kernel_optimize_test/include/linux/tracehook.h
Tejun Heo b23afb93d3 memcg: punt high overage reclaim to return-to-userland path
Currently, try_charge() tries to reclaim memory synchronously when the
high limit is breached; however, if the allocation doesn't have
__GFP_WAIT, synchronous reclaim is skipped.  If a process performs only
speculative allocations, it can blow way past the high limit.  This is
actually easily reproducible by simply doing "find /".  slab/slub
allocator tries speculative allocations first, so as long as there's
memory which can be consumed without blocking, it can keep allocating
memory regardless of the high limit.

This patch makes try_charge() always punt the over-high reclaim to the
return-to-userland path.  If try_charge() detects that high limit is
breached, it adds the overage to current->memcg_nr_pages_over_high and
schedules execution of mem_cgroup_handle_over_high() which performs
synchronous reclaim from the return-to-userland path.

As long as kernel doesn't have a run-away allocation spree, this should
provide enough protection while making kmemcg behave more consistently.
It also has the following benefits.

- All over-high reclaims can use GFP_KERNEL regardless of the specific
  gfp mask in use, e.g. GFP_NOFS, when the limit was breached.

- It copes with prio inversion.  Previously, a low-prio task with
  small memory.high might perform over-high reclaim with a bunch of
  locks held.  If a higher prio task needed any of these locks, it
  would have to wait until the low prio task finished reclaim and
  released the locks.  By handing over-high reclaim to the task exit
  path this issue can be avoided.

Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Michal Hocko <mhocko@kernel.org>
Reviewed-by: Vladimir Davydov <vdavydov@parallels.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-11-05 19:34:48 -08:00

197 lines
7.1 KiB
C

/*
* Tracing hooks
*
* Copyright (C) 2008-2009 Red Hat, Inc. All rights reserved.
*
* This copyrighted material is made available to anyone wishing to use,
* modify, copy, or redistribute it subject to the terms and conditions
* of the GNU General Public License v.2.
*
* This file defines hook entry points called by core code where
* user tracing/debugging support might need to do something. These
* entry points are called tracehook_*(). Each hook declared below
* has a detailed kerneldoc comment giving the context (locking et
* al) from which it is called, and the meaning of its return value.
*
* Each function here typically has only one call site, so it is ok
* to have some nontrivial tracehook_*() inlines. In all cases, the
* fast path when no tracing is enabled should be very short.
*
* The purpose of this file and the tracehook_* layer is to consolidate
* the interface that the kernel core and arch code uses to enable any
* user debugging or tracing facility (such as ptrace). The interfaces
* here are carefully documented so that maintainers of core and arch
* code do not need to think about the implementation details of the
* tracing facilities. Likewise, maintainers of the tracing code do not
* need to understand all the calling core or arch code in detail, just
* documented circumstances of each call, such as locking conditions.
*
* If the calling core code changes so that locking is different, then
* it is ok to change the interface documented here. The maintainer of
* core code changing should notify the maintainers of the tracing code
* that they need to work out the change.
*
* Some tracehook_*() inlines take arguments that the current tracing
* implementations might not necessarily use. These function signatures
* are chosen to pass in all the information that is on hand in the
* caller and might conceivably be relevant to a tracer, so that the
* core code won't have to be updated when tracing adds more features.
* If a call site changes so that some of those parameters are no longer
* already on hand without extra work, then the tracehook_* interface
* can change so there is no make-work burden on the core code. The
* maintainer of core code changing should notify the maintainers of the
* tracing code that they need to work out the change.
*/
#ifndef _LINUX_TRACEHOOK_H
#define _LINUX_TRACEHOOK_H 1
#include <linux/sched.h>
#include <linux/ptrace.h>
#include <linux/security.h>
#include <linux/task_work.h>
#include <linux/memcontrol.h>
struct linux_binprm;
/*
* ptrace report for syscall entry and exit looks identical.
*/
static inline int ptrace_report_syscall(struct pt_regs *regs)
{
int ptrace = current->ptrace;
if (!(ptrace & PT_PTRACED))
return 0;
ptrace_notify(SIGTRAP | ((ptrace & PT_TRACESYSGOOD) ? 0x80 : 0));
/*
* this isn't the same as continuing with a signal, but it will do
* for normal use. strace only continues with a signal if the
* stopping signal is not SIGTRAP. -brl
*/
if (current->exit_code) {
send_sig(current->exit_code, current, 1);
current->exit_code = 0;
}
return fatal_signal_pending(current);
}
/**
* tracehook_report_syscall_entry - task is about to attempt a system call
* @regs: user register state of current task
*
* This will be called if %TIF_SYSCALL_TRACE has been set, when the
* current task has just entered the kernel for a system call.
* Full user register state is available here. Changing the values
* in @regs can affect the system call number and arguments to be tried.
* It is safe to block here, preventing the system call from beginning.
*
* Returns zero normally, or nonzero if the calling arch code should abort
* the system call. That must prevent normal entry so no system call is
* made. If @task ever returns to user mode after this, its register state
* is unspecified, but should be something harmless like an %ENOSYS error
* return. It should preserve enough information so that syscall_rollback()
* can work (see asm-generic/syscall.h).
*
* Called without locks, just after entering kernel mode.
*/
static inline __must_check int tracehook_report_syscall_entry(
struct pt_regs *regs)
{
return ptrace_report_syscall(regs);
}
/**
* tracehook_report_syscall_exit - task has just finished a system call
* @regs: user register state of current task
* @step: nonzero if simulating single-step or block-step
*
* This will be called if %TIF_SYSCALL_TRACE has been set, when the
* current task has just finished an attempted system call. Full
* user register state is available here. It is safe to block here,
* preventing signals from being processed.
*
* If @step is nonzero, this report is also in lieu of the normal
* trap that would follow the system call instruction because
* user_enable_block_step() or user_enable_single_step() was used.
* In this case, %TIF_SYSCALL_TRACE might not be set.
*
* Called without locks, just before checking for pending signals.
*/
static inline void tracehook_report_syscall_exit(struct pt_regs *regs, int step)
{
if (step) {
siginfo_t info;
user_single_step_siginfo(current, regs, &info);
force_sig_info(SIGTRAP, &info, current);
return;
}
ptrace_report_syscall(regs);
}
/**
* tracehook_signal_handler - signal handler setup is complete
* @stepping: nonzero if debugger single-step or block-step in use
*
* Called by the arch code after a signal handler has been set up.
* Register and stack state reflects the user handler about to run.
* Signal mask changes have already been made.
*
* Called without locks, shortly before returning to user mode
* (or handling more signals).
*/
static inline void tracehook_signal_handler(int stepping)
{
if (stepping)
ptrace_notify(SIGTRAP);
}
/**
* set_notify_resume - cause tracehook_notify_resume() to be called
* @task: task that will call tracehook_notify_resume()
*
* Calling this arranges that @task will call tracehook_notify_resume()
* before returning to user mode. If it's already running in user mode,
* it will enter the kernel and call tracehook_notify_resume() soon.
* If it's blocked, it will not be woken.
*/
static inline void set_notify_resume(struct task_struct *task)
{
#ifdef TIF_NOTIFY_RESUME
if (!test_and_set_tsk_thread_flag(task, TIF_NOTIFY_RESUME))
kick_process(task);
#endif
}
/**
* tracehook_notify_resume - report when about to return to user mode
* @regs: user-mode registers of @current task
*
* This is called when %TIF_NOTIFY_RESUME has been set. Now we are
* about to return to user mode, and the user state in @regs can be
* inspected or adjusted. The caller in arch code has cleared
* %TIF_NOTIFY_RESUME before the call. If the flag gets set again
* asynchronously, this will be called again before we return to
* user mode.
*
* Called without locks.
*/
static inline void tracehook_notify_resume(struct pt_regs *regs)
{
/*
* The caller just cleared TIF_NOTIFY_RESUME. This barrier
* pairs with task_work_add()->set_notify_resume() after
* hlist_add_head(task->task_works);
*/
smp_mb__after_atomic();
if (unlikely(current->task_works))
task_work_run();
mem_cgroup_handle_over_high();
}
#endif /* <linux/tracehook.h> */