Skip to contentSkip to conversation

perf: Use pre-empt safe cpu_get/put insted of smp_processor_id

PATCH Superseded Roger Quadros 3 participants 4 messages Sep 10, 2012 – Sep 11, 2012

Subsystems (5)
Arm (32-bit) Arm sub-architectures OMAP2+ support OMAP power management support +1 more subsystems
lore
Share & tools
lore

Roger Quadros

author
#0 lore raw

PATCH v2 1/1 perf: Use pre-empt safe cpu_get/put insted of smp_processor_id

gets rid of below messages with CONFIG_DEBUG_PREEMPT enabled

[   28.832916] debug_smp_processor_id: 18 callbacks suppressed
[   28.832946] BUG: using smp_processor_id() in preemptible [00000000] code: modprobe/1763
[   28.841491] caller is pwrdm_set_next_pwrst+0x54/0x120

changes in v2:
- rebased on 3.6-rc5
- use put_cpu() immediately after get_cpu() in omap3_pm_idle()

Signed-off-by Roger Quadros <[email protected]>
---
 arch/arm/mach-omap2/clock.c       |    9 ++++++---
 arch/arm/mach-omap2/pm34xx.c      |   12 ++++++++----
 arch/arm/mach-omap2/powerdomain.c |    6 ++++--
 3 files changed, 18 insertions(+), 9 deletions(-)
arch/arm/mach-omap2/clock.c +6 -3
index ea3f565..06747b6 100644
@@ -285,7 +285,8 @@ void omap2_clk_disable(struct clk *clk)
 	pr_debug("clock: %s: disabling in hardware\n", clk->name);
 
 	if (clk->ops && clk->ops->disable) {
-		trace_clock_disable(clk->name, 0, smp_processor_id());
+		trace_clock_disable(clk->name, 0, get_cpu());
+		put_cpu();
 		clk->ops->disable(clk);
 	}
 

@@ -339,7 +340,8 @@ int omap2_clk_enable(struct clk *clk)
 	}
 
 	if (clk->ops && clk->ops->enable) {
-		trace_clock_enable(clk->name, 1, smp_processor_id());
+		trace_clock_enable(clk->name, 1, get_cpu());
+		put_cpu();
 		ret = clk->ops->enable(clk);
 		if (ret) {
 			WARN(1, "clock: %s: could not enable: %d\n",

@@ -380,7 +382,8 @@ int omap2_clk_set_rate(struct clk *clk, unsigned long rate)
 
 	/* dpll_ck, core_ck, virt_prcm_set; plus all clksel clocks */
 	if (clk->set_rate) {
-		trace_clock_set_rate(clk->name, rate, smp_processor_id());
+		trace_clock_set_rate(clk->name, rate, get_cpu());
+		put_cpu();
 		ret = clk->set_rate(clk, rate);
 	}
arch/arm/mach-omap2/pm34xx.c +8 -4
index 05bd8f0..5aa0a11 100644
@@ -346,18 +346,22 @@ void omap_sram_idle(void)
 
 static void omap3_pm_idle(void)
 {
+	unsigned cpu;
+
 	local_fiq_disable();
 
 	if (omap_irq_pending())
 		goto out;
 
-	trace_power_start(POWER_CSTATE, 1, smp_processor_id());
-	trace_cpu_idle(1, smp_processor_id());
+	cpu = get_cpu();
+	put_cpu();
+	trace_power_start(POWER_CSTATE, 1, cpu);
+	trace_cpu_idle(1, cpu);
 
 	omap_sram_idle();
 
-	trace_power_end(smp_processor_id());
-	trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
+	trace_power_end(cpu);
+	trace_cpu_idle(PWR_EVENT_EXIT, cpu);
 
 out:
 	local_fiq_enable();
arch/arm/mach-omap2/powerdomain.c +4 -2
index 69b36e1..138bf86 100644
@@ -169,7 +169,8 @@ static int _pwrdm_state_switch(struct powerdomain *pwrdm, int flag)
 				       ((state & OMAP_POWERSTATE_MASK) << 8) |
 				       ((prev & OMAP_POWERSTATE_MASK) << 0));
 			trace_power_domain_target(pwrdm->name, trace_state,
-						  smp_processor_id());
+						  get_cpu());
+			put_cpu();
 		}
 		break;
 	default:

@@ -491,7 +492,8 @@ int pwrdm_set_next_pwrst(struct powerdomain *pwrdm, u8 pwrst)
 	if (arch_pwrdm && arch_pwrdm->pwrdm_set_next_pwrst) {
 		/* Trace the pwrdm desired target state */
 		trace_power_domain_target(pwrdm->name, pwrst,
-					  smp_processor_id());
+					  get_cpu());
+		put_cpu();
 		/* Program the pwrdm desired target state */
 		ret = arch_pwrdm->pwrdm_set_next_pwrst(pwrdm, pwrst);
 	}
Signature · 1 line
1.7.4.1
#1 lore raw

PATCH v2 1/1 perf: Use pre-empt safe cpu_get/put insted of smp_processor_id

Quoted 11 lines · Roger Quadros

gets rid of below messages with CONFIG_DEBUG_PREEMPT enabled

[   28.832916] debug_smp_processor_id: 18 callbacks suppressed
[   28.832946] BUG: using smp_processor_id() in preemptible [00000000] code: modprobe/1763
[   28.841491] caller is pwrdm_set_next_pwrst+0x54/0x120

changes in v2:
- rebased on 3.6-rc5
- use put_cpu() immediately after get_cpu() in omap3_pm_idle()

Signed-off-by Roger Quadros <[email protected]>

Can you also add what platforms this was tested on?

Otherwise, this looks good, and I'll queue up for v3.7 (since it's not a regression.)

Thanks,

Kevin

Quoted 95 lines
---
 arch/arm/mach-omap2/clock.c       |    9 ++++++---
 arch/arm/mach-omap2/pm34xx.c      |   12 ++++++++----
 arch/arm/mach-omap2/powerdomain.c |    6 ++++--
 3 files changed, 18 insertions(+), 9 deletions(-)
arch/arm/mach-omap2/clock.c +6 -3
index ea3f565..06747b6 100644
@@ -285,7 +285,8 @@ void omap2_clk_disable(struct clk *clk)
 	pr_debug("clock: %s: disabling in hardware\n", clk->name);
 
 	if (clk->ops && clk->ops->disable) {
-		trace_clock_disable(clk->name, 0, smp_processor_id());
+		trace_clock_disable(clk->name, 0, get_cpu());
+		put_cpu();
 		clk->ops->disable(clk);
 	}
 

@@ -339,7 +340,8 @@ int omap2_clk_enable(struct clk *clk)
 	}
 
 	if (clk->ops && clk->ops->enable) {
-		trace_clock_enable(clk->name, 1, smp_processor_id());
+		trace_clock_enable(clk->name, 1, get_cpu());
+		put_cpu();
 		ret = clk->ops->enable(clk);
 		if (ret) {
 			WARN(1, "clock: %s: could not enable: %d\n",

@@ -380,7 +382,8 @@ int omap2_clk_set_rate(struct clk *clk, unsigned long rate)
 
 	/* dpll_ck, core_ck, virt_prcm_set; plus all clksel clocks */
 	if (clk->set_rate) {
-		trace_clock_set_rate(clk->name, rate, smp_processor_id());
+		trace_clock_set_rate(clk->name, rate, get_cpu());
+		put_cpu();
 		ret = clk->set_rate(clk, rate);
 	}
arch/arm/mach-omap2/pm34xx.c +8 -4
index 05bd8f0..5aa0a11 100644
@@ -346,18 +346,22 @@ void omap_sram_idle(void)
 
 static void omap3_pm_idle(void)
 {
+	unsigned cpu;
+
 	local_fiq_disable();
 
 	if (omap_irq_pending())
 		goto out;
 
-	trace_power_start(POWER_CSTATE, 1, smp_processor_id());
-	trace_cpu_idle(1, smp_processor_id());
+	cpu = get_cpu();
+	put_cpu();
+	trace_power_start(POWER_CSTATE, 1, cpu);
+	trace_cpu_idle(1, cpu);
 
 	omap_sram_idle();
 
-	trace_power_end(smp_processor_id());
-	trace_cpu_idle(PWR_EVENT_EXIT, smp_processor_id());
+	trace_power_end(cpu);
+	trace_cpu_idle(PWR_EVENT_EXIT, cpu);
 
 out:
 	local_fiq_enable();
arch/arm/mach-omap2/powerdomain.c +4 -2
index 69b36e1..138bf86 100644
@@ -169,7 +169,8 @@ static int _pwrdm_state_switch(struct powerdomain *pwrdm, int flag)
 				       ((state & OMAP_POWERSTATE_MASK) << 8) |
 				       ((prev & OMAP_POWERSTATE_MASK) << 0));
 			trace_power_domain_target(pwrdm->name, trace_state,
-						  smp_processor_id());
+						  get_cpu());
+			put_cpu();
 		}
 		break;
 	default:

@@ -491,7 +492,8 @@ int pwrdm_set_next_pwrst(struct powerdomain *pwrdm, u8 pwrst)
 	if (arch_pwrdm && arch_pwrdm->pwrdm_set_next_pwrst) {
 		/* Trace the pwrdm desired target state */
 		trace_power_domain_target(pwrdm->name, pwrst,
-					  smp_processor_id());
+					  get_cpu());
+		put_cpu();
 		/* Program the pwrdm desired target state */
 		ret = arch_pwrdm->pwrdm_set_next_pwrst(pwrdm, pwrst);
 	}
#2 lore raw

PATCH v2 1/1 perf: Use pre-empt safe cpu_get/put insted of smp_processor_id

In reply to Roger Quadros
Quoted 22 lines · Roger Quadros

gets rid of below messages with CONFIG_DEBUG_PREEMPT enabled

[   28.832916] debug_smp_processor_id: 18 callbacks suppressed
[   28.832946] BUG: using smp_processor_id() in preemptible [00000000] code: modprobe/1763
[   28.841491] caller is pwrdm_set_next_pwrst+0x54/0x120

changes in v2:
- rebased on 3.6-rc5
- use put_cpu() immediately after get_cpu() in omap3_pm_idle()

Signed-off-by Roger Quadros <[email protected]>
---
 arch/arm/mach-omap2/clock.c       |    9 ++++++---
 arch/arm/mach-omap2/pm34xx.c      |   12 ++++++++----
 arch/arm/mach-omap2/powerdomain.c |    6 ++++--
 3 files changed, 18 insertions(+), 9 deletions(-)
arch/arm/mach-omap2/clock.c +0 -0
index ea3f565..06747b6 100644

arch/arm/mach-omap2/clock.c void omap2_clk_disable(struct clk *clk) · line 289

Quoted 6 lines
arch/arm/mach-omap2/clock.c +2 -1
@@ -285,7 +285,8 @@ void omap2_clk_disable(struct clk *clk)
 	pr_debug("clock: %s: disabling in hardware\n", clk->name);
 
 	if (clk->ops && clk->ops->disable) {
-		trace_clock_disable(clk->name, 0, smp_processor_id());
+		trace_clock_disable(clk->name, 0, get_cpu());
+		put_cpu();

Why are you doing this? Why not just use raw_smp_processor_id()? Do you really care about the CPU number? get_cpu() and put_cpu() are about non-preemptible sections where you want to ensure the CPU the code is operating on is actually on that CPU.

How about just put 0 all the time because the CPU number is already part of the trace event?

Signature · 2 lines
Qualcomm Innovation Center, Inc. is a member of Code Aurora Forum,
hosted by The Linux Foundation
author
#3 lore raw

PATCH v2 1/1 perf: Use pre-empt safe cpu_get/put insted of smp_processor_id

Quoted 25 lines · Stephen Boyd
Quoted 24 lines · Roger Quadros

gets rid of below messages with CONFIG_DEBUG_PREEMPT enabled

[   28.832916] debug_smp_processor_id: 18 callbacks suppressed
[   28.832946] BUG: using smp_processor_id() in preemptible [00000000] code: modprobe/1763
[   28.841491] caller is pwrdm_set_next_pwrst+0x54/0x120

changes in v2:
- rebased on 3.6-rc5
- use put_cpu() immediately after get_cpu() in omap3_pm_idle()

Signed-off-by Roger Quadros <[email protected]>
---
 arch/arm/mach-omap2/clock.c       |    9 ++++++---
 arch/arm/mach-omap2/pm34xx.c      |   12 ++++++++----
 arch/arm/mach-omap2/powerdomain.c |    6 ++++--
 3 files changed, 18 insertions(+), 9 deletions(-)
arch/arm/mach-omap2/clock.c +0 -0
index ea3f565..06747b6 100644
@@ -285,7 +285,8 @@ void omap2_clk_disable(struct clk *clk)
 	pr_debug("clock: %s: disabling in hardware\n", clk->name);
Quoted 9 lines

arch/arm/mach-omap2/clock.c void omap2_clk_disable(struct clk *clk) · line 289

Quoted 4 lines
arch/arm/mach-omap2/clock.c +2 -1
 	if (clk->ops && clk->ops->disable) {
-		trace_clock_disable(clk->name, 0, smp_processor_id());
+		trace_clock_disable(clk->name, 0, get_cpu());
+		put_cpu();

Why are you doing this? Why not just use raw_smp_processor_id()? Do you really care about the CPU number? get_cpu() and put_cpu() are about non-preemptible sections where you want to ensure the CPU the code is operating on is actually on that CPU.

You are right. raw_smp_processor_id() makes perfect sense here.

Quoted 4 lines

How about just put 0 all the time because the CPU number is already part of the trace event?

I'm not sure about this. I can see that removing the cpu_id field from trace_cpu_idle() was discussed here [1] and it was decided to keep it. There was no conclusion about removing it from trace_clock_enable/disable though.

For now, I will resend the patch with raw_smp_processor_id().

regards,
-roger

[1] https://lkml.org/lkml/2011/8/19/316