X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/bd504ef0e0b883cdd7917b73b3574eb9ce669905..490019cf9519204c5fb36b2fba54ceb983bb6b72:/bsd/kern/kdebug.c diff --git a/bsd/kern/kdebug.c b/bsd/kern/kdebug.c index 79896dbbe..556e0d0bb 100644 --- a/bsd/kern/kdebug.c +++ b/bsd/kern/kdebug.c @@ -1,5 +1,5 @@ /* - * Copyright (c) 2000-2006 Apple Computer, Inc. All rights reserved. + * Copyright (c) 2000-2013 Apple Inc. All rights reserved. * * @Apple_LICENSE_HEADER_START@ * @@ -32,6 +32,8 @@ #include #include #include +#include +#include #define HZ 100 #include @@ -53,6 +55,8 @@ #include #include #include +#include +#include #include #include @@ -72,27 +76,96 @@ #include +extern boolean_t kdebug_serial; +#if KDEBUG_MOJO_TRACE +#include +static void kdebug_serial_print( /* forward */ + uint32_t, uint32_t, uint64_t, + uintptr_t, uintptr_t, uintptr_t, uintptr_t, uintptr_t); +#endif + +/* + * IOP(s) + * + * https://coreoswiki.apple.com/wiki/pages/U6z3i0q9/Consistent_Logging_Implementers_Guide.html + * + * IOP(s) are auxiliary cores that want to participate in kdebug event logging. + * They are registered dynamically. Each is assigned a cpu_id at registration. + * + * NOTE: IOP trace events may not use the same clock hardware as "normal" + * cpus. There is an effort made to synchronize the IOP timebase with the + * AP, but it should be understood that there may be discrepancies. + * + * Once registered, an IOP is permanent, it cannot be unloaded/unregistered. + * The current implementation depends on this for thread safety. + * + * New registrations occur by allocating an kd_iop struct and assigning + * a provisional cpu_id of list_head->cpu_id + 1. Then a CAS to claim the + * list_head pointer resolves any races. + * + * You may safely walk the kd_iops list at any time, without holding locks. + * + * When allocating buffers, the current kd_iops head is captured. Any operations + * that depend on the buffer state (such as flushing IOP traces on reads, + * etc.) should use the captured list head. This will allow registrations to + * take place while trace is in use. + */ + +typedef struct kd_iop { + kd_callback_t callback; + uint32_t cpu_id; + uint64_t last_timestamp; /* Prevent timer rollback */ + struct kd_iop* next; +} kd_iop_t; + +static kd_iop_t* kd_iops = NULL; + /* XXX should have prototypes, but Mach does not provide one */ void task_act_iterate_wth_args(task_t, void(*)(thread_t, void *), void *); int cpu_number(void); /* XXX include path broken */ +void commpage_update_kdebug_enable(void); /* XXX sign */ /* XXX should probably be static, but it's debugging code... */ -int kdbg_read(user_addr_t, size_t *, vnode_t, vfs_context_t); +int kdbg_read(user_addr_t, size_t *, vnode_t, vfs_context_t, uint32_t); void kdbg_control_chud(int, void *); int kdbg_control(int *, u_int, user_addr_t, size_t *); -int kdbg_getentropy (user_addr_t, size_t *, int); -int kdbg_readmap(user_addr_t, size_t *, vnode_t, vfs_context_t); -int kdbg_getreg(kd_regtype *); +int kdbg_readcpumap(user_addr_t, size_t *); +int kdbg_readcurcpumap(user_addr_t, size_t *); +int kdbg_readthrmap(user_addr_t, size_t *, vnode_t, vfs_context_t); +int kdbg_readthrmap_v3(user_addr_t, size_t *, int); +int kdbg_readcurthrmap(user_addr_t, size_t *); int kdbg_setreg(kd_regtype *); int kdbg_setrtcdec(kd_regtype *); int kdbg_setpidex(kd_regtype *); int kdbg_setpid(kd_regtype *); -void kdbg_mapinit(void); +void kdbg_thrmap_init(void); int kdbg_reinit(boolean_t); int kdbg_bootstrap(boolean_t); +int kdbg_cpumap_init_internal(kd_iop_t* iops, uint32_t cpu_count, + uint8_t** cpumap, uint32_t* cpumap_size); + +kd_threadmap* kdbg_thrmap_init_internal(unsigned int count, + unsigned int *mapsize, + unsigned int *mapcount); + +static boolean_t kdebug_current_proc_enabled(uint32_t debugid); +static boolean_t kdebug_debugid_enabled(uint32_t debugid); +static errno_t kdebug_check_trace_string(uint32_t debugid, uint64_t str_id); + +int kdbg_write_v3_header(user_addr_t, size_t *, int); +int kdbg_write_v3_chunk_header(user_addr_t buffer, uint32_t tag, + uint32_t sub_tag, uint64_t length, + vnode_t vp, vfs_context_t ctx); + +user_addr_t kdbg_write_v3_event_chunk_header(user_addr_t buffer, uint32_t tag, + uint64_t length, vnode_t vp, + vfs_context_t ctx); + static int kdbg_enable_typefilter(void); static int kdbg_disable_typefilter(void); +static int kdbg_allocate_typefilter(void); +static int kdbg_deallocate_typefilter(void); static int create_buffers(boolean_t); static void delete_buffers(void); @@ -102,30 +175,21 @@ extern void IOSleep(int); /* trace enable status */ unsigned int kdebug_enable = 0; -/* track timestamps for security server's entropy needs */ -uint64_t * kd_entropy_buffer = 0; -unsigned int kd_entropy_bufsize = 0; -unsigned int kd_entropy_count = 0; -unsigned int kd_entropy_indx = 0; -vm_offset_t kd_entropy_buftomem = 0; - -#define MAX_ENTROPY_COUNT (128 * 1024) - +/* A static buffer to record events prior to the start of regular logging */ +#define KD_EARLY_BUFFER_MAX 64 +static kd_buf kd_early_buffer[KD_EARLY_BUFFER_MAX]; +static int kd_early_index = 0; +static boolean_t kd_early_overflow = FALSE; #define SLOW_NOLOG 0x01 #define SLOW_CHECKS 0x02 -#define SLOW_ENTROPY 0x04 #define SLOW_CHUD 0x08 -unsigned int kd_cpus; - #define EVENTS_PER_STORAGE_UNIT 2048 #define MIN_STORAGE_UNITS_PER_CPU 4 #define POINTER_FROM_KDS_PTR(x) (&kd_bufs[x.buffer_index].kdsb_addr[x.offset]) -#define NATIVE_TRACE_FACILITY - union kds_ptr { struct { uint32_t buffer_index:21; @@ -159,7 +223,6 @@ int n_storage_units = 0; int n_storage_buffers = 0; int n_storage_threshold = 0; int kds_waiter = 0; -int kde_waiter = 0; #pragma pack(0) struct kd_bufinfo { @@ -169,8 +232,13 @@ struct kd_bufinfo { uint32_t _pad; uint64_t kd_prev_timebase; uint32_t num_bufs; -} __attribute__(( aligned(CPU_CACHE_SIZE) )); +} __attribute__(( aligned(MAX_CPU_CACHE_LINE_SIZE) )); + +/* + * In principle, this control block can be shared in DRAM with other + * coprocessors and runtimes, for configuring what tracing is enabled. + */ struct kd_ctrl_page_t { union kds_ptr kds_free_list; uint32_t enabled :1; @@ -178,22 +246,28 @@ struct kd_ctrl_page_t { int kds_inuse_count; uint32_t kdebug_flags; uint32_t kdebug_slowcheck; - uint32_t _pad1; - struct { - uint64_t tsc_base; - uint64_t ns_base; - } cpu_timebase[32]; // should be max number of actual logical cpus -} kd_ctrl_page = {.kds_free_list = {.raw = KDS_PTR_NULL}, .enabled = 0, .kds_inuse_count = 0, .kdebug_flags = 0, .kdebug_slowcheck = SLOW_NOLOG}; + /* + * The number of kd_bufinfo structs allocated may not match the current + * number of active cpus. We capture the iops list head at initialization + * which we could use to calculate the number of cpus we allocated data for, + * unless it happens to be null. To avoid that case, we explicitly also + * capture a cpu count. + */ + kd_iop_t* kdebug_iops; + uint32_t kdebug_cpus; +} kd_ctrl_page = { .kds_free_list = {.raw = KDS_PTR_NULL}, .kdebug_slowcheck = SLOW_NOLOG }; + #pragma pack() struct kd_bufinfo *kdbip = NULL; #define KDCOPYBUF_COUNT 8192 #define KDCOPYBUF_SIZE (KDCOPYBUF_COUNT * sizeof(kd_buf)) -kd_buf *kdcopybuf = NULL; +#define PAGE_4KB 4096 +#define PAGE_16KB 16384 -int kdlog_sched_events = 0; +kd_buf *kdcopybuf = NULL; boolean_t kdlog_bg_trace = FALSE; boolean_t kdlog_bg_trace_running = FALSE; @@ -214,57 +288,52 @@ static lck_grp_t * kd_trace_mtx_sysctl_grp; static lck_attr_t * kd_trace_mtx_sysctl_attr; static lck_grp_attr_t *kd_trace_mtx_sysctl_grp_attr; -static lck_grp_t *stackshot_subsys_lck_grp; -static lck_grp_attr_t *stackshot_subsys_lck_grp_attr; -static lck_attr_t *stackshot_subsys_lck_attr; -static lck_mtx_t stackshot_subsys_mutex; +extern kern_return_t stack_snapshot2(int pid, user_addr_t tracebuf, uint32_t tracebuf_size, uint32_t flags, int32_t *retval); -void *stackshot_snapbuf = NULL; +#if CONFIG_TELEMETRY +extern kern_return_t stack_microstackshot(user_addr_t tracebuf, uint32_t tracebuf_size, uint32_t flags, int32_t *retval); +#endif /* CONFIG_TELEMETRY */ -int -stack_snapshot2(pid_t pid, user_addr_t tracebuf, uint32_t tracebuf_size, uint32_t flags, uint32_t dispatch_offset, int32_t *retval); +extern kern_return_t kern_stack_snapshot_with_reason(char* reason); + +extern kern_return_t kern_stack_snapshot_internal(int stackshot_config_version, void *stackshot_config, size_t stackshot_config_size, boolean_t stackshot_from_user); -extern void -kdp_snapshot_preflight(int pid, void *tracebuf, uint32_t tracebuf_size, uint32_t flags, uint32_t dispatch_offset); +extern kern_return_t stack_snapshot_from_kernel_internal(int pid, void *buf, uint32_t size, uint32_t flags, unsigned *bytes_traced); -extern int -kdp_stack_snapshot_geterror(void); -extern unsigned int -kdp_stack_snapshot_bytes_traced(void); +int stack_snapshot_from_kernel(pid_t pid, void *buf, uint32_t size, uint32_t flags, unsigned *bytes_traced); kd_threadmap *kd_mapptr = 0; unsigned int kd_mapsize = 0; unsigned int kd_mapcount = 0; -vm_offset_t kd_maptomem = 0; off_t RAW_file_offset = 0; int RAW_file_written = 0; #define RAW_FLUSH_SIZE (2 * 1024 * 1024) - pid_t global_state_pid = -1; /* Used to control exclusive use of kd_buffer */ -#define DBG_FUNC_MASK 0xfffffffc +/* + * A globally increasing counter for identifying strings in trace. Starts at + * 1 because 0 is a reserved return value. + */ +__attribute__((aligned(MAX_CPU_CACHE_LINE_SIZE))) +static uint64_t g_curr_str_id = 1; -/* TODO: move to kdebug.h */ -#define CLASS_MASK 0xff000000 -#define CLASS_OFFSET 24 -#define SUBCLASS_MASK 0x00ff0000 -#define SUBCLASS_OFFSET 16 -#define CSC_MASK 0xffff0000 /* class and subclass mask */ -#define CSC_OFFSET SUBCLASS_OFFSET +#define STR_ID_SIG_OFFSET (48) +#define STR_ID_MASK ((1ULL << STR_ID_SIG_OFFSET) - 1) +#define STR_ID_SIG_MASK (~STR_ID_MASK) -#define EXTRACT_CLASS(debugid) ( (uint8_t) ( ((debugid) & CLASS_MASK ) >> CLASS_OFFSET ) ) -#define EXTRACT_SUBCLASS(debugid) ( (uint8_t) ( ((debugid) & SUBCLASS_MASK) >> SUBCLASS_OFFSET ) ) -#define EXTRACT_CSC(debugid) ( (uint16_t)( ((debugid) & CSC_MASK ) >> CSC_OFFSET ) ) +/* + * A bit pattern for identifying string IDs generated by + * kdebug_trace_string(2). + */ +static uint64_t g_str_id_signature = (0x70acULL << STR_ID_SIG_OFFSET); #define INTERRUPT 0x01050000 #define MACH_vmfault 0x01300008 #define BSC_SysCall 0x040c0000 #define MACH_SysCall 0x010c0000 -#define DBG_SCALL_MASK 0xffff0000 - /* task to string structure */ struct tts @@ -293,27 +362,91 @@ typedef void (*kd_chudhook_fn) (uint32_t debugid, uintptr_t arg1, volatile kd_chudhook_fn kdebug_chudhook = 0; /* pointer to CHUD toolkit function */ -__private_extern__ void stackshot_lock_init( void ) __attribute__((section("__TEXT, initcode"))); - static uint8_t *type_filter_bitmap; +/* + * This allows kperf to swap out the global state pid when kperf ownership is + * passed from one process to another. It checks the old global state pid so + * that kperf can't accidentally steal control of trace when a non-kperf trace user has + * control of trace. + */ +void +kdbg_swap_global_state_pid(pid_t old_pid, pid_t new_pid); + +void +kdbg_swap_global_state_pid(pid_t old_pid, pid_t new_pid) +{ + if (!(kd_ctrl_page.kdebug_flags & KDBG_LOCKINIT)) + return; + + lck_mtx_lock(kd_trace_mtx_sysctl); + + if (old_pid == global_state_pid) + global_state_pid = new_pid; + + lck_mtx_unlock(kd_trace_mtx_sysctl); +} + +static uint32_t +kdbg_cpu_count(boolean_t early_trace) +{ + if (early_trace) { + /* + * we've started tracing before the IOKit has even + * started running... just use the static max value + */ + return max_ncpus; + } + + host_basic_info_data_t hinfo; + mach_msg_type_number_t count = HOST_BASIC_INFO_COUNT; + host_info((host_t)1 /* BSD_HOST */, HOST_BASIC_INFO, (host_info_t)&hinfo, &count); + assert(hinfo.logical_cpu_max > 0); + return hinfo.logical_cpu_max; +} + +#if MACH_ASSERT +#endif /* MACH_ASSERT */ + +static void +kdbg_iop_list_callback(kd_iop_t* iop, kd_callback_type type, void* arg) +{ + while (iop) { + iop->callback.func(iop->callback.context, type, arg); + iop = iop->next; + } +} + static void kdbg_set_tracing_enabled(boolean_t enabled, uint32_t trace_type) { int s = ml_set_interrupts_enabled(FALSE); lck_spin_lock(kds_spin_lock); - if (enabled) { kdebug_enable |= trace_type; kd_ctrl_page.kdebug_slowcheck &= ~SLOW_NOLOG; kd_ctrl_page.enabled = 1; + commpage_update_kdebug_enable(); } else { kdebug_enable &= ~(KDEBUG_ENABLE_TRACE|KDEBUG_ENABLE_PPT); kd_ctrl_page.kdebug_slowcheck |= SLOW_NOLOG; kd_ctrl_page.enabled = 0; + commpage_update_kdebug_enable(); } lck_spin_unlock(kds_spin_lock); ml_set_interrupts_enabled(s); + + if (enabled) { + kdbg_iop_list_callback(kd_ctrl_page.kdebug_iops, KD_CALLBACK_KDEBUG_ENABLED, NULL); + } else { + /* + * If you do not flush the IOP trace buffers, they can linger + * for a considerable period; consider code which disables and + * deallocates without a final sync flush. + */ + kdbg_iop_list_callback(kd_ctrl_page.kdebug_iops, KD_CALLBACK_KDEBUG_DISABLED, NULL); + kdbg_iop_list_callback(kd_ctrl_page.kdebug_iops, KD_CALLBACK_SYNC_FLUSH, NULL); + } } static void @@ -329,12 +462,11 @@ kdbg_set_flags(int slowflag, int enableflag, boolean_t enabled) kd_ctrl_page.kdebug_slowcheck &= ~slowflag; kdebug_enable &= ~enableflag; } + lck_spin_unlock(kds_spin_lock); ml_set_interrupts_enabled(s); } - -#ifdef NATIVE_TRACE_FACILITY void disable_wrap(uint32_t *old_slowcheck, uint32_t *old_flags) { @@ -369,26 +501,8 @@ enable_wrap(uint32_t old_slowcheck, boolean_t lostevents) ml_set_interrupts_enabled(s); } -void trace_set_timebases(__unused uint64_t tsc, __unused uint64_t ns) -{ -} -#else -/* Begin functions that are defined twice */ -void trace_set_timebases(uint64_t tsc, uint64_t ns) -{ - int cpu = cpu_number(); - kd_ctrl_page.cpu_timebase[cpu].tsc_base = tsc; - kd_ctrl_page.cpu_timebase[cpu].ns_base = ns; -} - -#endif - static int -#if defined(__i386__) || defined(__x86_64__) create_buffers(boolean_t early_trace) -#else -create_buffers(__unused boolean_t early_trace) -#endif { int i; int p_buffer_size; @@ -396,44 +510,31 @@ create_buffers(__unused boolean_t early_trace) int f_buffers; int error = 0; - /* - * get the number of cpus and cache it - */ -#if defined(__i386__) || defined(__x86_64__) - if (early_trace == TRUE) { - /* - * we've started tracing before the - * IOKit has even started running... just - * use the static max value - */ - kd_cpus = max_ncpus; - } else -#endif - { - host_basic_info_data_t hinfo; - mach_msg_type_number_t count = HOST_BASIC_INFO_COUNT; + /* + * For the duration of this allocation, trace code will only reference + * kdebug_iops. Any iops registered after this enabling will not be + * messaged until the buffers are reallocated. + * + * TLDR; Must read kd_iops once and only once! + */ + kd_ctrl_page.kdebug_iops = kd_iops; -#define BSD_HOST 1 - host_info((host_t)BSD_HOST, HOST_BASIC_INFO, (host_info_t)&hinfo, &count); - kd_cpus = hinfo.logical_cpu_max; - } - if (kmem_alloc(kernel_map, (vm_offset_t *)&kdbip, sizeof(struct kd_bufinfo) * kd_cpus) != KERN_SUCCESS) { - error = ENOSPC; - goto out; - } - trace_handler_map_bufinfo((uintptr_t)kdbip, sizeof(struct kd_bufinfo) * kd_cpus); + /* + * If the list is valid, it is sorted, newest -> oldest. Each iop entry + * has a cpu_id of "the older entry + 1", so the highest cpu_id will + * be the list head + 1. + */ + + kd_ctrl_page.kdebug_cpus = kd_ctrl_page.kdebug_iops ? kd_ctrl_page.kdebug_iops->cpu_id + 1 : kdbg_cpu_count(early_trace); -#if !defined(NATIVE_TRACE_FACILITY) - for(i=0;i<(int)kd_cpus;i++) { - get_nanotime_timebases(i, - &kd_ctrl_page.cpu_timebase[i].tsc_base, - &kd_ctrl_page.cpu_timebase[i].ns_base); + if (kmem_alloc(kernel_map, (vm_offset_t *)&kdbip, sizeof(struct kd_bufinfo) * kd_ctrl_page.kdebug_cpus, VM_KERN_MEMORY_DIAG) != KERN_SUCCESS) { + error = ENOSPC; + goto out; } -#endif - if (nkdbufs < (kd_cpus * EVENTS_PER_STORAGE_UNIT * MIN_STORAGE_UNITS_PER_CPU)) - n_storage_units = kd_cpus * MIN_STORAGE_UNITS_PER_CPU; + if (nkdbufs < (kd_ctrl_page.kdebug_cpus * EVENTS_PER_STORAGE_UNIT * MIN_STORAGE_UNITS_PER_CPU)) + n_storage_units = kd_ctrl_page.kdebug_cpus * MIN_STORAGE_UNITS_PER_CPU; else n_storage_units = nkdbufs / EVENTS_PER_STORAGE_UNIT; @@ -451,19 +552,19 @@ create_buffers(__unused boolean_t early_trace) kd_bufs = NULL; if (kdcopybuf == 0) { - if (kmem_alloc(kernel_map, (vm_offset_t *)&kdcopybuf, (vm_size_t)KDCOPYBUF_SIZE) != KERN_SUCCESS) { + if (kmem_alloc(kernel_map, (vm_offset_t *)&kdcopybuf, (vm_size_t)KDCOPYBUF_SIZE, VM_KERN_MEMORY_DIAG) != KERN_SUCCESS) { error = ENOSPC; goto out; } } - if (kmem_alloc(kernel_map, (vm_offset_t *)&kd_bufs, (vm_size_t)(n_storage_buffers * sizeof(struct kd_storage_buffers))) != KERN_SUCCESS) { + if (kmem_alloc(kernel_map, (vm_offset_t *)&kd_bufs, (vm_size_t)(n_storage_buffers * sizeof(struct kd_storage_buffers)), VM_KERN_MEMORY_DIAG) != KERN_SUCCESS) { error = ENOSPC; goto out; } bzero(kd_bufs, n_storage_buffers * sizeof(struct kd_storage_buffers)); for (i = 0; i < f_buffers; i++) { - if (kmem_alloc(kernel_map, (vm_offset_t *)&kd_bufs[i].kdsb_addr, (vm_size_t)f_buffer_size) != KERN_SUCCESS) { + if (kmem_alloc(kernel_map, (vm_offset_t *)&kd_bufs[i].kdsb_addr, (vm_size_t)f_buffer_size, VM_KERN_MEMORY_DIAG) != KERN_SUCCESS) { error = ENOSPC; goto out; } @@ -472,7 +573,7 @@ create_buffers(__unused boolean_t early_trace) kd_bufs[i].kdsb_size = f_buffer_size; } if (p_buffer_size) { - if (kmem_alloc(kernel_map, (vm_offset_t *)&kd_bufs[i].kdsb_addr, (vm_size_t)p_buffer_size) != KERN_SUCCESS) { + if (kmem_alloc(kernel_map, (vm_offset_t *)&kd_bufs[i].kdsb_addr, (vm_size_t)p_buffer_size, VM_KERN_MEMORY_DIAG) != KERN_SUCCESS) { error = ENOSPC; goto out; } @@ -490,8 +591,6 @@ create_buffers(__unused boolean_t early_trace) n_elements = kd_bufs[i].kdsb_size / sizeof(struct kd_storage); kds = kd_bufs[i].kdsb_addr; - trace_handler_map_buffer(i, (uintptr_t)kd_bufs[i].kdsb_addr, kd_bufs[i].kdsb_size); - for (n = 0; n < n_elements; n++) { kds[n].kds_next.buffer_index = kd_ctrl_page.kds_free_list.buffer_index; kds[n].kds_next.offset = kd_ctrl_page.kds_free_list.offset; @@ -502,15 +601,15 @@ create_buffers(__unused boolean_t early_trace) n_storage_units += n_elements; } - bzero((char *)kdbip, sizeof(struct kd_bufinfo) * kd_cpus); + bzero((char *)kdbip, sizeof(struct kd_bufinfo) * kd_ctrl_page.kdebug_cpus); - for (i = 0; i < (int)kd_cpus; i++) { + for (i = 0; i < (int)kd_ctrl_page.kdebug_cpus; i++) { kdbip[i].kd_list_head.raw = KDS_PTR_NULL; kdbip[i].kd_list_tail.raw = KDS_PTR_NULL; kdbip[i].kd_lostevents = FALSE; kdbip[i].num_bufs = 0; } - + kd_ctrl_page.kdebug_flags |= KDBG_BUFINIT; kd_ctrl_page.kds_inuse_count = 0; @@ -522,17 +621,15 @@ out: return(error); } - static void delete_buffers(void) { - int i; + int i; if (kd_bufs) { for (i = 0; i < n_storage_buffers; i++) { if (kd_bufs[i].kdsb_addr) { kmem_free(kernel_map, (vm_offset_t)kd_bufs[i].kdsb_addr, (vm_size_t)kd_bufs[i].kdsb_size); - trace_handler_unmap_buffer(i); } } kmem_free(kernel_map, (vm_offset_t)kd_bufs, (vm_size_t)(n_storage_buffers * sizeof(struct kd_storage_buffers))); @@ -548,17 +645,15 @@ delete_buffers(void) kd_ctrl_page.kds_free_list.raw = KDS_PTR_NULL; if (kdbip) { - trace_handler_unmap_bufinfo(); - - kmem_free(kernel_map, (vm_offset_t)kdbip, sizeof(struct kd_bufinfo) * kd_cpus); + kmem_free(kernel_map, (vm_offset_t)kdbip, sizeof(struct kd_bufinfo) * kd_ctrl_page.kdebug_cpus); kdbip = NULL; } + kd_ctrl_page.kdebug_iops = NULL; + kd_ctrl_page.kdebug_cpus = 0; kd_ctrl_page.kdebug_flags &= ~KDBG_BUFINIT; } - -#ifdef NATIVE_TRACE_FACILITY void release_storage_unit(int cpu, uint32_t kdsp_raw) { @@ -587,7 +682,7 @@ release_storage_unit(int cpu, uint32_t kdsp_raw) */ kdsp_actual = POINTER_FROM_KDS_PTR(kdsp); kdbp->kd_list_head = kdsp_actual->kds_next; - + kdsp_actual->kds_next = kd_ctrl_page.kds_free_list; kd_ctrl_page.kds_free_list = kdsp; @@ -636,7 +731,7 @@ allocate_storage_unit(int cpu) kdbp_vict = NULL; oldest_ts = (uint64_t)-1; - for (kdbp_try = &kdbip[0]; kdbp_try < &kdbip[kd_cpus]; kdbp_try++) { + for (kdbp_try = &kdbip[0]; kdbp_try < &kdbip[kd_ctrl_page.kdebug_cpus]; kdbp_try++) { if (kdbp_try->kd_list_head.raw == KDS_PTR_NULL) { /* @@ -671,6 +766,7 @@ allocate_storage_unit(int cpu) if (kdbp_vict == NULL) { kdebug_enable = 0; kd_ctrl_page.enabled = 0; + commpage_update_kdebug_enable(); retval = FALSE; goto out; } @@ -706,27 +802,214 @@ out: return (retval); } -#endif + +int +kernel_debug_register_callback(kd_callback_t callback) +{ + kd_iop_t* iop; + if (kmem_alloc(kernel_map, (vm_offset_t *)&iop, sizeof(kd_iop_t), VM_KERN_MEMORY_DIAG) == KERN_SUCCESS) { + memcpy(&iop->callback, &callback, sizeof(kd_callback_t)); + + /* + * Some IOP clients are not providing a name. + * + * Remove when fixed. + */ + { + boolean_t is_valid_name = FALSE; + for (uint32_t length=0; length 0x20 && callback.iop_name[length] < 0x7F) + continue; + if (callback.iop_name[length] == 0) { + if (length) + is_valid_name = TRUE; + break; + } + } + + if (!is_valid_name) { + strlcpy(iop->callback.iop_name, "IOP-???", sizeof(iop->callback.iop_name)); + } + } + + iop->last_timestamp = 0; + + do { + /* + * We use two pieces of state, the old list head + * pointer, and the value of old_list_head->cpu_id. + * If we read kd_iops more than once, it can change + * between reads. + * + * TLDR; Must not read kd_iops more than once per loop. + */ + iop->next = kd_iops; + iop->cpu_id = iop->next ? (iop->next->cpu_id+1) : kdbg_cpu_count(FALSE); + + /* + * Header says OSCompareAndSwapPtr has a memory barrier + */ + } while (!OSCompareAndSwapPtr(iop->next, iop, (void* volatile*)&kd_iops)); + + return iop->cpu_id; + } + + return 0; +} void -kernel_debug_internal( +kernel_debug_enter( + uint32_t coreid, uint32_t debugid, + uint64_t timestamp, uintptr_t arg1, uintptr_t arg2, uintptr_t arg3, uintptr_t arg4, - uintptr_t arg5, - int entropy_flag); + uintptr_t threadid + ) +{ + uint32_t bindx; + kd_buf *kd; + struct kd_bufinfo *kdbp; + struct kd_storage *kdsp_actual; + union kds_ptr kds_raw; + + if (kd_ctrl_page.kdebug_slowcheck) { + + if ( (kd_ctrl_page.kdebug_slowcheck & SLOW_NOLOG) || !(kdebug_enable & (KDEBUG_ENABLE_TRACE|KDEBUG_ENABLE_PPT))) + goto out1; + + if (kd_ctrl_page.kdebug_flags & KDBG_TYPEFILTER_CHECK) { + /* + * Recheck if TYPEFILTER is being used, and if so, + * dereference bitmap. If the trace facility is being + * disabled, we have ~100ms of preemption-free CPU + * usage to access the bitmap. + */ + disable_preemption(); + if (kd_ctrl_page.kdebug_flags & KDBG_TYPEFILTER_CHECK) { + if (isset(type_filter_bitmap, KDBG_EXTRACT_CSC(debugid))) + goto record_event_preempt_disabled; + } + enable_preemption(); + goto out1; + } + else if (kd_ctrl_page.kdebug_flags & KDBG_RANGECHECK) { + if (debugid >= kdlog_beg && debugid <= kdlog_end) + goto record_event; + goto out1; + } + else if (kd_ctrl_page.kdebug_flags & KDBG_VALCHECK) { + if ((debugid & KDBG_EVENTID_MASK) != kdlog_value1 && + (debugid & KDBG_EVENTID_MASK) != kdlog_value2 && + (debugid & KDBG_EVENTID_MASK) != kdlog_value3 && + (debugid & KDBG_EVENTID_MASK) != kdlog_value4) + goto out1; + } + } + +record_event: + + disable_preemption(); + +record_event_preempt_disabled: + if (kd_ctrl_page.enabled == 0) + goto out; + + kdbp = &kdbip[coreid]; + timestamp &= KDBG_TIMESTAMP_MASK; + +#if KDEBUG_MOJO_TRACE + if (kdebug_enable & KDEBUG_ENABLE_SERIAL) + kdebug_serial_print(coreid, debugid, timestamp, + arg1, arg2, arg3, arg4, threadid); +#endif + +retry_q: + kds_raw = kdbp->kd_list_tail; + + if (kds_raw.raw != KDS_PTR_NULL) { + kdsp_actual = POINTER_FROM_KDS_PTR(kds_raw); + bindx = kdsp_actual->kds_bufindx; + } else + kdsp_actual = NULL; + + if (kdsp_actual == NULL || bindx >= EVENTS_PER_STORAGE_UNIT) { + if (allocate_storage_unit(coreid) == FALSE) { + /* + * this can only happen if wrapping + * has been disabled + */ + goto out; + } + goto retry_q; + } + if ( !OSCompareAndSwap(bindx, bindx + 1, &kdsp_actual->kds_bufindx)) + goto retry_q; + + // IOP entries can be allocated before xnu allocates and inits the buffer + if (timestamp < kdsp_actual->kds_timestamp) + kdsp_actual->kds_timestamp = timestamp; + + kd = &kdsp_actual->kds_records[bindx]; + + kd->debugid = debugid; + kd->arg1 = arg1; + kd->arg2 = arg2; + kd->arg3 = arg3; + kd->arg4 = arg4; + kd->arg5 = threadid; + + kdbg_set_timestamp_and_cpu(kd, timestamp, coreid); + + OSAddAtomic(1, &kdsp_actual->kds_bufcnt); +out: + enable_preemption(); +out1: + if ((kds_waiter && kd_ctrl_page.kds_inuse_count >= n_storage_threshold)) { + boolean_t need_kds_wakeup = FALSE; + int s; + + /* + * try to take the lock here to synchronize with the + * waiter entering the blocked state... use the try + * mode to prevent deadlocks caused by re-entering this + * routine due to various trace points triggered in the + * lck_spin_sleep_xxxx routines used to actually enter + * our wait condition... no problem if we fail, + * there will be lots of additional events coming in that + * will eventually succeed in grabbing this lock + */ + s = ml_set_interrupts_enabled(FALSE); + + if (lck_spin_try_lock(kdw_spin_lock)) { + + if (kds_waiter && kd_ctrl_page.kds_inuse_count >= n_storage_threshold) { + kds_waiter = 0; + need_kds_wakeup = TRUE; + } + lck_spin_unlock(kdw_spin_lock); + } + + ml_set_interrupts_enabled(s); + + if (need_kds_wakeup == TRUE) + wakeup(&kds_waiter); + } +} + + -__attribute__((always_inline)) void +static void kernel_debug_internal( uint32_t debugid, uintptr_t arg1, uintptr_t arg2, uintptr_t arg3, uintptr_t arg4, - uintptr_t arg5, - int entropy_flag) + uintptr_t arg5) { struct proc *curproc; uint64_t now; @@ -755,30 +1038,6 @@ kernel_debug_internal( chudhook(debugid, arg1, arg2, arg3, arg4, arg5); ml_set_interrupts_enabled(s); } - if ((kdebug_enable & KDEBUG_ENABLE_ENTROPY) && entropy_flag) { - - now = mach_absolute_time(); - - s = ml_set_interrupts_enabled(FALSE); - lck_spin_lock(kds_spin_lock); - - if (kdebug_enable & KDEBUG_ENABLE_ENTROPY) { - - if (kd_entropy_indx < kd_entropy_count) { - kd_entropy_buffer[kd_entropy_indx] = now; - kd_entropy_indx++; - } - if (kd_entropy_indx == kd_entropy_count) { - /* - * Disable entropy collection - */ - kdebug_enable &= ~KDEBUG_ENABLE_ENTROPY; - kd_ctrl_page.kdebug_slowcheck &= ~SLOW_ENTROPY; - } - } - lck_spin_unlock(kds_spin_lock); - ml_set_interrupts_enabled(s); - } if ( (kd_ctrl_page.kdebug_slowcheck & SLOW_NOLOG) || !(kdebug_enable & (KDEBUG_ENABLE_TRACE|KDEBUG_ENABLE_PPT))) goto out1; @@ -809,33 +1068,60 @@ kernel_debug_internal( if (kd_ctrl_page.kdebug_flags & KDBG_TYPEFILTER_CHECK) { /* Always record trace system info */ - if (EXTRACT_CLASS(debugid) == DBG_TRACE) + if (KDBG_EXTRACT_CLASS(debugid) == DBG_TRACE) goto record_event; - if (isset(type_filter_bitmap, EXTRACT_CSC(debugid))) - goto record_event; + /* + * Recheck if TYPEFILTER is being used, and if so, + * dereference bitmap. If the trace facility is being + * disabled, we have ~100ms of preemption-free CPU + * usage to access the bitmap. + */ + disable_preemption(); + if (kd_ctrl_page.kdebug_flags & KDBG_TYPEFILTER_CHECK) { + if (isset(type_filter_bitmap, KDBG_EXTRACT_CSC(debugid))) + goto record_event_preempt_disabled; + } + enable_preemption(); goto out1; } else if (kd_ctrl_page.kdebug_flags & KDBG_RANGECHECK) { - if ((debugid >= kdlog_beg && debugid <= kdlog_end) || (debugid >> 24) == DBG_TRACE) - goto record_event; - if (kdlog_sched_events && (debugid & 0xffff0000) == (MACHDBG_CODE(DBG_MACH_SCHED, 0) | DBG_FUNC_NONE)) + /* Always record trace system info */ + if (KDBG_EXTRACT_CLASS(debugid) == DBG_TRACE) goto record_event; - goto out1; + + if (debugid < kdlog_beg || debugid > kdlog_end) + goto out1; } else if (kd_ctrl_page.kdebug_flags & KDBG_VALCHECK) { - if ((debugid & DBG_FUNC_MASK) != kdlog_value1 && - (debugid & DBG_FUNC_MASK) != kdlog_value2 && - (debugid & DBG_FUNC_MASK) != kdlog_value3 && - (debugid & DBG_FUNC_MASK) != kdlog_value4 && - (debugid >> 24 != DBG_TRACE)) + /* Always record trace system info */ + if (KDBG_EXTRACT_CLASS(debugid) == DBG_TRACE) + goto record_event; + + if ((debugid & KDBG_EVENTID_MASK) != kdlog_value1 && + (debugid & KDBG_EVENTID_MASK) != kdlog_value2 && + (debugid & KDBG_EVENTID_MASK) != kdlog_value3 && + (debugid & KDBG_EVENTID_MASK) != kdlog_value4) goto out1; } } record_event: disable_preemption(); + +record_event_preempt_disabled: + if (kd_ctrl_page.enabled == 0) + goto out; + cpu = cpu_number(); kdbp = &kdbip[cpu]; + +#if KDEBUG_MOJO_TRACE + if (kdebug_enable & KDEBUG_ENABLE_SERIAL) + kdebug_serial_print(cpu, debugid, + mach_absolute_time() & KDBG_TIMESTAMP_MASK, + arg1, arg2, arg3, arg4, arg5); +#endif + retry_q: kds_raw = kdbp->kd_list_tail; @@ -875,19 +1161,17 @@ retry_q: out: enable_preemption(); out1: - if ((kds_waiter && kd_ctrl_page.kds_inuse_count >= n_storage_threshold) || - (kde_waiter && kd_entropy_indx >= kd_entropy_count)) { + if (kds_waiter && kd_ctrl_page.kds_inuse_count >= n_storage_threshold) { uint32_t etype; uint32_t stype; - etype = debugid & DBG_FUNC_MASK; - stype = debugid & DBG_SCALL_MASK; + etype = debugid & KDBG_EVENTID_MASK; + stype = debugid & KDBG_CSC_MASK; if (etype == INTERRUPT || etype == MACH_vmfault || stype == BSC_SysCall || stype == MACH_SysCall) { boolean_t need_kds_wakeup = FALSE; - boolean_t need_kde_wakeup = FALSE; /* * try to take the lock here to synchronize with the @@ -907,18 +1191,12 @@ out1: kds_waiter = 0; need_kds_wakeup = TRUE; } - if (kde_waiter && kd_entropy_indx >= kd_entropy_count) { - kde_waiter = 0; - need_kde_wakeup = TRUE; - } lck_spin_unlock(kdw_spin_lock); } ml_set_interrupts_enabled(s); if (need_kds_wakeup == TRUE) wakeup(&kds_waiter); - if (need_kde_wakeup == TRUE) - wakeup(&kde_waiter); } } } @@ -932,7 +1210,7 @@ kernel_debug( uintptr_t arg4, __unused uintptr_t arg5) { - kernel_debug_internal(debugid, arg1, arg2, arg3, arg4, (uintptr_t)thread_tid(current_thread()), 1); + kernel_debug_internal(debugid, arg1, arg2, arg3, arg4, (uintptr_t)thread_tid(current_thread())); } void @@ -944,31 +1222,459 @@ kernel_debug1( uintptr_t arg4, uintptr_t arg5) { - kernel_debug_internal(debugid, arg1, arg2, arg3, arg4, arg5, 1); + kernel_debug_internal(debugid, arg1, arg2, arg3, arg4, arg5); +} + +void +kernel_debug_string_simple(const char *message) +{ + uintptr_t arg[4] = {0, 0, 0, 0}; + + /* Stuff the message string in the args and log it. */ + strncpy((char *)arg, message, MIN(sizeof(arg), strlen(message))); + KERNEL_DEBUG_EARLY( + TRACE_INFO_STRING, + arg[0], arg[1], arg[2], arg[3]); +} + +extern int master_cpu; /* MACH_KERNEL_PRIVATE */ +/* + * Used prior to start_kern_tracing() being called. + * Log temporarily into a static buffer. + */ +void +kernel_debug_early( + uint32_t debugid, + uintptr_t arg1, + uintptr_t arg2, + uintptr_t arg3, + uintptr_t arg4) +{ + /* If tracing is already initialized, use it */ + if (nkdbufs) { + KERNEL_DEBUG_CONSTANT(debugid, arg1, arg2, arg3, arg4, 0); + return; + } + + /* Do nothing if the buffer is full or we're not on the boot cpu */ + kd_early_overflow = kd_early_index >= KD_EARLY_BUFFER_MAX; + if (kd_early_overflow || + cpu_number() != master_cpu) + return; + + kd_early_buffer[kd_early_index].debugid = debugid; + kd_early_buffer[kd_early_index].timestamp = mach_absolute_time(); + kd_early_buffer[kd_early_index].arg1 = arg1; + kd_early_buffer[kd_early_index].arg2 = arg2; + kd_early_buffer[kd_early_index].arg3 = arg3; + kd_early_buffer[kd_early_index].arg4 = arg4; + kd_early_buffer[kd_early_index].arg5 = 0; + kd_early_index++; +} + +/* + * Transfen the contents of the temporary buffer into the trace buffers. + * Precede that by logging the rebase time (offset) - the TSC-based time (in ns) + * when mach_absolute_time is set to 0. + */ +static void +kernel_debug_early_end(void) +{ + int i; + + if (cpu_number() != master_cpu) + panic("kernel_debug_early_end() not call on boot processor"); + + /* Fake sentinel marking the start of kernel time relative to TSC */ + kernel_debug_enter( + 0, + TRACE_TIMESTAMPS, + 0, + (uint32_t)(tsc_rebase_abs_time >> 32), + (uint32_t)tsc_rebase_abs_time, + 0, + 0, + 0); + for (i = 0; i < kd_early_index; i++) { + kernel_debug_enter( + 0, + kd_early_buffer[i].debugid, + kd_early_buffer[i].timestamp, + kd_early_buffer[i].arg1, + kd_early_buffer[i].arg2, + kd_early_buffer[i].arg3, + kd_early_buffer[i].arg4, + 0); + } + + /* Cut events-lost event on overflow */ + if (kd_early_overflow) + KERNEL_DEBUG_CONSTANT( + TRACE_LOST_EVENTS, 0, 0, 0, 0, 0); + + /* This trace marks the start of kernel tracing */ + kernel_debug_string_simple("early trace done"); +} + +/* + * Returns non-zero if debugid is in a reserved class. + */ +static int +kdebug_validate_debugid(uint32_t debugid) +{ + uint8_t debugid_class; + + debugid_class = KDBG_EXTRACT_CLASS(debugid); + switch (debugid_class) { + case DBG_TRACE: + return EPERM; + } + + return 0; } /* - * Support syscall SYS_kdebug_trace + * Support syscall SYS_kdebug_trace. U64->K32 args may get truncated in kdebug_trace64 */ int -kdebug_trace(__unused struct proc *p, struct kdebug_trace_args *uap, __unused int32_t *retval) +kdebug_trace(struct proc *p, struct kdebug_trace_args *uap, int32_t *retval) +{ + struct kdebug_trace64_args uap64; + + uap64.code = uap->code; + uap64.arg1 = uap->arg1; + uap64.arg2 = uap->arg2; + uap64.arg3 = uap->arg3; + uap64.arg4 = uap->arg4; + + return kdebug_trace64(p, &uap64, retval); +} + +/* + * Support syscall SYS_kdebug_trace64. 64-bit args on K32 will get truncated to fit in 32-bit record format. + */ +int kdebug_trace64(__unused struct proc *p, struct kdebug_trace64_args *uap, __unused int32_t *retval) { + int err; + + if ((err = kdebug_validate_debugid(uap->code)) != 0) { + return err; + } + if ( __probable(kdebug_enable == 0) ) - return(EINVAL); - - kernel_debug_internal(uap->code, uap->arg1, uap->arg2, uap->arg3, uap->arg4, (uintptr_t)thread_tid(current_thread()), 0); + return(0); + + kernel_debug_internal(uap->code, (uintptr_t)uap->arg1, (uintptr_t)uap->arg2, (uintptr_t)uap->arg3, (uintptr_t)uap->arg4, (uintptr_t)thread_tid(current_thread())); return(0); } +/* + * Adding enough padding to contain a full tracepoint for the last + * portion of the string greatly simplifies the logic of splitting the + * string between tracepoints. Full tracepoints can be generated using + * the buffer itself, without having to manually add zeros to pad the + * arguments. + */ + +/* 2 string args in first tracepoint and 9 string data tracepoints */ +#define STR_BUF_ARGS (2 + (9 * 4)) +/* times the size of each arg on K64 */ +#define MAX_STR_LEN (STR_BUF_ARGS * sizeof(uint64_t)) +/* on K32, ending straddles a tracepoint, so reserve blanks */ +#define STR_BUF_SIZE (MAX_STR_LEN + (2 * sizeof(uint32_t))) + +/* + * This function does no error checking and assumes that it is called with + * the correct arguments, including that the buffer pointed to by str is at + * least STR_BUF_SIZE bytes. However, str must be aligned to word-size and + * be NUL-terminated. In cases where a string can fit evenly into a final + * tracepoint without its NUL-terminator, this function will not end those + * strings with a NUL in trace. It's up to clients to look at the function + * qualifier for DBG_FUNC_END in this case, to end the string. + */ +static uint64_t +kernel_debug_string_internal(uint32_t debugid, uint64_t str_id, void *vstr, + size_t str_len) +{ + /* str must be word-aligned */ + uintptr_t *str = vstr; + size_t written = 0; + uintptr_t thread_id; + int i; + uint32_t trace_debugid = TRACEDBG_CODE(DBG_TRACE_STRING, + TRACE_STRING_GLOBAL); + + thread_id = (uintptr_t)thread_tid(current_thread()); + + /* if the ID is being invalidated, just emit that */ + if (str_id != 0 && str_len == 0) { + kernel_debug_internal(trace_debugid | DBG_FUNC_START | DBG_FUNC_END, + (uintptr_t)debugid, (uintptr_t)str_id, 0, 0, + thread_id); + return str_id; + } + + /* generate an ID, if necessary */ + if (str_id == 0) { + str_id = OSIncrementAtomic64((SInt64 *)&g_curr_str_id); + str_id = (str_id & STR_ID_MASK) | g_str_id_signature; + } + + trace_debugid |= DBG_FUNC_START; + /* string can fit in a single tracepoint */ + if (str_len <= (2 * sizeof(uintptr_t))) { + trace_debugid |= DBG_FUNC_END; + } + + kernel_debug_internal(trace_debugid, (uintptr_t)debugid, + (uintptr_t)str_id, str[0], + str[1], thread_id); + + trace_debugid &= KDBG_EVENTID_MASK; + i = 2; + written += 2 * sizeof(uintptr_t); + + for (; written < str_len; i += 4, written += 4 * sizeof(uintptr_t)) { + if ((written + (4 * sizeof(uintptr_t))) >= str_len) { + trace_debugid |= DBG_FUNC_END; + } + kernel_debug_internal(trace_debugid, str[i], + str[i + 1], + str[i + 2], + str[i + 3], thread_id); + } + + return str_id; +} + +/* + * Returns true if the current process can emit events, and false otherwise. + * Trace system and scheduling events circumvent this check, as do events + * emitted in interrupt context. + */ +static boolean_t +kdebug_current_proc_enabled(uint32_t debugid) +{ + /* can't determine current process in interrupt context */ + if (ml_at_interrupt_context()) { + return TRUE; + } + + /* always emit trace system and scheduling events */ + if ((KDBG_EXTRACT_CLASS(debugid) == DBG_TRACE || + (debugid & KDBG_CSC_MASK) == MACHDBG_CODE(DBG_MACH_SCHED, 0))) + { + return TRUE; + } + + if (kd_ctrl_page.kdebug_flags & KDBG_PIDCHECK) { + proc_t cur_proc = current_proc(); + + /* only the process with the kdebug bit set is allowed */ + if (cur_proc && !(cur_proc->p_kdebug)) { + return FALSE; + } + } else if (kd_ctrl_page.kdebug_flags & KDBG_PIDEXCLUDE) { + proc_t cur_proc = current_proc(); + + /* every process except the one with the kdebug bit set is allowed */ + if (cur_proc && cur_proc->p_kdebug) { + return FALSE; + } + } + + return TRUE; +} + +/* + * Returns true if the debugid is disabled by filters, and false if the + * debugid is allowed to be traced. A debugid may not be traced if the + * typefilter disables its class and subclass, it's outside a range + * check, or if it's not an allowed debugid in a value check. Trace + * system events bypass this check. + */ +static boolean_t +kdebug_debugid_enabled(uint32_t debugid) +{ + boolean_t is_enabled = TRUE; + + /* if no filtering is enabled */ + if (!kd_ctrl_page.kdebug_slowcheck) { + return TRUE; + } + + if (KDBG_EXTRACT_CLASS(debugid) == DBG_TRACE) { + return TRUE; + } + + if (kd_ctrl_page.kdebug_flags & KDBG_TYPEFILTER_CHECK) { + disable_preemption(); + + /* + * Recheck if typefilter is still being used. If tracing is being + * disabled, there's a 100ms sleep on the other end to keep the + * bitmap around for this check. + */ + if (kd_ctrl_page.kdebug_flags & KDBG_TYPEFILTER_CHECK) { + if (!(isset(type_filter_bitmap, KDBG_EXTRACT_CSC(debugid)))) { + is_enabled = FALSE; + } + } + + enable_preemption(); + } else if (kd_ctrl_page.kdebug_flags & KDBG_RANGECHECK) { + if (debugid < kdlog_beg || debugid > kdlog_end) { + is_enabled = FALSE; + } + } else if (kd_ctrl_page.kdebug_flags & KDBG_VALCHECK) { + if ((debugid & KDBG_EVENTID_MASK) != kdlog_value1 && + (debugid & KDBG_EVENTID_MASK) != kdlog_value2 && + (debugid & KDBG_EVENTID_MASK) != kdlog_value3 && + (debugid & KDBG_EVENTID_MASK) != kdlog_value4) + { + is_enabled = FALSE; + } + } + + return is_enabled; +} + +/* + * Returns 0 if a string can be traced with these arguments. Returns errno + * value if error occurred. + */ +static errno_t +kdebug_check_trace_string(uint32_t debugid, uint64_t str_id) +{ + /* if there are function qualifiers on the debugid */ + if (debugid & ~KDBG_EVENTID_MASK) { + return EINVAL; + } + + if (kdebug_validate_debugid(debugid)) { + return EPERM; + } + + if (str_id != 0 && (str_id & STR_ID_SIG_MASK) != g_str_id_signature) { + return EINVAL; + } + + return 0; +} + +/* + * Implementation of KPI kernel_debug_string. + */ +int +kernel_debug_string(uint32_t debugid, uint64_t *str_id, const char *str) +{ + /* arguments to tracepoints must be word-aligned */ + __attribute__((aligned(sizeof(uintptr_t)))) char str_buf[STR_BUF_SIZE]; + assert_static(sizeof(str_buf) > MAX_STR_LEN); + vm_size_t len_copied; + int err; + + assert(str_id); + + if (__probable(kdebug_enable == 0)) { + return 0; + } + + if (!kdebug_current_proc_enabled(debugid)) { + return 0; + } + + if (!kdebug_debugid_enabled(debugid)) { + return 0; + } + + if ((err = kdebug_check_trace_string(debugid, *str_id)) != 0) { + return err; + } + + if (str == NULL) { + if (str_id == 0) { + return EINVAL; + } + + *str_id = kernel_debug_string_internal(debugid, *str_id, NULL, 0); + return 0; + } + + memset(str_buf, 0, sizeof(str_buf)); + len_copied = strlcpy(str_buf, str, MAX_STR_LEN + 1); + *str_id = kernel_debug_string_internal(debugid, *str_id, str_buf, + len_copied); + return 0; +} + +/* + * Support syscall kdebug_trace_string. + */ +int +kdebug_trace_string(__unused struct proc *p, + struct kdebug_trace_string_args *uap, + uint64_t *retval) +{ + __attribute__((aligned(sizeof(uintptr_t)))) char str_buf[STR_BUF_SIZE]; + assert_static(sizeof(str_buf) > MAX_STR_LEN); + size_t len_copied; + int err; + + if (__probable(kdebug_enable == 0)) { + return 0; + } + + if (!kdebug_current_proc_enabled(uap->debugid)) { + return 0; + } + + if (!kdebug_debugid_enabled(uap->debugid)) { + return 0; + } + + if ((err = kdebug_check_trace_string(uap->debugid, uap->str_id)) != 0) { + return err; + } + + if (uap->str == USER_ADDR_NULL) { + if (uap->str_id == 0) { + return EINVAL; + } + + *retval = kernel_debug_string_internal(uap->debugid, uap->str_id, + NULL, 0); + return 0; + } + + memset(str_buf, 0, sizeof(str_buf)); + err = copyinstr(uap->str, str_buf, MAX_STR_LEN + 1, &len_copied); + + /* it's alright to truncate the string, so allow ENAMETOOLONG */ + if (err == ENAMETOOLONG) { + str_buf[MAX_STR_LEN] = '\0'; + } else if (err) { + return err; + } + + if (len_copied <= 1) { + return EINVAL; + } + + /* convert back to a length */ + len_copied--; + + *retval = kernel_debug_string_internal(uap->debugid, uap->str_id, str_buf, + len_copied); + return 0; +} static void kdbg_lock_init(void) { if (kd_ctrl_page.kdebug_flags & KDBG_LOCKINIT) return; - - trace_handler_map_ctrl_page((uintptr_t)&kd_ctrl_page, sizeof(kd_ctrl_page), sizeof(struct kd_storage), sizeof(union kds_ptr)); /* * allocate lock group attribute and group @@ -1026,7 +1732,7 @@ kdbg_reinit(boolean_t early_trace) kmem_free(kernel_map, (vm_offset_t)kd_mapptr, kd_mapsize); kd_ctrl_page.kdebug_flags &= ~KDBG_MAPINIT; kd_mapsize = 0; - kd_mapptr = (kd_threadmap *) 0; + kd_mapptr = NULL; kd_mapcount = 0; } ret = kdbg_bootstrap(early_trace); @@ -1107,19 +1813,105 @@ kdbg_resolve_map(thread_t th_act, void *opaque) } } +/* + * + * Writes a cpumap for the given iops_list/cpu_count to the provided buffer. + * + * You may provide a buffer and size, or if you set the buffer to NULL, a + * buffer of sufficient size will be allocated. + * + * If you provide a buffer and it is too small, sets cpumap_size to the number + * of bytes required and returns EINVAL. + * + * On success, if you provided a buffer, cpumap_size is set to the number of + * bytes written. If you did not provide a buffer, cpumap is set to the newly + * allocated buffer and cpumap_size is set to the number of bytes allocated. + * + * NOTE: It may seem redundant to pass both iops and a cpu_count. + * + * We may be reporting data from "now", or from the "past". + * + * The "now" data would be for something like kdbg_readcurcpumap(). + * The "past" data would be for kdbg_readcpumap(). + * + * If we do not pass both iops and cpu_count, and iops is NULL, this function + * will need to read "now" state to get the number of cpus, which would be in + * error if we were reporting "past" state. + */ + +int +kdbg_cpumap_init_internal(kd_iop_t* iops, uint32_t cpu_count, uint8_t** cpumap, uint32_t* cpumap_size) +{ + assert(cpumap); + assert(cpumap_size); + assert(cpu_count); + assert(!iops || iops->cpu_id + 1 == cpu_count); + + uint32_t bytes_needed = sizeof(kd_cpumap_header) + cpu_count * sizeof(kd_cpumap); + uint32_t bytes_available = *cpumap_size; + *cpumap_size = bytes_needed; + + if (*cpumap == NULL) { + if (kmem_alloc(kernel_map, (vm_offset_t*)cpumap, (vm_size_t)*cpumap_size, VM_KERN_MEMORY_DIAG) != KERN_SUCCESS) { + return ENOMEM; + } + } else if (bytes_available < bytes_needed) { + return EINVAL; + } + + kd_cpumap_header* header = (kd_cpumap_header*)(uintptr_t)*cpumap; + + header->version_no = RAW_VERSION1; + header->cpu_count = cpu_count; + + kd_cpumap* cpus = (kd_cpumap*)&header[1]; + + int32_t index = cpu_count - 1; + while (iops) { + cpus[index].cpu_id = iops->cpu_id; + cpus[index].flags = KDBG_CPUMAP_IS_IOP; + bzero(cpus[index].name, sizeof(cpus->name)); + strlcpy(cpus[index].name, iops->callback.iop_name, sizeof(cpus->name)); + + iops = iops->next; + index--; + } + + while (index >= 0) { + cpus[index].cpu_id = index; + cpus[index].flags = 0; + bzero(cpus[index].name, sizeof(cpus->name)); + strlcpy(cpus[index].name, "AP", sizeof(cpus->name)); + + index--; + } + + return KERN_SUCCESS; +} + void -kdbg_mapinit(void) +kdbg_thrmap_init(void) { + if (kd_ctrl_page.kdebug_flags & KDBG_MAPINIT) + return; + + kd_mapptr = kdbg_thrmap_init_internal(0, &kd_mapsize, &kd_mapcount); + + if (kd_mapptr) + kd_ctrl_page.kdebug_flags |= KDBG_MAPINIT; +} + + +kd_threadmap* kdbg_thrmap_init_internal(unsigned int count, unsigned int *mapsize, unsigned int *mapcount) +{ + kd_threadmap *mapptr; struct proc *p; struct krt akrt; int tts_count; /* number of task-to-string structures */ struct tts *tts_mapptr; unsigned int tts_mapsize = 0; - vm_offset_t tts_maptomem=0; int i; - - if (kd_ctrl_page.kdebug_flags & KDBG_MAPINIT) - return; + vm_offset_t kaddr; /* * need to use PROC_SCANPROCLIST with proc_iterate @@ -1129,8 +1921,8 @@ kdbg_mapinit(void) /* * Calculate the sizes of map buffers */ - for (p = allproc.lh_first, kd_mapcount=0, tts_count=0; p; p = p->p_list.le_next) { - kd_mapcount += get_task_numacts((task_t)p->task); + for (p = allproc.lh_first, *mapcount=0, tts_count=0; p; p = p->p_list.le_next) { + *mapcount += get_task_numacts((task_t)p->task); tts_count++; } proc_list_unlock(); @@ -1139,83 +1931,88 @@ kdbg_mapinit(void) * The proc count could change during buffer allocation, * so introduce a small fudge factor to bump up the * buffer sizes. This gives new tasks some chance of - * making into the tables. Bump up by 10%. + * making into the tables. Bump up by 25%. */ - kd_mapcount += kd_mapcount/10; - tts_count += tts_count/10; + *mapcount += *mapcount/4; + tts_count += tts_count/4; + + *mapsize = *mapcount * sizeof(kd_threadmap); - kd_mapsize = kd_mapcount * sizeof(kd_threadmap); + if (count && count < *mapcount) + return (0); - if ((kmem_alloc(kernel_map, & kd_maptomem, (vm_size_t)kd_mapsize) == KERN_SUCCESS)) { - kd_mapptr = (kd_threadmap *) kd_maptomem; - bzero(kd_mapptr, kd_mapsize); + if ((kmem_alloc(kernel_map, &kaddr, (vm_size_t)*mapsize, VM_KERN_MEMORY_DIAG) == KERN_SUCCESS)) { + bzero((void *)kaddr, *mapsize); + mapptr = (kd_threadmap *)kaddr; } else - kd_mapptr = (kd_threadmap *) 0; + return (0); tts_mapsize = tts_count * sizeof(struct tts); - if ((kmem_alloc(kernel_map, & tts_maptomem, (vm_size_t)tts_mapsize) == KERN_SUCCESS)) { - tts_mapptr = (struct tts *) tts_maptomem; - bzero(tts_mapptr, tts_mapsize); - } else - tts_mapptr = (struct tts *) 0; + if ((kmem_alloc(kernel_map, &kaddr, (vm_size_t)tts_mapsize, VM_KERN_MEMORY_DIAG) == KERN_SUCCESS)) { + bzero((void *)kaddr, tts_mapsize); + tts_mapptr = (struct tts *)kaddr; + } else { + kmem_free(kernel_map, (vm_offset_t)mapptr, *mapsize); + return (0); + } /* * We need to save the procs command string * and take a reference for each task associated * with a valid process */ - if (tts_mapptr) { - /* - * should use proc_iterate - */ - proc_list_lock(); - - for (p = allproc.lh_first, i=0; p && i < tts_count; p = p->p_list.le_next) { - if (p->p_lflag & P_LEXIT) - continue; - - if (p->task) { - task_reference(p->task); - tts_mapptr[i].task = p->task; - tts_mapptr[i].pid = p->p_pid; - (void)strlcpy(tts_mapptr[i].task_comm, p->p_comm, sizeof(tts_mapptr[i].task_comm)); - i++; - } - } - tts_count = i; - proc_list_unlock(); + proc_list_lock(); + + /* + * should use proc_iterate + */ + for (p = allproc.lh_first, i=0; p && i < tts_count; p = p->p_list.le_next) { + if (p->p_lflag & P_LEXIT) + continue; + + if (p->task) { + task_reference(p->task); + tts_mapptr[i].task = p->task; + tts_mapptr[i].pid = p->p_pid; + (void)strlcpy(tts_mapptr[i].task_comm, p->p_comm, sizeof(tts_mapptr[i].task_comm)); + i++; + } } + tts_count = i; - if (kd_mapptr && tts_mapptr) { - kd_ctrl_page.kdebug_flags |= KDBG_MAPINIT; + proc_list_unlock(); - /* - * Initialize thread map data - */ - akrt.map = kd_mapptr; - akrt.count = 0; - akrt.maxcount = kd_mapcount; + /* + * Initialize thread map data + */ + akrt.map = mapptr; + akrt.count = 0; + akrt.maxcount = *mapcount; - for (i = 0; i < tts_count; i++) { - akrt.atts = &tts_mapptr[i]; - task_act_iterate_wth_args(tts_mapptr[i].task, kdbg_resolve_map, &akrt); - task_deallocate((task_t) tts_mapptr[i].task); - } - kmem_free(kernel_map, (vm_offset_t)tts_mapptr, tts_mapsize); + for (i = 0; i < tts_count; i++) { + akrt.atts = &tts_mapptr[i]; + task_act_iterate_wth_args(tts_mapptr[i].task, kdbg_resolve_map, &akrt); + task_deallocate((task_t) tts_mapptr[i].task); } + kmem_free(kernel_map, (vm_offset_t)tts_mapptr, tts_mapsize); + + *mapcount = akrt.count; + + return (mapptr); } static void kdbg_clear(void) { - /* + /* * Clean up the trace buffer * First make sure we're not in * the middle of cutting a trace */ kdbg_set_tracing_enabled(FALSE, KDEBUG_ENABLE_TRACE); + kdbg_disable_typefilter(); /* * make sure the SLOW_NOLOG is seen @@ -1224,14 +2021,12 @@ kdbg_clear(void) */ IOSleep(100); - kdlog_sched_events = 0; - global_state_pid = -1; + global_state_pid = -1; kd_ctrl_page.kdebug_flags &= (unsigned int)~KDBG_CKTYPES; kd_ctrl_page.kdebug_flags &= ~(KDBG_NOWRAP | KDBG_RANGECHECK | KDBG_VALCHECK); kd_ctrl_page.kdebug_flags &= ~(KDBG_PIDCHECK | KDBG_PIDEXCLUDE); - kdbg_disable_typefilter(); - + kdbg_deallocate_typefilter(); delete_buffers(); nkdbufs = 0; @@ -1354,16 +2149,13 @@ kdbg_setrtcdec(kd_regtype *kdr) int kdbg_enable_typefilter(void) { - if (kd_ctrl_page.kdebug_flags & KDBG_TYPEFILTER_CHECK) { - /* free the old filter */ - kdbg_disable_typefilter(); - } - - if (kmem_alloc(kernel_map, (vm_offset_t *)&type_filter_bitmap, KDBG_TYPEFILTER_BITMAP_SIZE) != KERN_SUCCESS) { - return ENOSPC; + int ret; + + /* Allocate memory for bitmap if not already allocated */ + ret = kdbg_allocate_typefilter(); + if (ret) { + return ret; } - - bzero(type_filter_bitmap, KDBG_TYPEFILTER_BITMAP_SIZE); /* Turn off range and value checks */ kd_ctrl_page.kdebug_flags &= ~(KDBG_RANGECHECK | KDBG_VALCHECK); @@ -1379,20 +2171,55 @@ kdbg_disable_typefilter(void) { /* Disable filter checking */ kd_ctrl_page.kdebug_flags &= ~KDBG_TYPEFILTER_CHECK; - + /* Turn off slow checks unless pid checks are using them */ if ( (kd_ctrl_page.kdebug_flags & (KDBG_PIDCHECK | KDBG_PIDEXCLUDE)) ) kdbg_set_flags(SLOW_CHECKS, 0, TRUE); else kdbg_set_flags(SLOW_CHECKS, 0, FALSE); - - if(type_filter_bitmap == NULL) - return 0; - vm_offset_t old_bitmap = (vm_offset_t)type_filter_bitmap; - type_filter_bitmap = NULL; + /* typefilter bitmap will be deallocated later */ + + return 0; +} + +static int +kdbg_allocate_typefilter(void) +{ + if (type_filter_bitmap == NULL) { + vm_offset_t bitmap = 0; + + if (kmem_alloc(kernel_map, &bitmap, KDBG_TYPEFILTER_BITMAP_SIZE, VM_KERN_MEMORY_DIAG) != KERN_SUCCESS) { + return ENOSPC; + } + + bzero((void *)bitmap, KDBG_TYPEFILTER_BITMAP_SIZE); + + if (!OSCompareAndSwapPtr(NULL, (void *)bitmap, &type_filter_bitmap)) { + kmem_free(kernel_map, bitmap, KDBG_TYPEFILTER_BITMAP_SIZE); + return 0; /* someone assigned a buffer */ + } + } else { + bzero(type_filter_bitmap, KDBG_TYPEFILTER_BITMAP_SIZE); + } + + return 0; +} + +static int +kdbg_deallocate_typefilter(void) +{ + if(type_filter_bitmap) { + vm_offset_t bitmap = (vm_offset_t)type_filter_bitmap; + + if (OSCompareAndSwapPtr((void *)bitmap, NULL, &type_filter_bitmap)) { + kmem_free(kernel_map, bitmap, KDBG_TYPEFILTER_BITMAP_SIZE); + return 0; + } else { + /* already swapped */ + } + } - kmem_free(kernel_map, old_bitmap, KDBG_TYPEFILTER_BITMAP_SIZE); return 0; } @@ -1401,18 +2228,11 @@ kdbg_setreg(kd_regtype * kdr) { int ret=0; unsigned int val_1, val_2, val; - - kdlog_sched_events = 0; - switch (kdr->type) { case KDBG_CLASSTYPE : val_1 = (kdr->value1 & 0xff); val_2 = (kdr->value2 & 0xff); - - if (val_1 == DBG_FSYSTEM && val_2 == (DBG_FSYSTEM + 1)) - kdlog_sched_events = 1; - kdlog_beg = (val_1<<24); kdlog_end = (val_2<<24); kd_ctrl_page.kdebug_flags &= (unsigned int)~KDBG_CKTYPES; @@ -1469,114 +2289,500 @@ kdbg_setreg(kd_regtype * kdr) return(ret); } +static int +kdbg_write_to_vnode(caddr_t buffer, size_t size, vnode_t vp, vfs_context_t ctx, off_t file_offset) +{ + return vn_rdwr(UIO_WRITE, vp, buffer, size, file_offset, UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, + vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx)); +} + int -kdbg_getreg(__unused kd_regtype * kdr) +kdbg_write_v3_chunk_header(user_addr_t buffer, uint32_t tag, uint32_t sub_tag, uint64_t length, vnode_t vp, vfs_context_t ctx) { -#if 0 - int i,j, ret=0; - unsigned int val_1, val_2, val; + int ret = KERN_SUCCESS; + kd_chunk_header_v3 header; + + header.tag = tag; + header.sub_tag = sub_tag; + header.length = length; + + // Check that only one of them is valid + assert(!buffer ^ !vp); + assert((vp == NULL) || (ctx != NULL)); + + // Write the 8-byte future_chunk_timestamp field in the payload + if (buffer || vp) { + if (vp) { + ret = kdbg_write_to_vnode((caddr_t)&header, sizeof(kd_chunk_header_v3), vp, ctx, RAW_file_offset); + if (ret) { + goto write_error; + } + RAW_file_offset += (sizeof(kd_chunk_header_v3)); + } + else { + ret = copyout(&header, buffer, sizeof(kd_chunk_header_v3)); + if (ret) { + goto write_error; + } + } + } +write_error: + return ret; +} - switch (kdr->type) { - case KDBG_CLASSTYPE : - val_1 = (kdr->value1 & 0xff); - val_2 = val_1 + 1; - kdlog_beg = (val_1<<24); - kdlog_end = (val_2<<24); - kd_ctrl_page.kdebug_flags &= (unsigned int)~KDBG_CKTYPES; - kd_ctrl_page.kdebug_flags |= (KDBG_RANGECHECK | KDBG_CLASSTYPE); - break; - case KDBG_SUBCLSTYPE : - val_1 = (kdr->value1 & 0xff); - val_2 = (kdr->value2 & 0xff); - val = val_2 + 1; - kdlog_beg = ((val_1<<24) | (val_2 << 16)); - kdlog_end = ((val_1<<24) | (val << 16)); - kd_ctrl_page.kdebug_flags &= (unsigned int)~KDBG_CKTYPES; - kd_ctrl_page.kdebug_flags |= (KDBG_RANGECHECK | KDBG_SUBCLSTYPE); - break; - case KDBG_RANGETYPE : - kdlog_beg = (kdr->value1); - kdlog_end = (kdr->value2); - kd_ctrl_page.kdebug_flags &= (unsigned int)~KDBG_CKTYPES; - kd_ctrl_page.kdebug_flags |= (KDBG_RANGECHECK | KDBG_RANGETYPE); - break; - case KDBG_TYPENONE : - kd_ctrl_page.kdebug_flags &= (unsigned int)~KDBG_CKTYPES; - kdlog_beg = 0; - kdlog_end = 0; - break; - default : - ret = EINVAL; - break; +int +kdbg_write_v3_chunk_header_to_buffer(void * buffer, uint32_t tag, uint32_t sub_tag, uint64_t length) +{ + kd_chunk_header_v3 header; + + header.tag = tag; + header.sub_tag = sub_tag; + header.length = length; + + if (!buffer) { + return 0; + } + + memcpy(buffer, &header, sizeof(kd_chunk_header_v3)); + + return (sizeof(kd_chunk_header_v3)); +} + +int +kdbg_write_v3_chunk_to_fd(uint32_t tag, uint32_t sub_tag, uint64_t length, void *payload, uint64_t payload_size, int fd) +{ + proc_t p; + struct vfs_context context; + struct fileproc *fp; + vnode_t vp; + p = current_proc(); + + proc_fdlock(p); + if ( (fp_lookup(p, fd, &fp, 1)) ) { + proc_fdunlock(p); + return EFAULT; + } + + context.vc_thread = current_thread(); + context.vc_ucred = fp->f_fglob->fg_cred; + + if (FILEGLOB_DTYPE(fp->f_fglob) != DTYPE_VNODE) { + fp_drop(p, fd, fp, 1); + proc_fdunlock(p); + return EBADF; } -#endif /* 0 */ - return(EINVAL); + vp = (struct vnode *) fp->f_fglob->fg_data; + proc_fdunlock(p); + + if ( (vnode_getwithref(vp)) == 0 ) { + RAW_file_offset = fp->f_fglob->fg_offset; + + kd_chunk_header_v3 chunk_header = { .tag = tag, .sub_tag = sub_tag, .length = length }; + + int ret = kdbg_write_to_vnode((caddr_t) &chunk_header, sizeof(kd_chunk_header_v3), vp, &context, RAW_file_offset); + if (!ret) { + RAW_file_offset += sizeof(kd_chunk_header_v3); + } + + ret = kdbg_write_to_vnode((caddr_t) payload, (size_t) payload_size, vp, &context, RAW_file_offset); + if (!ret) { + RAW_file_offset += payload_size; + } + + fp->f_fglob->fg_offset = RAW_file_offset; + vnode_put(vp); + } + + fp_drop(p, fd, fp, 0); + return KERN_SUCCESS; +} + +user_addr_t +kdbg_write_v3_event_chunk_header(user_addr_t buffer, uint32_t tag, uint64_t length, vnode_t vp, vfs_context_t ctx) +{ + uint64_t future_chunk_timestamp = 0; + length += sizeof(uint64_t); + + if (kdbg_write_v3_chunk_header(buffer, tag, V3_EVENT_DATA_VERSION, length, vp, ctx)) { + return 0; + } + if (buffer) { + buffer += sizeof(kd_chunk_header_v3); + } + + // Check that only one of them is valid + assert(!buffer ^ !vp); + assert((vp == NULL) || (ctx != NULL)); + + // Write the 8-byte future_chunk_timestamp field in the payload + if (buffer || vp) { + if (vp) { + int ret = kdbg_write_to_vnode((caddr_t)&future_chunk_timestamp, sizeof(uint64_t), vp, ctx, RAW_file_offset); + if (!ret) { + RAW_file_offset += (sizeof(uint64_t)); + } + } + else { + if (copyout(&future_chunk_timestamp, buffer, sizeof(uint64_t))) { + return 0; + } + } + } + + return (buffer + sizeof(uint64_t)); +} + +int +kdbg_write_v3_header(user_addr_t user_header, size_t *user_header_size, int fd) +{ + int ret = KERN_SUCCESS; + kd_header_v3 header; + + uint8_t* cpumap = 0; + uint32_t cpumap_size = 0; + uint32_t thrmap_size = 0; + + size_t bytes_needed = 0; + + // Check that only one of them is valid + assert(!user_header ^ !fd); + assert(user_header_size); + + if ( !(kd_ctrl_page.kdebug_flags & KDBG_BUFINIT) ) { + ret = EINVAL; + goto bail; + } + + if ( !(user_header || fd) ) { + ret = EINVAL; + goto bail; + } + + // Initialize the cpu map + ret = kdbg_cpumap_init_internal(kd_ctrl_page.kdebug_iops, kd_ctrl_page.kdebug_cpus, &cpumap, &cpumap_size); + if (ret != KERN_SUCCESS) { + goto bail; + } + + // Check if a thread map is initialized + if ( !kd_mapptr ) { + ret = EINVAL; + goto bail; + } + thrmap_size = kd_mapcount * sizeof(kd_threadmap); + + // Setup the header. + // See v3 header description in sys/kdebug.h for more inforamtion. + + header.tag = RAW_VERSION3; + header.sub_tag = V3_HEADER_VERSION; + header.length = ( sizeof(kd_header_v3) + cpumap_size - sizeof(kd_cpumap_header)); + + mach_timebase_info_data_t timebase = {0, 0}; + clock_timebase_info(&timebase); + header.timebase_numer = timebase.numer; + header.timebase_denom = timebase.denom; + header.timestamp = 0; + header.walltime_secs = 0; + header.walltime_usecs = 0; + header.timezone_minuteswest = 0; + header.timezone_dst = 0; + +#if defined __LP64__ + header.flags = 1; +#else + header.flags = 0; +#endif + + // If its a buffer, check if we have enough space to copy the header and the maps. + if (user_header) { + bytes_needed = header.length + thrmap_size + (2 * sizeof(kd_chunk_header_v3)); + if ( !user_header_size ) { + ret = EINVAL; + goto bail; + } + if (*user_header_size < bytes_needed) { + ret = EINVAL; + goto bail; + } + } + + // Start writing the header + if (fd) { + void *hdr_ptr = (void *)(((uintptr_t) &header) + sizeof(kd_chunk_header_v3)); + size_t payload_size = (sizeof(kd_header_v3) - sizeof(kd_chunk_header_v3)); + + ret = kdbg_write_v3_chunk_to_fd(RAW_VERSION3, V3_HEADER_VERSION, header.length, hdr_ptr, payload_size, fd); + if (ret) { + goto bail; + } + } + else { + if (copyout(&header, user_header, sizeof(kd_header_v3))) { + ret = EFAULT; + goto bail; + } + // Update the user pointer + user_header += sizeof(kd_header_v3); + } + + // Write a cpu map. This is a sub chunk of the header + cpumap = (uint8_t*)((uintptr_t) cpumap + sizeof(kd_cpumap_header)); + size_t payload_size = (size_t)(cpumap_size - sizeof(kd_cpumap_header)); + if (fd) { + ret = kdbg_write_v3_chunk_to_fd(V3_CPU_MAP, V3_CPUMAP_VERSION, payload_size, (void *)cpumap, payload_size, fd); + if (ret) { + goto bail; + } + } + else { + ret = kdbg_write_v3_chunk_header(user_header, V3_CPU_MAP, V3_CPUMAP_VERSION, payload_size, NULL, NULL); + if (ret) { + goto bail; + } + user_header += sizeof(kd_chunk_header_v3); + if (copyout(cpumap, user_header, payload_size)) { + ret = EFAULT; + goto bail; + } + // Update the user pointer + user_header += payload_size; + } + + // Write a thread map + if (fd) { + ret = kdbg_write_v3_chunk_to_fd(V3_THREAD_MAP, V3_THRMAP_VERSION, thrmap_size, (void *)kd_mapptr, thrmap_size, fd); + if (ret) { + goto bail; + } + } + else { + ret = kdbg_write_v3_chunk_header(user_header, V3_THREAD_MAP, V3_THRMAP_VERSION, thrmap_size, NULL, NULL); + if (ret) { + goto bail; + } + user_header += sizeof(kd_chunk_header_v3); + if (copyout(kd_mapptr, user_header, thrmap_size)) { + ret = EFAULT; + goto bail; + } + user_header += thrmap_size; + } + + if (fd) { + RAW_file_written += bytes_needed; + } + + *user_header_size = bytes_needed; +bail: + if (cpumap) { + kmem_free(kernel_map, (vm_offset_t)cpumap, cpumap_size); + } + return (ret); } +int +kdbg_readcpumap(user_addr_t user_cpumap, size_t *user_cpumap_size) +{ + uint8_t* cpumap = NULL; + uint32_t cpumap_size = 0; + int ret = KERN_SUCCESS; + + if (kd_ctrl_page.kdebug_flags & KDBG_BUFINIT) { + if (kdbg_cpumap_init_internal(kd_ctrl_page.kdebug_iops, kd_ctrl_page.kdebug_cpus, &cpumap, &cpumap_size) == KERN_SUCCESS) { + if (user_cpumap) { + size_t bytes_to_copy = (*user_cpumap_size >= cpumap_size) ? cpumap_size : *user_cpumap_size; + if (copyout(cpumap, user_cpumap, (size_t)bytes_to_copy)) { + ret = EFAULT; + } + } + *user_cpumap_size = cpumap_size; + kmem_free(kernel_map, (vm_offset_t)cpumap, cpumap_size); + } else + ret = EINVAL; + } else + ret = EINVAL; + + return (ret); +} int -kdbg_readmap(user_addr_t buffer, size_t *number, vnode_t vp, vfs_context_t ctx) +kdbg_readcurthrmap(user_addr_t buffer, size_t *bufsize) { - int avail = *number; + kd_threadmap *mapptr; + unsigned int mapsize; + unsigned int mapcount; + unsigned int count = 0; int ret = 0; - uint32_t count = 0; - count = avail/sizeof (kd_threadmap); + count = *bufsize/sizeof(kd_threadmap); + *bufsize = 0; - if (count && (count <= kd_mapcount)) - { - if ((kd_ctrl_page.kdebug_flags & KDBG_MAPINIT) && kd_mapsize && kd_mapptr) - { - if (*number < kd_mapsize) - ret = EINVAL; - else - { - if (vp) - { - RAW_header header; - clock_sec_t secs; - clock_usec_t usecs; - char *pad_buf; - int pad_size; - - header.version_no = RAW_VERSION1; - header.thread_count = count; - - clock_get_calendar_microtime(&secs, &usecs); - header.TOD_secs = secs; - header.TOD_usecs = usecs; - - ret = vn_rdwr(UIO_WRITE, vp, (caddr_t)&header, sizeof(RAW_header), RAW_file_offset, - UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx)); - if (ret) - goto write_error; - RAW_file_offset += sizeof(RAW_header); + if ( (mapptr = kdbg_thrmap_init_internal(count, &mapsize, &mapcount)) ) { + if (copyout(mapptr, buffer, mapcount * sizeof(kd_threadmap))) + ret = EFAULT; + else + *bufsize = (mapcount * sizeof(kd_threadmap)); - ret = vn_rdwr(UIO_WRITE, vp, (caddr_t)kd_mapptr, kd_mapsize, RAW_file_offset, - UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx)); - if (ret) - goto write_error; - RAW_file_offset += kd_mapsize; + kmem_free(kernel_map, (vm_offset_t)mapptr, mapsize); + } else + ret = EINVAL; - pad_size = PAGE_SIZE - (RAW_file_offset & PAGE_MASK_64); + return (ret); +} - if (pad_size) - { - pad_buf = (char *)kalloc(pad_size); - memset(pad_buf, 0, pad_size); +static int +kdbg_write_v1_plus_header(uint32_t count, vnode_t vp, vfs_context_t ctx) +{ + int ret = 0; + RAW_header header; + clock_sec_t secs; + clock_usec_t usecs; + char *pad_buf; + uint32_t pad_size; + uint32_t extra_thread_count = 0; + uint32_t cpumap_size; + unsigned int mapsize = kd_mapcount * sizeof(kd_threadmap); - ret = vn_rdwr(UIO_WRITE, vp, (caddr_t)pad_buf, pad_size, RAW_file_offset, - UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx)); - kfree(pad_buf, pad_size); - - if (ret) - goto write_error; - RAW_file_offset += pad_size; - } - RAW_file_written += sizeof(RAW_header) + kd_mapsize + pad_size; + /* + * To write a RAW_VERSION1+ file, we + * must embed a cpumap in the "padding" + * used to page align the events following + * the threadmap. If the threadmap happens + * to not require enough padding, we + * artificially increase its footprint + * until it needs enough padding. + */ - } else { - if (copyout(kd_mapptr, buffer, kd_mapsize)) + assert(vp); + assert(ctx); + + pad_size = PAGE_16KB - ((sizeof(RAW_header) + (count * sizeof(kd_threadmap))) & PAGE_MASK_64); + cpumap_size = sizeof(kd_cpumap_header) + kd_ctrl_page.kdebug_cpus * sizeof(kd_cpumap); + + if (cpumap_size > pad_size) { + /* If the cpu map doesn't fit in the current available pad_size, + * we increase the pad_size by 16K. We do this so that the event + * data is always available on a page aligned boundary for both + * 4k and 16k systems. We enforce this alignment for the event + * data so that we can take advantage of optimized file/disk writes.*/ + pad_size += PAGE_16KB; + } + + /* The way we are silently embedding a cpumap in the "padding" is by artificially + * increasing the number of thread entries. However, we'll also need to ensure that + * the cpumap is embedded in the last 4K page before when the event data is expected. + * This way the tools can read the data starting the next page boundary on both + * 4K and 16K systems preserving compatibility with older versions of the tools + */ + if (pad_size > PAGE_4KB) { + pad_size -= PAGE_4KB; + extra_thread_count = (pad_size / sizeof(kd_threadmap)) + 1; + } + + header.version_no = RAW_VERSION1; + header.thread_count = count + extra_thread_count; + + clock_get_calendar_microtime(&secs, &usecs); + header.TOD_secs = secs; + header.TOD_usecs = usecs; + + ret = vn_rdwr(UIO_WRITE, vp, (caddr_t)&header, sizeof(RAW_header), RAW_file_offset, + UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx)); + if (ret) + goto write_error; + RAW_file_offset += sizeof(RAW_header); + + ret = vn_rdwr(UIO_WRITE, vp, (caddr_t)kd_mapptr, mapsize, RAW_file_offset, + UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx)); + if (ret) + goto write_error; + RAW_file_offset += mapsize; + + if (extra_thread_count) { + pad_size = extra_thread_count * sizeof(kd_threadmap); + pad_buf = (char *)kalloc(pad_size); + if (!pad_buf) { + ret = ENOMEM; + goto write_error; + } + memset(pad_buf, 0, pad_size); + + ret = vn_rdwr(UIO_WRITE, vp, (caddr_t)pad_buf, pad_size, RAW_file_offset, + UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx)); + kfree(pad_buf, pad_size); + + if (ret) + goto write_error; + RAW_file_offset += pad_size; + + } + + pad_size = PAGE_SIZE - (RAW_file_offset & PAGE_MASK_64); + if (pad_size) { + pad_buf = (char *)kalloc(pad_size); + if (!pad_buf) { + ret = ENOMEM; + goto write_error; + } + memset(pad_buf, 0, pad_size); + + /* + * embed a cpumap in the padding bytes. + * older code will skip this. + * newer code will know how to read it. + */ + uint32_t temp = pad_size; + if (kdbg_cpumap_init_internal(kd_ctrl_page.kdebug_iops, kd_ctrl_page.kdebug_cpus, (uint8_t**)&pad_buf, &temp) != KERN_SUCCESS) { + memset(pad_buf, 0, pad_size); + } + + ret = vn_rdwr(UIO_WRITE, vp, (caddr_t)pad_buf, pad_size, RAW_file_offset, + UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx)); + kfree(pad_buf, pad_size); + + if (ret) + goto write_error; + RAW_file_offset += pad_size; + } + RAW_file_written += sizeof(RAW_header) + mapsize + pad_size; + +write_error: + return ret; +} + +int +kdbg_readthrmap(user_addr_t buffer, size_t *number, vnode_t vp, vfs_context_t ctx) +{ + + int avail = 0; + int ret = 0; + uint32_t count = 0; + unsigned int mapsize; + + if ((!vp && !buffer) || (vp && buffer)) { + return EINVAL; + } + + assert(number); + assert((vp == NULL) || (ctx != NULL)); + + avail = *number; + count = avail/sizeof (kd_threadmap); + mapsize = kd_mapcount * sizeof(kd_threadmap); + + if (count && (count <= kd_mapcount)) { + if ((kd_ctrl_page.kdebug_flags & KDBG_MAPINIT) && kd_mapsize && kd_mapptr) { + if (*number < mapsize) + ret = EINVAL; + else { + if (vp) { + ret = kdbg_write_v1_plus_header(count, vp, ctx); + if (ret) + goto write_error; + } + else { + if (copyout(kd_mapptr, buffer, mapsize)) ret = EINVAL; } } @@ -1591,10 +2797,11 @@ kdbg_readmap(user_addr_t buffer, size_t *number, vnode_t vp, vfs_context_t ctx) { count = 0; - vn_rdwr(UIO_WRITE, vp, (caddr_t)&count, sizeof(uint32_t), RAW_file_offset, - UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx)); - RAW_file_offset += sizeof(uint32_t); - RAW_file_written += sizeof(uint32_t); + ret = kdbg_write_to_vnode((caddr_t)&count, sizeof(uint32_t), vp, ctx, RAW_file_offset); + if (!ret) { + RAW_file_offset += sizeof(uint32_t); + RAW_file_written += sizeof(uint32_t); + } } write_error: if ((kd_ctrl_page.kdebug_flags & KDBG_MAPINIT) && kd_mapsize && kd_mapptr) @@ -1609,111 +2816,50 @@ write_error: } int -kdbg_getentropy (user_addr_t buffer, size_t *number, int ms_timeout) +kdbg_readthrmap_v3(user_addr_t buffer, size_t *number, int fd) { - int avail = *number; + int avail = 0; int ret = 0; - int s; - u_int64_t abstime; - u_int64_t ns; - int wait_result = THREAD_AWAKENED; - - - if (kd_entropy_buffer) - return(EBUSY); - - if (ms_timeout < 0) - return(EINVAL); - - kd_entropy_count = avail/sizeof(uint64_t); - - if (kd_entropy_count > MAX_ENTROPY_COUNT || kd_entropy_count == 0) { - /* - * Enforce maximum entropy entries - */ - return(EINVAL); - } - kd_entropy_bufsize = kd_entropy_count * sizeof(uint64_t); - - /* - * allocate entropy buffer - */ - if (kmem_alloc(kernel_map, &kd_entropy_buftomem, (vm_size_t)kd_entropy_bufsize) == KERN_SUCCESS) { - kd_entropy_buffer = (uint64_t *) kd_entropy_buftomem; - } else { - kd_entropy_buffer = (uint64_t *) 0; - kd_entropy_count = 0; - - return (ENOMEM); - } - kd_entropy_indx = 0; - - KERNEL_DEBUG_CONSTANT(0xbbbbf000 | DBG_FUNC_START, ms_timeout, kd_entropy_count, 0, 0, 0); - - /* - * Enable entropy sampling - */ - kdbg_set_flags(SLOW_ENTROPY, KDEBUG_ENABLE_ENTROPY, TRUE); - - if (ms_timeout) { - ns = (u_int64_t)ms_timeout * (u_int64_t)(1000 * 1000); - nanoseconds_to_absolutetime(ns, &abstime ); - clock_absolutetime_interval_to_deadline( abstime, &abstime ); - } else - abstime = 0; - - s = ml_set_interrupts_enabled(FALSE); - lck_spin_lock(kdw_spin_lock); - - while (wait_result == THREAD_AWAKENED && kd_entropy_indx < kd_entropy_count) { - - kde_waiter = 1; + uint32_t count = 0; + unsigned int mapsize; - if (abstime) { - /* - * wait for the specified timeout or - * until we've hit our sample limit - */ - wait_result = lck_spin_sleep_deadline(kdw_spin_lock, 0, &kde_waiter, THREAD_ABORTSAFE, abstime); - } else { - /* - * wait until we've hit our sample limit - */ - wait_result = lck_spin_sleep(kdw_spin_lock, 0, &kde_waiter, THREAD_ABORTSAFE); - } - kde_waiter = 0; + if ((!fd && !buffer) || (fd && buffer)) { + return EINVAL; } - lck_spin_unlock(kdw_spin_lock); - ml_set_interrupts_enabled(s); - /* - * Disable entropy sampling - */ - kdbg_set_flags(SLOW_ENTROPY, KDEBUG_ENABLE_ENTROPY, FALSE); + assert(number); - KERNEL_DEBUG_CONSTANT(0xbbbbf000 | DBG_FUNC_END, ms_timeout, kd_entropy_indx, 0, 0, 0); - - *number = 0; - ret = 0; + avail = *number; + count = avail/sizeof (kd_threadmap); + mapsize = kd_mapcount * sizeof(kd_threadmap); - if (kd_entropy_indx > 0) { - /* - * copyout the buffer - */ - if (copyout(kd_entropy_buffer, buffer, kd_entropy_indx * sizeof(uint64_t))) + if (count && (count <= kd_mapcount)) { + if ((kd_ctrl_page.kdebug_flags & KDBG_MAPINIT) && kd_mapsize && kd_mapptr) { + if (*number < mapsize) { + ret = EINVAL; + } + else { + ret = kdbg_write_v3_header(buffer, number, fd); + if (ret) { + goto write_error; + } + } + } + else { ret = EINVAL; - else - *number = kd_entropy_indx * sizeof(uint64_t); + } } - /* - * Always cleanup - */ - kd_entropy_count = 0; - kd_entropy_indx = 0; - kd_entropy_buftomem = 0; - kmem_free(kernel_map, (vm_offset_t)kd_entropy_buffer, kd_entropy_bufsize); - kd_entropy_buffer = (uint64_t *) 0; - + else { + ret = EINVAL; + } +write_error: + if ((kd_ctrl_page.kdebug_flags & KDBG_MAPINIT) && kd_mapsize && kd_mapptr) { + kmem_free(kernel_map, (vm_offset_t)kd_mapptr, kd_mapsize); + kd_ctrl_page.kdebug_flags &= ~KDBG_MAPINIT; + kd_mapsize = 0; + kd_mapptr = (kd_threadmap *) 0; + kd_mapcount = 0; + } return(ret); } @@ -1734,15 +2880,21 @@ kdbg_set_nkdbufs(unsigned int value) } -static void +static int kdbg_enable_bg_trace(void) { + int ret = 0; + if (kdlog_bg_trace == TRUE && kdlog_bg_trace_running == FALSE && n_storage_buffers == 0) { nkdbufs = bg_nkdbufs; - kdbg_reinit(FALSE); - kdbg_set_tracing_enabled(TRUE, KDEBUG_ENABLE_TRACE); - kdlog_bg_trace_running = TRUE; + ret = kdbg_reinit(FALSE); + if (0 == ret) { + kdbg_set_tracing_enabled(TRUE, KDEBUG_ENABLE_TRACE); + kdlog_bg_trace_running = TRUE; + } + wakeup(&kdlog_bg_trace); } + return ret; } static void @@ -1795,12 +2947,14 @@ kdbg_control(int *name, u_int namelen, user_addr_t where, size_t *sizep) proc_t p, curproc; if (name[0] == KERN_KDGETENTROPY || - name[0] == KERN_KDWRITETR || - name[0] == KERN_KDWRITEMAP || + name[0] == KERN_KDWRITETR || + name[0] == KERN_KDWRITETR_V3 || + name[0] == KERN_KDWRITEMAP || + name[0] == KERN_KDWRITEMAP_V3 || name[0] == KERN_KDEFLAGS || name[0] == KERN_KDDFLAGS || name[0] == KERN_KDENABLE || - name[0] == KERN_KDENABLE_BG_TRACE || + name[0] == KERN_KDENABLE_BG_TRACE || name[0] == KERN_KDSETBUF) { if ( namelen < 2 ) @@ -1811,76 +2965,109 @@ kdbg_control(int *name, u_int namelen, user_addr_t where, size_t *sizep) kdbg_lock_init(); if ( !(kd_ctrl_page.kdebug_flags & KDBG_LOCKINIT)) - return(ENOSPC); + return(ENOSPC); lck_mtx_lock(kd_trace_mtx_sysctl); switch(name[0]) { - - case KERN_KDGETBUF: - /* - * Does not alter the global_state_pid - * This is a passive request. - */ - if (size < sizeof(kd_bufinfo.nkdbufs)) { - /* - * There is not enough room to return even - * the first element of the info structure. + case KERN_KDGETBUF: + /* + * Does not alter the global_state_pid + * This is a passive request. */ - ret = EINVAL; + if (size < sizeof(kd_bufinfo.nkdbufs)) { + /* + * There is not enough room to return even + * the first element of the info structure. + */ + ret = EINVAL; + goto out; + } + kd_bufinfo.nkdbufs = nkdbufs; + kd_bufinfo.nkdthreads = kd_mapcount; + + if ( (kd_ctrl_page.kdebug_slowcheck & SLOW_NOLOG) ) + kd_bufinfo.nolog = 1; + else + kd_bufinfo.nolog = 0; + + kd_bufinfo.flags = kd_ctrl_page.kdebug_flags; +#if defined(__LP64__) + kd_bufinfo.flags |= KDBG_LP64; +#endif + kd_bufinfo.bufid = global_state_pid; + + if (size >= sizeof(kd_bufinfo)) { + /* + * Provide all the info we have + */ + if (copyout(&kd_bufinfo, where, sizeof(kd_bufinfo))) + ret = EINVAL; + } else { + /* + * For backwards compatibility, only provide + * as much info as there is room for. + */ + if (copyout(&kd_bufinfo, where, size)) + ret = EINVAL; + } + goto out; + case KERN_KDGETENTROPY: { + /* Obsolescent - just fake with a random buffer */ + char *buffer = (char *) kalloc(size); + read_frandom((void *) buffer, size); + ret = copyout(buffer, where, size); + kfree(buffer, size); goto out; } - kd_bufinfo.nkdbufs = nkdbufs; - kd_bufinfo.nkdthreads = kd_mapsize / sizeof(kd_threadmap); - - if ( (kd_ctrl_page.kdebug_slowcheck & SLOW_NOLOG) ) - kd_bufinfo.nolog = 1; - else - kd_bufinfo.nolog = 0; + + case KERN_KDENABLE_BG_TRACE: + bg_nkdbufs = kdbg_set_nkdbufs(value); + kdlog_bg_trace = TRUE; + ret = kdbg_enable_bg_trace(); + goto out; + + case KERN_KDDISABLE_BG_TRACE: + kdlog_bg_trace = FALSE; + kdbg_disable_bg_trace(); + goto out; - kd_bufinfo.flags = kd_ctrl_page.kdebug_flags; -#if defined(__LP64__) - kd_bufinfo.flags |= KDBG_LP64; -#endif - kd_bufinfo.bufid = global_state_pid; - - if (size >= sizeof(kd_bufinfo)) { - /* - * Provide all the info we have - */ - if (copyout(&kd_bufinfo, where, sizeof(kd_bufinfo))) + case KERN_KDWAIT_BG_TRACE_RESET: + if (!kdlog_bg_trace){ ret = EINVAL; - } else { - /* - * For backwards compatibility, only provide - * as much info as there is room for. - */ - if (copyout(&kd_bufinfo, where, size)) + goto out; + } + wait_result_t wait_result = assert_wait(&kdlog_bg_trace, THREAD_ABORTSAFE); + lck_mtx_unlock(kd_trace_mtx_sysctl); + if (wait_result == THREAD_WAITING) + wait_result = thread_block(THREAD_CONTINUE_NULL); + if (wait_result == THREAD_INTERRUPTED) + ret = EINTR; + lck_mtx_lock(kd_trace_mtx_sysctl); + goto out; + + case KERN_KDSET_BG_TYPEFILTER: + if (!kdlog_bg_trace || !kdlog_bg_trace_running){ ret = EINVAL; - } - goto out; - break; + goto out; + } - case KERN_KDGETENTROPY: - if (kd_entropy_buffer) - ret = EBUSY; - else - ret = kdbg_getentropy(where, sizep, value); - goto out; - break; + if (size != KDBG_TYPEFILTER_BITMAP_SIZE) { + ret = EINVAL; + goto out; + } - case KERN_KDENABLE_BG_TRACE: - bg_nkdbufs = kdbg_set_nkdbufs(value); - kdlog_bg_trace = TRUE; - kdbg_enable_bg_trace(); - goto out; - break; + if ((kd_ctrl_page.kdebug_flags & KDBG_TYPEFILTER_CHECK) == 0){ + if ((ret = kdbg_enable_typefilter())) + goto out; + } - case KERN_KDDISABLE_BG_TRACE: - kdlog_bg_trace = FALSE; - kdbg_disable_bg_trace(); - goto out; - break; + if (copyin(where, type_filter_bitmap, KDBG_TYPEFILTER_BITMAP_SIZE)) { + ret = EINVAL; + goto out; + } + kdbg_iop_list_callback(kd_ctrl_page.kdebug_iops, KD_CALLBACK_TYPEFILTER_CHANGED, type_filter_bitmap); + goto out; } if ((curproc = current_proc()) != NULL) @@ -1889,7 +3076,7 @@ kdbg_control(int *name, u_int namelen, user_addr_t where, size_t *sizep) ret = ESRCH; goto out; } - if (global_state_pid == -1) + if (global_state_pid == -1) global_state_pid = curpid; else if (global_state_pid != curpid) { if ((p = proc_find(global_state_pid)) == NULL) { @@ -1937,7 +3124,7 @@ kdbg_control(int *name, u_int namelen, user_addr_t where, size_t *sizep) ret = EINVAL; break; } - kdbg_mapinit(); + kdbg_thrmap_init(); kdbg_set_tracing_enabled(TRUE, value); } @@ -1958,7 +3145,7 @@ kdbg_control(int *name, u_int namelen, user_addr_t where, size_t *sizep) break; case KERN_KDREMOVE: kdbg_clear(); - kdbg_enable_bg_trace(); + ret = kdbg_enable_bg_trace(); break; case KERN_KDSETREG: if(size < sizeof(kd_regtype)) { @@ -1974,22 +3161,16 @@ kdbg_control(int *name, u_int namelen, user_addr_t where, size_t *sizep) ret = kdbg_setreg(&kd_Reg); break; case KERN_KDGETREG: - if (size < sizeof(kd_regtype)) { - ret = EINVAL; - break; - } - ret = kdbg_getreg(&kd_Reg); - if (copyout(&kd_Reg, where, sizeof(kd_regtype))) { - ret = EINVAL; - } kdbg_disable_bg_trace(); - + ret = EINVAL; break; case KERN_KDREADTR: - ret = kdbg_read(where, sizep, NULL, NULL); + ret = kdbg_read(where, sizep, NULL, NULL, RAW_VERSION1); break; - case KERN_KDWRITETR: - case KERN_KDWRITEMAP: + case KERN_KDWRITETR: + case KERN_KDWRITETR_V3: + case KERN_KDWRITEMAP: + case KERN_KDWRITEMAP_V3: { struct vfs_context context; struct fileproc *fp; @@ -1997,9 +3178,7 @@ kdbg_control(int *name, u_int namelen, user_addr_t where, size_t *sizep) vnode_t vp; int fd; - kdbg_disable_bg_trace(); - - if (name[0] == KERN_KDWRITETR) { + if (name[0] == KERN_KDWRITETR || name[0] == KERN_KDWRITETR_V3) { int s; int wait_result = THREAD_AWAKENED; u_int64_t abstime; @@ -2040,7 +3219,7 @@ kdbg_control(int *name, u_int namelen, user_addr_t where, size_t *sizep) context.vc_thread = current_thread(); context.vc_ucred = fp->f_fglob->fg_cred; - if (fp->f_fglob->fg_type != DTYPE_VNODE) { + if (FILEGLOB_DTYPE(fp->f_fglob) != DTYPE_VNODE) { fp_drop(p, fd, fp, 1); proc_fdunlock(p); @@ -2051,25 +3230,87 @@ kdbg_control(int *name, u_int namelen, user_addr_t where, size_t *sizep) proc_fdunlock(p); if ((ret = vnode_getwithref(vp)) == 0) { - - if (name[0] == KERN_KDWRITETR) { + RAW_file_offset = fp->f_fglob->fg_offset; + if (name[0] == KERN_KDWRITETR || name[0] == KERN_KDWRITETR_V3) { number = nkdbufs * sizeof(kd_buf); - KERNEL_DEBUG_CONSTANT((TRACEDBG_CODE(DBG_TRACE_INFO, 3)) | DBG_FUNC_START, 0, 0, 0, 0, 0); - ret = kdbg_read(0, &number, vp, &context); - KERNEL_DEBUG_CONSTANT((TRACEDBG_CODE(DBG_TRACE_INFO, 3)) | DBG_FUNC_END, number, 0, 0, 0, 0); + KERNEL_DEBUG_CONSTANT(TRACE_WRITING_EVENTS | DBG_FUNC_START, 0, 0, 0, 0, 0); + if (name[0] == KERN_KDWRITETR_V3) + ret = kdbg_read(0, &number, vp, &context, RAW_VERSION3); + else + ret = kdbg_read(0, &number, vp, &context, RAW_VERSION1); + KERNEL_DEBUG_CONSTANT(TRACE_WRITING_EVENTS | DBG_FUNC_END, number, 0, 0, 0, 0); *sizep = number; } else { - number = kd_mapsize; - kdbg_readmap(0, &number, vp, &context); + number = kd_mapcount * sizeof(kd_threadmap); + if (name[0] == KERN_KDWRITEMAP_V3) + kdbg_readthrmap_v3(0, &number, fd); + else + kdbg_readthrmap(0, &number, vp, &context); } + fp->f_fglob->fg_offset = RAW_file_offset; vnode_put(vp); } fp_drop(p, fd, fp, 0); break; } + case KERN_KDBUFWAIT: + { + /* WRITETR lite -- just block until there's data */ + int s; + int wait_result = THREAD_AWAKENED; + u_int64_t abstime; + u_int64_t ns; + size_t number = 0; + + kdbg_disable_bg_trace(); + + + if (*sizep) { + ns = ((u_int64_t)*sizep) * (u_int64_t)(1000 * 1000); + nanoseconds_to_absolutetime(ns, &abstime ); + clock_absolutetime_interval_to_deadline( abstime, &abstime ); + } else + abstime = 0; + + s = ml_set_interrupts_enabled(FALSE); + if( !s ) + panic("trying to wait with interrupts off"); + lck_spin_lock(kdw_spin_lock); + + /* drop the mutex so don't exclude others from + * accessing trace + */ + lck_mtx_unlock(kd_trace_mtx_sysctl); + + while (wait_result == THREAD_AWAKENED && + kd_ctrl_page.kds_inuse_count < n_storage_threshold) { + + kds_waiter = 1; + + if (abstime) + wait_result = lck_spin_sleep_deadline(kdw_spin_lock, 0, &kds_waiter, THREAD_ABORTSAFE, abstime); + else + wait_result = lck_spin_sleep(kdw_spin_lock, 0, &kds_waiter, THREAD_ABORTSAFE); + + kds_waiter = 0; + } + + /* check the count under the spinlock */ + number = (kd_ctrl_page.kds_inuse_count >= n_storage_threshold); + + lck_spin_unlock(kdw_spin_lock); + ml_set_interrupts_enabled(s); + + /* pick the mutex back up again */ + lck_mtx_lock(kd_trace_mtx_sysctl); + + /* write out whether we've exceeded the threshold */ + *sizep = number; + break; + } case KERN_KDPIDTR: if (size < sizeof(kd_regtype)) { ret = EINVAL; @@ -2096,10 +3337,16 @@ kdbg_control(int *name, u_int namelen, user_addr_t where, size_t *sizep) ret = kdbg_setpidex(&kd_Reg); break; - case KERN_KDTHRMAP: - ret = kdbg_readmap(where, sizep, NULL, NULL); - break; - case KERN_KDSETRTCDEC: + case KERN_KDCPUMAP: + ret = kdbg_readcpumap(where, sizep); + break; + case KERN_KDTHRMAP: + ret = kdbg_readthrmap(where, sizep, NULL, NULL); + break; + case KERN_KDREADCURTHRMAP: + ret = kdbg_readcurthrmap(where, sizep); + break; + case KERN_KDSETRTCDEC: if (size < sizeof(kd_regtype)) { ret = EINVAL; break; @@ -2115,20 +3362,21 @@ kdbg_control(int *name, u_int namelen, user_addr_t where, size_t *sizep) case KERN_KDSET_TYPEFILTER: kdbg_disable_bg_trace(); - if ((kd_ctrl_page.kdebug_flags & KDBG_TYPEFILTER_CHECK) == 0){ - if ((ret = kdbg_enable_typefilter())) - break; - } - if (size != KDBG_TYPEFILTER_BITMAP_SIZE) { ret = EINVAL; break; } + if ((kd_ctrl_page.kdebug_flags & KDBG_TYPEFILTER_CHECK) == 0){ + if ((ret = kdbg_enable_typefilter())) + break; + } + if (copyin(where, type_filter_bitmap, KDBG_TYPEFILTER_BITMAP_SIZE)) { ret = EINVAL; break; } + kdbg_iop_list_callback(kd_ctrl_page.kdebug_iops, KD_CALLBACK_TYPEFILTER_CHANGED, type_filter_bitmap); break; default: ret = EINVAL; @@ -2147,11 +3395,11 @@ out: * move through the lists w/o use of any locks */ int -kdbg_read(user_addr_t buffer, size_t *number, vnode_t vp, vfs_context_t ctx) +kdbg_read(user_addr_t buffer, size_t *number, vnode_t vp, vfs_context_t ctx, uint32_t file_version) { unsigned int count; unsigned int cpu, min_cpu; - uint64_t mintime, t; + uint64_t mintime, t, barrier = 0; int error = 0; kd_buf *tempbuf; uint32_t rcursor; @@ -2167,6 +3415,7 @@ kdbg_read(user_addr_t buffer, size_t *number, vnode_t vp, vfs_context_t ctx) boolean_t lostevents = FALSE; boolean_t out_of_events = FALSE; + assert(number); count = *number/sizeof(kd_buf); *number = 0; @@ -2174,7 +3423,22 @@ kdbg_read(user_addr_t buffer, size_t *number, vnode_t vp, vfs_context_t ctx) return EINVAL; memset(&lostevent, 0, sizeof(lostevent)); - lostevent.debugid = TRACEDBG_CODE(DBG_TRACE_INFO, 2); + lostevent.debugid = TRACE_LOST_EVENTS; + + /* Capture timestamp. Only sort events that have occured before the timestamp. + * Since the iop is being flushed here, its possible that events occur on the AP + * while running live tracing. If we are disabled, no new events should + * occur on the AP. + */ + + if (kd_ctrl_page.enabled) + { + // timestamp is non-zero value + barrier = mach_absolute_time() & KDBG_TIMESTAMP_MASK; + } + + // Request each IOP to provide us with up to date entries before merging buffers together. + kdbg_iop_list_callback(kd_ctrl_page.kdebug_iops, KD_CALLBACK_SYNC_FLUSH, NULL); /* * because we hold kd_trace_mtx_sysctl, no other control threads can @@ -2193,25 +3457,35 @@ kdbg_read(user_addr_t buffer, size_t *number, vnode_t vp, vfs_context_t ctx) tempbuf_count = KDCOPYBUF_COUNT; while (count) { - tempbuf = kdcopybuf; + tempbuf = kdcopybuf; tempbuf_number = 0; // While space - while (tempbuf_count) { + while (tempbuf_count) { mintime = 0xffffffffffffffffULL; min_kdbp = NULL; min_cpu = 0; // Check all CPUs - for (cpu = 0, kdbp = &kdbip[0]; cpu < kd_cpus; cpu++, kdbp++) { + for (cpu = 0, kdbp = &kdbip[0]; cpu < kd_ctrl_page.kdebug_cpus; cpu++, kdbp++) { // Find one with raw data if ((kdsp = kdbp->kd_list_head).raw == KDS_PTR_NULL) continue; + /* Debugging aid: maintain a copy of the "kdsp" + * index. + */ + volatile union kds_ptr kdsp_shadow; + + kdsp_shadow = kdsp; // Get from cpu data to buffer header to buffer kdsp_actual = POINTER_FROM_KDS_PTR(kdsp); + volatile struct kd_storage *kdsp_actual_shadow; + + kdsp_actual_shadow = kdsp_actual; + // See if there are actual data left in this buffer rcursor = kdsp_actual->kds_readlast; @@ -2220,6 +3494,14 @@ kdbg_read(user_addr_t buffer, size_t *number, vnode_t vp, vfs_context_t ctx) t = kdbg_get_timestamp(&kdsp_actual->kds_records[rcursor]); + if ((t > barrier) && (barrier > 0)) { + /* + * Need to wait to flush iop again before we + * sort any more data from the buffers + */ + out_of_events = TRUE; + break; + } if (t < kdsp_actual->kds_timestamp) { /* * indicates we've not yet completed filling @@ -2289,12 +3571,23 @@ nextevent: break; } if (tempbuf_number) { + if (file_version == RAW_VERSION3) { + if ( !(kdbg_write_v3_event_chunk_header(buffer, V3_RAW_EVENTS, (tempbuf_number * sizeof(kd_buf)), vp, ctx))) { + error = EFAULT; + goto check_error; + } + if (buffer) + buffer += (sizeof(kd_chunk_header_v3) + sizeof(uint64_t)); + assert(count >= (sizeof(kd_chunk_header_v3) + sizeof(uint64_t))); + count -= (sizeof(kd_chunk_header_v3) + sizeof(uint64_t)); + *number += (sizeof(kd_chunk_header_v3) + sizeof(uint64_t)); + } if (vp) { - error = vn_rdwr(UIO_WRITE, vp, (caddr_t)kdcopybuf, tempbuf_number * sizeof(kd_buf), RAW_file_offset, - UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx)); - - RAW_file_offset += (tempbuf_number * sizeof(kd_buf)); + size_t write_size = tempbuf_number * sizeof(kd_buf); + error = kdbg_write_to_vnode((caddr_t)kdcopybuf, write_size, vp, ctx, RAW_file_offset); + if (!error) + RAW_file_offset += write_size; if (RAW_file_written >= RAW_FLUSH_SIZE) { cluster_push(vp, 0); @@ -2305,6 +3598,7 @@ nextevent: error = copyout(kdcopybuf, buffer, tempbuf_number * sizeof(kd_buf)); buffer += (tempbuf_number * sizeof(kd_buf)); } +check_error: if (error) { *number = 0; error = EINVAL; @@ -2336,28 +3630,33 @@ unsigned char *getProcName(struct proc *proc) { } -#define STACKSHOT_SUBSYS_LOCK() lck_mtx_lock(&stackshot_subsys_mutex) -#define STACKSHOT_SUBSYS_UNLOCK() lck_mtx_unlock(&stackshot_subsys_mutex) -#if defined(__i386__) || defined (__x86_64__) -#define TRAP_DEBUGGER __asm__ volatile("int3"); -#endif - -#define SANE_TRACEBUF_SIZE (8 * 1024 * 1024) - -/* Initialize the mutex governing access to the stack snapshot subsystem */ -__private_extern__ void -stackshot_lock_init( void ) +static int +stackshot_kern_return_to_bsd_error(kern_return_t kr) { - stackshot_subsys_lck_grp_attr = lck_grp_attr_alloc_init(); - - stackshot_subsys_lck_grp = lck_grp_alloc_init("stackshot_subsys_lock", stackshot_subsys_lck_grp_attr); - - stackshot_subsys_lck_attr = lck_attr_alloc_init(); - - lck_mtx_init(&stackshot_subsys_mutex, stackshot_subsys_lck_grp, stackshot_subsys_lck_attr); + switch (kr) { + case KERN_SUCCESS: + return 0; + case KERN_RESOURCE_SHORTAGE: + return ENOMEM; + case KERN_NO_SPACE: + return ENOSPC; + case KERN_NO_ACCESS: + return EPERM; + case KERN_MEMORY_PRESENT: + return EEXIST; + case KERN_NOT_SUPPORTED: + return ENOTSUP; + case KERN_NOT_IN_SET: + return ENOENT; + default: + return EINVAL; + } } + /* + * DEPRECATION WARNING: THIS SYSCALL IS BEING REPLACED WITH SYS_stack_snapshot_with_config and SYS_microstackshot. + * * stack_snapshot: Obtains a coherent set of stack traces for all threads * on the system, tracing both kernel and user stacks * where available. Uses machine specific trace routines @@ -2382,103 +3681,215 @@ stackshot_lock_init( void ) int stack_snapshot(struct proc *p, register struct stack_snapshot_args *uap, int32_t *retval) { int error = 0; + kern_return_t kr; if ((error = suser(kauth_cred_get(), &p->p_acflag))) return(error); - return stack_snapshot2(uap->pid, uap->tracebuf, uap->tracebuf_size, - uap->flags, uap->dispatch_offset, retval); + kr = stack_snapshot2(uap->pid, uap->tracebuf, uap->tracebuf_size, uap->flags, retval); + return stackshot_kern_return_to_bsd_error(kr); } +/* + * stack_snapshot_with_config: Obtains a coherent set of stack traces for specified threads on the sysem, + * tracing both kernel and user stacks where available. Allocates a buffer from the + * kernel and maps the buffer into the calling task's address space. + * + * Inputs: uap->stackshot_config_version - version of the stackshot config that is being passed + * uap->stackshot_config - pointer to the stackshot config + * uap->stackshot_config_size- size of the stackshot config being passed + * Outputs: EINVAL if there is a problem with the arguments + * EFAULT if we failed to copy in the arguments succesfully + * EPERM if the caller is not privileged + * ENOTSUP if the caller is passing a version of arguments that is not supported by the kernel + * (indicates libsyscall:kernel mismatch) or if the caller is requesting unsupported flags + * ENOENT if the caller is requesting an existing buffer that doesn't exist or if the + * requested PID isn't found + * ENOMEM if the kernel is unable to allocate enough memory to serve the request + * ENOSPC if there isn't enough space in the caller's address space to remap the buffer + * ESRCH if the target PID isn't found + * returns KERN_SUCCESS on success + */ int -stack_snapshot2(pid_t pid, user_addr_t tracebuf, uint32_t tracebuf_size, uint32_t flags, uint32_t dispatch_offset, int32_t *retval) +stack_snapshot_with_config(struct proc *p, struct stack_snapshot_with_config_args *uap, __unused int *retval) { int error = 0; - unsigned bytesTraced = 0; - boolean_t istate; + kern_return_t kr; - *retval = -1; -/* Serialize tracing */ - STACKSHOT_SUBSYS_LOCK(); - - if ((tracebuf_size <= 0) || (tracebuf_size > SANE_TRACEBUF_SIZE)) { - error = EINVAL; - goto error_exit; - } + if ((error = suser(kauth_cred_get(), &p->p_acflag))) + return(error); - assert(stackshot_snapbuf == NULL); - if (kmem_alloc_kobject(kernel_map, (vm_offset_t *)&stackshot_snapbuf, tracebuf_size) != KERN_SUCCESS) { - error = ENOMEM; - goto error_exit; + if((void*)uap->stackshot_config == NULL) { + return EINVAL; } - if (panic_active()) { - error = ENOMEM; - goto error_exit; + switch (uap->stackshot_config_version) { + case STACKSHOT_CONFIG_TYPE: + if (uap->stackshot_config_size != sizeof(stackshot_config_t)) { + return EINVAL; + } + stackshot_config_t config; + error = copyin(uap->stackshot_config, &config, sizeof(stackshot_config_t)); + if (error != KERN_SUCCESS) + { + return EFAULT; + } + kr = kern_stack_snapshot_internal(uap->stackshot_config_version, &config, sizeof(stackshot_config_t), TRUE); + return stackshot_kern_return_to_bsd_error(kr); + default: + return ENOTSUP; } +} - istate = ml_set_interrupts_enabled(FALSE); -/* Preload trace parameters*/ - kdp_snapshot_preflight(pid, stackshot_snapbuf, tracebuf_size, flags, dispatch_offset); - -/* Trap to the debugger to obtain a coherent stack snapshot; this populates - * the trace buffer +#if CONFIG_TELEMETRY +/* + * microstackshot: Catch all system call for microstackshot related operations, including + * enabling/disabling both global and windowed microstackshots as well + * as retrieving windowed or global stackshots and the boot profile. + * Inputs: uap->tracebuf - address of the user space destination + * buffer + * uap->tracebuf_size - size of the user space trace buffer + * uap->flags - various flags + * Outputs: EPERM if the caller is not privileged + * EINVAL if the supplied mss_args is NULL, mss_args.tracebuf is NULL or mss_args.tracebuf_size is not sane + * ENOMEM if we don't have enough memory to satisfy the request + * *retval contains the number of bytes traced, if successful + * and -1 otherwise. */ +int +microstackshot(struct proc *p, struct microstackshot_args *uap, int32_t *retval) +{ + int error = 0; + kern_return_t kr; - TRAP_DEBUGGER; + if ((error = suser(kauth_cred_get(), &p->p_acflag))) + return(error); - ml_set_interrupts_enabled(istate); + kr = stack_microstackshot(uap->tracebuf, uap->tracebuf_size, uap->flags, retval); + return stackshot_kern_return_to_bsd_error(kr); +} +#endif /* CONFIG_TELEMETRY */ - bytesTraced = kdp_stack_snapshot_bytes_traced(); - - if (bytesTraced > 0) { - if ((error = copyout(stackshot_snapbuf, tracebuf, - ((bytesTraced < tracebuf_size) ? - bytesTraced : tracebuf_size)))) - goto error_exit; - *retval = bytesTraced; - } - else { - error = ENOENT; - goto error_exit; - } +/* + * kern_stack_snapshot_with_reason: Obtains a coherent set of stack traces for specified threads on the sysem, + * tracing both kernel and user stacks where available. Allocates a buffer from the + * kernel and stores the address of this buffer. + * + * Inputs: reason - the reason for triggering a stackshot (unused at the moment, but in the + * future will be saved in the stackshot) + * Outputs: EINVAL/ENOTSUP if there is a problem with the arguments + * EPERM if the caller doesn't pass at least one KERNEL stackshot flag + * ENOMEM if the kernel is unable to allocate enough memory to serve the request + * ESRCH if the target PID isn't found + * returns KERN_SUCCESS on success + */ +int +kern_stack_snapshot_with_reason(__unused char *reason) +{ + stackshot_config_t config; + kern_return_t kr; + + config.sc_pid = -1; + config.sc_flags = (STACKSHOT_SAVE_LOADINFO | STACKSHOT_GET_GLOBAL_MEM_STATS | STACKSHOT_SAVE_IN_KERNEL_BUFFER | + STACKSHOT_KCDATA_FORMAT); + config.sc_since_timestamp = 0; + config.sc_out_buffer_addr = 0; + config.sc_out_size_addr = 0; + + kr = kern_stack_snapshot_internal(STACKSHOT_CONFIG_TYPE, &config, sizeof(stackshot_config_t), FALSE); + return stackshot_kern_return_to_bsd_error(kr); +} - error = kdp_stack_snapshot_geterror(); - if (error == -1) { - error = ENOSPC; - *retval = -1; - goto error_exit; +/* + * stack_snapshot_from_kernel: Stackshot function for kernel consumers who have their own buffer. + * + * Inputs: pid - the PID to be traced or -1 for the whole system + * buf - a pointer to the buffer where the stackshot should be written + * size - the size of the buffer + * flags - flags to be passed to the stackshot + * *bytes_traced - a pointer to be filled with the length of the stackshot + * Outputs: -1 if there is a problem with the arguments + * the error returned by the stackshot code otherwise + */ +int +stack_snapshot_from_kernel(pid_t pid, void *buf, uint32_t size, uint32_t flags, unsigned *bytes_traced) +{ + kern_return_t kr; + + kr = stack_snapshot_from_kernel_internal(pid, buf, size, flags, bytes_traced); + if (kr == KERN_FAILURE) { + return -1; } -error_exit: - if (stackshot_snapbuf != NULL) - kmem_free(kernel_map, (vm_offset_t) stackshot_snapbuf, tracebuf_size); - stackshot_snapbuf = NULL; - STACKSHOT_SUBSYS_UNLOCK(); - return error; + return kr; } void -start_kern_tracing(unsigned int new_nkdbufs, boolean_t need_map) { +start_kern_tracing(unsigned int new_nkdbufs, boolean_t need_map) +{ if (!new_nkdbufs) return; nkdbufs = kdbg_set_nkdbufs(new_nkdbufs); kdbg_lock_init(); - kdbg_reinit(TRUE); - if (need_map == TRUE) - kdbg_mapinit(); - kdbg_set_tracing_enabled(TRUE, KDEBUG_ENABLE_TRACE); -#if defined(__i386__) || defined(__x86_64__) - uint64_t now = mach_absolute_time(); + kernel_debug_string_simple("start_kern_tracing"); + + if (0 == kdbg_reinit(TRUE)) { + + if (need_map == TRUE) { + uint32_t old1, old2; + + kdbg_thrmap_init(); + + disable_wrap(&old1, &old2); + } + + /* Hold off interrupts until the early traces are cut */ + boolean_t s = ml_set_interrupts_enabled(FALSE); - KERNEL_DEBUG_CONSTANT((TRACEDBG_CODE(DBG_TRACE_INFO, 1)) | DBG_FUNC_NONE, - (uint32_t)(tsc_rebase_abs_time >> 32), (uint32_t)tsc_rebase_abs_time, - (uint32_t)(now >> 32), (uint32_t)now, - 0); + kdbg_set_tracing_enabled( + TRUE, + kdebug_serial ? + (KDEBUG_ENABLE_TRACE | KDEBUG_ENABLE_SERIAL) : + KDEBUG_ENABLE_TRACE); + + /* + * Transfer all very early events from the static buffer + * into the real buffers. + */ + kernel_debug_early_end(); + + ml_set_interrupts_enabled(s); + + printf("kernel tracing started\n"); +#if KDEBUG_MOJO_TRACE + if (kdebug_serial) { + printf("serial output enabled with %lu named events\n", + sizeof(kd_events)/sizeof(kd_event_t)); + } #endif - printf("kernel tracing started\n"); + } else { + printf("error from kdbg_reinit, kernel tracing not started\n"); + } +} + +void +start_kern_tracing_with_typefilter(unsigned int new_nkdbufs, + boolean_t need_map, + unsigned int typefilter) +{ + /* startup tracing */ + start_kern_tracing(new_nkdbufs, need_map); + + /* check that tracing was actually enabled */ + if (!(kdebug_enable & KDEBUG_ENABLE_TRACE)) + return; + + /* setup the typefiltering */ + if (0 == kdbg_enable_typefilter()) + setbit(type_filter_bitmap, + typefilter & (KDBG_CSC_MASK >> KDBG_CSC_OFFSET)); } void @@ -2499,25 +3910,31 @@ kdbg_dump_trace_to_file(const char *filename) * The global pid exists, we're running * due to fs_usage, latency, etc... * don't cut the panic/shutdown trace file + * Disable tracing from this point to avoid + * perturbing state. */ + kdebug_enable = 0; + kd_ctrl_page.enabled = 0; + commpage_update_kdebug_enable(); return; } } - KERNEL_DEBUG_CONSTANT((TRACEDBG_CODE(DBG_TRACE_INFO, 0)) | DBG_FUNC_NONE, 0, 0, 0, 0, 0); + KERNEL_DEBUG_CONSTANT(TRACE_PANIC | DBG_FUNC_NONE, 0, 0, 0, 0, 0); kdebug_enable = 0; kd_ctrl_page.enabled = 0; + commpage_update_kdebug_enable(); ctx = vfs_context_kernel(); if ((error = vnode_open(filename, (O_CREAT | FWRITE | O_NOFOLLOW), 0600, 0, &vp, ctx))) return; - number = kd_mapsize; - kdbg_readmap(0, &number, vp, ctx); + number = kd_mapcount * sizeof(kd_threadmap); + kdbg_readthrmap(0, &number, vp, ctx); number = nkdbufs*sizeof(kd_buf); - kdbg_read(0, &number, vp, ctx); + kdbg_read(0, &number, vp, ctx, RAW_VERSION1); vnode_close(vp, FWRITE, ctx); @@ -2545,22 +3962,145 @@ void kdbg_get_task_name(char* name_buf, int len, task_t task) snprintf(name_buf, len, "%p [!bsd]", task); } +#if KDEBUG_MOJO_TRACE +static kd_event_t * +binary_search(uint32_t id) +{ + int low, high, mid; + low = 0; + high = sizeof(kd_events)/sizeof(kd_event_t) - 1; -#if defined(NATIVE_TRACE_FACILITY) -void trace_handler_map_ctrl_page(__unused uintptr_t addr, __unused size_t ctrl_page_size, __unused size_t storage_size, __unused size_t kds_ptr_size) -{ -} -void trace_handler_map_bufinfo(__unused uintptr_t addr, __unused size_t size) -{ -} -void trace_handler_unmap_bufinfo(void) -{ + while (TRUE) + { + mid = (low + high) / 2; + + if (low > high) + return NULL; /* failed */ + else if ( low + 1 >= high) { + /* We have a match */ + if (kd_events[high].id == id) + return &kd_events[high]; + else if (kd_events[low].id == id) + return &kd_events[low]; + else + return NULL; /* search failed */ + } + else if (id < kd_events[mid].id) + high = mid; + else + low = mid; + } } -void trace_handler_map_buffer(__unused int index, __unused uintptr_t addr, __unused size_t size) + +/* + * Look up event id to get name string. + * Using a per-cpu cache of a single entry + * before resorting to a binary search of the full table. + */ +#define NCACHE 1 +static kd_event_t *last_hit[MAX_CPUS]; +static kd_event_t * +event_lookup_cache(uint32_t cpu, uint32_t id) { + if (last_hit[cpu] == NULL || last_hit[cpu]->id != id) + last_hit[cpu] = binary_search(id); + return last_hit[cpu]; } -void trace_handler_unmap_buffer(__unused int index) + +static uint64_t kd_last_timstamp; + +static void +kdebug_serial_print( + uint32_t cpunum, + uint32_t debugid, + uint64_t timestamp, + uintptr_t arg1, + uintptr_t arg2, + uintptr_t arg3, + uintptr_t arg4, + uintptr_t threadid + ) { + char kprintf_line[192]; + char event[40]; + uint64_t us = timestamp / NSEC_PER_USEC; + uint64_t us_tenth = (timestamp % NSEC_PER_USEC) / 100; + uint64_t delta = timestamp - kd_last_timstamp; + uint64_t delta_us = delta / NSEC_PER_USEC; + uint64_t delta_us_tenth = (delta % NSEC_PER_USEC) / 100; + uint32_t event_id = debugid & KDBG_EVENTID_MASK; + const char *command; + const char *bra; + const char *ket; + kd_event_t *ep; + + /* event time and delta from last */ + snprintf(kprintf_line, sizeof(kprintf_line), + "%11llu.%1llu %8llu.%1llu ", + us, us_tenth, delta_us, delta_us_tenth); + + + /* event (id or name) - start prefixed by "[", end postfixed by "]" */ + bra = (debugid & DBG_FUNC_START) ? "[" : " "; + ket = (debugid & DBG_FUNC_END) ? "]" : " "; + ep = event_lookup_cache(cpunum, event_id); + if (ep) { + if (strlen(ep->name) < sizeof(event) - 3) + snprintf(event, sizeof(event), "%s%s%s", + bra, ep->name, ket); + else + snprintf(event, sizeof(event), "%s%x(name too long)%s", + bra, event_id, ket); + } else { + snprintf(event, sizeof(event), "%s%x%s", + bra, event_id, ket); + } + snprintf(kprintf_line + strlen(kprintf_line), + sizeof(kprintf_line) - strlen(kprintf_line), + "%-40s ", event); + + /* arg1 .. arg4 with special cases for strings */ + switch (event_id) { + case VFS_LOOKUP: + case VFS_LOOKUP_DONE: + if (debugid & DBG_FUNC_START) { + /* arg1 hex then arg2..arg4 chars */ + snprintf(kprintf_line + strlen(kprintf_line), + sizeof(kprintf_line) - strlen(kprintf_line), + "%-16lx %-8s%-8s%-8s ", + arg1, (char*)&arg2, (char*)&arg3, (char*)&arg4); + break; + } + /* else fall through for arg1..arg4 chars */ + case TRACE_STRING_EXEC: + case TRACE_STRING_NEWTHREAD: + case TRACE_INFO_STRING: + snprintf(kprintf_line + strlen(kprintf_line), + sizeof(kprintf_line) - strlen(kprintf_line), + "%-8s%-8s%-8s%-8s ", + (char*)&arg1, (char*)&arg2, (char*)&arg3, (char*)&arg4); + break; + default: + snprintf(kprintf_line + strlen(kprintf_line), + sizeof(kprintf_line) - strlen(kprintf_line), + "%-16lx %-16lx %-16lx %-16lx", + arg1, arg2, arg3, arg4); + } + + /* threadid, cpu and command name */ + if (threadid == (uintptr_t)thread_tid(current_thread()) && + current_proc() && + current_proc()->p_comm[0]) + command = current_proc()->p_comm; + else + command = "-"; + snprintf(kprintf_line + strlen(kprintf_line), + sizeof(kprintf_line) - strlen(kprintf_line), + " %-16lx %-2d %s\n", + threadid, cpunum, command); + + kprintf("%s", kprintf_line); + kd_last_timstamp = timestamp; } #endif