ARM: tegra: update the cache maintenance order for CPU shutdown
Subsystems (3)
PATCH 1/2 ARM: tegra: update the cache maintenance order for CPU shutdown
Updating the cache maintenance order before CPU shutdown when doing CPU hotplug.
The old order:
* clean L1 by flush_cache_all
* exit SMP
* CPU shutdown
Adapt to:
* disable L1 data cache by clear C bit
* clean L1 by v7_flush_dcache_louis
* exit SMP
* CPU shutdown
For CPU hotplug case, it's no need to do "flush_cache_all". And we should disable L1 data cache before clean L1 data cache. Then leaving the SMP coherency.
--- arch/arm/mach-tegra/hotplug.c | 4 ++-- arch/arm/mach-tegra/sleep-tegra20.S | 3 --- arch/arm/mach-tegra/sleep-tegra30.S | 3 --- arch/arm/mach-tegra/sleep.S | 4 +++- arch/arm/mach-tegra/sleep.h | 1 + 5 files changed, 6 insertions(+), 9 deletions(-)
arch/arm/mach-tegra/hotplug.c +2 -2
@@ -28,8 +28,8 @@ void __ref tegra_cpu_die(unsigned int cpu) { cpu =cpu_logical_map (cpu); -/* Flush the L1 data cache. */ -flush_cache_all (); +/* Clean L1 data cache */ +tegra_disable_clean_inv_dcache ();/* Shut down the current CPU. */ tegra_hotplug_shutdown ();
arch/arm/mach-tegra/sleep-tegra20.S +0 -3
@@ -33,9 +33,6 @@ * should never return */ ENTRY(tegra20_hotplug_shutdown) - /* Turn off SMP coherency */ - exit_smp r4, r5 - /* Put this CPU down */ cpu_id r0 bl tegra20_cpu_shutdown
arch/arm/mach-tegra/sleep-tegra30.S +0 -3
@@ -32,9 +32,6 @@ * Should never return. */ ENTRY(tegra30_hotplug_shutdown) - /* Turn off SMP coherency */ - exit_smp r4, r5 - /* Powergate this CPU */ mov r0, #TEGRA30_POWER_HOTPLUG_SHUTDOWN bl tegra30_cpu_shutdown
arch/arm/mach-tegra/sleep.S +3 -1
@@ -34,7 +34,7 @@ #include "flowctrl.h" #include "sleep.h" -#ifdef CONFIG_PM_SLEEP +#if defined(CONFIG_HOTPLUG_CPU) || defined(CONFIG_PM_SLEEP) /* * tegra_disable_clean_inv_dcache *
@@ -60,7 +60,9 @@ ENTRY(tegra_disable_clean_inv_dcache) ldmfd sp!, {r0, r4-r5, r7, r9-r11, pc} ENDPROC(tegra_disable_clean_inv_dcache) +#endif +#ifdef CONFIG_PM_SLEEP /* * tegra_sleep_cpu_finish(unsigned long v2p) *
arch/arm/mach-tegra/sleep.h +1 -0
@@ -106,6 +106,7 @@ exit_l2_resume:#else void tegra_resume (void );int tegra_sleep_cpu_finish (unsigned long ); +void tegra_disable_clean_inv_dcache (void );#ifdef CONFIG_HOTPLUG_CPU void tegra20_hotplug_init (void );
Signature · 1 line
1.7.0.4
PATCH 2/2 ARM: tegra: moving the clock gating procedure to tegra_cpu_kill
The tegra_cpu_die was be executed by the CPU itslf. So the clock gating procedure won't be executed after the CPU hardware shutdown code. Moving the clock gating procedure to tegra_cpu_kill that will be run by another CPU after the CPU died.
--- arch/arm/mach-tegra/common.h | 1 + arch/arm/mach-tegra/hotplug.c | 15 +++++++++++---- arch/arm/mach-tegra/platsmp.c | 1 + 3 files changed, 13 insertions(+), 4 deletions(-)
arch/arm/mach-tegra/common.h +1 -0
@@ -1,4 +1,5 @@extern struct smp_operations tegra_smp_ops; +extern int tegra_cpu_kill (unsigned int cpu);extern void tegra_cpu_die (unsigned int cpu);extern int tegra_cpu_disable (unsigned int cpu);
arch/arm/mach-tegra/hotplug.c +11 -4
@@ -19,6 +19,17 @@static void (*tegra_hotplug_shutdown)(void ); +int tegra_cpu_kill (unsigned cpu) +{ + cpu =cpu_logical_map (cpu); + +/* Clock gate the CPU */ +tegra_wait_cpu_in_reset (cpu); +tegra_disable_cpu_clock (cpu); + +return 1 ; +} +/* * platform-specific code to shutdown a CPU *
@@ -34,10 +45,6 @@ void __ref tegra_cpu_die(unsigned int cpu)/* Shut down the current CPU. */ tegra_hotplug_shutdown (); -/* Clock gate the CPU */ -tegra_wait_cpu_in_reset (cpu); -tegra_disable_cpu_clock (cpu); -/* Should never return here. */ BUG (); }
arch/arm/mach-tegra/platsmp.c +1 -0
@@ -175,6 +175,7 @@ struct smp_operations tegra_smp_ops __initdata = { .smp_secondary_init = tegra_secondary_init, .smp_boot_secondary = tegra_boot_secondary, #ifdef CONFIG_HOTPLUG_CPU + .cpu_kill = tegra_cpu_kill, .cpu_die = tegra_cpu_die, .cpu_disable = tegra_cpu_disable, #endif
Signature · 1 line
1.7.0.4
PATCH 2/2 ARM: tegra: moving the clock gating procedure to tegra_cpu_kill
Quoted 4 lines · Joseph Lo
The
tegra_cpu_diewas be executed by the CPU itslf. So the clock gating procedure won't be executed after the CPU hardware shutdown code. Moving the clock gating procedure totegra_cpu_killthat will be run by another CPU after the CPU died.
Hmmm. I wonder if this is enough to make kexec-on-Tegra-with-SMP-enabled work without explicitly hot-unplugging all the CPUs first... An implementation of cpu_kill() was a major part of what was missing. I thought an implementation of cpu_kill() would require a bunch of code from cpu_die() too. Does this patch assume cpu_die() has executed first, and only then cpu_kill() will work, or can cpu_kill() be used on its own?
arch/arm/mach-tegra/hotplug.c void __ref tegra_cpu_die(unsigned int cpu) · line 39
Quoted 9 lines
arch/arm/mach-tegra/hotplug.c +0
-3@@ -34,10 +45,6 @@ void __ref tegra_cpu_die(unsigned int cpu)/* Shut down the current CPU. */ tegra_hotplug_shutdown (); -/* Clock gate the CPU */ -tegra_wait_cpu_in_reset (cpu); -tegra_disable_cpu_clock (cpu);
I believe that's the last use of the calculation performed right at the start of this function:
cpu = cpu_logical_map(cpu);
Can you remove that?
Also, Peter, can you review these 2 patches as well?
PATCH 2/2 ARM: tegra: moving the clock gating procedure to tegra_cpu_kill
Quoted 6 lines · Stephen Warren
Quoted 4 lines · Joseph Lo
The
tegra_cpu_diewas be executed by the CPU itslf. So the clock gating procedure won't be executed after the CPU hardware shutdown code. Moving the clock gating procedure totegra_cpu_killthat will be run by another CPU after the CPU died.
Quoted 7 lines
Hmmm. I wonder if this is enough to make kexec-on-Tegra-with-SMP-enabled work without explicitly hot-unplugging all the CPUs first... An implementation of
cpu_kill()was a major part of what was missing. I thought an implementation ofcpu_kill()would require a bunch of code fromcpu_die()too. Does this patch assumecpu_die()has executed first, and only thencpu_kill()will work, or cancpu_kill()be used on its own?
Yes, the tegra_cpu_die must be run first then tegra_cpu_kill. Because the tegra_cpu_die shutdown the CPU by itself, then we can clock gate the CPU in tegra_cpu_kill. If the tegra_cpu_kill be run itself, the system will be hung in wait_for_reset(CPU).
About the kexec issue, it's complicate. Your solution should be the right solution. And we should implement the hardware shutdown code in cpu_die not cpu_kill. I suspect this issue can be reproduced on all ARM platform. Because the smp_send_stop can't really shutdown the secondary CPU core if just call machine_shutdown without disable_nonboot_cpu. And in platform_shutdown and platform_reboot case, the secondary CPU already been offlined and killed before this function be called. Does any other ARM SMP platform can run kexec without any issue?
(I will check if we can do some HW shutdown in tegra_cpu_kill. For ex, just clock gate the CPU without wait_for_reset.)
Thanks,
Joseph
PATCH 2/2 ARM: tegra: moving the clock gating procedure to tegra_cpu_kill
Quoted 28 lines · Joseph Lo
Quoted 6 lines · Stephen Warren
Quoted 4 lines · Joseph Lo
The
tegra_cpu_diewas be executed by the CPU itslf. So the clock gating procedure won't be executed after the CPU hardware shutdown code. Moving the clock gating procedure totegra_cpu_killthat will be run by another CPU after the CPU died.Quoted 7 lines
Hmmm. I wonder if this is enough to make kexec-on-Tegra-with-SMP-enabled work without explicitly hot-unplugging all the CPUs first... An implementation of
cpu_kill()was a major part of what was missing. I thought an implementation ofcpu_kill()would require a bunch of code fromcpu_die()too. Does this patch assumecpu_die()has executed first, and only thencpu_kill()will work, or cancpu_kill()be used on its own?Yes, the
tegra_cpu_diemust be run first thentegra_cpu_kill. Because thetegra_cpu_dieshutdown the CPU by itself, then we can clock gate the CPU integra_cpu_kill. If thetegra_cpu_killbe run itself, the system will be hung inwait_for_reset(CPU).About the kexec issue, it's complicate. Your solution should be the right solution. And we should implement the hardware shutdown code in
cpu_dienotcpu_kill. I suspect this issue can be reproduced on all ARM platform. Because thesmp_send_stopcan't really shutdown the secondary CPU core if just callmachine_shutdownwithoutdisable_nonboot_cpu. And inplatform_shutdownandplatform_rebootcase, the secondary CPU already been offlined and killed before this function be called. Does any other ARM SMP platform can run kexec without any issue?
Quoted 2 lines
(I will check if we can do some HW shutdown in
tegra_cpu_kill. For ex, just clock gate the CPU withoutwait_for_reset.)
Will, Joseph's thoughts above pretty much match mine re: kexec. In the kexec thread I started earlier, I talked about replacing:
void machine_shutdown(void)
{
#ifdef CONFIG_SMP smp_send_stop();
#endif
}
with something like:
void machine_shutdown(void)
{
#ifdef CONFIG_HOTPLUG_CPU
disable_nonboot_cpus();
#elifdef CONFIG_SMP
smp_send_stop();
#endif
}
and you'd responded "I think you're better off using what we currently have and hanging your code off platform_cpu_kill.".
What exactly did you mean by "what we currently have"; did you mean that cpu_kill() should work without cpu_die() having executed on the target CPU itself first, so Tegra should simply implement cpu_kill()? As Joseph says above, I'm not sure that will work. Any more detailed thoughts you have here would be greatly appreciated. Thanks.
PATCH 2/2 ARM: tegra: moving the clock gating procedure to tegra_cpu_kill
Quoted 30 lines · Stephen Warren
Quoted 11 lines · Joseph Lo
About the kexec issue, it's complicate. Your solution should be the right solution. And we should implement the hardware shutdown code in
cpu_dienotcpu_kill. I suspect this issue can be reproduced on all ARM platform. Because thesmp_send_stopcan't really shutdown the secondary CPU core if just callmachine_shutdownwithoutdisable_nonboot_cpu. And inplatform_shutdownandplatform_rebootcase, the secondary CPU already been offlined and killed before this function be called. Does any other ARM SMP platform can run kexec without any issue?(I will check if we can do some HW shutdown in
tegra_cpu_kill. For ex, just clock gate the CPU withoutwait_for_reset.)Will, Joseph's thoughts above pretty much match mine re: kexec. In the kexec thread I started earlier, I talked about replacing:
void
machine_shutdown(void)
{#ifdef CONFIG_SMP smp_send_stop();#endif
}with something like:
void
machine_shutdown(void)
{
#ifdefCONFIG_HOTPLUG_CPU
disable_nonboot_cpus();
#elifdefCONFIG_SMP
Quoted 6 lines
smp_send_stop();#endif
}and you'd responded "I think you're better off using what we currently have and hanging your code off
platform_cpu_kill.".
Sorry for being vague: I just meant that the code as it stands in mainline is better than calling disable_nonboot_cpus(), because the latter seems to require all the secondary cores to hit the idle loop, notice that they are not online, and then call cpu_die. We really need this sequence to be synchronous wrt the CPU initiating the kexec.
Quoted 5 lines
What exactly did you mean by "what we currently have"; did you mean that
cpu_kill()should work withoutcpu_die()having executed on the target CPU itself first, so Tegra should simply implementcpu_kill()? As Joseph says above, I'm not sure that will work. Any more detailed thoughts you have here would be greatly appreciated. Thanks.
I guess this comes down to the different between cpu_die/cpu_kill and whether you actually need cpu_die if you're not planning on returning to the same kernel that you left. If you really need to call cpu_die, then we need to figure out a way to do that whilst ensuring that only one CPU is left standing at the time we start copying the new kernel into place.
Will
PATCH 2/2 ARM: tegra: moving the clock gating procedure to tegra_cpu_kill
Quoted 7 lines · Will Deacon
Quoted 5 lines · Stephen Warren
What exactly did you mean by "what we currently have"; did you mean that
cpu_kill()should work withoutcpu_die()having executed on the target CPU itself first, so Tegra should simply implementcpu_kill()? As Joseph says above, I'm not sure that will work. Any more detailed thoughts you have here would be greatly appreciated. Thanks.
Quoted 5 lines
I guess this comes down to the different between cpu_die/cpu_kill and whether you actually need
cpu_dieif you're not planning on returning to the same kernel that you left. If you really need to callcpu_die, then we need to figure out a way to do that whilst ensuring that only one CPU is left standing at the time we start copying the new kernel into place.
I took a closer look at the code to try and figure out a way to solve this...
First of all, disable_nonboot_cpus *does* actually wait for the secondaries to get out of the way, because it calls __cpu_die and we end up waiting on a completion. The problem is that we IPI the secondaries in smp_send_stop, which puts then into a while(1) cpu_relax(); loop with interrupts disabled, so that completion won't be able to finish (the cores won't pass through idle and call cpu_die() themselves).
This means we must call disable_nonboot_cpus before sending the IPI. My first attempt was to make a hotplug-aware smp_send_stop, using disable_nonboot_cpus and smp_kill_cpus but this is also problematic because it's called by panic, where the simpler IPI-based code is probably a lot better.
The best bet is probably to call disable_nonboot_cpus in machine shutdown, before sending the stop and subsequent kill. It's very similar to your initial suggestion, but without the #elif.
Does that work for you?
Will
PATCH 2/2 ARM: tegra: moving the clock gating procedure to tegra_cpu_kill
On 12/23/2012 05:12 AM, Will Deacon wrote:
... [kexec on ARM SMP discussion]
Quoted 5 lines
The best bet is probably to call
disable_nonboot_cpusin machine shutdown, before sending the stop and subsequent kill. It's very similar to your initial suggestion, but without the #elif.Does that work for you?
Yes, it does. I just sent a patch implementing that. Thanks.