| Age | Commit message (Collapse) | Author |
|---|
|
* sverker/prealloc-race-bug/OTP-14491:
erts: Fix bug in quick alloc
|
|
The effect of the race is that a pre-allocated memory block
is inserted last without updating tail.data.last, which will cause
all subsequent insertions to also fail to update tail.data.last.
Hence all pre-allocation for this quick alloc instance is leaked
for this thread and will fallback on erts_alloc.
|
|
|
|
|
|
The enqueue operation have been re-written to behave better
during heavy contention by spreading writes over multiple
locations. This enqueue operation also take advantage of the
delayed aux work wake up functionality and can by this omit
one memory barrier.
|
|
As an optimization old thread progress data was kept and used in
handle_aux_work() in erl_process.c. This could cause memory to be
deallocated at a later time than intended, which is quite harmless.
This has, however, now been fixed.
|
|
- Document barrier semantics
- Introduce ddrb suffix on atomic ops
- Barrier macros for both non-SMP and SMP case
- Make the thread progress API a bit more intuitive
|
|
A number of memory allocation optimizations have been implemented. Most
optimizations reduce contention caused by synchronization between
threads during allocation and deallocation of memory. Most notably:
* Synchronization of memory management in scheduler specific allocator
instances has been rewritten to use lock-free synchronization.
* Synchronization of memory management in scheduler specific
pre-allocators has been rewritten to use lock-free synchronization.
* The 'mseg_alloc' memory segment allocator now use scheduler specific
instances instead of one instance. Apart from reducing contention
this also ensures that memory allocators always create memory
segments on the local NUMA node on a NUMA system.
|
