1berry/ruby
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Correctly calculate initial pages

Browse Source 
The old algorithm could calculate an undercount for the initial pages
due to two issues:

1. It did not take into account that some heap pages will have one less
   slot due to alignment. It assumed that every heap page would be able
   to be fully filled with slots. Pages that are unaligned with the slot
   size will lose one slot. The new algorithm assumes that every page
   will be unaligned.
2. It performed integer division, which truncates down. This means that
   the number of pages might not actually satisfy the number of slots.

This can cause the heap to grow in `gc_sweep_finish_size_pool` after
allocating all of the allocatable pages because the total number of
slots would be less than the initial configured number of slots.
This commit is contained in:
Peter Zhu 2023-08-30 15:28:22 -04:00
parent eb3d94f4ba
commit 4f0d58260a
Notes: git 2023-08-31 13:28:51 +00:00 
Merged: https://github.com/ruby/ruby/pull/8333
2 changed files with 53 additions and 23 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

30
gc.c
View File

@ -2305,11 +2305,22 @@ heap_add_pages(rb_objspace_t *objspace, rb_size_pool_t *size_pool, rb_heap_t *he
#endif
static size_t
minimum_pages_for_size_pool(rb_objspace_t *objspace, int size_pool_idx)
slots_to_pages_for_size_pool(rb_objspace_t *objspace, rb_size_pool_t *size_pool, size_t slots)
{
rb_size_pool_t *size_pool = &size_pools[size_pool_idx];
int multiple = size_pool->slot_size / BASE_SLOT_SIZE;
return gc_params.size_pool_init_slots[size_pool_idx] * multiple / HEAP_PAGE_OBJ_LIMIT;
size_t multiple = size_pool->slot_size / BASE_SLOT_SIZE;
/* Due to alignment, heap pages may have one less slot. We should
* ensure there is enough pages to guarantee that we will have at
* least the required number of slots after allocating all the pages. */
size_t slots_per_page = (HEAP_PAGE_OBJ_LIMIT / multiple) - 1;
return CEILDIV(slots, slots_per_page);
}
static size_t
minimum_pages_for_size_pool(rb_objspace_t *objspace, rb_size_pool_t *size_pool)
{
size_t size_pool_idx = size_pool - size_pools;
size_t init_slots = gc_params.size_pool_init_slots[size_pool_idx];
return slots_to_pages_for_size_pool(objspace, size_pool, init_slots);
}
static size_t
@ -2322,7 +2333,7 @@ heap_extend_pages(rb_objspace_t *objspace, rb_size_pool_t *size_pool, size_t fre
next_used = (size_t)(used * gc_params.growth_factor);
}
else if (total_slots == 0) {
next_used = minimum_pages_for_size_pool(objspace, (int)(size_pool - size_pools));
next_used = minimum_pages_for_size_pool(objspace, size_pool);
}
else {
/* Find `f' where free_slots = f * total_slots * goal_ratio
@ -3728,10 +3739,12 @@ Init_heap(void)
/* Set size pools allocatable pages. */
for (int i = 0; i < SIZE_POOL_COUNT; i++) {
rb_size_pool_t *size_pool = &size_pools[i];
/* Set the default value of size_pool_init_slots. */
gc_params.size_pool_init_slots[i] = GC_HEAP_INIT_SLOTS;
size_pools[i].allocatable_pages = minimum_pages_for_size_pool(objspace, i);
size_pool->allocatable_pages = minimum_pages_for_size_pool(objspace, size_pool);
}
heap_pages_expand_sorted(objspace);
@ -10829,7 +10842,7 @@ gc_verify_compaction_references(rb_execution_context_t *ec, VALUE self, VALUE do
size_t minimum_pages = 0;
if (RTEST(expand_heap)) {
minimum_pages = minimum_pages_for_size_pool(objspace, i);
minimum_pages = minimum_pages_for_size_pool(objspace, size_pool);
}
heap_add_pages(objspace, size_pool, heap, MAX(minimum_pages, heap->total_pages));
@ -11643,8 +11656,7 @@ gc_set_initial_pages(rb_objspace_t *objspace)
if (size_pool_init_slots > size_pool->eden_heap.total_slots) {
size_t slots = size_pool_init_slots - size_pool->eden_heap.total_slots;
int multiple = size_pool->slot_size / BASE_SLOT_SIZE;
size_pool->allocatable_pages = slots * multiple / HEAP_PAGE_OBJ_LIMIT;
size_pool->allocatable_pages = slots_to_pages_for_size_pool(objspace, size_pool, slots);
}
else {
/* We already have more slots than size_pool_init_slots allows, so

46
test/ruby/test_gc.rb
View File

@ -444,12 +444,12 @@ class TestGc < Test::Unit::TestCase
# Constant from gc.c.
GC_HEAP_INIT_SLOTS = 10_000
GC.stat_heap.each do |_, s|
# Sometimes pages will have 1 less slot due to alignment, so always increase slots_per_page by 1.
slots_per_page = (s[:heap_eden_slots] / s[:heap_eden_pages]) + 1
multiple = s[:slot_size] / (GC::INTERNAL_CONSTANTS[:BASE_SLOT_SIZE] + GC::INTERNAL_CONSTANTS[:RVALUE_OVERHEAD])
# Allocatable pages are assumed to have lost 1 slot due to alignment.
slots_per_page = (GC::INTERNAL_CONSTANTS[:HEAP_PAGE_OBJ_LIMIT] / multiple) - 1
total_slots = s[:heap_eden_slots] + s[:heap_allocatable_pages] * slots_per_page
# Give a 0.9x delta because integer division in minimum_pages_for_size_pool can sometimes cause number to be
# less than GC_HEAP_INIT_SLOTS.
assert_operator(total_slots, :>=, GC_HEAP_INIT_SLOTS * 0.9, s)
assert_operator(total_slots, :>=, GC_HEAP_INIT_SLOTS, s)
end
RUBY
@ -462,10 +462,16 @@ class TestGc < Test::Unit::TestCase
assert_separately([env, "-W0"], __FILE__, __LINE__, <<~RUBY)
SIZES = #{sizes}
GC.stat_heap.each do |i, s|
# Sometimes pages will have 1 less slot due to alignment, so always increase slots_per_page by 1.
slots_per_page = (s[:heap_eden_slots] / s[:heap_eden_pages]) + 1
multiple = s[:slot_size] / (GC::INTERNAL_CONSTANTS[:BASE_SLOT_SIZE] + GC::INTERNAL_CONSTANTS[:RVALUE_OVERHEAD])
# Allocatable pages are assumed to have lost 1 slot due to alignment.
slots_per_page = (GC::INTERNAL_CONSTANTS[:HEAP_PAGE_OBJ_LIMIT] / multiple) - 1
total_slots = s[:heap_eden_slots] + s[:heap_allocatable_pages] * slots_per_page
assert_in_epsilon(SIZES[i], total_slots, 0.01, s)
# The delta is calculated as follows:
# - For allocated pages, each page can vary by 1 slot due to alignment.
# - For allocatable pages, we can end up with at most 1 extra page of slots.
assert_in_delta(SIZES[i], total_slots, s[:heap_eden_pages] + slots_per_page, s)
end
RUBY
@ -486,10 +492,16 @@ class TestGc < Test::Unit::TestCase
# Check that we still have the same number of slots as initially configured.
GC.stat_heap.each do |i, s|
# Sometimes pages will have 1 less slot due to alignment, so always increase slots_per_page by 1.
slots_per_page = (s[:heap_eden_slots] / s[:heap_eden_pages]) + 1
multiple = s[:slot_size] / (GC::INTERNAL_CONSTANTS[:BASE_SLOT_SIZE] + GC::INTERNAL_CONSTANTS[:RVALUE_OVERHEAD])
# Allocatable pages are assumed to have lost 1 slot due to alignment.
slots_per_page = (GC::INTERNAL_CONSTANTS[:HEAP_PAGE_OBJ_LIMIT] / multiple) - 1
total_slots = s[:heap_eden_slots] + s[:heap_allocatable_pages] * slots_per_page
assert_in_epsilon(SIZES[i], total_slots, 0.01, s)
# The delta is calculated as follows:
# - For allocated pages, each page can vary by 1 slot due to alignment.
# - For allocatable pages, we can end up with at most 1 extra page of slots.
assert_in_delta(SIZES[i], total_slots, s[:heap_eden_pages] + slots_per_page, s)
end
RUBY
@ -525,10 +537,16 @@ class TestGc < Test::Unit::TestCase
# Check that we still have the same number of slots as initially configured.
GC.stat_heap.each do |i, s|
# Sometimes pages will have 1 less slot due to alignment, so always increase slots_per_page by 1.
slots_per_page = (s[:heap_eden_slots] / s[:heap_eden_pages]) + 1
multiple = s[:slot_size] / (GC::INTERNAL_CONSTANTS[:BASE_SLOT_SIZE] + GC::INTERNAL_CONSTANTS[:RVALUE_OVERHEAD])
# Allocatable pages are assumed to have lost 1 slot due to alignment.
slots_per_page = (GC::INTERNAL_CONSTANTS[:HEAP_PAGE_OBJ_LIMIT] / multiple) - 1
total_slots = s[:heap_eden_slots] + s[:heap_allocatable_pages] * slots_per_page
assert_in_epsilon(SIZES[i], total_slots, 0.01, s)
# The delta is calculated as follows:
# - For allocated pages, each page can vary by 1 slot due to alignment.
# - For allocatable pages, we can end up with at most 1 extra page of slots.
assert_in_delta(SIZES[i], total_slots, s[:heap_eden_pages] + slots_per_page, s)
end
RUBY
end