Re: [PATCH 2/2] powernv:idle:Implement lite variant of power_enter_stop
Hi Michael, On Fri, Sep 23, 2016 at 09:03:45PM +1000, Michael Ellerman wrote: > "Gautham R. Shenoy"writes: > > > From: "Gautham R. Shenoy" > > > > This patch adds a function named power_enter_stop_lite() that can > > execute a stop instruction when ESL and EC bits are set to zero in the > > PSSCR. The function handles the wake-up from idle at the instruction > > immediately after the stop instruction. > > > > If the flag OPAL_PM_WAKEUP_AT_NEXT_INST[1] is set in the device tree > > for a stop state, then use the lite variant for that particular stop > > state. > > Hi Gautham, > > It seems to me that this would be cleaner if it was modelled as a new > stop state? Surely it must have different power saving characteristics > to the existing state? It is supposed to be a new stop state, whose behaviour is different from the existing stop states in terms of where it wakes up from stop. You are right, it has different power saving characteristics to the existing state. > > > [1] : The corresponding patch in skiboot that defines > > OPAL_PM_WAKEUP_AT_NEXT_INST and enables it in the device tree > > can be found here: > > https://lists.ozlabs.org/pipermail/skiboot/2016-September/004805.html > > Which says "This will reduce the exit latency of the idle state", and > yet it doesn't change any of the latency/residency values? > > If all it does is reduce the exit latency, then shouldn't we always use > it? Presumably it also saves less power? It does save less power compared to the corresponding variant where ESL=EC=1. > > > diff --git a/arch/powerpc/kernel/idle_book3s.S > > b/arch/powerpc/kernel/idle_book3s.S > > index 32d666b..47ee106 100644 > > --- a/arch/powerpc/kernel/idle_book3s.S > > +++ b/arch/powerpc/kernel/idle_book3s.S > > @@ -43,6 +43,8 @@ > > #define PSSCR_HV_TEMPLATE PSSCR_ESL | PSSCR_EC | \ > > PSSCR_PSLL_MASK | PSSCR_TR_MASK | \ > > PSSCR_MTL_MASK > > +#define PSSCR_HV_TEMPLATE_LITE PSSCR_PSLL_MASK | PSSCR_TR_MASK | \ > > +PSSCR_MTL_MASK > > Why do we have these at all? Firmware tells us the PSSCR values to use > in the "ibm,cpu-idle-state-psscr" property. > > If we just used what firmware gave us then we wouldn't need the above, > or the juggling below. Let me rework the patch to use the cpu-idle-state-psscr property which is currently set to 0 in the firmware and cpu-idle-state-psscr-mask which is set to 0xF for all the stop states. I agree, we can do without the hardcoding of the mask in the kernel. > > > @@ -333,13 +349,19 @@ ALT_FTR_SECTION_END_NESTED_IFSET(CPU_FTR_ARCH_207S, > > 66); \ > > > > /* > > * r3 - requested stop state > > + * r4 - Indicates if the lite variant with ESL=EC=0 should be executed. > > */ > > _GLOBAL(power9_idle_stop) > > - LOAD_REG_IMMEDIATE(r4, PSSCR_HV_TEMPLATE) > > - or r4,r4,r3 > > + cmpdi r4, 1 > > + bne 4f > > + LOAD_REG_IMMEDIATE(r4, PSSCR_HV_TEMPLATE_LITE) > > + LOAD_REG_ADDR(r5,power_enter_stop_lite) > > + b 5f > > +4: LOAD_REG_IMMEDIATE(r4, PSSCR_HV_TEMPLATE) > > + LOAD_REG_ADDR(r5,power_enter_stop) > > +5: or r4,r4,r3 > > mtspr SPRN_PSSCR, r4 > > li r4, 1 > > - LOAD_REG_ADDR(r5,power_enter_stop) > > b pnv_powersave_common > > /* No return */ > > /* > > > diff --git a/arch/powerpc/platforms/powernv/idle.c > > b/arch/powerpc/platforms/powernv/idle.c > > index 479c256..c3d3fed 100644 > > --- a/arch/powerpc/platforms/powernv/idle.c > > +++ b/arch/powerpc/platforms/powernv/idle.c > > @@ -244,8 +244,15 @@ static DEVICE_ATTR(fastsleep_workaround_applyonce, > > 0600, > > static void power9_idle(void) > > { > > /* Requesting stop state 0 */ > > - power9_idle_stop(0); > > } > > That seems like the root of the problem, why aren't we passing a PSSCR > value here? > > I also don't see us using the psscr-mask property anywhere. Why isn't > that a bug? Because we are only setting the Requested Level field which is the bottom 4 bits. Everything else is taken from the hardcoded mask, which is not incorrect, but not a very flexible design. Thanks for the comments Michael. I will come back with a cleaner patch. > > cheers > -- Thanks and Regards gautham.
Re: [PATCH 2/2] powernv:idle:Implement lite variant of power_enter_stop
Hi Michael, On Fri, Sep 23, 2016 at 09:03:45PM +1000, Michael Ellerman wrote: > "Gautham R. Shenoy" writes: > > > From: "Gautham R. Shenoy" > > > > This patch adds a function named power_enter_stop_lite() that can > > execute a stop instruction when ESL and EC bits are set to zero in the > > PSSCR. The function handles the wake-up from idle at the instruction > > immediately after the stop instruction. > > > > If the flag OPAL_PM_WAKEUP_AT_NEXT_INST[1] is set in the device tree > > for a stop state, then use the lite variant for that particular stop > > state. > > Hi Gautham, > > It seems to me that this would be cleaner if it was modelled as a new > stop state? Surely it must have different power saving characteristics > to the existing state? It is supposed to be a new stop state, whose behaviour is different from the existing stop states in terms of where it wakes up from stop. You are right, it has different power saving characteristics to the existing state. > > > [1] : The corresponding patch in skiboot that defines > > OPAL_PM_WAKEUP_AT_NEXT_INST and enables it in the device tree > > can be found here: > > https://lists.ozlabs.org/pipermail/skiboot/2016-September/004805.html > > Which says "This will reduce the exit latency of the idle state", and > yet it doesn't change any of the latency/residency values? > > If all it does is reduce the exit latency, then shouldn't we always use > it? Presumably it also saves less power? It does save less power compared to the corresponding variant where ESL=EC=1. > > > diff --git a/arch/powerpc/kernel/idle_book3s.S > > b/arch/powerpc/kernel/idle_book3s.S > > index 32d666b..47ee106 100644 > > --- a/arch/powerpc/kernel/idle_book3s.S > > +++ b/arch/powerpc/kernel/idle_book3s.S > > @@ -43,6 +43,8 @@ > > #define PSSCR_HV_TEMPLATE PSSCR_ESL | PSSCR_EC | \ > > PSSCR_PSLL_MASK | PSSCR_TR_MASK | \ > > PSSCR_MTL_MASK > > +#define PSSCR_HV_TEMPLATE_LITE PSSCR_PSLL_MASK | PSSCR_TR_MASK | \ > > +PSSCR_MTL_MASK > > Why do we have these at all? Firmware tells us the PSSCR values to use > in the "ibm,cpu-idle-state-psscr" property. > > If we just used what firmware gave us then we wouldn't need the above, > or the juggling below. Let me rework the patch to use the cpu-idle-state-psscr property which is currently set to 0 in the firmware and cpu-idle-state-psscr-mask which is set to 0xF for all the stop states. I agree, we can do without the hardcoding of the mask in the kernel. > > > @@ -333,13 +349,19 @@ ALT_FTR_SECTION_END_NESTED_IFSET(CPU_FTR_ARCH_207S, > > 66); \ > > > > /* > > * r3 - requested stop state > > + * r4 - Indicates if the lite variant with ESL=EC=0 should be executed. > > */ > > _GLOBAL(power9_idle_stop) > > - LOAD_REG_IMMEDIATE(r4, PSSCR_HV_TEMPLATE) > > - or r4,r4,r3 > > + cmpdi r4, 1 > > + bne 4f > > + LOAD_REG_IMMEDIATE(r4, PSSCR_HV_TEMPLATE_LITE) > > + LOAD_REG_ADDR(r5,power_enter_stop_lite) > > + b 5f > > +4: LOAD_REG_IMMEDIATE(r4, PSSCR_HV_TEMPLATE) > > + LOAD_REG_ADDR(r5,power_enter_stop) > > +5: or r4,r4,r3 > > mtspr SPRN_PSSCR, r4 > > li r4, 1 > > - LOAD_REG_ADDR(r5,power_enter_stop) > > b pnv_powersave_common > > /* No return */ > > /* > > > diff --git a/arch/powerpc/platforms/powernv/idle.c > > b/arch/powerpc/platforms/powernv/idle.c > > index 479c256..c3d3fed 100644 > > --- a/arch/powerpc/platforms/powernv/idle.c > > +++ b/arch/powerpc/platforms/powernv/idle.c > > @@ -244,8 +244,15 @@ static DEVICE_ATTR(fastsleep_workaround_applyonce, > > 0600, > > static void power9_idle(void) > > { > > /* Requesting stop state 0 */ > > - power9_idle_stop(0); > > } > > That seems like the root of the problem, why aren't we passing a PSSCR > value here? > > I also don't see us using the psscr-mask property anywhere. Why isn't > that a bug? Because we are only setting the Requested Level field which is the bottom 4 bits. Everything else is taken from the hardcoded mask, which is not incorrect, but not a very flexible design. Thanks for the comments Michael. I will come back with a cleaner patch. > > cheers > -- Thanks and Regards gautham.
Re: [PATCH 2/2] powernv:idle:Implement lite variant of power_enter_stop
"Gautham R. Shenoy"writes: > From: "Gautham R. Shenoy" > > This patch adds a function named power_enter_stop_lite() that can > execute a stop instruction when ESL and EC bits are set to zero in the > PSSCR. The function handles the wake-up from idle at the instruction > immediately after the stop instruction. > > If the flag OPAL_PM_WAKEUP_AT_NEXT_INST[1] is set in the device tree > for a stop state, then use the lite variant for that particular stop > state. Hi Gautham, It seems to me that this would be cleaner if it was modelled as a new stop state? Surely it must have different power saving characteristics to the existing state? > [1] : The corresponding patch in skiboot that defines > OPAL_PM_WAKEUP_AT_NEXT_INST and enables it in the device tree > can be found here: > https://lists.ozlabs.org/pipermail/skiboot/2016-September/004805.html Which says "This will reduce the exit latency of the idle state", and yet it doesn't change any of the latency/residency values? If all it does is reduce the exit latency, then shouldn't we always use it? Presumably it also saves less power? > diff --git a/arch/powerpc/kernel/idle_book3s.S > b/arch/powerpc/kernel/idle_book3s.S > index 32d666b..47ee106 100644 > --- a/arch/powerpc/kernel/idle_book3s.S > +++ b/arch/powerpc/kernel/idle_book3s.S > @@ -43,6 +43,8 @@ > #define PSSCR_HV_TEMPLATEPSSCR_ESL | PSSCR_EC | \ > PSSCR_PSLL_MASK | PSSCR_TR_MASK | \ > PSSCR_MTL_MASK > +#define PSSCR_HV_TEMPLATE_LITE PSSCR_PSLL_MASK | PSSCR_TR_MASK | \ > + PSSCR_MTL_MASK Why do we have these at all? Firmware tells us the PSSCR values to use in the "ibm,cpu-idle-state-psscr" property. If we just used what firmware gave us then we wouldn't need the above, or the juggling below. > @@ -333,13 +349,19 @@ ALT_FTR_SECTION_END_NESTED_IFSET(CPU_FTR_ARCH_207S, > 66);\ > > /* > * r3 - requested stop state > + * r4 - Indicates if the lite variant with ESL=EC=0 should be executed. > */ > _GLOBAL(power9_idle_stop) > - LOAD_REG_IMMEDIATE(r4, PSSCR_HV_TEMPLATE) > - or r4,r4,r3 > + cmpdi r4, 1 > + bne 4f > + LOAD_REG_IMMEDIATE(r4, PSSCR_HV_TEMPLATE_LITE) > + LOAD_REG_ADDR(r5,power_enter_stop_lite) > + b 5f > +4: LOAD_REG_IMMEDIATE(r4, PSSCR_HV_TEMPLATE) > + LOAD_REG_ADDR(r5,power_enter_stop) > +5: or r4,r4,r3 > mtspr SPRN_PSSCR, r4 > li r4, 1 > - LOAD_REG_ADDR(r5,power_enter_stop) > b pnv_powersave_common > /* No return */ > /* > diff --git a/arch/powerpc/platforms/powernv/idle.c > b/arch/powerpc/platforms/powernv/idle.c > index 479c256..c3d3fed 100644 > --- a/arch/powerpc/platforms/powernv/idle.c > +++ b/arch/powerpc/platforms/powernv/idle.c > @@ -244,8 +244,15 @@ static DEVICE_ATTR(fastsleep_workaround_applyonce, 0600, > static void power9_idle(void) > { > /* Requesting stop state 0 */ > - power9_idle_stop(0); > } That seems like the root of the problem, why aren't we passing a PSSCR value here? I also don't see us using the psscr-mask property anywhere. Why isn't that a bug? cheers
Re: [PATCH 2/2] powernv:idle:Implement lite variant of power_enter_stop
"Gautham R. Shenoy" writes: > From: "Gautham R. Shenoy" > > This patch adds a function named power_enter_stop_lite() that can > execute a stop instruction when ESL and EC bits are set to zero in the > PSSCR. The function handles the wake-up from idle at the instruction > immediately after the stop instruction. > > If the flag OPAL_PM_WAKEUP_AT_NEXT_INST[1] is set in the device tree > for a stop state, then use the lite variant for that particular stop > state. Hi Gautham, It seems to me that this would be cleaner if it was modelled as a new stop state? Surely it must have different power saving characteristics to the existing state? > [1] : The corresponding patch in skiboot that defines > OPAL_PM_WAKEUP_AT_NEXT_INST and enables it in the device tree > can be found here: > https://lists.ozlabs.org/pipermail/skiboot/2016-September/004805.html Which says "This will reduce the exit latency of the idle state", and yet it doesn't change any of the latency/residency values? If all it does is reduce the exit latency, then shouldn't we always use it? Presumably it also saves less power? > diff --git a/arch/powerpc/kernel/idle_book3s.S > b/arch/powerpc/kernel/idle_book3s.S > index 32d666b..47ee106 100644 > --- a/arch/powerpc/kernel/idle_book3s.S > +++ b/arch/powerpc/kernel/idle_book3s.S > @@ -43,6 +43,8 @@ > #define PSSCR_HV_TEMPLATEPSSCR_ESL | PSSCR_EC | \ > PSSCR_PSLL_MASK | PSSCR_TR_MASK | \ > PSSCR_MTL_MASK > +#define PSSCR_HV_TEMPLATE_LITE PSSCR_PSLL_MASK | PSSCR_TR_MASK | \ > + PSSCR_MTL_MASK Why do we have these at all? Firmware tells us the PSSCR values to use in the "ibm,cpu-idle-state-psscr" property. If we just used what firmware gave us then we wouldn't need the above, or the juggling below. > @@ -333,13 +349,19 @@ ALT_FTR_SECTION_END_NESTED_IFSET(CPU_FTR_ARCH_207S, > 66);\ > > /* > * r3 - requested stop state > + * r4 - Indicates if the lite variant with ESL=EC=0 should be executed. > */ > _GLOBAL(power9_idle_stop) > - LOAD_REG_IMMEDIATE(r4, PSSCR_HV_TEMPLATE) > - or r4,r4,r3 > + cmpdi r4, 1 > + bne 4f > + LOAD_REG_IMMEDIATE(r4, PSSCR_HV_TEMPLATE_LITE) > + LOAD_REG_ADDR(r5,power_enter_stop_lite) > + b 5f > +4: LOAD_REG_IMMEDIATE(r4, PSSCR_HV_TEMPLATE) > + LOAD_REG_ADDR(r5,power_enter_stop) > +5: or r4,r4,r3 > mtspr SPRN_PSSCR, r4 > li r4, 1 > - LOAD_REG_ADDR(r5,power_enter_stop) > b pnv_powersave_common > /* No return */ > /* > diff --git a/arch/powerpc/platforms/powernv/idle.c > b/arch/powerpc/platforms/powernv/idle.c > index 479c256..c3d3fed 100644 > --- a/arch/powerpc/platforms/powernv/idle.c > +++ b/arch/powerpc/platforms/powernv/idle.c > @@ -244,8 +244,15 @@ static DEVICE_ATTR(fastsleep_workaround_applyonce, 0600, > static void power9_idle(void) > { > /* Requesting stop state 0 */ > - power9_idle_stop(0); > } That seems like the root of the problem, why aren't we passing a PSSCR value here? I also don't see us using the psscr-mask property anywhere. Why isn't that a bug? cheers
Re: [PATCH 2/2] powernv:idle:Implement lite variant of power_enter_stop
Hi Balbir, On Tue, Sep 20, 2016 at 03:54:43PM +1000, Balbir Singh wrote: > > diff --git a/arch/powerpc/platforms/powernv/idle.c > > b/arch/powerpc/platforms/powernv/idle.c > > index 479c256..c3d3fed 100644 > > --- a/arch/powerpc/platforms/powernv/idle.c > > +++ b/arch/powerpc/platforms/powernv/idle.c > > @@ -244,8 +244,15 @@ static DEVICE_ATTR(fastsleep_workaround_applyonce, > > 0600, > > static void power9_idle(void) > > { > > /* Requesting stop state 0 */ > > - power9_idle_stop(0); > > + power9_idle_stop(0, 0); > > } > > + > > +static void power9_idle_lite(void) > > +{ > > + /* Requesting stop state 0 with ESL=EC=0 */ > > + power9_idle_stop(0, 1); > > +} > > + > > /* > > * First deep stop state. Used to figure out when to save/restore > > * hypervisor context. > > @@ -414,8 +421,12 @@ static int __init pnv_init_idle_states(void) > > > > if (supported_cpuidle_states & OPAL_PM_NAP_ENABLED) > > ppc_md.power_save = power7_idle; > > - else if (supported_cpuidle_states & OPAL_PM_STOP_INST_FAST) > > - ppc_md.power_save = power9_idle; > > + else if (supported_cpuidle_states & OPAL_PM_STOP_INST_FAST) { > > + if (supported_cpuidle_states & OPAL_PM_WAKEUP_AT_NEXT_INST) > > + ppc_md.power_save = power9_idle_lite; > > + else > > + ppc_md.power_save = power9_idle; > > + } > > If I am reading this correctly we decide at boot time whether we support > wakeup at next instruction and make that the default sleep state. > I am a little surprised that these are exclusive. I was expecting > power9_idle_lite to be one of the states to go into before > power9_idle At boot time, we initialize ppc_md.power_save to power9_idle/power9_idle_lite which ends up being the default idle function in the absence of the cpuidle subsystem. When cpuidle is available, idle code will call cpuidle governors which will determine the appropriate idle state that can be entered into. Each of these idle states has an associated callback function. In case of the idle-states without OPAL_PM_STOP_INST_FAST associated with them, the callback is stop_loop() and when the flag is set, the callback function is stop_lite_loop(). So when cpuidle is present, these states are not exclusive. Note that both power9_idle() and power9_idle_lite() call stop0. Just that the former executes stop0 with ESL=EC=1 and latter with ESL=EC=0. That said, you're right that in the absence of the cpuidle governor, if the lite variant of stop is advertised by the firmware through the device-tree, we end up picking the liter version of stop0 as the default idle state. Do you suggest that we retain power9_idle which calls stop0 with ESL=EC=1 ? > Balbir Singh. > -- Thanks and Regards gautham.
Re: [PATCH 2/2] powernv:idle:Implement lite variant of power_enter_stop
Hi Balbir, On Tue, Sep 20, 2016 at 03:54:43PM +1000, Balbir Singh wrote: > > diff --git a/arch/powerpc/platforms/powernv/idle.c > > b/arch/powerpc/platforms/powernv/idle.c > > index 479c256..c3d3fed 100644 > > --- a/arch/powerpc/platforms/powernv/idle.c > > +++ b/arch/powerpc/platforms/powernv/idle.c > > @@ -244,8 +244,15 @@ static DEVICE_ATTR(fastsleep_workaround_applyonce, > > 0600, > > static void power9_idle(void) > > { > > /* Requesting stop state 0 */ > > - power9_idle_stop(0); > > + power9_idle_stop(0, 0); > > } > > + > > +static void power9_idle_lite(void) > > +{ > > + /* Requesting stop state 0 with ESL=EC=0 */ > > + power9_idle_stop(0, 1); > > +} > > + > > /* > > * First deep stop state. Used to figure out when to save/restore > > * hypervisor context. > > @@ -414,8 +421,12 @@ static int __init pnv_init_idle_states(void) > > > > if (supported_cpuidle_states & OPAL_PM_NAP_ENABLED) > > ppc_md.power_save = power7_idle; > > - else if (supported_cpuidle_states & OPAL_PM_STOP_INST_FAST) > > - ppc_md.power_save = power9_idle; > > + else if (supported_cpuidle_states & OPAL_PM_STOP_INST_FAST) { > > + if (supported_cpuidle_states & OPAL_PM_WAKEUP_AT_NEXT_INST) > > + ppc_md.power_save = power9_idle_lite; > > + else > > + ppc_md.power_save = power9_idle; > > + } > > If I am reading this correctly we decide at boot time whether we support > wakeup at next instruction and make that the default sleep state. > I am a little surprised that these are exclusive. I was expecting > power9_idle_lite to be one of the states to go into before > power9_idle At boot time, we initialize ppc_md.power_save to power9_idle/power9_idle_lite which ends up being the default idle function in the absence of the cpuidle subsystem. When cpuidle is available, idle code will call cpuidle governors which will determine the appropriate idle state that can be entered into. Each of these idle states has an associated callback function. In case of the idle-states without OPAL_PM_STOP_INST_FAST associated with them, the callback is stop_loop() and when the flag is set, the callback function is stop_lite_loop(). So when cpuidle is present, these states are not exclusive. Note that both power9_idle() and power9_idle_lite() call stop0. Just that the former executes stop0 with ESL=EC=1 and latter with ESL=EC=0. That said, you're right that in the absence of the cpuidle governor, if the lite variant of stop is advertised by the firmware through the device-tree, we end up picking the liter version of stop0 as the default idle state. Do you suggest that we retain power9_idle which calls stop0 with ESL=EC=1 ? > Balbir Singh. > -- Thanks and Regards gautham.
Re: [PATCH 2/2] powernv:idle:Implement lite variant of power_enter_stop
On 16/09/16 19:47, Gautham R. Shenoy wrote: > From: "Gautham R. Shenoy"> > This patch adds a function named power_enter_stop_lite() that can > execute a stop instruction when ESL and EC bits are set to zero in the > PSSCR. The function handles the wake-up from idle at the instruction > immediately after the stop instruction. > > If the flag OPAL_PM_WAKEUP_AT_NEXT_INST[1] is set in the device tree > for a stop state, then use the lite variant for that particular stop > state. > > [1] : The corresponding patch in skiboot that defines > OPAL_PM_WAKEUP_AT_NEXT_INST and enables it in the device tree > can be found here: > https://lists.ozlabs.org/pipermail/skiboot/2016-September/004805.html > > Signed-off-by: Gautham R. Shenoy > --- > arch/powerpc/include/asm/opal-api.h | 1 + > arch/powerpc/include/asm/processor.h | 3 ++- > arch/powerpc/kernel/idle_book3s.S | 28 +--- > arch/powerpc/platforms/powernv/idle.c | 17 ++--- > arch/powerpc/platforms/powernv/smp.c | 2 +- > drivers/cpuidle/cpuidle-powernv.c | 24 ++-- > 6 files changed, 65 insertions(+), 10 deletions(-) > > diff --git a/arch/powerpc/include/asm/opal-api.h > b/arch/powerpc/include/asm/opal-api.h > index 0e2e57b..6e5741e 100644 > --- a/arch/powerpc/include/asm/opal-api.h > +++ b/arch/powerpc/include/asm/opal-api.h > @@ -179,6 +179,7 @@ > #define OPAL_PM_TIMEBASE_STOP0x0002 > #define OPAL_PM_LOSE_HYP_CONTEXT 0x2000 > #define OPAL_PM_LOSE_FULL_CONTEXT0x4000 > +#define OPAL_PM_WAKEUP_AT_NEXT_INST 0x8000 > #define OPAL_PM_NAP_ENABLED 0x0001 > #define OPAL_PM_SLEEP_ENABLED0x0002 > #define OPAL_PM_WINKLE_ENABLED 0x0004 > diff --git a/arch/powerpc/include/asm/processor.h > b/arch/powerpc/include/asm/processor.h > index 68e3bf5..e0549a0 100644 > --- a/arch/powerpc/include/asm/processor.h > +++ b/arch/powerpc/include/asm/processor.h > @@ -460,7 +460,8 @@ 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 power9_idle_stop(unsigned long stop_level); > +extern unsigned long power9_idle_stop(unsigned long stop_level, > + unsigned long exec_lite); > > extern void flush_instruction_cache(void); > extern void hard_reset_now(void); > diff --git a/arch/powerpc/kernel/idle_book3s.S > b/arch/powerpc/kernel/idle_book3s.S > index 32d666b..47ee106 100644 > --- a/arch/powerpc/kernel/idle_book3s.S > +++ b/arch/powerpc/kernel/idle_book3s.S > @@ -43,6 +43,8 @@ > #define PSSCR_HV_TEMPLATEPSSCR_ESL | PSSCR_EC | \ > PSSCR_PSLL_MASK | PSSCR_TR_MASK | \ > PSSCR_MTL_MASK > +#define PSSCR_HV_TEMPLATE_LITE PSSCR_PSLL_MASK | PSSCR_TR_MASK | \ > + PSSCR_MTL_MASK > > .text > > @@ -246,6 +248,20 @@ enter_winkle: > > IDLE_STATE_ENTER_SEQ_NORET(PPC_WINKLE) > > + > +/* > + * power_enter_stop_lite : This will resume the wake up from > + * idle at the subsequent instruction. > + * > + * Caller should set ESL=EC=0 in PSSCR before calling > + * this function. > + * > + */ > +power_enter_stop_lite: > + IDLE_STATE_ENTER_SEQ(PPC_STOP) > +7: li r3,0 /* Since we didn't lose state, return 0 */ > + b pnv_wakeup_noloss > + > /* > * r3 - requested stop state > */ > @@ -333,13 +349,19 @@ ALT_FTR_SECTION_END_NESTED_IFSET(CPU_FTR_ARCH_207S, > 66);\ > > /* > * r3 - requested stop state > + * r4 - Indicates if the lite variant with ESL=EC=0 should be executed. > */ > _GLOBAL(power9_idle_stop) > - LOAD_REG_IMMEDIATE(r4, PSSCR_HV_TEMPLATE) > - or r4,r4,r3 > + cmpdi r4, 1 > + bne 4f > + LOAD_REG_IMMEDIATE(r4, PSSCR_HV_TEMPLATE_LITE) > + LOAD_REG_ADDR(r5,power_enter_stop_lite) > + b 5f > +4: LOAD_REG_IMMEDIATE(r4, PSSCR_HV_TEMPLATE) > + LOAD_REG_ADDR(r5,power_enter_stop) > +5: or r4,r4,r3 > mtspr SPRN_PSSCR, r4 > li r4, 1 > - LOAD_REG_ADDR(r5,power_enter_stop) > b pnv_powersave_common > /* No return */ > /* > diff --git a/arch/powerpc/platforms/powernv/idle.c > b/arch/powerpc/platforms/powernv/idle.c > index 479c256..c3d3fed 100644 > --- a/arch/powerpc/platforms/powernv/idle.c > +++ b/arch/powerpc/platforms/powernv/idle.c > @@ -244,8 +244,15 @@ static DEVICE_ATTR(fastsleep_workaround_applyonce, 0600, > static void power9_idle(void) > { > /* Requesting stop state 0 */ > - power9_idle_stop(0); > + power9_idle_stop(0, 0); > } > + > +static void power9_idle_lite(void) > +{ > + /* Requesting stop state 0 with ESL=EC=0 */ > + power9_idle_stop(0, 1); > +} > + >
Re: [PATCH 2/2] powernv:idle:Implement lite variant of power_enter_stop
On 16/09/16 19:47, Gautham R. Shenoy wrote: > From: "Gautham R. Shenoy" > > This patch adds a function named power_enter_stop_lite() that can > execute a stop instruction when ESL and EC bits are set to zero in the > PSSCR. The function handles the wake-up from idle at the instruction > immediately after the stop instruction. > > If the flag OPAL_PM_WAKEUP_AT_NEXT_INST[1] is set in the device tree > for a stop state, then use the lite variant for that particular stop > state. > > [1] : The corresponding patch in skiboot that defines > OPAL_PM_WAKEUP_AT_NEXT_INST and enables it in the device tree > can be found here: > https://lists.ozlabs.org/pipermail/skiboot/2016-September/004805.html > > Signed-off-by: Gautham R. Shenoy > --- > arch/powerpc/include/asm/opal-api.h | 1 + > arch/powerpc/include/asm/processor.h | 3 ++- > arch/powerpc/kernel/idle_book3s.S | 28 +--- > arch/powerpc/platforms/powernv/idle.c | 17 ++--- > arch/powerpc/platforms/powernv/smp.c | 2 +- > drivers/cpuidle/cpuidle-powernv.c | 24 ++-- > 6 files changed, 65 insertions(+), 10 deletions(-) > > diff --git a/arch/powerpc/include/asm/opal-api.h > b/arch/powerpc/include/asm/opal-api.h > index 0e2e57b..6e5741e 100644 > --- a/arch/powerpc/include/asm/opal-api.h > +++ b/arch/powerpc/include/asm/opal-api.h > @@ -179,6 +179,7 @@ > #define OPAL_PM_TIMEBASE_STOP0x0002 > #define OPAL_PM_LOSE_HYP_CONTEXT 0x2000 > #define OPAL_PM_LOSE_FULL_CONTEXT0x4000 > +#define OPAL_PM_WAKEUP_AT_NEXT_INST 0x8000 > #define OPAL_PM_NAP_ENABLED 0x0001 > #define OPAL_PM_SLEEP_ENABLED0x0002 > #define OPAL_PM_WINKLE_ENABLED 0x0004 > diff --git a/arch/powerpc/include/asm/processor.h > b/arch/powerpc/include/asm/processor.h > index 68e3bf5..e0549a0 100644 > --- a/arch/powerpc/include/asm/processor.h > +++ b/arch/powerpc/include/asm/processor.h > @@ -460,7 +460,8 @@ 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 power9_idle_stop(unsigned long stop_level); > +extern unsigned long power9_idle_stop(unsigned long stop_level, > + unsigned long exec_lite); > > extern void flush_instruction_cache(void); > extern void hard_reset_now(void); > diff --git a/arch/powerpc/kernel/idle_book3s.S > b/arch/powerpc/kernel/idle_book3s.S > index 32d666b..47ee106 100644 > --- a/arch/powerpc/kernel/idle_book3s.S > +++ b/arch/powerpc/kernel/idle_book3s.S > @@ -43,6 +43,8 @@ > #define PSSCR_HV_TEMPLATEPSSCR_ESL | PSSCR_EC | \ > PSSCR_PSLL_MASK | PSSCR_TR_MASK | \ > PSSCR_MTL_MASK > +#define PSSCR_HV_TEMPLATE_LITE PSSCR_PSLL_MASK | PSSCR_TR_MASK | \ > + PSSCR_MTL_MASK > > .text > > @@ -246,6 +248,20 @@ enter_winkle: > > IDLE_STATE_ENTER_SEQ_NORET(PPC_WINKLE) > > + > +/* > + * power_enter_stop_lite : This will resume the wake up from > + * idle at the subsequent instruction. > + * > + * Caller should set ESL=EC=0 in PSSCR before calling > + * this function. > + * > + */ > +power_enter_stop_lite: > + IDLE_STATE_ENTER_SEQ(PPC_STOP) > +7: li r3,0 /* Since we didn't lose state, return 0 */ > + b pnv_wakeup_noloss > + > /* > * r3 - requested stop state > */ > @@ -333,13 +349,19 @@ ALT_FTR_SECTION_END_NESTED_IFSET(CPU_FTR_ARCH_207S, > 66);\ > > /* > * r3 - requested stop state > + * r4 - Indicates if the lite variant with ESL=EC=0 should be executed. > */ > _GLOBAL(power9_idle_stop) > - LOAD_REG_IMMEDIATE(r4, PSSCR_HV_TEMPLATE) > - or r4,r4,r3 > + cmpdi r4, 1 > + bne 4f > + LOAD_REG_IMMEDIATE(r4, PSSCR_HV_TEMPLATE_LITE) > + LOAD_REG_ADDR(r5,power_enter_stop_lite) > + b 5f > +4: LOAD_REG_IMMEDIATE(r4, PSSCR_HV_TEMPLATE) > + LOAD_REG_ADDR(r5,power_enter_stop) > +5: or r4,r4,r3 > mtspr SPRN_PSSCR, r4 > li r4, 1 > - LOAD_REG_ADDR(r5,power_enter_stop) > b pnv_powersave_common > /* No return */ > /* > diff --git a/arch/powerpc/platforms/powernv/idle.c > b/arch/powerpc/platforms/powernv/idle.c > index 479c256..c3d3fed 100644 > --- a/arch/powerpc/platforms/powernv/idle.c > +++ b/arch/powerpc/platforms/powernv/idle.c > @@ -244,8 +244,15 @@ static DEVICE_ATTR(fastsleep_workaround_applyonce, 0600, > static void power9_idle(void) > { > /* Requesting stop state 0 */ > - power9_idle_stop(0); > + power9_idle_stop(0, 0); > } > + > +static void power9_idle_lite(void) > +{ > + /* Requesting stop state 0 with ESL=EC=0 */ > + power9_idle_stop(0, 1); > +} > + > /* > * First deep stop state. Used to figure out
[PATCH 2/2] powernv:idle:Implement lite variant of power_enter_stop
From: "Gautham R. Shenoy"This patch adds a function named power_enter_stop_lite() that can execute a stop instruction when ESL and EC bits are set to zero in the PSSCR. The function handles the wake-up from idle at the instruction immediately after the stop instruction. If the flag OPAL_PM_WAKEUP_AT_NEXT_INST[1] is set in the device tree for a stop state, then use the lite variant for that particular stop state. [1] : The corresponding patch in skiboot that defines OPAL_PM_WAKEUP_AT_NEXT_INST and enables it in the device tree can be found here: https://lists.ozlabs.org/pipermail/skiboot/2016-September/004805.html Signed-off-by: Gautham R. Shenoy --- arch/powerpc/include/asm/opal-api.h | 1 + arch/powerpc/include/asm/processor.h | 3 ++- arch/powerpc/kernel/idle_book3s.S | 28 +--- arch/powerpc/platforms/powernv/idle.c | 17 ++--- arch/powerpc/platforms/powernv/smp.c | 2 +- drivers/cpuidle/cpuidle-powernv.c | 24 ++-- 6 files changed, 65 insertions(+), 10 deletions(-) diff --git a/arch/powerpc/include/asm/opal-api.h b/arch/powerpc/include/asm/opal-api.h index 0e2e57b..6e5741e 100644 --- a/arch/powerpc/include/asm/opal-api.h +++ b/arch/powerpc/include/asm/opal-api.h @@ -179,6 +179,7 @@ #define OPAL_PM_TIMEBASE_STOP 0x0002 #define OPAL_PM_LOSE_HYP_CONTEXT 0x2000 #define OPAL_PM_LOSE_FULL_CONTEXT 0x4000 +#define OPAL_PM_WAKEUP_AT_NEXT_INST0x8000 #define OPAL_PM_NAP_ENABLED0x0001 #define OPAL_PM_SLEEP_ENABLED 0x0002 #define OPAL_PM_WINKLE_ENABLED 0x0004 diff --git a/arch/powerpc/include/asm/processor.h b/arch/powerpc/include/asm/processor.h index 68e3bf5..e0549a0 100644 --- a/arch/powerpc/include/asm/processor.h +++ b/arch/powerpc/include/asm/processor.h @@ -460,7 +460,8 @@ 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 power9_idle_stop(unsigned long stop_level); +extern unsigned long power9_idle_stop(unsigned long stop_level, + unsigned long exec_lite); extern void flush_instruction_cache(void); extern void hard_reset_now(void); diff --git a/arch/powerpc/kernel/idle_book3s.S b/arch/powerpc/kernel/idle_book3s.S index 32d666b..47ee106 100644 --- a/arch/powerpc/kernel/idle_book3s.S +++ b/arch/powerpc/kernel/idle_book3s.S @@ -43,6 +43,8 @@ #define PSSCR_HV_TEMPLATE PSSCR_ESL | PSSCR_EC | \ PSSCR_PSLL_MASK | PSSCR_TR_MASK | \ PSSCR_MTL_MASK +#define PSSCR_HV_TEMPLATE_LITE PSSCR_PSLL_MASK | PSSCR_TR_MASK | \ +PSSCR_MTL_MASK .text @@ -246,6 +248,20 @@ enter_winkle: IDLE_STATE_ENTER_SEQ_NORET(PPC_WINKLE) + +/* + * power_enter_stop_lite : This will resume the wake up from + * idle at the subsequent instruction. + * + * Caller should set ESL=EC=0 in PSSCR before calling + * this function. + * + */ +power_enter_stop_lite: + IDLE_STATE_ENTER_SEQ(PPC_STOP) +7: li r3,0 /* Since we didn't lose state, return 0 */ + b pnv_wakeup_noloss + /* * r3 - requested stop state */ @@ -333,13 +349,19 @@ ALT_FTR_SECTION_END_NESTED_IFSET(CPU_FTR_ARCH_207S, 66); \ /* * r3 - requested stop state + * r4 - Indicates if the lite variant with ESL=EC=0 should be executed. */ _GLOBAL(power9_idle_stop) - LOAD_REG_IMMEDIATE(r4, PSSCR_HV_TEMPLATE) - or r4,r4,r3 + cmpdi r4, 1 + bne 4f + LOAD_REG_IMMEDIATE(r4, PSSCR_HV_TEMPLATE_LITE) + LOAD_REG_ADDR(r5,power_enter_stop_lite) + b 5f +4: LOAD_REG_IMMEDIATE(r4, PSSCR_HV_TEMPLATE) + LOAD_REG_ADDR(r5,power_enter_stop) +5: or r4,r4,r3 mtspr SPRN_PSSCR, r4 li r4, 1 - LOAD_REG_ADDR(r5,power_enter_stop) b pnv_powersave_common /* No return */ /* diff --git a/arch/powerpc/platforms/powernv/idle.c b/arch/powerpc/platforms/powernv/idle.c index 479c256..c3d3fed 100644 --- a/arch/powerpc/platforms/powernv/idle.c +++ b/arch/powerpc/platforms/powernv/idle.c @@ -244,8 +244,15 @@ static DEVICE_ATTR(fastsleep_workaround_applyonce, 0600, static void power9_idle(void) { /* Requesting stop state 0 */ - power9_idle_stop(0); + power9_idle_stop(0, 0); } + +static void power9_idle_lite(void) +{ + /* Requesting stop state 0 with ESL=EC=0 */ + power9_idle_stop(0, 1); +} + /* * First deep stop state. Used to figure out when to save/restore * hypervisor context. @@ -414,8 +421,12 @@ static int __init pnv_init_idle_states(void) if (supported_cpuidle_states & OPAL_PM_NAP_ENABLED) ppc_md.power_save = power7_idle; -
[PATCH 2/2] powernv:idle:Implement lite variant of power_enter_stop
From: "Gautham R. Shenoy" This patch adds a function named power_enter_stop_lite() that can execute a stop instruction when ESL and EC bits are set to zero in the PSSCR. The function handles the wake-up from idle at the instruction immediately after the stop instruction. If the flag OPAL_PM_WAKEUP_AT_NEXT_INST[1] is set in the device tree for a stop state, then use the lite variant for that particular stop state. [1] : The corresponding patch in skiboot that defines OPAL_PM_WAKEUP_AT_NEXT_INST and enables it in the device tree can be found here: https://lists.ozlabs.org/pipermail/skiboot/2016-September/004805.html Signed-off-by: Gautham R. Shenoy --- arch/powerpc/include/asm/opal-api.h | 1 + arch/powerpc/include/asm/processor.h | 3 ++- arch/powerpc/kernel/idle_book3s.S | 28 +--- arch/powerpc/platforms/powernv/idle.c | 17 ++--- arch/powerpc/platforms/powernv/smp.c | 2 +- drivers/cpuidle/cpuidle-powernv.c | 24 ++-- 6 files changed, 65 insertions(+), 10 deletions(-) diff --git a/arch/powerpc/include/asm/opal-api.h b/arch/powerpc/include/asm/opal-api.h index 0e2e57b..6e5741e 100644 --- a/arch/powerpc/include/asm/opal-api.h +++ b/arch/powerpc/include/asm/opal-api.h @@ -179,6 +179,7 @@ #define OPAL_PM_TIMEBASE_STOP 0x0002 #define OPAL_PM_LOSE_HYP_CONTEXT 0x2000 #define OPAL_PM_LOSE_FULL_CONTEXT 0x4000 +#define OPAL_PM_WAKEUP_AT_NEXT_INST0x8000 #define OPAL_PM_NAP_ENABLED0x0001 #define OPAL_PM_SLEEP_ENABLED 0x0002 #define OPAL_PM_WINKLE_ENABLED 0x0004 diff --git a/arch/powerpc/include/asm/processor.h b/arch/powerpc/include/asm/processor.h index 68e3bf5..e0549a0 100644 --- a/arch/powerpc/include/asm/processor.h +++ b/arch/powerpc/include/asm/processor.h @@ -460,7 +460,8 @@ 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 power9_idle_stop(unsigned long stop_level); +extern unsigned long power9_idle_stop(unsigned long stop_level, + unsigned long exec_lite); extern void flush_instruction_cache(void); extern void hard_reset_now(void); diff --git a/arch/powerpc/kernel/idle_book3s.S b/arch/powerpc/kernel/idle_book3s.S index 32d666b..47ee106 100644 --- a/arch/powerpc/kernel/idle_book3s.S +++ b/arch/powerpc/kernel/idle_book3s.S @@ -43,6 +43,8 @@ #define PSSCR_HV_TEMPLATE PSSCR_ESL | PSSCR_EC | \ PSSCR_PSLL_MASK | PSSCR_TR_MASK | \ PSSCR_MTL_MASK +#define PSSCR_HV_TEMPLATE_LITE PSSCR_PSLL_MASK | PSSCR_TR_MASK | \ +PSSCR_MTL_MASK .text @@ -246,6 +248,20 @@ enter_winkle: IDLE_STATE_ENTER_SEQ_NORET(PPC_WINKLE) + +/* + * power_enter_stop_lite : This will resume the wake up from + * idle at the subsequent instruction. + * + * Caller should set ESL=EC=0 in PSSCR before calling + * this function. + * + */ +power_enter_stop_lite: + IDLE_STATE_ENTER_SEQ(PPC_STOP) +7: li r3,0 /* Since we didn't lose state, return 0 */ + b pnv_wakeup_noloss + /* * r3 - requested stop state */ @@ -333,13 +349,19 @@ ALT_FTR_SECTION_END_NESTED_IFSET(CPU_FTR_ARCH_207S, 66); \ /* * r3 - requested stop state + * r4 - Indicates if the lite variant with ESL=EC=0 should be executed. */ _GLOBAL(power9_idle_stop) - LOAD_REG_IMMEDIATE(r4, PSSCR_HV_TEMPLATE) - or r4,r4,r3 + cmpdi r4, 1 + bne 4f + LOAD_REG_IMMEDIATE(r4, PSSCR_HV_TEMPLATE_LITE) + LOAD_REG_ADDR(r5,power_enter_stop_lite) + b 5f +4: LOAD_REG_IMMEDIATE(r4, PSSCR_HV_TEMPLATE) + LOAD_REG_ADDR(r5,power_enter_stop) +5: or r4,r4,r3 mtspr SPRN_PSSCR, r4 li r4, 1 - LOAD_REG_ADDR(r5,power_enter_stop) b pnv_powersave_common /* No return */ /* diff --git a/arch/powerpc/platforms/powernv/idle.c b/arch/powerpc/platforms/powernv/idle.c index 479c256..c3d3fed 100644 --- a/arch/powerpc/platforms/powernv/idle.c +++ b/arch/powerpc/platforms/powernv/idle.c @@ -244,8 +244,15 @@ static DEVICE_ATTR(fastsleep_workaround_applyonce, 0600, static void power9_idle(void) { /* Requesting stop state 0 */ - power9_idle_stop(0); + power9_idle_stop(0, 0); } + +static void power9_idle_lite(void) +{ + /* Requesting stop state 0 with ESL=EC=0 */ + power9_idle_stop(0, 1); +} + /* * First deep stop state. Used to figure out when to save/restore * hypervisor context. @@ -414,8 +421,12 @@ static int __init pnv_init_idle_states(void) if (supported_cpuidle_states & OPAL_PM_NAP_ENABLED) ppc_md.power_save = power7_idle; - else if (supported_cpuidle_states &