x86_64: implement bootstrap function for pmap interface

arch/x86_64/include/arch/paging.h

@@ -0,0 +1,54 @@
+#ifndef ARCH_PAGING_H_
+#define ARCH_PAGING_H_
+
+#include <socks/types.h>
+#include <socks/compiler.h>
+
+#define PTE_PRESENT      0x01ULL
+#define PTE_RW           0x02ULL
+#define PTE_USR          0x04ULL
+#define PTE_WRITETHROUGH 0x08ULL
+#define PTE_NOCACHE      0x10ULL
+#define PTE_PAGESIZE     0x80ULL
+#define PTE_NX           0x8000000000000000ULL
+
+typedef phys_addr_t pml4t_ptr_t;
+typedef uint64_t pte_t;
+
+typedef struct pml4t {
+	phys_addr_t p_entries[512];
+} __packed pml4t_t;
+
+typedef struct pdpt {
+	union {
+		/* 4KiB and 2MiB pages */
+		phys_addr_t p_entries[512];
+		/* 1GiB pages */
+		pte_t p_pages[512];
+	};
+} __packed pdpt_t;
+
+typedef struct pdir {
+	union {
+		/* 4KiB pages */
+		phys_addr_t p_entries[512];
+		/* 2MiB pages */
+		pte_t p_pages[512];
+	};
+} __packed pdir_t;
+
+typedef struct ptab {
+	pte_t p_pages[512];
+} __packed ptab_t;
+
+typedef enum page_size {
+	PS_4K,
+	PS_2M,
+	PS_1G,
+} page_size_t;
+
+/* returns 1 if gigabyte pages are supported by the CPU, 0 otherwise.
+   defined in pmap_ctrl.S */
+extern int gigabyte_pages(void);
+
+#endif

arch/x86_64/include/socks/machine/pmap.h

@@ -0,0 +1,9 @@
+#ifndef SOCKS_X86_64_PMAP_H_
+#define SOCKS_X86_64_PMAP_H_
+
+#include <arch/paging.h>
+
+typedef pml4t_ptr_t ml_pmap_t;
+typedef uint64_t ml_pfn_t;
+
+#endif

arch/x86_64/init.c

@@ -1,4 +1,5 @@
-#include "socks/types.h"
+#include <socks/pmap.h>
+#include <socks/types.h>
 #include <arch/e820.h>
 #include <socks/init.h>
 #include <socks/memblock.h>
@@ -47,5 +48,7 @@ int ml_init(uintptr_t arg)
 
 	e820_scan(PTR32(mb->mmap_addr), mb->mmap_length);
 
+	pmap_bootstrap();
+
 	return 0;
 }

arch/x86_64/pmap.c

@@ -0,0 +1,280 @@
+#include "socks/types.h"
+#include <socks/memblock.h>
+#include <socks/vm.h>
+#include <socks/printk.h>
+#include <socks/status.h>
+#include <socks/compiler.h>
+#include <socks/pmap.h>
+
+/* some helpful datasize constants */
+#define C_1GiB 0x40000000ULL
+#define C_2GiB (2 * C_1GiB)
+
+#define BAD_INDEX ((unsigned int)-1)
+#define PTR_TO_ENTRY(x) (((x) & ~VM_PAGE_MASK) | PTE_PRESENT | PTE_RW)
+#define ENTRY_TO_PTR(x) ((x) & ~VM_PAGE_MASK)
+
+#define PFN(x) ((x) >> VM_PAGE_SHIFT)
+
+static int can_use_gbpages = 0;
+static pmap_t kernel_pmap;
+
+static size_t ps_size(page_size_t ps)
+{
+	switch (ps) {
+	case PS_4K:
+		return 0x1000;
+	case PS_2M:
+		return 0x200000;
+	case PS_1G:
+		return 0x40000000;
+	default:
+		return 0;
+	}
+}
+
+static pmap_t alloc_pmap(pml4t_t **pt)
+{
+	pml4t_t *p = kzalloc(sizeof *p, 0);
+	*pt = p;
+	return vm_virt_to_phys(p);
+}
+
+static pte_t make_pte(pfn_t pfn, vm_prot_t prot, page_size_t size)
+{
+	pte_t v = pfn;
+
+	switch (size) {
+	case PS_1G:
+		/* pfn_t is in terms of 4KiB pages, convert to 1GiB page frame number */
+		pfn >>= 18;
+		v = (pfn & 0x3FFFFF) << 30;
+		break;
+	case PS_2M:
+		/* pfn_t is in terms of 4KiB pages, convert to 2MiB page frame number */
+		pfn >>= 9;
+		v = (pfn & 0x7FFFFFFF) << 21;
+		break;
+	case PS_4K:
+		v = (pfn & 0xFFFFFFFFFF) << 12;
+		break;
+	default:
+		return 0;
+	}
+
+	v |= PTE_PRESENT;
+
+	if (size != PS_4K) {
+		v |= PTE_PAGESIZE;
+	}
+
+	if (prot & VM_PROT_WRITE) {
+		v |= PTE_RW;
+	}
+
+	if (!(prot & VM_PROT_EXEC)) {
+		v |= PTE_RW;
+	}
+
+	if ((prot & VM_PROT_USER) && !(prot & VM_PROT_SVR)) {
+		v |= PTE_USR;
+	}
+
+	return v;
+}
+
+static void delete_ptab(phys_addr_t pt)
+{
+	if (pt & PTE_PAGESIZE) {
+		/* this entry points to a hugepage, nothing to delete */
+		return;
+	}
+
+	pt &= ~VM_PAGE_MASK;
+	if (!pt) {
+		return;
+	}
+	
+	ptab_t *ptab = vm_phys_to_virt(pt);
+	kfree(ptab);
+}
+
+static void delete_pdir(phys_addr_t pd)
+{
+	if (pd & PTE_PAGESIZE) {
+		/* this entry points to a hugepage, nothing to delete */
+		return;
+	}
+
+	pd &= ~0x1FFFFFULL;
+		
+	pdir_t *pdir = vm_phys_to_virt(pd);
+	for (int i = 0; i < 512; i++) {
+		if (pdir->p_pages[i] & PTE_PAGESIZE) {
+			/* this is a hugepage, there is nothing to delete */
+			continue;
+		} 
+
+		delete_ptab(pdir->p_entries[i]);
+	}
+
+	kfree(pdir);
+}
+
+static kern_status_t do_pmap_add(pmap_t pmap, void *p, pfn_t pfn, vm_prot_t prot, page_size_t size)
+{
+	uintptr_t pv = (uintptr_t)p;
+	unsigned int
+		pml4t_index = BAD_INDEX,
+		pdpt_index = BAD_INDEX,
+		pd_index = BAD_INDEX,
+		pt_index = BAD_INDEX;
+
+	switch (size) {
+	case PS_4K:
+		pml4t_index = (pv >> 39) & 0x1FF;
+		pdpt_index = (pv >> 30) & 0x1FF;
+		pd_index = (pv >> 21) & 0x1FF;
+		pt_index = (pv >> 12) & 0x1FF;
+		break;
+	case PS_2M:
+		pml4t_index = (pv >> 39) & 0x1FF;
+		pdpt_index = (pv >> 30) & 0x1FF;
+		pd_index = (pv >> 21) & 0x1FF;
+		break;
+	case PS_1G:
+		if (!can_use_gbpages) {
+			return KERN_UNSUPPORTED;
+		}
+		pml4t_index = (pv >> 39) & 0x1FF;
+		pdpt_index = (pv >> 30) & 0x1FF;
+		break;
+	default:
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	/* 1. get PML4T (mandatory) */
+	pml4t_t *pml4t = vm_phys_to_virt(ENTRY_TO_PTR(pmap));
+	if (!pml4t) {
+		return KERN_INVALID_ARGUMENT;
+	}
+
+	/* 2. traverse PML4T, get PDPT (mandatory) */
+	pdpt_t *pdpt = NULL;
+	if (!pml4t->p_entries[pml4t_index]) {
+		pdpt = kzalloc(sizeof *pdpt, 0);
+		pml4t->p_entries[pml4t_index] = PTR_TO_ENTRY(vm_virt_to_phys(pdpt));
+	} else {
+		pdpt = vm_phys_to_virt(ENTRY_TO_PTR(pml4t->p_entries[pml4t_index]));
+	}
+
+	/* if we're mapping a 1GiB page, we stop here */
+	if (size == PS_1G) {
+		if (pdpt->p_entries[pdpt_index] != 0) {
+			/* this slot points to a ptab, delete it.
+			   if this slot points to a hugepage, this does nothing */
+			delete_pdir(pdpt->p_entries[pdpt_index]);
+		}
+		
+		pdpt->p_pages[pdpt_index] = make_pte(pfn, prot, size);
+
+		return KERN_OK;
+	}
+		
+		
+	/* 3. traverse PDPT, get PDIR (optional, 4K and 2M only) */
+	pdir_t *pdir = NULL;
+	if (!pdpt->p_entries[pdpt_index] || pdpt->p_pages[pdpt_index] & PTE_PAGESIZE) {
+		/* entry is null, or points to a hugepage */
+		pdir = kzalloc(sizeof *pdir, 0);
+		pdpt->p_entries[pdpt_index] = PTR_TO_ENTRY(vm_virt_to_phys(pdir));
+	} else {
+		pdir = vm_phys_to_virt(ENTRY_TO_PTR(pdpt->p_entries[pdpt_index]));
+	}
+
+	/* if we're ampping a 2MiB page, we stop here */
+	if (size == PS_2M) {
+		if (pdir->p_entries[pd_index] != 0) {
+			/* this slot points to a ptab, delete it.
+			   if this slot points to a hugepage, this does nothing */
+			delete_ptab(pdir->p_entries[pd_index]);
+		}
+
+		pdir->p_pages[pd_index] = make_pte(pfn, prot, size);
+		return KERN_OK;
+	}
+
+	/* 4. traverse PDIR, get PTAB (optional, 4K only) */
+	ptab_t *ptab = NULL;
+	if (!pdir->p_entries[pd_index] || pdir->p_pages[pd_index] & PTE_PAGESIZE) {
+		/* entry is null, or points to a hugepage */
+		ptab = kzalloc(sizeof *ptab, 0);
+		pdir->p_entries[pd_index] = PTR_TO_ENTRY(vm_virt_to_phys(ptab));
+	} else {
+		ptab = vm_phys_to_virt(ENTRY_TO_PTR(pdir->p_entries[pd_index]));
+	}
+
+	ptab->p_pages[pt_index] = make_pte(pfn, prot, size);
+	return KERN_OK;
+}
+
+void pmap_bootstrap(void)
+{
+	can_use_gbpages = gigabyte_pages();
+	printk("pmap: gigabyte pages %sabled", can_use_gbpages == 1 ? "en" : "dis");
+
+	page_size_t hugepage = PS_2M;
+	if (can_use_gbpages) {
+		hugepage = PS_1G;
+	}
+	size_t hugepage_sz = ps_size(hugepage);
+
+	pml4t_t *kernel_pml4t;
+	kernel_pmap = alloc_pmap(&kernel_pml4t);
+
+	//do_pmap_add(kernel_pmap, NULL, 0, VM_PROT_READ, PS_2M);
+
+	/* map 2GiB at the end of the address space to cover
+	   the kernel and memblock-allocated data */
+	uintptr_t vbase = VM_KERNEL_VOFFSET;
+	for (size_t i = 0; i < C_2GiB; i += hugepage_sz) {
+		do_pmap_add(kernel_pmap,
+			    (void *)(vbase + i),
+			    PFN(i),
+			    VM_PROT_READ | VM_PROT_WRITE | VM_PROT_EXEC | VM_PROT_SVR,
+			    hugepage);
+	}
+
+	pmap_switch(kernel_pmap);
+	printk("pmap: kernel pmap initialised");
+}
+
+pmap_t pmap_create(void)
+{
+	return 0;
+}
+
+void pmap_destroy(pmap_t pmap)
+{
+
+}
+
+kern_status_t pmap_add(pmap_t pmap, void *p, pfn_t pfn, vm_prot_t prot, pmap_flags_t flags)
+{
+	return KERN_OK;
+}
+
+kern_status_t pmap_add_block(pmap_t pmap, void *p, pfn_t pfn, size_t len, vm_prot_t prot, pmap_flags_t flags)
+{
+	return KERN_OK;
+}
+
+kern_status_t pmap_remove(pmap_t pmap, void *p)
+{
+	return KERN_OK;
+}
+
+kern_status_t pmap_remove_range(pmap_t pmap, void *p, size_t len)
+{
+	return KERN_OK;
+}

arch/x86_64/pmap_ctrl.S

@@ -0,0 +1,31 @@
+
+	.global pmap_switch
+	.type pmap_switch, @function
+
+pmap_switch:
+	mov %rdi, %cr3
+	ret
+
+	.global gigabyte_pages
+	.type gigabyte_pages, @function
+
+gigabyte_pages:
+	push %rbp
+	mov %rsp, %rbp
+
+	push %rbx
+
+	mov $0x80000001, %eax
+	cpuid
+	andl $0x4000000, %edx
+
+	jz 2f
+
+1:	mov $0x1, %rax
+	jmp 3f
+
+2:	mov $0x0, %rax	
+
+3:	pop %rbx
+	pop %rbp
+	ret

include/socks/pmap.h

@@ -0,0 +1,32 @@
+#ifndef SOCKS_PMAP_H_
+#define SOCKS_PMAP_H_
+
+/* all the functions declared in this file are defined in arch/xyz/pmap.c */
+
+#include <socks/vm.h>
+#include <socks/status.h>
+#include <socks/machine/pmap.h>
+#include <stddef.h>
+
+#define PFN(x) ((x) >> VM_PAGE_SHIFT)
+
+typedef ml_pmap_t pmap_t;
+typedef ml_pfn_t pfn_t;
+
+typedef enum pmap_flags {
+	PMAP_HUGEPAGE = 0x01u,
+} pmap_flags_t;
+
+extern void pmap_bootstrap(void);
+
+extern pmap_t pmap_create(void);
+extern void pmap_destroy(pmap_t pmap);
+extern void pmap_switch(pmap_t pmap);
+
+extern kern_status_t pmap_add(pmap_t pmap, void *p, pfn_t pfn, vm_prot_t prot, pmap_flags_t flags);
+extern kern_status_t pmap_add_block(pmap_t pmap, void *p, pfn_t pfn, size_t len, vm_prot_t prot, pmap_flags_t flags);
+
+extern kern_status_t pmap_remove(pmap_t pmap, void *p);
+extern kern_status_t pmap_remove_range(pmap_t pmap, void *p, size_t len);
+
+#endif

include/socks/status.h

@@ -3,8 +3,10 @@
 
 typedef unsigned int kern_status_t;
 
-#define KERN_OK            (0)
-#define KERN_UNIMPLEMENTED (1)
-#define KERN_NAME_EXISTS   (2)
+#define KERN_OK               (0)
+#define KERN_UNIMPLEMENTED    (1)
+#define KERN_NAME_EXISTS      (2)
+#define KERN_INVALID_ARGUMENT (3)
+#define KERN_UNSUPPORTED      (4)
 
 #endif

include/socks/vm.h

@@ -30,6 +30,14 @@ typedef struct vm_object {
 	unsigned int reserved;
 } vm_object_t;
 
+typedef enum vm_prot {
+	VM_PROT_READ  = 0x01u,
+	VM_PROT_WRITE = 0x02u,
+	VM_PROT_EXEC  = 0x04u,
+	VM_PROT_USER  = 0x08u,
+	VM_PROT_SVR   = 0x10u,
+} vm_prot_t;
+
 typedef enum vm_flags {
 	VM_GET_DMA = 0x01u,
 } vm_flags_t;