/**
* \brief Map a region of physical memory into physical memory address space.
*
* Maps the region of physical memory, based at base and sized size bytes
* to the same-sized virtual memory region. All pages are flagged according to
* bitmap. This function automatically fills the needed page directory entries
* in the page hierarchy rooted at pml4. base and size will be made
* page-aligned by this function.
*
* \param base Physical base address of memory region
* \param size Size in bytes of memory region
* \param bitmap Bitmap of flags for page tables/directories
*
* \return 0 on success, -1 on error (out of range)
*/
static int
paging_map_mem(lpaddr_t base,
size_t size,
uint64_t bitmap)
{
lvaddr_t vaddr, vbase = local_phys_to_mem(base);
lpaddr_t addr;
// Align given physical base address
if (base & X86_64_MEM_PAGE_MASK) {
base -= base & X86_64_MEM_PAGE_MASK;
}
paging_align(&vbase, &base, &size, X86_64_MEM_PAGE_SIZE);
// Is mapped region out of range?
assert(base + size <= (lpaddr_t)K1OM_PADDR_SPACE_LIMIT);
if (base + size > (lpaddr_t) K1OM_PADDR_SPACE_LIMIT) {
return -1;
}
// Map pages, tables and directories
for (vaddr = vbase, addr = base; vaddr < vbase + size; vaddr +=
X86_64_MEM_PAGE_SIZE, addr += X86_64_MEM_PAGE_SIZE) {
union x86_64_pdir_entry *pml4_base = &pml4[X86_64_PML4_BASE(vaddr)];
union x86_64_pdir_entry *pdpt_base =
&mem_pdpt[X86_64_PML4_BASE(addr)][X86_64_PDPT_BASE(vaddr)];
union x86_64_ptable_entry *pdir_base =
&mem_pdir[X86_64_PML4_BASE(addr)][X86_64_PDPT_BASE(addr)][X86_64_PDIR_BASE(
vaddr)];
debug(SUBSYS_PAGING,
"Mapping 2M page: vaddr = 0x%"PRIxLVADDR"x, addr = 0x%lx, " "PML4_BASE = %lu, PDPT_BASE = %lu, PDIR_BASE = %lu -- ",
vaddr, addr, X86_64_PML4_BASE(vaddr), X86_64_PDPT_BASE(vaddr),
X86_64_PDIR_BASE(vaddr));
mapit(pml4_base, pdpt_base, pdir_base, addr, bitmap);
}
// XXX FIXME: get rid of this TBL flush code, or move it elsewhere
// uint64_t cr3;
// __asm__ __volatile__("mov %%cr3,%0" : "=a" (cr3) : );
// __asm__ __volatile__("mov %0,%%cr3" : : "a" (cr3));
return 0;
}
void paging_dump_tables(struct dcb *dispatcher)
{
lvaddr_t root_pt = local_phys_to_mem(dispatcher->vspace);
// loop over pdpts
union x86_64_ptable_entry *pt;
size_t kernel_pml4e = X86_64_PML4_BASE(X86_64_MEMORY_OFFSET);
for (int pdpt_index = 0; pdpt_index < kernel_pml4e; pdpt_index++) {
union x86_64_pdir_entry *pdpt = (union x86_64_pdir_entry *)root_pt + pdpt_index;
if (!pdpt->raw) { continue; }
genpaddr_t pdpt_gp = pdpt->d.base_addr << BASE_PAGE_BITS;
lvaddr_t pdpt_lv = local_phys_to_mem(gen_phys_to_local_phys(pdpt_gp));
for (int pdir_index = 0; pdir_index < X86_64_PTABLE_SIZE; pdir_index++) {
// get pdir
union x86_64_pdir_entry *pdir = (union x86_64_pdir_entry *)pdpt_lv + pdir_index;
pt = (union x86_64_ptable_entry*)pdir;
if (!pdir->raw) { continue; }
// check if pdir or huge page
if (pt->huge.always1) {
// is huge page mapping
genpaddr_t paddr = (genpaddr_t)pt->huge.base_addr << HUGE_PAGE_BITS;
printf("%d.%d: 0x%"PRIxGENPADDR"\n", pdpt_index, pdir_index, paddr);
// goto next pdpt entry
continue;
}
genpaddr_t pdir_gp = pdir->d.base_addr << BASE_PAGE_BITS;
lvaddr_t pdir_lv = local_phys_to_mem(gen_phys_to_local_phys(pdir_gp));
for (int ptable_index = 0; ptable_index < X86_64_PTABLE_SIZE; ptable_index++) {
// get ptable
union x86_64_pdir_entry *ptable = (union x86_64_pdir_entry *)pdir_lv + ptable_index;
pt = (union x86_64_ptable_entry *)ptable;
if (!ptable->raw) { continue; }
// check if ptable or large page
if (pt->large.always1) {
// is large page mapping
genpaddr_t paddr = (genpaddr_t)pt->large.base_addr << LARGE_PAGE_BITS;
printf("%d.%d.%d: 0x%"PRIxGENPADDR"\n", pdpt_index, pdir_index, ptable_index, paddr);
// goto next pdir entry
continue;
}
genpaddr_t ptable_gp = ptable->d.base_addr << BASE_PAGE_BITS;
lvaddr_t ptable_lv = local_phys_to_mem(gen_phys_to_local_phys(ptable_gp));
for (int entry = 0; entry < X86_64_PTABLE_SIZE; entry++) {
union x86_64_ptable_entry *e =
(union x86_64_ptable_entry *)ptable_lv + entry;
genpaddr_t paddr = (genpaddr_t)e->base.base_addr << BASE_PAGE_BITS;
if (!paddr) {
continue;
}
printf("%d.%d.%d.%d: 0x%"PRIxGENPADDR"\n", pdpt_index, pdir_index, ptable_index, entry, paddr);
}
}
}
}
}
/**
* \brief Make a "good" PML4 table out of a page table.
*
* A "good" PML4 table is one that has all physical address space and
* the kernel mapped in. This function modifies the passed PML4, based
* at physical address 'base' accordingly. It does this by taking out
* the corresponding entries of the kernel's pristine PML4 table.
*
* \param base Physical base address of PML4 table to make "good".
*/
void paging_x86_64_make_good_pml4(lpaddr_t base)
{
union x86_64_pdir_entry *newpml4 =
(union x86_64_pdir_entry *)local_phys_to_mem(base);
int i;
// XXX: Disabled till vaddr_t is figured out
debug(SUBSYS_PAGING, "Is now a PML4: table = 0x%"PRIxLPADDR"\n", base);
// Map memory
for(i = X86_64_PML4_BASE(MEMORY_OFFSET); i < X86_64_PTABLE_SIZE; i++) {
newpml4[i] = pml4[i];
}
}
static inline lvaddr_t get_leaf_ptable_for_vaddr(genvaddr_t vaddr)
{
lvaddr_t root_pt = local_phys_to_mem(dcb_current->vspace);
// get pdpt
union x86_64_pdir_entry *pdpt = (union x86_64_pdir_entry *)root_pt + X86_64_PML4_BASE(vaddr);
if (!pdpt->raw) { return 0; }
genpaddr_t pdpt_gp = pdpt->d.base_addr << BASE_PAGE_BITS;
lvaddr_t pdpt_lv = local_phys_to_mem(gen_phys_to_local_phys(pdpt_gp));
// get pdir
union x86_64_pdir_entry *pdir = (union x86_64_pdir_entry *)pdpt_lv + X86_64_PDPT_BASE(vaddr);
if (!pdir->raw) { return 0; }
genpaddr_t pdir_gp = pdir->d.base_addr << BASE_PAGE_BITS;
lvaddr_t pdir_lv = local_phys_to_mem(gen_phys_to_local_phys(pdir_gp));
// get ptable
union x86_64_ptable_entry *ptable = (union x86_64_ptable_entry *)pdir_lv + X86_64_PDIR_BASE(vaddr);
if (!ptable->raw) { return 0; }
genpaddr_t ptable_gp = ptable->base.base_addr << BASE_PAGE_BITS;
lvaddr_t ptable_lv = local_phys_to_mem(gen_phys_to_local_phys(ptable_gp));
return ptable_lv;
}
/// Map within a x86_64 non leaf ptable
static errval_t x86_64_non_ptable(struct capability *dest, cslot_t slot,
struct capability *src, uintptr_t flags,
uintptr_t offset, size_t pte_count)
{
//printf("page_mappings_arch:x86_64_non_ptable\n");
if (slot >= X86_64_PTABLE_SIZE) { // Within pagetable
return SYS_ERR_VNODE_SLOT_INVALID;
}
if (type_is_vnode(src->type) && pte_count != 1) { // only allow single ptable mappings
printf("src type and count mismatch\n");
return SYS_ERR_VM_MAP_SIZE;
}
if (slot + pte_count > X86_64_PTABLE_SIZE) { // mapping size ok
printf("mapping size invalid (%zd)\n", pte_count);
return SYS_ERR_VM_MAP_SIZE;
}
size_t page_size = 0;
paging_x86_64_flags_t flags_large = 0;
switch (dest->type) {
case ObjType_VNode_x86_64_pml4:
if (src->type != ObjType_VNode_x86_64_pdpt) { // Right mapping
printf("src type invalid\n");
return SYS_ERR_WRONG_MAPPING;
}
if(slot >= X86_64_PML4_BASE(MEMORY_OFFSET)) { // Kernel mapped here
return SYS_ERR_VNODE_SLOT_RESERVED;
}
break;
case ObjType_VNode_x86_64_pdpt:
// huge page support
if (src->type != ObjType_VNode_x86_64_pdir) { // Right mapping
// TODO: check if the system allows 1GB mappings
page_size = X86_64_HUGE_PAGE_SIZE;
// check offset within frame
genpaddr_t off = offset;
if (off + pte_count * X86_64_HUGE_PAGE_SIZE > get_size(src)) {
return SYS_ERR_FRAME_OFFSET_INVALID;
}
// Calculate page access protection flags /
// Get frame cap rights
flags_large = paging_x86_64_cap_to_page_flags(src->rights);
// Mask with provided access rights mask
flags_large = paging_x86_64_mask_attrs(flags, X86_64_PTABLE_ACCESS(flags));
// Add additional arch-specific flags
flags_large |= X86_64_PTABLE_FLAGS(flags);
// Unconditionally mark the page present
flags_large |= X86_64_PTABLE_PRESENT;
}
break;
case ObjType_VNode_x86_64_pdir:
// superpage support
if (src->type != ObjType_VNode_x86_64_ptable) { // Right mapping
page_size = X86_64_LARGE_PAGE_SIZE;
// check offset within frame
genpaddr_t off = offset;
if (off + pte_count * X86_64_LARGE_PAGE_SIZE > get_size(src)) {
return SYS_ERR_FRAME_OFFSET_INVALID;
}
// Calculate page access protection flags /
// Get frame cap rights
flags_large = paging_x86_64_cap_to_page_flags(src->rights);
// Mask with provided access rights mask
flags_large = paging_x86_64_mask_attrs(flags, X86_64_PTABLE_ACCESS(flags));
// Add additional arch-specific flags
flags_large |= X86_64_PTABLE_FLAGS(flags);
// Unconditionally mark the page present
flags_large |= X86_64_PTABLE_PRESENT;
}
break;
default:
printf("dest type invalid\n");
return SYS_ERR_DEST_TYPE_INVALID;
}
// Convert destination base address
genpaddr_t dest_gp = get_address(dest);
lpaddr_t dest_lp = gen_phys_to_local_phys(dest_gp);
lvaddr_t dest_lv = local_phys_to_mem(dest_lp);
// Convert source base address
genpaddr_t src_gp = get_address(src);
lpaddr_t src_lp = gen_phys_to_local_phys(src_gp);
// set metadata
struct cte *src_cte = cte_for_cap(src);
src_cte->mapping_info.pte = dest_lp + slot * sizeof(union x86_64_ptable_entry);
src_cte->mapping_info.pte_count = pte_count;
src_cte->mapping_info.offset = offset;
cslot_t last_slot = slot + pte_count;
for (; slot < last_slot; slot++, offset += page_size) {
// Destination
union x86_64_pdir_entry *entry = (union x86_64_pdir_entry *)dest_lv + slot;
if (X86_64_IS_PRESENT(entry)) {
// cleanup mapping info
// TODO: cleanup already mapped pages
memset(&src_cte->mapping_info, 0, sizeof(struct mapping_info));
printf("slot in use\n");
return SYS_ERR_VNODE_SLOT_INUSE;
}
// determine if we map a large/huge page or a normal entry
if (page_size == X86_64_LARGE_PAGE_SIZE)
{
//a large page is mapped
paging_x86_64_map_large((union x86_64_ptable_entry *)entry, src_lp + offset, flags_large);
} else if (page_size == X86_64_HUGE_PAGE_SIZE) {
// a huge page is mapped
paging_x86_64_map_huge((union x86_64_ptable_entry *)entry, src_lp + offset, flags_large);
} else {
//a normal paging structure entry is mapped
paging_x86_64_map_table(entry, src_lp + offset);
}
}
return SYS_ERR_OK;
}