X-Git-Url: https://git.saurik.com/apple/xnu.git/blobdiff_plain/490019cf9519204c5fb36b2fba54ceb983bb6b72..a991bd8d3e7fe02dbca0644054bab73c5b75324a:/bsd/kern/kdebug.c diff --git a/bsd/kern/kdebug.c b/bsd/kern/kdebug.c index 556e0d0bb..f0ca4b75c 100644 --- a/bsd/kern/kdebug.c +++ b/bsd/kern/kdebug.c @@ -1,14 +1,14 @@ /* - * Copyright (c) 2000-2013 Apple Inc. All rights reserved. + * Copyright (c) 2000-2019 Apple Inc. All rights reserved. * * @Apple_LICENSE_HEADER_START@ - * + * * The contents of this file constitute Original Code as defined in and * are subject to the Apple Public Source License Version 1.1 (the * "License"). You may not use this file except in compliance with the * License. Please obtain a copy of the License at * http://www.apple.com/publicsource and read it before using this file. - * + * * This Original Code and all software distributed under the License are * distributed on an "AS IS" basis, WITHOUT WARRANTY OF ANY KIND, EITHER * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, @@ -16,13 +16,10 @@ * FITNESS FOR A PARTICULAR PURPOSE OR NON-INFRINGEMENT. Please see the * License for the specific language governing rights and limitations * under the License. - * + * * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ */ - -#include - #include #include #include @@ -30,21 +27,27 @@ #include #include #include +#include +#include #include #include #include -#include -#define HZ 100 #include #include #include +#include +#include #include +#include +#include + #if defined(__i386__) || defined(__x86_64__) #include #include #include +#include #endif #include @@ -53,42 +56,38 @@ #include #include #include +#include #include #include #include #include #include #include +#include +#include #include #include -#include #include #include #include #include #include -#include /* for isset() */ +#include /* for isset() */ -#include /* for host_info() */ +#include /* for host_info() */ #include #include +#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 +extern unsigned int wake_nkdbufs; +extern unsigned int trace_wrap; /* * 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. * @@ -112,81 +111,238 @@ static void kdebug_serial_print( /* forward */ */ typedef struct kd_iop { - kd_callback_t callback; - uint32_t cpu_id; - uint64_t last_timestamp; /* Prevent timer rollback */ - struct kd_iop* next; + 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 */ +/* + * Typefilter(s) + * + * A typefilter is a 8KB bitmap that is used to selectively filter events + * being recorded. It is able to individually address every class & subclass. + * + * There is a shared typefilter in the kernel which is lazily allocated. Once + * allocated, the shared typefilter is never deallocated. The shared typefilter + * is also mapped on demand into userspace processes that invoke kdebug_trace + * API from Libsyscall. When mapped into a userspace process, the memory is + * read only, and does not have a fixed address. + * + * It is a requirement that the kernel's shared typefilter always pass DBG_TRACE + * events. This is enforced automatically, by having the needed bits set any + * time the shared typefilter is mutated. + */ + +typedef uint8_t* typefilter_t; + +static typefilter_t kdbg_typefilter; +static mach_port_t kdbg_typefilter_memory_entry; + +/* + * There are 3 combinations of page sizes: + * + * 4KB / 4KB + * 4KB / 16KB + * 16KB / 16KB + * + * The typefilter is exactly 8KB. In the first two scenarios, we would like + * to use 2 pages exactly; in the third scenario we must make certain that + * a full page is allocated so we do not inadvertantly share 8KB of random + * data to userspace. The round_page_32 macro rounds to kernel page size. + */ +#define TYPEFILTER_ALLOC_SIZE MAX(round_page_32(KDBG_TYPEFILTER_BITMAP_SIZE), KDBG_TYPEFILTER_BITMAP_SIZE) + +static typefilter_t +typefilter_create(void) +{ + typefilter_t tf; + if (KERN_SUCCESS == kmem_alloc(kernel_map, (vm_offset_t*)&tf, TYPEFILTER_ALLOC_SIZE, VM_KERN_MEMORY_DIAG)) { + memset(&tf[KDBG_TYPEFILTER_BITMAP_SIZE], 0, TYPEFILTER_ALLOC_SIZE - KDBG_TYPEFILTER_BITMAP_SIZE); + return tf; + } + return NULL; +} + +static void +typefilter_deallocate(typefilter_t tf) +{ + assert(tf != NULL); + assert(tf != kdbg_typefilter); + kmem_free(kernel_map, (vm_offset_t)tf, TYPEFILTER_ALLOC_SIZE); +} + +static void +typefilter_copy(typefilter_t dst, typefilter_t src) +{ + assert(src != NULL); + assert(dst != NULL); + memcpy(dst, src, KDBG_TYPEFILTER_BITMAP_SIZE); +} + +static void +typefilter_reject_all(typefilter_t tf) +{ + assert(tf != NULL); + memset(tf, 0, KDBG_TYPEFILTER_BITMAP_SIZE); +} + +static void +typefilter_allow_all(typefilter_t tf) +{ + assert(tf != NULL); + memset(tf, ~0, KDBG_TYPEFILTER_BITMAP_SIZE); +} + +static void +typefilter_allow_class(typefilter_t tf, uint8_t class) +{ + assert(tf != NULL); + const uint32_t BYTES_PER_CLASS = 256 / 8; // 256 subclasses, 1 bit each + memset(&tf[class * BYTES_PER_CLASS], 0xFF, BYTES_PER_CLASS); +} + +static void +typefilter_allow_csc(typefilter_t tf, uint16_t csc) +{ + assert(tf != NULL); + setbit(tf, csc); +} + +static bool +typefilter_is_debugid_allowed(typefilter_t tf, uint32_t id) +{ + assert(tf != NULL); + return isset(tf, KDBG_EXTRACT_CSC(id)); +} + +static mach_port_t +typefilter_create_memory_entry(typefilter_t tf) +{ + assert(tf != NULL); + + mach_port_t memory_entry = MACH_PORT_NULL; + memory_object_size_t size = TYPEFILTER_ALLOC_SIZE; + + mach_make_memory_entry_64(kernel_map, + &size, + (memory_object_offset_t)tf, + VM_PROT_READ, + &memory_entry, + MACH_PORT_NULL); + + return memory_entry; +} + +static int kdbg_copyin_typefilter(user_addr_t addr, size_t size); +static void kdbg_enable_typefilter(void); +static void kdbg_disable_typefilter(void); + +/* + * External prototypes + */ + +void task_act_iterate_wth_args(task_t, void (*)(thread_t, void *), void *); +void commpage_update_kdebug_state(void); /* XXX sign */ + +extern int log_leaks; + +/* + * This flag is for testing purposes only -- it's highly experimental and tools + * have not been updated to support it. + */ +static bool kdbg_continuous_time = false; + +static inline uint64_t +kdbg_timestamp(void) +{ + if (kdbg_continuous_time) { + return mach_continuous_time(); + } else { + return mach_absolute_time(); + } +} + +static int kdbg_debug = 0; -/* XXX should probably be static, but it's debugging code... */ -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_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_thrmap_init(void); -int kdbg_reinit(boolean_t); -int kdbg_bootstrap(boolean_t); + +static int kdbg_read(user_addr_t, size_t *, vnode_t, vfs_context_t, uint32_t); +static int kdbg_readcpumap(user_addr_t, size_t *); +static int kdbg_readthrmap_v3(user_addr_t, size_t, int); +static int kdbg_readcurthrmap(user_addr_t, size_t *); +static int kdbg_setreg(kd_regtype *); +static int kdbg_setpidex(kd_regtype *); +static int kdbg_setpid(kd_regtype *); +static void kdbg_thrmap_init(void); +static int kdbg_reinit(bool); +static int kdbg_bootstrap(bool); +static int kdbg_test(size_t flavor); + +static int kdbg_write_v1_header(bool write_thread_map, vnode_t vp, vfs_context_t ctx); +static int kdbg_write_thread_map(vnode_t vp, vfs_context_t ctx); +static int kdbg_copyout_thread_map(user_addr_t buffer, size_t *buffer_size); +static void kdbg_clear_thread_map(void); + +static bool kdbg_wait(uint64_t timeout_ms, bool locked_wait); +static void kdbg_wakeup(void); int kdbg_cpumap_init_internal(kd_iop_t* iops, uint32_t cpu_count, - uint8_t** cpumap, uint32_t* cpumap_size); + uint8_t** cpumap, uint32_t* cpumap_size); -kd_threadmap* kdbg_thrmap_init_internal(unsigned int count, - unsigned int *mapsize, - unsigned int *mapcount); +static kd_threadmap *kdbg_thrmap_init_internal(size_t max_count, + vm_size_t *map_size, vm_size_t *map_count); -static boolean_t kdebug_current_proc_enabled(uint32_t debugid); -static boolean_t kdebug_debugid_enabled(uint32_t debugid); +static bool kdebug_current_proc_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); + 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); + 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); +// Helper functions -static int create_buffers(boolean_t); +static int create_buffers(bool); static void delete_buffers(void); +extern int tasks_count; +extern int threads_count; extern void IOSleep(int); /* trace enable status */ unsigned int kdebug_enable = 0; /* 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_CHUD 0x08 +#define KD_EARLY_BUFFER_SIZE (16 * 1024) +#define KD_EARLY_BUFFER_NBUFS (KD_EARLY_BUFFER_SIZE / sizeof(kd_buf)) +#if defined(__x86_64__) +__attribute__((aligned(KD_EARLY_BUFFER_SIZE))) +static kd_buf kd_early_buffer[KD_EARLY_BUFFER_NBUFS]; +#else /* defined(__x86_64__) */ +/* + * On ARM, the space for this is carved out by osfmk/arm/data.s -- clang + * has problems aligning to greater than 4K. + */ +extern kd_buf kd_early_buffer[KD_EARLY_BUFFER_NBUFS]; +#endif /* !defined(__x86_64__) */ + +static unsigned int kd_early_index = 0; +static bool kd_early_overflow = false; +static bool kd_early_done = false; -#define EVENTS_PER_STORAGE_UNIT 2048 -#define MIN_STORAGE_UNITS_PER_CPU 4 +#define SLOW_NOLOG 0x01 +#define SLOW_CHECKS 0x02 + +#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]) @@ -199,40 +355,42 @@ union kds_ptr { }; struct kd_storage { - union kds_ptr kds_next; + union kds_ptr kds_next; uint32_t kds_bufindx; uint32_t kds_bufcnt; uint32_t kds_readlast; - boolean_t kds_lostevents; + bool kds_lostevents; uint64_t kds_timestamp; - kd_buf kds_records[EVENTS_PER_STORAGE_UNIT]; + kd_buf kds_records[EVENTS_PER_STORAGE_UNIT]; }; -#define MAX_BUFFER_SIZE (1024 * 1024 * 128) -#define N_STORAGE_UNITS_PER_BUFFER (MAX_BUFFER_SIZE / sizeof(struct kd_storage)) +#define MAX_BUFFER_SIZE (1024 * 1024 * 128) +#define N_STORAGE_UNITS_PER_BUFFER (MAX_BUFFER_SIZE / sizeof(struct kd_storage)) +static_assert(N_STORAGE_UNITS_PER_BUFFER <= 0x7ff, + "shoudn't overflow kds_ptr.offset"); struct kd_storage_buffers { - struct kd_storage *kdsb_addr; - uint32_t kdsb_size; + struct kd_storage *kdsb_addr; + uint32_t kdsb_size; }; #define KDS_PTR_NULL 0xffffffff struct kd_storage_buffers *kd_bufs = NULL; -int n_storage_units = 0; -int n_storage_buffers = 0; -int n_storage_threshold = 0; -int kds_waiter = 0; +int n_storage_units = 0; +unsigned int n_storage_buffers = 0; +int n_storage_threshold = 0; +int kds_waiter = 0; #pragma pack(0) struct kd_bufinfo { union kds_ptr kd_list_head; union kds_ptr kd_list_tail; - boolean_t kd_lostevents; + bool kd_lostevents; uint32_t _pad; uint64_t kd_prev_timebase; uint32_t num_bufs; -} __attribute__(( aligned(MAX_CPU_CACHE_LINE_SIZE) )); +} __attribute__((aligned(MAX_CPU_CACHE_LINE_SIZE))); /* @@ -241,11 +399,12 @@ struct kd_bufinfo { */ struct kd_ctrl_page_t { union kds_ptr kds_free_list; - uint32_t enabled :1; - uint32_t _pad0 :31; - int kds_inuse_count; + uint32_t enabled :1; + uint32_t _pad0 :31; + int kds_inuse_count; uint32_t kdebug_flags; uint32_t kdebug_slowcheck; + uint64_t oldest_time; /* * The number of kd_bufinfo structs allocated may not match the current * number of active cpus. We capture the iops list head at initialization @@ -255,63 +414,43 @@ struct kd_ctrl_page_t { */ kd_iop_t* kdebug_iops; uint32_t kdebug_cpus; -} kd_ctrl_page = { .kds_free_list = {.raw = KDS_PTR_NULL}, .kdebug_slowcheck = SLOW_NOLOG }; +} kd_ctrl_page = { + .kds_free_list = {.raw = KDS_PTR_NULL}, + .kdebug_slowcheck = SLOW_NOLOG, + .oldest_time = 0 +}; #pragma pack() struct kd_bufinfo *kdbip = NULL; -#define KDCOPYBUF_COUNT 8192 -#define KDCOPYBUF_SIZE (KDCOPYBUF_COUNT * sizeof(kd_buf)) +#define KDCOPYBUF_COUNT 8192 +#define KDCOPYBUF_SIZE (KDCOPYBUF_COUNT * sizeof(kd_buf)) -#define PAGE_4KB 4096 -#define PAGE_16KB 16384 +#define PAGE_4KB 4096 +#define PAGE_16KB 16384 kd_buf *kdcopybuf = NULL; -boolean_t kdlog_bg_trace = FALSE; -boolean_t kdlog_bg_trace_running = FALSE; -unsigned int bg_nkdbufs = 0; - unsigned int nkdbufs = 0; -unsigned int kdlog_beg=0; -unsigned int kdlog_end=0; -unsigned int kdlog_value1=0; -unsigned int kdlog_value2=0; -unsigned int kdlog_value3=0; -unsigned int kdlog_value4=0; +unsigned int kdlog_beg = 0; +unsigned int kdlog_end = 0; +unsigned int kdlog_value1 = 0; +unsigned int kdlog_value2 = 0; +unsigned int kdlog_value3 = 0; +unsigned int kdlog_value4 = 0; static lck_spin_t * kdw_spin_lock; static lck_spin_t * kds_spin_lock; -static lck_mtx_t * kd_trace_mtx_sysctl; -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; - -extern kern_return_t stack_snapshot2(int pid, user_addr_t tracebuf, uint32_t tracebuf_size, uint32_t flags, int32_t *retval); - -#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 */ - -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 kern_return_t stack_snapshot_from_kernel_internal(int pid, void *buf, uint32_t size, uint32_t flags, unsigned *bytes_traced); - -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; - -off_t RAW_file_offset = 0; -int RAW_file_written = 0; +vm_size_t kd_mapsize = 0; +vm_size_t kd_mapcount = 0; -#define RAW_FLUSH_SIZE (2 * 1024 * 1024) +off_t RAW_file_offset = 0; +int RAW_file_written = 0; -pid_t global_state_pid = -1; /* Used to control exclusive use of kd_buffer */ +#define RAW_FLUSH_SIZE (2 * 1024 * 1024) /* * A globally increasing counter for identifying strings in trace. Starts at @@ -330,82 +469,191 @@ static uint64_t g_curr_str_id = 1; */ 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 INTERRUPT 0x01050000 +#define MACH_vmfault 0x01300008 +#define BSC_SysCall 0x040c0000 +#define MACH_SysCall 0x010c0000 -/* task to string structure */ -struct tts -{ - task_t task; /* from procs task */ - pid_t pid; /* from procs p_pid */ - char task_comm[20]; /* from procs p_comm */ +struct kd_task_name { + task_t ktn_task; + pid_t ktn_pid; + char ktn_name[20]; }; -typedef struct tts tts_t; - -struct krt -{ - kd_threadmap *map; /* pointer to the map buffer */ - int count; - int maxcount; - struct tts *atts; +struct kd_resolver { + kd_threadmap *krs_map; + vm_size_t krs_count; + vm_size_t krs_maxcount; + struct kd_task_name *krs_task; }; -typedef struct krt krt_t; - -/* This is for the CHUD toolkit call */ -typedef void (*kd_chudhook_fn) (uint32_t debugid, uintptr_t arg1, - uintptr_t arg2, uintptr_t arg3, - uintptr_t arg4, uintptr_t arg5); - -volatile kd_chudhook_fn kdebug_chudhook = 0; /* pointer to CHUD toolkit function */ - -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. + * TRACE file formats... + * + * RAW_VERSION0 + * + * uint32_t #threadmaps + * kd_threadmap[] + * kd_buf[] + * + * RAW_VERSION1 + * + * RAW_header, with version_no set to RAW_VERSION1 + * kd_threadmap[] + * Empty space to pad alignment to the nearest page boundary. + * kd_buf[] + * + * RAW_VERSION1+ + * + * RAW_header, with version_no set to RAW_VERSION1 + * kd_threadmap[] + * kd_cpumap_header, with version_no set to RAW_VERSION1 + * kd_cpumap[] + * Empty space to pad alignment to the nearest page boundary. + * kd_buf[] + * + * V1+ implementation details... + * + * It would have been nice to add the cpumap data "correctly", but there were + * several obstacles. Existing code attempts to parse both V1 and V0 files. + * Due to the fact that V0 has no versioning or header, the test looks like + * this: + * + * // Read header + * if (header.version_no != RAW_VERSION1) { // Assume V0 } + * + * If we add a VERSION2 file format, all existing code is going to treat that + * as a VERSION0 file when reading it, and crash terribly when trying to read + * RAW_VERSION2 threadmap entries. + * + * To differentiate between a V1 and V1+ file, read as V1 until you reach + * the padding bytes. Then: + * + * boolean_t is_v1plus = FALSE; + * if (padding_bytes >= sizeof(kd_cpumap_header)) { + * kd_cpumap_header header = // read header; + * if (header.version_no == RAW_VERSION1) { + * is_v1plus = TRUE; + * } + * } + * */ -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); +#define RAW_VERSION3 0x00001000 - if (old_pid == global_state_pid) - global_state_pid = new_pid; +// Version 3 header +// The header chunk has the tag 0x00001000 which also serves as a magic word +// that identifies the file as a version 3 trace file. The header payload is +// a set of fixed fields followed by a variable number of sub-chunks: +/* + * ____________________________________________________________________________ + | Offset | Size | Field | + | ---------------------------------------------------------------------------- + | 0 | 4 | Tag (0x00001000) | + | 4 | 4 | Sub-tag. Represents the version of the header. | + | 8 | 8 | Length of header payload (40+8x) | + | 16 | 8 | Time base info. Two 32-bit numbers, numer/denom, | + | | | for converting timestamps to nanoseconds. | + | 24 | 8 | Timestamp of trace start. | + | 32 | 8 | Wall time seconds since Unix epoch. | + | | | As returned by gettimeofday(). | + | 40 | 4 | Wall time microseconds. As returned by gettimeofday(). | + | 44 | 4 | Local time zone offset in minutes. ( " ) | + | 48 | 4 | Type of daylight savings time correction to apply. ( " ) | + | 52 | 4 | Flags. 1 = 64-bit. Remaining bits should be written | + | | | as 0 and ignored when reading. | + | 56 | 8x | Variable number of sub-chunks. None are required. | + | | | Ignore unknown chunks. | + | ---------------------------------------------------------------------------- + */ +// NOTE: The header sub-chunks are considered part of the header chunk, +// so they must be included in the header chunk’s length field. +// The CPU map is an optional sub-chunk of the header chunk. It provides +// information about the CPUs that are referenced from the trace events. +typedef struct { + uint32_t tag; + uint32_t sub_tag; + uint64_t length; + uint32_t timebase_numer; + uint32_t timebase_denom; + uint64_t timestamp; + uint64_t walltime_secs; + uint32_t walltime_usecs; + uint32_t timezone_minuteswest; + uint32_t timezone_dst; + uint32_t flags; +} __attribute__((packed)) kd_header_v3; + +typedef struct { + uint32_t tag; + uint32_t sub_tag; + uint64_t length; +} __attribute__((packed)) kd_chunk_header_v3; + +#define V3_CONFIG 0x00001b00 +#define V3_CPU_MAP 0x00001c00 +#define V3_THREAD_MAP 0x00001d00 +#define V3_RAW_EVENTS 0x00001e00 +#define V3_NULL_CHUNK 0x00002000 + +// The current version of all kernel managed chunks is 1. The +// V3_CURRENT_CHUNK_VERSION is added to ease the simple case +// when most/all the kernel managed chunks have the same version. + +#define V3_CURRENT_CHUNK_VERSION 1 +#define V3_HEADER_VERSION V3_CURRENT_CHUNK_VERSION +#define V3_CPUMAP_VERSION V3_CURRENT_CHUNK_VERSION +#define V3_THRMAP_VERSION V3_CURRENT_CHUNK_VERSION +#define V3_EVENT_DATA_VERSION V3_CURRENT_CHUNK_VERSION - lck_mtx_unlock(kd_trace_mtx_sysctl); -} +typedef struct krt krt_t; static uint32_t -kdbg_cpu_count(boolean_t early_trace) +kdbg_cpu_count(bool early_trace) { if (early_trace) { - /* - * we've started tracing before the IOKit has even - * started running... just use the static max value - */ +#if defined(__x86_64__) return max_ncpus; +#else /* defined(__x86_64__) */ + return ml_get_cpu_count(); +#endif /* !defined(__x86_64__) */ } +#if defined(__x86_64__) 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; +#else /* defined(__x86_64__) */ + return ml_get_topology_info()->max_cpu_id + 1; +#endif /* !defined(__x86_64__) */ } #if MACH_ASSERT + +static bool +kdbg_iop_list_is_valid(kd_iop_t* iop) +{ + if (iop) { + /* Is list sorted by cpu_id? */ + kd_iop_t* temp = iop; + do { + assert(!temp->next || temp->next->cpu_id == temp->cpu_id - 1); + assert(temp->next || (temp->cpu_id == kdbg_cpu_count(false) || temp->cpu_id == kdbg_cpu_count(true))); + } while ((temp = temp->next)); + + /* Does each entry have a function and a name? */ + temp = iop; + do { + assert(temp->callback.func); + assert(strlen(temp->callback.iop_name) < sizeof(temp->callback.iop_name)); + } while ((temp = temp->next)); + } + + return true; +} + #endif /* MACH_ASSERT */ static void @@ -417,43 +665,54 @@ kdbg_iop_list_callback(kd_iop_t* iop, kd_callback_type type, void* arg) } } +static lck_grp_t *kdebug_lck_grp = NULL; + static void -kdbg_set_tracing_enabled(boolean_t enabled, uint32_t trace_type) +kdbg_set_tracing_enabled(bool enabled, uint32_t trace_type) { - int s = ml_set_interrupts_enabled(FALSE); - lck_spin_lock(kds_spin_lock); + /* + * Drain any events from IOPs before making the state change. On + * enabling, this removes any stale events from before tracing. On + * disabling, this saves any events up to the point tracing is disabled. + */ + kdbg_iop_list_callback(kd_ctrl_page.kdebug_iops, KD_CALLBACK_SYNC_FLUSH, + NULL); + + int s = ml_set_interrupts_enabled(false); + lck_spin_lock_grp(kds_spin_lock, kdebug_lck_grp); + if (enabled) { + /* + * The oldest valid time is now; reject past events from IOPs. + */ + kd_ctrl_page.oldest_time = kdbg_timestamp(); kdebug_enable |= trace_type; kd_ctrl_page.kdebug_slowcheck &= ~SLOW_NOLOG; kd_ctrl_page.enabled = 1; - commpage_update_kdebug_enable(); + commpage_update_kdebug_state(); } else { - kdebug_enable &= ~(KDEBUG_ENABLE_TRACE|KDEBUG_ENABLE_PPT); + kdebug_enable &= ~(KDEBUG_ENABLE_TRACE | KDEBUG_ENABLE_PPT); kd_ctrl_page.kdebug_slowcheck |= SLOW_NOLOG; kd_ctrl_page.enabled = 0; - commpage_update_kdebug_enable(); + commpage_update_kdebug_state(); } 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); + 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); + kdbg_iop_list_callback(kd_ctrl_page.kdebug_iops, + KD_CALLBACK_KDEBUG_DISABLED, NULL); } } static void -kdbg_set_flags(int slowflag, int enableflag, boolean_t enabled) +kdbg_set_flags(int slowflag, int enableflag, bool enabled) { - int s = ml_set_interrupts_enabled(FALSE); - lck_spin_lock(kds_spin_lock); + int s = ml_set_interrupts_enabled(false); + lck_spin_lock_grp(kds_spin_lock, kdebug_lck_grp); if (enabled) { kd_ctrl_page.kdebug_slowcheck |= slowflag; @@ -462,53 +721,58 @@ 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); } -void +/* + * Disable wrapping and return true if trace wrapped, false otherwise. + */ +static bool disable_wrap(uint32_t *old_slowcheck, uint32_t *old_flags) { - int s = ml_set_interrupts_enabled(FALSE); - lck_spin_lock(kds_spin_lock); + bool wrapped; + int s = ml_set_interrupts_enabled(false); + lck_spin_lock_grp(kds_spin_lock, kdebug_lck_grp); *old_slowcheck = kd_ctrl_page.kdebug_slowcheck; *old_flags = kd_ctrl_page.kdebug_flags; + wrapped = kd_ctrl_page.kdebug_flags & KDBG_WRAPPED; kd_ctrl_page.kdebug_flags &= ~KDBG_WRAPPED; kd_ctrl_page.kdebug_flags |= KDBG_NOWRAP; lck_spin_unlock(kds_spin_lock); ml_set_interrupts_enabled(s); + + return wrapped; } -void -enable_wrap(uint32_t old_slowcheck, boolean_t lostevents) +static void +enable_wrap(uint32_t old_slowcheck) { - int s = ml_set_interrupts_enabled(FALSE); - lck_spin_lock(kds_spin_lock); + int s = ml_set_interrupts_enabled(false); + lck_spin_lock_grp(kds_spin_lock, kdebug_lck_grp); kd_ctrl_page.kdebug_flags &= ~KDBG_NOWRAP; - if ( !(old_slowcheck & SLOW_NOLOG)) + if (!(old_slowcheck & SLOW_NOLOG)) { kd_ctrl_page.kdebug_slowcheck &= ~SLOW_NOLOG; - - if (lostevents == TRUE) - kd_ctrl_page.kdebug_flags |= KDBG_WRAPPED; + } lck_spin_unlock(kds_spin_lock); ml_set_interrupts_enabled(s); } static int -create_buffers(boolean_t early_trace) +create_buffers(bool early_trace) { - int i; - int p_buffer_size; - int f_buffer_size; - int f_buffers; - int error = 0; + unsigned int i; + unsigned int p_buffer_size; + unsigned int f_buffer_size; + unsigned int f_buffers; + int error = 0; /* * For the duration of this allocation, trace code will only reference @@ -519,6 +783,7 @@ create_buffers(boolean_t early_trace) */ kd_ctrl_page.kdebug_iops = kd_iops; + assert(kdbg_iop_list_is_valid(kd_ctrl_page.kdebug_iops)); /* * If the list is valid, it is sorted, newest -> oldest. Each iop entry @@ -533,10 +798,11 @@ create_buffers(boolean_t early_trace) goto out; } - if (nkdbufs < (kd_ctrl_page.kdebug_cpus * EVENTS_PER_STORAGE_UNIT * 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 + } else { n_storage_units = nkdbufs / EVENTS_PER_STORAGE_UNIT; + } nkdbufs = n_storage_units * EVENTS_PER_STORAGE_UNIT; @@ -546,13 +812,14 @@ create_buffers(boolean_t early_trace) f_buffer_size = N_STORAGE_UNITS_PER_BUFFER * sizeof(struct kd_storage); p_buffer_size = (n_storage_units % N_STORAGE_UNITS_PER_BUFFER) * sizeof(struct kd_storage); - if (p_buffer_size) + if (p_buffer_size) { n_storage_buffers++; + } kd_bufs = NULL; if (kdcopybuf == 0) { - if (kmem_alloc(kernel_map, (vm_offset_t *)&kdcopybuf, (vm_size_t)KDCOPYBUF_SIZE, VM_KERN_MEMORY_DIAG) != 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; } @@ -585,13 +852,15 @@ create_buffers(boolean_t early_trace) for (i = 0; i < n_storage_buffers; i++) { struct kd_storage *kds; - int n_elements; - int n; + uint16_t n_elements; + static_assert(N_STORAGE_UNITS_PER_BUFFER <= UINT16_MAX); + assert(kd_bufs[i].kdsb_size <= N_STORAGE_UNITS_PER_BUFFER * + sizeof(struct kd_storage)); n_elements = kd_bufs[i].kdsb_size / sizeof(struct kd_storage); kds = kd_bufs[i].kdsb_addr; - for (n = 0; n < n_elements; n++) { + for (uint16_t 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; @@ -603,29 +872,30 @@ create_buffers(boolean_t early_trace) bzero((char *)kdbip, sizeof(struct kd_bufinfo) * kd_ctrl_page.kdebug_cpus); - for (i = 0; i < (int)kd_ctrl_page.kdebug_cpus; i++) { + for (i = 0; i < 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].kd_lostevents = false; kdbip[i].num_bufs = 0; } - + kd_ctrl_page.kdebug_flags |= KDBG_BUFINIT; kd_ctrl_page.kds_inuse_count = 0; n_storage_threshold = n_storage_units / 2; out: - if (error) + if (error) { delete_buffers(); + } - return(error); + return error; } static void delete_buffers(void) { - int i; - + unsigned int i; + if (kd_bufs) { for (i = 0; i < n_storage_buffers; i++) { if (kd_bufs[i].kdsb_addr) { @@ -646,10 +916,10 @@ delete_buffers(void) if (kdbip) { 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_iops = NULL; kd_ctrl_page.kdebug_cpus = 0; kd_ctrl_page.kdebug_flags &= ~KDBG_BUFINIT; } @@ -658,14 +928,14 @@ void release_storage_unit(int cpu, uint32_t kdsp_raw) { int s = 0; - struct kd_storage *kdsp_actual; + struct kd_storage *kdsp_actual; struct kd_bufinfo *kdbp; union kds_ptr kdsp; kdsp.raw = kdsp_raw; - s = ml_set_interrupts_enabled(FALSE); - lck_spin_lock(kds_spin_lock); + s = ml_set_interrupts_enabled(false); + lck_spin_lock_grp(kds_spin_lock, kdebug_lck_grp); kdbp = &kdbip[cpu]; @@ -674,7 +944,7 @@ release_storage_unit(int cpu, uint32_t kdsp_raw) * it's possible for the storage unit pointed to * by kdsp to have already been stolen... so * check to see if it's still the head of the list - * now that we're behind the lock that protects + * now that we're behind the lock that protects * adding and removing from the queue... * since we only ever release and steal units from * that position, if it's no longer the head @@ -692,19 +962,18 @@ release_storage_unit(int cpu, uint32_t kdsp_raw) ml_set_interrupts_enabled(s); } - -boolean_t +bool allocate_storage_unit(int cpu) { - union kds_ptr kdsp; - struct kd_storage *kdsp_actual, *kdsp_next_actual; - struct kd_bufinfo *kdbp, *kdbp_vict, *kdbp_try; - uint64_t oldest_ts, ts; - boolean_t retval = TRUE; - int s = 0; - - s = ml_set_interrupts_enabled(FALSE); - lck_spin_lock(kds_spin_lock); + union kds_ptr kdsp; + struct kd_storage *kdsp_actual, *kdsp_next_actual; + struct kd_bufinfo *kdbp, *kdbp_vict, *kdbp_try; + uint64_t oldest_ts, ts; + bool retval = true; + int s = 0; + + s = ml_set_interrupts_enabled(false); + lck_spin_lock_grp(kds_spin_lock, kdebug_lck_grp); kdbp = &kdbip[cpu]; @@ -712,27 +981,34 @@ allocate_storage_unit(int cpu) if (kdbp->kd_list_tail.raw != KDS_PTR_NULL) { kdsp_actual = POINTER_FROM_KDS_PTR(kdbp->kd_list_tail); - if (kdsp_actual->kds_bufindx < EVENTS_PER_STORAGE_UNIT) + if (kdsp_actual->kds_bufindx < EVENTS_PER_STORAGE_UNIT) { goto out; + } } - + if ((kdsp = kd_ctrl_page.kds_free_list).raw != KDS_PTR_NULL) { + /* + * If there's a free page, grab it from the free list. + */ kdsp_actual = POINTER_FROM_KDS_PTR(kdsp); kd_ctrl_page.kds_free_list = kdsp_actual->kds_next; kd_ctrl_page.kds_inuse_count++; } else { + /* + * Otherwise, we're going to lose events and repurpose the oldest + * storage unit we can find. + */ if (kd_ctrl_page.kdebug_flags & KDBG_NOWRAP) { kd_ctrl_page.kdebug_slowcheck |= SLOW_NOLOG; - kdbp->kd_lostevents = TRUE; - retval = FALSE; + kdbp->kd_lostevents = true; + retval = false; goto out; } kdbp_vict = NULL; - oldest_ts = (uint64_t)-1; + oldest_ts = UINT64_MAX; 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) { /* * no storage unit to steal @@ -751,14 +1027,17 @@ allocate_storage_unit(int cpu) */ continue; } - ts = kdbg_get_timestamp(&kdsp_actual->kds_records[0]); + /* + * When wrapping, steal the storage unit with the + * earliest timestamp on its last event, instead of the + * earliest timestamp on the first event. This allows a + * storage unit with more recent events to be preserved, + * even if the storage unit contains events that are + * older than those found in other CPUs. + */ + ts = kdbg_get_timestamp(&kdsp_actual->kds_records[EVENTS_PER_STORAGE_UNIT - 1]); if (ts < oldest_ts) { - /* - * when 'wrapping', we want to steal the - * storage unit that has the 'earliest' time - * associated with it (first event time) - */ oldest_ts = ts; kdbp_vict = kdbp_try; } @@ -766,8 +1045,8 @@ allocate_storage_unit(int cpu) if (kdbp_vict == NULL) { kdebug_enable = 0; kd_ctrl_page.enabled = 0; - commpage_update_kdebug_enable(); - retval = FALSE; + commpage_update_kdebug_state(); + retval = false; goto out; } kdsp = kdbp_vict->kd_list_head; @@ -776,31 +1055,36 @@ allocate_storage_unit(int cpu) if (kdbp_vict->kd_list_head.raw != KDS_PTR_NULL) { kdsp_next_actual = POINTER_FROM_KDS_PTR(kdbp_vict->kd_list_head); - kdsp_next_actual->kds_lostevents = TRUE; - } else - kdbp_vict->kd_lostevents = TRUE; + kdsp_next_actual->kds_lostevents = true; + } else { + kdbp_vict->kd_lostevents = true; + } + if (kd_ctrl_page.oldest_time < oldest_ts) { + kd_ctrl_page.oldest_time = oldest_ts; + } kd_ctrl_page.kdebug_flags |= KDBG_WRAPPED; } - kdsp_actual->kds_timestamp = mach_absolute_time(); + kdsp_actual->kds_timestamp = kdbg_timestamp(); kdsp_actual->kds_next.raw = KDS_PTR_NULL; - kdsp_actual->kds_bufcnt = 0; + kdsp_actual->kds_bufcnt = 0; kdsp_actual->kds_readlast = 0; kdsp_actual->kds_lostevents = kdbp->kd_lostevents; - kdbp->kd_lostevents = FALSE; - kdsp_actual->kds_bufindx = 0; + kdbp->kd_lostevents = false; + kdsp_actual->kds_bufindx = 0; - if (kdbp->kd_list_head.raw == KDS_PTR_NULL) + if (kdbp->kd_list_head.raw == KDS_PTR_NULL) { kdbp->kd_list_head = kdsp; - else + } else { POINTER_FROM_KDS_PTR(kdbp->kd_list_tail)->kds_next = kdsp; + } kdbp->kd_list_tail = kdsp; out: lck_spin_unlock(kds_spin_lock); ml_set_interrupts_enabled(s); - return (retval); + return retval; } int @@ -809,32 +1093,34 @@ 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) + if (callback.iop_name[length] > 0x20 && callback.iop_name[length] < 0x7F) { continue; + } if (callback.iop_name[length] == 0) { - if (length) - is_valid_name = TRUE; + 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 @@ -845,7 +1131,7 @@ kernel_debug_register_callback(kd_callback_t callback) * 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); + iop->cpu_id = iop->next ? (iop->next->cpu_id + 1) : kdbg_cpu_count(false); /* * Header says OSCompareAndSwapPtr has a memory barrier @@ -860,84 +1146,74 @@ kernel_debug_register_callback(kd_callback_t callback) void 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 threadid + uint32_t coreid, + uint32_t debugid, + uint64_t timestamp, + uintptr_t arg1, + uintptr_t arg2, + uintptr_t arg3, + uintptr_t arg4, + uintptr_t threadid ) { - uint32_t bindx; - kd_buf *kd; + 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))) + 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; + if (typefilter_is_debugid_allowed(kdbg_typefilter, debugid)) { + goto record_event; } - enable_preemption(); goto out1; - } - else if (kd_ctrl_page.kdebug_flags & KDBG_RANGECHECK) { - if (debugid >= kdlog_beg && debugid <= kdlog_end) + } 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) { + } 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) + (debugid & KDBG_EVENTID_MASK) != kdlog_value2 && + (debugid & KDBG_EVENTID_MASK) != kdlog_value3 && + (debugid & KDBG_EVENTID_MASK) != kdlog_value4) { goto out1; + } } } - + record_event: + if (timestamp < kd_ctrl_page.oldest_time) { + goto out1; + } disable_preemption(); -record_event_preempt_disabled: - if (kd_ctrl_page.enabled == 0) + 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 + } else { kdsp_actual = NULL; - + bindx = EVENTS_PER_STORAGE_UNIT; + } + if (kdsp_actual == NULL || bindx >= EVENTS_PER_STORAGE_UNIT) { - if (allocate_storage_unit(coreid) == FALSE) { + if (allocate_storage_unit(coreid) == false) { /* * this can only happen if wrapping * has been disabled @@ -946,12 +1222,14 @@ retry_q: } goto retry_q; } - if ( !OSCompareAndSwap(bindx, bindx + 1, &kdsp_actual->kds_bufindx)) + 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) + if (timestamp < kdsp_actual->kds_timestamp) { kdsp_actual->kds_timestamp = timestamp; + } kd = &kdsp_actual->kds_records[bindx]; @@ -961,7 +1239,7 @@ retry_q: kd->arg3 = arg3; kd->arg4 = arg4; kd->arg5 = threadid; - + kdbg_set_timestamp_and_cpu(kd, timestamp, coreid); OSAddAtomic(1, &kdsp_actual->kds_bufcnt); @@ -969,170 +1247,145 @@ out: enable_preemption(); out1: if ((kds_waiter && kd_ctrl_page.kds_inuse_count >= n_storage_threshold)) { - boolean_t need_kds_wakeup = FALSE; - int s; + kdbg_wakeup(); + } +} - /* - * 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); +/* + * Check if the given debug ID is allowed to be traced on the current process. + * + * Returns true if allowed and false otherwise. + */ +static inline bool +kdebug_debugid_procfilt_allowed(uint32_t debugid) +{ + uint32_t procfilt_flags = kd_ctrl_page.kdebug_flags & + (KDBG_PIDCHECK | KDBG_PIDEXCLUDE); - if (lck_spin_try_lock(kdw_spin_lock)) { + if (!procfilt_flags) { + return true; + } - 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); - } + /* + * DBG_TRACE and MACH_SCHED tracepoints ignore the process filter. + */ + if ((debugid & 0xffff0000) == MACHDBG_CODE(DBG_MACH_SCHED, 0) || + (debugid >> 24 == DBG_TRACE)) { + return true; + } - ml_set_interrupts_enabled(s); + struct proc *curproc = current_proc(); + /* + * If the process is missing (early in boot), allow it. + */ + if (!curproc) { + return true; + } - if (need_kds_wakeup == TRUE) - wakeup(&kds_waiter); + if (procfilt_flags & KDBG_PIDCHECK) { + /* + * Allow only processes marked with the kdebug bit. + */ + return curproc->p_kdebug; + } else if (procfilt_flags & KDBG_PIDEXCLUDE) { + /* + * Exclude any process marked with the kdebug bit. + */ + return !curproc->p_kdebug; + } else { + panic("kdebug: invalid procfilt flags %x", kd_ctrl_page.kdebug_flags); + __builtin_unreachable(); } } - - static void kernel_debug_internal( - uint32_t debugid, - uintptr_t arg1, - uintptr_t arg2, - uintptr_t arg3, - uintptr_t arg4, - uintptr_t arg5) + uint32_t debugid, + uintptr_t arg1, + uintptr_t arg2, + uintptr_t arg3, + uintptr_t arg4, + uintptr_t arg5, + uint64_t flags) { - struct proc *curproc; - uint64_t now; - uint32_t bindx; - boolean_t s; - kd_buf *kd; - int cpu; + uint64_t now; + uint32_t bindx; + kd_buf *kd; + int cpu; struct kd_bufinfo *kdbp; struct kd_storage *kdsp_actual; - union kds_ptr kds_raw; - - + union kds_ptr kds_raw; + bool only_filter = flags & KDBG_FLAG_FILTERED; + bool observe_procfilt = !(flags & KDBG_FLAG_NOPROCFILT); if (kd_ctrl_page.kdebug_slowcheck) { - - if (kdebug_enable & KDEBUG_ENABLE_CHUD) { - kd_chudhook_fn chudhook; - /* - * Mask interrupts to minimize the interval across - * which the driver providing the hook could be - * unloaded. - */ - s = ml_set_interrupts_enabled(FALSE); - chudhook = kdebug_chudhook; - if (chudhook) - chudhook(debugid, arg1, arg2, arg3, arg4, arg5); - ml_set_interrupts_enabled(s); - } - if ( (kd_ctrl_page.kdebug_slowcheck & SLOW_NOLOG) || !(kdebug_enable & (KDEBUG_ENABLE_TRACE|KDEBUG_ENABLE_PPT))) + if ((kd_ctrl_page.kdebug_slowcheck & SLOW_NOLOG) || + !(kdebug_enable & (KDEBUG_ENABLE_TRACE | KDEBUG_ENABLE_PPT))) { goto out1; - - if ( !ml_at_interrupt_context()) { - if (kd_ctrl_page.kdebug_flags & KDBG_PIDCHECK) { - /* - * If kdebug flag is not set for current proc, return - */ - curproc = current_proc(); - - if ((curproc && !(curproc->p_kdebug)) && - ((debugid & 0xffff0000) != (MACHDBG_CODE(DBG_MACH_SCHED, 0) | DBG_FUNC_NONE)) && - (debugid >> 24 != DBG_TRACE)) - goto out1; - } - else if (kd_ctrl_page.kdebug_flags & KDBG_PIDEXCLUDE) { - /* - * If kdebug flag is set for current proc, return - */ - curproc = current_proc(); + } - if ((curproc && curproc->p_kdebug) && - ((debugid & 0xffff0000) != (MACHDBG_CODE(DBG_MACH_SCHED, 0) | DBG_FUNC_NONE)) && - (debugid >> 24 != DBG_TRACE)) - goto out1; - } + if (!ml_at_interrupt_context() && observe_procfilt && + !kdebug_debugid_procfilt_allowed(debugid)) { + goto out1; } if (kd_ctrl_page.kdebug_flags & KDBG_TYPEFILTER_CHECK) { - /* Always record trace system info */ - if (KDBG_EXTRACT_CLASS(debugid) == DBG_TRACE) + if (typefilter_is_debugid_allowed(kdbg_typefilter, 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) { + } else if (only_filter) { + goto out1; + } else if (kd_ctrl_page.kdebug_flags & KDBG_RANGECHECK) { /* Always record trace system info */ - if (KDBG_EXTRACT_CLASS(debugid) == DBG_TRACE) + if (KDBG_EXTRACT_CLASS(debugid) == DBG_TRACE) { goto record_event; - - if (debugid < kdlog_beg || debugid > kdlog_end) + } + + if (debugid < kdlog_beg || debugid > kdlog_end) { goto out1; - } - else if (kd_ctrl_page.kdebug_flags & KDBG_VALCHECK) { + } + } else if (kd_ctrl_page.kdebug_flags & KDBG_VALCHECK) { /* Always record trace system info */ - if (KDBG_EXTRACT_CLASS(debugid) == DBG_TRACE) + if (KDBG_EXTRACT_CLASS(debugid) == DBG_TRACE) { goto record_event; - - if ((debugid & KDBG_EVENTID_MASK) != kdlog_value1 && + } + + 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) + (debugid & KDBG_EVENTID_MASK) != kdlog_value4) { goto out1; + } } + } else if (only_filter) { + goto out1; } + record_event: disable_preemption(); -record_event_preempt_disabled: - if (kd_ctrl_page.enabled == 0) + 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; if (kds_raw.raw != KDS_PTR_NULL) { kdsp_actual = POINTER_FROM_KDS_PTR(kds_raw); bindx = kdsp_actual->kds_bufindx; - } else + } else { kdsp_actual = NULL; - + bindx = EVENTS_PER_STORAGE_UNIT; + } + if (kdsp_actual == NULL || bindx >= EVENTS_PER_STORAGE_UNIT) { - if (allocate_storage_unit(cpu) == FALSE) { + if (allocate_storage_unit(cpu) == false) { /* * this can only happen if wrapping * has been disabled @@ -1141,10 +1394,12 @@ retry_q: } goto retry_q; } - now = mach_absolute_time() & KDBG_TIMESTAMP_MASK; - if ( !OSCompareAndSwap(bindx, bindx + 1, &kdsp_actual->kds_bufindx)) + now = kdbg_timestamp() & KDBG_TIMESTAMP_MASK; + + if (!OSCompareAndSwap(bindx, bindx + 1, &kdsp_actual->kds_bufindx)) { goto retry_q; + } kd = &kdsp_actual->kds_records[bindx]; @@ -1154,113 +1409,175 @@ retry_q: kd->arg3 = arg3; kd->arg4 = arg4; kd->arg5 = arg5; - + kdbg_set_timestamp_and_cpu(kd, now, cpu); OSAddAtomic(1, &kdsp_actual->kds_bufcnt); + +#if KPERF + kperf_kdebug_callback(debugid, __builtin_frame_address(0)); +#endif out: enable_preemption(); out1: if (kds_waiter && kd_ctrl_page.kds_inuse_count >= n_storage_threshold) { - uint32_t etype; - uint32_t stype; - + uint32_t etype; + uint32_t stype; + 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; - - /* - * 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 - * one of our 2 wait conditions... 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); + kdbg_wakeup(); } } } +__attribute__((noinline)) void kernel_debug( - uint32_t debugid, - uintptr_t arg1, - uintptr_t arg2, - uintptr_t arg3, - uintptr_t arg4, + uint32_t debugid, + uintptr_t arg1, + uintptr_t arg2, + uintptr_t arg3, + uintptr_t arg4, __unused uintptr_t arg5) { - kernel_debug_internal(debugid, arg1, arg2, arg3, arg4, (uintptr_t)thread_tid(current_thread())); + kernel_debug_internal(debugid, arg1, arg2, arg3, arg4, + (uintptr_t)thread_tid(current_thread()), 0); } +__attribute__((noinline)) void kernel_debug1( - uint32_t debugid, - uintptr_t arg1, - uintptr_t arg2, - uintptr_t arg3, - uintptr_t arg4, - uintptr_t arg5) + uint32_t debugid, + uintptr_t arg1, + uintptr_t arg2, + uintptr_t arg3, + uintptr_t arg4, + uintptr_t arg5) +{ + kernel_debug_internal(debugid, arg1, arg2, arg3, arg4, arg5, 0); +} + +__attribute__((noinline)) +void +kernel_debug_flags( + uint32_t debugid, + uintptr_t arg1, + uintptr_t arg2, + uintptr_t arg3, + uintptr_t arg4, + uint64_t flags) +{ + kernel_debug_internal(debugid, arg1, arg2, arg3, arg4, + (uintptr_t)thread_tid(current_thread()), flags); +} + +__attribute__((noinline)) +void +kernel_debug_filtered( + uint32_t debugid, + uintptr_t arg1, + uintptr_t arg2, + uintptr_t arg3, + uintptr_t arg4) { - kernel_debug_internal(debugid, arg1, arg2, arg3, arg4, arg5); + kernel_debug_flags(debugid, arg1, arg2, arg3, arg4, KDBG_FLAG_FILTERED); } void -kernel_debug_string_simple(const char *message) +kernel_debug_string_early(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))); + 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 */ +#define SIMPLE_STR_LEN (64) +static_assert(SIMPLE_STR_LEN % sizeof(uintptr_t) == 0); + +void +kernel_debug_string_simple(uint32_t eventid, const char *str) +{ + if (!kdebug_enable) { + return; + } + + /* array of uintptr_ts simplifies emitting the string as arguments */ + uintptr_t str_buf[(SIMPLE_STR_LEN / sizeof(uintptr_t)) + 1] = { 0 }; + size_t len = strlcpy((char *)str_buf, str, SIMPLE_STR_LEN + 1); + + uintptr_t thread_id = (uintptr_t)thread_tid(current_thread()); + uint32_t debugid = eventid | DBG_FUNC_START; + + /* string can fit in a single tracepoint */ + if (len <= (4 * sizeof(uintptr_t))) { + debugid |= DBG_FUNC_END; + } + + kernel_debug_internal(debugid, str_buf[0], + str_buf[1], + str_buf[2], + str_buf[3], thread_id, 0); + + debugid &= KDBG_EVENTID_MASK; + int i = 4; + size_t written = 4 * sizeof(uintptr_t); + + for (; written < len; i += 4, written += 4 * sizeof(uintptr_t)) { + /* if this is the last tracepoint to be emitted */ + if ((written + (4 * sizeof(uintptr_t))) >= len) { + debugid |= DBG_FUNC_END; + } + kernel_debug_internal(debugid, str_buf[i], + str_buf[i + 1], + str_buf[i + 2], + str_buf[i + 3], thread_id, 0); + } +} + +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) + 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); +#if defined(__x86_64__) + extern int early_boot; + /* + * Note that "early" isn't early enough in some cases where + * we're invoked before gsbase is set on x86, hence the + * check of "early_boot". + */ + if (early_boot) { + return; + } +#endif + + /* If early tracing is over, use the normal path. */ + if (kd_early_done) { + KDBG_RELEASE(debugid, arg1, arg2, arg3, arg4); 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) + /* Do nothing if the buffer is full or we're not on the boot cpu. */ + kd_early_overflow = kd_early_index >= KD_EARLY_BUFFER_NBUFS; + 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(); @@ -1273,47 +1590,54 @@ kernel_debug_early( } /* - * Transfen the contents of the temporary buffer into the trace buffers. + * Transfer 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) + if (cpu_number() != master_cpu) { panic("kernel_debug_early_end() not call on boot processor"); + } + + /* reset the current oldest time to allow early events */ + kd_ctrl_page.oldest_time = 0; +#if defined(__x86_64__) /* 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); + kernel_debug_enter(0, TRACE_TIMESTAMPS, 0, + (uint32_t)(tsc_rebase_abs_time >> 32), (uint32_t)tsc_rebase_abs_time, + tsc_at_boot, 0, 0); +#endif /* defined(__x86_64__) */ + for (unsigned int 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); + if (kd_early_overflow) { + KDBG_RELEASE(TRACE_LOST_EVENTS, 1); + } + + kd_early_done = true; /* This trace marks the start of kernel tracing */ - kernel_debug_string_simple("early trace done"); + kernel_debug_string_early("early trace done"); +} + +void +kernel_debug_disable(void) +{ + if (kdebug_enable) { + kdbg_set_tracing_enabled(false, 0); + } } /* @@ -1326,13 +1650,92 @@ kdebug_validate_debugid(uint32_t debugid) debugid_class = KDBG_EXTRACT_CLASS(debugid); switch (debugid_class) { - case DBG_TRACE: - return EPERM; + case DBG_TRACE: + return EPERM; } return 0; } +/* + * Support syscall SYS_kdebug_typefilter. + */ +int +kdebug_typefilter(__unused struct proc* p, + struct kdebug_typefilter_args* uap, + __unused int *retval) +{ + int ret = KERN_SUCCESS; + + if (uap->addr == USER_ADDR_NULL || + uap->size == USER_ADDR_NULL) { + return EINVAL; + } + + /* + * The atomic load is to close a race window with setting the typefilter + * and memory entry values. A description follows: + * + * Thread 1 (writer) + * + * Allocate Typefilter + * Allocate MemoryEntry + * Write Global MemoryEntry Ptr + * Atomic Store (Release) Global Typefilter Ptr + * + * Thread 2 (reader, AKA us) + * + * if ((Atomic Load (Acquire) Global Typefilter Ptr) == NULL) + * return; + * + * Without the atomic store, it isn't guaranteed that the write of + * Global MemoryEntry Ptr is visible before we can see the write of + * Global Typefilter Ptr. + * + * Without the atomic load, it isn't guaranteed that the loads of + * Global MemoryEntry Ptr aren't speculated. + * + * The global pointers transition from NULL -> valid once and only once, + * and never change after becoming valid. This means that having passed + * the first atomic load test of Global Typefilter Ptr, this function + * can then safely use the remaining global state without atomic checks. + */ + if (!os_atomic_load(&kdbg_typefilter, acquire)) { + return EINVAL; + } + + assert(kdbg_typefilter_memory_entry); + + mach_vm_offset_t user_addr = 0; + vm_map_t user_map = current_map(); + + ret = mach_to_bsd_errno( + mach_vm_map_kernel(user_map, // target map + &user_addr, // [in, out] target address + TYPEFILTER_ALLOC_SIZE, // initial size + 0, // mask (alignment?) + VM_FLAGS_ANYWHERE, // flags + VM_MAP_KERNEL_FLAGS_NONE, + VM_KERN_MEMORY_NONE, + kdbg_typefilter_memory_entry, // port (memory entry!) + 0, // offset (in memory entry) + false, // should copy + VM_PROT_READ, // cur_prot + VM_PROT_READ, // max_prot + VM_INHERIT_SHARE)); // inherit behavior on fork + + if (ret == KERN_SUCCESS) { + vm_size_t user_ptr_size = vm_map_is_64bit(user_map) ? 8 : 4; + ret = copyout(CAST_DOWN(void *, &user_addr), uap->addr, user_ptr_size ); + + if (ret != KERN_SUCCESS) { + mach_vm_deallocate(user_map, user_addr, TYPEFILTER_ALLOC_SIZE); + } + } + + return ret; +} + /* * Support syscall SYS_kdebug_trace. U64->K32 args may get truncated in kdebug_trace64 */ @@ -1351,22 +1754,31 @@ kdebug_trace(struct proc *p, struct kdebug_trace_args *uap, int32_t *retval) } /* - * Support syscall SYS_kdebug_trace64. 64-bit args on K32 will get truncated to fit in 32-bit record format. + * Support syscall SYS_kdebug_trace64. 64-bit args on K32 will get truncated + * to fit in 32-bit record format. + * + * It is intentional that error conditions are not checked until kdebug is + * enabled. This is to match the userspace wrapper behavior, which is optimizing + * for non-error case performance. */ -int kdebug_trace64(__unused struct proc *p, struct kdebug_trace64_args *uap, __unused int32_t *retval) +int +kdebug_trace64(__unused struct proc *p, struct kdebug_trace64_args *uap, __unused int32_t *retval) { int err; + if (__probable(kdebug_enable == 0)) { + return 0; + } + if ((err = kdebug_validate_debugid(uap->code)) != 0) { return err; } - if ( __probable(kdebug_enable == 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())); + 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()), 0); - return(0); + return 0; } /* @@ -1395,7 +1807,7 @@ int kdebug_trace64(__unused struct proc *p, struct kdebug_trace64_args *uap, __u */ static uint64_t kernel_debug_string_internal(uint32_t debugid, uint64_t str_id, void *vstr, - size_t str_len) + size_t str_len) { /* str must be word-aligned */ uintptr_t *str = vstr; @@ -1403,15 +1815,14 @@ kernel_debug_string_internal(uint32_t debugid, uint64_t str_id, void *vstr, uintptr_t thread_id; int i; uint32_t trace_debugid = TRACEDBG_CODE(DBG_TRACE_STRING, - TRACE_STRING_GLOBAL); + 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); + (uintptr_t)debugid, (uintptr_t)str_id, 0, 0, thread_id, 0); return str_id; } @@ -1427,9 +1838,8 @@ kernel_debug_string_internal(uint32_t debugid, uint64_t str_id, void *vstr, trace_debugid |= DBG_FUNC_END; } - kernel_debug_internal(trace_debugid, (uintptr_t)debugid, - (uintptr_t)str_id, str[0], - str[1], thread_id); + kernel_debug_internal(trace_debugid, (uintptr_t)debugid, (uintptr_t)str_id, + str[0], str[1], thread_id, 0); trace_debugid &= KDBG_EVENTID_MASK; i = 2; @@ -1440,9 +1850,9 @@ kernel_debug_string_internal(uint32_t debugid, uint64_t str_id, void *vstr, trace_debugid |= DBG_FUNC_END; } kernel_debug_internal(trace_debugid, str[i], - str[i + 1], - str[i + 2], - str[i + 3], thread_id); + str[i + 1], + str[i + 2], + str[i + 3], thread_id, 0); } return str_id; @@ -1453,19 +1863,18 @@ kernel_debug_string_internal(uint32_t debugid, uint64_t str_id, void *vstr, * Trace system and scheduling events circumvent this check, as do events * emitted in interrupt context. */ -static boolean_t +static bool kdebug_current_proc_enabled(uint32_t debugid) { /* can't determine current process in interrupt context */ if (ml_at_interrupt_context()) { - return TRUE; + 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; + (debugid & KDBG_CSC_MASK) == MACHDBG_CODE(DBG_MACH_SCHED, 0))) { + return true; } if (kd_ctrl_page.kdebug_flags & KDBG_PIDCHECK) { @@ -1473,71 +1882,58 @@ kdebug_current_proc_enabled(uint32_t debugid) /* only the process with the kdebug bit set is allowed */ if (cur_proc && !(cur_proc->p_kdebug)) { - return FALSE; + 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 false; } } - return TRUE; + 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 +bool kdebug_debugid_enabled(uint32_t debugid) { - boolean_t is_enabled = TRUE; - /* if no filtering is enabled */ if (!kd_ctrl_page.kdebug_slowcheck) { - return TRUE; + return true; } - if (KDBG_EXTRACT_CLASS(debugid) == DBG_TRACE) { - return TRUE; - } + return kdebug_debugid_explicitly_enabled(debugid); +} +bool +kdebug_debugid_explicitly_enabled(uint32_t debugid) +{ 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(); + return typefilter_is_debugid_allowed(kdbg_typefilter, debugid); + } else if (KDBG_EXTRACT_CLASS(debugid) == DBG_TRACE) { + return true; } else if (kd_ctrl_page.kdebug_flags & KDBG_RANGECHECK) { if (debugid < kdlog_beg || debugid > kdlog_end) { - is_enabled = FALSE; + return 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; + (debugid & KDBG_EVENTID_MASK) != kdlog_value2 && + (debugid & KDBG_EVENTID_MASK) != kdlog_value3 && + (debugid & KDBG_EVENTID_MASK) != kdlog_value4) { + return false; } } - return is_enabled; + return true; +} + +bool +kdebug_using_continuous_time(void) +{ + return kdebug_enable & KDEBUG_ENABLE_CONT_TIME; } /* @@ -1571,7 +1967,7 @@ 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); + static_assert(sizeof(str_buf) > MAX_STR_LEN); vm_size_t len_copied; int err; @@ -1605,7 +2001,7 @@ kernel_debug_string(uint32_t debugid, uint64_t *str_id, const char *str) 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); + len_copied); return 0; } @@ -1614,11 +2010,11 @@ kernel_debug_string(uint32_t debugid, uint64_t *str_id, const char *str) */ int kdebug_trace_string(__unused struct proc *p, - struct kdebug_trace_string_args *uap, - uint64_t *retval) + 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); + static_assert(sizeof(str_buf) > MAX_STR_LEN); size_t len_copied; int err; @@ -1644,7 +2040,7 @@ kdebug_trace_string(__unused struct proc *p, } *retval = kernel_debug_string_internal(uap->debugid, uap->str_id, - NULL, 0); + NULL, 0); return 0; } @@ -1666,49 +2062,41 @@ kdebug_trace_string(__unused struct proc *p, len_copied--; *retval = kernel_debug_string_internal(uap->debugid, uap->str_id, str_buf, - len_copied); + len_copied); return 0; } static void kdbg_lock_init(void) { - if (kd_ctrl_page.kdebug_flags & KDBG_LOCKINIT) + static lck_grp_attr_t *kdebug_lck_grp_attr = NULL; + static lck_attr_t *kdebug_lck_attr = NULL; + + if (kd_ctrl_page.kdebug_flags & KDBG_LOCKINIT) { return; - - /* - * allocate lock group attribute and group - */ - kd_trace_mtx_sysctl_grp_attr = lck_grp_attr_alloc_init(); - kd_trace_mtx_sysctl_grp = lck_grp_alloc_init("kdebug", kd_trace_mtx_sysctl_grp_attr); - - /* - * allocate the lock attribute - */ - kd_trace_mtx_sysctl_attr = lck_attr_alloc_init(); + } + assert(kdebug_lck_grp_attr == NULL); + kdebug_lck_grp_attr = lck_grp_attr_alloc_init(); + kdebug_lck_grp = lck_grp_alloc_init("kdebug", kdebug_lck_grp_attr); + kdebug_lck_attr = lck_attr_alloc_init(); - /* - * allocate and initialize mutex's - */ - kd_trace_mtx_sysctl = lck_mtx_alloc_init(kd_trace_mtx_sysctl_grp, kd_trace_mtx_sysctl_attr); - kds_spin_lock = lck_spin_alloc_init(kd_trace_mtx_sysctl_grp, kd_trace_mtx_sysctl_attr); - kdw_spin_lock = lck_spin_alloc_init(kd_trace_mtx_sysctl_grp, kd_trace_mtx_sysctl_attr); + kds_spin_lock = lck_spin_alloc_init(kdebug_lck_grp, kdebug_lck_attr); + kdw_spin_lock = lck_spin_alloc_init(kdebug_lck_grp, kdebug_lck_attr); kd_ctrl_page.kdebug_flags |= KDBG_LOCKINIT; } - int -kdbg_bootstrap(boolean_t early_trace) +kdbg_bootstrap(bool early_trace) { - kd_ctrl_page.kdebug_flags &= ~KDBG_WRAPPED; + kd_ctrl_page.kdebug_flags &= ~KDBG_WRAPPED; - return (create_buffers(early_trace)); + return create_buffers(early_trace); } int -kdbg_reinit(boolean_t early_trace) +kdbg_reinit(bool early_trace) { int ret = 0; @@ -1717,7 +2105,7 @@ kdbg_reinit(boolean_t early_trace) * First make sure we're not in * the middle of cutting a trace */ - kdbg_set_tracing_enabled(FALSE, KDEBUG_ENABLE_TRACE); + kernel_debug_disable(); /* * make sure the SLOW_NOLOG is seen @@ -1728,38 +2116,36 @@ kdbg_reinit(boolean_t early_trace) delete_buffers(); - 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 = NULL; - kd_mapcount = 0; - } + kdbg_clear_thread_map(); ret = kdbg_bootstrap(early_trace); RAW_file_offset = 0; RAW_file_written = 0; - return(ret); + return ret; } void -kdbg_trace_data(struct proc *proc, long *arg_pid) +kdbg_trace_data(struct proc *proc, long *arg_pid, long *arg_uniqueid) { - if (!proc) + if (!proc) { *arg_pid = 0; - else + *arg_uniqueid = 0; + } else { *arg_pid = proc->p_pid; + /* Fit in a trace point */ + *arg_uniqueid = (long)proc->p_uniqueid; + if ((uint64_t) *arg_uniqueid != proc->p_uniqueid) { + *arg_uniqueid = 0; + } + } } void -kdbg_trace_string(struct proc *proc, long *arg1, long *arg2, long *arg3, long *arg4) +kdbg_trace_string(struct proc *proc, long *arg1, long *arg2, long *arg3, + long *arg4) { - char *dbg_nameptr; - int dbg_namelen; - long dbg_parms[4]; - if (!proc) { *arg1 = 0; *arg2 = 0; @@ -1767,50 +2153,22 @@ kdbg_trace_string(struct proc *proc, long *arg1, long *arg2, long *arg3, long *a *arg4 = 0; return; } - /* - * Collect the pathname for tracing - */ - dbg_nameptr = proc->p_comm; - dbg_namelen = (int)strlen(proc->p_comm); - dbg_parms[0]=0L; - dbg_parms[1]=0L; - dbg_parms[2]=0L; - dbg_parms[3]=0L; - - if(dbg_namelen > (int)sizeof(dbg_parms)) - dbg_namelen = (int)sizeof(dbg_parms); - - strncpy((char *)dbg_parms, dbg_nameptr, dbg_namelen); - - *arg1=dbg_parms[0]; - *arg2=dbg_parms[1]; - *arg3=dbg_parms[2]; - *arg4=dbg_parms[3]; -} -static void -kdbg_resolve_map(thread_t th_act, void *opaque) -{ - kd_threadmap *mapptr; - krt_t *t = (krt_t *)opaque; - - if (t->count < t->maxcount) { - mapptr = &t->map[t->count]; - mapptr->thread = (uintptr_t)thread_tid(th_act); + const char *procname = proc_best_name(proc); + size_t namelen = strlen(procname); - (void) strlcpy (mapptr->command, t->atts->task_comm, - sizeof(t->atts->task_comm)); - /* - * Some kernel threads have no associated pid. - * We still need to mark the entry as valid. - */ - if (t->atts->pid) - mapptr->valid = t->atts->pid; - else - mapptr->valid = 1; + long args[4] = { 0 }; - t->count++; + if (namelen > sizeof(args)) { + namelen = sizeof(args); } + + strncpy((char *)args, procname, namelen); + + *arg1 = args[0]; + *arg2 = args[1]; + *arg3 = args[2]; + *arg4 = args[3]; } /* @@ -1831,7 +2189,6 @@ kdbg_resolve_map(thread_t th_act, void *opaque) * * 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 @@ -1850,11 +2207,12 @@ kdbg_cpumap_init_internal(kd_iop_t* iops, uint32_t cpu_count, uint8_t** cpumap, 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; } + bzero(*cpumap, *cpumap_size); } else if (bytes_available < bytes_needed) { return EINVAL; } @@ -1870,137 +2228,152 @@ kdbg_cpumap_init_internal(kd_iop_t* iops, uint32_t cpu_count, uint8_t** cpumap, 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_thrmap_init(void) { - if (kd_ctrl_page.kdebug_flags & KDBG_MAPINIT) + ktrace_assert_lock_held(); + + if (kd_ctrl_page.kdebug_flags & KDBG_MAPINIT) { return; + } kd_mapptr = kdbg_thrmap_init_internal(0, &kd_mapsize, &kd_mapcount); - if (kd_mapptr) + if (kd_mapptr) { kd_ctrl_page.kdebug_flags |= KDBG_MAPINIT; + } } +static void +kd_resolve_map(thread_t thread, void *opaque) +{ + struct kd_resolver *resolve = opaque; + + if (resolve->krs_count < resolve->krs_maxcount) { + kd_threadmap *map = &resolve->krs_map[resolve->krs_count]; + struct kd_task_name *task_name = resolve->krs_task; + map->thread = (uintptr_t)thread_tid(thread); + + (void)strlcpy(map->command, task_name->ktn_name, sizeof(map->command)); + /* + * Kernel threads should still be marked with non-zero valid bit. + */ + pid_t pid = resolve->krs_task->ktn_pid; + map->valid = pid == 0 ? 1 : pid; + resolve->krs_count++; + } +} -kd_threadmap* kdbg_thrmap_init_internal(unsigned int count, unsigned int *mapsize, unsigned int *mapcount) +static vm_size_t +kd_resolve_tasks(struct kd_task_name *task_names, vm_size_t ntasks) { - 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; - int i; - vm_offset_t kaddr; + vm_size_t i = 0; + proc_t p = PROC_NULL; - /* - * need to use PROC_SCANPROCLIST with proc_iterate - */ proc_list_lock(); - - /* - * Calculate the sizes of map buffers - */ - 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++; + ALLPROC_FOREACH(p) { + if (i >= ntasks) { + break; + } + /* + * Only record processes that can be referenced and are not exiting. + */ + if (p->task && (p->p_lflag & P_LEXIT) == 0) { + task_reference(p->task); + task_names[i].ktn_task = p->task; + task_names[i].ktn_pid = p->p_pid; + (void)strlcpy(task_names[i].ktn_name, proc_best_name(p), + sizeof(task_names[i].ktn_name)); + i++; + } } proc_list_unlock(); - /* - * 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 25%. - */ - *mapcount += *mapcount/4; - tts_count += tts_count/4; - - *mapsize = *mapcount * sizeof(kd_threadmap); + return i; +} - if (count && count < *mapcount) - return (0); +static vm_size_t +kd_resolve_threads(kd_threadmap *map, struct kd_task_name *task_names, + vm_size_t ntasks, vm_size_t nthreads) +{ + struct kd_resolver resolver = { + .krs_map = map, .krs_count = 0, .krs_maxcount = nthreads, + }; - 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 - return (0); + for (int i = 0; i < ntasks; i++) { + struct kd_task_name *cur_task = &task_names[i]; + resolver.krs_task = cur_task; + task_act_iterate_wth_args(cur_task->ktn_task, kd_resolve_map, + &resolver); + task_deallocate(cur_task->ktn_task); + } - tts_mapsize = tts_count * sizeof(struct tts); + return resolver.krs_count; +} - 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); +static kd_threadmap * +kdbg_thrmap_init_internal(size_t maxthreads, vm_size_t *mapsize, + vm_size_t *mapcount) +{ + kd_threadmap *thread_map = NULL; + struct kd_task_name *task_names; + vm_size_t names_size = 0; - return (0); - } - /* - * We need to save the procs command string - * and take a reference for each task associated - * with a valid process - */ + assert(mapsize != NULL); + assert(mapcount != NULL); - proc_list_lock(); + vm_size_t nthreads = threads_count; + vm_size_t ntasks = tasks_count; /* - * should use proc_iterate + * Allow 25% more threads and tasks to be created between now and taking the + * 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++; - } + if (os_add_overflow(nthreads, nthreads / 4, &nthreads) || + os_add_overflow(ntasks, ntasks / 4, &ntasks)) { + return NULL; } - tts_count = i; - proc_list_unlock(); + *mapcount = nthreads; + if (os_mul_overflow(nthreads, sizeof(kd_threadmap), mapsize)) { + return NULL; + } + if (os_mul_overflow(ntasks, sizeof(task_names[0]), &names_size)) { + return NULL; + } /* - * Initialize thread map data + * Wait until the out-parameters have been filled with the needed size to + * do the bounds checking on the provided maximum. */ - 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); + if (maxthreads != 0 && maxthreads < nthreads) { + return NULL; } - kmem_free(kernel_map, (vm_offset_t)tts_mapptr, tts_mapsize); - *mapcount = akrt.count; - - return (mapptr); + thread_map = kalloc_tag(*mapsize, VM_KERN_MEMORY_DIAG); + bzero(thread_map, *mapsize); + task_names = kheap_alloc(KHEAP_TEMP, names_size, Z_WAITOK | Z_ZERO); + ntasks = kd_resolve_tasks(task_names, ntasks); + *mapcount = kd_resolve_threads(thread_map, task_names, ntasks, nthreads); + kheap_free(KHEAP_TEMP, task_names, names_size); + return thread_map; } static void @@ -2011,7 +2384,7 @@ kdbg_clear(void) * First make sure we're not in * the middle of cutting a trace */ - kdbg_set_tracing_enabled(FALSE, KDEBUG_ENABLE_TRACE); + kernel_debug_disable(); kdbg_disable_typefilter(); /* @@ -2021,49 +2394,79 @@ kdbg_clear(void) */ IOSleep(100); - global_state_pid = -1; + /* reset kdebug state for each process */ + if (kd_ctrl_page.kdebug_flags & (KDBG_PIDCHECK | KDBG_PIDEXCLUDE)) { + proc_list_lock(); + proc_t p; + ALLPROC_FOREACH(p) { + p->p_kdebug = 0; + } + proc_list_unlock(); + } + 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_deallocate_typefilter(); + + kd_ctrl_page.oldest_time = 0; + delete_buffers(); - nkdbufs = 0; + nkdbufs = 0; /* Clean up the thread map buffer */ - kd_ctrl_page.kdebug_flags &= ~KDBG_MAPINIT; - if (kd_mapptr) { - kmem_free(kernel_map, (vm_offset_t)kd_mapptr, kd_mapsize); - kd_mapptr = (kd_threadmap *) 0; - } - kd_mapsize = 0; - kd_mapcount = 0; + kdbg_clear_thread_map(); RAW_file_offset = 0; RAW_file_written = 0; } +void +kdebug_reset(void) +{ + ktrace_assert_lock_held(); + + kdbg_lock_init(); + + kdbg_clear(); + if (kdbg_typefilter) { + typefilter_reject_all(kdbg_typefilter); + typefilter_allow_class(kdbg_typefilter, DBG_TRACE); + } +} + +void +kdebug_free_early_buf(void) +{ +#if defined(__x86_64__) + /* + * Make Intel aware that the early buffer is no longer being used. ARM + * handles this as part of the BOOTDATA segment. + */ + ml_static_mfree((vm_offset_t)&kd_early_buffer, sizeof(kd_early_buffer)); +#endif /* defined(__x86_64__) */ +} + int kdbg_setpid(kd_regtype *kdr) { pid_t pid; - int flag, ret=0; + int flag, ret = 0; struct proc *p; pid = (pid_t)kdr->value1; flag = (int)kdr->value2; - if (pid > 0) { - if ((p = proc_find(pid)) == NULL) + if (pid >= 0) { + if ((p = proc_find(pid)) == NULL) { ret = ESRCH; - else { + } else { if (flag == 1) { /* * turn on pid check for this and all pids */ kd_ctrl_page.kdebug_flags |= KDBG_PIDCHECK; kd_ctrl_page.kdebug_flags &= ~KDBG_PIDEXCLUDE; - kdbg_set_flags(SLOW_CHECKS, 0, TRUE); + kdbg_set_flags(SLOW_CHECKS, 0, true); p->p_kdebug = 1; } else { @@ -2072,16 +2475,16 @@ kdbg_setpid(kd_regtype *kdr) * Don't turn off all pid checking though * * kd_ctrl_page.kdebug_flags &= ~KDBG_PIDCHECK; - */ + */ p->p_kdebug = 0; } proc_rele(p); } - } - else + } else { ret = EINVAL; + } - return(ret); + return ret; } /* This is for pid exclusion in the trace buffer */ @@ -2089,135 +2492,180 @@ int kdbg_setpidex(kd_regtype *kdr) { pid_t pid; - int flag, ret=0; + int flag, ret = 0; struct proc *p; pid = (pid_t)kdr->value1; flag = (int)kdr->value2; - if (pid > 0) { - if ((p = proc_find(pid)) == NULL) + if (pid >= 0) { + if ((p = proc_find(pid)) == NULL) { ret = ESRCH; - else { + } else { if (flag == 1) { /* * turn on pid exclusion */ kd_ctrl_page.kdebug_flags |= KDBG_PIDEXCLUDE; kd_ctrl_page.kdebug_flags &= ~KDBG_PIDCHECK; - kdbg_set_flags(SLOW_CHECKS, 0, TRUE); + kdbg_set_flags(SLOW_CHECKS, 0, true); p->p_kdebug = 1; - } - else { + } else { /* * turn off pid exclusion for this pid value * Don't turn off all pid exclusion though * * kd_ctrl_page.kdebug_flags &= ~KDBG_PIDEXCLUDE; - */ + */ p->p_kdebug = 0; } proc_rele(p); } - } else + } else { ret = EINVAL; + } - return(ret); + return ret; } +/* + * The following functions all operate on the "global" typefilter singleton. + */ /* - * This is for setting a maximum decrementer value + * The tf param is optional, you may pass either a valid typefilter or NULL. + * If you pass a valid typefilter, you release ownership of that typefilter. */ -int -kdbg_setrtcdec(kd_regtype *kdr) +static int +kdbg_initialize_typefilter(typefilter_t tf) { - int ret = 0; - natural_t decval; + ktrace_assert_lock_held(); + assert(!kdbg_typefilter); + assert(!kdbg_typefilter_memory_entry); + typefilter_t deallocate_tf = NULL; - decval = (natural_t)kdr->value1; + if (!tf && ((tf = deallocate_tf = typefilter_create()) == NULL)) { + return ENOMEM; + } - if (decval && decval < KDBG_MINRTCDEC) - ret = EINVAL; - else - ret = ENOTSUP; + if ((kdbg_typefilter_memory_entry = typefilter_create_memory_entry(tf)) == MACH_PORT_NULL) { + if (deallocate_tf) { + typefilter_deallocate(deallocate_tf); + } + return ENOMEM; + } - return(ret); + /* + * The atomic store closes a race window with + * the kdebug_typefilter syscall, which assumes + * that any non-null kdbg_typefilter means a + * valid memory_entry is available. + */ + os_atomic_store(&kdbg_typefilter, tf, release); + + return KERN_SUCCESS; } -int -kdbg_enable_typefilter(void) +static int +kdbg_copyin_typefilter(user_addr_t addr, size_t size) { - int ret; + int ret = ENOMEM; + typefilter_t tf; - /* Allocate memory for bitmap if not already allocated */ - ret = kdbg_allocate_typefilter(); - if (ret) { - return ret; + ktrace_assert_lock_held(); + + if (size != KDBG_TYPEFILTER_BITMAP_SIZE) { + return EINVAL; } - /* Turn off range and value checks */ - kd_ctrl_page.kdebug_flags &= ~(KDBG_RANGECHECK | KDBG_VALCHECK); - - /* Enable filter checking */ - kd_ctrl_page.kdebug_flags |= KDBG_TYPEFILTER_CHECK; - kdbg_set_flags(SLOW_CHECKS, 0, TRUE); - return 0; -} + if ((tf = typefilter_create())) { + if ((ret = copyin(addr, tf, KDBG_TYPEFILTER_BITMAP_SIZE)) == 0) { + /* The kernel typefilter must always allow DBG_TRACE */ + typefilter_allow_class(tf, DBG_TRACE); -int -kdbg_disable_typefilter(void) -{ - /* Disable filter checking */ - kd_ctrl_page.kdebug_flags &= ~KDBG_TYPEFILTER_CHECK; + /* + * If this is the first typefilter; claim it. + * Otherwise copy and deallocate. + * + * Allocating a typefilter for the copyin allows + * the kernel to hold the invariant that DBG_TRACE + * must always be allowed. + */ + if (!kdbg_typefilter) { + if ((ret = kdbg_initialize_typefilter(tf))) { + return ret; + } + tf = NULL; + } else { + typefilter_copy(kdbg_typefilter, tf); + } - /* 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); + kdbg_enable_typefilter(); + kdbg_iop_list_callback(kd_ctrl_page.kdebug_iops, KD_CALLBACK_TYPEFILTER_CHANGED, kdbg_typefilter); + } - /* typefilter bitmap will be deallocated later */ + if (tf) { + typefilter_deallocate(tf); + } + } - return 0; + return ret; } -static int -kdbg_allocate_typefilter(void) +/* + * Enable the flags in the control page for the typefilter. Assumes that + * kdbg_typefilter has already been allocated, so events being written + * don't see a bad typefilter. + */ +static void +kdbg_enable_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; - } + assert(kdbg_typefilter); + kd_ctrl_page.kdebug_flags &= ~(KDBG_RANGECHECK | KDBG_VALCHECK); + kd_ctrl_page.kdebug_flags |= KDBG_TYPEFILTER_CHECK; + kdbg_set_flags(SLOW_CHECKS, 0, true); + commpage_update_kdebug_state(); +} - bzero((void *)bitmap, KDBG_TYPEFILTER_BITMAP_SIZE); +/* + * Disable the flags in the control page for the typefilter. The typefilter + * may be safely deallocated shortly after this function returns. + */ +static void +kdbg_disable_typefilter(void) +{ + bool notify_iops = kd_ctrl_page.kdebug_flags & KDBG_TYPEFILTER_CHECK; + kd_ctrl_page.kdebug_flags &= ~KDBG_TYPEFILTER_CHECK; - if (!OSCompareAndSwapPtr(NULL, (void *)bitmap, &type_filter_bitmap)) { - kmem_free(kernel_map, bitmap, KDBG_TYPEFILTER_BITMAP_SIZE); - return 0; /* someone assigned a buffer */ - } + if ((kd_ctrl_page.kdebug_flags & (KDBG_PIDCHECK | KDBG_PIDEXCLUDE))) { + kdbg_set_flags(SLOW_CHECKS, 0, true); } else { - bzero(type_filter_bitmap, KDBG_TYPEFILTER_BITMAP_SIZE); + kdbg_set_flags(SLOW_CHECKS, 0, false); } + commpage_update_kdebug_state(); - return 0; + if (notify_iops) { + /* + * Notify IOPs that the typefilter will now allow everything. + * Otherwise, they won't know a typefilter is no longer in + * effect. + */ + typefilter_allow_all(kdbg_typefilter); + kdbg_iop_list_callback(kd_ctrl_page.kdebug_iops, + KD_CALLBACK_TYPEFILTER_CHANGED, kdbg_typefilter); + } } -static int -kdbg_deallocate_typefilter(void) +uint32_t +kdebug_commpage_state(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 */ + if (kdebug_enable) { + if (kd_ctrl_page.kdebug_flags & KDBG_TYPEFILTER_CHECK) { + return KDEBUG_COMMPAGE_ENABLE_TYPEFILTER | KDEBUG_COMMPAGE_ENABLE_TRACE; } + + return KDEBUG_COMMPAGE_ENABLE_TRACE; } return 0; @@ -2226,38 +2674,37 @@ kdbg_deallocate_typefilter(void) int kdbg_setreg(kd_regtype * kdr) { - int ret=0; + int ret = 0; unsigned int val_1, val_2, val; switch (kdr->type) { - - case KDBG_CLASSTYPE : + case KDBG_CLASSTYPE: val_1 = (kdr->value1 & 0xff); val_2 = (kdr->value2 & 0xff); - kdlog_beg = (val_1<<24); - kdlog_end = (val_2<<24); + 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_VALCHECK; /* Turn off specific value check */ kd_ctrl_page.kdebug_flags |= (KDBG_RANGECHECK | KDBG_CLASSTYPE); - kdbg_set_flags(SLOW_CHECKS, 0, TRUE); + kdbg_set_flags(SLOW_CHECKS, 0, true); break; - case KDBG_SUBCLSTYPE : + 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)); + 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_VALCHECK; /* Turn off specific value check */ kd_ctrl_page.kdebug_flags |= (KDBG_RANGECHECK | KDBG_SUBCLSTYPE); - kdbg_set_flags(SLOW_CHECKS, 0, TRUE); + kdbg_set_flags(SLOW_CHECKS, 0, true); break; - case KDBG_RANGETYPE : + 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_VALCHECK; /* Turn off specific value check */ kd_ctrl_page.kdebug_flags |= (KDBG_RANGECHECK | KDBG_RANGETYPE); - kdbg_set_flags(SLOW_CHECKS, 0, TRUE); + kdbg_set_flags(SLOW_CHECKS, 0, true); break; case KDBG_VALCHECK: kdlog_value1 = (kdr->value1); @@ -2267,44 +2714,46 @@ kdbg_setreg(kd_regtype * kdr) kd_ctrl_page.kdebug_flags &= (unsigned int)~KDBG_CKTYPES; kd_ctrl_page.kdebug_flags &= ~KDBG_RANGECHECK; /* Turn off range check */ kd_ctrl_page.kdebug_flags |= KDBG_VALCHECK; /* Turn on specific value check */ - kdbg_set_flags(SLOW_CHECKS, 0, TRUE); + kdbg_set_flags(SLOW_CHECKS, 0, true); break; - case KDBG_TYPENONE : + case KDBG_TYPENONE: kd_ctrl_page.kdebug_flags &= (unsigned int)~KDBG_CKTYPES; - if ( (kd_ctrl_page.kdebug_flags & (KDBG_RANGECHECK | KDBG_VALCHECK | - KDBG_PIDCHECK | KDBG_PIDEXCLUDE | - KDBG_TYPEFILTER_CHECK)) ) - kdbg_set_flags(SLOW_CHECKS, 0, TRUE); - else - kdbg_set_flags(SLOW_CHECKS, 0, FALSE); + if ((kd_ctrl_page.kdebug_flags & (KDBG_RANGECHECK | KDBG_VALCHECK | + KDBG_PIDCHECK | KDBG_PIDEXCLUDE | + KDBG_TYPEFILTER_CHECK))) { + kdbg_set_flags(SLOW_CHECKS, 0, true); + } else { + kdbg_set_flags(SLOW_CHECKS, 0, false); + } kdlog_beg = 0; kdlog_end = 0; break; - default : + default: ret = EINVAL; break; } - return(ret); + 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)); + assert(size < INT_MAX); + return vn_rdwr(UIO_WRITE, vp, buffer, (int)size, file_offset, UIO_SYSSPACE, IO_NODELOCKED | IO_UNIT, + vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx)); } 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) { int ret = KERN_SUCCESS; - kd_chunk_header_v3 header; - - header.tag = tag; - header.sub_tag = sub_tag; - header.length = length; + kd_chunk_header_v3 header = { + .tag = tag, + .sub_tag = sub_tag, + .length = length, + }; // Check that only one of them is valid assert(!buffer ^ !vp); @@ -2318,8 +2767,7 @@ kdbg_write_v3_chunk_header(user_addr_t buffer, uint32_t tag, uint32_t sub_tag, u goto write_error; } RAW_file_offset += (sizeof(kd_chunk_header_v3)); - } - else { + } else { ret = copyout(&header, buffer, sizeof(kd_chunk_header_v3)); if (ret) { goto write_error; @@ -2330,25 +2778,7 @@ write_error: return ret; } -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 +static 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; @@ -2357,27 +2787,22 @@ kdbg_write_v3_chunk_to_fd(uint32_t tag, uint32_t sub_tag, uint64_t length, void vnode_t vp; p = current_proc(); - proc_fdlock(p); - if ( (fp_lookup(p, fd, &fp, 1)) ) { - proc_fdunlock(p); - return EFAULT; + if (fp_get_ftype(p, fd, DTYPE_VNODE, EBADF, &fp)) { + return EBADF; } + vp = fp->fp_glob->fg_data; 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; - } - vp = (struct vnode *) fp->f_fglob->fg_data; - proc_fdunlock(p); + context.vc_ucred = fp->fp_glob->fg_cred; - if ( (vnode_getwithref(vp)) == 0 ) { - RAW_file_offset = fp->f_fglob->fg_offset; + if ((vnode_getwithref(vp)) == 0) { + RAW_file_offset = fp->fp_glob->fg_offset; - kd_chunk_header_v3 chunk_header = { .tag = tag, .sub_tag = sub_tag, .length = length }; + 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) { @@ -2389,7 +2814,7 @@ kdbg_write_v3_chunk_to_fd(uint32_t tag, uint32_t sub_tag, uint64_t length, void RAW_file_offset += payload_size; } - fp->f_fglob->fg_offset = RAW_file_offset; + fp->fp_glob->fg_offset = RAW_file_offset; vnode_put(vp); } @@ -2400,185 +2825,179 @@ kdbg_write_v3_chunk_to_fd(uint32_t tag, uint32_t sub_tag, uint64_t length, void 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)); + 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; + int ret = KERN_SUCCESS; + + 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; + } + if (os_mul_overflow(kd_mapcount, sizeof(kd_threadmap), &thrmap_size)) { + ret = ERANGE; + goto bail; + } + + mach_timebase_info_data_t timebase = {0, 0}; + clock_timebase_info(&timebase); + + // Setup the header. + // See v3 header description in sys/kdebug.h for more inforamtion. + kd_header_v3 header = { + .tag = RAW_VERSION3, + .sub_tag = V3_HEADER_VERSION, + .length = (sizeof(kd_header_v3) + cpumap_size - sizeof(kd_cpumap_header)), + .timebase_numer = timebase.numer, + .timebase_denom = timebase.denom, + .timestamp = 0, /* FIXME rdar://problem/22053009 */ + .walltime_secs = 0, + .walltime_usecs = 0, + .timezone_minuteswest = 0, + .timezone_dst = 0, +#if defined(__LP64__) + .flags = 1, #else - header.flags = 0; + .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 = (size_t)header.length + thrmap_size + (2 * sizeof(kd_chunk_header_v3)); + 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; + } - // 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; + *user_header_size = bytes_needed; bail: - if (cpumap) { - kmem_free(kernel_map, (vm_offset_t)cpumap, cpumap_size); - } - return (ret); + if (cpumap) { + kmem_free(kernel_map, (vm_offset_t)cpumap, cpumap_size); + } + return ret; } int @@ -2598,66 +3017,89 @@ kdbg_readcpumap(user_addr_t user_cpumap, size_t *user_cpumap_size) } *user_cpumap_size = cpumap_size; kmem_free(kernel_map, (vm_offset_t)cpumap, cpumap_size); - } else + } else { ret = EINVAL; - } else + } + } else { ret = EINVAL; + } - return (ret); + return ret; } int kdbg_readcurthrmap(user_addr_t buffer, size_t *bufsize) { kd_threadmap *mapptr; - unsigned int mapsize; - unsigned int mapcount; - unsigned int count = 0; + vm_size_t mapsize; + vm_size_t mapcount; int ret = 0; + size_t count = *bufsize / sizeof(kd_threadmap); - count = *bufsize/sizeof(kd_threadmap); *bufsize = 0; - if ( (mapptr = kdbg_thrmap_init_internal(count, &mapsize, &mapcount)) ) { - if (copyout(mapptr, buffer, mapcount * sizeof(kd_threadmap))) + if ((mapptr = kdbg_thrmap_init_internal(count, &mapsize, &mapcount))) { + if (copyout(mapptr, buffer, mapcount * sizeof(kd_threadmap))) { ret = EFAULT; - else + } else { *bufsize = (mapcount * sizeof(kd_threadmap)); + } - kmem_free(kernel_map, (vm_offset_t)mapptr, mapsize); - } else + kfree(mapptr, mapsize); + } else { ret = EINVAL; + } - return (ret); + return ret; } static int -kdbg_write_v1_plus_header(uint32_t count, vnode_t vp, vfs_context_t ctx) +kdbg_write_v1_header(bool write_thread_map, vnode_t vp, vfs_context_t ctx) { int ret = 0; - RAW_header header; - clock_sec_t secs; - clock_usec_t usecs; - char *pad_buf; + 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); + size_t map_size = 0; + uint32_t map_count = 0; + + if (write_thread_map) { + assert(kd_ctrl_page.kdebug_flags & KDBG_MAPINIT); + if (kd_mapcount > UINT32_MAX) { + return ERANGE; + } + map_count = (uint32_t)kd_mapcount; + if (os_mul_overflow(map_count, sizeof(kd_threadmap), &map_size)) { + return ERANGE; + } + if (map_size >= INT_MAX) { + return ERANGE; + } + } /* - * 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. + * Without the buffers initialized, we cannot construct a CPU map or a + * thread map, and cannot write a header. + */ + if (!(kd_ctrl_page.kdebug_flags & KDBG_BUFINIT)) { + return EINVAL; + } + + /* + * 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. */ - assert(vp); - assert(ctx); + assert(vp); + assert(ctx); - pad_size = PAGE_16KB - ((sizeof(RAW_header) + (count * sizeof(kd_threadmap))) & PAGE_MASK_64); + pad_size = PAGE_16KB - ((sizeof(RAW_header) + map_size) & PAGE_MASK); cpumap_size = sizeof(kd_cpumap_header) + kd_ctrl_page.kdebug_cpus * sizeof(kd_cpumap); if (cpumap_size > pad_size) { @@ -2665,7 +3107,8 @@ kdbg_write_v1_plus_header(uint32_t count, vnode_t vp, vfs_context_t ctx) * 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.*/ + * data so that we can take advantage of optimized file/disk writes. + */ pad_size += PAGE_16KB; } @@ -2674,58 +3117,67 @@ kdbg_write_v1_plus_header(uint32_t count, vnode_t vp, vfs_context_t ctx) * 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; } + memset(&header, 0, sizeof(header)); header.version_no = RAW_VERSION1; - header.thread_count = count + extra_thread_count; + header.thread_count = map_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) + ret = vn_rdwr(UIO_WRITE, vp, (caddr_t)&header, (int)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); + RAW_file_written += 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 (write_thread_map) { + assert(map_size < INT_MAX); + ret = vn_rdwr(UIO_WRITE, vp, (caddr_t)kd_mapptr, (int)map_size, 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 += map_size; + RAW_file_written += map_size; + } if (extra_thread_count) { pad_size = extra_thread_count * sizeof(kd_threadmap); - pad_buf = (char *)kalloc(pad_size); + pad_buf = kheap_alloc(KHEAP_TEMP, pad_size, Z_WAITOK | Z_ZERO); 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) + assert(pad_size < INT_MAX); + ret = vn_rdwr(UIO_WRITE, vp, (caddr_t)pad_buf, (int)pad_size, RAW_file_offset, + UIO_SYSSPACE, IO_NODELOCKED | IO_UNIT, vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx)); + kheap_free(KHEAP_TEMP, pad_buf, pad_size); + if (ret) { goto write_error; - RAW_file_offset += pad_size; + } + RAW_file_offset += pad_size; + RAW_file_written += pad_size; } - pad_size = PAGE_SIZE - (RAW_file_offset & PAGE_MASK_64); + pad_size = PAGE_SIZE - (RAW_file_offset & PAGE_MASK); if (pad_size) { - pad_buf = (char *)kalloc(pad_size); + pad_buf = (char *)kheap_alloc(KHEAP_TEMP, pad_size, Z_WAITOK | Z_ZERO); if (!pad_buf) { ret = ENOMEM; goto write_error; } - memset(pad_buf, 0, pad_size); /* * embed a cpumap in the padding bytes. @@ -2737,204 +3189,241 @@ kdbg_write_v1_plus_header(uint32_t count, vnode_t vp, vfs_context_t ctx) 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) + assert(pad_size < INT_MAX); + ret = vn_rdwr(UIO_WRITE, vp, (caddr_t)pad_buf, (int)pad_size, RAW_file_offset, + UIO_SYSSPACE, IO_NODELOCKED | IO_UNIT, vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx)); + kheap_free(KHEAP_TEMP, pad_buf, pad_size); + if (ret) { goto write_error; + } + RAW_file_offset += pad_size; + RAW_file_written += 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) +static void +kdbg_clear_thread_map(void) { + ktrace_assert_lock_held(); + + if (kd_ctrl_page.kdebug_flags & KDBG_MAPINIT) { + assert(kd_mapptr != NULL); + kfree(kd_mapptr, kd_mapsize); + kd_mapptr = NULL; + kd_mapsize = 0; + kd_mapcount = 0; + kd_ctrl_page.kdebug_flags &= ~KDBG_MAPINIT; + } +} - int avail = 0; +/* + * Write out a version 1 header and the thread map, if it is initialized, to a + * vnode. Used by KDWRITEMAP and kdbg_dump_trace_to_file. + * + * Returns write errors from vn_rdwr if a write fails. Returns ENODATA if the + * thread map has not been initialized, but the header will still be written. + * Returns ENOMEM if padding could not be allocated. Returns 0 otherwise. + */ +static int +kdbg_write_thread_map(vnode_t vp, vfs_context_t ctx) +{ int ret = 0; - uint32_t count = 0; - unsigned int mapsize; + bool map_initialized; - if ((!vp && !buffer) || (vp && buffer)) { - return EINVAL; + ktrace_assert_lock_held(); + assert(ctx != NULL); + + map_initialized = (kd_ctrl_page.kdebug_flags & KDBG_MAPINIT); + + ret = kdbg_write_v1_header(map_initialized, vp, ctx); + if (ret == 0) { + if (map_initialized) { + kdbg_clear_thread_map(); + } else { + ret = ENODATA; + } } - assert(number); - assert((vp == NULL) || (ctx != NULL)); + return ret; +} + +/* + * Copy out the thread map to a user space buffer. Used by KDTHRMAP. + * + * Returns copyout errors if the copyout fails. Returns ENODATA if the thread + * map has not been initialized. Returns EINVAL if the buffer provided is not + * large enough for the entire thread map. Returns 0 otherwise. + */ +static int +kdbg_copyout_thread_map(user_addr_t buffer, size_t *buffer_size) +{ + bool map_initialized; + size_t map_size; + int ret = 0; - avail = *number; - count = avail/sizeof (kd_threadmap); - mapsize = kd_mapcount * sizeof(kd_threadmap); + ktrace_assert_lock_held(); + assert(buffer_size != NULL); - 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; - } - } - } - else - ret = EINVAL; + map_initialized = (kd_ctrl_page.kdebug_flags & KDBG_MAPINIT); + if (!map_initialized) { + return ENODATA; } - else - ret = EINVAL; - if (ret && vp) - { - count = 0; + map_size = kd_mapcount * sizeof(kd_threadmap); + if (*buffer_size < map_size) { + return EINVAL; + } - 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); - } + ret = copyout(kd_mapptr, buffer, map_size); + if (ret == 0) { + kdbg_clear_thread_map(); } -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); + + return ret; } int -kdbg_readthrmap_v3(user_addr_t buffer, size_t *number, int fd) +kdbg_readthrmap_v3(user_addr_t buffer, size_t buffer_size, int fd) { - int avail = 0; int ret = 0; - uint32_t count = 0; - unsigned int mapsize; + bool map_initialized; + size_t map_size; + + ktrace_assert_lock_held(); if ((!fd && !buffer) || (fd && buffer)) { return EINVAL; } - assert(number); + map_initialized = (kd_ctrl_page.kdebug_flags & KDBG_MAPINIT); + map_size = kd_mapcount * sizeof(kd_threadmap); - avail = *number; - count = avail/sizeof (kd_threadmap); - mapsize = kd_mapcount * sizeof(kd_threadmap); + if (map_initialized && (buffer_size >= map_size)) { + ret = kdbg_write_v3_header(buffer, &buffer_size, fd); - 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; + if (ret == 0) { + kdbg_clear_thread_map(); } - } - else { + } 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); -} + return ret; +} -static int -kdbg_set_nkdbufs(unsigned int value) +static void +kdbg_set_nkdbufs(unsigned int req_nkdbufs) { - /* - * We allow a maximum buffer size of 50% of either ram or max mapped address, whichever is smaller - * 'value' is the desired number of trace entries + /* + * Only allow allocation up to half the available memory (sane_size). */ - unsigned int max_entries = (sane_size/2) / sizeof(kd_buf); - - if (value <= max_entries) - return (value); - else - return (max_entries); + uint64_t max_nkdbufs = (sane_size / 2) / sizeof(kd_buf); + nkdbufs = (req_nkdbufs > max_nkdbufs) ? (unsigned int)max_nkdbufs : + req_nkdbufs; } - -static int -kdbg_enable_bg_trace(void) +/* + * Block until there are `n_storage_threshold` storage units filled with + * events or `timeout_ms` milliseconds have passed. If `locked_wait` is true, + * `ktrace_lock` is held while waiting. This is necessary while waiting to + * write events out of the buffers. + * + * Returns true if the threshold was reached and false otherwise. + * + * Called with `ktrace_lock` locked and interrupts enabled. + */ +static bool +kdbg_wait(uint64_t timeout_ms, bool locked_wait) { - int ret = 0; + int wait_result = THREAD_AWAKENED; + uint64_t abstime = 0; - if (kdlog_bg_trace == TRUE && kdlog_bg_trace_running == FALSE && n_storage_buffers == 0) { - nkdbufs = bg_nkdbufs; - ret = kdbg_reinit(FALSE); - if (0 == ret) { - kdbg_set_tracing_enabled(TRUE, KDEBUG_ENABLE_TRACE); - kdlog_bg_trace_running = TRUE; - } - wakeup(&kdlog_bg_trace); + ktrace_assert_lock_held(); + + if (timeout_ms != 0) { + uint64_t ns = timeout_ms * NSEC_PER_MSEC; + nanoseconds_to_absolutetime(ns, &abstime); + clock_absolutetime_interval_to_deadline(abstime, &abstime); } - return ret; -} -static void -kdbg_disable_bg_trace(void) -{ - if (kdlog_bg_trace_running == TRUE) { - kdlog_bg_trace_running = FALSE; - kdbg_clear(); + bool s = ml_set_interrupts_enabled(false); + if (!s) { + panic("kdbg_wait() called with interrupts disabled"); } -} + lck_spin_lock_grp(kdw_spin_lock, kdebug_lck_grp); + if (!locked_wait) { + /* drop the mutex to allow others to access trace */ + ktrace_unlock(); + } + while (wait_result == THREAD_AWAKENED && + kd_ctrl_page.kds_inuse_count < n_storage_threshold) { + kds_waiter = 1; -/* - * This function is provided for the CHUD toolkit only. - * int val: - * zero disables kdebug_chudhook function call - * non-zero enables kdebug_chudhook function call - * char *fn: - * address of the enabled kdebug_chudhook function -*/ + 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); + } -void -kdbg_control_chud(int val, void *fn) + kds_waiter = 0; + } + + /* check the count under the spinlock */ + bool threshold_exceeded = (kd_ctrl_page.kds_inuse_count >= n_storage_threshold); + + lck_spin_unlock(kdw_spin_lock); + ml_set_interrupts_enabled(s); + + if (!locked_wait) { + /* pick the mutex back up again */ + ktrace_lock(); + } + + /* write out whether we've exceeded the threshold */ + return threshold_exceeded; +} + +/* + * Wakeup a thread waiting using `kdbg_wait` if there are at least + * `n_storage_threshold` storage units in use. + */ +static void +kdbg_wakeup(void) { - kdbg_lock_init(); - - if (val) { - /* enable chudhook */ - kdebug_chudhook = fn; - kdbg_set_flags(SLOW_CHUD, KDEBUG_ENABLE_CHUD, TRUE); + bool need_kds_wakeup = false; + + /* + * 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 one of our 2 wait + * conditions. No problem if we fail, there will be lots of additional + * events coming in that will eventually succeed in grabbing this lock. + */ + bool 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); } - else { - /* disable chudhook */ - kdbg_set_flags(SLOW_CHUD, KDEBUG_ENABLE_CHUD, FALSE); - kdebug_chudhook = 0; + + ml_set_interrupts_enabled(s); + + if (need_kds_wakeup == true) { + wakeup(&kds_waiter); } } - int kdbg_control(int *name, u_int namelen, user_addr_t where, size_t *sizep) { @@ -2943,448 +3432,279 @@ kdbg_control(int *name, u_int namelen, user_addr_t where, size_t *sizep) unsigned int value = 0; kd_regtype kd_Reg; kbufinfo_t kd_bufinfo; - pid_t curpid; - proc_t p, curproc; - - if (name[0] == KERN_KDGETENTROPY || - 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_KDSETBUF) { - - if ( namelen < 2 ) - return(EINVAL); + proc_t p; + + if (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_KDSETBUF) { + if (namelen < 2) { + return EINVAL; + } value = name[1]; } - - kdbg_lock_init(); - if ( !(kd_ctrl_page.kdebug_flags & KDBG_LOCKINIT)) - return(ENOSPC); + kdbg_lock_init(); + assert(kd_ctrl_page.kdebug_flags & KDBG_LOCKINIT); - lck_mtx_lock(kd_trace_mtx_sysctl); + ktrace_lock(); - 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. - */ - 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); + /* + * Some requests only require "read" access to kdebug trace. Regardless, + * tell ktrace that a configuration or read is occurring (and see if it's + * allowed). + */ + if (name[0] != KERN_KDGETBUF && + name[0] != KERN_KDGETREG && + name[0] != KERN_KDREADCURTHRMAP) { + if ((ret = ktrace_configure(KTRACE_KDEBUG))) { goto out; } - - 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(); + } else { + if ((ret = ktrace_read_check())) { goto out; + } + } - case KERN_KDWAIT_BG_TRACE_RESET: - if (!kdlog_bg_trace){ - ret = EINVAL; - 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; + switch (name[0]) { + case KERN_KDGETBUF: + if (size < sizeof(kd_bufinfo.nkdbufs)) { + /* + * There is not enough room to return even + * the first element of the info structure. + */ + ret = EINVAL; + break; + } - case KERN_KDSET_BG_TYPEFILTER: - if (!kdlog_bg_trace || !kdlog_bg_trace_running){ - ret = EINVAL; - goto out; - } + memset(&kd_bufinfo, 0, sizeof(kd_bufinfo)); - if (size != KDBG_TYPEFILTER_BITMAP_SIZE) { - ret = EINVAL; - goto out; - } + kd_bufinfo.nkdbufs = nkdbufs; + kd_bufinfo.nkdthreads = kd_mapcount < INT_MAX ? (int)kd_mapcount : + INT_MAX; + if ((kd_ctrl_page.kdebug_slowcheck & SLOW_NOLOG)) { + kd_bufinfo.nolog = 1; + } else { + kd_bufinfo.nolog = 0; + } - if ((kd_ctrl_page.kdebug_flags & KDBG_TYPEFILTER_CHECK) == 0){ - if ((ret = kdbg_enable_typefilter())) - goto out; - } + kd_bufinfo.flags = kd_ctrl_page.kdebug_flags; +#if defined(__LP64__) + kd_bufinfo.flags |= KDBG_LP64; +#endif + { + int pid = ktrace_get_owning_pid(); + kd_bufinfo.bufid = (pid == 0 ? -1 : pid); + } - 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) - curpid = curproc->p_pid; - else { - ret = ESRCH; - goto out; - } - if (global_state_pid == -1) - global_state_pid = curpid; - else if (global_state_pid != curpid) { - if ((p = proc_find(global_state_pid)) == NULL) { + if (size >= sizeof(kd_bufinfo)) { /* - * The global pid no longer exists + * Provide all the info we have */ - global_state_pid = curpid; + if (copyout(&kd_bufinfo, where, sizeof(kd_bufinfo))) { + ret = EINVAL; + } } else { /* - * The global pid exists, deny this request + * For backwards compatibility, only provide + * as much info as there is room for. */ - proc_rele(p); - - ret = EBUSY; - goto out; + if (copyout(&kd_bufinfo, where, size)) { + ret = EINVAL; + } } - } + break; - switch(name[0]) { - case KERN_KDEFLAGS: - kdbg_disable_bg_trace(); + case KERN_KDREADCURTHRMAP: + ret = kdbg_readcurthrmap(where, sizep); + break; - value &= KDBG_USERFLAGS; - kd_ctrl_page.kdebug_flags |= value; - break; - case KERN_KDDFLAGS: - kdbg_disable_bg_trace(); + case KERN_KDEFLAGS: + value &= KDBG_USERFLAGS; + kd_ctrl_page.kdebug_flags |= value; + break; - value &= KDBG_USERFLAGS; - kd_ctrl_page.kdebug_flags &= ~value; - break; - case KERN_KDENABLE: + case KERN_KDDFLAGS: + value &= KDBG_USERFLAGS; + kd_ctrl_page.kdebug_flags &= ~value; + break; + + case KERN_KDENABLE: + /* + * Enable tracing mechanism. Two types: + * KDEBUG_TRACE is the standard one, + * and KDEBUG_PPT which is a carefully + * chosen subset to avoid performance impact. + */ + if (value) { /* - * Enable tracing mechanism. Two types: - * KDEBUG_TRACE is the standard one, - * and KDEBUG_PPT which is a carefully - * chosen subset to avoid performance impact. + * enable only if buffer is initialized */ - if (value) { - /* - * enable only if buffer is initialized - */ - if (!(kd_ctrl_page.kdebug_flags & KDBG_BUFINIT) || - !(value == KDEBUG_ENABLE_TRACE || value == KDEBUG_ENABLE_PPT)) { - ret = EINVAL; - break; - } - kdbg_thrmap_init(); - - kdbg_set_tracing_enabled(TRUE, value); - } - else - { - kdbg_set_tracing_enabled(FALSE, 0); - } - break; - case KERN_KDSETBUF: - kdbg_disable_bg_trace(); - - nkdbufs = kdbg_set_nkdbufs(value); - break; - case KERN_KDSETUP: - kdbg_disable_bg_trace(); - - ret = kdbg_reinit(FALSE); - break; - case KERN_KDREMOVE: - kdbg_clear(); - ret = kdbg_enable_bg_trace(); - break; - case KERN_KDSETREG: - if(size < sizeof(kd_regtype)) { + if (!(kd_ctrl_page.kdebug_flags & KDBG_BUFINIT) || + !(value == KDEBUG_ENABLE_TRACE || value == KDEBUG_ENABLE_PPT)) { ret = EINVAL; break; } - if (copyin(where, &kd_Reg, sizeof(kd_regtype))) { - ret = EINVAL; + kdbg_thrmap_init(); + + kdbg_set_tracing_enabled(true, value); + } else { + if (!kdebug_enable) { break; } - kdbg_disable_bg_trace(); - ret = kdbg_setreg(&kd_Reg); - break; - case KERN_KDGETREG: - kdbg_disable_bg_trace(); + kernel_debug_disable(); + } + break; + + case KERN_KDSETBUF: + kdbg_set_nkdbufs(value); + break; + + case KERN_KDSETUP: + ret = kdbg_reinit(false); + break; + + case KERN_KDREMOVE: + ktrace_reset(KTRACE_KDEBUG); + break; + + case KERN_KDSETREG: + if (size < sizeof(kd_regtype)) { ret = EINVAL; break; - case KERN_KDREADTR: - ret = kdbg_read(where, sizep, NULL, NULL, RAW_VERSION1); + } + if (copyin(where, &kd_Reg, sizeof(kd_regtype))) { + ret = EINVAL; break; - case KERN_KDWRITETR: - case KERN_KDWRITETR_V3: - case KERN_KDWRITEMAP: - case KERN_KDWRITEMAP_V3: - { - struct vfs_context context; - struct fileproc *fp; - size_t number; - vnode_t vp; - int fd; + } - if (name[0] == KERN_KDWRITETR || name[0] == KERN_KDWRITETR_V3) { - int s; - int wait_result = THREAD_AWAKENED; - u_int64_t abstime; - u_int64_t ns; - - 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); - lck_spin_lock(kdw_spin_lock); - - 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; - } - lck_spin_unlock(kdw_spin_lock); - ml_set_interrupts_enabled(s); - } - p = current_proc(); - fd = value; + ret = kdbg_setreg(&kd_Reg); + break; - proc_fdlock(p); - if ( (ret = fp_lookup(p, fd, &fp, 1)) ) { - proc_fdunlock(p); - break; - } - context.vc_thread = current_thread(); - context.vc_ucred = fp->f_fglob->fg_cred; + case KERN_KDGETREG: + ret = EINVAL; + break; - if (FILEGLOB_DTYPE(fp->f_fglob) != DTYPE_VNODE) { - fp_drop(p, fd, fp, 1); - proc_fdunlock(p); + case KERN_KDREADTR: + ret = kdbg_read(where, sizep, NULL, NULL, RAW_VERSION1); + break; + + case KERN_KDWRITETR: + case KERN_KDWRITETR_V3: + case KERN_KDWRITEMAP: + case KERN_KDWRITEMAP_V3: + { + struct vfs_context context; + struct fileproc *fp; + size_t number; + vnode_t vp; + int fd; + + if (name[0] == KERN_KDWRITETR || name[0] == KERN_KDWRITETR_V3) { + (void)kdbg_wait(size, true); + } + p = current_proc(); + fd = value; - ret = EBADF; - break; - } - vp = (struct vnode *)fp->f_fglob->fg_data; - proc_fdunlock(p); - - if ((ret = vnode_getwithref(vp)) == 0) { - 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(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_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); + if (fp_get_ftype(p, fd, DTYPE_VNODE, EBADF, &fp)) { + ret = EBADF; 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; + vp = fp->fp_glob->fg_data; + context.vc_thread = current_thread(); + context.vc_ucred = fp->fp_glob->fg_cred; - 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; - } + if ((ret = vnode_getwithref(vp)) == 0) { + RAW_file_offset = fp->fp_glob->fg_offset; + if (name[0] == KERN_KDWRITETR || name[0] == KERN_KDWRITETR_V3) { + number = nkdbufs * sizeof(kd_buf); - /* check the count under the spinlock */ - number = (kd_ctrl_page.kds_inuse_count >= n_storage_threshold); + KDBG_RELEASE(TRACE_WRITING_EVENTS | DBG_FUNC_START); + 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); + } + KDBG_RELEASE(TRACE_WRITING_EVENTS | DBG_FUNC_END, number); - lck_spin_unlock(kdw_spin_lock); - ml_set_interrupts_enabled(s); + *sizep = number; + } else { + number = kd_mapcount * sizeof(kd_threadmap); + if (name[0] == KERN_KDWRITEMAP_V3) { + ret = kdbg_readthrmap_v3(0, number, fd); + } else { + ret = kdbg_write_thread_map(vp, &context); + } + } + fp->fp_glob->fg_offset = RAW_file_offset; + vnode_put(vp); + } + fp_drop(p, fd, fp, 0); - /* pick the mutex back up again */ - lck_mtx_lock(kd_trace_mtx_sysctl); + break; + } + case KERN_KDBUFWAIT: + *sizep = kdbg_wait(size, false); + break; - /* write out whether we've exceeded the threshold */ - *sizep = number; + case KERN_KDPIDTR: + if (size < sizeof(kd_regtype)) { + ret = EINVAL; break; } - case KERN_KDPIDTR: - if (size < sizeof(kd_regtype)) { - ret = EINVAL; - break; - } - if (copyin(where, &kd_Reg, sizeof(kd_regtype))) { - ret = EINVAL; - break; - } - kdbg_disable_bg_trace(); - - ret = kdbg_setpid(&kd_Reg); + if (copyin(where, &kd_Reg, sizeof(kd_regtype))) { + ret = EINVAL; break; - case KERN_KDPIDEX: - if (size < sizeof(kd_regtype)) { - ret = EINVAL; - break; - } - if (copyin(where, &kd_Reg, sizeof(kd_regtype))) { - ret = EINVAL; - break; - } - kdbg_disable_bg_trace(); + } - ret = kdbg_setpidex(&kd_Reg); - break; - case KERN_KDCPUMAP: - ret = kdbg_readcpumap(where, sizep); - break; - case KERN_KDTHRMAP: - ret = kdbg_readthrmap(where, sizep, NULL, NULL); + ret = kdbg_setpid(&kd_Reg); + break; + + case KERN_KDPIDEX: + if (size < sizeof(kd_regtype)) { + ret = EINVAL; break; - case KERN_KDREADCURTHRMAP: - ret = kdbg_readcurthrmap(where, sizep); + } + if (copyin(where, &kd_Reg, sizeof(kd_regtype))) { + ret = EINVAL; break; - case KERN_KDSETRTCDEC: - if (size < sizeof(kd_regtype)) { - ret = EINVAL; - break; - } - if (copyin(where, &kd_Reg, sizeof(kd_regtype))) { - ret = EINVAL; - break; - } - kdbg_disable_bg_trace(); + } - ret = kdbg_setrtcdec(&kd_Reg); - break; - case KERN_KDSET_TYPEFILTER: - kdbg_disable_bg_trace(); + ret = kdbg_setpidex(&kd_Reg); + break; - if (size != KDBG_TYPEFILTER_BITMAP_SIZE) { - ret = EINVAL; - break; - } + case KERN_KDCPUMAP: + ret = kdbg_readcpumap(where, sizep); + break; - if ((kd_ctrl_page.kdebug_flags & KDBG_TYPEFILTER_CHECK) == 0){ - if ((ret = kdbg_enable_typefilter())) - break; - } + case KERN_KDTHRMAP: + ret = kdbg_copyout_thread_map(where, sizep); + 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; + case KERN_KDSET_TYPEFILTER: { + ret = kdbg_copyin_typefilter(where, size); + break; + } + + case KERN_KDTEST: + ret = kdbg_test(size); + break; + + default: + ret = EINVAL; + break; } out: - lck_mtx_unlock(kd_trace_mtx_sysctl); + ktrace_unlock(); - return(ret); + return ret; } @@ -3397,187 +3717,315 @@ out: int kdbg_read(user_addr_t buffer, size_t *number, vnode_t vp, vfs_context_t ctx, uint32_t file_version) { - unsigned int count; + size_t count; unsigned int cpu, min_cpu; - uint64_t mintime, t, barrier = 0; + uint64_t barrier_min = 0, barrier_max = 0, t, earliest_time; int error = 0; kd_buf *tempbuf; uint32_t rcursor; kd_buf lostevent; union kds_ptr kdsp; + bool traced_retrograde = false; struct kd_storage *kdsp_actual; struct kd_bufinfo *kdbp; struct kd_bufinfo *min_kdbp; - uint32_t tempbuf_count; + size_t tempbuf_count; uint32_t tempbuf_number; uint32_t old_kdebug_flags; uint32_t old_kdebug_slowcheck; - boolean_t lostevents = FALSE; - boolean_t out_of_events = FALSE; + bool out_of_events = false; + bool wrapped = false; - assert(number); - count = *number/sizeof(kd_buf); + assert(number != NULL); + count = *number / sizeof(kd_buf); *number = 0; - if (count == 0 || !(kd_ctrl_page.kdebug_flags & KDBG_BUFINIT) || kdcopybuf == 0) + ktrace_assert_lock_held(); + + if (count == 0 || !(kd_ctrl_page.kdebug_flags & KDBG_BUFINIT) || kdcopybuf == 0) { return EINVAL; + } + + thread_set_eager_preempt(current_thread()); memset(&lostevent, 0, sizeof(lostevent)); 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. + /* + * 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 - * be playing with kdebug_flags... the code that cuts new events could - * be running, but it grabs kds_spin_lock if it needs to acquire a new - * storage chunk which is where it examines kdebug_flags... it its adding - * to the same chunk we're reading from, no problem... + * Capture the current time. Only sort events that have occured + * before now. Since the IOPs are being flushed here, it is possible + * that events occur on the AP while running live tracing. */ + barrier_max = kdbg_timestamp() & KDBG_TIMESTAMP_MASK; - disable_wrap(&old_kdebug_slowcheck, &old_kdebug_flags); + /* + * Disable wrap so storage units cannot be stolen out from underneath us + * while merging events. + * + * Because we hold ktrace_lock, no other control threads can be playing + * with kdebug_flags. The code that emits new events could be running, + * but it grabs kds_spin_lock if it needs to acquire a new storage + * chunk, which is where it examines kdebug_flags. If it is adding to + * the same chunk we're reading from, check for that below. + */ + wrapped = disable_wrap(&old_kdebug_slowcheck, &old_kdebug_flags); - if (count > nkdbufs) + if (count > nkdbufs) { count = nkdbufs; + } - if ((tempbuf_count = count) > KDCOPYBUF_COUNT) - tempbuf_count = KDCOPYBUF_COUNT; + if ((tempbuf_count = count) > KDCOPYBUF_COUNT) { + tempbuf_count = KDCOPYBUF_COUNT; + } + + /* + * If the buffers have wrapped, do not emit additional lost events for the + * oldest storage units. + */ + if (wrapped) { + kd_ctrl_page.kdebug_flags &= ~KDBG_WRAPPED; + + for (cpu = 0, kdbp = &kdbip[0]; cpu < kd_ctrl_page.kdebug_cpus; cpu++, kdbp++) { + if ((kdsp = kdbp->kd_list_head).raw == KDS_PTR_NULL) { + continue; + } + kdsp_actual = POINTER_FROM_KDS_PTR(kdsp); + kdsp_actual->kds_lostevents = false; + } + } + /* + * Capture the earliest time where there are events for all CPUs and don't + * emit events with timestamps prior. + */ + barrier_min = kd_ctrl_page.oldest_time; while (count) { tempbuf = kdcopybuf; tempbuf_number = 0; - // While space + if (wrapped) { + /* + * Emit a lost events tracepoint to indicate that previous events + * were lost -- the thread map cannot be trusted. A new one must + * be taken so tools can analyze the trace in a backwards-facing + * fashion. + */ + kdbg_set_timestamp_and_cpu(&lostevent, barrier_min, 0); + *tempbuf = lostevent; + wrapped = false; + goto nextevent; + } + + /* While space left in merged events scratch buffer. */ while (tempbuf_count) { - mintime = 0xffffffffffffffffULL; + bool lostevents = false; + int lostcpu = 0; + earliest_time = UINT64_MAX; min_kdbp = NULL; min_cpu = 0; - // Check all CPUs + /* Check each CPU's buffers for the earliest event. */ 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 + /* Skip CPUs without data in their oldest storage unit. */ + if ((kdsp = kdbp->kd_list_head).raw == KDS_PTR_NULL) { +next_cpu: + continue; + } + /* 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 +next_event: + /* The next event to be read from this buffer. */ rcursor = kdsp_actual->kds_readlast; - if (rcursor == kdsp_actual->kds_bufindx) + /* Skip this buffer if there are no events left. */ + if (rcursor == kdsp_actual->kds_bufindx) { continue; + } + + /* + * Check that this storage unit wasn't stolen and events were + * lost. This must have happened while wrapping was disabled + * in this function. + */ + if (kdsp_actual->kds_lostevents) { + lostevents = true; + kdsp_actual->kds_lostevents = false; + + /* + * The earliest event we can trust is the first one in this + * stolen storage unit. + */ + uint64_t lost_time = + kdbg_get_timestamp(&kdsp_actual->kds_records[0]); + if (kd_ctrl_page.oldest_time < lost_time) { + /* + * If this is the first time we've seen lost events for + * this gap, record its timestamp as the oldest + * timestamp we're willing to merge for the lost events + * tracepoint. + */ + kd_ctrl_page.oldest_time = barrier_min = lost_time; + lostcpu = cpu; + } + } 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 > barrier_max) { + if (kdbg_debug) { + printf("kdebug: FUTURE EVENT: debugid %#8x: " + "time %lld from CPU %u " + "(barrier at time %lld, read %lu events)\n", + kdsp_actual->kds_records[rcursor].debugid, + t, cpu, barrier_max, *number + tempbuf_number); + } + goto next_cpu; + } if (t < kdsp_actual->kds_timestamp) { /* - * indicates we've not yet completed filling - * in this event... - * this should only occur when we're looking - * at the buf that the record head is utilizing - * we'll pick these events up on the next - * call to kdbg_read - * we bail at this point so that we don't - * get an out-of-order timestream by continuing - * to read events from the other CPUs' timestream(s) + * This indicates the event emitter hasn't completed + * filling in the event (becuase we're looking at the + * buffer that the record head is using). The max barrier + * timestamp should have saved us from seeing these kinds + * of things, but other CPUs might be slow on the up-take. + * + * Bail out so we don't get out-of-order events by + * continuing to read events from other CPUs' events. */ - out_of_events = TRUE; + out_of_events = true; break; } - if (t < mintime) { - mintime = t; + + /* + * Ignore events that have aged out due to wrapping or storage + * unit exhaustion while merging events. + */ + if (t < barrier_min) { + kdsp_actual->kds_readlast++; + if (kdbg_debug) { + printf("kdebug: PAST EVENT: debugid %#8x: " + "time %lld from CPU %u " + "(barrier at time %lld)\n", + kdsp_actual->kds_records[rcursor].debugid, + t, cpu, barrier_min); + } + + if (kdsp_actual->kds_readlast >= EVENTS_PER_STORAGE_UNIT) { + release_storage_unit(cpu, kdsp.raw); + + if ((kdsp = kdbp->kd_list_head).raw == KDS_PTR_NULL) { + goto next_cpu; + } + kdsp_actual = POINTER_FROM_KDS_PTR(kdsp); + } + + goto next_event; + } + + /* + * Don't worry about merging any events -- just walk through + * the CPUs and find the latest timestamp of lost events. + */ + if (lostevents) { + continue; + } + + if (t < earliest_time) { + earliest_time = t; min_kdbp = kdbp; min_cpu = cpu; } } - if (min_kdbp == NULL || out_of_events == TRUE) { + if (lostevents) { /* - * all buffers ran empty + * If any lost events were hit in the buffers, emit an event + * with the latest timestamp. */ - out_of_events = TRUE; + kdbg_set_timestamp_and_cpu(&lostevent, barrier_min, lostcpu); + *tempbuf = lostevent; + tempbuf->arg1 = 1; + goto nextevent; + } + if (min_kdbp == NULL) { + /* All buffers ran empty. */ + out_of_events = true; + } + if (out_of_events) { break; } - // Get data kdsp = min_kdbp->kd_list_head; kdsp_actual = POINTER_FROM_KDS_PTR(kdsp); - if (kdsp_actual->kds_lostevents == TRUE) { - kdbg_set_timestamp_and_cpu(&lostevent, kdsp_actual->kds_records[kdsp_actual->kds_readlast].timestamp, min_cpu); - *tempbuf = lostevent; - - kdsp_actual->kds_lostevents = FALSE; - lostevents = TRUE; - - goto nextevent; - } - - // Copy into buffer + /* Copy earliest event into merged events scratch buffer. */ *tempbuf = kdsp_actual->kds_records[kdsp_actual->kds_readlast++]; - if (kdsp_actual->kds_readlast == EVENTS_PER_STORAGE_UNIT) + if (kdsp_actual->kds_readlast == EVENTS_PER_STORAGE_UNIT) { release_storage_unit(min_cpu, kdsp.raw); + } /* - * Watch for out of order timestamps - */ - if (mintime < min_kdbp->kd_prev_timebase) { + * Watch for out of order timestamps (from IOPs). + */ + if (earliest_time < min_kdbp->kd_prev_timebase) { /* - * if so, use the previous timestamp + 1 cycle + * If we haven't already, emit a retrograde events event. + * Otherwise, ignore this event. */ - min_kdbp->kd_prev_timebase++; + if (traced_retrograde) { + continue; + } + if (kdbg_debug) { + printf("kdebug: RETRO EVENT: debugid %#8x: " + "time %lld from CPU %u " + "(barrier at time %lld)\n", + kdsp_actual->kds_records[rcursor].debugid, + t, cpu, barrier_min); + } + kdbg_set_timestamp_and_cpu(tempbuf, min_kdbp->kd_prev_timebase, kdbg_get_cpu(tempbuf)); - } else - min_kdbp->kd_prev_timebase = mintime; + tempbuf->arg1 = tempbuf->debugid; + tempbuf->arg2 = (kd_buf_argtype)earliest_time; + tempbuf->arg3 = 0; + tempbuf->arg4 = 0; + tempbuf->debugid = TRACE_RETROGRADE_EVENTS; + traced_retrograde = true; + } else { + min_kdbp->kd_prev_timebase = earliest_time; + } nextevent: tempbuf_count--; tempbuf_number++; tempbuf++; - if ((RAW_file_written += sizeof(kd_buf)) >= RAW_FLUSH_SIZE) + if ((RAW_file_written += sizeof(kd_buf)) >= RAW_FLUSH_SIZE) { break; + } } if (tempbuf_number) { + /* + * Remember the latest timestamp of events that we've merged so we + * don't think we've lost events later. + */ + uint64_t latest_time = kdbg_get_timestamp(tempbuf - 1); + if (kd_ctrl_page.oldest_time < latest_time) { + kd_ctrl_page.oldest_time = latest_time; + } if (file_version == RAW_VERSION3) { - if ( !(kdbg_write_v3_event_chunk_header(buffer, V3_RAW_EVENTS, (tempbuf_number * sizeof(kd_buf)), vp, ctx))) { + 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) + 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)); @@ -3586,11 +4034,12 @@ nextevent: if (vp) { 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) + if (!error) { RAW_file_offset += write_size; - + } + if (RAW_file_written >= RAW_FLUSH_SIZE) { - cluster_push(vp, 0); + error = VNOP_FSYNC(vp, MNT_NOWAIT, ctx); RAW_file_written = 0; } @@ -3607,500 +4056,393 @@ check_error: count -= tempbuf_number; *number += tempbuf_number; } - if (out_of_events == TRUE) - /* - * all trace buffers are empty - */ - break; + if (out_of_events == true) { + /* + * all trace buffers are empty + */ + break; + } - if ((tempbuf_count = count) > KDCOPYBUF_COUNT) - tempbuf_count = KDCOPYBUF_COUNT; + if ((tempbuf_count = count) > KDCOPYBUF_COUNT) { + tempbuf_count = KDCOPYBUF_COUNT; + } } - if ( !(old_kdebug_flags & KDBG_NOWRAP)) { - enable_wrap(old_kdebug_slowcheck, lostevents); + if (!(old_kdebug_flags & KDBG_NOWRAP)) { + enable_wrap(old_kdebug_slowcheck); } - return (error); + thread_clear_eager_preempt(current_thread()); + return error; } +#define KDEBUG_TEST_CODE(code) BSDDBG_CODE(DBG_BSD_KDEBUG_TEST, (code)) -unsigned char *getProcName(struct proc *proc); -unsigned char *getProcName(struct proc *proc) { +/* + * A test IOP for the SYNC_FLUSH callback. + */ - return (unsigned char *) &proc->p_comm; /* Return pointer to the proc name */ +static int sync_flush_iop = 0; +static void +sync_flush_callback(void * __unused context, kd_callback_type reason, + void * __unused arg) +{ + assert(sync_flush_iop > 0); + + if (reason == KD_CALLBACK_SYNC_FLUSH) { + kernel_debug_enter(sync_flush_iop, KDEBUG_TEST_CODE(0xff), + kdbg_timestamp(), 0, 0, 0, 0, 0); + } } +static struct kd_callback sync_flush_kdcb = { + .func = sync_flush_callback, + .iop_name = "test_sf", +}; + static int -stackshot_kern_return_to_bsd_error(kern_return_t kr) +kdbg_test(size_t flavor) { - 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; + int code = 0; + int dummy_iop = 0; + + switch (flavor) { + case 1: + /* try each macro */ + KDBG(KDEBUG_TEST_CODE(code)); code++; + KDBG(KDEBUG_TEST_CODE(code), 1); code++; + KDBG(KDEBUG_TEST_CODE(code), 1, 2); code++; + KDBG(KDEBUG_TEST_CODE(code), 1, 2, 3); code++; + KDBG(KDEBUG_TEST_CODE(code), 1, 2, 3, 4); code++; + + KDBG_RELEASE(KDEBUG_TEST_CODE(code)); code++; + KDBG_RELEASE(KDEBUG_TEST_CODE(code), 1); code++; + KDBG_RELEASE(KDEBUG_TEST_CODE(code), 1, 2); code++; + KDBG_RELEASE(KDEBUG_TEST_CODE(code), 1, 2, 3); code++; + KDBG_RELEASE(KDEBUG_TEST_CODE(code), 1, 2, 3, 4); code++; + + KDBG_FILTERED(KDEBUG_TEST_CODE(code)); code++; + KDBG_FILTERED(KDEBUG_TEST_CODE(code), 1); code++; + KDBG_FILTERED(KDEBUG_TEST_CODE(code), 1, 2); code++; + KDBG_FILTERED(KDEBUG_TEST_CODE(code), 1, 2, 3); code++; + KDBG_FILTERED(KDEBUG_TEST_CODE(code), 1, 2, 3, 4); code++; + + KDBG_RELEASE_NOPROCFILT(KDEBUG_TEST_CODE(code)); code++; + KDBG_RELEASE_NOPROCFILT(KDEBUG_TEST_CODE(code), 1); code++; + KDBG_RELEASE_NOPROCFILT(KDEBUG_TEST_CODE(code), 1, 2); code++; + KDBG_RELEASE_NOPROCFILT(KDEBUG_TEST_CODE(code), 1, 2, 3); code++; + KDBG_RELEASE_NOPROCFILT(KDEBUG_TEST_CODE(code), 1, 2, 3, 4); code++; + + KDBG_DEBUG(KDEBUG_TEST_CODE(code)); code++; + KDBG_DEBUG(KDEBUG_TEST_CODE(code), 1); code++; + KDBG_DEBUG(KDEBUG_TEST_CODE(code), 1, 2); code++; + KDBG_DEBUG(KDEBUG_TEST_CODE(code), 1, 2, 3); code++; + KDBG_DEBUG(KDEBUG_TEST_CODE(code), 1, 2, 3, 4); code++; + break; + + case 2: + if (kd_ctrl_page.kdebug_iops) { + /* avoid the assertion in kernel_debug_enter for a valid IOP */ + dummy_iop = kd_ctrl_page.kdebug_iops[0].cpu_id; + } + + /* ensure old timestamps are not emitted from kernel_debug_enter */ + kernel_debug_enter(dummy_iop, KDEBUG_TEST_CODE(code), + 100 /* very old timestamp */, 0, 0, 0, 0, 0); + code++; + kernel_debug_enter(dummy_iop, KDEBUG_TEST_CODE(code), + kdbg_timestamp(), 0, 0, 0, 0, 0); + code++; + break; + + case 3: + if (kd_ctrl_page.kdebug_iops) { + dummy_iop = kd_ctrl_page.kdebug_iops[0].cpu_id; + } + kernel_debug_enter(dummy_iop, KDEBUG_TEST_CODE(code), + kdbg_timestamp() * 2 /* !!! */, 0, 0, 0, 0, 0); + break; + + case 4: + if (!sync_flush_iop) { + sync_flush_iop = kernel_debug_register_callback( + sync_flush_kdcb); + assert(sync_flush_iop > 0); + } + break; + + default: + return ENOTSUP; } + + return 0; } +#undef KDEBUG_TEST_CODE -/* - * 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 - * for ppc, ppc64 and x86. - * Inputs: uap->pid - process id of process to be traced, or -1 - * for the entire system - * uap->tracebuf - address of the user space destination - * buffer - * uap->tracebuf_size - size of the user space trace buffer - * uap->options - various options, including the maximum - * number of frames to trace. - * Outputs: EPERM if the caller is not privileged - * EINVAL if the supplied trace buffer isn't sanely sized - * ENOMEM if we don't have enough memory to satisfy the - * request - * ENOENT if the target pid isn't found - * ENOSPC if the supplied buffer is insufficient - * *retval contains the number of bytes traced, if successful - * and -1 otherwise. If the request failed due to - * tracebuffer exhaustion, we copyout as much as possible. - */ -int -stack_snapshot(struct proc *p, register struct stack_snapshot_args *uap, int32_t *retval) { - int error = 0; - kern_return_t kr; +void +kdebug_init(unsigned int n_events, char *filter_desc, enum kdebug_opts opts) +{ + assert(filter_desc != NULL); - if ((error = suser(kauth_cred_get(), &p->p_acflag))) - return(error); + if (log_leaks && n_events == 0) { + n_events = 200000; + } - kr = stack_snapshot2(uap->pid, uap->tracebuf, uap->tracebuf_size, uap->flags, retval); - return stackshot_kern_return_to_bsd_error(kr); + kdebug_trace_start(n_events, filter_desc, opts); } -/* - * 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_snapshot_with_config(struct proc *p, struct stack_snapshot_with_config_args *uap, __unused int *retval) +static void +kdbg_set_typefilter_string(const char *filter_desc) { - int error = 0; - kern_return_t kr; + char *end = NULL; - if ((error = suser(kauth_cred_get(), &p->p_acflag))) - return(error); + ktrace_assert_lock_held(); - if((void*)uap->stackshot_config == NULL) { - return EINVAL; + assert(filter_desc != NULL); + + typefilter_reject_all(kdbg_typefilter); + typefilter_allow_class(kdbg_typefilter, DBG_TRACE); + + /* if the filter description starts with a number, assume it's a csc */ + if (filter_desc[0] >= '0' && filter_desc[0] <= '9') { + unsigned long csc = strtoul(filter_desc, NULL, 0); + if (filter_desc != end && csc <= KDBG_CSC_MAX) { + typefilter_allow_csc(kdbg_typefilter, (uint16_t)csc); + } + return; } - switch (uap->stackshot_config_version) { - case STACKSHOT_CONFIG_TYPE: - if (uap->stackshot_config_size != sizeof(stackshot_config_t)) { - return EINVAL; + while (filter_desc[0] != '\0') { + unsigned long allow_value; + + char filter_type = filter_desc[0]; + if (filter_type != 'C' && filter_type != 'S') { + printf("kdebug: unexpected filter type `%c'\n", filter_type); + return; + } + filter_desc++; + + allow_value = strtoul(filter_desc, &end, 0); + if (filter_desc == end) { + printf("kdebug: cannot parse `%s' as integer\n", filter_desc); + return; + } + + switch (filter_type) { + case 'C': + if (allow_value > KDBG_CLASS_MAX) { + printf("kdebug: class 0x%lx is invalid\n", allow_value); + return; } - stackshot_config_t config; - error = copyin(uap->stackshot_config, &config, sizeof(stackshot_config_t)); - if (error != KERN_SUCCESS) - { - return EFAULT; + printf("kdebug: C 0x%lx\n", allow_value); + typefilter_allow_class(kdbg_typefilter, (uint8_t)allow_value); + break; + case 'S': + if (allow_value > KDBG_CSC_MAX) { + printf("kdebug: class-subclass 0x%lx is invalid\n", allow_value); + return; } - kr = kern_stack_snapshot_internal(uap->stackshot_config_version, &config, sizeof(stackshot_config_t), TRUE); - return stackshot_kern_return_to_bsd_error(kr); + printf("kdebug: S 0x%lx\n", allow_value); + typefilter_allow_csc(kdbg_typefilter, (uint16_t)allow_value); + break; default: - return ENOTSUP; - } -} - -#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; - - if ((error = suser(kauth_cred_get(), &p->p_acflag))) - return(error); + __builtin_unreachable(); + } - kr = stack_microstackshot(uap->tracebuf, uap->tracebuf_size, uap->flags, retval); - return stackshot_kern_return_to_bsd_error(kr); + /* advance to next filter entry */ + filter_desc = end; + if (filter_desc[0] == ',') { + filter_desc++; + } + } } -#endif /* CONFIG_TELEMETRY */ -/* - * 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) +uint64_t +kdebug_wake(void) { - 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); + if (!wake_nkdbufs) { + return 0; + } + uint64_t start = mach_absolute_time(); + kdebug_trace_start(wake_nkdbufs, NULL, trace_wrap ? KDOPT_WRAPPING : 0); + return mach_absolute_time() - start; } /* - * 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 + * This function is meant to be called from the bootstrap thread or kdebug_wake. */ -int -stack_snapshot_from_kernel(pid_t pid, void *buf, uint32_t size, uint32_t flags, unsigned *bytes_traced) +void +kdebug_trace_start(unsigned int n_events, const char *filter_desc, + enum kdebug_opts opts) { - kern_return_t kr; - - kr = stack_snapshot_from_kernel_internal(pid, buf, size, flags, bytes_traced); - if (kr == KERN_FAILURE) { - return -1; + if (!n_events) { + kd_early_done = true; + return; } - return kr; -} + ktrace_start_single_threaded(); -void -start_kern_tracing(unsigned int new_nkdbufs, boolean_t need_map) -{ - - if (!new_nkdbufs) - return; - nkdbufs = kdbg_set_nkdbufs(new_nkdbufs); kdbg_lock_init(); - kernel_debug_string_simple("start_kern_tracing"); + ktrace_kernel_configure(KTRACE_KDEBUG); - if (0 == kdbg_reinit(TRUE)) { + kdbg_set_nkdbufs(n_events); - if (need_map == TRUE) { - uint32_t old1, old2; + kernel_debug_string_early("start_kern_tracing"); - kdbg_thrmap_init(); + if (kdbg_reinit((opts & KDOPT_ATBOOT))) { + printf("error from kdbg_reinit, kernel tracing not started\n"); + goto out; + } - disable_wrap(&old1, &old2); + /* + * Wrapping is disabled because boot and wake tracing is interested in + * the earliest events, at the expense of later ones. + */ + if (!(opts & KDOPT_WRAPPING)) { + uint32_t old1, old2; + (void)disable_wrap(&old1, &old2); + } + + if (filter_desc && filter_desc[0] != '\0') { + if (kdbg_initialize_typefilter(NULL) == KERN_SUCCESS) { + kdbg_set_typefilter_string(filter_desc); + kdbg_enable_typefilter(); } + } + + /* + * Hold off interrupts between getting a thread map and enabling trace + * and until the early traces are recorded. + */ + bool s = ml_set_interrupts_enabled(false); - /* Hold off interrupts until the early traces are cut */ - boolean_t s = ml_set_interrupts_enabled(FALSE); + if (!(opts & KDOPT_ATBOOT)) { + kdbg_thrmap_init(); + } - kdbg_set_tracing_enabled( - TRUE, - kdebug_serial ? - (KDEBUG_ENABLE_TRACE | KDEBUG_ENABLE_SERIAL) : - KDEBUG_ENABLE_TRACE); + kdbg_set_tracing_enabled(true, KDEBUG_ENABLE_TRACE); + if ((opts & KDOPT_ATBOOT)) { /* - * Transfer all very early events from the static buffer - * into the real buffers. + * 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 - } 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); + ml_set_interrupts_enabled(s); - /* check that tracing was actually enabled */ - if (!(kdebug_enable & KDEBUG_ENABLE_TRACE)) - return; + printf("kernel tracing started with %u events, filter = %s\n", n_events, + filter_desc ?: "none"); - /* setup the typefiltering */ - if (0 == kdbg_enable_typefilter()) - setbit(type_filter_bitmap, - typefilter & (KDBG_CSC_MASK >> KDBG_CSC_OFFSET)); +out: + ktrace_end_single_threaded(); } void kdbg_dump_trace_to_file(const char *filename) { - vfs_context_t ctx; - vnode_t vp; - int error; - size_t number; + vfs_context_t ctx; + vnode_t vp; + size_t write_size; + int ret; + ktrace_lock(); - if ( !(kdebug_enable & KDEBUG_ENABLE_TRACE)) - return; + if (!(kdebug_enable & KDEBUG_ENABLE_TRACE)) { + goto out; + } - if (global_state_pid != -1) { - if ((proc_find(global_state_pid)) != NULL) { - /* - * 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; - } + if (ktrace_get_owning_pid() != 0) { + /* + * Another process owns ktrace and is still active, disable tracing to + * prevent wrapping. + */ + kdebug_enable = 0; + kd_ctrl_page.enabled = 0; + commpage_update_kdebug_state(); + goto out; } - KERNEL_DEBUG_CONSTANT(TRACE_PANIC | DBG_FUNC_NONE, 0, 0, 0, 0, 0); + + KDBG_RELEASE(TRACE_WRITING_EVENTS | DBG_FUNC_START); kdebug_enable = 0; kd_ctrl_page.enabled = 0; - commpage_update_kdebug_enable(); + commpage_update_kdebug_state(); ctx = vfs_context_kernel(); - if ((error = vnode_open(filename, (O_CREAT | FWRITE | O_NOFOLLOW), 0600, 0, &vp, ctx))) - return; + if (vnode_open(filename, (O_CREAT | FWRITE | O_NOFOLLOW), 0600, 0, &vp, ctx)) { + goto out; + } - number = kd_mapcount * sizeof(kd_threadmap); - kdbg_readthrmap(0, &number, vp, ctx); + kdbg_write_thread_map(vp, ctx); - number = nkdbufs*sizeof(kd_buf); - kdbg_read(0, &number, vp, ctx, RAW_VERSION1); - - vnode_close(vp, FWRITE, ctx); + write_size = nkdbufs * sizeof(kd_buf); + ret = kdbg_read(0, &write_size, vp, ctx, RAW_VERSION1); + if (ret) { + goto out_close; + } - sync(current_proc(), (void *)NULL, (int *)NULL); -} + /* + * Wait to synchronize the file to capture the I/O in the + * TRACE_WRITING_EVENTS interval. + */ + ret = VNOP_FSYNC(vp, MNT_WAIT, ctx); -/* Helper function for filling in the BSD name for an address space - * Defined here because the machine bindings know only Mach threads - * and nothing about BSD processes. - * - * FIXME: need to grab a lock during this? - */ -void kdbg_get_task_name(char* name_buf, int len, task_t task) -{ - proc_t proc; - - /* Note: we can't use thread->task (and functions that rely on it) here - * because it hasn't been initialized yet when this function is called. - * We use the explicitly-passed task parameter instead. + /* + * Balance the starting TRACE_WRITING_EVENTS tracepoint manually. */ - proc = get_bsdtask_info(task); - if (proc != PROC_NULL) - snprintf(name_buf, len, "%s/%d", proc->p_comm, proc->p_pid); - else - snprintf(name_buf, len, "%p [!bsd]", task); -} + kd_buf end_event = { + .debugid = TRACE_WRITING_EVENTS | DBG_FUNC_END, + .arg1 = write_size, + .arg2 = ret, + .arg5 = (kd_buf_argtype)thread_tid(current_thread()), + }; + kdbg_set_timestamp_and_cpu(&end_event, kdbg_timestamp(), + cpu_number()); -#if KDEBUG_MOJO_TRACE -static kd_event_t * -binary_search(uint32_t id) -{ - int low, high, mid; + /* this is best effort -- ignore any errors */ + (void)kdbg_write_to_vnode((caddr_t)&end_event, sizeof(kd_buf), vp, ctx, + RAW_file_offset); - low = 0; - high = sizeof(kd_events)/sizeof(kd_event_t) - 1; +out_close: + vnode_close(vp, FWRITE, ctx); + sync(current_proc(), (void *)NULL, (int *)NULL); - 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; - } +out: + ktrace_unlock(); } -/* - * 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) +static int +kdbg_sysctl_continuous SYSCTL_HANDLER_ARGS { - if (last_hit[cpu] == NULL || last_hit[cpu]->id != id) - last_hit[cpu] = binary_search(id); - return last_hit[cpu]; -} +#pragma unused(oidp, arg1, arg2) + int value = kdbg_continuous_time; + int ret = sysctl_io_number(req, value, sizeof(value), &value, NULL); -static uint64_t kd_last_timstamp; + if (ret || !req->newptr) { + return ret; + } -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; + kdbg_continuous_time = value; + return 0; } -#endif + +SYSCTL_NODE(_kern, OID_AUTO, kdbg, CTLFLAG_RD | CTLFLAG_LOCKED, 0, + "kdbg"); + +SYSCTL_PROC(_kern_kdbg, OID_AUTO, experimental_continuous, + CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_LOCKED, 0, + sizeof(int), kdbg_sysctl_continuous, "I", + "Set kdebug to use mach_continuous_time"); + +SYSCTL_INT(_kern_kdbg, OID_AUTO, debug, + CTLFLAG_RW | CTLFLAG_LOCKED, + &kdbg_debug, 0, "Set kdebug debug mode"); + +SYSCTL_QUAD(_kern_kdbg, OID_AUTO, oldest_time, + CTLTYPE_QUAD | CTLFLAG_RD | CTLFLAG_LOCKED, + &kd_ctrl_page.oldest_time, + "Find the oldest timestamp still in trace");