| Age | Commit message (Collapse) | Author |
|---|
|
Communication between Erlang processes has conceptually always been
performed through asynchronous signaling. The runtime system
implementation has however previously preformed most operation
synchronously. In a system with only one true thread of execution, this
is not problematic (often the opposite). In a system with multiple threads
of execution (as current runtime system implementation with SMP support)
it becomes problematic. This since it often involves locking of structures
when updating them which in turn cause resource contention. Utilizing
true asynchronous communication often avoids these resource contention
issues.
The case that triggered this change was contention on the link lock due
to frequent updates of the monitor trees during communication with a
frequently used server. The signal order delivery guarantees of the
language makes it hard to change the implementation of only some signals
to use true asynchronous signaling. Therefore the implementations
of (almost) all signals have been changed.
Currently the following signals have been implemented as true
asynchronous signals:
- Message signals
- Exit signals
- Monitor signals
- Demonitor signals
- Monitor triggered signals (DOWN, CHANGE, etc)
- Link signals
- Unlink signals
- Group leader signals
All of the above already defined as asynchronous signals in the
language. The implementation of messages signals was quite
asynchronous to begin with, but had quite strict delivery constraints
due to the ordering guarantees of signals between a pair of processes.
The previously used message queue partitioned into two halves has been
replaced by a more general signal queue partitioned into three parts
that service all kinds of signals. More details regarding the signal
queue can be found in comments in the erl_proc_sig_queue.h file.
The monitor and link implementations have also been completely replaced
in order to fit the new asynchronous signaling implementation as good
as possible. More details regarding the new monitor and link
implementations can be found in the erl_monitor_link.h file.
|
|
|
|
|
|
This refactor was done using the unifdef tool like this:
for file in $(find erts/ -name *.[ch]); do unifdef -t -f defile -o $file $file; done
where defile contained:
#define ERTS_SMP 1
#define USE_THREADS 1
#define DDLL_SMP 1
#define ERTS_HAVE_SMP_EMU 1
#define SMP 1
#define ERL_BITS_REENTRANT 1
#define ERTS_USE_ASYNC_READY_Q 1
#define FDBLOCK 1
#undef ERTS_POLL_NEED_ASYNC_INTERRUPT_SUPPORT
#define ERTS_POLL_ASYNC_INTERRUPT_SUPPORT 0
#define ERTS_POLL_USE_WAKEUP_PIPE 1
#define ERTS_POLL_USE_UPDATE_REQUESTS_QUEUE 1
#undef ERTS_HAVE_PLAIN_EMU
#undef ERTS_SIGNAL_STATE
|
|
The implementation is still hidden behind ERTS_ENABLE_LOCK_COUNT, and
all categories are still enabled by default, but the actual counting can be
toggled at will.
OTP-13170
|
|
This allows us to enable/disable lock counting at will, and greatly improves
the performance of erts_debug:lock_counters/1 since we no longer have to
worry about the lock counters "dying" while we're enumerating them.
OTP-14412
|
|
|
|
* sverker/ASSERT_IN_ENV:
erts: Add macro ERTS_PROC_LOCKS_HIGHER_THAN
erts: Cleanup and extra assertions in nif_SUITE.c
erts: Cleanup enif_make_reverse_list
erts: Add assertions for correct ErlNifEnv
erts: Make erts_dbg_within_proc available
# Conflicts:
# erts/emulator/beam/erl_gc.h
|
|
|
|
to safeguard against bugs due to future proc lock changes.
The two places now using ERTS_PROC_LOCKS_HIGHER_THAN were
kind of bugs as BTM and TRACE locks were missing. But there was
probably no way to get there with BTM or TRACE locked.
|
|
Rickards said that this was ok
|
|
Add the possibility to use modules as trace data receivers. The functions
in the module have to be nifs as otherwise complex trace probes will be
very hard to handle (complex means trace probes for ports for example).
This commit changes the way that the ptab->tracer field works from always
being an immediate, to now be NIL if no tracer is present or else be
the tuple {TracerModule, TracerState} where TracerModule is an atom that
is later used to lookup the appropriate tracer callbacks to call and
TracerState is just passed to the tracer callback. The default process and
port tracers have been rewritten to use the new API.
This commit also changes the order which trace messages are delivered to the
potential tracer process. Any enif_send done in a tracer module may be delayed
indefinitely because of lock order issues. If a message is delayed any other
trace message send from that process is also delayed so that order is preserved
for each traced entity. This means that for some trace events (i.e. send/receive)
the events may come in an unintuitive order (receive before send) to the
trace receiver. Timestamps are taken when the trace message is generated so
trace messages from differented processes may arrive with the timestamp
out of order.
Both the erlang:trace and seq_trace:set_system_tracer accept the new tracer
module tracers and also the backwards compatible arguments.
OTP-10267
|
|
|
|
Do lock order check *before* trying to seize lock... duh!
|
|
Another process may already have been placed in this slot
since the free och the process struct can be scheduled for
later.
|
|
|
|
|
|
|
|
rickard/r16/port-optimizations/OTP-10336
* rickard/port-optimizations/OTP-10336:
Change annotate level for emacs-22 in cerl
Update etp-commands
Add documentation on communication in Erlang
Add support for busy port message queue
Add driver callback epilogue
Implement true asynchronous signaling between processes and ports
Add erl_drv_[send|output]_term
Move busy port flag
Use rwlock for driver list
Optimize management of port tasks
Improve configuration of process and port tables
Remove R9 compatibility features
Use ptab functionality also for ports
Prepare for use of ptab functionality also for ports
Atomic port state
Generalize process table implementation
Implement functionality for delaying thread progress from unmanaged threads
Conflicts:
erts/doc/src/erl_driver.xml
erts/doc/src/erlang.xml
erts/emulator/beam/beam_bif_load.c
erts/emulator/beam/beam_bp.c
erts/emulator/beam/beam_emu.c
erts/emulator/beam/bif.c
erts/emulator/beam/copy.c
erts/emulator/beam/erl_alloc.c
erts/emulator/beam/erl_alloc.types
erts/emulator/beam/erl_bif_info.c
erts/emulator/beam/erl_bif_port.c
erts/emulator/beam/erl_bif_trace.c
erts/emulator/beam/erl_init.c
erts/emulator/beam/erl_message.c
erts/emulator/beam/erl_port_task.c
erts/emulator/beam/erl_process.c
erts/emulator/beam/erl_process.h
erts/emulator/beam/erl_process_lock.c
erts/emulator/beam/erl_trace.c
erts/emulator/beam/export.h
erts/emulator/beam/global.h
erts/emulator/beam/io.c
erts/emulator/sys/unix/sys.c
erts/emulator/sys/vxworks/sys.c
erts/emulator/test/port_SUITE.erl
erts/etc/unix/cerl.src
erts/preloaded/ebin/erlang.beam
erts/preloaded/ebin/prim_inet.beam
erts/preloaded/src/prim_inet.erl
lib/hipe/cerl/erl_bif_types.erl
lib/kernel/doc/src/inet.xml
lib/kernel/src/inet.erl
|
|
|
|
|
|
Conflicts:
lib/diameter/autoconf/vxworks/sed.general
xcomp/README.md
|
|
|
|
* maint:
Use static allocation of process lock queues
Conflicts:
erts/emulator/beam/erl_process_lock.c
erts/emulator/beam/erl_process_lock.h
|
|
* rickard/proc-lock-queues/OTP-10163:
Use static allocation of process lock queues
|
|
By using statically allocated lock queues there is no longer
any need for locking corresponding pix lock when process
locks have been transferred after a wait. This costs us 3 words
extra in process structure, but improves performance during
contention.
|
|
* rickard/proc-sched/OTP-9892:
Teach etp-commands to understand new emulator internal data structures
Optimize process state changes
Optimize process table access
Implement possibility to use ordinary mutexes as process locks
Conflicts:
erts/emulator/beam/erl_alloc.types
|
|
|
|
|
|
|
|
|
|
Seen with valgrind running ets_SUITE:delete_large_tab
or delete_large_named_table.
|
|
The spinlocks used implementing pix-locks have been replaced with
mutexes since they perform better during heavy contention.
|
|
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.
|
|
All uses of the old deprecated atomic API in the runtime system
have been replaced with the use of the new atomic API. In a lot of
places this change imply a relaxation of memory barriers used.
|
|
The ethread atomics API now also provide double word size atomics.
Double word size atomics are implemented using native atomic
instructions on x86 (when the cmpxchg8b instruction is available)
and on x86_64 (when the cmpxchg16b instruction is available). On
other hardware where 32-bit atomics or word size atomics are
available, an optimized fallback is used; otherwise, a spinlock,
or a mutex based fallback is used.
The ethread library now performs runtime tests for presence of
hardware features, such as for example SSE2 instructions, instead
of requiring this to be determined at compile time.
There are now functions implementing each atomic operation with the
following implied memory barrier semantics: none, read, write,
acquire, release, and full. Some of the operation-barrier
combinations aren't especially useful. But instead of filtering
useful ones out, and potentially miss a useful one, we implement
them all.
A much smaller set of functionality for native atomics are required
to be implemented than before. More or less only cmpxchg and a
membar macro are required to be implemented for each atomic size.
Other functions will automatically be constructed from these. It is,
of course, often wise to implement more that this if possible from a
performance perspective.
|
|
|
|
|
|
|
|
Large parts of the ethread library have been rewritten. The
ethread library is an Erlang runtime system internal, portable
thread library used by the runtime system itself.
Most notable improvement is a reader optimized rwlock
implementation which dramatically improve the performance of
read-lock/read-unlock operations on multi processor systems by
avoiding ping-ponging of the rwlock cache lines. The reader
optimized rwlock implementation is used by miscellaneous
rwlocks in the runtime system that are known to be read-locked
frequently, and can be enabled on ETS tables by passing the
`{read_concurrency, true}' option upon table creation. See the
documentation of `ets:new/2' for more information.
The ethread library can now also use the libatomic_ops library
for atomic memory accesses. This makes it possible for the
Erlang runtime system to utilize optimized atomic operations
on more platforms than before. Use the
`--with-libatomic_ops=PATH' configure command line argument
when specifying where the libatomic_ops installation is
located. The libatomic_ops library can be downloaded from:
http://www.hpl.hp.com/research/linux/atomic_ops/
The changed API of the ethread library has also caused
modifications in the Erlang runtime system. Preparations for
the to come "delayed deallocation" feature has also been done
since it depends on the ethread library.
Note: When building for x86, the ethread library will now use
instructions that first appeared on the pentium 4 processor. If
you want the runtime system to be compatible with older
processors (back to 486) you need to pass the
`--enable-ethread-pre-pentium4-compatibility' configure command
line argument when configuring the system.
|
|
|
