PRIVATE void fpu_init(void)
{
unsigned short cw, sw;
fninit();
sw = fnstsw();
fnstcw(&cw);
if((sw & 0xff) == 0 &&
(cw & 0x103f) == 0x3f) {
/* We have some sort of FPU, but don't check exact model.
* Set CR0_NE and CR0_MP to handle fpu exceptions
* in native mode. */
write_cr0(read_cr0() | CR0_MP_NE);
fpu_presence = 1;
if(_cpufeature(_CPUF_I386_FXSR)) {
register struct proc *rp;
phys_bytes aligned_fp_area;
u32_t cr4 = read_cr4() | CR4_OSFXSR; /* Enable FXSR. */
/* OSXMMEXCPT if supported
* FXSR feature can be available without SSE
*/
if(_cpufeature(_CPUF_I386_SSE))
cr4 |= CR4_OSXMMEXCPT;
write_cr4(cr4);
osfxsr_feature = 1;
for (rp = BEG_PROC_ADDR; rp < END_PROC_ADDR; ++rp) {
/* FXSR requires 16-byte alignment of memory
* image, but unfortunately some old tools
* (probably linker) ignores ".balign 16"
* applied to our memory image.
* Thus we have to do manual alignment.
*/
aligned_fp_area =
(phys_bytes) &rp->p_fpu_state.fpu_image;
if(aligned_fp_area % FPUALIGN) {
aligned_fp_area += FPUALIGN -
(aligned_fp_area % FPUALIGN);
}
rp->p_fpu_state.fpu_save_area_p =
(void *) aligned_fp_area;
}
} else {
osfxsr_feature = 0;
}
} else {
/* No FPU presents. */
fpu_presence = 0;
osfxsr_feature = 0;
return;
}
}
PUBLIC void fpu_init(void)
{
unsigned short cw, sw;
fninit();
sw = fnstsw();
fnstcw(&cw);
if((sw & 0xff) == 0 &&
(cw & 0x103f) == 0x3f) {
/* We have some sort of FPU, but don't check exact model.
* Set CR0_NE and CR0_MP to handle fpu exceptions
* in native mode. */
write_cr0(read_cr0() | CR0_MP_NE);
get_cpulocal_var(fpu_presence) = 1;
if(_cpufeature(_CPUF_I386_FXSR)) {
u32_t cr4 = read_cr4() | CR4_OSFXSR; /* Enable FXSR. */
/* OSXMMEXCPT if supported
* FXSR feature can be available without SSE
*/
if(_cpufeature(_CPUF_I386_SSE))
cr4 |= CR4_OSXMMEXCPT;
write_cr4(cr4);
osfxsr_feature = 1;
} else {
osfxsr_feature = 0;
}
} else {
/* No FPU presents. */
get_cpulocal_var(fpu_presence) = 0;
osfxsr_feature = 0;
return;
}
}
static void vm_enable_paging(void)
{
u32_t cr0, cr4;
int pgeok;
psok = _cpufeature(_CPUF_I386_PSE);
pgeok = _cpufeature(_CPUF_I386_PGE);
cr0= read_cr0();
cr4= read_cr4();
/* First clear PG and PGE flag, as PGE must be enabled after PG. */
write_cr0(cr0 & ~I386_CR0_PG);
write_cr4(cr4 & ~(I386_CR4_PGE | I386_CR4_PSE));
cr0= read_cr0();
cr4= read_cr4();
/* Our first page table contains 4MB entries. */
if(psok)
cr4 |= I386_CR4_PSE;
write_cr4(cr4);
/* First enable paging, then enable global page flag. */
cr0 |= I386_CR0_PG;
write_cr0(cr0 );
cr0 |= I386_CR0_WP;
write_cr0(cr0);
/* May we enable these features? */
if(pgeok)
cr4 |= I386_CR4_PGE;
write_cr4(cr4);
}
void vm_enable_paging(void)
{
u32_t cr0, cr4;
int pgeok;
pgeok = _cpufeature(_CPUF_I386_PGE);
cr0= read_cr0();
cr4= read_cr4();
/* The boot loader should have put us in protected mode. */
assert(cr0 & I386_CR0_PE);
/* First clear PG and PGE flag, as PGE must be enabled after PG. */
write_cr0(cr0 & ~I386_CR0_PG);
write_cr4(cr4 & ~(I386_CR4_PGE | I386_CR4_PSE));
cr0= read_cr0();
cr4= read_cr4();
/* Our page table contains 4MB entries. */
cr4 |= I386_CR4_PSE;
write_cr4(cr4);
/* First enable paging, then enable global page flag. */
cr0 |= I386_CR0_PG;
write_cr0(cr0);
cr0 |= I386_CR0_WP;
write_cr0(cr0);
/* May we enable these features? */
if(pgeok)
cr4 |= I386_CR4_PGE;
write_cr4(cr4);
}
/*===========================================================================*
* pt_init *
*===========================================================================*/
void pt_init(void)
{
pt_t *newpt;
int s, r, p;
vir_bytes sparepages_mem;
#if defined(__arm__)
vir_bytes sparepagedirs_mem;
#endif
static u32_t currentpagedir[ARCH_VM_DIR_ENTRIES];
int m = kernel_boot_info.kern_mod;
#if defined(__i386__)
int global_bit_ok = 0;
u32_t mypdbr; /* Page Directory Base Register (cr3) value */
#elif defined(__arm__)
u32_t myttbr;
#endif
/* Find what the physical location of the kernel is. */
assert(m >= 0);
assert(m < kernel_boot_info.mods_with_kernel);
assert(kernel_boot_info.mods_with_kernel < MULTIBOOT_MAX_MODS);
kern_mb_mod = &kernel_boot_info.module_list[m];
kern_size = kern_mb_mod->mod_end - kern_mb_mod->mod_start;
assert(!(kern_mb_mod->mod_start % ARCH_BIG_PAGE_SIZE));
assert(!(kernel_boot_info.vir_kern_start % ARCH_BIG_PAGE_SIZE));
kern_start_pde = kernel_boot_info.vir_kern_start / ARCH_BIG_PAGE_SIZE;
/* Get ourselves spare pages. */
sparepages_mem = (vir_bytes) static_sparepages;
assert(!(sparepages_mem % VM_PAGE_SIZE));
#if defined(__arm__)
/* Get ourselves spare pagedirs. */
sparepagedirs_mem = (vir_bytes) static_sparepagedirs;
assert(!(sparepagedirs_mem % ARCH_PAGEDIR_SIZE));
#endif
/* Spare pages are used to allocate memory before VM has its own page
* table that things (i.e. arbitrary physical memory) can be mapped into.
* We get it by pre-allocating it in our bss (allocated and mapped in by
* the kernel) in static_sparepages. We also need the physical addresses
* though; we look them up now so they are ready for use.
*/
#if defined(__arm__)
missing_sparedirs = 0;
assert(STATIC_SPAREPAGEDIRS <= SPAREPAGEDIRS);
for(s = 0; s < SPAREPAGEDIRS; s++) {
vir_bytes v = (sparepagedirs_mem + s*ARCH_PAGEDIR_SIZE);;
phys_bytes ph;
if((r=sys_umap(SELF, VM_D, (vir_bytes) v,
ARCH_PAGEDIR_SIZE, &ph)) != OK)
panic("pt_init: sys_umap failed: %d", r);
if(s >= STATIC_SPAREPAGEDIRS) {
sparepagedirs[s].pagedir = NULL;
missing_sparedirs++;
continue;
}
sparepagedirs[s].pagedir = (void *) v;
sparepagedirs[s].phys = ph;
}
#endif
if(!(spare_pagequeue = reservedqueue_new(SPAREPAGES, 1, 1, 0)))
panic("reservedqueue_new for single pages failed");
assert(STATIC_SPAREPAGES < SPAREPAGES);
for(s = 0; s < STATIC_SPAREPAGES; s++) {
void *v = (void *) (sparepages_mem + s*VM_PAGE_SIZE);
phys_bytes ph;
if((r=sys_umap(SELF, VM_D, (vir_bytes) v,
VM_PAGE_SIZE*SPAREPAGES, &ph)) != OK)
panic("pt_init: sys_umap failed: %d", r);
reservedqueue_add(spare_pagequeue, v, ph);
}
#if defined(__i386__)
/* global bit and 4MB pages available? */
global_bit_ok = _cpufeature(_CPUF_I386_PGE);
bigpage_ok = _cpufeature(_CPUF_I386_PSE);
/* Set bit for PTE's and PDE's if available. */
if(global_bit_ok)
global_bit = I386_VM_GLOBAL;
#endif
/* Now reserve another pde for kernel's own mappings. */
{
int kernmap_pde;
phys_bytes addr, len;
int flags, index = 0;
u32_t offset = 0;
kernmap_pde = freepde();
offset = kernmap_pde * ARCH_BIG_PAGE_SIZE;
while(sys_vmctl_get_mapping(index, &addr, &len,
&flags) == OK) {
int usedpde;
vir_bytes vir;
if(index >= MAX_KERNMAPPINGS)
panic("VM: too many kernel mappings: %d", index);
kern_mappings[index].phys_addr = addr;
kern_mappings[index].len = len;
kern_mappings[index].flags = flags;
kern_mappings[index].vir_addr = offset;
kern_mappings[index].flags =
ARCH_VM_PTE_PRESENT;
if(flags & VMMF_UNCACHED)
#if defined(__i386__)
kern_mappings[index].flags |= PTF_NOCACHE;
#elif defined(__arm__)
kern_mappings[index].flags |= ARM_VM_PTE_DEVICE;
#endif
if(flags & VMMF_USER)
kern_mappings[index].flags |= ARCH_VM_PTE_USER;
#if defined(__arm__)
else
kern_mappings[index].flags |= ARM_VM_PTE_SUPER;
#endif
if(flags & VMMF_WRITE)
kern_mappings[index].flags |= ARCH_VM_PTE_RW;
#if defined(__i386__)
if(flags & VMMF_GLO)
kern_mappings[index].flags |= I386_VM_GLOBAL;
#elif defined(__arm__)
else
kern_mappings[index].flags |= ARCH_VM_PTE_RO;
#endif
if(addr % VM_PAGE_SIZE)
panic("VM: addr unaligned: %d", addr);
if(len % VM_PAGE_SIZE)
panic("VM: len unaligned: %d", len);
vir = offset;
if(sys_vmctl_reply_mapping(index, vir) != OK)
panic("VM: reply failed");
offset += len;
index++;
kernmappings++;
usedpde = ARCH_VM_PDE(offset);
while(usedpde > kernmap_pde) {
int newpde = freepde();
assert(newpde == kernmap_pde+1);
kernmap_pde = newpde;
}
}
}
/* Reserve PDEs available for mapping in the page directories. */
{
int pd;
for(pd = 0; pd < MAX_PAGEDIR_PDES; pd++) {
struct pdm *pdm = &pagedir_mappings[pd];
pdm->pdeno = freepde();
phys_bytes ph;
/* Allocate us a page table in which to
* remember page directory pointers.
*/
if(!(pdm->page_directories =
vm_allocpage(&ph, VMP_PAGETABLE))) {
panic("no virt addr for vm mappings");
}
memset(pdm->page_directories, 0, VM_PAGE_SIZE);
pdm->phys = ph;
#if defined(__i386__)
pdm->val = (ph & ARCH_VM_ADDR_MASK) |
ARCH_VM_PDE_PRESENT | ARCH_VM_PTE_RW;
#elif defined(__arm__)
pdm->val = (ph & ARCH_VM_PDE_MASK)
| ARCH_VM_PDE_PRESENT
| ARM_VM_PDE_DOMAIN; //LSC FIXME
#endif
}
}
/* Allright. Now. We have to make our own page directory and page tables,
* that the kernel has already set up, accessible to us. It's easier to
* understand if we just copy all the required pages (i.e. page directory
* and page tables), and set up the pointers as if VM had done it itself.
*
* This allocation will happen without using any page table, and just
* uses spare pages.
*/
newpt = &vmprocess->vm_pt;
if(pt_new(newpt) != OK)
panic("vm pt_new failed");
/* Get our current pagedir so we can see it. */
#if defined(__i386__)
if(sys_vmctl_get_pdbr(SELF, &mypdbr) != OK)
#elif defined(__arm__)
if(sys_vmctl_get_pdbr(SELF, &myttbr) != OK)
#endif
panic("VM: sys_vmctl_get_pdbr failed");
#if defined(__i386__)
if(sys_vircopy(NONE, mypdbr, SELF,
(vir_bytes) currentpagedir, VM_PAGE_SIZE) != OK)
#elif defined(__arm__)
if(sys_vircopy(NONE, myttbr, SELF,
(vir_bytes) currentpagedir, ARCH_PAGEDIR_SIZE) != OK)
#endif
panic("VM: sys_vircopy failed");
/* We have mapped in kernel ourselves; now copy mappings for VM
* that kernel made, including allocations for BSS. Skip identity
* mapping bits; just map in VM.
*/
for(p = 0; p < ARCH_VM_DIR_ENTRIES; p++) {
u32_t entry = currentpagedir[p];
phys_bytes ptaddr_kern, ptaddr_us;
/* BIGPAGEs are kernel mapping (do ourselves) or boot
* identity mapping (don't want).
*/
if(!(entry & ARCH_VM_PDE_PRESENT)) continue;
if((entry & ARCH_VM_BIGPAGE)) continue;
if(pt_ptalloc(newpt, p, 0) != OK)
panic("pt_ptalloc failed");
assert(newpt->pt_dir[p] & ARCH_VM_PDE_PRESENT);
#if defined(__i386__)
ptaddr_kern = entry & ARCH_VM_ADDR_MASK;
ptaddr_us = newpt->pt_dir[p] & ARCH_VM_ADDR_MASK;
#elif defined(__arm__)
ptaddr_kern = entry & ARCH_VM_PDE_MASK;
ptaddr_us = newpt->pt_dir[p] & ARCH_VM_PDE_MASK;
#endif
/* Copy kernel-initialized pagetable contents into our
* normally accessible pagetable.
*/
if(sys_abscopy(ptaddr_kern, ptaddr_us, VM_PAGE_SIZE) != OK)
panic("pt_init: abscopy failed");
}
/* Inform kernel vm has a newly built page table. */
assert(vmproc[VM_PROC_NR].vm_endpoint == VM_PROC_NR);
pt_bind(newpt, &vmproc[VM_PROC_NR]);
pt_init_done = 1;
/* All OK. */
return;
}
/*===========================================================================*
* pt_init *
*===========================================================================*/
PUBLIC void pt_init(phys_bytes usedlimit)
{
/* By default, the kernel gives us a data segment with pre-allocated
* memory that then can't grow. We want to be able to allocate memory
* dynamically, however. So here we copy the part of the page table
* that's ours, so we get a private page table. Then we increase the
* hardware segment size so we can allocate memory above our stack.
*/
pt_t *newpt;
int s, r;
vir_bytes v;
phys_bytes lo, hi;
vir_bytes extra_clicks;
u32_t moveup = 0;
int global_bit_ok = 0;
int free_pde;
int p;
struct vm_ep_data ep_data;
vir_bytes sparepages_mem;
phys_bytes sparepages_ph;
vir_bytes ptr;
/* Shorthand. */
newpt = &vmprocess->vm_pt;
/* Get ourselves spare pages. */
ptr = (vir_bytes) static_sparepages;
ptr += I386_PAGE_SIZE - (ptr % I386_PAGE_SIZE);
if(!(sparepages_mem = ptr))
panic("pt_init: aalloc for spare failed");
if((r=sys_umap(SELF, VM_D, (vir_bytes) sparepages_mem,
I386_PAGE_SIZE*SPAREPAGES, &sparepages_ph)) != OK)
panic("pt_init: sys_umap failed: %d", r);
missing_spares = 0;
assert(STATIC_SPAREPAGES < SPAREPAGES);
for(s = 0; s < SPAREPAGES; s++) {
if(s >= STATIC_SPAREPAGES) {
sparepages[s].page = NULL;
missing_spares++;
continue;
}
sparepages[s].page = (void *) (sparepages_mem + s*I386_PAGE_SIZE);
sparepages[s].phys = sparepages_ph + s*I386_PAGE_SIZE;
}
/* global bit and 4MB pages available? */
global_bit_ok = _cpufeature(_CPUF_I386_PGE);
bigpage_ok = _cpufeature(_CPUF_I386_PSE);
/* Set bit for PTE's and PDE's if available. */
if(global_bit_ok)
global_bit = I386_VM_GLOBAL;
/* The kernel and boot time processes need an identity mapping.
* We use full PDE's for this without separate page tables.
* Figure out which pde we can start using for other purposes.
*/
id_map_high_pde = usedlimit / I386_BIG_PAGE_SIZE;
/* We have to make mappings up till here. */
free_pde = id_map_high_pde+1;
/* Initial (current) range of our virtual address space. */
lo = CLICK2ABS(vmprocess->vm_arch.vm_seg[T].mem_phys);
hi = CLICK2ABS(vmprocess->vm_arch.vm_seg[S].mem_phys +
vmprocess->vm_arch.vm_seg[S].mem_len);
assert(!(lo % I386_PAGE_SIZE));
assert(!(hi % I386_PAGE_SIZE));
if(lo < VM_PROCSTART) {
moveup = VM_PROCSTART - lo;
assert(!(VM_PROCSTART % I386_PAGE_SIZE));
assert(!(lo % I386_PAGE_SIZE));
assert(!(moveup % I386_PAGE_SIZE));
}
/* Make new page table for ourselves, partly copied
* from the current one.
*/
if(pt_new(newpt) != OK)
panic("pt_init: pt_new failed");
/* Set up mappings for VM process. */
for(v = lo; v < hi; v += I386_PAGE_SIZE) {
phys_bytes addr;
u32_t flags;
/* We have to write the new position in the PT,
* so we can move our segments.
*/
if(pt_writemap(vmprocess, newpt, v+moveup, v, I386_PAGE_SIZE,
I386_VM_PRESENT|I386_VM_WRITE|I386_VM_USER, 0) != OK)
panic("pt_init: pt_writemap failed");
}
/* Move segments up too. */
vmprocess->vm_arch.vm_seg[T].mem_phys += ABS2CLICK(moveup);
vmprocess->vm_arch.vm_seg[D].mem_phys += ABS2CLICK(moveup);
vmprocess->vm_arch.vm_seg[S].mem_phys += ABS2CLICK(moveup);
/* Allocate us a page table in which to remember page directory
* pointers.
*/
if(!(page_directories = vm_allocpage(&page_directories_phys,
VMP_PAGETABLE)))
panic("no virt addr for vm mappings");
memset(page_directories, 0, I386_PAGE_SIZE);
/* Increase our hardware data segment to create virtual address
* space above our stack. We want to increase it to VM_DATATOP,
* like regular processes have.
*/
extra_clicks = ABS2CLICK(VM_DATATOP - hi);
vmprocess->vm_arch.vm_seg[S].mem_len += extra_clicks;
/* We pretend to the kernel we have a huge stack segment to
* increase our data segment.
*/
vmprocess->vm_arch.vm_data_top =
(vmprocess->vm_arch.vm_seg[S].mem_vir +
vmprocess->vm_arch.vm_seg[S].mem_len) << CLICK_SHIFT;
/* Where our free virtual address space starts.
* This is only a hint to the VM system.
*/
newpt->pt_virtop = 0;
/* Let other functions know VM now has a private page table. */
vmprocess->vm_flags |= VMF_HASPT;
/* Now reserve another pde for kernel's own mappings. */
{
int kernmap_pde;
phys_bytes addr, len;
int flags, index = 0;
u32_t offset = 0;
kernmap_pde = free_pde++;
offset = kernmap_pde * I386_BIG_PAGE_SIZE;
while(sys_vmctl_get_mapping(index, &addr, &len,
&flags) == OK) {
vir_bytes vir;
if(index >= MAX_KERNMAPPINGS)
panic("VM: too many kernel mappings: %d", index);
kern_mappings[index].phys_addr = addr;
kern_mappings[index].len = len;
kern_mappings[index].flags = flags;
kern_mappings[index].lin_addr = offset;
kern_mappings[index].flags =
I386_VM_PRESENT | I386_VM_USER | I386_VM_WRITE |
global_bit;
if(flags & VMMF_UNCACHED)
kern_mappings[index].flags |= PTF_NOCACHE;
if(addr % I386_PAGE_SIZE)
panic("VM: addr unaligned: %d", addr);
if(len % I386_PAGE_SIZE)
panic("VM: len unaligned: %d", len);
vir = arch_map2vir(&vmproc[VMP_SYSTEM], offset);
if(sys_vmctl_reply_mapping(index, vir) != OK)
panic("VM: reply failed");
offset += len;
index++;
kernmappings++;
}
}
/* Find a PDE below processes available for mapping in the
* page directories (readonly).
*/
pagedir_pde = free_pde++;
pagedir_pde_val = (page_directories_phys & I386_VM_ADDR_MASK) |
I386_VM_PRESENT | I386_VM_USER | I386_VM_WRITE;
/* Tell kernel about free pde's. */
while(free_pde*I386_BIG_PAGE_SIZE < VM_PROCSTART) {
if((r=sys_vmctl(SELF, VMCTL_I386_FREEPDE, free_pde++)) != OK) {
panic("VMCTL_I386_FREEPDE failed: %d", r);
}
}
/* first pde in use by process. */
proc_pde = free_pde;
/* Give our process the new, copied, private page table. */
pt_mapkernel(newpt); /* didn't know about vm_dir pages earlier */
pt_bind(newpt, vmprocess);
/* new segment limit for the kernel after paging is enabled */
ep_data.data_seg_limit = free_pde*I386_BIG_PAGE_SIZE;
/* the memory map which must be installed after paging is enabled */
ep_data.mem_map = vmprocess->vm_arch.vm_seg;
/* Now actually enable paging. */
if(sys_vmctl_enable_paging(&ep_data) != OK)
panic("pt_init: enable paging failed");
/* Back to reality - this is where the stack actually is. */
vmprocess->vm_arch.vm_seg[S].mem_len -= extra_clicks;
/* Pretend VM stack top is the same as any regular process, not to
* have discrepancies with new VM instances later on.
*/
vmprocess->vm_stacktop = VM_STACKTOP;
/* All OK. */
return;
}