| Message ID | 1427365095-26396-3-git-send-email-chenhui.zhao@freescale.com (mailing list archive) |
|---|---|
| State | Changes Requested |
| Delegated to: | Scott Wood |
| Headers | show |
On Thu, Mar 26, 2015 at 06:18:14PM +0800, chenhui zhao wrote: > Implemented CPU hotplug on e500mc, e5500 and e6500, and support > multiple threads mode and 64-bits mode. > > For e6500 with two threads, if one thread is online, it can > enable/disable the other thread in the same core. If two threads of > one core are offline, the core will enter the PH20 state (a low power > state). When the core is up again, Thread0 is up first, and it will be > bound with the present booting cpu. This way, all CPUs can hotplug > separately. > > Signed-off-by: Chenhui Zhao <chenhui.zhao@freescale.com> > --- > arch/powerpc/Kconfig | 2 +- > arch/powerpc/include/asm/fsl_pm.h | 4 + > arch/powerpc/include/asm/smp.h | 2 + > arch/powerpc/kernel/head_64.S | 20 +++-- > arch/powerpc/kernel/smp.c | 5 ++ > arch/powerpc/platforms/85xx/smp.c | 182 +++++++++++++++++++++++++++++--------- > arch/powerpc/sysdev/fsl_rcpm.c | 56 ++++++++++++ > 7 files changed, 220 insertions(+), 51 deletions(-) Please factor out changes to generic code (including but not limited to cur_boot_cpu and PIR handling) into separate patches with clear explanations. > diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig > index 22b0940..9846c83 100644 > --- a/arch/powerpc/Kconfig > +++ b/arch/powerpc/Kconfig > @@ -380,7 +380,7 @@ config SWIOTLB > config HOTPLUG_CPU > bool "Support for enabling/disabling CPUs" > depends on SMP && (PPC_PSERIES || \ > - PPC_PMAC || PPC_POWERNV || (PPC_85xx && !PPC_E500MC)) > + PPC_PMAC || PPC_POWERNV || FSL_SOC_BOOKE) > ---help--- > Say Y here to be able to disable and re-enable individual > CPUs at runtime on SMP machines. > diff --git a/arch/powerpc/include/asm/fsl_pm.h b/arch/powerpc/include/asm/fsl_pm.h > index bbe6089..579f495 100644 > --- a/arch/powerpc/include/asm/fsl_pm.h > +++ b/arch/powerpc/include/asm/fsl_pm.h > @@ -34,6 +34,10 @@ struct fsl_pm_ops { > void (*cpu_enter_state)(int cpu, int state); > /* exit the CPU from the specified state */ > void (*cpu_exit_state)(int cpu, int state); > + /* cpu up */ > + void (*cpu_up)(int cpu); Again, this sort of comment is useless. Tell us what "cpu up" *does*, when it should be called, etc. > @@ -189,16 +193,14 @@ _GLOBAL(fsl_secondary_thread_init) > isync > > /* > - * Fix PIR to match the linear numbering in the device tree. > - * > - * On e6500, the reset value of PIR uses the low three bits for > - * the thread within a core, and the upper bits for the core > - * number. There are two threads per core, so shift everything > - * but the low bit right by two bits so that the cpu numbering is > - * continuous. Why are you getting rid of this? If it's to avoid doing it twice on the same thread, in my work-in-progress kexec patches I instead check to see whether BUCSR has already been set up -- if it has, I assume we've already been here. > + * The current thread has been in 64-bit mode, > + * see the value of TMRN_IMSR. I don't see what the relevance of this comment is here. > + * compute the address of __cur_boot_cpu > */ > - mfspr r3, SPRN_PIR > - rlwimi r3, r3, 30, 2, 30 > + bl 10f > +10: mflr r22 > + addi r22,r22,(__cur_boot_cpu - 10b) > + lwz r3,0(r22) Please save non-volatile registers for things that need to stick around for a while. > mtspr SPRN_PIR, r3 If __cur_boot_cpu is meant to be the PIR of the currently booting CPU, it's a misleading. It looks like it's supposed to have something to do with the boot cpu (not "booting"). Also please don't put leading underscores on symbols just because the adjacent symbols have them. > -#ifdef CONFIG_HOTPLUG_CPU > +#ifdef CONFIG_PPC_E500MC > +static void qoriq_cpu_wait_die(void) > +{ > + unsigned int cpu = smp_processor_id(); > + > + hard_irq_disable(); > + /* mask all irqs to prevent cpu wakeup */ > + qoriq_pm_ops->irq_mask(cpu); > + idle_task_exit(); > + > + mtspr(SPRN_TCR, 0); > + mtspr(SPRN_TSR, mfspr(SPRN_TSR)); > + > + cur_cpu_spec->cpu_flush_caches(); > + > + generic_set_cpu_dead(cpu); > + smp_mb(); Comment memory barriers, as checkpatch says. > + while (1) > + ; Indent the ; > @@ -174,17 +232,29 @@ static inline u32 read_spin_table_addr_l(void *spin_table) > static void wake_hw_thread(void *info) > { > void fsl_secondary_thread_init(void); > - unsigned long imsr1, inia1; > + unsigned long imsr, inia; > int nr = *(const int *)info; > - > - imsr1 = MSR_KERNEL; > - inia1 = *(unsigned long *)fsl_secondary_thread_init; > - > - mttmr(TMRN_IMSR1, imsr1); > - mttmr(TMRN_INIA1, inia1); > - mtspr(SPRN_TENS, TEN_THREAD(1)); > + int hw_cpu = get_hard_smp_processor_id(nr); > + int thread_idx = cpu_thread_in_core(hw_cpu); > + > + __cur_boot_cpu = (u32)hw_cpu; > + imsr = MSR_KERNEL; > + inia = *(unsigned long *)fsl_secondary_thread_init; > + smp_mb(); > + if (thread_idx == 0) { > + mttmr(TMRN_IMSR0, imsr); > + mttmr(TMRN_INIA0, inia); > + } else { > + mttmr(TMRN_IMSR1, imsr); > + mttmr(TMRN_INIA1, inia); > + } > + isync(); > + mtspr(SPRN_TENS, TEN_THREAD(thread_idx)); Support for waking a secondary core should be a separate patch (I have similar code on the way for kexec). Likewise adding smp_mb()/isync() if it's really needed. In general, this patch tries to do too much at once. > smp_generic_kick_cpu(nr); > +#ifdef CONFIG_HOTPLUG_CPU > + generic_set_cpu_up(nr); > +#endif > } > #endif > > @@ -203,28 +273,46 @@ static int smp_85xx_kick_cpu(int nr) > > pr_debug("smp_85xx_kick_cpu: kick CPU #%d\n", nr); > > +#ifdef CONFIG_HOTPLUG_CPU > + sync_tb = 0; > + smp_mb(); > +#endif Timebase synchronization should also be separate. > #ifdef CONFIG_PPC64 > - /* Threads don't use the spin table */ > - if (cpu_thread_in_core(nr) != 0) { > + if (threads_per_core > 1) { > int primary = cpu_first_thread_sibling(nr); > > if (WARN_ON_ONCE(!cpu_has_feature(CPU_FTR_SMT))) > return -ENOENT; > > - if (cpu_thread_in_core(nr) != 1) { > - pr_err("%s: cpu %d: invalid hw thread %d\n", > - __func__, nr, cpu_thread_in_core(nr)); > - return -ENOENT; > + /* > + * If either one of threads in the same core is online, > + * use the online one to start the other. > + */ > + if (cpu_online(primary) || cpu_online(primary + 1)) { > + qoriq_pm_ops->cpu_up(nr); What if we don't have qoriq_pm_ops (e.g. VM guest, or some failure)? > + if (cpu_online(primary)) > + smp_call_function_single(primary, > + wake_hw_thread, &nr, 1); > + else > + smp_call_function_single(primary + 1, > + wake_hw_thread, &nr, 1); > + return 0; > } > - > - if (!cpu_online(primary)) { > - pr_err("%s: cpu %d: primary %d not online\n", > - __func__, nr, primary); > - return -ENOENT; > + /* > + * If both threads are offline, reset core to start. > + * When core is up, Thread 0 always gets up first, > + * so bind the current logical cpu with Thread 0. > + */ What if the core is not in a PM state that requires a reset? Where does this reset occur? > + if (hw_cpu != cpu_first_thread_sibling(hw_cpu)) { > + int hw_cpu1, hw_cpu2; > + > + hw_cpu1 = get_hard_smp_processor_id(primary); > + hw_cpu2 = get_hard_smp_processor_id(primary + 1); > + set_hard_smp_processor_id(primary, hw_cpu2); > + set_hard_smp_processor_id(primary + 1, hw_cpu1); > + /* get new physical cpu id */ > + hw_cpu = get_hard_smp_processor_id(nr); Why are you swapping the hard smp ids? > } > - > - smp_call_function_single(primary, wake_hw_thread, &nr, 0); > - return 0; > } > #endif > > @@ -252,11 +340,7 @@ static int smp_85xx_kick_cpu(int nr) > spin_table = phys_to_virt(*cpu_rel_addr); > > local_irq_save(flags); > -#ifdef CONFIG_PPC32 > #ifdef CONFIG_HOTPLUG_CPU > - /* Corresponding to generic_set_cpu_dead() */ > - generic_set_cpu_up(nr); > - Why did you move this? > if (system_state == SYSTEM_RUNNING) { > /* > * To keep it compatible with old boot program which uses > @@ -269,11 +353,16 @@ static int smp_85xx_kick_cpu(int nr) > out_be32(&spin_table->addr_l, 0); > flush_spin_table(spin_table); > > +#ifdef CONFIG_PPC_E500MC > + qoriq_pm_ops->cpu_up(nr); > +#endif Again, you've killed a VM guest kernel (this time, even if the guest doesn't see SMT). > @@ -489,13 +586,16 @@ void __init mpc85xx_smp_init(void) > __func__); > return; > } > - smp_85xx_ops.give_timebase = mpc85xx_give_timebase; > - smp_85xx_ops.take_timebase = mpc85xx_take_timebase; > -#ifdef CONFIG_HOTPLUG_CPU > - ppc_md.cpu_die = smp_85xx_mach_cpu_die; > -#endif You're moving this from a place that only runs when guts is found... > } > > + smp_85xx_ops.cpu_die = generic_cpu_die; > + ppc_md.cpu_die = smp_85xx_mach_cpu_die; > +#endif > + smp_85xx_ops.give_timebase = mpc85xx_give_timebase; > + smp_85xx_ops.take_timebase = mpc85xx_take_timebase; > + smp_85xx_ops.cpu_disable = generic_cpu_disable; > +#endif /* CONFIG_HOTPLUG_CPU */ ...to a place that runs unconditionally. Again, you're breaking VM guests. -Scott
On Thu, 2015-04-02 at 06:16 -0500, Zhao Chenhui-B35336 wrote: > > ________________________________________ > From: Wood Scott-B07421 > Sent: Tuesday, March 31, 2015 10:07 > To: Zhao Chenhui-B35336 > Cc: linuxppc-dev@lists.ozlabs.org; devicetree@vger.kernel.org; linux-kernel@vger.kernel.org; Jin Zhengxiong-R64188 > Subject: Re: [3/4] powerpc: support CPU hotplug for e500mc, e5500 and e6500 > > On Thu, Mar 26, 2015 at 06:18:14PM +0800, chenhui zhao wrote: > > @@ -189,16 +193,14 @@ _GLOBAL(fsl_secondary_thread_init) > > isync > > > > /* > > - * Fix PIR to match the linear numbering in the device tree. > > - * > > - * On e6500, the reset value of PIR uses the low three bits for > > - * the thread within a core, and the upper bits for the core > > - * number. There are two threads per core, so shift everything > > - * but the low bit right by two bits so that the cpu numbering is > > - * continuous. > > Why are you getting rid of this? If it's to avoid doing it twice on the > same thread, in my work-in-progress kexec patches I instead check to see > whether BUCSR has already been set up -- if it has, I assume we've > already been here. > > [chenhui] I didn't delete the branch prediction related code. I didn't say you did. I'm saying that you can check whether BUCSR has been set up, to determine whether PIR has already been adjusted, if your concern is avoiding running this twice on a thread between core resets. If that's not your concern, then please explain. > > + /* > > + * If both threads are offline, reset core to start. > > + * When core is up, Thread 0 always gets up first, > > + * so bind the current logical cpu with Thread 0. > > + */ > > What if the core is not in a PM state that requires a reset? > Where does this reset occur? > > [chenhui] Reset occurs in the function mpic_reset_core(). > > > + if (hw_cpu != cpu_first_thread_sibling(hw_cpu)) { > > + int hw_cpu1, hw_cpu2; > > + > > + hw_cpu1 = get_hard_smp_processor_id(primary); > > + hw_cpu2 = get_hard_smp_processor_id(primary + 1); > > + set_hard_smp_processor_id(primary, hw_cpu2); > > + set_hard_smp_processor_id(primary + 1, hw_cpu1); > > + /* get new physical cpu id */ > > + hw_cpu = get_hard_smp_processor_id(nr); > > Why are you swapping the hard smp ids? > > [chenhui] For example, Core1 has two threads, Thread0 and Thread1. In normal boot, Thread0 is CPU2, and Thread1 is CPU3. > But, if CPU2 and CPU3 are all off, user wants CPU3 up first. we need to call Thread0 as CPU3 and Thead1 as CPU2, considering > the limitation, after core is reset, only Thread0 is up, then Thread0 kicks up Thread1. There's no need for this. I have booting from a thread1, and having it kick its thread0, working locally without messing with the hwid/cpu mapping. > > @@ -252,11 +340,7 @@ static int smp_85xx_kick_cpu(int nr) > > spin_table = phys_to_virt(*cpu_rel_addr); > > > > local_irq_save(flags); > > -#ifdef CONFIG_PPC32 > > #ifdef CONFIG_HOTPLUG_CPU > > - /* Corresponding to generic_set_cpu_dead() */ > > - generic_set_cpu_up(nr); > > - > > Why did you move this? > > [chenhui] It would be better to set this after CPU is really up. Please make it a separate patch with an explanation. -Scott
diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig index 22b0940..9846c83 100644 --- a/arch/powerpc/Kconfig +++ b/arch/powerpc/Kconfig @@ -380,7 +380,7 @@ config SWIOTLB config HOTPLUG_CPU bool "Support for enabling/disabling CPUs" depends on SMP && (PPC_PSERIES || \ - PPC_PMAC || PPC_POWERNV || (PPC_85xx && !PPC_E500MC)) + PPC_PMAC || PPC_POWERNV || FSL_SOC_BOOKE) ---help--- Say Y here to be able to disable and re-enable individual CPUs at runtime on SMP machines. diff --git a/arch/powerpc/include/asm/fsl_pm.h b/arch/powerpc/include/asm/fsl_pm.h index bbe6089..579f495 100644 --- a/arch/powerpc/include/asm/fsl_pm.h +++ b/arch/powerpc/include/asm/fsl_pm.h @@ -34,6 +34,10 @@ struct fsl_pm_ops { void (*cpu_enter_state)(int cpu, int state); /* exit the CPU from the specified state */ void (*cpu_exit_state)(int cpu, int state); + /* cpu up */ + void (*cpu_up)(int cpu); + /* cpu die */ + void (*cpu_die)(int cpu); /* place the platform in the sleep state */ int (*plat_enter_sleep)(void); /* freeze the time base */ diff --git a/arch/powerpc/include/asm/smp.h b/arch/powerpc/include/asm/smp.h index d607df5..1e500ed 100644 --- a/arch/powerpc/include/asm/smp.h +++ b/arch/powerpc/include/asm/smp.h @@ -67,6 +67,7 @@ void generic_cpu_die(unsigned int cpu); void generic_set_cpu_dead(unsigned int cpu); void generic_set_cpu_up(unsigned int cpu); int generic_check_cpu_restart(unsigned int cpu); +int generic_check_cpu_dead(unsigned int cpu); #endif #ifdef CONFIG_PPC64 @@ -198,6 +199,7 @@ extern void generic_secondary_thread_init(void); extern unsigned long __secondary_hold_spinloop; extern unsigned long __secondary_hold_acknowledge; extern char __secondary_hold; +extern unsigned int __cur_boot_cpu; extern void __early_start(void); #endif /* __ASSEMBLY__ */ diff --git a/arch/powerpc/kernel/head_64.S b/arch/powerpc/kernel/head_64.S index d48125d..ac89050 100644 --- a/arch/powerpc/kernel/head_64.S +++ b/arch/powerpc/kernel/head_64.S @@ -181,6 +181,10 @@ exception_marker: #endif #ifdef CONFIG_PPC_BOOK3E + .globl __cur_boot_cpu +__cur_boot_cpu: + .long 0x0 + .align 3 _GLOBAL(fsl_secondary_thread_init) /* Enable branch prediction */ lis r3,BUCSR_INIT@h @@ -189,16 +193,14 @@ _GLOBAL(fsl_secondary_thread_init) isync /* - * Fix PIR to match the linear numbering in the device tree. - * - * On e6500, the reset value of PIR uses the low three bits for - * the thread within a core, and the upper bits for the core - * number. There are two threads per core, so shift everything - * but the low bit right by two bits so that the cpu numbering is - * continuous. + * The current thread has been in 64-bit mode, + * see the value of TMRN_IMSR. + * compute the address of __cur_boot_cpu */ - mfspr r3, SPRN_PIR - rlwimi r3, r3, 30, 2, 30 + bl 10f +10: mflr r22 + addi r22,r22,(__cur_boot_cpu - 10b) + lwz r3,0(r22) mtspr SPRN_PIR, r3 #endif diff --git a/arch/powerpc/kernel/smp.c b/arch/powerpc/kernel/smp.c index ec9ec20..2cca27a 100644 --- a/arch/powerpc/kernel/smp.c +++ b/arch/powerpc/kernel/smp.c @@ -454,6 +454,11 @@ int generic_check_cpu_restart(unsigned int cpu) return per_cpu(cpu_state, cpu) == CPU_UP_PREPARE; } +int generic_check_cpu_dead(unsigned int cpu) +{ + return per_cpu(cpu_state, cpu) == CPU_DEAD; +} + static bool secondaries_inhibited(void) { return kvm_hv_mode_active(); diff --git a/arch/powerpc/platforms/85xx/smp.c b/arch/powerpc/platforms/85xx/smp.c index fba474f..f51441b 100644 --- a/arch/powerpc/platforms/85xx/smp.c +++ b/arch/powerpc/platforms/85xx/smp.c @@ -2,7 +2,7 @@ * Author: Andy Fleming <afleming@freescale.com> * Kumar Gala <galak@kernel.crashing.org> * - * Copyright 2006-2008, 2011-2012 Freescale Semiconductor Inc. + * Copyright 2006-2008, 2011-2012, 2015 Freescale Semiconductor Inc. * * This program is free software; you can redistribute it and/or modify it * under the terms of the GNU General Public License as published by the @@ -29,6 +29,7 @@ #include <asm/fsl_guts.h> #include <asm/code-patching.h> #include <asm/cputhreads.h> +#include <asm/fsl_pm.h> #include <sysdev/fsl_soc.h> #include <sysdev/mpic.h> @@ -43,10 +44,20 @@ struct epapr_spin_table { u32 pir; }; -static struct ccsr_guts __iomem *guts; +#ifdef CONFIG_HOTPLUG_CPU static u64 timebase; static int tb_req; static int tb_valid; +/* if it is non-zero, synchronize time base */ +static int sync_tb; + +#ifdef CONFIG_PPC_E500MC +static void mpc85xx_timebase_freeze(int freeze) +{ + qoriq_pm_ops->freeze_time_base(freeze); +} +#else +static struct ccsr_guts __iomem *guts; static void mpc85xx_timebase_freeze(int freeze) { @@ -60,11 +71,15 @@ static void mpc85xx_timebase_freeze(int freeze) in_be32(&guts->devdisr); } +#endif static void mpc85xx_give_timebase(void) { unsigned long flags; + if (!sync_tb) + return; + local_irq_save(flags); while (!tb_req) @@ -113,6 +128,9 @@ static void mpc85xx_take_timebase(void) { unsigned long flags; + if (!sync_tb) + return; + local_irq_save(flags); tb_req = 1; @@ -126,7 +144,46 @@ static void mpc85xx_take_timebase(void) local_irq_restore(flags); } -#ifdef CONFIG_HOTPLUG_CPU +#ifdef CONFIG_PPC_E500MC +static void qoriq_cpu_wait_die(void) +{ + unsigned int cpu = smp_processor_id(); + + hard_irq_disable(); + /* mask all irqs to prevent cpu wakeup */ + qoriq_pm_ops->irq_mask(cpu); + idle_task_exit(); + + mtspr(SPRN_TCR, 0); + mtspr(SPRN_TSR, mfspr(SPRN_TSR)); + + cur_cpu_spec->cpu_flush_caches(); + + generic_set_cpu_dead(cpu); + smp_mb(); + while (1) + ; +} + +static void qoriq_real_cpu_die(unsigned int cpu) +{ + int i; + + for (i = 0; i < 100; i++) { + smp_rmb(); + if (generic_check_cpu_dead(cpu)) { + qoriq_pm_ops->cpu_die(cpu); +#ifdef CONFIG_PPC64 + paca[cpu].cpu_start = 0; +#endif + return; + } + msleep(10); + } + pr_err("%s: CPU%d didn't die...\n", __func__, cpu); +} + +#else static void smp_85xx_mach_cpu_die(void) { unsigned int cpu = smp_processor_id(); @@ -156,6 +213,7 @@ static void smp_85xx_mach_cpu_die(void) ; } #endif +#endif /* CONFIG_HOTPLUG_CPU */ static inline void flush_spin_table(void *spin_table) { @@ -174,17 +232,29 @@ static inline u32 read_spin_table_addr_l(void *spin_table) static void wake_hw_thread(void *info) { void fsl_secondary_thread_init(void); - unsigned long imsr1, inia1; + unsigned long imsr, inia; int nr = *(const int *)info; - - imsr1 = MSR_KERNEL; - inia1 = *(unsigned long *)fsl_secondary_thread_init; - - mttmr(TMRN_IMSR1, imsr1); - mttmr(TMRN_INIA1, inia1); - mtspr(SPRN_TENS, TEN_THREAD(1)); + int hw_cpu = get_hard_smp_processor_id(nr); + int thread_idx = cpu_thread_in_core(hw_cpu); + + __cur_boot_cpu = (u32)hw_cpu; + imsr = MSR_KERNEL; + inia = *(unsigned long *)fsl_secondary_thread_init; + smp_mb(); + if (thread_idx == 0) { + mttmr(TMRN_IMSR0, imsr); + mttmr(TMRN_INIA0, inia); + } else { + mttmr(TMRN_IMSR1, imsr); + mttmr(TMRN_INIA1, inia); + } + isync(); + mtspr(SPRN_TENS, TEN_THREAD(thread_idx)); smp_generic_kick_cpu(nr); +#ifdef CONFIG_HOTPLUG_CPU + generic_set_cpu_up(nr); +#endif } #endif @@ -203,28 +273,46 @@ static int smp_85xx_kick_cpu(int nr) pr_debug("smp_85xx_kick_cpu: kick CPU #%d\n", nr); +#ifdef CONFIG_HOTPLUG_CPU + sync_tb = 0; + smp_mb(); +#endif #ifdef CONFIG_PPC64 - /* Threads don't use the spin table */ - if (cpu_thread_in_core(nr) != 0) { + if (threads_per_core > 1) { int primary = cpu_first_thread_sibling(nr); if (WARN_ON_ONCE(!cpu_has_feature(CPU_FTR_SMT))) return -ENOENT; - if (cpu_thread_in_core(nr) != 1) { - pr_err("%s: cpu %d: invalid hw thread %d\n", - __func__, nr, cpu_thread_in_core(nr)); - return -ENOENT; + /* + * If either one of threads in the same core is online, + * use the online one to start the other. + */ + if (cpu_online(primary) || cpu_online(primary + 1)) { + qoriq_pm_ops->cpu_up(nr); + if (cpu_online(primary)) + smp_call_function_single(primary, + wake_hw_thread, &nr, 1); + else + smp_call_function_single(primary + 1, + wake_hw_thread, &nr, 1); + return 0; } - - if (!cpu_online(primary)) { - pr_err("%s: cpu %d: primary %d not online\n", - __func__, nr, primary); - return -ENOENT; + /* + * If both threads are offline, reset core to start. + * When core is up, Thread 0 always gets up first, + * so bind the current logical cpu with Thread 0. + */ + if (hw_cpu != cpu_first_thread_sibling(hw_cpu)) { + int hw_cpu1, hw_cpu2; + + hw_cpu1 = get_hard_smp_processor_id(primary); + hw_cpu2 = get_hard_smp_processor_id(primary + 1); + set_hard_smp_processor_id(primary, hw_cpu2); + set_hard_smp_processor_id(primary + 1, hw_cpu1); + /* get new physical cpu id */ + hw_cpu = get_hard_smp_processor_id(nr); } - - smp_call_function_single(primary, wake_hw_thread, &nr, 0); - return 0; } #endif @@ -252,11 +340,7 @@ static int smp_85xx_kick_cpu(int nr) spin_table = phys_to_virt(*cpu_rel_addr); local_irq_save(flags); -#ifdef CONFIG_PPC32 #ifdef CONFIG_HOTPLUG_CPU - /* Corresponding to generic_set_cpu_dead() */ - generic_set_cpu_up(nr); - if (system_state == SYSTEM_RUNNING) { /* * To keep it compatible with old boot program which uses @@ -269,11 +353,16 @@ static int smp_85xx_kick_cpu(int nr) out_be32(&spin_table->addr_l, 0); flush_spin_table(spin_table); +#ifdef CONFIG_PPC_E500MC + qoriq_pm_ops->cpu_up(nr); +#endif /* * We don't set the BPTR register here since it already points * to the boot page properly. */ mpic_reset_core(nr); + sync_tb = 1; + smp_mb(); /* * wait until core is ready... @@ -292,7 +381,12 @@ static int smp_85xx_kick_cpu(int nr) /* clear the acknowledge status */ __secondary_hold_acknowledge = -1; } + + /* Corresponding to generic_set_cpu_dead() */ + generic_set_cpu_up(nr); #endif + +#ifdef CONFIG_PPC32 flush_spin_table(spin_table); out_be32(&spin_table->pir, hw_cpu); out_be32(&spin_table->addr_l, __pa(__early_start)); @@ -304,9 +398,7 @@ static int smp_85xx_kick_cpu(int nr) pr_err("%s: timeout waiting for core %d to ack\n", __func__, hw_cpu); ret = -ENOENT; - goto out; } -out: #else smp_generic_kick_cpu(nr); @@ -317,6 +409,9 @@ out: flush_spin_table(spin_table); #endif +#ifdef CONFIG_HOTPLUG_CPU +out: +#endif local_irq_restore(flags); if (ioremappable) @@ -328,10 +423,6 @@ out: struct smp_ops_t smp_85xx_ops = { .kick_cpu = smp_85xx_kick_cpu, .cpu_bootable = smp_generic_cpu_bootable, -#ifdef CONFIG_HOTPLUG_CPU - .cpu_disable = generic_cpu_disable, - .cpu_die = generic_cpu_die, -#endif #ifdef CONFIG_KEXEC .give_timebase = smp_generic_give_timebase, .take_timebase = smp_generic_take_timebase, @@ -447,6 +538,7 @@ static void smp_85xx_setup_cpu(int cpu_nr) smp_85xx_basic_setup(cpu_nr); } +#ifdef CONFIG_HOTPLUG_CPU static const struct of_device_id mpc85xx_smp_guts_ids[] = { { .compatible = "fsl,mpc8572-guts", }, { .compatible = "fsl,p1020-guts", }, @@ -456,12 +548,12 @@ static const struct of_device_id mpc85xx_smp_guts_ids[] = { { .compatible = "fsl,p2020-guts", }, {}, }; +#endif void __init mpc85xx_smp_init(void) { struct device_node *np; - np = of_find_node_by_type(NULL, "open-pic"); if (np) { smp_85xx_ops.probe = smp_mpic_probe; @@ -480,6 +572,11 @@ void __init mpc85xx_smp_init(void) smp_85xx_ops.probe = NULL; } +#ifdef CONFIG_HOTPLUG_CPU +#ifdef CONFIG_PPC_E500MC + smp_85xx_ops.cpu_die = qoriq_real_cpu_die; + ppc_md.cpu_die = qoriq_cpu_wait_die; +#else np = of_find_matching_node(NULL, mpc85xx_smp_guts_ids); if (np) { guts = of_iomap(np, 0); @@ -489,13 +586,16 @@ void __init mpc85xx_smp_init(void) __func__); return; } - smp_85xx_ops.give_timebase = mpc85xx_give_timebase; - smp_85xx_ops.take_timebase = mpc85xx_take_timebase; -#ifdef CONFIG_HOTPLUG_CPU - ppc_md.cpu_die = smp_85xx_mach_cpu_die; -#endif } + smp_85xx_ops.cpu_die = generic_cpu_die; + ppc_md.cpu_die = smp_85xx_mach_cpu_die; +#endif + smp_85xx_ops.give_timebase = mpc85xx_give_timebase; + smp_85xx_ops.take_timebase = mpc85xx_take_timebase; + smp_85xx_ops.cpu_disable = generic_cpu_disable; +#endif /* CONFIG_HOTPLUG_CPU */ + smp_ops = &smp_85xx_ops; #ifdef CONFIG_KEXEC diff --git a/arch/powerpc/sysdev/fsl_rcpm.c b/arch/powerpc/sysdev/fsl_rcpm.c index e30f1bc..a507fd0 100644 --- a/arch/powerpc/sysdev/fsl_rcpm.c +++ b/arch/powerpc/sysdev/fsl_rcpm.c @@ -131,6 +131,46 @@ static void rcpm_v2_cpu_enter_state(int cpu, int state) } } +static void rcpm_v1_cpu_die(int cpu) +{ + rcpm_v1_cpu_enter_state(cpu, E500_PM_PH15); +} + +static void qoriq_disable_thread(void *info) +{ + int hw_cpu = get_hard_smp_processor_id(*(const int *)info); + int thread = cpu_thread_in_core(hw_cpu); + + mtspr(SPRN_TENC, TEN_THREAD(thread)); +} + +static void rcpm_v2_cpu_die(int cpu) +{ + int primary; + + if (threads_per_core == 1) { + rcpm_v2_cpu_enter_state(cpu, E500_PM_PH20); + return; + } + + primary = cpu_first_thread_sibling(cpu); + if (cpu_is_offline(primary) && cpu_is_offline(primary + 1)) { + /* when two threads are all offline, put core in PH20 */ + rcpm_v2_cpu_enter_state(cpu, E500_PM_PH20); + } else { + /* + * When one thread is offline, disable the thread + * by running qoriq_disable_thread() on the other thread. + */ + if (cpu_online(primary)) + smp_call_function_single(primary, + qoriq_disable_thread, &cpu, 1); + else + smp_call_function_single(primary + 1, + qoriq_disable_thread, &cpu, 1); + } +} + static void rcpm_v1_cpu_exit_state(int cpu, int state) { int hw_cpu = get_hard_smp_processor_id(cpu); @@ -149,6 +189,12 @@ static void rcpm_v1_cpu_exit_state(int cpu, int state) } } +static void rcpm_v1_cpu_up(int cpu) +{ + rcpm_v1_cpu_exit_state(cpu, E500_PM_PH15); + rcpm_v1_irq_unmask(cpu); +} + static void rcpm_v2_cpu_exit_state(int cpu, int state) { int hw_cpu = get_hard_smp_processor_id(cpu); @@ -172,6 +218,12 @@ static void rcpm_v2_cpu_exit_state(int cpu, int state) } } +static void rcpm_v2_cpu_up(int cpu) +{ + rcpm_v2_cpu_exit_state(cpu, E500_PM_PH20); + rcpm_v2_irq_unmask(cpu); +} + static int rcpm_v1_plat_enter_state(int state) { u32 *pmcsr_reg = &rcpm_v1_regs->powmgtcsr; @@ -280,6 +332,8 @@ static const struct fsl_pm_ops qoriq_rcpm_v1_ops = { .irq_unmask = rcpm_v1_irq_unmask, .cpu_enter_state = rcpm_v1_cpu_enter_state, .cpu_exit_state = rcpm_v1_cpu_exit_state, + .cpu_up = rcpm_v1_cpu_up, + .cpu_die = rcpm_v1_cpu_die, .plat_enter_sleep = rcpm_v1_plat_enter_sleep, .set_ip_power = rcpm_v1_set_ip_power, .freeze_time_base = rcpm_v1_freeze_time_base, @@ -291,6 +345,8 @@ static const struct fsl_pm_ops qoriq_rcpm_v2_ops = { .irq_unmask = rcpm_v2_irq_unmask, .cpu_enter_state = rcpm_v2_cpu_enter_state, .cpu_exit_state = rcpm_v2_cpu_exit_state, + .cpu_up = rcpm_v2_cpu_up, + .cpu_die = rcpm_v2_cpu_die, .plat_enter_sleep = rcpm_v2_plat_enter_sleep, .set_ip_power = rcpm_v2_set_ip_power, .freeze_time_base = rcpm_v2_freeze_time_base,
Implemented CPU hotplug on e500mc, e5500 and e6500, and support multiple threads mode and 64-bits mode. For e6500 with two threads, if one thread is online, it can enable/disable the other thread in the same core. If two threads of one core are offline, the core will enter the PH20 state (a low power state). When the core is up again, Thread0 is up first, and it will be bound with the present booting cpu. This way, all CPUs can hotplug separately. Signed-off-by: Chenhui Zhao <chenhui.zhao@freescale.com> --- arch/powerpc/Kconfig | 2 +- arch/powerpc/include/asm/fsl_pm.h | 4 + arch/powerpc/include/asm/smp.h | 2 + arch/powerpc/kernel/head_64.S | 20 +++-- arch/powerpc/kernel/smp.c | 5 ++ arch/powerpc/platforms/85xx/smp.c | 182 +++++++++++++++++++++++++++++--------- arch/powerpc/sysdev/fsl_rcpm.c | 56 ++++++++++++ 7 files changed, 220 insertions(+), 51 deletions(-)