]> git.saurik.com Git - apple/xnu.git/blobdiff - osfmk/kperf/action.c
xnu-4903.270.47.tar.gz
[apple/xnu.git] / osfmk / kperf / action.c
index dbaf90b2155b5e413d397085081fb552282f58ca..2ff723f937b328f7846cef30d30411d638b97f08 100644 (file)
@@ -2,7 +2,7 @@
  * Copyright (c) 2011 Apple Computer, Inc. All rights reserved.
  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
- * 
+ *
  * This file contains Original Code and/or Modifications of Original Code
  * as defined in and that are subject to the Apple Public Source License
  * Version 2.0 (the 'License'). You may not use this file except in
  * unlawful or unlicensed copies of an Apple operating system, or to
  * circumvent, violate, or enable the circumvention or violation of, any
  * terms of an Apple operating system software license agreement.
- * 
+ *
  * Please obtain a copy of the License at
  * http://www.opensource.apple.com/apsl/ and read it before using this file.
- * 
+ *
  * The Original Code and all software distributed under the License are
  * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
  * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
@@ -22,7 +22,7 @@
  * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
  * Please see the License for the specific language governing rights and
  * limitations under the License.
- * 
+ *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  */
 
 
 #include <mach/mach_types.h>
 #include <machine/machine_routines.h>
-// #include <libkern/libkern.h>
 #include <kern/kalloc.h>
 #include <kern/debug.h> /* panic */
 #include <kern/thread.h>
 #include <sys/errno.h>
+#include <sys/vm.h>
+#include <vm/vm_object.h>
+#include <vm/vm_page.h>
+#include <vm/vm_pageout.h>
 
-#include <chud/chud_xnu.h>
-#include <kperf/kperf.h>
-
+#include <kperf/action.h>
+#include <kperf/ast.h>
 #include <kperf/buffer.h>
-#include <kperf/timetrigger.h>
-#include <kperf/threadinfo.h>
 #include <kperf/callstack.h>
-#include <kperf/sample.h>
-#include <kperf/action.h>
 #include <kperf/context.h>
-#include <kperf/ast.h>
+#include <kperf/kdebug_trigger.h>
+#include <kperf/kperf.h>
+#include <kperf/kperf_kpc.h>
+#include <kperf/kperf_timer.h>
+#include <kperf/pet.h>
+#include <kperf/sample.h>
+#include <kperf/thread_samplers.h>
 
-#define ACTION_MAX 32
+#define ACTION_MAX (32)
 
 /* the list of different actions to take */
-struct action
-{
+struct action {
        uint32_t sample;
+       uint32_t ucallstack_depth;
+       uint32_t kcallstack_depth;
        uint32_t userdata;
        int pid_filter;
 };
 
 /* the list of actions */
-static unsigned actionc = 0;
+static unsigned int actionc = 0;
 static struct action *actionv = NULL;
 
-/* whether to record callstacks on kdebug events */
-static int kdebug_callstack_action = 0;
+/* should emit tracepoint on context switch */
+int kperf_kdebug_cswitch = 0;
 
-/* whether we get a callback on a thread switch */
-int  kperf_cswitch_hook = 0;
+bool
+kperf_action_has_non_system(unsigned int actionid)
+{
+       if (actionid > actionc) {
+               return false;
+       }
 
-/* indirect hooks to play nice with CHUD for the transition to kperf */
-kern_return_t chudxnu_kdebug_callback_enter(chudxnu_kdebug_callback_func_t fn);
-kern_return_t chudxnu_kdebug_callback_cancel(void);
+       if (actionv[actionid - 1].sample & ~SAMPLER_SYS_MEM) {
+               return true;
+       } else {
+               return false;
+       }
+}
+
+bool
+kperf_action_has_task(unsigned int actionid)
+{
+       if (actionid > actionc) {
+               return false;
+       }
+
+       return actionv[actionid - 1].sample & SAMPLER_TASK_MASK;
+}
+
+bool
+kperf_action_has_thread(unsigned int actionid)
+{
+       if (actionid > actionc) {
+               return false;
+       }
+
+       return actionv[actionid - 1].sample & SAMPLER_THREAD_MASK;
+}
+
+static void
+kperf_system_memory_log(void)
+{
+       BUF_DATA(PERF_MI_SYS_DATA, (uintptr_t)vm_page_free_count,
+           (uintptr_t)vm_page_wire_count, (uintptr_t)vm_page_external_count,
+           (uintptr_t)(vm_page_active_count + vm_page_inactive_count +
+           vm_page_speculative_count));
+       BUF_DATA(PERF_MI_SYS_DATA_2, (uintptr_t)vm_page_anonymous_count,
+           (uintptr_t)vm_page_internal_count,
+           (uintptr_t)vm_pageout_vminfo.vm_pageout_compressions,
+           (uintptr_t)VM_PAGE_COMPRESSOR_COUNT);
+}
 
-/* Do the real work! */
-/* this can be called in any context ... right? */
 static kern_return_t
-kperf_sample_internal( struct kperf_sample *sbuf,
-                       struct kperf_context *context,
-                       unsigned sample_what, unsigned sample_flags,
-                       unsigned actionid )
+kperf_sample_internal(struct kperf_sample *sbuf,
+    struct kperf_context *context,
+    unsigned sample_what, unsigned sample_flags,
+    unsigned actionid, uint32_t ucallstack_depth)
 {
-       boolean_t enabled;
-       int did_ucallstack = 0, did_tinfo_extra = 0;
-       uint32_t userdata;
+       int pended_ucallstack = 0;
+       int pended_th_dispatch = 0;
+       bool on_idle_thread = false;
+       uint32_t userdata = actionid;
+       bool task_only = false;
 
        /* not much point continuing here, but what to do ? return
         * Shutdown? cut a tracepoint and continue?
         */
-       if( sample_what == 0 )
+       if (sample_what == 0) {
                return SAMPLE_CONTINUE;
+       }
 
-       int is_kernel = (context->cur_pid == 0);
+       /* callstacks should be explicitly ignored */
+       if (sample_flags & SAMPLE_FLAG_EMPTY_CALLSTACK) {
+               sample_what &= ~(SAMPLER_KSTACK | SAMPLER_USTACK);
+       }
+
+       if (sample_flags & SAMPLE_FLAG_ONLY_SYSTEM) {
+               sample_what &= SAMPLER_SYS_MEM;
+       }
+
+       assert((sample_flags & (SAMPLE_FLAG_THREAD_ONLY | SAMPLE_FLAG_TASK_ONLY))
+           != (SAMPLE_FLAG_THREAD_ONLY | SAMPLE_FLAG_TASK_ONLY));
+       if (sample_flags & SAMPLE_FLAG_THREAD_ONLY) {
+               sample_what &= SAMPLER_THREAD_MASK;
+       }
+       if (sample_flags & SAMPLE_FLAG_TASK_ONLY) {
+               task_only = true;
+               sample_what &= SAMPLER_TASK_MASK;
+       }
+
+       if (!task_only) {
+               context->cur_thread->kperf_pet_gen = kperf_pet_gen;
+       }
+       bool is_kernel = (context->cur_pid == 0);
+
+       if (actionid && actionid <= actionc) {
+               sbuf->kcallstack.nframes = actionv[actionid - 1].kcallstack_depth;
+       } else {
+               sbuf->kcallstack.nframes = MAX_CALLSTACK_FRAMES;
+       }
+
+       if (ucallstack_depth) {
+               sbuf->ucallstack.nframes = ucallstack_depth;
+       } else {
+               sbuf->ucallstack.nframes = MAX_CALLSTACK_FRAMES;
+       }
 
-       sbuf->kcallstack.nframes = 0;
        sbuf->kcallstack.flags = CALLSTACK_VALID;
-       sbuf->ucallstack.nframes = 0;
        sbuf->ucallstack.flags = CALLSTACK_VALID;
 
-       /*  an event occurred. Sample everything and dump it in a
-        *  buffer.
+       /* an event occurred. Sample everything and dump it in a
+        * buffer.
         */
 
        /* collect data from samplers */
-       if( sample_what & SAMPLER_TINFO ) {
-               kperf_threadinfo_sample( &sbuf->threadinfo, context );
-               
+       if (sample_what & SAMPLER_TH_INFO) {
+               kperf_thread_info_sample(&sbuf->th_info, context);
+
                /* See if we should drop idle thread samples */
-               if( !(sample_flags & SAMPLE_FLAG_IDLE_THREADS) )
-                       if (sbuf->threadinfo.runmode & 0x40)
-                               return SAMPLE_CONTINUE;
+               if (!(sample_flags & SAMPLE_FLAG_IDLE_THREADS)) {
+                       if (sbuf->th_info.kpthi_runmode & 0x40) {
+                               on_idle_thread = true;
+                               goto log_sample;
+                       }
+               }
        }
 
-       if( (sample_what & SAMPLER_KSTACK) && !(sample_flags & SAMPLE_FLAG_EMPTY_CALLSTACK) )
-               kperf_kcallstack_sample( &sbuf->kcallstack, context );
+       if (sample_what & SAMPLER_TH_SNAPSHOT) {
+               kperf_thread_snapshot_sample(&(sbuf->th_snapshot), context);
+       }
+       if (sample_what & SAMPLER_TH_SCHEDULING) {
+               kperf_thread_scheduling_sample(&(sbuf->th_scheduling), context);
+       }
+       if (sample_what & SAMPLER_KSTACK) {
+               if (sample_flags & SAMPLE_FLAG_CONTINUATION) {
+                       kperf_continuation_sample(&(sbuf->kcallstack), context);
+                       /* outside of interrupt context, backtrace the current thread */
+               } else if (sample_flags & SAMPLE_FLAG_NON_INTERRUPT) {
+                       kperf_backtrace_sample(&(sbuf->kcallstack), context);
+               } else {
+                       kperf_kcallstack_sample(&(sbuf->kcallstack), context);
+               }
+       }
+       if (sample_what & SAMPLER_TK_SNAPSHOT) {
+               kperf_task_snapshot_sample(context->cur_task, &(sbuf->tk_snapshot));
+       }
 
        /* sensitive ones */
-       if ( !is_kernel ) {
-               if( sample_flags & SAMPLE_FLAG_PEND_USER )
-               {
-                       if( (sample_what & SAMPLER_USTACK) && !(sample_flags & SAMPLE_FLAG_EMPTY_CALLSTACK) )
-                               did_ucallstack = kperf_ucallstack_pend( context );
-
-                       if( sample_what & SAMPLER_TINFOEX )
-                               did_tinfo_extra = kperf_threadinfo_extra_pend( context );
+       if (!is_kernel) {
+               if (sample_what & SAMPLER_MEMINFO) {
+                       kperf_meminfo_sample(context->cur_task, &(sbuf->meminfo));
                }
-               else
-               {
-                       if( (sample_what & SAMPLER_USTACK) && !(sample_flags & SAMPLE_FLAG_EMPTY_CALLSTACK) )
-                               kperf_ucallstack_sample( &sbuf->ucallstack, context );
-
-                       if( sample_what & SAMPLER_TINFOEX )
-                               kperf_threadinfo_extra_sample( &sbuf->tinfo_ex,
-                                                              context );
+
+               if (sample_flags & SAMPLE_FLAG_PEND_USER) {
+                       if (sample_what & SAMPLER_USTACK) {
+                               pended_ucallstack = kperf_ucallstack_pend(context, sbuf->ucallstack.nframes);
+                       }
+
+                       if (sample_what & SAMPLER_TH_DISPATCH) {
+                               pended_th_dispatch = kperf_thread_dispatch_pend(context);
+                       }
+               } else {
+                       if (sample_what & SAMPLER_USTACK) {
+                               kperf_ucallstack_sample(&(sbuf->ucallstack), context);
+                       }
+
+                       if (sample_what & SAMPLER_TH_DISPATCH) {
+                               kperf_thread_dispatch_sample(&(sbuf->th_dispatch), context);
+                       }
                }
        }
 
-#if KPC
-       if ( sample_what & SAMPLER_PMC_CPU )
-               kperf_kpc_cpu_sample( &sbuf->kpcdata, 
-                                     (sample_what & SAMPLER_PMC_CPU) != 0 );
-#endif
+       if (sample_what & SAMPLER_PMC_THREAD) {
+               kperf_kpc_thread_sample(&(sbuf->kpcdata), sample_what);
+       } else if (sample_what & SAMPLER_PMC_CPU) {
+               kperf_kpc_cpu_sample(&(sbuf->kpcdata), sample_what);
+       }
 
+log_sample:
        /* lookup the user tag, if any */
-       if( actionid 
-           && (actionid <= actionc) )
-               userdata = actionv[actionid-1].userdata;
-       else
-               userdata = actionid;
+       if (actionid && (actionid <= actionc)) {
+               userdata = actionv[actionid - 1].userdata;
+       }
+
+       /* avoid logging if this sample only pended samples */
+       if (sample_flags & SAMPLE_FLAG_PEND_USER &&
+           !(sample_what & ~(SAMPLER_USTACK | SAMPLER_TH_DISPATCH))) {
+               return SAMPLE_CONTINUE;
+       }
 
        /* stash the data into the buffer
         * interrupts off to ensure we don't get split
         */
-       enabled = ml_set_interrupts_enabled(FALSE);
+       boolean_t enabled = ml_set_interrupts_enabled(FALSE);
 
-       BUF_DATA( PERF_GEN_EVENT | DBG_FUNC_START, sample_what, 
-                  actionid, userdata, sample_flags );
+       BUF_DATA(PERF_GEN_EVENT | DBG_FUNC_START, sample_what,
+           actionid, userdata, sample_flags);
 
-       /* dump threadinfo */
-       if( sample_what & SAMPLER_TINFO )
-               kperf_threadinfo_log( &sbuf->threadinfo );
-
-       /* dump kcallstack */
-       if( sample_what & SAMPLER_KSTACK )
-               kperf_kcallstack_log( &sbuf->kcallstack );
+       if (sample_flags & SAMPLE_FLAG_SYSTEM) {
+               if (sample_what & SAMPLER_SYS_MEM) {
+                       kperf_system_memory_log();
+               }
+       }
+       if (on_idle_thread) {
+               goto log_sample_end;
+       }
 
+       if (sample_what & SAMPLER_TH_INFO) {
+               kperf_thread_info_log(&sbuf->th_info);
+       }
+       if (sample_what & SAMPLER_TH_SCHEDULING) {
+               kperf_thread_scheduling_log(&(sbuf->th_scheduling));
+       }
+       if (sample_what & SAMPLER_TH_SNAPSHOT) {
+               kperf_thread_snapshot_log(&(sbuf->th_snapshot));
+       }
+       if (sample_what & SAMPLER_KSTACK) {
+               kperf_kcallstack_log(&sbuf->kcallstack);
+       }
+       if (sample_what & SAMPLER_TH_INSCYC) {
+               kperf_thread_inscyc_log(context);
+       }
+       if (sample_what & SAMPLER_TK_SNAPSHOT) {
+               kperf_task_snapshot_log(&(sbuf->tk_snapshot));
+       }
+       if (sample_what & SAMPLER_TK_INFO) {
+               kperf_task_info_log(context);
+       }
 
        /* dump user stuff */
-       if ( !is_kernel ) {
-               if ( sample_flags & SAMPLE_FLAG_PEND_USER )
-               {
-                       if ( did_ucallstack )
-                               BUF_INFO1( PERF_CS_UPEND, 0 );
-
-                       if ( did_tinfo_extra )
-                               BUF_INFO1( PERF_TI_XPEND, 0 );
+       if (!is_kernel) {
+               /* dump meminfo */
+               if (sample_what & SAMPLER_MEMINFO) {
+                       kperf_meminfo_log(&(sbuf->meminfo));
                }
-               else
-               {
-                       if( sample_what & SAMPLER_USTACK )
-                               kperf_ucallstack_log( &sbuf->ucallstack );
 
-                       if( sample_what & SAMPLER_TINFOEX )
-                               kperf_threadinfo_extra_log( &sbuf->tinfo_ex );
+               if (sample_flags & SAMPLE_FLAG_PEND_USER) {
+                       if (pended_ucallstack) {
+                               BUF_INFO(PERF_CS_UPEND);
+                       }
+
+                       if (pended_th_dispatch) {
+                               BUF_INFO(PERF_TI_DISPPEND);
+                       }
+               } else {
+                       if (sample_what & SAMPLER_USTACK) {
+                               kperf_ucallstack_log(&(sbuf->ucallstack));
+                       }
+
+                       if (sample_what & SAMPLER_TH_DISPATCH) {
+                               kperf_thread_dispatch_log(&(sbuf->th_dispatch));
+                       }
                }
        }
 
-#if KPC
-       if ( sample_what & SAMPLER_PMC_CPU )
-               kperf_kpc_cpu_log( &sbuf->kpcdata );
-       
-#endif
+       if (sample_what & SAMPLER_PMC_THREAD) {
+               kperf_kpc_thread_log(&(sbuf->kpcdata));
+       } else if (sample_what & SAMPLER_PMC_CPU) {
+               kperf_kpc_cpu_log(&(sbuf->kpcdata));
+       }
 
-       BUF_DATA1( PERF_GEN_EVENT | DBG_FUNC_END, sample_what );
+log_sample_end:
+       BUF_DATA(PERF_GEN_EVENT | DBG_FUNC_END, sample_what, on_idle_thread ? 1 : 0);
 
        /* intrs back on */
        ml_set_interrupts_enabled(enabled);
@@ -204,369 +340,411 @@ kperf_sample_internal( struct kperf_sample *sbuf,
 
 /* Translate actionid into sample bits and take a sample */
 kern_return_t
-kperf_sample( struct kperf_sample *sbuf,
-             struct kperf_context *context,
-              unsigned actionid, unsigned sample_flags )
+kperf_sample(struct kperf_sample *sbuf,
+    struct kperf_context *context,
+    unsigned actionid, unsigned sample_flags)
 {
-       unsigned sample_what = 0;
-       int pid_filter;
-
        /* work out what to sample, if anything */
-       if( (actionid > actionc) || (actionid == 0) )
+       if ((actionid > actionc) || (actionid == 0)) {
                return SAMPLE_SHUTDOWN;
+       }
 
        /* check the pid filter against the context's current pid.
         * filter pid == -1 means any pid
         */
-       pid_filter = actionv[actionid-1].pid_filter;
-       if( (pid_filter != -1)
-           && (pid_filter != context->cur_pid) )
+       int pid_filter = actionv[actionid - 1].pid_filter;
+       if ((pid_filter != -1) && (pid_filter != context->cur_pid)) {
                return SAMPLE_CONTINUE;
+       }
 
        /* the samplers to run */
-       sample_what = actionv[actionid-1].sample;
+       unsigned int sample_what = actionv[actionid - 1].sample;
 
        /* do the actual sample operation */
-       return kperf_sample_internal( sbuf, context, sample_what, 
-                                     sample_flags, actionid );
+       return kperf_sample_internal(sbuf, context, sample_what,
+                  sample_flags, actionid,
+                  actionv[actionid - 1].ucallstack_depth);
 }
 
-/* ast callback on a thread */
 void
-kperf_thread_ast_handler( thread_t thread )
+kperf_kdebug_handler(uint32_t debugid, uintptr_t *starting_fp)
 {
-       int r;
-       uint32_t t_chud;
-       unsigned sample_what = 0;
-       /* we know we're on a thread, so let's do stuff */
-       task_t task = NULL;
+       uint32_t sample_flags = SAMPLE_FLAG_PEND_USER;
+       struct kperf_sample *sample = NULL;
+       kern_return_t kr = KERN_SUCCESS;
+       int s;
+
+       if (!kperf_kdebug_should_trigger(debugid)) {
+               return;
+       }
 
-       BUF_INFO1(PERF_AST_HNDLR | DBG_FUNC_START, thread);
+       BUF_VERB(PERF_KDBG_HNDLR | DBG_FUNC_START, debugid);
 
-       /* use ~2kb of the stack for the sample, should be ok since we're in the ast */
+       thread_t thread = current_thread();
+       task_t task = get_threadtask(thread);
+       struct kperf_context ctx = {
+               .cur_thread = thread,
+               .cur_task = task,
+               .cur_pid = task_pid(task),
+               .trigger_type = TRIGGER_TYPE_KDEBUG,
+               .trigger_id = 0,
+       };
+
+       s = ml_set_interrupts_enabled(0);
+
+       sample = kperf_intr_sample_buffer();
+
+       if (!ml_at_interrupt_context()) {
+               sample_flags |= SAMPLE_FLAG_NON_INTERRUPT;
+               ctx.starting_fp = starting_fp;
+       }
+
+       kr = kperf_sample(sample, &ctx, kperf_kdebug_get_action(), sample_flags);
+
+       ml_set_interrupts_enabled(s);
+       BUF_VERB(PERF_KDBG_HNDLR | DBG_FUNC_END, kr);
+}
+
+/*
+ * This function allocates >2.3KB of the stack.  Prevent the compiler from
+ * inlining this function into ast_taken and ensure the stack memory is only
+ * allocated for the kperf AST.
+ */
+__attribute__((noinline))
+void
+kperf_thread_ast_handler(thread_t thread)
+{
+       BUF_INFO(PERF_AST_HNDLR | DBG_FUNC_START, thread, kperf_get_thread_flags(thread));
+
+       /* ~2KB of the stack for the sample since this is called from AST */
        struct kperf_sample sbuf;
-       bzero(&sbuf, sizeof(struct kperf_sample));
+       memset(&sbuf, 0, sizeof(struct kperf_sample));
+
+       task_t task = get_threadtask(thread);
+
+       if (task_did_exec(task) || task_is_exec_copy(task)) {
+               BUF_INFO(PERF_AST_HNDLR | DBG_FUNC_END, SAMPLE_CONTINUE);
+               return;
+       }
 
        /* make a context, take a sample */
-       struct kperf_context ctx;
-       ctx.cur_thread = thread;
-       ctx.cur_pid = -1;
-
-       task = chudxnu_task_for_thread(thread);
-       if(task)
-               ctx.cur_pid = chudxnu_pid_for_task(task);
-
-       /* decode the chud bits so we know what to sample */
-       t_chud = kperf_get_thread_bits(thread);
-       
-       if (t_chud & T_AST_NAME)
-               sample_what |= SAMPLER_TINFOEX;
-       
-       if (t_chud & T_AST_CALLSTACK)
-       {
+       struct kperf_context ctx = {
+               .cur_thread = thread,
+               .cur_task = task,
+               .cur_pid = task_pid(task),
+       };
+
+       /* decode the flags to determine what to sample */
+       unsigned int sample_what = 0;
+       uint32_t flags = kperf_get_thread_flags(thread);
+
+       if (flags & T_KPERF_AST_DISPATCH) {
+               sample_what |= SAMPLER_TH_DISPATCH;
+       }
+       if (flags & T_KPERF_AST_CALLSTACK) {
                sample_what |= SAMPLER_USTACK;
-               sample_what |= SAMPLER_TINFO;
+               sample_what |= SAMPLER_TH_INFO;
        }
 
-       /* do the sample, just of the user stuff */
-       r = kperf_sample_internal( &sbuf, &ctx, sample_what, 0, 0 );
+       uint32_t ucallstack_depth = T_KPERF_GET_CALLSTACK_DEPTH(flags);
+
+       int r = kperf_sample_internal(&sbuf, &ctx, sample_what, 0, 0, ucallstack_depth);
 
-       BUF_INFO1(PERF_AST_HNDLR | DBG_FUNC_END, r);
+       BUF_INFO(PERF_AST_HNDLR | DBG_FUNC_END, r);
 }
 
 /* register AST bits */
 int
-kperf_ast_pend( thread_t cur_thread, uint32_t check_bits,
-               uint32_t set_bits )
+kperf_ast_pend(thread_t thread, uint32_t set_flags)
 {
-       /* pend on the thread */
-       uint32_t t_chud, set_done = 0;
        /* can only pend on the current thread */
-       if( cur_thread != chudxnu_current_thread() )
+       if (thread != current_thread()) {
                panic("pending to non-current thread");
+       }
 
        /* get our current bits */
-       t_chud = kperf_get_thread_bits(cur_thread);
+       uint32_t flags = kperf_get_thread_flags(thread);
 
        /* see if it's already been done or pended */
-       if( !(t_chud & check_bits ) )
-       {
+       if (!(flags & set_flags)) {
                /* set the bit on the thread */
-               t_chud |= set_bits;
-               kperf_set_thread_bits(cur_thread, t_chud);
+               flags |= set_flags;
+               kperf_set_thread_flags(thread, flags);
 
                /* set the actual AST */
-               kperf_set_thread_ast( cur_thread );
-
-               set_done = 1;
+               act_set_kperf(thread);
+               return 1;
        }
 
-       return set_done;
-
-//     BUF_INFO3( dbg_code, (uintptr_t)cur_thread, t_chud, set_done );
+       return 0;
 }
 
-/*
- * kdebug callback & stack management
- */
-
-#define IS_END(debugid)           ((debugid & 3) == DBG_FUNC_END)
-#define IS_MIG(debugid)           (IS_END(debugid) && ((debugid & 0xff000000U) == KDBG_CLASS_ENCODE((unsigned)DBG_MIG, 0U)))
-#define IS_MACH_SYSCALL(debugid)  (IS_END(debugid) && (KDBG_CLASS_DECODE(debugid) == KDBG_CLASS_ENCODE(DBG_MACH, DBG_MACH_EXCP_SC)))
-#define IS_VM_FAULT(debugid)      (IS_END(debugid) && (KDBG_CLASS_DECODE(debugid) == KDBG_CLASS_ENCODE(DBG_MACH, DBG_MACH_VM)))
-#define IS_BSD_SYSCTLL(debugid)   (IS_END(debugid) && (KDBG_CLASS_DECODE(debugid) == KDBG_CLASS_ENCODE(DBG_BSD, DBG_BSD_EXCP_SC)))
-#define IS_APPS_SIGNPOST(debugid) (IS_END(debugid) && (KDBG_CLASS_DECODE(debugid) == KDBG_CLASS_ENCODE(DBG_APPS, DBG_MACH_CHUD)))
-#define IS_MACH_SIGNPOST(debugid) (IS_END(debugid) && (KDBG_CLASS_DECODE(debugid) == KDBG_CLASS_ENCODE(DBG_MACH, DBG_MACH_CHUD)))
-
 void
-kperf_kdebug_callback(uint32_t debugid)
+kperf_ast_set_callstack_depth(thread_t thread, uint32_t depth)
 {
-       int cur_pid = 0;
-       task_t task = NULL;
-
-       /* if we're not doing kperf callback stacks, return */
-       if( !kdebug_callstack_action )
-               return;
-
-       /* if we're looking at a kperf tracepoint, don't recurse */
-       if( (debugid & 0xff000000) == KDBG_CLASS_ENCODE(DBG_PERF, 0) )
-               return;
-
-       /* ensure interrupts are already off thanks to kdebug */
-       if( ml_get_interrupts_enabled() )
-               return;
-
-       /* make sure we're not being called recursively.  */
-#if NOTYET
-       if( kperf_kdbg_recurse(KPERF_RECURSE_IN) )
-               return;
-#endif
-
-       /* check the happy list of trace codes */
-       if( !( IS_MIG(debugid)
-              || IS_MACH_SYSCALL(debugid)
-              || IS_VM_FAULT(debugid)
-              || IS_BSD_SYSCTLL(debugid)
-              || IS_MACH_SIGNPOST(debugid)
-              || IS_APPS_SIGNPOST(debugid) ) )
-               return;
-
-       /* check for kernel */
-       thread_t thread = chudxnu_current_thread();
-       task = chudxnu_task_for_thread(thread);
-       if(task)
-               cur_pid = chudxnu_pid_for_task(task);
-       if( !cur_pid )
-               return;
+       uint32_t ast_flags = kperf_get_thread_flags(thread);
+       uint32_t existing_callstack_depth = T_KPERF_GET_CALLSTACK_DEPTH(ast_flags);
 
-#if NOTYET
-       /* setup a context */
-       struct kperf_context ctx;
-       struct kperf_sample *intbuf = NULL;
-
-       ctx.cur_thread = thread;
-       ctx.cur_pid = cur_pid;
-       ctx.trigger_type = TRIGGER_TYPE_TRACE;
-       ctx.trigger_id = 0;
-
-       /* CPU sample buffer -- only valid with interrupts off (above)
-        * Technically this isn't true -- tracepoints can, and often
-        * are, cut from interrupt handlers, but none of those tracepoints
-        * should make it this far.
-        */
-       intbuf = kperf_intr_sample_buffer();
-
-       /* do the sample */
-       kperf_sample( intbuf, &ctx, kdebug_callstack_action, SAMPLE_FLAG_PEND_USER );
-       
-       /* no longer recursive */
-       kperf_kdbg_recurse(KPERF_RECURSE_OUT);
-#else
-       /* dicing with death */
-       BUF_INFO2(PERF_KDBG_HNDLR, debugid, cur_pid);
-
-       /* pend the AST */
-       kperf_ast_pend( thread, T_AST_CALLSTACK, T_AST_CALLSTACK );
-#endif
+       if (existing_callstack_depth != depth) {
+               ast_flags &= ~T_KPERF_SET_CALLSTACK_DEPTH(depth);
+               ast_flags |= T_KPERF_SET_CALLSTACK_DEPTH(depth);
 
+               kperf_set_thread_flags(thread, ast_flags);
+       }
 }
 
 int
-kperf_kdbg_get_stacks(void)
+kperf_kdbg_cswitch_get(void)
 {
-       return kdebug_callstack_action;
+       return kperf_kdebug_cswitch;
 }
 
 int
-kperf_kdbg_set_stacks(int newval)
+kperf_kdbg_cswitch_set(int newval)
 {
-       /* set the value */
-       kdebug_callstack_action = newval;
-
-       /* enable the callback from kdebug */
-       if( newval )
-               chudxnu_kdebug_callback_enter(NULL);
-       else
-               chudxnu_kdebug_callback_cancel();
+       kperf_kdebug_cswitch = newval;
+       kperf_on_cpu_update();
 
        return 0;
 }
 
-/*
- * Thread switch
- */
-
-/* called from context switch handler */
-void
-kperf_switch_context( __unused thread_t old, thread_t new )
-{
-       task_t task = get_threadtask(new);
-       int pid = chudxnu_pid_for_task(task);
-
-       /* cut a tracepoint to tell us what the new thread's PID is
-        * for Instruments
-        */
-       BUF_DATA2( PERF_TI_CSWITCH, thread_tid(new), pid );
-}
-
 /*
  * Action configuration
  */
-unsigned
+unsigned int
 kperf_action_get_count(void)
 {
        return actionc;
 }
 
 int
-kperf_action_set_samplers( unsigned actionid, uint32_t samplers )
+kperf_action_set_samplers(unsigned actionid, uint32_t samplers)
 {
-       if( (actionid > actionc) || (actionid == 0) )
+       if ((actionid > actionc) || (actionid == 0)) {
                return EINVAL;
+       }
+
+       /* disallow both CPU and thread counters to be sampled in the same
+        * action */
+       if ((samplers & SAMPLER_PMC_THREAD) && (samplers & SAMPLER_PMC_CPU)) {
+               return EINVAL;
+       }
 
-       actionv[actionid-1].sample = samplers;
+       actionv[actionid - 1].sample = samplers;
 
        return 0;
 }
 
 int
-kperf_action_get_samplers( unsigned actionid, uint32_t *samplers_out )
+kperf_action_get_samplers(unsigned actionid, uint32_t *samplers_out)
 {
-       if( (actionid > actionc) )
+       if ((actionid > actionc)) {
                return EINVAL;
+       }
 
-       if( actionid == 0 )
+       if (actionid == 0) {
                *samplers_out = 0; /* "NULL" action */
-       else
-               *samplers_out = actionv[actionid-1].sample;
+       } else {
+               *samplers_out = actionv[actionid - 1].sample;
+       }
 
        return 0;
 }
 
 int
-kperf_action_set_userdata( unsigned actionid, uint32_t userdata )
+kperf_action_set_userdata(unsigned actionid, uint32_t userdata)
 {
-       if( (actionid > actionc) || (actionid == 0) )
+       if ((actionid > actionc) || (actionid == 0)) {
                return EINVAL;
+       }
 
-       actionv[actionid-1].userdata = userdata;
+       actionv[actionid - 1].userdata = userdata;
 
        return 0;
 }
 
 int
-kperf_action_get_userdata( unsigned actionid, uint32_t *userdata_out )
+kperf_action_get_userdata(unsigned actionid, uint32_t *userdata_out)
 {
-       if( (actionid > actionc) )
+       if ((actionid > actionc)) {
                return EINVAL;
+       }
 
-       if( actionid == 0 )
+       if (actionid == 0) {
                *userdata_out = 0; /* "NULL" action */
-       else
-               *userdata_out = actionv[actionid-1].userdata;
+       } else {
+               *userdata_out = actionv[actionid - 1].userdata;
+       }
 
        return 0;
 }
 
 int
-kperf_action_set_filter( unsigned actionid,
-                        int pid )
+kperf_action_set_filter(unsigned actionid, int pid)
 {
-       if( (actionid > actionc) || (actionid == 0) )
+       if ((actionid > actionc) || (actionid == 0)) {
                return EINVAL;
+       }
 
-       actionv[actionid-1].pid_filter = pid;
+       actionv[actionid - 1].pid_filter = pid;
 
        return 0;
 }
 
 int
-kperf_action_get_filter( unsigned actionid,
-                        int *pid_out )
+kperf_action_get_filter(unsigned actionid, int *pid_out)
 {
-       if( (actionid > actionc) )
+       if ((actionid > actionc)) {
                return EINVAL;
+       }
 
-       if( actionid == 0 )
+       if (actionid == 0) {
                *pid_out = -1; /* "NULL" action */
-       else
-               *pid_out = actionv[actionid-1].pid_filter;
+       } else {
+               *pid_out = actionv[actionid - 1].pid_filter;
+       }
 
        return 0;
 }
 
+void
+kperf_action_reset(void)
+{
+       for (unsigned int i = 0; i < actionc; i++) {
+               kperf_action_set_samplers(i + 1, 0);
+               kperf_action_set_userdata(i + 1, 0);
+               kperf_action_set_filter(i + 1, -1);
+               kperf_action_set_ucallstack_depth(i + 1, MAX_CALLSTACK_FRAMES);
+               kperf_action_set_kcallstack_depth(i + 1, MAX_CALLSTACK_FRAMES);
+       }
+}
+
 int
 kperf_action_set_count(unsigned count)
 {
        struct action *new_actionv = NULL, *old_actionv = NULL;
-       unsigned old_count, i;
+       unsigned old_count;
 
        /* easy no-op */
-       if( count == actionc )
+       if (count == actionc) {
                return 0;
+       }
 
        /* TODO: allow shrinking? */
-       if( count < actionc )
+       if (count < actionc) {
                return EINVAL;
+       }
 
        /* cap it for good measure */
-       if( count > ACTION_MAX )
+       if (count > ACTION_MAX) {
                return EINVAL;
+       }
 
        /* creating the action arror for the first time. create a few
         * more things, too.
         */
-               if( actionc == 0 )
-       {
+       if (actionc == 0) {
                int r;
-               r = kperf_init();
-
-               if( r != 0 )
+               if ((r = kperf_init())) {
                        return r;
+               }
        }
 
        /* create a new array */
-       new_actionv = kalloc( count * sizeof(*new_actionv) );
-       if( new_actionv == NULL )
+       new_actionv = kalloc_tag(count * sizeof(*new_actionv), VM_KERN_MEMORY_DIAG);
+       if (new_actionv == NULL) {
                return ENOMEM;
+       }
 
        old_actionv = actionv;
        old_count = actionc;
 
-       if( old_actionv != NULL )
-               bcopy( actionv, new_actionv, actionc * sizeof(*actionv) );
+       if (old_actionv != NULL) {
+               memcpy(new_actionv, actionv, actionc * sizeof(*actionv));
+       }
 
-       bzero( &new_actionv[actionc], (count - old_count) * sizeof(*actionv) );
+       memset(&(new_actionv[actionc]), 0, (count - old_count) * sizeof(*actionv));
 
-       for( i = old_count; i < count; i++ )
+       for (unsigned int i = old_count; i < count; i++) {
                new_actionv[i].pid_filter = -1;
+               new_actionv[i].ucallstack_depth = MAX_CALLSTACK_FRAMES;
+               new_actionv[i].kcallstack_depth = MAX_CALLSTACK_FRAMES;
+       }
 
        actionv = new_actionv;
        actionc = count;
 
-       if( old_actionv != NULL )
-               kfree( old_actionv, old_count * sizeof(*actionv) );
+       if (old_actionv != NULL) {
+               kfree(old_actionv, old_count * sizeof(*actionv));
+       }
+
+       return 0;
+}
+
+int
+kperf_action_set_ucallstack_depth(unsigned action_id, uint32_t depth)
+{
+       if ((action_id > actionc) || (action_id == 0)) {
+               return EINVAL;
+       }
+
+       if (depth > MAX_CALLSTACK_FRAMES) {
+               return EINVAL;
+       }
+
+       actionv[action_id - 1].ucallstack_depth = depth;
+
+       return 0;
+}
+
+int
+kperf_action_set_kcallstack_depth(unsigned action_id, uint32_t depth)
+{
+       if ((action_id > actionc) || (action_id == 0)) {
+               return EINVAL;
+       }
+
+       if (depth > MAX_CALLSTACK_FRAMES) {
+               return EINVAL;
+       }
+
+       actionv[action_id - 1].kcallstack_depth = depth;
+
+       return 0;
+}
+
+int
+kperf_action_get_ucallstack_depth(unsigned action_id, uint32_t * depth_out)
+{
+       if ((action_id > actionc)) {
+               return EINVAL;
+       }
+
+       assert(depth_out);
+
+       if (action_id == 0) {
+               *depth_out = MAX_CALLSTACK_FRAMES;
+       } else {
+               *depth_out = actionv[action_id - 1].ucallstack_depth;
+       }
+
+       return 0;
+}
+
+int
+kperf_action_get_kcallstack_depth(unsigned action_id, uint32_t * depth_out)
+{
+       if ((action_id > actionc)) {
+               return EINVAL;
+       }
+
+       assert(depth_out);
+
+       if (action_id == 0) {
+               *depth_out = MAX_CALLSTACK_FRAMES;
+       } else {
+               *depth_out = actionv[action_id - 1].kcallstack_depth;
+       }
 
        return 0;
 }