This is the mail archive of the
libc-alpha@sourceware.org
mailing list for the glibc project.
Re: [PATCH v3] lowlevellock comments
- From: Bernie Ogden <bernie dot ogden at linaro dot org>
- To: libc-alpha <libc-alpha at sourceware dot org>
- Cc: Torvald Riegel <triegel at redhat dot com>, "Carlos O'Donell" <carlos at redhat dot com>, Bernard Ogden <bernie dot ogden at linaro dot org>
- Date: Mon, 17 Nov 2014 13:56:26 +0000
- Subject: Re: [PATCH v3] lowlevellock comments
- Authentication-results: sourceware.org; auth=none
- References: <CALE0ps2oQnDZxfc7hPVyVAY33bzs=apEtwyUH6HjYO_oxjYRNw at mail dot gmail dot com> <1414066371-394-1-git-send-email-bernie dot ogden at linaro dot org> <CALE0ps3asLLDmyX7F9qW9R5nGCJbKeV+th+eLdqG5Tda0Ov8XQ at mail dot gmail dot com>
Ping^2
On 10 November 2014 09:25, Bernie Ogden <bernie.ogden@linaro.org> wrote:
> Ping
>
> On 23 October 2014 13:12, Bernard Ogden <bernie.ogden@linaro.org> wrote:
>> Here's a v3. All comments addressed - replies to comments I have something to
>> say about inline below.
>>
>> On 6 October 2014 13:26, Torvald Riegel <triegel@redhat.com> wrote:
>>>> To make the >1 state easier to refer to in comments I've gone for:
>>>> >1 - locked-with-waiters, there _may_ be waiters
>>>>
>>>> If I'm right then strictly >1 means locked-possibly-with-waiters, but that's
>>>> really getting too much of a mouthful (as 'potentially contended' would have
>>>> been in the earlier terminology).
>>>
>>> But "locked possibly with waiters" is the right description, so I'd
>>> rather use this.
>>
>> OK, changed throughout to 'acquired, possibly with waiters' (based on the
>> suggested text below).
>>
>>>> > Please use locked, or unlocked terms everywhere.
>>>>
>>>> I've gone for unlocked, locked and locked with waiters.
>>>
>>> My vote would be for "acquired" / "not acquired", as in "the lock has
>>> been acquired".
>>
>> Switched to acquired/not acquired, as no-one defended locked/unlocked.
>>
>>>> + thread ID while the clone is running and is reset to zero
>>>> + afterwards. */
>>>
>>> Could you add who resets it?
>>
>> Done (it's the kernel).
>>
>>>>
>>>> +/* Uses futexes to avoid unnecessary calls into the kernel. Likely paths are
>>>
>>> I'd say it uses futexes to block using the kernel. That's the purpose
>>> of the futex call, so that we don't need to just spin-wait.
>> <snip>
>>>> + the cases with no waiters and hence do not involve syscalls. Unlikely paths
>>>> + usually will involve syscalls to wait or to wake.
>>>
>>> The "likely" path is not necessarily likely, but we try to make
>>> uncontended mutexes fast, so that's why we avoid the syscalls in these
>>> cases. I would suggest:
>>>
>>> Low-level locks use a combination of atomic operations (to acquire and
>>> release lock ownership) and futex operations (to block until the state
>>> of a lock changes). A lock can be in one of three states:
>>> 0: not acquired,
>>> 1: acquired; no other threads are blocked or about to block for changes
>>> to the lock state,
>>>>1: acquired, other threads are potentially blocked or about to block
>>> for changes to the lock state.
>>>
>>> We expect that the common case is an uncontended lock, so we just need
>>> to transition the lock between states 0 and 1; releasing the lock does
>>> not need to wake any other blocked threads. If the lock is contended
>>> and a thread decides to block using a futex operation, then this thread
>>> needs to first change the state to >1; if this state is observed during
>>> lock release, the releasing thread will wake one of the potentially
>>> blocked threads.
>>
>> Used your text, tweaked slightly to make the difference between 1 and >1 even
>> more explicit. Also reworded the >1 description to a form that makes a bit
>> more sense to the pedant in me.
>>
>>>> + cond_locks are never in the locked state, they are either unlocked or
>>>> + locked with waiters.
>>>
>>> Change that to the following?:
>>>
>>> Condition variables contain an optimization for broadcasts that requeues
>>> waiting threads on a lock's futex. Therefore, there is a special
>>> variant of the locks (whose name contains "cond") that makes sure to
>>> always set the lock state to >1 and not just 1.
>>
>> And just used this as is.
>>
>>>> +/* If LOCK is 0 (unlocked), set to 1 (locked) and return 0 to indicate the
>>>> + lock was acquired. Otherwise leave the lock unchanged and return non-zero
>>>> + to indicate the lock was not acquired. */
>>>
>>> If you use acquired / not acquired instead of locked / unlocked, this
>>> gets simpler.
>>
>> In that I can be a bit less verbose about the meaning of returns? Done.
>>
>>>> + 2) The previous owner of the lock dies. FUTEX will be
>>>> + ID | FUTEX_OWNER_DIED. FUTEX_WAITERS may also be set. Return FUTEX to
>>>> + indicate that the lock is not acquired. */
>>>
>>> I'd say "this value" instead of the second occurence of "FUTEX" to
>>> highlight that we do NOT read FUTEX twice.
>>
>> Fixed - also put the FUTEX_WAITERS part in brackets and cleared up some
>> ambiguity.
>>
>>
>> 2014-10-23 Bernard Ogden <bernie.ogden@linaro.org>
>>
>> * nptl/lowlevellock.c: Add comments.
>> * nptl/lowlevelrobustlock.c: Likewise.
>> * sysdeps/nptl/lowlevellock.h: Likewise.
>>
>> ---
>>
>> diff --git a/nptl/lowlevellock.c b/nptl/lowlevellock.c
>> index e198af7..99cca8e 100644
>> --- a/nptl/lowlevellock.c
>> +++ b/nptl/lowlevellock.c
>> @@ -27,10 +27,10 @@ void
>> __lll_lock_wait_private (int *futex)
>> {
>> if (*futex == 2)
>> - lll_futex_wait (futex, 2, LLL_PRIVATE);
>> + lll_futex_wait (futex, 2, LLL_PRIVATE); /* Wait if *futex == 2. */
>>
>> while (atomic_exchange_acq (futex, 2) != 0)
>> - lll_futex_wait (futex, 2, LLL_PRIVATE);
>> + lll_futex_wait (futex, 2, LLL_PRIVATE); /* Wait if *futex == 2. */
>> }
>>
>>
>> @@ -40,10 +40,10 @@ void
>> __lll_lock_wait (int *futex, int private)
>> {
>> if (*futex == 2)
>> - lll_futex_wait (futex, 2, private);
>> + lll_futex_wait (futex, 2, private); /* Wait if *futex == 2. */
>>
>> while (atomic_exchange_acq (futex, 2) != 0)
>> - lll_futex_wait (futex, 2, private);
>> + lll_futex_wait (futex, 2, private); /* Wait if *futex == 2. */
>> }
>>
>>
>> @@ -75,7 +75,7 @@ __lll_timedlock_wait (int *futex, const struct timespec *abstime, int private)
>> if (rt.tv_sec < 0)
>> return ETIMEDOUT;
>>
>> - /* Wait. */
>> + /* If *futex == 2, wait until woken or timeout. */
>> lll_futex_timed_wait (futex, 2, &rt, private);
>> }
>>
>> @@ -83,6 +83,11 @@ __lll_timedlock_wait (int *futex, const struct timespec *abstime, int private)
>> }
>>
>>
>> +/* The kernel notifies a process which uses CLONE_CHILD_CLEARTID via futex
>> + wake-up when the clone terminates. The memory location contains the
>> + thread ID while the clone is running and is reset to zero by the kernel
>> + afterwards. The kernel up to version 3.16.3 does not use the private futex
>> + operations for futex wake-up when the clone terminates. */
>> int
>> __lll_timedwait_tid (int *tidp, const struct timespec *abstime)
>> {
>> @@ -113,8 +118,10 @@ __lll_timedwait_tid (int *tidp, const struct timespec *abstime)
>> if (rt.tv_sec < 0)
>> return ETIMEDOUT;
>>
>> - /* Wait until thread terminates. The kernel so far does not use
>> - the private futex operations for this. */
>> + /* If *tidp == tid, wait until thread terminates or the wait times out.
>> + The kernel to version 3.16.3 does not use the private futex
>> + operations for futex wake-up when the clone terminates.
>> + */
>> if (lll_futex_timed_wait (tidp, tid, &rt, LLL_SHARED) == -ETIMEDOUT)
>> return ETIMEDOUT;
>> }
>> diff --git a/nptl/lowlevelrobustlock.c b/nptl/lowlevelrobustlock.c
>> index 3525807..2f4f314 100644
>> --- a/nptl/lowlevelrobustlock.c
>> +++ b/nptl/lowlevelrobustlock.c
>> @@ -36,14 +36,17 @@ __lll_robust_lock_wait (int *futex, int private)
>>
>> do
>> {
>> + /* If the owner died, return the present value of the futex. */
>> if (__glibc_unlikely (oldval & FUTEX_OWNER_DIED))
>> return oldval;
>>
>> + /* Attempt to put lock into state 'acquired, possibly with waiters'. */
>> int newval = oldval | FUTEX_WAITERS;
>> if (oldval != newval
>> && atomic_compare_and_exchange_bool_acq (futex, newval, oldval))
>> continue;
>>
>> + /* If *futex == 2, wait until woken. */
>> lll_futex_wait (futex, newval, private);
>>
>> try:
>> @@ -100,15 +103,17 @@ __lll_robust_timedlock_wait (int *futex, const struct timespec *abstime,
>> return ETIMEDOUT;
>> #endif
>>
>> - /* Wait. */
>> + /* If the owner died, return the present value of the futex. */
>> if (__glibc_unlikely (oldval & FUTEX_OWNER_DIED))
>> return oldval;
>>
>> + /* Attempt to put lock into state 'acquired, possibly with waiters'. */
>> int newval = oldval | FUTEX_WAITERS;
>> if (oldval != newval
>> && atomic_compare_and_exchange_bool_acq (futex, newval, oldval))
>> continue;
>>
>> + /* If *futex == 2, wait until woken or timeout. */
>> #if (!defined __ASSUME_FUTEX_CLOCK_REALTIME \
>> || !defined lll_futex_timed_wait_bitset)
>> lll_futex_timed_wait (futex, newval, &rt, private);
>> diff --git a/sysdeps/nptl/lowlevellock.h b/sysdeps/nptl/lowlevellock.h
>> index 28f4ba3..8f6e270 100644
>> --- a/sysdeps/nptl/lowlevellock.h
>> +++ b/sysdeps/nptl/lowlevellock.h
>> @@ -22,9 +22,53 @@
>> #include <atomic.h>
>> #include <lowlevellock-futex.h>
>>
>> +/* Low-level locks use a combination of atomic operations (to acquire and
>> + release lock ownership) and futex operations (to block until the state
>> + of a lock changes). A lock can be in one of three states:
>> + 0: not acquired,
>> + 1: acquired with no waiters; no other threads are blocked or about to block
>> + for changes to the lock state,
>> + >1: acquired, possibly with waiters; there may be other threads blocked or
>> + about to block for changes to the lock state.
>> +
>> + We expect that the common case is an uncontended lock, so we just need
>> + to transition the lock between states 0 and 1; releasing the lock does
>> + not need to wake any other blocked threads. If the lock is contended
>> + and a thread decides to block using a futex operation, then this thread
>> + needs to first change the state to >1; if this state is observed during
>> + lock release, the releasing thread will wake one of the potentially
>> + blocked threads.
>> +
>> + Much of this code takes a 'private' parameter. This may be:
>> + LLL_PRIVATE: lock only shared within a process
>> + LLL_SHARED: lock may be shared across processes.
>> +
>> + Condition variables contain an optimization for broadcasts that requeues
>> + waiting threads on a lock's futex. Therefore, there is a special
>> + variant of the locks (whose name contains "cond") that makes sure to
>> + always set the lock state to >1 and not just 1.
>> +
>> + Robust locks set the lock to the id of the owner. This allows detection
>> + of the case where the owner exits without releasing the lock. Flags are
>> + OR'd with the owner id to record additional information about lock state.
>> + Therefore the states of robust locks are:
>> + 0: not acquired
>> + id: acquired (by user identified by id & FUTEX_TID_MASK)
>> +
>> + The following flags may be set in the robust lock value:
>> + FUTEX_WAITERS - possibly has waiters
>> + FUTEX_OWNER_DIED - owning user has exited without releasing the futex. */
>> +
>> +
>> +/* If LOCK is 0 (not acquired), set to 1 (acquired) and return 0. Otherwise
>> + leave lock unchanged and return non-zero to indicate that the lock was not
>> + acquired. */
>> #define lll_trylock(lock) \
>> atomic_compare_and_exchange_bool_acq (&(lock), 1, 0)
>>
>> +/* If LOCK is 0 (not acquired), set to 2 (acquired, possibly with waiters) and
>> + return 0. Otherwise leave lock unchanged and return non-zero to indicate
>> + that the lock was not acquired. */
>> #define lll_cond_trylock(lock) \
>> atomic_compare_and_exchange_bool_acq (&(lock), 2, 0)
>>
>> @@ -35,6 +79,13 @@ extern int __lll_robust_lock_wait (int *futex, int private) attribute_hidden;
>> /* This is an expression rather than a statement even though its value is
>> void, so that it can be used in a comma expression or as an expression
>> that's cast to void. */
>> +/* The inner conditional compiles to a call to __lll_lock_wait_private if
>> + private is known at compile time to be LLL_PRIVATE, and to a call to
>> + __lll_lock_wait otherwise. */
>> +/* If FUTEX is 0 (not acquired), set to 1 (acquired with no waiters) and
>> + return. Otherwise block until we acquire the lock, at which point FUTEX is
>> + 2 (acquired, possibly with waiters), then return. The lock is aways
>> + acquired on return. */
>> #define __lll_lock(futex, private) \
>> ((void) \
>> ({ \
>> @@ -52,6 +103,14 @@ extern int __lll_robust_lock_wait (int *futex, int private) attribute_hidden;
>> __lll_lock (&(futex), private)
>>
>>
>> +/* If FUTEX is 0 (not acquired), set to ID (acquired with no waiters) and
>> + return 0. Otherwise, block until either:
>> + 1) We acquire the lock, at which point FUTEX will be ID | FUTEX_WAITERS
>> + (acquired, possibly with waiters), then return 0.
>> + 2) The previous owner of the lock dies. FUTEX will be the value of id as
>> + set by the previous owner, with FUTEX_OWNER_DIED set (FUTEX_WAITERS may
>> + also be set). Return this value to indicate that the lock is not
>> + acquired. */
>> #define __lll_robust_lock(futex, id, private) \
>> ({ \
>> int *__futex = (futex); \
>> @@ -69,6 +128,12 @@ extern int __lll_robust_lock_wait (int *futex, int private) attribute_hidden;
>> /* This is an expression rather than a statement even though its value is
>> void, so that it can be used in a comma expression or as an expression
>> that's cast to void. */
>> +/* Unconditionally set FUTEX to 2 (acquired, possibly with waiters). condvar
>> + locks only have 'not acquired' and 'acquired, possibly with waiters' states,
>> + so there is no need to check FUTEX before setting.
>> + If FUTEX was 0 (not acquired) then return. Otherwise, block until the lock
>> + is acquired, at which point FUTEX is 2 (acquired, possibly with waiters).
>> + The lock is always acquired on return. */
>> #define __lll_cond_lock(futex, private) \
>> ((void) \
>> ({ \
>> @@ -79,6 +144,14 @@ extern int __lll_robust_lock_wait (int *futex, int private) attribute_hidden;
>> #define lll_cond_lock(futex, private) __lll_cond_lock (&(futex), private)
>>
>>
>> +/* If FUTEX is 0 (not acquired), set to ID | FUTEX_WAITERS (acquired, possibly
>> + with waiters) and return 0. Otherwise, block until either:
>> + 1) We acquire the lock, at which point FUTEX will be ID | FUTEX_WAITERS
>> + (acquired, possibly with waiters), then return 0.
>> + 2) The previous owner of the lock dies. FUTEX will be the value of id as
>> + set by the previous owner, with FUTEX_OWNER_DIED set (FUTEX_WAITERS may
>> + also be set). Return this value to indicate that the lock is not
>> + acquired. */
>> #define lll_robust_cond_lock(futex, id, private) \
>> __lll_robust_lock (&(futex), (id) | FUTEX_WAITERS, private)
>>
>> @@ -88,8 +161,9 @@ extern int __lll_timedlock_wait (int *futex, const struct timespec *,
>> extern int __lll_robust_timedlock_wait (int *futex, const struct timespec *,
>> int private) attribute_hidden;
>>
>> -/* Take futex if it is untaken.
>> - Otherwise block until either we get the futex or abstime runs out. */
>> +
>> +/* As __lll_lock, but with a timeout. If the timeout occurs then return
>> + ETIMEDOUT. If ABSTIME is invalid, return EINVAL. */
>> #define __lll_timedlock(futex, abstime, private) \
>> ({ \
>> int *__futex = (futex); \
>> @@ -104,6 +178,8 @@ extern int __lll_robust_timedlock_wait (int *futex, const struct timespec *,
>> __lll_timedlock (&(futex), abstime, private)
>>
>>
>> +/* As __lll_robust_lock, but with a timeout. If the timeout occurs then return
>> + ETIMEDOUT. If ABSTIME is invalid, return EINVAL. */
>> #define __lll_robust_timedlock(futex, abstime, id, private) \
>> ({ \
>> int *__futex = (futex); \
>> @@ -121,6 +197,9 @@ extern int __lll_robust_timedlock_wait (int *futex, const struct timespec *,
>> /* This is an expression rather than a statement even though its value is
>> void, so that it can be used in a comma expression or as an expression
>> that's cast to void. */
>> +/* Unconditionally set FUTEX to 0 (not acquired), releasing the lock. If FUTEX
>> + was >1 (acquired, possibly with waiters), then wake any waiters. The waiter
>> + who acquires the lock will set FUTEX to >1. */
>> #define __lll_unlock(futex, private) \
>> ((void) \
>> ({ \
>> @@ -136,6 +215,9 @@ extern int __lll_robust_timedlock_wait (int *futex, const struct timespec *,
>> /* This is an expression rather than a statement even though its value is
>> void, so that it can be used in a comma expression or as an expression
>> that's cast to void. */
>> +/* Unconditionally set FUTEX to 0 (not acquired), releasing the lock. If FUTEX
>> + had FUTEX_WAITERS set then wake any waiters. The waiter who acquires the
>> + lock will set FUTEX_WAITERS. */
>> #define __lll_robust_unlock(futex, private) \
>> ((void) \
>> ({ \
>> @@ -159,15 +241,12 @@ extern int __lll_robust_timedlock_wait (int *futex, const struct timespec *,
>> #define LLL_LOCK_INITIALIZER (0)
>> #define LLL_LOCK_INITIALIZER_LOCKED (1)
>>
>> -/* The states of a lock are:
>> - 0 - untaken
>> - 1 - taken by one user
>> - >1 - taken by more users */
>>
>> /* The kernel notifies a process which uses CLONE_CHILD_CLEARTID via futex
>> - wakeup when the clone terminates. The memory location contains the
>> - thread ID while the clone is running and is reset to zero
>> - afterwards. */
>> + wake-up when the clone terminates. The memory location contains the
>> + thread ID while the clone is running and is reset to zero by the kernel
>> + afterwards. The kernel up to version 3.16.3 does not use the private futex
>> + operations for futex wake-up when the clone terminates. */
>> #define lll_wait_tid(tid) \
>> do { \
>> __typeof (tid) __tid; \
>> @@ -178,6 +257,8 @@ extern int __lll_robust_timedlock_wait (int *futex, const struct timespec *,
>> extern int __lll_timedwait_tid (int *, const struct timespec *)
>> attribute_hidden;
>>
>> +/* As lll_wait_tid, but with a timeout. If the timeout occurs then return
>> + ETIMEDOUT. If ABSTIME is invalid, return EINVAL. */
>> #define lll_timedwait_tid(tid, abstime) \
>> ({ \
>> int __res = 0; \