The old RUBY_GC_HEAP_INIT_SLOTS isn't really usable anymore as
it initalize all the pools by the same factor, but it's unlikely
that pools will need similar sizes.
In production our 40B pool is 5 to 6 times bigger than our 80B pool.
It's not uncommon for simple binding to wrap structs without
any Ruby object references. Hence with no `mark` function.
Might as well mark them as protected by a write barrier.
There's a memory leak in ObjectSpace::WeakMap due to not freeing
the `struct weakmap`. It can be seen in the following script:
```
100.times do
10000.times do
ObjectSpace::WeakMap.new
end
# Output the Resident Set Size (memory usage, in KB) of the current Ruby process
puts `ps -o rss= -p #{$$}`
end
```
Instance variables held in gen_ivtbl are marked with rb_gc_mark. It
prevents the referenced objects from moving, which is bad for copying
garbage collectors.
This commit allows those instance variables to be updated during
gc_update_object_references.
This commit adds rb_gc_mark_and_move which takes a pointer to an object
and marks it during marking phase and updates references during compaction.
This allows for marking and reference updating to be combined into a
single function, which reduces code duplication and prevents bugs if
marking and reference updating goes out of sync.
This commit also implements rb_gc_mark_and_move on iseq as an example.
This commit moves the classpath (and tmp_classpath) from instance
variables to the rb_classext_t. This improves performance as we no
longer need to set an instance variable when assigning a classpath to
a class.
I benchmarked with the following script:
```ruby
name = :MyClass
puts(Benchmark.measure do
10_000_000.times do |i|
Object.const_set(name, Class.new)
Object.send(:remove_const, name)
end
end)
```
Before this patch:
```
5.440119 0.025264 5.465383 ( 5.467105)
```
After this patch:
```
4.889646 0.028325 4.917971 ( 4.942678)
```
The reference updating code for strings is not re-embedding strings
because the code is incorrectly wrapped inside of a
`if (STR_SHARED_P(obj))` clause. Shared strings can't be re-embedded
so this ends up being a no-op. This means that strings can be moved to a
large size pool during compaction, but won't be re-embedded, which would
waste the space.
There is an integer underflow when the environment variable
RUBY_GC_HEAP_INIT_SLOTS is less than the number of slots currently
in the Ruby heap.
[Bug #19284]
If a size pooll is small, then `min_free_slots < heap_init_slots` is true.
This means that min_free_slots will be set to heap_init_slots. This
causes `swept_slots < min_free_slots` to be true in a later if statement.
The if statement could trigger a major GC which could cause major GC
thrashing.
gc_compact_move incorrectly returns false when destination heap is full
after sweeping. It returns false even if destination heap is different
than source heap (returning false means that the source heap has
finished compacting). This causes the source page to get locked, which
causes a read barrier fire when we try to compact the source heap again.
We eagerly set the new shape of an object when moving an object during
compaction. This new shape may have a different capacity than the
current original shape capacity. This means that we cannot copy from the
original buffer using size of the new capacity. Instead, we should use
the ivar count (which is less than or equal to both the new and original
capacities).
Co-Authored-By: Matt Valentine-House <matt@eightbitraptor.com>
Allocating memory (xmalloc and xrealloc) during GC could cause GC to
trigger, which would crash with `[BUG] during_gc != 0`. This is an
intermittent bug which could be hard to debug.
This commit changes it so that any memory allocation during GC will
emit a warning. When debug flags are enabled it will also cause a crash.
When moving Objects between size pools we have to assign a new shape.
This happened during updating references - we tried to create a new shape
tree that mirrored the existing tree, but based on the root shape of the
new size pool.
This causes allocations to happen if the new tree doesn't already exist,
potentially triggering a GC, during GC.
This commit changes object movement to look for a pre-existing new tree
during object movement, and if that tree does not exist, we don't move
the object to the new pool.
This allows us to remove the shape allocation from update references.
Co-Authored-By: Peter Zhu <peter@peterzhu.ca>
When an object becomes "too complex" (in other words it has too many
variations in the shape tree), we transition it to use a "too complex"
shape and use a hash for storing instance variables.
Without this patch, there were rare cases where shape tree growth could
"explode" and cause performance degradation on what would otherwise have
been cached fast paths.
This patch puts a limit on shape tree growth, and gracefully degrades in
the rare case where there could be a factorial growth in the shape tree.
For example:
```ruby
class NG; end
HUGE_NUMBER.times do
NG.new.instance_variable_set(:"@unique_ivar_#{_1}", 1)
end
```
We consider objects to be "too complex" when the object's class has more
than SHAPE_MAX_VARIATIONS (currently 8) leaf nodes in the shape tree and
the object introduces a new variation (a new leaf node) associated with
that class.
For example, new variations on instances of the following class would be
considered "too complex" because those instances create more than 8
leaves in the shape tree:
```ruby
class Foo; end
9.times { Foo.new.instance_variable_set(":@uniq_#{_1}", 1) }
```
However, the following class is *not* too complex because it only has
one leaf in the shape tree:
```ruby
class Foo
def initialize
@a = @b = @c = @d = @e = @f = @g = @h = @i = nil
end
end
9.times { Foo.new }
``
This case is rare, so we don't expect this change to impact performance
of most applications, but it needs to be handled.
Co-Authored-By: Aaron Patterson <tenderlove@ruby-lang.org>
When moving Objects between size pools we have to assign a new shape.
This happened during updating references - we tried to create a new shape
tree that mirrored the existing tree, but based on the root shape of the
new size pool.
This causes allocations to happen if the new tree doesn't already exist,
potentially triggering a GC, during GC.
This commit changes object movement to look for a pre-existing new tree
during object movement, and if that tree does not exist, we don't move
the object to the new pool.
This allows us to remove the shape allocation from update references.
Co-Authored-By: Peter Zhu <peter@peterzhu.ca>
We can loosely predict the number of ivar sets on a class based on the
number of iv set instructions in the initialize method. This should give
us a more accurate estimate to use for initial size pool allocation,
which should in turn give us more cache hits.
This commit adds RVALUE_OVERHEAD for storing metadata at the end of the
slot. This commit moves the ractor_belonging_id in debug builds from the
flags to RVALUE_OVERHEAD which frees the 16 bits in the headers for
object shapes.
We would like to differentiate types of objects via their shape. This
commit adds a special T_OBJECT shape when we allocate an instance of
T_OBJECT. This allows us to avoid testing whether an object is an
instance of a T_OBJECT or not, we can just check the shape.
Since object shapes store the capacity of an object, we no longer
need the numiv field on RObjects. This gives us one extra slot which
we can use to give embedded objects one more instance variable (for a
total of 3 ivs). This commit removes the concept of numiv from RObject.
This commit adds a `capacity` field to shapes, and adds shape
transitions whenever an object's capacity changes. Objects which are
allocated out of a bigger size pool will also make a transition from the
root shape to the shape with the correct capacity for their size pool
when they are allocated.
This commit will allow us to remove numiv from objects completely, and
will also mean we can guarantee that if two objects share shapes, their
IVs are in the same positions (an embedded and extended object cannot
share shapes). This will enable us to implement ivar sets in YJIT using
object shapes.
Co-Authored-By: Aaron Patterson <tenderlove@ruby-lang.org>
fiber machine stack is placed outside of C stack allocated by wasm-ld,
so highest stack address recorded by `rb_wasm_record_stack_base` is
invalid when running on non-main fiber.
Therefore, we should scan `stack_{start,end}` which always point a valid
stack range in any context.
We were previously incrementing the max_iv_count on a class in gc
freeing. By the time we free an object though, we're not guaranteed its
class is still valid. Instead, we can do this when marking and we're
guaranteed the object still knows its class.
* Avoid RCLASS_IV_TBL in marshal.c
* Avoid RCLASS_IV_TBL for class names
* Avoid RCLASS_IV_TBL for autoload
* Avoid RCLASS_IV_TBL for class variables
* Avoid copying RCLASS_IV_TBL onto ICLASSes
* Use object shapes for Class and Module IVs
`iv_count` is a misleading name because when IVs are unset, the new
shape doesn't decrement this value. `next_iv_count` is an accurate, and
more descriptive name.
Before object shapes, we were using class serial to invalidate
inline caches. Now that we use shape_id for inline cache keys,
the class serial is unnecessary.
Co-Authored-By: Aaron Patterson <tenderlove@ruby-lang.org>
Shapes gives us an almost exact count of instance variables on an
object. Since we know the number of instance variables that have been
set, we will never access slots that haven't been initialized with an
IV.
Shapes provides us with an (almost) exact count of instance variables.
We only need to check for Qundef when an IV has been "undefined"
Prefer to use ROBJECT_IV_COUNT when iterating IVs
GCC 12 introduced a new warning flag `-Wuse-after-free`, however it
has a false positive at `realloc` when optimization is disabled, since
the memory requested for reallocation is guaranteed to not be touched.
This workaround is very unclear why the false warning is suppressed by
a statement-expression GCC extension.
Object Shapes is used for accessing instance variables and representing the
"frozenness" of objects. Object instances have a "shape" and the shape
represents some attributes of the object (currently which instance variables are
set and the "frozenness"). Shapes form a tree data structure, and when a new
instance variable is set on an object, that object "transitions" to a new shape
in the shape tree. Each shape has an ID that is used for caching. The shape
structure is independent of class, so objects of different types can have the
same shape.
For example:
```ruby
class Foo
def initialize
# Starts with shape id 0
@a = 1 # transitions to shape id 1
@b = 1 # transitions to shape id 2
end
end
class Bar
def initialize
# Starts with shape id 0
@a = 1 # transitions to shape id 1
@b = 1 # transitions to shape id 2
end
end
foo = Foo.new # `foo` has shape id 2
bar = Bar.new # `bar` has shape id 2
```
Both `foo` and `bar` instances have the same shape because they both set
instance variables of the same name in the same order.
This technique can help to improve inline cache hits as well as generate more
efficient machine code in JIT compilers.
This commit also adds some methods for debugging shapes on objects. See
`RubyVM::Shape` for more details.
For more context on Object Shapes, see [Feature: #18776]
Co-Authored-By: Aaron Patterson <tenderlove@ruby-lang.org>
Co-Authored-By: Eileen M. Uchitelle <eileencodes@gmail.com>
Co-Authored-By: John Hawthorn <john@hawthorn.email>
Tabs were expanded because the file did not have any tab indentation in unedited lines.
Please update your editor config, and use misc/expand_tabs.rb in the pre-commit hook.
Object Shapes is used for accessing instance variables and representing the
"frozenness" of objects. Object instances have a "shape" and the shape
represents some attributes of the object (currently which instance variables are
set and the "frozenness"). Shapes form a tree data structure, and when a new
instance variable is set on an object, that object "transitions" to a new shape
in the shape tree. Each shape has an ID that is used for caching. The shape
structure is independent of class, so objects of different types can have the
same shape.
For example:
```ruby
class Foo
def initialize
# Starts with shape id 0
@a = 1 # transitions to shape id 1
@b = 1 # transitions to shape id 2
end
end
class Bar
def initialize
# Starts with shape id 0
@a = 1 # transitions to shape id 1
@b = 1 # transitions to shape id 2
end
end
foo = Foo.new # `foo` has shape id 2
bar = Bar.new # `bar` has shape id 2
```
Both `foo` and `bar` instances have the same shape because they both set
instance variables of the same name in the same order.
This technique can help to improve inline cache hits as well as generate more
efficient machine code in JIT compilers.
This commit also adds some methods for debugging shapes on objects. See
`RubyVM::Shape` for more details.
For more context on Object Shapes, see [Feature: #18776]
Co-Authored-By: Aaron Patterson <tenderlove@ruby-lang.org>
Co-Authored-By: Eileen M. Uchitelle <eileencodes@gmail.com>
Co-Authored-By: John Hawthorn <john@hawthorn.email>
Poisoned regions cannot be accessed without unpoisoning outside gc.c.
Specifically, debug.gem is terminated by AddressSanitizer.
```
SUMMARY: AddressSanitizer: use-after-poison iseq_collector.c:39 in iseq_i
```
Tabs were expanded because the file did not have any tab indentation in unedited lines.
Please update your editor config, and use misc/expand_tabs.rb in the pre-commit hook.
rb_ary_tmp_new suggests that the array is temporary in some way, but
that's not true, it just creates an array that's hidden and not on the
transient heap. This commit renames it to rb_ary_hidden_new.
Before this commit, if we don't have enough slots after sweeping but
had pages on the tomb heap, then the GC would frequently allocate and
deallocate pages. This is because after sweeping it would set
allocatable pages (since there were not enough slots) but free the
pages on the tomb heap.
This commit reuses pages on the tomb heap if there's not enough slots
after sweeping.
Prior to this commit it was possible to call `ObjectSpace._id2ref` with
an offset static symbol object_id and get back a new, incorrectly tagged
symbol:
```
> sensible_sym = ObjectSpace._id2ref(:a.object_id)
=> :a
> nonsense_sym = ObjectSpace._id2ref(:a.object_id + 40)
=> :a
> sensible_sym == nonsense_sym
=> false
```
`nonsense_sym` ends up tagged with `RUBY_ID_INSTANCE` instead of
`RB_ID_LOCAL`. That means we can do silly things like:
```
> foo = Object.new
> foo.instance_variable_set(:a, 123)
(irb):2:in `instance_variable_set': `a' is not allowed as an instance variable name (NameError)
> foo.instance_variable_set(ObjectSpace._id2ref(:a.object_id + 40), 123)
=> 123
> foo.instance_variables
=> [:a]
```
This was happening because `get_id_entry` ignores the tag bits when
looking up the symbol. So `rb_id2str(symid)` would return a value and
then we'd continue on with the nonsense `symid`.
This commit prevents the situation by checking that the `symid` actually
matches what we get back from `get_id_entry`. Now we get a `RangeError`
for the nonsense id:
```
> ObjectSpace._id2ref(:a.object_id)
=> :a
> ObjectSpace._id2ref(:a.object_id + 40)
(irb):1:in `_id2ref': 0x000000000013f408 is not symbol id value (RangeError)
```
Co-authored-by: John Hawthorn <jhawthorn@github.com>
In wmap_live_p, if is_pointer_to_heap returns false, then the page is
either in the tomb or has already been freed, so the object is dead. In
this case, wmap_live_p should return false.
This commit implements Objects on Variable Width Allocation. This allows
Objects with more ivars to be embedded (i.e. contents directly follow the
object header) which improves performance through better cache locality.
This commit enables Arrays to move between size pools during compaction.
This can occur if the array is mutated such that it would fit in a
different size pool when embedded.
The move is carried out in two stages:
1. The RVALUE is moved to a destination heap during object movement
phase of compaction
2. The array data is re-embedded and the original buffer free'd if
required. This happens during the update references step
In order to reliably test compaction we need to be able to move objects
between size pools.
In order for this to happen there must be pages in a size pool into
which we can allocate.
The existing implementation of `double_heap` only doubled the existing
number of pages in the heap, so if a size pool had a low number of pages
(or 0) it's not guaranteed that enough space will be created to move
objects into that size pool.
This commit deprecates the `double_heap` option and replaces it with
`expand_heap` instead.
expand heap will expand each heap by enough pages to hold a number of
slots defined by `GC_HEAP_INIT_SLOTS` or by `heap->total_pags` whichever
is larger.
If both `double_heap` and `expand_heap` are present, a deprecation
warning will be shown for `double_heap` and the `expand_heap` behaviour
will take precedence
Given that this is an API intended for debugging and testing GC
compaction I'm not concerned about the extra memory usage or time taken
to create the pages. However, for completeness:
Running the following `test.rb` and using `time` on my Macbook Pro shows
the following memory usage and time impact:
pp "RSS (kb): #{`ps -o rss #{Process.pid}`.lines.last.to_i}"
GC.verify_compaction_references(double_heap: true, toward: :empty)
pp "RSS (kb): #{`ps -o rss #{Process.pid}`.lines.last.to_i}"
❯ time make run
./miniruby -I./lib -I. -I.ext/common -r./arm64-darwin21-fake ./test.rb
"RSS (kb): 24000"
<internal:gc>:251: warning: double_heap is deprecated and will be removed
"RSS (kb): 25232"
________________________________________________________
Executed in 124.37 millis fish external
usr time 82.22 millis 0.09 millis 82.12 millis
sys time 28.76 millis 2.61 millis 26.15 millis
❯ time make run
./miniruby -I./lib -I. -I.ext/common -r./arm64-darwin21-fake ./test.rb
"RSS (kb): 24000"
"RSS (kb): 49040"
________________________________________________________
Executed in 150.13 millis fish external
usr time 103.32 millis 0.10 millis 103.22 millis
sys time 35.73 millis 2.59 millis 33.14 millis
If the page_body is a null pointer, then read_barrier_handler will
crash with an unrelated message. This commit improves the error message.
Before:
test.rb:1: [BUG] Couldn't unprotect page 0x0000000000000000, errno: Cannot allocate memory
After:
test.rb:1: [BUG] read_barrier_handler: segmentation fault at 0x14
The GC compaction mechanism implements a kind of read barrier by marking
some (OS) pages as unreadable, and installing a SIGBUS/SIGSEGV handler
to detect when they're accessed and invalidate an attempt to move the
object.
Unfortunately, when a debugger is attached to the Ruby interpreter on
Mac OS, the debugger will trap the EXC_BAD_ACCES mach exception before
the runtime can transform that into a SIGBUS signal and dispatch it.
Thus, execution gets stuck; any attempt to continue from the debugger
re-executes the line that caused the exception and no forward progress
can be made.
This makes it impossible to debug either the Ruby interpreter or a C
extension whilst compaction is in use.
To fix this, we disable the EXC_BAD_ACCESS handler when installing the
SIGBUS/SIGSEGV handlers, and re-enable them once the compaction is done.
The debugger will still trap on the attempt to read the bad page, but it
will be trapping the SIGBUS signal, rather than the EXC_BAD_ACCESS mach
exception. It's possible to continue from this in the debugger, which
invokes the signal handler and allows forward progress to be made.
Commit 0c36ba5319 changed GC compaction
methods to not be implemented when not supported. However, that commit
only does compile time checks (which currently only checks for WASM),
but there are additional compaction support checks during run time.
This commit changes it so that GC compaction methods aren't defined
during run time if the platform does not support GC compaction.
[Bug #18829]
Only growth heaps are allowed to start major GCs. Before this patch,
growth heaps are defined as size pools that freed more slots than had
empty slots (i.e. there were more dead objects that empty space).
But if the size pool is relatively stable and tightly packed with mostly
old objects and has allocatable pages, then it would be incorrectly
classified as a growth heap and trigger major GC. But since it's stable,
it would not use any of the allocatable pages and forever be classified
as a growth heap, causing major GC thrashing. This commit changes the
definition of growth heap to require that the size pool to have no
allocatable pages.
Having a while loop over `heap_prepare` makes the GC logic difficult to
understand (it is difficult to understand when and why `heap_prepare`
yields a free page). It is also a source of bugs and can cause an infinite
loop if `heap_page` never yields a free page.
Fixes [Bug #18779]
Define the following methods as `rb_f_notimplement` on unsupported
platforms:
- GC.compact
- GC.auto_compact
- GC.auto_compact=
- GC.latest_compact_info
- GC.verify_compaction_references
This change allows users to call `GC.respond_to?(:compact)` to
properly test for compaction support. Previously, it was necessary to
invoke `GC.compact` or `GC.verify_compaction_references` and check if
those methods raised `NotImplementedError` to determine if compaction
was supported.
This follows the precedent set for other platform-specific
methods. For example, in `process.c` for methods such as
`Process.fork`, `Process.setpgid`, and `Process.getpriority`.
These methods are removed from gc.rb and added to gc.c:
- GC.compact
- GC.auto_compact
- GC.auto_compact=
- GC.latest_compact_info
- GC.verify_compaction_references
This is a prefactor to allow setting these methods to
`rb_f_notimplement` in a followup commit.
Some size pools may not have any pages/slots, so total_slots is 0. This
causes a divide-by-zero in the calculation. This commit adds a special
case to catch the case when total_slots is 0 and returns the number of
pages for heap_init_slots.
If the size pool has no or few pages/slots, then min_free_slots will
be a very small number (or even 0). Then the heap won't be eligible to
grow, causing GC thrashing or infinite loops.
Size pools with no pages won't be swept so gc_sweep_finish_size_pool
will never be called on it, but gc_sweep_finish_size_pool must be called
to grow the size pool.
Depending on alignment, the last bitmap plane may not used. Then it will
appear as if all of the objects on that plane is unmarked, which will
cause a buffer overrun when we try to free the object. This commit
changes the loop to calculate the number of planes used
(bitmap_plane_count).
Since 4d8f76286b, we need to dereference
the includer field on iclasses, so we need to mark it to make sure
it's alive.
Sometimes during compaction we crash because the field is dangling,
though I have a hard time constructing such a situation. See
http://ci.rvm.jp/results/trunk@ruby-iga/3947725
We didn't update the includer field during compaction so it could become
a dangling pointer after compaction. It's only recently that we started
to dereference the field, and we were only comparing the pointer before
then, so the omission only recently started to cause crashes.
By instrumenting object.c:833 with `rp(includer);`, you can see the
includer field become `T_NONE` with the following script:
```ruby
mod = Module.new do
protected def foo = 1
end
klass = Class.new do
include Module.new
def run
foo
end
end
klass.include(mod)
GC.verify_compaction_references(double_heap: true, toward: :empty)
klass.new.run
```
I found a crash in a private application that this patch fixes, but
wasn't able to develop a small reproducer. Hence the above demo that
requires instrumentation.
During VM startup, rb_objspace_alloc sets malloc_limit
(objspace->malloc_params.limit) before ruby_gc_set_params is called, thus
nullifying the effect of RUBY_GC_MALLOC_LIMIT before the initial GC run.
The call sequence is as follows:
main.c::main()
ruby_init
ruby_setup
Init_BareVM
rb_objspace_alloc // malloc_limit = gc_params.malloc_limit_min;
ruby_options
ruby_process_options
process_options
ruby_gc_set_params // RUBY_GC_MALLOC_LIMIT => gc_params.malloc_limit_min
With ruby_gc_set_params setting malloc_limit, RUBY_GC_MALLOC_LIMIT
affects the process sooner.
[ruby-core:107170]
Commit dde164e968 decoupled incremental
marking from page sizes. This commit changes Ruby heap page sizes to
64KiB. Doing so will have several benefits:
1. We can use compaction on systems with 64KiB system page sizes (e.g.
PowerPC).
2. Larger page sizes will allow Variable Width Allocation to increase
slot sizes and embed larger objects.
3. Since commit 002fa28599, macOS has 64
KiB pages. Making page sizes 64 KiB will bring these systems to
parity.
I have attached some bechmark results below.
Discourse:
On Discourse, we saw much better p99 performance (e.g. for "categories"
it went from 214ms on master to 134ms on branch, for "home" it went
from 265ms to 251ms). We don’t see much change in p60, p75, and p90
performance. We also see a slight decrease in memory usage by 1.04x.
Branch RSS: 354.9MB
Master RSS: 368.2MB
railsbench:
On rails bench, we don’t see a big change in RPS or p99
performance. We don’t see a big difference in memory usage.
Branch RPS: 826.27
Master RPS: 824.85
Branch p99: 1.67
Master p99: 1.72
Branch RSS: 88.72MB
Master RSS: 88.48MB
liquid:
We don’t see a significant change in liquid performance.
Branch parse & render: 28.653 I/s
Master parse & render: 28.563 i/s
Currently, rb_aligned_malloc uses mmap if Ruby heap pages can be
allocated through mmap (when system heap page size <= Ruby heap page
size). If Ruby heap page sizes is increased to 64KiB, then mmap will
be used on systems with 64KiB system page sizes. However, the transient
heap also uses rb_aligned_malloc and requires 32KiB alignment. This
would break in the current implementation since it would allocate sizes
through mmap that is not a multiple of the system page size.
This commit adds heap_page_body_allocate which will use mmap when
possible and changes rb_aligned_malloc to not use mmap (and only
use posix_memalign).
This commit changes the way compaction moves objects and sweeps pages in
order to better facilitate object movement between size pools.
Previously we would move the scan cursor first until we found an empty
slot and then we'd decrement the compact cursor until we found something
to move into that slot. We would sweep the page that contained the scan
cursor before trying to fill it
In this algorithm we first move the compact cursor down until we find an
object to move - We then take a free page from the desired destination
heap (always the same heap in this current iteration of the code).
If there is no free page we sweep the page at the sweeping_page cursor,
add it to the free pages, and advance the cursor to the next page, and
try again.
We sweep one page from each size pool in this way, and then repeat that
process until all the size pools are compacted (all the cursors have
met), and then we update references and sweep the rest of the heap.
Currently, the number of incremental marking steps is calculated based
on the number of pooled pages available. This means that if we make Ruby
heap pages larger, it would run fewer incremental marking steps (which
would mean each incremental marking step takes longer).
This commit changes incremental marking to run after every
INCREMENTAL_MARK_STEP_ALLOCATIONS number of allocations. This means that
the behaviour of incremental marking remains the same regardless of the
Ruby heap page size.
I've benchmarked against discourse benchmarks and did not get a
significant change in response times beyond the margin of error. This is
expected as this new incremental marking algorithm behaves very
similarly to the previous one.
Use ISEQ_BODY macro to get the rb_iseq_constant_body of the ISeq. Using
this macro will make it easier for us to change the allocation strategy
of rb_iseq_constant_body when using Variable Width Allocation.
Previously, we would build a new `superclasses` array for each class,
even though for all immediate subclasses of a class, the array is
identical.
This avoids duplicating the arrays on leaf classes (those without
subclasses) by calculating and storing a "superclasses including self"
array on a class when it's first inherited and sharing that among all
superclasses.
An additional trick used is that the "superclass array including self"
is valid as "self"'s superclass array. It just has it's own class at the
end. We can use this to avoid an extra pointer of storage and can use
one bit of a flag to track that we've "upgraded" the array.
Previously when checking ancestors, we would walk all the way up the
ancestry chain checking each parent for a matching class or module.
I believe this was especially unfriendly to CPU cache since for each
step we need to check two cache lines (the class and class ext).
This check is used quite often in:
* case statements
* rescue statements
* Calling protected methods
* Class#is_a?
* Module#===
* Module#<=>
I believe it's most common to check a class against a parent class, to
this commit aims to improve that (unfortunately does not help checking
for an included Module).
This is done by storing on each class the number and an array of all
parent classes, in order (BasicObject is at index 0). Using this we can
check whether a class is a subclass of another in constant time since we
know the location to expect it in the hierarchy.
(1) gc_verify_internal_consistency() use barrier locking
for consistency while `during_gc == true` at the end
of the sweep on `RGENGC_CHECK_MODE >= 2`.
(2) `rb_objspace_reachable_objects_from()` is called without
VM synchronization and it checks `during_gc != true`.
So (1) and (2) causes BUG because of `during_gc == true`.
To prevent this error, wait for VM barrier on `during_gc == false`
and introduce VM locking on `rb_objspace_reachable_objects_from()`.
http://ci.rvm.jp/results/trunk-asserts@phosphorus-docker/3830088
gc_marks_continue will start sweeping when it finishes marking. However,
if the heap we are trying to allocate into is full, then the sweeping
may not yield any free slots. If we don't call gc_sweep_continue
immediate after this, then another GC will be started halfway during
lazy sweeping. gc_sweep_continue will either grow the heap or finish
sweeping.
Add a new macro BASE_SLOT_SIZE that determines the slot size.
For Variable Width Allocation (compiled with USE_RVARGC=1), all slot
sizes are powers-of-2 multiples of BASE_SLOT_SIZE.
For USE_RVARGC=0, BASE_SLOT_SIZE is set to sizeof(RVALUE).
Renames rb_id_table_foreach_with_replace to
rb_id_table_foreach_values_with_replace and passes only the value to the
callback. We can use this in GC compaction when we cannot access the
global symbol array.
NUM_IN_PAGE could return a value much larger than 64. According to the
C11 spec 6.5.7 paragraph 3 this is undefined behavior:
> If the value of the right operand is negative or is greater than or
> equal to the width of the promoted left operand, the behavior is
> undefined.
On most platforms, this is usually not a problem as the architecture
will mask off all out-of-range bits.
WebAssembly has function local infinite registers and stack values, but
there is no way to scan the values in a call stack for now.
This implementation uses Asyncify to spilling out wasm locals into
linear memory.
On 32-bit systems, VWA causes class_serial to not be aligned (it only
guarantees 4 byte alignment but class_serial is 8 bytes and requires 8
byte alignment). This commit uses a hack to allocate class_serial
through malloc. Once VWA allocates with 8 byte alignment in the future,
we will revert this commit.
This commit switches from a custom implemented bsearch algorithm to
use the one provided by the C standard library.
Because `is_pointer_to_heap` will only return true if the pointer
being searched for is a valid slot starting address within the heap
page body, we've extracted the bsearch call site into a more general
function so we can use it elsewhere.
The new function `heap_page_for_ptr` returns the heap page for any heap
page pointer, regardless of whether that is at the start of a slot or
in the middle of one.
We then use this function as the basis of `is_pointer_to_heap`.
Some callable method entries (cme) can be a key of `overloaded_cme_table`
and the keys should be pinned because the table is numtable (VALUE is a key).
Before the patch GC checks the cme is in `overloaded_cme_table` by looking up
the table, but it needs VM locking.
It works well in normal GC marking because it is protected by the VM lock,
but it doesn't work on `rb_objspace_reachable_objects_from` because it doesn't
use VM lock.
Now, the number of target cmes are small enough, I decide to pin down
all possible cmes instead of using looking up the table.
`overloaded_cme_table` keeps cme -> monly_cme pairs to manage
corresponding `monly_cme` for `cme`. The lifetime of the `monly_cme`
should be longer than `monly_cme`, but the previous patch losts the
reference to the living `monly_cme`.
Now `overloaded_cme_table` values are always root (keys are only weak
reference), it means `monly_cme` does not freed until corresponding
`cme` is invalidated.
To make managing easy, move `overloaded_cme_table` to `rb_vm_t`.
`def` (`rb_method_definition_t`) is shared by multiple callable
method entries (cme, `rb_callable_method_entry_t`).
There are two issues:
* old -> young reference: `cme1->def->mandatory_only_cme = monly_cme`
if `cme1` is young and `monly_cme` is young, there is no problem.
Howevr, another old `cme2` can refer `def`, in this case, old `cme2`
points young `monly_cme` and it violates gengc assumption.
* cme can have different `defined_class` but `monly_cme` only has
one `defined_class`. It does not make sense and `monly_cme`
should be created for a cme (not `def`).
To solve these issues, this patch allocates `monly_cme` per `cme`.
`cme` does not have another room to store a pointer to the `monly_cme`,
so this patch introduces `overloaded_cme_table`, which is weak key map
`[cme] -> [monly_cme]`.
`def::body::iseqptr::monly_cme` is deleted.
The first issue is reported by Alan Wu.
When using `rp(obj)` for debugging during development, it may be
useful to know that an object is soon to be swept. Add a new letter to
the object dump for whether the object is garbage. It's easy to forget
about lazy sweep.
Except on Windows and MinGW, we can only use compaction on systems that
use mmap (only systems that use mmap can use the read barrier that
compaction requires). We don't need to separately detect whether we can
support compaction or not.
* Lazily create singletons on instance_{exec,eval}
Previously when instance_exec or instance_eval was called on an object,
that object would be given a singleton class so that method
definitions inside the block would be added to the object rather than
its class.
This commit aims to improve performance by delaying the creation of the
singleton class unless/until one is needed for method definition. Most
of the time instance_eval is used without any method definition.
This was implemented by adding a flag to the cref indicating that it
represents a singleton of the object rather than a class itself. In this
case CREF_CLASS returns the object's existing class, but in cases that
we are defining a method (either via definemethod or
VM_SPECIAL_OBJECT_CBASE which is used for undef and alias).
This also happens to fix what I believe is a bug. Previously
instance_eval behaved differently with regards to constant access for
true/false/nil than for all other objects. I don't think this was
intentional.
String::Foo = "foo"
"".instance_eval("Foo") # => "foo"
Integer::Foo = "foo"
123.instance_eval("Foo") # => "foo"
TrueClass::Foo = "foo"
true.instance_eval("Foo") # NameError: uninitialized constant Foo
This also slightly changes the error message when trying to define a method
through instance_eval on an object which can't have a singleton class.
Before:
$ ruby -e '123.instance_eval { def foo; end }'
-e:1:in `block in <main>': no class/module to add method (TypeError)
After:
$ ./ruby -e '123.instance_eval { def foo; end }'
-e:1:in `block in <main>': can't define singleton (TypeError)
IMO this error is a small improvement on the original and better matches
the (both old and new) message when definging a method using `def self.`
$ ruby -e '123.instance_eval{ def self.foo; end }'
-e:1:in `block in <main>': can't define singleton (TypeError)
Co-authored-by: Matthew Draper <matthew@trebex.net>
* Remove "under" argument from yield_under
* Move CREF_SINGLETON_SET into vm_cref_new
* Simplify vm_get_const_base
* Fix leaf VM_SPECIAL_OBJECT_CONST_BASE
Co-authored-by: Matthew Draper <matthew@trebex.net>
suseconds_t, which is the type of tv_usec, may be defined with a longer
size type than tv_nsec's type (long). So usec to nsec conversion needs
an explicit casting.
This commit adds a Ractor cache for every size pool. Previously, all VWA
allocated objects used the slowpath and locked the VM.
On a micro-benchmark that benchmarks String allocation:
VWA turned off:
29.196591 0.889709 30.086300 ( 9.434059)
VWA before this commit:
29.279486 41.477869 70.757355 ( 12.527379)
VWA after this commit:
16.782903 0.557117 17.340020 ( 4.255603)
Updating RCLASS_PARENT_SUBCLASSES and RCLASS_MODULE_SUBCLASSES while
compacting can trigger the read barrier. This commit makes
RCLASS_SUBCLASSES a doubly linked list with a dedicated head object so
that we can add and remove entries from the list without having to touch
an object in the Ruby heap
* `GC.measure_total_time = true` enables total time measurement (default: true)
* `GC.measure_total_time` returns current flag.
* `GC.total_time` returns measured total time in nano seconds.
* `GC.stat(:time)` (and Hash) returns measured total time in milli seconds.
Compare with the C methods, A built-in methods written in Ruby is
slower if only mandatory parameters are given because it needs to
check the argumens and fill default values for optional and keyword
parameters (C methods can check the number of parameters with `argc`,
so there are no overhead). Passing mandatory arguments are common
(optional arguments are exceptional, in many cases) so it is important
to provide the fast path for such common cases.
`Primitive.mandatory_only?` is a special builtin function used with
`if` expression like that:
```ruby
def self.at(time, subsec = false, unit = :microsecond, in: nil)
if Primitive.mandatory_only?
Primitive.time_s_at1(time)
else
Primitive.time_s_at(time, subsec, unit, Primitive.arg!(:in))
end
end
```
and it makes two ISeq,
```
def self.at(time, subsec = false, unit = :microsecond, in: nil)
Primitive.time_s_at(time, subsec, unit, Primitive.arg!(:in))
end
def self.at(time)
Primitive.time_s_at1(time)
end
```
and (2) is pointed by (1). Note that `Primitive.mandatory_only?`
should be used only in a condition of an `if` statement and the
`if` statement should be equal to the methdo body (you can not
put any expression before and after the `if` statement).
A method entry with `mandatory_only?` (`Time.at` on the above case)
is marked as `iseq_overload`. When the method will be dispatch only
with mandatory arguments (`Time.at(0)` for example), make another
method entry with ISeq (2) as mandatory only method entry and it
will be cached in an inline method cache.
The idea is similar discussed in https://bugs.ruby-lang.org/issues/16254
but it only checks mandatory parameters or more, because many cases
only mandatory parameters are given. If we find other cases (optional
or keyword parameters are used frequently and it hurts performance),
we can extend the feature.
With RVARGC we always store the rb_classext_t in the same slot as the
RClass struct that refers to it. So we don't need to store the pointer
or access through the pointer anymore and can switch the RCLASS_EXT
macro to use an offset
This commit fixes a memory leak introduced in an early part of the
variable width allocation project that would prevent the rb_classext_t
struct from being free'd when the class is swept.
This commit deprecates rb_gc_force_recycle and coverts it to a no-op
function. Also removes invalidate_mark_stack_chunk since only
rb_gc_force_recycle uses it.
```
../gc.c:2342:45: warning: comparison of integers of different signs: 'short' and 'size_t' (aka 'unsigned long') [-Wsign-compare]
GC_ASSERT(size_pools[pool_id].slot_size == slot_size);
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^~~~~~~~~~~~~
```
Add cast to short, because `GC_ASSERT`s in `size_pool_for_size`
already use cast to short.
Matt Valentine-House
Peter Zhu
Jean Boussier
Jean byroot Boussier
Jean Boussier
Kunshan Wang
Nobuyoshi Nakada
Jemma Issroff
Mirek Klimos
Aaron Patterson
S-H-GAMELINKS
Yuta Saito
John Hawthorn
Sergey Fedorov
Samuel Williams
git
Takashi Kokubun
Daniel Colson
KJ Tsanaktsidis
Mike Dalessio
Kaíque Kandy Koga
Alan Wu
Kazuhiro NISHIYAMA
Eric Wong
Koichi Sasada
Yusuke Endoh