X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/7ddcb079202367355dddccdfa4318e57d50318be..a1c7dba18ef36983396c282fe85292db066e39db:/osfmk/i386/i386_vm_init.c?ds=sidebyside diff --git a/osfmk/i386/i386_vm_init.c b/osfmk/i386/i386_vm_init.c index 866dfa1fb..8a1d753b5 100644 --- a/osfmk/i386/i386_vm_init.c +++ b/osfmk/i386/i386_vm_init.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2003-2008 Apple Computer, Inc. All rights reserved. + * Copyright (c) 2003-2012 Apple Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ * @@ -54,8 +54,6 @@ * the rights to redistribute these changes. */ -#include -#include #include @@ -79,18 +77,39 @@ #include #include #include -#include +#include #include #include #include + vm_size_t mem_size = 0; pmap_paddr_t first_avail = 0;/* first after page tables */ uint64_t max_mem; /* Size of physical memory (bytes), adjusted by maxmem */ uint64_t mem_actual; -uint64_t sane_size = 0; /* Memory size to use for defaults calculations */ +uint64_t sane_size = 0; /* Memory size for defaults calculations */ + +/* + * KASLR parameters + */ +ppnum_t vm_kernel_base_page; +vm_offset_t vm_kernel_base; +vm_offset_t vm_kernel_top; +vm_offset_t vm_kernel_stext; +vm_offset_t vm_kernel_etext; +vm_offset_t vm_kernel_slide; +vm_offset_t vm_hib_base; +vm_offset_t vm_kext_base = VM_MIN_KERNEL_AND_KEXT_ADDRESS; +vm_offset_t vm_kext_top = VM_MIN_KERNEL_ADDRESS; + +vm_offset_t vm_prelink_stext; +vm_offset_t vm_prelink_etext; +vm_offset_t vm_prelink_sinfo; +vm_offset_t vm_prelink_einfo; +vm_offset_t vm_slinkedit; +vm_offset_t vm_elinkedit; #define MAXLORESERVE (32 * 1024 * 1024) @@ -112,21 +131,24 @@ vm_offset_t virtual_avail, virtual_end; static pmap_paddr_t avail_remaining; vm_offset_t static_memory_end = 0; -vm_offset_t sHIB, eHIB, stext, etext, sdata, edata, end; +vm_offset_t sHIB, eHIB, stext, etext, sdata, edata, sconstdata, econstdata, end; /* * _mh_execute_header is the mach_header for the currently executing kernel */ -void *sectTEXTB; unsigned long sectSizeTEXT; -void *sectDATAB; unsigned long sectSizeDATA; -void *sectOBJCB; unsigned long sectSizeOBJC; -void *sectLINKB; unsigned long sectSizeLINK; -void *sectPRELINKB; unsigned long sectSizePRELINK; -void *sectHIBB; unsigned long sectSizeHIB; -void *sectINITPTB; unsigned long sectSizeINITPT; +vm_offset_t segTEXTB; unsigned long segSizeTEXT; +vm_offset_t segDATAB; unsigned long segSizeDATA; +vm_offset_t segLINKB; unsigned long segSizeLINK; +vm_offset_t segPRELINKB; unsigned long segSizePRELINK; +vm_offset_t segPRELINKINFOB; unsigned long segSizePRELINKINFO; +vm_offset_t segHIBB; unsigned long segSizeHIB; +vm_offset_t sectCONSTB; unsigned long sectSizeConst; + +boolean_t doconstro_override = FALSE; -kernel_segment_command_t *segTEXT; -kernel_section_t *cursectTEXT, *lastsectTEXT; +static kernel_segment_command_t *segTEXT, *segDATA; +static kernel_section_t *cursectTEXT, *lastsectTEXT; +static kernel_section_t *sectDCONST; extern uint64_t firmware_Conventional_bytes; extern uint64_t firmware_RuntimeServices_bytes; @@ -138,8 +160,16 @@ extern uint64_t firmware_Unusable_bytes; extern uint64_t firmware_other_bytes; uint64_t firmware_MMIO_bytes; +/* + * Linker magic to establish the highest address in the kernel. + */ +extern void *last_kernel_symbol; + #if DEBUG #define PRINT_PMAP_MEMORY_TABLE +#define DBG(x...) kprintf(x) +#else +#define DBG(x...) #endif /* DEBUG */ /* * Basic VM initialization. @@ -156,7 +186,6 @@ i386_vm_init(uint64_t maxmem, unsigned int msize; ppnum_t fap; unsigned int i; - unsigned int safeboot; ppnum_t maxpg = 0; uint32_t pmap_type; uint32_t maxloreserve; @@ -164,64 +193,129 @@ i386_vm_init(uint64_t maxmem, uint32_t mbuf_reserve = 0; boolean_t mbuf_override = FALSE; boolean_t coalescing_permitted; -#if DEBUG - kprintf("Boot args revision: %d version: %d", - args->Revision, args->Version); - kprintf(" commandline: \""); - for(i=0; iCommandLine[i]); - kprintf("\"\n"); -#endif + vm_kernel_base_page = i386_btop(args->kaddr); + vm_offset_t base_address; + vm_offset_t static_base_address; + + /* + * Establish the KASLR parameters. + */ + static_base_address = ml_static_ptovirt(KERNEL_BASE_OFFSET); + base_address = ml_static_ptovirt(args->kaddr); + vm_kernel_slide = base_address - static_base_address; + if (args->kslide) { + kprintf("KASLR slide: 0x%016lx dynamic\n", vm_kernel_slide); + if (vm_kernel_slide != ((vm_offset_t)args->kslide)) + panic("Kernel base inconsistent with slide - rebased?"); + } else { + /* No slide relative to on-disk symbols */ + kprintf("KASLR slide: 0x%016lx static and ignored\n", + vm_kernel_slide); + vm_kernel_slide = 0; + } + + /* + * Zero out local relocations to avoid confusing kxld. + * TODO: might be better to move this code to OSKext::initialize + */ + if (_mh_execute_header.flags & MH_PIE) { + struct load_command *loadcmd; + uint32_t cmd; + + loadcmd = (struct load_command *)((uintptr_t)&_mh_execute_header + + sizeof (_mh_execute_header)); + + for (cmd = 0; cmd < _mh_execute_header.ncmds; cmd++) { + if (loadcmd->cmd == LC_DYSYMTAB) { + struct dysymtab_command *dysymtab; + + dysymtab = (struct dysymtab_command *)loadcmd; + dysymtab->nlocrel = 0; + dysymtab->locreloff = 0; + kprintf("Hiding local relocations\n"); + break; + } + loadcmd = (struct load_command *)((uintptr_t)loadcmd + loadcmd->cmdsize); + } + } /* * Now retrieve addresses for end, edata, and etext * from MACH-O headers. */ - - sectTEXTB = (void *) getsegdatafromheader( - &_mh_execute_header, "__TEXT", §SizeTEXT); - sectDATAB = (void *) getsegdatafromheader( - &_mh_execute_header, "__DATA", §SizeDATA); - sectOBJCB = (void *) getsegdatafromheader( - &_mh_execute_header, "__OBJC", §SizeOBJC); - sectLINKB = (void *) getsegdatafromheader( - &_mh_execute_header, "__LINKEDIT", §SizeLINK); - sectHIBB = (void *)getsegdatafromheader( - &_mh_execute_header, "__HIB", §SizeHIB); - sectINITPTB = (void *)getsegdatafromheader( - &_mh_execute_header, "__INITPT", §SizeINITPT); - sectPRELINKB = (void *) getsegdatafromheader( - &_mh_execute_header, "__PRELINK_TEXT", §SizePRELINK); - - segTEXT = getsegbynamefromheader(&_mh_execute_header, "__TEXT"); + segTEXTB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, + "__TEXT", &segSizeTEXT); + segDATAB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, + "__DATA", &segSizeDATA); + segLINKB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, + "__LINKEDIT", &segSizeLINK); + segHIBB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, + "__HIB", &segSizeHIB); + segPRELINKB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, + "__PRELINK_TEXT", &segSizePRELINK); + segPRELINKINFOB = (vm_offset_t) getsegdatafromheader(&_mh_execute_header, + "__PRELINK_INFO", &segSizePRELINKINFO); + segTEXT = getsegbynamefromheader(&_mh_execute_header, + "__TEXT"); + segDATA = getsegbynamefromheader(&_mh_execute_header, + "__DATA"); + sectDCONST = getsectbynamefromheader(&_mh_execute_header, + "__DATA", "__const"); cursectTEXT = lastsectTEXT = firstsect(segTEXT); /* Discover the last TEXT section within the TEXT segment */ while ((cursectTEXT = nextsect(segTEXT, cursectTEXT)) != NULL) { lastsectTEXT = cursectTEXT; } - sHIB = (vm_offset_t) sectHIBB; - eHIB = (vm_offset_t) sectHIBB + sectSizeHIB; + sHIB = segHIBB; + eHIB = segHIBB + segSizeHIB; + vm_hib_base = sHIB; /* Zero-padded from ehib to stext if text is 2M-aligned */ - stext = (vm_offset_t) sectTEXTB; + stext = segTEXTB; + lowGlo.lgStext = stext; etext = (vm_offset_t) round_page_64(lastsectTEXT->addr + lastsectTEXT->size); /* Zero-padded from etext to sdata if text is 2M-aligned */ - sdata = (vm_offset_t) sectDATAB; - edata = (vm_offset_t) sectDATAB + sectSizeDATA; - -#if DEBUG - kprintf("sectTEXTB = %p\n", sectTEXTB); - kprintf("sectDATAB = %p\n", sectDATAB); - kprintf("sectOBJCB = %p\n", sectOBJCB); - kprintf("sectLINKB = %p\n", sectLINKB); - kprintf("sectHIBB = %p\n", sectHIBB); - kprintf("sectPRELINKB = %p\n", sectPRELINKB); - kprintf("eHIB = %p\n", (void *) eHIB); - kprintf("stext = %p\n", (void *) stext); - kprintf("etext = %p\n", (void *) etext); - kprintf("sdata = %p\n", (void *) sdata); - kprintf("edata = %p\n", (void *) edata); -#endif + sdata = segDATAB; + edata = segDATAB + segSizeDATA; + + sectCONSTB = (vm_offset_t) sectDCONST->addr; + sectSizeConst = sectDCONST->size; + sconstdata = sectCONSTB; + econstdata = sectCONSTB + sectSizeConst; + + if (sectSizeConst & PAGE_MASK) { + kernel_section_t *ns = nextsect(segDATA, sectDCONST); + if (ns && !(ns->addr & PAGE_MASK)) + doconstro_override = TRUE; + } else + doconstro_override = TRUE; + + DBG("segTEXTB = %p\n", (void *) segTEXTB); + DBG("segDATAB = %p\n", (void *) segDATAB); + DBG("segLINKB = %p\n", (void *) segLINKB); + DBG("segHIBB = %p\n", (void *) segHIBB); + DBG("segPRELINKB = %p\n", (void *) segPRELINKB); + DBG("segPRELINKINFOB = %p\n", (void *) segPRELINKINFOB); + DBG("sHIB = %p\n", (void *) sHIB); + DBG("eHIB = %p\n", (void *) eHIB); + DBG("stext = %p\n", (void *) stext); + DBG("etext = %p\n", (void *) etext); + DBG("sdata = %p\n", (void *) sdata); + DBG("edata = %p\n", (void *) edata); + DBG("sconstdata = %p\n", (void *) sconstdata); + DBG("econstdata = %p\n", (void *) econstdata); + DBG("kernel_top = %p\n", (void *) &last_kernel_symbol); + + vm_kernel_base = sHIB; + vm_kernel_top = (vm_offset_t) &last_kernel_symbol; + vm_kernel_stext = stext; + vm_kernel_etext = etext; + vm_prelink_stext = segPRELINKB; + vm_prelink_etext = segPRELINKB + segSizePRELINK; + vm_prelink_sinfo = segPRELINKINFOB; + vm_prelink_einfo = segPRELINKINFOB + segSizePRELINKINFO; + vm_slinkedit = segLINKB; + vm_elinkedit = segLINKB + segSizePRELINK; vm_set_page_size(); @@ -229,9 +323,6 @@ i386_vm_init(uint64_t maxmem, * Compute the memory size. */ - if ((1 == vm_himemory_mode) || PE_parse_boot_argn("-x", &safeboot, sizeof (safeboot))) { - maxpg = 1 << (32 - I386_PGSHIFT); - } avail_remaining = 0; avail_end = 0; pmptr = pmap_memory_regions; @@ -259,6 +350,32 @@ i386_vm_init(uint64_t maxmem, base = (ppnum_t) (mptr->PhysicalStart >> I386_PGSHIFT); top = (ppnum_t) (((mptr->PhysicalStart) >> I386_PGSHIFT) + mptr->NumberOfPages - 1); + if (base == 0) { + /* + * Avoid having to deal with the edge case of the + * very first possible physical page and the roll-over + * to -1; just ignore that page. + */ + kprintf("WARNING: ignoring first page in [0x%llx:0x%llx]\n", (uint64_t) base, (uint64_t) top); + base++; + } + if (top + 1 == 0) { + /* + * Avoid having to deal with the edge case of the + * very last possible physical page and the roll-over + * to 0; just ignore that page. + */ + kprintf("WARNING: ignoring last page in [0x%llx:0x%llx]\n", (uint64_t) base, (uint64_t) top); + top--; + } + if (top < base) { + /* + * That was the only page in that region, so + * ignore the whole region. + */ + continue; + } + #if MR_RSV_TEST static uint32_t nmr = 0; if ((base > 0x20000) && (nmr++ < 4)) @@ -328,10 +445,10 @@ i386_vm_init(uint64_t maxmem, break; } -#if DEBUG - kprintf("EFI region %d: type %u/%d, base 0x%x, top 0x%x\n", - i, mptr->Type, pmap_type, base, top); -#endif + DBG("EFI region %d: type %u/%d, base 0x%x, top 0x%x %s\n", + i, mptr->Type, pmap_type, base, top, + (mptr->Attribute&EFI_MEMORY_KERN_RESERVED)? "RESERVED" : + (mptr->Attribute&EFI_MEMORY_RUNTIME)? "RUNTIME" : ""); if (maxpg) { if (base >= maxpg) @@ -384,15 +501,17 @@ i386_vm_init(uint64_t maxmem, if ((mptr->Attribute & EFI_MEMORY_KERN_RESERVED) && - (top < I386_KERNEL_IMAGE_BASE_PAGE)) { - pmptr->alloc = pmptr->base; + (top < vm_kernel_base_page)) { + pmptr->alloc_up = pmptr->base; + pmptr->alloc_down = pmptr->end; pmap_reserved_range_indices[pmap_last_reserved_range_index++] = pmap_memory_region_count; } else { /* * mark as already mapped */ - pmptr->alloc = top; + pmptr->alloc_up = top + 1; + pmptr->alloc_down = top; } pmptr->type = pmap_type; pmptr->attribute = mptr->Attribute; @@ -404,7 +523,9 @@ i386_vm_init(uint64_t maxmem, * mark already allocated */ pmptr->base = base; - pmptr->alloc = pmptr->end = (fap - 1); + pmptr->end = (fap - 1); + pmptr->alloc_up = pmptr->end + 1; + pmptr->alloc_down = pmptr->end; pmptr->type = pmap_type; pmptr->attribute = mptr->Attribute; /* @@ -414,10 +535,10 @@ i386_vm_init(uint64_t maxmem, pmptr++; pmap_memory_region_count++; - pmptr->alloc = pmptr->base = fap; + pmptr->alloc_up = pmptr->base = fap; pmptr->type = pmap_type; pmptr->attribute = mptr->Attribute; - pmptr->end = top; + pmptr->alloc_down = pmptr->end = top; if (mptr->Attribute & EFI_MEMORY_KERN_RESERVED) pmap_reserved_range_indices[pmap_last_reserved_range_index++] = pmap_memory_region_count; @@ -425,10 +546,10 @@ i386_vm_init(uint64_t maxmem, /* * entire range useable */ - pmptr->alloc = pmptr->base = base; + pmptr->alloc_up = pmptr->base = base; pmptr->type = pmap_type; pmptr->attribute = mptr->Attribute; - pmptr->end = top; + pmptr->alloc_down = pmptr->end = top; if (mptr->Attribute & EFI_MEMORY_KERN_RESERVED) pmap_reserved_range_indices[pmap_last_reserved_range_index++] = pmap_memory_region_count; } @@ -444,12 +565,12 @@ i386_vm_init(uint64_t maxmem, if (prev_pmptr && (pmptr->type == prev_pmptr->type) && (coalescing_permitted) && - (pmptr->base == pmptr->alloc) && + (pmptr->base == pmptr->alloc_up) && + (prev_pmptr->end == prev_pmptr->alloc_down) && (pmptr->base == (prev_pmptr->end + 1))) { - if (prev_pmptr->end == prev_pmptr->alloc) - prev_pmptr->alloc = pmptr->base; prev_pmptr->end = pmptr->end; + prev_pmptr->alloc_down = pmptr->alloc_down; } else { pmap_memory_region_count++; prev_pmptr = pmptr; @@ -465,10 +586,11 @@ i386_vm_init(uint64_t maxmem, addr64_t region_start, region_end; addr64_t efi_start, efi_end; for (j=0;jtype, (addr64_t) p->base << I386_PGSHIFT, - (addr64_t) p->alloc << I386_PGSHIFT, + (addr64_t) p->alloc_up << I386_PGSHIFT, + (addr64_t) p->alloc_down << I386_PGSHIFT, (addr64_t) p->end << I386_PGSHIFT); region_start = (addr64_t) p->base << I386_PGSHIFT; region_end = ((addr64_t) p->end << I386_PGSHIFT) - 1; @@ -518,7 +640,7 @@ i386_vm_init(uint64_t maxmem, if ( (maxmem > (uint64_t)first_avail) && (maxmem < sane_size)) { ppnum_t discarded_pages = (ppnum_t)((sane_size - maxmem) >> I386_PGSHIFT); ppnum_t highest_pn = 0; - ppnum_t cur_alloc = 0; + ppnum_t cur_end = 0; uint64_t pages_to_use; unsigned cur_region = 0; @@ -532,15 +654,17 @@ i386_vm_init(uint64_t maxmem, pages_to_use = avail_remaining; while (cur_region < pmap_memory_region_count && pages_to_use) { - for (cur_alloc = pmap_memory_regions[cur_region].alloc; - cur_alloc < pmap_memory_regions[cur_region].end && pages_to_use; - cur_alloc++) { - if (cur_alloc > highest_pn) - highest_pn = cur_alloc; + for (cur_end = pmap_memory_regions[cur_region].base; + cur_end < pmap_memory_regions[cur_region].end && pages_to_use; + cur_end++) { + if (cur_end > highest_pn) + highest_pn = cur_end; pages_to_use--; } - if (pages_to_use == 0) - pmap_memory_regions[cur_region].end = cur_alloc; + if (pages_to_use == 0) { + pmap_memory_regions[cur_region].end = cur_end; + pmap_memory_regions[cur_region].alloc_down = cur_end; + } cur_region++; } @@ -582,7 +706,9 @@ i386_vm_init(uint64_t maxmem, else maxloreserve = MAXLORESERVE / PAGE_SIZE; +#if SOCKETS mbuf_reserve = bsd_mbuf_cluster_reserve(&mbuf_override) / PAGE_SIZE; +#endif } else maxloreserve = (maxloreserve * (1024 * 1024)) / PAGE_SIZE; @@ -630,8 +756,8 @@ pmap_next_page_reserved(ppnum_t *pn) { for (n = 0; n < pmap_last_reserved_range_index; n++) { uint32_t reserved_index = pmap_reserved_range_indices[n]; region = &pmap_memory_regions[reserved_index]; - if (region->alloc < region->end) { - *pn = region->alloc++; + if (region->alloc_up <= region->alloc_down) { + *pn = region->alloc_up++; avail_remaining--; if (*pn > max_ppnum) @@ -642,7 +768,7 @@ pmap_next_page_reserved(ppnum_t *pn) { pmap_reserved_pages_allocated++; #if DEBUG - if (region->alloc == region->end) { + if (region->alloc_up > region->alloc_down) { kprintf("Exhausted reserved range index: %u, base: 0x%x end: 0x%x, type: 0x%x, attribute: 0x%llx\n", reserved_index, region->base, region->end, region->type, region->attribute); } #endif @@ -668,8 +794,8 @@ pmap_next_page_hi( for (n = pmap_memory_region_count - 1; n >= 0; n--) { region = &pmap_memory_regions[n]; - if (region->alloc != region->end) { - *pn = region->alloc++; + if (region->alloc_down >= region->alloc_up) { + *pn = region->alloc_down--; avail_remaining--; if (*pn > max_ppnum) @@ -697,12 +823,12 @@ pmap_next_page( ppnum_t *pn) { if (avail_remaining) while (pmap_memory_region_current < pmap_memory_region_count) { - if (pmap_memory_regions[pmap_memory_region_current].alloc == - pmap_memory_regions[pmap_memory_region_current].end) { + if (pmap_memory_regions[pmap_memory_region_current].alloc_up > + pmap_memory_regions[pmap_memory_region_current].alloc_down) { pmap_memory_region_current++; continue; } - *pn = pmap_memory_regions[pmap_memory_region_current].alloc++; + *pn = pmap_memory_regions[pmap_memory_region_current].alloc_up++; avail_remaining--; if (*pn > max_ppnum)