X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/6601e61aa18bf4f09af135ff61fc7f4771d23b06..0c530ab8987f0ae6a1a3d9284f40182b88852816:/osfmk/i386/pmap.c diff --git a/osfmk/i386/pmap.c b/osfmk/i386/pmap.c index e43dac464..7224204c9 100644 --- a/osfmk/i386/pmap.c +++ b/osfmk/i386/pmap.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000-2004 Apple Computer, Inc. All rights reserved. + * Copyright (c) 2000-2006 Apple Computer, Inc. All rights reserved. * * @APPLE_LICENSE_HEADER_START@ * @@ -116,6 +116,8 @@ #include #include #include +#include +#include #if MACH_KDB #include @@ -129,17 +131,35 @@ #include #include +#include + +#include + +#ifdef IWANTTODEBUG +#undef DEBUG +#define DEBUG 1 +#define POSTCODE_DELAY 1 +#include +#endif /* IWANTTODEBUG */ /* * Forward declarations for internal functions. */ +void pmap_expand_pml4( + pmap_t map, + vm_map_offset_t v); + +void pmap_expand_pdpt( + pmap_t map, + vm_map_offset_t v); + void pmap_expand( pmap_t map, - vm_offset_t v); + vm_map_offset_t v); -extern void pmap_remove_range( +static void pmap_remove_range( pmap_t pmap, - vm_offset_t va, + vm_map_offset_t va, pt_entry_t *spte, pt_entry_t *epte); @@ -155,9 +175,6 @@ void phys_attribute_set( ppnum_t phys, int bits); -void pmap_growkernel( - vm_offset_t addr); - void pmap_set_reference( ppnum_t pn); @@ -166,24 +183,21 @@ void pmap_movepage( unsigned long to, vm_size_t size); -pt_entry_t * pmap_mapgetpte( - vm_map_t map, - vm_offset_t v); - boolean_t phys_page_exists( ppnum_t pn); -#ifndef set_dirbase -void set_dirbase(vm_offset_t dirbase); -#endif /* set_dirbase */ +#ifdef PMAP_DEBUG +void dump_pmap(pmap_t); +void dump_4GB_pdpt(pmap_t p); +void dump_4GB_pdpt_thread(thread_t tp); +#endif #define iswired(pte) ((pte) & INTEL_PTE_WIRED) -#define WRITE_PTE(pte_p, pte_entry) *(pte_p) = (pte_entry); -#define WRITE_PTE_FAST(pte_p, pte_entry) *(pte_p) = (pte_entry); +int nx_enabled = 1; /* enable no-execute protection */ + +int cpu_64bit = 0; -#define value_64bit(value) ((value) & 0xFFFFFFFF00000000LL) -#define low32(x) ((unsigned int)((x) & 0x00000000ffffffffLL)) /* * Private data structures. @@ -198,7 +212,7 @@ void set_dirbase(vm_offset_t dirbase); typedef struct pv_entry { struct pv_entry *next; /* next pv_entry */ pmap_t pmap; /* pmap where mapping lies */ - vm_offset_t va; /* virtual address for mapping */ + vm_map_offset_t va; /* virtual address for mapping */ } *pv_entry_t; #define PV_ENTRY_NULL ((pv_entry_t) 0) @@ -225,7 +239,7 @@ int mappingrecurse = 0; pv_free_list = pv_e->next; \ pv_free_count--; \ if (pv_free_count < PV_LOW_WATER_MARK) \ - if (hw_compare_and_store(0,1,&mappingrecurse)) \ + if (hw_compare_and_store(0,1,(u_int *)&mappingrecurse)) \ thread_call_enter(mapping_adjust_call); \ } \ simple_unlock(&pv_free_list_lock); \ @@ -241,10 +255,7 @@ int mappingrecurse = 0; zone_t pv_list_zone; /* zone of pv_entry structures */ -#ifdef PAE static zone_t pdpt_zone; -#endif - /* * Each entry in the pv_head_table is locked by a bit in the @@ -264,14 +275,8 @@ pmap_paddr_t vm_first_phys = (pmap_paddr_t) 0; pmap_paddr_t vm_last_phys = (pmap_paddr_t) 0; boolean_t pmap_initialized = FALSE;/* Has pmap_init completed? */ -pmap_paddr_t kernel_vm_end = (pmap_paddr_t)0; - -#define GROW_KERNEL_FUNCTION_IMPLEMENTED 1 -#if GROW_KERNEL_FUNCTION_IMPLEMENTED /* not needed until growing kernel pmap */ static struct vm_object kptobj_object_store; static vm_object_t kptobj; -#endif - /* * Index into pv_head table, its lock bits, and the modify/reference @@ -302,6 +307,7 @@ char *pmap_phys_attributes; * page-directory entry. */ #define PDE_MAPPED_SIZE (pdetova(1)) +uint64_t pde_mapped_size; /* * Locking and TLB invalidation @@ -342,23 +348,21 @@ char *pmap_phys_attributes; /* * We raise the interrupt level to splvm, to block interprocessor - * interrupts during pmap operations. We must take the CPU out of - * the cpus_active set while interrupts are blocked. + * interrupts during pmap operations. We mark the cpu's cr3 inactive + * while interrupts are blocked. */ -#define SPLVM(spl) { \ - spl = splhigh(); \ - mp_disable_preemption(); \ - i_bit_clear(cpu_number(), &cpus_active); \ - mp_enable_preemption(); \ +#define SPLVM(spl) { \ + spl = splhigh(); \ + CPU_CR3_MARK_INACTIVE(); \ } -#define SPLX(spl) { \ - mp_disable_preemption(); \ - i_bit_set(cpu_number(), &cpus_active); \ - mp_enable_preemption(); \ - splx(spl); \ +#define SPLX(spl) { \ + if (current_cpu_datap()->cpu_tlb_invalid) \ + process_pmap_updates(); \ + CPU_CR3_MARK_ACTIVE(); \ + splx(spl); \ } - + /* * Lock on pmap system */ @@ -405,8 +409,8 @@ extern int disableSerialOuput; #define LOOP_CHECK(msg, pmap) \ if (--loop_count == 0) { \ mp_disable_preemption(); \ - kprintf("%s: cpu %d pmap %x, cpus_active 0x%x\n", \ - msg, cpu_number(), pmap, cpus_active); \ + kprintf("%s: cpu %d pmap %x\n", \ + msg, cpu_number(), pmap); \ Debugger("deadlock detection"); \ mp_enable_preemption(); \ loop_count = max_lock_loops; \ @@ -416,76 +420,15 @@ extern int disableSerialOuput; #define LOOP_CHECK(msg, pmap) #endif /* USLOCK_DEBUG */ -#define PMAP_UPDATE_TLBS(pmap, s, e) \ -{ \ - cpu_set cpu_mask; \ - cpu_set users; \ - \ - mp_disable_preemption(); \ - cpu_mask = 1 << cpu_number(); \ - \ - /* Since the pmap is locked, other updates are locked */ \ - /* out, and any pmap_activate has finished. */ \ - \ - /* find other cpus using the pmap */ \ - users = (pmap)->cpus_using & ~cpu_mask; \ - if (users) { \ - LOOP_VAR; \ - /* signal them, and wait for them to finish */ \ - /* using the pmap */ \ - signal_cpus(users, (pmap), (s), (e)); \ - while (((pmap)->cpus_using & cpus_active & ~cpu_mask)) { \ - LOOP_CHECK("PMAP_UPDATE_TLBS", pmap); \ - cpu_pause(); \ - } \ - } \ - /* invalidate our own TLB if pmap is in use */ \ - \ - if ((pmap)->cpus_using & cpu_mask) { \ - INVALIDATE_TLB((pmap), (s), (e)); \ - } \ - \ - mp_enable_preemption(); \ -} - -#define MAX_TBIS_SIZE 32 /* > this -> TBIA */ /* XXX */ - -#define INVALIDATE_TLB(m, s, e) { \ - flush_tlb(); \ -} - -/* - * Structures to keep track of pending TLB invalidations - */ -cpu_set cpus_active; -cpu_set cpus_idle; -#define UPDATE_LIST_SIZE 4 +static void pmap_flush_tlbs(pmap_t pmap); -struct pmap_update_item { - pmap_t pmap; /* pmap to invalidate */ - vm_offset_t start; /* start address to invalidate */ - vm_offset_t end; /* end address to invalidate */ -}; +#define PMAP_UPDATE_TLBS(pmap, s, e) \ + pmap_flush_tlbs(pmap) -typedef struct pmap_update_item *pmap_update_item_t; -/* - * List of pmap updates. If the list overflows, - * the last entry is changed to invalidate all. - */ -struct pmap_update_list { - decl_simple_lock_data(,lock) - int count; - struct pmap_update_item item[UPDATE_LIST_SIZE]; -} ; -typedef struct pmap_update_list *pmap_update_list_t; +#define MAX_TBIS_SIZE 32 /* > this -> TBIA */ /* XXX */ -extern void signal_cpus( - cpu_set use_list, - pmap_t pmap, - vm_offset_t start, - vm_offset_t end); pmap_memory_region_t pmap_memory_regions[PMAP_MEMORY_REGIONS_SIZE]; @@ -493,14 +436,12 @@ pmap_memory_region_t pmap_memory_regions[PMAP_MEMORY_REGIONS_SIZE]; * Other useful macros. */ #define current_pmap() (vm_map_pmap(current_thread()->map)) -#define pmap_in_use(pmap, cpu) (((pmap)->cpus_using & (1 << (cpu))) != 0) struct pmap kernel_pmap_store; pmap_t kernel_pmap; -#ifdef PMAP_QUEUE -decl_simple_lock_data(,free_pmap_lock) -#endif +pd_entry_t high_shared_pde; +pd_entry_t commpage64_pde; struct zone *pmap_zone; /* zone of pmap structures */ @@ -508,6 +449,9 @@ int pmap_debug = 0; /* flag for debugging prints */ unsigned int inuse_ptepages_count = 0; /* debugging */ +addr64_t kernel64_cr3; +boolean_t no_shared_cr3 = FALSE; /* -no_shared_cr3 boot arg */ + /* * Pmap cache. Cache is threaded through ref_count field of pmap. * Max will eventually be constant -- variable for experimentation. @@ -518,11 +462,11 @@ pmap_t pmap_cache_list; int pmap_cache_count; decl_simple_lock_data(,pmap_cache_lock) -extern vm_offset_t hole_start, hole_end; - extern char end; static int nkpt; +extern uint32_t lowGlo; +extern void *version; pt_entry_t *DMAP1, *DMAP2; caddr_t DADDR1; @@ -533,7 +477,7 @@ caddr_t DADDR2; struct pmap_alias { vm_offset_t rpc; pmap_t pmap; - vm_offset_t va; + vm_map_offset_t va; int cookie; #define PMAP_ALIAS_COOKIE 0xdeadbeef } pmap_aliasbuf[PMAP_ALIAS_MAX]; @@ -542,73 +486,183 @@ extern vm_offset_t get_rpc(); #endif /* DEBUG_ALIAS */ -#define pmap_pde(m, v) (&((m)->dirbase[(vm_offset_t)(v) >> PDESHIFT])) -#define pdir_pde(d, v) (d[(vm_offset_t)(v) >> PDESHIFT]) - -static __inline int -pmap_is_current(pmap_t pmap) +/* + * for legacy, returns the address of the pde entry. + * for 64 bit, causes the pdpt page containing the pde entry to be mapped, + * then returns the mapped address of the pde entry in that page + */ +pd_entry_t * +pmap_pde(pmap_t m, vm_map_offset_t v) { - return (pmap == kernel_pmap || - (pmap->dirbase[PTDPTDI] & PG_FRAME) == (PTDpde[0] & PG_FRAME)); + pd_entry_t *pde; + if (!cpu_64bit || (m == kernel_pmap)) { + pde = (&((m)->dirbase[(vm_offset_t)(v) >> PDESHIFT])); + } else { + assert(m); + assert(ml_get_interrupts_enabled() == 0 || get_preemption_level() != 0); + pde = pmap64_pde(m, v); + } + return pde; } /* - * return address of mapped pte for vaddr va in pmap pmap. + * the single pml4 page per pmap is allocated at pmap create time and exists + * for the duration of the pmap. we allocate this page in kernel vm (to save us one + * level of page table dynamic mapping. + * this returns the address of the requested pml4 entry in the top level page. */ -pt_entry_t * -pmap_pte(pmap_t pmap, vm_offset_t va) -{ - pd_entry_t *pde; - pd_entry_t newpf; - - pde = pmap_pde(pmap, va); - if (*pde != 0) { - if (pmap_is_current(pmap)) - return( vtopte(va)); - newpf = *pde & PG_FRAME; - if (((*CM4) & PG_FRAME) != newpf) { - *CM4 = newpf | INTEL_PTE_RW | INTEL_PTE_VALID; - invlpg((u_int)CA4); - } - return (pt_entry_t *)CA4 + (i386_btop(va) & (NPTEPG-1)); +static inline +pml4_entry_t * +pmap64_pml4(pmap_t pmap, vm_map_offset_t vaddr) +{ + return ((pml4_entry_t *)pmap->pm_hold + ((vm_offset_t)((vaddr>>PML4SHIFT)&(NPML4PG-1)))); +} + +/* + * maps in the pml4 page, if any, containing the pdpt entry requested + * and returns the address of the pdpt entry in that mapped page + */ +pdpt_entry_t * +pmap64_pdpt(pmap_t pmap, vm_map_offset_t vaddr) +{ + pml4_entry_t newpf; + pml4_entry_t *pml4; + int i; + + assert(pmap); + assert(ml_get_interrupts_enabled() == 0 || get_preemption_level() != 0); + if ((vaddr > 0x00007FFFFFFFFFFFULL) && (vaddr < 0xFFFF800000000000ULL)) { + return(0); } - return(0); + + pml4 = pmap64_pml4(pmap, vaddr); + + if (pml4 && ((*pml4 & INTEL_PTE_VALID))) { + + newpf = *pml4 & PG_FRAME; + + + for (i=PMAP_PDPT_FIRST_WINDOW; i < PMAP_PDPT_FIRST_WINDOW+PMAP_PDPT_NWINDOWS; i++) { + if (((*(current_cpu_datap()->cpu_pmap->mapwindow[i].prv_CMAP)) & PG_FRAME) == newpf) { + return((pdpt_entry_t *)(current_cpu_datap()->cpu_pmap->mapwindow[i].prv_CADDR) + + ((vm_offset_t)((vaddr>>PDPTSHIFT)&(NPDPTPG-1)))); + } + } + + current_cpu_datap()->cpu_pmap->pdpt_window_index++; + if (current_cpu_datap()->cpu_pmap->pdpt_window_index > (PMAP_PDPT_FIRST_WINDOW+PMAP_PDPT_NWINDOWS-1)) + current_cpu_datap()->cpu_pmap->pdpt_window_index = PMAP_PDPT_FIRST_WINDOW; + pmap_store_pte( + (current_cpu_datap()->cpu_pmap->mapwindow[current_cpu_datap()->cpu_pmap->pdpt_window_index].prv_CMAP), + newpf | INTEL_PTE_RW | INTEL_PTE_VALID); + invlpg((u_int)(current_cpu_datap()->cpu_pmap->mapwindow[current_cpu_datap()->cpu_pmap->pdpt_window_index].prv_CADDR)); + return ((pdpt_entry_t *)(current_cpu_datap()->cpu_pmap->mapwindow[current_cpu_datap()->cpu_pmap->pdpt_window_index].prv_CADDR) + + ((vm_offset_t)((vaddr>>PDPTSHIFT)&(NPDPTPG-1)))); + } + + return (0); } - -#define DEBUG_PTE_PAGE 0 -#if DEBUG_PTE_PAGE -void -ptep_check( - ptep_t ptep) +/* + * maps in the pdpt page, if any, containing the pde entry requested + * and returns the address of the pde entry in that mapped page + */ +pd_entry_t * +pmap64_pde(pmap_t pmap, vm_map_offset_t vaddr) { - register pt_entry_t *pte, *epte; - int ctu, ctw; + pdpt_entry_t newpf; + pdpt_entry_t *pdpt; + int i; - /* check the use and wired counts */ - if (ptep == PTE_PAGE_NULL) - return; - pte = pmap_pte(ptep->pmap, ptep->va); - epte = pte + INTEL_PGBYTES/sizeof(pt_entry_t); - ctu = 0; - ctw = 0; - while (pte < epte) { - if (pte->pfn != 0) { - ctu++; - if (pte->wired) - ctw++; + assert(pmap); + assert(ml_get_interrupts_enabled() == 0 || get_preemption_level() != 0); + if ((vaddr > 0x00007FFFFFFFFFFFULL) && (vaddr < 0xFFFF800000000000ULL)) { + return(0); + } + + /* if (vaddr & (1ULL << 63)) panic("neg addr");*/ + pdpt = pmap64_pdpt(pmap, vaddr); + + if (pdpt && ((*pdpt & INTEL_PTE_VALID))) { + + newpf = *pdpt & PG_FRAME; + + for (i=PMAP_PDE_FIRST_WINDOW; i < PMAP_PDE_FIRST_WINDOW+PMAP_PDE_NWINDOWS; i++) { + if (((*(current_cpu_datap()->cpu_pmap->mapwindow[i].prv_CMAP)) & PG_FRAME) == newpf) { + return((pd_entry_t *)(current_cpu_datap()->cpu_pmap->mapwindow[i].prv_CADDR) + + ((vm_offset_t)((vaddr>>PDSHIFT)&(NPDPG-1)))); + } } - pte++; + + current_cpu_datap()->cpu_pmap->pde_window_index++; + if (current_cpu_datap()->cpu_pmap->pde_window_index > (PMAP_PDE_FIRST_WINDOW+PMAP_PDE_NWINDOWS-1)) + current_cpu_datap()->cpu_pmap->pde_window_index = PMAP_PDE_FIRST_WINDOW; + pmap_store_pte( + (current_cpu_datap()->cpu_pmap->mapwindow[current_cpu_datap()->cpu_pmap->pde_window_index].prv_CMAP), + newpf | INTEL_PTE_RW | INTEL_PTE_VALID); + invlpg((u_int)(current_cpu_datap()->cpu_pmap->mapwindow[current_cpu_datap()->cpu_pmap->pde_window_index].prv_CADDR)); + return ((pd_entry_t *)(current_cpu_datap()->cpu_pmap->mapwindow[current_cpu_datap()->cpu_pmap->pde_window_index].prv_CADDR) + + ((vm_offset_t)((vaddr>>PDSHIFT)&(NPDPG-1)))); } - if (ctu != ptep->use_count || ctw != ptep->wired_count) { - printf("use %d wired %d - actual use %d wired %d\n", - ptep->use_count, ptep->wired_count, ctu, ctw); - panic("pte count"); + return (0); +} + + + +/* + * return address of mapped pte for vaddr va in pmap pmap. + * must be called with pre-emption or interrupts disabled + * if targeted pmap is not the kernel pmap + * since we may be passing back a virtual address that is + * associated with this cpu... pre-emption or interrupts + * must remain disabled until the caller is done using + * the pointer that was passed back . + * + * maps the pde page, if any, containing the pte in and returns + * the address of the pte in that mapped page + */ +pt_entry_t * +pmap_pte(pmap_t pmap, vm_map_offset_t vaddr) +{ + pd_entry_t *pde; + pd_entry_t newpf; + int i; + + assert(pmap); + pde = pmap_pde(pmap,vaddr); + + if (pde && ((*pde & INTEL_PTE_VALID))) { + if (pmap == kernel_pmap) { + return (vtopte(vaddr)); /* compat kernel still has pte's mapped */ + } + + assert(ml_get_interrupts_enabled() == 0 || get_preemption_level() != 0); + + newpf = *pde & PG_FRAME; + + for (i=PMAP_PTE_FIRST_WINDOW; i < PMAP_PTE_FIRST_WINDOW+PMAP_PTE_NWINDOWS; i++) { + if (((*(current_cpu_datap()->cpu_pmap->mapwindow[i].prv_CMAP)) & PG_FRAME) == newpf) { + return((pt_entry_t *)(current_cpu_datap()->cpu_pmap->mapwindow[i].prv_CADDR) + + ((vm_offset_t)i386_btop(vaddr) & (NPTEPG-1))); + } + } + + current_cpu_datap()->cpu_pmap->pte_window_index++; + if (current_cpu_datap()->cpu_pmap->pte_window_index > (PMAP_PTE_FIRST_WINDOW+PMAP_PTE_NWINDOWS-1)) + current_cpu_datap()->cpu_pmap->pte_window_index = PMAP_PTE_FIRST_WINDOW; + pmap_store_pte( + (current_cpu_datap()->cpu_pmap->mapwindow[current_cpu_datap()->cpu_pmap->pte_window_index].prv_CMAP), + newpf | INTEL_PTE_RW | INTEL_PTE_VALID); + invlpg((u_int)(current_cpu_datap()->cpu_pmap->mapwindow[current_cpu_datap()->cpu_pmap->pte_window_index].prv_CADDR)); + return ((pt_entry_t *)(current_cpu_datap()->cpu_pmap->mapwindow[current_cpu_datap()->cpu_pmap->pte_window_index].prv_CADDR) + + ((vm_offset_t)i386_btop(vaddr) & (NPTEPG-1))); } + + return(0); } -#endif /* DEBUG_PTE_PAGE */ + /* * Map memory at initialization. The physical addresses being @@ -619,17 +673,18 @@ ptep_check( */ vm_offset_t pmap_map( - register vm_offset_t virt, - register vm_offset_t start_addr, - register vm_offset_t end_addr, - register vm_prot_t prot) + vm_offset_t virt, + vm_map_offset_t start_addr, + vm_map_offset_t end_addr, + vm_prot_t prot, + unsigned int flags) { - register int ps; + int ps; ps = PAGE_SIZE; while (start_addr < end_addr) { - pmap_enter(kernel_pmap, - virt, (ppnum_t) i386_btop(start_addr), prot, 0, FALSE); + pmap_enter(kernel_pmap, (vm_map_offset_t)virt, + (ppnum_t) i386_btop(start_addr), prot, flags, FALSE); virt += ps; start_addr += ps; } @@ -645,30 +700,36 @@ pmap_map( */ vm_offset_t pmap_map_bd( - register vm_offset_t virt, - register vm_offset_t start_addr, - register vm_offset_t end_addr, - vm_prot_t prot) + vm_offset_t virt, + vm_map_offset_t start_addr, + vm_map_offset_t end_addr, + vm_prot_t prot, + unsigned int flags) { - register pt_entry_t template; - register pt_entry_t *pte; + pt_entry_t template; + pt_entry_t *pte; template = pa_to_pte(start_addr) - | INTEL_PTE_NCACHE | INTEL_PTE_REF | INTEL_PTE_MOD | INTEL_PTE_WIRED | INTEL_PTE_VALID; + + if(flags & (VM_MEM_NOT_CACHEABLE | VM_WIMG_USE_DEFAULT)) { + template |= INTEL_PTE_NCACHE; + if(!(flags & (VM_MEM_GUARDED | VM_WIMG_USE_DEFAULT))) + template |= INTEL_PTE_PTA; + } + if (prot & VM_PROT_WRITE) template |= INTEL_PTE_WRITE; - /* XXX move pmap_pte out of loop, once one pte mapped, all are */ while (start_addr < end_addr) { - pte = pmap_pte(kernel_pmap, virt); + pte = pmap_pte(kernel_pmap, (vm_map_offset_t)virt); if (pte == PT_ENTRY_NULL) { panic("pmap_map_bd: Invalid kernel address\n"); } - WRITE_PTE_FAST(pte, template) + pmap_store_pte(pte, template); pte_increment_pa(template); virt += PAGE_SIZE; start_addr += PAGE_SIZE; @@ -685,6 +746,142 @@ extern vm_offset_t etext; extern void *sectHIBB; extern int sectSizeHIB; + +vm_offset_t +pmap_high_shared_remap(enum high_fixed_addresses e, vm_offset_t va, int sz) +{ + vm_offset_t ve = pmap_index_to_virt(e); + pt_entry_t *ptep; + pmap_paddr_t pa; + int i; + + assert(0 == (va & PAGE_MASK)); /* expecting page aligned */ + ptep = pmap_pte(kernel_pmap, (vm_map_offset_t)ve); + + for (i=0; i< sz; i++) { + pa = (pmap_paddr_t) kvtophys(va); + pmap_store_pte(ptep, (pa & PG_FRAME) + | INTEL_PTE_VALID + | INTEL_PTE_GLOBAL + | INTEL_PTE_RW + | INTEL_PTE_REF + | INTEL_PTE_MOD); + va+= PAGE_SIZE; + ptep++; + } + return ve; +} + +vm_offset_t +pmap_cpu_high_shared_remap(int cpu, enum high_cpu_types e, vm_offset_t va, int sz) +{ + enum high_fixed_addresses a = e + HIGH_CPU_END * cpu; + return pmap_high_shared_remap(HIGH_FIXED_CPUS_BEGIN + a, va, sz); +} + +void pmap_init_high_shared(void); + +extern vm_offset_t gdtptr, idtptr; + +extern uint32_t low_intstack; + +extern struct fake_descriptor ldt_desc_pattern; +extern struct fake_descriptor tss_desc_pattern; + +extern char hi_remap_text, hi_remap_etext; +extern char t_zero_div; + +pt_entry_t *pte_unique_base; + +void +pmap_init_high_shared(void) +{ + + vm_offset_t haddr; + struct __gdt_desc_struct gdt_desc = {0,0,0}; + struct __idt_desc_struct idt_desc = {0,0,0}; +#if MACH_KDB + struct i386_tss *ttss; +#endif + + kprintf("HIGH_MEM_BASE 0x%x fixed per-cpu begin 0x%x\n", + HIGH_MEM_BASE,pmap_index_to_virt(HIGH_FIXED_CPUS_BEGIN)); + pte_unique_base = pmap_pte(kernel_pmap, (vm_map_offset_t)pmap_index_to_virt(HIGH_FIXED_CPUS_BEGIN)); + + if (i386_btop(&hi_remap_etext - &hi_remap_text + 1) > + HIGH_FIXED_TRAMPS_END - HIGH_FIXED_TRAMPS + 1) + panic("tramps too large"); + haddr = pmap_high_shared_remap(HIGH_FIXED_TRAMPS, + (vm_offset_t) &hi_remap_text, 3); + kprintf("tramp: 0x%x, ",haddr); + printf("hi mem tramps at 0x%x\n",haddr); + /* map gdt up high and update ptr for reload */ + haddr = pmap_high_shared_remap(HIGH_FIXED_GDT, + (vm_offset_t) master_gdt, 1); + __asm__ __volatile__("sgdt %0": "=m" (gdt_desc): :"memory"); + gdt_desc.address = haddr; + kprintf("GDT: 0x%x, ",haddr); + /* map ldt up high */ + haddr = pmap_high_shared_remap(HIGH_FIXED_LDT_BEGIN, + (vm_offset_t) master_ldt, + HIGH_FIXED_LDT_END - HIGH_FIXED_LDT_BEGIN + 1); + kprintf("LDT: 0x%x, ",haddr); + /* put new ldt addr into gdt */ + master_gdt[sel_idx(KERNEL_LDT)] = ldt_desc_pattern; + master_gdt[sel_idx(KERNEL_LDT)].offset = (vm_offset_t) haddr; + fix_desc(&master_gdt[sel_idx(KERNEL_LDT)], 1); + master_gdt[sel_idx(USER_LDT)] = ldt_desc_pattern; + master_gdt[sel_idx(USER_LDT)].offset = (vm_offset_t) haddr; + fix_desc(&master_gdt[sel_idx(USER_LDT)], 1); + + /* map idt up high */ + haddr = pmap_high_shared_remap(HIGH_FIXED_IDT, + (vm_offset_t) master_idt, 1); + __asm__ __volatile__("sidt %0" : "=m" (idt_desc)); + idt_desc.address = haddr; + kprintf("IDT: 0x%x, ", haddr); + /* remap ktss up high and put new high addr into gdt */ + haddr = pmap_high_shared_remap(HIGH_FIXED_KTSS, + (vm_offset_t) &master_ktss, 1); + master_gdt[sel_idx(KERNEL_TSS)] = tss_desc_pattern; + master_gdt[sel_idx(KERNEL_TSS)].offset = (vm_offset_t) haddr; + fix_desc(&master_gdt[sel_idx(KERNEL_TSS)], 1); + kprintf("KTSS: 0x%x, ",haddr); +#if MACH_KDB + /* remap dbtss up high and put new high addr into gdt */ + haddr = pmap_high_shared_remap(HIGH_FIXED_DBTSS, + (vm_offset_t) &master_dbtss, 1); + master_gdt[sel_idx(DEBUG_TSS)] = tss_desc_pattern; + master_gdt[sel_idx(DEBUG_TSS)].offset = (vm_offset_t) haddr; + fix_desc(&master_gdt[sel_idx(DEBUG_TSS)], 1); + ttss = (struct i386_tss *)haddr; + kprintf("DBTSS: 0x%x, ",haddr); +#endif /* MACH_KDB */ + + /* remap dftss up high and put new high addr into gdt */ + haddr = pmap_high_shared_remap(HIGH_FIXED_DFTSS, + (vm_offset_t) &master_dftss, 1); + master_gdt[sel_idx(DF_TSS)] = tss_desc_pattern; + master_gdt[sel_idx(DF_TSS)].offset = (vm_offset_t) haddr; + fix_desc(&master_gdt[sel_idx(DF_TSS)], 1); + kprintf("DFTSS: 0x%x\n",haddr); + + /* remap mctss up high and put new high addr into gdt */ + haddr = pmap_high_shared_remap(HIGH_FIXED_DFTSS, + (vm_offset_t) &master_mctss, 1); + master_gdt[sel_idx(MC_TSS)] = tss_desc_pattern; + master_gdt[sel_idx(MC_TSS)].offset = (vm_offset_t) haddr; + fix_desc(&master_gdt[sel_idx(MC_TSS)], 1); + kprintf("MCTSS: 0x%x\n",haddr); + + __asm__ __volatile__("lgdt %0": "=m" (gdt_desc)); + __asm__ __volatile__("lidt %0": "=m" (idt_desc)); + kprintf("gdt/idt reloaded, "); + set_tr(KERNEL_TSS); + kprintf("tr reset to KERNEL_TSS\n"); +} + + /* * Bootstrap the system enough to run with virtual memory. * Map the kernel's code and data, and allocate the system page table. @@ -706,50 +903,60 @@ extern int sectSizeHIB; void pmap_bootstrap( - __unused vm_offset_t load_start) + __unused vm_offset_t load_start, + boolean_t IA32e) { vm_offset_t va; pt_entry_t *pte; int i; int wpkernel, boot_arg; + pdpt_entry_t *pdpt; vm_last_addr = VM_MAX_KERNEL_ADDRESS; /* Set the highest address * known to VM */ - /* * The kernel's pmap is statically allocated so we don't * have to use pmap_create, which is unlikely to work * correctly at this part of the boot sequence. */ + kernel_pmap = &kernel_pmap_store; -#ifdef PMAP_QUEUE - kernel_pmap->pmap_link.next = (queue_t)kernel_pmap; /* Set up anchor forward */ - kernel_pmap->pmap_link.prev = (queue_t)kernel_pmap; /* Set up anchor reverse */ -#endif kernel_pmap->ref_count = 1; + kernel_pmap->nx_enabled = FALSE; + kernel_pmap->pm_64bit = 0; kernel_pmap->pm_obj = (vm_object_t) NULL; kernel_pmap->dirbase = (pd_entry_t *)((unsigned int)IdlePTD | KERNBASE); - kernel_pmap->pdirbase = (pd_entry_t *)IdlePTD; -#ifdef PAE - kernel_pmap->pm_pdpt = (pd_entry_t *)((unsigned int)IdlePDPT | KERNBASE ); - kernel_pmap->pm_ppdpt = (vm_offset_t)IdlePDPT; -#endif + kernel_pmap->pdirbase = (pmap_paddr_t)((int)IdlePTD); + pdpt = (pd_entry_t *)((unsigned int)IdlePDPT | KERNBASE ); + kernel_pmap->pm_pdpt = pdpt; + kernel_pmap->pm_cr3 = (pmap_paddr_t)((int)IdlePDPT); va = (vm_offset_t)kernel_pmap->dirbase; /* setup self referential mapping(s) */ - for (i = 0; i< NPGPTD; i++ ) { + for (i = 0; i< NPGPTD; i++, pdpt++) { pmap_paddr_t pa; pa = (pmap_paddr_t) kvtophys(va + i386_ptob(i)); - * (pd_entry_t *) (kernel_pmap->dirbase + PTDPTDI + i) = + pmap_store_pte( + (pd_entry_t *) (kernel_pmap->dirbase + PTDPTDI + i), (pa & PG_FRAME) | INTEL_PTE_VALID | INTEL_PTE_RW | INTEL_PTE_REF | - INTEL_PTE_MOD | INTEL_PTE_WIRED ; -#ifdef PAE - kernel_pmap->pm_pdpt[i] = pa | INTEL_PTE_VALID; -#endif + INTEL_PTE_MOD | INTEL_PTE_WIRED) ; + pmap_store_pte(pdpt, pa | INTEL_PTE_VALID); } + cpu_64bit = IA32e; + + lo_kernel_cr3 = kernel_pmap->pm_cr3; + current_cpu_datap()->cpu_kernel_cr3 = (addr64_t) kernel_pmap->pm_cr3; + + /* save the value we stuff into created pmaps to share the gdts etc */ + high_shared_pde = *pmap_pde(kernel_pmap, HIGH_MEM_BASE); + /* make sure G bit is on for high shared pde entry */ + high_shared_pde |= INTEL_PTE_GLOBAL; + pmap_store_pte(pmap_pde(kernel_pmap, HIGH_MEM_BASE), high_shared_pde); + nkpt = NKPT; + inuse_ptepages_count += NKPT; virtual_avail = (vm_offset_t)VADDR(KPTDI,0) + (vm_offset_t)first_avail; virtual_end = (vm_offset_t)(VM_MAX_KERNEL_ADDRESS); @@ -759,25 +966,18 @@ pmap_bootstrap( * mapping of pages. */ #define SYSMAP(c, p, v, n) \ - v = (c)va; va += ((n)*INTEL_PGBYTES); p = pte; pte += (n); + v = (c)va; va += ((n)*INTEL_PGBYTES); p = pte; pte += (n) va = virtual_avail; - pte = (pt_entry_t *) pmap_pte(kernel_pmap, va); - - /* - * CMAP1/CMAP2 are used for zeroing and copying pages. - * CMAP3 is used for ml_phys_read/write. - */ - SYSMAP(caddr_t, CM1, CA1, 1) - * (pt_entry_t *) CM1 = 0; - SYSMAP(caddr_t, CM2, CA2, 1) - * (pt_entry_t *) CM2 = 0; - SYSMAP(caddr_t, CM3, CA3, 1) - * (pt_entry_t *) CM3 = 0; + pte = vtopte(va); - /* used by pmap_pte */ - SYSMAP(caddr_t, CM4, CA4, 1) - * (pt_entry_t *) CM4 = 0; + for (i=0; icpu_pmap->mapwindow[i].prv_CMAP), + (current_cpu_datap()->cpu_pmap->mapwindow[i].prv_CADDR), + 1); + *current_cpu_datap()->cpu_pmap->mapwindow[i].prv_CMAP = 0; + } /* DMAP user for debugger */ SYSMAP(caddr_t, DMAP1, DADDR1, 1); @@ -791,45 +991,109 @@ pmap_bootstrap( virtual_avail = va; wpkernel = 1; - if (PE_parse_boot_arg("debug", &boot_arg)) { - if (boot_arg & DB_PRT) wpkernel = 0; - if (boot_arg & DB_NMI) wpkernel = 0; + if (PE_parse_boot_arg("wpkernel", &boot_arg)) { + if (boot_arg == 0) + wpkernel = 0; } - /* remap kernel text readonly if not debugging or kprintfing */ + /* Remap kernel text readonly unless the "wpkernel" boot-arg is present + * and set to 0. + */ if (wpkernel) { vm_offset_t myva; pt_entry_t *ptep; - for (myva = i386_round_page(VM_MIN_KERNEL_ADDRESS + MP_BOOT + MP_BOOTSTACK); myva < etext; myva += PAGE_SIZE) { + for (myva = i386_round_page(MP_BOOT + MP_BOOTSTACK); myva < etext; myva += PAGE_SIZE) { if (myva >= (vm_offset_t)sectHIBB && myva < ((vm_offset_t)sectHIBB + sectSizeHIB)) continue; - ptep = pmap_pte(kernel_pmap, myva); + ptep = pmap_pte(kernel_pmap, (vm_map_offset_t)myva); if (ptep) - *ptep &= ~INTEL_PTE_RW; + pmap_store_pte(ptep, *ptep & ~INTEL_PTE_RW); } - flush_tlb(); } + /* no matter what, kernel page zero is not accessible */ + pte = pmap_pte(kernel_pmap, 0); + pmap_store_pte(pte, INTEL_PTE_INVALID); + + /* map lowmem global page into fixed addr 0x2000 */ + if (0 == (pte = pmap_pte(kernel_pmap,0x2000))) panic("lowmem pte"); + + pmap_store_pte(pte, kvtophys((vm_offset_t)&lowGlo)|INTEL_PTE_VALID|INTEL_PTE_REF|INTEL_PTE_MOD|INTEL_PTE_WIRED|INTEL_PTE_RW); + flush_tlb(); + simple_lock_init(&kernel_pmap->lock, 0); simple_lock_init(&pv_free_list_lock, 0); - /* invalidate user virtual addresses */ - memset((char *)kernel_pmap->dirbase, - 0, - (KPTDI) * sizeof(pd_entry_t)); + pmap_init_high_shared(); + + pde_mapped_size = PDE_MAPPED_SIZE; + + if (cpu_64bit) { + pdpt_entry_t *ppdpt = (pdpt_entry_t *)IdlePDPT; + pdpt_entry_t *ppdpt64 = (pdpt_entry_t *)IdlePDPT64; + pdpt_entry_t *ppml4 = (pdpt_entry_t *)IdlePML4; + int istate = ml_set_interrupts_enabled(FALSE); + + /* + * Clone a new 64-bit 3rd-level page table directory, IdlePML4, + * with page bits set for the correct IA-32e operation and so that + * the legacy-mode IdlePDPT is retained for slave processor start-up. + * This is necessary due to the incompatible use of page bits between + * 64-bit and legacy modes. + */ + kernel_pmap->pm_cr3 = (pmap_paddr_t)((int)IdlePML4); /* setup in start.s for us */ + kernel_pmap->pm_pml4 = IdlePML4; + kernel_pmap->pm_pdpt = (pd_entry_t *) + ((unsigned int)IdlePDPT64 | KERNBASE ); +#define PAGE_BITS INTEL_PTE_VALID|INTEL_PTE_RW|INTEL_PTE_USER|INTEL_PTE_REF + pmap_store_pte(kernel_pmap->pm_pml4, + (uint32_t)IdlePDPT64 | PAGE_BITS); + pmap_store_pte((ppdpt64+0), *(ppdpt+0) | PAGE_BITS); + pmap_store_pte((ppdpt64+1), *(ppdpt+1) | PAGE_BITS); + pmap_store_pte((ppdpt64+2), *(ppdpt+2) | PAGE_BITS); + pmap_store_pte((ppdpt64+3), *(ppdpt+3) | PAGE_BITS); + + /* + * The kernel is also mapped in the uber-sapce at the 4GB starting + * 0xFFFFFF80:00000000. This is the highest entry in the 4th-level. + */ + pmap_store_pte((ppml4+KERNEL_UBER_PML4_INDEX), *(ppml4+0)); + + kernel64_cr3 = (addr64_t) kernel_pmap->pm_cr3; + cpu_IA32e_enable(current_cpu_datap()); + current_cpu_datap()->cpu_is64bit = TRUE; + /* welcome to a 64 bit world */ + + /* Re-initialize and load descriptors */ + cpu_desc_init64(&cpu_data_master, TRUE); + cpu_desc_load64(&cpu_data_master); + fast_syscall_init64(); + + pde_mapped_size = 512*4096 ; + + ml_set_interrupts_enabled(istate); + + } + kernel_pmap->pm_hold = (vm_offset_t)kernel_pmap->pm_pml4; kprintf("Kernel virtual space from 0x%x to 0x%x.\n", VADDR(KPTDI,0), virtual_end); -#ifdef PAE - kprintf("Available physical space from 0x%llx to 0x%llx\n", - avail_start, avail_end); printf("PAE enabled\n"); -#else - kprintf("Available physical space from 0x%x to 0x%x\n", + if (cpu_64bit){ + printf("64 bit mode enabled\n");kprintf("64 bit mode enabled\n"); } + + kprintf("Available physical space from 0x%llx to 0x%llx\n", avail_start, avail_end); -#endif + + /* + * By default for 64-bit users loaded at 4GB, share kernel mapping. + * But this may be overridden by the -no_shared_cr3 boot-arg. + */ + if (PE_parse_boot_arg("-no_shared_cr3", &no_shared_cr3)) { + kprintf("Shared kernel address space disabled\n"); + } } void @@ -852,8 +1116,8 @@ pmap_init(void) register long npages; vm_offset_t addr; register vm_size_t s; - vm_offset_t vaddr; - ppnum_t ppn; + vm_map_offset_t vaddr; + ppnum_t ppn; /* * Allocate memory for the pv_head_table and its lock bits, @@ -892,11 +1156,8 @@ pmap_init(void) pmap_zone = zinit(s, 400*s, 4096, "pmap"); /* XXX */ s = (vm_size_t) sizeof(struct pv_entry); pv_list_zone = zinit(s, 10000*s, 4096, "pv_list"); /* XXX */ -#ifdef PAE - // s = (vm_size_t) (sizeof(pdpt_entry_t) * NPGPTD); s = 63; pdpt_zone = zinit(s, 400*s, 4096, "pdpt"); /* XXX */ -#endif /* * Only now, when all of the data structures are allocated, @@ -909,23 +1170,22 @@ pmap_init(void) vm_first_phys = 0; vm_last_phys = avail_end; -#if GROW_KERNEL_FUNCTION_IMPLEMENTED kptobj = &kptobj_object_store; _vm_object_allocate((vm_object_size_t)NKPDE, kptobj); kernel_pmap->pm_obj = kptobj; -#endif /* create pv entries for kernel pages mapped by low level startup code. these have to exist so we can pmap_remove() e.g. kext pages from the middle of our addr space */ - vaddr = (vm_offset_t)VM_MIN_KERNEL_ADDRESS; + vaddr = (vm_map_offset_t)0; for (ppn = 0; ppn < i386_btop(avail_start) ; ppn++ ) { pv_entry_t pv_e; pv_e = pai_to_pvh(ppn); pv_e->va = vaddr; vaddr += PAGE_SIZE; + kernel_pmap->stats.resident_count++; pv_e->pmap = kernel_pmap; pv_e->next = PV_ENTRY_NULL; } @@ -938,10 +1198,6 @@ pmap_init(void) pmap_cache_list = PMAP_NULL; pmap_cache_count = 0; simple_lock_init(&pmap_cache_lock, 0); -#ifdef PMAP_QUEUE - simple_lock_init(&free_pmap_lock, 0); -#endif - } void @@ -952,7 +1208,7 @@ x86_lowmem_free(void) the actual pages that are released are determined by which pages the memory sizing code puts into the region table */ - ml_static_mfree((vm_offset_t) i386_ptob(pmap_memory_regions[0].base)|VM_MIN_KERNEL_ADDRESS, + ml_static_mfree((vm_offset_t) i386_ptob(pmap_memory_regions[0].base), (vm_size_t) i386_ptob(pmap_memory_regions[0].end - pmap_memory_regions[0].base)); } @@ -1002,15 +1258,20 @@ pmap_verify_free( */ pmap_t pmap_create( - vm_size_t size) + vm_map_size_t sz, + boolean_t is_64bit) { - register pmap_t p; -#ifdef PMAP_QUEUE - register pmap_t pro; - spl_t s; -#endif - register int i; - register vm_offset_t va; + register pmap_t p; + int i; + vm_offset_t va; + vm_size_t size; + pdpt_entry_t *pdpt; + pml4_entry_t *pml4p; + int template; + pd_entry_t *pdp; + spl_t s; + + size = (vm_size_t) sz; /* * A software use-only map doesn't even need a map. @@ -1023,61 +1284,168 @@ pmap_create( p = (pmap_t) zalloc(pmap_zone); if (PMAP_NULL == p) panic("pmap_create zalloc"); - if (KERN_SUCCESS != kmem_alloc_wired(kernel_map, (vm_offset_t *)(&p->dirbase), NBPTD)) - panic("pmap_create kmem_alloc_wired"); -#ifdef PAE - p->pm_hold = (vm_offset_t)zalloc(pdpt_zone); - if ((vm_offset_t)NULL == p->pm_hold) { - panic("pdpt zalloc"); - } - p->pm_pdpt = (pdpt_entry_t *) (( p->pm_hold + 31) & ~31); - p->pm_ppdpt = kvtophys((vm_offset_t)p->pm_pdpt); /* XXX */ -#endif - if (NULL == (p->pm_obj = vm_object_allocate((vm_object_size_t)(NPGPTD*NPDEPG)))) - panic("pmap_create vm_object_allocate"); - memcpy(p->dirbase, - (void *)((unsigned int)IdlePTD | KERNBASE), - NBPTD); - va = (vm_offset_t)p->dirbase; - p->pdirbase = (pd_entry_t *)(kvtophys(va)); - simple_lock_init(&p->lock, 0); - - /* setup self referential mapping(s) */ - for (i = 0; i< NPGPTD; i++ ) { - pmap_paddr_t pa; - pa = (pmap_paddr_t) kvtophys(va + i386_ptob(i)); - * (pd_entry_t *) (p->dirbase + PTDPTDI + i) = - (pa & PG_FRAME) | INTEL_PTE_VALID | INTEL_PTE_RW | INTEL_PTE_REF | - INTEL_PTE_MOD | INTEL_PTE_WIRED ; -#ifdef PAE - p->pm_pdpt[i] = pa | INTEL_PTE_VALID; -#endif - } - p->cpus_using = 0; + /* init counts now since we'll be bumping some */ + simple_lock_init(&p->lock, 0); p->stats.resident_count = 0; p->stats.wired_count = 0; p->ref_count = 1; + p->nx_enabled = 1; + p->pm_64bit = is_64bit; + p->pm_kernel_cr3 = FALSE; + p->pm_shared = FALSE; + + if (!cpu_64bit) { + /* legacy 32 bit setup */ + /* in the legacy case the pdpt layer is hardwired to 4 entries and each + * entry covers 1GB of addr space */ + if (KERN_SUCCESS != kmem_alloc_wired(kernel_map, (vm_offset_t *)(&p->dirbase), NBPTD)) + panic("pmap_create kmem_alloc_wired"); + p->pm_hold = (vm_offset_t)zalloc(pdpt_zone); + if ((vm_offset_t)NULL == p->pm_hold) { + panic("pdpt zalloc"); + } + pdpt = (pdpt_entry_t *) (( p->pm_hold + 31) & ~31); + p->pm_cr3 = (pmap_paddr_t)kvtophys((vm_offset_t)pdpt); + if (NULL == (p->pm_obj = vm_object_allocate((vm_object_size_t)(NPGPTD*NPTDPG)))) + panic("pmap_create vm_object_allocate"); + + memset((char *)p->dirbase, 0, NBPTD); + + va = (vm_offset_t)p->dirbase; + p->pdirbase = kvtophys(va); + + template = cpu_64bit ? INTEL_PTE_VALID|INTEL_PTE_RW|INTEL_PTE_USER|INTEL_PTE_REF : INTEL_PTE_VALID; + for (i = 0; i< NPGPTD; i++, pdpt++) { + pmap_paddr_t pa; + pa = (pmap_paddr_t) kvtophys(va + i386_ptob(i)); + pmap_store_pte(pdpt, pa | template); + } + + /* map the high shared pde */ + pmap_store_pte(pmap_pde(p, HIGH_MEM_BASE), high_shared_pde); -#ifdef PMAP_QUEUE - /* insert new pmap at head of queue hanging off kernel_pmap */ - SPLVM(s); - simple_lock(&free_pmap_lock); - p->pmap_link.next = (queue_t)kernel_pmap->pmap_link.next; - kernel_pmap->pmap_link.next = (queue_t)p; + } else { - pro = (pmap_t) p->pmap_link.next; - p->pmap_link.prev = (queue_t)pro->pmap_link.prev; - pro->pmap_link.prev = (queue_t)p; + /* 64 bit setup */ - - simple_unlock(&free_pmap_lock); - SPLX(s); -#endif + /* alloc the pml4 page in kernel vm */ + if (KERN_SUCCESS != kmem_alloc_wired(kernel_map, (vm_offset_t *)(&p->pm_hold), PAGE_SIZE)) + panic("pmap_create kmem_alloc_wired pml4"); + + memset((char *)p->pm_hold, 0, PAGE_SIZE); + p->pm_cr3 = (pmap_paddr_t)kvtophys((vm_offset_t)p->pm_hold); + + inuse_ptepages_count++; + p->stats.resident_count++; + p->stats.wired_count++; + + /* allocate the vm_objs to hold the pdpt, pde and pte pages */ + + if (NULL == (p->pm_obj_pml4 = vm_object_allocate((vm_object_size_t)(NPML4PGS)))) + panic("pmap_create pdpt obj"); + + if (NULL == (p->pm_obj_pdpt = vm_object_allocate((vm_object_size_t)(NPDPTPGS)))) + panic("pmap_create pdpt obj"); + + if (NULL == (p->pm_obj = vm_object_allocate((vm_object_size_t)(NPDEPGS)))) + panic("pmap_create pte obj"); + + /* uber space points to uber mapped kernel */ + s = splhigh(); + pml4p = pmap64_pml4(p, 0ULL); + pmap_store_pte((pml4p+KERNEL_UBER_PML4_INDEX),*kernel_pmap->pm_pml4); + if (!is_64bit) { + while ((pdp = pmap64_pde(p, (uint64_t)HIGH_MEM_BASE)) == PD_ENTRY_NULL) { + splx(s); + pmap_expand_pdpt(p, (uint64_t)HIGH_MEM_BASE); /* need room for another pde entry */ + s = splhigh(); + } + pmap_store_pte(pdp, high_shared_pde); + } + + splx(s); + } return(p); } +void +pmap_set_4GB_pagezero(pmap_t p) +{ + int spl; + pdpt_entry_t *user_pdptp; + pdpt_entry_t *kern_pdptp; + + assert(p->pm_64bit); + + /* Kernel-shared cr3 may be disabled by boot arg. */ + if (no_shared_cr3) + return; + + /* + * Set the bottom 4 3rd-level pte's to be the kernel's. + */ + spl = splhigh(); + while ((user_pdptp = pmap64_pdpt(p, 0x0)) == PDPT_ENTRY_NULL) { + splx(spl); + pmap_expand_pml4(p, 0x0); + spl = splhigh(); + } + kern_pdptp = kernel_pmap->pm_pdpt; + pmap_store_pte(user_pdptp+0, *(kern_pdptp+0)); + pmap_store_pte(user_pdptp+1, *(kern_pdptp+1)); + pmap_store_pte(user_pdptp+2, *(kern_pdptp+2)); + pmap_store_pte(user_pdptp+3, *(kern_pdptp+3)); + + p->pm_kernel_cr3 = TRUE; + + splx(spl); + +} + +void +pmap_load_kernel_cr3(void) +{ + uint32_t kernel_cr3; + + assert(!ml_get_interrupts_enabled()); + + /* + * Reload cr3 with the true kernel cr3. + * Note: kernel's pml4 resides below 4GB physical. + */ + kernel_cr3 = current_cpu_datap()->cpu_kernel_cr3; + set_cr3(kernel_cr3); + current_cpu_datap()->cpu_active_cr3 = kernel_cr3; + current_cpu_datap()->cpu_task_map = TASK_MAP_32BIT; + current_cpu_datap()->cpu_tlb_invalid = FALSE; + __asm__ volatile("mfence"); +} + +void +pmap_clear_4GB_pagezero(pmap_t p) +{ + int spl; + pdpt_entry_t *user_pdptp; + + if (!p->pm_kernel_cr3) + return; + + spl = splhigh(); + user_pdptp = pmap64_pdpt(p, 0x0); + pmap_store_pte(user_pdptp+0, 0); + pmap_store_pte(user_pdptp+1, 0); + pmap_store_pte(user_pdptp+2, 0); + pmap_store_pte(user_pdptp+3, 0); + + p->pm_kernel_cr3 = FALSE; + + pmap_load_kernel_cr3(); + + splx(spl); +} + /* * Retire the given physical map from service. * Should only be called if the map contains @@ -1088,47 +1456,29 @@ void pmap_destroy( register pmap_t p) { - register pt_entry_t *pdep; register int c; spl_t s; +#if 0 + register pt_entry_t *pdep; register vm_page_t m; -#ifdef PMAP_QUEUE - register pmap_t pre,pro; #endif if (p == PMAP_NULL) return; - SPLVM(s); simple_lock(&p->lock); c = --p->ref_count; if (c == 0) { - register int my_cpu; - - mp_disable_preemption(); - my_cpu = cpu_number(); - /* * If some cpu is not using the physical pmap pointer that it * is supposed to be (see set_dirbase), we might be using the * pmap that is being destroyed! Make sure we are * physically on the right pmap: */ - /* force pmap/cr3 update */ PMAP_UPDATE_TLBS(p, VM_MIN_ADDRESS, VM_MAX_KERNEL_ADDRESS); - if (PMAP_REAL(my_cpu) == p) { - PMAP_CPU_CLR(p, my_cpu); - PMAP_REAL(my_cpu) = kernel_pmap; -#ifdef PAE - set_cr3((unsigned int)kernel_pmap->pm_ppdpt); -#else - set_cr3((unsigned int)kernel_pmap->pdirbase); -#endif - } - mp_enable_preemption(); } simple_unlock(&p->lock); SPLX(s); @@ -1137,31 +1487,21 @@ pmap_destroy( return; /* still in use */ } -#ifdef PMAP_QUEUE - /* remove from pmap queue */ - SPLVM(s); - simple_lock(&free_pmap_lock); - - pre = (pmap_t)p->pmap_link.prev; - pre->pmap_link.next = (queue_t)p->pmap_link.next; - pro = (pmap_t)p->pmap_link.next; - pro->pmap_link.prev = (queue_t)p->pmap_link.prev; - - simple_unlock(&free_pmap_lock); - SPLX(s); -#endif - /* * Free the memory maps, then the * pmap structure. */ + if (!cpu_64bit) { +#if 0 pdep = (pt_entry_t *)p->dirbase; while (pdep < (pt_entry_t *)&p->dirbase[(UMAXPTDI+1)]) { - int ind; + int ind; + if (*pdep & INTEL_PTE_VALID) { - ind = pdep - (pt_entry_t *)&p->dirbase[0]; + ind = pdep - (pt_entry_t *)&p->dirbase[0]; + vm_object_lock(p->pm_obj); m = vm_page_lookup(p->pm_obj, (vm_object_offset_t)ind); if (m == VM_PAGE_NULL) { @@ -1170,25 +1510,49 @@ pmap_destroy( vm_page_lock_queues(); vm_page_free(m); inuse_ptepages_count--; + vm_object_unlock(p->pm_obj); vm_page_unlock_queues(); /* * Clear pdes, this might be headed for the cache. */ - *pdep++ = 0; + pmap_store_pte(pdep, 0); + pdep++; } else { - *pdep++ = 0; + pmap_store_pte(pdep, 0); + pdep++; } } - - vm_object_deallocate(p->pm_obj); - kmem_free(kernel_map, (vm_offset_t)p->dirbase, NBPTD); -#ifdef PAE - zfree(pdpt_zone, (void *)p->pm_hold); +#else + inuse_ptepages_count -= p->pm_obj->resident_page_count; #endif + vm_object_deallocate(p->pm_obj); + kmem_free(kernel_map, (vm_offset_t)p->dirbase, NBPTD); + zfree(pdpt_zone, (void *)p->pm_hold); + } else { + + /* 64 bit */ + + pmap_unmap_sharedpage(p); + + /* free 64 bit mode structs */ + inuse_ptepages_count--; + kmem_free(kernel_map, (vm_offset_t)p->pm_hold, PAGE_SIZE); + + inuse_ptepages_count -= p->pm_obj_pml4->resident_page_count; + vm_object_deallocate(p->pm_obj_pml4); + + inuse_ptepages_count -= p->pm_obj_pdpt->resident_page_count; + vm_object_deallocate(p->pm_obj_pdpt); + + inuse_ptepages_count -= p->pm_obj->resident_page_count; + vm_object_deallocate(p->pm_obj); + + } + zfree(pmap_zone, p); } @@ -1223,34 +1587,33 @@ pmap_reference( * Assumes that the pte-page exists. */ -/* static */ -void +static void pmap_remove_range( pmap_t pmap, - vm_offset_t va, + vm_map_offset_t start_vaddr, pt_entry_t *spte, pt_entry_t *epte) { register pt_entry_t *cpte; - int num_removed, num_unwired; + int num_removed, num_unwired, num_found; int pai; pmap_paddr_t pa; + vm_map_offset_t vaddr; -#if DEBUG_PTE_PAGE - if (pmap != kernel_pmap) - ptep_check(get_pte_page(spte)); -#endif /* DEBUG_PTE_PAGE */ num_removed = 0; num_unwired = 0; + num_found = 0; - for (cpte = spte; cpte < epte; - cpte++, va += PAGE_SIZE) { + /* invalidate the PTEs first to "freeze" them */ + for (cpte = spte, vaddr = start_vaddr; + cpte < epte; + cpte++, vaddr += PAGE_SIZE_64) { pa = pte_to_pa(*cpte); if (pa == 0) continue; + num_found++; - num_removed++; if (iswired(*cpte)) num_unwired++; @@ -1260,27 +1623,46 @@ pmap_remove_range( * Outside range of managed physical memory. * Just remove the mappings. */ - register pt_entry_t *lpte = cpte; - - *lpte = 0; + pmap_store_pte(cpte, 0); continue; } - pai = pa_index(pa); - LOCK_PVH(pai); + /* invalidate the PTE */ + pmap_update_pte(cpte, *cpte, (*cpte & ~INTEL_PTE_VALID)); + } - /* - * Get the modify and reference bits. - */ - { - register pt_entry_t *lpte; + if (0 == num_found) { + /* nothing was changed, we're done */ + goto update_counts; + } - lpte = cpte; - pmap_phys_attributes[pai] |= - *lpte & (PHYS_MODIFIED|PHYS_REFERENCED); - *lpte = 0; + /* propagate the invalidates to other CPUs */ - } + PMAP_UPDATE_TLBS(pmap, start_vaddr, vaddr); + + for (cpte = spte, vaddr = start_vaddr; + cpte < epte; + cpte++, vaddr += PAGE_SIZE_64) { + + pa = pte_to_pa(*cpte); + if (pa == 0) + continue; + + pai = pa_index(pa); + + LOCK_PVH(pai); + + num_removed++; + + /* + * Get the modify and reference bits, then + * nuke the entry in the page table + */ + /* remember reference and change */ + pmap_phys_attributes[pai] |= + (char)(*cpte & (PHYS_MODIFIED|PHYS_REFERENCED)); + /* completely invalidate the PTE */ + pmap_store_pte(cpte, 0); /* * Remove the mapping from the pvlist for @@ -1293,7 +1675,7 @@ pmap_remove_range( if (pv_h->pmap == PMAP_NULL) { panic("pmap_remove: null pv_list!"); } - if (pv_h->va == va && pv_h->pmap == pmap) { + if (pv_h->va == vaddr && pv_h->pmap == pmap) { /* * Header is the pv_entry. Copy the next one * to header and free the next one (we cannot @@ -1315,7 +1697,7 @@ pmap_remove_range( if ((cur = prev->next) == PV_ENTRY_NULL) { panic("pmap-remove: mapping not in pv_list!"); } - } while (cur->va != va || cur->pmap != pmap); + } while (cur->va != vaddr || cur->pmap != pmap); prev->next = cur->next; PV_FREE(cur); } @@ -1323,6 +1705,7 @@ pmap_remove_range( } } + update_counts: /* * Update the counts */ @@ -1330,6 +1713,7 @@ pmap_remove_range( pmap->stats.resident_count -= num_removed; assert(pmap->stats.wired_count >= num_unwired); pmap->stats.wired_count -= num_unwired; + return; } /* @@ -1364,39 +1748,31 @@ pmap_remove( spl_t spl; register pt_entry_t *pde; register pt_entry_t *spte, *epte; - vm_offset_t l; - vm_offset_t s, e; - vm_offset_t orig_s; + addr64_t l64; + addr64_t orig_s64; - if (map == PMAP_NULL) + if (map == PMAP_NULL || s64 == e64) return; PMAP_READ_LOCK(map, spl); - if (value_64bit(s64) || value_64bit(e64)) { - panic("pmap_remove addr overflow"); - } - - orig_s = s = (vm_offset_t)low32(s64); - e = (vm_offset_t)low32(e64); - - pde = pmap_pde(map, s); - - while (s < e) { - l = (s + PDE_MAPPED_SIZE) & ~(PDE_MAPPED_SIZE-1); - if (l > e) - l = e; - if (*pde & INTEL_PTE_VALID) { - spte = (pt_entry_t *)pmap_pte(map, (s & ~(PDE_MAPPED_SIZE-1))); - spte = &spte[ptenum(s)]; - epte = &spte[intel_btop(l-s)]; - pmap_remove_range(map, s, spte, epte); + orig_s64 = s64; + + while (s64 < e64) { + l64 = (s64 + pde_mapped_size) & ~(pde_mapped_size-1); + if (l64 > e64) + l64 = e64; + pde = pmap_pde(map, s64); + if (pde && (*pde & INTEL_PTE_VALID)) { + spte = (pt_entry_t *)pmap_pte(map, (s64 & ~(pde_mapped_size-1))); + spte = &spte[ptenum(s64)]; + epte = &spte[intel_btop(l64-s64)]; + pmap_remove_range(map, s64, spte, epte); } - s = l; + s64 = l64; pde++; } - - PMAP_UPDATE_TLBS(map, orig_s, e); + PMAP_UPDATE_TLBS(map, orig_s64, e64); PMAP_READ_UNLOCK(map, spl); } @@ -1420,10 +1796,10 @@ pmap_page_protect( register pmap_t pmap; spl_t spl; boolean_t remove; - pmap_paddr_t phys; + pmap_paddr_t phys; assert(pn != vm_page_fictitious_addr); - phys = (pmap_paddr_t)i386_ptob(pn); + if (!valid_page(pn)) { /* * Not a managed page. @@ -1445,17 +1821,17 @@ pmap_page_protect( remove = TRUE; break; } + phys = (pmap_paddr_t)i386_ptob(pn); + pai = pa_index(phys); + pv_h = pai_to_pvh(pai); + /* * Lock the pmap system first, since we will be changing * several pmaps. */ - PMAP_WRITE_LOCK(spl); - pai = pa_index(phys); - pv_h = pai_to_pvh(pai); - /* * Walk down PV list, changing or removing all mappings. * We do not have to lock the pv_list because we have @@ -1463,93 +1839,94 @@ pmap_page_protect( */ if (pv_h->pmap != PMAP_NULL) { - prev = pv_e = pv_h; - do { - register vm_offset_t va; - pmap = pv_e->pmap; - /* - * Lock the pmap to block pmap_extract and similar routines. - */ - simple_lock(&pmap->lock); - - { - - va = pv_e->va; - pte = pmap_pte(pmap, va); - - /* - * Consistency checks. - */ - /* assert(*pte & INTEL_PTE_VALID); XXX */ - /* assert(pte_to_phys(*pte) == phys); */ + prev = pv_e = pv_h; - } - - /* - * Remove the mapping if new protection is NONE - * or if write-protecting a kernel mapping. - */ - if (remove || pmap == kernel_pmap) { - /* - * Remove the mapping, collecting any modify bits. - */ - { - pmap_phys_attributes[pai] |= - *pte & (PHYS_MODIFIED|PHYS_REFERENCED); - *pte++ = 0; - PMAP_UPDATE_TLBS(pmap, va, va + PAGE_SIZE); - } + do { + register vm_map_offset_t vaddr; - assert(pmap->stats.resident_count >= 1); - pmap->stats.resident_count--; + pmap = pv_e->pmap; + /* + * Lock the pmap to block pmap_extract and similar routines. + */ + simple_lock(&pmap->lock); - /* - * Remove the pv_entry. - */ - if (pv_e == pv_h) { + vaddr = pv_e->va; + pte = pmap_pte(pmap, vaddr); + if(0 == pte) { + kprintf("pmap_page_protect pmap 0x%x pn 0x%x vaddr 0x%llx\n",pmap, pn, vaddr); + panic("pmap_page_protect"); + } /* - * Fix up head later. + * Consistency checks. */ - pv_h->pmap = PMAP_NULL; - } - else { + /* assert(*pte & INTEL_PTE_VALID); XXX */ + /* assert(pte_to_phys(*pte) == phys); */ + + /* - * Delete this entry. + * Remove the mapping if new protection is NONE + * or if write-protecting a kernel mapping. */ - prev->next = pv_e->next; - PV_FREE(pv_e); - } - } - else { - /* - * Write-protect. - */ + if (remove || pmap == kernel_pmap) { + /* + * Remove the mapping, collecting any modify bits. + */ + pmap_update_pte(pte, *pte, (*pte & ~INTEL_PTE_VALID)); - *pte &= ~INTEL_PTE_WRITE; - pte++; - PMAP_UPDATE_TLBS(pmap, va, va + PAGE_SIZE); - /* - * Advance prev. - */ - prev = pv_e; - } + PMAP_UPDATE_TLBS(pmap, vaddr, vaddr + PAGE_SIZE); - simple_unlock(&pmap->lock); + pmap_phys_attributes[pai] |= *pte & (PHYS_MODIFIED|PHYS_REFERENCED); - } while ((pv_e = prev->next) != PV_ENTRY_NULL); + pmap_store_pte(pte, 0); - /* - * If pv_head mapping was removed, fix it up. - */ - if (pv_h->pmap == PMAP_NULL) { - pv_e = pv_h->next; - if (pv_e != PV_ENTRY_NULL) { - *pv_h = *pv_e; - PV_FREE(pv_e); + + //XXX breaks DEBUG build assert(pmap->stats.resident_count >= 1); + pmap->stats.resident_count--; + + /* + * Remove the pv_entry. + */ + if (pv_e == pv_h) { + /* + * Fix up head later. + */ + pv_h->pmap = PMAP_NULL; + } + else { + /* + * Delete this entry. + */ + prev->next = pv_e->next; + PV_FREE(pv_e); + } + } else { + /* + * Write-protect. + */ + pmap_update_pte(pte, *pte, (*pte & ~INTEL_PTE_WRITE)); + PMAP_UPDATE_TLBS(pmap, vaddr, vaddr + PAGE_SIZE); + /* + * Advance prev. + */ + prev = pv_e; + } + + simple_unlock(&pmap->lock); + + } while ((pv_e = prev->next) != PV_ENTRY_NULL); + + /* + * If pv_head mapping was removed, fix it up. + */ + if (pv_h->pmap == PMAP_NULL) { + pv_e = pv_h->next; + + if (pv_e != PV_ENTRY_NULL) { + *pv_h = *pv_e; + PV_FREE(pv_e); + } } - } } - PMAP_WRITE_UNLOCK(spl); } @@ -1577,64 +1954,92 @@ unsigned int pmap_disconnect( void pmap_protect( pmap_t map, - vm_offset_t s, - vm_offset_t e, + vm_map_offset_t sva, + vm_map_offset_t eva, vm_prot_t prot) { register pt_entry_t *pde; register pt_entry_t *spte, *epte; - vm_offset_t l; + vm_map_offset_t lva; + vm_map_offset_t orig_sva; spl_t spl; - vm_offset_t orig_s = s; - + boolean_t set_NX; + int num_found = 0; if (map == PMAP_NULL) return; - /* - * Determine the new protection. - */ - switch (prot) { - case VM_PROT_READ: - case VM_PROT_READ|VM_PROT_EXECUTE: - break; - case VM_PROT_READ|VM_PROT_WRITE: - case VM_PROT_ALL: - return; /* nothing to do */ - default: - pmap_remove(map, (addr64_t)s, (addr64_t)e); + if (prot == VM_PROT_NONE) { + pmap_remove(map, sva, eva); return; } + if ( (prot & VM_PROT_EXECUTE) || !nx_enabled || !map->nx_enabled ) + set_NX = FALSE; + else + set_NX = TRUE; + SPLVM(spl); simple_lock(&map->lock); - pde = pmap_pde(map, s); - while (s < e) { - l = (s + PDE_MAPPED_SIZE) & ~(PDE_MAPPED_SIZE-1); - if (l > e) - l = e; - if (*pde & INTEL_PTE_VALID) { - spte = (pt_entry_t *)pmap_pte(map, (s & ~(PDE_MAPPED_SIZE-1))); - spte = &spte[ptenum(s)]; - epte = &spte[intel_btop(l-s)]; + orig_sva = sva; + while (sva < eva) { + lva = (sva + pde_mapped_size) & ~(pde_mapped_size-1); + if (lva > eva) + lva = eva; + pde = pmap_pde(map, sva); + if (pde && (*pde & INTEL_PTE_VALID)) { + spte = (pt_entry_t *)pmap_pte(map, (sva & ~(pde_mapped_size-1))); + spte = &spte[ptenum(sva)]; + epte = &spte[intel_btop(lva-sva)]; while (spte < epte) { - if (*spte & INTEL_PTE_VALID) - *spte &= ~INTEL_PTE_WRITE; + if (*spte & INTEL_PTE_VALID) { + + if (prot & VM_PROT_WRITE) + pmap_update_pte(spte, *spte, (*spte | INTEL_PTE_WRITE)); + else + pmap_update_pte(spte, *spte, (*spte & ~INTEL_PTE_WRITE)); + + if (set_NX == TRUE) + pmap_update_pte(spte, *spte, (*spte | INTEL_PTE_NX)); + else + pmap_update_pte(spte, *spte, (*spte & ~INTEL_PTE_NX)); + + num_found++; + + } spte++; } } - s = l; - pde++; + sva = lva; } - - PMAP_UPDATE_TLBS(map, orig_s, e); + if (num_found) + PMAP_UPDATE_TLBS(map, orig_sva, eva); simple_unlock(&map->lock); SPLX(spl); } +/* Map a (possibly) autogenned block */ +void +pmap_map_block( + pmap_t pmap, + addr64_t va, + ppnum_t pa, + uint32_t size, + vm_prot_t prot, + int attr, + __unused unsigned int flags) +{ + uint32_t page; + + for (page = 0; page < size; page++) { + pmap_enter(pmap, va, pa, prot, attr, TRUE); + va += PAGE_SIZE; + pa++; + } +} /* @@ -1652,7 +2057,7 @@ pmap_protect( void pmap_enter( register pmap_t pmap, - vm_offset_t v, + vm_map_offset_t vaddr, ppnum_t pn, vm_prot_t prot, unsigned int flags, @@ -1665,19 +2070,27 @@ pmap_enter( pt_entry_t template; spl_t spl; pmap_paddr_t old_pa; - pmap_paddr_t pa = (pmap_paddr_t)i386_ptob(pn); + pmap_paddr_t pa = (pmap_paddr_t)i386_ptob(pn); + boolean_t need_tlbflush = FALSE; + boolean_t set_NX; + char oattr; - XPR(0x80000000, "%x/%x: pmap_enter %x/%x/%x\n", + XPR(0x80000000, "%x/%x: pmap_enter %x/%qx/%x\n", current_thread(), current_thread(), - pmap, v, pn); + pmap, vaddr, pn); assert(pn != vm_page_fictitious_addr); if (pmap_debug) - printf("pmap(%x, %x)\n", v, pn); + printf("pmap(%qx, %x)\n", vaddr, pn); if (pmap == PMAP_NULL) return; + if ( (prot & VM_PROT_EXECUTE) || !nx_enabled || !pmap->nx_enabled ) + set_NX = FALSE; + else + set_NX = TRUE; + /* * Must allocate a new pvlist entry while we're unlocked; * zalloc may cause pageout (which will lock the pmap system). @@ -1695,13 +2108,13 @@ pmap_enter( * pages to map one VM page. */ - while ((pte = pmap_pte(pmap, v)) == PT_ENTRY_NULL) { + while ((pte = pmap_pte(pmap, vaddr)) == PT_ENTRY_NULL) { /* * Must unlock to expand the pmap. */ PMAP_READ_UNLOCK(pmap, spl); - pmap_expand(pmap, v); + pmap_expand(pmap, vaddr); /* going to grow pde level page(s) */ PMAP_READ_LOCK(pmap, spl); } @@ -1717,7 +2130,7 @@ pmap_enter( template = pa_to_pte(pa) | INTEL_PTE_VALID; - if(flags & VM_MEM_NOT_CACHEABLE) { + if(VM_MEM_NOT_CACHEABLE == (flags & (VM_MEM_NOT_CACHEABLE | VM_WIMG_USE_DEFAULT))) { if(!(flags & VM_MEM_GUARDED)) template |= INTEL_PTE_PTA; template |= INTEL_PTE_NCACHE; @@ -1727,6 +2140,10 @@ pmap_enter( template |= INTEL_PTE_USER; if (prot & VM_PROT_WRITE) template |= INTEL_PTE_WRITE; + + if (set_NX == TRUE) + template |= INTEL_PTE_NX; + if (wired) { template |= INTEL_PTE_WIRED; if (!iswired(*pte)) @@ -1739,11 +2156,9 @@ pmap_enter( } } - if (*pte & INTEL_PTE_MOD) - template |= INTEL_PTE_MOD; - WRITE_PTE(pte, template) - pte++; - + /* store modified PTE and preserve RC bits */ + pmap_update_pte(pte, *pte, template | (*pte & (INTEL_PTE_REF | INTEL_PTE_MOD))); + need_tlbflush = TRUE; goto Done; } @@ -1754,6 +2169,7 @@ pmap_enter( * 2) Add pvlist entry for new mapping * 3) Enter new mapping. * + * SHARING FAULTS IS HORRIBLY BROKEN * SHARING_FAULTS complicates this slightly in that it cannot * replace the mapping, but must remove it (because adding the * pvlist entry for the new mapping may remove others), and @@ -1767,18 +2183,21 @@ pmap_enter( if (old_pa != (pmap_paddr_t) 0) { - -#if DEBUG_PTE_PAGE - if (pmap != kernel_pmap) - ptep_check(get_pte_page(pte)); -#endif /* DEBUG_PTE_PAGE */ - /* * Don't do anything to pages outside valid memory here. * Instead convince the code that enters a new mapping * to overwrite the old one. */ + /* invalidate the PTE */ + pmap_update_pte(pte, *pte, (*pte & ~INTEL_PTE_VALID)); + /* propagate the invalidate everywhere */ + PMAP_UPDATE_TLBS(pmap, vaddr, vaddr + PAGE_SIZE); + /* remember reference and change */ + oattr = (char)(*pte & (PHYS_MODIFIED | PHYS_REFERENCED)); + /* completely invalidate the PTE */ + pmap_store_pte(pte,0); + if (valid_page(i386_btop(old_pa))) { pai = pa_index(old_pa); @@ -1791,10 +2210,7 @@ pmap_enter( pmap->stats.wired_count--; } - pmap_phys_attributes[pai] |= - *pte & (PHYS_MODIFIED|PHYS_REFERENCED); - WRITE_PTE(pte, 0) - + pmap_phys_attributes[pai] |= oattr; /* * Remove the mapping from the pvlist for * this physical page. @@ -1806,7 +2222,8 @@ pmap_enter( if (pv_h->pmap == PMAP_NULL) { panic("pmap_enter: null pv_list!"); } - if (pv_h->va == v && pv_h->pmap == pmap) { + + if (pv_h->va == vaddr && pv_h->pmap == pmap) { /* * Header is the pv_entry. Copy the next one * to header and free the next one (we cannot @@ -1828,7 +2245,7 @@ pmap_enter( if ((cur = prev->next) == PV_ENTRY_NULL) { panic("pmap_enter: mapping not in pv_list!"); } - } while (cur->va != v || cur->pmap != pmap); + } while (cur->va != vaddr || cur->pmap != pmap); prev->next = cur->next; pv_e = cur; } @@ -1843,13 +2260,13 @@ pmap_enter( * one). Do removal part of accounting. */ old_pa = (pmap_paddr_t) 0; - assert(pmap->stats.resident_count >= 1); - pmap->stats.resident_count--; + if (iswired(*pte)) { assert(pmap->stats.wired_count >= 1); pmap->stats.wired_count--; } } + need_tlbflush = TRUE; } @@ -1863,7 +2280,7 @@ pmap_enter( pai = pa_index(pa); -#if SHARING_FAULTS +#if SHARING_FAULTS /* this is horribly broken , do not enable */ RetryPvList: /* * We can return here from the sharing fault code below @@ -1878,7 +2295,7 @@ RetryPvList: /* * No mappings yet */ - pv_h->va = v; + pv_h->va = vaddr; pv_h->pmap = pmap; pv_h->next = PV_ENTRY_NULL; } @@ -1891,13 +2308,13 @@ RetryPvList: */ pv_entry_t e = pv_h; while (e != PV_ENTRY_NULL) { - if (e->pmap == pmap && e->va == v) + if (e->pmap == pmap && e->va == vaddr) panic("pmap_enter: already in pv_list"); e = e->next; } } #endif /* DEBUG */ -#if SHARING_FAULTS +#if SHARING_FAULTS /* broken, do not enable */ { /* * do sharing faults. @@ -1923,6 +2340,7 @@ RetryPvList: */ pmap_remove_range(pmap, e->va, opte, opte + 1); + PMAP_UPDATE_TLBS(pmap, e->va, e->va + PAGE_SIZE); /* @@ -1976,7 +2394,7 @@ RetryPvList: if (logit) { pma = &pmap_aliasbuf[pmap_alias_index]; pma->pmap = pmap; - pma->va = v; + pma->va = vaddr; pma->rpc = rpc; pma->cookie = PMAP_ALIAS_COOKIE; if (++pmap_alias_index >= PMAP_ALIAS_MAX) @@ -1996,7 +2414,7 @@ RetryPvList: panic("pmap no pv_e's"); } } - pv_e->va = v; + pv_e->va = vaddr; pv_e->pmap = pmap; pv_e->next = pv_h->next; pv_h->next = pv_e; @@ -2006,13 +2424,18 @@ RetryPvList: pv_e = PV_ENTRY_NULL; } UNLOCK_PVH(pai); + + /* + * only count the mapping + * for 'managed memory' + */ + pmap->stats.resident_count++; } /* - * Step 3) Enter and count the mapping. + * Step 3) Enter the mapping. */ - pmap->stats.resident_count++; /* * Build a template to speed up entering - @@ -2030,18 +2453,22 @@ RetryPvList: template |= INTEL_PTE_USER; if (prot & VM_PROT_WRITE) template |= INTEL_PTE_WRITE; + + if (set_NX == TRUE) + template |= INTEL_PTE_NX; + if (wired) { template |= INTEL_PTE_WIRED; pmap->stats.wired_count++; } - - WRITE_PTE(pte, template) + pmap_store_pte(pte, template); Done: - PMAP_UPDATE_TLBS(pmap, v, v + PAGE_SIZE); + if (need_tlbflush == TRUE) + PMAP_UPDATE_TLBS(pmap, vaddr, vaddr + PAGE_SIZE); if (pv_e != PV_ENTRY_NULL) { - PV_FREE(pv_e); + PV_FREE(pv_e); } PMAP_READ_UNLOCK(pmap, spl); @@ -2057,7 +2484,7 @@ Done: void pmap_change_wiring( register pmap_t map, - vm_offset_t v, + vm_map_offset_t vaddr, boolean_t wired) { register pt_entry_t *pte; @@ -2070,7 +2497,7 @@ pmap_change_wiring( */ PMAP_READ_LOCK(map, spl); - if ((pte = pmap_pte(map, v)) == PT_ENTRY_NULL) + if ((pte = pmap_pte(map, vaddr)) == PT_ENTRY_NULL) panic("pmap_change_wiring: pte missing"); if (wired && !iswired(*pte)) { @@ -2078,7 +2505,8 @@ pmap_change_wiring( * wiring down mapping */ map->stats.wired_count++; - *pte++ |= INTEL_PTE_WIRED; + pmap_update_pte(pte, *pte, (*pte | INTEL_PTE_WIRED)); + pte++; } else if (!wired && iswired(*pte)) { /* @@ -2086,7 +2514,8 @@ pmap_change_wiring( */ assert(map->stats.wired_count >= 1); map->stats.wired_count--; - *pte++ &= ~INTEL_PTE_WIRED; + pmap_update_pte(pte, *pte, (*pte & ~INTEL_PTE_WIRED)); + pte++; } PMAP_READ_UNLOCK(map, spl); @@ -2101,18 +2530,18 @@ ppnum_t pmap_find_phys(pmap_t pmap, addr64_t va) { pt_entry_t *ptp; - vm_offset_t a32; ppnum_t ppn; - if (value_64bit(va)) - panic("pmap_find_phys 64 bit value"); - a32 = (vm_offset_t) low32(va); - ptp = pmap_pte(pmap, a32); + mp_disable_preemption(); + + ptp = pmap_pte(pmap, va); if (PT_ENTRY_NULL == ptp) { ppn = 0; } else { ppn = (ppnum_t) i386_btop(pte_to_pa(*ptp)); } + mp_enable_preemption(); + return ppn; } @@ -2129,54 +2558,40 @@ pmap_find_phys(pmap_t pmap, addr64_t va) vm_offset_t pmap_extract( register pmap_t pmap, - vm_offset_t va) + vm_map_offset_t vaddr) { - ppnum_t ppn; - vm_offset_t vaddr; + ppnum_t ppn; + vm_offset_t paddr; - vaddr = (vm_offset_t)0; - ppn = pmap_find_phys(pmap, (addr64_t)va); - if (ppn) { - vaddr = ((vm_offset_t)i386_ptob(ppn)) | (va & INTEL_OFFMASK); - } - return (vaddr); + paddr = (vm_offset_t)0; + ppn = pmap_find_phys(pmap, vaddr); + if (ppn) { + paddr = ((vm_offset_t)i386_ptob(ppn)) | (vaddr & INTEL_OFFMASK); + } + return (paddr); } - -/* - * Routine: pmap_expand - * - * Expands a pmap to be able to map the specified virtual address. - * - * Allocates new virtual memory for the P0 or P1 portion of the - * pmap, then re-maps the physical pages that were in the old - * pmap to be in the new pmap. - * - * Must be called with the pmap system and the pmap unlocked, - * since these must be unlocked to use vm_allocate or vm_deallocate. - * Thus it must be called in a loop that checks whether the map - * has been expanded enough. - * (We won't loop forever, since page tables aren't shrunk.) - */ void -pmap_expand( - register pmap_t map, - register vm_offset_t v) +pmap_expand_pml4( + pmap_t map, + vm_map_offset_t vaddr) { - pt_entry_t *pdp; register vm_page_t m; register pmap_paddr_t pa; - register int i; + uint64_t i; spl_t spl; ppnum_t pn; + pml4_entry_t *pml4p; - if (map == kernel_pmap) { - pmap_growkernel(v); - return; - } + if (kernel_pmap == map) panic("expand kernel pml4"); + + spl = splhigh(); + pml4p = pmap64_pml4(map, vaddr); + splx(spl); + if (PML4_ENTRY_NULL == pml4p) panic("pmap_expand_pml4 no pml4p"); /* - * Allocate a VM page for the level 2 page table entries. + * Allocate a VM page for the pml4 page */ while ((m = vm_page_grab()) == VM_PAGE_NULL) VM_PAGE_WAIT(); @@ -2187,14 +2602,25 @@ pmap_expand( */ pn = m->phys_page; pa = i386_ptob(pn); - i = pdenum(map, v); - vm_object_lock(map->pm_obj); - vm_page_insert(m, map->pm_obj, (vm_object_offset_t)i); + i = pml4idx(map, vaddr); + + vm_object_lock(map->pm_obj_pml4); +#if 0 /* DEBUG */ + if (0 != vm_page_lookup(map->pm_obj_pml4, (vm_object_offset_t)i)) { + kprintf("pmap_expand_pml4: obj_pml4 not empty, pmap 0x%x pm_obj_pml4 0x%x vaddr 0x%llx i 0x%llx\n", + map, map->pm_obj_pml4, vaddr, i); + } +#endif + vm_page_insert(m, map->pm_obj_pml4, (vm_object_offset_t)i); + vm_page_lock_queues(); vm_page_wire(m); - inuse_ptepages_count++; - vm_object_unlock(map->pm_obj); + vm_page_unlock_queues(); + vm_object_unlock(map->pm_obj_pml4); + inuse_ptepages_count++; + map->stats.resident_count++; + map->stats.wired_count++; /* * Zero the page. @@ -2205,14 +2631,17 @@ pmap_expand( /* * See if someone else expanded us first */ - if (pmap_pte(map, v) != PT_ENTRY_NULL) { + if (pmap64_pdpt(map, vaddr) != PDPT_ENTRY_NULL) { PMAP_READ_UNLOCK(map, spl); - vm_object_lock(map->pm_obj); + vm_object_lock(map->pm_obj_pml4); vm_page_lock_queues(); vm_page_free(m); inuse_ptepages_count--; + map->stats.resident_count--; + map->stats.wired_count--; + vm_page_unlock_queues(); - vm_object_unlock(map->pm_obj); + vm_object_unlock(map->pm_obj_pml4); return; } @@ -2222,103 +2651,300 @@ pmap_expand( * set several page directory entries. */ - pdp = &map->dirbase[pdenum(map, v)]; - *pdp = pa_to_pte(pa) - | INTEL_PTE_VALID - | INTEL_PTE_USER - | INTEL_PTE_WRITE; + pml4p = pmap64_pml4(map, vaddr); /* refetch under lock */ - PMAP_READ_UNLOCK(map, spl); - return; -} + pmap_store_pte(pml4p, pa_to_pte(pa) + | INTEL_PTE_VALID + | INTEL_PTE_USER + | INTEL_PTE_WRITE); -/* - * Copy the range specified by src_addr/len - * from the source map to the range dst_addr/len - * in the destination map. - * - * This routine is only advisory and need not do anything. - */ -#if 0 -void -pmap_copy( - pmap_t dst_pmap, - pmap_t src_pmap, - vm_offset_t dst_addr, - vm_size_t len, - vm_offset_t src_addr) -{ -#ifdef lint - dst_pmap++; src_pmap++; dst_addr++; len++; src_addr++; -#endif /* lint */ -} -#endif/* 0 */ + PMAP_READ_UNLOCK(map, spl); -/* - * pmap_sync_page_data_phys(ppnum_t pa) - * - * Invalidates all of the instruction cache on a physical page and - * pushes any dirty data from the data cache for the same physical page - * Not required in i386. - */ -void -pmap_sync_page_data_phys(__unused ppnum_t pa) -{ return; + } -/* - * pmap_sync_page_attributes_phys(ppnum_t pa) - * - * Write back and invalidate all cachelines on a physical page. - */ void -pmap_sync_page_attributes_phys(ppnum_t pa) +pmap_expand_pdpt( + pmap_t map, + vm_map_offset_t vaddr) { - cache_flush_page_phys(pa); -} + register vm_page_t m; + register pmap_paddr_t pa; + uint64_t i; + spl_t spl; + ppnum_t pn; + pdpt_entry_t *pdptp; -int collect_ref; -int collect_unref; + if (kernel_pmap == map) panic("expand kernel pdpt"); -/* - * Routine: pmap_collect - * Function: - * Garbage collects the physical map system for - * pages which are no longer used. - * Success need not be guaranteed -- that is, there - * may well be pages which are not referenced, but - * others may be collected. - * Usage: - * Called by the pageout daemon when pages are scarce. - */ -void -pmap_collect( - pmap_t p) -{ - register pt_entry_t *pdp, *ptp; - pt_entry_t *eptp; - int wired; - spl_t spl; + spl = splhigh(); + while ((pdptp = pmap64_pdpt(map, vaddr)) == PDPT_ENTRY_NULL) { + splx(spl); + pmap_expand_pml4(map, vaddr); /* need room for another pdpt entry */ + spl = splhigh(); + } + splx(spl); - if (p == PMAP_NULL) - return; - if (p == kernel_pmap) - return; + /* + * Allocate a VM page for the pdpt page + */ + while ((m = vm_page_grab()) == VM_PAGE_NULL) + VM_PAGE_WAIT(); /* - * Garbage collect map. + * put the page into the pmap's obj list so it + * can be found later. */ - PMAP_READ_LOCK(p, spl); + pn = m->phys_page; + pa = i386_ptob(pn); + i = pdptidx(map, vaddr); - for (pdp = (pt_entry_t *)p->dirbase; + vm_object_lock(map->pm_obj_pdpt); +#if 0 /* DEBUG */ + if (0 != vm_page_lookup(map->pm_obj_pdpt, (vm_object_offset_t)i)) { + kprintf("pmap_expand_pdpt: obj_pdpt not empty, pmap 0x%x pm_obj_pdpt 0x%x vaddr 0x%llx i 0x%llx\n", + map, map->pm_obj_pdpt, vaddr, i); + } +#endif + vm_page_insert(m, map->pm_obj_pdpt, (vm_object_offset_t)i); + + vm_page_lock_queues(); + vm_page_wire(m); + + vm_page_unlock_queues(); + vm_object_unlock(map->pm_obj_pdpt); + inuse_ptepages_count++; + map->stats.resident_count++; + map->stats.wired_count++; + + /* + * Zero the page. + */ + pmap_zero_page(pn); + + PMAP_READ_LOCK(map, spl); + /* + * See if someone else expanded us first + */ + if (pmap64_pde(map, vaddr) != PD_ENTRY_NULL) { + PMAP_READ_UNLOCK(map, spl); + vm_object_lock(map->pm_obj_pdpt); + vm_page_lock_queues(); + vm_page_free(m); + inuse_ptepages_count--; + map->stats.resident_count--; + map->stats.wired_count--; + + vm_page_unlock_queues(); + vm_object_unlock(map->pm_obj_pdpt); + return; + } + + /* + * Set the page directory entry for this page table. + * If we have allocated more than one hardware page, + * set several page directory entries. + */ + + pdptp = pmap64_pdpt(map, vaddr); /* refetch under lock */ + + pmap_store_pte(pdptp, pa_to_pte(pa) + | INTEL_PTE_VALID + | INTEL_PTE_USER + | INTEL_PTE_WRITE); + + PMAP_READ_UNLOCK(map, spl); + + return; + +} + + + +/* + * Routine: pmap_expand + * + * Expands a pmap to be able to map the specified virtual address. + * + * Allocates new virtual memory for the P0 or P1 portion of the + * pmap, then re-maps the physical pages that were in the old + * pmap to be in the new pmap. + * + * Must be called with the pmap system and the pmap unlocked, + * since these must be unlocked to use vm_allocate or vm_deallocate. + * Thus it must be called in a loop that checks whether the map + * has been expanded enough. + * (We won't loop forever, since page tables aren't shrunk.) + */ +void +pmap_expand( + pmap_t map, + vm_map_offset_t vaddr) +{ + pt_entry_t *pdp; + register vm_page_t m; + register pmap_paddr_t pa; + uint64_t i; + spl_t spl; + ppnum_t pn; + + /* + * if not the kernel map (while we are still compat kernel mode) + * and we are 64 bit, propagate expand upwards + */ + + if (cpu_64bit && (map != kernel_pmap)) { + spl = splhigh(); + while ((pdp = pmap64_pde(map, vaddr)) == PD_ENTRY_NULL) { + splx(spl); + pmap_expand_pdpt(map, vaddr); /* need room for another pde entry */ + spl = splhigh(); + } + splx(spl); + } else { + pdp = pmap_pde(map, vaddr); + } + + + /* + * Allocate a VM page for the pde entries. + */ + while ((m = vm_page_grab()) == VM_PAGE_NULL) + VM_PAGE_WAIT(); + + /* + * put the page into the pmap's obj list so it + * can be found later. + */ + pn = m->phys_page; + pa = i386_ptob(pn); + i = pdeidx(map, vaddr); + + vm_object_lock(map->pm_obj); +#if 0 /* DEBUG */ + if (0 != vm_page_lookup(map->pm_obj, (vm_object_offset_t)i)) { + kprintf("pmap_expand: obj not empty, pmap 0x%x pm_obj 0x%x vaddr 0x%llx i 0x%llx\n", + map, map->pm_obj, vaddr, i); + } +#endif + vm_page_insert(m, map->pm_obj, (vm_object_offset_t)i); + + vm_page_lock_queues(); + vm_page_wire(m); + inuse_ptepages_count++; + + vm_page_unlock_queues(); + vm_object_unlock(map->pm_obj); + + /* + * Zero the page. + */ + pmap_zero_page(pn); + + PMAP_READ_LOCK(map, spl); + /* + * See if someone else expanded us first + */ + if (pmap_pte(map, vaddr) != PT_ENTRY_NULL) { + PMAP_READ_UNLOCK(map, spl); + vm_object_lock(map->pm_obj); + + vm_page_lock_queues(); + vm_page_free(m); + inuse_ptepages_count--; + + vm_page_unlock_queues(); + vm_object_unlock(map->pm_obj); + return; + } + + pdp = pmap_pde(map, vaddr); /* refetch while locked */ + + /* + * Set the page directory entry for this page table. + * If we have allocated more than one hardware page, + * set several page directory entries. + */ + + pmap_store_pte(pdp, pa_to_pte(pa) + | INTEL_PTE_VALID + | INTEL_PTE_USER + | INTEL_PTE_WRITE); + + + PMAP_READ_UNLOCK(map, spl); + + return; +} + + +/* + * pmap_sync_page_data_phys(ppnum_t pa) + * + * Invalidates all of the instruction cache on a physical page and + * pushes any dirty data from the data cache for the same physical page + * Not required in i386. + */ +void +pmap_sync_page_data_phys(__unused ppnum_t pa) +{ + return; +} + +/* + * pmap_sync_page_attributes_phys(ppnum_t pa) + * + * Write back and invalidate all cachelines on a physical page. + */ +void +pmap_sync_page_attributes_phys(ppnum_t pa) +{ + cache_flush_page_phys(pa); +} + +int collect_ref; +int collect_unref; + +/* + * Routine: pmap_collect + * Function: + * Garbage collects the physical map system for + * pages which are no longer used. + * Success need not be guaranteed -- that is, there + * may well be pages which are not referenced, but + * others may be collected. + * Usage: + * Called by the pageout daemon when pages are scarce. + */ +void +pmap_collect( + pmap_t p) +{ + register pt_entry_t *pdp, *ptp; + pt_entry_t *eptp; + int wired; + spl_t spl; + + if (p == PMAP_NULL) + return; + + if (p == kernel_pmap) + return; + + /* + * Garbage collect map. + */ + PMAP_READ_LOCK(p, spl); + + for (pdp = (pt_entry_t *)p->dirbase; pdp < (pt_entry_t *)&p->dirbase[(UMAXPTDI+1)]; pdp++) { if (*pdp & INTEL_PTE_VALID) { if(*pdp & INTEL_PTE_REF) { - *pdp &= ~INTEL_PTE_REF; + pmap_store_pte(pdp, *pdp & ~INTEL_PTE_REF); collect_ref++; } else { collect_unref++; @@ -2351,7 +2977,7 @@ pmap_collect( /* * Invalidate the page directory pointer. */ - *pdp = 0x0; + pmap_store_pte(pdp, 0x0); PMAP_READ_UNLOCK(p, spl); @@ -2378,32 +3004,21 @@ pmap_collect( } } PMAP_UPDATE_TLBS(p, VM_MIN_ADDRESS, VM_MAX_ADDRESS); + PMAP_READ_UNLOCK(p, spl); return; } -/* - * Routine: pmap_kernel - * Function: - * Returns the physical map handle for the kernel. - */ -#if 0 -pmap_t -pmap_kernel(void) -{ - return (kernel_pmap); -} -#endif/* 0 */ void pmap_copy_page(src, dst) ppnum_t src; ppnum_t dst; { - bcopy_phys((addr64_t)i386_ptob(src), - (addr64_t)i386_ptob(dst), - PAGE_SIZE); + bcopy_phys((addr64_t)i386_ptob(src), + (addr64_t)i386_ptob(dst), + PAGE_SIZE); } @@ -2424,8 +3039,8 @@ pmap_copy_page(src, dst) void pmap_pageable( __unused pmap_t pmap, - __unused vm_offset_t start_addr, - __unused vm_offset_t end_addr, + __unused vm_map_offset_t start_addr, + __unused vm_map_offset_t end_addr, __unused boolean_t pageable) { #ifdef lint @@ -2485,25 +3100,24 @@ phys_attribute_clear( simple_lock(&pmap->lock); { - register vm_offset_t va; + register vm_map_offset_t va; va = pv_e->va; - pte = pmap_pte(pmap, va); -#if 0 /* - * Consistency checks. + * first make sure any processor actively + * using this pmap fluses its TLB state */ - assert(*pte & INTEL_PTE_VALID); - /* assert(pte_to_phys(*pte) == phys); */ -#endif + + PMAP_UPDATE_TLBS(pmap, va, va + PAGE_SIZE); /* * Clear modify or reference bits. */ - *pte++ &= ~bits; - PMAP_UPDATE_TLBS(pmap, va, va + PAGE_SIZE); + pte = pmap_pte(pmap, va); + pmap_update_pte(pte, *pte, (*pte & ~bits)); + } simple_unlock(&pmap->lock); @@ -2539,15 +3153,24 @@ phys_attribute_test( return (FALSE); } + phys = i386_ptob(pn); + pai = pa_index(phys); + /* + * super fast check... if bits already collected + * no need to take any locks... + * if not set, we need to recheck after taking + * the lock in case they got pulled in while + * we were waiting for the lock + */ + if (pmap_phys_attributes[pai] & bits) + return (TRUE); + pv_h = pai_to_pvh(pai); + /* * Lock the pmap system first, since we will be checking * several pmaps. */ - PMAP_WRITE_LOCK(spl); - phys = i386_ptob(pn); - pai = pa_index(phys); - pv_h = pai_to_pvh(pai); if (pmap_phys_attributes[pai] & bits) { PMAP_WRITE_UNLOCK(spl); @@ -2572,7 +3195,7 @@ phys_attribute_test( simple_lock(&pmap->lock); { - register vm_offset_t va; + register vm_map_offset_t va; va = pv_e->va; pte = pmap_pte(pmap, va); @@ -2738,45 +3361,49 @@ pmap_clear_refmod(ppnum_t pa, unsigned int mask) void pmap_modify_pages( pmap_t map, - vm_offset_t s, - vm_offset_t e) + vm_map_offset_t sva, + vm_map_offset_t eva) { spl_t spl; register pt_entry_t *pde; register pt_entry_t *spte, *epte; - vm_offset_t l; - vm_offset_t orig_s = s; + vm_map_offset_t lva; + vm_map_offset_t orig_sva; if (map == PMAP_NULL) return; PMAP_READ_LOCK(map, spl); - pde = pmap_pde(map, s); - while (s && s < e) { - l = (s + PDE_MAPPED_SIZE) & ~(PDE_MAPPED_SIZE-1); - if (l > e) - l = e; - if (*pde & INTEL_PTE_VALID) { - spte = (pt_entry_t *)pmap_pte(map, (s & ~(PDE_MAPPED_SIZE-1))); - if (l) { - spte = &spte[ptenum(s)]; - epte = &spte[intel_btop(l-s)]; + orig_sva = sva; + while (sva && sva < eva) { + lva = (sva + pde_mapped_size) & ~(pde_mapped_size-1); + if (lva > eva) + lva = eva; + pde = pmap_pde(map, sva); + if (pde && (*pde & INTEL_PTE_VALID)) { + spte = (pt_entry_t *)pmap_pte(map, (sva & ~(pde_mapped_size-1))); + if (lva) { + spte = &spte[ptenum(sva)]; + epte = &spte[intel_btop(lva-sva)]; } else { - epte = &spte[intel_btop(PDE_MAPPED_SIZE)]; - spte = &spte[ptenum(s)]; + epte = &spte[intel_btop(pde_mapped_size)]; + spte = &spte[ptenum(sva)]; } while (spte < epte) { if (*spte & INTEL_PTE_VALID) { - *spte |= (INTEL_PTE_MOD | INTEL_PTE_WRITE); + pmap_store_pte(spte, *spte + | INTEL_PTE_MOD + | INTEL_PTE_WRITE); } spte++; } } - s = l; + sva = lva; pde++; } - PMAP_UPDATE_TLBS(map, orig_s, e); + PMAP_UPDATE_TLBS(map, orig_sva, eva); + PMAP_READ_UNLOCK(map, spl); } @@ -2796,225 +3423,21 @@ flush_dcache(__unused vm_offset_t addr, return; } -/* -* TLB Coherence Code (TLB "shootdown" code) -* -* Threads that belong to the same task share the same address space and -* hence share a pmap. However, they may run on distinct cpus and thus -* have distinct TLBs that cache page table entries. In order to guarantee -* the TLBs are consistent, whenever a pmap is changed, all threads that -* are active in that pmap must have their TLB updated. To keep track of -* this information, the set of cpus that are currently using a pmap is -* maintained within each pmap structure (cpus_using). Pmap_activate() and -* pmap_deactivate add and remove, respectively, a cpu from this set. -* Since the TLBs are not addressable over the bus, each processor must -* flush its own TLB; a processor that needs to invalidate another TLB -* needs to interrupt the processor that owns that TLB to signal the -* update. -* -* Whenever a pmap is updated, the lock on that pmap is locked, and all -* cpus using the pmap are signaled to invalidate. All threads that need -* to activate a pmap must wait for the lock to clear to await any updates -* in progress before using the pmap. They must ACQUIRE the lock to add -* their cpu to the cpus_using set. An implicit assumption made -* throughout the TLB code is that all kernel code that runs at or higher -* than splvm blocks out update interrupts, and that such code does not -* touch pageable pages. -* -* A shootdown interrupt serves another function besides signaling a -* processor to invalidate. The interrupt routine (pmap_update_interrupt) -* waits for the both the pmap lock (and the kernel pmap lock) to clear, -* preventing user code from making implicit pmap updates while the -* sending processor is performing its update. (This could happen via a -* user data write reference that turns on the modify bit in the page -* table). It must wait for any kernel updates that may have started -* concurrently with a user pmap update because the IPC code -* changes mappings. -* Spinning on the VALUES of the locks is sufficient (rather than -* having to acquire the locks) because any updates that occur subsequent -* to finding the lock unlocked will be signaled via another interrupt. -* (This assumes the interrupt is cleared before the low level interrupt code -* calls pmap_update_interrupt()). -* -* The signaling processor must wait for any implicit updates in progress -* to terminate before continuing with its update. Thus it must wait for an -* acknowledgement of the interrupt from each processor for which such -* references could be made. For maintaining this information, a set -* cpus_active is used. A cpu is in this set if and only if it can -* use a pmap. When pmap_update_interrupt() is entered, a cpu is removed from -* this set; when all such cpus are removed, it is safe to update. -* -* Before attempting to acquire the update lock on a pmap, a cpu (A) must -* be at least at the priority of the interprocessor interrupt -* (splip<=splvm). Otherwise, A could grab a lock and be interrupted by a -* kernel update; it would spin forever in pmap_update_interrupt() trying -* to acquire the user pmap lock it had already acquired. Furthermore A -* must remove itself from cpus_active. Otherwise, another cpu holding -* the lock (B) could be in the process of sending an update signal to A, -* and thus be waiting for A to remove itself from cpus_active. If A is -* spinning on the lock at priority this will never happen and a deadlock -* will result. -*/ - -/* - * Signal another CPU that it must flush its TLB - */ -void -signal_cpus( - cpu_set use_list, - pmap_t pmap, - vm_offset_t start_addr, - vm_offset_t end_addr) -{ - register int which_cpu, j; - register pmap_update_list_t update_list_p; - - while ((which_cpu = ffs((unsigned long)use_list)) != 0) { - which_cpu -= 1; /* convert to 0 origin */ +#if MACH_KDB - update_list_p = cpu_update_list(which_cpu); - simple_lock(&update_list_p->lock); +/* show phys page mappings and attributes */ - j = update_list_p->count; - if (j >= UPDATE_LIST_SIZE) { - /* - * list overflowed. Change last item to - * indicate overflow. - */ - update_list_p->item[UPDATE_LIST_SIZE-1].pmap = kernel_pmap; - update_list_p->item[UPDATE_LIST_SIZE-1].start = VM_MIN_ADDRESS; - update_list_p->item[UPDATE_LIST_SIZE-1].end = VM_MAX_KERNEL_ADDRESS; - } - else { - update_list_p->item[j].pmap = pmap; - update_list_p->item[j].start = start_addr; - update_list_p->item[j].end = end_addr; - update_list_p->count = j+1; - } - cpu_update_needed(which_cpu) = TRUE; - simple_unlock(&update_list_p->lock); - - /* if its the kernel pmap, ignore cpus_idle */ - if (((cpus_idle & (1 << which_cpu)) == 0) || - (pmap == kernel_pmap) || PMAP_REAL(which_cpu) == pmap) - { - i386_signal_cpu(which_cpu, MP_TLB_FLUSH, ASYNC); - } - use_list &= ~(1 << which_cpu); - } -} +extern void db_show_page(pmap_paddr_t pa); void -process_pmap_updates( - register pmap_t my_pmap) +db_show_page(pmap_paddr_t pa) { - register int my_cpu; - register pmap_update_list_t update_list_p; - register int j; - register pmap_t pmap; - - mp_disable_preemption(); - my_cpu = cpu_number(); - update_list_p = cpu_update_list(my_cpu); - simple_lock(&update_list_p->lock); - - for (j = 0; j < update_list_p->count; j++) { - pmap = update_list_p->item[j].pmap; - if (pmap == my_pmap || - pmap == kernel_pmap) { - - if (pmap->ref_count <= 0) { - PMAP_CPU_CLR(pmap, my_cpu); - PMAP_REAL(my_cpu) = kernel_pmap; -#ifdef PAE - set_cr3((unsigned int)kernel_pmap->pm_ppdpt); -#else - set_cr3((unsigned int)kernel_pmap->pdirbase); -#endif - } else - INVALIDATE_TLB(pmap, - update_list_p->item[j].start, - update_list_p->item[j].end); - } - } - update_list_p->count = 0; - cpu_update_needed(my_cpu) = FALSE; - simple_unlock(&update_list_p->lock); - mp_enable_preemption(); -} - -/* - * Interrupt routine for TBIA requested from other processor. - * This routine can also be called at all interrupts time if - * the cpu was idle. Some driver interrupt routines might access - * newly allocated vm. (This is the case for hd) - */ -void -pmap_update_interrupt(void) -{ - register int my_cpu; - spl_t s; - register pmap_t my_pmap; - - mp_disable_preemption(); - my_cpu = cpu_number(); - - /* - * Raise spl to splvm (above splip) to block out pmap_extract - * from IO code (which would put this cpu back in the active - * set). - */ - s = splhigh(); - - my_pmap = PMAP_REAL(my_cpu); - - if (!(my_pmap && pmap_in_use(my_pmap, my_cpu))) - my_pmap = kernel_pmap; - - do { - LOOP_VAR; - - /* - * Indicate that we're not using either user or kernel - * pmap. - */ - i_bit_clear(my_cpu, &cpus_active); - - /* - * Wait for any pmap updates in progress, on either user - * or kernel pmap. - */ - while (*(volatile int *)(&my_pmap->lock.interlock.lock_data) || - *(volatile int *)(&kernel_pmap->lock.interlock.lock_data)) { - LOOP_CHECK("pmap_update_interrupt", my_pmap); - cpu_pause(); - } - - process_pmap_updates(my_pmap); - - i_bit_set(my_cpu, &cpus_active); - - } while (cpu_update_needed(my_cpu)); - - splx(s); - mp_enable_preemption(); -} - -#if MACH_KDB - -/* show phys page mappings and attributes */ - -extern void db_show_page(pmap_paddr_t pa); - -void -db_show_page(pmap_paddr_t pa) -{ - pv_entry_t pv_h; - int pai; - char attr; - - pai = pa_index(pa); - pv_h = pai_to_pvh(pai); + pv_entry_t pv_h; + int pai; + char attr; + + pai = pa_index(pa); + pv_h = pai_to_pvh(pai); attr = pmap_phys_attributes[pai]; printf("phys page %x ", pa); @@ -3044,7 +3467,7 @@ void db_kvtophys( vm_offset_t vaddr) { - db_printf("0x%x", kvtophys(vaddr)); + db_printf("0x%qx", kvtophys(vaddr)); } /* @@ -3055,7 +3478,7 @@ db_show_vaddrs( pt_entry_t *dirbase) { pt_entry_t *ptep, *pdep, tmp; - int x, y, pdecnt, ptecnt; + unsigned int x, y, pdecnt, ptecnt; if (dirbase == 0) { dirbase = kernel_pmap->dirbase; @@ -3064,7 +3487,7 @@ db_show_vaddrs( db_printf("need a dirbase...\n"); return; } - dirbase = (pt_entry_t *) ((unsigned long) dirbase & ~INTEL_OFFMASK); + dirbase = (pt_entry_t *) (int) ((unsigned long) dirbase & ~INTEL_OFFMASK); db_printf("dirbase: 0x%x\n", dirbase); @@ -3109,153 +3532,24 @@ pmap_list_resident_pages( } #endif /* MACH_VM_DEBUG */ -#ifdef MACH_BSD -/* - * pmap_pagemove - * - * BSD support routine to reassign virtual addresses. - */ - -void -pmap_movepage(unsigned long from, unsigned long to, vm_size_t size) -{ - spl_t spl; - pt_entry_t *pte, saved_pte; - - /* Lock the kernel map */ - PMAP_READ_LOCK(kernel_pmap, spl); - - - while (size > 0) { - pte = pmap_pte(kernel_pmap, from); - if (pte == NULL) - panic("pmap_pagemove from pte NULL"); - saved_pte = *pte; - PMAP_READ_UNLOCK(kernel_pmap, spl); - - pmap_enter(kernel_pmap, to, (ppnum_t)i386_btop(i386_trunc_page(*pte)), - VM_PROT_READ|VM_PROT_WRITE, 0, *pte & INTEL_PTE_WIRED); - pmap_remove(kernel_pmap, (addr64_t)from, (addr64_t)(from+PAGE_SIZE)); - - PMAP_READ_LOCK(kernel_pmap, spl); - pte = pmap_pte(kernel_pmap, to); - if (pte == NULL) - panic("pmap_pagemove 'to' pte NULL"); - - *pte = saved_pte; - - from += PAGE_SIZE; - to += PAGE_SIZE; - size -= PAGE_SIZE; - } - - /* Get the processors to update the TLBs */ - PMAP_UPDATE_TLBS(kernel_pmap, from, from+size); - PMAP_UPDATE_TLBS(kernel_pmap, to, to+size); - - PMAP_READ_UNLOCK(kernel_pmap, spl); - -} -#endif /* MACH_BSD */ /* temporary workaround */ boolean_t -coredumpok(vm_map_t map, vm_offset_t va) +coredumpok(__unused vm_map_t map, __unused vm_offset_t va) { +#if 0 pt_entry_t *ptep; ptep = pmap_pte(map->pmap, va); if (0 == ptep) return FALSE; return ((*ptep & (INTEL_PTE_NCACHE | INTEL_PTE_WIRED)) != (INTEL_PTE_NCACHE | INTEL_PTE_WIRED)); -} - -/* - * grow the number of kernel page table entries, if needed - */ -void -pmap_growkernel(vm_offset_t addr) -{ -#if GROW_KERNEL_FUNCTION_IMPLEMENTED - struct pmap *pmap; - int s; - vm_offset_t ptppaddr; - ppnum_t ppn; - vm_page_t nkpg; - pd_entry_t newpdir = 0; - - /* - * Serialize. - * Losers return to try again until the winner completes the work. - */ - if (kptobj == 0) panic("growkernel 0"); - if (!vm_object_lock_try(kptobj)) { - return; - } - - vm_page_lock_queues(); - - s = splhigh(); - - /* - * If this is the first time thru, locate the end of the - * kernel page table entries and set nkpt to the current - * number of kernel page table pages - */ - - if (kernel_vm_end == 0) { - kernel_vm_end = KERNBASE; - nkpt = 0; - - while (pdir_pde(kernel_pmap->dirbase, kernel_vm_end)) { - kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); - nkpt++; - } - } - - /* - * Now allocate and map the required number of page tables - */ - addr = (addr + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); - while (kernel_vm_end < addr) { - if (pdir_pde(kernel_pmap->dirbase, kernel_vm_end)) { - kernel_vm_end = (kernel_vm_end + PAGE_SIZE * NPTEPG) & ~(PAGE_SIZE * NPTEPG - 1); - continue; /* someone already filled this one */ - } - - nkpg = vm_page_alloc(kptobj, nkpt); - if (!nkpg) - panic("pmap_growkernel: no memory to grow kernel"); - - nkpt++; - vm_page_wire(nkpg); - ppn = nkpg->phys_page; - pmap_zero_page(ppn); - ptppaddr = i386_ptob(ppn); - newpdir = (pd_entry_t) (ptppaddr | INTEL_PTE_VALID | - INTEL_PTE_RW | INTEL_PTE_REF | INTEL_PTE_MOD); - pdir_pde(kernel_pmap->dirbase, kernel_vm_end) = newpdir; - - simple_lock(&free_pmap_lock); - for (pmap = (struct pmap *)kernel_pmap->pmap_link.next; - pmap != kernel_pmap ; - pmap = (struct pmap *)pmap->pmap_link.next ) { - *pmap_pde(pmap, kernel_vm_end) = newpdir; - } - simple_unlock(&free_pmap_lock); - } - splx(s); - vm_page_unlock_queues(); - vm_object_unlock(kptobj); +#else + return TRUE; #endif } -pt_entry_t * -pmap_mapgetpte(vm_map_t map, vm_offset_t v) -{ - return pmap_pte(map->pmap, v); -} boolean_t phys_page_exists( @@ -3312,27 +3606,88 @@ mapping_adjust() } void -pmap_commpage_init(vm_offset_t kernel_commpage, vm_offset_t user_commpage, int cnt) +pmap_commpage32_init(vm_offset_t kernel_commpage, vm_offset_t user_commpage, int cnt) { int i; pt_entry_t *opte, *npte; pt_entry_t pte; + for (i = 0; i < cnt; i++) { - opte = pmap_pte(kernel_pmap, kernel_commpage); + opte = pmap_pte(kernel_pmap, (vm_map_offset_t)kernel_commpage); if (0 == opte) panic("kernel_commpage"); - npte = pmap_pte(kernel_pmap, user_commpage); - if (0 == npte) panic("user_commpage"); pte = *opte | INTEL_PTE_USER|INTEL_PTE_GLOBAL; pte &= ~INTEL_PTE_WRITE; // ensure read only - WRITE_PTE_FAST(npte, pte); + npte = pmap_pte(kernel_pmap, (vm_map_offset_t)user_commpage); + if (0 == npte) panic("user_commpage"); + pmap_store_pte(npte, pte); kernel_commpage += INTEL_PGBYTES; user_commpage += INTEL_PGBYTES; } } +#define PMAP_COMMPAGE64_CNT (_COMM_PAGE64_AREA_USED/PAGE_SIZE) +pt_entry_t pmap_commpage64_ptes[PMAP_COMMPAGE64_CNT]; + +void +pmap_commpage64_init(vm_offset_t kernel_commpage, __unused vm_map_offset_t user_commpage, int cnt) +{ + spl_t s; + int i; + pt_entry_t *kptep; + + s = splhigh(); + for (i = 0; i< cnt; i++) { + kptep = pmap_pte(kernel_pmap, (uint64_t)kernel_commpage + (i*PAGE_SIZE)); + if ((0 == kptep) || (0 == (*kptep & INTEL_PTE_VALID))) panic("pmap_commpage64_init pte"); + pmap_commpage64_ptes[i] = ((*kptep & ~INTEL_PTE_WRITE) | INTEL_PTE_USER); + } + splx(s); + +} + +void +pmap_map_sharedpage(__unused task_t task, pmap_t p) +{ + pt_entry_t *ptep; + spl_t s; + int i; + + if (!p->pm_64bit) return; + /* setup high 64 bit commpage */ + s = splhigh(); + while ((ptep = pmap_pte(p, (uint64_t)_COMM_PAGE64_BASE_ADDRESS)) == PD_ENTRY_NULL) { + splx(s); + pmap_expand(p, (uint64_t)_COMM_PAGE64_BASE_ADDRESS); + s = splhigh(); + } + + for (i = 0; i< PMAP_COMMPAGE64_CNT; i++) { + ptep = pmap_pte(p, (uint64_t)_COMM_PAGE64_BASE_ADDRESS + (i*PAGE_SIZE)); + if (0 == ptep) panic("pmap_map_sharedpage"); + pmap_store_pte(ptep, pmap_commpage64_ptes[i]); + } + splx(s); + +} + +void +pmap_unmap_sharedpage(pmap_t pmap) +{ + spl_t s; + pt_entry_t *ptep; + int i; + + if (!pmap->pm_64bit) return; + s = splhigh(); + for (i = 0; i< PMAP_COMMPAGE64_CNT; i++) { + ptep = pmap_pte(pmap, (uint64_t)_COMM_PAGE64_BASE_ADDRESS + (i*PAGE_SIZE)); + if (ptep) pmap_store_pte(ptep, 0); + } + splx(s); +} + static cpu_pmap_t cpu_pmap_master; -static struct pmap_update_list cpu_update_list_master; struct cpu_pmap * pmap_cpu_alloc(boolean_t is_boot_cpu) @@ -3340,13 +3695,13 @@ pmap_cpu_alloc(boolean_t is_boot_cpu) int ret; int i; cpu_pmap_t *cp; - pmap_update_list_t up; vm_offset_t address; + vm_map_address_t mapaddr; vm_map_entry_t entry; + pt_entry_t *pte; if (is_boot_cpu) { cp = &cpu_pmap_master; - up = &cpu_update_list_master; } else { /* * The per-cpu pmap data structure itself. @@ -3360,46 +3715,31 @@ pmap_cpu_alloc(boolean_t is_boot_cpu) bzero((void *)cp, sizeof(cpu_pmap_t)); /* - * The tlb flush update list. + * The temporary windows used for copy/zero - see loose_ends.c */ - ret = kmem_alloc(kernel_map, - (vm_offset_t *) &up, sizeof(*up)); + ret = vm_map_find_space(kernel_map, + &mapaddr, PMAP_NWINDOWS*PAGE_SIZE, (vm_map_offset_t)0, 0, &entry); if (ret != KERN_SUCCESS) { - printf("pmap_cpu_alloc() failed ret=%d\n", ret); + printf("pmap_cpu_alloc() " + "vm_map_find_space ret=%d\n", ret); pmap_cpu_free(cp); return NULL; } + address = (vm_offset_t)mapaddr; - /* - * The temporary windows used for copy/zero - see loose_ends.c - */ - for (i = 0; i < PMAP_NWINDOWS; i++) { - ret = vm_map_find_space(kernel_map, - &address, PAGE_SIZE, 0, &entry); - if (ret != KERN_SUCCESS) { - printf("pmap_cpu_alloc() " - "vm_map_find_space ret=%d\n", ret); - pmap_cpu_free(cp); - return NULL; - } - vm_map_unlock(kernel_map); - + for (i = 0; i < PMAP_NWINDOWS; i++, address += PAGE_SIZE) { + while ((pte = pmap_pte(kernel_pmap, (vm_map_offset_t)address)) == 0) + pmap_expand(kernel_pmap, (vm_map_offset_t)address); + * (int *) pte = 0; cp->mapwindow[i].prv_CADDR = (caddr_t) address; - cp->mapwindow[i].prv_CMAP = vtopte(address); - * (int *) cp->mapwindow[i].prv_CMAP = 0; - - kprintf("pmap_cpu_alloc() " - "window=%d CADDR=0x%x CMAP=0x%x\n", - i, address, vtopte(address)); + cp->mapwindow[i].prv_CMAP = pte; } + vm_map_unlock(kernel_map); } - /* - * Set up the pmap request list - */ - cp->update_list = up; - simple_lock_init(&up->lock, 0); - up->count = 0; + cp->pdpt_window_index = PMAP_PDPT_FIRST_WINDOW; + cp->pde_window_index = PMAP_PDE_FIRST_WINDOW; + cp->pte_window_index = PMAP_PTE_FIRST_WINDOW; return cp; } @@ -3408,9 +3748,461 @@ void pmap_cpu_free(struct cpu_pmap *cp) { if (cp != NULL && cp != &cpu_pmap_master) { - if (cp->update_list != NULL) - kfree((void *) cp->update_list, - sizeof(*cp->update_list)); kfree((void *) cp, sizeof(cpu_pmap_t)); } } + + +mapwindow_t * +pmap_get_mapwindow(pt_entry_t pentry) +{ + mapwindow_t *mp; + int i; + boolean_t istate; + + /* + * can be called from hardware interrupt context + * so we need to protect the lookup process + */ + istate = ml_set_interrupts_enabled(FALSE); + + /* + * Note: 0th map reserved for pmap_pte() + */ + for (i = PMAP_NWINDOWS_FIRSTFREE; i < PMAP_NWINDOWS; i++) { + mp = ¤t_cpu_datap()->cpu_pmap->mapwindow[i]; + + if (*mp->prv_CMAP == 0) { + *mp->prv_CMAP = pentry; + break; + } + } + if (i >= PMAP_NWINDOWS) + mp = NULL; + (void) ml_set_interrupts_enabled(istate); + + return (mp); +} + + +/* + * kern_return_t pmap_nest(grand, subord, vstart, size) + * + * grand = the pmap that we will nest subord into + * subord = the pmap that goes into the grand + * vstart = start of range in pmap to be inserted + * nstart = start of range in pmap nested pmap + * size = Size of nest area (up to 16TB) + * + * Inserts a pmap into another. This is used to implement shared segments. + * + * on x86 this is very limited right now. must be exactly 1 segment. + * + * Note that we depend upon higher level VM locks to insure that things don't change while + * we are doing this. For example, VM should not be doing any pmap enters while it is nesting + * or do 2 nests at once. + */ + + +kern_return_t pmap_nest(pmap_t grand, pmap_t subord, addr64_t vstart, addr64_t nstart, uint64_t size) { + + vm_map_offset_t vaddr, nvaddr; + pd_entry_t *pde,*npde; + unsigned int i, need_flush; + unsigned int num_pde; + spl_t s; + + // do validity tests + + if(size & 0x0FFFFFFFULL) return KERN_INVALID_VALUE; /* We can only do this for multiples of 256MB */ + if((size >> 28) > 65536) return KERN_INVALID_VALUE; /* Max size we can nest is 16TB */ + if(vstart & 0x0FFFFFFFULL) return KERN_INVALID_VALUE; /* We can only do this aligned to 256MB */ + if(nstart & 0x0FFFFFFFULL) return KERN_INVALID_VALUE; /* We can only do this aligned to 256MB */ + if(size == 0) { + panic("pmap_nest: size is invalid - %016llX\n", size); + } + if ((size >> 28) != 1) panic("pmap_nest: size 0x%llx must be 0x%x", size, NBPDE); + + subord->pm_shared = TRUE; + + // prepopulate subord pmap pde's if necessary + + if (cpu_64bit) { + s = splhigh(); + while (PD_ENTRY_NULL == (npde = pmap_pde(subord, nstart))) { + splx(s); + pmap_expand(subord, nstart); + s = splhigh(); + } + splx(s); + } + + PMAP_READ_LOCK(subord,s); + nvaddr = (vm_map_offset_t)nstart; + need_flush = 0; + num_pde = size >> PDESHIFT; + + for (i=0;i> PDESHIFT; + + for (i=0;inx_enabled = 0; +} + +void +pt_fake_zone_info(int *count, vm_size_t *cur_size, vm_size_t *max_size, vm_size_t *elem_size, + vm_size_t *alloc_size, int *collectable, int *exhaustable) +{ + *count = inuse_ptepages_count; + *cur_size = PAGE_SIZE * inuse_ptepages_count; + *max_size = PAGE_SIZE * (inuse_ptepages_count + vm_page_inactive_count + vm_page_active_count + vm_page_free_count); + *elem_size = PAGE_SIZE; + *alloc_size = PAGE_SIZE; + + *collectable = 1; + *exhaustable = 0; +} + +vm_offset_t pmap_cpu_high_map_vaddr(int cpu, enum high_cpu_types e) +{ + enum high_fixed_addresses a; + a = e + HIGH_CPU_END * cpu; + return pmap_index_to_virt(HIGH_FIXED_CPUS_BEGIN + a); +} + +vm_offset_t pmap_high_map_vaddr(enum high_cpu_types e) +{ + return pmap_cpu_high_map_vaddr(cpu_number(), e); +} + +vm_offset_t pmap_high_map(pt_entry_t pte, enum high_cpu_types e) +{ + enum high_fixed_addresses a; + vm_offset_t vaddr; + + a = e + HIGH_CPU_END * cpu_number(); + vaddr = (vm_offset_t)pmap_index_to_virt(HIGH_FIXED_CPUS_BEGIN + a); + *(pte_unique_base + a) = pte; + + /* TLB flush for this page for this cpu */ + invlpg((uintptr_t)vaddr); + + return vaddr; +} + + +/* + * Called with pmap locked, we: + * - scan through per-cpu data to see which other cpus need to flush + * - send an IPI to each non-idle cpu to be flushed + * - wait for all to signal back that they are inactive or we see that + * they are in an interrupt handler or at a safe point + * - flush the local tlb is active for this pmap + * - return ... the caller will unlock the pmap + */ +void +pmap_flush_tlbs(pmap_t pmap) +{ + unsigned int cpu; + unsigned int cpu_bit; + cpu_set cpus_to_signal; + unsigned int my_cpu = cpu_number(); + pmap_paddr_t pmap_cr3 = pmap->pm_cr3; + boolean_t flush_self = FALSE; + uint64_t deadline; + + assert(!ml_get_interrupts_enabled()); + + /* + * Scan other cpus for matching active or task CR3. + * For idle cpus (with no active map) we mark them invalid but + * don't signal -- they'll check as they go busy. + * Note: for the kernel pmap we look for 64-bit shared address maps. + */ + cpus_to_signal = 0; + for (cpu = 0, cpu_bit = 1; cpu < real_ncpus; cpu++, cpu_bit <<= 1) { + if (!cpu_datap(cpu)->cpu_running) + continue; + if ((cpu_datap(cpu)->cpu_task_cr3 == pmap_cr3) || + (CPU_GET_ACTIVE_CR3(cpu) == pmap_cr3) || + (pmap->pm_shared) || + ((pmap == kernel_pmap) && + (!CPU_CR3_IS_ACTIVE(cpu) || + cpu_datap(cpu)->cpu_task_map == TASK_MAP_64BIT_SHARED))) { + if (cpu == my_cpu) { + flush_self = TRUE; + continue; + } + cpu_datap(cpu)->cpu_tlb_invalid = TRUE; + __asm__ volatile("mfence"); + + if (CPU_CR3_IS_ACTIVE(cpu)) { + cpus_to_signal |= cpu_bit; + i386_signal_cpu(cpu, MP_TLB_FLUSH, ASYNC); + } + } + } + + if (cpus_to_signal) { + KERNEL_DEBUG(0xef800024 | DBG_FUNC_START, cpus_to_signal, 0, 0, 0, 0); + + deadline = mach_absolute_time() + LockTimeOut; + /* + * Wait for those other cpus to acknowledge + */ + for (cpu = 0, cpu_bit = 1; cpu < real_ncpus; cpu++, cpu_bit <<= 1) { + while ((cpus_to_signal & cpu_bit) != 0) { + if (!cpu_datap(cpu)->cpu_running || + cpu_datap(cpu)->cpu_tlb_invalid == FALSE || + !CPU_CR3_IS_ACTIVE(cpu)) { + cpus_to_signal &= ~cpu_bit; + break; + } + if (mach_absolute_time() > deadline) + panic("pmap_flush_tlbs() " + "timeout pmap=%p cpus_to_signal=%p", + pmap, cpus_to_signal); + cpu_pause(); + } + if (cpus_to_signal == 0) + break; + } + KERNEL_DEBUG(0xef800024 | DBG_FUNC_END, cpus_to_signal, 0, 0, 0, 0); + } + + /* + * Flush local tlb if required. + * We need this flush even if the pmap being changed + * is the user map... in case we do a copyin/out + * before returning to user mode. + */ + if (flush_self) + flush_tlb(); + +} + +void +process_pmap_updates(void) +{ + flush_tlb(); + + current_cpu_datap()->cpu_tlb_invalid = FALSE; + __asm__ volatile("mfence"); +} + +void +pmap_update_interrupt(void) +{ + KERNEL_DEBUG(0xef800028 | DBG_FUNC_START, 0, 0, 0, 0, 0); + + assert(!ml_get_interrupts_enabled()); + + process_pmap_updates(); + + KERNEL_DEBUG(0xef800028 | DBG_FUNC_END, 0, 0, 0, 0, 0); +} + + +unsigned int pmap_cache_attributes(ppnum_t pn) { + + if (!pmap_valid_page(pn)) + return (VM_WIMG_IO); + + return (VM_WIMG_COPYBACK); +} + +#ifdef PMAP_DEBUG +void +pmap_dump(pmap_t p) +{ + int i; + + kprintf("pmap 0x%x\n",p); + + kprintf(" pm_cr3 0x%llx\n",p->pm_cr3); + kprintf(" pm_pml4 0x%x\n",p->pm_pml4); + kprintf(" pm_pdpt 0x%x\n",p->pm_pdpt); + + kprintf(" pml4[0] 0x%llx\n",*p->pm_pml4); + for (i=0;i<8;i++) + kprintf(" pdpt[%d] 0x%llx\n",i, p->pm_pdpt[i]); +} + +void pmap_dump_wrap(void) +{ + pmap_dump(current_cpu_datap()->cpu_active_thread->task->map->pmap); +} + +void +dump_4GB_pdpt(pmap_t p) +{ + int spl; + pdpt_entry_t *user_pdptp; + pdpt_entry_t *kern_pdptp; + pdpt_entry_t *pml4p; + + spl = splhigh(); + while ((user_pdptp = pmap64_pdpt(p, 0x0)) == PDPT_ENTRY_NULL) { + splx(spl); + pmap_expand_pml4(p, 0x0); + spl = splhigh(); + } + kern_pdptp = kernel_pmap->pm_pdpt; + if (kern_pdptp == NULL) + panic("kern_pdptp == NULL"); + kprintf("dump_4GB_pdpt(%p)\n" + "kern_pdptp=%p (phys=0x%016llx)\n" + "\t 0x%08x: 0x%016llx\n" + "\t 0x%08x: 0x%016llx\n" + "\t 0x%08x: 0x%016llx\n" + "\t 0x%08x: 0x%016llx\n" + "\t 0x%08x: 0x%016llx\n" + "user_pdptp=%p (phys=0x%016llx)\n" + "\t 0x%08x: 0x%016llx\n" + "\t 0x%08x: 0x%016llx\n" + "\t 0x%08x: 0x%016llx\n" + "\t 0x%08x: 0x%016llx\n" + "\t 0x%08x: 0x%016llx\n", + p, kern_pdptp, kvtophys(kern_pdptp), + kern_pdptp+0, *(kern_pdptp+0), + kern_pdptp+1, *(kern_pdptp+1), + kern_pdptp+2, *(kern_pdptp+2), + kern_pdptp+3, *(kern_pdptp+3), + kern_pdptp+4, *(kern_pdptp+4), + user_pdptp, kvtophys(user_pdptp), + user_pdptp+0, *(user_pdptp+0), + user_pdptp+1, *(user_pdptp+1), + user_pdptp+2, *(user_pdptp+2), + user_pdptp+3, *(user_pdptp+3), + user_pdptp+4, *(user_pdptp+4)); + kprintf("user pm_cr3=0x%016llx pm_hold=0x%08x pm_pml4=0x%08x\n", + p->pm_cr3, p->pm_hold, p->pm_pml4); + pml4p = (pdpt_entry_t *)p->pm_hold; + if (pml4p == NULL) + panic("user pml4p == NULL"); + kprintf("\t 0x%08x: 0x%016llx\n" + "\t 0x%08x: 0x%016llx\n", + pml4p+0, *(pml4p), + pml4p+KERNEL_UBER_PML4_INDEX, *(pml4p+KERNEL_UBER_PML4_INDEX)); + kprintf("kern pm_cr3=0x%016llx pm_hold=0x%08x pm_pml4=0x%08x\n", + kernel_pmap->pm_cr3, kernel_pmap->pm_hold, kernel_pmap->pm_pml4); + pml4p = (pdpt_entry_t *)kernel_pmap->pm_hold; + if (pml4p == NULL) + panic("kern pml4p == NULL"); + kprintf("\t 0x%08x: 0x%016llx\n" + "\t 0x%08x: 0x%016llx\n", + pml4p+0, *(pml4p), + pml4p+511, *(pml4p+511)); + splx(spl); +} + +void dump_4GB_pdpt_thread(thread_t tp) +{ + dump_4GB_pdpt(tp->map->pmap); +} + + +#endif