RGBCube/serenity
1
Fork 0
mirror of https://github.com/RGBCube/serenity synced 2025-05-28 16:35:08 +00:00
Code Issues Activity Actions

Kernel: Fix partial munmap() deallocating still-in-use VM

Browse source 
We were always returning the full VM range of the partially-unmapped
Region to the range allocator. This caused us to re-use those addresses
for subsequent VM allocations.

This patch also skips creating a new VMObject in partial munmap().
Instead we just make split regions that point into the same VMObject.

This fixes the mysterious GCC ICE on large C++ programs.
This commit is contained in:
Andreas Kling 2019-09-27 20:17:41 +02:00
parent d5f3972012
commit 2584636d19
4 changed files with 17 additions and 12 deletions
Download patch file Download diff file Expand all files Collapse all files

20
Kernel/Process.cpp
View file

@ -108,9 +108,9 @@ static unsigned prot_to_region_access_flags(int prot)
return access;
}
Region& Process::allocate_split_region(const Region& source_region, const Range& range)
Region& Process::allocate_split_region(const Region& source_region, const Range& range, size_t offset_in_vmo)
{
m_regions.append(Region::create_user_accessible(range, source_region.name(), source_region.access()));
m_regions.append(Region::create_user_accessible(range, source_region.vmobject(), offset_in_vmo, source_region.name(), source_region.access()));
return m_regions.last();
}
@ -244,20 +244,24 @@ int Process::sys$munmap(void* addr, size_t size)
auto remaining_ranges_after_unmap = old_region_range.carve(range_to_unmap);
ASSERT(!remaining_ranges_after_unmap.is_empty());
auto make_replacement_region = [&](const Range& new_range) -> Region& {
auto& new_region = allocate_split_region(*old_region, new_range);
ASSERT(new_range.base() >= old_region_range.base());
ASSERT(new_range.end() <= old_region_range.end());
size_t new_range_offset_in_old_region = new_range.base().get() - old_region_range.base().get();
size_t first_physical_page_of_new_region_in_old_region = new_range_offset_in_old_region / PAGE_SIZE;
for (size_t i = 0; i < new_region.page_count(); ++i) {
new_region.vmobject().physical_pages()[i] = old_region->vmobject().physical_pages()[first_physical_page_of_new_region_in_old_region + i];
}
return new_region;
return allocate_split_region(*old_region, new_range, new_range_offset_in_old_region);
};
Vector<Region*, 2> new_regions;
for (auto& new_range : remaining_ranges_after_unmap) {
new_regions.unchecked_append(&make_replacement_region(new_range));
}
// We manually unmap the old region here, specifying that we *don't* want the VM deallocated.
MM.unmap_region(*old_region, false);
deallocate_region(*old_region);
// Instead we give back the unwanted VM manually.
page_directory().range_allocator().deallocate(range_to_unmap);
// And finally we map the new region(s).
for (auto* new_region : new_regions) {
MM.map_region(*this, *new_region);
}