Re: [PATCH] intel_idle: set the state_tables array into __initdata to save mem

* Re: [PATCH] intel_idle: set the state_tables array into __initdata to save mem
  2013-03-07 16:42 [PATCH] intel_idle: set the state_tables array into __initdata to save mem Chuansheng Liu
@ 2013-03-07  9:49 ` Daniel Lezcano
  2013-03-08  0:44     ` Liu, Chuansheng
  2013-03-08 15:02 ` [PATCH 0/3] " Chuansheng Liu
  1 sibling, 1 reply; 15+ messages in thread
From: Daniel Lezcano @ 2013-03-07  9:49 UTC (permalink / raw)
  To: Chuansheng Liu; +Cc: lenb, len.brown, linux-pm, linux-kernel

On 03/07/2013 05:42 PM, Chuansheng Liu wrote:
> 
> Currently, in intel_idle.c, there are 5 state_tables array, every
> array size is sizeof(struct cpuidle_state) * CPUIDLE_STATE_MAX.
> 
> But after intel_idle_probe(), just only one array is useful.
> 
> Here we can just define one static state_table, and initialize it
> in intel_idle_probe(), and set other data state_tables as __initdata.
> 
> It can save about 3K, which also benefits mobile devices.
> 
> Signed-off-by: liu chuansheng <chuansheng.liu@intel.com>
> ---
>  drivers/idle/intel_idle.c |   19 ++++++++++++-------
>  1 files changed, 12 insertions(+), 7 deletions(-)
> 
> diff --git a/drivers/idle/intel_idle.c b/drivers/idle/intel_idle.c
> index 5d66750..0642bfe 100644
> --- a/drivers/idle/intel_idle.c
> +++ b/drivers/idle/intel_idle.c
> @@ -99,7 +99,7 @@ static int intel_idle(struct cpuidle_device *dev,
>  			struct cpuidle_driver *drv, int index);
>  static int intel_idle_cpu_init(int cpu);
>  
> -static struct cpuidle_state *cpuidle_state_table;
> +static struct cpuidle_state cpuidle_state_table[CPUIDLE_STATE_MAX];
>  
>  /*
>   * Set this flag for states where the HW flushes the TLB for us
> @@ -124,7 +124,7 @@ static struct cpuidle_state *cpuidle_state_table;
>   * which is also the index into the MWAIT hint array.
>   * Thus C0 is a dummy.
>   */
> -static struct cpuidle_state nehalem_cstates[CPUIDLE_STATE_MAX] = {
> +static struct cpuidle_state nehalem_cstates[CPUIDLE_STATE_MAX] __initdata = {
>  	{
>  		.name = "C1-NHM",
>  		.desc = "MWAIT 0x00",
> @@ -157,7 +157,7 @@ static struct cpuidle_state nehalem_cstates[CPUIDLE_STATE_MAX] = {
>  		.enter = NULL }
>  };
>  
> -static struct cpuidle_state snb_cstates[CPUIDLE_STATE_MAX] = {
> +static struct cpuidle_state snb_cstates[CPUIDLE_STATE_MAX] __initdata = {
>  	{
>  		.name = "C1-SNB",
>  		.desc = "MWAIT 0x00",
> @@ -197,7 +197,7 @@ static struct cpuidle_state snb_cstates[CPUIDLE_STATE_MAX] = {
>  		.enter = NULL }
>  };
>  
> -static struct cpuidle_state ivb_cstates[CPUIDLE_STATE_MAX] = {
> +static struct cpuidle_state ivb_cstates[CPUIDLE_STATE_MAX] __initdata = {
>  	{
>  		.name = "C1-IVB",
>  		.desc = "MWAIT 0x00",
> @@ -237,7 +237,7 @@ static struct cpuidle_state ivb_cstates[CPUIDLE_STATE_MAX] = {
>  		.enter = NULL }
>  };
>  
> -static struct cpuidle_state hsw_cstates[CPUIDLE_STATE_MAX] = {
> +static struct cpuidle_state hsw_cstates[CPUIDLE_STATE_MAX] __initdata = {
>  	{
>  		.name = "C1-HSW",
>  		.desc = "MWAIT 0x00",
> @@ -277,7 +277,7 @@ static struct cpuidle_state hsw_cstates[CPUIDLE_STATE_MAX] = {
>  		.enter = NULL }
>  };
>  
> -static struct cpuidle_state atom_cstates[CPUIDLE_STATE_MAX] = {
> +static struct cpuidle_state atom_cstates[CPUIDLE_STATE_MAX] __initdata = {
>  	{
>  		.name = "C1E-ATM",
>  		.desc = "MWAIT 0x00",
> @@ -504,7 +504,12 @@ static int intel_idle_probe(void)
>  	pr_debug(PREFIX "MWAIT substates: 0x%x\n", mwait_substates);
>  
>  	icpu = (const struct idle_cpu *)id->driver_data;
> -	cpuidle_state_table = icpu->state_table;
> +	/* Copy the icpu->state_table into cpuidle_state_table,
> +	 * The pointing array by icpu->state_table is with __initdata,
> +	 * which will be freed after kernel init ending.
> +	 */
> +	memcpy(cpuidle_state_table, icpu->state_table,
> +			sizeof(cpuidle_state_table));

The idea is good but it could be pushed a bit further.

Instead of changing cpuidle_state_table to an array, change it to an
initdata and the functions intel_idle_cpuidle_driver_init,
intel_idle_probe to init functions.

That would also be nice to get rid of a global variable to store the
current cpu data and let the intel_idle_probe to return the pointer to
the right table.

>  	if (boot_cpu_has(X86_FEATURE_ARAT))	/* Always Reliable APIC Timer */
>  		lapic_timer_reliable_states = LAPIC_TIMER_ALWAYS_RELIABLE;
> 

-- 
 <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

^ permalink raw reply	[flat|nested] 15+ messages in thread