Sinipelto
/
banan-os
forked from Bananymous/banan-os
1
0
Fork 0
Code Pull Requests Activity

Kernel: Rework physical memory allocation 

PhysicalRange is now much simpler bitmap. This makes expanding
PhysicalRange API much easier.
This commit is contained in:
Bananymous 2023-10-08 01:38:51 +03:00
parent f071240b33
commit 03d2bf4002
3 changed files with 85 additions and 128 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

34
kernel/include/kernel/Memory/PhysicalRange.h
View File

@ -3,7 +3,6 @@
#include <kernel/Memory/Types.h>
#include <stddef.h>
#include <stdint.h>
namespace Kernel
{
@ -12,42 +11,37 @@ namespace Kernel
{
public:
PhysicalRange(paddr_t, size_t);
paddr_t reserve_page();
void release_page(paddr_t);
paddr_t reserve_contiguous_pages(size_t pages);
void release_contiguous_pages(paddr_t paddr, size_t pages);
paddr_t start() const { return m_paddr; }
paddr_t end() const { return m_paddr + m_size; }
bool contains(paddr_t addr) const { return m_paddr <= addr && addr < m_paddr + m_size; }
size_t usable_memory() const { return m_reservable_pages * PAGE_SIZE; }
size_t usable_memory() const { return m_data_pages * PAGE_SIZE; }
size_t used_pages() const { return m_used_pages; }
size_t used_pages() const { return m_data_pages - m_free_pages; }
size_t free_pages() const { return m_free_pages; }
private:
struct node
{
node* next;
node* prev;
};
unsigned long long* ull_bitmap_ptr() { return (unsigned long long*)m_vaddr; }
paddr_t page_address(const node*) const;
node* node_address(paddr_t) const;
paddr_t paddr_for_bit(unsigned long long) const;
unsigned long long bit_for_paddr(paddr_t paddr) const;
private:
paddr_t m_paddr { 0 };
const paddr_t m_paddr { 0 };
const size_t m_size { 0 };
vaddr_t m_vaddr { 0 };
size_t m_size { 0 };
uint64_t m_total_pages { 0 };
uint64_t m_reservable_pages { 0 };
uint64_t m_list_pages { 0 };
size_t m_used_pages { 0 };
const size_t m_bitmap_pages { 0 };
const size_t m_data_pages { 0 };
size_t m_free_pages { 0 };
node* m_free_list { nullptr };
node* m_used_list { nullptr };
};
}

37
kernel/kernel/Memory/Heap.cpp
View File

@ -3,6 +3,8 @@
#include <kernel/Memory/PageTable.h>
#include <kernel/multiboot.h>
extern uint8_t g_kernel_end[];
namespace Kernel
{
@ -30,14 +32,22 @@ namespace Kernel
for (size_t i = 0; i < g_multiboot_info->mmap_length;)
{
multiboot_memory_map_t* mmmt = (multiboot_memory_map_t*)P2V(g_multiboot_info->mmap_addr + i);
if (mmmt->type == 1)
{
PhysicalRange range(mmmt->base_addr, mmmt->length);
if (range.usable_memory() > 0)
MUST(m_physical_ranges.push_back(range));
}
{
paddr_t start = mmmt->base_addr;
if (start < V2P(g_kernel_end))
start = V2P(g_kernel_end);
if (auto rem = start % PAGE_SIZE)
start += PAGE_SIZE - rem;
paddr_t end = mmmt->base_addr + mmmt->length;
if (auto rem = end % PAGE_SIZE)
end -= rem;
// Physical pages needs atleast 2 pages
if (end > start + PAGE_SIZE)
MUST(m_physical_ranges.emplace_back(start, end - start));
}
i += mmmt->size + sizeof(uint32_t);
}
@ -55,22 +65,17 @@ namespace Kernel
{
LockGuard _(m_lock);
for (auto& range : m_physical_ranges)
if (paddr_t page = range.reserve_page())
return page;
if (range.free_pages() >= 1)
return range.reserve_page();
return 0;
}
void Heap::release_page(paddr_t addr)
void Heap::release_page(paddr_t paddr)
{
LockGuard _(m_lock);
for (auto& range : m_physical_ranges)
{
if (range.contains(addr))
{
range.release_page(addr);
return;
}
}
if (range.contains(paddr))
return range.release_page(paddr);
ASSERT_NOT_REACHED();
}

142
kernel/kernel/Memory/PhysicalRange.cpp
View File

@ -3,123 +3,81 @@
#include <kernel/Memory/PageTable.h>
#include <kernel/Memory/PhysicalRange.h>
extern uint8_t g_kernel_end[];
namespace Kernel
{
PhysicalRange::PhysicalRange(paddr_t start, size_t size)
using ull = unsigned long long;
static constexpr ull ull_bits = sizeof(ull) * 8;
PhysicalRange::PhysicalRange(paddr_t paddr, size_t size)
: m_paddr(paddr)
, m_size(size)
, m_bitmap_pages(BAN::Math::div_round_up<size_t>(size / PAGE_SIZE, 8))
, m_data_pages((size / PAGE_SIZE) - m_bitmap_pages)
, m_free_pages(m_data_pages)
{
// We can't use the memory ovelapping with kernel
if (start + size <= V2P(g_kernel_end))
return;
ASSERT(paddr % PAGE_SIZE == 0);
ASSERT(size % PAGE_SIZE == 0);
ASSERT(m_bitmap_pages < size / PAGE_SIZE);
// Align start to page boundary and after the kernel memory
m_paddr = BAN::Math::max(start, V2P(g_kernel_end));
if (auto rem = m_paddr % PAGE_SIZE)
m_paddr += PAGE_SIZE - rem;
if (size <= m_paddr - start)
return;
// Align size to page boundary
m_size = size - (m_paddr - start);
if (auto rem = m_size % PAGE_SIZE)
m_size -= rem;
// We need atleast 2 pages
m_total_pages = m_size / PAGE_SIZE;
if (m_total_pages <= 1)
return;
// FIXME: if total pages is just over multiple of (PAGE_SIZE / sizeof(node)) we might make
// couple of pages unallocatable
m_list_pages = BAN::Math::div_round_up<uint64_t>(m_total_pages * sizeof(node), PAGE_SIZE);
m_reservable_pages = m_total_pages - m_list_pages;
m_used_pages = 0;
m_free_pages = m_reservable_pages;
m_vaddr = PageTable::kernel().reserve_free_contiguous_pages(m_list_pages, KERNEL_OFFSET);
m_vaddr = PageTable::kernel().reserve_free_contiguous_pages(m_bitmap_pages, KERNEL_OFFSET);
ASSERT(m_vaddr);
PageTable::kernel().map_range_at(m_paddr, m_vaddr, m_list_pages * PAGE_SIZE, PageTable::Flags::ReadWrite | PageTable::Flags::Present);
PageTable::kernel().map_range_at(m_paddr, m_vaddr, size, PageTable::Flags::ReadWrite | PageTable::Flags::Present);
// Initialize page list so that every page points to the next one
node* page_list = (node*)m_vaddr;
memset((void*)m_vaddr, 0x00, m_bitmap_pages * PAGE_SIZE);
memset((void*)m_vaddr, 0xFF, m_data_pages / 8);
for (int i = 0; i < m_data_pages % 8; i++)
((uint8_t*)m_vaddr)[m_data_pages / 8] |= 1 << i;
for (uint64_t i = 0; i < m_reservable_pages; i++)
page_list[i] = { page_list + i - 1, page_list + i + 1 };
page_list[ 0 ].next = nullptr;
page_list[m_reservable_pages - 1].prev = nullptr;
dprintln("physical range needs {} pages for bitmap", m_bitmap_pages);
}
m_free_list = page_list;
m_used_list = nullptr;
paddr_t PhysicalRange::paddr_for_bit(ull bit) const
{
return m_paddr + (m_bitmap_pages + bit) * PAGE_SIZE;
}
ull PhysicalRange::bit_for_paddr(paddr_t paddr) const
{
return (paddr - m_paddr) / PAGE_SIZE - m_bitmap_pages;
}
paddr_t PhysicalRange::reserve_page()
{
if (m_free_list == nullptr)
return 0;
ASSERT(free_pages() > 0);
node* page = m_free_list;
ASSERT(page->next == nullptr);
ull ull_count = BAN::Math::div_round_up<ull>(m_data_pages, ull_bits);
// Detatch page from top of the free list
m_free_list = m_free_list->prev;
if (m_free_list)
m_free_list->next = nullptr;
for (ull i = 0; i < ull_count; i++)
{
if (ull_bitmap_ptr()[i] == 0)
continue;
// Add page to used list
if (m_used_list)
m_used_list->next = page;
page->prev = m_used_list;
m_used_list = page;
int lsb = __builtin_ctzll(ull_bitmap_ptr()[i]);
m_used_pages++;
m_free_pages--;
ull_bitmap_ptr()[i] &= ~(1ull << lsb);
m_free_pages--;
return paddr_for_bit(i * ull_bits + lsb);
}
return page_address(page);
ASSERT_NOT_REACHED();
}
void PhysicalRange::release_page(paddr_t page_address)
void PhysicalRange::release_page(paddr_t paddr)
{
ASSERT(m_used_list);
ASSERT(paddr % PAGE_SIZE == 0);
ASSERT(paddr - m_paddr <= m_size);
node* page = node_address(page_address);
// Detach page from used list
if (page->prev)
page->prev->next = page->next;
if (page->next)
page->next->prev = page->prev;
if (m_used_list == page)
m_used_list = page->prev;
ull full_bit = bit_for_paddr(paddr);
ull off = full_bit / ull_bits;
ull bit = full_bit % ull_bits;
ull mask = 1ull << bit;
// Add page to the top of free list
page->prev = m_free_list;
page->next = nullptr;
if (m_free_list)
m_free_list->next = page;
m_free_list = page;
ASSERT(!(ull_bitmap_ptr()[off] & mask));
ull_bitmap_ptr()[off] |= mask;
m_used_pages--;
m_free_pages++;
}
paddr_t PhysicalRange::page_address(const node* page) const
{
ASSERT((vaddr_t)page <= m_vaddr + m_reservable_pages * sizeof(node));
uint64_t page_index = page - (node*)m_vaddr;
return m_paddr + (page_index + m_list_pages) * PAGE_SIZE;
}
PhysicalRange::node* PhysicalRange::node_address(paddr_t page_address) const
{
ASSERT(page_address % PAGE_SIZE == 0);
ASSERT(m_paddr + m_list_pages * PAGE_SIZE <= page_address && page_address < m_paddr + m_size);
uint64_t page_offset = page_address - (m_paddr + m_list_pages * PAGE_SIZE);
return (node*)m_vaddr + page_offset / PAGE_SIZE;
}
}