]> git.saurik.com Git - apple/xnu.git/blobdiff - osfmk/kperf/callstack.c
xnu-7195.50.7.100.1.tar.gz
[apple/xnu.git] / osfmk / kperf / callstack.c
index d0c1e39478b21f707e3c27269b622ef5c017c2bb..a6a367c3d258f69d1f1a6eca626c0c89f7a61917 100644 (file)
@@ -2,7 +2,7 @@
  * Copyright (c) 2011 Apple Computer, Inc. All rights reserved.
  *
  * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
  * 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
  * 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.
  * 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.
  * 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,
  * 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,
  * 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.
  * 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@
  */
 
 /* Collect kernel callstacks */
 
 #include <mach/mach_types.h>
  * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
  */
 
 /* Collect kernel callstacks */
 
 #include <mach/mach_types.h>
-#include <machine/machine_routines.h>  /* XXX: remove me */
 #include <kern/thread.h>
 #include <kern/thread.h>
-
-#include <chud/chud_xnu.h>
-
+#include <kern/backtrace.h>
+#include <kern/cambria_layout.h>
+#include <vm/vm_map.h>
 #include <kperf/buffer.h>
 #include <kperf/context.h>
 #include <kperf/callstack.h>
 #include <kperf/ast.h>
 #include <kperf/buffer.h>
 #include <kperf/context.h>
 #include <kperf/callstack.h>
 #include <kperf/ast.h>
+#include <sys/errno.h>
+
+#if defined(__arm__) || defined(__arm64__)
+#include <arm/cpu_data.h>
+#include <arm/cpu_data_internal.h>
+#endif
 
 static void
 
 static void
-callstack_sample( struct callstack *cs, 
-                  struct kperf_context *context,
-                  uint32_t is_user )
+callstack_fixup_user(struct kp_ucallstack *cs, thread_t thread)
 {
 {
-       kern_return_t kr;
-       mach_msg_type_number_t nframes; /* WTF with the type? */
-       uint32_t code;
+       uint64_t fixup_val = 0;
+       assert(cs->kpuc_nframes < MAX_UCALLSTACK_FRAMES);
+
+#if defined(__x86_64__)
+       user_addr_t sp_user;
+       bool user_64;
+       x86_saved_state_t *state;
 
 
-       if( is_user )
-               code = PERF_CS_USAMPLE;
-       else
-               code = PERF_CS_KSAMPLE;
+       state = get_user_regs(thread);
+       if (!state) {
+               goto out;
+       }
 
 
-       BUF_INFO1( code, (uintptr_t)context->cur_thread );
+       user_64 = is_saved_state64(state);
+       if (user_64) {
+               sp_user = saved_state64(state)->isf.rsp;
+       } else {
+               sp_user = saved_state32(state)->uesp;
+       }
 
 
-       /* fill out known flags */
-       cs->flags = 0;
-       if( !is_user )
-       {
-               cs->flags |= CALLSTACK_KERNEL;
-#ifdef __LP64__
-               cs->flags |= CALLSTACK_64BIT;
+       if (thread == current_thread()) {
+               (void)copyin(sp_user, (char *)&fixup_val,
+                   user_64 ? sizeof(uint64_t) : sizeof(uint32_t));
+       } else {
+               (void)vm_map_read_user(get_task_map(get_threadtask(thread)), sp_user,
+                   &fixup_val, user_64 ? sizeof(uint64_t) : sizeof(uint32_t));
+       }
+
+#elif defined(__arm64__) || defined(__arm__)
+
+       struct arm_saved_state *state = get_user_regs(thread);
+       if (!state) {
+               goto out;
+       }
+
+       /* encode thumb mode into low bit of PC */
+       if (get_saved_state_cpsr(state) & PSR_TF) {
+               cs->kpuc_frames[0] |= 1ULL;
+       }
+
+
+       fixup_val = get_saved_state_lr(state);
+
+#else
+#error "callstack_fixup_user: unsupported architecture"
 #endif
 #endif
+
+out:
+       cs->kpuc_frames[cs->kpuc_nframes++] = fixup_val;
+}
+
+#if defined(__x86_64__)
+
+__attribute__((used))
+static kern_return_t
+interrupted_kernel_sp_value(uintptr_t *sp_val)
+{
+       x86_saved_state_t *state;
+       uintptr_t sp;
+       bool state_64;
+       uint64_t cs;
+       uintptr_t top, bottom;
+
+       state = current_cpu_datap()->cpu_int_state;
+       if (!state) {
+               return KERN_FAILURE;
+       }
+
+       state_64 = is_saved_state64(state);
+
+       if (state_64) {
+               cs = saved_state64(state)->isf.cs;
+       } else {
+               cs = saved_state32(state)->cs;
        }
        }
-       else
-       {
-               /* FIXME: detect 32 vs 64-bit? */
+       /* return early if interrupted a thread in user space */
+       if ((cs & SEL_PL) == SEL_PL_U) {
+               return KERN_FAILURE;
        }
 
        }
 
-       /* collect the callstack */
-       nframes = MAX_CALLSTACK_FRAMES;
-       kr = chudxnu_thread_get_callstack64( context->cur_thread, 
-                                            cs->frames, 
-                                            &nframes,
-                                            is_user );
+       if (state_64) {
+               sp = saved_state64(state)->isf.rsp;
+       } else {
+               sp = saved_state32(state)->uesp;
+       }
 
 
-       /* check for overflow */
-       if( kr == KERN_SUCCESS )
-       {
-               cs->flags |= CALLSTACK_VALID;
-               cs->nframes = nframes;
+       /* make sure the stack pointer is pointing somewhere in this stack */
+       bottom = current_thread()->kernel_stack;
+       top = bottom + kernel_stack_size;
+       if (sp >= bottom && sp < top) {
+               return KERN_FAILURE;
        }
        }
-       else if( kr == KERN_RESOURCE_SHORTAGE )
-       {
-               /* FIXME: more here */
-               cs->flags |= CALLSTACK_TRUNCATED;
-               cs->flags |= CALLSTACK_VALID;
-               cs->nframes = nframes;
+
+       *sp_val = *(uintptr_t *)sp;
+       return KERN_SUCCESS;
+}
+
+#elif defined(__arm64__)
+
+__attribute__((used))
+static kern_return_t
+interrupted_kernel_lr(uintptr_t *lr)
+{
+       struct arm_saved_state *state;
+
+       state = getCpuDatap()->cpu_int_state;
+
+       /* return early if interrupted a thread in user space */
+       if (PSR64_IS_USER(get_saved_state_cpsr(state))) {
+               return KERN_FAILURE;
        }
        }
-       else
-       {
-               BUF_INFO2(PERF_PET_ERROR, ERR_GETSTACK, kr);
-               cs->nframes = 0;
+
+       *lr = get_saved_state_lr(state);
+       return KERN_SUCCESS;
+}
+
+#elif defined(__arm__)
+
+__attribute__((used))
+static kern_return_t
+interrupted_kernel_lr(uintptr_t *lr)
+{
+       struct arm_saved_state *state;
+
+       state = getCpuDatap()->cpu_int_state;
+
+       /* return early if interrupted a thread in user space */
+       if (PSR_IS_USER(get_saved_state_cpsr(state))) {
+               return KERN_FAILURE;
        }
 
        }
 
-       if( cs->nframes >= MAX_CALLSTACK_FRAMES )
-       {
-               /* necessary? */
-               BUF_INFO1(PERF_PET_ERROR, ERR_FRAMES);
-               cs->nframes = 0;
+       *lr = get_saved_state_lr(state);
+       return KERN_SUCCESS;
+}
+
+#else /* defined(__arm__) */
+#error "interrupted_kernel_{sp,lr}: unsupported architecture"
+#endif /* !defined(__arm__) */
+
+
+static void
+callstack_fixup_interrupted(struct kp_kcallstack *cs)
+{
+       uintptr_t fixup_val = 0;
+       assert(cs->kpkc_nframes < MAX_KCALLSTACK_FRAMES);
+
+       /*
+        * Only provide arbitrary data on development or debug kernels.
+        */
+#if DEVELOPMENT || DEBUG
+#if defined(__x86_64__)
+       (void)interrupted_kernel_sp_value(&fixup_val);
+#elif defined(__arm64__) || defined(__arm__)
+       (void)interrupted_kernel_lr(&fixup_val);
+#endif /* defined(__x86_64__) */
+#endif /* DEVELOPMENT || DEBUG */
+
+       assert(cs->kpkc_flags & CALLSTACK_KERNEL);
+       cs->kpkc_frames[cs->kpkc_nframes++] = fixup_val;
+}
+
+void
+kperf_continuation_sample(struct kp_kcallstack *cs, struct kperf_context *context)
+{
+       thread_t thread;
+
+       assert(cs != NULL);
+       assert(context != NULL);
+
+       thread = context->cur_thread;
+       assert(thread != NULL);
+       assert(thread->continuation != NULL);
+
+       cs->kpkc_flags = CALLSTACK_CONTINUATION | CALLSTACK_VALID | CALLSTACK_KERNEL;
+#ifdef __LP64__
+       cs->kpkc_flags |= CALLSTACK_64BIT;
+#endif
+
+       cs->kpkc_nframes = 1;
+       cs->kpkc_frames[0] = VM_KERNEL_UNSLIDE(thread->continuation);
+}
+
+void
+kperf_backtrace_sample(struct kp_kcallstack *cs, struct kperf_context *context)
+{
+       assert(cs != NULL);
+       assert(context != NULL);
+       assert(context->cur_thread == current_thread());
+
+       cs->kpkc_flags = CALLSTACK_KERNEL | CALLSTACK_KERNEL_WORDS;
+#ifdef __LP64__
+       cs->kpkc_flags |= CALLSTACK_64BIT;
+#endif
+
+       BUF_VERB(PERF_CS_BACKTRACE | DBG_FUNC_START, 1);
+
+       bool trunc = false;
+       cs->kpkc_nframes = backtrace_frame(cs->kpkc_word_frames,
+           cs->kpkc_nframes - 1, context->starting_fp, &trunc);
+       if (cs->kpkc_nframes > 0) {
+               cs->kpkc_flags |= CALLSTACK_VALID;
+               /*
+                * Fake the value pointed to by the stack pointer or the link
+                * register for symbolicators.
+                */
+               cs->kpkc_word_frames[cs->kpkc_nframes + 1] = 0;
+               cs->kpkc_nframes += 1;
+       }
+       if (trunc) {
+               cs->kpkc_flags |= CALLSTACK_TRUNCATED;
        }
 
        }
 
+       BUF_VERB(PERF_CS_BACKTRACE | DBG_FUNC_END, cs->kpkc_nframes);
 }
 
 }
 
+kern_return_t chudxnu_thread_get_callstack64_kperf(thread_t thread,
+    uint64_t *callStack, mach_msg_type_number_t *count,
+    boolean_t user_only);
+
 void
 void
-kperf_kcallstack_sample( struct callstack *cs, struct kperf_context *context )
+kperf_kcallstack_sample(struct kp_kcallstack *cs, struct kperf_context *context)
 {
 {
-       callstack_sample( cs, context, 0 );
+       thread_t thread;
+
+       assert(cs != NULL);
+       assert(context != NULL);
+       assert(cs->kpkc_nframes <= MAX_KCALLSTACK_FRAMES);
+
+       thread = context->cur_thread;
+       assert(thread != NULL);
+
+       BUF_INFO(PERF_CS_KSAMPLE | DBG_FUNC_START, (uintptr_t)thread_tid(thread),
+           cs->kpkc_nframes);
+
+       cs->kpkc_flags = CALLSTACK_KERNEL;
+#ifdef __LP64__
+       cs->kpkc_flags |= CALLSTACK_64BIT;
+#endif
+
+       if (ml_at_interrupt_context()) {
+               assert(thread == current_thread());
+               cs->kpkc_flags |= CALLSTACK_KERNEL_WORDS;
+               bool trunc = false;
+               cs->kpkc_nframes = backtrace_interrupted(
+                       cs->kpkc_word_frames, cs->kpkc_nframes - 1, &trunc);
+               if (cs->kpkc_nframes != 0) {
+                       callstack_fixup_interrupted(cs);
+               }
+               if (trunc) {
+                       cs->kpkc_flags |= CALLSTACK_TRUNCATED;
+               }
+       } else {
+               /*
+                * Rely on legacy CHUD backtracer to backtrace kernel stacks on
+                * other threads.
+                */
+               kern_return_t kr;
+               kr = chudxnu_thread_get_callstack64_kperf(thread,
+                   cs->kpkc_frames, &cs->kpkc_nframes, FALSE);
+               if (kr == KERN_SUCCESS) {
+                       cs->kpkc_flags |= CALLSTACK_VALID;
+               } else if (kr == KERN_RESOURCE_SHORTAGE) {
+                       cs->kpkc_flags |= CALLSTACK_VALID;
+                       cs->kpkc_flags |= CALLSTACK_TRUNCATED;
+               } else {
+                       cs->kpkc_nframes = 0;
+               }
+       }
+
+       if (!(cs->kpkc_flags & CALLSTACK_VALID)) {
+               BUF_INFO(PERF_CS_ERROR, ERR_GETSTACK);
+       }
+
+       BUF_INFO(PERF_CS_KSAMPLE | DBG_FUNC_END, (uintptr_t)thread_tid(thread),
+           cs->kpkc_flags, cs->kpkc_nframes);
 }
 
 void
 }
 
 void
-kperf_ucallstack_sample( struct callstack *cs, struct kperf_context *context )
+kperf_ucallstack_sample(struct kp_ucallstack *cs, struct kperf_context *context)
+{
+       assert(ml_get_interrupts_enabled() == TRUE);
+
+       thread_t thread = context->cur_thread;
+       assert(thread != NULL);
+
+       BUF_INFO(PERF_CS_USAMPLE | DBG_FUNC_START,
+           (uintptr_t)thread_tid(thread), cs->kpuc_nframes);
+
+       bool user64 = false;
+       bool trunc = false;
+       int error = 0;
+       /*
+        * Leave space for the fixup information.
+        */
+       unsigned int maxnframes = cs->kpuc_nframes - 1;
+       unsigned int nframes = backtrace_thread_user(thread, cs->kpuc_frames,
+           maxnframes, &error, &user64, &trunc, true);
+       cs->kpuc_nframes = MIN(maxnframes, nframes);
+
+       /*
+        * Ignore EFAULT to get as much of the stack as possible.  It will be
+        * marked as truncated, below.
+        */
+       if (error == 0 || error == EFAULT) {
+               callstack_fixup_user(cs, thread);
+               cs->kpuc_flags |= CALLSTACK_VALID;
+       } else {
+               cs->kpuc_nframes = 0;
+               BUF_INFO(PERF_CS_ERROR, ERR_GETSTACK, error);
+       }
+
+       cs->kpuc_flags |= CALLSTACK_KERNEL_WORDS | (user64 ? CALLSTACK_64BIT : 0) |
+           (trunc ? CALLSTACK_TRUNCATED : 0);
+
+       BUF_INFO(PERF_CS_USAMPLE | DBG_FUNC_END, (uintptr_t)thread_tid(thread),
+           cs->kpuc_flags, cs->kpuc_nframes);
+}
+
+static inline uintptr_t
+scrub_word(uintptr_t *bt, int n_frames, int frame, bool kern)
+{
+       if (frame < n_frames) {
+               if (kern) {
+                       return VM_KERNEL_UNSLIDE(bt[frame]);
+               } else {
+                       return bt[frame];
+               }
+       } else {
+               return 0;
+       }
+}
+
+static inline uintptr_t
+scrub_frame(uint64_t *bt, int n_frames, int frame)
 {
 {
-       callstack_sample( cs, context, 1 );
+       if (frame < n_frames) {
+               return (uintptr_t)(bt[frame]);
+       } else {
+               return 0;
+       }
 }
 
 static void
 }
 
 static void
-callstack_log( struct callstack *cs, uint32_t code )
+callstack_log(uint32_t hdrid, uint32_t dataid, void *vframes,
+    unsigned int nframes, unsigned int flags)
 {
 {
-       unsigned int i, j, n, of = 4;
+       BUF_VERB(PERF_CS_LOG | DBG_FUNC_START, flags, nframes);
+
+       BUF_DATA(hdrid, flags, nframes);
 
 
-       /* Header on the stack */
-       BUF_DATA2( code, cs->flags, cs->nframes );
+       unsigned int nevts = nframes / 4;
+       unsigned int ovf = nframes % 4;
+       if (ovf != 0) {
+               nevts++;
+       }
 
 
-       /* look for how many batches of 4 */
-       n  = cs->nframes / 4;
-       of = cs->nframes % 4;
-       if( of != 0 )
-               n++;
+       bool kern = flags & CALLSTACK_KERNEL;
 
 
-       /* print all the stack data, and zero the overflow */
-       for( i = 0; i < n; i++ )
-       {
-#define SCRUB_FRAME(x) (((x)<cs->nframes)?cs->frames[x]:0)
-               j = i * 4;
-               BUF_DATA ( code, 
-                          SCRUB_FRAME(j+0),
-                          SCRUB_FRAME(j+1),
-                          SCRUB_FRAME(j+2),
-                          SCRUB_FRAME(j+3) );
-#undef SCRUB_FRAME
+       if (flags & CALLSTACK_KERNEL_WORDS) {
+               uintptr_t *frames = vframes;
+               for (unsigned int i = 0; i < nevts; i++) {
+                       unsigned int j = i * 4;
+                       BUF_DATA(dataid,
+                           scrub_word(frames, nframes, j + 0, kern),
+                           scrub_word(frames, nframes, j + 1, kern),
+                           scrub_word(frames, nframes, j + 2, kern),
+                           scrub_word(frames, nframes, j + 3, kern));
+               }
+       } else {
+               for (unsigned int i = 0; i < nevts; i++) {
+                       uint64_t *frames = vframes;
+                       unsigned int j = i * 4;
+                       BUF_DATA(dataid,
+                           scrub_frame(frames, nframes, j + 0),
+                           scrub_frame(frames, nframes, j + 1),
+                           scrub_frame(frames, nframes, j + 2),
+                           scrub_frame(frames, nframes, j + 3));
+               }
        }
        }
+
+       BUF_VERB(PERF_CS_LOG | DBG_FUNC_END, flags, nframes);
 }
 
 void
 }
 
 void
-kperf_kcallstack_log( struct callstack *cs )
+kperf_kcallstack_log(struct kp_kcallstack *cs)
 {
 {
-       callstack_log( cs, PERF_CS_KDATA );
+       callstack_log(PERF_CS_KHDR, PERF_CS_KDATA, cs->kpkc_frames,
+           cs->kpkc_nframes, cs->kpkc_flags);
 }
 
 void
 }
 
 void
-kperf_ucallstack_log( struct callstack *cs )
+kperf_ucallstack_log(struct kp_ucallstack *cs)
 {
 {
-       callstack_log( cs, PERF_CS_UDATA );
+       callstack_log(PERF_CS_UHDR, PERF_CS_UDATA, cs->kpuc_frames,
+           cs->kpuc_nframes, cs->kpuc_flags);
 }
 
 int
 }
 
 int
-kperf_ucallstack_pend( struct kperf_context * context )
+kperf_ucallstack_pend(struct kperf_context * context, uint32_t depth,
+    unsigned int actionid)
+{
+       if (depth < 2) {
+               panic("HUH");
+       }
+       kperf_ast_set_callstack_depth(context->cur_thread, depth);
+       return kperf_ast_pend(context->cur_thread, T_KPERF_AST_CALLSTACK,
+                  actionid);
+}
+
+static kern_return_t
+chudxnu_kern_read(void *dstaddr, vm_offset_t srcaddr, vm_size_t size)
+{
+       return (ml_nofault_copy(srcaddr, (vm_offset_t)dstaddr, size) == size) ?
+              KERN_SUCCESS : KERN_FAILURE;
+}
+
+static kern_return_t
+chudxnu_task_read(
+       task_t      task,
+       void        *kernaddr,
+       uint64_t    usraddr,
+       vm_size_t   size)
+{
+       //ppc version ported to arm
+       kern_return_t ret = KERN_SUCCESS;
+
+       if (ml_at_interrupt_context()) {
+               return KERN_FAILURE;    // can't look at tasks on interrupt stack
+       }
+
+       if (current_task() == task) {
+               if (copyin(usraddr, kernaddr, size)) {
+                       ret = KERN_FAILURE;
+               }
+       } else {
+               vm_map_t map = get_task_map(task);
+               ret = vm_map_read_user(map, usraddr, kernaddr, size);
+       }
+
+       return ret;
+}
+
+static inline uint64_t
+chudxnu_vm_unslide( uint64_t ptr, int kaddr )
+{
+       if (!kaddr) {
+               return ptr;
+       }
+
+       return VM_KERNEL_UNSLIDE(ptr);
+}
+
+#if __arm__
+#define ARM_SUPERVISOR_MODE(cpsr) ((((cpsr) & PSR_MODE_MASK) != PSR_USER_MODE) ? TRUE : FALSE)
+#define CS_FLAG_EXTRASP  1  // capture extra sp register
+static kern_return_t
+chudxnu_thread_get_callstack64_internal(
+       thread_t                thread,
+       uint64_t                *callStack,
+       mach_msg_type_number_t  *count,
+       boolean_t               user_only,
+       int flags)
+{
+       kern_return_t kr;
+       task_t                  task;
+       uint64_t                currPC = 0ULL, currLR = 0ULL, currSP = 0ULL;
+       uint64_t                prevPC = 0ULL;
+       uint32_t                kernStackMin = thread->kernel_stack;
+       uint32_t                kernStackMax = kernStackMin + kernel_stack_size;
+       uint64_t       *buffer = callStack;
+       uint32_t                frame[2];
+       int             bufferIndex = 0;
+       int             bufferMaxIndex = 0;
+       boolean_t       supervisor = FALSE;
+       struct arm_saved_state *state = NULL;
+       uint32_t                *fp = NULL, *nextFramePointer = NULL, *topfp = NULL;
+       uint64_t                pc = 0ULL;
+
+       task = get_threadtask(thread);
+
+       bufferMaxIndex = *count;
+       //get thread state
+       if (user_only) {
+               state = find_user_regs(thread);
+       } else {
+               state = find_kern_regs(thread);
+       }
+
+       if (!state) {
+               *count = 0;
+               return KERN_FAILURE;
+       }
+
+       /* make sure it is safe to dereference before you do it */
+       supervisor = ARM_SUPERVISOR_MODE(state->cpsr);
+
+       /* can't take a kernel callstack if we've got a user frame */
+       if (!user_only && !supervisor) {
+               return KERN_FAILURE;
+       }
+
+       /*
+        * Reserve space for saving LR (and sometimes SP) at the end of the
+        * backtrace.
+        */
+       if (flags & CS_FLAG_EXTRASP) {
+               bufferMaxIndex -= 2;
+       } else {
+               bufferMaxIndex -= 1;
+       }
+
+       if (bufferMaxIndex < 2) {
+               *count = 0;
+               return KERN_RESOURCE_SHORTAGE;
+       }
+
+       currPC = (uint64_t)state->pc; /* r15 */
+       if (state->cpsr & PSR_TF) {
+               currPC |= 1ULL; /* encode thumb mode into low bit of PC */
+       }
+       currLR = (uint64_t)state->lr; /* r14 */
+       currSP = (uint64_t)state->sp; /* r13 */
+
+       fp = (uint32_t *)state->r[7]; /* frame pointer */
+       topfp = fp;
+
+       bufferIndex = 0;  // start with a stack of size zero
+       buffer[bufferIndex++] = chudxnu_vm_unslide(currPC, supervisor); // save PC in position 0.
+
+       // Now, fill buffer with stack backtraces.
+       while (bufferIndex < bufferMaxIndex) {
+               pc = 0ULL;
+               /*
+                * Below the frame pointer, the following values are saved:
+                * -> FP
+                */
+
+               /*
+                * Note that we read the pc even for the first stack frame
+                * (which, in theory, is always empty because the callee fills
+                * it in just before it lowers the stack.  However, if we
+                * catch the program in between filling in the return address
+                * and lowering the stack, we want to still have a valid
+                * backtrace. FixupStack correctly disregards this value if
+                * necessary.
+                */
+
+               if ((uint32_t)fp == 0 || ((uint32_t)fp & 0x3) != 0) {
+                       /* frame pointer is invalid - stop backtracing */
+                       pc = 0ULL;
+                       break;
+               }
+
+               if (supervisor) {
+                       if (((uint32_t)fp > kernStackMax) ||
+                           ((uint32_t)fp < kernStackMin)) {
+                               kr = KERN_FAILURE;
+                       } else {
+                               kr = chudxnu_kern_read(&frame,
+                                   (vm_offset_t)fp,
+                                   (vm_size_t)sizeof(frame));
+                               if (kr == KERN_SUCCESS) {
+                                       pc = (uint64_t)frame[1];
+                                       nextFramePointer = (uint32_t *) (frame[0]);
+                               } else {
+                                       pc = 0ULL;
+                                       nextFramePointer = 0ULL;
+                                       kr = KERN_FAILURE;
+                               }
+                       }
+               } else {
+                       kr = chudxnu_task_read(task,
+                           &frame,
+                           (((uint64_t)(uint32_t)fp) & 0x00000000FFFFFFFFULL),
+                           sizeof(frame));
+                       if (kr == KERN_SUCCESS) {
+                               pc = (uint64_t) frame[1];
+                               nextFramePointer = (uint32_t *) (frame[0]);
+                       } else {
+                               pc = 0ULL;
+                               nextFramePointer = 0ULL;
+                               kr = KERN_FAILURE;
+                       }
+               }
+
+               if (kr != KERN_SUCCESS) {
+                       pc = 0ULL;
+                       break;
+               }
+
+               if (nextFramePointer) {
+                       buffer[bufferIndex++] = chudxnu_vm_unslide(pc, supervisor);
+                       prevPC = pc;
+               }
+
+               if (nextFramePointer < fp) {
+                       break;
+               } else {
+                       fp = nextFramePointer;
+               }
+       }
+
+       if (bufferIndex >= bufferMaxIndex) {
+               bufferIndex = bufferMaxIndex;
+               kr = KERN_RESOURCE_SHORTAGE;
+       } else {
+               kr = KERN_SUCCESS;
+       }
+
+       // Save link register and R13 (sp) at bottom of stack (used for later fixup).
+       buffer[bufferIndex++] = chudxnu_vm_unslide(currLR, supervisor);
+       if (flags & CS_FLAG_EXTRASP) {
+               buffer[bufferIndex++] = chudxnu_vm_unslide(currSP, supervisor);
+       }
+
+       *count = bufferIndex;
+       return kr;
+}
+
+kern_return_t
+chudxnu_thread_get_callstack64_kperf(
+       thread_t                thread,
+       uint64_t                *callStack,
+       mach_msg_type_number_t  *count,
+       boolean_t               user_only)
 {
 {
-       return kperf_ast_pend( context->cur_thread, T_AST_CALLSTACK,
-                              T_AST_CALLSTACK );
+       return chudxnu_thread_get_callstack64_internal( thread, callStack, count, user_only, 0 );
 }
 }
+#elif __arm64__
+
+#if defined(HAS_APPLE_PAC)
+#include <ptrauth.h>
+#endif
+
+// chudxnu_thread_get_callstack gathers a raw callstack along with any information needed to
+// fix it up later (in case we stopped program as it was saving values into prev stack frame, etc.)
+// after sampling has finished.
+//
+// For an N-entry callstack:
+//
+// [0]      current pc
+// [1..N-3] stack frames (including current one)
+// [N-2]    current LR (return value if we're in a leaf function)
+// [N-1]    current r0 (in case we've saved LR in r0) (optional)
+//
+//
+#define ARM_SUPERVISOR_MODE(cpsr) ((((cpsr) & PSR_MODE_MASK) != PSR_USER_MODE) ? TRUE : FALSE)
 
 
-//     kr = chudxnu_thread_get_callstack(context->generic->threadID, 
-//              (uint32_t*)frames, &frameCount, !collectingSupervisorStack);
+#define CS_FLAG_EXTRASP  1  // capture extra sp register
+
+static kern_return_t
+chudxnu_thread_get_callstack64_internal(
+       thread_t                thread,
+       uint64_t                *callStack,
+       mach_msg_type_number_t  *count,
+       boolean_t               user_only,
+       int flags)
+{
+       kern_return_t   kr = KERN_SUCCESS;
+       task_t                  task;
+       uint64_t                currPC = 0ULL, currLR = 0ULL, currSP = 0ULL;
+       uint64_t                prevPC = 0ULL;
+       uint64_t                kernStackMin = thread->kernel_stack;
+       uint64_t                kernStackMax = kernStackMin + kernel_stack_size;
+       uint64_t       *buffer = callStack;
+       int             bufferIndex = 0;
+       int             bufferMaxIndex = 0;
+       boolean_t       kernel = FALSE;
+       struct arm_saved_state *sstate = NULL;
+       uint64_t                pc = 0ULL;
+
+       task = get_threadtask(thread);
+       bufferMaxIndex = *count;
+       //get thread state
+       if (user_only) {
+               sstate = find_user_regs(thread);
+       } else {
+               sstate = find_kern_regs(thread);
+       }
+
+       if (!sstate) {
+               *count = 0;
+               return KERN_FAILURE;
+       }
+
+       if (is_saved_state64(sstate)) {
+               struct arm_saved_state64 *state = NULL;
+               uint64_t *fp = NULL, *nextFramePointer = NULL, *topfp = NULL;
+               uint64_t frame[2];
+
+               state = saved_state64(sstate);
+
+               /* make sure it is safe to dereference before you do it */
+               kernel = PSR64_IS_KERNEL(state->cpsr);
+
+               /* can't take a kernel callstack if we've got a user frame */
+               if (!user_only && !kernel) {
+                       return KERN_FAILURE;
+               }
+
+               /*
+                * Reserve space for saving LR (and sometimes SP) at the end of the
+                * backtrace.
+                */
+               if (flags & CS_FLAG_EXTRASP) {
+                       bufferMaxIndex -= 2;
+               } else {
+                       bufferMaxIndex -= 1;
+               }
+
+               if (bufferMaxIndex < 2) {
+                       *count = 0;
+                       return KERN_RESOURCE_SHORTAGE;
+               }
+
+               currPC = state->pc;
+               currLR = state->lr;
+               currSP = state->sp;
+
+               fp = (uint64_t *)state->fp; /* frame pointer */
+               topfp = fp;
+
+               bufferIndex = 0;  // start with a stack of size zero
+               buffer[bufferIndex++] = chudxnu_vm_unslide(currPC, kernel); // save PC in position 0.
+
+               BUF_VERB(PERF_CS_BACKTRACE | DBG_FUNC_START, kernel, 0);
+
+               // Now, fill buffer with stack backtraces.
+               while (bufferIndex < bufferMaxIndex) {
+                       pc = 0ULL;
+                       /*
+                        * Below the frame pointer, the following values are saved:
+                        * -> FP
+                        */
+
+                       /*
+                        * Note that we read the pc even for the first stack frame
+                        * (which, in theory, is always empty because the callee fills
+                        * it in just before it lowers the stack.  However, if we
+                        * catch the program in between filling in the return address
+                        * and lowering the stack, we want to still have a valid
+                        * backtrace. FixupStack correctly disregards this value if
+                        * necessary.
+                        */
+
+                       if ((uint64_t)fp == 0 || ((uint64_t)fp & 0x3) != 0) {
+                               /* frame pointer is invalid - stop backtracing */
+                               pc = 0ULL;
+                               break;
+                       }
+
+                       if (kernel) {
+                               if (((uint64_t)fp > kernStackMax) ||
+                                   ((uint64_t)fp < kernStackMin)) {
+                                       kr = KERN_FAILURE;
+                               } else {
+                                       kr = chudxnu_kern_read(&frame,
+                                           (vm_offset_t)fp,
+                                           (vm_size_t)sizeof(frame));
+                                       if (kr == KERN_SUCCESS) {
+#if defined(HAS_APPLE_PAC)
+                                               /* return addresses on stack will be signed by arm64e ABI */
+                                               pc = (uint64_t)ptrauth_strip((void *)frame[1], ptrauth_key_return_address);
+#else
+                                               pc = frame[1];
+#endif
+                                               nextFramePointer = (uint64_t *)frame[0];
+                                       } else {
+                                               pc = 0ULL;
+                                               nextFramePointer = 0ULL;
+                                               kr = KERN_FAILURE;
+                                       }
+                               }
+                       } else {
+                               kr = chudxnu_task_read(task,
+                                   &frame,
+                                   (vm_offset_t)fp,
+                                   (vm_size_t)sizeof(frame));
+                               if (kr == KERN_SUCCESS) {
+#if defined(HAS_APPLE_PAC)
+                                       /* return addresses on stack will be signed by arm64e ABI */
+                                       pc = (uint64_t)ptrauth_strip((void *)frame[1], ptrauth_key_return_address);
+#else
+                                       pc = frame[1];
+#endif
+                                       nextFramePointer = (uint64_t *)(frame[0]);
+                               } else {
+                                       pc = 0ULL;
+                                       nextFramePointer = 0ULL;
+                                       kr = KERN_FAILURE;
+                               }
+                       }
+
+                       if (kr != KERN_SUCCESS) {
+                               pc = 0ULL;
+                               break;
+                       }
+
+                       if (nextFramePointer) {
+                               buffer[bufferIndex++] = chudxnu_vm_unslide(pc, kernel);
+                               prevPC = pc;
+                       }
+
+                       if (nextFramePointer < fp) {
+                               break;
+                       } else {
+                               fp = nextFramePointer;
+                       }
+               }
+
+               BUF_VERB(PERF_CS_BACKTRACE | DBG_FUNC_END, bufferIndex);
+
+               if (bufferIndex >= bufferMaxIndex) {
+                       bufferIndex = bufferMaxIndex;
+                       kr = KERN_RESOURCE_SHORTAGE;
+               } else {
+                       kr = KERN_SUCCESS;
+               }
+
+               // Save link register and SP at bottom of stack (used for later fixup).
+               buffer[bufferIndex++] = chudxnu_vm_unslide(currLR, kernel);
+               if (flags & CS_FLAG_EXTRASP) {
+                       buffer[bufferIndex++] = chudxnu_vm_unslide(currSP, kernel);
+               }
+       } else {
+               struct arm_saved_state32 *state = NULL;
+               uint32_t *fp = NULL, *nextFramePointer = NULL, *topfp = NULL;
+
+               /* 64-bit kernel stacks, 32-bit user stacks */
+               uint64_t frame[2];
+               uint32_t frame32[2];
+
+               state = saved_state32(sstate);
+
+               /* make sure it is safe to dereference before you do it */
+               kernel = ARM_SUPERVISOR_MODE(state->cpsr);
+
+               /* can't take a kernel callstack if we've got a user frame */
+               if (!user_only && !kernel) {
+                       return KERN_FAILURE;
+               }
+
+               /*
+                * Reserve space for saving LR (and sometimes SP) at the end of the
+                * backtrace.
+                */
+               if (flags & CS_FLAG_EXTRASP) {
+                       bufferMaxIndex -= 2;
+               } else {
+                       bufferMaxIndex -= 1;
+               }
+
+               if (bufferMaxIndex < 2) {
+                       *count = 0;
+                       return KERN_RESOURCE_SHORTAGE;
+               }
+
+               currPC = (uint64_t)state->pc; /* r15 */
+               if (state->cpsr & PSR_TF) {
+                       currPC |= 1ULL; /* encode thumb mode into low bit of PC */
+               }
+               currLR = (uint64_t)state->lr; /* r14 */
+               currSP = (uint64_t)state->sp; /* r13 */
+
+               fp = (uint32_t *)(uintptr_t)state->r[7]; /* frame pointer */
+               topfp = fp;
+
+               bufferIndex = 0;  // start with a stack of size zero
+               buffer[bufferIndex++] = chudxnu_vm_unslide(currPC, kernel); // save PC in position 0.
+
+               BUF_VERB(PERF_CS_BACKTRACE | DBG_FUNC_START, kernel, 1);
+
+               // Now, fill buffer with stack backtraces.
+               while (bufferIndex < bufferMaxIndex) {
+                       pc = 0ULL;
+                       /*
+                        * Below the frame pointer, the following values are saved:
+                        * -> FP
+                        */
+
+                       /*
+                        * Note that we read the pc even for the first stack frame
+                        * (which, in theory, is always empty because the callee fills
+                        * it in just before it lowers the stack.  However, if we
+                        * catch the program in between filling in the return address
+                        * and lowering the stack, we want to still have a valid
+                        * backtrace. FixupStack correctly disregards this value if
+                        * necessary.
+                        */
+
+                       if ((uint32_t)fp == 0 || ((uint32_t)fp & 0x3) != 0) {
+                               /* frame pointer is invalid - stop backtracing */
+                               pc = 0ULL;
+                               break;
+                       }
+
+                       if (kernel) {
+                               if (((uint32_t)fp > kernStackMax) ||
+                                   ((uint32_t)fp < kernStackMin)) {
+                                       kr = KERN_FAILURE;
+                               } else {
+                                       kr = chudxnu_kern_read(&frame,
+                                           (vm_offset_t)fp,
+                                           (vm_size_t)sizeof(frame));
+                                       if (kr == KERN_SUCCESS) {
+                                               pc = (uint64_t)frame[1];
+                                               nextFramePointer = (uint32_t *) (frame[0]);
+                                       } else {
+                                               pc = 0ULL;
+                                               nextFramePointer = 0ULL;
+                                               kr = KERN_FAILURE;
+                                       }
+                               }
+                       } else {
+                               kr = chudxnu_task_read(task,
+                                   &frame32,
+                                   (((uint64_t)(uint32_t)fp) & 0x00000000FFFFFFFFULL),
+                                   sizeof(frame32));
+                               if (kr == KERN_SUCCESS) {
+                                       pc = (uint64_t)frame32[1];
+                                       nextFramePointer = (uint32_t *)(uintptr_t)(frame32[0]);
+                               } else {
+                                       pc = 0ULL;
+                                       nextFramePointer = 0ULL;
+                                       kr = KERN_FAILURE;
+                               }
+                       }
+
+                       if (kr != KERN_SUCCESS) {
+                               pc = 0ULL;
+                               break;
+                       }
+
+                       if (nextFramePointer) {
+                               buffer[bufferIndex++] = chudxnu_vm_unslide(pc, kernel);
+                               prevPC = pc;
+                       }
+
+                       if (nextFramePointer < fp) {
+                               break;
+                       } else {
+                               fp = nextFramePointer;
+                       }
+               }
+
+               BUF_VERB(PERF_CS_BACKTRACE | DBG_FUNC_END, bufferIndex);
+
+               /* clamp callstack size to max */
+               if (bufferIndex >= bufferMaxIndex) {
+                       bufferIndex = bufferMaxIndex;
+                       kr = KERN_RESOURCE_SHORTAGE;
+               } else {
+                       /* ignore all other failures */
+                       kr = KERN_SUCCESS;
+               }
+
+               // Save link register and R13 (sp) at bottom of stack (used for later fixup).
+               buffer[bufferIndex++] = chudxnu_vm_unslide(currLR, kernel);
+               if (flags & CS_FLAG_EXTRASP) {
+                       buffer[bufferIndex++] = chudxnu_vm_unslide(currSP, kernel);
+               }
+       }
+
+       *count = bufferIndex;
+       return kr;
+}
+
+kern_return_t
+chudxnu_thread_get_callstack64_kperf(
+       thread_t                thread,
+       uint64_t                *callStack,
+       mach_msg_type_number_t  *count,
+       boolean_t               user_only)
+{
+       return chudxnu_thread_get_callstack64_internal( thread, callStack, count, user_only, 0 );
+}
+#elif __x86_64__
+
+#define VALID_STACK_ADDRESS(supervisor, addr, minKernAddr, maxKernAddr)   (supervisor ? (addr>=minKernAddr && addr<=maxKernAddr) : TRUE)
+// don't try to read in the hole
+#define VALID_STACK_ADDRESS64(supervisor, addr, minKernAddr, maxKernAddr) \
+(supervisor ? ((uint64_t)addr >= minKernAddr && (uint64_t)addr <= maxKernAddr) : \
+((uint64_t)addr != 0ULL && ((uint64_t)addr <= 0x00007FFFFFFFFFFFULL || (uint64_t)addr >= 0xFFFF800000000000ULL)))
+
+typedef struct _cframe64_t {
+       uint64_t        prevFP;         // can't use a real pointer here until we're a 64 bit kernel
+       uint64_t        caller;
+       uint64_t        args[0];
+}cframe64_t;
+
+
+typedef struct _cframe_t {
+       uint32_t                prev;   // this is really a user32-space pointer to the previous frame
+       uint32_t                caller;
+       uint32_t                args[0];
+} cframe_t;
+
+extern void * find_user_regs(thread_t);
+extern x86_saved_state32_t *find_kern_regs(thread_t);
+
+static kern_return_t
+do_kernel_backtrace(
+       thread_t thread,
+       struct x86_kernel_state *regs,
+       uint64_t *frames,
+       mach_msg_type_number_t *start_idx,
+       mach_msg_type_number_t max_idx)
+{
+       uint64_t kernStackMin = (uint64_t)thread->kernel_stack;
+       uint64_t kernStackMax = (uint64_t)kernStackMin + kernel_stack_size;
+       mach_msg_type_number_t ct = *start_idx;
+       kern_return_t kr = KERN_FAILURE;
+
+#if __LP64__
+       uint64_t currPC = 0ULL;
+       uint64_t currFP = 0ULL;
+       uint64_t prevPC = 0ULL;
+       uint64_t prevFP = 0ULL;
+       if (KERN_SUCCESS != chudxnu_kern_read(&currPC, (vm_offset_t)&(regs->k_rip), sizeof(uint64_t))) {
+               return KERN_FAILURE;
+       }
+       if (KERN_SUCCESS != chudxnu_kern_read(&currFP, (vm_offset_t)&(regs->k_rbp), sizeof(uint64_t))) {
+               return KERN_FAILURE;
+       }
+#else
+       uint32_t currPC = 0U;
+       uint32_t currFP = 0U;
+       uint32_t prevPC = 0U;
+       uint32_t prevFP = 0U;
+       if (KERN_SUCCESS != chudxnu_kern_read(&currPC, (vm_offset_t)&(regs->k_eip), sizeof(uint32_t))) {
+               return KERN_FAILURE;
+       }
+       if (KERN_SUCCESS != chudxnu_kern_read(&currFP, (vm_offset_t)&(regs->k_ebp), sizeof(uint32_t))) {
+               return KERN_FAILURE;
+       }
+#endif
+
+       if (*start_idx >= max_idx) {
+               return KERN_RESOURCE_SHORTAGE;  // no frames traced
+       }
+       if (!currPC) {
+               return KERN_FAILURE;
+       }
+
+       frames[ct++] = chudxnu_vm_unslide((uint64_t)currPC, 1);
+
+       // build a backtrace of this kernel state
+#if __LP64__
+       while (VALID_STACK_ADDRESS64(TRUE, currFP, kernStackMin, kernStackMax)) {
+               // this is the address where caller lives in the user thread
+               uint64_t caller = currFP + sizeof(uint64_t);
+#else
+       while (VALID_STACK_ADDRESS(TRUE, currFP, kernStackMin, kernStackMax)) {
+               uint32_t caller = (uint32_t)currFP + sizeof(uint32_t);
+#endif
+
+               if (!currFP || !currPC) {
+                       currPC = 0;
+                       break;
+               }
+
+               if (ct >= max_idx) {
+                       *start_idx = ct;
+                       return KERN_RESOURCE_SHORTAGE;
+               }
+
+               /* read our caller */
+               kr = chudxnu_kern_read(&currPC, (vm_offset_t)caller, sizeof(currPC));
+
+               if (kr != KERN_SUCCESS || !currPC) {
+                       currPC = 0UL;
+                       break;
+               }
+
+               /*
+                * retrive contents of the frame pointer and advance to the next stack
+                * frame if it's valid
+                */
+               prevFP = 0;
+               kr = chudxnu_kern_read(&prevFP, (vm_offset_t)currFP, sizeof(currPC));
+
+#if __LP64__
+               if (VALID_STACK_ADDRESS64(TRUE, prevFP, kernStackMin, kernStackMax)) {
+#else
+               if (VALID_STACK_ADDRESS(TRUE, prevFP, kernStackMin, kernStackMax)) {
+#endif
+                       frames[ct++] = chudxnu_vm_unslide((uint64_t)currPC, 1);
+                       prevPC = currPC;
+               }
+               if (prevFP <= currFP) {
+                       break;
+               } else {
+                       currFP = prevFP;
+               }
+       }
+
+       *start_idx = ct;
+       return KERN_SUCCESS;
+}
+
+
+
+static kern_return_t
+do_backtrace32(
+       task_t task,
+       thread_t thread,
+       x86_saved_state32_t *regs,
+       uint64_t *frames,
+       mach_msg_type_number_t *start_idx,
+       mach_msg_type_number_t max_idx,
+       boolean_t supervisor)
+{
+       uint32_t tmpWord = 0UL;
+       uint64_t currPC = (uint64_t) regs->eip;
+       uint64_t currFP = (uint64_t) regs->ebp;
+       uint64_t prevPC = 0ULL;
+       uint64_t prevFP = 0ULL;
+       uint64_t kernStackMin = thread->kernel_stack;
+       uint64_t kernStackMax = kernStackMin + kernel_stack_size;
+       mach_msg_type_number_t ct = *start_idx;
+       kern_return_t kr = KERN_FAILURE;
+
+       if (ct >= max_idx) {
+               return KERN_RESOURCE_SHORTAGE;  // no frames traced
+       }
+       frames[ct++] = chudxnu_vm_unslide(currPC, supervisor);
+
+       // build a backtrace of this 32 bit state.
+       while (VALID_STACK_ADDRESS(supervisor, currFP, kernStackMin, kernStackMax)) {
+               cframe_t *fp = (cframe_t *) (uintptr_t) currFP;
+
+               if (!currFP) {
+                       currPC = 0;
+                       break;
+               }
+
+               if (ct >= max_idx) {
+                       *start_idx = ct;
+                       return KERN_RESOURCE_SHORTAGE;
+               }
+
+               /* read our caller */
+               if (supervisor) {
+                       kr = chudxnu_kern_read(&tmpWord, (vm_offset_t) &fp->caller, sizeof(uint32_t));
+               } else {
+                       kr = chudxnu_task_read(task, &tmpWord, (vm_offset_t) &fp->caller, sizeof(uint32_t));
+               }
+
+               if (kr != KERN_SUCCESS) {
+                       currPC = 0ULL;
+                       break;
+               }
+
+               currPC = (uint64_t) tmpWord;    // promote 32 bit address
+
+               /*
+                * retrive contents of the frame pointer and advance to the next stack
+                * frame if it's valid
+                */
+               prevFP = 0;
+               if (supervisor) {
+                       kr = chudxnu_kern_read(&tmpWord, (vm_offset_t)&fp->prev, sizeof(uint32_t));
+               } else {
+                       kr = chudxnu_task_read(task, &tmpWord, (vm_offset_t)&fp->prev, sizeof(uint32_t));
+               }
+               prevFP = (uint64_t) tmpWord;    // promote 32 bit address
+
+               if (prevFP) {
+                       frames[ct++] = chudxnu_vm_unslide(currPC, supervisor);
+                       prevPC = currPC;
+               }
+               if (prevFP < currFP) {
+                       break;
+               } else {
+                       currFP = prevFP;
+               }
+       }
+
+       *start_idx = ct;
+       return KERN_SUCCESS;
+}
+
+static kern_return_t
+do_backtrace64(
+       task_t task,
+       thread_t thread,
+       x86_saved_state64_t *regs,
+       uint64_t *frames,
+       mach_msg_type_number_t *start_idx,
+       mach_msg_type_number_t max_idx,
+       boolean_t supervisor)
+{
+       uint64_t currPC = regs->isf.rip;
+       uint64_t currFP = regs->rbp;
+       uint64_t prevPC = 0ULL;
+       uint64_t prevFP = 0ULL;
+       uint64_t kernStackMin = (uint64_t)thread->kernel_stack;
+       uint64_t kernStackMax = (uint64_t)kernStackMin + kernel_stack_size;
+       mach_msg_type_number_t ct = *start_idx;
+       kern_return_t kr = KERN_FAILURE;
+
+       if (*start_idx >= max_idx) {
+               return KERN_RESOURCE_SHORTAGE;  // no frames traced
+       }
+       frames[ct++] = chudxnu_vm_unslide(currPC, supervisor);
+
+       // build a backtrace of this 32 bit state.
+       while (VALID_STACK_ADDRESS64(supervisor, currFP, kernStackMin, kernStackMax)) {
+               // this is the address where caller lives in the user thread
+               uint64_t caller = currFP + sizeof(uint64_t);
+
+               if (!currFP) {
+                       currPC = 0;
+                       break;
+               }
+
+               if (ct >= max_idx) {
+                       *start_idx = ct;
+                       return KERN_RESOURCE_SHORTAGE;
+               }
+
+               /* read our caller */
+               if (supervisor) {
+                       kr = chudxnu_kern_read(&currPC, (vm_offset_t)caller, sizeof(uint64_t));
+               } else {
+                       kr = chudxnu_task_read(task, &currPC, caller, sizeof(uint64_t));
+               }
+
+               if (kr != KERN_SUCCESS) {
+                       currPC = 0ULL;
+                       break;
+               }
+
+               /*
+                * retrive contents of the frame pointer and advance to the next stack
+                * frame if it's valid
+                */
+               prevFP = 0;
+               if (supervisor) {
+                       kr = chudxnu_kern_read(&prevFP, (vm_offset_t)currFP, sizeof(uint64_t));
+               } else {
+                       kr = chudxnu_task_read(task, &prevFP, currFP, sizeof(uint64_t));
+               }
+
+               if (VALID_STACK_ADDRESS64(supervisor, prevFP, kernStackMin, kernStackMax)) {
+                       frames[ct++] = chudxnu_vm_unslide(currPC, supervisor);
+                       prevPC = currPC;
+               }
+               if (prevFP < currFP) {
+                       break;
+               } else {
+                       currFP = prevFP;
+               }
+       }
+
+       *start_idx = ct;
+       return KERN_SUCCESS;
+}
+
+static kern_return_t
+chudxnu_thread_get_callstack64_internal(
+       thread_t                thread,
+       uint64_t                *callstack,
+       mach_msg_type_number_t  *count,
+       boolean_t               user_only,
+       boolean_t               kern_only)
+{
+       kern_return_t kr = KERN_FAILURE;
+       task_t task = thread->task;
+       uint64_t currPC = 0ULL;
+       boolean_t supervisor = FALSE;
+       mach_msg_type_number_t bufferIndex = 0;
+       mach_msg_type_number_t bufferMaxIndex = *count;
+       x86_saved_state_t *tagged_regs = NULL;          // kernel register state
+       x86_saved_state64_t *regs64 = NULL;
+       x86_saved_state32_t *regs32 = NULL;
+       x86_saved_state32_t *u_regs32 = NULL;
+       x86_saved_state64_t *u_regs64 = NULL;
+       struct x86_kernel_state *kregs = NULL;
+
+       if (ml_at_interrupt_context()) {
+               if (user_only) {
+                       /* can't backtrace user state on interrupt stack. */
+                       return KERN_FAILURE;
+               }
+
+               /* backtracing at interrupt context? */
+               if (thread == current_thread() && current_cpu_datap()->cpu_int_state) {
+                       /*
+                        * Locate the registers for the interrupted thread, assuming it is
+                        * current_thread().
+                        */
+                       tagged_regs = current_cpu_datap()->cpu_int_state;
+
+                       if (is_saved_state64(tagged_regs)) {
+                               /* 64 bit registers */
+                               regs64 = saved_state64(tagged_regs);
+                               supervisor = ((regs64->isf.cs & SEL_PL) != SEL_PL_U);
+                       } else {
+                               /* 32 bit registers */
+                               regs32 = saved_state32(tagged_regs);
+                               supervisor = ((regs32->cs & SEL_PL) != SEL_PL_U);
+                       }
+               }
+       }
+
+       if (!ml_at_interrupt_context() && kernel_task == task) {
+               if (!thread->kernel_stack) {
+                       return KERN_FAILURE;
+               }
+
+               // Kernel thread not at interrupt context
+               kregs = (struct x86_kernel_state *)NULL;
+
+               // nofault read of the thread->kernel_stack pointer
+               if (KERN_SUCCESS != chudxnu_kern_read(&kregs, (vm_offset_t)&(thread->kernel_stack), sizeof(void *))) {
+                       return KERN_FAILURE;
+               }
+
+               // Adjust to find the saved kernel state
+               kregs = STACK_IKS((vm_offset_t)(uintptr_t)kregs);
+
+               supervisor = TRUE;
+       } else if (!tagged_regs) {
+               /*
+                * not at interrupt context, or tracing a different thread than
+                * current_thread() at interrupt context
+                */
+               tagged_regs = USER_STATE(thread);
+               if (is_saved_state64(tagged_regs)) {
+                       /* 64 bit registers */
+                       regs64 = saved_state64(tagged_regs);
+                       supervisor = ((regs64->isf.cs & SEL_PL) != SEL_PL_U);
+               } else {
+                       /* 32 bit registers */
+                       regs32 = saved_state32(tagged_regs);
+                       supervisor = ((regs32->cs & SEL_PL) != SEL_PL_U);
+               }
+       }
+
+       *count = 0;
+
+       if (supervisor) {
+               // the caller only wants a user callstack.
+               if (user_only) {
+                       // bail - we've only got kernel state
+                       return KERN_FAILURE;
+               }
+       } else {
+               // regs32(64) is not in supervisor mode.
+               u_regs32 = regs32;
+               u_regs64 = regs64;
+               regs32 = NULL;
+               regs64 = NULL;
+       }
+
+       if (user_only) {
+               /* we only want to backtrace the user mode */
+               if (!(u_regs32 || u_regs64)) {
+                       /* no user state to look at */
+                       return KERN_FAILURE;
+               }
+       }
+
+       /*
+        * Order of preference for top of stack:
+        * 64 bit kernel state (not likely)
+        * 32 bit kernel state
+        * 64 bit user land state
+        * 32 bit user land state
+        */
+
+       if (kregs) {
+               /*
+                * nofault read of the registers from the kernel stack (as they can
+                * disappear on the fly).
+                */
+
+               if (KERN_SUCCESS != chudxnu_kern_read(&currPC, (vm_offset_t)&(kregs->k_rip), sizeof(uint64_t))) {
+                       return KERN_FAILURE;
+               }
+       } else if (regs64) {
+               currPC = regs64->isf.rip;
+       } else if (regs32) {
+               currPC = (uint64_t) regs32->eip;
+       } else if (u_regs64) {
+               currPC = u_regs64->isf.rip;
+       } else if (u_regs32) {
+               currPC = (uint64_t) u_regs32->eip;
+       }
+
+       if (!currPC) {
+               /* no top of the stack, bail out */
+               return KERN_FAILURE;
+       }
+
+       bufferIndex = 0;
+
+       if (bufferMaxIndex < 1) {
+               *count = 0;
+               return KERN_RESOURCE_SHORTAGE;
+       }
+
+       /* backtrace kernel */
+       if (kregs) {
+               addr64_t address = 0ULL;
+               size_t size = 0UL;
+
+               // do the backtrace
+               kr = do_kernel_backtrace(thread, kregs, callstack, &bufferIndex, bufferMaxIndex);
+
+               // and do a nofault read of (r|e)sp
+               uint64_t rsp = 0ULL;
+               size = sizeof(uint64_t);
+
+               if (KERN_SUCCESS != chudxnu_kern_read(&address, (vm_offset_t)&(kregs->k_rsp), size)) {
+                       address = 0ULL;
+               }
+
+               if (address && KERN_SUCCESS == chudxnu_kern_read(&rsp, (vm_offset_t)address, size) && bufferIndex < bufferMaxIndex) {
+                       callstack[bufferIndex++] = (uint64_t)rsp;
+               }
+       } else if (regs64) {
+               uint64_t rsp = 0ULL;
+
+               // backtrace the 64bit side.
+               kr = do_backtrace64(task, thread, regs64, callstack, &bufferIndex,
+                   bufferMaxIndex - 1, TRUE);
+
+               if (KERN_SUCCESS == chudxnu_kern_read(&rsp, (vm_offset_t) regs64->isf.rsp, sizeof(uint64_t)) &&
+                   bufferIndex < bufferMaxIndex) {
+                       callstack[bufferIndex++] = rsp;
+               }
+       } else if (regs32) {
+               uint32_t esp = 0UL;
+
+               // backtrace the 32bit side.
+               kr = do_backtrace32(task, thread, regs32, callstack, &bufferIndex,
+                   bufferMaxIndex - 1, TRUE);
+
+               if (KERN_SUCCESS == chudxnu_kern_read(&esp, (vm_offset_t) regs32->uesp, sizeof(uint32_t)) &&
+                   bufferIndex < bufferMaxIndex) {
+                       callstack[bufferIndex++] = (uint64_t) esp;
+               }
+       } else if (u_regs64 && !kern_only) {
+               /* backtrace user land */
+               uint64_t rsp = 0ULL;
+
+               kr = do_backtrace64(task, thread, u_regs64, callstack, &bufferIndex,
+                   bufferMaxIndex - 1, FALSE);
+
+               if (KERN_SUCCESS == chudxnu_task_read(task, &rsp, (addr64_t) u_regs64->isf.rsp, sizeof(uint64_t)) &&
+                   bufferIndex < bufferMaxIndex) {
+                       callstack[bufferIndex++] = rsp;
+               }
+       } else if (u_regs32 && !kern_only) {
+               uint32_t esp = 0UL;
+
+               kr = do_backtrace32(task, thread, u_regs32, callstack, &bufferIndex,
+                   bufferMaxIndex - 1, FALSE);
+
+               if (KERN_SUCCESS == chudxnu_task_read(task, &esp, (addr64_t) u_regs32->uesp, sizeof(uint32_t)) &&
+                   bufferIndex < bufferMaxIndex) {
+                       callstack[bufferIndex++] = (uint64_t) esp;
+               }
+       }
+
+       *count = bufferIndex;
+       return kr;
+}
+
+__private_extern__
+kern_return_t
+chudxnu_thread_get_callstack64_kperf(
+       thread_t                thread,
+       uint64_t                *callstack,
+       mach_msg_type_number_t  *count,
+       boolean_t               is_user)
+{
+       return chudxnu_thread_get_callstack64_internal(thread, callstack, count, is_user, !is_user);
+}
+#else /* !__arm__ && !__arm64__ && !__x86_64__ */
+#error kperf: unsupported architecture
+#endif /* !__arm__ && !__arm64__ && !__x86_64__ */