From mboxrd@z Thu Jan  1 00:00:00 1970
From: jszhang@marvell.com (Jisheng Zhang)
Date: Wed, 8 Jul 2015 20:42:18 +0800
Subject: [PATCH 4/4] arm: psci: add cpuidle_ops support
In-Reply-To: <20150708193842.1ca1c55a@xhacker>
References: <1436350417-6712-1-git-send-email-jszhang@marvell.com>
 <1436350417-6712-5-git-send-email-jszhang@marvell.com>
 <20150708103429.GL7557@n2100.arm.linux.org.uk>
 <20150708193842.1ca1c55a@xhacker>
Message-ID: <20150708204218.50c3cfc3@xhacker>
To: linux-arm-kernel@lists.infradead.org
List-Id: linux-arm-kernel.lists.infradead.org

Dear Russell,

On Wed, 8 Jul 2015 19:38:42 +0800
Jisheng Zhang <jszhang@marvell.com> wrote:

> Dear Russell,
> 
> On Wed, 8 Jul 2015 11:34:29 +0100
> Russell King - ARM Linux <linux@arm.linux.org.uk> wrote:
> 
> > On Wed, Jul 08, 2015 at 06:13:37PM +0800, Jisheng Zhang wrote:
> > > This patch implements cpuidle_ops using psci. After this patch, we can use
> > > cpuidle-arm.c with psci backend for both arm and arm64.
> > 
> > I really don't see the point of most of the patches in this series.
> > 
> > To summarise, what you're doing is:
> > 
> > 1. Renaming arch/arm/kernel/psci_smp.c to arch/arm/kernel/psci.c
> > 2. Adding a #ifdef CONFIG_SMP around _all_ the code in psci.c
> > 3. Adding cpuidle code with an #ifdef CONFIG_CPU_IDLE around all the
> >    CPU idle code.
> > 
> > So, we end up with a file which contains:
> > 
> > /*
> >  header
> >  */
> > #include statements
> > 
> > /*
> >  some commentry relevant to SMP code
> >  */
> > #ifdef CONFIG_CPU_IDLE
> > ... cpu idle code ...
> > #endif
> > #ifdef CONFIG_SMP
> > ... smp code ...
> > #endif
> > 
> > which (a) is a mess, and (b) is unnecessary.  The only relevant bits which
> > are shared are the #include statements.
> > 
> > Please try this alternative approach:
> > 
> > 1. Leave psci_smp.c alone.
> > 2. Add arch/arm/kernel/psci_cpuidle.c containing the #include statements
> >    you need, and the CPU idle code.

After more consideration, I have one concern.

Currently, cpuidle_ops is defined as the following,

> struct cpuidle_ops {
>        int (*suspend)(int cpu, unsigned long arg);

the cpu may not be necessary, because to-be-suspended cpu is always the calling
cpu itself.

>	int (*init)(struct device_node *, int cpu);

the device_node here may not be necessary either, because we can get the node
via. of_get_cpu_node.

So if we refine cpuidle_ops defintion, the code would be as simple as

static struct cpuidle_ops psci_cpuidle_ops __initdata = {
        .suspend = cpu_psci_cpu_suspend,
        .init = cpu_psci_cpu_init_idle,
};
CPUIDLE_METHOD_OF_DECLARE(psci_idle, "psci", &psci_cpuidle_ops);

I'm not sure a new psci_cpuidle.c only with above 5 lines is acceptable or not.

Thanks,
Jisheng




> > 
> > I think such an approach will reduce your patch series to two patches,
> > one moving the ARM64 code, and one adding the cpuidle code.
> > 
> 
> Good idea! Will refine the patches as you suggested.
>