From 86e13684aa77f07c77db352f437d9e53a84dde90 Mon Sep 17 00:00:00 2001 From: Thomas Renninger Date: Wed, 22 Apr 2009 13:48:30 +0200 Subject: [CPUFREQ] powernow-k8: Set transition latency to 1 if ACPI tables export 0 This doesn't fix anything, but it's expected that a transition latency of 0 could cause trouble in the future. Signed-off-by: Thomas Renninger Cc: Langsdorf, Mark Signed-off-by: Dave Jones --- arch/x86/kernel/cpu/cpufreq/powernow-k8.c | 13 +++++++++++++ 1 file changed, 13 insertions(+) (limited to 'arch') diff --git a/arch/x86/kernel/cpu/cpufreq/powernow-k8.c b/arch/x86/kernel/cpu/cpufreq/powernow-k8.c index cf52215d9eb..935989693a6 100644 --- a/arch/x86/kernel/cpu/cpufreq/powernow-k8.c +++ b/arch/x86/kernel/cpu/cpufreq/powernow-k8.c @@ -1046,6 +1046,19 @@ static int get_transition_latency(struct powernow_k8_data *data) if (cur_latency > max_latency) max_latency = cur_latency; } + if (max_latency == 0) { + /* + * Fam 11h always returns 0 as transition latency. + * This is intended and means "very fast". While cpufreq core + * and governors currently can handle that gracefully, better + * set it to 1 to avoid problems in the future. + * For all others it's a BIOS bug. + */ + if (!boot_cpu_data.x86 == 0x11) + printk(KERN_ERR FW_WARN PFX "Invalid zero transition " + "latency\n"); + max_latency = 1; + } /* value in usecs, needs to be in nanoseconds */ return 1000 * max_latency; } -- cgit v1.2.3 From 21335d021464c3ba3c20fc7207ffe2bdd2458568 Mon Sep 17 00:00:00 2001 From: Luis Henriques Date: Thu, 23 Apr 2009 19:45:04 +0100 Subject: [CPUFREQ] powernow-k8.c: mess cleanup Mess cleanup in powernow_k8_acpi_pst_values() function. Signed-off-by: Luis Henriques Signed-off-by: Dave Jones --- arch/x86/kernel/cpu/cpufreq/powernow-k8.c | 16 +++++++++------- 1 file changed, 9 insertions(+), 7 deletions(-) (limited to 'arch') diff --git a/arch/x86/kernel/cpu/cpufreq/powernow-k8.c b/arch/x86/kernel/cpu/cpufreq/powernow-k8.c index 935989693a6..a9d7ecbf64b 100644 --- a/arch/x86/kernel/cpu/cpufreq/powernow-k8.c +++ b/arch/x86/kernel/cpu/cpufreq/powernow-k8.c @@ -1,3 +1,4 @@ + /* * (c) 2003-2006 Advanced Micro Devices, Inc. * Your use of this code is subject to the terms and conditions of the @@ -823,13 +824,14 @@ static void powernow_k8_acpi_pst_values(struct powernow_k8_data *data, if (!data->acpi_data.state_count || (cpu_family == CPU_HW_PSTATE)) return; - control = data->acpi_data.states[index].control; data->irt = (control - >> IRT_SHIFT) & IRT_MASK; data->rvo = (control >> - RVO_SHIFT) & RVO_MASK; data->exttype = (control - >> EXT_TYPE_SHIFT) & EXT_TYPE_MASK; - data->plllock = (control >> PLL_L_SHIFT) & PLL_L_MASK; data->vidmvs = 1 - << ((control >> MVS_SHIFT) & MVS_MASK); data->vstable = - (control >> VST_SHIFT) & VST_MASK; } + control = data->acpi_data.states[index].control; + data->irt = (control >> IRT_SHIFT) & IRT_MASK; + data->rvo = (control >> RVO_SHIFT) & RVO_MASK; + data->exttype = (control >> EXT_TYPE_SHIFT) & EXT_TYPE_MASK; + data->plllock = (control >> PLL_L_SHIFT) & PLL_L_MASK; + data->vidmvs = 1 << ((control >> MVS_SHIFT) & MVS_MASK); + data->vstable = (control >> VST_SHIFT) & VST_MASK; +} static int powernow_k8_cpu_init_acpi(struct powernow_k8_data *data) { -- cgit v1.2.3 From 931db6a32dbfaad627e89d0524979ce9cb894691 Mon Sep 17 00:00:00 2001 From: Dave Jones Date: Mon, 1 Jun 2009 12:29:55 -0400 Subject: [CPUFREQ] Clean up convoluted code in arch/x86/kernel/tsc.c:time_cpufreq_notifier() Christoph Hellwig noticed the following potential uninitialised use: > arch/x86/kernel/tsc.c: In function 'time_cpufreq_notifier': > arch/x86/kernel/tsc.c:634: warning: 'dummy' may be used uninitialized in this function > > where we do have CONFIG_SMP set, freq->flags & CPUFREQ_CONST_LOOPS is > true and ref_freq is false. It seems plausable, though the circumstances for hitting it are really low. Nearly all SMP capable cpufreq drivers set CPUFREQ_CONST_LOOPS. powernow-k8 is really the only exception. The older CPUs were typically only ever UP. (powernow-k7 never supported SMP for eg) It's worth fixing regardless, as it cleans up the code. Fix possible uninitialized use of dummy, by just removing it, and making the setting of lpj more obvious. Signed-off-by: Dave Jones --- arch/x86/kernel/tsc.c | 8 +++----- 1 file changed, 3 insertions(+), 5 deletions(-) (limited to 'arch') diff --git a/arch/x86/kernel/tsc.c b/arch/x86/kernel/tsc.c index 3e1c057e98f..3fbd3206ecc 100644 --- a/arch/x86/kernel/tsc.c +++ b/arch/x86/kernel/tsc.c @@ -631,17 +631,15 @@ static int time_cpufreq_notifier(struct notifier_block *nb, unsigned long val, void *data) { struct cpufreq_freqs *freq = data; - unsigned long *lpj, dummy; + unsigned long *lpj; if (cpu_has(&cpu_data(freq->cpu), X86_FEATURE_CONSTANT_TSC)) return 0; - lpj = &dummy; - if (!(freq->flags & CPUFREQ_CONST_LOOPS)) + lpj = &boot_cpu_data.loops_per_jiffy; #ifdef CONFIG_SMP + if (!(freq->flags & CPUFREQ_CONST_LOOPS)) lpj = &cpu_data(freq->cpu).loops_per_jiffy; -#else - lpj = &boot_cpu_data.loops_per_jiffy; #endif if (!ref_freq) { -- cgit v1.2.3 From b394f1dfc070e6f50f5f33694000815e50ef319d Mon Sep 17 00:00:00 2001 From: Andrew Morton Date: Wed, 10 Jun 2009 12:41:37 -0700 Subject: [CPUFREQ] reduce scope of ACPI_PSS_BIOS_BUG_MSG[] This symbol doesn't need file-global scope. Cc: "Zhang, Rui" Cc: Dave Jones Cc: Ingo Molnar Cc: Langsdorf, Mark Cc: Leo Milano Cc: Thomas Renninger Signed-off-by: Andrew Morton Signed-off-by: Dave Jones --- arch/x86/kernel/cpu/cpufreq/powernow-k8.c | 7 +++---- 1 file changed, 3 insertions(+), 4 deletions(-) (limited to 'arch') diff --git a/arch/x86/kernel/cpu/cpufreq/powernow-k8.c b/arch/x86/kernel/cpu/cpufreq/powernow-k8.c index a9d7ecbf64b..2709b3c183b 100644 --- a/arch/x86/kernel/cpu/cpufreq/powernow-k8.c +++ b/arch/x86/kernel/cpu/cpufreq/powernow-k8.c @@ -1250,13 +1250,12 @@ static int powernowk8_verify(struct cpufreq_policy *pol) return cpufreq_frequency_table_verify(pol, data->powernow_table); } -static const char ACPI_PSS_BIOS_BUG_MSG[] = - KERN_ERR FW_BUG PFX "No compatible ACPI _PSS objects found.\n" - KERN_ERR FW_BUG PFX "Try again with latest BIOS.\n"; - /* per CPU init entry point to the driver */ static int __cpuinit powernowk8_cpu_init(struct cpufreq_policy *pol) { + static const char ACPI_PSS_BIOS_BUG_MSG[] = + KERN_ERR FW_BUG PFX "No compatible ACPI _PSS objects found.\n" + KERN_ERR FW_BUG PFX "Try again with latest BIOS.\n"; struct powernow_k8_data *data; cpumask_t oldmask; int rc; -- cgit v1.2.3 From 532cfee6ba0a8efcf7c3ce38b9881292d79d516e Mon Sep 17 00:00:00 2001 From: Naga Chumbalkar Date: Thu, 11 Jun 2009 15:26:48 +0000 Subject: [CPUFREQ] powernow-k8: read P-state from HW By definition, "cpuinfo_cur_freq" should report the value from HW. So, don't depend on the cached value. Instead read P-state directly from HW, while taking into account the erratum 311 workaround for Fam 11h processors. Cc: Andreas Herrmann Cc: Langsdorf, Mark Cc: Thomas Renninger Signed-off-by: Naga Chumbalkar Reviewed-by: Andreas Herrmann Tested-by: Andreas Herrmann Acked-by: Langsdorf, Mark Signed-off-by: Thomas Renninger Signed-off-by: Dave Jones --- arch/x86/kernel/cpu/cpufreq/powernow-k8.c | 25 +++++++++++-------------- 1 file changed, 11 insertions(+), 14 deletions(-) (limited to 'arch') diff --git a/arch/x86/kernel/cpu/cpufreq/powernow-k8.c b/arch/x86/kernel/cpu/cpufreq/powernow-k8.c index 2709b3c183b..331021112f2 100644 --- a/arch/x86/kernel/cpu/cpufreq/powernow-k8.c +++ b/arch/x86/kernel/cpu/cpufreq/powernow-k8.c @@ -118,20 +118,17 @@ static int query_current_values_with_pending_wait(struct powernow_k8_data *data) u32 i = 0; if (cpu_family == CPU_HW_PSTATE) { - if (data->currpstate == HW_PSTATE_INVALID) { - /* read (initial) hw pstate if not yet set */ - rdmsr(MSR_PSTATE_STATUS, lo, hi); - i = lo & HW_PSTATE_MASK; - - /* - * a workaround for family 11h erratum 311 might cause - * an "out-of-range Pstate if the core is in Pstate-0 - */ - if (i >= data->numps) - data->currpstate = HW_PSTATE_0; - else - data->currpstate = i; - } + rdmsr(MSR_PSTATE_STATUS, lo, hi); + i = lo & HW_PSTATE_MASK; + data->currpstate = i; + + /* + * a workaround for family 11h erratum 311 might cause + * an "out-of-range Pstate if the core is in Pstate-0 + */ + if ((boot_cpu_data.x86 == 0x11) && (i >= data->numps)) + data->currpstate = HW_PSTATE_0; + return 0; } do { -- cgit v1.2.3 From e15bc4559b397a611441a135b1f5992f07d0f436 Mon Sep 17 00:00:00 2001 From: Naga Chumbalkar Date: Thu, 11 Jun 2009 15:26:54 +0000 Subject: [CPUFREQ] powernow-k8: get drv data for correct CPU Make powernowk8_get() similar to powernowk8_target() and powernowk8_verify() in the way it obtains "powernow_data" for a given CPU. Cc: Andreas Herrmann Cc: Langsdorf, Mark Cc: Thomas Renninger Signed-off-by: Naga Chumbalkar Reviewed-by: Andreas Herrmann Tested-by: Andreas Herrmann Acked-by: Langsdorf, Mark Signed-off-by: Thomas Renninger Signed-off-by: Dave Jones --- arch/x86/kernel/cpu/cpufreq/powernow-k8.c | 6 +----- 1 file changed, 1 insertion(+), 5 deletions(-) (limited to 'arch') diff --git a/arch/x86/kernel/cpu/cpufreq/powernow-k8.c b/arch/x86/kernel/cpu/cpufreq/powernow-k8.c index 331021112f2..20c7b99d7ba 100644 --- a/arch/x86/kernel/cpu/cpufreq/powernow-k8.c +++ b/arch/x86/kernel/cpu/cpufreq/powernow-k8.c @@ -1385,13 +1385,9 @@ static int __devexit powernowk8_cpu_exit(struct cpufreq_policy *pol) static unsigned int powernowk8_get(unsigned int cpu) { - struct powernow_k8_data *data; + struct powernow_k8_data *data = per_cpu(powernow_data, cpu); cpumask_t oldmask = current->cpus_allowed; unsigned int khz = 0; - unsigned int first; - - first = cpumask_first(cpu_core_mask(cpu)); - data = per_cpu(powernow_data, first); if (!data) return -EINVAL; -- cgit v1.2.3 From 394122ab144dae4b276d74644a2f11c44a60ac5c Mon Sep 17 00:00:00 2001 From: Rusty Russell Date: Thu, 11 Jun 2009 22:59:58 +0930 Subject: [CPUFREQ] cpumask: avoid playing with cpus_allowed in speedstep-ich.c Impact: don't play with current's cpumask It's generally a very bad idea to mug some process's cpumask: it could legitimately and reasonably be changed by root, which could break us (if done before our code) or them (if we restore the wrong value). We use smp_call_function_single: this had the advantage of being more efficient, too. Signed-off-by: Rusty Russell To: cpufreq@vger.kernel.org Cc: Dominik Brodowski Signed-off-by: Dave Jones --- arch/x86/kernel/cpu/cpufreq/speedstep-ich.c | 93 +++++++++++++++++------------ arch/x86/kernel/cpu/cpufreq/speedstep-lib.c | 1 + 2 files changed, 55 insertions(+), 39 deletions(-) (limited to 'arch') diff --git a/arch/x86/kernel/cpu/cpufreq/speedstep-ich.c b/arch/x86/kernel/cpu/cpufreq/speedstep-ich.c index 016c1a4fa3f..6911e91fb4f 100644 --- a/arch/x86/kernel/cpu/cpufreq/speedstep-ich.c +++ b/arch/x86/kernel/cpu/cpufreq/speedstep-ich.c @@ -89,7 +89,8 @@ static int speedstep_find_register(void) * speedstep_set_state - set the SpeedStep state * @state: new processor frequency state (SPEEDSTEP_LOW or SPEEDSTEP_HIGH) * - * Tries to change the SpeedStep state. + * Tries to change the SpeedStep state. Can be called from + * smp_call_function_single. */ static void speedstep_set_state(unsigned int state) { @@ -143,6 +144,11 @@ static void speedstep_set_state(unsigned int state) return; } +/* Wrapper for smp_call_function_single. */ +static void _speedstep_set_state(void *_state) +{ + speedstep_set_state(*(unsigned int *)_state); +} /** * speedstep_activate - activate SpeedStep control in the chipset @@ -226,22 +232,28 @@ static unsigned int speedstep_detect_chipset(void) return 0; } -static unsigned int _speedstep_get(const struct cpumask *cpus) -{ +struct get_freq_data { unsigned int speed; - cpumask_t cpus_allowed; - - cpus_allowed = current->cpus_allowed; - set_cpus_allowed_ptr(current, cpus); - speed = speedstep_get_frequency(speedstep_processor); - set_cpus_allowed_ptr(current, &cpus_allowed); - dprintk("detected %u kHz as current frequency\n", speed); - return speed; + unsigned int processor; +}; + +static void get_freq_data(void *_data) +{ + struct get_freq_data *data = _data; + + data->speed = speedstep_get_frequency(data->processor); } static unsigned int speedstep_get(unsigned int cpu) { - return _speedstep_get(cpumask_of(cpu)); + struct get_freq_data data = { .processor = cpu }; + + /* You're supposed to ensure CPU is online. */ + if (smp_call_function_single(cpu, get_freq_data, &data, 1) != 0) + BUG(); + + dprintk("detected %u kHz as current frequency\n", data.speed); + return data.speed; } /** @@ -257,16 +269,16 @@ static int speedstep_target(struct cpufreq_policy *policy, unsigned int target_freq, unsigned int relation) { - unsigned int newstate = 0; + unsigned int newstate = 0, policy_cpu; struct cpufreq_freqs freqs; - cpumask_t cpus_allowed; int i; if (cpufreq_frequency_table_target(policy, &speedstep_freqs[0], target_freq, relation, &newstate)) return -EINVAL; - freqs.old = _speedstep_get(policy->cpus); + policy_cpu = cpumask_any_and(policy->cpus, cpu_online_mask); + freqs.old = speedstep_get(policy_cpu); freqs.new = speedstep_freqs[newstate].frequency; freqs.cpu = policy->cpu; @@ -276,20 +288,13 @@ static int speedstep_target(struct cpufreq_policy *policy, if (freqs.old == freqs.new) return 0; - cpus_allowed = current->cpus_allowed; - for_each_cpu(i, policy->cpus) { freqs.cpu = i; cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE); } - /* switch to physical CPU where state is to be changed */ - set_cpus_allowed_ptr(current, policy->cpus); - - speedstep_set_state(newstate); - - /* allow to be run on all CPUs */ - set_cpus_allowed_ptr(current, &cpus_allowed); + smp_call_function_single(policy_cpu, _speedstep_set_state, &newstate, + true); for_each_cpu(i, policy->cpus) { freqs.cpu = i; @@ -312,33 +317,43 @@ static int speedstep_verify(struct cpufreq_policy *policy) return cpufreq_frequency_table_verify(policy, &speedstep_freqs[0]); } +struct get_freqs { + struct cpufreq_policy *policy; + int ret; +}; + +static void get_freqs_on_cpu(void *_get_freqs) +{ + struct get_freqs *get_freqs = _get_freqs; + + get_freqs->ret = + speedstep_get_freqs(speedstep_processor, + &speedstep_freqs[SPEEDSTEP_LOW].frequency, + &speedstep_freqs[SPEEDSTEP_HIGH].frequency, + &get_freqs->policy->cpuinfo.transition_latency, + &speedstep_set_state); +} static int speedstep_cpu_init(struct cpufreq_policy *policy) { - int result = 0; - unsigned int speed; - cpumask_t cpus_allowed; + int result; + unsigned int policy_cpu, speed; + struct get_freqs gf; /* only run on CPU to be set, or on its sibling */ #ifdef CONFIG_SMP cpumask_copy(policy->cpus, cpu_sibling_mask(policy->cpu)); #endif - - cpus_allowed = current->cpus_allowed; - set_cpus_allowed_ptr(current, policy->cpus); + policy_cpu = cpumask_any_and(policy->cpus, cpu_online_mask); /* detect low and high frequency and transition latency */ - result = speedstep_get_freqs(speedstep_processor, - &speedstep_freqs[SPEEDSTEP_LOW].frequency, - &speedstep_freqs[SPEEDSTEP_HIGH].frequency, - &policy->cpuinfo.transition_latency, - &speedstep_set_state); - set_cpus_allowed_ptr(current, &cpus_allowed); - if (result) - return result; + gf.policy = policy; + smp_call_function_single(policy_cpu, get_freqs_on_cpu, &gf, 1); + if (gf.ret) + return gf.ret; /* get current speed setting */ - speed = _speedstep_get(policy->cpus); + speed = speedstep_get(policy_cpu); if (!speed) return -EIO; diff --git a/arch/x86/kernel/cpu/cpufreq/speedstep-lib.c b/arch/x86/kernel/cpu/cpufreq/speedstep-lib.c index 2e3c6862657..f4c290b8482 100644 --- a/arch/x86/kernel/cpu/cpufreq/speedstep-lib.c +++ b/arch/x86/kernel/cpu/cpufreq/speedstep-lib.c @@ -226,6 +226,7 @@ static unsigned int pentium4_get_frequency(void) } +/* Warning: may get called from smp_call_function_single. */ unsigned int speedstep_get_frequency(unsigned int processor) { switch (processor) { -- cgit v1.2.3 From e3f996c26ff6c4c084aaaa64dce6e54d31f517be Mon Sep 17 00:00:00 2001 From: Rusty Russell Date: Thu, 11 Jun 2009 22:59:58 +0930 Subject: [CPUFREQ] cpumask: avoid cpumask games in arch/x86/kernel/cpu/cpufreq/speedstep-centrino.c Impact: don't play with current's cpumask It's generally a very bad idea to mug some process's cpumask: it could legitimately and reasonably be changed by root, which could break us (if done before our code) or them (if we restore the wrong value). Use rdmsr_on_cpu and wrmsr_on_cpu instead. Signed-off-by: Rusty Russell To: cpufreq@vger.kernel.org Cc: Jeremy Fitzhardinge Signed-off-by: Dave Jones --- arch/x86/kernel/cpu/cpufreq/speedstep-centrino.c | 60 +++++++----------------- 1 file changed, 17 insertions(+), 43 deletions(-) (limited to 'arch') diff --git a/arch/x86/kernel/cpu/cpufreq/speedstep-centrino.c b/arch/x86/kernel/cpu/cpufreq/speedstep-centrino.c index 55c831ed71c..8d672ef162c 100644 --- a/arch/x86/kernel/cpu/cpufreq/speedstep-centrino.c +++ b/arch/x86/kernel/cpu/cpufreq/speedstep-centrino.c @@ -323,14 +323,8 @@ static unsigned int get_cur_freq(unsigned int cpu) { unsigned l, h; unsigned clock_freq; - cpumask_t saved_mask; - saved_mask = current->cpus_allowed; - set_cpus_allowed_ptr(current, &cpumask_of_cpu(cpu)); - if (smp_processor_id() != cpu) - return 0; - - rdmsr(MSR_IA32_PERF_STATUS, l, h); + rdmsr_on_cpu(cpu, MSR_IA32_PERF_STATUS, &l, &h); clock_freq = extract_clock(l, cpu, 0); if (unlikely(clock_freq == 0)) { @@ -340,11 +334,9 @@ static unsigned int get_cur_freq(unsigned int cpu) * P-state transition (like TM2). Get the last freq set * in PERF_CTL. */ - rdmsr(MSR_IA32_PERF_CTL, l, h); + rdmsr_on_cpu(cpu, MSR_IA32_PERF_CTL, &l, &h); clock_freq = extract_clock(l, cpu, 1); } - - set_cpus_allowed_ptr(current, &saved_mask); return clock_freq; } @@ -467,15 +459,10 @@ static int centrino_target (struct cpufreq_policy *policy, struct cpufreq_freqs freqs; int retval = 0; unsigned int j, k, first_cpu, tmp; - cpumask_var_t saved_mask, covered_cpus; + cpumask_var_t covered_cpus; - if (unlikely(!alloc_cpumask_var(&saved_mask, GFP_KERNEL))) - return -ENOMEM; - if (unlikely(!zalloc_cpumask_var(&covered_cpus, GFP_KERNEL))) { - free_cpumask_var(saved_mask); + if (unlikely(!zalloc_cpumask_var(&covered_cpus, GFP_KERNEL))) return -ENOMEM; - } - cpumask_copy(saved_mask, ¤t->cpus_allowed); if (unlikely(per_cpu(centrino_model, cpu) == NULL)) { retval = -ENODEV; @@ -493,7 +480,7 @@ static int centrino_target (struct cpufreq_policy *policy, first_cpu = 1; for_each_cpu(j, policy->cpus) { - const struct cpumask *mask; + int good_cpu; /* cpufreq holds the hotplug lock, so we are safe here */ if (!cpu_online(j)) @@ -504,32 +491,30 @@ static int centrino_target (struct cpufreq_policy *policy, * Make sure we are running on CPU that wants to change freq */ if (policy->shared_type == CPUFREQ_SHARED_TYPE_ANY) - mask = policy->cpus; + good_cpu = cpumask_any_and(policy->cpus, + cpu_online_mask); else - mask = cpumask_of(j); + good_cpu = j; - set_cpus_allowed_ptr(current, mask); - preempt_disable(); - if (unlikely(!cpu_isset(smp_processor_id(), *mask))) { + if (good_cpu >= nr_cpu_ids) { dprintk("couldn't limit to CPUs in this domain\n"); retval = -EAGAIN; if (first_cpu) { /* We haven't started the transition yet. */ - goto migrate_end; + goto out; } - preempt_enable(); break; } msr = per_cpu(centrino_model, cpu)->op_points[newstate].index; if (first_cpu) { - rdmsr(MSR_IA32_PERF_CTL, oldmsr, h); + rdmsr_on_cpu(good_cpu, MSR_IA32_PERF_CTL, &oldmsr, &h); if (msr == (oldmsr & 0xffff)) { dprintk("no change needed - msr was and needs " "to be %x\n", oldmsr); retval = 0; - goto migrate_end; + goto out; } freqs.old = extract_clock(oldmsr, cpu, 0); @@ -553,14 +538,11 @@ static int centrino_target (struct cpufreq_policy *policy, oldmsr |= msr; } - wrmsr(MSR_IA32_PERF_CTL, oldmsr, h); - if (policy->shared_type == CPUFREQ_SHARED_TYPE_ANY) { - preempt_enable(); + wrmsr_on_cpu(good_cpu, MSR_IA32_PERF_CTL, oldmsr, h); + if (policy->shared_type == CPUFREQ_SHARED_TYPE_ANY) break; - } - cpu_set(j, *covered_cpus); - preempt_enable(); + cpumask_set_cpu(j, covered_cpus); } for_each_cpu(k, policy->cpus) { @@ -578,10 +560,8 @@ static int centrino_target (struct cpufreq_policy *policy, * Best effort undo.. */ - for_each_cpu_mask_nr(j, *covered_cpus) { - set_cpus_allowed_ptr(current, &cpumask_of_cpu(j)); - wrmsr(MSR_IA32_PERF_CTL, oldmsr, h); - } + for_each_cpu(j, covered_cpus) + wrmsr_on_cpu(j, MSR_IA32_PERF_CTL, oldmsr, h); tmp = freqs.new; freqs.new = freqs.old; @@ -593,15 +573,9 @@ static int centrino_target (struct cpufreq_policy *policy, cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE); } } - set_cpus_allowed_ptr(current, saved_mask); retval = 0; - goto out; -migrate_end: - preempt_enable(); - set_cpus_allowed_ptr(current, saved_mask); out: - free_cpumask_var(saved_mask); free_cpumask_var(covered_cpus); return retval; } -- cgit v1.2.3 From 1ff6e97f1d993dff2f9b6f4a9173687370660232 Mon Sep 17 00:00:00 2001 From: Rusty Russell Date: Fri, 12 Jun 2009 20:55:37 +0930 Subject: [CPUFREQ] cpumask: avoid playing with cpus_allowed in powernow-k8.c cpumask: avoid playing with cpus_allowed in powernow-k8.c It's generally a very bad idea to mug some process's cpumask: it could legitimately and reasonably be changed by root, which could break us (if done before our code) or them (if we restore the wrong value). I did not replace powernowk8_target; it needs fixing, but it grabs a mutex (so no smp_call_function_single here) but Mark points out it can be called multiple times per second, so work_on_cpu is too heavy. Signed-off-by: Rusty Russell To: cpufreq@vger.kernel.org Acked-by: Mark Langsdorf Tested-by: Mark Langsdorf Signed-off-by: Dave Jones --- arch/x86/kernel/cpu/cpufreq/powernow-k8.c | 118 +++++++++++++++--------------- 1 file changed, 60 insertions(+), 58 deletions(-) (limited to 'arch') diff --git a/arch/x86/kernel/cpu/cpufreq/powernow-k8.c b/arch/x86/kernel/cpu/cpufreq/powernow-k8.c index 20c7b99d7ba..1f55547d5b3 100644 --- a/arch/x86/kernel/cpu/cpufreq/powernow-k8.c +++ b/arch/x86/kernel/cpu/cpufreq/powernow-k8.c @@ -508,41 +508,34 @@ static int core_voltage_post_transition(struct powernow_k8_data *data, return 0; } -static int check_supported_cpu(unsigned int cpu) +static void check_supported_cpu(void *_rc) { - cpumask_t oldmask; u32 eax, ebx, ecx, edx; - unsigned int rc = 0; - - oldmask = current->cpus_allowed; - set_cpus_allowed_ptr(current, &cpumask_of_cpu(cpu)); + int *rc = _rc; - if (smp_processor_id() != cpu) { - printk(KERN_ERR PFX "limiting to cpu %u failed\n", cpu); - goto out; - } + *rc = -ENODEV; if (current_cpu_data.x86_vendor != X86_VENDOR_AMD) - goto out; + return; eax = cpuid_eax(CPUID_PROCESSOR_SIGNATURE); if (((eax & CPUID_XFAM) != CPUID_XFAM_K8) && ((eax & CPUID_XFAM) < CPUID_XFAM_10H)) - goto out; + return; if ((eax & CPUID_XFAM) == CPUID_XFAM_K8) { if (((eax & CPUID_USE_XFAM_XMOD) != CPUID_USE_XFAM_XMOD) || ((eax & CPUID_XMOD) > CPUID_XMOD_REV_MASK)) { printk(KERN_INFO PFX "Processor cpuid %x not supported\n", eax); - goto out; + return; } eax = cpuid_eax(CPUID_GET_MAX_CAPABILITIES); if (eax < CPUID_FREQ_VOLT_CAPABILITIES) { printk(KERN_INFO PFX "No frequency change capabilities detected\n"); - goto out; + return; } cpuid(CPUID_FREQ_VOLT_CAPABILITIES, &eax, &ebx, &ecx, &edx); @@ -550,21 +543,17 @@ static int check_supported_cpu(unsigned int cpu) != P_STATE_TRANSITION_CAPABLE) { printk(KERN_INFO PFX "Power state transitions not supported\n"); - goto out; + return; } } else { /* must be a HW Pstate capable processor */ cpuid(CPUID_FREQ_VOLT_CAPABILITIES, &eax, &ebx, &ecx, &edx); if ((edx & USE_HW_PSTATE) == USE_HW_PSTATE) cpu_family = CPU_HW_PSTATE; else - goto out; + return; } - rc = 1; - -out: - set_cpus_allowed_ptr(current, &oldmask); - return rc; + *rc = 0; } static int check_pst_table(struct powernow_k8_data *data, struct pst_s *pst, @@ -1247,6 +1236,32 @@ static int powernowk8_verify(struct cpufreq_policy *pol) return cpufreq_frequency_table_verify(pol, data->powernow_table); } +struct init_on_cpu { + struct powernow_k8_data *data; + int rc; +}; + +static void __cpuinit powernowk8_cpu_init_on_cpu(void *_init_on_cpu) +{ + struct init_on_cpu *init_on_cpu = _init_on_cpu; + + if (pending_bit_stuck()) { + printk(KERN_ERR PFX "failing init, change pending bit set\n"); + init_on_cpu->rc = -ENODEV; + return; + } + + if (query_current_values_with_pending_wait(init_on_cpu->data)) { + init_on_cpu->rc = -ENODEV; + return; + } + + if (cpu_family == CPU_OPTERON) + fidvid_msr_init(); + + init_on_cpu->rc = 0; +} + /* per CPU init entry point to the driver */ static int __cpuinit powernowk8_cpu_init(struct cpufreq_policy *pol) { @@ -1254,13 +1269,14 @@ static int __cpuinit powernowk8_cpu_init(struct cpufreq_policy *pol) KERN_ERR FW_BUG PFX "No compatible ACPI _PSS objects found.\n" KERN_ERR FW_BUG PFX "Try again with latest BIOS.\n"; struct powernow_k8_data *data; - cpumask_t oldmask; + struct init_on_cpu init_on_cpu; int rc; if (!cpu_online(pol->cpu)) return -ENODEV; - if (!check_supported_cpu(pol->cpu)) + smp_call_function_single(pol->cpu, check_supported_cpu, &rc, 1); + if (rc) return -ENODEV; data = kzalloc(sizeof(struct powernow_k8_data), GFP_KERNEL); @@ -1300,27 +1316,12 @@ static int __cpuinit powernowk8_cpu_init(struct cpufreq_policy *pol) pol->cpuinfo.transition_latency = get_transition_latency(data); /* only run on specific CPU from here on */ - oldmask = current->cpus_allowed; - set_cpus_allowed_ptr(current, &cpumask_of_cpu(pol->cpu)); - - if (smp_processor_id() != pol->cpu) { - printk(KERN_ERR PFX "limiting to cpu %u failed\n", pol->cpu); - goto err_out_unmask; - } - - if (pending_bit_stuck()) { - printk(KERN_ERR PFX "failing init, change pending bit set\n"); - goto err_out_unmask; - } - - if (query_current_values_with_pending_wait(data)) - goto err_out_unmask; - - if (cpu_family == CPU_OPTERON) - fidvid_msr_init(); - - /* run on any CPU again */ - set_cpus_allowed_ptr(current, &oldmask); + init_on_cpu.data = data; + smp_call_function_single(data->cpu, powernowk8_cpu_init_on_cpu, + &init_on_cpu, 1); + rc = init_on_cpu.rc; + if (rc != 0) + goto err_out_exit_acpi; if (cpu_family == CPU_HW_PSTATE) cpumask_copy(pol->cpus, cpumask_of(pol->cpu)); @@ -1357,8 +1358,7 @@ static int __cpuinit powernowk8_cpu_init(struct cpufreq_policy *pol) return 0; -err_out_unmask: - set_cpus_allowed_ptr(current, &oldmask); +err_out_exit_acpi: powernow_k8_cpu_exit_acpi(data); err_out: @@ -1383,24 +1383,25 @@ static int __devexit powernowk8_cpu_exit(struct cpufreq_policy *pol) return 0; } +static void query_values_on_cpu(void *_err) +{ + int *err = _err; + struct powernow_k8_data *data = __get_cpu_var(powernow_data); + + *err = query_current_values_with_pending_wait(data); +} + static unsigned int powernowk8_get(unsigned int cpu) { struct powernow_k8_data *data = per_cpu(powernow_data, cpu); - cpumask_t oldmask = current->cpus_allowed; unsigned int khz = 0; + int err; if (!data) return -EINVAL; - set_cpus_allowed_ptr(current, &cpumask_of_cpu(cpu)); - if (smp_processor_id() != cpu) { - printk(KERN_ERR PFX - "limiting to CPU %d failed in powernowk8_get\n", cpu); - set_cpus_allowed_ptr(current, &oldmask); - return 0; - } - - if (query_current_values_with_pending_wait(data)) + smp_call_function_single(cpu, query_values_on_cpu, &err, true); + if (err) goto out; if (cpu_family == CPU_HW_PSTATE) @@ -1411,7 +1412,6 @@ static unsigned int powernowk8_get(unsigned int cpu) out: - set_cpus_allowed_ptr(current, &oldmask); return khz; } @@ -1437,7 +1437,9 @@ static int __cpuinit powernowk8_init(void) unsigned int i, supported_cpus = 0; for_each_online_cpu(i) { - if (check_supported_cpu(i)) + int rc; + smp_call_function_single(i, check_supported_cpu, &rc, 1); + if (rc == 0) supported_cpus++; } -- cgit v1.2.3 From 8e7c25971b1590776a90b249de3d859dd45e7414 Mon Sep 17 00:00:00 2001 From: Rusty Russell Date: Fri, 12 Jun 2009 20:58:37 +0930 Subject: [CPUFREQ] cpumask: new cpumask operators for arch/x86/kernel/cpu/cpufreq/powernow-k8.c Remove all old-style cpumask operators, and cpumask_t. Also: get rid of the unused define_siblings function. Signed-off-by: Rusty Russell Acked-by: Mark Langsdorf Tested-by: Mark Langsdorf Signed-off-by: Dave Jones --- arch/x86/kernel/cpu/cpufreq/powernow-k8.c | 8 ++++---- arch/x86/kernel/cpu/cpufreq/powernow-k8.h | 11 ----------- 2 files changed, 4 insertions(+), 15 deletions(-) (limited to 'arch') diff --git a/arch/x86/kernel/cpu/cpufreq/powernow-k8.c b/arch/x86/kernel/cpu/cpufreq/powernow-k8.c index 1f55547d5b3..81cbe64ed6b 100644 --- a/arch/x86/kernel/cpu/cpufreq/powernow-k8.c +++ b/arch/x86/kernel/cpu/cpufreq/powernow-k8.c @@ -1094,7 +1094,7 @@ static int transition_frequency_fidvid(struct powernow_k8_data *data, freqs.old = find_khz_freq_from_fid(data->currfid); freqs.new = find_khz_freq_from_fid(fid); - for_each_cpu_mask_nr(i, *(data->available_cores)) { + for_each_cpu(i, data->available_cores) { freqs.cpu = i; cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE); } @@ -1102,7 +1102,7 @@ static int transition_frequency_fidvid(struct powernow_k8_data *data, res = transition_fid_vid(data, fid, vid); freqs.new = find_khz_freq_from_fid(data->currfid); - for_each_cpu_mask_nr(i, *(data->available_cores)) { + for_each_cpu(i, data->available_cores) { freqs.cpu = i; cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE); } @@ -1127,7 +1127,7 @@ static int transition_frequency_pstate(struct powernow_k8_data *data, data->currpstate); freqs.new = find_khz_freq_from_pstate(data->powernow_table, pstate); - for_each_cpu_mask_nr(i, *(data->available_cores)) { + for_each_cpu(i, data->available_cores) { freqs.cpu = i; cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE); } @@ -1135,7 +1135,7 @@ static int transition_frequency_pstate(struct powernow_k8_data *data, res = transition_pstate(data, pstate); freqs.new = find_khz_freq_from_pstate(data->powernow_table, pstate); - for_each_cpu_mask_nr(i, *(data->available_cores)) { + for_each_cpu(i, data->available_cores) { freqs.cpu = i; cpufreq_notify_transition(&freqs, CPUFREQ_POSTCHANGE); } diff --git a/arch/x86/kernel/cpu/cpufreq/powernow-k8.h b/arch/x86/kernel/cpu/cpufreq/powernow-k8.h index 6c6698feade..c9c1190b5e1 100644 --- a/arch/x86/kernel/cpu/cpufreq/powernow-k8.h +++ b/arch/x86/kernel/cpu/cpufreq/powernow-k8.h @@ -223,14 +223,3 @@ static void powernow_k8_acpi_pst_values(struct powernow_k8_data *data, unsigned static int fill_powernow_table_pstate(struct powernow_k8_data *data, struct cpufreq_frequency_table *powernow_table); static int fill_powernow_table_fidvid(struct powernow_k8_data *data, struct cpufreq_frequency_table *powernow_table); - -#ifdef CONFIG_SMP -static inline void define_siblings(int cpu, cpumask_t cpu_sharedcore_mask[]) -{ -} -#else -static inline void define_siblings(int cpu, cpumask_t cpu_sharedcore_mask[]) -{ - cpu_set(0, cpu_sharedcore_mask[0]); -} -#endif -- cgit v1.2.3