Small kernel heap for VMM internals, kmalloc/kfree

Makefile

@@ -1,4 +1,4 @@
-SOURCES = src/mem/paging/paging.c src/mem/paging/pmm.c src/string/string.c src/io/kbd/ps2.c src/io/serial/serial.c src/io/term/printf.c src/io/term/term.c src/idt/idt.c src/mem/gdt/gdt.c src/mem/misc/utils.c src/time/timer.c src/kmain.c
+SOURCES = src/mem/heap/kheap.c src/mem/paging/vmm.c src/mem/paging/paging.c src/mem/paging/pmm.c src/string/string.c src/io/kbd/ps2.c src/io/serial/serial.c src/io/term/printf.c src/io/term/term.c src/idt/idt.c src/mem/gdt/gdt.c src/mem/misc/utils.c src/time/timer.c src/kmain.c
 
 build:
 	rm -f *.o
@@ -30,7 +30,7 @@ build-iso: limine/limine build
 	./limine/limine bios-install pepper.iso
 
 debug:
-	qemu-system-x86_64 -drive file=pepper.iso -s -S -d int -no-reboot &
+	qemu-system-x86_64 -drive file=pepper.iso -s -S -d int -no-reboot -no-shutdown &
 	gdb pepperk --command=debug.gdb
 
 run: build-iso

src/kernel.h

@@ -16,4 +16,8 @@ enum ErrorCodes
 #define DEBUG(log, ...) fctprintf((void*)&skputc, 0, "debug: [%s]: " log "\r\n", __FILE__, ##__VA_ARGS__)
 
 // printf("debug: [%s]: " log "\n", __FILE__, ##__VA_ARGS__);
+
+void hcf();
+#define assert(check) do { if(!(check)) hcf(); } while(0)
+
 #endif

src/kmain.c

@@ -12,6 +12,8 @@
 #include "io/kbd/ps2.h"
 #include "mem/paging/pmm.h"
 #include "mem/paging/paging.h"
+#include "mem/paging/vmm.h"
+#include "mem/heap/kheap.h"
 
 // Limine version used
 __attribute__((used, section(".limine_requests")))
@@ -53,8 +55,8 @@ static volatile LIMINE_REQUESTS_END_MARKER;
 
 struct limine_framebuffer* framebuffer;
 
-// Panic 
+// Panic (should dump registers etc. in the future)
-static void hcf()
+void hcf()
 {
 	for (;;)
 	{
@@ -92,8 +94,19 @@ void kmain()
 	SET_INTERRUPTS;
 
 	pmm_init(memmap_request.response, hhdm_request.response);
+
+	// Remap kernel , HHDM and framebuffer
 	paging_init(kerneladdr_request.response, framebuffer);
 	
+	kheap_init();
+
+	void* ptr = kmalloc(10); DEBUG("(KMALLOC TEST) Allocated 10 bytes at 0x%p", ptr);
+	void* ptr2 = kmalloc(200); DEBUG("(KMALLOC TEST) Allocated 200 bytes at 0x%p", ptr2);
+	kfree(ptr);
+	void* ptr3 = kmalloc(5); DEBUG("(KMALLOC TEST) Allocated 5 bytes at 0x%p", ptr3);
+
+	vmm_init();
+
 	keyboard_init(FR);
 
 	term_init();

src/mem/heap/kheap.c

@@ -0,0 +1,106 @@
+#include "kheap.h"
+#include "mem/paging/paging.h"
+#include "mem/paging/pmm.h"
+#include <stddef.h>
+#include <kernel.h>
+
+extern uint64_t kernel_phys_base;
+extern uint64_t kernel_virt_base;
+
+uintptr_t kheap_start;
+
+static struct heap_block_t* head = NULL;
+static uintptr_t end;
+
+// Kernel root table (level 4)
+extern uint64_t *kernel_pml4;
+
+static void kheap_map_page()
+{
+    uintptr_t phys = pmm_alloc();
+    paging_map_page(kernel_pml4, end, phys, PTE_PRESENT | PTE_WRITABLE | PTE_NOEXEC);
+    end += PAGE_SIZE;
+    DEBUG("Mapped first kheap page");
+}
+
+void kheap_init()
+{
+    kheap_start = ALIGN_UP(kernel_virt_base + KERNEL_SIZE, PAGE_SIZE);
+    end = kheap_start;
+
+    // At least 1 page must be mapped for it to work
+    kheap_map_page();
+
+    // Give linked list head its properties
+    head = (struct heap_block_t*)kheap_start;
+    head->size = PAGE_SIZE - sizeof(struct heap_block_t);
+    head->free = true;
+    head->next = NULL;
+    DEBUG("kheap initialized, head=0x%p, size=%u", head, head->size);
+}
+
+void* kmalloc(size_t size)
+{
+    // No size, no memory allocated!
+    if (!size) return NULL;
+
+    struct heap_block_t* curr = head;
+
+    while (curr)
+    {
+        // Is block free and big enough for us?
+        if (curr->free && curr->size >= size)
+        {
+            // We split the block if it is big enough
+            if (curr->size > size + sizeof(struct heap_block_t))
+            {
+                struct heap_block_t* new_block = (struct heap_block_t*)((uintptr_t)curr + sizeof(struct heap_block_t) + size);
+                // We have to subtract the size of our block struct
+                new_block->size = curr->size - size - sizeof(struct heap_block_t);
+                new_block->free = true;
+
+                // Then we chain up the block in the list
+                new_block->next = curr->next;
+                curr->next = new_block;
+
+                curr->size = size;
+            }
+
+            // Found a good block, we return it
+            curr->free = false;
+            return (void*)((uintptr_t)curr + sizeof(struct heap_block_t));
+        }
+        // Continue browsing the list if nothing good was found yet
+        curr = curr->next;
+    }
+
+    // If we're hear it means we didn't have enough memory
+    // for the block allocation. So we will allocate more..
+    uintptr_t old_end = end;
+    kheap_map_page();
+
+    struct heap_block_t* block = (struct heap_block_t*)old_end;
+    block->size = PAGE_SIZE - sizeof(struct heap_block_t);
+    block->free = false;
+    block->next = NULL;
+
+    // Put the block at the end of the list
+    curr = head;
+    while (curr->next)
+    {
+        curr = curr->next;
+    }
+    curr->next = block;
+
+    return (void*)((uintptr_t)block + sizeof(struct heap_block_t));
+}
+
+void kfree(void* ptr)
+{
+    // Nothing to free
+    if (!ptr) return;
+
+    // Set it free!
+    struct heap_block_t* block = (struct heap_block_t*)((uintptr_t)ptr - sizeof(struct heap_block_t));
+    block->free = true;
+}

src/mem/heap/kheap.h

@@ -0,0 +1,26 @@
+#ifndef KHEAP_H
+#define KHEAP_H
+
+// We need some kind of simple kernel heap to make our linked list
+// for the VMM, as we need "malloc" and "free" for that data structure.
+// When the kernel heap is ready, we can alloc our VM object linked list
+// and then continue working on the VMM.
+
+// 16MB should be enough for some linked lists
+#define KHEAP_SIZE (16*1024*1024)
+
+#include <stdbool.h>
+#include <stddef.h>
+
+struct heap_block_t
+{
+    size_t size;
+    bool free;
+    struct heap_block_t* next;
+};
+
+void kheap_init();
+void* kmalloc(size_t size);
+void kfree(void* ptr);
+
+#endif

src/mem/paging/paging.c

@@ -39,10 +39,14 @@ static uint64_t* alloc_page_table()
     return virt;
 }
 
+// Kernel paging root table, that will be placed in cr3
 __attribute__((aligned(4096)))
-static uint64_t *kernel_pml4;
+uint64_t *kernel_pml4;
 
-void map_page(uint64_t virt, uint64_t phys, uint64_t flags)
+// Will map a page ONLY according to the kernel_pml4 root table.
+// For kernel initialization/mapping only
+// Deprecated, will be removed
+/* void paging_kmap_page(uint64_t virt, uint64_t phys, uint64_t flags)
 {
     virt = PAGE_ALIGN_DOWN(virt);
     phys = PAGE_ALIGN_DOWN(phys);
@@ -92,35 +96,85 @@ void map_page(uint64_t virt, uint64_t phys, uint64_t flags)
 
     // Flush TLB (apply changes)
     invlpg((void *)virt);
+} */
+
+// Same as above, only this one takes any root table (not only kernel)
+// Duplicate code but don't worry about it, I'll refactor one day
+void paging_map_page(uint64_t* root_table, uint64_t virt, uint64_t phys, uint64_t flags)
+{
+    virt = PAGE_ALIGN_DOWN(virt);
+    phys = PAGE_ALIGN_DOWN(phys);
+
+    // Translate the virt address into page table indexes
+    uint64_t pml4_i = PML4_INDEX(virt);
+    uint64_t pdpt_i = PDPT_INDEX(virt);
+    uint64_t pd_i = PD_INDEX(virt);
+    uint64_t pt_i = PT_INDEX(virt);
+
+    uint64_t *pdpt, *pd, *pt;
+
+    // PML4
+    // If the entry at index is not present, allocate enough space for it
+    // then populate the entry with correct addr + flags
+    if (!(root_table[pml4_i] & PTE_PRESENT))
+    {
+        pdpt = alloc_page_table();
+        root_table[pml4_i] = VIRT_TO_PHYS(pdpt) | PTE_PRESENT | PTE_WRITABLE;
     }
+    else {
+        pdpt = (uint64_t *)PHYS_TO_VIRT(root_table[pml4_i] & ~0xFFFULL);
+    }
+
+    // PDPT: same here
+    if (!(pdpt[pdpt_i] & PTE_PRESENT))
+    {
+        pd = alloc_page_table();
+        pdpt[pdpt_i] = VIRT_TO_PHYS(pd) | PTE_PRESENT | PTE_WRITABLE;
+    } 
+    else {
+        pd = (uint64_t *)PHYS_TO_VIRT(pdpt[pdpt_i] & ~0xFFFULL);
+    }
+
+    // PD: and here
+    if (!(pd[pd_i] & PTE_PRESENT))
+    {
+        pt = alloc_page_table();
+        pd[pd_i] = VIRT_TO_PHYS(pt) | PTE_PRESENT | PTE_WRITABLE;
+    }
+    else {
+        pt = (uint64_t *)PHYS_TO_VIRT(pd[pd_i] & ~0xFFFULL);
+    }
+
+    // PT: finally, populate the page table entry
+    pt[pt_i] = phys | flags | PTE_PRESENT;
+
+    // Flush TLB (apply changes)
+    invlpg((void *)virt);
+}
+
+uint64_t kernel_phys_base;
+uint64_t kernel_virt_base;
 
 void paging_init(struct limine_kernel_address_response* kaddr, struct limine_framebuffer* fb)
 {
     // We should map the kernel, GDT, IDT, stack, framebuffer.
     // Optionally we could map ACPI tables (we can find them in the Limine memmap)
 
-    uint64_t kernel_phys_base = kaddr->physical_base;
+    kernel_phys_base = kaddr->physical_base;
-    uint64_t kernel_virt_base = kaddr->virtual_base;
+    kernel_virt_base = kaddr->virtual_base;
+
+    DEBUG("Kernel lives at virt=0x%p phys=0x%p", kernel_virt_base, kernel_phys_base);
 
     kernel_pml4 = alloc_page_table();
 
     // for debug
     uint64_t page_count = 0;
 
-    // First 16 MB identity-mapped (phys = virt)
-    // This is because there might be some leftover stuff in the lower phys addresses
-    // from boot/bios/acpi/...
-    for (uint64_t i=0; i<0x1000000; i += PAGE_SIZE)
-    {
-        map_page(i, i, PTE_WRITABLE);
-        page_count++;
-    }
-    DEBUG("Mapped %u pages for the identity-mapping of the first 16 MB", page_count); page_count = 0;
-
     // HHDM map first 1 GB using given offset
     for (uint64_t i=0; i<0x40000000; i += PAGE_SIZE)
     {
-        map_page(i+hhdm_off, i, PTE_WRITABLE);
+        //paging_kmap_page(i+hhdm_off, i, PTE_WRITABLE);
+        paging_map_page(kernel_pml4, i+hhdm_off, i, PTE_WRITABLE);
         page_count++;
     }
     DEBUG("Mapped %u pages for first 1GB (HHDM)", page_count); page_count = 0;
@@ -131,7 +185,8 @@ void paging_init(struct limine_kernel_address_response* kaddr, struct limine_fra
     // For now who gives a shit, let's RWX all kernel
     for (uint64_t i = 0; i < KERNEL_SIZE; i += PAGE_SIZE)
     {
-        map_page(kernel_virt_base+i, kernel_phys_base+i, PTE_WRITABLE);
+        //paging_kmap_page(kernel_virt_base+i, kernel_phys_base+i, PTE_WRITABLE);
+        paging_map_page(kernel_pml4, kernel_virt_base+i, kernel_phys_base+i, PTE_WRITABLE);
         page_count++;
     }
     DEBUG("Mapped %u pages for kernel", page_count); page_count = 0;
@@ -145,7 +200,8 @@ void paging_init(struct limine_kernel_address_response* kaddr, struct limine_fra
     // Map the framebuffer (with cache-disable & write-through)
     for (uint64_t i=0; i<fb_pages; i++)
     {
-        map_page(fb_virt+i*PAGE_SIZE, fb_phys+i*PAGE_SIZE, PTE_WRITABLE | PTE_PCD | PTE_PWT);
+        //paging_kmap_page(fb_virt+i*PAGE_SIZE, fb_phys+i*PAGE_SIZE, PTE_WRITABLE | PTE_PCD | PTE_PWT);
+        paging_map_page(kernel_pml4, fb_virt+i*PAGE_SIZE, fb_phys+i*PAGE_SIZE, PTE_WRITABLE | PTE_PCD | PTE_PWT);
         page_count++;
     }
     DEBUG("Mapped %u pages for framebuffer", page_count);

src/mem/paging/paging.h

@@ -8,6 +8,7 @@
 #include <limine.h>
 
 void paging_init(struct limine_kernel_address_response* kaddr, struct limine_framebuffer* fb);
+void paging_map_page(uint64_t* root_table, uint64_t virt, uint64_t phys, uint64_t flags);
 
 extern uint64_t hhdm_off;
 

src/mem/paging/pmm.c

@@ -25,7 +25,6 @@ We will look for the biggest usable physical memory region
 and use this for the bitmap. The reserved memory will be ignored.
 */
 
-struct usable_memory* usable_mem;
 struct limine_memmap_entry* biggest_entry;
 
 static void pmm_find_biggest_usable_region(struct limine_memmap_response* memmap, struct limine_hhdm_response* hhdm)

src/mem/paging/pmm.h

@@ -7,12 +7,4 @@ void pmm_init(struct limine_memmap_response* memmap, struct limine_hhdm_response
 void pmm_free(uintptr_t addr);
 uintptr_t pmm_alloc();
 
-// Might be upgraded to a freelist later.
-// For now, we can take the biggest usable region and we will be fine.
-struct usable_memory
-{
-    uint64_t base; // physical
-    uint64_t length;
-};
-
 #endif

src/mem/paging/vmm.c

@@ -0,0 +1,66 @@
+/*
+The VMM (virtual memory manager) will have two roles:
+- mapping pages
+- unmapping pages
+in a specified virtual space
+
+compared to the PMM which allocs/frees 4kb frames ("physical pages").
+*/
+
+#include "vmm.h"
+#include "paging.h"
+#include <stddef.h>
+#include "pmm.h"
+#include <kernel.h>
+
+void* vmm_pt_root = 0;
+
+// Linked list head for virtual memory objects
+struct vm_object* vm_objs = NULL;
+
+
+uint64_t convert_x86_vm_flags(size_t flags)
+{
+    uint64_t value = 0;
+    if (flags & VM_FLAG_WRITE)
+    {
+        value |= PTE_WRITABLE;
+    }
+    if (flags & VM_FLAG_USER)
+    {
+        value |= PTE_USER;
+    }
+    if ((flags & VM_FLAG_EXEC) == 0)
+    {
+        value |= PTE_NOEXEC;
+    }
+    return value;
+}
+
+extern uint64_t *kernel_pml4;
+
+void vmm_setup_pt_root()
+{
+    // We alloc a physical page (frame) for the pointer, then map it
+    // to virt (pointer)
+    uintptr_t phys = pmm_alloc();
+    vmm_pt_root = (void*)kernel_pml4;
+    paging_map_page(kernel_pml4, (uint64_t)vmm_pt_root, phys, convert_x86_vm_flags(VM_FLAG_WRITE | VM_FLAG_EXEC));
+    DEBUG("VMM setup: vmm_pt_root=0x%p (phys=0x%p)", vmm_pt_root, phys);
+}
+
+void* vmm_alloc(size_t length, size_t flags)
+{
+    // We will try to allocate at least length bytes, which have to be rounded UP to
+    // the next page so its coherent with the PMM
+    size_t len = ALIGN_UP(length, PAGE_SIZE);
+
+    // Some linked list shenanigans will be here
+    // but for now we'd need some kheap to kmalloc the linked list items
+    // else we can't do it
+}
+
+void vmm_init()
+{
+    vmm_setup_pt_root();
+}

src/mem/paging/vmm.h

@@ -0,0 +1,29 @@
+#ifndef VMM_H
+#define VMM_H
+
+#include <stdint.h>
+#include <stddef.h>
+
+/*
+This will be our linked list of virtual memory objects.
+Flags here aren't x86 flags, they are platform-agnostic
+kernel-defined flags.
+*/
+
+struct vm_object
+{
+    uintptr_t base;
+    size_t length;
+    size_t flags;
+    struct vm_object* next;
+};
+
+// Flags bitfield
+#define VM_FLAG_NONE 0
+#define VM_FLAG_WRITE (1 << 0)
+#define VM_FLAG_EXEC (1 << 1)
+#define VM_FLAG_USER (1 << 2)
+
+void vmm_init();
+
+#endif