Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1360330 > unrolled thread

Re: [PATCH v8 3/6] cpufreq: powernv: Remove cpu_to_chip_id() from hot-path

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  Re: [PATCH v8 3/6] cpufreq: powernv: Remove cpu_to_chip_id() from  hot-path Michael Neuling <mikey@neuling.org> - 2016-03-18 05:10 +0100
    Re: [PATCH v8 3/6] cpufreq: powernv: Remove cpu_to_chip_id() from  hot-path Michael Neuling <mikey@neuling.org> - 2016-03-18 05:20 +0100
      Re: [PATCH v8 3/6] cpufreq: powernv: Remove cpu_to_chip_id() from hot-path "Rafael J. Wysocki" <rafael@kernel.org> - 2016-03-18 14:20 +0100
      [PATCH] cpufreq: powernv: Define per_cpu chip pointer to optimize hot-path Shilpasri G Bhat <shilpa.bhat@linux.vnet.ibm.com> - 2016-03-18 16:00 +0100
        Re: [PATCH] cpufreq: powernv: Define per_cpu chip pointer to  optimize hot-path Viresh Kumar <viresh.kumar@linaro.org> - 2016-03-21 08:30 +0100
          Re: [PATCH] cpufreq: powernv: Define per_cpu chip pointer to optimize hot-path "Rafael J. Wysocki" <rjw@rjwysocki.net> - 2016-03-21 15:20 +0100
    Re: [PATCH v8 3/6] cpufreq: powernv: Remove cpu_to_chip_id() from  hot-path Benjamin Herrenschmidt <benh@au1.ibm.com> - 2016-03-18 23:40 +0100
      Re: [PATCH v8 3/6] cpufreq: powernv: Remove cpu_to_chip_id() from  hot-path Michael Neuling <mikey@neuling.org> - 2016-03-19 00:30 +0100

#1360330 — Re: [PATCH v8 3/6] cpufreq: powernv: Remove cpu_to_chip_id() from hot-path

On Wed, 2016-02-03 at 01:11 +0530, Shilpasri G Bhat wrote:

> cpu_to_chip_id() does a DT walk through to find out the chip id by
> taking a contended device tree lock. This adds an unnecessary overhead
> in a hot path. So instead of calling cpu_to_chip_id() everytime cache
> the chip ids for all cores in the array 'core_to_chip_map' and use it
> in the hotpath.
> 
> Reported-by: Anton Blanchard <anton@samba.org>
> Signed-off-by: Shilpasri G Bhat <shilpa.bhat@linux.vnet.ibm.com>
> Reviewed-by: Gautham R. Shenoy <ego@linux.vnet.ibm.com>
> Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
> ---
> No changes from v7.

How about this instead?  It removes the linear lookup and seems a lot
less complex.

Mikey

diff --git a/drivers/cpufreq/powernv-cpufreq.c b/drivers/cpufreq/powernv-cpufreq.c
index 547890f..d63d2cb 100644
--- a/drivers/cpufreq/powernv-cpufreq.c
+++ b/drivers/cpufreq/powernv-cpufreq.c
@@ -52,6 +52,7 @@ static struct chip {
 } *chips;
 
 static int nr_chips;
+static DEFINE_PER_CPU(unsigned int, chip_id);
 
 /*
  * Note: The set of pstates consists of contiguous integers, the
@@ -317,9 +318,7 @@ static void powernv_cpufreq_throttle_check(void *data)
 
 	pmsr = get_pmspr(SPRN_PMSR);
 
-	for (i = 0; i < nr_chips; i++)
-		if (chips[i].id == cpu_to_chip_id(cpu))
-			break;
+	i = this_cpu_read(chip_id);
 
 	/* Check for Pmax Capping */
 	pmsr_pmax = (s8)PMSR_MAX(pmsr);
@@ -560,6 +559,7 @@ static int init_chip_info(void)
 	for_each_possible_cpu(cpu) {
 		unsigned int id = cpu_to_chip_id(cpu);
 
+		per_cpu(chip_id, cpu) = nr_chips;
 		if (prev_chip_id != id) {
 			prev_chip_id = id;
 			chip[nr_chips++] = id;

[toc] | [next] | [standalone]

#1360333

On Fri, 2016-03-18 at 15:04 +1100, Michael Neuling wrote:

> On Wed, 2016-02-03 at 01:11 +0530, Shilpasri G Bhat wrote:
> 

> > cpu_to_chip_id() does a DT walk through to find out the chip id by
> > taking a contended device tree lock. This adds an unnecessary
> > overhead
> > in a hot path. So instead of calling cpu_to_chip_id() everytime
> > cache
> > the chip ids for all cores in the array 'core_to_chip_map' and use
> > it
> > in the hotpath.
> > 
> > Reported-by: Anton Blanchard <anton@samba.org>
> > Signed-off-by: Shilpasri G Bhat <shilpa.bhat@linux.vnet.ibm.com>
> > Reviewed-by: Gautham R. Shenoy <ego@linux.vnet.ibm.com>
> > Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
> > ---
> > No changes from v7.
> 
> How about this instead?  It removes the linear lookup and seems a lot
> less complex.

BTW we never init nr_chips before using it.  We also need something
like.

diff --git a/drivers/cpufreq/powernv-cpufreq.c b/drivers/cpufreq/powernv-cpufreq.c
index d63d2cb..c819ed4 100644
--- a/drivers/cpufreq/powernv-cpufreq.c
+++ b/drivers/cpufreq/powernv-cpufreq.c
@@ -556,6 +556,8 @@ static int init_chip_info(void)
 	unsigned int cpu, i;
 	unsigned int prev_chip_id = UINT_MAX;
 
+	nr_chips = 0;
+
 	for_each_possible_cpu(cpu) {
 		unsigned int id = cpu_to_chip_id(cpu);
 

[toc] | [prev] | [next] | [standalone]

#1360611 — Re: [PATCH v8 3/6] cpufreq: powernv: Remove cpu_to_chip_id() from hot-path

On Fri, Mar 18, 2016 at 5:11 AM, Michael Neuling <mikey@neuling.org> wrote:
> On Fri, 2016-03-18 at 15:04 +1100, Michael Neuling wrote:
>
>> On Wed, 2016-02-03 at 01:11 +0530, Shilpasri G Bhat wrote:
>>
>
>> > cpu_to_chip_id() does a DT walk through to find out the chip id by
>> > taking a contended device tree lock. This adds an unnecessary
>> > overhead
>> > in a hot path. So instead of calling cpu_to_chip_id() everytime
>> > cache
>> > the chip ids for all cores in the array 'core_to_chip_map' and use
>> > it
>> > in the hotpath.
>> >
>> > Reported-by: Anton Blanchard <anton@samba.org>
>> > Signed-off-by: Shilpasri G Bhat <shilpa.bhat@linux.vnet.ibm.com>
>> > Reviewed-by: Gautham R. Shenoy <ego@linux.vnet.ibm.com>
>> > Acked-by: Viresh Kumar <viresh.kumar@linaro.org>
>> > ---
>> > No changes from v7.
>>
>> How about this instead?  It removes the linear lookup and seems a lot
>> less complex.

This has gone in already.  Can you please send a patch on top of it?

> BTW we never init nr_chips before using it.  We also need something
> like.
>
> diff --git a/drivers/cpufreq/powernv-cpufreq.c b/drivers/cpufreq/powernv-cpufreq.c
> index d63d2cb..c819ed4 100644
> --- a/drivers/cpufreq/powernv-cpufreq.c
> +++ b/drivers/cpufreq/powernv-cpufreq.c
> @@ -556,6 +556,8 @@ static int init_chip_info(void)
>         unsigned int cpu, i;
>         unsigned int prev_chip_id = UINT_MAX;
>
> +       nr_chips = 0;
> +
>         for_each_possible_cpu(cpu) {
>                 unsigned int id = cpu_to_chip_id(cpu);
>
>

Including this part too maybe?

Thanks,
Rafael

[toc] | [prev] | [next] | [standalone]

#1360711 — [PATCH] cpufreq: powernv: Define per_cpu chip pointer to optimize hot-path

From: Michael Neuling <mikey@neuling.org>

"cpufreq: powernv: Remove cpu_to_chip_id() from hot-path" introduced
'core_to_chip_map' array to cache the chip-id of all cores. Replace
this with per_cpu variable that stores the pointer to the chip-array.
This removes the linear lookup and provides a neater and simpler
solution.

Signed-off-by: Michael Neuling <mikey@neuling.org>
Tested-by: Shilpasri G Bhat <shilpa.bhat@linux.vnet.ibm.com>
---

- Rebased the patch on top of linux-pm/linux-next
- nr_chips is defined static, so it will be initialized to zero
- Moved the initialization of the per_cpu variable after 'chips' is
  allocated
- Removed 'core_to_chip_map'

 drivers/cpufreq/powernv-cpufreq.c | 50 +++++++++++++--------------------------
 1 file changed, 17 insertions(+), 33 deletions(-)

diff --git a/drivers/cpufreq/powernv-cpufreq.c b/drivers/cpufreq/powernv-cpufreq.c
index 50bf120..a00bcc2 100644
--- a/drivers/cpufreq/powernv-cpufreq.c
+++ b/drivers/cpufreq/powernv-cpufreq.c
@@ -44,7 +44,6 @@
 
 static struct cpufreq_frequency_table powernv_freqs[POWERNV_MAX_PSTATES+1];
 static bool rebooting, throttled, occ_reset;
-static unsigned int *core_to_chip_map;
 
 static const char * const throttle_reason[] = {
 	"No throttling",
@@ -65,6 +64,7 @@ static struct chip {
 } *chips;
 
 static int nr_chips;
+static DEFINE_PER_CPU(struct chip *, chip_info);
 
 /*
  * Note: The set of pstates consists of contiguous integers, the
@@ -324,34 +324,31 @@ static inline unsigned int get_nominal_index(void)
 
 static void powernv_cpufreq_throttle_check(void *data)
 {
+	struct chip *chip;
 	unsigned int cpu = smp_processor_id();
-	unsigned int chip_id = core_to_chip_map[cpu_core_index_of_thread(cpu)];
 	unsigned long pmsr;
-	int pmsr_pmax, i;
+	int pmsr_pmax;
 
 	pmsr = get_pmspr(SPRN_PMSR);
-
-	for (i = 0; i < nr_chips; i++)
-		if (chips[i].id == chip_id)
-			break;
+	chip = this_cpu_read(chip_info);
 
 	/* Check for Pmax Capping */
 	pmsr_pmax = (s8)PMSR_MAX(pmsr);
 	if (pmsr_pmax != powernv_pstate_info.max) {
-		if (chips[i].throttled)
+		if (chip->throttled)
 			goto next;
-		chips[i].throttled = true;
+		chip->throttled = true;
 		if (pmsr_pmax < powernv_pstate_info.nominal)
 			pr_warn_once("CPU %d on Chip %u has Pmax reduced below nominal frequency (%d < %d)\n",
-				     cpu, chips[i].id, pmsr_pmax,
+				     cpu, chip->id, pmsr_pmax,
 				     powernv_pstate_info.nominal);
-		trace_powernv_throttle(chips[i].id,
-				      throttle_reason[chips[i].throttle_reason],
+		trace_powernv_throttle(chip->id,
+				      throttle_reason[chip->throttle_reason],
 				      pmsr_pmax);
-	} else if (chips[i].throttled) {
-		chips[i].throttled = false;
-		trace_powernv_throttle(chips[i].id,
-				      throttle_reason[chips[i].throttle_reason],
+	} else if (chip->throttled) {
+		chip->throttled = false;
+		trace_powernv_throttle(chip->id,
+				      throttle_reason[chip->throttle_reason],
 				      pmsr_pmax);
 	}
 
@@ -558,47 +555,34 @@ static int init_chip_info(void)
 	unsigned int chip[256];
 	unsigned int cpu, i;
 	unsigned int prev_chip_id = UINT_MAX;
-	cpumask_t cpu_mask;
-	int ret = -ENOMEM;
-
-	core_to_chip_map = kcalloc(cpu_nr_cores(), sizeof(unsigned int),
-				   GFP_KERNEL);
-	if (!core_to_chip_map)
-		goto out;
 
-	cpumask_copy(&cpu_mask, cpu_possible_mask);
-	for_each_cpu(cpu, &cpu_mask) {
+	for_each_possible_cpu(cpu) {
 		unsigned int id = cpu_to_chip_id(cpu);
 
 		if (prev_chip_id != id) {
 			prev_chip_id = id;
 			chip[nr_chips++] = id;
 		}
-		core_to_chip_map[cpu_core_index_of_thread(cpu)] = id;
-		cpumask_andnot(&cpu_mask, &cpu_mask, cpu_sibling_mask(cpu));
 	}
 
 	chips = kcalloc(nr_chips, sizeof(struct chip), GFP_KERNEL);
 	if (!chips)
-		goto free_chip_map;
+		return -ENOMEM;
 
 	for (i = 0; i < nr_chips; i++) {
 		chips[i].id = chip[i];
 		cpumask_copy(&chips[i].mask, cpumask_of_node(chip[i]));
 		INIT_WORK(&chips[i].throttle, powernv_cpufreq_work_fn);
+		for_each_cpu(cpu, &chips[i].mask)
+			per_cpu(chip_info, cpu) =  &chips[i];
 	}
 
 	return 0;
-free_chip_map:
-	kfree(core_to_chip_map);
-out:
-	return ret;
 }
 
 static inline void clean_chip_info(void)
 {
 	kfree(chips);
-	kfree(core_to_chip_map);
 }
 
 static inline void unregister_all_notifiers(void)
-- 
1.9.3

[toc] | [prev] | [next] | [standalone]

#1361559 — Re: [PATCH] cpufreq: powernv: Define per_cpu chip pointer to optimize hot-path

On 18-03-16, 20:28, Shilpasri G Bhat wrote:
> From: Michael Neuling <mikey@neuling.org>
> 
> "cpufreq: powernv: Remove cpu_to_chip_id() from hot-path" introduced

If the patch is already committed, you should provide its commit id as well.

> 'core_to_chip_map' array to cache the chip-id of all cores. Replace
> this with per_cpu variable that stores the pointer to the chip-array.
> This removes the linear lookup and provides a neater and simpler
> solution.
> 
> Signed-off-by: Michael Neuling <mikey@neuling.org>
> Tested-by: Shilpasri G Bhat <shilpa.bhat@linux.vnet.ibm.com>

You are sending this patch and you have updated it as well.. So, you should have
had your signed-off here.

Acked-by: Viresh Kumar <viresh.kumar@linaro.org>

-- 
viresh

[toc] | [prev] | [next] | [standalone]

#1361877 — Re: [PATCH] cpufreq: powernv: Define per_cpu chip pointer to optimize hot-path

On Monday, March 21, 2016 12:52:55 PM Viresh Kumar wrote:
> On 18-03-16, 20:28, Shilpasri G Bhat wrote:
> > From: Michael Neuling <mikey@neuling.org>
> > 
> > "cpufreq: powernv: Remove cpu_to_chip_id() from hot-path" introduced
> 
> If the patch is already committed, you should provide its commit id as well.
> 
> > 'core_to_chip_map' array to cache the chip-id of all cores. Replace
> > this with per_cpu variable that stores the pointer to the chip-array.
> > This removes the linear lookup and provides a neater and simpler
> > solution.
> > 
> > Signed-off-by: Michael Neuling <mikey@neuling.org>
> > Tested-by: Shilpasri G Bhat <shilpa.bhat@linux.vnet.ibm.com>
> 
> You are sending this patch and you have updated it as well.. So, you should have
> had your signed-off here.

Right.  Plus if you send a patch from anyone else, you should add your
sigh-off to it anyway, even if it hasn't been modified.

So Shilpasri, please resend with your S-o-b and with the ACK from Viresh.

Thanks,
Rafael

[toc] | [prev] | [next] | [standalone]

#1360975

On Fri, 2016-03-18 at 15:04 +1100, Michael Neuling wrote:
> 
>  static int nr_chips;
> +static DEFINE_PER_CPU(unsigned int, chip_id);
>  
>  /*
>   * Note: The set of pstates consists of contiguous integers, the
> @@ -317,9 +318,7 @@ static void powernv_cpufreq_throttle_check(void
> *data)
>  
>         pmsr = get_pmspr(SPRN_PMSR);
>  
> -       for (i = 0; i < nr_chips; i++)
> -               if (chips[i].id == cpu_to_chip_id(cpu))
> -                       break;
> +       i = this_cpu_read(chip_id);

Except it's not a chip_id, so your patch confused me for a good 2mn ...
Call it chip_idx maybe ? ie, index.

Cheers,
Ben.

>         /* Check for Pmax Capping */
>         pmsr_pmax = (s8)PMSR_MAX(pmsr);
> @@ -560,6 +559,7 @@ static int init_chip_info(void)
>         for_each_possible_cpu(cpu) {
>                 unsigned int id = cpu_to_chip_id(cpu);
>  
> +               per_cpu(chip_id, cpu) = nr_chips;
>                 if (prev_chip_id != id) {
>                         prev_chip_id = id;
>                         chip[nr_chips++] = id;
> _______________________________________________

[toc] | [prev] | [next] | [standalone]

#1360998

On Sat, 2016-03-19 at 09:37 +1100, Benjamin Herrenschmidt wrote:
> On Fri, 2016-03-18 at 15:04 +1100, Michael Neuling wrote:
> > 
> >  static int nr_chips;
> > +static DEFINE_PER_CPU(unsigned int, chip_id);
> >  
> >  /*
> >   * Note: The set of pstates consists of contiguous integers, the
> > @@ -317,9 +318,7 @@ static void powernv_cpufreq_throttle_check(void
> > *data)
> >  
> >         pmsr = get_pmspr(SPRN_PMSR);
> >  
> > -       for (i = 0; i < nr_chips; i++)
> > -               if (chips[i].id == cpu_to_chip_id(cpu))
> > -                       break;
> > +       i = this_cpu_read(chip_id);
> 
> Except it's not a chip_id, so your patch confused me for a good 2mn
> ...
> Call it chip_idx maybe ? ie, index.

Yeah, it was a badly named variable but I changed it even more and
Shilpasri rebased it here:

http://patchwork.ozlabs.org/patch/599523/

Mikey

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web