X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/2d21ac55c334faf3a56e5634905ed6987fc787d4..HEAD:/osfmk/i386/machine_routines.c diff --git a/osfmk/i386/machine_routines.c b/osfmk/i386/machine_routines.c index 4ffb4ddc3..7d0e21dac 100644 --- a/osfmk/i386/machine_routines.c +++ b/osfmk/i386/machine_routines.c @@ -1,8 +1,8 @@ /* - * Copyright (c) 2000-2007 Apple Inc. All rights reserved. + * Copyright (c) 2000-2012 Apple Inc. All rights reserved. * * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ - * + * * This file contains Original Code and/or Modifications of Original Code * as defined in and that are subject to the Apple Public Source License * Version 2.0 (the 'License'). You may not use this file except in @@ -11,10 +11,10 @@ * unlawful or unlicensed copies of an Apple operating system, or to * circumvent, violate, or enable the circumvention or violation of, any * terms of an Apple operating system software license agreement. - * + * * Please obtain a copy of the License at * http://www.opensource.apple.com/apsl/ and read it before using this file. - * + * * The Original Code and all software distributed under the License are * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, @@ -22,7 +22,7 @@ * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. * Please see the License for the specific language governing rights and * limitations under the License. - * + * * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ */ @@ -33,130 +33,244 @@ #include #include #include -#include + #include #include -#include +#include +#include + +#include +#include #include -#include +#include +#include #include -#include -#include #include -#include +#include #include #include +#include #include -#if MACH_KDB -#include -#include -#include -#include -#include -#include -#include -#include +#include +#include +#include +#include +#include +#if KPC +#include #endif - +#include +#include #if DEBUG -#define DBG(x...) kprintf("DBG: " x) +#define DBG(x...) kprintf("DBG: " x) #else #define DBG(x...) #endif -extern thread_t Shutdown_context(thread_t thread, void (*doshutdown)(processor_t),processor_t processor); -extern void wakeup(void *); -extern unsigned KernelRelocOffset; +#if MONOTONIC +#include +#endif /* MONOTONIC */ -static int max_cpus_initialized = 0; +extern void wakeup(void *); + +uint64_t LockTimeOut; +uint64_t TLBTimeOut; +uint64_t LockTimeOutTSC; +uint32_t LockTimeOutUsec; +uint64_t MutexSpin; +uint64_t low_MutexSpin; +int64_t high_MutexSpin; +uint64_t LastDebuggerEntryAllowance; +uint64_t delay_spin_threshold; + +extern uint64_t panic_restart_timeout; + +boolean_t virtualized = FALSE; -unsigned int LockTimeOut; -unsigned int LockTimeOutTSC; -unsigned int MutexSpin; +decl_simple_lock_data(static, ml_timer_evaluation_slock); +uint32_t ml_timer_eager_evaluations; +uint64_t ml_timer_eager_evaluation_max; +static boolean_t ml_timer_evaluation_in_progress = FALSE; +LCK_GRP_DECLARE(max_cpus_grp, "max_cpus"); +LCK_MTX_DECLARE(max_cpus_lock, &max_cpus_grp); +static int max_cpus_initialized = 0; #define MAX_CPUS_SET 0x1 #define MAX_CPUS_WAIT 0x2 /* IO memory map services */ /* Map memory map IO space */ -vm_offset_t ml_io_map( - vm_offset_t phys_addr, +vm_offset_t +ml_io_map( + vm_offset_t phys_addr, vm_size_t size) { - return(io_map(phys_addr,size,VM_WIMG_IO)); + return io_map(phys_addr, size, VM_WIMG_IO); } /* boot memory allocation */ -vm_offset_t ml_static_malloc( - __unused vm_size_t size) +vm_offset_t +ml_static_malloc( + __unused vm_size_t size) { - return((vm_offset_t)NULL); + return (vm_offset_t)NULL; } -void ml_get_bouncepool_info(vm_offset_t *phys_addr, vm_size_t *size) +void +ml_get_bouncepool_info(vm_offset_t *phys_addr, vm_size_t *size) { - *phys_addr = bounce_pool_base; - *size = bounce_pool_size; + *phys_addr = 0; + *size = 0; } vm_offset_t -ml_boot_ptovirt( +ml_static_ptovirt( vm_offset_t paddr) { - return (vm_offset_t)((paddr-KernelRelocOffset) | LINEAR_KERNEL_ADDRESS); -} +#if defined(__x86_64__) + return (vm_offset_t)(((unsigned long) paddr) | VM_MIN_KERNEL_ADDRESS); +#else + return (vm_offset_t)((paddr) | LINEAR_KERNEL_ADDRESS); +#endif +} vm_offset_t -ml_static_ptovirt( - vm_offset_t paddr) +ml_static_slide( + vm_offset_t vaddr) { - return (vm_offset_t)((unsigned) paddr | LINEAR_KERNEL_ADDRESS); -} + return VM_KERNEL_SLIDE(vaddr); +} + +/* + * base must be page-aligned, and size must be a multiple of PAGE_SIZE + */ +kern_return_t +ml_static_verify_page_protections( + uint64_t base, uint64_t size, vm_prot_t prot) +{ + vm_prot_t pageprot; + uint64_t offset; + + DBG("ml_static_verify_page_protections: vaddr 0x%llx sz 0x%llx prot 0x%x\n", base, size, prot); + + /* + * base must be within the static bounds, defined to be: + * (vm_kernel_stext, kc_highest_nonlinkedit_vmaddr) + */ +#if DEVELOPMENT || DEBUG || KASAN + assert(kc_highest_nonlinkedit_vmaddr > 0 && base > vm_kernel_stext && base < kc_highest_nonlinkedit_vmaddr); +#else /* On release kernels, assume this is a protection mismatch failure. */ + if (kc_highest_nonlinkedit_vmaddr == 0 || base < vm_kernel_stext || base >= kc_highest_nonlinkedit_vmaddr) { + return KERN_FAILURE; + } +#endif + + for (offset = 0; offset < size; offset += PAGE_SIZE) { + if (pmap_get_prot(kernel_pmap, base + offset, &pageprot) == KERN_FAILURE) { + return KERN_FAILURE; + } + if ((pageprot & prot) != prot) { + return KERN_FAILURE; + } + } + + return KERN_SUCCESS; +} +vm_offset_t +ml_static_unslide( + vm_offset_t vaddr) +{ + return VM_KERNEL_UNSLIDE(vaddr); +} /* - * Routine: ml_static_mfree - * Function: + * Reclaim memory, by virtual address, that was used in early boot that is no longer needed + * by the kernel. */ void ml_static_mfree( vm_offset_t vaddr, vm_size_t size) { - vm_offset_t vaddr_cur; + addr64_t vaddr_cur; ppnum_t ppn; + uint32_t freed_pages = 0; + vm_size_t map_size; + + assert(vaddr >= VM_MIN_KERNEL_ADDRESS); -// if (vaddr < VM_MIN_KERNEL_ADDRESS) return; + assert((vaddr & (PAGE_SIZE - 1)) == 0); /* must be page aligned */ - assert((vaddr & (PAGE_SIZE-1)) == 0); /* must be page aligned */ + for (vaddr_cur = vaddr; vaddr_cur < round_page_64(vaddr + size);) { + map_size = pmap_query_pagesize(kernel_pmap, vaddr_cur); - for (vaddr_cur = vaddr; - vaddr_cur < round_page_32(vaddr+size); - vaddr_cur += PAGE_SIZE) { - ppn = pmap_find_phys(kernel_pmap, (addr64_t)vaddr_cur); - if (ppn != (vm_offset_t)NULL) { - kernel_pmap->stats.resident_count++; - if (kernel_pmap->stats.resident_count > - kernel_pmap->stats.resident_max) { - kernel_pmap->stats.resident_max = - kernel_pmap->stats.resident_count; + /* just skip if nothing mapped here */ + if (map_size == 0) { + vaddr_cur += PAGE_SIZE; + continue; + } + + /* + * Can't free from the middle of a large page. + */ + assert((vaddr_cur & (map_size - 1)) == 0); + + ppn = pmap_find_phys(kernel_pmap, vaddr_cur); + assert(ppn != (ppnum_t)NULL); + + pmap_remove(kernel_pmap, vaddr_cur, vaddr_cur + map_size); + while (map_size > 0) { + if (++kernel_pmap->stats.resident_count > kernel_pmap->stats.resident_max) { + kernel_pmap->stats.resident_max = kernel_pmap->stats.resident_count; } - pmap_remove(kernel_pmap, (addr64_t)vaddr_cur, (addr64_t)(vaddr_cur+PAGE_SIZE)); - vm_page_create(ppn,(ppn+1)); - vm_page_wire_count--; + + assert(pmap_valid_page(ppn)); + if (IS_MANAGED_PAGE(ppn)) { + vm_page_create(ppn, (ppn + 1)); + freed_pages++; + } + map_size -= PAGE_SIZE; + vaddr_cur += PAGE_SIZE; + ppn++; } } + vm_page_lockspin_queues(); + vm_page_wire_count -= freed_pages; + vm_page_wire_count_initial -= freed_pages; + if (vm_page_wire_count_on_boot != 0) { + assert(vm_page_wire_count_on_boot >= freed_pages); + vm_page_wire_count_on_boot -= freed_pages; + } + vm_page_unlock_queues(); + +#if DEBUG + kprintf("ml_static_mfree: Released 0x%x pages at VA %p, size:0x%llx, last ppn: 0x%x\n", freed_pages, (void *)vaddr, (uint64_t)size, ppn); +#endif } +/* Change page protections for addresses previously loaded by efiboot */ +kern_return_t +ml_static_protect(vm_offset_t vmaddr, vm_size_t size, vm_prot_t prot) +{ + boolean_t NX = !!!(prot & VM_PROT_EXECUTE), ro = !!!(prot & VM_PROT_WRITE); + + assert(prot & VM_PROT_READ); + + pmap_mark_range(kernel_pmap, vmaddr, size, NX, ro); + + return KERN_SUCCESS; +} /* virtual to physical on wired pages */ -vm_offset_t ml_vtophys( +vm_offset_t +ml_vtophys( vm_offset_t vaddr) { - return kvtophys(vaddr); + return (vm_offset_t)kvtophys(vaddr); } /* @@ -169,24 +283,30 @@ vm_offset_t ml_vtophys( * the duration of the copy process. */ -vm_size_t ml_nofault_copy( +vm_size_t +ml_nofault_copy( vm_offset_t virtsrc, vm_offset_t virtdst, vm_size_t size) { addr64_t cur_phys_dst, cur_phys_src; uint32_t count, nbytes = 0; while (size > 0) { - if (!(cur_phys_src = kvtophys(virtsrc))) + if (!(cur_phys_src = kvtophys(virtsrc))) { break; - if (!(cur_phys_dst = kvtophys(virtdst))) + } + if (!(cur_phys_dst = kvtophys(virtdst))) { break; - if (!pmap_valid_page(i386_btop(cur_phys_dst)) || !pmap_valid_page(i386_btop(cur_phys_src))) + } + if (!pmap_valid_page(i386_btop(cur_phys_dst)) || !pmap_valid_page(i386_btop(cur_phys_src))) { break; - count = PAGE_SIZE - (cur_phys_src & PAGE_MASK); - if (count > (PAGE_SIZE - (cur_phys_dst & PAGE_MASK))) - count = PAGE_SIZE - (cur_phys_dst & PAGE_MASK); - if (count > size) - count = size; + } + count = (uint32_t)(PAGE_SIZE - (cur_phys_src & PAGE_MASK)); + if (count > (PAGE_SIZE - (cur_phys_dst & PAGE_MASK))) { + count = (uint32_t)(PAGE_SIZE - (cur_phys_dst & PAGE_MASK)); + } + if (count > size) { + count = (uint32_t)size; + } bcopy_phys(cur_phys_src, cur_phys_dst, count); @@ -199,160 +319,205 @@ vm_size_t ml_nofault_copy( return nbytes; } +/* + * Routine: ml_validate_nofault + * Function: Validate that ths address range has a valid translations + * in the kernel pmap. If translations are present, they are + * assumed to be wired; i.e. no attempt is made to guarantee + * that the translation persist after the check. + * Returns: TRUE if the range is mapped and will not cause a fault, + * FALSE otherwise. + */ + +boolean_t +ml_validate_nofault( + vm_offset_t virtsrc, vm_size_t size) +{ + addr64_t cur_phys_src; + uint32_t count; + + while (size > 0) { + if (!(cur_phys_src = kvtophys(virtsrc))) { + return FALSE; + } + if (!pmap_valid_page(i386_btop(cur_phys_src))) { + return FALSE; + } + count = (uint32_t)(PAGE_SIZE - (cur_phys_src & PAGE_MASK)); + if (count > size) { + count = (uint32_t)size; + } + + virtsrc += count; + size -= count; + } + + return TRUE; +} + /* Interrupt handling */ /* Initialize Interrupts */ -void ml_init_interrupt(void) +void +ml_init_interrupt(void) { (void) ml_set_interrupts_enabled(TRUE); } + /* Get Interrupts Enabled */ -boolean_t ml_get_interrupts_enabled(void) +boolean_t +ml_get_interrupts_enabled(void) { - unsigned long flags; + unsigned long flags; - __asm__ volatile("pushf; popl %0" : "=r" (flags)); - return (flags & EFL_IF) != 0; + __asm__ volatile ("pushf; pop %0": "=r" (flags)); + return (flags & EFL_IF) != 0; } /* Set Interrupts Enabled */ -boolean_t ml_set_interrupts_enabled(boolean_t enable) +boolean_t +ml_set_interrupts_enabled(boolean_t enable) { - unsigned long flags; + unsigned long flags; + boolean_t istate; + + __asm__ volatile ("pushf; pop %0" : "=r" (flags)); - __asm__ volatile("pushf; popl %0" : "=r" (flags)); + assert(get_interrupt_level() ? (enable == FALSE) : TRUE); - if (enable) { - ast_t *myast; + istate = ((flags & EFL_IF) != 0); - myast = ast_pending(); + if (enable) { + __asm__ volatile ("sti;nop"); - if ( (get_preemption_level() == 0) && (*myast & AST_URGENT) ) { - __asm__ volatile("sti"); - __asm__ volatile ("int $0xff"); - } else { - __asm__ volatile ("sti"); + if ((get_preemption_level() == 0) && (*ast_pending() & AST_URGENT)) { + __asm__ volatile ("int %0" :: "N" (T_PREEMPT)); + } + } else { + if (istate) { + __asm__ volatile ("cli"); + } + } + + return istate; +} + +/* Early Set Interrupts Enabled */ +boolean_t +ml_early_set_interrupts_enabled(boolean_t enable) +{ + if (enable == TRUE) { + kprintf("Caller attempted to enable interrupts too early in " + "kernel startup. Halting.\n"); + hlt(); + /*NOTREACHED*/ } - } - else { - __asm__ volatile("cli"); - } - return (flags & EFL_IF) != 0; + /* On x86, do not allow interrupts to be enabled very early */ + return FALSE; } /* Check if running at interrupt context */ -boolean_t ml_at_interrupt_context(void) +boolean_t +ml_at_interrupt_context(void) { return get_interrupt_level() != 0; } +void +ml_get_power_state(boolean_t *icp, boolean_t *pidlep) +{ + *icp = (get_interrupt_level() != 0); + /* These will be technically inaccurate for interrupts that occur + * successively within a single "idle exit" event, but shouldn't + * matter statistically. + */ + *pidlep = (current_cpu_datap()->lcpu.package->num_idle == topoParms.nLThreadsPerPackage); +} + /* Generate a fake interrupt */ -void ml_cause_interrupt(void) +__dead2 +void +ml_cause_interrupt(void) { panic("ml_cause_interrupt not defined yet on Intel"); } -void ml_thread_policy( +/* + * TODO: transition users of this to kernel_thread_start_priority + * ml_thread_policy is an unsupported KPI + */ +void +ml_thread_policy( thread_t thread, -__unused unsigned policy_id, + __unused unsigned policy_id, unsigned policy_info) { if (policy_info & MACHINE_NETWORK_WORKLOOP) { - spl_t s = splsched(); + thread_precedence_policy_data_t info; + __assert_only kern_return_t kret; - thread_lock(thread); + info.importance = 1; - set_priority(thread, thread->priority + 1); - - thread_unlock(thread); - splx(s); + kret = thread_policy_set_internal(thread, THREAD_PRECEDENCE_POLICY, + (thread_policy_t)&info, + THREAD_PRECEDENCE_POLICY_COUNT); + assert(kret == KERN_SUCCESS); } } /* Initialize Interrupts */ -void ml_install_interrupt_handler( +void +ml_install_interrupt_handler( void *nub, int source, void *target, IOInterruptHandler handler, - void *refCon) + void *refCon) { boolean_t current_state; - current_state = ml_get_interrupts_enabled(); + current_state = ml_set_interrupts_enabled(FALSE); PE_install_interrupt_handler(nub, source, target, - (IOInterruptHandler) handler, refCon); + (IOInterruptHandler) handler, refCon); (void) ml_set_interrupts_enabled(current_state); - - initialize_screen(NULL, kPEAcquireScreen); } void -machine_idle(void) +machine_signal_idle( + processor_t processor) { - x86_core_t *my_core = x86_core(); - cpu_data_t *my_cpu = current_cpu_datap(); - int others_active; - - /* - * We halt this cpu thread - * unless kernel param idlehalt is false and no other thread - * in the same core is active - if so, don't halt so that this - * core doesn't go into a low-power mode. - * For 4/4, we set a null "active cr3" while idle. - */ - if (my_core == NULL || my_cpu == NULL) - goto out; - - others_active = !atomic_decl_and_test( - (long *) &my_core->active_lcpus, 1); - my_cpu->lcpu.idle = TRUE; - if (idlehalt || others_active) { - DBGLOG(cpu_handle, cpu_number(), MP_IDLE); - MARK_CPU_IDLE(cpu_number()); - machine_idle_cstate(FALSE); - MARK_CPU_ACTIVE(cpu_number()); - DBGLOG(cpu_handle, cpu_number(), MP_UNIDLE); - } - my_cpu->lcpu.idle = FALSE; - atomic_incl((long *) &my_core->active_lcpus, 1); - out: - __asm__ volatile("sti"); + cpu_interrupt(processor->cpu_id); } +__dead2 void -machine_signal_idle( - processor_t processor) +machine_signal_idle_deferred( + __unused processor_t processor) { - cpu_interrupt(PROCESSOR_DATA(processor, slot_num)); + panic("Unimplemented"); } -thread_t -machine_processor_shutdown( - thread_t thread, - void (*doshutdown)(processor_t), - processor_t processor) +__dead2 +void +machine_signal_idle_cancel( + __unused processor_t processor) { - vmx_suspend(); - fpu_save_context(thread); - return(Shutdown_context(thread, doshutdown, processor)); + panic("Unimplemented"); } -kern_return_t -ml_processor_register( - cpu_id_t cpu_id, - uint32_t lapic_id, - processor_t *processor_out, - ipi_handler_t *ipi_handler, - boolean_t boot_cpu) +static kern_return_t +register_cpu( + uint32_t lapic_id, + processor_t *processor_out, + boolean_t boot_cpu ) { - int target_cpu; - cpu_data_t *this_cpu_datap; + int target_cpu; + cpu_data_t *this_cpu_datap; this_cpu_datap = cpu_data_alloc(boot_cpu); if (this_cpu_datap == NULL) { @@ -360,155 +525,301 @@ ml_processor_register( } target_cpu = this_cpu_datap->cpu_number; assert((boot_cpu && (target_cpu == 0)) || - (!boot_cpu && (target_cpu != 0))); + (!boot_cpu && (target_cpu != 0))); lapic_cpu_map(lapic_id, target_cpu); - this_cpu_datap->cpu_id = cpu_id; + /* The cpu_id is not known at registration phase. Just do + * lapic_id for now + */ this_cpu_datap->cpu_phys_number = lapic_id; this_cpu_datap->cpu_console_buf = console_cpu_alloc(boot_cpu); - if (this_cpu_datap->cpu_console_buf == NULL) + if (this_cpu_datap->cpu_console_buf == NULL) { goto failed; + } - this_cpu_datap->cpu_chud = chudxnu_cpu_alloc(boot_cpu); - if (this_cpu_datap->cpu_chud == NULL) +#if KPC + if (kpc_register_cpu(this_cpu_datap) != TRUE) { goto failed; + } +#endif if (!boot_cpu) { - this_cpu_datap->lcpu.core = cpu_thread_alloc(this_cpu_datap->cpu_number); - if (this_cpu_datap->lcpu.core == NULL) - goto failed; - - pmCPUStateInit(); - - this_cpu_datap->cpu_pmap = pmap_cpu_alloc(boot_cpu); - if (this_cpu_datap->cpu_pmap == NULL) + cpu_thread_alloc(this_cpu_datap->cpu_number); + if (this_cpu_datap->lcpu.core == NULL) { goto failed; - - this_cpu_datap->cpu_processor = cpu_processor_alloc(boot_cpu); - if (this_cpu_datap->cpu_processor == NULL) - goto failed; - /* - * processor_init() deferred to topology start - * because "slot numbers" a.k.a. logical processor numbers - * are not yet finalized. - */ + } } + /* + * processor_init() deferred to topology start + * because "slot numbers" a.k.a. logical processor numbers + * are not yet finalized. + */ *processor_out = this_cpu_datap->cpu_processor; - *ipi_handler = NULL; - - if (target_cpu == machine_info.max_cpus - 1) { - /* - * All processors are now registered but not started (except - * for this "in-limbo" boot processor). We call to the machine - * topology code to finalize and activate the topology. - */ - cpu_topology_start(); - } return KERN_SUCCESS; failed: - cpu_processor_free(this_cpu_datap->cpu_processor); - pmap_cpu_free(this_cpu_datap->cpu_pmap); - chudxnu_cpu_free(this_cpu_datap->cpu_chud); console_cpu_free(this_cpu_datap->cpu_console_buf); +#if KPC + kpc_unregister_cpu(this_cpu_datap); +#endif /* KPC */ + return KERN_FAILURE; } + +kern_return_t +ml_processor_register( + cpu_id_t cpu_id, + uint32_t lapic_id, + processor_t *processor_out, + boolean_t boot_cpu, + boolean_t start ) +{ + static boolean_t done_topo_sort = FALSE; + static uint32_t num_registered = 0; + + /* Register all CPUs first, and track max */ + if (start == FALSE) { + num_registered++; + + DBG( "registering CPU lapic id %d\n", lapic_id ); + + return register_cpu( lapic_id, processor_out, boot_cpu ); + } + + /* Sort by topology before we start anything */ + if (!done_topo_sort) { + DBG( "about to start CPUs. %d registered\n", num_registered ); + + cpu_topology_sort( num_registered ); + done_topo_sort = TRUE; + } + + /* Assign the cpu ID */ + uint32_t cpunum = -1; + cpu_data_t *this_cpu_datap = NULL; + + /* find cpu num and pointer */ + cpunum = ml_get_cpuid( lapic_id ); + + if (cpunum == 0xFFFFFFFF) { /* never heard of it? */ + panic( "trying to start invalid/unregistered CPU %d\n", lapic_id ); + } + + this_cpu_datap = cpu_datap(cpunum); + + /* fix the CPU id */ + this_cpu_datap->cpu_id = cpu_id; + + /* allocate and initialize other per-cpu structures */ + if (!boot_cpu) { + mp_cpus_call_cpu_init(cpunum); + random_cpu_init(cpunum); + } + + /* output arg */ + *processor_out = this_cpu_datap->cpu_processor; + + /* OK, try and start this CPU */ + return cpu_topology_start_cpu( cpunum ); +} + + void ml_cpu_get_info(ml_cpu_info_t *cpu_infop) { - boolean_t os_supports_sse; + boolean_t os_supports_sse; i386_cpu_info_t *cpuid_infop; - if (cpu_infop == NULL) + if (cpu_infop == NULL) { return; - + } + /* * Are we supporting MMX/SSE/SSE2/SSE3? * As distinct from whether the cpu has these capabilities. */ - os_supports_sse = get_cr4() & CR4_XMM; - if ((cpuid_features() & CPUID_FEATURE_SSE4_2) && os_supports_sse) + os_supports_sse = !!(get_cr4() & CR4_OSXMM); + + if (ml_fpu_avx_enabled()) { + cpu_infop->vector_unit = 9; + } else if ((cpuid_features() & CPUID_FEATURE_SSE4_2) && os_supports_sse) { cpu_infop->vector_unit = 8; - else if ((cpuid_features() & CPUID_FEATURE_SSE4_1) && os_supports_sse) + } else if ((cpuid_features() & CPUID_FEATURE_SSE4_1) && os_supports_sse) { cpu_infop->vector_unit = 7; - else if ((cpuid_features() & CPUID_FEATURE_SSSE3) && os_supports_sse) + } else if ((cpuid_features() & CPUID_FEATURE_SSSE3) && os_supports_sse) { cpu_infop->vector_unit = 6; - else if ((cpuid_features() & CPUID_FEATURE_SSE3) && os_supports_sse) + } else if ((cpuid_features() & CPUID_FEATURE_SSE3) && os_supports_sse) { cpu_infop->vector_unit = 5; - else if ((cpuid_features() & CPUID_FEATURE_SSE2) && os_supports_sse) + } else if ((cpuid_features() & CPUID_FEATURE_SSE2) && os_supports_sse) { cpu_infop->vector_unit = 4; - else if ((cpuid_features() & CPUID_FEATURE_SSE) && os_supports_sse) + } else if ((cpuid_features() & CPUID_FEATURE_SSE) && os_supports_sse) { cpu_infop->vector_unit = 3; - else if (cpuid_features() & CPUID_FEATURE_MMX) + } else if (cpuid_features() & CPUID_FEATURE_MMX) { cpu_infop->vector_unit = 2; - else + } else { cpu_infop->vector_unit = 0; + } cpuid_infop = cpuid_info(); - cpu_infop->cache_line_size = cpuid_infop->cache_linesize; + cpu_infop->cache_line_size = cpuid_infop->cache_linesize; cpu_infop->l1_icache_size = cpuid_infop->cache_size[L1I]; cpu_infop->l1_dcache_size = cpuid_infop->cache_size[L1D]; - - if (cpuid_infop->cache_size[L2U] > 0) { - cpu_infop->l2_settings = 1; - cpu_infop->l2_cache_size = cpuid_infop->cache_size[L2U]; - } else { - cpu_infop->l2_settings = 0; - cpu_infop->l2_cache_size = 0xFFFFFFFF; - } - if (cpuid_infop->cache_size[L3U] > 0) { - cpu_infop->l3_settings = 1; - cpu_infop->l3_cache_size = cpuid_infop->cache_size[L3U]; - } else { - cpu_infop->l3_settings = 0; - cpu_infop->l3_cache_size = 0xFFFFFFFF; - } + if (cpuid_infop->cache_size[L2U] > 0) { + cpu_infop->l2_settings = 1; + cpu_infop->l2_cache_size = cpuid_infop->cache_size[L2U]; + } else { + cpu_infop->l2_settings = 0; + cpu_infop->l2_cache_size = 0xFFFFFFFF; + } + + if (cpuid_infop->cache_size[L3U] > 0) { + cpu_infop->l3_settings = 1; + cpu_infop->l3_cache_size = cpuid_infop->cache_size[L3U]; + } else { + cpu_infop->l3_settings = 0; + cpu_infop->l3_cache_size = 0xFFFFFFFF; + } } -void -ml_init_max_cpus(unsigned long max_cpus) +int +ml_early_cpu_max_number(void) { - boolean_t current_state; + int n = max_ncpus; + + assert(startup_phase >= STARTUP_SUB_TUNABLES); + if (max_cpus_from_firmware) { + n = MIN(n, max_cpus_from_firmware); + } + return n - 1; +} - current_state = ml_set_interrupts_enabled(FALSE); - if (max_cpus_initialized != MAX_CPUS_SET) { - if (max_cpus > 0 && max_cpus <= MAX_CPUS) { +void +ml_set_max_cpus(unsigned int max_cpus) +{ + lck_mtx_lock(&max_cpus_lock); + if (max_cpus_initialized != MAX_CPUS_SET) { + if (max_cpus > 0 && max_cpus <= MAX_CPUS) { /* * Note: max_cpus is the number of enabled processors * that ACPI found; max_ncpus is the maximum number * that the kernel supports or that the "cpus=" * boot-arg has set. Here we take int minimum. */ - machine_info.max_cpus = MIN(max_cpus, max_ncpus); + machine_info.max_cpus = (integer_t)MIN(max_cpus, max_ncpus); } - if (max_cpus_initialized == MAX_CPUS_WAIT) - wakeup((event_t)&max_cpus_initialized); - max_cpus_initialized = MAX_CPUS_SET; - } - (void) ml_set_interrupts_enabled(current_state); + if (max_cpus_initialized == MAX_CPUS_WAIT) { + thread_wakeup((event_t) &max_cpus_initialized); + } + max_cpus_initialized = MAX_CPUS_SET; + } + lck_mtx_unlock(&max_cpus_lock); } -int -ml_get_max_cpus(void) +unsigned int +ml_wait_max_cpus(void) { - boolean_t current_state; + lck_mtx_lock(&max_cpus_lock); + while (max_cpus_initialized != MAX_CPUS_SET) { + max_cpus_initialized = MAX_CPUS_WAIT; + lck_mtx_sleep(&max_cpus_lock, LCK_SLEEP_DEFAULT, &max_cpus_initialized, THREAD_UNINT); + } + lck_mtx_unlock(&max_cpus_lock); + return machine_info.max_cpus; +} - current_state = ml_set_interrupts_enabled(FALSE); - if (max_cpus_initialized != MAX_CPUS_SET) { - max_cpus_initialized = MAX_CPUS_WAIT; - assert_wait((event_t)&max_cpus_initialized, THREAD_UNINT); - (void)thread_block(THREAD_CONTINUE_NULL); - } - (void) ml_set_interrupts_enabled(current_state); - return(machine_info.max_cpus); +void +ml_panic_trap_to_debugger(__unused const char *panic_format_str, + __unused va_list *panic_args, + __unused unsigned int reason, + __unused void *ctx, + __unused uint64_t panic_options_mask, + __unused unsigned long panic_caller) +{ + return; +} + +static uint64_t +virtual_timeout_inflate64(unsigned int vti, uint64_t timeout, uint64_t max_timeout) +{ + if (vti >= 64) { + return max_timeout; + } + + if ((timeout << vti) >> vti != timeout) { + return max_timeout; + } + + if ((timeout << vti) > max_timeout) { + return max_timeout; + } + + return timeout << vti; +} + +static uint32_t +virtual_timeout_inflate32(unsigned int vti, uint32_t timeout, uint32_t max_timeout) +{ + if (vti >= 32) { + return max_timeout; + } + + if ((timeout << vti) >> vti != timeout) { + return max_timeout; + } + + return timeout << vti; +} + +/* + * Some timeouts are later adjusted or used in calculations setting + * other values. In order to avoid overflow, cap the max timeout as + * 2^47ns (~39 hours). + */ +static const uint64_t max_timeout_ns = 1ULL << 47; + +/* + * Inflate a timeout in absolutetime. + */ +static uint64_t +virtual_timeout_inflate_abs(unsigned int vti, uint64_t timeout) +{ + uint64_t max_timeout; + nanoseconds_to_absolutetime(max_timeout_ns, &max_timeout); + return virtual_timeout_inflate64(vti, timeout, max_timeout); +} + +/* + * Inflate a value in TSC ticks. + */ +static uint64_t +virtual_timeout_inflate_tsc(unsigned int vti, uint64_t timeout) +{ + const uint64_t max_timeout = tmrCvt(max_timeout_ns, tscFCvtn2t); + return virtual_timeout_inflate64(vti, timeout, max_timeout); +} + +/* + * Inflate a timeout in microseconds. + */ +static uint32_t +virtual_timeout_inflate_us(unsigned int vti, uint64_t timeout) +{ + const uint32_t max_timeout = ~0; + return virtual_timeout_inflate32(vti, timeout, max_timeout); +} + +uint64_t +ml_get_timebase_entropy(void) +{ + return __builtin_ia32_rdtsc(); } /* @@ -518,26 +829,160 @@ ml_get_max_cpus(void) void ml_init_lock_timeout(void) { - uint64_t abstime; - uint32_t mtxspin; + uint64_t abstime; + uint32_t mtxspin; +#if DEVELOPMENT || DEBUG + uint64_t default_timeout_ns = NSEC_PER_SEC >> 2; +#else + uint64_t default_timeout_ns = NSEC_PER_SEC >> 1; +#endif + uint32_t slto; + uint32_t prt; - /* LockTimeOut is absolutetime, LockTimeOutTSC is in TSC ticks */ - nanoseconds_to_absolutetime(NSEC_PER_SEC>>2, &abstime); - LockTimeOut = (uint32_t) abstime; - LockTimeOutTSC = (uint32_t) tmrCvt(abstime, tscFCvtn2t); + if (PE_parse_boot_argn("slto_us", &slto, sizeof(slto))) { + default_timeout_ns = slto * NSEC_PER_USEC; + } + + /* + * LockTimeOut is absolutetime, LockTimeOutTSC is in TSC ticks, + * and LockTimeOutUsec is in microseconds and it's 32-bits. + */ + LockTimeOutUsec = (uint32_t) (default_timeout_ns / NSEC_PER_USEC); + nanoseconds_to_absolutetime(default_timeout_ns, &abstime); + LockTimeOut = abstime; + LockTimeOutTSC = tmrCvt(abstime, tscFCvtn2t); - if (PE_parse_boot_arg("mtxspin", &mtxspin)) { - if (mtxspin > USEC_PER_SEC>>4) - mtxspin = USEC_PER_SEC>>4; - nanoseconds_to_absolutetime(mtxspin*NSEC_PER_USEC, &abstime); + /* + * TLBTimeOut dictates the TLB flush timeout period. It defaults to + * LockTimeOut but can be overriden separately. In particular, a + * zero value inhibits the timeout-panic and cuts a trace evnt instead + * - see pmap_flush_tlbs(). + */ + if (PE_parse_boot_argn("tlbto_us", &slto, sizeof(slto))) { + default_timeout_ns = slto * NSEC_PER_USEC; + nanoseconds_to_absolutetime(default_timeout_ns, &abstime); + TLBTimeOut = (uint32_t) abstime; } else { - nanoseconds_to_absolutetime(10*NSEC_PER_USEC, &abstime); + TLBTimeOut = LockTimeOut; + } + +#if DEVELOPMENT || DEBUG + reportphyreaddelayabs = LockTimeOut >> 1; +#endif + if (PE_parse_boot_argn("phyreadmaxus", &slto, sizeof(slto))) { + default_timeout_ns = slto * NSEC_PER_USEC; + nanoseconds_to_absolutetime(default_timeout_ns, &abstime); + reportphyreaddelayabs = abstime; + } + + if (PE_parse_boot_argn("phywritemaxus", &slto, sizeof(slto))) { + nanoseconds_to_absolutetime((uint64_t)slto * NSEC_PER_USEC, &abstime); + reportphywritedelayabs = abstime; + } + + if (PE_parse_boot_argn("tracephyreadus", &slto, sizeof(slto))) { + nanoseconds_to_absolutetime((uint64_t)slto * NSEC_PER_USEC, &abstime); + tracephyreaddelayabs = abstime; + } + + if (PE_parse_boot_argn("tracephywriteus", &slto, sizeof(slto))) { + nanoseconds_to_absolutetime((uint64_t)slto * NSEC_PER_USEC, &abstime); + tracephywritedelayabs = abstime; + } + + if (PE_parse_boot_argn("mtxspin", &mtxspin, sizeof(mtxspin))) { + if (mtxspin > USEC_PER_SEC >> 4) { + mtxspin = USEC_PER_SEC >> 4; + } + nanoseconds_to_absolutetime(mtxspin * NSEC_PER_USEC, &abstime); + } else { + nanoseconds_to_absolutetime(10 * NSEC_PER_USEC, &abstime); } MutexSpin = (unsigned int)abstime; + low_MutexSpin = MutexSpin; + /* + * high_MutexSpin should be initialized as low_MutexSpin * real_ncpus, but + * real_ncpus is not set at this time + */ + high_MutexSpin = -1; + + nanoseconds_to_absolutetime(4ULL * NSEC_PER_SEC, &LastDebuggerEntryAllowance); + if (PE_parse_boot_argn("panic_restart_timeout", &prt, sizeof(prt))) { + nanoseconds_to_absolutetime(prt * NSEC_PER_SEC, &panic_restart_timeout); + } + + virtualized = ((cpuid_features() & CPUID_FEATURE_VMM) != 0); + if (virtualized) { + unsigned int vti; + + if (!PE_parse_boot_argn("vti", &vti, sizeof(vti))) { + vti = 6; + } + printf("Timeouts adjusted for virtualization (<<%d)\n", vti); + kprintf("Timeouts adjusted for virtualization (<<%d):\n", vti); +#define VIRTUAL_TIMEOUT_INFLATE_ABS(_timeout) \ +MACRO_BEGIN \ + kprintf("%24s: 0x%016llx ", #_timeout, _timeout); \ + _timeout = virtual_timeout_inflate_abs(vti, _timeout); \ + kprintf("-> 0x%016llx\n", _timeout); \ +MACRO_END + +#define VIRTUAL_TIMEOUT_INFLATE_TSC(_timeout) \ +MACRO_BEGIN \ + kprintf("%24s: 0x%016llx ", #_timeout, _timeout); \ + _timeout = virtual_timeout_inflate_tsc(vti, _timeout); \ + kprintf("-> 0x%016llx\n", _timeout); \ +MACRO_END +#define VIRTUAL_TIMEOUT_INFLATE_US(_timeout) \ +MACRO_BEGIN \ + kprintf("%24s: 0x%08x ", #_timeout, _timeout); \ + _timeout = virtual_timeout_inflate_us(vti, _timeout); \ + kprintf("-> 0x%08x\n", _timeout); \ +MACRO_END + VIRTUAL_TIMEOUT_INFLATE_US(LockTimeOutUsec); + VIRTUAL_TIMEOUT_INFLATE_ABS(LockTimeOut); + VIRTUAL_TIMEOUT_INFLATE_TSC(LockTimeOutTSC); + VIRTUAL_TIMEOUT_INFLATE_ABS(TLBTimeOut); + VIRTUAL_TIMEOUT_INFLATE_ABS(MutexSpin); + VIRTUAL_TIMEOUT_INFLATE_ABS(low_MutexSpin); + VIRTUAL_TIMEOUT_INFLATE_ABS(reportphyreaddelayabs); + } + + interrupt_latency_tracker_setup(); + simple_lock_init(&ml_timer_evaluation_slock, 0); } /* - * This is called from the machine-independent routine cpu_up() + * Threshold above which we should attempt to block + * instead of spinning for clock_delay_until(). + */ + +void +ml_init_delay_spin_threshold(int threshold_us) +{ + nanoseconds_to_absolutetime(threshold_us * NSEC_PER_USEC, &delay_spin_threshold); +} + +boolean_t +ml_delay_should_spin(uint64_t interval) +{ + return (interval < delay_spin_threshold) ? TRUE : FALSE; +} + +TUNABLE(uint32_t, yield_delay_us, "yield_delay_us", 0); + +void +ml_delay_on_yield(void) +{ +#if DEVELOPMENT || DEBUG + if (yield_delay_us) { + delay(yield_delay_us); + } +#endif +} + +/* + * This is called from the machine-independent layer * to perform machine-dependent info updates. Defer to cpu_thread_init(). */ void @@ -547,12 +992,14 @@ ml_cpu_up(void) } /* - * This is called from the machine-independent routine cpu_down() + * This is called from the machine-independent layer * to perform machine-dependent info updates. */ void ml_cpu_down(void) { + i386_deactivate_cpu(); + return; } @@ -560,108 +1007,262 @@ ml_cpu_down(void) * The following are required for parts of the kernel * that cannot resolve these functions as inlines: */ -extern thread_t current_act(void); +extern thread_t current_act(void) __attribute__((const)); thread_t current_act(void) { - return(current_thread_fast()); + return current_thread_fast(); } #undef current_thread -extern thread_t current_thread(void); +extern thread_t current_thread(void) __attribute__((const)); thread_t current_thread(void) { - return(current_thread_fast()); + return current_thread_fast(); } -boolean_t ml_is64bit(void) { - - return (cpu_mode_is64bit()); +boolean_t +ml_is64bit(void) +{ + return cpu_mode_is64bit(); } -boolean_t ml_thread_is64bit(thread_t thread) { - - return (thread_is_64bit(thread)); +boolean_t +ml_thread_is64bit(thread_t thread) +{ + return thread_is_64bit_addr(thread); } -boolean_t ml_state_is64bit(void *saved_state) { - +boolean_t +ml_state_is64bit(void *saved_state) +{ return is_saved_state64(saved_state); } -void ml_cpu_set_ldt(int selector) +void +ml_cpu_set_ldt(int selector) { /* * Avoid loading the LDT * if we're setting the KERNEL LDT and it's already set. */ if (selector == KERNEL_LDT && - current_cpu_datap()->cpu_ldt == KERNEL_LDT) + current_cpu_datap()->cpu_ldt == KERNEL_LDT) { return; + } - /* - * If 64bit this requires a mode switch (and back). - */ - if (cpu_mode_is64bit()) - ml_64bit_lldt(selector); - else - lldt(selector); - current_cpu_datap()->cpu_ldt = selector; + lldt(selector); + current_cpu_datap()->cpu_ldt = selector; } -void ml_fp_setvalid(boolean_t value) +void +ml_fp_setvalid(boolean_t value) { - fp_setvalid(value); + fp_setvalid(value); } -uint64_t ml_cpu_int_event_time(void) +uint64_t +ml_cpu_int_event_time(void) { return current_cpu_datap()->cpu_int_event_time; } - -#if MACH_KDB - -/* - * Display the global msrs - * * - * ms - */ -void -db_msr(__unused db_expr_t addr, - __unused int have_addr, - __unused db_expr_t count, - __unused char *modif) +vm_offset_t +ml_stack_remaining(void) { + uintptr_t local = (uintptr_t) &local; - uint32_t i, msrlow, msrhigh; + if (ml_at_interrupt_context() != 0) { + return local - (current_cpu_datap()->cpu_int_stack_top - INTSTACK_SIZE); + } else { + return local - current_thread()->kernel_stack; + } +} - /* Try all of the first 4096 msrs */ - for (i = 0; i < 4096; i++) { - if (!rdmsr_carefully(i, &msrlow, &msrhigh)) { - db_printf("%08X - %08X.%08X\n", i, msrhigh, msrlow); - } +#if KASAN +vm_offset_t ml_stack_base(void); +vm_size_t ml_stack_size(void); + +vm_offset_t +ml_stack_base(void) +{ + if (ml_at_interrupt_context()) { + return current_cpu_datap()->cpu_int_stack_top - INTSTACK_SIZE; + } else { + return current_thread()->kernel_stack; } +} - /* Try all of the 4096 msrs at 0x0C000000 */ - for (i = 0; i < 4096; i++) { - if (!rdmsr_carefully(0x0C000000 | i, &msrlow, &msrhigh)) { - db_printf("%08X - %08X.%08X\n", - 0x0C000000 | i, msrhigh, msrlow); - } +vm_size_t +ml_stack_size(void) +{ + if (ml_at_interrupt_context()) { + return INTSTACK_SIZE; + } else { + return kernel_stack_size; } +} +#endif + +void +kernel_preempt_check(void) +{ + boolean_t intr; + unsigned long flags; + + assert(get_preemption_level() == 0); + + if (__improbable(*ast_pending() & AST_URGENT)) { + /* + * can handle interrupts and preemptions + * at this point + */ + __asm__ volatile ("pushf; pop %0" : "=r" (flags)); + + intr = ((flags & EFL_IF) != 0); - /* Try all of the 4096 msrs at 0xC0000000 */ - for (i = 0; i < 4096; i++) { - if (!rdmsr_carefully(0xC0000000 | i, &msrlow, &msrhigh)) { - db_printf("%08X - %08X.%08X\n", - 0xC0000000 | i, msrhigh, msrlow); + /* + * now cause the PRE-EMPTION trap + */ + if (intr == TRUE) { + __asm__ volatile ("int %0" :: "N" (T_PREEMPT)); } } } -#endif +boolean_t +machine_timeout_suspended(void) +{ + return pmap_tlb_flush_timeout || spinlock_timed_out || panic_active() || mp_recent_debugger_activity() || ml_recent_wake(); +} + +/* Eagerly evaluate all pending timer and thread callouts + */ +void +ml_timer_evaluate(void) +{ + KERNEL_DEBUG_CONSTANT(DECR_TIMER_RESCAN | DBG_FUNC_START, 0, 0, 0, 0, 0); + + uint64_t te_end, te_start = mach_absolute_time(); + simple_lock(&ml_timer_evaluation_slock, LCK_GRP_NULL); + ml_timer_evaluation_in_progress = TRUE; + thread_call_delayed_timer_rescan_all(); + mp_cpus_call(CPUMASK_ALL, ASYNC, timer_queue_expire_rescan, NULL); + ml_timer_evaluation_in_progress = FALSE; + ml_timer_eager_evaluations++; + te_end = mach_absolute_time(); + ml_timer_eager_evaluation_max = MAX(ml_timer_eager_evaluation_max, (te_end - te_start)); + simple_unlock(&ml_timer_evaluation_slock); + + KERNEL_DEBUG_CONSTANT(DECR_TIMER_RESCAN | DBG_FUNC_END, 0, 0, 0, 0, 0); +} + +boolean_t +ml_timer_forced_evaluation(void) +{ + return ml_timer_evaluation_in_progress; +} + +uint64_t +ml_energy_stat(__unused thread_t t) +{ + return 0; +} + +void +ml_gpu_stat_update(uint64_t gpu_ns_delta) +{ + current_thread()->machine.thread_gpu_ns += gpu_ns_delta; +} + +uint64_t +ml_gpu_stat(thread_t t) +{ + return t->machine.thread_gpu_ns; +} + +int plctrace_enabled = 0; + +void +_disable_preemption(void) +{ + disable_preemption_internal(); +} + +void +_enable_preemption(void) +{ + enable_preemption_internal(); +} + +void +plctrace_disable(void) +{ + plctrace_enabled = 0; +} + +static boolean_t ml_quiescing; + +void +ml_set_is_quiescing(boolean_t quiescing) +{ + ml_quiescing = quiescing; +} + +boolean_t +ml_is_quiescing(void) +{ + return ml_quiescing; +} + +uint64_t +ml_get_booter_memory_size(void) +{ + return 0; +} + +void +machine_lockdown(void) +{ + x86_64_protect_data_const(); +} + +bool +ml_cpu_can_exit(__unused int cpu_id) +{ + return true; +} + +void +ml_cpu_begin_state_transition(__unused int cpu_id) +{ +} + +void +ml_cpu_end_state_transition(__unused int cpu_id) +{ +} + +void +ml_cpu_begin_loop(void) +{ +} + +void +ml_cpu_end_loop(void) +{ +} + +size_t +ml_get_vm_reserved_regions(bool vm_is64bit, struct vm_reserved_region **regions) +{ +#pragma unused(vm_is64bit) + assert(regions != NULL); + + *regions = NULL; + return 0; +}