/*
- * 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,
* 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 <machine/spl.h>
-
#include <sys/errno.h>
#include <sys/param.h>
#include <sys/systm.h>
#include <sys/vm.h>
#include <sys/sysctl.h>
#include <sys/kdebug.h>
+#include <sys/kauth.h>
+#include <sys/ktrace.h>
#include <sys/sysproto.h>
#include <sys/bsdtask_info.h>
+#include <sys/random.h>
-#define HZ 100
#include <mach/clock_types.h>
#include <mach/mach_types.h>
#include <mach/mach_time.h>
+#include <mach/mach_vm.h>
+#include <machine/atomic.h>
#include <machine/machine_routines.h>
+#include <mach/machine.h>
+#include <mach/vm_map.h>
+
#if defined(__i386__) || defined(__x86_64__)
#include <i386/rtclock_protos.h>
#include <i386/mp.h>
#include <i386/machine_routines.h>
+#include <i386/tsc.h>
#endif
#include <kern/clock.h>
#include <kern/task.h>
#include <kern/debug.h>
#include <kern/kalloc.h>
+#include <kern/cpu_number.h>
#include <kern/cpu_data.h>
#include <kern/assert.h>
#include <kern/telemetry.h>
+#include <kern/sched_prim.h>
#include <vm/vm_kern.h>
#include <sys/lock.h>
+#include <kperf/kperf.h>
+#include <pexpert/device_tree.h>
#include <sys/malloc.h>
#include <sys/mcache.h>
-#include <sys/kauth.h>
#include <sys/vnode.h>
#include <sys/vnode_internal.h>
#include <sys/fcntl.h>
#include <sys/file_internal.h>
#include <sys/ubc.h>
-#include <sys/param.h> /* for isset() */
+#include <sys/param.h> /* for isset() */
-#include <mach/mach_host.h> /* for host_info() */
+#include <mach/mach_host.h> /* for host_info() */
#include <libkern/OSAtomic.h>
#include <machine/pal_routines.h>
+#include <machine/atomic.h>
+
+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.
*
*/
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 <machine/...> include path broken */
+/*
+ * 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);
-void kdbg_control_chud(int, void *);
int kdbg_control(int *, u_int, user_addr_t, size_t *);
-int kdbg_getentropy (user_addr_t, size_t *, int);
-int kdbg_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_readcurthrmap(user_addr_t, size_t *);
-int kdbg_getreg(kd_regtype *);
-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);
-
-int kdbg_cpumap_init_internal(kd_iop_t* iops, uint32_t cpu_count, uint8_t** cpumap, uint32_t* cpumap_size);
-kd_threadmap* kdbg_thrmap_init_internal(unsigned int count, unsigned int *mapsize, unsigned int *mapcount);
-
-static int kdbg_enable_typefilter(void);
-static int kdbg_disable_typefilter(void);
-
-static int create_buffers(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);
+
+static kd_threadmap *kdbg_thrmap_init_internal(size_t max_count,
+ vm_size_t *map_size, vm_size_t *map_count);
+
+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);
+
+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);
+
+// Helper functions
+
+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;
-/* track timestamps for security server's entropy needs */
-uint64_t * kd_entropy_buffer = 0;
-unsigned int kd_entropy_bufsize = 0;
-unsigned int kd_entropy_count = 0;
-unsigned int kd_entropy_indx = 0;
-vm_offset_t kd_entropy_buftomem = 0;
+/* A static buffer to record events prior to the start of regular logging */
-#define MAX_ENTROPY_COUNT (128 * 1024)
+#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__) */
-#define SLOW_NOLOG 0x01
-#define SLOW_CHECKS 0x02
-#define SLOW_ENTROPY 0x04
-#define SLOW_CHUD 0x08
+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])
};
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 kde_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)));
+
+/*
+ * In principle, this control block can be shared in DRAM with other
+ * coprocessors and runtimes, for configuring what tracing is enabled.
+ */
struct kd_ctrl_page_t {
union kds_ptr kds_free_list;
- uint32_t enabled :1;
- 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
*/
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))
-kd_buf *kdcopybuf = NULL;
+#define KDCOPYBUF_COUNT 8192
+#define KDCOPYBUF_SIZE (KDCOPYBUF_COUNT * sizeof(kd_buf))
-boolean_t kdlog_bg_trace = FALSE;
-boolean_t kdlog_bg_trace_running = FALSE;
-unsigned int bg_nkdbufs = 0;
+#define PAGE_4KB 4096
+#define PAGE_16KB 16384
+
+kd_buf *kdcopybuf = NULL;
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;
-
-static lck_grp_t *stackshot_subsys_lck_grp;
-static lck_grp_attr_t *stackshot_subsys_lck_grp_attr;
-static lck_attr_t *stackshot_subsys_lck_attr;
-static lck_mtx_t stackshot_subsys_mutex;
-
-void *stackshot_snapbuf = NULL;
-
-int
-stack_snapshot2(pid_t pid, user_addr_t tracebuf, uint32_t tracebuf_size, uint32_t flags, uint32_t dispatch_offset, int32_t *retval);
-
-int
-stack_snapshot_from_kernel(pid_t pid, void *buf, uint32_t size, uint32_t flags, unsigned *bytesTraced);
-extern void
-kdp_snapshot_preflight(int pid, void *tracebuf, uint32_t tracebuf_size, uint32_t flags, uint32_t dispatch_offset);
-
-extern int
-kdp_stack_snapshot_geterror(void);
-extern unsigned int
-kdp_stack_snapshot_bytes_traced(void);
kd_threadmap *kd_mapptr = 0;
-unsigned int kd_mapsize = 0;
-unsigned int kd_mapcount = 0;
+vm_size_t kd_mapsize = 0;
+vm_size_t kd_mapcount = 0;
-off_t RAW_file_offset = 0;
-int RAW_file_written = 0;
+off_t RAW_file_offset = 0;
+int RAW_file_written = 0;
-#define RAW_FLUSH_SIZE (2 * 1024 * 1024)
+#define RAW_FLUSH_SIZE (2 * 1024 * 1024)
-pid_t global_state_pid = -1; /* Used to control exclusive use of kd_buffer */
-
-#define DBG_FUNC_MASK 0xfffffffc
-
-/* TODO: move to kdebug.h */
-#define CLASS_MASK 0xff000000
-#define CLASS_OFFSET 24
-#define SUBCLASS_MASK 0x00ff0000
-#define SUBCLASS_OFFSET 16
-#define CSC_MASK 0xffff0000 /* class and subclass mask */
-#define CSC_OFFSET SUBCLASS_OFFSET
+/*
+ * A globally increasing counter for identifying strings in trace. Starts at
+ * 1 because 0 is a reserved return value.
+ */
+__attribute__((aligned(MAX_CPU_CACHE_LINE_SIZE)))
+static uint64_t g_curr_str_id = 1;
-#define EXTRACT_CLASS(debugid) ( (uint8_t) ( ((debugid) & CLASS_MASK ) >> CLASS_OFFSET ) )
-#define EXTRACT_SUBCLASS(debugid) ( (uint8_t) ( ((debugid) & SUBCLASS_MASK) >> SUBCLASS_OFFSET ) )
-#define EXTRACT_CSC(debugid) ( (uint16_t)( ((debugid) & CSC_MASK ) >> CSC_OFFSET ) )
+#define STR_ID_SIG_OFFSET (48)
+#define STR_ID_MASK ((1ULL << STR_ID_SIG_OFFSET) - 1)
+#define STR_ID_SIG_MASK (~STR_ID_MASK)
-#define INTERRUPT 0x01050000
-#define MACH_vmfault 0x01300008
-#define BSC_SysCall 0x040c0000
-#define MACH_SysCall 0x010c0000
-#define DBG_SCALL_MASK 0xffff0000
+/*
+ * A bit pattern for identifying string IDs generated by
+ * kdebug_trace_string(2).
+ */
+static uint64_t g_str_id_signature = (0x70acULL << STR_ID_SIG_OFFSET);
+#define INTERRUPT 0x01050000
+#define MACH_vmfault 0x01300008
+#define BSC_SysCall 0x040c0000
+#define MACH_SysCall 0x010c0000
-/* 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);
+/*
+ * 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;
+ * }
+ * }
+ *
+ */
-volatile kd_chudhook_fn kdebug_chudhook = 0; /* pointer to CHUD toolkit function */
+#define RAW_VERSION3 0x00001000
-__private_extern__ void stackshot_lock_init( void );
+// 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
-static uint8_t *type_filter_bitmap;
+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 boolean_t
+
+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;
-}
+ 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));
-static boolean_t
-kdbg_iop_list_contains_cpu_id(kd_iop_t* list, uint32_t cpu_id)
-{
- while (list) {
- if (list->cpu_id == cpu_id)
- return TRUE;
- list = list->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 FALSE;
+ return true;
}
-/*
- * This is a temporary workaround for <rdar://problem/13512084>
- *
- * DO NOT CALL IN RELEASE BUILD, LEAKS ADDRESS INFORMATION!
- */
-static boolean_t
-kdbg_iop_list_check_for_timestamp_rollback(kd_iop_t* list, uint32_t cpu_id, uint64_t timestamp)
-{
- while (list) {
- if (list->cpu_id == cpu_id) {
- if (list->last_timestamp > timestamp) {
- kprintf("%s is sending trace events that have gone backwards in time. Run the following command: \"symbols -2 -lookup 0x%p\" and file a radar against the matching kext.\n", list->callback.iop_name, (void*)list->callback.func);
- }
- /* Unconditional set mitigates syslog spam */
- list->last_timestamp = timestamp;
- return TRUE;
- }
- list = list->next;
- }
-
- return FALSE;
-}
#endif /* MACH_ASSERT */
static void
}
}
+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_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_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;
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
kd_ctrl_page.kdebug_cpus = kd_ctrl_page.kdebug_iops ? kd_ctrl_page.kdebug_iops->cpu_id + 1 : kdbg_cpu_count(early_trace);
- if (kmem_alloc(kernel_map, (vm_offset_t *)&kdbip, sizeof(struct kd_bufinfo) * kd_ctrl_page.kdebug_cpus) != KERN_SUCCESS) {
+ if (kmem_alloc(kernel_map, (vm_offset_t *)&kdbip, sizeof(struct kd_bufinfo) * kd_ctrl_page.kdebug_cpus, VM_KERN_MEMORY_DIAG) != KERN_SUCCESS) {
error = ENOSPC;
goto out;
}
- 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;
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) != KERN_SUCCESS) {
+ if (kmem_alloc(kernel_map, (vm_offset_t *)&kdcopybuf, (vm_size_t)KDCOPYBUF_SIZE, VM_KERN_MEMORY_DIAG) != KERN_SUCCESS) {
error = ENOSPC;
goto out;
}
}
- if (kmem_alloc(kernel_map, (vm_offset_t *)&kd_bufs, (vm_size_t)(n_storage_buffers * sizeof(struct kd_storage_buffers))) != KERN_SUCCESS) {
+ if (kmem_alloc(kernel_map, (vm_offset_t *)&kd_bufs, (vm_size_t)(n_storage_buffers * sizeof(struct kd_storage_buffers)), VM_KERN_MEMORY_DIAG) != KERN_SUCCESS) {
error = ENOSPC;
goto out;
}
bzero(kd_bufs, n_storage_buffers * sizeof(struct kd_storage_buffers));
for (i = 0; i < f_buffers; i++) {
- if (kmem_alloc(kernel_map, (vm_offset_t *)&kd_bufs[i].kdsb_addr, (vm_size_t)f_buffer_size) != KERN_SUCCESS) {
+ if (kmem_alloc(kernel_map, (vm_offset_t *)&kd_bufs[i].kdsb_addr, (vm_size_t)f_buffer_size, VM_KERN_MEMORY_DIAG) != KERN_SUCCESS) {
error = ENOSPC;
goto out;
}
kd_bufs[i].kdsb_size = f_buffer_size;
}
if (p_buffer_size) {
- if (kmem_alloc(kernel_map, (vm_offset_t *)&kd_bufs[i].kdsb_addr, (vm_size_t)p_buffer_size) != KERN_SUCCESS) {
+ if (kmem_alloc(kernel_map, (vm_offset_t *)&kd_bufs[i].kdsb_addr, (vm_size_t)p_buffer_size, VM_KERN_MEMORY_DIAG) != KERN_SUCCESS) {
error = ENOSPC;
goto out;
}
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;
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) {
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;
}
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];
* 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
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];
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
*/
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;
}
if (kdbp_vict == NULL) {
kdebug_enable = 0;
kd_ctrl_page.enabled = 0;
- retval = FALSE;
+ commpage_update_kdebug_state();
+ retval = false;
goto out;
}
kdsp = kdbp_vict->kd_list_head;
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
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)) == KERN_SUCCESS) {
+ 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));
-
+
/*
* <rdar://problem/13351477> Some IOP clients are not providing a name.
*
* Remove when fixed.
*/
{
- boolean_t is_valid_name = FALSE;
- for (uint32_t length=0; length<sizeof(callback.iop_name); ++length) {
+ bool is_valid_name = false;
+ for (uint32_t length = 0; length < sizeof(callback.iop_name); ++length) {
/* This is roughly isprintable(c) */
- if (callback.iop_name[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
* 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
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) {
- if (isset(type_filter_bitmap, EXTRACT_CSC(debugid)))
+ if (typefilter_is_debugid_allowed(kdbg_typefilter, debugid)) {
goto record_event;
+ }
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) {
- if ((debugid & DBG_FUNC_MASK) != kdlog_value1 &&
- (debugid & DBG_FUNC_MASK) != kdlog_value2 &&
- (debugid & DBG_FUNC_MASK) != kdlog_value3 &&
- (debugid & DBG_FUNC_MASK) != kdlog_value4)
+ } else if (kd_ctrl_page.kdebug_flags & KDBG_VALCHECK) {
+ if ((debugid & KDBG_EVENTID_MASK) != kdlog_value1 &&
+ (debugid & KDBG_EVENTID_MASK) != kdlog_value2 &&
+ (debugid & KDBG_EVENTID_MASK) != kdlog_value3 &&
+ (debugid & KDBG_EVENTID_MASK) != kdlog_value4) {
goto out1;
+ }
}
}
-
+
record_event:
- assert(kdbg_iop_list_contains_cpu_id(kd_ctrl_page.kdebug_iops, coreid));
- /* Remove when <rdar://problem/13512084> is closed. */
- assert(kdbg_iop_list_check_for_timestamp_rollback(kd_ctrl_page.kdebug_iops, coreid, timestamp));
+ if (timestamp < kd_ctrl_page.oldest_time) {
+ goto out1;
+ }
disable_preemption();
- if (kd_ctrl_page.enabled == 0)
+ if (kd_ctrl_page.enabled == 0) {
goto out;
+ }
kdbp = &kdbip[coreid];
timestamp &= KDBG_TIMESTAMP_MASK;
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
}
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];
kd->arg3 = arg3;
kd->arg4 = arg4;
kd->arg5 = threadid;
-
+
kdbg_set_timestamp_and_cpu(kd, timestamp, coreid);
OSAddAtomic(1, &kdsp_actual->kds_bufcnt);
enable_preemption();
out1:
if ((kds_waiter && kd_ctrl_page.kds_inuse_count >= n_storage_threshold)) {
- boolean_t need_kds_wakeup = FALSE;
- int s;
-
- /*
- * try to take the lock here to synchronize with the
- * waiter entering the blocked state... use the try
- * mode to prevent deadlocks caused by re-entering this
- * routine due to various trace points triggered in the
- * lck_spin_sleep_xxxx routines used to actually enter
- * our wait condition... no problem if we fail,
- * there will be lots of additional events coming in that
- * will eventually succeed in grabbing this lock
- */
- s = ml_set_interrupts_enabled(FALSE);
+ kdbg_wakeup();
+ }
+}
- if (lck_spin_try_lock(kdw_spin_lock)) {
+/*
+ * 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 (kds_waiter && kd_ctrl_page.kds_inuse_count >= n_storage_threshold) {
- kds_waiter = 0;
- need_kds_wakeup = TRUE;
- }
- lck_spin_unlock(kdw_spin_lock);
+ if (!procfilt_flags) {
+ return true;
+ }
- ml_set_interrupts_enabled(s);
-
- if (need_kds_wakeup == TRUE)
- wakeup(&kds_waiter);
- }
+ /*
+ * 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;
}
-}
+ struct proc *curproc = current_proc();
+ /*
+ * If the process is missing (early in boot), allow it.
+ */
+ if (!curproc) {
+ return true;
+ }
+ 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();
+ }
+}
-void
-kernel_debug_internal(
- uint32_t debugid,
- uintptr_t arg1,
- uintptr_t arg2,
- uintptr_t arg3,
- uintptr_t arg4,
- uintptr_t arg5,
- int entropy_flag);
-
-__attribute__((always_inline)) void
+static void
kernel_debug_internal(
- uint32_t debugid,
- uintptr_t arg1,
- uintptr_t arg2,
- uintptr_t arg3,
- uintptr_t arg4,
- uintptr_t arg5,
- int entropy_flag)
-{
- struct proc *curproc;
- uint64_t now;
- uint32_t bindx;
- boolean_t s;
- kd_buf *kd;
- int cpu;
+ uint32_t debugid,
+ uintptr_t arg1,
+ uintptr_t arg2,
+ uintptr_t arg3,
+ uintptr_t arg4,
+ uintptr_t arg5,
+ uint64_t flags)
+{
+ 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))) {
+ goto out1;
}
- if ((kdebug_enable & KDEBUG_ENABLE_ENTROPY) && entropy_flag) {
-
- now = mach_absolute_time();
- s = ml_set_interrupts_enabled(FALSE);
- lck_spin_lock(kds_spin_lock);
-
- if (kdebug_enable & KDEBUG_ENABLE_ENTROPY) {
-
- if (kd_entropy_indx < kd_entropy_count) {
- kd_entropy_buffer[kd_entropy_indx] = now;
- kd_entropy_indx++;
- }
- if (kd_entropy_indx == kd_entropy_count) {
- /*
- * Disable entropy collection
- */
- kdebug_enable &= ~KDEBUG_ENABLE_ENTROPY;
- kd_ctrl_page.kdebug_slowcheck &= ~SLOW_ENTROPY;
- }
- }
- lck_spin_unlock(kds_spin_lock);
- ml_set_interrupts_enabled(s);
- }
- if ( (kd_ctrl_page.kdebug_slowcheck & SLOW_NOLOG) || !(kdebug_enable & (KDEBUG_ENABLE_TRACE|KDEBUG_ENABLE_PPT)))
- goto out1;
-
- 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 (EXTRACT_CLASS(debugid) == DBG_TRACE)
+ if (typefilter_is_debugid_allowed(kdbg_typefilter, debugid)) {
goto record_event;
+ }
- if (isset(type_filter_bitmap, EXTRACT_CSC(debugid)))
- goto record_event;
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 (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 (EXTRACT_CLASS(debugid) == DBG_TRACE)
+ if (KDBG_EXTRACT_CLASS(debugid) == DBG_TRACE) {
goto record_event;
-
- if ((debugid & DBG_FUNC_MASK) != kdlog_value1 &&
- (debugid & DBG_FUNC_MASK) != kdlog_value2 &&
- (debugid & DBG_FUNC_MASK) != kdlog_value3 &&
- (debugid & DBG_FUNC_MASK) != kdlog_value4)
+ }
+
+ if ((debugid & KDBG_EVENTID_MASK) != kdlog_value1 &&
+ (debugid & KDBG_EVENTID_MASK) != kdlog_value2 &&
+ (debugid & KDBG_EVENTID_MASK) != kdlog_value3 &&
+ (debugid & KDBG_EVENTID_MASK) != kdlog_value4) {
goto out1;
+ }
}
+ } else if (only_filter) {
+ goto out1;
}
+
record_event:
disable_preemption();
- if (kd_ctrl_page.enabled == 0)
+ if (kd_ctrl_page.enabled == 0) {
goto out;
+ }
cpu = cpu_number();
kdbp = &kdbip[cpu];
+
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
}
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];
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) ||
- (kde_waiter && kd_entropy_indx >= kd_entropy_count)) {
- uint32_t etype;
- uint32_t stype;
-
- etype = debugid & DBG_FUNC_MASK;
- stype = debugid & DBG_SCALL_MASK;
+ if (kds_waiter && kd_ctrl_page.kds_inuse_count >= n_storage_threshold) {
+ 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;
- boolean_t need_kde_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;
- }
- if (kde_waiter && kd_entropy_indx >= kd_entropy_count) {
- kde_waiter = 0;
- need_kde_wakeup = TRUE;
- }
- lck_spin_unlock(kdw_spin_lock);
- }
- ml_set_interrupts_enabled(s);
-
- if (need_kds_wakeup == TRUE)
- wakeup(&kds_waiter);
- if (need_kde_wakeup == TRUE)
- wakeup(&kde_waiter);
+ 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()), 1);
+ 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, 1);
+ kernel_debug_internal(debugid, arg1, arg2, arg3, arg4, arg5, 0);
}
-/*
- * Support syscall SYS_kdebug_trace
- */
-int
-kdebug_trace(__unused struct proc *p, struct kdebug_trace_args *uap, __unused int32_t *retval)
+__attribute__((noinline))
+void
+kernel_debug_flags(
+ uint32_t debugid,
+ uintptr_t arg1,
+ uintptr_t arg2,
+ uintptr_t arg3,
+ uintptr_t arg4,
+ uint64_t flags)
{
- if ( __probable(kdebug_enable == 0) )
- return(EINVAL);
-
- kernel_debug_internal(uap->code, uap->arg1, uap->arg2, uap->arg3, uap->arg4, (uintptr_t)thread_tid(current_thread()), 0);
-
- return(0);
+ kernel_debug_internal(debugid, arg1, arg2, arg3, arg4,
+ (uintptr_t)thread_tid(current_thread()), flags);
}
-
-static void
-kdbg_lock_init(void)
+__attribute__((noinline))
+void
+kernel_debug_filtered(
+ uint32_t debugid,
+ uintptr_t arg1,
+ uintptr_t arg2,
+ uintptr_t arg3,
+ uintptr_t arg4)
{
- 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();
-
-
- /*
- * 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);
-
- kd_ctrl_page.kdebug_flags |= KDBG_LOCKINIT;
+ kernel_debug_flags(debugid, arg1, arg2, arg3, arg4, KDBG_FLAG_FILTERED);
}
-
-int
-kdbg_bootstrap(boolean_t early_trace)
+void
+kernel_debug_string_early(const char *message)
{
- kd_ctrl_page.kdebug_flags &= ~KDBG_WRAPPED;
+ uintptr_t arg[4] = {0, 0, 0, 0};
- return (create_buffers(early_trace));
+ /* Stuff the message string in the args and log it. */
+ strncpy((char *)arg, message, MIN(sizeof(arg), strlen(message)));
+ KERNEL_DEBUG_EARLY(
+ TRACE_INFO_STRING,
+ arg[0], arg[1], arg[2], arg[3]);
}
-int
-kdbg_reinit(boolean_t early_trace)
+#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)
{
- int ret = 0;
+ if (!kdebug_enable) {
+ return;
+ }
- /*
- * Disable trace collecting
- * First make sure we're not in
- * the middle of cutting a trace
- */
- kdbg_set_tracing_enabled(FALSE, KDEBUG_ENABLE_TRACE);
+ /* 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);
- /*
- * make sure the SLOW_NOLOG is seen
- * by everyone that might be trying
- * to cut a trace..
- */
- IOSleep(100);
+ uintptr_t thread_id = (uintptr_t)thread_tid(current_thread());
+ uint32_t debugid = eventid | DBG_FUNC_START;
- delete_buffers();
+ /* string can fit in a single tracepoint */
+ if (len <= (4 * sizeof(uintptr_t))) {
+ debugid |= DBG_FUNC_END;
+ }
- 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;
- }
- ret = kdbg_bootstrap(early_trace);
+ kernel_debug_internal(debugid, str_buf[0],
+ str_buf[1],
+ str_buf[2],
+ str_buf[3], thread_id, 0);
- RAW_file_offset = 0;
- RAW_file_written = 0;
+ debugid &= KDBG_EVENTID_MASK;
+ int i = 4;
+ size_t written = 4 * sizeof(uintptr_t);
- return(ret);
+ 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
-kdbg_trace_data(struct proc *proc, long *arg_pid)
+kernel_debug_early(
+ uint32_t debugid,
+ uintptr_t arg1,
+ uintptr_t arg2,
+ uintptr_t arg3,
+ uintptr_t arg4)
{
- if (!proc)
- *arg_pid = 0;
- else
- *arg_pid = proc->p_pid;
-}
-
+#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
-void
-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 early tracing is over, use the normal path. */
+ if (kd_early_done) {
+ KDBG_RELEASE(debugid, arg1, arg2, arg3, arg4);
+ return;
+ }
- if (!proc) {
- *arg1 = 0;
- *arg2 = 0;
- *arg3 = 0;
- *arg4 = 0;
+ /* 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;
}
- /*
- * 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];
+
+ kd_early_buffer[kd_early_index].debugid = debugid;
+ kd_early_buffer[kd_early_index].timestamp = mach_absolute_time();
+ kd_early_buffer[kd_early_index].arg1 = arg1;
+ kd_early_buffer[kd_early_index].arg2 = arg2;
+ kd_early_buffer[kd_early_index].arg3 = arg3;
+ kd_early_buffer[kd_early_index].arg4 = arg4;
+ kd_early_buffer[kd_early_index].arg5 = 0;
+ kd_early_index++;
}
+/*
+ * 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
-kdbg_resolve_map(thread_t th_act, void *opaque)
+kernel_debug_early_end(void)
{
- kd_threadmap *mapptr;
- krt_t *t = (krt_t *)opaque;
+ if (cpu_number() != master_cpu) {
+ panic("kernel_debug_early_end() not call on boot processor");
+ }
- if (t->count < t->maxcount) {
- mapptr = &t->map[t->count];
- mapptr->thread = (uintptr_t)thread_tid(th_act);
+ /* 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,
+ 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);
+ }
- (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;
+ /* Cut events-lost event on overflow */
+ if (kd_early_overflow) {
+ KDBG_RELEASE(TRACE_LOST_EVENTS, 1);
+ }
- t->count++;
+ kd_early_done = true;
+
+ /* This trace marks the start of kernel tracing */
+ kernel_debug_string_early("early trace done");
+}
+
+void
+kernel_debug_disable(void)
+{
+ if (kdebug_enable) {
+ kdbg_set_tracing_enabled(false, 0);
}
}
/*
- *
- * Writes a cpumap for the given iops_list/cpu_count to the provided buffer.
- *
- * You may provide a buffer and size, or if you set the buffer to NULL, a
- * buffer of sufficient size will be allocated.
- *
- * If you provide a buffer and it is too small, sets cpumap_size to the number
- * of bytes required and returns EINVAL.
- *
- * On success, if you provided a buffer, cpumap_size is set to the number of
- * bytes written. If you did not provide a buffer, cpumap is set to the newly
- * allocated buffer and cpumap_size is set to the number of bytes allocated.
- *
- * NOTE: It may seem redundant to pass both iops and a cpu_count.
- *
- * We may be reporting data from "now", or from the "past".
- *
- * The "now" data would be for something like kdbg_readcurcpumap().
- * The "past" data would be for kdbg_readcpumap().
- *
- * If we do not pass both iops and cpu_count, and iops is NULL, this function
- * will need to read "now" state to get the number of cpus, which would be in
- * error if we were reporting "past" state.
+ * Returns non-zero if debugid is in a reserved class.
*/
-
-int
-kdbg_cpumap_init_internal(kd_iop_t* iops, uint32_t cpu_count, uint8_t** cpumap, uint32_t* cpumap_size)
+static int
+kdebug_validate_debugid(uint32_t debugid)
{
- assert(cpumap);
- assert(cpumap_size);
- assert(cpu_count);
- assert(!iops || iops->cpu_id + 1 == cpu_count);
+ uint8_t debugid_class;
- 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) != KERN_SUCCESS) {
- return ENOMEM;
- }
- } else if (bytes_available < bytes_needed) {
- return EINVAL;
+ debugid_class = KDBG_EXTRACT_CLASS(debugid);
+ switch (debugid_class) {
+ case DBG_TRACE:
+ return EPERM;
}
- kd_cpumap_header* header = (kd_cpumap_header*)(uintptr_t)*cpumap;
+ return 0;
+}
- header->version_no = RAW_VERSION1;
- header->cpu_count = cpu_count;
+/*
+ * Support syscall SYS_kdebug_typefilter.
+ */
+int
+kdebug_typefilter(__unused struct proc* p,
+ struct kdebug_typefilter_args* uap,
+ __unused int *retval)
+{
+ int ret = KERN_SUCCESS;
- kd_cpumap* cpus = (kd_cpumap*)&header[1];
+ if (uap->addr == USER_ADDR_NULL ||
+ uap->size == USER_ADDR_NULL) {
+ return EINVAL;
+ }
- int32_t index = cpu_count - 1;
- while (iops) {
- cpus[index].cpu_id = iops->cpu_id;
- cpus[index].flags = KDBG_CPUMAP_IS_IOP;
- bzero(cpus[index].name, sizeof(cpus->name));
- strlcpy(cpus[index].name, iops->callback.iop_name, sizeof(cpus->name));
-
- iops = iops->next;
- index--;
+ /*
+ * 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;
}
-
- 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--;
+ 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 KERN_SUCCESS;
+
+ return ret;
}
-void
-kdbg_thrmap_init(void)
+/*
+ * Support syscall SYS_kdebug_trace. U64->K32 args may get truncated in kdebug_trace64
+ */
+int
+kdebug_trace(struct proc *p, struct kdebug_trace_args *uap, int32_t *retval)
{
- if (kd_ctrl_page.kdebug_flags & KDBG_MAPINIT)
- return;
+ struct kdebug_trace64_args uap64;
- kd_mapptr = kdbg_thrmap_init_internal(0, &kd_mapsize, &kd_mapcount);
+ uap64.code = uap->code;
+ uap64.arg1 = uap->arg1;
+ uap64.arg2 = uap->arg2;
+ uap64.arg3 = uap->arg3;
+ uap64.arg4 = uap->arg4;
- if (kd_mapptr)
- kd_ctrl_page.kdebug_flags |= KDBG_MAPINIT;
+ return kdebug_trace64(p, &uap64, retval);
}
-
-kd_threadmap* kdbg_thrmap_init_internal(unsigned int count, unsigned int *mapsize, unsigned int *mapcount)
+/*
+ * 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)
{
- 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;
+ int err;
- /*
- * need to use PROC_SCANPROCLIST with proc_iterate
- */
- proc_list_lock();
+ if (__probable(kdebug_enable == 0)) {
+ return 0;
+ }
- /*
- * 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++;
+ if ((err = kdebug_validate_debugid(uap->code)) != 0) {
+ return err;
}
- 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;
+ 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;
+}
- *mapsize = *mapcount * sizeof(kd_threadmap);
+/*
+ * Adding enough padding to contain a full tracepoint for the last
+ * portion of the string greatly simplifies the logic of splitting the
+ * string between tracepoints. Full tracepoints can be generated using
+ * the buffer itself, without having to manually add zeros to pad the
+ * arguments.
+ */
- if (count && count < *mapcount)
- return (0);
+/* 2 string args in first tracepoint and 9 string data tracepoints */
+#define STR_BUF_ARGS (2 + (9 * 4))
+/* times the size of each arg on K64 */
+#define MAX_STR_LEN (STR_BUF_ARGS * sizeof(uint64_t))
+/* on K32, ending straddles a tracepoint, so reserve blanks */
+#define STR_BUF_SIZE (MAX_STR_LEN + (2 * sizeof(uint32_t)))
- if ((kmem_alloc(kernel_map, &kaddr, (vm_size_t)*mapsize) == KERN_SUCCESS)) {
- bzero((void *)kaddr, *mapsize);
- mapptr = (kd_threadmap *)kaddr;
- } else
- return (0);
+/*
+ * This function does no error checking and assumes that it is called with
+ * the correct arguments, including that the buffer pointed to by str is at
+ * least STR_BUF_SIZE bytes. However, str must be aligned to word-size and
+ * be NUL-terminated. In cases where a string can fit evenly into a final
+ * tracepoint without its NUL-terminator, this function will not end those
+ * strings with a NUL in trace. It's up to clients to look at the function
+ * qualifier for DBG_FUNC_END in this case, to end the string.
+ */
+static uint64_t
+kernel_debug_string_internal(uint32_t debugid, uint64_t str_id, void *vstr,
+ size_t str_len)
+{
+ /* str must be word-aligned */
+ uintptr_t *str = vstr;
+ size_t written = 0;
+ uintptr_t thread_id;
+ int i;
+ uint32_t trace_debugid = TRACEDBG_CODE(DBG_TRACE_STRING,
+ TRACE_STRING_GLOBAL);
- tts_mapsize = tts_count * sizeof(struct tts);
+ thread_id = (uintptr_t)thread_tid(current_thread());
- if ((kmem_alloc(kernel_map, &kaddr, (vm_size_t)tts_mapsize) == KERN_SUCCESS)) {
- bzero((void *)kaddr, tts_mapsize);
- tts_mapptr = (struct tts *)kaddr;
- } else {
- kmem_free(kernel_map, (vm_offset_t)mapptr, *mapsize);
+ /* 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, 0);
+ return str_id;
+ }
- return (0);
+ /* generate an ID, if necessary */
+ if (str_id == 0) {
+ str_id = OSIncrementAtomic64((SInt64 *)&g_curr_str_id);
+ str_id = (str_id & STR_ID_MASK) | g_str_id_signature;
}
- /*
- * We need to save the procs command string
- * and take a reference for each task associated
- * with a valid process
- */
- proc_list_lock();
+ trace_debugid |= DBG_FUNC_START;
+ /* string can fit in a single tracepoint */
+ if (str_len <= (2 * sizeof(uintptr_t))) {
+ trace_debugid |= DBG_FUNC_END;
+ }
- /*
- * should use proc_iterate
- */
- for (p = allproc.lh_first, i=0; p && i < tts_count; p = p->p_list.le_next) {
- if (p->p_lflag & P_LEXIT)
- continue;
+ kernel_debug_internal(trace_debugid, (uintptr_t)debugid, (uintptr_t)str_id,
+ str[0], str[1], thread_id, 0);
- 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++;
+ trace_debugid &= KDBG_EVENTID_MASK;
+ i = 2;
+ written += 2 * sizeof(uintptr_t);
+
+ for (; written < str_len; i += 4, written += 4 * sizeof(uintptr_t)) {
+ if ((written + (4 * sizeof(uintptr_t))) >= str_len) {
+ trace_debugid |= DBG_FUNC_END;
}
+ kernel_debug_internal(trace_debugid, str[i],
+ str[i + 1],
+ str[i + 2],
+ str[i + 3], thread_id, 0);
}
- tts_count = i;
+ return str_id;
+}
+
+/*
+ * Returns true if the current process can emit events, and false otherwise.
+ * Trace system and scheduling events circumvent this check, as do events
+ * emitted in interrupt context.
+ */
+static bool
+kdebug_current_proc_enabled(uint32_t debugid)
+{
+ /* can't determine current process in interrupt context */
+ if (ml_at_interrupt_context()) {
+ return true;
+ }
+
+ /* always emit trace system and scheduling events */
+ if ((KDBG_EXTRACT_CLASS(debugid) == DBG_TRACE ||
+ (debugid & KDBG_CSC_MASK) == MACHDBG_CODE(DBG_MACH_SCHED, 0))) {
+ return true;
+ }
+
+ if (kd_ctrl_page.kdebug_flags & KDBG_PIDCHECK) {
+ proc_t cur_proc = current_proc();
+
+ /* only the process with the kdebug bit set is allowed */
+ if (cur_proc && !(cur_proc->p_kdebug)) {
+ return false;
+ }
+ } else if (kd_ctrl_page.kdebug_flags & KDBG_PIDEXCLUDE) {
+ proc_t cur_proc = current_proc();
+
+ /* every process except the one with the kdebug bit set is allowed */
+ if (cur_proc && cur_proc->p_kdebug) {
+ return false;
+ }
+ }
+
+ return true;
+}
+
+bool
+kdebug_debugid_enabled(uint32_t debugid)
+{
+ /* if no filtering is enabled */
+ if (!kd_ctrl_page.kdebug_slowcheck) {
+ 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) {
+ 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) {
+ 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) {
+ return false;
+ }
+ }
+
+ return true;
+}
+
+bool
+kdebug_using_continuous_time(void)
+{
+ return kdebug_enable & KDEBUG_ENABLE_CONT_TIME;
+}
+
+/*
+ * Returns 0 if a string can be traced with these arguments. Returns errno
+ * value if error occurred.
+ */
+static errno_t
+kdebug_check_trace_string(uint32_t debugid, uint64_t str_id)
+{
+ /* if there are function qualifiers on the debugid */
+ if (debugid & ~KDBG_EVENTID_MASK) {
+ return EINVAL;
+ }
+
+ if (kdebug_validate_debugid(debugid)) {
+ return EPERM;
+ }
+
+ if (str_id != 0 && (str_id & STR_ID_SIG_MASK) != g_str_id_signature) {
+ return EINVAL;
+ }
+
+ return 0;
+}
+
+/*
+ * Implementation of KPI kernel_debug_string.
+ */
+int
+kernel_debug_string(uint32_t debugid, uint64_t *str_id, const char *str)
+{
+ /* arguments to tracepoints must be word-aligned */
+ __attribute__((aligned(sizeof(uintptr_t)))) char str_buf[STR_BUF_SIZE];
+ static_assert(sizeof(str_buf) > MAX_STR_LEN);
+ vm_size_t len_copied;
+ int err;
+
+ assert(str_id);
+
+ if (__probable(kdebug_enable == 0)) {
+ return 0;
+ }
+
+ if (!kdebug_current_proc_enabled(debugid)) {
+ return 0;
+ }
+
+ if (!kdebug_debugid_enabled(debugid)) {
+ return 0;
+ }
+
+ if ((err = kdebug_check_trace_string(debugid, *str_id)) != 0) {
+ return err;
+ }
+
+ if (str == NULL) {
+ if (str_id == 0) {
+ return EINVAL;
+ }
+
+ *str_id = kernel_debug_string_internal(debugid, *str_id, NULL, 0);
+ return 0;
+ }
+
+ memset(str_buf, 0, sizeof(str_buf));
+ len_copied = strlcpy(str_buf, str, MAX_STR_LEN + 1);
+ *str_id = kernel_debug_string_internal(debugid, *str_id, str_buf,
+ len_copied);
+ return 0;
+}
+
+/*
+ * Support syscall kdebug_trace_string.
+ */
+int
+kdebug_trace_string(__unused struct proc *p,
+ struct kdebug_trace_string_args *uap,
+ uint64_t *retval)
+{
+ __attribute__((aligned(sizeof(uintptr_t)))) char str_buf[STR_BUF_SIZE];
+ static_assert(sizeof(str_buf) > MAX_STR_LEN);
+ size_t len_copied;
+ int err;
+
+ if (__probable(kdebug_enable == 0)) {
+ return 0;
+ }
+
+ if (!kdebug_current_proc_enabled(uap->debugid)) {
+ return 0;
+ }
+
+ if (!kdebug_debugid_enabled(uap->debugid)) {
+ return 0;
+ }
+
+ if ((err = kdebug_check_trace_string(uap->debugid, uap->str_id)) != 0) {
+ return err;
+ }
+
+ if (uap->str == USER_ADDR_NULL) {
+ if (uap->str_id == 0) {
+ return EINVAL;
+ }
+
+ *retval = kernel_debug_string_internal(uap->debugid, uap->str_id,
+ NULL, 0);
+ return 0;
+ }
+
+ memset(str_buf, 0, sizeof(str_buf));
+ err = copyinstr(uap->str, str_buf, MAX_STR_LEN + 1, &len_copied);
+
+ /* it's alright to truncate the string, so allow ENAMETOOLONG */
+ if (err == ENAMETOOLONG) {
+ str_buf[MAX_STR_LEN] = '\0';
+ } else if (err) {
+ return err;
+ }
+
+ if (len_copied <= 1) {
+ return EINVAL;
+ }
+
+ /* convert back to a length */
+ len_copied--;
+
+ *retval = kernel_debug_string_internal(uap->debugid, uap->str_id, str_buf,
+ len_copied);
+ return 0;
+}
+
+static void
+kdbg_lock_init(void)
+{
+ 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;
+ }
+
+ 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();
+
+ 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(bool early_trace)
+{
+ kd_ctrl_page.kdebug_flags &= ~KDBG_WRAPPED;
+
+ return create_buffers(early_trace);
+}
+
+int
+kdbg_reinit(bool early_trace)
+{
+ int ret = 0;
+
+ /*
+ * Disable trace collecting
+ * First make sure we're not in
+ * the middle of cutting a trace
+ */
+ kernel_debug_disable();
+
+ /*
+ * make sure the SLOW_NOLOG is seen
+ * by everyone that might be trying
+ * to cut a trace..
+ */
+ IOSleep(100);
+
+ delete_buffers();
+
+ kdbg_clear_thread_map();
+ ret = kdbg_bootstrap(early_trace);
+
+ RAW_file_offset = 0;
+ RAW_file_written = 0;
+
+ return ret;
+}
+
+void
+kdbg_trace_data(struct proc *proc, long *arg_pid, long *arg_uniqueid)
+{
+ if (!proc) {
+ *arg_pid = 0;
+ *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)
+{
+ if (!proc) {
+ *arg1 = 0;
+ *arg2 = 0;
+ *arg3 = 0;
+ *arg4 = 0;
+ return;
+ }
+
+ const char *procname = proc_best_name(proc);
+ size_t namelen = strlen(procname);
+
+ long args[4] = { 0 };
+
+ if (namelen > sizeof(args)) {
+ namelen = sizeof(args);
+ }
+
+ strncpy((char *)args, procname, namelen);
+
+ *arg1 = args[0];
+ *arg2 = args[1];
+ *arg3 = args[2];
+ *arg4 = args[3];
+}
+
+/*
+ *
+ * Writes a cpumap for the given iops_list/cpu_count to the provided buffer.
+ *
+ * You may provide a buffer and size, or if you set the buffer to NULL, a
+ * buffer of sufficient size will be allocated.
+ *
+ * If you provide a buffer and it is too small, sets cpumap_size to the number
+ * of bytes required and returns EINVAL.
+ *
+ * On success, if you provided a buffer, cpumap_size is set to the number of
+ * bytes written. If you did not provide a buffer, cpumap is set to the newly
+ * allocated buffer and cpumap_size is set to the number of bytes allocated.
+ *
+ * NOTE: It may seem redundant to pass both iops and a cpu_count.
+ *
+ * We may be reporting data from "now", or from the "past".
+ *
+ * The "past" data would be for kdbg_readcpumap().
+ *
+ * If we do not pass both iops and cpu_count, and iops is NULL, this function
+ * will need to read "now" state to get the number of cpus, which would be in
+ * error if we were reporting "past" state.
+ */
+
+int
+kdbg_cpumap_init_internal(kd_iop_t* iops, uint32_t cpu_count, uint8_t** cpumap, uint32_t* cpumap_size)
+{
+ assert(cpumap);
+ assert(cpumap_size);
+ assert(cpu_count);
+ assert(!iops || iops->cpu_id + 1 == cpu_count);
+
+ uint32_t bytes_needed = sizeof(kd_cpumap_header) + cpu_count * sizeof(kd_cpumap);
+ uint32_t bytes_available = *cpumap_size;
+ *cpumap_size = bytes_needed;
+
+ if (*cpumap == NULL) {
+ if (kmem_alloc(kernel_map, (vm_offset_t*)cpumap, (vm_size_t)*cpumap_size, VM_KERN_MEMORY_DIAG) != KERN_SUCCESS) {
+ return ENOMEM;
+ }
+ bzero(*cpumap, *cpumap_size);
+ } else if (bytes_available < bytes_needed) {
+ return EINVAL;
+ }
+
+ kd_cpumap_header* header = (kd_cpumap_header*)(uintptr_t)*cpumap;
+
+ header->version_no = RAW_VERSION1;
+ header->cpu_count = cpu_count;
+
+ kd_cpumap* cpus = (kd_cpumap*)&header[1];
+
+ int32_t index = cpu_count - 1;
+ while (iops) {
+ cpus[index].cpu_id = iops->cpu_id;
+ cpus[index].flags = KDBG_CPUMAP_IS_IOP;
+ 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;
+ strlcpy(cpus[index].name, "AP", sizeof(cpus->name));
+
+ index--;
+ }
+
+ return KERN_SUCCESS;
+}
+
+void
+kdbg_thrmap_init(void)
+{
+ 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) {
+ 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++;
+ }
+}
+
+static vm_size_t
+kd_resolve_tasks(struct kd_task_name *task_names, vm_size_t ntasks)
+{
+ vm_size_t i = 0;
+ proc_t p = PROC_NULL;
+
+ proc_list_lock();
+ 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();
+ return i;
+}
+
+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,
+ };
+
+ 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);
+ }
+
+ return resolver.krs_count;
+}
+
+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;
+
+ assert(mapsize != NULL);
+ assert(mapcount != NULL);
+
+ vm_size_t nthreads = threads_count;
+ vm_size_t ntasks = tasks_count;
+
/*
- * Initialize thread map data
+ * Allow 25% more threads and tasks to be created between now and taking the
+ * proc_list_lock.
*/
- 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 (os_add_overflow(nthreads, nthreads / 4, &nthreads) ||
+ os_add_overflow(ntasks, ntasks / 4, &ntasks)) {
+ return NULL;
}
- kmem_free(kernel_map, (vm_offset_t)tts_mapptr, tts_mapsize);
- *mapcount = akrt.count;
+ *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;
+ }
+
+ /*
+ * Wait until the out-parameters have been filled with the needed size to
+ * do the bounds checking on the provided maximum.
+ */
+ if (maxthreads != 0 && maxthreads < nthreads) {
+ return NULL;
+ }
- 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
kdbg_clear(void)
{
- /*
+ /*
* Clean up the trace buffer
* First make sure we're not in
* the middle of cutting a trace
*/
- kdbg_set_tracing_enabled(FALSE, KDEBUG_ENABLE_TRACE);
+ kernel_debug_disable();
+ kdbg_disable_typefilter();
/*
* make sure the SLOW_NOLOG is seen
*/
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_disable_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 {
* 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 */
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;
+}
+
+/*
+ * The following functions all operate on the "global" typefilter singleton.
+ */
+
+/*
+ * 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.
+ */
+static int
+kdbg_initialize_typefilter(typefilter_t tf)
+{
+ ktrace_assert_lock_held();
+ assert(!kdbg_typefilter);
+ assert(!kdbg_typefilter_memory_entry);
+ typefilter_t deallocate_tf = NULL;
+
+ if (!tf && ((tf = deallocate_tf = typefilter_create()) == NULL)) {
+ return ENOMEM;
+ }
+
+ if ((kdbg_typefilter_memory_entry = typefilter_create_memory_entry(tf)) == MACH_PORT_NULL) {
+ if (deallocate_tf) {
+ typefilter_deallocate(deallocate_tf);
+ }
+ return ENOMEM;
+ }
+
+ /*
+ * 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;
+}
+
+static int
+kdbg_copyin_typefilter(user_addr_t addr, size_t size)
+{
+ int ret = ENOMEM;
+ typefilter_t tf;
+
+ ktrace_assert_lock_held();
- return(ret);
-}
+ if (size != KDBG_TYPEFILTER_BITMAP_SIZE) {
+ return EINVAL;
+ }
+ 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);
-/*
- * This is for setting a maximum decrementer value
- */
-int
-kdbg_setrtcdec(kd_regtype *kdr)
-{
- int ret = 0;
- natural_t decval;
+ /*
+ * 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);
+ }
- decval = (natural_t)kdr->value1;
+ kdbg_enable_typefilter();
+ kdbg_iop_list_callback(kd_ctrl_page.kdebug_iops, KD_CALLBACK_TYPEFILTER_CHANGED, kdbg_typefilter);
+ }
- if (decval && decval < KDBG_MINRTCDEC)
- ret = EINVAL;
- else
- ret = ENOTSUP;
+ if (tf) {
+ typefilter_deallocate(tf);
+ }
+ }
- return(ret);
+ return ret;
}
-int
+/*
+ * 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 (kd_ctrl_page.kdebug_flags & KDBG_TYPEFILTER_CHECK) {
- /* free the old filter */
- kdbg_disable_typefilter();
- }
-
- if (kmem_alloc(kernel_map, (vm_offset_t *)&type_filter_bitmap, KDBG_TYPEFILTER_BITMAP_SIZE) != KERN_SUCCESS) {
- return ENOSPC;
- }
-
- bzero(type_filter_bitmap, KDBG_TYPEFILTER_BITMAP_SIZE);
-
- /* Turn off range and value checks */
+ assert(kdbg_typefilter);
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;
+ kdbg_set_flags(SLOW_CHECKS, 0, true);
+ commpage_update_kdebug_state();
}
-int
+/*
+ * 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)
{
- /* Disable filter checking */
+ bool notify_iops = kd_ctrl_page.kdebug_flags & KDBG_TYPEFILTER_CHECK;
kd_ctrl_page.kdebug_flags &= ~KDBG_TYPEFILTER_CHECK;
-
- /* Turn off slow checks unless pid checks are using them */
- if ( (kd_ctrl_page.kdebug_flags & (KDBG_PIDCHECK | KDBG_PIDEXCLUDE)) )
- kdbg_set_flags(SLOW_CHECKS, 0, TRUE);
- else
- kdbg_set_flags(SLOW_CHECKS, 0, FALSE);
-
- if(type_filter_bitmap == NULL)
- return 0;
- vm_offset_t old_bitmap = (vm_offset_t)type_filter_bitmap;
- type_filter_bitmap = NULL;
+ 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);
+ }
+ commpage_update_kdebug_state();
+
+ 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);
+ }
+}
+
+uint32_t
+kdebug_commpage_state(void)
+{
+ 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;
+ }
- kmem_free(kernel_map, old_bitmap, KDBG_TYPEFILTER_BITMAP_SIZE);
return 0;
}
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);
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)
+{
+ 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_getreg(__unused kd_regtype * kdr)
+kdbg_write_v3_chunk_header(user_addr_t buffer, uint32_t tag, uint32_t sub_tag, uint64_t length, vnode_t vp, vfs_context_t ctx)
{
-#if 0
- int i,j, ret=0;
- unsigned int val_1, val_2, val;
+ int ret = KERN_SUCCESS;
+ kd_chunk_header_v3 header = {
+ .tag = tag,
+ .sub_tag = sub_tag,
+ .length = length,
+ };
- switch (kdr->type) {
- case KDBG_CLASSTYPE :
- val_1 = (kdr->value1 & 0xff);
- val_2 = val_1 + 1;
- kdlog_beg = (val_1<<24);
- kdlog_end = (val_2<<24);
- kd_ctrl_page.kdebug_flags &= (unsigned int)~KDBG_CKTYPES;
- kd_ctrl_page.kdebug_flags |= (KDBG_RANGECHECK | KDBG_CLASSTYPE);
- break;
- case KDBG_SUBCLSTYPE :
- val_1 = (kdr->value1 & 0xff);
- val_2 = (kdr->value2 & 0xff);
- val = val_2 + 1;
- kdlog_beg = ((val_1<<24) | (val_2 << 16));
- kdlog_end = ((val_1<<24) | (val << 16));
- kd_ctrl_page.kdebug_flags &= (unsigned int)~KDBG_CKTYPES;
- kd_ctrl_page.kdebug_flags |= (KDBG_RANGECHECK | KDBG_SUBCLSTYPE);
- break;
- case KDBG_RANGETYPE :
- kdlog_beg = (kdr->value1);
- kdlog_end = (kdr->value2);
- kd_ctrl_page.kdebug_flags &= (unsigned int)~KDBG_CKTYPES;
- kd_ctrl_page.kdebug_flags |= (KDBG_RANGECHECK | KDBG_RANGETYPE);
- break;
- case KDBG_TYPENONE :
- kd_ctrl_page.kdebug_flags &= (unsigned int)~KDBG_CKTYPES;
- kdlog_beg = 0;
- kdlog_end = 0;
- break;
- default :
+ // Check that only one of them is valid
+ assert(!buffer ^ !vp);
+ assert((vp == NULL) || (ctx != NULL));
+
+ // Write the 8-byte future_chunk_timestamp field in the payload
+ if (buffer || vp) {
+ if (vp) {
+ ret = kdbg_write_to_vnode((caddr_t)&header, sizeof(kd_chunk_header_v3), vp, ctx, RAW_file_offset);
+ if (ret) {
+ goto write_error;
+ }
+ RAW_file_offset += (sizeof(kd_chunk_header_v3));
+ } else {
+ ret = copyout(&header, buffer, sizeof(kd_chunk_header_v3));
+ if (ret) {
+ goto write_error;
+ }
+ }
+ }
+write_error:
+ return ret;
+}
+
+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;
+ struct vfs_context context;
+ struct fileproc *fp;
+ vnode_t vp;
+ p = current_proc();
+
+ 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->fp_glob->fg_cred;
+
+ 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,
+ };
+
+ int ret = kdbg_write_to_vnode((caddr_t) &chunk_header, sizeof(kd_chunk_header_v3), vp, &context, RAW_file_offset);
+ if (!ret) {
+ RAW_file_offset += sizeof(kd_chunk_header_v3);
+ }
+
+ ret = kdbg_write_to_vnode((caddr_t) payload, (size_t) payload_size, vp, &context, RAW_file_offset);
+ if (!ret) {
+ RAW_file_offset += payload_size;
+ }
+
+ fp->fp_glob->fg_offset = RAW_file_offset;
+ vnode_put(vp);
+ }
+
+ fp_drop(p, fd, fp, 0);
+ return KERN_SUCCESS;
+}
+
+user_addr_t
+kdbg_write_v3_event_chunk_header(user_addr_t buffer, uint32_t tag, uint64_t length, vnode_t vp, vfs_context_t ctx)
+{
+ uint64_t future_chunk_timestamp = 0;
+ length += sizeof(uint64_t);
+
+ if (kdbg_write_v3_chunk_header(buffer, tag, V3_EVENT_DATA_VERSION, length, vp, ctx)) {
+ return 0;
+ }
+ if (buffer) {
+ buffer += sizeof(kd_chunk_header_v3);
+ }
+
+ // Check that only one of them is valid
+ assert(!buffer ^ !vp);
+ assert((vp == NULL) || (ctx != NULL));
+
+ // Write the 8-byte future_chunk_timestamp field in the payload
+ if (buffer || vp) {
+ if (vp) {
+ int ret = kdbg_write_to_vnode((caddr_t)&future_chunk_timestamp, sizeof(uint64_t), vp, ctx, RAW_file_offset);
+ if (!ret) {
+ RAW_file_offset += (sizeof(uint64_t));
+ }
+ } else {
+ if (copyout(&future_chunk_timestamp, buffer, sizeof(uint64_t))) {
+ return 0;
+ }
+ }
+ }
+
+ return buffer + sizeof(uint64_t);
+}
+
+int
+kdbg_write_v3_header(user_addr_t user_header, size_t *user_header_size, int fd)
+{
+ int ret = KERN_SUCCESS;
+
+ 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;
- break;
+ 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
+ .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;
}
-#endif /* 0 */
- return(EINVAL);
+
+ *user_header_size = bytes_needed;
+bail:
+ if (cpumap) {
+ kmem_free(kernel_map, (vm_offset_t)cpumap, cpumap_size);
+ }
+ return ret;
}
int
}
*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;
}
-int
-kdbg_readthrmap(user_addr_t buffer, size_t *number, vnode_t vp, vfs_context_t ctx)
+static int
+kdbg_write_v1_header(bool write_thread_map, vnode_t vp, vfs_context_t ctx)
{
- int avail = *number;
int ret = 0;
- uint32_t count = 0;
- unsigned int mapsize;
+ 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;
+ 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;
+ }
+ }
+
+ /*
+ * 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;
+ }
- count = avail/sizeof (kd_threadmap);
+ /*
+ * 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.
+ */
- mapsize = kd_mapcount * sizeof(kd_threadmap);
+ assert(vp);
+ assert(ctx);
- 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)
- {
- 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;
-
- /*
- * To write a RAW_VERSION1+ file, we
- * must embed a cpumap in the "padding"
- * used to page align the events folloing
- * the threadmap. If the threadmap happens
- * to not require enough padding, we
- * artificially increase its footprint
- * until it needs enough padding.
- */
+ 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) {
+ /* If the cpu map doesn't fit in the current available pad_size,
+ * we increase the pad_size by 16K. We do this so that the event
+ * data is always available on a page aligned boundary for both
+ * 4k and 16k systems. We enforce this alignment for the event
+ * data so that we can take advantage of optimized file/disk writes.
+ */
+ pad_size += PAGE_16KB;
+ }
+
+ /* The way we are silently embedding a cpumap in the "padding" is by artificially
+ * increasing the number of thread entries. However, we'll also need to ensure that
+ * the cpumap is embedded in the last 4K page before when the event data is expected.
+ * This way the tools can read the data starting the next page boundary on both
+ * 4K and 16K systems preserving compatibility with older versions of the tools
+ */
+ if (pad_size > PAGE_4KB) {
+ pad_size -= PAGE_4KB;
+ extra_thread_count = (pad_size / sizeof(kd_threadmap)) + 1;
+ }
- pad_size = PAGE_SIZE - ((sizeof(RAW_header) + (count * sizeof(kd_threadmap))) & PAGE_MASK_64);
- cpumap_size = sizeof(kd_cpumap_header) + kd_ctrl_page.kdebug_cpus * sizeof(kd_cpumap);
+ memset(&header, 0, sizeof(header));
+ header.version_no = RAW_VERSION1;
+ header.thread_count = map_count + extra_thread_count;
- if (cpumap_size > pad_size) {
- /* Force an overflow onto the next page, we get a full page of padding */
- extra_thread_count = (pad_size / sizeof(kd_threadmap)) + 1;
- }
+ clock_get_calendar_microtime(&secs, &usecs);
+ header.TOD_secs = secs;
+ header.TOD_usecs = usecs;
+
+ 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);
+
+ 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;
+ }
- header.version_no = RAW_VERSION1;
- header.thread_count = count + extra_thread_count;
+ RAW_file_offset += map_size;
+ RAW_file_written += map_size;
+ }
- clock_get_calendar_microtime(&secs, &usecs);
- header.TOD_secs = secs;
- header.TOD_usecs = usecs;
+ if (extra_thread_count) {
+ pad_size = extra_thread_count * sizeof(kd_threadmap);
+ pad_buf = kheap_alloc(KHEAP_TEMP, pad_size, Z_WAITOK | Z_ZERO);
+ if (!pad_buf) {
+ ret = ENOMEM;
+ goto write_error;
+ }
- ret = vn_rdwr(UIO_WRITE, vp, (caddr_t)&header, sizeof(RAW_header), RAW_file_offset,
- UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx));
- if (ret)
- goto write_error;
- RAW_file_offset += sizeof(RAW_header);
+ 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;
+ }
- 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;
+ RAW_file_offset += pad_size;
+ RAW_file_written += pad_size;
+ }
- if (extra_thread_count) {
- pad_size = extra_thread_count * sizeof(kd_threadmap);
- pad_buf = (char *)kalloc(pad_size);
- memset(pad_buf, 0, pad_size);
+ pad_size = PAGE_SIZE - (RAW_file_offset & PAGE_MASK);
+ if (pad_size) {
+ pad_buf = (char *)kheap_alloc(KHEAP_TEMP, pad_size, Z_WAITOK | Z_ZERO);
+ if (!pad_buf) {
+ ret = ENOMEM;
+ goto write_error;
+ }
- 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);
+ /*
+ * embed a cpumap in the padding bytes.
+ * older code will skip this.
+ * newer code will know how to read it.
+ */
+ uint32_t temp = pad_size;
+ if (kdbg_cpumap_init_internal(kd_ctrl_page.kdebug_iops, kd_ctrl_page.kdebug_cpus, (uint8_t**)&pad_buf, &temp) != KERN_SUCCESS) {
+ memset(pad_buf, 0, pad_size);
+ }
- if (ret)
- goto write_error;
- RAW_file_offset += pad_size;
+ 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;
+ }
- }
-
- pad_size = PAGE_SIZE - (RAW_file_offset & PAGE_MASK_64);
- if (pad_size) {
- pad_buf = (char *)kalloc(pad_size);
- memset(pad_buf, 0, pad_size);
+ RAW_file_offset += pad_size;
+ RAW_file_written += pad_size;
+ }
- /*
- * embed a cpumap in the padding bytes.
- * older code will skip this.
- * newer code will know how to read it.
- */
- uint32_t temp = pad_size;
- if (kdbg_cpumap_init_internal(kd_ctrl_page.kdebug_iops, kd_ctrl_page.kdebug_cpus, (uint8_t**)&pad_buf, &temp) != KERN_SUCCESS) {
- memset(pad_buf, 0, pad_size);
- }
+write_error:
+ return ret;
+}
+
+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;
+ }
+}
+
+/*
+ * 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;
+ bool map_initialized;
+
+ 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;
+ }
+ }
- 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);
+ 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;
- if (ret)
- goto write_error;
- RAW_file_offset += pad_size;
- }
- RAW_file_written += sizeof(RAW_header) + mapsize + pad_size;
+ ktrace_assert_lock_held();
+ assert(buffer_size != NULL);
- } 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;
+ }
- vn_rdwr(UIO_WRITE, vp, (caddr_t)&count, sizeof(uint32_t), RAW_file_offset,
- UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx));
- RAW_file_offset += sizeof(uint32_t);
- RAW_file_written += sizeof(uint32_t);
+ ret = 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_getentropy (user_addr_t buffer, size_t *number, int ms_timeout)
+kdbg_readthrmap_v3(user_addr_t buffer, size_t buffer_size, int fd)
{
- int avail = *number;
int ret = 0;
- int s;
- u_int64_t abstime;
- u_int64_t ns;
- int wait_result = THREAD_AWAKENED;
-
+ bool map_initialized;
+ size_t map_size;
- if (kd_entropy_buffer)
- return(EBUSY);
+ ktrace_assert_lock_held();
- if (ms_timeout < 0)
- return(EINVAL);
+ if ((!fd && !buffer) || (fd && buffer)) {
+ return EINVAL;
+ }
- kd_entropy_count = avail/sizeof(uint64_t);
+ map_initialized = (kd_ctrl_page.kdebug_flags & KDBG_MAPINIT);
+ map_size = kd_mapcount * sizeof(kd_threadmap);
- if (kd_entropy_count > MAX_ENTROPY_COUNT || kd_entropy_count == 0) {
- /*
- * Enforce maximum entropy entries
- */
- return(EINVAL);
- }
- kd_entropy_bufsize = kd_entropy_count * sizeof(uint64_t);
+ if (map_initialized && (buffer_size >= map_size)) {
+ ret = kdbg_write_v3_header(buffer, &buffer_size, fd);
- /*
- * allocate entropy buffer
- */
- if (kmem_alloc(kernel_map, &kd_entropy_buftomem, (vm_size_t)kd_entropy_bufsize) == KERN_SUCCESS) {
- kd_entropy_buffer = (uint64_t *) kd_entropy_buftomem;
+ if (ret == 0) {
+ kdbg_clear_thread_map();
+ }
} else {
- kd_entropy_buffer = (uint64_t *) 0;
- kd_entropy_count = 0;
-
- return (ENOMEM);
+ ret = EINVAL;
}
- kd_entropy_indx = 0;
- KERNEL_DEBUG_CONSTANT(0xbbbbf000 | DBG_FUNC_START, ms_timeout, kd_entropy_count, 0, 0, 0);
+ return ret;
+}
+static void
+kdbg_set_nkdbufs(unsigned int req_nkdbufs)
+{
/*
- * Enable entropy sampling
+ * Only allow allocation up to half the available memory (sane_size).
*/
- kdbg_set_flags(SLOW_ENTROPY, KDEBUG_ENABLE_ENTROPY, TRUE);
-
- if (ms_timeout) {
- ns = (u_int64_t)ms_timeout * (u_int64_t)(1000 * 1000);
- nanoseconds_to_absolutetime(ns, &abstime );
- clock_absolutetime_interval_to_deadline( abstime, &abstime );
- } else
- abstime = 0;
-
- s = ml_set_interrupts_enabled(FALSE);
- lck_spin_lock(kdw_spin_lock);
+ uint64_t max_nkdbufs = (sane_size / 2) / sizeof(kd_buf);
+ nkdbufs = (req_nkdbufs > max_nkdbufs) ? (unsigned int)max_nkdbufs :
+ req_nkdbufs;
+}
- while (wait_result == THREAD_AWAKENED && kd_entropy_indx < kd_entropy_count) {
+/*
+ * 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 wait_result = THREAD_AWAKENED;
+ uint64_t abstime = 0;
- kde_waiter = 1;
+ ktrace_assert_lock_held();
- if (abstime) {
- /*
- * wait for the specified timeout or
- * until we've hit our sample limit
- */
- wait_result = lck_spin_sleep_deadline(kdw_spin_lock, 0, &kde_waiter, THREAD_ABORTSAFE, abstime);
- } else {
- /*
- * wait until we've hit our sample limit
- */
- wait_result = lck_spin_sleep(kdw_spin_lock, 0, &kde_waiter, THREAD_ABORTSAFE);
- }
- kde_waiter = 0;
+ if (timeout_ms != 0) {
+ uint64_t ns = timeout_ms * NSEC_PER_MSEC;
+ nanoseconds_to_absolutetime(ns, &abstime);
+ clock_absolutetime_interval_to_deadline(abstime, &abstime);
}
- lck_spin_unlock(kdw_spin_lock);
- ml_set_interrupts_enabled(s);
-
- /*
- * Disable entropy sampling
- */
- kdbg_set_flags(SLOW_ENTROPY, KDEBUG_ENABLE_ENTROPY, FALSE);
-
- KERNEL_DEBUG_CONSTANT(0xbbbbf000 | DBG_FUNC_END, ms_timeout, kd_entropy_indx, 0, 0, 0);
- *number = 0;
- ret = 0;
+ 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 (kd_entropy_indx > 0) {
- /*
- * copyout the buffer
- */
- if (copyout(kd_entropy_buffer, buffer, kd_entropy_indx * sizeof(uint64_t)))
- ret = EINVAL;
- else
- *number = kd_entropy_indx * sizeof(uint64_t);
+ if (!locked_wait) {
+ /* drop the mutex to allow others to access trace */
+ ktrace_unlock();
}
- /*
- * Always cleanup
- */
- kd_entropy_count = 0;
- kd_entropy_indx = 0;
- kd_entropy_buftomem = 0;
- kmem_free(kernel_map, (vm_offset_t)kd_entropy_buffer, kd_entropy_bufsize);
- kd_entropy_buffer = (uint64_t *) 0;
-
- return(ret);
-}
+ while (wait_result == THREAD_AWAKENED &&
+ kd_ctrl_page.kds_inuse_count < n_storage_threshold) {
+ kds_waiter = 1;
-static int
-kdbg_set_nkdbufs(unsigned int value)
-{
- /*
- * 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
- */
- unsigned int max_entries = (sane_size/2) / sizeof(kd_buf);
+ 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);
+ }
- if (value <= max_entries)
- return (value);
- else
- return (max_entries);
-}
+ kds_waiter = 0;
+ }
+ /* check the count under the spinlock */
+ bool threshold_exceeded = (kd_ctrl_page.kds_inuse_count >= n_storage_threshold);
-static int
-kdbg_enable_bg_trace(void)
-{
- int ret = 0;
+ lck_spin_unlock(kdw_spin_lock);
+ ml_set_interrupts_enabled(s);
- 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;
- }
+ if (!locked_wait) {
+ /* pick the mutex back up again */
+ ktrace_lock();
}
- return ret;
+
+ /* 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_disable_bg_trace(void)
+kdbg_wakeup(void)
{
- if (kdlog_bg_trace_running == TRUE) {
- kdlog_bg_trace_running = FALSE;
- kdbg_clear();
- }
-}
+ 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);
+ }
-/*
- * 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
-*/
+ ml_set_interrupts_enabled(s);
-void
-kdbg_control_chud(int val, void *fn)
-{
- kdbg_lock_init();
-
- if (val) {
- /* enable chudhook */
- kdebug_chudhook = fn;
- kdbg_set_flags(SLOW_CHUD, KDEBUG_ENABLE_CHUD, TRUE);
- }
- else {
- /* disable chudhook */
- kdbg_set_flags(SLOW_CHUD, KDEBUG_ENABLE_CHUD, FALSE);
- kdebug_chudhook = 0;
+ if (need_kds_wakeup == true) {
+ wakeup(&kds_waiter);
}
}
-
int
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_KDWRITEMAP ||
- 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;
- break;
-
- case KERN_KDGETENTROPY:
- if (kd_entropy_buffer)
- ret = EBUSY;
- else
- ret = kdbg_getentropy(where, sizep, value);
- goto out;
- break;
-
- case KERN_KDENABLE_BG_TRACE:
- bg_nkdbufs = kdbg_set_nkdbufs(value);
- kdlog_bg_trace = TRUE;
- ret = kdbg_enable_bg_trace();
+ /*
+ * 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;
- break;
-
- case KERN_KDDISABLE_BG_TRACE:
- kdlog_bg_trace = FALSE;
- kdbg_disable_bg_trace();
+ }
+ } else {
+ if ((ret = ktrace_read_check())) {
goto out;
- break;
- }
-
- 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) {
- /*
- * The global pid no longer exists
- */
- global_state_pid = curpid;
- } else {
+
+ switch (name[0]) {
+ case KERN_KDGETBUF:
+ if (size < sizeof(kd_bufinfo.nkdbufs)) {
/*
- * The global pid exists, deny this request
+ * There is not enough room to return even
+ * the first element of the info structure.
*/
- proc_rele(p);
-
- ret = EBUSY;
- goto out;
+ ret = EINVAL;
+ break;
}
- }
- switch(name[0]) {
- case KERN_KDEFLAGS:
- kdbg_disable_bg_trace();
+ memset(&kd_bufinfo, 0, sizeof(kd_bufinfo));
- value &= KDBG_USERFLAGS;
- kd_ctrl_page.kdebug_flags |= value;
- break;
- case KERN_KDDFLAGS:
- kdbg_disable_bg_trace();
+ 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;
+ }
- value &= KDBG_USERFLAGS;
- kd_ctrl_page.kdebug_flags &= ~value;
- break;
- case KERN_KDENABLE:
+ 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 (size >= sizeof(kd_bufinfo)) {
/*
- * Enable tracing mechanism. Two types:
- * KDEBUG_TRACE is the standard one,
- * and KDEBUG_PPT which is a carefully
- * chosen subset to avoid performance impact.
+ * Provide all the info we have
*/
- 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 (copyout(&kd_bufinfo, where, sizeof(kd_bufinfo))) {
ret = EINVAL;
- break;
}
- if (copyin(where, &kd_Reg, sizeof(kd_regtype))) {
+ } else {
+ /*
+ * For backwards compatibility, only provide
+ * as much info as there is room for.
+ */
+ if (copyout(&kd_bufinfo, where, size)) {
ret = EINVAL;
- break;
}
- kdbg_disable_bg_trace();
+ }
+ break;
- ret = kdbg_setreg(&kd_Reg);
- break;
- case KERN_KDGETREG:
- if (size < sizeof(kd_regtype)) {
- ret = EINVAL;
- break;
- }
- ret = kdbg_getreg(&kd_Reg);
- if (copyout(&kd_Reg, where, sizeof(kd_regtype))) {
- ret = EINVAL;
- }
- kdbg_disable_bg_trace();
+ case KERN_KDREADCURTHRMAP:
+ ret = kdbg_readcurthrmap(where, sizep);
+ break;
- break;
- case KERN_KDREADTR:
- ret = kdbg_read(where, sizep, NULL, NULL);
- break;
- case KERN_KDWRITETR:
- case KERN_KDWRITEMAP:
- {
- struct vfs_context context;
- struct fileproc *fp;
- size_t number;
- vnode_t vp;
- int fd;
-
- kdbg_disable_bg_trace();
-
- if (name[0] == KERN_KDWRITETR) {
- 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;
+ case KERN_KDEFLAGS:
+ value &= KDBG_USERFLAGS;
+ kd_ctrl_page.kdebug_flags |= value;
+ break;
- proc_fdlock(p);
- if ( (ret = fp_lookup(p, fd, &fp, 1)) ) {
- proc_fdunlock(p);
+ 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 only if buffer is initialized
+ */
+ if (!(kd_ctrl_page.kdebug_flags & KDBG_BUFINIT) ||
+ !(value == KDEBUG_ENABLE_TRACE || value == KDEBUG_ENABLE_PPT)) {
+ ret = EINVAL;
break;
}
- 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);
+ kdbg_thrmap_init();
- ret = EBADF;
+ kdbg_set_tracing_enabled(true, value);
+ } else {
+ if (!kdebug_enable) {
break;
}
- vp = (struct vnode *)fp->f_fglob->fg_data;
- proc_fdunlock(p);
- if ((ret = vnode_getwithref(vp)) == 0) {
+ kernel_debug_disable();
+ }
+ break;
- if (name[0] == KERN_KDWRITETR) {
- number = nkdbufs * sizeof(kd_buf);
+ case KERN_KDSETBUF:
+ kdbg_set_nkdbufs(value);
+ break;
- KERNEL_DEBUG_CONSTANT((TRACEDBG_CODE(DBG_TRACE_INFO, 3)) | DBG_FUNC_START, 0, 0, 0, 0, 0);
- ret = kdbg_read(0, &number, vp, &context);
- KERNEL_DEBUG_CONSTANT((TRACEDBG_CODE(DBG_TRACE_INFO, 3)) | DBG_FUNC_END, number, 0, 0, 0, 0);
+ case KERN_KDSETUP:
+ ret = kdbg_reinit(false);
+ break;
- *sizep = number;
- } else {
- number = kd_mapcount * sizeof(kd_threadmap);
- kdbg_readthrmap(0, &number, vp, &context);
- }
- vnode_put(vp);
- }
- fp_drop(p, fd, fp, 0);
+ case KERN_KDREMOVE:
+ ktrace_reset(KTRACE_KDEBUG);
+ break;
+ case KERN_KDSETREG:
+ if (size < sizeof(kd_regtype)) {
+ ret = EINVAL;
+ break;
+ }
+ if (copyin(where, &kd_Reg, sizeof(kd_regtype))) {
+ ret = EINVAL;
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;
+ ret = kdbg_setreg(&kd_Reg);
+ break;
- 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;
- }
+ case KERN_KDGETREG:
+ ret = EINVAL;
+ break;
- /* check the count under the spinlock */
- number = (kd_ctrl_page.kds_inuse_count >= n_storage_threshold);
+ case KERN_KDREADTR:
+ ret = kdbg_read(where, sizep, NULL, NULL, RAW_VERSION1);
+ break;
- lck_spin_unlock(kdw_spin_lock);
- ml_set_interrupts_enabled(s);
+ 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;
- /* pick the mutex back up again */
- lck_mtx_lock(kd_trace_mtx_sysctl);
- /* write out whether we've exceeded the threshold */
- *sizep = number;
+ if (fp_get_ftype(p, fd, DTYPE_VNODE, EBADF, &fp)) {
+ ret = EBADF;
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);
- break;
- case KERN_KDPIDEX:
- if (size < sizeof(kd_regtype)) {
- ret = EINVAL;
- break;
- }
- if (copyin(where, &kd_Reg, sizeof(kd_regtype))) {
- ret = EINVAL;
- break;
+ vp = fp->fp_glob->fg_data;
+ context.vc_thread = current_thread();
+ context.vc_ucred = fp->fp_glob->fg_cred;
+
+ 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);
+
+ 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);
+
+ *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);
+ }
}
- kdbg_disable_bg_trace();
+ fp->fp_glob->fg_offset = RAW_file_offset;
+ vnode_put(vp);
+ }
+ fp_drop(p, fd, fp, 0);
+
+ break;
+ }
+ case KERN_KDBUFWAIT:
+ *sizep = kdbg_wait(size, false);
+ break;
- ret = kdbg_setpidex(&kd_Reg);
+ case KERN_KDPIDTR:
+ if (size < sizeof(kd_regtype)) {
+ ret = EINVAL;
break;
- case KERN_KDCPUMAP:
- ret = kdbg_readcpumap(where, sizep);
+ }
+ if (copyin(where, &kd_Reg, sizeof(kd_regtype))) {
+ ret = EINVAL;
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;
+
+ case KERN_KDCPUMAP:
+ ret = kdbg_readcpumap(where, sizep);
+ break;
+
+ case KERN_KDTHRMAP:
+ ret = kdbg_copyout_thread_map(where, sizep);
+ break;
- if ((kd_ctrl_page.kdebug_flags & KDBG_TYPEFILTER_CHECK) == 0){
- if ((ret = kdbg_enable_typefilter()))
- break;
- }
+ case KERN_KDSET_TYPEFILTER: {
+ ret = kdbg_copyin_typefilter(where, size);
+ break;
+ }
- if (size != KDBG_TYPEFILTER_BITMAP_SIZE) {
- ret = EINVAL;
- break;
- }
+ case KERN_KDTEST:
+ ret = kdbg_test(size);
+ 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;
+ default:
+ ret = EINVAL;
+ break;
}
out:
- lck_mtx_unlock(kd_trace_mtx_sysctl);
+ ktrace_unlock();
- return(ret);
+ return ret;
}
* move through the lists w/o use of any locks
*/
int
-kdbg_read(user_addr_t buffer, size_t *number, vnode_t vp, vfs_context_t ctx)
+kdbg_read(user_addr_t buffer, size_t *number, vnode_t vp, vfs_context_t ctx, uint32_t file_version)
{
- unsigned int count;
+ 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;
- 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 = TRACEDBG_CODE(DBG_TRACE_INFO, 2);
-
- /* 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.
+ lostevent.debugid = TRACE_LOST_EVENTS;
+
+ /*
+ * 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 the buffers have wrapped, do not emit additional lost events for the
+ * oldest storage units.
+ */
+ if (wrapped) {
+ kd_ctrl_page.kdebug_flags &= ~KDBG_WRAPPED;
- if ((tempbuf_count = count) > KDCOPYBUF_COUNT)
- tempbuf_count = KDCOPYBUF_COUNT;
+ 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))) {
+ error = EFAULT;
+ goto check_error;
+ }
+ if (buffer) {
+ buffer += (sizeof(kd_chunk_header_v3) + sizeof(uint64_t));
+ }
+ assert(count >= (sizeof(kd_chunk_header_v3) + sizeof(uint64_t)));
+ count -= (sizeof(kd_chunk_header_v3) + sizeof(uint64_t));
+ *number += (sizeof(kd_chunk_header_v3) + sizeof(uint64_t));
+ }
if (vp) {
- error = vn_rdwr(UIO_WRITE, vp, (caddr_t)kdcopybuf, tempbuf_number * sizeof(kd_buf), RAW_file_offset,
- UIO_SYSSPACE, IO_NODELOCKED|IO_UNIT, vfs_context_ucred(ctx), (int *) 0, vfs_context_proc(ctx));
+ size_t write_size = tempbuf_number * sizeof(kd_buf);
+ error = kdbg_write_to_vnode((caddr_t)kdcopybuf, write_size, vp, ctx, RAW_file_offset);
+ if (!error) {
+ RAW_file_offset += write_size;
+ }
- RAW_file_offset += (tempbuf_number * sizeof(kd_buf));
-
if (RAW_file_written >= RAW_FLUSH_SIZE) {
- cluster_push(vp, 0);
+ error = VNOP_FSYNC(vp, MNT_NOWAIT, ctx);
RAW_file_written = 0;
}
error = copyout(kdcopybuf, buffer, tempbuf_number * sizeof(kd_buf));
buffer += (tempbuf_number * sizeof(kd_buf));
}
+check_error:
if (error) {
*number = 0;
error = EINVAL;
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) {
-
- return (unsigned char *) &proc->p_comm; /* Return pointer to the proc name */
-
-}
-
-#define STACKSHOT_SUBSYS_LOCK() lck_mtx_lock(&stackshot_subsys_mutex)
-#define STACKSHOT_SUBSYS_UNLOCK() lck_mtx_unlock(&stackshot_subsys_mutex)
-#if defined(__i386__) || defined (__x86_64__)
-#define TRAP_DEBUGGER __asm__ volatile("int3");
-#else
-#error No TRAP_DEBUGGER definition for this architecture
-#endif
+/*
+ * A test IOP for the SYNC_FLUSH callback.
+ */
-#define SANE_TRACEBUF_SIZE (8 * 1024 * 1024)
-#define SANE_BOOTPROFILE_TRACEBUF_SIZE (64 * 1024 * 1024)
+static int sync_flush_iop = 0;
-/* Initialize the mutex governing access to the stack snapshot subsystem */
-__private_extern__ void
-stackshot_lock_init( void )
+static void
+sync_flush_callback(void * __unused context, kd_callback_type reason,
+ void * __unused arg)
{
- stackshot_subsys_lck_grp_attr = lck_grp_attr_alloc_init();
-
- stackshot_subsys_lck_grp = lck_grp_alloc_init("stackshot_subsys_lock", stackshot_subsys_lck_grp_attr);
+ assert(sync_flush_iop > 0);
- stackshot_subsys_lck_attr = lck_attr_alloc_init();
-
- lck_mtx_init(&stackshot_subsys_mutex, stackshot_subsys_lck_grp, stackshot_subsys_lck_attr);
+ if (reason == KD_CALLBACK_SYNC_FLUSH) {
+ kernel_debug_enter(sync_flush_iop, KDEBUG_TEST_CODE(0xff),
+ kdbg_timestamp(), 0, 0, 0, 0, 0);
+ }
}
-/*
- * 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;
-
- if ((error = suser(kauth_cred_get(), &p->p_acflag)))
- return(error);
-
- return stack_snapshot2(uap->pid, uap->tracebuf, uap->tracebuf_size,
- uap->flags, uap->dispatch_offset, retval);
-}
+static struct kd_callback sync_flush_kdcb = {
+ .func = sync_flush_callback,
+ .iop_name = "test_sf",
+};
-int
-stack_snapshot_from_kernel(pid_t pid, void *buf, uint32_t size, uint32_t flags, unsigned *bytesTraced)
+static int
+kdbg_test(size_t flavor)
{
- int error = 0;
- boolean_t istate;
-
- if ((buf == NULL) || (size <= 0) || (bytesTraced == NULL)) {
- return -1;
- }
+ 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;
- /* cap in individual stackshot to SANE_TRACEBUF_SIZE */
- if (size > SANE_TRACEBUF_SIZE) {
- size = SANE_TRACEBUF_SIZE;
- }
+ 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;
+ }
-/* Serialize tracing */
- STACKSHOT_SUBSYS_LOCK();
- istate = ml_set_interrupts_enabled(FALSE);
+ /* 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;
-/* Preload trace parameters*/
- kdp_snapshot_preflight(pid, buf, size, flags, 0);
+ case 4:
+ if (!sync_flush_iop) {
+ sync_flush_iop = kernel_debug_register_callback(
+ sync_flush_kdcb);
+ assert(sync_flush_iop > 0);
+ }
+ break;
-/* Trap to the debugger to obtain a coherent stack snapshot; this populates
- * the trace buffer
- */
- TRAP_DEBUGGER;
+ default:
+ return ENOTSUP;
+ }
- ml_set_interrupts_enabled(istate);
+ return 0;
+}
- *bytesTraced = kdp_stack_snapshot_bytes_traced();
+#undef KDEBUG_TEST_CODE
- error = kdp_stack_snapshot_geterror();
-
- STACKSHOT_SUBSYS_UNLOCK();
+void
+kdebug_init(unsigned int n_events, char *filter_desc, enum kdebug_opts opts)
+{
+ assert(filter_desc != NULL);
- return error;
+ if (log_leaks && n_events == 0) {
+ n_events = 200000;
+ }
+ kdebug_trace_start(n_events, filter_desc, opts);
}
-int
-stack_snapshot2(pid_t pid, user_addr_t tracebuf, uint32_t tracebuf_size, uint32_t flags, uint32_t dispatch_offset, int32_t *retval)
+static void
+kdbg_set_typefilter_string(const char *filter_desc)
{
- boolean_t istate;
- int error = 0;
- unsigned bytesTraced = 0;
-
-#if CONFIG_TELEMETRY
- if (flags & STACKSHOT_GLOBAL_MICROSTACKSHOT_ENABLE) {
- telemetry_global_ctl(1);
- *retval = 0;
- return (0);
- } else if (flags & STACKSHOT_GLOBAL_MICROSTACKSHOT_DISABLE) {
- telemetry_global_ctl(0);
- *retval = 0;
- return (0);
- }
-#endif
+ char *end = NULL;
- *retval = -1;
-/* Serialize tracing */
- STACKSHOT_SUBSYS_LOCK();
-
- if (tracebuf_size <= 0) {
- error = EINVAL;
- goto error_exit;
- }
+ ktrace_assert_lock_held();
-#if CONFIG_TELEMETRY
- if (flags & STACKSHOT_GET_MICROSTACKSHOT) {
+ assert(filter_desc != NULL);
- if (tracebuf_size > SANE_TRACEBUF_SIZE) {
- error = EINVAL;
- goto error_exit;
- }
+ typefilter_reject_all(kdbg_typefilter);
+ typefilter_allow_class(kdbg_typefilter, DBG_TRACE);
- bytesTraced = tracebuf_size;
- error = telemetry_gather(tracebuf, &bytesTraced,
- (flags & STACKSHOT_SET_MICROSTACKSHOT_MARK) ? TRUE : FALSE);
- if (error == KERN_NO_SPACE) {
- error = ENOSPC;
+ /* 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);
}
-
- *retval = (int)bytesTraced;
- goto error_exit;
+ return;
}
- if (flags & STACKSHOT_GET_BOOT_PROFILE) {
+ while (filter_desc[0] != '\0') {
+ unsigned long allow_value;
- if (tracebuf_size > SANE_BOOTPROFILE_TRACEBUF_SIZE) {
- error = EINVAL;
- goto error_exit;
+ 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++;
- bytesTraced = tracebuf_size;
- error = bootprofile_gather(tracebuf, &bytesTraced);
- if (error == KERN_NO_SPACE) {
- error = ENOSPC;
+ allow_value = strtoul(filter_desc, &end, 0);
+ if (filter_desc == end) {
+ printf("kdebug: cannot parse `%s' as integer\n", filter_desc);
+ return;
}
- *retval = (int)bytesTraced;
- goto error_exit;
- }
-#endif
+ switch (filter_type) {
+ case 'C':
+ if (allow_value > KDBG_CLASS_MAX) {
+ printf("kdebug: class 0x%lx is invalid\n", allow_value);
+ return;
+ }
+ 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;
+ }
+ printf("kdebug: S 0x%lx\n", allow_value);
+ typefilter_allow_csc(kdbg_typefilter, (uint16_t)allow_value);
+ break;
+ default:
+ __builtin_unreachable();
+ }
- if (tracebuf_size > SANE_TRACEBUF_SIZE) {
- error = EINVAL;
- goto error_exit;
+ /* advance to next filter entry */
+ filter_desc = end;
+ if (filter_desc[0] == ',') {
+ filter_desc++;
+ }
}
+}
- assert(stackshot_snapbuf == NULL);
- if (kmem_alloc_kobject(kernel_map, (vm_offset_t *)&stackshot_snapbuf, tracebuf_size) != KERN_SUCCESS) {
- error = ENOMEM;
- goto error_exit;
+uint64_t
+kdebug_wake(void)
+{
+ 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;
+}
- if (panic_active()) {
- error = ENOMEM;
- goto error_exit;
+/*
+ * This function is meant to be called from the bootstrap thread or kdebug_wake.
+ */
+void
+kdebug_trace_start(unsigned int n_events, const char *filter_desc,
+ enum kdebug_opts opts)
+{
+ if (!n_events) {
+ kd_early_done = true;
+ return;
}
- istate = ml_set_interrupts_enabled(FALSE);
-/* Preload trace parameters*/
- kdp_snapshot_preflight(pid, stackshot_snapbuf, tracebuf_size, flags, dispatch_offset);
+ ktrace_start_single_threaded();
-/* Trap to the debugger to obtain a coherent stack snapshot; this populates
- * the trace buffer
- */
+ kdbg_lock_init();
- TRAP_DEBUGGER;
+ ktrace_kernel_configure(KTRACE_KDEBUG);
- ml_set_interrupts_enabled(istate);
+ kdbg_set_nkdbufs(n_events);
- bytesTraced = kdp_stack_snapshot_bytes_traced();
-
- if (bytesTraced > 0) {
- if ((error = copyout(stackshot_snapbuf, tracebuf,
- ((bytesTraced < tracebuf_size) ?
- bytesTraced : tracebuf_size))))
- goto error_exit;
- *retval = bytesTraced;
- }
- else {
- error = ENOENT;
- goto error_exit;
- }
+ kernel_debug_string_early("start_kern_tracing");
- error = kdp_stack_snapshot_geterror();
- if (error == -1) {
- error = ENOSPC;
- *retval = -1;
- goto error_exit;
+ if (kdbg_reinit((opts & KDOPT_ATBOOT))) {
+ printf("error from kdbg_reinit, kernel tracing not started\n");
+ goto out;
}
-error_exit:
- if (stackshot_snapbuf != NULL)
- kmem_free(kernel_map, (vm_offset_t) stackshot_snapbuf, tracebuf_size);
- stackshot_snapbuf = NULL;
- STACKSHOT_SUBSYS_UNLOCK();
- return error;
-}
+ /*
+ * 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);
+ }
-void
-start_kern_tracing(unsigned int new_nkdbufs, boolean_t need_map) {
+ if (filter_desc && filter_desc[0] != '\0') {
+ if (kdbg_initialize_typefilter(NULL) == KERN_SUCCESS) {
+ kdbg_set_typefilter_string(filter_desc);
+ kdbg_enable_typefilter();
+ }
+ }
- if (!new_nkdbufs)
- return;
- nkdbufs = kdbg_set_nkdbufs(new_nkdbufs);
- kdbg_lock_init();
+ /*
+ * 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);
- if (0 == kdbg_reinit(TRUE)) {
+ if (!(opts & KDOPT_ATBOOT)) {
+ kdbg_thrmap_init();
+ }
- if (need_map == TRUE) {
- uint32_t old1, old2;
+ kdbg_set_tracing_enabled(true, KDEBUG_ENABLE_TRACE);
- kdbg_thrmap_init();
+ if ((opts & KDOPT_ATBOOT)) {
+ /*
+ * Transfer all very early events from the static buffer into the real
+ * buffers.
+ */
+ kernel_debug_early_end();
+ }
- disable_wrap(&old1, &old2);
- }
- kdbg_set_tracing_enabled(TRUE, KDEBUG_ENABLE_TRACE);
+ ml_set_interrupts_enabled(s);
-#if defined(__i386__) || defined(__x86_64__)
- uint64_t now = mach_absolute_time();
+ printf("kernel tracing started with %u events, filter = %s\n", n_events,
+ filter_desc ?: "none");
- KERNEL_DEBUG_CONSTANT((TRACEDBG_CODE(DBG_TRACE_INFO, 1)) | DBG_FUNC_NONE,
- (uint32_t)(tsc_rebase_abs_time >> 32), (uint32_t)tsc_rebase_abs_time,
- (uint32_t)(now >> 32), (uint32_t)now,
- 0);
-#endif
- printf("kernel tracing started\n");
- } else {
- printf("error from kdbg_reinit,kernel tracing not started\n");
- }
+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;
- 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((TRACEDBG_CODE(DBG_TRACE_INFO, 0)) | 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_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);
-
- 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;
+ }
+
+ /*
+ * Wait to synchronize the file to capture the I/O in the
+ * TRACE_WRITING_EVENTS interval.
+ */
+ ret = VNOP_FSYNC(vp, MNT_WAIT, ctx);
+
+ /*
+ * Balance the starting TRACE_WRITING_EVENTS tracepoint manually.
+ */
+ 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());
+ /* this is best effort -- ignore any errors */
+ (void)kdbg_write_to_vnode((caddr_t)&end_event, sizeof(kd_buf), vp, ctx,
+ RAW_file_offset);
+
+out_close:
+ vnode_close(vp, FWRITE, ctx);
sync(current_proc(), (void *)NULL, (int *)NULL);
+
+out:
+ ktrace_unlock();
}
-/* 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)
+static int
+kdbg_sysctl_continuous SYSCTL_HANDLER_ARGS
{
- 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.
- */
- 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);
+#pragma unused(oidp, arg1, arg2)
+ int value = kdbg_continuous_time;
+ int ret = sysctl_io_number(req, value, sizeof(value), &value, NULL);
+
+ if (ret || !req->newptr) {
+ return ret;
+ }
+
+ kdbg_continuous_time = value;
+ return 0;
}
+
+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");
projects / apple / xnu.git / blobdiff