.intel.com;
mgross@linux.intel.com; Thokala, Srikanth [off-list ref];
Raja Subramanian, Lakshmi Bai [off-list ref];
Sangannavar, Mallikarjunappa [off-list ref];
Sanil, Shruthi [off-list ref]
Subject: Re: [PATCH v6 2/2] clocksource: Add Intel Keem Bay timer support

On Tue, Sep 07 2021 at 00:06, shruthi sanil wrote:

quoted

+
+/* Provides a unique ID for each timer */ static
+DEFINE_IDA(keembay_timer_ida);

quoted

+
+	timer_id = ida_alloc(&keembay_timer_ida, GFP_KERNEL);
+	if (timer_id < 0) {
+		ret = timer_id;
+		goto err_keembay_ce_to_free;
+	}

May I ask what the purpose of the IDA, which is backed by a full blown
xarray, is here?

AFAICT all you want is a unique number for the timer name for up to 8
timers.

quoted

+	timer_name = kasprintf(GFP_KERNEL, "keembay_timer%d",

timer_id);

So what's wrong about:

static unsigned int keembay_timer_id;

	timer_name = kasprintf(GFP_KERNEL, "keembay_timer%d",
keembay_timer_id++);

Hmm?

Yes, we had initially implemented it in the similar way, 
but in the course of review it got changed to use IDA.

quoted

quoted

+
+	clockevents_config_and_register(&keembay_ce_to->clkevt,
+					timer_of_rate(keembay_ce_to),
+					1,
+					U32_MAX);

Aside of that what's the point of registering more than one of those timers as
clock event? The core will only use one and the rest is just going to use
memory for no value.

Instead of
keembay_ce_to->clkevt.cpumask = cpumask_of(0); 
can I update it as 
keembay_ce_to->clkevt.cpumask = cpu_possible_mask; 
so that each timer would be associated with different cores?

Let me try to clarify:

The Intel Keem bay Soc is a 4 x Cortex-A53

The arch ARM timer is per CPU on this platform.

Case 1:
-------
 - the architected timer is not desired and this timer is wanted to be
used instead (but rating tells the opposite) => rewrite per cpu code

Case 2:
-------
 - the architected timer are desired and this timer is used as a
broadcast timer when a core is going done with cpuidle. One timer is needed.

 - In order to prevent useless wakeup, the timer uses the flag DYNIRQ.
However, cpumask_of(0) is set and makes inoperative this flag. In order
to make full use of it, clkevt.cpumask must be cpu_possible_mask

Hope that helps

  -- Daniel








-- 
<http://www.linaro.org/> Linaro.org │ Open source software for ARM SoCs

Follow Linaro:  <http://www.facebook.com/pages/Linaro> Facebook |
<http://twitter.com/#!/linaroorg> Twitter |
<http://www.linaro.org/linaro-blog/> Blog