The membarrier_state field is located within the mm_struct, which
is not guaranteed to exist when used from runqueue-lock-free iteration
on runqueues by the membarrier system call.
Copy the membarrier_state from the mm_struct into the scheduler runqueue
when the scheduler switches between mm.
When registering membarrier for mm, after setting the registration bit
in the mm membarrier state, issue a synchronize_rcu() to ensure the
scheduler observes the change. In order to take care of the case
where a runqueue keeps executing the target mm without swapping to
other mm, iterate over each runqueue and issue an IPI to copy the
membarrier_state from the mm_struct into each runqueue which have the
same mm which state has just been modified.
Move the mm membarrier_state field closer to pgd in mm_struct to use
a cache line already touched by the scheduler switch_mm.
The membarrier_execve() (now membarrier_exec_mmap) hook now needs to
clear the runqueue's membarrier state in addition to clear the mm
membarrier state, so move its implementation into the scheduler
membarrier code so it can access the runqueue structure.
Add memory barrier in membarrier_exec_mmap() prior to clearing
the membarrier state, ensuring memory accesses executed prior to exec
are not reordered with the stores clearing the membarrier state.
As suggested by Linus, move all membarrier.c RCU read-side locks outside
of the for each cpu loops.
Suggested-by: Linus Torvalds <torvalds@linux-foundation.org>
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Chris Metcalf <cmetcalf@ezchip.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Kirill Tkhai <tkhai@yandex.ru>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Paul E. McKenney <paulmck@linux.ibm.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Russell King - ARM Linux admin <linux@armlinux.org.uk>
Cc: Thomas Gleixner <tglx@linutronix.de>
Link: https://lkml.kernel.org/r/20190919173705.2181-5-mathieu.desnoyers@efficios.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
When going through execve(), zero out the NUMA fault statistics instead of
freeing them.
During execve, the task is reachable through procfs and the scheduler. A
concurrent /proc/*/sched reader can read data from a freed ->numa_faults
allocation (confirmed by KASAN) and write it back to userspace.
I believe that it would also be possible for a use-after-free read to occur
through a race between a NUMA fault and execve(): task_numa_fault() can
lead to task_numa_compare(), which invokes task_weight() on the currently
running task of a different CPU.
Another way to fix this would be to make ->numa_faults RCU-managed or add
extra locking, but it seems easier to wipe the NUMA fault statistics on
execve.
Signed-off-by: Jann Horn <jannh@google.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Petr Mladek <pmladek@suse.com>
Cc: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will@kernel.org>
Fixes: 82727018b0 ("sched/numa: Call task_numa_free() from do_execve()")
Link: https://lkml.kernel.org/r/20190716152047.14424-1-jannh@google.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Pull force_sig() argument change from Eric Biederman:
"A source of error over the years has been that force_sig has taken a
task parameter when it is only safe to use force_sig with the current
task.
The force_sig function is built for delivering synchronous signals
such as SIGSEGV where the userspace application caused a synchronous
fault (such as a page fault) and the kernel responded with a signal.
Because the name force_sig does not make this clear, and because the
force_sig takes a task parameter the function force_sig has been
abused for sending other kinds of signals over the years. Slowly those
have been fixed when the oopses have been tracked down.
This set of changes fixes the remaining abusers of force_sig and
carefully rips out the task parameter from force_sig and friends
making this kind of error almost impossible in the future"
* 'siginfo-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/user-namespace: (27 commits)
signal/x86: Move tsk inside of CONFIG_MEMORY_FAILURE in do_sigbus
signal: Remove the signal number and task parameters from force_sig_info
signal: Factor force_sig_info_to_task out of force_sig_info
signal: Generate the siginfo in force_sig
signal: Move the computation of force into send_signal and correct it.
signal: Properly set TRACE_SIGNAL_LOSE_INFO in __send_signal
signal: Remove the task parameter from force_sig_fault
signal: Use force_sig_fault_to_task for the two calls that don't deliver to current
signal: Explicitly call force_sig_fault on current
signal/unicore32: Remove tsk parameter from __do_user_fault
signal/arm: Remove tsk parameter from __do_user_fault
signal/arm: Remove tsk parameter from ptrace_break
signal/nds32: Remove tsk parameter from send_sigtrap
signal/riscv: Remove tsk parameter from do_trap
signal/sh: Remove tsk parameter from force_sig_info_fault
signal/um: Remove task parameter from send_sigtrap
signal/x86: Remove task parameter from send_sigtrap
signal: Remove task parameter from force_sig_mceerr
signal: Remove task parameter from force_sig
signal: Remove task parameter from force_sigsegv
...
The function force_sigsegv is always called on the current task
so passing in current is redundant and not passing in current
makes this fact obvious.
This also makes it clear force_sigsegv always calls force_sig
on the current task.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Add SPDX license identifiers to all files which:
- Have no license information of any form
- Have EXPORT_.*_SYMBOL_GPL inside which was used in the
initial scan/conversion to ignore the file
These files fall under the project license, GPL v2 only. The resulting SPDX
license identifier is:
GPL-2.0-only
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Pull scheduler updates from Ingo Molnar:
"The main changes in this cycle were:
- refcount conversions
- Solve the rq->leaf_cfs_rq_list can of worms for real.
- improve power-aware scheduling
- add sysctl knob for Energy Aware Scheduling
- documentation updates
- misc other changes"
* 'sched-core-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (34 commits)
kthread: Do not use TIMER_IRQSAFE
kthread: Convert worker lock to raw spinlock
sched/fair: Use non-atomic cpumask_{set,clear}_cpu()
sched/fair: Remove unused 'sd' parameter from select_idle_smt()
sched/wait: Use freezable_schedule() when possible
sched/fair: Prune, fix and simplify the nohz_balancer_kick() comment block
sched/fair: Explain LLC nohz kick condition
sched/fair: Simplify nohz_balancer_kick()
sched/topology: Fix percpu data types in struct sd_data & struct s_data
sched/fair: Simplify post_init_entity_util_avg() by calling it with a task_struct pointer argument
sched/fair: Fix O(nr_cgroups) in the load balancing path
sched/fair: Optimize update_blocked_averages()
sched/fair: Fix insertion in rq->leaf_cfs_rq_list
sched/fair: Add tmp_alone_branch assertion
sched/core: Use READ_ONCE()/WRITE_ONCE() in move_queued_task()/task_rq_lock()
sched/debug: Initialize sd_sysctl_cpus if !CONFIG_CPUMASK_OFFSTACK
sched/pelt: Skip updating util_est when utilization is higher than CPU's capacity
sched/fair: Update scale invariance of PELT
sched/fair: Move the rq_of() helper function
sched/core: Convert task_struct.stack_refcount to refcount_t
...
atomic_t variables are currently used to implement reference
counters with the following properties:
- counter is initialized to 1 using atomic_set()
- a resource is freed upon counter reaching zero
- once counter reaches zero, its further
increments aren't allowed
- counter schema uses basic atomic operations
(set, inc, inc_not_zero, dec_and_test, etc.)
Such atomic variables should be converted to a newly provided
refcount_t type and API that prevents accidental counter overflows
and underflows. This is important since overflows and underflows
can lead to use-after-free situation and be exploitable.
The variable sighand_struct.count is used as pure reference counter.
Convert it to refcount_t and fix up the operations.
** Important note for maintainers:
Some functions from refcount_t API defined in lib/refcount.c
have different memory ordering guarantees than their atomic
counterparts.
The full comparison can be seen in
https://lkml.org/lkml/2017/11/15/57 and it is hopefully soon
in state to be merged to the documentation tree.
Normally the differences should not matter since refcount_t provides
enough guarantees to satisfy the refcounting use cases, but in
some rare cases it might matter.
Please double check that you don't have some undocumented
memory guarantees for this variable usage.
For the sighand_struct.count it might make a difference
in following places:
- __cleanup_sighand: decrement in refcount_dec_and_test() only
provides RELEASE ordering and control dependency on success
vs. fully ordered atomic counterpart
Suggested-by: Kees Cook <keescook@chromium.org>
Signed-off-by: Elena Reshetova <elena.reshetova@intel.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Reviewed-by: David Windsor <dwindsor@gmail.com>
Reviewed-by: Hans Liljestrand <ishkamiel@gmail.com>
Reviewed-by: Andrea Parri <andrea.parri@amarulasolutions.com>
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: akpm@linux-foundation.org
Cc: viro@zeniv.linux.org.uk
Link: https://lkml.kernel.org/r/1547814450-18902-2-git-send-email-elena.reshetova@intel.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Pull trivial vfs updates from Al Viro:
"A few cleanups + Neil's namespace_unlock() optimization"
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/viro/vfs:
exec: make prepare_bprm_creds static
genheaders: %-<width>s had been there since v6; %-*s - since v7
VFS: use synchronize_rcu_expedited() in namespace_unlock()
iov_iter: reduce code duplication
get_arg_page() checks bprm->rlim_stack.rlim_cur and re-calculates the
"extra" size for argv/envp pointers every time, this is a bit ugly and
even not strictly correct: acct_arg_size() must not account this size.
Remove all the rlimit code in get_arg_page(). Instead, add bprm->argmin
calculated once at the start of __do_execve_file() and change
copy_strings to check bprm->p >= bprm->argmin.
The patch adds the new helper, prepare_arg_pages() which initializes
bprm->argc/envc and bprm->argmin.
[oleg@redhat.com: fix !CONFIG_MMU version of get_arg_page()]
Link: http://lkml.kernel.org/r/20181126122307.GA1660@redhat.com
[akpm@linux-foundation.org: use max_t]
Link: http://lkml.kernel.org/r/20181112160910.GA28440@redhat.com
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Kees Cook <keescook@chromium.org>
Tested-by: Guenter Roeck <linux@roeck-us.net>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Michal Hocko <mhocko@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
prepare_bprm_creds is not used outside exec.c, so there's no reason for it
to have external linkage.
Signed-off-by: Chanho Min <chanho.min@lge.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Revert commit c22397888f "exec: make de_thread() freezable" as
requested by Ingo Molnar:
"So there's a new regression in v4.20-rc4, my desktop produces this
lockdep splat:
[ 1772.588771] WARNING: pkexec/4633 still has locks held!
[ 1772.588773] 4.20.0-rc4-custom-00213-g93a49841322b #1 Not tainted
[ 1772.588775] ------------------------------------
[ 1772.588776] 1 lock held by pkexec/4633:
[ 1772.588778] #0: 00000000ed85fbf8 (&sig->cred_guard_mutex){+.+.}, at: prepare_bprm_creds+0x2a/0x70
[ 1772.588786] stack backtrace:
[ 1772.588789] CPU: 7 PID: 4633 Comm: pkexec Not tainted 4.20.0-rc4-custom-00213-g93a49841322b #1
[ 1772.588792] Call Trace:
[ 1772.588800] dump_stack+0x85/0xcb
[ 1772.588803] flush_old_exec+0x116/0x890
[ 1772.588807] ? load_elf_phdrs+0x72/0xb0
[ 1772.588809] load_elf_binary+0x291/0x1620
[ 1772.588815] ? sched_clock+0x5/0x10
[ 1772.588817] ? search_binary_handler+0x6d/0x240
[ 1772.588820] search_binary_handler+0x80/0x240
[ 1772.588823] load_script+0x201/0x220
[ 1772.588825] search_binary_handler+0x80/0x240
[ 1772.588828] __do_execve_file.isra.32+0x7d2/0xa60
[ 1772.588832] ? strncpy_from_user+0x40/0x180
[ 1772.588835] __x64_sys_execve+0x34/0x40
[ 1772.588838] do_syscall_64+0x60/0x1c0
The warning gets triggered by an ancient lockdep check in the freezer:
(gdb) list *0xffffffff812ece06
0xffffffff812ece06 is in flush_old_exec (./include/linux/freezer.h:57).
52 * DO NOT ADD ANY NEW CALLERS OF THIS FUNCTION
53 * If try_to_freeze causes a lockdep warning it means the caller may deadlock
54 */
55 static inline bool try_to_freeze_unsafe(void)
56 {
57 might_sleep();
58 if (likely(!freezing(current)))
59 return false;
60 return __refrigerator(false);
61 }
I reviewed the ->cred_guard_mutex code, and the mutex is held across all
of exec() - and we always did this.
But there's this recent -rc4 commit:
> Chanho Min (1):
> exec: make de_thread() freezable
c22397888f: exec: make de_thread() freezable
I believe this commit is bogus, you cannot call try_to_freeze() from
de_thread(), because it's holding the ->cred_guard_mutex."
Reported-by: Ingo Molnar <mingo@kernel.org>
Tested-by: Ingo Molnar <mingo@kernel.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Suspend fails due to the exec family of functions blocking the freezer.
The casue is that de_thread() sleeps in TASK_UNINTERRUPTIBLE waiting for
all sub-threads to die, and we have the deadlock if one of them is frozen.
This also can occur with the schedule() waiting for the group thread leader
to exit if it is frozen.
In our machine, it causes freeze timeout as bellows.
Freezing of tasks failed after 20.010 seconds (1 tasks refusing to freeze, wq_busy=0):
setcpushares-ls D ffffffc00008ed70 0 5817 1483 0x0040000d
Call trace:
[<ffffffc00008ed70>] __switch_to+0x88/0xa0
[<ffffffc000d1c30c>] __schedule+0x1bc/0x720
[<ffffffc000d1ca90>] schedule+0x40/0xa8
[<ffffffc0001cd784>] flush_old_exec+0xdc/0x640
[<ffffffc000220360>] load_elf_binary+0x2a8/0x1090
[<ffffffc0001ccff4>] search_binary_handler+0x9c/0x240
[<ffffffc00021c584>] load_script+0x20c/0x228
[<ffffffc0001ccff4>] search_binary_handler+0x9c/0x240
[<ffffffc0001ce8e0>] do_execveat_common.isra.14+0x4f8/0x6e8
[<ffffffc0001cedd0>] compat_SyS_execve+0x38/0x48
[<ffffffc00008de30>] el0_svc_naked+0x24/0x28
To fix this, make de_thread() freezable. It looks safe and works fine.
Suggested-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Chanho Min <chanho.min@lge.com>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Pavel Machek <pavel@ucw.cz>
Acked-by: Michal Hocko <mhocko@suse.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
On 32-bit systems, the buffer allocated by kernel_read_file() is too
small if the file size is > SIZE_MAX, due to truncation to size_t.
Fortunately, since the 'count' argument to kernel_read() is also
truncated to size_t, only the allocated space is filled; then, -EIO is
returned since 'pos != i_size' after the read loop.
But this is not obvious and seems incidental. We should be more
explicit about this case. So, fail early if i_size > SIZE_MAX.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Mimi Zohar <zohar@linux.ibm.com>
Pull core signal handling updates from Eric Biederman:
"It was observed that a periodic timer in combination with a
sufficiently expensive fork could prevent fork from every completing.
This contains the changes to remove the need for that restart.
This set of changes is split into several parts:
- The first part makes PIDTYPE_TGID a proper pid type instead
something only for very special cases. The part starts using
PIDTYPE_TGID enough so that in __send_signal where signals are
actually delivered we know if the signal is being sent to a a group
of processes or just a single process.
- With that prep work out of the way the logic in fork is modified so
that fork logically makes signals received while it is running
appear to be received after the fork completes"
* 'siginfo-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/user-namespace: (22 commits)
signal: Don't send signals to tasks that don't exist
signal: Don't restart fork when signals come in.
fork: Have new threads join on-going signal group stops
fork: Skip setting TIF_SIGPENDING in ptrace_init_task
signal: Add calculate_sigpending()
fork: Unconditionally exit if a fatal signal is pending
fork: Move and describe why the code examines PIDNS_ADDING
signal: Push pid type down into complete_signal.
signal: Push pid type down into __send_signal
signal: Push pid type down into send_signal
signal: Pass pid type into do_send_sig_info
signal: Pass pid type into send_sigio_to_task & send_sigurg_to_task
signal: Pass pid type into group_send_sig_info
signal: Pass pid and pid type into send_sigqueue
posix-timers: Noralize good_sigevent
signal: Use PIDTYPE_TGID to clearly store where file signals will be sent
pid: Implement PIDTYPE_TGID
pids: Move the pgrp and session pid pointers from task_struct to signal_struct
kvm: Don't open code task_pid in kvm_vcpu_ioctl
pids: Compute task_tgid using signal->leader_pid
...
Like vm_area_dup(), it initializes the anon_vma_chain head, and the
basic mm pointer.
The rest of the fields end up being different for different users,
although the plan is to also initialize the 'vm_ops' field to a dummy
entry.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The vm_area_struct is one of the most fundamental memory management
objects, but the management of it is entirely open-coded evertwhere,
ranging from allocation and freeing (using kmem_cache_[z]alloc and
kmem_cache_free) to initializing all the fields.
We want to unify this in order to end up having some unified
initialization of the vmas, and the first step to this is to at least
have basic allocation functions.
Right now those functions are literally just wrappers around the
kmem_cache_*() calls. This is a purely mechanical conversion:
# new vma:
kmem_cache_zalloc(vm_area_cachep, GFP_KERNEL) -> vm_area_alloc()
# copy old vma
kmem_cache_alloc(vm_area_cachep, GFP_KERNEL) -> vm_area_dup(old)
# free vma
kmem_cache_free(vm_area_cachep, vma) -> vm_area_free(vma)
to the point where the old vma passed in to the vm_area_dup() function
isn't even used yet (because I've left all the old manual initialization
alone).
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Everywhere except in the pid array we distinguish between a tasks pid and
a tasks tgid (thread group id). Even in the enumeration we want that
distinction sometimes so we have added __PIDTYPE_TGID. With leader_pid
we almost have an implementation of PIDTYPE_TGID in struct signal_struct.
Add PIDTYPE_TGID as a first class member of the pid_type enumeration and
into the pids array. Then remove the __PIDTYPE_TGID special case and the
leader_pid in signal_struct.
The net size increase is just an extra pointer added to struct pid and
an extra pair of pointers of an hlist_node added to task_struct.
The effect on code maintenance is the removal of a number of special
cases today and the potential to remove many more special cases as
PIDTYPE_TGID gets used to it's fullest. The long term potential
is allowing zombie thread group leaders to exit, which will remove
a lot more special cases in the code.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Pull restartable sequence support from Thomas Gleixner:
"The restartable sequences syscall (finally):
After a lot of back and forth discussion and massive delays caused by
the speculative distraction of maintainers, the core set of
restartable sequences has finally reached a consensus.
It comes with the basic non disputed core implementation along with
support for arm, powerpc and x86 and a full set of selftests
It was exposed to linux-next earlier this week, so it does not fully
comply with the merge window requirements, but there is really no
point to drag it out for yet another cycle"
* 'core-rseq-for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
rseq/selftests: Provide Makefile, scripts, gitignore
rseq/selftests: Provide parametrized tests
rseq/selftests: Provide basic percpu ops test
rseq/selftests: Provide basic test
rseq/selftests: Provide rseq library
selftests/lib.mk: Introduce OVERRIDE_TARGETS
powerpc: Wire up restartable sequences system call
powerpc: Add syscall detection for restartable sequences
powerpc: Add support for restartable sequences
x86: Wire up restartable sequence system call
x86: Add support for restartable sequences
arm: Wire up restartable sequences system call
arm: Add syscall detection for restartable sequences
arm: Add restartable sequences support
rseq: Introduce restartable sequences system call
uapi/headers: Provide types_32_64.h
Expose a new system call allowing each thread to register one userspace
memory area to be used as an ABI between kernel and user-space for two
purposes: user-space restartable sequences and quick access to read the
current CPU number value from user-space.
* Restartable sequences (per-cpu atomics)
Restartables sequences allow user-space to perform update operations on
per-cpu data without requiring heavy-weight atomic operations.
The restartable critical sections (percpu atomics) work has been started
by Paul Turner and Andrew Hunter. It lets the kernel handle restart of
critical sections. [1] [2] The re-implementation proposed here brings a
few simplifications to the ABI which facilitates porting to other
architectures and speeds up the user-space fast path.
Here are benchmarks of various rseq use-cases.
Test hardware:
arm32: ARMv7 Processor rev 4 (v7l) "Cubietruck", 2-core
x86-64: Intel E5-2630 v3@2.40GHz, 16-core, hyperthreading
The following benchmarks were all performed on a single thread.
* Per-CPU statistic counter increment
getcpu+atomic (ns/op) rseq (ns/op) speedup
arm32: 344.0 31.4 11.0
x86-64: 15.3 2.0 7.7
* LTTng-UST: write event 32-bit header, 32-bit payload into tracer
per-cpu buffer
getcpu+atomic (ns/op) rseq (ns/op) speedup
arm32: 2502.0 2250.0 1.1
x86-64: 117.4 98.0 1.2
* liburcu percpu: lock-unlock pair, dereference, read/compare word
getcpu+atomic (ns/op) rseq (ns/op) speedup
arm32: 751.0 128.5 5.8
x86-64: 53.4 28.6 1.9
* jemalloc memory allocator adapted to use rseq
Using rseq with per-cpu memory pools in jemalloc at Facebook (based on
rseq 2016 implementation):
The production workload response-time has 1-2% gain avg. latency, and
the P99 overall latency drops by 2-3%.
* Reading the current CPU number
Speeding up reading the current CPU number on which the caller thread is
running is done by keeping the current CPU number up do date within the
cpu_id field of the memory area registered by the thread. This is done
by making scheduler preemption set the TIF_NOTIFY_RESUME flag on the
current thread. Upon return to user-space, a notify-resume handler
updates the current CPU value within the registered user-space memory
area. User-space can then read the current CPU number directly from
memory.
Keeping the current cpu id in a memory area shared between kernel and
user-space is an improvement over current mechanisms available to read
the current CPU number, which has the following benefits over
alternative approaches:
- 35x speedup on ARM vs system call through glibc
- 20x speedup on x86 compared to calling glibc, which calls vdso
executing a "lsl" instruction,
- 14x speedup on x86 compared to inlined "lsl" instruction,
- Unlike vdso approaches, this cpu_id value can be read from an inline
assembly, which makes it a useful building block for restartable
sequences.
- The approach of reading the cpu id through memory mapping shared
between kernel and user-space is portable (e.g. ARM), which is not the
case for the lsl-based x86 vdso.
On x86, yet another possible approach would be to use the gs segment
selector to point to user-space per-cpu data. This approach performs
similarly to the cpu id cache, but it has two disadvantages: it is
not portable, and it is incompatible with existing applications already
using the gs segment selector for other purposes.
Benchmarking various approaches for reading the current CPU number:
ARMv7 Processor rev 4 (v7l)
Machine model: Cubietruck
- Baseline (empty loop): 8.4 ns
- Read CPU from rseq cpu_id: 16.7 ns
- Read CPU from rseq cpu_id (lazy register): 19.8 ns
- glibc 2.19-0ubuntu6.6 getcpu: 301.8 ns
- getcpu system call: 234.9 ns
x86-64 Intel(R) Xeon(R) CPU E5-2630 v3 @ 2.40GHz:
- Baseline (empty loop): 0.8 ns
- Read CPU from rseq cpu_id: 0.8 ns
- Read CPU from rseq cpu_id (lazy register): 0.8 ns
- Read using gs segment selector: 0.8 ns
- "lsl" inline assembly: 13.0 ns
- glibc 2.19-0ubuntu6 getcpu: 16.6 ns
- getcpu system call: 53.9 ns
- Speed (benchmark taken on v8 of patchset)
Running 10 runs of hackbench -l 100000 seems to indicate, contrary to
expectations, that enabling CONFIG_RSEQ slightly accelerates the
scheduler:
Configuration: 2 sockets * 8-core Intel(R) Xeon(R) CPU E5-2630 v3 @
2.40GHz (directly on hardware, hyperthreading disabled in BIOS, energy
saving disabled in BIOS, turboboost disabled in BIOS, cpuidle.off=1
kernel parameter), with a Linux v4.6 defconfig+localyesconfig,
restartable sequences series applied.
* CONFIG_RSEQ=n
avg.: 41.37 s
std.dev.: 0.36 s
* CONFIG_RSEQ=y
avg.: 40.46 s
std.dev.: 0.33 s
- Size
On x86-64, between CONFIG_RSEQ=n/y, the text size increase of vmlinux is
567 bytes, and the data size increase of vmlinux is 5696 bytes.
[1] https://lwn.net/Articles/650333/
[2] http://www.linuxplumbersconf.org/2013/ocw/system/presentations/1695/original/LPC%20-%20PerCpu%20Atomics.pdf
Signed-off-by: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Joel Fernandes <joelaf@google.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Dave Watson <davejwatson@fb.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: "H . Peter Anvin" <hpa@zytor.com>
Cc: Chris Lameter <cl@linux.com>
Cc: Russell King <linux@arm.linux.org.uk>
Cc: Andrew Hunter <ahh@google.com>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Cc: "Paul E . McKenney" <paulmck@linux.vnet.ibm.com>
Cc: Paul Turner <pjt@google.com>
Cc: Boqun Feng <boqun.feng@gmail.com>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Ben Maurer <bmaurer@fb.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: linux-api@vger.kernel.org
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: http://lkml.kernel.org/r/20151027235635.16059.11630.stgit@pjt-glaptop.roam.corp.google.com
Link: http://lkml.kernel.org/r/20150624222609.6116.86035.stgit@kitami.mtv.corp.google.com
Link: https://lkml.kernel.org/r/20180602124408.8430-3-mathieu.desnoyers@efficios.com
Mathieu Desnoyers
Jann Horn
Linus Torvalds
Eric W. Biederman
Thomas Gleixner
Alexey Dobriyan
Oleg Nesterov
Vineet Gupta
YueHaibing
Elena Reshetova
Davidlohr Bueso
Chanho Min
Rafael J. Wysocki
Eric Biggers
Kirill A. Shutemov