linux/kernel/context_tracking.c

177 lines
5.7 KiB
C
Raw Normal View History

/*
* Context tracking: Probe on high level context boundaries such as kernel
* and userspace. This includes syscalls and exceptions entry/exit.
*
* This is used by RCU to remove its dependency on the timer tick while a CPU
* runs in userspace.
*
* Started by Frederic Weisbecker:
*
* Copyright (C) 2012 Red Hat, Inc., Frederic Weisbecker <fweisbec@redhat.com>
*
* Many thanks to Gilad Ben-Yossef, Paul McKenney, Ingo Molnar, Andrew Morton,
* Steven Rostedt, Peter Zijlstra for suggestions and improvements.
*
*/
#include <linux/context_tracking.h>
#include <linux/rcupdate.h>
#include <linux/sched.h>
#include <linux/hardirq.h>
#include <linux/export.h>
#include <linux/kprobes.h>
#define CREATE_TRACE_POINTS
#include <trace/events/context_tracking.h>
struct static_key context_tracking_enabled = STATIC_KEY_INIT_FALSE;
EXPORT_SYMBOL_GPL(context_tracking_enabled);
DEFINE_PER_CPU(struct context_tracking, context_tracking);
EXPORT_SYMBOL_GPL(context_tracking);
void context_tracking_cpu_set(int cpu)
{
if (!per_cpu(context_tracking.active, cpu)) {
per_cpu(context_tracking.active, cpu) = true;
static_key_slow_inc(&context_tracking_enabled);
}
}
/**
* context_tracking_user_enter - Inform the context tracking that the CPU is going to
* enter userspace mode.
*
* This function must be called right before we switch from the kernel
* to userspace, when it's guaranteed the remaining kernel instructions
* to execute won't use any RCU read side critical section because this
* function sets RCU in extended quiescent state.
*/
void context_tracking_user_enter(void)
{
unsigned long flags;
/*
* Repeat the user_enter() check here because some archs may be calling
* this from asm and if no CPU needs context tracking, they shouldn't
* go further. Repeat the check here until they support the inline static
* key check.
*/
if (!context_tracking_is_enabled())
return;
/*
* Some contexts may involve an exception occuring in an irq,
* leading to that nesting:
* rcu_irq_enter() rcu_user_exit() rcu_user_exit() rcu_irq_exit()
* This would mess up the dyntick_nesting count though. And rcu_irq_*()
* helpers are enough to protect RCU uses inside the exception. So
* just return immediately if we detect we are in an IRQ.
*/
if (in_interrupt())
return;
/* Kernel threads aren't supposed to go to userspace */
WARN_ON_ONCE(!current->mm);
local_irq_save(flags);
context_tracking: Fix runtime CPU off-case As long as the context tracking is enabled on any CPU, even a single one, all other CPUs need to keep track of their user <-> kernel boundaries cross as well. This is because a task can sleep while servicing an exception that happened in the kernel or in userspace. Then when the task eventually wakes up and return from the exception, the CPU needs to know if we resume in userspace or in the kernel. exception_exit() get this information from exception_enter() that saved the previous state. If the CPU where the exception happened didn't keep track of these informations, exception_exit() doesn't know which state tracking to restore on the CPU where the task got migrated and we may return to userspace with the context tracking subsystem thinking that we are in kernel mode. This can be fixed in the long term if we move our context tracking probes on very low level arch fast path user <-> kernel boundary, although even that is worrisome as an exception can still happen in the few instructions between the probe and the actual iret. Also we are not yet ready to set these probes in the fast path given the potential overhead problem it induces. So let's fix this by always enable context tracking even on CPUs that are not in the full dynticks range. OTOH we can spare the rcu_user_*() and vtime_user_*() calls there because the tick runs on these CPUs and we can handle RCU state machine and cputime accounting through it. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Borislav Petkov <bp@alien8.de> Cc: Li Zhong <zhong@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Kevin Hilman <khilman@linaro.org>
2013-07-11 23:59:33 +02:00
if ( __this_cpu_read(context_tracking.state) != IN_USER) {
if (__this_cpu_read(context_tracking.active)) {
trace_user_enter(0);
context_tracking: Fix runtime CPU off-case As long as the context tracking is enabled on any CPU, even a single one, all other CPUs need to keep track of their user <-> kernel boundaries cross as well. This is because a task can sleep while servicing an exception that happened in the kernel or in userspace. Then when the task eventually wakes up and return from the exception, the CPU needs to know if we resume in userspace or in the kernel. exception_exit() get this information from exception_enter() that saved the previous state. If the CPU where the exception happened didn't keep track of these informations, exception_exit() doesn't know which state tracking to restore on the CPU where the task got migrated and we may return to userspace with the context tracking subsystem thinking that we are in kernel mode. This can be fixed in the long term if we move our context tracking probes on very low level arch fast path user <-> kernel boundary, although even that is worrisome as an exception can still happen in the few instructions between the probe and the actual iret. Also we are not yet ready to set these probes in the fast path given the potential overhead problem it induces. So let's fix this by always enable context tracking even on CPUs that are not in the full dynticks range. OTOH we can spare the rcu_user_*() and vtime_user_*() calls there because the tick runs on these CPUs and we can handle RCU state machine and cputime accounting through it. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Borislav Petkov <bp@alien8.de> Cc: Li Zhong <zhong@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Kevin Hilman <khilman@linaro.org>
2013-07-11 23:59:33 +02:00
/*
* At this stage, only low level arch entry code remains and
* then we'll run in userspace. We can assume there won't be
* any RCU read-side critical section until the next call to
* user_exit() or rcu_irq_enter(). Let's remove RCU's dependency
* on the tick.
*/
vtime_user_enter(current);
rcu_user_enter();
}
/*
context_tracking: Fix runtime CPU off-case As long as the context tracking is enabled on any CPU, even a single one, all other CPUs need to keep track of their user <-> kernel boundaries cross as well. This is because a task can sleep while servicing an exception that happened in the kernel or in userspace. Then when the task eventually wakes up and return from the exception, the CPU needs to know if we resume in userspace or in the kernel. exception_exit() get this information from exception_enter() that saved the previous state. If the CPU where the exception happened didn't keep track of these informations, exception_exit() doesn't know which state tracking to restore on the CPU where the task got migrated and we may return to userspace with the context tracking subsystem thinking that we are in kernel mode. This can be fixed in the long term if we move our context tracking probes on very low level arch fast path user <-> kernel boundary, although even that is worrisome as an exception can still happen in the few instructions between the probe and the actual iret. Also we are not yet ready to set these probes in the fast path given the potential overhead problem it induces. So let's fix this by always enable context tracking even on CPUs that are not in the full dynticks range. OTOH we can spare the rcu_user_*() and vtime_user_*() calls there because the tick runs on these CPUs and we can handle RCU state machine and cputime accounting through it. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Borislav Petkov <bp@alien8.de> Cc: Li Zhong <zhong@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Kevin Hilman <khilman@linaro.org>
2013-07-11 23:59:33 +02:00
* Even if context tracking is disabled on this CPU, because it's outside
* the full dynticks mask for example, we still have to keep track of the
* context transitions and states to prevent inconsistency on those of
* other CPUs.
* If a task triggers an exception in userspace, sleep on the exception
* handler and then migrate to another CPU, that new CPU must know where
* the exception returns by the time we call exception_exit().
* This information can only be provided by the previous CPU when it called
* exception_enter().
* OTOH we can spare the calls to vtime and RCU when context_tracking.active
* is false because we know that CPU is not tickless.
*/
cputime: Generic on-demand virtual cputime accounting If we want to stop the tick further idle, we need to be able to account the cputime without using the tick. Virtual based cputime accounting solves that problem by hooking into kernel/user boundaries. However implementing CONFIG_VIRT_CPU_ACCOUNTING require low level hooks and involves more overhead. But we already have a generic context tracking subsystem that is required for RCU needs by archs which plan to shut down the tick outside idle. This patch implements a generic virtual based cputime accounting that relies on these generic kernel/user hooks. There are some upsides of doing this: - This requires no arch code to implement CONFIG_VIRT_CPU_ACCOUNTING if context tracking is already built (already necessary for RCU in full tickless mode). - We can rely on the generic context tracking subsystem to dynamically (de)activate the hooks, so that we can switch anytime between virtual and tick based accounting. This way we don't have the overhead of the virtual accounting when the tick is running periodically. And one downside: - There is probably more overhead than a native virtual based cputime accounting. But this relies on hooks that are already set anyway. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Ingo Molnar <mingo@kernel.org> Cc: Li Zhong <zhong@linux.vnet.ibm.com> Cc: Namhyung Kim <namhyung.kim@lge.com> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Paul Gortmaker <paul.gortmaker@windriver.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Thomas Gleixner <tglx@linutronix.de>
2012-07-25 07:56:04 +02:00
__this_cpu_write(context_tracking.state, IN_USER);
}
local_irq_restore(flags);
}
NOKPROBE_SYMBOL(context_tracking_user_enter);
/**
* context_tracking_user_exit - Inform the context tracking that the CPU is
* exiting userspace mode and entering the kernel.
*
* This function must be called after we entered the kernel from userspace
* before any use of RCU read side critical section. This potentially include
* any high level kernel code like syscalls, exceptions, signal handling, etc...
*
* This call supports re-entrancy. This way it can be called from any exception
* handler without needing to know if we came from userspace or not.
*/
void context_tracking_user_exit(void)
{
unsigned long flags;
if (!context_tracking_is_enabled())
return;
if (in_interrupt())
return;
local_irq_save(flags);
if (__this_cpu_read(context_tracking.state) == IN_USER) {
context_tracking: Fix runtime CPU off-case As long as the context tracking is enabled on any CPU, even a single one, all other CPUs need to keep track of their user <-> kernel boundaries cross as well. This is because a task can sleep while servicing an exception that happened in the kernel or in userspace. Then when the task eventually wakes up and return from the exception, the CPU needs to know if we resume in userspace or in the kernel. exception_exit() get this information from exception_enter() that saved the previous state. If the CPU where the exception happened didn't keep track of these informations, exception_exit() doesn't know which state tracking to restore on the CPU where the task got migrated and we may return to userspace with the context tracking subsystem thinking that we are in kernel mode. This can be fixed in the long term if we move our context tracking probes on very low level arch fast path user <-> kernel boundary, although even that is worrisome as an exception can still happen in the few instructions between the probe and the actual iret. Also we are not yet ready to set these probes in the fast path given the potential overhead problem it induces. So let's fix this by always enable context tracking even on CPUs that are not in the full dynticks range. OTOH we can spare the rcu_user_*() and vtime_user_*() calls there because the tick runs on these CPUs and we can handle RCU state machine and cputime accounting through it. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Borislav Petkov <bp@alien8.de> Cc: Li Zhong <zhong@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Kevin Hilman <khilman@linaro.org>
2013-07-11 23:59:33 +02:00
if (__this_cpu_read(context_tracking.active)) {
/*
* We are going to run code that may use RCU. Inform
* RCU core about that (ie: we may need the tick again).
*/
rcu_user_exit();
vtime_user_exit(current);
trace_user_exit(0);
context_tracking: Fix runtime CPU off-case As long as the context tracking is enabled on any CPU, even a single one, all other CPUs need to keep track of their user <-> kernel boundaries cross as well. This is because a task can sleep while servicing an exception that happened in the kernel or in userspace. Then when the task eventually wakes up and return from the exception, the CPU needs to know if we resume in userspace or in the kernel. exception_exit() get this information from exception_enter() that saved the previous state. If the CPU where the exception happened didn't keep track of these informations, exception_exit() doesn't know which state tracking to restore on the CPU where the task got migrated and we may return to userspace with the context tracking subsystem thinking that we are in kernel mode. This can be fixed in the long term if we move our context tracking probes on very low level arch fast path user <-> kernel boundary, although even that is worrisome as an exception can still happen in the few instructions between the probe and the actual iret. Also we are not yet ready to set these probes in the fast path given the potential overhead problem it induces. So let's fix this by always enable context tracking even on CPUs that are not in the full dynticks range. OTOH we can spare the rcu_user_*() and vtime_user_*() calls there because the tick runs on these CPUs and we can handle RCU state machine and cputime accounting through it. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Borislav Petkov <bp@alien8.de> Cc: Li Zhong <zhong@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Kevin Hilman <khilman@linaro.org>
2013-07-11 23:59:33 +02:00
}
cputime: Generic on-demand virtual cputime accounting If we want to stop the tick further idle, we need to be able to account the cputime without using the tick. Virtual based cputime accounting solves that problem by hooking into kernel/user boundaries. However implementing CONFIG_VIRT_CPU_ACCOUNTING require low level hooks and involves more overhead. But we already have a generic context tracking subsystem that is required for RCU needs by archs which plan to shut down the tick outside idle. This patch implements a generic virtual based cputime accounting that relies on these generic kernel/user hooks. There are some upsides of doing this: - This requires no arch code to implement CONFIG_VIRT_CPU_ACCOUNTING if context tracking is already built (already necessary for RCU in full tickless mode). - We can rely on the generic context tracking subsystem to dynamically (de)activate the hooks, so that we can switch anytime between virtual and tick based accounting. This way we don't have the overhead of the virtual accounting when the tick is running periodically. And one downside: - There is probably more overhead than a native virtual based cputime accounting. But this relies on hooks that are already set anyway. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Andrew Morton <akpm@linux-foundation.org> Cc: Ingo Molnar <mingo@kernel.org> Cc: Li Zhong <zhong@linux.vnet.ibm.com> Cc: Namhyung Kim <namhyung.kim@lge.com> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Paul Gortmaker <paul.gortmaker@windriver.com> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Thomas Gleixner <tglx@linutronix.de>
2012-07-25 07:56:04 +02:00
__this_cpu_write(context_tracking.state, IN_KERNEL);
}
local_irq_restore(flags);
}
NOKPROBE_SYMBOL(context_tracking_user_exit);
/**
* __context_tracking_task_switch - context switch the syscall callbacks
* @prev: the task that is being switched out
* @next: the task that is being switched in
*
* The context tracking uses the syscall slow path to implement its user-kernel
* boundaries probes on syscalls. This way it doesn't impact the syscall fast
* path on CPUs that don't do context tracking.
*
* But we need to clear the flag on the previous task because it may later
* migrate to some CPU that doesn't do the context tracking. As such the TIF
* flag may not be desired there.
*/
void __context_tracking_task_switch(struct task_struct *prev,
struct task_struct *next)
{
context_tracking: Fix runtime CPU off-case As long as the context tracking is enabled on any CPU, even a single one, all other CPUs need to keep track of their user <-> kernel boundaries cross as well. This is because a task can sleep while servicing an exception that happened in the kernel or in userspace. Then when the task eventually wakes up and return from the exception, the CPU needs to know if we resume in userspace or in the kernel. exception_exit() get this information from exception_enter() that saved the previous state. If the CPU where the exception happened didn't keep track of these informations, exception_exit() doesn't know which state tracking to restore on the CPU where the task got migrated and we may return to userspace with the context tracking subsystem thinking that we are in kernel mode. This can be fixed in the long term if we move our context tracking probes on very low level arch fast path user <-> kernel boundary, although even that is worrisome as an exception can still happen in the few instructions between the probe and the actual iret. Also we are not yet ready to set these probes in the fast path given the potential overhead problem it induces. So let's fix this by always enable context tracking even on CPUs that are not in the full dynticks range. OTOH we can spare the rcu_user_*() and vtime_user_*() calls there because the tick runs on these CPUs and we can handle RCU state machine and cputime accounting through it. Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com> Cc: Steven Rostedt <rostedt@goodmis.org> Cc: Paul E. McKenney <paulmck@linux.vnet.ibm.com> Cc: Ingo Molnar <mingo@kernel.org> Cc: Thomas Gleixner <tglx@linutronix.de> Cc: Peter Zijlstra <peterz@infradead.org> Cc: Borislav Petkov <bp@alien8.de> Cc: Li Zhong <zhong@linux.vnet.ibm.com> Cc: Mike Galbraith <efault@gmx.de> Cc: Kevin Hilman <khilman@linaro.org>
2013-07-11 23:59:33 +02:00
clear_tsk_thread_flag(prev, TIF_NOHZ);
set_tsk_thread_flag(next, TIF_NOHZ);
}
#ifdef CONFIG_CONTEXT_TRACKING_FORCE
void __init context_tracking_init(void)
{
int cpu;
for_each_possible_cpu(cpu)
context_tracking_cpu_set(cpu);
}
#endif