[RFC][PATCH 0/2] process interrupts from idle wakeup

[RFC][PATCH 0/2] process interrupts from idle wakeup

[Nicholas Piggin]

Hi,

Just wanted to get some opinions on this approach. I will wait until
current batch of idle fixes are merged before submitting properly,
and still need to test on POWER9, but on POWER8 the speedup is
quite good.

Thanks,
Nick

Nicholas Piggin (2):
  powerpc/powernv: process interrupts from system reset wakeup
  powerpc/64s: msgclr when handling doorbell exceptions

 arch/powerpc/include/asm/dbell.h      | 13 +++++++++++++
 arch/powerpc/include/asm/machdep.h    |  1 -
 arch/powerpc/include/asm/ppc-opcode.h |  3 +++
 arch/powerpc/include/asm/processor.h  |  1 +
 arch/powerpc/kernel/dbell.c           |  9 +++++++++
 arch/powerpc/platforms/powernv/idle.c | 11 ++++++++---
 drivers/cpuidle/cpuidle-powernv.c     | 35 +++++++++++++++++++++++++++++++++--
 7 files changed, 67 insertions(+), 6 deletions(-)

-- 
2.11.0

[PATCH 1/2] powerpc/powernv: process interrupts from system reset wakeup

[Nicholas Piggin]

When the CPU wakes from low power state, it begins at the system reset
interrupt with the exception that caused the wakeup encoded in SRR1.

Today, powernv idle wakeup ignores the wakeup reason (except a special
case for HMI), and the regular interrupt corresponding to the exception
will fire after the idle wakeup exits.

Change this to replay the interrupt from the idle wakeup before
interrupts are hard-enabled.

Test on POWER8 of context_switch selftests benchmark with polling idle
disabled (e.g., always nap) gives the following results:

                                original         wakeup direct
Different threads, same core:   315k/s           264k/s
Different cores:                235k/s           242k/s

There is a slowdown for doorbell IPI (same core) case because system
reset wakeup does not clear the message and the doorbell interrupt fires
again needlessly.
---
 arch/powerpc/include/asm/machdep.h    |  1 -
 arch/powerpc/include/asm/processor.h  |  1 +
 arch/powerpc/platforms/powernv/idle.c | 11 ++++++++---
 drivers/cpuidle/cpuidle-powernv.c     | 35 +++++++++++++++++++++++++++++++++--
 4 files changed, 42 insertions(+), 6 deletions(-)

diff --git a/arch/powerpc/include/asm/machdep.h b/arch/powerpc/include/asm/machdep.h
index 5011b69107a7..c0d9fd2e8c04 100644
--- a/arch/powerpc/include/asm/machdep.h
+++ b/arch/powerpc/include/asm/machdep.h
@@ -223,7 +223,6 @@ struct machdep_calls {
 
 extern void e500_idle(void);
 extern void power4_idle(void);
-extern void power7_idle(void);
 extern void ppc6xx_idle(void);
 extern void book3e_idle(void);
 
diff --git a/arch/powerpc/include/asm/processor.h b/arch/powerpc/include/asm/processor.h
index e0fecbcea2a2..8190943d2619 100644
--- a/arch/powerpc/include/asm/processor.h
+++ b/arch/powerpc/include/asm/processor.h
@@ -454,6 +454,7 @@ extern int powersave_nap;	/* set if nap mode can be used in idle loop */
 extern unsigned long power7_nap(int check_irq);
 extern unsigned long power7_sleep(void);
 extern unsigned long power7_winkle(void);
+extern unsigned long power7_idle(void);
 extern unsigned long power9_idle_stop(unsigned long stop_psscr_val,
 				      unsigned long stop_psscr_mask);
 
diff --git a/arch/powerpc/platforms/powernv/idle.c b/arch/powerpc/platforms/powernv/idle.c
index 54546b632026..6b28a4f9c1fd 100644
--- a/arch/powerpc/platforms/powernv/idle.c
+++ b/arch/powerpc/platforms/powernv/idle.c
@@ -234,11 +234,16 @@ u64 pnv_default_stop_mask;
 /*
  * Used for ppc_md.power_save which needs a function with no parameters
  */
-static void power9_idle(void)
+static void power9_power_save(void)
 {
 	power9_idle_stop(pnv_default_stop_val, pnv_default_stop_mask);
 }
 
+static void power7_power_save(void)
+{
+	power7_idle();
+}
+
 /*
  * First deep stop state. Used to figure out when to save/restore
  * hypervisor context.
@@ -534,9 +539,9 @@ static int __init pnv_init_idle_states(void)
 	}
 
 	if (supported_cpuidle_states & OPAL_PM_NAP_ENABLED)
-		ppc_md.power_save = power7_idle;
+		ppc_md.power_save = power7_power_save;
 	else if (supported_cpuidle_states & OPAL_PM_STOP_INST_FAST)
-		ppc_md.power_save = power9_idle;
+		ppc_md.power_save = power9_power_save;
 
 out:
 	return 0;
diff --git a/drivers/cpuidle/cpuidle-powernv.c b/drivers/cpuidle/cpuidle-powernv.c
index 370593006f5f..e7e080c0790c 100644
--- a/drivers/cpuidle/cpuidle-powernv.c
+++ b/drivers/cpuidle/cpuidle-powernv.c
@@ -70,13 +70,37 @@ static int snooze_loop(struct cpuidle_device *dev,
 	return index;
 }
 
+/*
+ * Table to convert shifted SRR1 wakeup reason for POWER7, POWER8, POWER9
+ * to __replay_interrupt vector.
+ */
+static const unsigned short srr1_wakeup_to_replay_table[0x10] =
+{ 0, 0, 0, 0xe80,	/* 0x3 = hv doorbell */
+  0, 0xa00,		/* 0x5 = doorbell */
+  0x900,		/* 0x6 = decrementer */
+  0, 0x500,		/* 0x8 = external */
+  0xea0,		/* 0x9 = hv virt (POWER9) */
+  0xe60,		/* 0xa = hmi */
+  0, 0, 0, 0, 0, };
+
+/* Shift SRR1_WAKEMASK_P8 down and convert to __replay_interrupt vector */
+#define SRR1_TO_REPLAY(srr1) \
+	((unsigned int)srr1_wakeup_to_replay_table[((srr1) >> 18) & 0xf])
+
 static int nap_loop(struct cpuidle_device *dev,
 			struct cpuidle_driver *drv,
 			int index)
 {
+	u64 srr1;
+	unsigned int reason;
+
 	ppc64_runlatch_off();
-	power7_idle();
+	srr1 = power7_idle();
 	ppc64_runlatch_on();
+
+	reason = SRR1_TO_REPLAY(srr1);
+	__replay_interrupt(reason);
+
 	return index;
 }
 
@@ -111,10 +135,17 @@ static int stop_loop(struct cpuidle_device *dev,
 		     struct cpuidle_driver *drv,
 		     int index)
 {
+	u64 srr1;
+	unsigned int reason;
+
 	ppc64_runlatch_off();
-	power9_idle_stop(stop_psscr_table[index].val,
+	srr1 = power9_idle_stop(stop_psscr_table[index].val,
 			 stop_psscr_table[index].mask);
 	ppc64_runlatch_on();
+
+	reason = SRR1_TO_REPLAY(srr1);
+	__replay_interrupt(reason);
+
 	return index;
 }
 
-- 
2.11.0

[PATCH 2/2] powerpc/64s: msgclr when handling doorbell exceptions

[Nicholas Piggin]

msgsnd doorbell exceptions are cleared when the doorbell interrupt is
taken. However if a doorbell exception causes a system reset interrupt
wake from power saving state, the message is not cleared. Processing
the doorbell from the system reset interrupt requires msgclr to avoid
taking the exception twice.

Testing this plus the previous wakup direct patch gives:

                                original         wakeup direct     msgclr
Different threads, same core:   315k/s           264k/s            345k/s
Different cores:                235k/s           242k/s            242k/s

Net speedup is +10% for same core, and +3% for different core.
---
 arch/powerpc/include/asm/dbell.h      | 13 +++++++++++++
 arch/powerpc/include/asm/ppc-opcode.h |  3 +++
 arch/powerpc/kernel/dbell.c           |  9 +++++++++
 3 files changed, 25 insertions(+)

diff --git a/arch/powerpc/include/asm/dbell.h b/arch/powerpc/include/asm/dbell.h
index 378167377065..b242badb2af7 100644
--- a/arch/powerpc/include/asm/dbell.h
+++ b/arch/powerpc/include/asm/dbell.h
@@ -46,6 +46,19 @@ static inline void _ppc_msgsnd(u32 msg)
 		__asm__ __volatile__ (PPC_MSGSNDP(%0) : : "r" (msg));
 }
 
+static inline void _ppc_msgclr(u32 msg)
+{
+	__asm__ __volatile__ (ASM_FTR_IFSET(PPC_MSGCLR(%1), PPC_MSGCLRP(%1), %0)
+				: : "i" (CPU_FTR_HVMODE), "r" (msg));
+}
+
+static inline void ppc_msgclr(enum ppc_dbell type)
+{
+	u32 msg = PPC_DBELL_TYPE(type);
+
+	_ppc_msgclr(msg);
+}
+
 #else /* CONFIG_PPC_BOOK3S */
 
 #define PPC_DBELL_MSGTYPE		PPC_DBELL
diff --git a/arch/powerpc/include/asm/ppc-opcode.h b/arch/powerpc/include/asm/ppc-opcode.h
index e7d6d86563ee..eb8c7d5fcfbe 100644
--- a/arch/powerpc/include/asm/ppc-opcode.h
+++ b/arch/powerpc/include/asm/ppc-opcode.h
@@ -162,6 +162,7 @@
 #define PPC_INST_MSGSND			0x7c00019c
 #define PPC_INST_MSGCLR			0x7c0001dc
 #define PPC_INST_MSGSNDP		0x7c00011c
+#define PPC_INST_MSGCLRP		0x7c00015c
 #define PPC_INST_MTTMR			0x7c0003dc
 #define PPC_INST_NOP			0x60000000
 #define PPC_INST_PASTE			0x7c00070c
@@ -349,6 +350,8 @@
 					___PPC_RB(b))
 #define PPC_MSGSNDP(b)		stringify_in_c(.long PPC_INST_MSGSNDP | \
 					___PPC_RB(b))
+#define PPC_MSGCLRP(b)		stringify_in_c(.long PPC_INST_MSGCLRP | \
+					___PPC_RB(b))
 #define PPC_POPCNTB(a, s)	stringify_in_c(.long PPC_INST_POPCNTB | \
 					__PPC_RA(a) | __PPC_RS(s))
 #define PPC_POPCNTD(a, s)	stringify_in_c(.long PPC_INST_POPCNTD | \
diff --git a/arch/powerpc/kernel/dbell.c b/arch/powerpc/kernel/dbell.c
index 2128f3a96c32..2919f11b6afe 100644
--- a/arch/powerpc/kernel/dbell.c
+++ b/arch/powerpc/kernel/dbell.c
@@ -40,6 +40,15 @@ void doorbell_exception(struct pt_regs *regs)
 
 	irq_enter();
 
+	/*
+	 * May have come from doorbell triggered sreset wakeup, which does
+	 * not clear the message (tested on POWER8). Clear it here so we
+	 * don't get a doorbell interrupt when enabling hard irqs. This
+	 * msgclear could be moved into the sreset wakeup replay specific
+	 * code if it is expensive.
+	 */
+	ppc_msgclr(PPC_DBELL_MSGTYPE);
+
 	may_hard_irq_enable();
 
 	kvmppc_set_host_ipi(smp_processor_id(), 0);
-- 
2.11.0

Re: [PATCH 1/2] powerpc/powernv: process interrupts from system reset wakeup

[Michael Ellerman]

Nicholas Piggin <npiggin@gmail.com> writes:

> When the CPU wakes from low power state, it begins at the system reset
> interrupt with the exception that caused the wakeup encoded in SRR1.
>
> Today, powernv idle wakeup ignores the wakeup reason (except a special
> case for HMI), and the regular interrupt corresponding to the exception
> will fire after the idle wakeup exits.
>
> Change this to replay the interrupt from the idle wakeup before
> interrupts are hard-enabled.
>
> Test on POWER8 of context_switch selftests benchmark with polling idle
> disabled (e.g., always nap) gives the following results:
>
>                                 original         wakeup direct
> Different threads, same core:   315k/s           264k/s
> Different cores:                235k/s           242k/s
>
> There is a slowdown for doorbell IPI (same core) case because system
> reset wakeup does not clear the message and the doorbell interrupt fires
> again needlessly.

Seems like a win.

> diff --git a/arch/powerpc/include/asm/machdep.h b/arch/powerpc/include/asm/machdep.h
> index 5011b69107a7..c0d9fd2e8c04 100644
> --- a/arch/powerpc/include/asm/machdep.h
> +++ b/arch/powerpc/include/asm/machdep.h
> @@ -223,7 +223,6 @@ struct machdep_calls {
>  
>  extern void e500_idle(void);
>  extern void power4_idle(void);
> -extern void power7_idle(void);
>  extern void ppc6xx_idle(void);
>  extern void book3e_idle(void);
>  
> diff --git a/arch/powerpc/include/asm/processor.h b/arch/powerpc/include/asm/processor.h
> index e0fecbcea2a2..8190943d2619 100644
> --- a/arch/powerpc/include/asm/processor.h
> +++ b/arch/powerpc/include/asm/processor.h
> @@ -454,6 +454,7 @@ extern int powersave_nap;	/* set if nap mode can be used in idle loop */
>  extern unsigned long power7_nap(int check_irq);
>  extern unsigned long power7_sleep(void);
>  extern unsigned long power7_winkle(void);
> +extern unsigned long power7_idle(void);
>  extern unsigned long power9_idle_stop(unsigned long stop_psscr_val,
>  				      unsigned long stop_psscr_mask);
>  
> diff --git a/arch/powerpc/platforms/powernv/idle.c b/arch/powerpc/platforms/powernv/idle.c
> index 54546b632026..6b28a4f9c1fd 100644
> --- a/arch/powerpc/platforms/powernv/idle.c
> +++ b/arch/powerpc/platforms/powernv/idle.c
> @@ -234,11 +234,16 @@ u64 pnv_default_stop_mask;
>  /*
>   * Used for ppc_md.power_save which needs a function with no parameters
>   */
> -static void power9_idle(void)
> +static void power9_power_save(void)
>  {
>  	power9_idle_stop(pnv_default_stop_val, pnv_default_stop_mask);
>  }
>  
> +static void power7_power_save(void)
> +{
> +	power7_idle();
> +}

Erk. This makes me wonder if we can just mandate using cpuidle for
powernv and drop ppc_md.power_save. I wonder who if anyone has ever
tested powernv without cpuidle.

I notice we already have:

config CPU_IDLE
	bool "CPU idle PM support"
	default y if ACPI || PPC_PSERIES


But not your problem for this patch.

> @@ -534,9 +539,9 @@ static int __init pnv_init_idle_states(void)
>  	}
>  
>  	if (supported_cpuidle_states & OPAL_PM_NAP_ENABLED)
> -		ppc_md.power_save = power7_idle;
> +		ppc_md.power_save = power7_power_save;
>  	else if (supported_cpuidle_states & OPAL_PM_STOP_INST_FAST)
> -		ppc_md.power_save = power9_idle;
> +		ppc_md.power_save = power9_power_save;
>  
>  out:
>  	return 0;
> diff --git a/drivers/cpuidle/cpuidle-powernv.c b/drivers/cpuidle/cpuidle-powernv.c
> index 370593006f5f..e7e080c0790c 100644
> --- a/drivers/cpuidle/cpuidle-powernv.c
> +++ b/drivers/cpuidle/cpuidle-powernv.c
> @@ -70,13 +70,37 @@ static int snooze_loop(struct cpuidle_device *dev,
>  	return index;
>  }
>  
> +/*
> + * Table to convert shifted SRR1 wakeup reason for POWER7, POWER8, POWER9
> + * to __replay_interrupt vector.
> + */
> +static const unsigned short srr1_wakeup_to_replay_table[0x10] =
> +{ 0, 0, 0, 0xe80,	/* 0x3 = hv doorbell */
> +  0, 0xa00,		/* 0x5 = doorbell */
> +  0x900,		/* 0x6 = decrementer */
> +  0, 0x500,		/* 0x8 = external */
> +  0xea0,		/* 0x9 = hv virt (POWER9) */
> +  0xe60,		/* 0xa = hmi */
> +  0, 0, 0, 0, 0, };
> +
> +/* Shift SRR1_WAKEMASK_P8 down and convert to __replay_interrupt vector */
> +#define SRR1_TO_REPLAY(srr1) \
> +	((unsigned int)srr1_wakeup_to_replay_table[((srr1) >> 18) & 0xf])

This is a bit hairy, I'd just use a switch, but I guess this generates
vastly better code?

> +
>  static int nap_loop(struct cpuidle_device *dev,
>  			struct cpuidle_driver *drv,
>  			int index)
>  {
> +	u64 srr1;
> +	unsigned int reason;
> +
>  	ppc64_runlatch_off();
> -	power7_idle();
> +	srr1 = power7_idle();
>  	ppc64_runlatch_on();
> +
> +	reason = SRR1_TO_REPLAY(srr1);
> +	__replay_interrupt(reason);
> +
>  	return index;
>  }
>  
> @@ -111,10 +135,17 @@ static int stop_loop(struct cpuidle_device *dev,
>  		     struct cpuidle_driver *drv,
>  		     int index)
>  {
> +	u64 srr1;
> +	unsigned int reason;
> +
>  	ppc64_runlatch_off();
> -	power9_idle_stop(stop_psscr_table[index].val,
> +	srr1 = power9_idle_stop(stop_psscr_table[index].val,
>  			 stop_psscr_table[index].mask);
>  	ppc64_runlatch_on();
> +
> +	reason = SRR1_TO_REPLAY(srr1);
> +	__replay_interrupt(reason);
> +
>  	return index;
>  }

We end up with a bit of duplicated code there, but I guess it's not
worth worrying about.

cheers